[LU-11591] MDS n003 crash, LustreError: 20558:0:(osp_sync.c:1258:osp_sync_thread()) ASSERTION( thread->t_flags != SVC_RUNNING ) failed: 4632 changes, 16403 in progress, 0 in flight

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.13.0, Lustre 2.12.1
• Affects Version/s: Lustre 2.11.0
• Labels: patch
• Severity: 3
• Assignee: Alexander Boyko
• Reporter: Alexander Boyko

Description

2018-10-28 14:17:05 [357676.303755] LustreError: 20558:0:(llog_cat.c:269:llog_cat_id2handle()) snx11205-OST0003-osc-MDT0001: error opening log id [0x36ba7:0x1:0x0]:0: rc = -2
2018-10-28 14:17:05 [357676.320844] LustreError: 20558:0:(osp_sync.c:1258:osp_sync_thread()) ASSERTION( thread->t_flags != SVC_RUNNING ) failed: 4632 changes, 16403 in progress, 0 in flight
2018-10-28 14:17:05 [357676.339185] LustreError: 20558:0:(osp_sync.c:1258:osp_sync_thread()) LBUG
2018-10-28 14:17:05 [357676.347801] Pid: 20558, comm: osp-syn-3-1
2018-10-28 14:17:05 [357676.353657]
2018-10-28 14:17:05 [357676.353657] Call Trace:
2018-10-28 14:17:05 [357676.361029]  [<ffffffffc09aa7ae>] libcfs_call_trace+0x4e/0x60 [libcfs]
2018-10-28 14:17:05 [357676.369345]  [<ffffffffc09aa83c>] lbug_with_loc+0x4c/0xb0 [libcfs]
2018-10-28 14:17:05 [357676.377231]  [<ffffffffc157705b>] osp_sync_thread+0xa1b/0xa70 [osp]
2018-10-28 14:17:05 [357676.385441]  [<ffffffff816b3e2c>] ? __schedule+0x47c/0xa30
2018-10-28 14:17:05 [357676.392679]  [<ffffffffc1576640>] ? osp_sync_thread+0x0/0xa70 [osp]
2018-10-28 14:17:05 [357676.400867]  [<ffffffff810b4031>] kthread+0xd1/0xe0
2018-10-28 14:17:05 [357676.407508]  [<ffffffff810b3f60>] ? kthread+0x0/0xe0
2018-10-28 14:17:05 [357676.414352]  [<ffffffff816c155d>] ret_from_fork+0x5d/0xb0
2018-10-28 14:17:05 [357676.421426]  [<ffffffff810b3f60>] ? kthread+0x0/0xe0
2018-10-28 14:17:05 [357676.428211]
2018-10-28 14:17:05 [357676.431329] Kernel panic - not syncing: LBUG

catalog llog_handle

crash> llog_handle 0xffff880fcb784200
struct llog_handle {
  lgh_lock = {
    {
      count = {
  lgh_id = {
    lgl_oi = {
      {
        oi = {
          oi_id = 8,
          oi_seq = 1
        },
        oi_fid = {
          f_seq = 8,
          f_oid = 1,
          f_ver = 0
        }
      }
    },
    lgl_ogen = 0
  },
  lgh_hdr = 0xffff880f60a70000,
  lgh_hdr_size = 8192,
  lgh_obj = 0xffff880fcb784d00,
  lgh_last_idx = 54914,
  lgh_cur_offset = 3530688,
  lgh_ctxt = 0xffff880f6bd4e840,
  u = {
    phd = {
      phd_entry = {
        next = 0xffff880f0dd95a90,
        prev = 0xffff88051ca0b290
      },
crash> llog_log_hdr 0xffff880f60a70000
struct llog_log_hdr {
  llh_hdr = {
    lrh_len = 8192,
    lrh_index = 0,
    lrh_type = 275010873,
    lrh_id = 0
  },
  llh_timestamp = 1519928037,
  llh_count = 2,
  llh_bitmap_offset = 88,
  llh_size = 64,
  llh_flags = 18,
  llh_cat_idx = 54912,
  llh_tgtuuid = {
    uuid = "\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000"
  },
  llh_reserved = {0},
  llh_bitmap = {1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0...},
  llh_tail = {
    lrt_len = 8192,
    lrt_index = 54914
  }
}

The catalog processing and writing operates the same record. So the race is very close to LU-11392/LU-7001. I've identify one more place for race.

1 Thread with write                            2 Thread with read
llog_osd_write_rec()                           llog_process_thread()
lgh_last_idx++                             
lock lgh_hdr_mutex
ext2_set_bit() 
dt_write_rec (write header)                ext2_test_bit()
                                                           check lgh_last_idx was changed
                                                           dt_read_rec() reread the record, and here we got the old value of record
unlock lgh_hdr_mutex
dt_write_rec (write the record)

When llog_process_thread test the bit from bitmap and reread the record, we cannot be sure that the record was written by another thread. The lgh_hdr_mutex guarantee only the consistent of llog header, not the record itself.