lkml.org 
[lkml]   [2021]   [Oct]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH -next v2 2/6] ext4: introduce last_check_time record previous check time
On Fri 08-10-21 10:38:31, yebin wrote:
> On 2021/10/8 9:56, yebin wrote:
> > On 2021/10/7 20:31, Jan Kara wrote:
> > > On Sat 11-09-21 17:00:55, Ye Bin wrote:
> > > > kmmpd:
> > > > ...
> > > > diff = jiffies - last_update_time;
> > > > if (diff > mmp_check_interval * HZ) {
> > > > ...
> > > > As "mmp_check_interval = 2 * mmp_update_interval", 'diff' always little
> > > > than 'mmp_update_interval', so there will never trigger detection.
> > > > Introduce last_check_time record previous check time.
> > > >
> > > > Signed-off-by: Ye Bin <yebin10@huawei.com>
> > > I think the check is there only for the case where write_mmp_block() +
> > > sleep took longer than mmp_check_interval. I agree that should rarely
> > > happen but on a really busy system it is possible and in that case
> > > we would
> > > miss updating mmp block for too long and so another node could have
> > > started
> > > using the filesystem. I actually don't see a reason why kmmpd should be
> > > checking the block each mmp_check_interval as you do -
> > > mmp_check_interval
> > > is just for ext4_multi_mount_protect() to know how long it should wait
> > > before considering mmp block stale... Am I missing something?
> > >
> > > Honza
> > I'm sorry, I didn't understand the detection mechanism here before. Now
> > I understand
> > the detection mechanism here.
> > As you said, it's just an abnormal protection. There's really no problem.
> >
> Yeah, i did test as following steps
> hostA hostB
> mount
> ext4_multi_mount_protect -> seq == EXT4_MMP_SEQ_CLEAN
> delay 5s after label "skip" so hostB will see seq is
> EXT4_MMP_SEQ_CLEAN
> mount
> ext4_multi_mount_protect -> seq == EXT4_MMP_SEQ_CLEAN
> run kmmpd
> run kmmpd
>
> Actually,in this situation kmmpd will not detect confliction.
> In ext4_multi_mount_protect function we write mmp data first and wait
> 'wait_time * HZ' seconds,
> read mmp data do check. Most of the time, If 'wait_time' is zero, it can pass
> check.

But how can be wait_time zero? As far as I'm reading the code, wait_time
must be at least EXT4_MMP_MIN_CHECK_INTERVAL...

Honza

--
Jan Kara <jack@suse.com>
SUSE Labs, CR

\
 
 \ /
  Last update: 2021-10-12 10:47    [W:0.313 / U:0.072 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site