[PATCH] md: better handling for read error in raid1 during resync
authorNeilBrown <neilb@suse.de>
Fri, 6 Jan 2006 08:20:22 +0000 (00:20 -0800)
committerLinus Torvalds <torvalds@g5.osdl.org>
Fri, 6 Jan 2006 16:34:04 +0000 (08:34 -0800)
Handling of read errors during resync is separate from handling of read errors
during normal IO in raid1.  A previous patch added support for read errors
during normal IO.  This one adds support for read errors during resync or
recovery.

The key differences are that we don't need to freeze the array, because the
normal handling of resync means that this part of the array will be idle
except for resync, and the read/overwrite/re-read is needed in a separate
piece of code.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
drivers/md/raid1.c

index ea1f1eb93c774fce472d210ba0092354835c5f55..14a8fe0349c75140d5f08e0bb6a151a7784c0f7b 100644 (file)
@@ -1072,9 +1072,7 @@ abort:
 
 static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
 {
-       int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
        r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
-       conf_t *conf = mddev_to_conf(r1_bio->mddev);
 
        if (bio->bi_size)
                return 1;
@@ -1087,10 +1085,7 @@ static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
         * or re-read if the read failed.
         * We don't do much here, just schedule handling by raid1d
         */
-       if (!uptodate) {
-               md_error(r1_bio->mddev,
-                        conf->mirrors[r1_bio->read_disk].rdev);
-       } else
+       if (test_bit(BIO_UPTODATE, &bio->bi_flags))
                set_bit(R1BIO_Uptodate, &r1_bio->state);
        reschedule_retry(r1_bio);
        return 0;
@@ -1134,27 +1129,89 @@ static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
 
        bio = r1_bio->bios[r1_bio->read_disk];
 
-/*
-       if (r1_bio->sector == 0) printk("First sync write startss\n");
-*/
+
        /*
         * schedule writes
         */
        if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
-               /*
-                * There is no point trying a read-for-reconstruct as
-                * reconstruct is about to be aborted
+               /* ouch - failed to read all of that.
+                * Try some synchronous reads of other devices to get
+                * good data, much like with normal read errors.  Only
+                * read into the pages we already have so they we don't
+                * need to re-issue the read request.
+                * We don't need to freeze the array, because being in an
+                * active sync request, there is no normal IO, and
+                * no overlapping syncs.
                 */
-               char b[BDEVNAME_SIZE];
-               printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
-                       " for block %llu\n",
-                       bdevname(bio->bi_bdev,b), 
-                       (unsigned long long)r1_bio->sector);
-               md_done_sync(mddev, r1_bio->sectors, 0);
-               put_buf(r1_bio);
-               return;
+               sector_t sect = r1_bio->sector;
+               int sectors = r1_bio->sectors;
+               int idx = 0;
+
+               while(sectors) {
+                       int s = sectors;
+                       int d = r1_bio->read_disk;
+                       int success = 0;
+                       mdk_rdev_t *rdev;
+
+                       if (s > (PAGE_SIZE>>9))
+                               s = PAGE_SIZE >> 9;
+                       do {
+                               if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
+                                       rdev = conf->mirrors[d].rdev;
+                                       if (sync_page_io(rdev->bdev,
+                                                        sect + rdev->data_offset,
+                                                        s<<9,
+                                                        bio->bi_io_vec[idx].bv_page,
+                                                        READ)) {
+                                               success = 1;
+                                               break;
+                                       }
+                               }
+                               d++;
+                               if (d == conf->raid_disks)
+                                       d = 0;
+                       } while (!success && d != r1_bio->read_disk);
+
+                       if (success) {
+                               /* write it back and re-read */
+                               set_bit(R1BIO_Uptodate, &r1_bio->state);
+                               while (d != r1_bio->read_disk) {
+                                       if (d == 0)
+                                               d = conf->raid_disks;
+                                       d--;
+                                       if (r1_bio->bios[d]->bi_end_io != end_sync_read)
+                                               continue;
+                                       rdev = conf->mirrors[d].rdev;
+                                       if (sync_page_io(rdev->bdev,
+                                                        sect + rdev->data_offset,
+                                                        s<<9,
+                                                        bio->bi_io_vec[idx].bv_page,
+                                                        WRITE) == 0 ||
+                                           sync_page_io(rdev->bdev,
+                                                        sect + rdev->data_offset,
+                                                        s<<9,
+                                                        bio->bi_io_vec[idx].bv_page,
+                                                        READ) == 0) {
+                                               md_error(mddev, rdev);
+                                       }
+                               }
+                       } else {
+                               char b[BDEVNAME_SIZE];
+                               /* Cannot read from anywhere, array is toast */
+                               md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
+                               printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
+                                      " for block %llu\n",
+                                      bdevname(bio->bi_bdev,b),
+                                      (unsigned long long)r1_bio->sector);
+                               md_done_sync(mddev, r1_bio->sectors, 0);
+                               put_buf(r1_bio);
+                               return;
+                       }
+                       sectors -= s;
+                       sect += s;
+                       idx ++;
+               }
        }
-
        atomic_set(&r1_bio->remaining, 1);
        for (i = 0; i < disks ; i++) {
                wbio = r1_bio->bios[i];