f2fs: fix potential data inconsistence of checkpoint
authorChao Yu <yuchao0@huawei.com>
Sat, 2 Feb 2019 09:33:01 +0000 (17:33 +0800)
committerJaegeuk Kim <jaegeuk@kernel.org>
Wed, 6 Mar 2019 03:58:06 +0000 (19:58 -0800)
commitc42d28ce3e16dbd88e575c0acfda96d221dae2c9
treef96e5e3a8f42d59e2764d1e1e3ec25c3552a4e29
parent025cdb166c1e6a142fdec46bcd0873e00f0be0fd
f2fs: fix potential data inconsistence of checkpoint

Previously, we changed lock from cp_rwsem to node_change, it solved
the deadlock issue which was caused by below race condition:

Thread A Thread B
- f2fs_setattr
 - f2fs_lock_op  -- read_lock
 - dquot_transfer
  - __dquot_transfer
   - dquot_acquire
    - commit_dqblk
     - f2fs_quota_write
      - f2fs_write_begin
       - f2fs_write_failed
- write_checkpoint
 - block_operations
  - f2fs_lock_all  -- write_lock
        - f2fs_truncate_blocks
         - f2fs_lock_op  -- read_lock

But it breaks the sematics of cp_rwsem, in other callers like:
- f2fs_file_write_iter -> f2fs_write_begin -> f2fs_write_failed
- f2fs_direct_IO -> f2fs_write_failed

We allow to truncate dnode w/o cp_rwsem held, result in incorrect sit
bitmap update, which can cause further data corruption.

So this patch reverts previous fix implementation, and try to fix
deadlock by skipping calling f2fs_truncate_blocks() in f2fs_write_failed()
only for quota file, and keep the preallocated data/node in the tail of
quota file, we can expecte that the preallocated space can be used to
store quota info latter soon.

Fixes: af033b2aa8a8 ("f2fs: guarantee journalled quota data by checkpoint")
Signed-off-by: Gao Xiang <gaoxiang25@huawei.com>
Signed-off-by: Sheng Yong <shengyong1@huawei.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
fs/f2fs/data.c
fs/f2fs/f2fs.h
fs/f2fs/file.c
fs/f2fs/inline.c