mbox series

[-next,0/6] Fix some issues about mmp

Message ID 20210906144754.2601607-1-yebin10@huawei.com
Headers show
Series Fix some issues about mmp | expand

Message

yebin (H) Sept. 6, 2021, 2:47 p.m. UTC 
I test mmp function as follow steps:
1. Inject delay 5s in ext4_multi_mount_protect function after
"skip:" label.
2. Share HostA block device(sda) with HostB(nbd0) by NBD.
3. Enable mmp feature when mkfs.ext4 sda.
4. Mount sda and nbd0 at the same time.

I found kmmpd never trigger detect multi-mount. Reason is as follow:
1. Kmmpd init seq with 0, if two host have same nodename, will lead to
detect confliction very slow, even never detect confliction.
2. When detect confliction in kmmpd, we get 'check_bh' is same with 'bh'.
so we compare data with itself.
3. We only trigger detect when ”diff > mmp_check_interval * HZ“,
'mmp_check_interval' is double of 'mmp_update_interval', 'diff' is
about 'mmp_update_interval'. So 'diff' is little than 'mmp_check_interval * HZ'
normaly.

And i also found that 'check_interval' value store in disk  is not sure
after umount.

Ye Bin (6):
  ext4: init seq with random value in kmmpd
  ext4: introduce last_check_time record previous check time
  ext4: compare to local seq and nodename when check conflict
  ext4: avoid to re-read mmp check data get from page cache
  ext4: avoid to double free s_mmp_bh
  ext4: fix possible store wrong check interval value in disk when
    umount

 fs/ext4/mmp.c | 77 ++++++++++++++++++++++++++++-----------------------
 1 file changed, 43 insertions(+), 34 deletions(-)