]> git.hungrycats.org Git - linux/commitdiff
f2fs: fix to don't set SB_RDONLY in f2fs_handle_critical_error()
authorChao Yu <chao@kernel.org>
Tue, 10 Sep 2024 03:07:13 +0000 (11:07 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 4 Oct 2024 14:33:16 +0000 (16:33 +0200)
[ Upstream commit 930c6ab93492c4b15436524e704950b364b2930c ]

syzbot reports a f2fs bug as below:

------------[ cut here ]------------
WARNING: CPU: 1 PID: 58 at kernel/rcu/sync.c:177 rcu_sync_dtor+0xcd/0x180 kernel/rcu/sync.c:177
CPU: 1 UID: 0 PID: 58 Comm: kworker/1:2 Not tainted 6.10.0-syzkaller-12562-g1722389b0d86 #0
Workqueue: events destroy_super_work
RIP: 0010:rcu_sync_dtor+0xcd/0x180 kernel/rcu/sync.c:177
Call Trace:
 percpu_free_rwsem+0x41/0x80 kernel/locking/percpu-rwsem.c:42
 destroy_super_work+0xec/0x130 fs/super.c:282
 process_one_work kernel/workqueue.c:3231 [inline]
 process_scheduled_works+0xa2c/0x1830 kernel/workqueue.c:3312
 worker_thread+0x86d/0xd40 kernel/workqueue.c:3390
 kthread+0x2f0/0x390 kernel/kthread.c:389
 ret_from_fork+0x4b/0x80 arch/x86/kernel/process.c:147
 ret_from_fork_asm+0x1a/0x30 arch/x86/entry/entry_64.S:244

As Christian Brauner pointed out [1]: the root cause is f2fs sets
SB_RDONLY flag in internal function, rather than setting the flag
covered w/ sb->s_umount semaphore via remount procedure, then below
race condition causes this bug:

- freeze_super()
 - sb_wait_write(sb, SB_FREEZE_WRITE)
 - sb_wait_write(sb, SB_FREEZE_PAGEFAULT)
 - sb_wait_write(sb, SB_FREEZE_FS)
- f2fs_handle_critical_error
 - sb->s_flags |= SB_RDONLY
- thaw_super
 - thaw_super_locked
  - sb_rdonly() is true, so it skips
    sb_freeze_unlock(sb, SB_FREEZE_FS)
  - deactivate_locked_super

Since f2fs has almost the same logic as ext4 [2] when handling critical
error in filesystem if it mounts w/ errors=remount-ro option:
- set CP_ERROR_FLAG flag which indicates filesystem is stopped
- record errors to superblock
- set SB_RDONLY falg
Once we set CP_ERROR_FLAG flag, all writable interfaces can detect the
flag and stop any further updates on filesystem. So, it is safe to not
set SB_RDONLY flag, let's remove the logic and keep in line w/ ext4 [3].

[1] https://lore.kernel.org/all/20240729-himbeeren-funknetz-96e62f9c7aee@brauner
[2] https://lore.kernel.org/all/20240729132721.hxih6ehigadqf7wx@quack3
[3] https://lore.kernel.org/linux-ext4/20240805201241.27286-1-jack@suse.cz

Fixes: b62e71be2110 ("f2fs: support errors=remount-ro|continue|panic mountoption")
Reported-by: syzbot+20d7e439f76bbbd863a7@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/000000000000b90a8e061e21d12f@google.com/
Cc: Jan Kara <jack@suse.cz>
Cc: Christian Brauner <brauner@kernel.org>
Signed-off-by: Chao Yu <chao@kernel.org>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
fs/f2fs/super.c

index 832c052dd94bcd347c44a7514b64e688d4efd5c5..f897e416b7e6d96873cab2431867e13f2e3b2cf0 100644 (file)
@@ -4183,12 +4183,14 @@ void f2fs_handle_critical_error(struct f2fs_sb_info *sbi, unsigned char reason,
        }
 
        f2fs_warn(sbi, "Remounting filesystem read-only");
+
        /*
-        * Make sure updated value of ->s_mount_flags will be visible before
-        * ->s_flags update
+        * We have already set CP_ERROR_FLAG flag to stop all updates
+        * to filesystem, so it doesn't need to set SB_RDONLY flag here
+        * because the flag should be set covered w/ sb->s_umount semaphore
+        * via remount procedure, otherwise, it will confuse code like
+        * freeze_super() which will lead to deadlocks and other problems.
         */
-       smp_wmb();
-       sb->s_flags |= SB_RDONLY;
 }
 
 static void f2fs_record_error_work(struct work_struct *work)