[lvc-project] [PATCH] jffs2: fix use-after-free in jffs2_garbage_collect_thread()

[Zhihao Cheng]

在 2026/3/23 17:21, Dmitriy Chumachenko 写道:
> During fuzz testing, the following issue was discovered.
>                                                                                  
> BUG: KASAN: use-after-free in __lock_acquire+0x3f22/0x53c0 kernel/locking/lockdep.c:4825
> Read of size 8 at addr ffff888053cfa098 by task jffs2_gcd_mtd0/11093
>                                                                                  
> CPU: 1 PID: 11093 Comm: jffs2_gcd_mtd0 Not tainted 5.10.232-syzkaller #0
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014
> Call Trace:
>   __dump_stack lib/dump_stack.c:77 [inline]
>   dump_stack+0x107/0x167 lib/dump_stack.c:118
>   print_address_description.constprop.0+0x1c/0x220 mm/kasan/report.c:377
>   __kasan_report mm/kasan/report.c:537 [inline]
>   kasan_report.cold+0x1f/0x37 mm/kasan/report.c:554
>   __lock_acquire+0x3f22/0x53c0 kernel/locking/lockdep.c:4825
>   lock_acquire kernel/locking/lockdep.c:5566 [inline]
>   lock_acquire+0x197/0x480 kernel/locking/lockdep.c:5531
>   __raw_spin_lock_irqsave include/linux/spinlock_api_smp.h:110 [inline]
>   _raw_spin_lock_irqsave+0x36/0x60 kernel/locking/spinlock.c:159
>   complete+0x13/0x60 kernel/sched/completion.c:32
>   complete_and_exit+0x20/0x40 kernel/exit.c:943
>   jffs2_garbage_collect_thread+0x554/0x750 fs/jffs2/background.c:164
>   kthread+0x3a9/0x490 kernel/kthread.c:328
>   ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:298
>                                                                                  
> Allocated by task 11091:
>   kasan_save_stack+0x1b/0x40 mm/kasan/common.c:48
>   kasan_set_track mm/kasan/common.c:56 [inline]
>   __kasan_kmalloc.constprop.0+0xc9/0xd0 mm/kasan/common.c:461
>   kmalloc include/linux/slab.h:552 [inline]
>   kzalloc include/linux/slab.h:664 [inline]
>   jffs2_init_fs_context+0x41/0xd0 fs/jffs2/super.c:314
>   alloc_fs_context+0x4f9/0x840 fs/fs_context.c:267
>   do_new_mount fs/namespace.c:2896 [inline]
>   path_mount+0xb99/0x2140 fs/namespace.c:3247
>   do_mount fs/namespace.c:3260 [inline]
>   __do_sys_mount fs/namespace.c:3468 [inline]
>   __se_sys_mount fs/namespace.c:3445 [inline]
>   __x64_sys_mount+0x283/0x300 fs/namespace.c:3445
>   do_syscall_64+0x30/0x40 arch/x86/entry/common.c:46
>   entry_SYSCALL_64_after_hwframe+0x67/0xd1
>                                                                                  
> Freed by task 28546:
>   kasan_save_stack+0x1b/0x40 mm/kasan/common.c:48
>   kasan_set_track+0x1c/0x30 mm/kasan/common.c:56
>   kasan_set_free_info+0x1b/0x30 mm/kasan/generic.c:355
>   __kasan_slab_free+0x112/0x170 mm/kasan/common.c:422
>   slab_free_hook mm/slub.c:1542 [inline]
>   slab_free_freelist_hook+0xb8/0x1b0 mm/slub.c:1576
>   slab_free mm/slub.c:3149 [inline]
>   kfree+0xd9/0x360 mm/slub.c:4125
>   deactivate_locked_super+0x96/0x170 fs/super.c:335
>   deactivate_super+0xb2/0xd0 fs/super.c:366
>   cleanup_mnt+0x3a3/0x530 fs/namespace.c:1118
>   task_work_run+0xdf/0x1a0 kernel/task_work.c:185
>   tracehook_notify_resume include/linux/tracehook.h:188 [inline]
>   exit_to_user_mode_loop kernel/entry/common.c:172 [inline]
>   exit_to_user_mode_prepare+0x1de/0x1f0 kernel/entry/common.c:199
>   syscall_exit_to_user_mode+0x38/0x1e0 kernel/entry/common.c:274
> 
> In jffs2_garbage_collect_thread() gc_task is set to NULL and then
> kthread_complete_and_exit() calls complete() on gc_thread_exit. These operations
> are not atomic: stop path can see gc_task == NULL, skip wait_for_completion(),
> and the caller frees jffs2_sb_info while the GC thread still accesses
> gc_thread_exit in complete().
> 
> Fix moving complete() under erase_completion_lock together with gc_task =
> NULL, and replacing kthread_complete_and_exit() with kthread_exit(). The
> conditional wait in stop path is preserved as it is needed when
> jffs2_do_fill_super() fails before start().
> 
> Found by Linux Verification Center (linuxtesting.org) with Syzkaller.

I guess the race window is small and it won't happen in normal mode, the 
jffs2_garbage_collect_thread is killed by 
jffs2_stop_garbage_collect_thread normally. For above case, the 
jffs2_garbage_collect_thread is stopped by other reasons(eg. 
jffs2_garbage_collect_pass returns ENOSPC, or user sends SIGKILL to 
jffs2_garbage_collect_thread task).
> 
> Fixes: e2d48b1a98bb ("[JFFS2] Fix cleanup in case of GC-Task not started")
> Signed-off-by: Dmitriy Chumachenko <Dmitry.Chumachenko at cyberprotect.ru>
> ---
>   fs/jffs2/background.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/jffs2/background.c b/fs/jffs2/background.c
> index bb0ee1a59e71..abf0572dfd3c 100644
> --- a/fs/jffs2/background.c
> +++ b/fs/jffs2/background.c
> @@ -160,6 +160,7 @@ static int jffs2_garbage_collect_thread(void *_c)
>    die:
>   	spin_lock(&c->erase_completion_lock);
>   	c->gc_task = NULL;
> +	complete(&c->gc_thread_exit);
>   	spin_unlock(&c->erase_completion_lock);
> -	kthread_complete_and_exit(&c->gc_thread_exit, 0);
> +	kthread_exit(0);
>   }
> 

Hi, I think the UAF could still happen even this patch is applied.

jffs2_kill_sb            jffs2_garbage_collect_thread
  jffs2_stop_garbage_collect_thread
   spin_lock(&c->erase_completion_lock)
     send_sig(SIGKILL, c->gc_task, 1)
     wait = 1
   spin_unlock(&c->erase_completion_lock)
                               goto die // receive kill SIGKILL
                               spin_lock(&c->erase_completion_lock)
                               complete(&c->gc_thread_exit)
   wait_for_completion(&c->gc_thread_exit)  // don't wait
  kfree(c)
                               spin_unlock(&c->erase_completion_lock)