Re: [PATCH -next v2 3/7] md: delay choosing sync direction to md_start_sync()

[Yu Kuai]

Hi,

在 2023/08/16 14:38, Xiao Ni 写道:
> On Tue, Aug 15, 2023 at 11:13 AM Yu Kuai <yukuai1@xxxxxxxxxxxxxxx> wrote:
> > From: Yu Kuai <yukuai3@xxxxxxxxxx>
> >
> > Before this patch, for read-write array:
> >
> > 1) md_check_recover() found that something need to be done, and it'll
> >     try to grab 'reconfig_mutex'. The case that md_check_recover() need
> >     to do something:
> >     - array is not suspend;
> >     - super_block need to be updated;
> >     - 'MD_RECOVERY_NEEDED' or ''MD_RECOVERY_DONE' is set;
> >     - unusual case related to safemode;
> >
> > 2) if 'MD_RECOVERY_RUNNING' is not set, and 'MD_RECOVERY_NEEDED' is set,
> >     md_check_recover() will try to choose a sync direction, and then
> >     queue a work md_start_sync().
> >
> > 3) md_start_sync() register sync_thread;
> >
> > After this patch,
> >
> > 1) is the same;
> > 2) if 'MD_RECOVERY_RUNNING' is not set, and 'MD_RECOVERY_NEEDED' is set,
> >     queue a work md_start_sync() directly;
> > 3) md_start_sync() will try to choose a sync direction, and then
> >     register sync_thread();
> >
> > Because 'MD_RECOVERY_RUNNING' is cleared when sync_thread is done, 2)
> > and 3) is always ran in serial and they can never concurrent, this
> > change should not introduce any behavior change for now.
> >
> > Also fix a problem that md_start_sync() can clear 'MD_RECOVERY_RUNNING'
> > without protection in error path, which might affect the logical in
> > md_check_recovery().
> >
> > The advantage to change this is that array reconfiguration is
> > independent from daemon now, and it'll be much easier to synchronize it
> > with io, consider that io may rely on daemon thread to be done.
>
> Hi Kuai
>
> What's the meaning of "array reconfiguration" here? "mdadm -f/-r/-a"
> something like this, right?. Because before and after this patch, only
> one sync thread can be running, so If we don't do this change, are
> there bugs or performance problems?

As we discussed([1]), and explained in patch 0, the purpose of this
change is to prepare to synchronize io with array reconfiguration(add
or remove rdev from array, for example, modify
conf->mirrors[].rdev/replacement for raid10).

Without this change, normal io can rely on daemon thread, while daemone
thread can change array configuration. raid1/raid10 record such io as
'io_queued', and can use freeze_array() to do synchronization in daemon
thread. However, other personalities have to implement such logical as
well, and I found it quite complicated, at least for raid456.

[1] 
https://lore.kernel.org/all/cb390b39-1b1e-04e1-55ad-2ff8afc47e1b@xxxxxxxxxx/

Thanks,
Kuai

> If it's only a patch that wants to make md_check_recovery more clearer
> and easier, I'm good with this idea too.