Before this fix, the deadlock would prevent the leader pod from giving
up its lease, which would make it take several minutes for new pods to
be allowed to elect a new leader. During that time, no Pinniped
controllers could write to the Kube API, so important resources were not
being updated during that window. It would also make pod shutdown take
about 1 minute.
After this fix, the leader gives up its lease immediately, and pod
shutdown takes about 1 second. This improves restart/upgrade time and
also fixes the problem where there was no leader for several minutes
after a restart/upgrade.
The deadlock was between the post-start hook and the pre-shutdown hook.
The pre-shutdown hook blocked until a certain background goroutine in
the post-start hook finished, but that goroutine could not finish until
the pre-shutdown hook finished. Thus, they were both blocked, waiting
for each other infinitely. Eventually the process would be externally
killed.
This deadlock was most likely introduced by some change in Kube's
generic api server package related to how the many complex channels used
during server shutdown interact with each other, and was not noticed
when we upgraded to the version which introduced the change.
- Change update-codegen.sh script to also generated openapi code for the
aggregated API types
- Update both aggregated API servers' configuration to make them serve
the openapi docs for the aggregated APIs
- Add new integration test which runs `kubectl explain` for all Pinniped
API resources, and all fields and subfields of those resources
- Update some the comments on the API structs
- Change some names of the tmpl files to make the filename better match
the struct names
Also fix some tests that were broken by bumping golang and dependencies
in the previous commits.
Note that in addition to changes made to satisfy the linter which do not
impact the behavior of the code, this commit also adds ReadHeaderTimeout
to all usages of http.Server to satisfy the linter (and because it
seemed like a good suggestion).
Highlights from this dep bump:
1. Made a copy of the v0.4.0 github.com/go-logr/stdr implementation
for use in tests. We must bump this dep as Kube code uses a
newer version now. We would have to rewrite hundreds of test log
assertions without this copy.
2. Use github.com/felixge/httpsnoop to undo the changes made by
ory/fosite#636 for CLI based login flows. This is required for
backwards compatibility with older versions of our CLI. A
separate change after this will update the CLI to be more
flexible (it is purposefully not part of this change to confirm
that we did not break anything). For all browser login flows, we
now redirect using http.StatusSeeOther instead of http.StatusFound.
3. Drop plog.RemoveKlogGlobalFlags as klog no longer mutates global
process flags
4. Only bump github.com/ory/x to v0.0.297 instead of the latest
v0.0.321 because v0.0.298+ pulls in a newer version of
go.opentelemetry.io/otel/semconv which breaks k8s.io/apiserver.
We should update k8s.io/apiserver to use the newer code.
5. Migrate all code from k8s.io/apimachinery/pkg/util/clock to
k8s.io/utils/clock and k8s.io/utils/clock/testing
6. Delete testutil.NewDeleteOptionsRecorder and migrate to the new
kubetesting.NewDeleteActionWithOptions
7. Updated ExpectedAuthorizeCodeSessionJSONFromFuzzing caused by
fosite's new rotated_secrets OAuth client field. This new field
is currently not relevant to us as we have no private clients.
Signed-off-by: Monis Khan <mok@vmware.com>
This change updates the TLS config used by all pinniped components.
There are no configuration knobs associated with this change. Thus
this change tightens our static defaults.
There are four TLS config levels:
1. Secure (TLS 1.3 only)
2. Default (TLS 1.2+ best ciphers that are well supported)
3. Default LDAP (TLS 1.2+ with less good ciphers)
4. Legacy (currently unused, TLS 1.2+ with all non-broken ciphers)
Highlights per component:
1. pinniped CLI
- uses "secure" config against KAS
- uses "default" for all other connections
2. concierge
- uses "secure" config as an aggregated API server
- uses "default" config as a impersonation proxy API server
- uses "secure" config against KAS
- uses "default" config for JWT authenticater (mostly, see code)
- no changes to webhook authenticater (see code)
3. supervisor
- uses "default" config as a server
- uses "secure" config against KAS
- uses "default" config against OIDC IDPs
- uses "default LDAP" config against LDAP IDPs
Signed-off-by: Monis Khan <mok@vmware.com>
- Used to determine on which port the impersonation proxy will bind
- Defaults to 8444, which is the old hard-coded port value
- Allow the port number to be configured to any value within the
range 1024 to 65535
- This commit does not include adding new config knobs to the ytt
values file, so while it is possible to change this port without
needing to recompile, it is not convenient
- Allow the port number to be configured to any value within the
range 1024 to 65535
- This commit does not include adding new config knobs to the ytt
values file, so while it is possible to change this port without
needing to recompile, it is not convenient
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>
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>
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>
This change updates the impersonator logic to pass through requests
that authenticated via a bearer token that asserts a UID. This
allows us to support service account tokens (as well as any other
form of token based authentication).
Signed-off-by: Monis Khan <mok@vmware.com>
This change updates the impersonator logic to use the delegated
authorizer for all non-rest verbs such as impersonate. This allows
it to correctly perform authorization checks for incoming requests
that set impersonation headers while not performing unnecessary
checks that are already handled by KAS.
The audit layer is enabled to track the original user who made the
request. This information is then included in a reserved extra
field original-user-info.impersonation-proxy.concierge.pinniped.dev
as a JSON blob.
Signed-off-by: Monis Khan <mok@vmware.com>
Instead of using the LongRunningFunc to determine if we can safely
use http2, follow the same logic as the aggregation proxy and only
use http2 when the request is not an upgrade.
Signed-off-by: Monis Khan <mok@vmware.com>
When the frontend connection to our proxy is closed, the proxy falls through to
a panic(), which means the HTTP handler goroutine is killed, so we were not
seeing this log statement.
Signed-off-by: Andrew Keesler <akeesler@vmware.com>
We were previously issuing both client certs and server certs with
both extended key usages included. Split the Issue*() methods into
separate methods for issuing server certs versus client certs so
they can have different extended key usages tailored for each use
case.
Also took the opportunity to clean up the parameters of the Issue*()
methods and New() methods to more closely match how we prefer to call
them. We were always only passing the common name part of the
pkix.Name to New(), so now the New() method just takes the common name
as a string. When making a server cert, we don't need to set the
deprecated common name field, so remove that param. When making a client
cert, we're always making it in the format expected by the Kube API
server, so just accept the username and group as parameters directly.
