]> git.ipfire.org Git - thirdparty/mdadm.git/commitdiff
mdmon: use activate spare for re-add
authorDan Williams <dan.j.williams@intel.com>
Tue, 12 Aug 2008 09:25:46 +0000 (02:25 -0700)
committerDan Williams <dan.j.williams@intel.com>
Tue, 12 Aug 2008 09:25:46 +0000 (02:25 -0700)
Disks that are not in-sync or failed are not assembled into member
arrays by mdadm.  Teach mdmon to resolve this situation by checking for
spares at start.  imsm_activate_spare() is updated to prefer devices
that can be re-added versus new spares.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
monitor.c
super-intel.c

index 7cce5a8b2f283eaf45d5f7c33e68fd293c9cf3a6..382cad44b76add76d50a78e8546096b1eec6928e 100644 (file)
--- a/monitor.c
+++ b/monitor.c
@@ -243,10 +243,15 @@ static int read_and_act(struct active_array *a)
                 * readonly ???
                 */
                get_resync_start(a);
-//             printf("Found a readonly array at %llu\n", a->resync_start);
-               if (a->resync_start == ~0ULL)
+               if (a->resync_start == ~0ULL) {
                        a->next_state = read_auto; /* array is clean */
-               else {
+                       /* give the metadata a chance to force active if
+                        * we have some recovery to do.  metadata sets
+                        * resync_start to !MaxSector in this case
+                        */
+                       a->container->ss->set_array_state(a, 1);
+               }
+               if (a->resync_start != ~0ULL) {
                        a->container->ss->set_array_state(a, 0);
                        a->next_state = active;
                }
index 62bd32da35737d7b3a0f657188f224f5e3d1cf0b..28b26d98a7b61595d0c3cd662502e2e88ce2cd43 100644 (file)
@@ -2169,6 +2169,10 @@ static void imsm_set_array_state(struct active_array *a, int consistent)
        failed = imsm_count_failed(super, map);
        map_state = imsm_check_degraded(super, inst, failed);
 
+       if (consistent && !dev->vol.dirty &&
+           (dev->vol.migr_state || map_state != IMSM_T_STATE_NORMAL))
+               a->resync_start = 0ULL;
+
        if (a->resync_start == ~0ULL) {
                /* complete recovery or initial resync */
                if (map->map_state != map_state) {
@@ -2191,6 +2195,7 @@ static void imsm_set_array_state(struct active_array *a, int consistent)
                dev->vol.migr_state = 1;
                dev->vol.migr_type = failed ? 1 : 0;
                dup_map(dev);
+               a->check_degraded = 1;
                super->updates_pending++;
        }
 
@@ -2309,6 +2314,93 @@ static void imsm_sync_metadata(struct supertype *container)
        super->updates_pending = 0;
 }
 
+static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
+{
+       struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
+       struct imsm_map *map = get_imsm_map(dev, 0);
+       int i = get_imsm_disk_idx(map, idx);
+       struct dl *dl;
+
+       for (dl = super->disks; dl; dl = dl->next)
+               if (dl->index == i)
+                       break;
+
+       if (__le32_to_cpu(dl->disk.status) & FAILED_DISK)
+               dl = NULL;
+
+       if (dl)
+               dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
+
+       return dl;
+}
+
+static struct dl *imsm_add_spare(struct intel_super *super, int idx, struct active_array *a)
+{
+       struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
+       struct imsm_map *map = get_imsm_map(dev, 0);
+       unsigned long long esize;
+       unsigned long long pos;
+       struct mdinfo *d;
+       struct extent *ex;
+       int j;
+       int found;
+       __u32 array_start;
+       struct dl *dl;
+
+       for (dl = super->disks; dl; dl = dl->next) {
+               /* If in this array, skip */
+               for (d = a->info.devs ; d ; d = d->next)
+                       if (d->disk.major == dl->major &&
+                           d->disk.minor == dl->minor) {
+                               dprintf("%x:%x already in array\n", dl->major, dl->minor);
+                               break;
+                       }
+               if (d)
+                       continue;
+
+               /* Does this unused device have the requisite free space?
+                * We need a->info.component_size sectors
+                */
+               ex = get_extents(super, dl);
+               if (!ex) {
+                       dprintf("cannot get extents\n");
+                       continue;
+               }
+               found = 0;
+               j = 0;
+               pos = 0;
+               array_start = __le32_to_cpu(map->pba_of_lba0);
+
+               do {
+                       /* check that we can start at pba_of_lba0 with
+                        * a->info.component_size of space
+                        */
+                       esize = ex[j].start - pos;
+                       if (array_start >= pos &&
+                           array_start + a->info.component_size < ex[j].start) {
+                               found = 1;
+                               break;
+                       }
+                       pos = ex[j].start + ex[j].size;
+                       j++;
+                           
+               } while (ex[j-1].size);
+
+               free(ex);
+               if (!found) {
+                       dprintf("%x:%x does not have %llu at %d\n",
+                               dl->major, dl->minor,
+                               a->info.component_size,
+                               __le32_to_cpu(map->pba_of_lba0));
+                       /* No room */
+                       continue;
+               } else
+                       break;
+       }
+
+       return dl;
+}
+
 static struct mdinfo *imsm_activate_spare(struct active_array *a,
                                          struct metadata_update **updates)
 {
@@ -2352,7 +2444,6 @@ static struct mdinfo *imsm_activate_spare(struct active_array *a,
                return NULL;
 
        /* For each slot, if it is not working, find a spare */
-       dl = super->disks;
        for (i = 0; i < a->info.array.raid_disks; i++) {
                for (d = a->info.devs ; d ; d = d->next)
                        if (d->disk.raid_disk == i)
@@ -2361,96 +2452,47 @@ static struct mdinfo *imsm_activate_spare(struct active_array *a,
                if (d && (d->state_fd >= 0))
                        continue;
 
-               /* OK, this device needs recovery.  Find a spare */
-               for ( ; dl ; dl = dl->next) {
-                       unsigned long long esize;
-                       unsigned long long pos;
-                       struct mdinfo *d2;
-                       struct extent *ex;
-                       int j;
-                       int found;
-                       __u32 array_start;
-
-                       /* If in this array, skip */
-                       for (d2 = a->info.devs ; d2 ; d2 = d2->next)
-                               if (d2->disk.major == dl->major &&
-                                   d2->disk.minor == dl->minor) {
-                                       dprintf("%x:%x already in array\n", dl->major, dl->minor);
-                                       break;
-                               }
-                       if (d2)
-                               continue;
-
-                       /* Does this unused device have the requisite free space?
-                        * We need a->info.component_size sectors
-                        */
-                       ex = get_extents(super, dl);
-                       if (!ex) {
-                               dprintf("cannot get extents\n");
-                               continue;
-                       }
-                       found = 0;
-                       j = 0;
-                       pos = 0;
-                       array_start = __le32_to_cpu(map->pba_of_lba0);
-
-                       do {
-                               /* check that we can start at pba_of_lba0 with
-                                * a->info.component_size of space
-                                */
-                               esize = ex[j].start - pos;
-                               if (array_start >= pos &&
-                                   array_start + a->info.component_size < ex[j].start) {
-                                       found = 1;
-                                       break;
-                               }
-                               pos = ex[j].start + ex[j].size;
-                               j++;
-                                   
-                       } while (ex[j-1].size);
-
-                       free(ex);
-                       if (!found) {
-                               dprintf("%x:%x does not have %llu at %d\n",
-                                       dl->major, dl->minor,
-                                       a->info.component_size,
-                                       __le32_to_cpu(map->pba_of_lba0));
-                               /* No room */
-                               continue;
-                       }
-
-                       /* found a usable disk with enough space */
-                       di = malloc(sizeof(*di));
-                       memset(di, 0, sizeof(*di));
-
-                       /* dl->index will be -1 in the case we are activating a
-                        * pristine spare.  imsm_process_update() will create a
-                        * new index in this case.  Once a disk is found to be
-                        * failed in all member arrays it is kicked from the
-                        * metadata
-                        */
-                       di->disk.number = dl->index;
+               /*
+                * OK, this device needs recovery.  Try to re-add the previous
+                * occupant of this slot, if this fails add a new spare
+                */
+               dl = imsm_readd(super, i, a);
+               if (!dl)
+                       dl = imsm_add_spare(super, i, a);
+               if (!dl)
+                       continue;
+               /* found a usable disk with enough space */
+               di = malloc(sizeof(*di));
+               memset(di, 0, sizeof(*di));
+
+               /* dl->index will be -1 in the case we are activating a
+                * pristine spare.  imsm_process_update() will create a
+                * new index in this case.  Once a disk is found to be
+                * failed in all member arrays it is kicked from the
+                * metadata
+                */
+               di->disk.number = dl->index;
 
-                       /* (ab)use di->devs to store a pointer to the device
-                        * we chose
-                        */
-                       di->devs = (struct mdinfo *) dl;
-
-                       di->disk.raid_disk = i;
-                       di->disk.major = dl->major;
-                       di->disk.minor = dl->minor;
-                       di->disk.state = 0;
-                       di->data_offset = array_start;
-                       di->component_size = a->info.component_size;
-                       di->container_member = inst;
-                       di->next = rv;
-                       rv = di;
-                       num_spares++;
-                       dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
-                               i, pos);
+               /* (ab)use di->devs to store a pointer to the device
+                * we chose
+                */
+               di->devs = (struct mdinfo *) dl;
+
+               di->disk.raid_disk = i;
+               di->disk.major = dl->major;
+               di->disk.minor = dl->minor;
+               di->disk.state = 0;
+               di->data_offset = __le32_to_cpu(map->pba_of_lba0);
+               di->component_size = a->info.component_size;
+               di->container_member = inst;
+               di->next = rv;
+               rv = di;
+               num_spares++;
+               dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
+                       i, di->data_offset);
 
-                       break;
-               }
+               break;
        }
 
        if (!rv)