open-consul/agent/consul/connect_ca_endpoint.go

432 lines
12 KiB
Go
Raw Normal View History

package consul
import (
"errors"
"fmt"
"reflect"
"strings"
"time"
"github.com/hashicorp/consul/acl"
"github.com/hashicorp/consul/agent/connect"
"github.com/hashicorp/consul/agent/consul/state"
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/go-memdb"
)
var ErrConnectNotEnabled = errors.New("Connect must be enabled in order to use this endpoint")
// ConnectCA manages the Connect CA.
type ConnectCA struct {
// srv is a pointer back to the server.
srv *Server
}
// 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() {
return acl.ErrPermissionDenied
}
state := s.srv.fsm.State()
_, config, err := state.CAConfig()
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() {
return acl.ErrPermissionDenied
}
// Exit early if it's a no-op change
state := s.srv.fsm.State()
confIdx, config, err := state.CAConfig()
if err != nil {
return err
}
// 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
}
// 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)
}
if err := newProvider.Configure(args.Config.ClusterID, true, args.Config.Config); 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
}
// 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)
2018-04-30 03:44:40 +00:00
s.srv.logger.Printf("[INFO] connect: 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.
// 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
}
// Have the old provider cross-sign the new intermediate
oldProvider, _ := s.srv.getCAProvider()
if oldProvider == nil {
return fmt.Errorf("internal error: CA provider is nil")
}
xcCert, err := oldProvider.CrossSignCA(newRoot)
if err != nil {
return err
}
// Add the cross signed cert to the new root's intermediates.
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.srv.logger.Printf("[WARN] connect: failed to clean up old provider %q", config.Provider)
}
s.srv.logger.Printf("[INFO] connect: CA rotated to new root under provider %q", 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
}
2018-05-08 13:23:44 +00:00
// Load the ClusterID to generate TrustDomain. We do this outside the loop
// since by definition this value should be immutable once set for lifetime of
// the cluster so we don't need to look it up more than once. We also don't
// have to worry about non-atomicity between the config fetch transaction and
// the CARoots transaction below since this field must remain immutable. Do
// not re-use this state/config for other logic that might care about changes
// of config during the blocking query below.
{
state := s.srv.fsm.State()
_, config, err := state.CAConfig()
if err != nil {
return err
}
2018-05-09 16:15:29 +00:00
// Check CA is actually bootstrapped...
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()
}
2018-05-08 13:23:44 +00:00
}
return s.srv.blockingQuery(
&args.QueryOptions, &reply.QueryMeta,
func(ws memdb.WatchSet, state *state.Store) error {
index, roots, err := state.CARoots(ws)
if err != nil {
return err
}
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 a whitelist 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,
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
}
serviceID, ok := spiffeID.(*connect.SpiffeIDService)
if !ok {
return fmt.Errorf("SPIFFE ID in CSR must be a service ID")
}
provider, caRoot := s.srv.getCAProvider()
if provider == nil {
return fmt.Errorf("internal error: CA provider is nil")
}
// Verify that the CSR entity is in the cluster's trust domain
state := s.srv.fsm.State()
_, config, err := state.CAConfig()
if err != nil {
return err
}
signingID := connect.SpiffeIDSigningForCluster(config)
if !signingID.CanSign(serviceID) {
return fmt.Errorf("SPIFFE ID in CSR from a different trust domain: %s, "+
"we are %s", serviceID.Host, signingID.Host())
}
// 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
}
if rule != nil && !rule.ServiceWrite(serviceID.Service, nil) {
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)
}
// All seems to be in order, actually sign it.
2018-04-24 23:31:42 +00:00
pem, err := provider.Sign(csr)
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. Right now we can rely on only
// the built-in provider being supported and the implementation detail that we
// have to write a SerialIndex update to the provider config table for every
// cert issued so in all cases this index will be higher than any previous
// sign response. This has to be reloaded after the provider.Sign call to
// observe the index update.
state = s.srv.fsm.State()
modIdx, _, err := state.CAConfig()
2018-04-24 23:31:42 +00:00
if err != nil {
return err
}
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{
2018-04-24 23:31:42 +00:00
SerialNumber: connect.HexString(cert.SerialNumber.Bytes()),
CertPEM: pem,
Service: serviceID.Service,
2018-04-24 23:31:42 +00:00
ServiceURI: cert.URIs[0].String(),
ValidAfter: cert.NotBefore,
ValidBefore: cert.NotAfter,
RaftIndex: structs.RaftIndex{
ModifyIndex: modIdx,
CreateIndex: modIdx,
},
2018-04-24 23:31:42 +00:00
}
return nil
}