src/patches/suse-2.6.27.25/patches.fixes/0028-md-add-explicit-method-to-signal-the-end-of-a-reshap.patch

   1 From cea9c22800773cecb1d41f4a6139f9eb6a95368b Mon Sep 17 00:00:00 2001
   2 From: NeilBrown <neilb@suse.de>
   3 Date: Tue, 31 Mar 2009 15:15:05 +1100
   4 Subject: [PATCH] md: add explicit method to signal the end of a reshape.
   5
   6 Currently raid5 (the only module that supports restriping)
   7 notices that the reshape has finished be sync_request being
   8 given a large value, and handles any cleanup them.
   9
  10 This patch changes it so md_check_recovery calls into an
  11 explicit finish_reshape method as well.
  12
  13 The clean-up from sync_request can do things that need to be
  14 done promptly, typically things local to the raid5_conf_t
  15 structure.
  16
  17 The "finish_reshape" method is called under the mddev_lock
  18 so it can do things involving reconfiguring the device.
  19
  20 This allows us to get rid of md_set_array_sectors_locked, which
  21 would have caused a deadlock if you tried to stop and array
  22 while a reshape was happening.
  23
  24 Signed-off-by: NeilBrown <neilb@suse.de>
  25 ---
  26  drivers/md/md.c           |   11 ++--------
  27  drivers/md/raid5.c        |   50 +++++++++++++++++++++++++++-------------------
  28  include/linux/raid/md_k.h |    2 -
  29  3 files changed, 34 insertions(+), 29 deletions(-)
  30
  31 --- linux-2.6.27-SLE11_BRANCH.orig/drivers/md/md.c
  32 +++ linux-2.6.27-SLE11_BRANCH/drivers/md/md.c
  33 @@ -4759,14 +4759,6 @@ void md_set_array_sectors(mddev_t *mddev
  34  }
  35  EXPORT_SYMBOL(md_set_array_sectors);
  36
  37 -void md_set_array_sectors_lock(mddev_t *mddev, sector_t array_sectors)
  38 -{
  39 -       mddev_lock(mddev);
  40 -       md_set_array_sectors(mddev, array_sectors);
  41 -       mddev_unlock(mddev);
  42 -}
  43 -EXPORT_SYMBOL(md_set_array_sectors_lock);
  44 -
  45  static int update_size(mddev_t *mddev, sector_t num_sectors)
  46  {
  47         mdk_rdev_t * rdev;
  48 @@ -6313,6 +6305,9 @@ void md_check_recovery(mddev_t *mddev)
  49                                         sysfs_notify(&mddev->kobj, NULL,
  50                                                      "degraded");
  51                         }
  52 +                       if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
  53 +                           mddev->pers->finish_reshape)
  54 +                               mddev->pers->finish_reshape(mddev);
  55                         md_update_sb(mddev, 1);
  56
  57                         /* if array is no-longer degraded, then any saved_raid_disk
  58 --- linux-2.6.27-SLE11_BRANCH.orig/drivers/md/raid5.c
  59 +++ linux-2.6.27-SLE11_BRANCH/drivers/md/raid5.c
  60 @@ -3853,6 +3853,7 @@ static inline sector_t sync_request(mdde
  61         if (sector_nr >= max_sector) {
  62                 /* just being told to finish up .. nothing much to do */
  63                 unplug_slaves(mddev);
  64 +
  65                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
  66                         end_reshape(conf);
  67                         return 0;
  68 @@ -4798,43 +4799,49 @@ static int raid5_start_reshape(mddev_t *
  69
  70  static void end_reshape(raid5_conf_t *conf)
  71  {
  72 -       struct block_device *bdev;
  73
  74         if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
  75 -               mddev_t *mddev = conf->mddev;
  76 -
  77 -               md_set_array_sectors_lock(mddev, raid5_size(mddev, 0,
  78 -                                                      conf->raid_disks));
  79 -               set_capacity(mddev->gendisk, mddev->array_sectors);
  80 -               mddev->changed = 1;
  81 -               conf->previous_raid_disks = conf->raid_disks;
  82
  83 -               bdev = bdget_disk(conf->mddev->gendisk, 0);
  84 -               if (bdev) {
  85 -                       mutex_lock(&bdev->bd_inode->i_mutex);
  86 -                       i_size_write(bdev->bd_inode,
  87 -                                    (loff_t)conf->mddev->array_sectors << 9);
  88 -                       mutex_unlock(&bdev->bd_inode->i_mutex);
  89 -                       bdput(bdev);
  90 -               }
  91                 spin_lock_irq(&conf->device_lock);
  92 +               conf->previous_raid_disks = conf->raid_disks;
  93                 conf->expand_progress = MaxSector;
  94                 spin_unlock_irq(&conf->device_lock);
  95 -               conf->mddev->reshape_position = MaxSector;
  96
  97                 /* read-ahead size must cover two whole stripes, which is
  98                  * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
  99                  */
 100                 {
 101 -                       int data_disks = conf->previous_raid_disks - conf->max_degraded;
 102 -                       int stripe = data_disks *
 103 -                               (conf->mddev->chunk_size / PAGE_SIZE);
 104 +                       int data_disks = conf->raid_disks - conf->max_degraded;
 105 +                       int stripe = data_disks * (conf->chunk_size
 106 +                                                  / PAGE_SIZE);
 107                         if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
 108                                 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
 109                 }
 110         }
 111  }
 112
 113 +static void raid5_finish_reshape(mddev_t *mddev)
 114 +{
 115 +       struct block_device *bdev;
 116 +
 117 +       if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
 118 +
 119 +               md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
 120 +               set_capacity(mddev->gendisk, mddev->array_sectors);
 121 +               mddev->changed = 1;
 122 +               mddev->reshape_position = MaxSector;
 123 +
 124 +               bdev = bdget_disk(mddev->gendisk, 0);
 125 +               if (bdev) {
 126 +                       mutex_lock(&bdev->bd_inode->i_mutex);
 127 +                       i_size_write(bdev->bd_inode,
 128 +                                    (loff_t)mddev->array_sectors << 9);
 129 +                       mutex_unlock(&bdev->bd_inode->i_mutex);
 130 +                       bdput(bdev);
 131 +               }
 132 +       }
 133 +}
 134 +
 135  static void raid5_quiesce(mddev_t *mddev, int state)
 136  {
 137         raid5_conf_t *conf = mddev_to_conf(mddev);
 138 @@ -4883,6 +4890,7 @@ static struct mdk_personality raid6_pers
 139  #ifdef CONFIG_MD_RAID5_RESHAPE
 140         .check_reshape  = raid5_check_reshape,
 141         .start_reshape  = raid5_start_reshape,
 142 +       .finish_reshape = raid5_finish_reshape,
 143  #endif
 144         .quiesce        = raid5_quiesce,
 145  };
 146 @@ -4905,6 +4913,7 @@ static struct mdk_personality raid5_pers
 147  #ifdef CONFIG_MD_RAID5_RESHAPE
 148         .check_reshape  = raid5_check_reshape,
 149         .start_reshape  = raid5_start_reshape,
 150 +       .finish_reshape = raid5_finish_reshape,
 151  #endif
 152         .quiesce        = raid5_quiesce,
 153  };
 154 @@ -4928,6 +4937,7 @@ static struct mdk_personality raid4_pers
 155  #ifdef CONFIG_MD_RAID5_RESHAPE
 156         .check_reshape  = raid5_check_reshape,
 157         .start_reshape  = raid5_start_reshape,
 158 +       .finish_reshape = raid5_finish_reshape,
 159  #endif
 160         .quiesce        = raid5_quiesce,
 161  };
 162 --- linux-2.6.27-SLE11_BRANCH.orig/include/linux/raid/md_k.h
 163 +++ linux-2.6.27-SLE11_BRANCH/include/linux/raid/md_k.h
 164 @@ -310,6 +310,7 @@ struct mdk_personality
 165         sector_t (*size) (mddev_t *mddev, sector_t sectors, int raid_disks);
 166         int (*check_reshape) (mddev_t *mddev);
 167         int (*start_reshape) (mddev_t *mddev);
 168 +       void (*finish_reshape) (mddev_t *mddev);
 169         int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
 170         /* quiesce moves between quiescence states
 171          * 0 - fully active
 172 @@ -396,4 +397,3 @@ static inline void safe_put_page(struct
 173  #endif /* CONFIG_BLOCK */
 174  #endif
 175
 176 -extern void md_set_array_sectors_lock(mddev_t *mddev, sector_t array_sectors);