]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - super-intel.c
Fix some issues with setting 'new' state of a reshape
[thirdparty/mdadm.git] / super-intel.c
index ffb03437697e55ede723d916862bf545b8d13df2..83fcb064f6abe896e6041b856babc02c00ca77db 100644 (file)
@@ -284,6 +284,13 @@ struct extent {
        unsigned long long start, size;
 };
 
+/* definitions of reshape process types */
+enum imsm_reshape_type {
+       CH_TAKEOVER,
+       CH_CHUNK_MIGR,
+       CH_LEVEL_MIGRATION
+};
+
 /* definition of messages passed to imsm_process_update */
 enum imsm_update_type {
        update_activate_spare,
@@ -3047,11 +3054,6 @@ static int load_super_imsm(struct supertype *st, int fd, char *devname)
        struct intel_super *super;
        int rv;
 
-#ifndef MDASSEMBLE
-       if (load_super_imsm_all(st, fd, &st->sb, devname) == 0)
-               return 0;
-#endif
-
        if (test_partition(fd))
                /* IMSM not allowed on partitions */
                return 1;
@@ -3741,7 +3743,6 @@ static int write_init_super_imsm(struct supertype *st)
        if (st->update_tail) {
                /* queue the recently created array / added disk
                 * as a metadata update */
-               struct dl *d;
                int rv;
 
                /* determine if we are creating a volume or adding a disk */
@@ -3753,11 +3754,6 @@ static int write_init_super_imsm(struct supertype *st)
                } else
                        rv = create_array(st, current_vol);
 
-               for (d = super->disks; d ; d = d->next) {
-                       close(d->fd);
-                       d->fd = -1;
-               }
-
                return rv;
        } else {
                struct dl *d;
@@ -4541,6 +4537,7 @@ static struct mdinfo *container_content_imsm(struct supertype *st, char *subarra
        for (i = 0; i < mpb->num_raid_devs; i++) {
                struct imsm_dev *dev;
                struct imsm_map *map;
+               struct imsm_map *map2;
                struct mdinfo *this;
                int slot;
                char *ep;
@@ -4551,6 +4548,7 @@ static struct mdinfo *container_content_imsm(struct supertype *st, char *subarra
 
                dev = get_imsm_dev(super, i);
                map = get_imsm_map(dev, 0);
+               map2 = get_imsm_map(dev, 1);
 
                /* do not publish arrays that are in the middle of an
                 * unsupported migration
@@ -4632,7 +4630,17 @@ static struct mdinfo *container_content_imsm(struct supertype *st, char *subarra
                        info_d->disk.minor = d->minor;
                        info_d->disk.raid_disk = slot;
                        info_d->recovery_start = recovery_start;
-
+                       if (map2) {
+                               if (slot < map2->num_members)
+                                       info_d->disk.state = (1 << MD_DISK_ACTIVE);
+                               else
+                                       this->array.spare_disks++;
+                       } else {
+                               if (slot < map->num_members)
+                                       info_d->disk.state = (1 << MD_DISK_ACTIVE);
+                               else
+                                       this->array.spare_disks++;
+                       }
                        if (info_d->recovery_start == MaxSector)
                                this->array.working_disks++;
 
@@ -4849,6 +4857,73 @@ static void handle_missing(struct intel_super *super, struct imsm_dev *dev)
 
 static void imsm_set_disk(struct active_array *a, int n, int state);
 
+static void imsm_progress_container_reshape(struct intel_super *super)
+{
+       /* if no device has a migr_state, but some device has a
+        * different number of members than the previous device, start
+        * changing the number of devices in this device to match
+        * previous.
+        */
+       struct imsm_super *mpb = super->anchor;
+       int prev_disks = -1;
+       int i;
+
+       for (i = 0; i < mpb->num_raid_devs; i++) {
+               struct imsm_dev *dev = get_imsm_dev(super, i);
+               struct imsm_map *map = get_imsm_map(dev, 0);
+               struct imsm_map *map2;
+               int prev_num_members;
+               int used_disks;
+
+               if (dev->vol.migr_state)
+                       return;
+
+               if (prev_disks == -1)
+                       prev_disks = map->num_members;
+               if (prev_disks == map->num_members)
+                       continue;
+
+               /* OK, this array needs to enter reshape mode.
+                * i.e it needs a migr_state
+                */
+
+               prev_num_members = map->num_members;
+               map->num_members = prev_disks;
+               dev->vol.migr_state = 1;
+               dev->vol.curr_migr_unit = 0;
+               dev->vol.migr_type = MIGR_GEN_MIGR;
+               for (i = prev_num_members;
+                    i < map->num_members; i++)
+                       set_imsm_ord_tbl_ent(map, i, i);
+               map2 = get_imsm_map(dev, 1);
+               /* Copy the current map */
+               memcpy(map2, map, sizeof_imsm_map(map));
+               map2->num_members = prev_num_members;
+
+               /* calculate new size
+                */
+               used_disks = imsm_num_data_members(dev, 0);
+               if (used_disks) {
+                       unsigned long long array_blocks;
+
+                       array_blocks =
+                               map->blocks_per_member
+                               * used_disks;
+                       /* round array size down to closest MB
+                        */
+                       array_blocks = (array_blocks
+                                       >> SECT_PER_MB_SHIFT)
+                               << SECT_PER_MB_SHIFT;
+                       dev->size_low =
+                               __cpu_to_le32((__u32)array_blocks);
+                       dev->size_high =
+                               __cpu_to_le32(
+                                       (__u32)(array_blocks >> 32));
+               }
+               super->updates_pending++;
+       }
+}
+
 /* Handle dirty -> clean transititions, resync and reshape.  Degraded and rebuild
  * states are handled in imsm_set_disk() with one exception, when a
  * resync is stopped due to a new failure this routine will set the
@@ -4918,6 +4993,7 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
                                                       * array size
                                                       */
                                super->updates_pending++;
+                               imsm_progress_container_reshape(super);
                        }                               
                }
        }
@@ -5335,6 +5411,12 @@ static struct mdinfo *imsm_activate_spare(struct active_array *a,
                /* No repair during migration */
                return NULL;
 
+       if (a->info.array.level == 4)
+               /* No repair for takeovered array
+                * imsm doesn't support raid4
+                */
+               return NULL;
+
        if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
                return NULL;
 
@@ -5563,6 +5645,119 @@ static int add_remove_disk_update(struct intel_super *super)
        return check_degraded;
 }
 
+static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
+                                               struct intel_super *super,
+                                               void ***space_list)
+{
+       struct dl *new_disk;
+       struct intel_dev *id;
+       int i;
+       int delta_disks = u->new_raid_disks - u->old_raid_disks;
+       int disk_count = u->old_raid_disks;
+       void **tofree = NULL;
+       int devices_to_reshape = 1;
+       struct imsm_super *mpb = super->anchor;
+       int ret_val = 0;
+
+       dprintf("imsm: imsm_process_update() for update_reshape\n");
+
+       /* enable spares to use in array */
+       for (i = 0; i < delta_disks; i++) {
+               new_disk = get_disk_super(super,
+                                         major(u->new_disks[i]),
+                                         minor(u->new_disks[i]));
+               dprintf("imsm: imsm_process_update(): new disk "
+                       "for reshape is: %i:%i (%p, index = %i)\n",
+                       major(u->new_disks[i]), minor(u->new_disks[i]),
+                       new_disk, new_disk->index);
+               if ((new_disk == NULL) ||
+                   ((new_disk->index >= 0) &&
+                    (new_disk->index < u->old_raid_disks)))
+                       goto update_reshape_exit;
+               new_disk->index = disk_count++;
+               /* slot to fill in autolayout
+                */
+               new_disk->raiddisk = new_disk->index;
+               new_disk->disk.status |=
+                       CONFIGURED_DISK;
+               new_disk->disk.status &= ~SPARE_DISK;
+       }
+
+       dprintf("imsm: process_update(): update_reshape: volume set"
+               " mpb->num_raid_devs = %i\n", mpb->num_raid_devs);
+       /* manage changes in volume
+        */
+       for (id = super->devlist ; id; id = id->next) {
+               void **sp = *space_list;
+               struct imsm_dev *newdev;
+               struct imsm_map *newmap, *oldmap;
+
+               if (!sp)
+                       continue;
+               *space_list = *sp;
+               newdev = (void*)sp;
+               /* Copy the dev, but not (all of) the map */
+               memcpy(newdev, id->dev, sizeof(*newdev));
+               oldmap = get_imsm_map(id->dev, 0);
+               newmap = get_imsm_map(newdev, 0);
+               /* Copy the current map */
+               memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
+               /* update one device only
+                */
+               if (devices_to_reshape) {
+                       int used_disks;
+
+                       dprintf("process_update(): modifying "
+                               "subdev: %i\n", id->index);
+                       devices_to_reshape--;
+                       newdev->vol.migr_state = 1;
+                       newdev->vol.curr_migr_unit = 0;
+                       newdev->vol.migr_type = MIGR_GEN_MIGR;
+                       newmap->num_members = u->new_raid_disks;
+                       for (i = 0; i < delta_disks; i++) {
+                               set_imsm_ord_tbl_ent(newmap,
+                                                    u->old_raid_disks + i,
+                                                    u->old_raid_disks + i);
+                       }
+                       /* New map is correct, now need to save old map
+                        */
+                       newmap = get_imsm_map(newdev, 1);
+                       memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
+
+                       /* calculate new size
+                        */
+                       used_disks = imsm_num_data_members(newdev, 0);
+                       if (used_disks) {
+                               unsigned long long array_blocks;
+
+                               array_blocks =
+                                       newmap->blocks_per_member * used_disks;
+                               /* round array size down to closest MB
+                                */
+                               array_blocks = (array_blocks
+                                               >> SECT_PER_MB_SHIFT)
+                                       << SECT_PER_MB_SHIFT;
+                               newdev->size_low =
+                                       __cpu_to_le32((__u32)array_blocks);
+                               newdev->size_high =
+                                       __cpu_to_le32((__u32)(array_blocks >> 32));
+                       }
+               }
+
+               sp = (void **)id->dev;
+               id->dev = newdev;
+               *sp = tofree;
+               tofree = sp;
+       }
+       if (tofree)
+               *space_list = tofree;
+       ret_val = 1;
+
+update_reshape_exit:
+
+       return ret_val;
+}
+
 static void imsm_process_update(struct supertype *st,
                                struct metadata_update *update)
 {
@@ -5607,86 +5802,9 @@ static void imsm_process_update(struct supertype *st,
        switch (type) {
        case update_reshape_container_disks: {
                struct imsm_update_reshape *u = (void *)update->buf;
-               struct dl *new_disk;
-               struct intel_dev *id;
-               int i;
-               int delta_disks = u->new_raid_disks - u->old_raid_disks;
-               void **tofree = NULL;
-               int devices_to_reshape = 1;
-
-               dprintf("imsm: imsm_process_update() for update_reshape\n");
-
-               /* enable spares to use in array */
-               for (i = 0; i < delta_disks; i++) {
-                       new_disk = get_disk_super(super,
-                                                 major(u->new_disks[i]),
-                                                 minor(u->new_disks[i]));
-                       dprintf("imsm: imsm_process_update(): new disk "\
-                               "for reshape is: %i:%i (%p, index = %i)\n",
-                               major(u->new_disks[i]), minor(u->new_disks[i]),
-                               new_disk, new_disk->index);
-                       if ((new_disk == NULL) ||
-                           ((new_disk->index >= 0) &&
-                            (new_disk->index < u->old_raid_disks)))
-                               goto update_reshape_exit;
-
-                       new_disk->index = mpb->num_disks++;
-                       /* slot to fill in autolayout */
-                       new_disk->raiddisk = new_disk->index;
-                       new_disk->disk.status |=
-                               CONFIGURED_DISK;
-                       new_disk->disk.status &= ~SPARE_DISK;
-               }
-
-               dprintf("imsm: process_update(): update_reshape: volume set"
-                       " mpb->num_raid_devs = %i\n", mpb->num_raid_devs);
-               /* manage changes in volume
-                */
-               for (id = super->devlist ; id; id = id->next) {
-                       void **sp = update->space_list;
-                       struct imsm_dev *newdev;
-                       struct imsm_map *newmap, *oldmap;
-
-                       if (!sp)
-                               continue;
-                       update->space_list = *sp;
-                       newdev = (void*)sp;
-                       /* Copy the dev, but not (all of) the map */
-                       memcpy(newdev, id->dev, sizeof(*newdev));
-                       oldmap = get_imsm_map(id->dev, 0);
-                       newmap = get_imsm_map(newdev, 0);
-                       /* Copy the current map */
-                       memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
-                       /* update one device only
-                        */
-                       if (devices_to_reshape) {
-                               dprintf("process_update(): modifying "\
-                                       "subdev: %i\n", id->index);
-                               devices_to_reshape--;
-                               newdev->vol.migr_state = 1;
-                               newdev->vol.curr_migr_unit = 0;
-                               newdev->vol.migr_type = MIGR_GEN_MIGR;
-                               newmap->num_members = u->new_raid_disks;
-                               for (i = 0; i < delta_disks; i++) {
-                                       set_imsm_ord_tbl_ent(newmap,
-                                                       u->old_raid_disks + i,
-                                                       u->old_raid_disks + i);
-                               }
-                               /* New map is correct, now need to save old map
-                                */
-                               newmap = get_imsm_map(newdev, 1);
-                               memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
-                       }
-
-                       sp = (void **)id->dev;
-                       id->dev = newdev;
-                       *sp = tofree;
-                       tofree = sp;
-               }
-
-               update->space_list = tofree;
-               super->updates_pending++;
-update_reshape_exit:
+               if (apply_reshape_container_disks_update(
+                           u, super, &update->space_list))
+                       super->updates_pending++;
                break;
        }
        case update_activate_spare: {
@@ -6207,6 +6325,10 @@ static int imsm_reshape_is_allowed_on_container(struct supertype *st,
                                                struct geo_params *geo,
                                                int *old_raid_disks)
 {
+       /* currently we only support increasing the number of devices
+        * for a container.  This increases the number of device for each
+        * member array.  They must all be RAID0 or RAID5.
+        */
        int ret_val = 0;
        struct mdinfo *info, *member;
        int devices_that_can_grow = 0;
@@ -6329,6 +6451,7 @@ static int imsm_create_metadata_update_for_reshape(
        struct mdinfo *spares = NULL;
        int i;
        int delta_disks = 0;
+       struct mdinfo *dev;
 
        dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
                geo->raid_disks);
@@ -6367,27 +6490,18 @@ static int imsm_create_metadata_update_for_reshape(
        dprintf("imsm: %i spares are available.\n\n",
                spares->array.spare_disks);
 
+       dev = spares->devs;
        for (i = 0; i < delta_disks; i++) {
-               struct mdinfo *dev = spares->devs;
                struct dl *dl;
 
+               if (dev == NULL)
+                       break;
                u->new_disks[i] = makedev(dev->disk.major,
                                          dev->disk.minor);
                dl = get_disk_super(super, dev->disk.major, dev->disk.minor);
-               dl->index = mpb->num_disks++;
-       }
-       /* Now update the metadata so that container_content will find
-        * the new devices
-        */
-       for (i = 0; i < mpb->num_raid_devs; i++) {
-               int d;
-               struct imsm_dev *dev = get_imsm_dev(super, i);
-               struct imsm_map *map = get_imsm_map(dev, 0);
-               map->num_members = geo->raid_disks;
-               for (d = 0; d < delta_disks; d++) {
-                       set_imsm_ord_tbl_ent(map, old_raid_disks + d,
-                                            mpb->num_disks - delta_disks + d);
-               }
+               dl->index = mpb->num_disks;
+               mpb->num_disks++;
+               dev = dev->next;
        }
 
 abort:
@@ -6407,16 +6521,139 @@ abort:
        return 0;
 }
 
+static void imsm_update_metadata_locally(struct supertype *st,
+                                        void *buf, int len)
+{
+       struct metadata_update mu;
+
+       mu.buf = buf;
+       mu.len = len;
+       mu.space = NULL;
+       mu.space_list = NULL;
+       mu.next = NULL;
+       imsm_prepare_update(st, &mu);
+       imsm_process_update(st, &mu);
+
+       while (mu.space_list) {
+               void **space = mu.space_list;
+               mu.space_list = *space;
+               free(space);
+       }
+}
+
+/***************************************************************************
+* Function:    imsm_analyze_change
+* Description: Function analyze change for single volume
+*              and validate if transition is supported
+* Parameters:  Geometry parameters, supertype structure
+* Returns:     Operation type code on success, -1 if fail
+****************************************************************************/
+enum imsm_reshape_type imsm_analyze_change(struct supertype *st,
+                                          struct geo_params *geo)
+{
+       struct mdinfo info;
+       int change = -1;
+       int check_devs = 0;
+
+       getinfo_super_imsm_volume(st, &info, NULL);
+
+       if ((geo->level != info.array.level) &&
+           (geo->level >= 0) &&
+           (geo->level != UnSet)) {
+               switch (info.array.level) {
+               case 0:
+                       if (geo->level == 5) {
+                               change = CH_LEVEL_MIGRATION;
+                               check_devs = 1;
+                       }
+                       if (geo->level == 10) {
+                               change = CH_TAKEOVER;
+                               check_devs = 1;
+                       }
+                       break;
+               case 5:
+                       if (geo->level != 0)
+                               change = CH_LEVEL_MIGRATION;
+                       break;
+               case 10:
+                       if (geo->level == 0) {
+                               change = CH_TAKEOVER;
+                               check_devs = 1;
+                       }
+                       break;
+               }
+               if (change == -1) {
+                       fprintf(stderr,
+                               Name " Error. Level Migration from %d to %d "
+                               "not supported!\n",
+                               info.array.level, geo->level);
+                       goto analyse_change_exit;
+               }
+       } else
+               geo->level = info.array.level;
+
+       if ((geo->layout != info.array.layout)
+           && ((geo->layout != UnSet) && (geo->layout != -1))) {
+               change = CH_LEVEL_MIGRATION;
+               if ((info.array.layout == 0)
+                   && (info.array.level == 5)
+                   && (geo->layout == 5)) {
+                       /* reshape 5 -> 4 */
+               } else if ((info.array.layout == 5)
+                          && (info.array.level == 5)
+                          && (geo->layout == 0)) {
+                       /* reshape 4 -> 5 */
+                       geo->layout = 0;
+                       geo->level = 5;
+               } else {
+                       fprintf(stderr,
+                               Name " Error. Layout Migration from %d to %d "
+                               "not supported!\n",
+                               info.array.layout, geo->layout);
+                       change = -1;
+                       goto analyse_change_exit;
+               }
+       } else
+               geo->layout = info.array.layout;
+
+       if ((geo->chunksize > 0) && (geo->chunksize != UnSet)
+           && (geo->chunksize != info.array.chunk_size))
+               change = CH_CHUNK_MIGR;
+       else
+               geo->chunksize = info.array.chunk_size;
+
+       if (!validate_geometry_imsm(st,
+                                   geo->level,
+                                   geo->layout,
+                                   geo->raid_disks,
+                                   (geo->chunksize / 1024),
+                                   geo->size,
+                                   0, 0, 1))
+               change = -1;
+
+       if (check_devs) {
+               struct intel_super *super = st->sb;
+               struct imsm_super *mpb = super->anchor;
+
+               if (mpb->num_raid_devs > 1) {
+                       fprintf(stderr,
+                               Name " Error. Cannot perform operation on %s"
+                               "- for this operation it MUST be single "
+                               "array in container\n",
+                               geo->dev_name);
+                       change = -1;
+               }
+       }
+
+analyse_change_exit:
+
+       return change;
+}
 
 static int imsm_reshape_super(struct supertype *st, long long size, int level,
                              int layout, int chunksize, int raid_disks,
                              char *backup, char *dev, int verbose)
 {
-       /* currently we only support increasing the number of devices
-        * for a container.  This increases the number of device for each
-        * member array.  They must all be RAID0 or RAID5.
-        */
-
        int ret_val = 1;
        struct geo_params geo;
 
@@ -6425,6 +6662,7 @@ static int imsm_reshape_super(struct supertype *st, long long size, int level,
        memset(&geo, sizeof(struct geo_params), 0);
 
        geo.dev_name = dev;
+       geo.dev_id = st->devnum;
        geo.size = size;
        geo.level = level;
        geo.layout = layout;
@@ -6437,11 +6675,9 @@ static int imsm_reshape_super(struct supertype *st, long long size, int level,
        if (experimental() == 0)
                return ret_val;
 
-       /* verify reshape conditions
-        * on container level we can only increase number of devices. */
        if (st->container_dev == st->devnum) {
-               /* check for delta_disks > 0
-                *and supported raid levels 0 and 5 only in container */
+               /* On container level we can only increase number of devices. */
+               dprintf("imsm: info: Container operation\n");
                int old_raid_disks = 0;
                if (imsm_reshape_is_allowed_on_container(
                            st, &geo, &old_raid_disks)) {
@@ -6451,22 +6687,80 @@ static int imsm_reshape_super(struct supertype *st, long long size, int level,
                        len = imsm_create_metadata_update_for_reshape(
                                st, &geo, old_raid_disks, &u);
 
-                       if (len) {
-                               ret_val = 0;
+                       if (len <= 0) {
+                               dprintf("imsm: Cannot prepare update\n");
+                               goto exit_imsm_reshape_super;
+                       }
+
+                       ret_val = 0;
+                       /* update metadata locally */
+                       imsm_update_metadata_locally(st, u, len);
+                       /* and possibly remotely */
+                       if (st->update_tail)
                                append_metadata_update(st, u, len);
-                       else
-                               fprintf(stderr, Name "imsm: Cannot prepare "
-                                       "update\n");
-               } else
+                       else
+                               free(u);
+
+               } else {
                        fprintf(stderr, Name "imsm: Operation is not allowed "
                                "on this container\n");
-       } else
-               fprintf(stderr, Name "imsm: not a container operation\n");
+               }
+       } else {
+               /* On volume level we support following operations
+                * - takeover: raid10 -> raid0; raid0 -> raid10
+                * - chunk size migration
+                * - migration: raid5 -> raid0; raid0 -> raid5
+                */
+               struct intel_super *super = st->sb;
+               struct intel_dev *dev = super->devlist;
+               int change, devnum;
+               dprintf("imsm: info: Volume operation\n");
+               /* find requested device */
+               while (dev) {
+                       imsm_find_array_minor_by_subdev(dev->index, st->container_dev, &devnum);
+                       if (devnum == geo.dev_id)
+                               break;
+                       dev = dev->next;
+               }
+               if (dev == NULL) {
+                       fprintf(stderr, Name " Cannot find %s (%i) subarray\n",
+                               geo.dev_name, geo.dev_id);
+                       goto exit_imsm_reshape_super;
+               }
+               super->current_vol = dev->index;
+               change = imsm_analyze_change(st, &geo);
+               switch (change) {
+               case CH_TAKEOVER:
+                       ret_val = 0;
+                       break;
+               case CH_CHUNK_MIGR:
+                       ret_val = 0;
+                       break;
+               case CH_LEVEL_MIGRATION:
+                       ret_val = 0;
+                       break;
+               default:
+                       ret_val = 1;
+               }
+       }
 
+exit_imsm_reshape_super:
        dprintf("imsm: reshape_super Exit code = %i\n", ret_val);
        return ret_val;
 }
 
+static int imsm_manage_reshape(
+       int afd, struct mdinfo *sra, struct reshape *reshape,
+       struct supertype *st, unsigned long stripes,
+       int *fds, unsigned long long *offsets,
+       int dests, int *destfd, unsigned long long *destoffsets)
+{
+       /* Just use child_monitor for now */
+       return child_monitor(
+               afd, sra, reshape, st, stripes,
+               fds, offsets, dests, destfd, destoffsets);
+}
+
 struct superswitch super_imsm = {
 #ifndef        MDASSEMBLE
        .examine_super  = examine_super_imsm,
@@ -6504,6 +6798,7 @@ struct superswitch super_imsm = {
        .default_geometry = default_geometry_imsm,
        .get_disk_controller_domain = imsm_get_disk_controller_domain,
        .reshape_super  = imsm_reshape_super,
+       .manage_reshape = imsm_manage_reshape,
 
        .external       = 1,
        .name = "imsm",