[PATCH] md: Fix issues with referencing rdev in md/raid1
authorNeilBrown <neilb@suse.de>
Fri, 1 Sep 2006 04:27:36 +0000 (21:27 -0700)
committerLinus Torvalds <torvalds@g5.osdl.org>
Fri, 1 Sep 2006 18:39:08 +0000 (11:39 -0700)
We need to be careful when referencing mirrors[i].rdev.  It can disappear
under us at various times.

So:
  fix a couple of problem places.
  comment a couple of non-problem places
  move an 'atomic_add' which deferences rdev down a little
    way to some where where it is sure to not be NULL.

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 87bfe9e7d8cad704fc598a727beb14cc978c0088..3b4d69c0562301b18039657ac0d47261a4eed6af 100644 (file)
@@ -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);
                }
        }
 
@@ -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) &&
@@ -1486,7 +1500,6 @@ static void raid1d(mddev_t *mddev)
                                                        d = conf->raid_disks;
                                                d--;
                                                rdev = conf->mirrors[d].rdev;
-                                               atomic_add(s, &rdev->corrected_errors);
                                                if (rdev &&
                                                    test_bit(In_sync, &rdev->flags)) {
                                                        if (sync_page_io(rdev->bdev,
@@ -1509,9 +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 {
@@ -1787,19 +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;
 }