On Wed, Sep 18, 2024 at 11:01:30PM -0700, Greg Thelen wrote:
Linux stable v5.10.226 suffers a lockdep warning when accessing
/proc/PID/cpuset. cset_cgroup_from_root() is called without cgroup_mutex
is held, which causes assertion failure.
Bisect blames 5.10.225 commit 688325078a8b ("cgroup/cpuset: Prevent UAF
in proc_cpuset_show()"). I've have not easily reproduced the problem
that this change fixes, so I'm not sure if it's best to revert the fix
or adapt it to meet the 5.10 locking expectations.
The lockdep complaint:
$ cat /proc/1/cpuset
$ dmesg
[ 198.744891] ------------[ cut here ]------------
[ 198.744918] WARNING: CPU: 4 PID: 9301 at
kernel/cgroup/cgroup.c:1395 cset_cgroup_from_root+0xb2/0xd0
[ 198.744957] RIP: 0010:cset_cgroup_from_root+0xb2/0xd0
[ 198.744960] Code: 02 00 00 74 11 48 8b 09 48 39 cb 75 eb eb 19 49
83 c6 10 4c 89 f0 48 85 c0 74 0d 5b 41 5e c3 48 8b 43 60 48 85 c0 75
f3 0f 0b <0f> 0b 83 3d 69 01 ee 01 00 0f 85 78 ff ff ff eb 8b 0f 0b eb
87 66
[ 198.744962] RSP: 0018:ffffb492608a7ce8 EFLAGS: 00010046
[ 198.744977] RAX: 0000000000000000 RBX: ffffffff8f4171b8 RCX:
cc949de848c33e00
[ 198.744979] RDX: 0000000000001000 RSI: ffffffff8f415450 RDI:
ffff92e5417c4dc0
[ 198.744981] RBP: ffff9303467e3f00 R08: 0000000000000008 R09:
ffffffff9122d568
[ 198.744983] R10: ffff92e5417c4380 R11: 0000000000000000 R12:
ffff92e3d9506000
[ 198.744984] R13: 0000000000000000 R14: ffff92e443a96000 R15:
ffff92e3d9506000
[ 198.744987] FS: 00007f15d94ed740(0000) GS:ffff9302bf500000(0000)
knlGS:0000000000000000
[ 198.744988] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 198.744990] CR2: 00007f15d94ca000 CR3: 00000002816ca003 CR4:
00000000001706e0
[ 198.744992] Call Trace:
[ 198.744996] ? __warn+0xcd/0x1c0
[ 198.745000] ? cset_cgroup_from_root+0xb2/0xd0
[ 198.745008] ? report_bug+0x87/0xf0
[ 198.745015] ? handle_bug+0x42/0x80
[ 198.745017] ? exc_invalid_op+0x16/0x70
[ 198.745021] ? asm_exc_invalid_op+0x12/0x20
[ 198.745030] ? cset_cgroup_from_root+0xb2/0xd0
[ 198.745034] ? cset_cgroup_from_root+0x28/0xd0
[ 198.745038] cgroup_path_ns_locked+0x23/0x50
[ 198.745044] proc_cpuset_show+0x115/0x210
[ 198.745049] proc_single_show+0x4a/0xa0
[ 198.745056] seq_read_iter+0x14d/0x400
[ 198.745063] seq_read+0x103/0x130
[ 198.745074] vfs_read+0xea/0x320
[ 198.745078] ? do_user_addr_fault+0x25b/0x390
[ 198.745085] ? do_user_addr_fault+0x25b/0x390
[ 198.745090] ksys_read+0x70/0xe0
[ 198.745096] do_syscall_64+0x2d/0x40
[ 198.745099] entry_SYSCALL_64_after_hwframe+0x61/0xcb
I'll queue up d23b5c577715 ("cgroup: Make operations on the cgroup
root_list RCU safe") onto 5.15/5.10. Thanks for reporting!
--
Thanks,
Sasha
