f2fs: fix to set superblock dirty correctly
authorChao Yu <yuchao0@huawei.com>
Wed, 31 Aug 2016 02:43:19 +0000 (10:43 +0800)
committerJaegeuk Kim <jaegeuk@kernel.org>
Thu, 8 Sep 2016 01:53:47 +0000 (18:53 -0700)
commitc2a080aefa94c4b3dad35fbd9fa6cd0c8039b128
tree84a190ec206006a19e4077b9ee31d128764f7703
parente7ba108a06216dae89a64c0243560502276b92d8
f2fs: fix to set superblock dirty correctly

tests/generic/251 of fstest suit complains us with below message:

------------[ cut here ]------------
invalid opcode: 0000 [#1] PREEMPT SMP
CPU: 2 PID: 7698 Comm: fstrim Tainted: G           O    4.7.0+ #21
task: e9f4e000 task.stack: e7262000
EIP: 0060:[<f89fcefe>] EFLAGS: 00010202 CPU: 2
EIP is at write_checkpoint+0xfde/0x1020 [f2fs]
EAX: f33eb300 EBX: eecac310 ECX: 00000001 EDX: ffff0001
ESI: eecac000 EDI: eecac5f0 EBP: e7263dec ESP: e7263d18
 DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
CR0: 80050033 CR2: b76ab01c CR3: 2eb89de0 CR4: 000406f0
Stack:
 00000001 a220fb7b e9f4e000 00000002 419ff2d3 b3a05151 00000002 e9f4e5d8
 e9f4e000 419ff2d3 b3a05151 eecac310 c10b8154 b3a05151 419ff2d3 c10b78bd
 e9f4e000 e9f4e000 e9f4e5d8 00000001 e9f4e000 ec409000 eecac2cc eecac288
Call Trace:
 [<c10b8154>] ? __lock_acquire+0x3c4/0x760
 [<c10b78bd>] ? mark_held_locks+0x5d/0x80
 [<f8a10632>] f2fs_trim_fs+0x1c2/0x2e0 [f2fs]
 [<f89e9f56>] f2fs_ioctl+0x6b6/0x10b0 [f2fs]
 [<c13d51df>] ? __this_cpu_preempt_check+0xf/0x20
 [<c10b4281>] ? trace_hardirqs_off_caller+0x91/0x120
 [<f89e98a0>] ? __exchange_data_block+0xd30/0xd30 [f2fs]
 [<c120b2e1>] do_vfs_ioctl+0x81/0x7f0
 [<c11d57c5>] ? kmem_cache_free+0x245/0x2e0
 [<c1217840>] ? get_unused_fd_flags+0x40/0x40
 [<c1206eec>] ? putname+0x4c/0x50
 [<c11f631e>] ? do_sys_open+0x16e/0x1d0
 [<c1001990>] ? do_fast_syscall_32+0x30/0x1c0
 [<c13d51df>] ? __this_cpu_preempt_check+0xf/0x20
 [<c120baa8>] SyS_ioctl+0x58/0x80
 [<c1001a01>] do_fast_syscall_32+0xa1/0x1c0
 [<c178cc54>] sysenter_past_esp+0x45/0x74
EIP: [<f89fcefe>] write_checkpoint+0xfde/0x1020 [f2fs] SS:ESP 0068:e7263d18
---[ end trace 4de95d7e6b3aa7c6 ]---

The reason is: with below call stack, we will encounter BUG_ON during
doing fstrim.

Thread A Thread B
- write_checkpoint
 - do_checkpoint
- f2fs_write_inode
 - update_inode_page
  - update_inode
   - set_page_dirty
    - f2fs_set_node_page_dirty
     - inc_page_count
      - percpu_counter_inc
      - set_sbi_flag(SBI_IS_DIRTY)
  - clear_sbi_flag(SBI_IS_DIRTY)

Thread C Thread D
- f2fs_write_node_page
 - set_node_addr
  - __set_nat_cache_dirty
   - nm_i->dirty_nat_cnt++
- do_vfs_ioctl
 - f2fs_ioctl
  - f2fs_trim_fs
   - write_checkpoint
    - f2fs_bug_on(nm_i->dirty_nat_cnt)

Fix it by setting superblock dirty correctly in do_checkpoint and
f2fs_write_node_page.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
fs/f2fs/checkpoint.c