md: Add support for Raid0->Raid10 takeover
[safe/jmp/linux-2.6] / drivers / md / dm-region-hash.c
index 36dbe29..168bd38 100644 (file)
@@ -79,6 +79,11 @@ struct dm_region_hash {
        struct list_head recovered_regions;
        struct list_head failed_recovered_regions;
 
+       /*
+        * If there was a barrier failure no regions can be marked clean.
+        */
+       int barrier_failure;
+
        void *context;
        sector_t target_begin;
 
@@ -211,6 +216,7 @@ struct dm_region_hash *dm_region_hash_create(
        INIT_LIST_HEAD(&rh->quiesced_regions);
        INIT_LIST_HEAD(&rh->recovered_regions);
        INIT_LIST_HEAD(&rh->failed_recovered_regions);
+       rh->barrier_failure = 0;
 
        rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
                                                      sizeof(struct dm_region));
@@ -377,8 +383,6 @@ static void complete_resync_work(struct dm_region *reg, int success)
 /* dm_rh_mark_nosync
  * @ms
  * @bio
- * @done
- * @error
  *
  * The bio was written on some mirror(s) but failed on other mirror(s).
  * We can successfully endio the bio but should avoid the region being
@@ -386,8 +390,7 @@ static void complete_resync_work(struct dm_region *reg, int success)
  *
  * This function is _not_ safe in interrupt context!
  */
-void dm_rh_mark_nosync(struct dm_region_hash *rh,
-                      struct bio *bio, unsigned done, int error)
+void dm_rh_mark_nosync(struct dm_region_hash *rh, struct bio *bio)
 {
        unsigned long flags;
        struct dm_dirty_log *log = rh->log;
@@ -395,6 +398,11 @@ void dm_rh_mark_nosync(struct dm_region_hash *rh,
        region_t region = dm_rh_bio_to_region(rh, bio);
        int recovering = 0;
 
+       if (bio_empty_barrier(bio)) {
+               rh->barrier_failure = 1;
+               return;
+       }
+
        /* We must inform the log that the sync count has changed. */
        log->type->set_region_sync(log, region, 0);
 
@@ -419,7 +427,6 @@ void dm_rh_mark_nosync(struct dm_region_hash *rh,
        BUG_ON(!list_empty(&reg->list));
        spin_unlock_irqrestore(&rh->region_lock, flags);
 
-       bio_endio(bio, error);
        if (recovering)
                complete_resync_work(reg, 0);
 }
@@ -515,8 +522,11 @@ void dm_rh_inc_pending(struct dm_region_hash *rh, struct bio_list *bios)
 {
        struct bio *bio;
 
-       for (bio = bios->head; bio; bio = bio->bi_next)
+       for (bio = bios->head; bio; bio = bio->bi_next) {
+               if (bio_empty_barrier(bio))
+                       continue;
                rh_inc(rh, dm_rh_bio_to_region(rh, bio));
+       }
 }
 EXPORT_SYMBOL_GPL(dm_rh_inc_pending);
 
@@ -544,7 +554,14 @@ void dm_rh_dec(struct dm_region_hash *rh, region_t region)
                 */
 
                /* do nothing for DM_RH_NOSYNC */
-               if (reg->state == DM_RH_RECOVERING) {
+               if (unlikely(rh->barrier_failure)) {
+                       /*
+                        * If a write barrier failed some time ago, we
+                        * don't know whether or not this write made it
+                        * to the disk, so we must resync the device.
+                        */
+                       reg->state = DM_RH_NOSYNC;
+               } else if (reg->state == DM_RH_RECOVERING) {
                        list_add_tail(&reg->list, &rh->quiesced_regions);
                } else if (reg->state == DM_RH_DIRTY) {
                        reg->state = DM_RH_CLEAN;
@@ -643,10 +660,9 @@ void dm_rh_recovery_end(struct dm_region *reg, int success)
        spin_lock_irq(&rh->region_lock);
        if (success)
                list_add(&reg->list, &reg->rh->recovered_regions);
-       else {
-               reg->state = DM_RH_NOSYNC;
+       else
                list_add(&reg->list, &reg->rh->failed_recovered_regions);
-       }
+
        spin_unlock_irq(&rh->region_lock);
 
        rh->wakeup_workers(rh->context);