]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - super-intel.c
imsm: fix activate_spare off-by-one
[thirdparty/mdadm.git] / super-intel.c
index 541365965448819c8db4dd696212f64a858e7fab..07e3d46c13b0236397d1d0d0349e517e362edeef 100644 (file)
@@ -53,6 +53,7 @@
 
 #define MPB_SECTOR_CNT 418
 #define IMSM_RESERVED_SECTORS 4096
+#define SECT_PER_MB_SHIFT 11
 
 /* Disk configuration info. */
 #define IMSM_MAX_DEVICES 255
@@ -88,7 +89,7 @@ struct imsm_map {
        __u8  num_members;      /* number of member disks */
        __u8  num_domains;      /* number of parity domains */
        __u8  failed_disk_num;  /* valid only when state is degraded */
-       __u8  reserved[1];
+       __u8  ddf;
        __u32 filler[7];        /* expansion area */
 #define IMSM_ORD_REBUILD (1 << 24)
        __u32 disk_ord_tbl[1];  /* disk_ord_tbl[num_members],
@@ -105,6 +106,7 @@ struct imsm_vol {
 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
 #define MIGR_GEN_MIGR 3
 #define MIGR_STATE_CHANGE 4
+#define MIGR_REPAIR 5
        __u8  migr_type;        /* Initializing, Rebuilding, ... */
        __u8  dirty;
        __u8  fs_state;         /* fast-sync state for CnG (0xff == disabled) */
@@ -193,6 +195,29 @@ struct bbm_log {
 static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" };
 #endif
 
+static __u8 migr_type(struct imsm_dev *dev)
+{
+       if (dev->vol.migr_type == MIGR_VERIFY &&
+           dev->status & DEV_VERIFY_AND_FIX)
+               return MIGR_REPAIR;
+       else
+               return dev->vol.migr_type;
+}
+
+static void set_migr_type(struct imsm_dev *dev, __u8 migr_type)
+{
+       /* for compatibility with older oroms convert MIGR_REPAIR, into
+        * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
+        */
+       if (migr_type == MIGR_REPAIR) {
+               dev->vol.migr_type = MIGR_VERIFY;
+               dev->status |= DEV_VERIFY_AND_FIX;
+       } else {
+               dev->vol.migr_type = migr_type;
+               dev->status &= ~DEV_VERIFY_AND_FIX;
+       }
+}
+
 static unsigned int sector_count(__u32 bytes)
 {
        return ((bytes + (512-1)) & (~(512-1))) / 512;
@@ -621,10 +646,23 @@ static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
        printf("     Chunk Size : %u KiB\n",
                __le16_to_cpu(map->blocks_per_strip) / 2);
        printf("       Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
-       printf("  Migrate State : %s", dev->vol.migr_state ? "migrating" : "idle");
-       if (dev->vol.migr_state)
-               printf(": %s", dev->vol.migr_type ? "rebuilding" : "initializing");
-       printf("\n");
+       printf("  Migrate State : %s", dev->vol.migr_state ? "migrating" : "idle\n");
+       if (dev->vol.migr_state) {
+               if (migr_type(dev) == MIGR_INIT)
+                       printf(": initializing\n");
+               else if (migr_type(dev) == MIGR_REBUILD)
+                       printf(": rebuilding\n");
+               else if (migr_type(dev) == MIGR_VERIFY)
+                       printf(": check\n");
+               else if (migr_type(dev) == MIGR_GEN_MIGR)
+                       printf(": general migration\n");
+               else if (migr_type(dev) == MIGR_STATE_CHANGE)
+                       printf(": state change\n");
+               else if (migr_type(dev) == MIGR_REPAIR)
+                       printf(": repair\n");
+               else
+                       printf(": <unknown:%d>\n", migr_type(dev));
+       }
        printf("      Map State : %s", map_state_str[map->map_state]);
        if (dev->vol.migr_state) {
                struct imsm_map *map = get_imsm_map(dev, 1);
@@ -679,7 +717,7 @@ static void examine_super_imsm(struct supertype *st, char *homehost)
        printf("         Family : %08x\n", __le32_to_cpu(mpb->family_num));
        printf("     Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
        getinfo_super_imsm(st, &info);
-       fname_from_uuid(st, &info, nbuf,'-');
+       fname_from_uuid(st, &info, nbuf, ':');
        printf("           UUID : %s\n", nbuf + 5);
        sum = __le32_to_cpu(mpb->check_sum);
        printf("       Checksum : %08x %s\n", sum,
@@ -697,7 +735,8 @@ static void examine_super_imsm(struct supertype *st, char *homehost)
                printf("      Signature : %x\n", __le32_to_cpu(log->signature));
                printf("    Entry Count : %d\n", __le32_to_cpu(log->entry_count));
                printf("   Spare Blocks : %d\n",  __le32_to_cpu(log->reserved_spare_block_count));
-               printf("    First Spare : %llx\n", __le64_to_cpu(log->first_spare_lba));
+               printf("    First Spare : %llx\n",
+                      (unsigned long long) __le64_to_cpu(log->first_spare_lba));
        }
        for (i = 0; i < mpb->num_raid_devs; i++) {
                struct mdinfo info;
@@ -705,7 +744,7 @@ static void examine_super_imsm(struct supertype *st, char *homehost)
 
                super->current_vol = i;
                getinfo_super_imsm(st, &info);
-               fname_from_uuid(st, &info, nbuf, '-');
+               fname_from_uuid(st, &info, nbuf, ':');
                print_imsm_dev(dev, nbuf + 5, super->disks->index);
        }
        for (i = 0; i < mpb->num_disks; i++) {
@@ -715,7 +754,7 @@ static void examine_super_imsm(struct supertype *st, char *homehost)
        }
 }
 
-static void brief_examine_super_imsm(struct supertype *st)
+static void brief_examine_super_imsm(struct supertype *st, int verbose)
 {
        /* We just write a generic IMSM ARRAY entry */
        struct mdinfo info;
@@ -724,22 +763,38 @@ static void brief_examine_super_imsm(struct supertype *st)
        struct intel_super *super = st->sb;
        int i;
 
-       if (!super->anchor->num_raid_devs)
+       if (!super->anchor->num_raid_devs) {
+               printf("ARRAY metadata=imsm\n");
                return;
+       }
 
        getinfo_super_imsm(st, &info);
-       fname_from_uuid(st, &info, nbuf,'-');
-       printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf + 5);
+       fname_from_uuid(st, &info, nbuf, ':');
        for (i = 0; i < super->anchor->num_raid_devs; i++) {
                struct imsm_dev *dev = get_imsm_dev(super, i);
 
                super->current_vol = i;
                getinfo_super_imsm(st, &info);
-               fname_from_uuid(st, &info, nbuf1,'-');
-               printf("ARRAY /dev/md/%.16s container=%s\n"
-                      "   member=%d auto=mdp UUID=%s\n",
+               fname_from_uuid(st, &info, nbuf1, ':');
+               printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
                       dev->volume, nbuf + 5, i, nbuf1 + 5);
        }
+       printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5);
+}
+
+static void export_examine_super_imsm(struct supertype *st)
+{
+       struct intel_super *super = st->sb;
+       struct imsm_super *mpb = super->anchor;
+       struct mdinfo info;
+       char nbuf[64];
+
+       getinfo_super_imsm(st, &info);
+       fname_from_uuid(st, &info, nbuf, ':');
+       printf("MD_METADATA=imsm\n");
+       printf("MD_LEVEL=container\n");
+       printf("MD_UUID=%s\n", nbuf+5);
+       printf("MD_DEVICES=%u\n", mpb->num_disks);
 }
 
 static void detail_super_imsm(struct supertype *st, char *homehost)
@@ -748,7 +803,7 @@ static void detail_super_imsm(struct supertype *st, char *homehost)
        char nbuf[64];
 
        getinfo_super_imsm(st, &info);
-       fname_from_uuid(st, &info, nbuf,'-');
+       fname_from_uuid(st, &info, nbuf, ':');
        printf("\n           UUID : %s\n", nbuf + 5);
 }
 
@@ -757,7 +812,7 @@ static void brief_detail_super_imsm(struct supertype *st)
        struct mdinfo info;
        char nbuf[64];
        getinfo_super_imsm(st, &info);
-       fname_from_uuid(st, &info, nbuf,'-');
+       fname_from_uuid(st, &info, nbuf, ':');
        printf(" UUID=%s", nbuf + 5);
 }
 
@@ -1151,6 +1206,9 @@ static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
        info->array.utime         = 0;
        info->array.chunk_size    = __le16_to_cpu(map->blocks_per_strip) << 9;
        info->array.state         = !dev->vol.dirty;
+       info->custom_array_size   = __le32_to_cpu(dev->size_high);
+       info->custom_array_size   <<= 32;
+       info->custom_array_size   |= __le32_to_cpu(dev->size_low);
 
        info->disk.major = 0;
        info->disk.minor = 0;
@@ -1166,7 +1224,8 @@ static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
        if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
                info->resync_start = 0;
        else if (dev->vol.migr_state)
-               info->resync_start = __le32_to_cpu(dev->vol.curr_migr_unit);
+               /* FIXME add curr_migr_unit to resync_start conversion */
+               info->resync_start = 0;
        else
                info->resync_start = ~0ULL;
 
@@ -1638,7 +1697,7 @@ load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
  * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
  * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
  *    map1state=unitialized)
- * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
+ * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR  map0state=normal
  *    map1state=normal)
  * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
  *    map1state=degraded)
@@ -1649,7 +1708,7 @@ static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type)
        struct imsm_map *src = get_imsm_map(dev, 0);
 
        dev->vol.migr_state = 1;
-       dev->vol.migr_type = migr_type;
+       set_migr_type(dev, migr_type);
        dev->vol.curr_migr_unit = 0;
        dest = get_imsm_map(dev, 1);
 
@@ -1731,7 +1790,8 @@ static int parse_raid_devices(struct intel_super *super)
                if (posix_memalign(&buf, 512, len) != 0)
                        return 1;
 
-               memcpy(buf, super->buf, len);
+               memcpy(buf, super->buf, super->len);
+               memset(buf + super->len, 0, len - super->len);
                free(super->buf);
                super->buf = buf;
                super->len = len;
@@ -2109,8 +2169,10 @@ static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
        if (st->subarray[0]) {
                if (atoi(st->subarray) <= super->anchor->num_raid_devs)
                        super->current_vol = atoi(st->subarray);
-               else
+               else {
+                       free_imsm(super);
                        return 1;
+               }
        }
 
        *sbp = super;
@@ -2135,8 +2197,8 @@ static int load_super_imsm(struct supertype *st, int fd, char *devname)
        if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
                return 0;
 #endif
-       if (st->subarray[0])
-               return 1; /* FIXME */
+
+       free_super_imsm(st);
 
        super = alloc_super(0);
        if (!super) {
@@ -2157,6 +2219,15 @@ static int load_super_imsm(struct supertype *st, int fd, char *devname)
                return rv;
        }
 
+       if (st->subarray[0]) {
+               if (atoi(st->subarray) <= super->anchor->num_raid_devs)
+                       super->current_vol = atoi(st->subarray);
+               else {
+                       free_imsm(super);
+                       return 1;
+               }
+       }
+
        st->sb = super;
        if (st->ss == NULL) {
                st->ss = &super_imsm;
@@ -2175,13 +2246,12 @@ static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
        return info->chunk_size >> 9;
 }
 
-static __u32 info_to_num_data_stripes(mdu_array_info_t *info)
+static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains)
 {
        __u32 num_stripes;
 
        num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
-       if (info->level == 1)
-               num_stripes /= 2;
+       num_stripes /= num_domains;
 
        return num_stripes;
 }
@@ -2258,6 +2328,7 @@ static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
        int i;
        unsigned long long array_blocks;
        size_t size_old, size_new;
+       __u32 num_data_stripes;
 
        if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
                fprintf(stderr, Name": This imsm-container already has the "
@@ -2320,36 +2391,44 @@ static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
                array_blocks = calc_array_size(info->level, info->raid_disks,
                                               info->layout, info->chunk_size,
                                               info->size*2);
+       /* 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));
        dev->status = __cpu_to_le32(0);
        dev->reserved_blocks = __cpu_to_le32(0);
        vol = &dev->vol;
        vol->migr_state = 0;
-       vol->migr_type = MIGR_INIT;
+       set_migr_type(dev, MIGR_INIT);
        vol->dirty = 0;
        vol->curr_migr_unit = 0;
        map = get_imsm_map(dev, 0);
        map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
        map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
        map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
-       map->num_data_stripes = __cpu_to_le32(info_to_num_data_stripes(info));
        map->failed_disk_num = ~0;
        map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
                                       IMSM_T_STATE_NORMAL;
+       map->ddf = 1;
 
        if (info->level == 1 && info->raid_disks > 2) {
                fprintf(stderr, Name": imsm does not support more than 2 disks"
                                "in a raid1 volume\n");
                return 0;
        }
+
+       map->raid_level = info->level;
        if (info->level == 10) {
                map->raid_level = 1;
                map->num_domains = info->raid_disks / 2;
-       } else {
-               map->raid_level = info->level;
-               map->num_domains = !!map->raid_level;
-       }
+       } else if (info->level == 1)
+               map->num_domains = info->raid_disks;
+       else
+               map->num_domains = 1;
+
+       num_data_stripes = info_to_num_data_stripes(info, map->num_domains);
+       map->num_data_stripes = __cpu_to_le32(num_data_stripes);
 
        map->num_members = info->raid_disks;
        for (i = 0; i < map->num_members; i++) {
@@ -2645,17 +2724,16 @@ static int write_super_imsm(struct intel_super *super, int doclose)
 }
 
 
-static int create_array(struct supertype *st)
+static int create_array(struct supertype *st, int dev_idx)
 {
        size_t len;
        struct imsm_update_create_array *u;
        struct intel_super *super = st->sb;
-       struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
+       struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
        struct imsm_map *map = get_imsm_map(dev, 0);
        struct disk_info *inf;
        struct imsm_disk *disk;
        int i;
-       int idx;
 
        len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
              sizeof(*inf) * map->num_members;
@@ -2667,11 +2745,12 @@ static int create_array(struct supertype *st)
        }
 
        u->type = update_create_array;
-       u->dev_idx = super->current_vol;
+       u->dev_idx = dev_idx;
        imsm_copy_dev(&u->dev, dev);
        inf = get_disk_info(u);
        for (i = 0; i < map->num_members; i++) {
-               idx = get_imsm_disk_idx(dev, i);
+               int idx = get_imsm_disk_idx(dev, i);
+
                disk = get_imsm_disk(super, idx);
                serialcpy(inf[i].serial, disk->serial);
        }
@@ -2705,21 +2784,26 @@ static int _add_disk(struct supertype *st)
 
 static int write_init_super_imsm(struct supertype *st)
 {
+       struct intel_super *super = st->sb;
+       int current_vol = super->current_vol;
+
+       /* we are done with current_vol reset it to point st at the container */
+       super->current_vol = -1;
+
        if (st->update_tail) {
                /* queue the recently created array / added disk
                 * as a metadata update */
-               struct intel_super *super = st->sb;
                struct dl *d;
                int rv;
 
                /* determine if we are creating a volume or adding a disk */
-               if (super->current_vol < 0) {
+               if (current_vol < 0) {
                        /* in the add disk case we are running in mdmon
                         * context, so don't close fd's
                         */
                        return _add_disk(st);
                } else
-                       rv = create_array(st);
+                       rv = create_array(st, current_vol);
 
                for (d = super->disks; d ; d = d->next) {
                        close(d->fd);
@@ -3301,6 +3385,18 @@ static struct mdinfo *container_content_imsm(struct supertype *st)
                struct mdinfo *this;
                int slot;
 
+               /* do not publish arrays that are in the middle of an
+                * unsupported migration
+                */
+               if (dev->vol.migr_state &&
+                   (migr_type(dev) == MIGR_GEN_MIGR ||
+                    migr_type(dev) == MIGR_STATE_CHANGE)) {
+                       fprintf(stderr, Name ": cannot assemble volume '%.16s':"
+                               " unsupported migration in progress\n",
+                               dev->volume);
+                       continue;
+               }
+
                this = malloc(sizeof(*this));
                memset(this, 0, sizeof(*this));
                this->next = rest;
@@ -3347,7 +3443,7 @@ static struct mdinfo *container_content_imsm(struct supertype *st)
                        info_d = malloc(sizeof(*info_d));
                        if (!info_d) {
                                fprintf(stderr, Name ": failed to allocate disk"
-                                       " for volume %s\n", (char *) dev->volume);
+                                       " for volume %.16s\n", dev->volume);
                                free(this);
                                this = rest;
                                break;
@@ -3501,7 +3597,8 @@ static int is_resyncing(struct imsm_dev *dev)
        if (!dev->vol.migr_state)
                return 0;
 
-       if (dev->vol.migr_type == MIGR_INIT)
+       if (migr_type(dev) == MIGR_INIT ||
+           migr_type(dev) == MIGR_REPAIR)
                return 1;
 
        migr_map = get_imsm_map(dev, 1);
@@ -3519,7 +3616,7 @@ static int is_rebuilding(struct imsm_dev *dev)
        if (!dev->vol.migr_state)
                return 0;
 
-       if (dev->vol.migr_type != MIGR_REBUILD)
+       if (migr_type(dev) != MIGR_REBUILD)
                return 0;
 
        migr_map = get_imsm_map(dev, 1);
@@ -3550,7 +3647,7 @@ static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
 
        disk->status |= FAILED_DISK;
        set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
-       if (map->failed_disk_num == ~0)
+       if (~map->failed_disk_num == 0)
                map->failed_disk_num = slot;
        return 1;
 }
@@ -3610,20 +3707,14 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
        } else if (!is_resyncing(dev) && !failed) {
                /* mark the start of the init process if nothing is failed */
                dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
-               if (map->map_state == IMSM_T_STATE_NORMAL)
-                       migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REBUILD);
-               else
+               if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
                        migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
+               else
+                       migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
                super->updates_pending++;
        }
 
-       /* check if we can update the migration checkpoint */
-       if (dev->vol.migr_state &&
-           __le32_to_cpu(dev->vol.curr_migr_unit) != a->resync_start) {
-               dprintf("imsm: checkpoint migration (%llu)\n", a->resync_start);
-               dev->vol.curr_migr_unit = __cpu_to_le32(a->resync_start);
-               super->updates_pending++;
-       }
+        /* FIXME check if we can update curr_migr_unit from resync_start */
 
        /* mark dirty / clean */
        if (dev->vol.dirty != !consistent) {
@@ -3767,14 +3858,13 @@ static struct dl *imsm_add_spare(struct intel_super *super, int slot,
        int idx = get_imsm_disk_idx(dev, slot);
        struct imsm_super *mpb = super->anchor;
        struct imsm_map *map;
-       unsigned long long esize;
        unsigned long long pos;
        struct mdinfo *d;
        struct extent *ex;
        int i, j;
        int found;
        __u32 array_start;
-       __u32 blocks;
+       __u32 array_end;
        struct dl *dl;
 
        for (dl = super->disks; dl; dl = dl->next) {
@@ -3826,15 +3916,14 @@ static struct dl *imsm_add_spare(struct intel_super *super, int slot,
                        j = 0;
                        pos = 0;
                        array_start = __le32_to_cpu(map->pba_of_lba0);
-                       blocks = __le32_to_cpu(map->blocks_per_member);
+                       array_end = array_start +
+                                   __le32_to_cpu(map->blocks_per_member) - 1;
 
                        do {
                                /* check that we can start at pba_of_lba0 with
                                 * blocks_per_member of space
                                 */
-                               esize = ex[j].start - pos;
-                               if (array_start >= pos &&
-                                   array_start + blocks < ex[j].start) {
+                               if (array_start >= pos && array_end < ex[j].start) {
                                        found = 1;
                                        break;
                                }
@@ -3848,9 +3937,8 @@ static struct dl *imsm_add_spare(struct intel_super *super, int slot,
 
                free(ex);
                if (i < mpb->num_raid_devs) {
-                       dprintf("%x:%x does not have %u at %u\n",
-                               dl->major, dl->minor,
-                               blocks, array_start);
+                       dprintf("%x:%x does not have %u to %u available\n",
+                               dl->major, dl->minor, array_start, array_end);
                        /* No room */
                        continue;
                }
@@ -4367,7 +4455,9 @@ static void imsm_prepare_update(struct supertype *st,
                        free(super->next_buf);
 
                super->next_len = buf_len;
-               if (posix_memalign(&super->next_buf, 512, buf_len) != 0)
+               if (posix_memalign(&super->next_buf, 512, buf_len) == 0)
+                       memset(super->next_buf, 0, buf_len);
+               else
                        super->next_buf = NULL;
        }
 }
@@ -4429,6 +4519,7 @@ struct superswitch super_imsm = {
 #ifndef        MDASSEMBLE
        .examine_super  = examine_super_imsm,
        .brief_examine_super = brief_examine_super_imsm,
+       .export_examine_super = export_examine_super_imsm,
        .detail_super   = detail_super_imsm,
        .brief_detail_super = brief_detail_super_imsm,
        .write_init_super = write_init_super_imsm,