open-consul/agent/consul/acl.go

2239 lines
63 KiB
Go
Raw Normal View History

2014-08-08 22:32:43 +00:00
package consul
import (
"fmt"
"sort"
"sync"
2014-08-08 22:32:43 +00:00
"time"
"github.com/armon/go-metrics"
"github.com/armon/go-metrics/prometheus"
2014-08-08 22:32:43 +00:00
"github.com/hashicorp/consul/acl"
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/consul/logging"
"github.com/hashicorp/go-hclog"
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
"golang.org/x/sync/singleflight"
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
"golang.org/x/time/rate"
2014-08-08 22:52:52 +00:00
)
var ACLCounters = []prometheus.CounterDefinition{
{
Name: []string{"acl", "token", "cache_hit"},
Help: "",
},
{
Name: []string{"acl", "token", "cache_miss"},
Help: "",
},
}
var ACLSummaries = []prometheus.SummaryDefinition{
{
Name: []string{"acl", "resolveTokenLegacy"},
Help: "",
},
{
Name: []string{"acl", "ResolveToken"},
Help: "",
},
{
Name: []string{"acl", "ResolveTokenToIdentity"},
Help: "",
},
}
// These must be kept in sync with the constants in command/agent/acl.go.
2014-08-08 22:52:52 +00:00
const (
// anonymousToken is the token ID we re-write to if there is no token ID
// provided.
anonymousToken = "anonymous"
Creates new "prepared-query" ACL type and new token capture behavior. Prior to this change, prepared queries had the following behavior for ACLs, which will need to change to support templates: 1. A management token, or a token with read access to the service being queried needed to be provided in order to create a prepared query. 2. The token used to create the prepared query was stored with the query in the state store and used to execute the query. 3. A management token, or the token used to create the query needed to be supplied to perform and CRUD operations on an existing prepared query. This was pretty subtle and complicated behavior, and won't work for templates since the service name is computed at execution time. To solve this, we introduce a new "prepared-query" ACL type, where the prefix applies to the query name for static prepared query types and to the prefix for template prepared query types. With this change, the new behavior is: 1. A management token, or a token with "prepared-query" write access to the query name or (soon) the given template prefix is required to do any CRUD operations on a prepared query, or to list prepared queries (the list is filtered by this ACL). 2. You will no longer need a management token to list prepared queries, but you will only be able to see prepared queries that you have access to (you get an empty list instead of permission denied). 3. When listing or getting a query, because it was easy to capture management tokens given the past behavior, this will always blank out the "Token" field (replacing the contents as <hidden>) for all tokens unless a management token is supplied. Going forward, we should discourage people from binding tokens for execution unless strictly necessary. 4. No token will be captured by default when a prepared query is created. If the user wishes to supply an execution token then can pass it in via the "Token" field in the prepared query definition. Otherwise, this field will default to empty. 5. At execution time, we will use the captured token if it exists with the prepared query definition, otherwise we will use the token that's passed in with the request, just like we do for other RPCs (or you can use the agent's configured token for DNS). 6. Prepared queries with no name (accessible only by ID) will not require ACLs to create or modify (execution time will depend on the service ACL configuration). Our argument here is that these are designed to be ephemeral and the IDs are as good as an ACL. Management tokens will be able to list all of these. These changes enable templates, but also enable delegation of authority to manage the prepared query namespace.
2016-02-23 08:12:58 +00:00
// redactedToken is shown in structures with embedded tokens when they
// are not allowed to be displayed.
Creates new "prepared-query" ACL type and new token capture behavior. Prior to this change, prepared queries had the following behavior for ACLs, which will need to change to support templates: 1. A management token, or a token with read access to the service being queried needed to be provided in order to create a prepared query. 2. The token used to create the prepared query was stored with the query in the state store and used to execute the query. 3. A management token, or the token used to create the query needed to be supplied to perform and CRUD operations on an existing prepared query. This was pretty subtle and complicated behavior, and won't work for templates since the service name is computed at execution time. To solve this, we introduce a new "prepared-query" ACL type, where the prefix applies to the query name for static prepared query types and to the prefix for template prepared query types. With this change, the new behavior is: 1. A management token, or a token with "prepared-query" write access to the query name or (soon) the given template prefix is required to do any CRUD operations on a prepared query, or to list prepared queries (the list is filtered by this ACL). 2. You will no longer need a management token to list prepared queries, but you will only be able to see prepared queries that you have access to (you get an empty list instead of permission denied). 3. When listing or getting a query, because it was easy to capture management tokens given the past behavior, this will always blank out the "Token" field (replacing the contents as <hidden>) for all tokens unless a management token is supplied. Going forward, we should discourage people from binding tokens for execution unless strictly necessary. 4. No token will be captured by default when a prepared query is created. If the user wishes to supply an execution token then can pass it in via the "Token" field in the prepared query definition. Otherwise, this field will default to empty. 5. At execution time, we will use the captured token if it exists with the prepared query definition, otherwise we will use the token that's passed in with the request, just like we do for other RPCs (or you can use the agent's configured token for DNS). 6. Prepared queries with no name (accessible only by ID) will not require ACLs to create or modify (execution time will depend on the service ACL configuration). Our argument here is that these are designed to be ephemeral and the IDs are as good as an ACL. Management tokens will be able to list all of these. These changes enable templates, but also enable delegation of authority to manage the prepared query namespace.
2016-02-23 08:12:58 +00:00
redactedToken = "<hidden>"
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// aclUpgradeBatchSize controls how many tokens we look at during each round of upgrading. Individual raft logs
// will be further capped using the aclBatchUpsertSize. This limit just prevents us from creating a single slice
// with all tokens in it.
aclUpgradeBatchSize = 128
// aclUpgradeRateLimit is the number of batch upgrade requests per second allowed.
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
aclUpgradeRateLimit rate.Limit = 1.0
// aclTokenReapingRateLimit is the number of batch token reaping requests per second allowed.
aclTokenReapingRateLimit rate.Limit = 1.0
// aclTokenReapingBurst is the number of batch token reaping requests per second
// that can burst after a period of idleness.
aclTokenReapingBurst = 5
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// aclBatchDeleteSize is the number of deletions to send in a single batch operation. 4096 should produce a batch that is <150KB
// in size but should be sufficiently large to handle 1 replication round in a single batch
aclBatchDeleteSize = 4096
// aclBatchUpsertSize is the target size in bytes we want to submit for a batch upsert request. We estimate the size at runtime
// due to the data being more variable in its size.
aclBatchUpsertSize = 256 * 1024
// DEPRECATED (ACL-Legacy-Compat) aclModeCheck* are all only for legacy usage
// aclModeCheckMinInterval is the minimum amount of time between checking if the
// agent should be using the new or legacy ACL system. All the places it is
// currently used will backoff as it detects that it is remaining in legacy mode.
// However the initial min value is kept small so that new cluster creation
// can enter into new ACL mode quickly.
aclModeCheckMinInterval = 50 * time.Millisecond
// aclModeCheckMaxInterval controls the maximum interval for how often the agent
// checks if it should be using the new or legacy ACL system.
aclModeCheckMaxInterval = 30 * time.Second
// Maximum number of re-resolution requests to be made if the token is modified between
// resolving the token and resolving its policies that would remove one of its policies.
tokenPolicyResolutionMaxRetries = 5
// Maximum number of re-resolution requests to be made if the token is modified between
// resolving the token and resolving its roles that would remove one of its roles.
tokenRoleResolutionMaxRetries = 5
2014-08-08 22:32:43 +00:00
)
// missingIdentity is used to return some identity in the event that the real identity cannot be ascertained
type missingIdentity struct {
reason string
token string
}
func (id *missingIdentity) ID() string {
return id.reason
}
func (id *missingIdentity) SecretToken() string {
return id.token
}
func (id *missingIdentity) PolicyIDs() []string {
return nil
}
func (id *missingIdentity) RoleIDs() []string {
return nil
}
func (id *missingIdentity) EmbeddedPolicy() *structs.ACLPolicy {
return nil
}
func (id *missingIdentity) ServiceIdentityList() []*structs.ACLServiceIdentity {
return nil
}
func (id *missingIdentity) NodeIdentityList() []*structs.ACLNodeIdentity {
return nil
}
func (id *missingIdentity) IsExpired(asOf time.Time) bool {
return false
}
func (id *missingIdentity) IsLocal() bool {
return false
}
func (id *missingIdentity) EnterpriseMetadata() *structs.EnterpriseMeta {
return structs.DefaultEnterpriseMeta()
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func minTTL(a time.Duration, b time.Duration) time.Duration {
if a < b {
return a
}
return b
2014-08-08 22:32:43 +00:00
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
type ACLRemoteError struct {
Err error
}
2014-08-12 17:38:57 +00:00
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func (e ACLRemoteError) Error() string {
return fmt.Sprintf("Error communicating with the ACL Datacenter: %v", e.Err)
}
2014-08-12 17:38:57 +00:00
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func IsACLRemoteError(err error) bool {
_, ok := err.(ACLRemoteError)
return ok
2014-08-08 22:32:43 +00:00
}
func tokenSecretCacheID(token string) string {
return "token-secret:" + token
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
type ACLResolverDelegate interface {
ACLDatacenter(legacy bool) string
UseLegacyACLs() bool
ResolveIdentityFromToken(token string) (bool, structs.ACLIdentity, error)
ResolvePolicyFromID(policyID string) (bool, *structs.ACLPolicy, error)
ResolveRoleFromID(roleID string) (bool, *structs.ACLRole, error)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
RPC(method string, args interface{}, reply interface{}) error
EnterpriseACLResolverDelegate
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
2014-08-08 22:32:43 +00:00
type policyOrRoleTokenError struct {
Err error
token string
}
func (e policyOrRoleTokenError) Error() string {
return e.Err.Error()
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// ACLResolverConfig holds all the configuration necessary to create an ACLResolver
type ACLResolverConfig struct {
Config *Config
Logger hclog.Logger
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// CacheConfig is a pass through configuration for ACL cache limits
CacheConfig *structs.ACLCachesConfig
// Delegate that implements some helper functionality that is server/client specific
Delegate ACLResolverDelegate
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// AutoDisable indicates that RPC responses should be checked and if they indicate ACLs are disabled
// remotely then disable them locally as well. This is particularly useful for the client agent
// so that it can detect when the servers have gotten ACLs enabled.
AutoDisable bool
// ACLConfig is the configuration necessary to pass through to the acl package when creating authorizers
// and when authorizing access
ACLConfig *acl.Config
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
// ACLResolver is the type to handle all your token and policy resolution needs.
//
// Supports:
// - Resolving tokens locally via the ACLResolverDelegate
// - Resolving policies locally via the ACLResolverDelegate
// - Resolving roles locally via the ACLResolverDelegate
// - Resolving legacy tokens remotely via an ACL.GetPolicy RPC
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// - Resolving tokens remotely via an ACL.TokenRead RPC
// - Resolving policies remotely via an ACL.PolicyResolve RPC
// - Resolving roles remotely via an ACL.RoleResolve RPC
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
//
// Remote Resolution:
// Remote resolution can be done synchronously or asynchronously depending
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// on the ACLDownPolicy in the Config passed to the resolver.
//
// When the down policy is set to async-cache and we have already cached values
// then go routines will be spawned to perform the RPCs in the background
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
// and then will update the cache with either the positive or negative result.
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
//
// When the down policy is set to extend-cache or the token/policy/role is not already
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// cached then the same go routines are spawned to do the RPCs in the background.
// However in this mode channels are created to receive the results of the RPC
// and are registered with the resolver. Those channels are immediately read/blocked
// upon.
//
type ACLResolver struct {
config *Config
logger hclog.Logger
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
delegate ACLResolverDelegate
aclConf *acl.Config
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
cache *structs.ACLCaches
identityGroup singleflight.Group
policyGroup singleflight.Group
roleGroup singleflight.Group
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
legacyGroup singleflight.Group
2016-08-04 00:01:32 +00:00
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
down acl.Authorizer
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
autoDisable bool
disabled time.Time
disabledLock sync.RWMutex
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func NewACLResolver(config *ACLResolverConfig) (*ACLResolver, error) {
if config == nil {
return nil, fmt.Errorf("ACL Resolver must be initialized with a config")
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if config.Config == nil {
return nil, fmt.Errorf("ACLResolverConfig.Config must not be nil")
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if config.Delegate == nil {
return nil, fmt.Errorf("ACL Resolver must be initialized with a valid delegate")
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if config.Logger == nil {
config.Logger = hclog.New(&hclog.LoggerOptions{})
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
cache, err := structs.NewACLCaches(config.CacheConfig)
if err != nil {
return nil, err
2014-08-08 22:32:43 +00:00
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
var down acl.Authorizer
switch config.Config.ACLDownPolicy {
case "allow":
down = acl.AllowAll()
case "deny":
down = acl.DenyAll()
case "async-cache", "extend-cache":
// Leave the down policy as nil to signal this.
default:
return nil, fmt.Errorf("invalid ACL down policy %q", config.Config.ACLDownPolicy)
2014-08-08 22:32:43 +00:00
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
return &ACLResolver{
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
config: config.Config,
logger: config.Logger.Named(logging.ACL),
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
delegate: config.Delegate,
aclConf: config.ACLConfig,
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
cache: cache,
autoDisable: config.AutoDisable,
down: down,
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}, nil
}
2014-08-08 22:32:43 +00:00
func (r *ACLResolver) Close() {
r.aclConf.Close()
}
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
func (r *ACLResolver) fetchAndCacheTokenLegacy(token string, cached *structs.AuthorizerCacheEntry) (acl.Authorizer, error) {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
req := structs.ACLPolicyResolveLegacyRequest{
Datacenter: r.delegate.ACLDatacenter(true),
ACL: token,
2014-08-08 22:52:52 +00:00
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
cacheTTL := r.config.ACLTokenTTL
if cached != nil {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
cacheTTL = cached.TTL
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
var reply structs.ACLPolicyResolveLegacyResponse
err := r.delegate.RPC("ACL.GetPolicy", &req, &reply)
2014-08-08 22:52:52 +00:00
if err == nil {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
parent := acl.RootAuthorizer(reply.Parent)
if parent == nil {
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
var authorizer acl.Authorizer
if cached != nil {
authorizer = cached.Authorizer
}
r.cache.PutAuthorizerWithTTL(token, authorizer, cacheTTL)
return authorizer, acl.ErrInvalidParent
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
var policies []*acl.Policy
policy := reply.Policy
if policy != nil {
policies = append(policies, policy.ConvertFromLegacy())
}
authorizer, err := acl.NewPolicyAuthorizerWithDefaults(parent, policies, r.aclConf)
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
r.cache.PutAuthorizerWithTTL(token, authorizer, reply.TTL)
return authorizer, err
2014-08-08 22:52:52 +00:00
}
if acl.IsErrNotFound(err) {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// Make sure to remove from the cache if it was deleted
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
r.cache.PutAuthorizerWithTTL(token, nil, cacheTTL)
return nil, acl.ErrNotFound
2014-08-08 22:52:52 +00:00
}
2014-08-08 22:32:43 +00:00
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// some other RPC error
switch r.config.ACLDownPolicy {
2014-08-08 22:32:43 +00:00
case "allow":
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
r.cache.PutAuthorizerWithTTL(token, acl.AllowAll(), cacheTTL)
return acl.AllowAll(), nil
case "async-cache", "extend-cache":
2014-08-08 22:32:43 +00:00
if cached != nil {
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
r.cache.PutAuthorizerWithTTL(token, cached.Authorizer, cacheTTL)
return cached.Authorizer, nil
2014-08-08 22:32:43 +00:00
}
fallthrough
default:
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
r.cache.PutAuthorizerWithTTL(token, acl.DenyAll(), cacheTTL)
return acl.DenyAll(), nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
}
func (r *ACLResolver) resolveTokenLegacy(token string) (structs.ACLIdentity, acl.Authorizer, error) {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
defer metrics.MeasureSince([]string{"acl", "resolveTokenLegacy"}, time.Now())
// Attempt to resolve locally first (local results are not cached)
// This is only useful for servers where either legacy replication is being
// done or the server is within the primary datacenter.
if done, identity, err := r.delegate.ResolveIdentityFromToken(token); done {
if err == nil && identity != nil {
policies, err := r.resolvePoliciesForIdentity(identity)
if err != nil {
return identity, nil, err
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
authz, err := policies.Compile(r.cache, r.aclConf)
if err != nil {
return identity, nil, err
}
return identity, acl.NewChainedAuthorizer([]acl.Authorizer{authz, acl.RootAuthorizer(r.config.ACLDefaultPolicy)}), nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
return nil, nil, err
}
identity := &missingIdentity{
reason: "legacy-token",
token: token,
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
// Look in the cache prior to making a RPC request
entry := r.cache.GetAuthorizer(token)
if entry != nil && entry.Age() <= minTTL(entry.TTL, r.config.ACLTokenTTL) {
metrics.IncrCounter([]string{"acl", "token", "cache_hit"}, 1)
if entry.Authorizer != nil {
return identity, entry.Authorizer, nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
return identity, nil, acl.ErrNotFound
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
metrics.IncrCounter([]string{"acl", "token", "cache_miss"}, 1)
// Resolve the token in the background and wait on the result if we must
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
waitChan := r.legacyGroup.DoChan(token, func() (interface{}, error) {
authorizer, err := r.fetchAndCacheTokenLegacy(token, entry)
return authorizer, err
})
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
waitForResult := entry == nil || r.config.ACLDownPolicy != "async-cache"
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if !waitForResult {
// waitForResult being false requires the cacheEntry to not be nil
if entry.Authorizer != nil {
return identity, entry.Authorizer, nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
return identity, nil, acl.ErrNotFound
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
// block waiting for the async RPC to finish.
res := <-waitChan
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
var authorizer acl.Authorizer
if res.Val != nil { // avoid a nil-not-nil bug
authorizer = res.Val.(acl.Authorizer)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
return identity, authorizer, res.Err
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
func (r *ACLResolver) fetchAndCacheIdentityFromToken(token string, cached *structs.IdentityCacheEntry) (structs.ACLIdentity, error) {
cacheID := tokenSecretCacheID(token)
req := structs.ACLTokenGetRequest{
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
Datacenter: r.delegate.ACLDatacenter(false),
TokenID: token,
TokenIDType: structs.ACLTokenSecret,
QueryOptions: structs.QueryOptions{
Token: token,
AllowStale: true,
},
}
var resp structs.ACLTokenResponse
err := r.delegate.RPC("ACL.TokenRead", &req, &resp)
if err == nil {
if resp.Token == nil {
r.cache.PutIdentity(cacheID, nil)
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
return nil, acl.ErrNotFound
} else if resp.Token.Local && r.config.Datacenter != resp.SourceDatacenter {
r.cache.PutIdentity(cacheID, nil)
return nil, acl.PermissionDeniedError{Cause: fmt.Sprintf("This is a local token in datacenter %q", resp.SourceDatacenter)}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
} else {
r.cache.PutIdentity(cacheID, resp.Token)
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
return resp.Token, nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
}
if acl.IsErrNotFound(err) {
// Make sure to remove from the cache if it was deleted
r.cache.PutIdentity(cacheID, nil)
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
return nil, acl.ErrNotFound
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// some other RPC error
if cached != nil && (r.config.ACLDownPolicy == "extend-cache" || r.config.ACLDownPolicy == "async-cache") {
// extend the cache
r.cache.PutIdentity(cacheID, cached.Identity)
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
return cached.Identity, nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
r.cache.PutIdentity(cacheID, nil)
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
return nil, err
}
// resolveIdentityFromToken takes a token secret as a string and returns an ACLIdentity.
// We read the value from ACLResolver's cache if available, and if the read misses
// we initiate an RPC for the value.
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func (r *ACLResolver) resolveIdentityFromToken(token string) (structs.ACLIdentity, error) {
// Attempt to resolve locally first (local results are not cached)
if done, identity, err := r.delegate.ResolveIdentityFromToken(token); done {
return identity, err
}
// Check the cache before making any RPC requests
cacheEntry := r.cache.GetIdentity(tokenSecretCacheID(token))
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if cacheEntry != nil && cacheEntry.Age() <= r.config.ACLTokenTTL {
metrics.IncrCounter([]string{"acl", "token", "cache_hit"}, 1)
return cacheEntry.Identity, nil
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
metrics.IncrCounter([]string{"acl", "token", "cache_miss"}, 1)
// Background a RPC request and wait on it if we must
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
waitChan := r.identityGroup.DoChan(token, func() (interface{}, error) {
identity, err := r.fetchAndCacheIdentityFromToken(token, cacheEntry)
return identity, err
})
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
waitForResult := cacheEntry == nil || r.config.ACLDownPolicy != "async-cache"
if !waitForResult {
// waitForResult being false requires the cacheEntry to not be nil
return cacheEntry.Identity, nil
}
// block on the read here, this is why we don't need chan buffering
res := <-waitChan
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
var identity structs.ACLIdentity
if res.Val != nil { // avoid a nil-not-nil bug
identity = res.Val.(structs.ACLIdentity)
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
if res.Err != nil && !acl.IsErrNotFound(res.Err) {
return identity, ACLRemoteError{Err: res.Err}
2014-08-08 22:32:43 +00:00
}
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
return identity, res.Err
2014-08-08 22:32:43 +00:00
}
2014-08-09 00:38:39 +00:00
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
func (r *ACLResolver) fetchAndCachePoliciesForIdentity(identity structs.ACLIdentity, policyIDs []string, cached map[string]*structs.PolicyCacheEntry) (map[string]*structs.ACLPolicy, error) {
req := structs.ACLPolicyBatchGetRequest{
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
Datacenter: r.delegate.ACLDatacenter(false),
PolicyIDs: policyIDs,
QueryOptions: structs.QueryOptions{
Token: identity.SecretToken(),
AllowStale: true,
},
}
var resp structs.ACLPolicyBatchResponse
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
err := r.delegate.RPC("ACL.PolicyResolve", &req, &resp)
if err == nil {
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
out := make(map[string]*structs.ACLPolicy)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
for _, policy := range resp.Policies {
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
out[policy.ID] = policy
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
for _, policyID := range policyIDs {
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
if policy, ok := out[policyID]; ok {
r.cache.PutPolicy(policyID, policy)
} else {
r.cache.PutPolicy(policyID, nil)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
2014-08-09 00:38:39 +00:00
}
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
return out, nil
2014-08-09 00:38:39 +00:00
}
if handledErr := r.maybeHandleIdentityErrorDuringFetch(identity, err); handledErr != nil {
return nil, handledErr
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
// other RPC error - use cache if available
extendCache := r.config.ACLDownPolicy == "extend-cache" || r.config.ACLDownPolicy == "async-cache"
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
out := make(map[string]*structs.ACLPolicy)
insufficientCache := false
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
for _, policyID := range policyIDs {
if entry, ok := cached[policyID]; extendCache && ok {
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
r.cache.PutPolicy(policyID, entry.Policy)
if entry.Policy != nil {
out[policyID] = entry.Policy
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
} else {
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
r.cache.PutPolicy(policyID, nil)
insufficientCache = true
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
}
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
if insufficientCache {
return nil, ACLRemoteError{Err: err}
}
return out, nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
func (r *ACLResolver) fetchAndCacheRolesForIdentity(identity structs.ACLIdentity, roleIDs []string, cached map[string]*structs.RoleCacheEntry) (map[string]*structs.ACLRole, error) {
req := structs.ACLRoleBatchGetRequest{
Datacenter: r.delegate.ACLDatacenter(false),
RoleIDs: roleIDs,
QueryOptions: structs.QueryOptions{
Token: identity.SecretToken(),
AllowStale: true,
},
}
var resp structs.ACLRoleBatchResponse
err := r.delegate.RPC("ACL.RoleResolve", &req, &resp)
if err == nil {
out := make(map[string]*structs.ACLRole)
for _, role := range resp.Roles {
out[role.ID] = role
}
for _, roleID := range roleIDs {
if role, ok := out[roleID]; ok {
r.cache.PutRole(roleID, role)
} else {
r.cache.PutRole(roleID, nil)
}
}
return out, nil
}
if handledErr := r.maybeHandleIdentityErrorDuringFetch(identity, err); handledErr != nil {
return nil, handledErr
}
// other RPC error - use cache if available
extendCache := r.config.ACLDownPolicy == "extend-cache" || r.config.ACLDownPolicy == "async-cache"
out := make(map[string]*structs.ACLRole)
insufficientCache := false
for _, roleID := range roleIDs {
if entry, ok := cached[roleID]; extendCache && ok {
r.cache.PutRole(roleID, entry.Role)
if entry.Role != nil {
out[roleID] = entry.Role
}
} else {
r.cache.PutRole(roleID, nil)
insufficientCache = true
}
}
if insufficientCache {
return nil, ACLRemoteError{Err: err}
}
return out, nil
}
func (r *ACLResolver) maybeHandleIdentityErrorDuringFetch(identity structs.ACLIdentity, err error) error {
if acl.IsErrNotFound(err) {
// make sure to indicate that this identity is no longer valid within
// the cache
r.cache.PutIdentity(tokenSecretCacheID(identity.SecretToken()), nil)
// Do not touch the cache. Getting a top level ACL not found error
// only indicates that the secret token used in the request
// no longer exists
return &policyOrRoleTokenError{acl.ErrNotFound, identity.SecretToken()}
}
if acl.IsErrPermissionDenied(err) {
// invalidate our ID cache so that identity resolution will take place
// again in the future
r.cache.RemoveIdentity(tokenSecretCacheID(identity.SecretToken()))
// Do not remove from the cache for permission denied
// what this does indicate is that our view of the token is out of date
return &policyOrRoleTokenError{acl.ErrPermissionDenied, identity.SecretToken()}
}
return nil
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func (r *ACLResolver) filterPoliciesByScope(policies structs.ACLPolicies) structs.ACLPolicies {
var out structs.ACLPolicies
for _, policy := range policies {
if len(policy.Datacenters) == 0 {
out = append(out, policy)
continue
}
for _, dc := range policy.Datacenters {
if dc == r.config.Datacenter {
out = append(out, policy)
continue
}
}
}
return out
}
func (r *ACLResolver) resolvePoliciesForIdentity(identity structs.ACLIdentity) (structs.ACLPolicies, error) {
policyIDs := identity.PolicyIDs()
roleIDs := identity.RoleIDs()
serviceIdentities := identity.ServiceIdentityList()
nodeIdentities := identity.NodeIdentityList()
if len(policyIDs) == 0 && len(serviceIdentities) == 0 && len(roleIDs) == 0 && len(nodeIdentities) == 0 {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
policy := identity.EmbeddedPolicy()
if policy != nil {
return []*structs.ACLPolicy{policy}, nil
}
// In this case the default policy will be all that is in effect.
return nil, nil
}
// Collect all of the roles tied to this token.
roles, err := r.collectRolesForIdentity(identity, roleIDs)
if err != nil {
return nil, err
}
// Merge the policies and service identities across Token and Role fields.
for _, role := range roles {
for _, link := range role.Policies {
policyIDs = append(policyIDs, link.ID)
}
serviceIdentities = append(serviceIdentities, role.ServiceIdentities...)
nodeIdentities = append(nodeIdentities, role.NodeIdentityList()...)
}
// Now deduplicate any policies or service identities that occur more than once.
policyIDs = dedupeStringSlice(policyIDs)
serviceIdentities = dedupeServiceIdentities(serviceIdentities)
nodeIdentities = dedupeNodeIdentities(nodeIdentities)
// Generate synthetic policies for all service identities in effect.
syntheticPolicies := r.synthesizePoliciesForServiceIdentities(serviceIdentities, identity.EnterpriseMetadata())
syntheticPolicies = append(syntheticPolicies, r.synthesizePoliciesForNodeIdentities(nodeIdentities)...)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// For the new ACLs policy replication is mandatory for correct operation on servers. Therefore
// we only attempt to resolve policies locally
policies, err := r.collectPoliciesForIdentity(identity, policyIDs, len(syntheticPolicies))
if err != nil {
return nil, err
}
policies = append(policies, syntheticPolicies...)
filtered := r.filterPoliciesByScope(policies)
return filtered, nil
}
func (r *ACLResolver) synthesizePoliciesForServiceIdentities(serviceIdentities []*structs.ACLServiceIdentity, entMeta *structs.EnterpriseMeta) []*structs.ACLPolicy {
if len(serviceIdentities) == 0 {
return nil
}
syntheticPolicies := make([]*structs.ACLPolicy, 0, len(serviceIdentities))
for _, s := range serviceIdentities {
syntheticPolicies = append(syntheticPolicies, s.SyntheticPolicy(entMeta))
}
return syntheticPolicies
}
func (r *ACLResolver) synthesizePoliciesForNodeIdentities(nodeIdentities []*structs.ACLNodeIdentity) []*structs.ACLPolicy {
if len(nodeIdentities) == 0 {
return nil
}
syntheticPolicies := make([]*structs.ACLPolicy, 0, len(nodeIdentities))
for _, n := range nodeIdentities {
syntheticPolicies = append(syntheticPolicies, n.SyntheticPolicy())
}
return syntheticPolicies
}
func dedupeServiceIdentities(in []*structs.ACLServiceIdentity) []*structs.ACLServiceIdentity {
// From: https://github.com/golang/go/wiki/SliceTricks#in-place-deduplicate-comparable
if len(in) <= 1 {
return in
}
sort.Slice(in, func(i, j int) bool {
return in[i].ServiceName < in[j].ServiceName
})
j := 0
for i := 1; i < len(in); i++ {
if in[j].ServiceName == in[i].ServiceName {
// Prefer increasing scope.
if len(in[j].Datacenters) == 0 || len(in[i].Datacenters) == 0 {
in[j].Datacenters = nil
} else {
in[j].Datacenters = mergeStringSlice(in[j].Datacenters, in[i].Datacenters)
}
continue
}
j++
in[j] = in[i]
}
// Discard the skipped items.
for i := j + 1; i < len(in); i++ {
in[i] = nil
}
return in[:j+1]
}
func dedupeNodeIdentities(in []*structs.ACLNodeIdentity) []*structs.ACLNodeIdentity {
// From: https://github.com/golang/go/wiki/SliceTricks#in-place-deduplicate-comparable
if len(in) <= 1 {
return in
}
sort.Slice(in, func(i, j int) bool {
if in[i].NodeName < in[j].NodeName {
return true
}
return in[i].Datacenter < in[j].Datacenter
})
j := 0
for i := 1; i < len(in); i++ {
if in[j].NodeName == in[i].NodeName && in[j].Datacenter == in[i].Datacenter {
continue
}
j++
in[j] = in[i]
}
// Discard the skipped items.
for i := j + 1; i < len(in); i++ {
in[i] = nil
}
return in[:j+1]
}
func mergeStringSlice(a, b []string) []string {
out := make([]string, 0, len(a)+len(b))
out = append(out, a...)
out = append(out, b...)
return dedupeStringSlice(out)
}
func dedupeStringSlice(in []string) []string {
// From: https://github.com/golang/go/wiki/SliceTricks#in-place-deduplicate-comparable
if len(in) <= 1 {
return in
}
sort.Strings(in)
j := 0
for i := 1; i < len(in); i++ {
if in[j] == in[i] {
continue
}
j++
in[j] = in[i]
}
return in[:j+1]
}
func (r *ACLResolver) collectPoliciesForIdentity(identity structs.ACLIdentity, policyIDs []string, extraCap int) ([]*structs.ACLPolicy, error) {
policies := make([]*structs.ACLPolicy, 0, len(policyIDs)+extraCap)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// Get all associated policies
var missing []string
var expired []*structs.ACLPolicy
expCacheMap := make(map[string]*structs.PolicyCacheEntry)
var accessorID string
if identity != nil {
accessorID = identity.ID()
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
for _, policyID := range policyIDs {
if done, policy, err := r.delegate.ResolvePolicyFromID(policyID); done {
if err != nil && !acl.IsErrNotFound(err) {
2014-08-12 17:54:56 +00:00
return nil, err
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if policy != nil {
policies = append(policies, policy)
} else {
r.logger.Warn("policy not found for identity",
"policy", policyID,
"accessorID", accessorID,
)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
continue
}
// create the missing list which we can execute an RPC to get all the missing policies at once
entry := r.cache.GetPolicy(policyID)
if entry == nil {
missing = append(missing, policyID)
continue
}
if entry.Policy == nil {
// this happens when we cache a negative response for the policy's existence
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
continue
}
if entry.Age() >= r.config.ACLPolicyTTL {
expired = append(expired, entry.Policy)
expCacheMap[policyID] = entry
} else {
policies = append(policies, entry.Policy)
}
}
// Hot-path if we have no missing or expired policies
if len(missing)+len(expired) == 0 {
return policies, nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
hasMissing := len(missing) > 0
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
fetchIDs := missing
for _, policy := range expired {
fetchIDs = append(fetchIDs, policy.ID)
}
// Background a RPC request and wait on it if we must
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
waitChan := r.policyGroup.DoChan(identity.SecretToken(), func() (interface{}, error) {
policies, err := r.fetchAndCachePoliciesForIdentity(identity, fetchIDs, expCacheMap)
return policies, err
})
2014-08-09 00:38:39 +00:00
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
waitForResult := hasMissing || r.config.ACLDownPolicy != "async-cache"
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if !waitForResult {
// waitForResult being false requires that all the policies were cached already
policies = append(policies, expired...)
return policies, nil
2014-08-09 00:38:39 +00:00
}
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
res := <-waitChan
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
if res.Err != nil {
return nil, res.Err
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
if res.Val != nil {
foundPolicies := res.Val.(map[string]*structs.ACLPolicy)
acl: reduce complexity of token resolution process with alternative singleflighting (#5480) acl: reduce complexity of token resolution process with alternative singleflighting Switches acl resolution to use golang.org/x/sync/singleflight. For the identity/legacy lookups this is a drop-in replacement with the same overall approach to request coalescing. For policies this is technically a change in behavior, but when considered holistically is approximately performance neutral (with the benefit of less code). There are two goals with this blob of code (speaking specifically of policy resolution here): 1) Minimize cross-DC requests. 2) Minimize client-to-server LAN requests. The previous iteration of this code was optimizing for the case of many possibly different tokens being resolved concurrently that have a significant overlap in linked policies such that deduplication would be worth the complexity. While this is laudable there are some things to consider that can help to adjust expectations: 1) For v1.4+ policies are always replicated, and once a single policy shows up in a secondary DC the replicated data is considered authoritative for requests made in that DC. This means that our earlier concerns about minimizing cross-DC requests are irrelevant because there will be no cross-DC policy reads that occur. 2) For Server nodes the in-memory ACL policy cache is capped at zero, meaning it has no caching. Only Client nodes run with a cache. This means that instead of having an entire DC's worth of tokens (what a Server might see) that can have policy resolutions coalesced these nodes will only ever be seeing node-local token resolutions. In a reasonable worst-case scenario where a scheduler like Kubernetes has "filled" a node with Connect services, even that will only schedule ~100 connect services per node. If every service has a unique token there will only be 100 tokens to coalesce and even then those requests have to occur concurrently AND be hitting an empty consul cache. Instead of seeing a great coalescing opportunity for cutting down on redundant Policy resolutions, in practice it's far more likely given node densities that you'd see requests for the same token concurrently than you would for two tokens sharing a policy concurrently (to a degree that would warrant the overhead of the current variation of singleflighting. Given that, this patch switches the Policy resolution process to only singleflight by requesting token (but keeps the cache as by-policy).
2019-03-14 14:35:34 +00:00
for _, policy := range foundPolicies {
policies = append(policies, policy)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
2014-08-09 00:38:39 +00:00
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
return policies, nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
func (r *ACLResolver) resolveRolesForIdentity(identity structs.ACLIdentity) (structs.ACLRoles, error) {
return r.collectRolesForIdentity(identity, identity.RoleIDs())
}
func (r *ACLResolver) collectRolesForIdentity(identity structs.ACLIdentity, roleIDs []string) (structs.ACLRoles, error) {
if len(roleIDs) == 0 {
return nil, nil
}
// For the new ACLs policy & role replication is mandatory for correct operation
// on servers. Therefore we only attempt to resolve roles locally
roles := make([]*structs.ACLRole, 0, len(roleIDs))
var missing []string
var expired []*structs.ACLRole
expCacheMap := make(map[string]*structs.RoleCacheEntry)
for _, roleID := range roleIDs {
if done, role, err := r.delegate.ResolveRoleFromID(roleID); done {
if err != nil && !acl.IsErrNotFound(err) {
return nil, err
}
if role != nil {
roles = append(roles, role)
} else {
var accessorID string
if identity != nil {
accessorID = identity.ID()
}
r.logger.Warn("role not found for identity",
"role", roleID,
"accessorID", accessorID,
)
}
continue
}
// create the missing list which we can execute an RPC to get all the missing roles at once
entry := r.cache.GetRole(roleID)
if entry == nil {
missing = append(missing, roleID)
continue
}
if entry.Role == nil {
// this happens when we cache a negative response for the role's existence
continue
}
if entry.Age() >= r.config.ACLRoleTTL {
expired = append(expired, entry.Role)
expCacheMap[roleID] = entry
} else {
roles = append(roles, entry.Role)
}
}
// Hot-path if we have no missing or expired roles
if len(missing)+len(expired) == 0 {
return roles, nil
}
hasMissing := len(missing) > 0
fetchIDs := missing
for _, role := range expired {
fetchIDs = append(fetchIDs, role.ID)
}
waitChan := r.roleGroup.DoChan(identity.SecretToken(), func() (interface{}, error) {
roles, err := r.fetchAndCacheRolesForIdentity(identity, fetchIDs, expCacheMap)
return roles, err
})
waitForResult := hasMissing || r.config.ACLDownPolicy != "async-cache"
if !waitForResult {
// waitForResult being false requires that all the roles were cached already
roles = append(roles, expired...)
return roles, nil
}
res := <-waitChan
if res.Err != nil {
return nil, res.Err
}
if res.Val != nil {
foundRoles := res.Val.(map[string]*structs.ACLRole)
for _, role := range foundRoles {
roles = append(roles, role)
}
}
return roles, nil
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func (r *ACLResolver) resolveTokenToPolicies(token string) (structs.ACLPolicies, error) {
_, policies, err := r.resolveTokenToIdentityAndPolicies(token)
return policies, err
}
func (r *ACLResolver) resolveTokenToIdentityAndPolicies(token string) (structs.ACLIdentity, structs.ACLPolicies, error) {
var lastErr error
var lastIdentity structs.ACLIdentity
for i := 0; i < tokenPolicyResolutionMaxRetries; i++ {
// Resolve the token to an ACLIdentity
identity, err := r.resolveIdentityFromToken(token)
if err != nil {
return nil, nil, err
} else if identity == nil {
return nil, nil, acl.ErrNotFound
} else if identity.IsExpired(time.Now()) {
return nil, nil, acl.ErrNotFound
}
lastIdentity = identity
policies, err := r.resolvePoliciesForIdentity(identity)
if err == nil {
return identity, policies, nil
}
lastErr = err
if tokenErr, ok := err.(*policyOrRoleTokenError); ok {
if acl.IsErrNotFound(err) && tokenErr.token == identity.SecretToken() {
// token was deleted while resolving policies
return nil, nil, acl.ErrNotFound
}
// other types of policyOrRoleTokenErrors should cause retrying the whole token
// resolution process
} else {
return identity, nil, err
}
}
return lastIdentity, nil, lastErr
}
func (r *ACLResolver) resolveTokenToIdentityAndRoles(token string) (structs.ACLIdentity, structs.ACLRoles, error) {
var lastErr error
var lastIdentity structs.ACLIdentity
for i := 0; i < tokenRoleResolutionMaxRetries; i++ {
// Resolve the token to an ACLIdentity
identity, err := r.resolveIdentityFromToken(token)
if err != nil {
return nil, nil, err
} else if identity == nil {
return nil, nil, acl.ErrNotFound
} else if identity.IsExpired(time.Now()) {
return nil, nil, acl.ErrNotFound
}
lastIdentity = identity
roles, err := r.resolveRolesForIdentity(identity)
if err == nil {
return identity, roles, nil
}
lastErr = err
if tokenErr, ok := err.(*policyOrRoleTokenError); ok {
if acl.IsErrNotFound(err) && tokenErr.token == identity.SecretToken() {
// token was deleted while resolving roles
return nil, nil, acl.ErrNotFound
}
// other types of policyOrRoleTokenErrors should cause retrying the whole token
// resolution process
} else {
return identity, nil, err
}
2014-08-09 00:38:39 +00:00
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
return lastIdentity, nil, lastErr
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
func (r *ACLResolver) disableACLsWhenUpstreamDisabled(err error) error {
if !r.autoDisable || err == nil || !acl.IsErrDisabled(err) {
return err
}
r.logger.Debug("ACLs disabled on upstream servers, will retry", "retry_interval", r.config.ACLDisabledTTL)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
r.disabledLock.Lock()
r.disabled = time.Now().Add(r.config.ACLDisabledTTL)
r.disabledLock.Unlock()
return err
}
func (r *ACLResolver) ResolveTokenToIdentityAndAuthorizer(token string) (structs.ACLIdentity, acl.Authorizer, error) {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if !r.ACLsEnabled() {
return nil, nil, nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
if acl.RootAuthorizer(token) != nil {
return nil, nil, acl.ErrRootDenied
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
// handle the anonymous token
if token == "" {
token = anonymousToken
}
if r.delegate.UseLegacyACLs() {
identity, authorizer, err := r.resolveTokenLegacy(token)
return identity, authorizer, r.disableACLsWhenUpstreamDisabled(err)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
defer metrics.MeasureSince([]string{"acl", "ResolveToken"}, time.Now())
identity, policies, err := r.resolveTokenToIdentityAndPolicies(token)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if err != nil {
r.disableACLsWhenUpstreamDisabled(err)
if IsACLRemoteError(err) {
r.logger.Error("Error resolving token", "error", err)
return &missingIdentity{reason: "primary-dc-down", token: token}, r.down, nil
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
return nil, nil, err
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
// Build the Authorizer
var chain []acl.Authorizer
authz, err := policies.Compile(r.cache, r.aclConf)
if err != nil {
return nil, nil, err
}
chain = append(chain, authz)
authz, err = r.resolveEnterpriseDefaultsForIdentity(identity)
if err != nil {
if IsACLRemoteError(err) {
r.logger.Error("Error resolving identity defaults", "error", err)
return identity, r.down, nil
}
return nil, nil, err
} else if authz != nil {
chain = append(chain, authz)
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
chain = append(chain, acl.RootAuthorizer(r.config.ACLDefaultPolicy))
return identity, acl.NewChainedAuthorizer(chain), nil
}
func (r *ACLResolver) ResolveToken(token string) (acl.Authorizer, error) {
_, authz, err := r.ResolveTokenToIdentityAndAuthorizer(token)
return authz, err
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
func (r *ACLResolver) ResolveTokenToIdentity(token string) (structs.ACLIdentity, error) {
if !r.ACLsEnabled() {
return nil, nil
}
if acl.RootAuthorizer(token) != nil {
return nil, acl.ErrRootDenied
}
// handle the anonymous token
if token == "" {
token = anonymousToken
}
if r.delegate.UseLegacyACLs() {
identity, _, err := r.resolveTokenLegacy(token)
return identity, r.disableACLsWhenUpstreamDisabled(err)
}
defer metrics.MeasureSince([]string{"acl", "ResolveTokenToIdentity"}, time.Now())
return r.resolveIdentityFromToken(token)
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func (r *ACLResolver) ACLsEnabled() bool {
// Whether we desire ACLs to be enabled according to configuration
if !r.config.ACLsEnabled {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
return false
}
if r.autoDisable {
// Whether ACLs are disabled according to RPCs failing with a ACLs Disabled error
r.disabledLock.RLock()
defer r.disabledLock.RUnlock()
return !time.Now().Before(r.disabled)
}
return true
}
func (r *ACLResolver) GetMergedPolicyForToken(token string) (structs.ACLIdentity, *acl.Policy, error) {
ident, policies, err := r.resolveTokenToIdentityAndPolicies(token)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if err != nil {
return nil, nil, err
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
if len(policies) == 0 {
return nil, nil, acl.ErrNotFound
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
policy, err := policies.Merge(r.cache, r.aclConf)
return ident, policy, err
2014-08-09 00:38:39 +00:00
}
// aclFilter is used to filter results from our state store based on ACL rules
// configured for the provided token.
type aclFilter struct {
authorizer acl.Authorizer
logger hclog.Logger
}
2017-04-21 00:02:42 +00:00
// newACLFilter constructs a new aclFilter.
func newACLFilter(authorizer acl.Authorizer, logger hclog.Logger) *aclFilter {
if logger == nil {
logger = hclog.New(&hclog.LoggerOptions{})
}
return &aclFilter{
authorizer: authorizer,
logger: logger,
}
}
// allowNode is used to determine if a node is accessible for an ACL.
func (f *aclFilter) allowNode(node string, ent *acl.AuthorizerContext) bool {
return f.authorizer.NodeRead(node, ent) == acl.Allow
}
// allowNode is used to determine if the gateway and service are accessible for an ACL
func (f *aclFilter) allowGateway(gs *structs.GatewayService) bool {
var authzContext acl.AuthorizerContext
// Need read on service and gateway. Gateway may have different EnterpriseMeta so we fill authzContext twice
gs.Gateway.FillAuthzContext(&authzContext)
if !f.allowService(gs.Gateway.Name, &authzContext) {
return false
}
gs.Service.FillAuthzContext(&authzContext)
if !f.allowService(gs.Service.Name, &authzContext) {
return false
}
return true
}
// allowService is used to determine if a service is accessible for an ACL.
func (f *aclFilter) allowService(service string, ent *acl.AuthorizerContext) bool {
if service == "" {
return true
}
return f.authorizer.ServiceRead(service, ent) == acl.Allow
}
// allowSession is used to determine if a session for a node is accessible for
// an ACL.
func (f *aclFilter) allowSession(node string, ent *acl.AuthorizerContext) bool {
return f.authorizer.SessionRead(node, ent) == acl.Allow
}
// filterHealthChecks is used to filter a set of health checks down based on
// the configured ACL rules for a token.
func (f *aclFilter) filterHealthChecks(checks *structs.HealthChecks) {
hc := *checks
var authzContext acl.AuthorizerContext
for i := 0; i < len(hc); i++ {
check := hc[i]
check.FillAuthzContext(&authzContext)
if f.allowNode(check.Node, &authzContext) && f.allowService(check.ServiceName, &authzContext) {
continue
}
f.logger.Debug("dropping check from result due to ACLs", "check", check.CheckID)
hc = append(hc[:i], hc[i+1:]...)
i--
}
*checks = hc
}
// filterServices is used to filter a set of services based on ACLs.
func (f *aclFilter) filterServices(services structs.Services, entMeta *structs.EnterpriseMeta) {
var authzContext acl.AuthorizerContext
entMeta.FillAuthzContext(&authzContext)
for svc := range services {
if f.allowService(svc, &authzContext) {
continue
}
f.logger.Debug("dropping service from result due to ACLs", "service", svc)
delete(services, svc)
}
}
// filterServiceNodes is used to filter a set of nodes for a given service
// based on the configured ACL rules.
func (f *aclFilter) filterServiceNodes(nodes *structs.ServiceNodes) {
sn := *nodes
var authzContext acl.AuthorizerContext
for i := 0; i < len(sn); i++ {
node := sn[i]
node.FillAuthzContext(&authzContext)
if f.allowNode(node.Node, &authzContext) && f.allowService(node.ServiceName, &authzContext) {
continue
}
f.logger.Debug("dropping node from result due to ACLs", "node", node.Node)
sn = append(sn[:i], sn[i+1:]...)
i--
}
*nodes = sn
}
// filterNodeServices is used to filter services on a given node base on ACLs.
func (f *aclFilter) filterNodeServices(services **structs.NodeServices) {
if *services == nil {
return
}
var authzContext acl.AuthorizerContext
structs.WildcardEnterpriseMeta().FillAuthzContext(&authzContext)
if !f.allowNode((*services).Node.Node, &authzContext) {
*services = nil
return
}
for svcName, svc := range (*services).Services {
svc.FillAuthzContext(&authzContext)
if f.allowNode((*services).Node.Node, &authzContext) && f.allowService(svcName, &authzContext) {
continue
}
f.logger.Debug("dropping service from result due to ACLs", "service", svc.CompoundServiceID())
delete((*services).Services, svcName)
}
}
// filterNodeServices is used to filter services on a given node base on ACLs.
func (f *aclFilter) filterNodeServiceList(services **structs.NodeServiceList) {
if services == nil || *services == nil {
return
}
var authzContext acl.AuthorizerContext
structs.WildcardEnterpriseMeta().FillAuthzContext(&authzContext)
if !f.allowNode((*services).Node.Node, &authzContext) {
*services = nil
return
}
svcs := (*services).Services
modified := false
for i := 0; i < len(svcs); i++ {
svc := svcs[i]
svc.FillAuthzContext(&authzContext)
if f.allowNode((*services).Node.Node, &authzContext) && f.allowService(svc.Service, &authzContext) {
continue
}
f.logger.Debug("dropping service from result due to ACLs", "service", svc.CompoundServiceID())
svcs = append(svcs[:i], svcs[i+1:]...)
i--
modified = true
}
if modified {
*services = &structs.NodeServiceList{
Node: (*services).Node,
Services: svcs,
}
}
}
// filterCheckServiceNodes is used to filter nodes based on ACL rules.
func (f *aclFilter) filterCheckServiceNodes(nodes *structs.CheckServiceNodes) {
csn := *nodes
var authzContext acl.AuthorizerContext
for i := 0; i < len(csn); i++ {
node := csn[i]
node.Service.FillAuthzContext(&authzContext)
if f.allowNode(node.Node.Node, &authzContext) && f.allowService(node.Service.Service, &authzContext) {
continue
}
f.logger.Debug("dropping node from result due to ACLs", "node", node.Node.Node)
csn = append(csn[:i], csn[i+1:]...)
i--
}
*nodes = csn
}
2020-09-29 01:41:47 +00:00
// filterServiceTopology is used to filter upstreams/downstreams based on ACL rules.
// this filter is unlike others in that it also returns whether the result was filtered by ACLs
2020-09-29 01:41:47 +00:00
func (f *aclFilter) filterServiceTopology(topology *structs.ServiceTopology) bool {
numUp := len(topology.Upstreams)
numDown := len(topology.Downstreams)
f.filterCheckServiceNodes(&topology.Upstreams)
f.filterCheckServiceNodes(&topology.Downstreams)
return numUp != len(topology.Upstreams) || numDown != len(topology.Downstreams)
2020-09-29 01:41:47 +00:00
}
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
// filterDatacenterCheckServiceNodes is used to filter nodes based on ACL rules.
func (f *aclFilter) filterDatacenterCheckServiceNodes(datacenterNodes *map[string]structs.CheckServiceNodes) {
dn := *datacenterNodes
out := make(map[string]structs.CheckServiceNodes)
for dc := range dn {
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
nodes := dn[dc]
f.filterCheckServiceNodes(&nodes)
if len(nodes) > 0 {
out[dc] = nodes
}
}
*datacenterNodes = out
}
// filterSessions is used to filter a set of sessions based on ACLs.
func (f *aclFilter) filterSessions(sessions *structs.Sessions) {
s := *sessions
for i := 0; i < len(s); i++ {
session := s[i]
var entCtx acl.AuthorizerContext
session.FillAuthzContext(&entCtx)
if f.allowSession(session.Node, &entCtx) {
continue
}
f.logger.Debug("dropping session from result due to ACLs", "session", session.ID)
s = append(s[:i], s[i+1:]...)
i--
}
*sessions = s
}
// filterCoordinates is used to filter nodes in a coordinate dump based on ACL
// rules.
func (f *aclFilter) filterCoordinates(coords *structs.Coordinates) {
c := *coords
var authzContext acl.AuthorizerContext
structs.WildcardEnterpriseMeta().FillAuthzContext(&authzContext)
for i := 0; i < len(c); i++ {
node := c[i].Node
if f.allowNode(node, &authzContext) {
continue
}
f.logger.Debug("dropping node from result due to ACLs", "node", node)
c = append(c[:i], c[i+1:]...)
i--
}
*coords = c
}
2018-03-04 19:53:52 +00:00
// filterIntentions is used to filter intentions based on ACL rules.
// We prune entries the user doesn't have access to, and we redact any tokens
// if the user doesn't have a management token.
func (f *aclFilter) filterIntentions(ixns *structs.Intentions) {
ret := make(structs.Intentions, 0, len(*ixns))
for _, ixn := range *ixns {
if !ixn.CanRead(f.authorizer) {
f.logger.Debug("dropping intention from result due to ACLs", "intention", ixn.ID)
2018-03-04 19:53:52 +00:00
continue
}
ret = append(ret, ixn)
}
*ixns = ret
}
// filterNodeDump is used to filter through all parts of a node dump and
// remove elements the provided ACL token cannot access.
func (f *aclFilter) filterNodeDump(dump *structs.NodeDump) {
nd := *dump
var authzContext acl.AuthorizerContext
for i := 0; i < len(nd); i++ {
info := nd[i]
// Filter nodes
structs.WildcardEnterpriseMeta().FillAuthzContext(&authzContext)
if node := info.Node; !f.allowNode(node, &authzContext) {
f.logger.Debug("dropping node from result due to ACLs", "node", node)
nd = append(nd[:i], nd[i+1:]...)
i--
continue
}
// Filter services
for j := 0; j < len(info.Services); j++ {
svc := info.Services[j].Service
info.Services[j].FillAuthzContext(&authzContext)
if f.allowNode(info.Node, &authzContext) && f.allowService(svc, &authzContext) {
continue
}
f.logger.Debug("dropping service from result due to ACLs", "service", svc)
info.Services = append(info.Services[:j], info.Services[j+1:]...)
j--
}
// Filter checks
for j := 0; j < len(info.Checks); j++ {
chk := info.Checks[j]
chk.FillAuthzContext(&authzContext)
if f.allowNode(info.Node, &authzContext) && f.allowService(chk.ServiceName, &authzContext) {
continue
}
f.logger.Debug("dropping check from result due to ACLs", "check", chk.CheckID)
info.Checks = append(info.Checks[:j], info.Checks[j+1:]...)
j--
}
}
*dump = nd
}
// filterServiceDump is used to filter nodes based on ACL rules.
func (f *aclFilter) filterServiceDump(services *structs.ServiceDump) {
svcs := *services
var authzContext acl.AuthorizerContext
for i := 0; i < len(svcs); i++ {
service := svcs[i]
if f.allowGateway(service.GatewayService) {
// ServiceDump might only have gateway config and no node information
if service.Node == nil {
continue
}
service.Service.FillAuthzContext(&authzContext)
if f.allowNode(service.Node.Node, &authzContext) {
continue
}
}
f.logger.Debug("dropping service from result due to ACLs", "service", service.GatewayService.Service)
svcs = append(svcs[:i], svcs[i+1:]...)
i--
}
*services = svcs
}
// filterNodes is used to filter through all parts of a node list and remove
// elements the provided ACL token cannot access.
func (f *aclFilter) filterNodes(nodes *structs.Nodes) {
n := *nodes
var authzContext acl.AuthorizerContext
structs.WildcardEnterpriseMeta().FillAuthzContext(&authzContext)
for i := 0; i < len(n); i++ {
node := n[i].Node
if f.allowNode(node, &authzContext) {
continue
}
f.logger.Debug("dropping node from result due to ACLs", "node", node)
n = append(n[:i], n[i+1:]...)
i--
}
*nodes = n
}
// redactPreparedQueryTokens will redact any tokens unless the client has a
// management token. This eases the transition to delegated authority over
// prepared queries, since it was easy to capture management tokens in Consul
// 0.6.3 and earlier, and we don't want to willy-nilly show those. This does
// have the limitation of preventing delegated non-management users from seeing
// captured tokens, but they can at least see whether or not a token is set.
func (f *aclFilter) redactPreparedQueryTokens(query **structs.PreparedQuery) {
// Management tokens can see everything with no filtering.
var authzContext acl.AuthorizerContext
structs.DefaultEnterpriseMeta().FillAuthzContext(&authzContext)
if f.authorizer.ACLWrite(&authzContext) == acl.Allow {
return
}
// Let the user see if there's a blank token, otherwise we need
// to redact it, since we know they don't have a management
// token.
if (*query).Token != "" {
// Redact the token, using a copy of the query structure
// since we could be pointed at a live instance from the
// state store so it's not safe to modify it. Note that
// this clone will still point to things like underlying
// arrays in the original, but for modifying just the
// token it will be safe to use.
clone := *(*query)
clone.Token = redactedToken
*query = &clone
}
}
Creates new "prepared-query" ACL type and new token capture behavior. Prior to this change, prepared queries had the following behavior for ACLs, which will need to change to support templates: 1. A management token, or a token with read access to the service being queried needed to be provided in order to create a prepared query. 2. The token used to create the prepared query was stored with the query in the state store and used to execute the query. 3. A management token, or the token used to create the query needed to be supplied to perform and CRUD operations on an existing prepared query. This was pretty subtle and complicated behavior, and won't work for templates since the service name is computed at execution time. To solve this, we introduce a new "prepared-query" ACL type, where the prefix applies to the query name for static prepared query types and to the prefix for template prepared query types. With this change, the new behavior is: 1. A management token, or a token with "prepared-query" write access to the query name or (soon) the given template prefix is required to do any CRUD operations on a prepared query, or to list prepared queries (the list is filtered by this ACL). 2. You will no longer need a management token to list prepared queries, but you will only be able to see prepared queries that you have access to (you get an empty list instead of permission denied). 3. When listing or getting a query, because it was easy to capture management tokens given the past behavior, this will always blank out the "Token" field (replacing the contents as <hidden>) for all tokens unless a management token is supplied. Going forward, we should discourage people from binding tokens for execution unless strictly necessary. 4. No token will be captured by default when a prepared query is created. If the user wishes to supply an execution token then can pass it in via the "Token" field in the prepared query definition. Otherwise, this field will default to empty. 5. At execution time, we will use the captured token if it exists with the prepared query definition, otherwise we will use the token that's passed in with the request, just like we do for other RPCs (or you can use the agent's configured token for DNS). 6. Prepared queries with no name (accessible only by ID) will not require ACLs to create or modify (execution time will depend on the service ACL configuration). Our argument here is that these are designed to be ephemeral and the IDs are as good as an ACL. Management tokens will be able to list all of these. These changes enable templates, but also enable delegation of authority to manage the prepared query namespace.
2016-02-23 08:12:58 +00:00
// filterPreparedQueries is used to filter prepared queries based on ACL rules.
// We prune entries the user doesn't have access to, and we redact any tokens
// if the user doesn't have a management token.
Creates new "prepared-query" ACL type and new token capture behavior. Prior to this change, prepared queries had the following behavior for ACLs, which will need to change to support templates: 1. A management token, or a token with read access to the service being queried needed to be provided in order to create a prepared query. 2. The token used to create the prepared query was stored with the query in the state store and used to execute the query. 3. A management token, or the token used to create the query needed to be supplied to perform and CRUD operations on an existing prepared query. This was pretty subtle and complicated behavior, and won't work for templates since the service name is computed at execution time. To solve this, we introduce a new "prepared-query" ACL type, where the prefix applies to the query name for static prepared query types and to the prefix for template prepared query types. With this change, the new behavior is: 1. A management token, or a token with "prepared-query" write access to the query name or (soon) the given template prefix is required to do any CRUD operations on a prepared query, or to list prepared queries (the list is filtered by this ACL). 2. You will no longer need a management token to list prepared queries, but you will only be able to see prepared queries that you have access to (you get an empty list instead of permission denied). 3. When listing or getting a query, because it was easy to capture management tokens given the past behavior, this will always blank out the "Token" field (replacing the contents as <hidden>) for all tokens unless a management token is supplied. Going forward, we should discourage people from binding tokens for execution unless strictly necessary. 4. No token will be captured by default when a prepared query is created. If the user wishes to supply an execution token then can pass it in via the "Token" field in the prepared query definition. Otherwise, this field will default to empty. 5. At execution time, we will use the captured token if it exists with the prepared query definition, otherwise we will use the token that's passed in with the request, just like we do for other RPCs (or you can use the agent's configured token for DNS). 6. Prepared queries with no name (accessible only by ID) will not require ACLs to create or modify (execution time will depend on the service ACL configuration). Our argument here is that these are designed to be ephemeral and the IDs are as good as an ACL. Management tokens will be able to list all of these. These changes enable templates, but also enable delegation of authority to manage the prepared query namespace.
2016-02-23 08:12:58 +00:00
func (f *aclFilter) filterPreparedQueries(queries *structs.PreparedQueries) {
var authzContext acl.AuthorizerContext
structs.DefaultEnterpriseMeta().FillAuthzContext(&authzContext)
// Management tokens can see everything with no filtering.
// TODO is this check even necessary - this looks like a search replace from
// the 1.4 ACL rewrite. The global-management token will provide unrestricted query privileges
// so asking for ACLWrite should be unnecessary.
if f.authorizer.ACLWrite(&authzContext) == acl.Allow {
return
}
// Otherwise, we need to see what the token has access to.
Creates new "prepared-query" ACL type and new token capture behavior. Prior to this change, prepared queries had the following behavior for ACLs, which will need to change to support templates: 1. A management token, or a token with read access to the service being queried needed to be provided in order to create a prepared query. 2. The token used to create the prepared query was stored with the query in the state store and used to execute the query. 3. A management token, or the token used to create the query needed to be supplied to perform and CRUD operations on an existing prepared query. This was pretty subtle and complicated behavior, and won't work for templates since the service name is computed at execution time. To solve this, we introduce a new "prepared-query" ACL type, where the prefix applies to the query name for static prepared query types and to the prefix for template prepared query types. With this change, the new behavior is: 1. A management token, or a token with "prepared-query" write access to the query name or (soon) the given template prefix is required to do any CRUD operations on a prepared query, or to list prepared queries (the list is filtered by this ACL). 2. You will no longer need a management token to list prepared queries, but you will only be able to see prepared queries that you have access to (you get an empty list instead of permission denied). 3. When listing or getting a query, because it was easy to capture management tokens given the past behavior, this will always blank out the "Token" field (replacing the contents as <hidden>) for all tokens unless a management token is supplied. Going forward, we should discourage people from binding tokens for execution unless strictly necessary. 4. No token will be captured by default when a prepared query is created. If the user wishes to supply an execution token then can pass it in via the "Token" field in the prepared query definition. Otherwise, this field will default to empty. 5. At execution time, we will use the captured token if it exists with the prepared query definition, otherwise we will use the token that's passed in with the request, just like we do for other RPCs (or you can use the agent's configured token for DNS). 6. Prepared queries with no name (accessible only by ID) will not require ACLs to create or modify (execution time will depend on the service ACL configuration). Our argument here is that these are designed to be ephemeral and the IDs are as good as an ACL. Management tokens will be able to list all of these. These changes enable templates, but also enable delegation of authority to manage the prepared query namespace.
2016-02-23 08:12:58 +00:00
ret := make(structs.PreparedQueries, 0, len(*queries))
for _, query := range *queries {
// If no prefix ACL applies to this query then filter it, since
// we know at this point the user doesn't have a management
// token, otherwise see what the policy says.
prefix, ok := query.GetACLPrefix()
if !ok || f.authorizer.PreparedQueryRead(prefix, &authzContext) != acl.Allow {
f.logger.Debug("dropping prepared query from result due to ACLs", "query", query.ID)
Creates new "prepared-query" ACL type and new token capture behavior. Prior to this change, prepared queries had the following behavior for ACLs, which will need to change to support templates: 1. A management token, or a token with read access to the service being queried needed to be provided in order to create a prepared query. 2. The token used to create the prepared query was stored with the query in the state store and used to execute the query. 3. A management token, or the token used to create the query needed to be supplied to perform and CRUD operations on an existing prepared query. This was pretty subtle and complicated behavior, and won't work for templates since the service name is computed at execution time. To solve this, we introduce a new "prepared-query" ACL type, where the prefix applies to the query name for static prepared query types and to the prefix for template prepared query types. With this change, the new behavior is: 1. A management token, or a token with "prepared-query" write access to the query name or (soon) the given template prefix is required to do any CRUD operations on a prepared query, or to list prepared queries (the list is filtered by this ACL). 2. You will no longer need a management token to list prepared queries, but you will only be able to see prepared queries that you have access to (you get an empty list instead of permission denied). 3. When listing or getting a query, because it was easy to capture management tokens given the past behavior, this will always blank out the "Token" field (replacing the contents as <hidden>) for all tokens unless a management token is supplied. Going forward, we should discourage people from binding tokens for execution unless strictly necessary. 4. No token will be captured by default when a prepared query is created. If the user wishes to supply an execution token then can pass it in via the "Token" field in the prepared query definition. Otherwise, this field will default to empty. 5. At execution time, we will use the captured token if it exists with the prepared query definition, otherwise we will use the token that's passed in with the request, just like we do for other RPCs (or you can use the agent's configured token for DNS). 6. Prepared queries with no name (accessible only by ID) will not require ACLs to create or modify (execution time will depend on the service ACL configuration). Our argument here is that these are designed to be ephemeral and the IDs are as good as an ACL. Management tokens will be able to list all of these. These changes enable templates, but also enable delegation of authority to manage the prepared query namespace.
2016-02-23 08:12:58 +00:00
continue
}
// Redact any tokens if necessary. We make a copy of just the
// pointer so we don't mess with the caller's slice.
final := query
f.redactPreparedQueryTokens(&final)
ret = append(ret, final)
Creates new "prepared-query" ACL type and new token capture behavior. Prior to this change, prepared queries had the following behavior for ACLs, which will need to change to support templates: 1. A management token, or a token with read access to the service being queried needed to be provided in order to create a prepared query. 2. The token used to create the prepared query was stored with the query in the state store and used to execute the query. 3. A management token, or the token used to create the query needed to be supplied to perform and CRUD operations on an existing prepared query. This was pretty subtle and complicated behavior, and won't work for templates since the service name is computed at execution time. To solve this, we introduce a new "prepared-query" ACL type, where the prefix applies to the query name for static prepared query types and to the prefix for template prepared query types. With this change, the new behavior is: 1. A management token, or a token with "prepared-query" write access to the query name or (soon) the given template prefix is required to do any CRUD operations on a prepared query, or to list prepared queries (the list is filtered by this ACL). 2. You will no longer need a management token to list prepared queries, but you will only be able to see prepared queries that you have access to (you get an empty list instead of permission denied). 3. When listing or getting a query, because it was easy to capture management tokens given the past behavior, this will always blank out the "Token" field (replacing the contents as <hidden>) for all tokens unless a management token is supplied. Going forward, we should discourage people from binding tokens for execution unless strictly necessary. 4. No token will be captured by default when a prepared query is created. If the user wishes to supply an execution token then can pass it in via the "Token" field in the prepared query definition. Otherwise, this field will default to empty. 5. At execution time, we will use the captured token if it exists with the prepared query definition, otherwise we will use the token that's passed in with the request, just like we do for other RPCs (or you can use the agent's configured token for DNS). 6. Prepared queries with no name (accessible only by ID) will not require ACLs to create or modify (execution time will depend on the service ACL configuration). Our argument here is that these are designed to be ephemeral and the IDs are as good as an ACL. Management tokens will be able to list all of these. These changes enable templates, but also enable delegation of authority to manage the prepared query namespace.
2016-02-23 08:12:58 +00:00
}
*queries = ret
}
func (f *aclFilter) filterToken(token **structs.ACLToken) {
var entCtx acl.AuthorizerContext
if token == nil || *token == nil || f == nil {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
return
}
(*token).FillAuthzContext(&entCtx)
if f.authorizer.ACLRead(&entCtx) != acl.Allow {
// no permissions to read
*token = nil
} else if f.authorizer.ACLWrite(&entCtx) != acl.Allow {
// no write permissions - redact secret
clone := *(*token)
clone.SecretID = redactedToken
*token = &clone
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
func (f *aclFilter) filterTokens(tokens *structs.ACLTokens) {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
ret := make(structs.ACLTokens, 0, len(*tokens))
for _, token := range *tokens {
final := token
f.filterToken(&final)
if final != nil {
ret = append(ret, final)
}
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
*tokens = ret
}
func (f *aclFilter) filterTokenStub(token **structs.ACLTokenListStub) {
var entCtx acl.AuthorizerContext
if token == nil || *token == nil || f == nil {
return
}
(*token).FillAuthzContext(&entCtx)
if f.authorizer.ACLRead(&entCtx) != acl.Allow {
*token = nil
}
}
func (f *aclFilter) filterTokenStubs(tokens *[]*structs.ACLTokenListStub) {
ret := make(structs.ACLTokenListStubs, 0, len(*tokens))
for _, token := range *tokens {
final := token
f.filterTokenStub(&final)
if final != nil {
ret = append(ret, final)
}
}
*tokens = ret
}
func (f *aclFilter) filterPolicy(policy **structs.ACLPolicy) {
var entCtx acl.AuthorizerContext
if policy == nil || *policy == nil || f == nil {
return
}
(*policy).FillAuthzContext(&entCtx)
if f.authorizer.ACLRead(&entCtx) != acl.Allow {
// no permissions to read
*policy = nil
}
}
func (f *aclFilter) filterPolicies(policies *structs.ACLPolicies) {
ret := make(structs.ACLPolicies, 0, len(*policies))
for _, policy := range *policies {
final := policy
f.filterPolicy(&final)
if final != nil {
ret = append(ret, final)
}
}
*policies = ret
}
func (f *aclFilter) filterRole(role **structs.ACLRole) {
var entCtx acl.AuthorizerContext
if role == nil || *role == nil || f == nil {
return
}
(*role).FillAuthzContext(&entCtx)
if f.authorizer.ACLRead(&entCtx) != acl.Allow {
// no permissions to read
*role = nil
}
}
func (f *aclFilter) filterRoles(roles *structs.ACLRoles) {
ret := make(structs.ACLRoles, 0, len(*roles))
for _, role := range *roles {
final := role
f.filterRole(&final)
if final != nil {
ret = append(ret, final)
}
}
*roles = ret
}
func (f *aclFilter) filterBindingRule(rule **structs.ACLBindingRule) {
var entCtx acl.AuthorizerContext
if rule == nil || *rule == nil || f == nil {
return
}
(*rule).FillAuthzContext(&entCtx)
if f.authorizer.ACLRead(&entCtx) != acl.Allow {
// no permissions to read
*rule = nil
}
}
func (f *aclFilter) filterBindingRules(rules *structs.ACLBindingRules) {
ret := make(structs.ACLBindingRules, 0, len(*rules))
for _, rule := range *rules {
final := rule
f.filterBindingRule(&final)
if final != nil {
ret = append(ret, final)
}
}
*rules = ret
}
func (f *aclFilter) filterAuthMethod(method **structs.ACLAuthMethod) {
var entCtx acl.AuthorizerContext
if method == nil || *method == nil || f == nil {
return
}
(*method).FillAuthzContext(&entCtx)
if f.authorizer.ACLRead(&entCtx) != acl.Allow {
// no permissions to read
*method = nil
}
}
func (f *aclFilter) filterAuthMethods(methods *structs.ACLAuthMethods) {
ret := make(structs.ACLAuthMethods, 0, len(*methods))
for _, method := range *methods {
final := method
f.filterAuthMethod(&final)
if final != nil {
ret = append(ret, final)
}
}
*methods = ret
}
func (f *aclFilter) filterServiceList(services *structs.ServiceList) {
ret := make(structs.ServiceList, 0, len(*services))
for _, svc := range *services {
var authzContext acl.AuthorizerContext
svc.FillAuthzContext(&authzContext)
if f.authorizer.ServiceRead(svc.Name, &authzContext) != acl.Allow {
sid := structs.NewServiceID(svc.Name, &svc.EnterpriseMeta)
f.logger.Debug("dropping service from result due to ACLs", "service", sid.String())
continue
}
ret = append(ret, svc)
}
*services = ret
}
// filterGatewayServices is used to filter gateway to service mappings based on ACL rules.
func (f *aclFilter) filterGatewayServices(mappings *structs.GatewayServices) {
ret := make(structs.GatewayServices, 0, len(*mappings))
for _, s := range *mappings {
// This filter only checks ServiceRead on the linked service.
// ServiceRead on the gateway is checked in the GatewayServices endpoint before filtering.
var authzContext acl.AuthorizerContext
s.Service.FillAuthzContext(&authzContext)
if f.authorizer.ServiceRead(s.Service.Name, &authzContext) != acl.Allow {
f.logger.Debug("dropping service from result due to ACLs", "service", s.Service.String())
continue
}
ret = append(ret, s)
}
*mappings = ret
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func (r *ACLResolver) filterACLWithAuthorizer(authorizer acl.Authorizer, subj interface{}) error {
if authorizer == nil {
return nil
}
// Create the filter
filt := newACLFilter(authorizer, r.logger)
switch v := subj.(type) {
case *structs.CheckServiceNodes:
filt.filterCheckServiceNodes(v)
case *structs.IndexedCheckServiceNodes:
filt.filterCheckServiceNodes(&v.Nodes)
2020-09-29 01:41:47 +00:00
case *structs.IndexedServiceTopology:
filtered := filt.filterServiceTopology(v.ServiceTopology)
if filtered {
v.FilteredByACLs = true
}
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
case *structs.DatacenterIndexedCheckServiceNodes:
filt.filterDatacenterCheckServiceNodes(&v.DatacenterNodes)
case *structs.IndexedCoordinates:
filt.filterCoordinates(&v.Coordinates)
case *structs.IndexedHealthChecks:
filt.filterHealthChecks(&v.HealthChecks)
2018-03-04 19:53:52 +00:00
case *structs.IndexedIntentions:
filt.filterIntentions(&v.Intentions)
case *structs.IndexedNodeDump:
filt.filterNodeDump(&v.Dump)
case *structs.IndexedServiceDump:
filt.filterServiceDump(&v.Dump)
case *structs.IndexedNodes:
filt.filterNodes(&v.Nodes)
case *structs.IndexedNodeServices:
filt.filterNodeServices(&v.NodeServices)
case **structs.NodeServiceList:
filt.filterNodeServiceList(v)
case *structs.IndexedServiceNodes:
filt.filterServiceNodes(&v.ServiceNodes)
Creates new "prepared-query" ACL type and new token capture behavior. Prior to this change, prepared queries had the following behavior for ACLs, which will need to change to support templates: 1. A management token, or a token with read access to the service being queried needed to be provided in order to create a prepared query. 2. The token used to create the prepared query was stored with the query in the state store and used to execute the query. 3. A management token, or the token used to create the query needed to be supplied to perform and CRUD operations on an existing prepared query. This was pretty subtle and complicated behavior, and won't work for templates since the service name is computed at execution time. To solve this, we introduce a new "prepared-query" ACL type, where the prefix applies to the query name for static prepared query types and to the prefix for template prepared query types. With this change, the new behavior is: 1. A management token, or a token with "prepared-query" write access to the query name or (soon) the given template prefix is required to do any CRUD operations on a prepared query, or to list prepared queries (the list is filtered by this ACL). 2. You will no longer need a management token to list prepared queries, but you will only be able to see prepared queries that you have access to (you get an empty list instead of permission denied). 3. When listing or getting a query, because it was easy to capture management tokens given the past behavior, this will always blank out the "Token" field (replacing the contents as <hidden>) for all tokens unless a management token is supplied. Going forward, we should discourage people from binding tokens for execution unless strictly necessary. 4. No token will be captured by default when a prepared query is created. If the user wishes to supply an execution token then can pass it in via the "Token" field in the prepared query definition. Otherwise, this field will default to empty. 5. At execution time, we will use the captured token if it exists with the prepared query definition, otherwise we will use the token that's passed in with the request, just like we do for other RPCs (or you can use the agent's configured token for DNS). 6. Prepared queries with no name (accessible only by ID) will not require ACLs to create or modify (execution time will depend on the service ACL configuration). Our argument here is that these are designed to be ephemeral and the IDs are as good as an ACL. Management tokens will be able to list all of these. These changes enable templates, but also enable delegation of authority to manage the prepared query namespace.
2016-02-23 08:12:58 +00:00
case *structs.IndexedServices:
filt.filterServices(v.Services, &v.EnterpriseMeta)
case *structs.IndexedSessions:
filt.filterSessions(&v.Sessions)
Creates new "prepared-query" ACL type and new token capture behavior. Prior to this change, prepared queries had the following behavior for ACLs, which will need to change to support templates: 1. A management token, or a token with read access to the service being queried needed to be provided in order to create a prepared query. 2. The token used to create the prepared query was stored with the query in the state store and used to execute the query. 3. A management token, or the token used to create the query needed to be supplied to perform and CRUD operations on an existing prepared query. This was pretty subtle and complicated behavior, and won't work for templates since the service name is computed at execution time. To solve this, we introduce a new "prepared-query" ACL type, where the prefix applies to the query name for static prepared query types and to the prefix for template prepared query types. With this change, the new behavior is: 1. A management token, or a token with "prepared-query" write access to the query name or (soon) the given template prefix is required to do any CRUD operations on a prepared query, or to list prepared queries (the list is filtered by this ACL). 2. You will no longer need a management token to list prepared queries, but you will only be able to see prepared queries that you have access to (you get an empty list instead of permission denied). 3. When listing or getting a query, because it was easy to capture management tokens given the past behavior, this will always blank out the "Token" field (replacing the contents as <hidden>) for all tokens unless a management token is supplied. Going forward, we should discourage people from binding tokens for execution unless strictly necessary. 4. No token will be captured by default when a prepared query is created. If the user wishes to supply an execution token then can pass it in via the "Token" field in the prepared query definition. Otherwise, this field will default to empty. 5. At execution time, we will use the captured token if it exists with the prepared query definition, otherwise we will use the token that's passed in with the request, just like we do for other RPCs (or you can use the agent's configured token for DNS). 6. Prepared queries with no name (accessible only by ID) will not require ACLs to create or modify (execution time will depend on the service ACL configuration). Our argument here is that these are designed to be ephemeral and the IDs are as good as an ACL. Management tokens will be able to list all of these. These changes enable templates, but also enable delegation of authority to manage the prepared query namespace.
2016-02-23 08:12:58 +00:00
case *structs.IndexedPreparedQueries:
filt.filterPreparedQueries(&v.Queries)
case **structs.PreparedQuery:
filt.redactPreparedQueryTokens(v)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
case *structs.ACLTokens:
filt.filterTokens(v)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
case **structs.ACLToken:
filt.filterToken(v)
case *[]*structs.ACLTokenListStub:
filt.filterTokenStubs(v)
case **structs.ACLTokenListStub:
filt.filterTokenStub(v)
case *structs.ACLPolicies:
filt.filterPolicies(v)
case **structs.ACLPolicy:
filt.filterPolicy(v)
case *structs.ACLRoles:
filt.filterRoles(v)
case **structs.ACLRole:
filt.filterRole(v)
case *structs.ACLBindingRules:
filt.filterBindingRules(v)
case **structs.ACLBindingRule:
filt.filterBindingRule(v)
case *structs.ACLAuthMethods:
filt.filterAuthMethods(v)
case **structs.ACLAuthMethod:
filt.filterAuthMethod(v)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
case *structs.IndexedServiceList:
filt.filterServiceList(&v.Services)
case *structs.GatewayServices:
filt.filterGatewayServices(v)
default:
panic(fmt.Errorf("Unhandled type passed to ACL filter: %T %#v", subj, subj))
}
return nil
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// filterACL is used to filter results from our service catalog based on the
// rules configured for the provided token.
func (r *ACLResolver) filterACL(token string, subj interface{}) error {
// Get the ACL from the token
authorizer, err := r.ResolveToken(token)
if err != nil {
return err
}
// Fast path if ACLs are not enabled
if authorizer == nil {
return nil
}
return r.filterACLWithAuthorizer(authorizer, subj)
}
// vetRegisterWithACL applies the given ACL's policy to the catalog update and
// determines if it is allowed. Since the catalog register request is so
// dynamic, this is a pretty complex algorithm and was worth breaking out of the
// endpoint. The NodeServices record for the node must be supplied, and can be
// nil.
//
// This is a bit racy because we have to check the state store outside of a
// transaction. It's the best we can do because we don't want to flow ACL
// checking down there. The node information doesn't change in practice, so this
// will be fine. If we expose ways to change node addresses in a later version,
// then we should split the catalog API at the node and service level so we can
// address this race better (even then it would be super rare, and would at
// worst let a service update revert a recent node update, so it doesn't open up
// too much abuse).
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func vetRegisterWithACL(rule acl.Authorizer, subj *structs.RegisterRequest,
ns *structs.NodeServices) error {
// Fast path if ACLs are not enabled.
if rule == nil {
return nil
}
var authzContext acl.AuthorizerContext
subj.FillAuthzContext(&authzContext)
// Vet the node info. This allows service updates to re-post the required
// node info for each request without having to have node "write"
// privileges.
needsNode := ns == nil || subj.ChangesNode(ns.Node)
if needsNode && rule.NodeWrite(subj.Node, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
// Vet the service change. This includes making sure they can register
// the given service, and that we can write to any existing service that
// is being modified by id (if any).
if subj.Service != nil {
if rule.ServiceWrite(subj.Service.Service, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
if ns != nil {
other, ok := ns.Services[subj.Service.ID]
if ok {
// This is effectively a delete, so we DO NOT apply the
// sentinel scope to the service we are overwriting, just
// the regular ACL policy.
var secondaryCtx acl.AuthorizerContext
other.FillAuthzContext(&secondaryCtx)
if rule.ServiceWrite(other.Service, &secondaryCtx) != acl.Allow {
return acl.ErrPermissionDenied
}
}
}
}
// Make sure that the member was flattened before we got there. This
// keeps us from having to verify this check as well.
if subj.Check != nil {
return fmt.Errorf("check member must be nil")
}
// Vet the checks. Node-level checks require node write, and
// service-level checks require service write.
for _, check := range subj.Checks {
// Make sure that the node matches - we don't allow you to mix
// checks from other nodes because we'd have to pull a bunch
// more state store data to check this. If ACLs are enabled then
// we simply require them to match in a given request. There's a
// note in state_store.go to ban this down there in Consul 0.8,
// but it's good to leave this here because it's required for
// correctness wrt. ACLs.
if check.Node != subj.Node {
return fmt.Errorf("Node '%s' for check '%s' doesn't match register request node '%s'",
check.Node, check.CheckID, subj.Node)
}
// Node-level check.
if check.ServiceID == "" {
if rule.NodeWrite(subj.Node, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
continue
}
// Service-level check, check the common case where it
// matches the service part of this request, which has
// already been vetted above, and might be being registered
// along with its checks.
if subj.Service != nil && subj.Service.ID == check.ServiceID {
continue
}
// Service-level check for some other service. Make sure they've
// got write permissions for that service.
if ns == nil {
return fmt.Errorf("Unknown service '%s' for check '%s'", check.ServiceID, check.CheckID)
}
other, ok := ns.Services[check.ServiceID]
if !ok {
return fmt.Errorf("Unknown service '%s' for check '%s'", check.ServiceID, check.CheckID)
}
// We are only adding a check here, so we don't add the scope,
// since the sentinel policy doesn't apply to adding checks at
// this time.
var secondaryCtx acl.AuthorizerContext
other.FillAuthzContext(&secondaryCtx)
if rule.ServiceWrite(other.Service, &secondaryCtx) != acl.Allow {
return acl.ErrPermissionDenied
}
}
return nil
}
// vetDeregisterWithACL applies the given ACL's policy to the catalog update and
// determines if it is allowed. Since the catalog deregister request is so
// dynamic, this is a pretty complex algorithm and was worth breaking out of the
// endpoint. The NodeService for the referenced service must be supplied, and can
// be nil; similar for the HealthCheck for the referenced health check.
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func vetDeregisterWithACL(rule acl.Authorizer, subj *structs.DeregisterRequest,
ns *structs.NodeService, nc *structs.HealthCheck) error {
// Fast path if ACLs are not enabled.
if rule == nil {
return nil
}
// We don't apply sentinel in this path, since at this time sentinel
// only applies to create and update operations.
var authzContext acl.AuthorizerContext
// fill with the defaults for use with the NodeWrite check
subj.FillAuthzContext(&authzContext)
// Allow service deregistration if the token has write permission for the node.
// This accounts for cases where the agent no longer has a token with write permission
// on the service to deregister it.
if rule.NodeWrite(subj.Node, &authzContext) == acl.Allow {
return nil
}
// This order must match the code in applyDeregister() in
// fsm/commands_oss.go since it also evaluates things in this order,
// and will ignore fields based on this precedence. This lets us also
// ignore them from an ACL perspective.
if subj.ServiceID != "" {
if ns == nil {
return fmt.Errorf("Unknown service '%s'", subj.ServiceID)
}
ns.FillAuthzContext(&authzContext)
if rule.ServiceWrite(ns.Service, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
} else if subj.CheckID != "" {
if nc == nil {
return fmt.Errorf("Unknown check '%s'", subj.CheckID)
}
nc.FillAuthzContext(&authzContext)
if nc.ServiceID != "" {
if rule.ServiceWrite(nc.ServiceName, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
} else {
if rule.NodeWrite(subj.Node, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
}
} else {
// Since NodeWrite is not given - otherwise the earlier check
// would've returned already - we can deny here.
return acl.ErrPermissionDenied
}
return nil
}
// vetNodeTxnOp applies the given ACL policy to a node transaction operation.
func vetNodeTxnOp(op *structs.TxnNodeOp, rule acl.Authorizer) error {
2018-12-12 13:22:25 +00:00
// Fast path if ACLs are not enabled.
if rule == nil {
return nil
}
var authzContext acl.AuthorizerContext
op.FillAuthzContext(&authzContext)
2020-08-21 15:42:37 +00:00
if rule.NodeWrite(op.Node.Node, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
return nil
}
// vetServiceTxnOp applies the given ACL policy to a service transaction operation.
func vetServiceTxnOp(op *structs.TxnServiceOp, rule acl.Authorizer) error {
2018-12-12 13:22:25 +00:00
// Fast path if ACLs are not enabled.
if rule == nil {
return nil
}
var authzContext acl.AuthorizerContext
op.FillAuthzContext(&authzContext)
if rule.ServiceWrite(op.Service.Service, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
return nil
}
// vetCheckTxnOp applies the given ACL policy to a check transaction operation.
func vetCheckTxnOp(op *structs.TxnCheckOp, rule acl.Authorizer) error {
2018-12-12 13:22:25 +00:00
// Fast path if ACLs are not enabled.
if rule == nil {
return nil
}
var authzContext acl.AuthorizerContext
op.FillAuthzContext(&authzContext)
if op.Check.ServiceID == "" {
// Node-level check.
if rule.NodeWrite(op.Check.Node, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
} else {
// Service-level check.
if rule.ServiceWrite(op.Check.ServiceName, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
}
return nil
}