The Nomad client expects certain cgroups paths to exist in order to
manage tasks. These paths are created when the agent first starts, but
if process fails the agent would just log the error and proceed with its
initialization, despite not being able to run tasks.
This commit surfaces the errors back to the client initialization so the
process can stop early and make clear to operators that something went
wrong.
Support for fingerprinting the Consul admin partition was added in #19485. But
when the client fingerprints Consul CE, it gets a valid fingerprint and working
Consul but with a warn-level log. Return "ok" from the partition extractor, but
also ensure that we only add the Consul attribute if it actually has a value.
Fixes: https://github.com/hashicorp/nomad/issues/19756
The Nomad client renders templates in the same privileged process used for most
other client operations. During internal testing, we discovered that a malicious
task can create a symlink that can cause template rendering to read and write to
arbitrary files outside the allocation sandbox. Because the Nomad agent can be
restarted without restarting tasks, we can't simply check that the path is safe
at the time we write without encountering a time-of-check/time-of-use race.
To protect Nomad client hosts from this attack, we'll now read and write
templates in a subprocess:
* On Linux/Unix, this subprocess is sandboxed via chroot to the allocation
directory. This requires that Nomad is running as a privileged process. A
non-root Nomad agent will warn that it cannot sandbox the template renderer.
* On Windows, this process is sandboxed via a Windows AppContainer which has
been granted access to only to the allocation directory. This does not require
special privileges on Windows. (Creating symlinks in the first place can be
prevented by running workloads as non-Administrator or
non-ContainerAdministrator users.)
Both sandboxes cause encountered symlinks to be evaluated in the context of the
sandbox, which will result in a "file not found" or "access denied" error,
depending on the platform. This change will also require an update to
Consul-Template to allow callers to inject a custom `ReaderFunc` and
`RenderFunc`.
This design is intended as a workaround to allow us to fix this bug without
creating backwards compatibility issues for running tasks. A future version of
Nomad may introduce a read-only mount specifically for templates and artifacts
so that tasks cannot write into the same location that the Nomad agent is.
Fixes: https://github.com/hashicorp/nomad/issues/19888
Fixes: CVE-2024-1329
During allocation directory migration, the client was not checking that any
symlinks in the archive aren't pointing to somewhere outside the allocation
directory. While task driver sandboxing will protect against processes inside
the task from reading/writing thru the symlink, this doesn't protect against the
client itself from performing unintended operations outside the sandbox.
This changeset includes two changes:
* Update the archive unpacking to check the source of symlinks and require that
they fall within the sandbox.
* Fix a bug in the symlink check where it was using `filepath.Rel` which doesn't
work for paths in the sibling directories of the sandbox directory. This bug
doesn't appear to be exploitable but caused errors in testing.
Fixes: https://github.com/hashicorp/nomad/issues/19887
Add new configuration option on task's volume_mounts, to give a fine grained control over SELinux "z" label
* Update website/content/docs/job-specification/volume_mount.mdx
Co-authored-by: Luiz Aoqui <luiz@hashicorp.com>
* fix: typo
* func: make volume mount verification happen even on mounts with no volume
---------
Co-authored-by: Luiz Aoqui <luiz@hashicorp.com>
Co-authored-by: Tim Gross <tgross@hashicorp.com>
We don't run the whole suite of unit tests on all platforms to keep CI times
reasonable, so the only things we've been running on Windows are
platform-specific.
I'm working on some platform-specific `template` related work and having these
tests run on Windows will reduce the risk of regressions. Our Windows CI box
doesn't have Consul or Vault, so I've skipped those tests for the time being,
and can follow up with that later. There's also a test with assertions looking
for specific paths, and the results are different on Windows. I've skipped those
for the moment as well and will follow up under a separate PR.
Also swap `testify` for `shoenig/test`
Fixes#19781
Do not mark the envoy bootstrap hook as done after successfully running once.
Since the bootstrap file is written to /secrets, which is a tmpfs on supported
platforms, it is not persisted across reboots. This causes the task and
allocation to fail on reboot (see #19781).
This fixes it by *always* rewriting the envoy bootstrap file every time the
Nomad agent starts. This does mean we may write a new bootstrap file to an
already running Envoy task, but in my testing that doesn't have any impact.
This commit doesn't necessarily fix every use of Done by hooks, but hopefully
improves the situation. The comment on Done has been expanded to hopefully
avoid misuse in the future.
Done assertions were removed from tests as they add more noise than value.
*Alternative 1: Use a regular file*
An alternative approach would be to write the bootstrap file somewhere
other than the tmpfs, but this is *unsafe* as when Consul ACLs are
enabled the file will contain a secret token:
https://developer.hashicorp.com/consul/commands/connect/envoy#bootstrap
*Alternative 2: Detect if file is already written*
An alternative approach would be to detect if the bootstrap file exists,
and only write it if it doesn't.
This is just a more complicated form of the current fix. I think in
general in the absence of other factors task hooks should be idempotent
and therefore able to rerun on any agent startup. This simplifies the
code and our ability to reason about task restarts vs agent restarts vs
node reboots by making them all take the same code path.
* client/allocdir: use an interface in place of AllocDir structs
This PR replace *allocdir.AllocDir with allocdir.Interface such that we
may eventually have another implementation of alloc directories. This is
in support of the exec2 driver, which will need an implementation of the
alloc directory incompatibile with the current version.
* use rlock
Some users with batch workloads or short-lived prestart tasks want to derive a
Vaul token, use it, and then allow it to expire without requiring a constant
refresh. Add the `vault.allow_token_expiration` field, which works only with the
Workload Identity workflow and not the legacy workflow.
When set to true, this disables the client's renewal loop in the
`vault_hook`. When Vault revokes the token lease, the token will no longer be
valid. The client will also now automatically detect if the Vault auth
configuration does not allow renewals and will disable the renewal loop
automatically.
Note this should only be used when a secret is requested from Vault once at the
start of a task or in a short-lived prestart task. Long-running tasks should
never set `allow_token_expiration=true` if they obtain Vault secrets via
`template` blocks, as the Vault token will expire and the template runner will
continue to make failing requests to Vault until the `vault_retry` attempts are
exhausted.
Fixes: https://github.com/hashicorp/nomad/issues/8690
* Move group into a separate helper module for reuse
* Add shutdownCh to worker
The shutdown channel is used to signal that worker has stopped.
* Make server shutdown block on workers' shutdownCh
* Fix waiting for eval broker state change blocking indefinitely
There was a race condition in the GenericNotifier between the
Run and WaitForChange functions, where WaitForChange blocks
trying to write to a full unsubscribeCh, but the Run function never
reads from the unsubscribeCh as it has already stopped.
This commit fixes it by unblocking if the notifier has been stopped.
* Bound the amount of time server shutdown waits on worker completion
* Fix lostcancel linter error
* Fix worker test using unexpected worker constructor
* Add changelog
---------
Co-authored-by: Marvin Chin <marvinchin@users.noreply.github.com>
Consul Enterprise agents all belong to an admin partition. Fingerprint this
attribute when available. When a Consul agent is not explicitly configured with
"default" it is in the default partition but will not report this in its
`/v1/agent/self` endpoint. Fallback to "default" when missing only for Consul
Enterprise.
This feature provides users the ability to add constraints for jobs to land on
Nomad nodes that have a Consul in that partition. Or it can allow cluster
administrators to pair Consul partitions 1:1 with Nomad node pools. We'll also
have the option to implement a future `partition` field in the jobspec's
`consul` block to create an implicit constraint.
Ref: https://github.com/hashicorp/nomad/issues/13139#issuecomment-1856479581
* numalib: provide a fallback for topology scanning on linux
* numalib: better package var names
* cl: add cl
* lint: fix my sloppy code
* cl: fixup wording
Unsupported environments like containers or guests OSs inside LXD can
incorrectly number of available cores thus leading to numalib having trouble
detecting cores and panicking. This code adds tests for linux sysfs detection
methods and fixes the panic.
It is often expected that a task that needs access to Vault defines a
`vault` block to specify the Vault policy to use to derive a token.
But in some scenarios, like when the Nomad client is connected to a
local Vault agent that is responsible for authn/authz, the task is not
required to defined a `vault` block.
In these situations, the `default` Vault cluster should be used to
render the template.
If an allocrunner is persisted to the client state but the client stops before
task runner can start, we end up with an allocation in the database with
allocrunner state but no taskrunner state. This ends up mimicking an old
pre-0.9.5 state where this state was not recorded and that hits a backwards
compatibility shim. This leaves allocations in the client state that can never
be restored, but won't ever be removed either.
Update the backwards compatibility shim so that we fail the restore for the
allocrunner and remove the allocation from the client state. Taskrunners persist
state during graceful shutdown, so it shouldn't be possible to leak tasks that
have actually started. This lets us "start over" with the allocation, if the
server still wants to place it on the client.
When a Connect service is registered with Consul, Nomad includes the nested
`Connect.SidecarService` field that includes health checks for the Envoy
proxy. Because these are not part of the job spec, the alloc health tracker
created by `health_hook` doesn't know to read the value of these checks.
In many circumstances this won't be noticed, but if the Envoy health check
happens to take longer than the `update.min_healthy_time` (perhaps because it's
been set low), it's possible for a deployment to progress too early such that
there will briefly be no healthy instances of the service available in Consul.
Update the Consul service client to find the nested sidecar service in the
service catalog and attach it to the results provided to the tracker. The
tracker can then check the sidecar health checks.
Fixes: https://github.com/hashicorp/nomad/issues/19269
This commit introduces the parameter preventRescheduleOnLost which indicates that the task group can't afford to have multiple instances running at the same time. In the case of a node going down, its allocations will be registered as unknown but no replacements will be rescheduled. If the lost node comes back up, the allocs will reconnect and continue to run.
In case of max_client_disconnect also being enabled, if there is a reschedule policy, an error will be returned.
Implements issue #10366
Co-authored-by: Dom Lavery <dom@circleci.com>
Co-authored-by: Tim Gross <tgross@hashicorp.com>
Co-authored-by: Luiz Aoqui <luiz@hashicorp.com>
The `script_check_hook` runs at the task level but can create script checks for
both task-level services and group-level services. Now that we allow the Consul
namespace to be set at the task-level `consul.namespace`, we need to have both
possible namespaces handy when creating and updating checks.
Refactor the JWT token derivation logic to only take a single request
since it was only ever called with a map of length one.
The original implementation received multiple requets to match the
legacy flow, but but legacy flow requests were batched from the Nomad
client to the server, which doesn't happen for JWT. Each JWT request
goes directly from the Nomad client to the Consul agent, so there is no
batching involved.
The task-level service hook is using the group-level method to get the provider
namespace, but this was not designed with task-level `consul` blocks in
mind. This leads to task-level services using the group-level
`consul.namespace`. Fix by creating a method to get the correct namespace and
move this into the service hook itself rather than in the outer `initHooks`
method.
The template hook must use the Consul token for the cluster defined in
the task-level `consul` block or, if `nil, in the group-level `consul`
block.
The Consul tokens are generated by the allocrunner consul hook, but
during the transition period we must fallback to the Nomad agent token
if workload identities are not being used.
So an empty token returned from `GetConsulTokens()` is not enough to
determine if we should use the legacy flow (either this is an old task
or the cluster is not configured for Consul WI), or if there is a
misconfiguration (task or group is `consul` block is using a cluster
that doesn't have an `identity` set).
In order to distinguish between the two scenarios we must iterate over
the task identities looking for one suitable for the Consul cluster
being used.
PRs #19034 and #19040 accidentally conflicted with each other without a merge
conflict when #19034 changes the method signature of `SetConsulTokens`. Because
CI doesn't rebase, both PRs tested fine and only were broken once they landed on
`main`. Fix that.
Add a `Postrun` and `Destroy` hook to the allocrunner's `consul_hook` to ensure
that Consul tokens we've created via WI get revoked via the logout API when
we're done with them. Also add the logout to the `Prerun` hook if we've hit an
error.
The allocrunner has a service registration handler that proxies various API
calls to Consul. With multi-cluster support (for ENT), the service registration
handler is what selects the correct Consul client. The name of this field in the
allocrunner and taskrunner code base looks like it's referring to the actual
Consul API client. This was actually the case before Nomad native service
discovery was implemented, but now the name is misleading.
A series of errors may happen when a token is invalidated while the
Vault client is waiting to renew it. The token may have been invalidated
for several reasons, such as the alloc finished running and it's now
terminal or the token may have been change directly on Vault
out-of-band.
Most of the errors are caused by retries that will never succeed until
Vault fully removes the token from its state.
This commit prevents the retries by making the error `invalid lease ID`
a fatal error.
In earlier versions of Vault, this case was covered by the error `lease
not found or lease is not renewable`, which is already considered to be
a fatal error by Nomad:
2d0cde4ccc/vault/expiration.go (L636-L639)
But https://github.com/hashicorp/vault/pull/5346 introduced an earlier
`nil` check that generates a different error message:
750ab337ea/vault/expiration.go (L1362-L1364)
Both errors happen for the same reason (`le == nil`) and so should be
considered fatal on renewal.
Previously, a Vault token could renewed either periodically via the
renewal loop or immediately by calling `RenewToken()`.
But a race condition in the renewal loop could cause an attempt to renew
an expired token. If both `updateCh` and `renewalCh` are active (such as
when a task stops at the same time its token is waiting for renewal),
the following `select` picks a `case` at random.
78f0c6b2a9/client/vaultclient/vaultclient.go (L557-L564)
If `case <-renewalCh` is picked, the token is incorrectly re-added to
the heap, causing unnecessary renewals of a token that is already expired.
1604dba508/client/vaultclient/vaultclient.go (L505-L510)
To prevent this situation, the `renew()` function should only renew
tokens that are currently in the heap, so `RenewToken()` must first push
the token to the heap and wait for the renewal to happen instead of
calling `renew()` directly since this could cause another race condition
where the token is renewed twice: once by `RenewToken()` calling
`renew()` directly and a second time if the renewal happens to pick the
token as soon as `RenewToken()` adds it to the heap.
Remove the now-unused original configuration blocks for Consul and Vault from
the client. When the client needs to refer to a Consul or Vault block it will
always be for a specific cluster for the task/service. Add a helper for
accessing the default clusters (for the client's own use).
This is two of three changesets for this work. The remainder will implement the
same changes in the `command/agent` package.
As part of this work I discovered and fixed two bugs:
* The gRPC proxy socket that we create for Envoy is only ever created using the
default Consul cluster's configuration. This will prevent Connect from being
used with the non-default cluster.
* The Consul configuration we use for templates always comes from the default
Consul cluster's configuration, but will use the correct Consul token for the
non-default cluster. This will prevent templates from being used with the
non-default cluster.
Ref: https://github.com/hashicorp/nomad/issues/18947
Ref: https://github.com/hashicorp/nomad/pull/18991
Fixes: https://github.com/hashicorp/nomad/issues/18984
Fixes: https://github.com/hashicorp/nomad/issues/18983
This PR tweaks the linux cpu fingerprinter to handle the case where no
NUMA node data is found under /sys/devices/system/, in which case we
need to assume just one node, one socket.
The template hook emits an error when the task has a Consul block that requires
WI but there's no WI. The exact error message we get depends on whether we're
running in CE or ENT. Update the test assertion so that we can tolerate this
difference without building ENT-specific test files.
When looking up the Consul or Vault cluster from a client hook, we should always
use an accessor function rather than trying to lookup the `Cluster` field, which
may be empty for jobs registered before Nomad 1.7.
* identity: support change_mode and change_signal
wip - just jobspec portion
* test struct
* cleanup some insignificant boogs
* actually implement change mode
* docs tweaks
* add changelog
* test identity.change_mode operations
* use more words in changelog
* job endpoint tests
* address comments from code review
---------
Co-authored-by: Tim Gross <tgross@hashicorp.com>
Allocations that were created before Nomad 1.7 will not have the cluster field
set for their Consul blocks. While this can be corrected server-side, that
doesn't help allocations already on clients.
Allocations that were created before Nomad 1.7 will not have the `cluster` field
set for their Vault blocks. While this can be corrected server-side, that
doesn't help allocations already on clients.
Also add extra safety on Consul cluster lookup too
OIDC mandates the support of the RS256 signing algorithm so in order to maximize workload identity's usefulness this change switches from using the EdDSA signing algorithm to RS256.
Old keys will continue to use EdDSA but new keys will use RS256. The EdDSA generation code was left in place because it's fast and cheap and I'm not going to lie I hope we get to use it again.
**Test Updates**
Most of our Variables and Keyring tests had a subtle assumption in them that the keyring would be initialized by the time the test server had elected a leader. ed25519 key generation is so fast that the fact that it was happening asynchronously with server startup didn't seem to cause problems. Sadly rsa key generation is so slow that basically all of these tests failed.
I added a new `testutil.WaitForKeyring` helper to replace `testutil.WaitForLeader` in cases where the keyring must be initialized before the test may continue. However this is mostly used in the `nomad/` package.
In the `api` and `command/agent` packages I decided to switch their helpers to wait for keyring initialization by default. This will slow down tests a bit, but allow those packages to not be as concerned with subtle server readiness details. On my machine rsa key generation takes 63ms, so hopefully the difference isn't significant on CI runners.
**TODO**
- Docs and changelog entries.
- Upgrades - right now upgrades won't get RS256 keys until their root key rotates either manually or after ~30 days.
- Observability - I'm not sure there's a way for operators to see if they're using EdDSA or RS256 unless they inspect a key. The JWKS endpoint can be inspected to see if EdDSA will be used for new identities, but it doesn't technically define which key is active. If upgrades can be fixed to automatically rotate keys, we probably don't need to worry about this.
**Requiem for ed25519**
When workload identities were first implemented we did not immediately consider OIDC compliance. Consul, Vault, and many other third parties support JWT auth methods without full OIDC compliance. For the machine<-->machine use cases workload identity is intended to fulfill, OIDC seemed like a bigger risk than asset.
EdDSA/ed25519 is the signing algorithm we chose for workload identity JWTs because of all these lovely properties:
1. Deterministic keys that can be derived from our preexisting root keys. This was perhaps the biggest factor since we already had a root encryption key around from which we could derive a signing key.
2. Wonderfully compact: 64 byte private key, 32 byte public key, 64 byte signatures. Just glorious.
3. No parameters. No choices of encodings. It's all well-defined by [RFC 8032](https://datatracker.ietf.org/doc/html/rfc8032).
4. Fastest performing signing algorithm! We don't even care that much about the performance of our chosen algorithm, but what a free bonus!
5. Arguably one of the most secure signing algorithms widely available. Not just from a cryptanalysis perspective, but from an API and usage perspective too.
Life was good with ed25519, but sadly it could not last.
[IDPs](https://en.wikipedia.org/wiki/Identity_provider), such as AWS's IAM OIDC Provider, love OIDC. They have OIDC implemented for humans, so why not reuse that OIDC support for machines as well? Since OIDC mandates RS256, many implementations don't bother implementing other signing algorithms (or at least not advertising their support). A quick survey of OIDC Discovery endpoints revealed only 2 out of 10 OIDC providers advertised support for anything other than RS256:
- [PayPal](https://www.paypalobjects.com/.well-known/openid-configuration) supports HS256
- [Yahoo](https://api.login.yahoo.com/.well-known/openid-configuration) supports ES256
RS256 only:
- [GitHub](https://token.actions.githubusercontent.com/.well-known/openid-configuration)
- [GitLab](https://gitlab.com/.well-known/openid-configuration)
- [Google](https://accounts.google.com/.well-known/openid-configuration)
- [Intuit](https://developer.api.intuit.com/.well-known/openid_configuration)
- [Microsoft](https://login.microsoftonline.com/fabrikamb2c.onmicrosoft.com/v2.0/.well-known/openid-configuration)
- [SalesForce](https://login.salesforce.com/.well-known/openid-configuration)
- [SimpleLogin (acquired by ProtonMail)](https://app.simplelogin.io/.well-known/openid-configuration/)
- [TFC](https://app.terraform.io/.well-known/openid-configuration)
The `group_service_hook` needs to supply the Consul service client with Consul
tokens for its services. The lookup in the hook resources was looking for the
wrong key. This would cause the service client to ignore the Consul token we've
received and use the agent's own token.
This changeset also moves the prefix formatting into `MakeUniqueIdentityName` method
to reduce the risk of this kind of bug in the future.
