]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - super-intel.c
Typo in dprintf
[thirdparty/mdadm.git] / super-intel.c
index 594d6691fc001bf24b00690fa15730056e13fe76..cd409e92cadbc19e936ae05c4128f6ff29913d8b 100644 (file)
@@ -233,6 +233,13 @@ struct intel_dev {
        unsigned index;
 };
 
+struct intel_hba {
+       enum sys_dev_type type;
+       char *path;
+       char *pci_id;
+       struct intel_hba *next;
+};
+
 enum action {
        DISK_REMOVE = 1,
        DISK_ADD
@@ -268,7 +275,7 @@ struct intel_super {
                                      active */
        struct dl *missing; /* disks removed while we weren't looking */
        struct bbm_log *bbm_log;
-       const char *hba; /* device path of the raid controller for this metadata */
+       struct intel_hba *hba; /* device path of the raid controller for this metadata */
        const struct imsm_orom *orom; /* platform firmware support */
        struct intel_super *next; /* (temp) list for disambiguating family_num */
 };
@@ -284,6 +291,12 @@ struct extent {
        unsigned long long start, size;
 };
 
+/* definitions of reshape process types */
+enum imsm_reshape_type {
+       CH_TAKEOVER,
+       CH_MIGRATION,
+};
+
 /* definition of messages passed to imsm_process_update */
 enum imsm_update_type {
        update_activate_spare,
@@ -292,6 +305,7 @@ enum imsm_update_type {
        update_rename_array,
        update_add_remove_disk,
        update_reshape_container_disks,
+       update_takeover
 };
 
 struct imsm_update_activate_spare {
@@ -312,6 +326,15 @@ struct geo_params {
        int raid_disks;
 };
 
+enum takeover_direction {
+       R10_TO_R0,
+       R0_TO_R10
+};
+struct imsm_update_takeover {
+       enum imsm_update_type type;
+       int subarray;
+       enum takeover_direction direction;
+};
 
 struct imsm_update_reshape {
        enum imsm_update_type type;
@@ -345,6 +368,115 @@ struct imsm_update_add_remove_disk {
        enum imsm_update_type type;
 };
 
+
+static const char *_sys_dev_type[] = {
+       [SYS_DEV_UNKNOWN] = "Unknown",
+       [SYS_DEV_SAS] = "SAS",
+       [SYS_DEV_SATA] = "SATA"
+};
+
+const char *get_sys_dev_type(enum sys_dev_type type)
+{
+       if (type >= SYS_DEV_MAX)
+               type = SYS_DEV_UNKNOWN;
+
+       return _sys_dev_type[type];
+}
+
+#ifndef MDASSEMBLE
+static struct intel_hba * alloc_intel_hba(struct sys_dev *device)
+{
+       struct intel_hba *result = malloc(sizeof(*result));
+       if (result) {
+               result->type = device->type;
+               result->path = strdup(device->path);
+               result->next = NULL;
+               if (result->path && (result->pci_id = strrchr(result->path, '/')) != NULL)
+                       result->pci_id++;
+       }
+       return result;
+}
+
+static struct intel_hba * find_intel_hba(struct intel_hba *hba, struct sys_dev *device)
+{
+       struct intel_hba *result=NULL;
+       for (result = hba; result; result = result->next) {
+               if (result->type == device->type && strcmp(result->path, device->path) == 0)
+                       break;
+       }
+       return result;
+}
+
+
+static int attach_hba_to_super(struct intel_super *super, struct sys_dev *device,
+                              const char *devname)
+{
+       struct intel_hba *hba;
+
+       /* check if disk attached to Intel HBA */
+       hba = find_intel_hba(super->hba, device);
+       if (hba != NULL)
+               return 1;
+       /* Check if HBA is already attached to super */
+       if (super->hba == NULL) {
+               super->hba = alloc_intel_hba(device);
+               return 1;
+       }
+
+       hba = super->hba;
+       /* Intel metadata allows for all disks attached to the same type HBA.
+        * Do not sypport odf HBA types mixing
+        */
+       if (device->type != hba->type)
+               return 2;
+
+       while (hba->next)
+               hba = hba->next;
+
+       hba->next = alloc_intel_hba(device);
+       return 1;
+}
+
+static struct sys_dev* find_disk_attached_hba(int fd, const char *devname)
+{
+       struct sys_dev *list, *elem, *prev;
+       char *disk_path;
+
+       if ((list = find_intel_devices()) == NULL)
+               return 0;
+
+       if (fd < 0)
+               disk_path  = (char *) devname;
+       else
+               disk_path = diskfd_to_devpath(fd);
+
+       if (!disk_path) {
+               free_sys_dev(&list);
+               return 0;
+       }
+
+       for (prev = NULL, elem = list; elem; prev = elem, elem = elem->next) {
+               if (path_attached_to_hba(disk_path, elem->path)) {
+                       if (prev == NULL)
+                               list = list->next;
+                       else
+                               prev->next = elem->next;
+                       elem->next = NULL;
+                       if (disk_path != devname)
+                               free(disk_path);
+                       free_sys_dev(&list);
+                       return elem;
+               }
+       }
+       if (disk_path != devname)
+               free(disk_path);
+       free_sys_dev(&list);
+
+       return NULL;
+}
+#endif /* MDASSEMBLE */
+
+
 static struct supertype *match_metadata_desc_imsm(char *arg)
 {
        struct supertype *st;
@@ -432,17 +564,24 @@ static size_t sizeof_imsm_map(struct imsm_map *map)
 
 struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
 {
+       /* A device can have 2 maps if it is in the middle of a migration.
+        * If second_map is:
+        *    0   - we return the first map
+        *    1   - we return the second map if it exists, else NULL
+        *   -1   - we return the second map if it exists, else the first
+        */
        struct imsm_map *map = &dev->vol.map[0];
 
-       if (second_map && !dev->vol.migr_state)
+       if (second_map == 1 && !dev->vol.migr_state)
                return NULL;
-       else if (second_map) {
+       else if (second_map == 1 ||
+                (second_map < 0 && dev->vol.migr_state)) {
                void *ptr = map;
 
                return ptr + sizeof_imsm_map(map);
        } else
                return map;
-               
+
 }
 
 /* return the size of the device.
@@ -521,14 +660,7 @@ static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev,
 {
        struct imsm_map *map;
 
-       if (second_map == -1) {
-               if (dev->vol.migr_state)
-                       map = get_imsm_map(dev, 1);
-               else
-                       map = get_imsm_map(dev, 0);
-       } else {
-               map = get_imsm_map(dev, second_map);
-       }
+       map = get_imsm_map(dev, second_map);
 
        /* top byte identifies disk under rebuild */
        return __le32_to_cpu(map->disk_ord_tbl[slot]);
@@ -765,6 +897,12 @@ static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
                printf("]");
        }
        printf("\n");
+       printf("    Failed disk : ");
+       if (map->failed_disk_num == 0xff)
+               printf("none");
+       else
+               printf("%i", map->failed_disk_num);
+       printf("\n");
        slot = get_imsm_disk_slot(map, disk_idx);
        if (slot >= 0) {
                ord = get_imsm_ord_tbl_ent(dev, slot, -1);
@@ -1001,10 +1139,10 @@ static void brief_detail_super_imsm(struct supertype *st)
 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
 static void fd2devname(int fd, char *name);
 
-static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
+static int ahci_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
 {
-       /* dump an unsorted list of devices attached to ahci, as well as
-        * non-connected ports
+       /* dump an unsorted list of devices attached to AHCI Intel storage
+        * controller, as well as non-connected ports
         */
        int hba_len = strlen(hba_path) + 1;
        struct dirent *ent;
@@ -1164,6 +1302,51 @@ static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_b
        return err;
 }
 
+
+
+static void print_found_intel_controllers(struct sys_dev *elem)
+{
+       for (; elem; elem = elem->next) {
+               fprintf(stderr, Name ": found Intel(R) ");
+               if (elem->type == SYS_DEV_SATA)
+                       fprintf(stderr, "SATA ");
+               else if (elem->type == SYS_DEV_SAS)
+                       fprintf(stderr, "SAS ");
+               fprintf(stderr, "RAID controller");
+               if (elem->pci_id)
+                       fprintf(stderr, " at %s", elem->pci_id);
+               fprintf(stderr, ".\n");
+       }
+       fflush(stderr);
+}
+
+static int ahci_get_port_count(const char *hba_path, int *port_count)
+{
+       struct dirent *ent;
+       DIR *dir;
+       int host_base = -1;
+
+       *port_count = 0;
+       if ((dir = opendir(hba_path)) == NULL)
+               return -1;
+
+       for (ent = readdir(dir); ent; ent = readdir(dir)) {
+               int host;
+
+               if (sscanf(ent->d_name, "host%d", &host) != 1)
+                       continue;
+               if (*port_count == 0)
+                       host_base = host;
+               else if (host < host_base)
+                       host_base = host;
+
+               if (host + 1 > *port_count + host_base)
+                       *port_count = host + 1 - host_base;
+       }
+       closedir(dir);
+       return host_base;
+}
+
 static int detail_platform_imsm(int verbose, int enumerate_only)
 {
        /* There are two components to imsm platform support, the ahci SATA
@@ -1179,11 +1362,9 @@ static int detail_platform_imsm(int verbose, int enumerate_only)
         */
        const struct imsm_orom *orom;
        struct sys_dev *list, *hba;
-       DIR *dir;
-       struct dirent *ent;
-       const char *hba_path;
        int host_base = 0;
        int port_count = 0;
+       int result=0;
 
        if (enumerate_only) {
                if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
@@ -1191,24 +1372,19 @@ static int detail_platform_imsm(int verbose, int enumerate_only)
                return 2;
        }
 
-       list = find_driver_devices("pci", "ahci");
-       for (hba = list; hba; hba = hba->next)
-               if (devpath_to_vendor(hba->path) == 0x8086)
-                       break;
-
-       if (!hba) {
+       list = find_intel_devices();
+       if (!list) {
                if (verbose)
-                       fprintf(stderr, Name ": unable to find active ahci controller\n");
+                       fprintf(stderr, Name ": no active Intel(R) RAID "
+                               "controller found.\n");
                free_sys_dev(&list);
                return 2;
        } else if (verbose)
-               fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
-       hba_path = hba->path;
-       hba->path = NULL;
-       free_sys_dev(&list);
+               print_found_intel_controllers(list);
 
        orom = find_imsm_orom();
        if (!orom) {
+               free_sys_dev(&list);
                if (verbose)
                        fprintf(stderr, Name ": imsm option-rom not found\n");
                return 2;
@@ -1242,35 +1418,29 @@ static int detail_platform_imsm(int verbose, int enumerate_only)
               imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
        printf("      Max Disks : %d\n", orom->tds);
        printf("    Max Volumes : %d\n", orom->vpa);
-       printf(" I/O Controller : %s\n", hba_path);
-
-       /* find the smallest scsi host number to determine a port number base */
-       dir = opendir(hba_path);
-       for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
-               int host;
 
-               if (sscanf(ent->d_name, "host%d", &host) != 1)
-                       continue;
-               if (port_count == 0)
-                       host_base = host;
-               else if (host < host_base)
-                       host_base = host;
-
-               if (host + 1 > port_count + host_base)
-                       port_count = host + 1 - host_base;
-
-       }
-       if (dir)
-               closedir(dir);
-
-       if (!port_count || imsm_enumerate_ports(hba_path, port_count,
-                                               host_base, verbose) != 0) {
-               if (verbose)
-                       fprintf(stderr, Name ": failed to enumerate ports\n");
-               return 2;
+       for (hba = list; hba; hba = hba->next) {
+               printf(" I/O Controller : %s (%s)\n",
+                       hba->path, get_sys_dev_type(hba->type));
+
+               if (hba->type == SYS_DEV_SATA) {
+                       host_base = ahci_get_port_count(hba->path, &port_count);
+                       if (ahci_enumerate_ports(hba->path, port_count, host_base, verbose)) {
+                               if (verbose)
+                                       fprintf(stderr, Name ": failed to enumerate "
+                                               "ports on SATA controller at %s.", hba->pci_id);
+                               result |= 2;
+                       }
+               } else if (hba->type == SYS_DEV_SAS) {
+                               if (verbose)
+                                       fprintf(stderr, Name ": failed to enumerate "
+                                               "devices on SAS controller at %s.", hba->pci_id);
+                               result |= 2;
+               }
        }
 
-       return 0;
+       free_sys_dev(&list);
+       return result;
 }
 #endif
 
@@ -1520,6 +1690,8 @@ static __u64 blocks_per_migr_unit(struct imsm_dev *dev)
                migr_chunk = migr_strip_blocks_resync(dev);
                disks = imsm_num_data_members(dev, 0);
                blocks_per_unit = stripes_per_unit * migr_chunk * disks;
+               if (migr_type(dev) == MIGR_GEN_MIGR)
+                       return blocks_per_unit;
                stripe = __le32_to_cpu(map->blocks_per_strip) * disks;
                segment = blocks_per_unit / stripe;
                block_rel = blocks_per_unit - segment * stripe;
@@ -1586,14 +1758,53 @@ static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info,
        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);
-       if (prev_map) {
+       if (prev_map && map->map_state == prev_map->map_state) {
+               info->reshape_active = 1;
                info->new_level = get_imsm_raid_level(map);
                info->new_layout = imsm_level_to_layout(info->new_level);
                info->new_chunk = __le16_to_cpu(map->blocks_per_strip) << 9;
+               info->delta_disks = map->num_members - prev_map->num_members;
+               if (info->delta_disks) {
+                       /* this needs to be applied to every array
+                        * in the container.
+                        */
+                       info->reshape_active = 2;
+               }
+               /* We shape information that we give to md might have to be
+                * modify to cope with md's requirement for reshaping arrays.
+                * For example, when reshaping a RAID0, md requires it to be
+                * presented as a degraded RAID4.
+                * Also if a RAID0 is migrating to a RAID5 we need to specify
+                * the array as already being RAID5, but the 'before' layout
+                * is a RAID4-like layout.
+                */
+               switch (info->array.level) {
+               case 0:
+                       switch(info->new_level) {
+                       case 0:
+                               /* conversion is happening as RAID4 */
+                               info->array.level = 4;
+                               info->array.raid_disks += 1;
+                               break;
+                       case 5:
+                               /* conversion is happening as RAID5 */
+                               info->array.level = 5;
+                               info->array.layout = ALGORITHM_PARITY_N;
+                               info->array.raid_disks += 1;
+                               info->delta_disks -= 1;
+                               break;
+                       default:
+                               /* FIXME error message */
+                               info->array.level = UnSet;
+                               break;
+                       }
+                       break;
+               }
        } else {
                info->new_level = UnSet;
                info->new_layout = UnSet;
                info->new_chunk = info->array.chunk_size;
+               info->delta_disks = 0;
        }
        info->disk.major = 0;
        info->disk.minor = 0;
@@ -1607,12 +1818,8 @@ static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info,
                __le32_to_cpu(map_to_analyse->blocks_per_member);
        memset(info->uuid, 0, sizeof(info->uuid));
        info->recovery_start = MaxSector;
-       info->reshape_active = (prev_map != NULL);
-       if (info->reshape_active)
-               info->delta_disks = map->num_members - prev_map->num_members;
-       else
-               info->delta_disks = 0;
 
+       info->reshape_progress = 0;
        if (map_to_analyse->map_state == IMSM_T_STATE_UNINITIALIZED ||
            dev->vol.dirty) {
                info->resync_start = 0;
@@ -1626,6 +1833,28 @@ static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info,
                        info->resync_start = blocks_per_unit * units;
                        break;
                }
+               case MIGR_GEN_MIGR: {
+                       __u64 blocks_per_unit = blocks_per_migr_unit(dev);
+                       __u64 units = __le32_to_cpu(dev->vol.curr_migr_unit);
+                       unsigned long long array_blocks;
+                       int used_disks;
+
+                       info->reshape_progress = blocks_per_unit * units;
+                       dprintf("IMSM: General Migration checkpoint : %llu "
+                              "(%llu) -> read reshape progress : %llu\n",
+                               units, blocks_per_unit, info->reshape_progress);
+
+                       used_disks = imsm_num_data_members(dev, 1);
+                       if (used_disks > 0) {
+                               array_blocks = map->blocks_per_member *
+                                       used_disks;
+                               /* round array size down to closest MB
+                                */
+                               info->custom_array_size = (array_blocks
+                                               >> SECT_PER_MB_SHIFT)
+                                               << SECT_PER_MB_SHIFT;
+                       }
+               }
                case MIGR_VERIFY:
                        /* we could emulate the checkpointing of
                         * 'sync_action=check' migrations, but for now
@@ -1633,7 +1862,6 @@ static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info,
                         */
                case MIGR_REBUILD:
                        /* this is handled by container_content_imsm() */
-               case MIGR_GEN_MIGR:
                case MIGR_STATE_CHANGE:
                        /* FIXME handle other migrations */
                default:
@@ -2307,6 +2535,7 @@ static int parse_raid_devices(struct intel_super *super)
        int i;
        struct imsm_dev *dev_new;
        size_t len, len_migr;
+       size_t max_len = 0;
        size_t space_needed = 0;
        struct imsm_super *mpb = super->anchor;
 
@@ -2322,7 +2551,11 @@ static int parse_raid_devices(struct intel_super *super)
                dv = malloc(sizeof(*dv));
                if (!dv)
                        return 1;
-               dev_new = malloc(len_migr);
+               if (max_len < len_migr)
+                       max_len = len_migr;
+               if (max_len > len_migr)
+                       space_needed += max_len - len_migr;
+               dev_new = malloc(max_len);
                if (!dev_new) {
                        free(dv);
                        return 1;
@@ -2523,6 +2756,11 @@ static void free_imsm_disks(struct intel_super *super)
                super->disks = d->next;
                __free_imsm_disk(d);
        }
+       while (super->disk_mgmt_list) {
+               d = super->disk_mgmt_list;
+               super->disk_mgmt_list = d->next;
+               __free_imsm_disk(d);
+       }
        while (super->missing) {
                d = super->missing;
                super->missing = d->next;
@@ -2534,6 +2772,8 @@ static void free_imsm_disks(struct intel_super *super)
 /* free all the pieces hanging off of a super pointer */
 static void __free_imsm(struct intel_super *super, int free_disks)
 {
+       struct intel_hba *elem, *next;
+
        if (super->buf) {
                free(super->buf);
                super->buf = NULL;
@@ -2541,10 +2781,15 @@ static void __free_imsm(struct intel_super *super, int free_disks)
        if (free_disks)
                free_imsm_disks(super);
        free_devlist(super);
-       if (super->hba) {
-               free((void *) super->hba);
-               super->hba = NULL;
+       elem = super->hba;
+       while (elem) {
+               if (elem->path)
+                       free((void *)elem->path);
+               next = elem->next;
+               free(elem);
+               elem = next;
        }
+       super->hba = NULL;
 }
 
 static void free_imsm(struct intel_super *super)
@@ -2574,20 +2819,6 @@ static struct intel_super *alloc_super(void)
                super->create_offset = ~((__u32 ) 0);
                if (!check_env("IMSM_NO_PLATFORM"))
                        super->orom = find_imsm_orom();
-               if (super->orom && !check_env("IMSM_TEST_OROM")) {
-                       struct sys_dev *list, *ent;
-
-                       /* find the first intel ahci controller */
-                       list = find_driver_devices("pci", "ahci");
-                       for (ent = list; ent; ent = ent->next)
-                               if (devpath_to_vendor(ent->path) == 0x8086)
-                                       break;
-                       if (ent) {
-                               super->hba = ent->path;
-                               ent->path = NULL;
-                       }
-                       free_sys_dev(&list);
-               }
        }
 
        return super;
@@ -3235,12 +3466,13 @@ static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
                fprintf(stderr, Name ": failed to allocate device list entry\n");
                return 0;
        }
-       dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
+       dev = calloc(1, sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
        if (!dev) {
                free(dv);
                fprintf(stderr, Name": could not allocate raid device\n");
                return 0;
        }
+
        strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
        if (info->level == 1)
                array_blocks = info_to_blocks_per_member(info);
@@ -3253,8 +3485,7 @@ static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
 
        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);
+       dev->status = (DEV_READ_COALESCING | DEV_WRITE_COALESCING);
        vol = &dev->vol;
        vol->migr_state = 0;
        set_migr_type(dev, MIGR_INIT);
@@ -3437,8 +3668,9 @@ static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
        return 0;
 }
 
+
 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
-                             int fd, char *devname)
+                            int fd, char *devname)
 {
        struct intel_super *super = st->sb;
        struct dl *dd;
@@ -3447,14 +3679,48 @@ static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
        int rv;
        struct stat stb;
 
-       /* if we are on an RAID enabled platform check that the disk is
-        * attached to the raid controller
+       /* If we are on an RAID enabled platform check that the disk is
+        * attached to the raid controller.
+        * We do not need to test disks attachment for container based additions,
+        * they shall be already tested when container was created/assembled.
         */
-       if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
-               fprintf(stderr,
-                       Name ": %s is not attached to the raid controller: %s\n",
-                       devname ? : "disk", super->hba);
-               return 1;
+       if ((fd != -1) && !check_env("IMSM_NO_PLATFORM")) {
+               struct sys_dev *hba_name;
+               struct intel_hba *hba;
+
+               hba_name = find_disk_attached_hba(fd, NULL);
+               if (!hba_name) {
+                       fprintf(stderr,
+                               Name ": %s is not attached to Intel(R) RAID controller.\n",
+                               devname ? : "disk");
+                       return 1;
+               }
+               rv = attach_hba_to_super(super, hba_name, devname);
+               switch (rv) {
+               case 2:
+                       fprintf(stderr, Name ": %s is attached to Intel(R) %s RAID "
+                               "controller (%s),\n    but the container is assigned to Intel(R) "
+                               "%s RAID controller (",
+                               devname,
+                               get_sys_dev_type(hba_name->type),
+                               hba_name->pci_id ? : "Err!",
+                               get_sys_dev_type(hba_name->type));
+
+                       hba = super->hba;
+                       while (hba) {
+                               fprintf(stderr, "%s", hba->pci_id ? : "Err!");
+                               if (hba->next)
+                                       fprintf(stderr, ", ");
+                               hba = hba->next;
+                       }
+
+                       fprintf(stderr, ").\n"
+                               "    Mixing devices attached to different controllers "
+                               "is not allowed.\n");
+                       free_sys_dev(&hba_name);
+                       return 1;
+               }
+               free_sys_dev(&hba_name);
        }
 
        if (super->current_vol >= 0)
@@ -3736,7 +4002,6 @@ 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 */
@@ -3748,11 +4013,6 @@ 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;
@@ -3962,17 +4222,21 @@ static int is_raid_level_supported(const struct imsm_orom *orom, int level, int
 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
 static int
 validate_geometry_imsm_orom(struct intel_super *super, int level, int layout,
-                           int raiddisks, int chunk, int verbose)
+                           int raiddisks, int *chunk, int verbose)
 {
        if (!is_raid_level_supported(super->orom, level, raiddisks)) {
                pr_vrb(": platform does not support raid%d with %d disk%s\n",
                        level, raiddisks, raiddisks > 1 ? "s" : "");
                return 0;
        }
-       if (super->orom && level != 1 &&
-           !imsm_orom_has_chunk(super->orom, chunk)) {
-               pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
-               return 0;
+       if (super->orom && level != 1) {
+               if (chunk && (*chunk == 0 || *chunk == UnSet))
+                       *chunk = imsm_orom_default_chunk(super->orom);
+               else if (chunk && !imsm_orom_has_chunk(super->orom, *chunk)) {
+                       pr_vrb(": platform does not support a chunk size of: "
+                              "%d\n", *chunk);
+                       return 0;
+               }
        }
        if (layout != imsm_level_to_layout(level)) {
                if (level == 5)
@@ -3992,7 +4256,7 @@ validate_geometry_imsm_orom(struct intel_super *super, int level, int layout,
  * FIX ME add ahci details
  */
 static int validate_geometry_imsm_volume(struct supertype *st, int level,
-                                        int layout, int raiddisks, int chunk,
+                                        int layout, int raiddisks, int *chunk,
                                         unsigned long long size, char *dev,
                                         unsigned long long *freesize,
                                         int verbose)
@@ -4181,7 +4445,8 @@ static int reserve_space(struct supertype *st, int raiddisks,
        maxsize = merge_extents(super, extent_cnt);
        minsize = size;
        if (size == 0)
-               minsize = chunk;
+               /* chunk is in K */
+               minsize = chunk * 2;
 
        if (cnt < raiddisks ||
            (super->orom && used && used != raiddisks) ||
@@ -4194,8 +4459,8 @@ static int reserve_space(struct supertype *st, int raiddisks,
        if (size == 0) {
                size = maxsize;
                if (chunk) {
-                       size /= chunk;
-                       size *= chunk;
+                       size /= 2 * chunk;
+                       size *= 2 * chunk;
                }
        }
 
@@ -4210,7 +4475,7 @@ static int reserve_space(struct supertype *st, int raiddisks,
 }
 
 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
-                                 int raiddisks, int chunk, unsigned long long size,
+                                 int raiddisks, int *chunk, unsigned long long size,
                                  char *dev, unsigned long long *freesize,
                                  int verbose)
 {
@@ -4224,7 +4489,8 @@ static int validate_geometry_imsm(struct supertype *st, int level, int layout,
        if (level == LEVEL_CONTAINER) {
                /* Must be a fresh device to add to a container */
                return validate_geometry_imsm_container(st, level, layout,
-                                                       raiddisks, chunk, size,
+                                                       raiddisks,
+                                                       chunk?*chunk:0, size,
                                                        dev, freesize,
                                                        verbose);
        }
@@ -4243,7 +4509,8 @@ static int validate_geometry_imsm(struct supertype *st, int level, int layout,
                                                         raiddisks, chunk,
                                                         verbose))
                                return 0;
-                       return reserve_space(st, raiddisks, size, chunk, freesize);
+                       return reserve_space(st, raiddisks, size,
+                                            chunk?*chunk:0, freesize);
                }
                return 1;
        }
@@ -4450,7 +4717,6 @@ static int update_subarray_imsm(struct supertype *st, char *subarray,
 
        return 0;
 }
-#endif /* MDASSEMBLE */
 
 static int is_gen_migration(struct imsm_dev *dev)
 {
@@ -4462,6 +4728,7 @@ static int is_gen_migration(struct imsm_dev *dev)
 
        return 0;
 }
+#endif /* MDASSEMBLE */
 
 static int is_rebuilding(struct imsm_dev *dev)
 {
@@ -4528,11 +4795,19 @@ static struct mdinfo *container_content_imsm(struct supertype *st, char *subarra
        struct mdinfo *rest = NULL;
        unsigned int i;
        int bbm_errors = 0;
+       struct dl *d;
+       int spare_disks = 0;
 
        /* check for bad blocks */
        if (imsm_bbm_log_size(super->anchor))
                bbm_errors = 1;
 
+       /* count spare devices, not used in maps
+        */
+       for (d = super->disks; d; d = d->next)
+               if (d->index == -1)
+                       spare_disks++;
+
        for (i = 0; i < mpb->num_raid_devs; i++) {
                struct imsm_dev *dev;
                struct imsm_map *map;
@@ -4581,7 +4856,7 @@ static struct mdinfo *container_content_imsm(struct supertype *st, char *subarra
 
                        skip = 0;
                        idx = get_imsm_disk_idx(dev, slot, 0);
-                       ord = get_imsm_ord_tbl_ent(dev, slot, 0);
+                       ord = get_imsm_ord_tbl_ent(dev, slot, -1);
                        for (d = super->disks; d ; d = d->next)
                                if (d->index == idx)
                                        break;
@@ -4632,9 +4907,13 @@ static struct mdinfo *container_content_imsm(struct supertype *st, char *subarra
                        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++;
@@ -4645,6 +4924,7 @@ static struct mdinfo *container_content_imsm(struct supertype *st, char *subarra
                }
                /* now that the disk list is up-to-date fixup recovery_start */
                update_recovery_start(dev, this);
+               this->array.spare_disks += spare_disks;
                rest = this;
        }
 
@@ -4801,6 +5081,8 @@ static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
        __u32 ord;
        int slot;
        struct imsm_map *map;
+       char buf[MAX_RAID_SERIAL_LEN+3];
+       unsigned int len, shift = 0;
 
        /* new failures are always set in map[0] */
        map = get_imsm_map(dev, 0);
@@ -4813,8 +5095,12 @@ static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
        if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
                return 0;
 
+       sprintf(buf, "%s:0", disk->serial);
+       if ((len = strlen(buf)) >= MAX_RAID_SERIAL_LEN)
+               shift = len - MAX_RAID_SERIAL_LEN + 1;
+       strncpy((char *)disk->serial, &buf[shift], MAX_RAID_SERIAL_LEN);
+
        disk->status |= FAILED_DISK;
-       disk->status &= ~CONFIGURED_DISK;
        set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
        if (map->failed_disk_num == 0xff)
                map->failed_disk_num = slot;
@@ -4850,6 +5136,37 @@ static void handle_missing(struct intel_super *super, struct imsm_dev *dev)
        super->updates_pending++;
 }
 
+static unsigned long long imsm_set_array_size(struct imsm_dev *dev)
+{
+       int used_disks = imsm_num_data_members(dev, 0);
+       unsigned long long array_blocks;
+       struct imsm_map *map;
+
+       if (used_disks == 0) {
+               /* when problems occures
+                * return current array_blocks value
+                */
+               array_blocks = __le32_to_cpu(dev->size_high);
+               array_blocks = array_blocks << 32;
+               array_blocks += __le32_to_cpu(dev->size_low);
+
+               return array_blocks;
+       }
+
+       /* set array size in metadata
+        */
+       map = get_imsm_map(dev, 0);
+       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));
+
+       return array_blocks;
+}
+
 static void imsm_set_disk(struct active_array *a, int n, int state);
 
 static void imsm_progress_container_reshape(struct intel_super *super)
@@ -4862,13 +5179,13 @@ static void imsm_progress_container_reshape(struct intel_super *super)
        struct imsm_super *mpb = super->anchor;
        int prev_disks = -1;
        int i;
+       int copy_map_size;
 
        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;
@@ -4882,6 +5199,7 @@ static void imsm_progress_container_reshape(struct intel_super *super)
                 * i.e it needs a migr_state
                 */
 
+               copy_map_size = sizeof_imsm_map(map);
                prev_num_members = map->num_members;
                map->num_members = prev_disks;
                dev->vol.migr_state = 1;
@@ -4892,29 +5210,10 @@ static void imsm_progress_container_reshape(struct intel_super *super)
                        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));
+               memcpy(map2, map, copy_map_size);
                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));
-               }
+               imsm_set_array_size(dev);
                super->updates_pending++;
        }
 }
@@ -4944,18 +5243,7 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
                 */
                if (a->curr_action == reshape) {
                        /* still reshaping, maybe update curr_migr_unit */
-                       long long blocks_per_unit = blocks_per_migr_unit(dev);
-                       long long unit = a->last_checkpoint;
-                       if (blocks_per_unit) {
-                               unit /= blocks_per_unit;
-                               if (unit >
-                                   __le32_to_cpu(dev->vol.curr_migr_unit)) {
-                                       dev->vol.curr_migr_unit =
-                                               __cpu_to_le32(unit);
-                                       super->updates_pending++;
-                               }
-                       }
-                       return 0;
+                       goto mark_checkpoint;
                } else {
                        if (a->last_checkpoint == 0 && a->prev_action == reshape) {
                                /* for some reason we aborted the reshape.
@@ -4971,25 +5259,33 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
                        if (a->last_checkpoint >= a->info.component_size) {
                                unsigned long long array_blocks;
                                int used_disks;
-                               /* it seems the reshape is all done */
-                               dev->vol.migr_state = 0;
-                               dev->vol.migr_type = 0;
-                               dev->vol.curr_migr_unit = 0;
+                               struct mdinfo *mdi;
+
+                               used_disks = imsm_num_data_members(dev, 0);
+                               if (used_disks > 0) {
+                                       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;
+                                       a->info.custom_array_size = array_blocks;
+                                       /* encourage manager to update array
+                                        * size
+                                        */
+
+                                       a->check_reshape = 1;
+                               }
+                               /* finalize online capacity expansion/reshape */
+                               for (mdi = a->info.devs; mdi; mdi = mdi->next)
+                                       imsm_set_disk(a,
+                                                     mdi->disk.raid_disk,
+                                                     mdi->curr_state);
 
-                               used_disks = imsm_num_data_members(dev, -1);
-                               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));
-                               a->info.custom_array_size = array_blocks;
-                               a->check_reshape = 1; /* encourage manager to update
-                                                      * array size
-                                                      */
-                               super->updates_pending++;
                                imsm_progress_container_reshape(super);
-                       }                               
+                       }
                }
        }
 
@@ -5024,6 +5320,7 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
                super->updates_pending++;
        }
 
+mark_checkpoint:
        /* check if we can update curr_migr_unit from resync_start, recovery_start */
        blocks_per_unit = blocks_per_migr_unit(dev);
        if (blocks_per_unit) {
@@ -5054,15 +5351,6 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
                super->updates_pending++;
        }
 
-       /* finalize online capacity expansion/reshape */
-       if ((a->curr_action != reshape) &&
-           (a->prev_action == reshape)) {
-               struct mdinfo *mdi;
-
-               for (mdi = a->info.devs; mdi; mdi = mdi->next)
-                       imsm_set_disk(a, mdi->disk.raid_disk, mdi->curr_state);
-       }
-
        return consistent;
 }
 
@@ -5653,16 +5941,17 @@ static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
        int devices_to_reshape = 1;
        struct imsm_super *mpb = super->anchor;
        int ret_val = 0;
+       unsigned int dev_id;
 
-       dprintf("imsm: imsm_process_update() for update_reshape\n");
+       dprintf("imsm: apply_reshape_container_disks_update()\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",
+               dprintf("imsm: 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) ||
@@ -5678,15 +5967,21 @@ static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
                new_disk->disk.status &= ~SPARE_DISK;
        }
 
-       dprintf("imsm: process_update(): update_reshape: volume set"
-               " mpb->num_raid_devs = %i\n", mpb->num_raid_devs);
+       dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
+               mpb->num_raid_devs);
        /* manage changes in volume
         */
-       for (id = super->devlist ; id; id = id->next) {
+       for (dev_id = 0; dev_id < mpb->num_raid_devs; dev_id++) {
                void **sp = *space_list;
                struct imsm_dev *newdev;
                struct imsm_map *newmap, *oldmap;
 
+               for (id = super->devlist ; id; id = id->next) {
+                       if (id->index == dev_id)
+                               break;
+               }
+               if (id == NULL)
+                       break;
                if (!sp)
                        continue;
                *space_list = *sp;
@@ -5700,10 +5995,8 @@ static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
                /* update one device only
                 */
                if (devices_to_reshape) {
-                       int used_disks;
-
-                       dprintf("process_update(): modifying "
-                               "subdev: %i\n", id->index);
+                       dprintf("imsm: modifying subdev: %i\n",
+                               id->index);
                        devices_to_reshape--;
                        newdev->vol.migr_state = 1;
                        newdev->vol.curr_migr_unit = 0;
@@ -5719,24 +6012,7 @@ static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
                        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));
-                       }
+                       imsm_set_array_size(newdev);
                }
 
                sp = (void **)id->dev;
@@ -5744,6 +6020,8 @@ static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
                *sp = tofree;
                tofree = sp;
        }
+       if (tofree)
+               *space_list = tofree;
        ret_val = 1;
 
 update_reshape_exit:
@@ -5751,6 +6029,111 @@ update_reshape_exit:
        return ret_val;
 }
 
+static int apply_takeover_update(struct imsm_update_takeover *u,
+                                struct intel_super *super,
+                                void ***space_list)
+{
+       struct imsm_dev *dev = NULL;
+       struct intel_dev *dv;
+       struct imsm_dev *dev_new;
+       struct imsm_map *map;
+       struct dl *dm, *du;
+       int i;
+
+       for (dv = super->devlist; dv; dv = dv->next)
+               if (dv->index == (unsigned int)u->subarray) {
+                       dev = dv->dev;
+                       break;
+               }
+
+       if (dev == NULL)
+               return 0;
+
+       map = get_imsm_map(dev, 0);
+
+       if (u->direction == R10_TO_R0) {
+               /* Number of failed disks must be half of initial disk number */
+               if (imsm_count_failed(super, dev) != (map->num_members / 2))
+                       return 0;
+
+               /* iterate through devices to mark removed disks as spare */
+               for (dm = super->disks; dm; dm = dm->next) {
+                       if (dm->disk.status & FAILED_DISK) {
+                               int idx = dm->index;
+                               /* update indexes on the disk list */
+/* FIXME this loop-with-the-loop looks wrong,  I'm not convinced
+   the index values will end up being correct.... NB */
+                               for (du = super->disks; du; du = du->next)
+                                       if (du->index > idx)
+                                               du->index--;
+                               /* mark as spare disk */
+                               dm->disk.status = SPARE_DISK;
+                               dm->index = -1;
+                       }
+               }
+               /* update map */
+               map->num_members = map->num_members / 2;
+               map->map_state = IMSM_T_STATE_NORMAL;
+               map->num_domains = 1;
+               map->raid_level = 0;
+               map->failed_disk_num = -1;
+       }
+
+       if (u->direction == R0_TO_R10) {
+               void **space;
+               /* update slots in current disk list */
+               for (dm = super->disks; dm; dm = dm->next) {
+                       if (dm->index >= 0)
+                               dm->index *= 2;
+               }
+               /* create new *missing* disks */
+               for (i = 0; i < map->num_members; i++) {
+                       space = *space_list;
+                       if (!space)
+                               continue;
+                       *space_list = *space;
+                       du = (void *)space;
+                       memcpy(du, super->disks, sizeof(*du));
+                       du->fd = -1;
+                       du->minor = 0;
+                       du->major = 0;
+                       du->index = (i * 2) + 1;
+                       sprintf((char *)du->disk.serial,
+                               " MISSING_%d", du->index);
+                       sprintf((char *)du->serial,
+                               "MISSING_%d", du->index);
+                       du->next = super->missing;
+                       super->missing = du;
+               }
+               /* create new dev and map */
+               space = *space_list;
+               if (!space)
+                       return 0;
+               *space_list = *space;
+               dev_new = (void *)space;
+               memcpy(dev_new, dev, sizeof(*dev));
+               /* update new map */
+               map = get_imsm_map(dev_new, 0);
+               map->num_members = map->num_members * 2;
+               map->map_state = IMSM_T_STATE_DEGRADED;
+               map->num_domains = 2;
+               map->raid_level = 1;
+               /* replace dev<->dev_new */
+               dv->dev = dev_new;
+       }
+       /* update disk order table */
+       for (du = super->disks; du; du = du->next)
+               if (du->index >= 0)
+                       set_imsm_ord_tbl_ent(map, du->index, du->index);
+       for (du = super->missing; du; du = du->next)
+               if (du->index >= 0) {
+                       set_imsm_ord_tbl_ent(map, du->index, du->index);
+                       mark_missing(dev_new, &du->disk, du->index);
+               }
+
+       return 1;
+}
+
 static void imsm_process_update(struct supertype *st,
                                struct metadata_update *update)
 {
@@ -5793,6 +6176,15 @@ static void imsm_process_update(struct supertype *st,
        mpb = super->anchor;
 
        switch (type) {
+       case update_takeover: {
+               struct imsm_update_takeover *u = (void *)update->buf;
+               if (apply_takeover_update(u, super, &update->space_list)) {
+                       imsm_update_version_info(super);
+                       super->updates_pending++;
+               }
+               break;
+       }
+
        case update_reshape_container_disks: {
                struct imsm_update_reshape *u = (void *)update->buf;
                if (apply_reshape_container_disks_update(
@@ -6113,6 +6505,53 @@ static void imsm_prepare_update(struct supertype *st,
        size_t len = 0;
 
        switch (type) {
+       case update_takeover: {
+               struct imsm_update_takeover *u = (void *)update->buf;
+               if (u->direction == R0_TO_R10) {
+                       void **tail = (void **)&update->space_list;
+                       struct imsm_dev *dev = get_imsm_dev(super, u->subarray);
+                       struct imsm_map *map = get_imsm_map(dev, 0);
+                       int num_members = map->num_members;
+                       void *space;
+                       int size, i;
+                       int err = 0;
+                       /* allocate memory for added disks */
+                       for (i = 0; i < num_members; i++) {
+                               size = sizeof(struct dl);
+                               space = malloc(size);
+                               if (!space) {
+                                       err++;
+                                       break;
+                               }
+                               *tail = space;
+                               tail = space;
+                               *tail = NULL;
+                       }
+                       /* allocate memory for new device */
+                       size = sizeof_imsm_dev(super->devlist->dev, 0) +
+                               (num_members * sizeof(__u32));
+                       space = malloc(size);
+                       if (!space)
+                               err++;
+                       else {
+                               *tail = space;
+                               tail = space;
+                               *tail = NULL;
+                       }
+                       if (!err) {
+                               len = disks_to_mpb_size(num_members * 2);
+                       } else {
+                               /* if allocation didn't success, free buffer */
+                               while (update->space_list) {
+                                       void **sp = update->space_list;
+                                       update->space_list = *sp;
+                                       free(sp);
+                               }
+                       }
+               }
+
+               break;
+       }
        case update_reshape_container_disks: {
                /* Every raid device in the container is about to
                 * gain some more devices, and we will enter a
@@ -6263,40 +6702,38 @@ static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned ind
                __free_imsm_disk(dl);
        }
 }
-#endif /* MDASSEMBLE */
 
 static char disk_by_path[] = "/dev/disk/by-path/";
 
 static const char *imsm_get_disk_controller_domain(const char *path)
 {
-       struct sys_dev *list, *hba = NULL;
        char disk_path[PATH_MAX];
-       int ahci = 0;
-       char *dpath = NULL;
-
-       list = find_driver_devices("pci", "ahci");
-       for (hba = list; hba; hba = hba->next)
-               if (devpath_to_vendor(hba->path) == 0x8086)
-                       break;
-
-       if (hba) {
-               struct stat st;
+       char *drv=NULL;
+       struct stat st;
 
-               strncpy(disk_path, disk_by_path, PATH_MAX - 1);
-               strncat(disk_path, path, PATH_MAX - strlen(disk_path) - 1);
-               if (stat(disk_path, &st) == 0) {
-                       dpath = devt_to_devpath(st.st_rdev);
-                       if (dpath)
-                               ahci = path_attached_to_hba(dpath, hba->path);
-               }
+       strncpy(disk_path, disk_by_path, PATH_MAX - 1);
+       strncat(disk_path, path, PATH_MAX - strlen(disk_path) - 1);
+       if (stat(disk_path, &st) == 0) {
+               struct sys_dev* hba;
+               char *path=NULL;
+
+               path = devt_to_devpath(st.st_rdev);
+               if (path == NULL)
+                       return "unknown";
+               hba = find_disk_attached_hba(-1, path);
+               if (hba && hba->type == SYS_DEV_SAS)
+                       drv = "isci";
+               else if (hba && hba->type == SYS_DEV_SATA)
+                       drv = "ahci";
+               else 
+                       drv = "unknown";
+               dprintf("path: %s hba: %s attached: %s\n",
+                       path, (hba) ? hba->path : "NULL", drv);
+               free(path);
+               if (hba)
+                       free_sys_dev(&hba);
        }
-       dprintf("path: %s(%s) hba: %s attached: %d\n",
-               path, dpath, (hba) ? hba->path : "NULL", ahci);
-       free_sys_dev(&list);
-       if (ahci)
-               return "ahci";
-       else
-               return NULL;
+       return drv;
 }
 
 static int imsm_find_array_minor_by_subdev(int subdev, int container, int *minor)
@@ -6318,6 +6755,10 @@ 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;
@@ -6344,7 +6785,7 @@ static int imsm_reshape_is_allowed_on_container(struct supertype *st,
                dprintf("imsm: checking device_num: %i\n",
                        member->container_member);
 
-               if (geo->raid_disks < member->array.raid_disks) {
+               if (geo->raid_disks <= member->array.raid_disks) {
                        /* we work on container for Online Capacity Expansion
                         * only so raid_disks has to grow
                         */
@@ -6469,7 +6910,8 @@ static int imsm_create_metadata_update_for_reshape(
 
        if (spares == NULL
            || delta_disks > spares->array.spare_disks) {
-               dprintf("imsm: ERROR: Cannot get spare devices.\n");
+               fprintf(stderr, Name ": imsm: ERROR: Cannot get spare devices "
+                       "for %s.\n", geo->dev_name);
                goto abort;
        }
 
@@ -6530,28 +6972,177 @@ static void imsm_update_metadata_locally(struct supertype *st,
        }
 }
 
+/***************************************************************************
+* 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;
+       int chunk;
+
+       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_MIGRATION;
+                               check_devs = 1;
+                       }
+                       if (geo->level == 10) {
+                               change = CH_TAKEOVER;
+                               check_devs = 1;
+                       }
+                       break;
+               case 1:
+                       if (geo->level == 0) {
+                               change = CH_TAKEOVER;
+                               check_devs = 1;
+                       }
+                       break;
+               case 5:
+                       if (geo->level == 0)
+                               change = CH_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_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_MIGRATION;
+       else
+               geo->chunksize = info.array.chunk_size;
+
+       chunk = geo->chunksize / 1024;
+       if (!validate_geometry_imsm(st,
+                                   geo->level,
+                                   geo->layout,
+                                   geo->raid_disks,
+                                   &chunk,
+                                   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;
+}
+
+int imsm_takeover(struct supertype *st, struct geo_params *geo)
+{
+       struct intel_super *super = st->sb;
+       struct imsm_update_takeover *u;
+
+       u = malloc(sizeof(struct imsm_update_takeover));
+       if (u == NULL)
+               return 1;
+
+       u->type = update_takeover;
+       u->subarray = super->current_vol;
+
+       /* 10->0 transition */
+       if (geo->level == 0)
+               u->direction = R10_TO_R0;
+
+       /* 0->10 transition */
+       if (geo->level == 10)
+               u->direction = R0_TO_R10;
+
+       /* update metadata locally */
+       imsm_update_metadata_locally(st, u,
+                                       sizeof(struct imsm_update_takeover));
+       /* and possibly remotely */
+       if (st->update_tail)
+               append_metadata_update(st, u,
+                                       sizeof(struct imsm_update_takeover));
+       else
+               free(u);
+
+       return 0;
+}
+
 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)
+                             int delta_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;
 
        dprintf("imsm: reshape_super called.\n");
 
-       memset(&geo, sizeof(struct geo_params), 0);
+       memset(&geo, 0, sizeof(struct geo_params));
 
        geo.dev_name = dev;
+       geo.dev_id = st->devnum;
        geo.size = size;
        geo.level = level;
        geo.layout = layout;
        geo.chunksize = chunksize;
        geo.raid_disks = raid_disks;
+       if (delta_disks != UnSet)
+               geo.raid_disks += delta_disks;
 
        dprintf("\tfor level      : %i\n", geo.level);
        dprintf("\tfor raid_disks : %i\n", geo.raid_disks);
@@ -6559,13 +7150,9 @@ 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)) {
@@ -6589,11 +7176,45 @@ static int imsm_reshape_super(struct supertype *st, long long size, int level,
                        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 {
+                       fprintf(stderr, Name ": (imsm) Operation "
+                               "is not allowed on this container\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 = imsm_takeover(st, &geo);
+                       break;
+               case CH_MIGRATION:
+                       ret_val = 0;
+                       break;
+               default:
+                       ret_val = 1;
+               }
+       }
 
 exit_imsm_reshape_super:
        dprintf("imsm: reshape_super Exit code = %i\n", ret_val);
@@ -6611,6 +7232,7 @@ static int imsm_manage_reshape(
                afd, sra, reshape, st, stripes,
                fds, offsets, dests, destfd, destoffsets);
 }
+#endif /* MDASSEMBLE */
 
 struct superswitch super_imsm = {
 #ifndef        MDASSEMBLE
@@ -6628,6 +7250,10 @@ struct superswitch super_imsm = {
        .kill_subarray = kill_subarray_imsm,
        .update_subarray = update_subarray_imsm,
        .load_container = load_container_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,
 #endif
        .match_home     = match_home_imsm,
        .uuid_from_super= uuid_from_super_imsm,
@@ -6646,10 +7272,6 @@ struct superswitch super_imsm = {
        .free_super     = free_super_imsm,
        .match_metadata_desc = match_metadata_desc_imsm,
        .container_content = container_content_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",