Merge to Fedora kernel-2.6.18-1.2224_FC5 patched with stable patch-2.6.18.1-vs2.0...
[linux-2.6.git] / drivers / md / raid1.c
index 5a54494..3b4d69c 100644 (file)
@@ -374,26 +374,26 @@ static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int
         * already.
         */
        if (atomic_dec_and_test(&r1_bio->remaining)) {
-               if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
+               if (test_bit(R1BIO_BarrierRetry, &r1_bio->state))
                        reschedule_retry(r1_bio);
-                       goto out;
-               }
-               /* it really is the end of this request */
-               if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
-                       /* free extra copy of the data pages */
-                       int i = bio->bi_vcnt;
-                       while (i--)
-                               safe_put_page(bio->bi_io_vec[i].bv_page);
+               else {
+                       /* it really is the end of this request */
+                       if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
+                               /* free extra copy of the data pages */
+                               int i = bio->bi_vcnt;
+                               while (i--)
+                                       safe_put_page(bio->bi_io_vec[i].bv_page);
+                       }
+                       /* clear the bitmap if all writes complete successfully */
+                       bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
+                                       r1_bio->sectors,
+                                       !test_bit(R1BIO_Degraded, &r1_bio->state),
+                                       behind);
+                       md_write_end(r1_bio->mddev);
+                       raid_end_bio_io(r1_bio);
                }
-               /* clear the bitmap if all writes complete successfully */
-               bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
-                               r1_bio->sectors,
-                               !test_bit(R1BIO_Degraded, &r1_bio->state),
-                               behind);
-               md_write_end(r1_bio->mddev);
-               raid_end_bio_io(r1_bio);
        }
- out:
+
        if (to_put)
                bio_put(to_put);
 
@@ -930,10 +930,13 @@ static void status(struct seq_file *seq, mddev_t *mddev)
 
        seq_printf(seq, " [%d/%d] [", conf->raid_disks,
                                                conf->working_disks);
-       for (i = 0; i < conf->raid_disks; i++)
+       rcu_read_lock();
+       for (i = 0; i < conf->raid_disks; i++) {
+               mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
                seq_printf(seq, "%s",
-                             conf->mirrors[i].rdev &&
-                             test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
+                          rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
+       }
+       rcu_read_unlock();
        seq_printf(seq, "]");
 }
 
@@ -975,7 +978,6 @@ static void error(mddev_t *mddev, mdk_rdev_t *rdev)
 static void print_conf(conf_t *conf)
 {
        int i;
-       mirror_info_t *tmp;
 
        printk("RAID1 conf printout:\n");
        if (!conf) {
@@ -985,14 +987,17 @@ static void print_conf(conf_t *conf)
        printk(" --- wd:%d rd:%d\n", conf->working_disks,
                conf->raid_disks);
 
+       rcu_read_lock();
        for (i = 0; i < conf->raid_disks; i++) {
                char b[BDEVNAME_SIZE];
-               tmp = conf->mirrors + i;
-               if (tmp->rdev)
+               mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
+               if (rdev)
                        printk(" disk %d, wo:%d, o:%d, dev:%s\n",
-                               i, !test_bit(In_sync, &tmp->rdev->flags), !test_bit(Faulty, &tmp->rdev->flags),
-                               bdevname(tmp->rdev->bdev,b));
+                              i, !test_bit(In_sync, &rdev->flags),
+                              !test_bit(Faulty, &rdev->flags),
+                              bdevname(rdev->bdev,b));
        }
+       rcu_read_unlock();
 }
 
 static void close_sync(conf_t *conf)
@@ -1008,20 +1013,20 @@ static int raid1_spare_active(mddev_t *mddev)
 {
        int i;
        conf_t *conf = mddev->private;
-       mirror_info_t *tmp;
 
        /*
         * Find all failed disks within the RAID1 configuration 
-        * and mark them readable
+        * and mark them readable.
+        * Called under mddev lock, so rcu protection not needed.
         */
        for (i = 0; i < conf->raid_disks; i++) {
-               tmp = conf->mirrors + i;
-               if (tmp->rdev 
-                   && !test_bit(Faulty, &tmp->rdev->flags)
-                   && !test_bit(In_sync, &tmp->rdev->flags)) {
+               mdk_rdev_t *rdev = conf->mirrors[i].rdev;
+               if (rdev
+                   && !test_bit(Faulty, &rdev->flags)
+                   && !test_bit(In_sync, &rdev->flags)) {
                        conf->working_disks++;
                        mddev->degraded--;
-                       set_bit(In_sync, &tmp->rdev->flags);
+                       set_bit(In_sync, &rdev->flags);
                }
        }
 
@@ -1145,7 +1150,7 @@ static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
                long sectors_to_go = r1_bio->sectors;
                /* make sure these bits doesn't get cleared. */
                do {
-                       bitmap_end_sync(mddev->bitmap, r1_bio->sector,
+                       bitmap_end_sync(mddev->bitmap, s,
                                        &sync_blocks, 1);
                        s += sync_blocks;
                        sectors_to_go -= sync_blocks;
@@ -1237,7 +1242,7 @@ static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
                /* 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
+                * read into the pages we already have so 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
@@ -1257,6 +1262,10 @@ static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
                                s = PAGE_SIZE >> 9;
                        do {
                                if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
+                                       /* No rcu protection needed here devices
+                                        * can only be removed when no resync is
+                                        * active, and resync is currently active
+                                        */
                                        rdev = conf->mirrors[d].rdev;
                                        if (sync_page_io(rdev->bdev,
                                                         sect + rdev->data_offset,
@@ -1463,6 +1472,11 @@ static void raid1d(mddev_t *mddev)
                                        s = PAGE_SIZE >> 9;
 
                                do {
+                                       /* Note: no rcu protection needed here
+                                        * as this is synchronous in the raid1d thread
+                                        * which is the thread that might remove
+                                        * a device.  If raid1d ever becomes multi-threaded....
+                                        */
                                        rdev = conf->mirrors[d].rdev;
                                        if (rdev &&
                                            test_bit(In_sync, &rdev->flags) &&
@@ -1508,9 +1522,11 @@ static void raid1d(mddev_t *mddev)
                                                                         s<<9, conf->tmppage, READ) == 0)
                                                                /* Well, this device is dead */
                                                                md_error(mddev, rdev);
-                                                       else
+                                                       else {
                                                                atomic_add(s, &rdev->corrected_errors);
-
+                                                               printk(KERN_INFO "raid1:%s: read error corrected (%d sectors at %llu on %s)\n",
+                                                                      mdname(mddev), s, (unsigned long long)(sect + rdev->data_offset), bdevname(rdev->bdev, b));
+                                                       }
                                                }
                                        }
                                } else {
@@ -1624,6 +1640,13 @@ static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, i
                return 0;
        }
 
+       if (mddev->bitmap == NULL &&
+           mddev->recovery_cp == MaxSector &&
+           !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
+           conf->fullsync == 0) {
+               *skipped = 1;
+               return max_sector - sector_nr;
+       }
        /* before building a request, check if we can skip these blocks..
         * This call the bitmap_start_sync doesn't actually record anything
         */
@@ -1779,19 +1802,17 @@ static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, i
                for (i=0; i<conf->raid_disks; i++) {
                        bio = r1_bio->bios[i];
                        if (bio->bi_end_io == end_sync_read) {
-                               md_sync_acct(conf->mirrors[i].rdev->bdev, nr_sectors);
+                               md_sync_acct(bio->bi_bdev, nr_sectors);
                                generic_make_request(bio);
                        }
                }
        } else {
                atomic_set(&r1_bio->remaining, 1);
                bio = r1_bio->bios[r1_bio->read_disk];
-               md_sync_acct(conf->mirrors[r1_bio->read_disk].rdev->bdev,
-                            nr_sectors);
+               md_sync_acct(bio->bi_bdev, nr_sectors);
                generic_make_request(bio);
 
        }
-
        return nr_sectors;
 }
 
@@ -1890,7 +1911,8 @@ static int run(mddev_t *mddev)
 
                disk = conf->mirrors + i;
 
-               if (!disk->rdev) {
+               if (!disk->rdev ||
+                   !test_bit(In_sync, &disk->rdev->flags)) {
                        disk->head_position = 0;
                        mddev->degraded++;
                }