Re: [PATCH] fs/jfs: Add a mutex named txEnd_lmLogClose_mutex to prevent a race condition between txEnd and lmLogClose functions

[Michal Koutný]

Hello Zheng.

On Mon, May 15, 2023 at 05:59:56PM +0800, Zheng Wang <zyytlz.wz@xxxxxxx> wrote:
> ==================================================================
> BUG: KASAN: slab-use-after-free in instrument_atomic_write include/linux/instrumented.h:87 [inline]
> BUG: KASAN: slab-use-after-free in clear_bit include/asm-generic/bitops/instrumented-atomic.h:41 [inline]
> BUG: KASAN: slab-use-after-free in txEnd+0x2a3/0x5a0 fs/jfs/jfs_txnmgr.c:535
> Write of size 8 at addr ffff888021bee840 by task jfsCommit/130
> 
> CPU: 3 PID: 130 Comm: jfsCommit Not tainted 6.3.0-rc7-pasta #1

Is this still pertinent with the current mainline? (There were some
changes to jfs.)

> Through analysis, it was found that a race condition occurred between two
> functions lmLogClose and txEnd, which were executed in different threads.
> The possible sequence is as follows:
> 
> -------------------------------------------------------------------------
> cpu1(free thread)        |        cpu2(use thread)
> -------------------------------------------------------------------------
> lmLogClose               |        txEnd
>                          |        log = JFS_SBI(tblk->sb)->log;
> sbi->log = NULL;         |
> kfree(log); [1] free log |
>                          |        clear_bit(log_FLUSH, &log->flag); [2] UAF

That looks sane to a by-passer.

> Fix it by add a mutex lock between lmLogClose and txEnd:

It doesn't feel right wrt "lock data, not code" heuristics.
And when I apply that, it turns out there's already jfs_log_mutex.
I'd suggest you explain more why a new lock is needed (if that's the
preferred solutino).

Thanks,
Michal

Attachment: signature.asc
Description: PGP signature