[RFC] block, bfq: use-after-free with bfq <==> cgroup interactions

From: Sasha Levin
Date: Sat Jan 14 2023 - 10:38:17 EST


I've observed the follow use after free after commit 64dc8c732f5c
("block, bfq: fix possible uaf for 'bfqq->bic'"):

[ 114.277139] BUG: unable to handle page fault for address: ffff9edd3a529f58
[ 114.284173] #PF: supervisor read access in kernel mode
[ 114.289338] #PF: error_code(0x0000) - not-present page
[ 114.294478] PGD a661c01067 P4D a661c01067 PUD c03f1c2067 PMD c03efef067 PTE 800ffffe85ad6060
[ 114.302947] Oops: 0000 [#1] SMP DEBUG_PAGEALLOC NOPTI
[ 114.308000] CPU: 153 PID: 4171 Comm: udevd Tainted: G W 5.15.88-dbg-DEV #5
[ 114.316215] Hardware name: Google, Inc.
Arcadia_IT_80/Arcadia_IT_80, BIOS 34.2.2 10/03/2022
[ 114.329301] RIP: 0010:bic_set_bfqq (./block/bfq-iosched.c:392)
[ 114.333599] Code: 38 5d c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00
0f 1f 44 00 00 55 48 89 e5 53 48 89 fb 89 d0 48 8b 4c c7 38 48 85 c9
74 14 <48> 39 99 98 01 00 00 75 0b 48 c7 81 98 01 00 00 00 00 00 00 48
89
All code
========
0: 38 5d c3 cmp %bl,-0x3d(%rbp)
3: 66 2e 0f 1f 84 00 00 cs nopw 0x0(%rax,%rax,1)
a: 00 00 00
d: 0f 1f 40 00 nopl 0x0(%rax)
11: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1)
16: 55 push %rbp
17: 48 89 e5 mov %rsp,%rbp
1a: 53 push %rbx
1b: 48 89 fb mov %rdi,%rbx
1e: 89 d0 mov %edx,%eax
20: 48 8b 4c c7 38 mov 0x38(%rdi,%rax,8),%rcx
25: 48 85 c9 test %rcx,%rcx
28: 74 14 je 0x3e
2a:* 48 39 99 98 01 00 00 cmp %rbx,0x198(%rcx) <-- trapping instruction
31: 75 0b jne 0x3e
33: 48 c7 81 98 01 00 00 movq $0x0,0x198(%rcx)
3a: 00 00 00 00
3e: 48 rex.W
3f: 89 .byte 0x89

Code starting with the faulting instruction
===========================================
0: 48 39 99 98 01 00 00 cmp %rbx,0x198(%rcx)
7: 75 0b jne 0x14
9: 48 c7 81 98 01 00 00 movq $0x0,0x198(%rcx)
10: 00 00 00 00
14: 48 rex.W
15: 89 .byte 0x89
[ 114.352382] RSP: 0018:ffffb94be342b5b8 EFLAGS: 00010086
[ 114.357607] RAX: 0000000000000001 RBX: ffff9edd3a4f5f08 RCX: ffff9edd3a529dc0
[ 114.364730] RDX: 0000000000000001 RSI: 0000000000000000 RDI: ffff9edd3a4f5f08
[ 114.371856] RBP: ffffb94be342b5c0 R08: 0000000000000000 R09: 0000000000000001
[ 114.378989] R10: 00000000000a0027 R11: ffffffff9f6c9500 R12: 0000000000000060
[ 114.386120] R13: ffff9edd3a529dc0 R14: ffff9edd3a4f5f08 R15: ffff9edcd12d3800
[ 114.393252] FS: 00007fa7ab3e5740(0000) GS:ffff9f3a8e440000(0000) knlGS:0000000000000000
[ 114.401340] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 114.407086] CR2: ffff9edd3a529f58 CR3: 000000011006e004 CR4: 0000000000370ee0
[ 114.414255] Call Trace:
[ 114.416706] <TASK>
[ 114.418821] bfq_bic_update_cgroup (././include/linux/blk-cgroup.h:401 ./block/bfq-cgroup.c:286 ./block/bfq-cgroup.c:774)
[ 114.425087] bfq_bio_merge (./block/bfq-iosched.c:?)
[ 114.430599] __blk_mq_sched_bio_merge (./block/blk-mq-sched.c:383)
[ 114.436950] blk_mq_submit_bio (./block/blk-mq.c:2220)
[ 114.442776] __submit_bio (./block/blk-core.c:928)
[ 114.448262] submit_bio_noacct (././include/linux/bio.h:618 ./block/blk-core.c:1009 ./block/blk-core.c:1038)
[ 114.454181] submit_bio (./block/blk-core.c:1101)
[ 114.459338] ext4_io_submit (./fs/ext4/page-io.c:383)
[ 114.464824] ext4_writepages (./fs/ext4/inode.c:?)
[ 114.470692] ? do_writepages (./mm/page-writeback.c:2364)
[ 114.476440] ? lock_is_held_type (./kernel/locking/lockdep.c:5365 ./kernel/locking/lockdep.c:5665)
[ 114.482468] ? lock_is_held_type (./kernel/locking/lockdep.c:5365 ./kernel/locking/lockdep.c:5665)
[ 114.488477] do_writepages (./mm/page-writeback.c:2364)
[ 114.494048] ? wbc_attach_and_unlock_inode (./fs/fs-writeback.c:719)
[ 114.500948] filemap_fdatawrite_wbc (./mm/filemap.c:400)
[ 114.507125] filemap_flush (./mm/filemap.c:? ./mm/filemap.c:439 ./mm/filemap.c:466)
[ 114.512526] ext4_alloc_da_blocks (./fs/ext4/inode.c:?)
[ 114.518555] ext4_rename2 (./fs/ext4/namei.c:? ./fs/ext4/namei.c:4191)
[ 114.524124] ? down_write_nested (./kernel/locking/rwsem.c:1643)
[ 114.530038] ? lock_two_nondirectories (./fs/inode.c:1044)
[ 114.536471] vfs_rename (./fs/namei.c:4680)
[ 114.541785] do_renameat2 (./fs/namei.c:?)
[ 114.547307] __x64_sys_rename (./fs/namei.c:4877 ./fs/namei.c:4875 ./fs/namei.c:4875)
[ 114.552988] do_syscall_64 (./arch/x86/entry/common.c:50 ./arch/x86/entry/common.c:80)
[ 114.558389] ? sysvec_call_function_single (./arch/x86/kernel/smp.c:243)
[ 114.565209] entry_SYSCALL_64_after_hwframe (./arch/x86/entry/entry_64.S:118)
[ 114.572083] RIP: 0033:0x7fa7ab490327

The proposed fix is based purely on an observation that earlier in the
same function the call order is reversed where we first bic_set_bfqq()
and only then bfq_release_process_ref().

And thus, this explanation is only a "how" and not a "why", which is why
the patch is going out as an RFC.

Fixes: 64dc8c732f5c ("block, bfq: fix possible uaf for 'bfqq->bic'")
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
block/bfq-cgroup.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/block/bfq-cgroup.c b/block/bfq-cgroup.c
index 1b2829e99dad..cec4d88f6de7 100644
--- a/block/bfq-cgroup.c
+++ b/block/bfq-cgroup.c
@@ -770,9 +770,9 @@ static void __bfq_bic_change_cgroup(struct bfq_data *bfqd,
* bfqq now so that we cannot merge bio to a
* request from the old cgroup.
*/
+ bic_set_bfqq(bic, NULL, true);
bfq_put_cooperator(sync_bfqq);
bfq_release_process_ref(bfqd, sync_bfqq);
- bic_set_bfqq(bic, NULL, true);
}
}
}
--
2.35.1