]> git.karo-electronics.de Git - linux-beck.git/commitdiff
dm raid: correct error messages on old metadata validation
authorHeinz Mauelshagen <heinzm@redhat.com>
Tue, 18 Oct 2016 15:46:45 +0000 (17:46 +0200)
committerMike Snitzer <snitzer@redhat.com>
Mon, 21 Nov 2016 14:52:05 +0000 (09:52 -0500)
When target 1.9.1 gets takeover/reshape requests on devices with old superblock
format not supporting such conversions and rejects them in super_init_validation(),
it logs bogus error message (e.g. Reshape when a takeover is requested).

Whilst on it, add messages for disk adding/removing and stripe sectors
reshape requests, use the newer rs_{takeover,reshape}_requested() API,
address a raid10 false positive in checking array positions and
remove rs_set_new() because device members are already set proper.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
drivers/md/dm-raid.c

index 6d53810963f7531a7e5048dad5c55ef53e8aa914..698b03e9d955b95f2d587834d37a05e39d5f9ae8 100644 (file)
@@ -2050,16 +2050,17 @@ static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
 
        mddev->reshape_position = MaxSector;
 
+       mddev->raid_disks = le32_to_cpu(sb->num_devices);
+       mddev->level = le32_to_cpu(sb->level);
+       mddev->layout = le32_to_cpu(sb->layout);
+       mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
+
        /*
         * Reshaping is supported, e.g. reshape_position is valid
         * in superblock and superblock content is authoritative.
         */
        if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
                /* Superblock is authoritative wrt given raid set layout! */
-               mddev->raid_disks = le32_to_cpu(sb->num_devices);
-               mddev->level = le32_to_cpu(sb->level);
-               mddev->layout = le32_to_cpu(sb->layout);
-               mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
                mddev->new_level = le32_to_cpu(sb->new_level);
                mddev->new_layout = le32_to_cpu(sb->new_layout);
                mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
@@ -2087,38 +2088,44 @@ static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
                /*
                 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
                 */
-               if (le32_to_cpu(sb->level) != mddev->new_level) {
-                       DMERR("Reshaping/takeover raid sets not yet supported. (raid level/stripes/size change)");
-                       return -EINVAL;
-               }
-               if (le32_to_cpu(sb->layout) != mddev->new_layout) {
-                       DMERR("Reshaping raid sets not yet supported. (raid layout change)");
-                       DMERR("  0x%X vs 0x%X", le32_to_cpu(sb->layout), mddev->layout);
-                       DMERR("  Old layout: %s w/ %d copies",
-                             raid10_md_layout_to_format(le32_to_cpu(sb->layout)),
-                             raid10_md_layout_to_copies(le32_to_cpu(sb->layout)));
-                       DMERR("  New layout: %s w/ %d copies",
-                             raid10_md_layout_to_format(mddev->layout),
-                             raid10_md_layout_to_copies(mddev->layout));
-                       return -EINVAL;
-               }
-               if (le32_to_cpu(sb->stripe_sectors) != mddev->new_chunk_sectors) {
-                       DMERR("Reshaping raid sets not yet supported. (stripe sectors change)");
-                       return -EINVAL;
-               }
+               struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
+               struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
 
-               /* We can only change the number of devices in raid1 with old (i.e. pre 1.0.7) metadata */
-               if (!rt_is_raid1(rs->raid_type) &&
-                   (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
-                       DMERR("Reshaping raid sets not yet supported. (device count change from %u to %u)",
-                             sb->num_devices, mddev->raid_disks);
+               if (rs_takeover_requested(rs)) {
+                       if (rt_cur && rt_new)
+                               DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
+                                     rt_cur->name, rt_new->name);
+                       else
+                               DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
+                       return -EINVAL;
+               } else if (rs_reshape_requested(rs)) {
+                       DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
+                       if (mddev->layout != mddev->new_layout) {
+                               if (rt_cur && rt_new)
+                                       DMERR("  current layout %s vs new layout %s",
+                                             rt_cur->name, rt_new->name);
+                               else
+                                       DMERR("  current layout 0x%X vs new layout 0x%X",
+                                             le32_to_cpu(sb->layout), mddev->new_layout);
+                       }
+                       if (mddev->chunk_sectors != mddev->new_chunk_sectors)
+                               DMERR("  current stripe sectors %u vs new stripe sectors %u",
+                                     mddev->chunk_sectors, mddev->new_chunk_sectors);
+                       if (rs->delta_disks)
+                               DMERR("  current %u disks vs new %u disks",
+                                     mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
+                       if (rs_is_raid10(rs)) {
+                               DMERR("  Old layout: %s w/ %u copies",
+                                     raid10_md_layout_to_format(mddev->layout),
+                                     raid10_md_layout_to_copies(mddev->layout));
+                               DMERR("  New layout: %s w/ %u copies",
+                                     raid10_md_layout_to_format(mddev->new_layout),
+                                     raid10_md_layout_to_copies(mddev->new_layout));
+                       }
                        return -EINVAL;
                }
 
                DMINFO("Discovered old metadata format; upgrading to extended metadata format");
-
-               /* Table line is checked vs. authoritative superblock */
-               rs_set_new(rs);
        }
 
        if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
@@ -2211,7 +2218,7 @@ static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
                                continue;
 
                        if (role != r->raid_disk) {
-                               if (__is_raid10_near(mddev->layout)) {
+                               if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
                                        if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
                                            rs->raid_disks % rs->raid10_copies) {
                                                rs->ti->error =