raid1 critical sections not protected in 2.4.x?

raid1 critical sections not protected in 2.4.x?

[Steven Dake]

Folks,

I have been looking at the raid1.c code.  There is a kernel thread
raid1d that could be scheduled on processor A.  Then on processor B,
other raid code could be scheduled as per each block I/O request.

The particular critical section is the usage of the conf variable, and
specifically conf->mirrors.  raid1d reads conf->mirrors[x] while other
parts of the md code alter it simulatenously.

Is this a bug?  It might explain how people see counts get out of wack
on the media...

Thanks
-steve

Re: raid1 critical sections not protected in 2.4.x?

[Lars Marowsky-Bree]

On 2003-10-09T10:41:59,
   Steven Dake <sdake@mvista.com> said:

> I have been looking at the raid1.c code.  There is a kernel thread
> raid1d that could be scheduled on processor A.  Then on processor B,
> other raid code could be scheduled as per each block I/O request.

Did you follow the recent locking fixes to md I discussed with Neil? I
think this fixes some of these; at least we have a confirmation from IBM
that some raid1 races (not resyncing after a hotadd, no progress during
resync etc) went away with it.


Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
High Availability & Clustering		ever tried. ever failed. no matter.
SuSE Labs				try again. fail again. fail better.
Research & Development, SUSE LINUX AG		-- Samuel Beckett

-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: raid1 critical sections not protected in 2.4.x?

[Steven Dake]

Could you provide a pointer to those md locking fixes?

Thanks!
-steve

On Fri, 2003-10-10 at 01:14, Lars Marowsky-Bree wrote:
> On 2003-10-09T10:41:59,
>    Steven Dake <sdake@mvista.com> said:
> 
> > I have been looking at the raid1.c code.  There is a kernel thread
> > raid1d that could be scheduled on processor A.  Then on processor B,
> > other raid code could be scheduled as per each block I/O request.
> 
> Did you follow the recent locking fixes to md I discussed with Neil? I
> think this fixes some of these; at least we have a confirmation from IBM
> that some raid1 races (not resyncing after a hotadd, no progress during
> resync etc) went away with it.
> 
> 
> Sincerely,
>     Lars Marowsky-Brée <lmb@suse.de>

-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: raid1 critical sections not protected in 2.4.x?

[Lars Marowsky-Bree]

On 2003-10-13T14:29:30,
   Steven Dake <sdake@mvista.com> said:

> Could you provide a pointer to those md locking fixes?

I'm on a train right now :), but I believe I put a copy of the most
recent version at ftp://ftp.suse.com/pub/people/lmb/md/md-locking. If
not, I'll put them there as soon as possible ;-)

Neil is currently travelling too, so he hadn't had time to review it
yet. My version deviates from his to fix some races which we found.

There's some more work I think needs doing, namely moving all of the md
global structs and arrays into a single struct md_whatever[] for
tidyness, but I'm not sure whether I still want to touch the 2.4 code at
all ;-)


Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
High Availability & Clustering		ever tried. ever failed. no matter.
SuSE Labs				try again. fail again. fail better.
Research & Development, SUSE LINUX AG		-- Samuel Beckett

-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html