open-vault/builtin/logical/aws/path_user.go
Joel Thompson ac18a44fae secret/aws: Pass policy ARNs to AssumedRole and FederationToken roles (#6789)
* secret/aws: Pass policy ARNs to AssumedRole and FederationToken roles

AWS now allows you to pass policy ARNs as well as, and in addition to,
policy documents for AssumeRole and GetFederationToken (see
https://aws.amazon.com/about-aws/whats-new/2019/05/session-permissions/).
Vault already collects policy ARNs for iam_user credential types; now it
will allow policy ARNs for assumed_role and federation_token credential
types and plumb them through to the appropriate AWS calls.

This brings along a minor breaking change. Vault roles of the
federation_token credential type are now required to have either a
policy_document or a policy_arns specified. This was implicit
previously; a missing policy_document would result in a validation error
from the AWS SDK when retrieving credentials. However, it would still
allow creating a role that didn't have a policy_document specified and
then later specifying it, after which retrieving the AWS credentials
would work. Similar workflows in which the Vault role didn't have a
policy_document specified for some period of time, such as deleting the
policy_document and then later adding it back, would also have worked
previously but will now be broken.

The reason for this breaking change is because a credential_type of
federation_token without either a policy_document or policy_arns
specified will return credentials that have equivalent permissions to
the credentials the Vault server itself is using. This is quite
dangerous (e.g., it could allow Vault clients access to retrieve
credentials that could modify Vault's underlying storage) and so should
be discouraged. This scenario is still possible when passing in an
appropriate policy_document or policy_arns parameter, but clients should
be explicitly aware of what they are doing and opt in to it by passing
in the appropriate role parameters.

* Error out on dangerous federation token retrieval

The AWS secrets role code now disallows creation of a dangerous role
configuration; however, pre-existing roles could have existed that would
trigger this now-dangerous code path, so also adding a check for this
configuration at credential retrieval time.

* Run makefmt

* Fix tests

* Fix comments/docs
2019-08-20 12:34:41 -07:00

292 lines
9.1 KiB
Go

package aws
import (
"context"
"fmt"
"strings"
"time"
"github.com/aws/aws-sdk-go/aws"
"github.com/aws/aws-sdk-go/aws/awserr"
"github.com/aws/aws-sdk-go/service/iam"
"github.com/hashicorp/errwrap"
"github.com/hashicorp/vault/sdk/framework"
"github.com/hashicorp/vault/sdk/helper/strutil"
"github.com/hashicorp/vault/sdk/logical"
"github.com/mitchellh/mapstructure"
)
func pathUser(b *backend) *framework.Path {
return &framework.Path{
Pattern: "(creds|sts)/" + framework.GenericNameRegex("name"),
Fields: map[string]*framework.FieldSchema{
"name": &framework.FieldSchema{
Type: framework.TypeString,
Description: "Name of the role",
},
"role_arn": &framework.FieldSchema{
Type: framework.TypeString,
Description: "ARN of role to assume when credential_type is " + assumedRoleCred,
},
"ttl": &framework.FieldSchema{
Type: framework.TypeDurationSecond,
Description: "Lifetime of the returned credentials in seconds",
Default: 3600,
},
},
Callbacks: map[logical.Operation]framework.OperationFunc{
logical.ReadOperation: b.pathCredsRead,
logical.UpdateOperation: b.pathCredsRead,
},
HelpSynopsis: pathUserHelpSyn,
HelpDescription: pathUserHelpDesc,
}
}
func (b *backend) pathCredsRead(ctx context.Context, req *logical.Request, d *framework.FieldData) (*logical.Response, error) {
roleName := d.Get("name").(string)
// Read the policy
role, err := b.roleRead(ctx, req.Storage, roleName, true)
if err != nil {
return nil, errwrap.Wrapf("error retrieving role: {{err}}", err)
}
if role == nil {
return logical.ErrorResponse(fmt.Sprintf(
"Role '%s' not found", roleName)), nil
}
var ttl int64
ttlRaw, ok := d.GetOk("ttl")
switch {
case ok:
ttl = int64(ttlRaw.(int))
case role.DefaultSTSTTL > 0:
ttl = int64(role.DefaultSTSTTL.Seconds())
default:
ttl = int64(d.Get("ttl").(int))
}
var maxTTL int64
if role.MaxSTSTTL > 0 {
maxTTL = int64(role.MaxSTSTTL.Seconds())
} else {
maxTTL = int64(b.System().MaxLeaseTTL().Seconds())
}
if ttl > maxTTL {
ttl = maxTTL
}
roleArn := d.Get("role_arn").(string)
var credentialType string
switch {
case len(role.CredentialTypes) == 1:
credentialType = role.CredentialTypes[0]
// There is only one way for the CredentialTypes to contain more than one entry, and that's an upgrade path
// where it contains iamUserCred and federationTokenCred
// This ambiguity can be resolved based on req.Path, so resolve it assuming CredentialTypes only has those values
case len(role.CredentialTypes) > 1:
if strings.HasPrefix(req.Path, "creds") {
credentialType = iamUserCred
} else {
credentialType = federationTokenCred
}
// sanity check on the assumption above
if !strutil.StrListContains(role.CredentialTypes, credentialType) {
return logical.ErrorResponse(fmt.Sprintf("requested credential type %q not in allowed credential types %#v", credentialType, role.CredentialTypes)), nil
}
}
// creds requested through the sts path shouldn't be allowed to get iamUserCred type creds
// when the role is created from legacy data because they might have more privileges in AWS.
// See https://github.com/hashicorp/vault/issues/4229#issuecomment-380316788 for details.
if role.ProhibitFlexibleCredPath {
if credentialType == iamUserCred && strings.HasPrefix(req.Path, "sts") {
return logical.ErrorResponse(fmt.Sprintf("attempted to retrieve %s credentials through the sts path; this is not allowed for legacy roles", iamUserCred)), nil
}
if credentialType != iamUserCred && strings.HasPrefix(req.Path, "creds") {
return logical.ErrorResponse(fmt.Sprintf("attempted to retrieve %s credentials through the creds path; this is not allowed for legacy roles", credentialType)), nil
}
}
switch credentialType {
case iamUserCred:
return b.secretAccessKeysCreate(ctx, req.Storage, req.DisplayName, roleName, role)
case assumedRoleCred:
switch {
case roleArn == "":
if len(role.RoleArns) != 1 {
return logical.ErrorResponse("did not supply a role_arn parameter and unable to determine one"), nil
}
roleArn = role.RoleArns[0]
case !strutil.StrListContains(role.RoleArns, roleArn):
return logical.ErrorResponse(fmt.Sprintf("role_arn %q not in allowed role arns for Vault role %q", roleArn, roleName)), nil
}
return b.assumeRole(ctx, req.Storage, req.DisplayName, roleName, roleArn, role.PolicyDocument, role.PolicyArns, ttl)
case federationTokenCred:
return b.getFederationToken(ctx, req.Storage, req.DisplayName, roleName, role.PolicyDocument, role.PolicyArns, ttl)
default:
return logical.ErrorResponse(fmt.Sprintf("unknown credential_type: %q", credentialType)), nil
}
}
func (b *backend) pathUserRollback(ctx context.Context, req *logical.Request, _kind string, data interface{}) error {
var entry walUser
if err := mapstructure.Decode(data, &entry); err != nil {
return err
}
username := entry.UserName
// Get the client
client, err := b.clientIAM(ctx, req.Storage)
if err != nil {
return err
}
// Get information about this user
groupsResp, err := client.ListGroupsForUser(&iam.ListGroupsForUserInput{
UserName: aws.String(username),
MaxItems: aws.Int64(1000),
})
if err != nil {
// This isn't guaranteed to be perfect; for example, an IAM user
// might have gotten put into the WAL but then the IAM user creation
// failed (e.g., Vault didn't have permissions) and then the WAL
// deletion failed as well. Then, if Vault doesn't have access to
// call iam:ListGroupsForUser, AWS will return an access denied error
// and the WAL will never get cleaned up. But this is better than
// just having Vault "forget" about a user it actually created.
//
// BEWARE a potential race condition -- where this is called
// immediately after a user is created. AWS eventual consistency
// might say the user doesn't exist when the user does in fact
// exist, and this could cause Vault to forget about the user.
// This won't happen if the user creation fails (because the WAL
// minimum age is 5 minutes, and AWS eventual consistency is, in
// practice, never that long), but it could happen if a lease holder
// asks immediately after getting a user to revoke the lease, causing
// Vault to leak the secret, which would be a Very Bad Thing to allow.
// So we make sure that, if there's an associated lease, it must be at
// least 5 minutes old as well.
if aerr, ok := err.(awserr.Error); ok {
acceptMissingIamUsers := false
if req.Secret == nil || time.Since(req.Secret.IssueTime) > time.Duration(minAwsUserRollbackAge) {
// WAL rollback
acceptMissingIamUsers = true
}
if aerr.Code() == iam.ErrCodeNoSuchEntityException && acceptMissingIamUsers {
return nil
}
}
return err
}
groups := groupsResp.Groups
// Inline (user) policies
policiesResp, err := client.ListUserPolicies(&iam.ListUserPoliciesInput{
UserName: aws.String(username),
MaxItems: aws.Int64(1000),
})
if err != nil {
return err
}
policies := policiesResp.PolicyNames
// Attached managed policies
manPoliciesResp, err := client.ListAttachedUserPolicies(&iam.ListAttachedUserPoliciesInput{
UserName: aws.String(username),
MaxItems: aws.Int64(1000),
})
if err != nil {
return err
}
manPolicies := manPoliciesResp.AttachedPolicies
keysResp, err := client.ListAccessKeys(&iam.ListAccessKeysInput{
UserName: aws.String(username),
MaxItems: aws.Int64(1000),
})
if err != nil {
return err
}
keys := keysResp.AccessKeyMetadata
// Revoke all keys
for _, k := range keys {
_, err = client.DeleteAccessKey(&iam.DeleteAccessKeyInput{
AccessKeyId: k.AccessKeyId,
UserName: aws.String(username),
})
if err != nil {
return err
}
}
// Detach managed policies
for _, p := range manPolicies {
_, err = client.DetachUserPolicy(&iam.DetachUserPolicyInput{
UserName: aws.String(username),
PolicyArn: p.PolicyArn,
})
if err != nil {
return err
}
}
// Delete any inline (user) policies
for _, p := range policies {
_, err = client.DeleteUserPolicy(&iam.DeleteUserPolicyInput{
UserName: aws.String(username),
PolicyName: p,
})
if err != nil {
return err
}
}
// Remove the user from all their groups
for _, g := range groups {
_, err = client.RemoveUserFromGroup(&iam.RemoveUserFromGroupInput{
GroupName: g.GroupName,
UserName: aws.String(username),
})
if err != nil {
return err
}
}
// Delete the user
_, err = client.DeleteUser(&iam.DeleteUserInput{
UserName: aws.String(username),
})
if err != nil {
return err
}
return nil
}
type walUser struct {
UserName string
}
const pathUserHelpSyn = `
Generate AWS credentials from a specific Vault role.
`
const pathUserHelpDesc = `
This path will generate new, never before used AWS credentials for
accessing AWS. The IAM policy used to back this key pair will be
the "name" parameter. For example, if this backend is mounted at "aws",
then "aws/creds/deploy" would generate access keys for the "deploy" role.
The access keys will have a lease associated with them. The access keys
can be revoked by using the lease ID when using the iam_user credential type.
When using AWS STS credential types (assumed_role or federation_token),
revoking the lease does not revoke the access keys.
`