md/raid10: clear bad-block record when write succeeds.
authorNeilBrown <neilb@suse.de>
Thu, 28 Jul 2011 01:39:24 +0000 (11:39 +1000)
committerNeilBrown <neilb@suse.de>
Thu, 28 Jul 2011 01:39:24 +0000 (11:39 +1000)
If we succeed in writing to a block that was recorded as
being bad, we clear the bad-block record.

This requires some delayed handling as the bad-block-list update has
to happen in process-context.

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

index 13077a3fd7d2fa610c657e8cd02ed347faedd915..39b2058845f5786fc2273d0ee7d30fb277547538 100644 (file)
@@ -181,7 +181,7 @@ static void put_all_bios(conf_t *conf, r10bio_t *r10_bio)
 
        for (i = 0; i < conf->copies; i++) {
                struct bio **bio = & r10_bio->devs[i].bio;
-               if (*bio && *bio != IO_BLOCKED)
+               if (!BIO_SPECIAL(*bio))
                        bio_put(*bio);
                *bio = NULL;
        }
@@ -267,7 +267,8 @@ static inline void update_head_pos(int slot, r10bio_t *r10_bio)
 /*
  * Find the disk number which triggered given bio
  */
-static int find_bio_disk(conf_t *conf, r10bio_t *r10_bio, struct bio *bio)
+static int find_bio_disk(conf_t *conf, r10bio_t *r10_bio,
+                        struct bio *bio, int *slotp)
 {
        int slot;
 
@@ -278,6 +279,8 @@ static int find_bio_disk(conf_t *conf, r10bio_t *r10_bio, struct bio *bio)
        BUG_ON(slot == conf->copies);
        update_head_pos(slot, r10_bio);
 
+       if (slotp)
+               *slotp = slot;
        return r10_bio->devs[slot].devnum;
 }
 
@@ -329,9 +332,11 @@ static void raid10_end_write_request(struct bio *bio, int error)
        int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
        r10bio_t *r10_bio = bio->bi_private;
        int dev;
+       int dec_rdev = 1;
        conf_t *conf = r10_bio->mddev->private;
+       int slot;
 
-       dev = find_bio_disk(conf, r10_bio, bio);
+       dev = find_bio_disk(conf, r10_bio, bio, &slot);
 
        /*
         * this branch is our 'one mirror IO has finished' event handler:
@@ -340,7 +345,7 @@ static void raid10_end_write_request(struct bio *bio, int error)
                md_error(r10_bio->mddev, conf->mirrors[dev].rdev);
                /* an I/O failed, we can't clear the bitmap */
                set_bit(R10BIO_Degraded, &r10_bio->state);
-       } else
+       } else {
                /*
                 * Set R10BIO_Uptodate in our master bio, so that
                 * we will return a good error code for to the higher
@@ -350,8 +355,23 @@ static void raid10_end_write_request(struct bio *bio, int error)
                 * user-side. So if something waits for IO, then it will
                 * wait for the 'master' bio.
                 */
+               sector_t first_bad;
+               int bad_sectors;
+
                set_bit(R10BIO_Uptodate, &r10_bio->state);
 
+               /* Maybe we can clear some bad blocks. */
+               if (is_badblock(conf->mirrors[dev].rdev,
+                               r10_bio->devs[slot].addr,
+                               r10_bio->sectors,
+                               &first_bad, &bad_sectors)) {
+                       bio_put(bio);
+                       r10_bio->devs[slot].bio = IO_MADE_GOOD;
+                       dec_rdev = 0;
+                       set_bit(R10BIO_MadeGood, &r10_bio->state);
+               }
+       }
+
        /*
         *
         * Let's see if all mirrored write operations have finished
@@ -364,10 +384,13 @@ static void raid10_end_write_request(struct bio *bio, int error)
                                !test_bit(R10BIO_Degraded, &r10_bio->state),
                                0);
                md_write_end(r10_bio->mddev);
-               raid_end_bio_io(r10_bio);
+               if (test_bit(R10BIO_MadeGood, &r10_bio->state))
+                       reschedule_retry(r10_bio);
+               else
+                       raid_end_bio_io(r10_bio);
        }
-
-       rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev);
+       if (dec_rdev)
+               rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev);
 }
 
 
@@ -1382,7 +1405,7 @@ static void end_sync_read(struct bio *bio, int error)
        conf_t *conf = r10_bio->mddev->private;
        int d;
 
-       d = find_bio_disk(conf, r10_bio, bio);
+       d = find_bio_disk(conf, r10_bio, bio, NULL);
 
        if (test_bit(BIO_UPTODATE, &bio->bi_flags))
                set_bit(R10BIO_Uptodate, &r10_bio->state);
@@ -1414,23 +1437,37 @@ static void end_sync_write(struct bio *bio, int error)
        mddev_t *mddev = r10_bio->mddev;
        conf_t *conf = mddev->private;
        int d;
+       sector_t first_bad;
+       int bad_sectors;
+       int slot;
 
-       d = find_bio_disk(conf, r10_bio, bio);
+       d = find_bio_disk(conf, r10_bio, bio, &slot);
 
        if (!uptodate)
                md_error(mddev, conf->mirrors[d].rdev);
+       else if (is_badblock(conf->mirrors[d].rdev,
+                            r10_bio->devs[slot].addr,
+                            r10_bio->sectors,
+                            &first_bad, &bad_sectors))
+               set_bit(R10BIO_MadeGood, &r10_bio->state);
 
        rdev_dec_pending(conf->mirrors[d].rdev, mddev);
        while (atomic_dec_and_test(&r10_bio->remaining)) {
                if (r10_bio->master_bio == NULL) {
                        /* the primary of several recovery bios */
                        sector_t s = r10_bio->sectors;
-                       put_buf(r10_bio);
+                       if (test_bit(R10BIO_MadeGood, &r10_bio->state))
+                               reschedule_retry(r10_bio);
+                       else
+                               put_buf(r10_bio);
                        md_done_sync(mddev, s, 1);
                        break;
                } else {
                        r10bio_t *r10_bio2 = (r10bio_t *)r10_bio->master_bio;
-                       put_buf(r10_bio);
+                       if (test_bit(R10BIO_MadeGood, &r10_bio->state))
+                               reschedule_retry(r10_bio);
+                       else
+                               put_buf(r10_bio);
                        r10_bio = r10_bio2;
                }
        }
@@ -1901,6 +1938,44 @@ read_more:
                generic_make_request(bio);
 }
 
+static void handle_write_completed(conf_t *conf, r10bio_t *r10_bio)
+{
+       /* Some sort of write request has finished and it
+        * succeeded in writing where we thought there was a
+        * bad block.  So forget the bad block.
+        */
+       int m;
+       mdk_rdev_t *rdev;
+
+       if (test_bit(R10BIO_IsSync, &r10_bio->state) ||
+           test_bit(R10BIO_IsRecover, &r10_bio->state)) {
+               for (m = 0; m < conf->copies; m++)
+                       if (r10_bio->devs[m].bio &&
+                           test_bit(BIO_UPTODATE,
+                                    &r10_bio->devs[m].bio->bi_flags)) {
+                               int dev = r10_bio->devs[m].devnum;
+                               rdev = conf->mirrors[dev].rdev;
+                               rdev_clear_badblocks(
+                                       rdev,
+                                       r10_bio->devs[m].addr,
+                                       r10_bio->sectors);
+                       }
+               put_buf(r10_bio);
+       } else {
+               for (m = 0; m < conf->copies; m++)
+                       if (r10_bio->devs[m].bio == IO_MADE_GOOD) {
+                               int dev = r10_bio->devs[m].devnum;
+                               rdev = conf->mirrors[dev].rdev;
+                               rdev_clear_badblocks(
+                                       rdev,
+                                       r10_bio->devs[m].addr,
+                                       r10_bio->sectors);
+                               rdev_dec_pending(rdev, conf->mddev);
+                       }
+               raid_end_bio_io(r10_bio);
+       }
+}
+
 static void raid10d(mddev_t *mddev)
 {
        r10bio_t *r10_bio;
@@ -1928,7 +2003,9 @@ static void raid10d(mddev_t *mddev)
 
                mddev = r10_bio->mddev;
                conf = mddev->private;
-               if (test_bit(R10BIO_IsSync, &r10_bio->state))
+               if (test_bit(R10BIO_MadeGood, &r10_bio->state))
+                       handle_write_completed(conf, r10_bio);
+               else if (test_bit(R10BIO_IsSync, &r10_bio->state))
                        sync_request_write(mddev, r10_bio);
                else if (test_bit(R10BIO_IsRecover, &r10_bio->state))
                        recovery_request_write(mddev, r10_bio);
index c646152ba4e48cb1b39ea306c49b60ce852c0725..d8b7f9af92d551111556d62ab7f676474dbafaa0 100644 (file)
@@ -118,6 +118,13 @@ struct r10bio_s {
  * level, we store IO_BLOCKED in the appropriate 'bios' pointer
  */
 #define IO_BLOCKED ((struct bio*)1)
+/* When we successfully write to a known bad-block, we need to remove the
+ * bad-block marking which must be done from process context.  So we record
+ * the success by setting devs[n].bio to IO_MADE_GOOD
+ */
+#define IO_MADE_GOOD ((struct bio *)2)
+
+#define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
 
 /* bits for r10bio.state */
 #define        R10BIO_Uptodate 0
@@ -128,4 +135,8 @@ struct r10bio_s {
  * so that raid10d knows what to do with them.
  */
 #define        R10BIO_ReadError 4
+/* If a write for this request means we can clear some
+ * known-bad-block records, we set this flag.
+ */
+#define        R10BIO_MadeGood 5
 #endif