]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
md/raid5: change reshape-progress measurement to cope with reshaping backwards.
authorNeilBrown <neilb@suse.de>
Tue, 31 Mar 2009 04:16:46 +0000 (15:16 +1100)
committerNeilBrown <neilb@suse.de>
Tue, 31 Mar 2009 04:16:46 +0000 (15:16 +1100)
When reducing the number of devices in a raid4/5/6, the reshape
process has to start at the end of the array and work down to the
beginning.  So we need to handle expand_progress and expand_lo
differently.

This patch renames "expand_progress" and "expand_lo" to avoid the
implication that anything is getting bigger (expand->reshape) and
every place they are used, we make sure that they are used the right
way depending on whether delta_disks is positive or negative.

Signed-off-by: NeilBrown <neilb@suse.de>
drivers/md/raid5.c
drivers/md/raid5.h

index a0f22dd33234a15857fc5ec4f397ca8a8bab5466..1023c4e48a916d7c71112329647f78fe5c3e6518 100644 (file)
@@ -3593,24 +3593,28 @@ static int make_request(struct request_queue *q, struct bio * bi)
        retry:
                previous = 0;
                prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
-               if (likely(conf->expand_progress == MaxSector))
+               if (likely(conf->reshape_progress == MaxSector))
                        disks = conf->raid_disks;
                else {
-                       /* spinlock is needed as expand_progress may be
+                       /* spinlock is needed as reshape_progress may be
                         * 64bit on a 32bit platform, and so it might be
                         * possible to see a half-updated value
-                        * Ofcourse expand_progress could change after
+                        * Ofcourse reshape_progress could change after
                         * the lock is dropped, so once we get a reference
                         * to the stripe that we think it is, we will have
                         * to check again.
                         */
                        spin_lock_irq(&conf->device_lock);
                        disks = conf->raid_disks;
-                       if (logical_sector >= conf->expand_progress) {
+                       if (mddev->delta_disks < 0
+                           ? logical_sector < conf->reshape_progress
+                           : logical_sector >= conf->reshape_progress) {
                                disks = conf->previous_raid_disks;
                                previous = 1;
                        } else {
-                               if (logical_sector >= conf->expand_lo) {
+                               if (mddev->delta_disks < 0
+                                   ? logical_sector < conf->reshape_safe
+                                   : logical_sector >= conf->reshape_safe) {
                                        spin_unlock_irq(&conf->device_lock);
                                        schedule();
                                        goto retry;
@@ -3630,7 +3634,7 @@ static int make_request(struct request_queue *q, struct bio * bi)
                sh = get_active_stripe(conf, new_sector, previous,
                                       (bi->bi_rw&RWA_MASK));
                if (sh) {
-                       if (unlikely(conf->expand_progress != MaxSector)) {
+                       if (unlikely(conf->reshape_progress != MaxSector)) {
                                /* expansion might have moved on while waiting for a
                                 * stripe, so we must do the range check again.
                                 * Expansion could still move past after this
@@ -3641,8 +3645,10 @@ static int make_request(struct request_queue *q, struct bio * bi)
                                 */
                                int must_retry = 0;
                                spin_lock_irq(&conf->device_lock);
-                               if (logical_sector <  conf->expand_progress &&
-                                   disks == conf->previous_raid_disks)
+                               if ((mddev->delta_disks < 0
+                                    ? logical_sector >= conf->reshape_progress
+                                    : logical_sector < conf->reshape_progress)
+                                   && disks == conf->previous_raid_disks)
                                        /* mismatch, need to try again */
                                        must_retry = 1;
                                spin_unlock_irq(&conf->device_lock);
@@ -3720,13 +3726,20 @@ static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped
        int dd_idx;
        sector_t writepos, safepos, gap;
 
-       if (sector_nr == 0 &&
-           conf->expand_progress != 0) {
-               /* restarting in the middle, skip the initial sectors */
-               sector_nr = conf->expand_progress;
+       if (sector_nr == 0) {
+               /* If restarting in the middle, skip the initial sectors */
+               if (mddev->delta_disks < 0 &&
+                   conf->reshape_progress < raid5_size(mddev, 0, 0)) {
+                       sector_nr = raid5_size(mddev, 0, 0)
+                               - conf->reshape_progress;
+               } else if (mddev->delta_disks > 0 &&
+                          conf->reshape_progress > 0)
+                       sector_nr = conf->reshape_progress;
                sector_div(sector_nr, new_data_disks);
-               *skipped = 1;
-               return sector_nr;
+               if (sector_nr) {
+                       *skipped = 1;
+                       return sector_nr;
+               }
        }
 
        /* we update the metadata when there is more than 3Meg
@@ -3734,28 +3747,37 @@ static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped
         * probably be time based) or when the data about to be
         * copied would over-write the source of the data at
         * the front of the range.
-        * i.e. one new_stripe forward from expand_progress new_maps
-        * to after where expand_lo old_maps to
+        * i.e. one new_stripe along from reshape_progress new_maps
+        * to after where reshape_safe old_maps to
         */
-       writepos = conf->expand_progress +
-               conf->chunk_size/512*(new_data_disks);
+       writepos = conf->reshape_progress;
        sector_div(writepos, new_data_disks);
-       safepos = conf->expand_lo;
+       safepos = conf->reshape_safe;
        sector_div(safepos, data_disks);
-       gap = conf->expand_progress - conf->expand_lo;
+       if (mddev->delta_disks < 0) {
+               writepos -= conf->chunk_size/512;
+               safepos += conf->chunk_size/512;
+               gap = conf->reshape_safe - conf->reshape_progress;
+       } else {
+               writepos += conf->chunk_size/512;
+               safepos -= conf->chunk_size/512;
+               gap = conf->reshape_progress - conf->reshape_safe;
+       }
 
-       if (writepos >= safepos ||
+       if ((mddev->delta_disks < 0
+            ? writepos < safepos
+            : writepos > safepos) ||
            gap > (new_data_disks)*3000*2 /*3Meg*/) {
                /* Cannot proceed until we've updated the superblock... */
                wait_event(conf->wait_for_overlap,
                           atomic_read(&conf->reshape_stripes)==0);
-               mddev->reshape_position = conf->expand_progress;
+               mddev->reshape_position = conf->reshape_progress;
                set_bit(MD_CHANGE_DEVS, &mddev->flags);
                md_wakeup_thread(mddev->thread);
                wait_event(mddev->sb_wait, mddev->flags == 0 ||
                           kthread_should_stop());
                spin_lock_irq(&conf->device_lock);
-               conf->expand_lo = mddev->reshape_position;
+               conf->reshape_safe = mddev->reshape_position;
                spin_unlock_irq(&conf->device_lock);
                wake_up(&conf->wait_for_overlap);
        }
@@ -3792,7 +3814,10 @@ static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped
                release_stripe(sh);
        }
        spin_lock_irq(&conf->device_lock);
-       conf->expand_progress = (sector_nr + i) * new_data_disks;
+       if (mddev->delta_disks < 0)
+               conf->reshape_progress -= i * new_data_disks;
+       else
+               conf->reshape_progress += i * new_data_disks;
        spin_unlock_irq(&conf->device_lock);
        /* Ok, those stripe are ready. We can start scheduling
         * reads on the source stripes.
@@ -3823,14 +3848,14 @@ static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped
                /* Cannot proceed until we've updated the superblock... */
                wait_event(conf->wait_for_overlap,
                           atomic_read(&conf->reshape_stripes) == 0);
-               mddev->reshape_position = conf->expand_progress;
+               mddev->reshape_position = conf->reshape_progress;
                set_bit(MD_CHANGE_DEVS, &mddev->flags);
                md_wakeup_thread(mddev->thread);
                wait_event(mddev->sb_wait,
                           !test_bit(MD_CHANGE_DEVS, &mddev->flags)
                           || kthread_should_stop());
                spin_lock_irq(&conf->device_lock);
-               conf->expand_lo = mddev->reshape_position;
+               conf->reshape_safe = mddev->reshape_position;
                spin_unlock_irq(&conf->device_lock);
                wake_up(&conf->wait_for_overlap);
        }
@@ -4283,7 +4308,7 @@ static raid5_conf_t *setup_conf(mddev_t *mddev)
                conf->max_degraded = 1;
        conf->algorithm = mddev->new_layout;
        conf->max_nr_stripes = NR_STRIPES;
-       conf->expand_progress = mddev->reshape_position;
+       conf->reshape_progress = mddev->reshape_position;
 
        memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
                 conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
@@ -4441,9 +4466,9 @@ static int run(mddev_t *mddev)
 
        print_raid5_conf(conf);
 
-       if (conf->expand_progress != MaxSector) {
+       if (conf->reshape_progress != MaxSector) {
                printk("...ok start reshape thread\n");
-               conf->expand_lo = conf->expand_progress;
+               conf->reshape_safe = conf->reshape_progress;
                atomic_set(&conf->reshape_stripes, 0);
                clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
                clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
@@ -4782,8 +4807,11 @@ static int raid5_start_reshape(mddev_t *mddev)
        spin_lock_irq(&conf->device_lock);
        conf->previous_raid_disks = conf->raid_disks;
        conf->raid_disks += mddev->delta_disks;
-       conf->expand_progress = 0;
-       conf->expand_lo = 0;
+       if (mddev->delta_disks < 0)
+               conf->reshape_progress = raid5_size(mddev, 0, 0);
+       else
+               conf->reshape_progress = 0;
+       conf->reshape_safe = conf->reshape_progress;
        spin_unlock_irq(&conf->device_lock);
 
        /* Add some new drives, as many as will fit.
@@ -4825,7 +4853,7 @@ static int raid5_start_reshape(mddev_t *mddev)
                mddev->recovery = 0;
                spin_lock_irq(&conf->device_lock);
                mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
-               conf->expand_progress = MaxSector;
+               conf->reshape_progress = MaxSector;
                spin_unlock_irq(&conf->device_lock);
                return -EAGAIN;
        }
@@ -4842,7 +4870,7 @@ static void end_reshape(raid5_conf_t *conf)
 
                spin_lock_irq(&conf->device_lock);
                conf->previous_raid_disks = conf->raid_disks;
-               conf->expand_progress = MaxSector;
+               conf->reshape_progress = MaxSector;
                spin_unlock_irq(&conf->device_lock);
 
                /* read-ahead size must cover two whole stripes, which is
index c2f37f25ef44ece83a0d6075a4b48c40a8fe838c..b2edcc434e4170ac81dabb09fe89fd734218649a 100644 (file)
@@ -337,11 +337,16 @@ struct raid5_private_data {
        int                     raid_disks;
        int                     max_nr_stripes;
 
-       /* used during an expand */
-       sector_t                expand_progress;        /* MaxSector when no expand happening */
-       sector_t                expand_lo; /* from here up to expand_progress it out-of-bounds
-                                           * as we haven't flushed the metadata yet
-                                           */
+       /* reshape_progress is the leading edge of a 'reshape'
+        * It has value MaxSector when no reshape is happening
+        * If delta_disks < 0, it is the last sector we started work on,
+        * else is it the next sector to work on.
+        */
+       sector_t                reshape_progress;
+       /* reshape_safe is the trailing edge of a reshape.  We know that
+        * before (or after) this address, all reshape has completed.
+        */
+       sector_t                reshape_safe;
        int                     previous_raid_disks;
 
        struct list_head        handle_list; /* stripes needing handling */