Stopping the Engine while a container with autoremove set is running may
leave behind dead containers on disk. These containers aren't reclaimed
on next start, appear as "dead" in `docker ps -a` and can't be
inspected or removed by the user.
This bug has existed since a long time but became user visible with
9f5f4f5a42. Prior to that commit,
containers with no rwlayer weren't added to the in-memory viewdb, so
they weren't visible in `docker ps -a`. However, some dangling files
would still live on disk (e.g. folder in /var/lib/docker/containers,
mount points, etc).
The underlying issue is that when the daemon stops, it tries to stop all
running containers and then closes the containerd client. This leaves a
small window of time where the Engine might receive 'task stop' events
from containerd, and trigger autoremove. If the containerd client is
closed in parallel, the Engine is unable to complete the removal,
leaving the container in 'dead' state. In such case, the Engine logs the
following error:
cannot remove container "bcbc98b4f5c2b072eb3c4ca673fa1c222d2a8af00bf58eae0f37085b9724ea46": Canceled: grpc: the client connection is closing: context canceled
Solving the underlying issue would require complex changes to the
shutdown sequence. Moreover, the same issue could also happen if the
daemon crashes while it deletes a container. Thus, add a cleanup step
on daemon startup to remove these dead containers.
Signed-off-by: Albin Kerouanton <albin.kerouanton@docker.com>
Commit 6aea8c2591 updated this part of the
test to account for BuildKit now being used as default, which doesn't use
regular containers to "commit" to an image, and therefore doesn't produce
events as part of the build.
However, it changed it to filter for `build` type events, which .. don't
exist. The test passed because it only checked for number of events produced,
which is approximated by checking the number of lines in the output. In
this case, the output would be empty, so "events" was a slice of strings
with length 1; an empty string.
Given that BuildKit doesn't produce meaningful events to test for, we could
either disable BuildKit, to test the legacy builder, or just remove this
part of the test. Let's do the latter.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
These tests used intermediate images (as produced by the classic builder)
as part of the test. When using BuildKit, such images are not produced,
and will only be in the build-cache.
From the tests, it looks like the extra checks were not critical to
verify the behavior, so let's simplify them to not depend on this.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Make sure the error is the one we expected, and slightly clarify
the behavior (named volumes not having a meaning in VOLUME, so will
be used as-is).
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
In a multi-module repository, the top-level LICENSE file is used for
all modules in the repository. However, when using a "replace" rule,
the vendored file is removed, but added back when removing the replace
rule.
This patch adds copies of the license to each of the modules to reduce
code-churn; these are vanilla copies downloaded from;
https://www.apache.org/licenses/LICENSE-2.0.txt
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
These releases include 2 security fixes following the security policy:
- crypto/x509: excessive resource consumption in printing error string for host certificate validation
Within HostnameError.Error(), when constructing an error string, there is no limit to the number of hosts that will be printed out.
Furthermore, the error string is constructed by repeated string concatenation, leading to quadratic runtime.
Therefore, a certificate provided by a malicious actor can result in excessive resource consumption.
HostnameError.Error() now limits the number of hosts and utilizes strings.Builder when constructing an error string.
Thanks to Philippe Antoine (Catena cyber) for reporting this issue.
This is CVE-2025-61729 and Go issue https://go.dev/issue/76445.
- crypto/x509: excluded subdomain constraint does not restrict wildcard SANs
An excluded subdomain constraint in a certificate chain does not restrict the
usage of wildcard SANs in the leaf certificate. For example a constraint that
excludes the subdomain test.example.com does not prevent a leaf certificate from
claiming the SAN *.example.com.
This is CVE-2025-61727 and Go issue https://go.dev/issue/76442.
View the release notes for more information:
https://go.dev/doc/devel/release#go1.25.5
Signed-off-by: Paweł Gronowski <pawel.gronowski@docker.com>
Includes a change to use go.dev/dl instead of storage.googleapis.com/golang
as fallback URL, because storage.googleapis.com/golang is deprecated.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
