releases/2.6.20.2/fix-various-bugs-with-aligned-reads-in-raid5.patch

   1 From stable-bounces@linux.kernel.org Tue Feb  6 15:45:04 2007
   2 From: Neil Brown <neilb@suse.de>
   3 Date: Wed, 7 Feb 2007 10:26:56 +1100
   4 Subject: Fix various bugs with aligned reads in RAID5.
   5 To: "Kai" <epimetreus@fastmail.fm>, Andrew Morton <akpm@linux-foundation.org>, stable@kernel.org
   6 Cc: linux-kernel@vger.kernel.org, Jens Axboe <jens.axboe@oracle.com>
   7 Message-ID: <17865.3776.511594.763544@notabene.brown>
   8
   9 From: Neil Brown <neilb@suse.de>
  10
  11 Fix various bugs with aligned reads in RAID5.
  12
  13 It is possible for raid5 to be sent a bio that is too big
  14 for an underlying device.  So if it is a READ that we
  15 pass stright down to a device, it will fail and confuse
  16 RAID5.
  17
  18 So in 'chunk_aligned_read' we check that the bio fits within the
  19 parameters for the target device and if it doesn't fit, fall back
  20 on reading through the stripe cache and making lots of one-page
  21 requests.
  22
  23 Note that this is the earliest time we can check against the device
  24 because earlier we don't have a lock on the device, so it could change
  25 underneath us.
  26
  27 Also, the code for handling a retry through the cache when a read
  28 fails has not been tested and was badly broken.  This patch fixes that
  29 code.
  30
  31 Signed-off-by: Neil Brown <neilb@suse.de>
  32 Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
  33
  34 ---
  35  drivers/md/raid5.c |   42 +++++++++++++++++++++++++++++++++++++++---
  36  1 file changed, 39 insertions(+), 3 deletions(-)
  37
  38 --- linux-2.6.20.1.orig/drivers/md/raid5.c
  39 +++ linux-2.6.20.1/drivers/md/raid5.c
  40 @@ -2620,7 +2620,7 @@ static struct bio *remove_bio_from_retry
  41         }
  42         bi = conf->retry_read_aligned_list;
  43         if(bi) {
  44 -               conf->retry_read_aligned = bi->bi_next;
  45 +               conf->retry_read_aligned_list = bi->bi_next;
  46                 bi->bi_next = NULL;
  47                 bi->bi_phys_segments = 1; /* biased count of active stripes */
  48                 bi->bi_hw_segments = 0; /* count of processed stripes */
  49 @@ -2669,6 +2669,27 @@ static int raid5_align_endio(struct bio
  50         return 0;
  51  }
  52
  53 +static int bio_fits_rdev(struct bio *bi)
  54 +{
  55 +       request_queue_t *q = bdev_get_queue(bi->bi_bdev);
  56 +
  57 +       if ((bi->bi_size>>9) > q->max_sectors)
  58 +               return 0;
  59 +       blk_recount_segments(q, bi);
  60 +       if (bi->bi_phys_segments > q->max_phys_segments ||
  61 +           bi->bi_hw_segments > q->max_hw_segments)
  62 +               return 0;
  63 +
  64 +       if (q->merge_bvec_fn)
  65 +               /* it's too hard to apply the merge_bvec_fn at this stage,
  66 +                * just just give up
  67 +                */
  68 +               return 0;
  69 +
  70 +       return 1;
  71 +}
  72 +
  73 +
  74  static int chunk_aligned_read(request_queue_t *q, struct bio * raid_bio)
  75  {
  76         mddev_t *mddev = q->queuedata;
  77 @@ -2715,6 +2736,13 @@ static int chunk_aligned_read(request_qu
  78                 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
  79                 align_bi->bi_sector += rdev->data_offset;
  80
  81 +               if (!bio_fits_rdev(align_bi)) {
  82 +                       /* too big in some way */
  83 +                       bio_put(align_bi);
  84 +                       rdev_dec_pending(rdev, mddev);
  85 +                       return 0;
  86 +               }
  87 +
  88                 spin_lock_irq(&conf->device_lock);
  89                 wait_event_lock_irq(conf->wait_for_stripe,
  90                                     conf->quiesce == 0,
  91 @@ -3107,7 +3135,9 @@ static int  retry_aligned_read(raid5_con
  92         last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
  93
  94         for (; logical_sector < last_sector;
  95 -            logical_sector += STRIPE_SECTORS, scnt++) {
  96 +            logical_sector += STRIPE_SECTORS,
  97 +                    sector += STRIPE_SECTORS,
  98 +                    scnt++) {
  99
 100                 if (scnt < raid_bio->bi_hw_segments)
 101                         /* already done this stripe */
 102 @@ -3123,7 +3153,13 @@ static int  retry_aligned_read(raid5_con
 103                 }
 104
 105                 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
 106 -               add_stripe_bio(sh, raid_bio, dd_idx, 0);
 107 +               if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
 108 +                       release_stripe(sh);
 109 +                       raid_bio->bi_hw_segments = scnt;
 110 +                       conf->retry_read_aligned = raid_bio;
 111 +                       return handled;
 112 +               }
 113 +
 114                 handle_stripe(sh, NULL);
 115                 release_stripe(sh);
 116                 handled++;