daemon: overlay2: remove world writable permission from the lower file

[Dzejrou]

• relates to / introduced in daemon: overlay2: Write layer metadata atomically #46471

In de2447c, the creation of the 'lower' file was changed from using os.Create to using ioutils.AtomicWriteFile, which ignores the system's umask. This means that even though the requested permission in the source code was always 0666, it was 0644 on systems with default umask of 0022 prior to de2447c, so the move to AtomicFile potentially increased the file's permissions.

This is not a security issue because the parent directory does not allow writes into the file, but it can confuse security scanners on Linux-based systems into giving false positives.

Reproduction steps:

1. Install a docker version that contains de2447c (I used docker-24.0.7_ce).
2. In one terminal start any container (I used alpine/sh for simplicity).
3. In a second terminal, execute:

germ85:~ # find /var/lib/docker/overlay2/ -type f -perm -222 -ls
   563855      4 -rw-rw-rw-   1 root     root           28 Mar  5 13:58 /var/lib/docker/overlay2/faeafe7e6f15289f0825aeffa2597f0b7ecf5807d7c21d5a9a2d3383090693b7-init/lower
   563878      4 -rw-rw-rw-   1 root     root           57 Mar  5 13:58 /var/lib/docker/overlay2/faeafe7e6f15289f0825aeffa2597f0b7ecf5807d7c21d5a9a2d3383090693b7/lower

With this patch added to the same docker version (24.0.7_ce):

germ85:~ # find /var/lib/docker/overlay2/ -type f -ls | grep lower
   564922      4 -rw-r--r--   1 root     root           28 Mar  5 14:46 /var/lib/docker/overlay2/488b492ef97cfdaa2a531d9e57aa5bd32425ca6bd9154b985eaed89a4fb00dc5-init/lower
   564945      4 -rw-r--r--   1 root     root           57 Mar  5 14:46 /var/lib/docker/overlay2/488b492ef97cfdaa2a531d9e57aa5bd32425ca6bd9154b985eaed89a4fb00dc5/lower

Output from the same system with docker-24.0.5_ce-150000.185.1 (before de2447c was added) installed shows the same behavior as 27.0.7_ce with this patch:

germ85:~ # rpm -qa docker
docker-24.0.5_ce-150000.185.1.x86_64
germ85:~ # find /var/lib/docker/overlay2/ -type f -ls | grep lower
   565976      4 -rw-r--r--   1 root     root           28 Mar  5 14:51 /var/lib/docker/overlay2/d16828005bf086a36eaf4d5d3d4822f652b34bfb0e3323b9c1173dae0d0418e7-init/lower
   565999      4 -rw-r--r--   1 root     root           57 Mar  5 14:51 /var/lib/docker/overlay2/d16828005bf086a36eaf4d5d3d4822f652b34bfb0e3323b9c1173dae0d0418e7/lower

The patch from this PR restores the behavior from 24.0.5_ce where the lower files were not world writable.

[thaJeztah]

In de2447c, the creation of the 'lower' file was changed from using os.Create to using ioutils.AtomicWriteFile, which ignores the system's umask.

Hm.. nice find. We should probably consider updating the GoDoc for those functions and call this out explicitly (either that, or make it align with os.Create if the original intent was for it to be a "drop-in" replacement).

[thaJeztah]

Looks like #46471 was part of v25.0, so I added a cherry-pick label for consideration.

[neersighted]

This looks good to me; it does appear that the intended semantics of the ioutils methods are a literal permissions mask not accounting for umask.

Would you mind updating the docstrings in the pkg/ioutils package to make clear how they handle the permissions bits?

[thaJeztah]

Would you mind updating the docstrings in the pkg/ioutils package to make clear how they handle the permissions bits?

I think it's fine to do that in a separate PR, as it's not directly related, and wouldn't be needed to backport, so separate concern.

[Dzejrou]

Looks like #46471 was part of v25.0, so I added a cherry-pick label for consideration.

Would a v24 backport also be possible? That's the version this was reported to me against and what we use (SLES). I can open the PRs for both backports if needed.

[neersighted]

There won't be a 24.0 release, unless a maintainer is interested in sponsoring that work (currently Mirantis/Microsoft are maintaining 23.0 and everyone else is on 25.0). If SLES is packaging 24.0, I think you would be better off asking them to take a patch (or an untagged commit) in their packaging/fork.

[Dzejrou]

There won't be a 24.0 release, ...

Thanks, that's enough for me (I will have that commit submitted, but we have an upstream-first policy and as such if upstream were to ever release another v24, I'd need to get the patch merged there before submitting it to SLES).

[thaJeztah]

Would a v24 backport also be possible? That's the version this was reported to me against and what we use (SLES). I can open the PRs for both backports if needed.

Ah! I missed that that patch was backported, so indeed it looks like 23.0 and 24.0 also would have this issue.

[thaJeztah]

LGTM

[thaJeztah]

Created a quick tracking ticket for the GoDoc changes in case someone wants to pick up that work;

• Update GoDoc for ioutils (AtomicWriteFile) to document expected behavior #47516