open-consul/agent/consul/connect_ca_endpoint.go

633 lines
19 KiB
Go
Raw Normal View History

package consul
import (
"context"
"errors"
"fmt"
"net/url"
"reflect"
"strings"
"time"
"github.com/hashicorp/go-hclog"
"golang.org/x/time/rate"
"github.com/hashicorp/consul/acl"
"github.com/hashicorp/consul/agent/connect"
"github.com/hashicorp/consul/agent/connect/ca"
"github.com/hashicorp/consul/agent/consul/state"
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/go-memdb"
)
var (
// Err strings. net/rpc doesn't have a way to transport typed/rich errors so
// we currently rely on sniffing the error string in a few cases where we need
// to change client behavior. These are the canonical error strings to use.
// Note though that client code can't use `err == consul.Err*` directly since
// the error returned by RPC will be a plain error.errorString created by
// net/rpc client so will not be the same _instance_ that this package
// variable points to. Clients need to compare using `err.Error() ==
// consul.ErrRateLimited.Error()` which is very sad. Short of replacing our
// RPC mechanism it's hard to know how to make that much better though.
ErrConnectNotEnabled = errors.New("Connect must be enabled in order to use this endpoint")
ErrRateLimited = errors.New("Rate limit reached, try again later")
ErrNotPrimaryDatacenter = errors.New("not the primary datacenter")
ErrStateReadOnly = errors.New("CA Provider State is read-only")
)
const (
// csrLimitWait is the maximum time we'll wait for a slot when CSR concurrency
// limiting or rate limiting is occurring. It's intentionally short so small
// batches of requests can be accommodated when server has capacity (assuming
// signing one cert takes much less than this) but failing requests fast when
// a thundering herd comes along.
csrLimitWait = 500 * time.Millisecond
)
// ConnectCA manages the Connect CA.
type ConnectCA struct {
// srv is a pointer back to the server.
srv *Server
logger hclog.Logger
}
// ConfigurationGet returns the configuration for the CA.
func (s *ConnectCA) ConfigurationGet(
args *structs.DCSpecificRequest,
reply *structs.CAConfiguration) error {
// Exit early if Connect hasn't been enabled.
if !s.srv.config.ConnectEnabled {
return ErrConnectNotEnabled
}
if done, err := s.srv.forward("ConnectCA.ConfigurationGet", args, args, reply); done {
return err
}
// This action requires operator read access.
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
rule, err := s.srv.ResolveToken(args.Token)
if err != nil {
return err
}
if rule != nil && rule.OperatorRead(nil) != acl.Allow {
return acl.ErrPermissionDenied
}
state := s.srv.fsm.State()
_, config, err := state.CAConfig(nil)
if err != nil {
return err
}
*reply = *config
return nil
}
// ConfigurationSet updates the configuration for the CA.
func (s *ConnectCA) ConfigurationSet(
args *structs.CARequest,
reply *interface{}) error {
// Exit early if Connect hasn't been enabled.
if !s.srv.config.ConnectEnabled {
return ErrConnectNotEnabled
}
if done, err := s.srv.forward("ConnectCA.ConfigurationSet", args, args, reply); done {
return err
}
// This action requires operator write access.
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
rule, err := s.srv.ResolveToken(args.Token)
if err != nil {
return err
}
if rule != nil && rule.OperatorWrite(nil) != acl.Allow {
return acl.ErrPermissionDenied
}
// Exit early if it's a no-op change
state := s.srv.fsm.State()
confIdx, config, err := state.CAConfig(nil)
if err != nil {
return err
}
// Don't allow state changes. Either it needs to be empty or the same to allow
// read-modify-write loops that don't touch the State field.
if len(args.Config.State) > 0 &&
!reflect.DeepEqual(args.Config.State, config.State) {
return ErrStateReadOnly
}
// Don't allow users to change the ClusterID.
args.Config.ClusterID = config.ClusterID
if args.Config.Provider == config.Provider && reflect.DeepEqual(args.Config.Config, config.Config) {
return nil
}
// If the provider hasn't changed, we need to load the current Provider state
// so it can decide if it needs to change resources or not based on the config
// change.
if args.Config.Provider == config.Provider {
// Note this is a shallow copy since the State method doc requires the
// provider return a map that will not be further modified and should not
// modify the one we pass to Configure.
args.Config.State = config.State
}
// Create a new instance of the provider described by the config
// and get the current active root CA. This acts as a good validation
// of the config and makes sure the provider is functioning correctly
// before we commit any changes to Raft.
newProvider, err := s.srv.createCAProvider(args.Config)
if err != nil {
return fmt.Errorf("could not initialize provider: %v", err)
}
pCfg := ca.ProviderConfig{
ClusterID: args.Config.ClusterID,
Datacenter: s.srv.config.Datacenter,
// This endpoint can be called in a secondary DC too so set this correctly.
IsPrimary: s.srv.config.Datacenter == s.srv.config.PrimaryDatacenter,
RawConfig: args.Config.Config,
State: args.Config.State,
}
if err := newProvider.Configure(pCfg); err != nil {
return fmt.Errorf("error configuring provider: %v", err)
}
if err := newProvider.GenerateRoot(); err != nil {
return fmt.Errorf("error generating CA root certificate: %v", err)
}
newRootPEM, err := newProvider.ActiveRoot()
if err != nil {
return err
}
newActiveRoot, err := parseCARoot(newRootPEM, args.Config.Provider, args.Config.ClusterID)
if err != nil {
2018-05-04 23:01:54 +00:00
return err
}
// See if the provider needs to persist any state along with the config
pState, err := newProvider.State()
if err != nil {
return fmt.Errorf("error getting provider state: %v", err)
}
args.Config.State = pState
// Compare the new provider's root CA ID to the current one. If they
// match, just update the existing provider with the new config.
// If they don't match, begin the root rotation process.
_, root, err := state.CARootActive(nil)
if err != nil {
return err
}
// If the root didn't change or if this is a secondary DC, just update the
// config and return.
if (s.srv.config.Datacenter != s.srv.config.PrimaryDatacenter) ||
root != nil && root.ID == newActiveRoot.ID {
args.Op = structs.CAOpSetConfig
resp, err := s.srv.raftApply(structs.ConnectCARequestType, args)
if err != nil {
return err
}
if respErr, ok := resp.(error); ok {
return respErr
}
// If the config has been committed, update the local provider instance
s.srv.setCAProvider(newProvider, newActiveRoot)
s.logger.Info("CA provider config updated")
return nil
}
// At this point, we know the config change has trigged a root rotation,
// either by swapping the provider type or changing the provider's config
// to use a different root certificate.
// First up, sanity check that the current provider actually supports
// cross-signing.
oldProvider, _ := s.srv.getCAProvider()
if oldProvider == nil {
return fmt.Errorf("internal error: CA provider is nil")
}
canXSign, err := oldProvider.SupportsCrossSigning()
if err != nil {
return fmt.Errorf("CA provider error: %s", err)
}
if !canXSign && !args.Config.ForceWithoutCrossSigning {
return errors.New("The current CA Provider does not support cross-signing. " +
"You can try again with ForceWithoutCrossSigningSet but this may cause " +
"disruption - see documentation for more.")
}
if !canXSign && args.Config.ForceWithoutCrossSigning {
s.logger.Warn("current CA doesn't support cross signing but " +
"CA reconfiguration forced anyway with ForceWithoutCrossSigning")
}
// If it's a config change that would trigger a rotation (different provider/root):
// 1. Get the root from the new provider.
// 2. Call CrossSignCA on the old provider to sign the new root with the old one to
// get a cross-signed certificate.
// 3. Take the active root for the new provider and append the intermediate from step 2
// to its list of intermediates.
newRoot, err := connect.ParseCert(newRootPEM)
if err != nil {
return err
}
if canXSign {
// Have the old provider cross-sign the new root
xcCert, err := oldProvider.CrossSignCA(newRoot)
if err != nil {
return err
}
// Add the cross signed cert to the new CA's intermediates (to be attached
// to leaf certs).
newActiveRoot.IntermediateCerts = []string{xcCert}
}
intermediate, err := newProvider.GenerateIntermediate()
if err != nil {
return err
}
if intermediate != newRootPEM {
newActiveRoot.IntermediateCerts = append(newActiveRoot.IntermediateCerts, intermediate)
}
// Update the roots and CA config in the state store at the same time
idx, roots, err := state.CARoots(nil)
if err != nil {
return err
}
var newRoots structs.CARoots
for _, r := range roots {
newRoot := *r
if newRoot.Active {
newRoot.Active = false
newRoot.RotatedOutAt = time.Now()
}
newRoots = append(newRoots, &newRoot)
}
newRoots = append(newRoots, newActiveRoot)
args.Op = structs.CAOpSetRootsAndConfig
args.Index = idx
args.Config.ModifyIndex = confIdx
args.Roots = newRoots
resp, err := s.srv.raftApply(structs.ConnectCARequestType, args)
if err != nil {
return err
}
if respErr, ok := resp.(error); ok {
return respErr
}
if respOk, ok := resp.(bool); ok && !respOk {
return fmt.Errorf("could not atomically update roots and config")
}
// If the config has been committed, update the local provider instance
// and call teardown on the old provider
s.srv.setCAProvider(newProvider, newActiveRoot)
if err := oldProvider.Cleanup(); err != nil {
s.logger.Warn("failed to clean up old provider", "provider", config.Provider)
}
s.logger.Info("CA rotated to new root under provider", "provider", args.Config.Provider)
return nil
}
// Roots returns the currently trusted root certificates.
func (s *ConnectCA) Roots(
args *structs.DCSpecificRequest,
reply *structs.IndexedCARoots) error {
// Forward if necessary
if done, err := s.srv.forward("ConnectCA.Roots", args, args, reply); done {
return err
}
// Exit early if Connect hasn't been enabled.
if !s.srv.config.ConnectEnabled {
return ErrConnectNotEnabled
}
return s.srv.blockingQuery(
&args.QueryOptions, &reply.QueryMeta,
func(ws memdb.WatchSet, state *state.Store) error {
index, roots, config, err := state.CARootsAndConfig(ws)
if err != nil {
return err
}
if config != nil {
// Build TrustDomain based on the ClusterID stored.
signingID := connect.SpiffeIDSigningForCluster(config)
if signingID == nil {
// If CA is bootstrapped at all then this should never happen but be
// defensive.
return errors.New("no cluster trust domain setup")
}
reply.TrustDomain = signingID.Host()
}
reply.Index, reply.Roots = index, roots
if reply.Roots == nil {
reply.Roots = make(structs.CARoots, 0)
}
// The API response must NEVER contain the secret information
// such as keys and so on. We use an allowlist below to copy the
// specific fields we want to expose.
for i, r := range reply.Roots {
// IMPORTANT: r must NEVER be modified, since it is a pointer
// directly to the structure in the memdb store.
reply.Roots[i] = &structs.CARoot{
ID: r.ID,
Name: r.Name,
SerialNumber: r.SerialNumber,
SigningKeyID: r.SigningKeyID,
ExternalTrustDomain: r.ExternalTrustDomain,
NotBefore: r.NotBefore,
NotAfter: r.NotAfter,
RootCert: r.RootCert,
IntermediateCerts: r.IntermediateCerts,
RaftIndex: r.RaftIndex,
Active: r.Active,
PrivateKeyType: r.PrivateKeyType,
PrivateKeyBits: r.PrivateKeyBits,
2018-03-20 17:36:05 +00:00
}
if r.Active {
reply.ActiveRootID = r.ID
}
}
return nil
},
)
}
// Sign signs a certificate for a service.
func (s *ConnectCA) Sign(
args *structs.CASignRequest,
reply *structs.IssuedCert) error {
// Exit early if Connect hasn't been enabled.
if !s.srv.config.ConnectEnabled {
return ErrConnectNotEnabled
}
if done, err := s.srv.forward("ConnectCA.Sign", args, args, reply); done {
return err
}
// Parse the CSR
csr, err := connect.ParseCSR(args.CSR)
if err != nil {
return err
}
// Parse the SPIFFE ID
spiffeID, err := connect.ParseCertURI(csr.URIs[0])
if err != nil {
return err
}
provider, caRoot := s.srv.getCAProvider()
if provider == nil {
return fmt.Errorf("internal error: CA provider is nil")
} else if caRoot == nil {
return fmt.Errorf("internal error: CA root is nil")
}
// Verify that the CSR entity is in the cluster's trust domain
state := s.srv.fsm.State()
_, config, err := state.CAConfig(nil)
if err != nil {
return err
}
signingID := connect.SpiffeIDSigningForCluster(config)
serviceID, isService := spiffeID.(*connect.SpiffeIDService)
agentID, isAgent := spiffeID.(*connect.SpiffeIDAgent)
if !isService && !isAgent {
return fmt.Errorf("SPIFFE ID in CSR must be a service or agent ID")
}
if isService {
if !signingID.CanSign(spiffeID) {
return fmt.Errorf("SPIFFE ID in CSR from a different trust domain: %s, "+
"we are %s", serviceID.Host, signingID.Host())
}
} else {
// isAgent - if we support more ID types then this would need to be an else if
// here we are just automatically fixing the trust domain. For auto-encrypt and
// auto-config they make certificate requests before learning about the roots
// so they will have a dummy trust domain in the CSR.
trustDomain := signingID.Host()
if agentID.Host != trustDomain {
originalURI := agentID.URI()
agentID.Host = trustDomain
csr.Subject.CommonName = connect.AgentCN(agentID.Agent, trustDomain)
// recreate the URIs list
uris := make([]*url.URL, len(csr.URIs))
for i, uri := range csr.URIs {
if originalURI.String() == uri.String() {
uris[i] = agentID.URI()
} else {
uris[i] = uri
}
}
csr.URIs = uris
}
}
// Verify that the ACL token provided has permission to act as this service
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
rule, err := s.srv.ResolveToken(args.Token)
if err != nil {
return err
}
var authzContext acl.AuthorizerContext
var entMeta structs.EnterpriseMeta
if isService {
entMeta.Merge(serviceID.GetEnterpriseMeta())
entMeta.FillAuthzContext(&authzContext)
if rule != nil && rule.ServiceWrite(serviceID.Service, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
// Verify that the DC in the service URI matches us. We might relax this
// requirement later but being restrictive for now is safer.
if serviceID.Datacenter != s.srv.config.Datacenter {
return fmt.Errorf("SPIFFE ID in CSR from a different datacenter: %s, "+
"we are %s", serviceID.Datacenter, s.srv.config.Datacenter)
}
} else if isAgent {
structs.DefaultEnterpriseMeta().FillAuthzContext(&authzContext)
if rule != nil && rule.NodeWrite(agentID.Agent, &authzContext) != acl.Allow {
return acl.ErrPermissionDenied
}
}
commonCfg, err := config.GetCommonConfig()
if err != nil {
return err
}
if commonCfg.CSRMaxPerSecond > 0 {
lim := s.srv.caLeafLimiter.getCSRRateLimiterWithLimit(rate.Limit(commonCfg.CSRMaxPerSecond))
// Wait up to the small threshold we allow for a token.
ctx, cancel := context.WithTimeout(context.Background(), csrLimitWait)
defer cancel()
if lim.Wait(ctx) != nil {
return ErrRateLimited
}
} else if commonCfg.CSRMaxConcurrent > 0 {
s.srv.caLeafLimiter.csrConcurrencyLimiter.SetSize(int64(commonCfg.CSRMaxConcurrent))
ctx, cancel := context.WithTimeout(context.Background(), csrLimitWait)
defer cancel()
if err := s.srv.caLeafLimiter.csrConcurrencyLimiter.Acquire(ctx); err != nil {
return ErrRateLimited
}
defer s.srv.caLeafLimiter.csrConcurrencyLimiter.Release()
}
// All seems to be in order, actually sign it.
2018-04-24 23:31:42 +00:00
pem, err := provider.Sign(csr)
if err == ca.ErrRateLimited {
return ErrRateLimited
}
2018-04-24 23:31:42 +00:00
if err != nil {
return err
}
// Append any intermediates needed by this root.
for _, p := range caRoot.IntermediateCerts {
pem = strings.TrimSpace(pem) + "\n" + p
}
// Append our local CA's intermediate if there is one.
inter, err := provider.ActiveIntermediate()
if err != nil {
return err
}
root, err := provider.ActiveRoot()
if err != nil {
return err
}
if inter != root {
pem = strings.TrimSpace(pem) + "\n" + inter
}
// TODO(banks): when we implement IssuedCerts table we can use the insert to
// that as the raft index to return in response.
//
// UPDATE(mkeeler): The original implementation relied on updating the CAConfig
// and using its index as the ModifyIndex for certs. This was buggy. The long
// term goal is still to insert some metadata into raft about the certificates
// and use that raft index for the ModifyIndex. This is a partial step in that
// direction except that we only are setting an index and not storing the
// metadata.
req := structs.CALeafRequest{
Op: structs.CALeafOpIncrementIndex,
Datacenter: s.srv.config.Datacenter,
WriteRequest: structs.WriteRequest{Token: args.Token},
}
resp, err := s.srv.raftApply(structs.ConnectCALeafRequestType|structs.IgnoreUnknownTypeFlag, &req)
2018-04-24 23:31:42 +00:00
if err != nil {
return err
}
modIdx, ok := resp.(uint64)
if !ok {
return fmt.Errorf("Invalid response from updating the leaf cert index")
}
2018-04-24 23:31:42 +00:00
cert, err := connect.ParseCert(pem)
if err != nil {
return err
}
// Set the response
2018-04-30 03:44:40 +00:00
*reply = structs.IssuedCert{
SerialNumber: connect.EncodeSerialNumber(cert.SerialNumber),
CertPEM: pem,
ValidAfter: cert.NotBefore,
ValidBefore: cert.NotAfter,
EnterpriseMeta: entMeta,
RaftIndex: structs.RaftIndex{
ModifyIndex: modIdx,
CreateIndex: modIdx,
},
2018-04-24 23:31:42 +00:00
}
if isService {
reply.Service = serviceID.Service
reply.ServiceURI = cert.URIs[0].String()
} else if isAgent {
reply.Agent = agentID.Agent
reply.AgentURI = cert.URIs[0].String()
}
return nil
}
// SignIntermediate signs an intermediate certificate for a remote datacenter.
func (s *ConnectCA) SignIntermediate(
args *structs.CASignRequest,
reply *string) error {
// Exit early if Connect hasn't been enabled.
if !s.srv.config.ConnectEnabled {
return ErrConnectNotEnabled
}
if done, err := s.srv.forward("ConnectCA.SignIntermediate", args, args, reply); done {
return err
}
// Verify we are allowed to serve this request
if s.srv.config.PrimaryDatacenter != s.srv.config.Datacenter {
return ErrNotPrimaryDatacenter
}
// This action requires operator write access.
rule, err := s.srv.ResolveToken(args.Token)
if err != nil {
return err
}
if rule != nil && rule.OperatorWrite(nil) != acl.Allow {
return acl.ErrPermissionDenied
}
provider, _ := s.srv.getCAProvider()
if provider == nil {
return fmt.Errorf("internal error: CA provider is nil")
}
csr, err := connect.ParseCSR(args.CSR)
if err != nil {
return err
}
cert, err := provider.SignIntermediate(csr)
if err != nil {
return err
}
*reply = cert
return nil
}