]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
md: simplify raid10 read_balance
authorNeilBrown <neilb@suse.de>
Wed, 11 May 2011 04:27:03 +0000 (14:27 +1000)
committerNeilBrown <neilb@suse.de>
Wed, 11 May 2011 04:27:03 +0000 (14:27 +1000)
raid10 read balance has two different loop for looking through
possible devices to chose the best.
Collapse those into one loop and generally make the code more
readable.

Signed-off-by: NeilBrown <neilb@suse.de>
drivers/md/raid10.c

index 8e9462626ec5cb8d7441b46ddae9a848772153fe..8ea0acad606bad395f8680d22a5a0de96b583986 100644 (file)
@@ -488,13 +488,19 @@ static int raid10_mergeable_bvec(struct request_queue *q,
 static int read_balance(conf_t *conf, r10bio_t *r10_bio)
 {
        const sector_t this_sector = r10_bio->sector;
-       int disk, slot, nslot;
+       int disk, slot;
        const int sectors = r10_bio->sectors;
-       sector_t new_distance, current_distance;
+       sector_t new_distance, best_dist;
        mdk_rdev_t *rdev;
+       int do_balance;
+       int best_slot;
 
        raid10_find_phys(conf, r10_bio);
        rcu_read_lock();
+retry:
+       best_slot = -1;
+       best_dist = MaxSector;
+       do_balance = 1;
        /*
         * Check if we can balance. We can balance on the whole
         * device if no resync is going on (recovery is ok), or below
@@ -502,86 +508,58 @@ static int read_balance(conf_t *conf, r10bio_t *r10_bio)
         * above the resync window.
         */
        if (conf->mddev->recovery_cp < MaxSector
-           && (this_sector + sectors >= conf->next_resync)) {
-               /* make sure that disk is operational */
-               slot = 0;
-               disk = r10_bio->devs[slot].devnum;
-
-               while ((rdev = rcu_dereference(conf->mirrors[disk].rdev)) == NULL ||
-                      r10_bio->devs[slot].bio == IO_BLOCKED ||
-                      !test_bit(In_sync, &rdev->flags)) {
-                       slot++;
-                       if (slot == conf->copies) {
-                               slot = 0;
-                               disk = -1;
-                               break;
-                       }
-                       disk = r10_bio->devs[slot].devnum;
-               }
-               goto rb_out;
-       }
-
+           && (this_sector + sectors >= conf->next_resync))
+               do_balance = 0;
 
-       /* make sure the disk is operational */
-       slot = 0;
-       disk = r10_bio->devs[slot].devnum;
-       while ((rdev=rcu_dereference(conf->mirrors[disk].rdev)) == NULL ||
-              r10_bio->devs[slot].bio == IO_BLOCKED ||
-              !test_bit(In_sync, &rdev->flags)) {
-               slot ++;
-               if (slot == conf->copies) {
-                       disk = -1;
-                       goto rb_out;
-               }
+       for (slot = 0; slot < conf->copies ; slot++) {
+               if (r10_bio->devs[slot].bio == IO_BLOCKED)
+                       continue;
                disk = r10_bio->devs[slot].devnum;
-       }
-
-
-       current_distance = abs(r10_bio->devs[slot].addr -
-                              conf->mirrors[disk].head_position);
-
-       /* Find the disk whose head is closest,
-        * or - for far > 1 - find the closest to partition beginning */
-
-       for (nslot = slot; nslot < conf->copies; nslot++) {
-               int ndisk = r10_bio->devs[nslot].devnum;
-
-
-               if ((rdev=rcu_dereference(conf->mirrors[ndisk].rdev)) == NULL ||
-                   r10_bio->devs[nslot].bio == IO_BLOCKED ||
-                   !test_bit(In_sync, &rdev->flags))
+               rdev = rcu_dereference(conf->mirrors[disk].rdev);
+               if (rdev == NULL)
+                       continue;
+               if (!test_bit(In_sync, &rdev->flags))
                        continue;
 
+               if (!do_balance)
+                       break;
+
                /* This optimisation is debatable, and completely destroys
                 * sequential read speed for 'far copies' arrays.  So only
                 * keep it for 'near' arrays, and review those later.
                 */
-               if (conf->near_copies > 1 && !atomic_read(&rdev->nr_pending)) {
-                       disk = ndisk;
-                       slot = nslot;
+               if (conf->near_copies > 1 && !atomic_read(&rdev->nr_pending))
                        break;
-               }
 
                /* for far > 1 always use the lowest address */
                if (conf->far_copies > 1)
-                       new_distance = r10_bio->devs[nslot].addr;
+                       new_distance = r10_bio->devs[slot].addr;
                else
-                       new_distance = abs(r10_bio->devs[nslot].addr -
-                                          conf->mirrors[ndisk].head_position);
-               if (new_distance < current_distance) {
-                       current_distance = new_distance;
-                       disk = ndisk;
-                       slot = nslot;
+                       new_distance = abs(r10_bio->devs[slot].addr -
+                                          conf->mirrors[disk].head_position);
+               if (new_distance < best_dist) {
+                       best_dist = new_distance;
+                       best_slot = slot;
                }
        }
+       if (slot == conf->copies)
+               slot = best_slot;
 
-rb_out:
-       r10_bio->read_slot = slot;
-/*     conf->next_seq_sect = this_sector + sectors;*/
-
-       if (disk >= 0 && (rdev=rcu_dereference(conf->mirrors[disk].rdev))!= NULL)
-               atomic_inc(&conf->mirrors[disk].rdev->nr_pending);
-       else
+       if (slot >= 0) {
+               disk = r10_bio->devs[slot].devnum;
+               rdev = rcu_dereference(conf->mirrors[disk].rdev);
+               if (!rdev)
+                       goto retry;
+               atomic_inc(&rdev->nr_pending);
+               if (test_bit(Faulty, &rdev->flags)) {
+                       /* Cannot risk returning a device that failed
+                        * before we inc'ed nr_pending
+                        */
+                       rdev_dec_pending(rdev, conf->mddev);
+                       goto retry;
+               }
+               r10_bio->read_slot = slot;
+       } else
                disk = -1;
        rcu_read_unlock();