Re: [Nbd] NBD, DISCARD/TRIM

[Paul Clements <paul.clements@...856...>]

On Fri, Mar 22, 2013 at 5:24 PM, folkert <folkert@...421...> wrote:
>> I had suspected the kill_bdev patch, as it's the only thing of any
>> significance that has changed recently related to nbd shutdown, but it
>
> As a note: this does not happen during shutdown. Well, maybe it does but
> that was not invoked at the time of the crash.
> The test was:
>  mount /dev/nbd0 /mnt -o noatime,discard,barrier=1  (or without -o)
>  and then on that mountpoint
>  iozone3 [...]

So maybe iozone is doing something odd here? Can you make this happen
without that?

>> seems that this is something that has just broken in ext3/ext4. Even
>> with kill_bdev calls removed, I still see BUGs (I think the same one
>> you reported [mine is fs/buffer.c:2958]) and NULL ptr derefs in jbd
>> when I abort the nbd connection with a mounted filesystem.
>> Do you know or can you (or anyone else) check when this started?
>
> Kernel 3.2 also exhibits this sudden:
>
>         [  489.655034] block nbd0: shutting down socket
>         [  489.655093] end_request: I/O error, dev nbd0, sector 358648
>         [  489.655571] block nbd0: queue cleared
>
> errors but no crashes.

OK, so no BUG...yes, I think the BUG specifically is due to the
kill_bdev patch, recently introduced. But there are other problems...


>> As for what is causing your client to stop in the first place, I'm not
>> sure. How do you have the FS mounted? Are you testing discard, i.e.,
>> you have the option enabled in the nbd-server?
>
> As I wrote above.
> But also with just plain mount /dev/nbd0 /mnt. This is always with a
> filesystem build with mkfs.ext4 -j. Furthermore, this is with the stock

You're saying _any_ mount of ext4 over nbd does this sudden shutdown,
with 3.2+ kernel? Surely that's not the case?

I certainly don't see that here, but I'm not running stock nbd tools.
Can anyone confirm or deny this?

Thanks,
Paul