Skip to content

RFC Per user namespace rlimits

Alexey Gladkov containers kernel-hardening lkml 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 RFC Latest4 patches 7 repliesIn review

Changes in v1 from the cover letter

* After discussion with Eric W. Biederman, I increased the size of ucounts to atomic_long_t.
* Added ucount_max to avoid the fork bomb.

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/4Increase size of ucounts to atomic_long_tinclude/linux/user_namespace.h kernel/ucount.c+9 -92 replies
2/4Move the user's process counter to ucountsfs/exec.c fs/io-wq.c include/linux/sched/user.h +7 more+71 -210 replies
3/4Do not allow fork if RLIMIT_NPROC is exceeded in the user namespace treefs/exec.c fs/io-wq.c include/linux/sched.h +2 more+32 -120 replies
4/4Allow to change the user namespace in which user rlimits are countedfs/exec.c fs/io-wq.c fs/io-wq.h +8 more+80 -164 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [RFC PATCH v1 0/4] Per user namespace rlimits

Preface
-------
These patches are for binding the rlimits to a user in the user namespace. This patch set can be applied on top of:

git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git v5.8-2-g43e210d68200

Problem
-------
Some rlimits are set per user: RLIMIT_NPROC, RLIMIT_MEMLOCK, RLIMIT_SIGPENDING, RLIMIT_MSGQUEUE. When several containers are created from one user then the processes inside the containers influence each other.

Eric W. Biederman mentioned this issue [1][2][3].

Introduced changes
------------------
To fix this problem, you can bind the counter of the specified rlimits to the user within the user namespace. By default, to preserve backward compatibility, only the initial user namespace is used. This patch adds one more prctl parameter to change the binding to the user namespace.

This will not cause the user to take more resources than allowed in the parent user namespace because it only virtualizes the rlimit counter. Limits in all parent user namespaces are taken into account.

For example, this allows us to run multiple containers by the same user and set the RLIMIT_NPROC to 1 inside.

ToDo
----
* RLIMIT_MEMLOCK, RLIMIT_SIGPENDING and RLIMIT_MSGQUEUE are not implemented.
* No documentation.
* No tests.

[1] https://lore.kernel.org/containers/[email protected]/
[2] https://lists.linuxfoundation.org/pipermail/containers/2020-August/042096.html
[3] https://lists.linuxfoundation.org/pipermail/containers/2020-October/042524.html

Changelog
---------
v1:
* After discussion with Eric W. Biederman, I increased the size of ucounts to atomic_long_t.
* Added ucount_max to avoid the fork bomb.

--

Alexey Gladkov (4):
Increase size of ucounts to atomic_long_t
Move the user's process counter to ucounts
Do not allow fork if RLIMIT_NPROC is exceeded in the user namespace
tree
Allow to change the user namespace in which user rlimits are counted

 fs/exec.c                      | 13 ++++++---
 fs/io-wq.c                     | 25 +++++++++++++-----
 fs/io-wq.h                     |  1 +
 fs/io_uring.c                  |  1 +
 include/linux/cred.h           |  8 ++++++
 include/linux/sched.h          |  3 +++
 include/linux/sched/user.h     |  1 -
 include/linux/user_namespace.h | 12 +++++++--
 include/uapi/linux/prctl.h     |  5 ++++
 kernel/cred.c                  | 44 ++++++++++++++++++++++++-------
 kernel/exit.c                  |  2 +-
 kernel/fork.c                  | 13 ++++++---
 kernel/sys.c                   | 26 ++++++++++++++++--
 kernel/ucount.c                | 48 +++++++++++++++++++++++++++++-----
 kernel/user.c                  |  3 ++-
 kernel/user_namespace.c        |  3 +++
 16 files changed, 171 insertions(+), 37 deletions(-)
Signature · 1 line
2.25.4