raid5: handle io error of batch list
authorshli@kernel.org <shli@kernel.org>
Mon, 15 Dec 2014 01:57:03 +0000 (12:57 +1100)
committerNeilBrown <neilb@suse.de>
Tue, 21 Apr 2015 22:00:41 +0000 (08:00 +1000)
If io error happens in any stripe of a batch list, the batch list will be
split, then normal process will run for the stripes in the list.

Signed-off-by: Shaohua Li <shli@fusionio.com>
Signed-off-by: NeilBrown <neilb@suse.de>
drivers/md/raid5.c
drivers/md/raid5.h

index 717189e742434a68d36e420b43e32183fff08324..54f3cb312b425011fc098c8ad5c0d3221a68febb 100644 (file)
@@ -1070,6 +1070,9 @@ again:
                        pr_debug("skip op %ld on disc %d for sector %llu\n",
                                bi->bi_rw, i, (unsigned long long)sh->sector);
                        clear_bit(R5_LOCKED, &sh->dev[i].flags);
+                       if (sh->batch_head)
+                               set_bit(STRIPE_BATCH_ERR,
+                                       &sh->batch_head->state);
                        set_bit(STRIPE_HANDLE, &sh->state);
                }
 
@@ -2380,6 +2383,9 @@ static void raid5_end_write_request(struct bio *bi, int error)
        }
        rdev_dec_pending(rdev, conf->mddev);
 
+       if (sh->batch_head && !uptodate)
+               set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
+
        if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
                clear_bit(R5_LOCKED, &sh->dev[i].flags);
        set_bit(STRIPE_HANDLE, &sh->state);
@@ -4124,6 +4130,46 @@ static int clear_batch_ready(struct stripe_head *sh)
        return 0;
 }
 
+static void check_break_stripe_batch_list(struct stripe_head *sh)
+{
+       struct stripe_head *head_sh, *next;
+       int i;
+
+       if (!test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state))
+               return;
+
+       head_sh = sh;
+       do {
+               sh = list_first_entry(&sh->batch_list,
+                                     struct stripe_head, batch_list);
+               BUG_ON(sh == head_sh);
+       } while (!test_bit(STRIPE_DEGRADED, &sh->state));
+
+       while (sh != head_sh) {
+               next = list_first_entry(&sh->batch_list,
+                                       struct stripe_head, batch_list);
+               list_del_init(&sh->batch_list);
+
+               sh->state = head_sh->state & ~((1 << STRIPE_ACTIVE) |
+                                              (1 << STRIPE_PREREAD_ACTIVE) |
+                                              (1 << STRIPE_DEGRADED));
+               sh->check_state = head_sh->check_state;
+               sh->reconstruct_state = head_sh->reconstruct_state;
+               for (i = 0; i < sh->disks; i++)
+                       sh->dev[i].flags = head_sh->dev[i].flags &
+                               (~((1 << R5_WriteError) | (1 << R5_Overlap)));
+
+               spin_lock_irq(&sh->stripe_lock);
+               sh->batch_head = NULL;
+               spin_unlock_irq(&sh->stripe_lock);
+
+               set_bit(STRIPE_HANDLE, &sh->state);
+               release_stripe(sh);
+
+               sh = next;
+       }
+}
+
 static void handle_stripe(struct stripe_head *sh)
 {
        struct stripe_head_state s;
@@ -4146,6 +4192,8 @@ static void handle_stripe(struct stripe_head *sh)
                return;
        }
 
+       check_break_stripe_batch_list(sh);
+
        if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
                spin_lock(&sh->stripe_lock);
                /* Cannot process 'sync' concurrently with 'discard' */
index c8d0004dca8f15f4449ff9042300fcaed091d51b..cf3562e99440c33b72f6f781575a4b7c7adc7b11 100644 (file)
@@ -336,6 +336,7 @@ enum {
        STRIPE_DISCARD,
        STRIPE_ON_RELEASE_LIST,
        STRIPE_BATCH_READY,
+       STRIPE_BATCH_ERR,
 };
 
 /*