Changes made to both components:
1. Logs are always flushed on process exit
2. Informer cache sync can no longer hang process start up forever
Changes made to concierge:
1. Add pre-shutdown hook that waits for controllers to exit cleanly
2. Informer caches are synced in post-start hook
Changes made to supervisor:
1. Add shutdown code that waits for controllers to exit cleanly
2. Add shutdown code that waits for active connections to become idle
Waiting for controllers to exit cleanly is critical as this allows
the leader election logic to release the lock on exit. This reduces
the time needed for the next leader to be elected.
Signed-off-by: Monis Khan <mok@vmware.com>
The kubelet will send the SIGTERM signal when it wants a process to
exit. After a grace period, it will send the SIGKILL signal to
force the process to terminate. The concierge has always handled
both SIGINT and SIGTERM as indicators for it to gracefully exit
(i.e. stop watches, controllers, etc). This change updates the
supervisor to do the same (previously it only handled SIGINT). This
is required to allow the leader election lock release logic to run.
Otherwise it can take a few minutes for new pods to acquire the
lease since they believe it is already held.
Signed-off-by: Monis Khan <mok@vmware.com>
For clusters where the control plane nodes aren't running a CNI, the
kube-cert-agent pods deployed by concierge cannot be scheduled as they
don't know to use `hostNetwork: true`. This change allows embedding the
host network setting in the Concierge configuration. (by copying it from
the kube-controller-manager pod spec when generating the kube-cert-agent
Deployment)
Also fixed a stray double comma in one of the nearby tests.
OpenShift has good defaults for these duration fields that we can
use instead of coming up with them ourselves:
e14e06ba8d/pkg/config/leaderelection/leaderelection.go (L87-L109)
Copied here for easy future reference:
// We want to be able to tolerate 60s of kube-apiserver disruption without causing pod restarts.
// We want the graceful lease re-acquisition fairly quick to avoid waits on new deployments and other rollouts.
// We want a single set of guidance for nearly every lease in openshift. If you're special, we'll let you know.
// 1. clock skew tolerance is leaseDuration-renewDeadline == 30s
// 2. kube-apiserver downtime tolerance is == 78s
// lastRetry=floor(renewDeadline/retryPeriod)*retryPeriod == 104
// downtimeTolerance = lastRetry-retryPeriod == 78s
// 3. worst non-graceful lease acquisition is leaseDuration+retryPeriod == 163s
// 4. worst graceful lease acquisition is retryPeriod == 26s
if ret.LeaseDuration.Duration == 0 {
ret.LeaseDuration.Duration = 137 * time.Second
}
if ret.RenewDeadline.Duration == 0 {
// this gives 107/26=4 retries and allows for 137-107=30 seconds of clock skew
// if the kube-apiserver is unavailable for 60s starting just before t=26 (the first renew),
// then we will retry on 26s intervals until t=104 (kube-apiserver came back up at 86), and there will
// be 33 seconds of extra time before the lease is lost.
ret.RenewDeadline.Duration = 107 * time.Second
}
if ret.RetryPeriod.Duration == 0 {
ret.RetryPeriod.Duration = 26 * time.Second
}
Signed-off-by: Monis Khan <mok@vmware.com>
This change fixes a small race condition that occurred when the
current leader failed to renew its lease. Before this change, the
leader would first release the lease via the Kube API and then would
update its in-memory status to reflect that change. Now those
events occur in the reverse (i.e. correct) order.
Signed-off-by: Monis Khan <mok@vmware.com>
- Change list of attributeParsingOverrides to a map
- Add unit test for sAMAccountName as group name without the override
- Change some comments in the the type definition.
The Kube API server code that we use will cast inputs in an attempt
to see if they implement optional interfaces. This change adds a
simple wrapper struct to prevent such casts from causing us any
issues.
Signed-off-by: Monis Khan <mok@vmware.com>
In the upstream dynamiccertificates package, we rely on two pieces
of code:
1. DynamicServingCertificateController.newTLSContent which calls
- clientCA.CurrentCABundleContent
- servingCert.CurrentCertKeyContent
2. unionCAContent.VerifyOptions which calls
- unionCAContent.CurrentCABundleContent
This results in calls to our tlsServingCertDynamicCertProvider and
impersonationSigningCertProvider. If we Unset these providers, we
subtly break these consumers. At best this results in test slowness
and flakes while we wait for reconcile loops to converge. At worst,
it results in actual errors during runtime. For example, we
previously would Unset the impersonationSigningCertProvider on any
sync loop error (even a transient one caused by a network blip or
a conflict between writes from different replicas of the concierge).
This would cause us to transiently fail to issue new certificates
from the token credential require API. It would also cause us to
transiently fail to authenticate previously issued client certs
(which results in occasional Unauthorized errors in CI).
Signed-off-by: Monis Khan <mok@vmware.com>
- Add `AllowPasswordGrant` boolean field to OIDCIdentityProvider's spec
- The oidc upstream watcher controller copies the value of
`AllowPasswordGrant` into the configuration of the cached provider
- Add password grant to the UpstreamOIDCIdentityProviderI interface
which is implemented by the cached provider instance for use in the
authorization endpoint
- Enhance the IDP discovery endpoint to return the supported "flows"
for each IDP ("cli_password" and/or "browser_authcode")
- Enhance `pinniped get kubeconfig` to help the user choose the desired
flow for the selected IDP, and to write the flow into the resulting
kubeconfg
- Enhance `pinniped login oidc` to have a flow flag to tell it which
client-side flow it should use for auth (CLI-based or browser-based)
- In the Dex config, allow the resource owner password grant, which Dex
implements to also return ID tokens, for use in integration tests
- Enhance the authorize endpoint to perform password grant when
requested by the incoming headers. This commit does not include unit
tests for the enhancements to the authorize endpoint, which will come
in the next commit
- Extract some shared helpers from the callback endpoint to share the
code with the authorize endpoint
- Add new integration tests
At a high level, it switches us to a distroless base container image, but that also includes several related bits:
- Add a writable /tmp but make the rest of our filesystems read-only at runtime.
- Condense our main server binaries into a single pinniped-server binary. This saves a bunch of space in
the image due to duplicated library code. The correct behavior is dispatched based on `os.Args[0]`, and
the `pinniped-server` binary is symlinked to `pinniped-concierge` and `pinniped-supervisor`.
- Strip debug symbols from our binaries. These aren't really useful in a distroless image anyway and all the
normal stuff you'd expect to work, such as stack traces, still does.
- Add a separate `pinniped-concierge-kube-cert-agent` binary with "sleep" and "print" functionality instead of
using builtin /bin/sleep and /bin/cat for the kube-cert-agent. This is split from the main server binary
because the loading/init time of the main server binary was too large for the tiny resource footprint we
established in our kube-cert-agent PodSpec. Using a separate binary eliminates this issue and the extra
binary adds only around 1.5MiB of image size.
- Switch the kube-cert-agent code to use a JSON `{"tls.crt": "<b64 cert>", "tls.key": "<b64 key>"}` format.
This is more robust to unexpected input formatting than the old code, which simply concatenated the files
with some extra newlines and split on whitespace.
- Update integration tests that made now-invalid assumptions about the `pinniped-server` image.
Signed-off-by: Matt Moyer <moyerm@vmware.com>
Prior to this fix, this controller did not correctly react to changes to the ClusterIP service. It would still eventually react with a long delay due to our 5 minute resync interval.
Signed-off-by: Matt Moyer <moyerm@vmware.com>
This change fixes a race that can occur because we have multiple
writers with no leader election lock.
1. TestAPIServingCertificateAutoCreationAndRotation/automatic
expires the current serving certificate
2. CertsExpirerController 1 deletes expired serving certificate
3. CertsExpirerController 2 starts deletion of expired serving
certificate but has not done so yet
4. CertsManagerController 1 creates new serving certificate
5. TestAPIServingCertificateAutoCreationAndRotation/automatic
records the new serving certificate
6. CertsExpirerController 2 finishes deletion, and thus deletes the
newly created serving certificate instead of the old one
7. CertsManagerController 2 creates new serving certificate
8. TestAPIServingCertificateAutoCreationAndRotation/automatic keeps
running and eventually times out because it is expecting the
serving certificate created by CertsManagerController 2 to match
the value it recorded from CertsManagerController 1 (which will
never happen since that certificate was incorrectly deleted).
Signed-off-by: Monis Khan <mok@vmware.com>
This functioned fine, but did not have the intended visual appearance when it came to how the text of the auth code wrapped inside the copy button in the manual flow.
The new styling behaves correctly on at least Chrome, Firefox, and Safari on macOS.
Signed-off-by: Matt Moyer <moyerm@vmware.com>
TestAgentController really runs the controller and evaluates multiple
calls to the controller's Sync with real informers caching updates.
There is a large amount of non-determinism in this unit test, and it
does not always behave the same way. Because it makes assertions about
the specific errors that should be returned by Sync, it was not
accounting for some errors that are only returned by Sync once in a
while depending on the exact (unpredictable) order of operations.
This commit doesn't fix the non-determinism in the test, but rather
tries to work around it by also allowing other (undesired but
inevitable) error messages to appear in the list of actual error
messages returned by the calls to the Sync function.
Signed-off-by: Margo Crawford <margaretc@vmware.com>
This is a new pacakge internal/oidc/provider/formposthtml containing a number of static files embedded using the relatively recent Go "//go:embed" functionality introduced in Go 1.16 (https://blog.golang.org/go1.16).
The Javascript and CSS files are minifiied and injected to make a single self-contained HTML response. There is a special Content-Security-Policy helper to calculate hash-based script-src and style-src rules.
This new code is covered by a new integration test that exercises the JS/HTML functionality in a real browser outside of the rest of the Supervisor.
Signed-off-by: Matt Moyer <moyerm@vmware.com>
Our Supervisor callback handler now needs to load JS and CSS from the provider endpoint, and this JS needs to make a `fetch()` call across origins (to post the form to the CLI callback). This requires a custom Content-Security-Policy compared to other pages we render.
Signed-off-by: Matt Moyer <moyerm@vmware.com>
After noticing that the upstream OIDC discovery calls can hang
indefinitely, I had tried to impose a one minute timeout on them
by giving them a timeout context. However, I hadn't noticed that the
context also gets passed into the JWKS fetching object, which gets
added to our cache and used later. Therefore the timeout context
was added to the cache and timed out while sitting in the cache,
causing later JWKS fetchers to fail.
This commit is trying again to impose a reasonable timeout on these
discovery and JWKS calls, but this time by using http.Client's Timeout
field, which is documented to be a timeout for *each* request/response
cycle, so hopefully this is a more appropriate way to impose a timeout
for this use case. The http.Client instance ends up in the cache on
the JWKS fetcher object, so the timeout should apply to each JWKS
request as well.
Requests that can hang forever are effectively a server-side resource
leak, which could theoretically be taken advantage of in a denial of
service attempt, so it would be nice to avoid having them.
- Add new optional ytt params for the Supervisor deployment.
- When the Supervisor is making calls to an upstream OIDC provider,
use these variables if they were provided.
- These settings are integration tested in the main CI pipeline by
sometimes setting them on deployments in certain cases, and then
letting the existing integration tests (e.g. TestE2EFullIntegration)
provide the coverage, so there are no explicit changes to the
integration tests themselves in this commit.
- this allows the oidc upsream watcher to honor the
HTTP_PROXY,HTTPS_PROXY,NO_PROXY environment variables
Co-authored-by: Christian Ang <angc@vmware.com>
We want the value of time.Now() to be calculated before the call to
IssueClientCertPEM to prevent the ExpirationTimestamp from being
later than the notAfter timestamp on the issued certificate.
Signed-off-by: Monis Khan <mok@vmware.com>
Before this change, we used the `fosite.DefaultOpenIDConnectClient{}` struct, which implements the `fosite.Client` and `fosite.OpenIDConnectClient` interfaces. For a future change, we also need to implement some additional optional interfaces, so we can no longer use the provided default types. Instead, we now use a custom `clientregistry.Client{}` struct, which implements all the requisite interfaces and can be extended to handle the new functionality (in a future change).
There is also a new `clientregistry.StaticRegistry{}` struct, which implements the `fosite.ClientManager` and looks up our single static client. We could potentially extend this in the future with a registry backed by Kubernetes API, for example.
This should be 100% refactor, with no user-observable change.
Signed-off-by: Matt Moyer <moyerm@vmware.com>
This change updates the impersonator to always authorize every
request instead of relying on the Kuberentes API server to perform
the check on the impersonated request. This protects us from
scenarios where we fail to correctly impersonate the user due to
some bug in our proxy logic. We still rely completely on the API
server to perform admission checks on the impersonated requests.
Signed-off-by: Monis Khan <mok@vmware.com>
This change updates the impersonation proxy code to run as a
distinct service account that only has permission to impersonate
identities. Thus any future vulnerability that causes the
impersonation headers to be dropped will fail closed instead of
escalating to the concierge's default service account which has
significantly more permissions.
Signed-off-by: Monis Khan <mok@vmware.com>
WithImpersonation already deletes impersonation headers and has done
so since the early days:
https://github.com/kubernetes/kubernetes/pull/36769
ensureNoImpersonationHeaders will still reject any request that has
impersonation headers set so we will always fail closed.
Signed-off-by: Monis Khan <mok@vmware.com>
When anonymous authentication is disabled, the impersonation proxy
will no longer authenticate anonymous requests other than calls to
the token credential request API (this API is used to retrieve
credentials and thus must be accessed anonymously).
Signed-off-by: Benjamin A. Petersen <ben@benjaminapetersen.me>
Signed-off-by: Monis Khan <mok@vmware.com>
When a CredentialIssuer is switched from one service type to another (or switched to disabled mode), the `impersonatorconfig` controller will delete the previous Service, if any. Normally one Concierge pod will succeed to delete this initially and any other pods will see a NotFound error.
Before this change, the NotFound would bubble up and cause the strategy to enter a ErrorDuringSetup status until the next reconcile loop. We now handle this case without reporting an error.
Signed-off-by: Matt Moyer <moyerm@vmware.com>
The new version has different behavior for the `nonce` claim, which is now omitted if it would be empty (see https://github.com/ory/fosite/pull/570).
Signed-off-by: Matt Moyer <moyerm@vmware.com>
These are tricky because a real load balancer controller (e.g., on GKE) will overwrite and set NodePort, so we can't blindly set the desired state of this fields.
For now, we will just skip reconciling these. In the future, we could be more clever about merging them together with the current state.
Signed-off-by: Margo Crawford <margaretc@vmware.com>
If the only thing that has changed about a strategy is the LastUpdated timestamp, then we should not update the object.
Signed-off-by: Margo Crawford <margaretc@vmware.com>
This is to allow the use of binary LDAP entry attributes as the UID.
For example, a user might like to configure AD’s objectGUID or maybe
objectSid attributes as the UID attribute.
This negatively impacts the readability of the UID when it did not come
from a binary value, but we're considering this an okay trade-off to
keep things simple for now. In the future, we may offer more
customizable encoding options for binary attributes.
These UIDs are currently only used in the downstream OIDC `sub` claim.
They do not effect the user's identity on the Kubernetes cluster,
which is only based on their mapped username and group memberships from
the upstream identity provider. We are not currently supporting any
special encoding for those username and group name LDAP attributes, so
their values in the LDAP entry must be ASCII or UTF-8 in order for them
to be interpreted correctly.
Margo Crawford