xfs: update the last_sync_lsn with ctx start lsn

hulk inclusion
category: bugfix
bugzilla: 189076, https://gitee.com/openeuler/kernel/issues/I76JSK


CVE: NA

--------------------------------

While performing the io fault injection test, I caught the following data
corruption report:

 XFS (dm-6): Internal error ltbno + ltlen > bno at line 1976 of file fs/xfs/libxfs/xfs_alloc.c.  Caller xfs_free_ag_extent+0x40b/0x930 [xfs]
 CPU: 7 PID: 184267 Comm: kworker/7:1 Kdump: loaded Tainted: G           O      5.10.0-136.12.0.86.h1179.eulerosv2r12.x86_64 #1
 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.1-0-ga5cab58-20211121_093514-szxrtosci10000 04/01/2014
 Workqueue: xfs-inodegc/dm-6 xfs_inodegc_worker [xfs]
 Call Trace:
  dump_stack+0x57/0x6e
  xfs_corruption_error+0x81/0x90 [xfs]
  xfs_free_ag_extent+0x43c/0x930 [xfs]
  xfs_free_agfl_block+0x3b/0xd0 [xfs]
  xfs_agfl_free_finish_item+0x14c/0x160 [xfs]
  xfs_defer_finish_one+0xd5/0x220 [xfs]
  xfs_defer_finish_noroll+0xb5/0x210 [xfs]
  xfs_defer_finish+0x11/0x70 [xfs]
  xfs_itruncate_extents_flags+0xc1/0x240 [xfs]
  xfs_inactive_truncate+0xab/0xe0 [xfs]
  xfs_inactive+0x154/0x170 [xfs]
  xfs_inodegc_inactivate+0x16/0x50 [xfs]
  xfs_inodegc_worker+0xa0/0x110 [xfs]
  process_one_work+0x1b5/0x350
  worker_thread+0x49/0x310
  kthread+0xfe/0x140
  ret_from_fork+0x22/0x30
 XFS (dm-6): Corruption detected. Unmount and run xfs_repair

After analyzing the disk image, I found that the cause of the problem was
that the transactions were not replayed.

The problem arises in that the iclog buffer IO completion updates the
l_last_sync_lsn with it's own LSN. Transactions can be large enough to
span many iclogs, only commit iclog goes to update l_last_sync_lsn. Since
the last_sync_lsn update and the insertion of the item into the ail list
releases the l_icloglock, if the ail is is empty in the meantime, the new
iclog gets the last_sync_lsn as tail lsn. If the new iclog is written to
disk and a shutdown occurs, the current iclog will not be able to replay
in the next mount.

  xlog_state_done_syncing
    xlog_state_do_callback
      spin_lock(&log->l_icloglock);
      xlog_state_do_iclog_callbacks
        xlog_state_iodone_process_iclog
          xlog_state_set_callback
            last_sync_lsn = iclog->ic_header.h_lsn
        spin_unlock(&log->l_icloglock);
        ====>AIL is empty and get tail lsn for new iclog
        xlog_cil_process_committed
          xlog_cil_committed
            xfs_trans_committed_bulk(ctx->start_lsn)
              xfs_log_item_batch_insert(commit_lsn)
        xlog_state_clean_iclog(log, iclog)
        spin_lock(&log->l_icloglock);
      spin_unlock(&log->l_icloglock);

Fix is simple, updates the l_last_sync_lsn with it's first ctx start lsn
when commit iclog buffer IO completion. Even if the above happens, the
iclog will be replayed as well.

Signed-off-by: Long Li <leo.lilong@huawei.com>