memcg: Fix possible use-after-free in memcg_write_event_control()
authorTejun Heo <tj@kernel.org>
Thu, 8 Dec 2022 02:53:15 +0000 (16:53 -1000)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 8 Dec 2022 18:40:58 +0000 (10:40 -0800)
commitfbf8321238bac04368f57af572e05a9c01347a0b
tree3dc1ff4425dbc725bc6116e3dde8c76afc4f8837
parent479174d402bcf60789106eedc4def3957c060bad
memcg: Fix possible use-after-free in memcg_write_event_control()

memcg_write_event_control() accesses the dentry->d_name of the specified
control fd to route the write call.  As a cgroup interface file can't be
renamed, it's safe to access d_name as long as the specified file is a
regular cgroup file.  Also, as these cgroup interface files can't be
removed before the directory, it's safe to access the parent too.

Prior to 347c4a874710 ("memcg: remove cgroup_event->cft"), there was a
call to __file_cft() which verified that the specified file is a regular
cgroupfs file before further accesses.  The cftype pointer returned from
__file_cft() was no longer necessary and the commit inadvertently
dropped the file type check with it allowing any file to slip through.
With the invarients broken, the d_name and parent accesses can now race
against renames and removals of arbitrary files and cause
use-after-free's.

Fix the bug by resurrecting the file type check in __file_cft().  Now
that cgroupfs is implemented through kernfs, checking the file
operations needs to go through a layer of indirection.  Instead, let's
check the superblock and dentry type.

Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: 347c4a874710 ("memcg: remove cgroup_event->cft")
Cc: stable@kernel.org # v3.14+
Reported-by: Jann Horn <jannh@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Roman Gushchin <roman.gushchin@linux.dev>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/cgroup.h
kernel/cgroup/cgroup-internal.h
mm/memcontrol.c