Peter,
"Peter T. Breuer" wrote:
>
> "A month of sundays ago Lars Kellogg-Stedman wrote:"
> > > Are you using Russell/Martin's md patch?
> >
> > Possibly. I'm using a patch Scott (Smyth) sent me, which I presume
> > incorporates this work -- the patch fixed the problem with the raid5 resync
> > process wedging when the volume was mounted.
>
> doesn't ever have null pointers), and in raid5 it converts
> raid5_sync_request to using sector counts instead of block counts.
> That's fine, but in one place it makes an additional change to sector
> counts in a call to md_done_sync that looks unwarranted to me (in my xfs
> snapshot) since md_done_sync uses blocks, not sectors:
>
> - md_done_sync(conf->mddev, (sh->size>>10) - sh->sync_redone,1);
> + md_done_sync(conf->mddev, (sh->size>>9) - sh->sync_redone, 1);
>
> I'm not using raid5 so I can't comment, but that bit seemed offbase
> to me when I read the patch.
>
> void md_done_sync(mddev_t *mddev, int blocks, int ok)
> // stuff that subtracts the blocks count from the
> // to-be-recovered list
Then you don't have the full patch. The patch from Russell/Martin
changed md.c also to use sectors instead of block in all cases. What
their patch didn't include was changes to raid1.c, which I stumbled
across when I first attempted to use it. I've sent a patch to them for
raid1.c, and also a patch to make the BLKSETSIZE ioctl work, too, but
I've not heard anything from either of them on it.
--
"Men occasionally stumble over the truth, but most of them pick
themselves up and hurry off as if nothing had happened."
-- Winston Churchill
Danny
|