]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - super-intel.c
Allow test to detect 'resync=DELAYED' state
[thirdparty/mdadm.git] / super-intel.c
index ffb03437697e55ede723d916862bf545b8d13df2..f578057d9842b297572bacd6ab11c8ef6cfc4cdc 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 +291,13 @@ struct extent {
        unsigned long long start, size;
 };
 
+/* definitions of reshape process types */
+enum imsm_reshape_type {
+       CH_TAKEOVER,
+       CH_CHUNK_MIGR,
+       CH_LEVEL_MIGRATION
+};
+
 /* definition of messages passed to imsm_process_update */
 enum imsm_update_type {
        update_activate_spare,
@@ -292,6 +306,7 @@ enum imsm_update_type {
        update_rename_array,
        update_add_remove_disk,
        update_reshape_container_disks,
+       update_takeover
 };
 
 struct imsm_update_activate_spare {
@@ -312,6 +327,15 @@ 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 +369,114 @@ 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];
+}
+
+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;
+}
+
+
 static struct supertype *match_metadata_desc_imsm(char *arg)
 {
        struct supertype *st;
@@ -1001,10 +1133,10 @@ 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 +1296,51 @@ 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 +1356,9 @@ 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 +1366,19 @@ 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 +1412,29 @@ 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
 
@@ -2523,6 +2687,11 @@ 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 +2703,8 @@ 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 +2712,15 @@ 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 +2750,6 @@ 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;
@@ -3047,11 +3209,6 @@ static int load_super_imsm(struct supertype *st, int fd, char *devname)
        struct intel_super *super;
        int rv;
 
-#ifndef MDASSEMBLE
-       if (load_super_imsm_all(st, fd, &st->sb, devname) == 0)
-               return 0;
-#endif
-
        if (test_partition(fd))
                /* IMSM not allowed on partitions */
                return 1;
@@ -3442,8 +3599,9 @@ 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;
@@ -3452,14 +3610,48 @@ 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)
@@ -3741,7 +3933,6 @@ static int write_init_super_imsm(struct supertype *st)
        if (st->update_tail) {
                /* queue the recently created array / added disk
                 * as a metadata update */
-               struct dl *d;
                int rv;
 
                /* determine if we are creating a volume or adding a disk */
@@ -3753,11 +3944,6 @@ static int write_init_super_imsm(struct supertype *st)
                } else
                        rv = create_array(st, current_vol);
 
-               for (d = super->disks; d ; d = d->next) {
-                       close(d->fd);
-                       d->fd = -1;
-               }
-
                return rv;
        } else {
                struct dl *d;
@@ -4541,6 +4727,7 @@ static struct mdinfo *container_content_imsm(struct supertype *st, char *subarra
        for (i = 0; i < mpb->num_raid_devs; i++) {
                struct imsm_dev *dev;
                struct imsm_map *map;
+               struct imsm_map *map2;
                struct mdinfo *this;
                int slot;
                char *ep;
@@ -4551,6 +4738,7 @@ static struct mdinfo *container_content_imsm(struct supertype *st, char *subarra
 
                dev = get_imsm_dev(super, i);
                map = get_imsm_map(dev, 0);
+               map2 = get_imsm_map(dev, 1);
 
                /* do not publish arrays that are in the middle of an
                 * unsupported migration
@@ -4632,7 +4820,17 @@ static struct mdinfo *container_content_imsm(struct supertype *st, char *subarra
                        info_d->disk.minor = d->minor;
                        info_d->disk.raid_disk = slot;
                        info_d->recovery_start = recovery_start;
-
+                       if (map2) {
+                               if (slot < map2->num_members)
+                                       info_d->disk.state = (1 << MD_DISK_ACTIVE);
+                               else
+                                       this->array.spare_disks++;
+                       } else {
+                               if (slot < map->num_members)
+                                       info_d->disk.state = (1 << MD_DISK_ACTIVE);
+                               else
+                                       this->array.spare_disks++;
+                       }
                        if (info_d->recovery_start == MaxSector)
                                this->array.working_disks++;
 
@@ -4849,6 +5047,73 @@ static void handle_missing(struct intel_super *super, struct imsm_dev *dev)
 
 static void imsm_set_disk(struct active_array *a, int n, int state);
 
+static void imsm_progress_container_reshape(struct intel_super *super)
+{
+       /* if no device has a migr_state, but some device has a
+        * different number of members than the previous device, start
+        * changing the number of devices in this device to match
+        * previous.
+        */
+       struct imsm_super *mpb = super->anchor;
+       int prev_disks = -1;
+       int i;
+
+       for (i = 0; i < mpb->num_raid_devs; i++) {
+               struct imsm_dev *dev = get_imsm_dev(super, i);
+               struct imsm_map *map = get_imsm_map(dev, 0);
+               struct imsm_map *map2;
+               int prev_num_members;
+               int used_disks;
+
+               if (dev->vol.migr_state)
+                       return;
+
+               if (prev_disks == -1)
+                       prev_disks = map->num_members;
+               if (prev_disks == map->num_members)
+                       continue;
+
+               /* OK, this array needs to enter reshape mode.
+                * i.e it needs a migr_state
+                */
+
+               prev_num_members = map->num_members;
+               map->num_members = prev_disks;
+               dev->vol.migr_state = 1;
+               dev->vol.curr_migr_unit = 0;
+               dev->vol.migr_type = MIGR_GEN_MIGR;
+               for (i = prev_num_members;
+                    i < map->num_members; i++)
+                       set_imsm_ord_tbl_ent(map, i, i);
+               map2 = get_imsm_map(dev, 1);
+               /* Copy the current map */
+               memcpy(map2, map, sizeof_imsm_map(map));
+               map2->num_members = prev_num_members;
+
+               /* calculate new size
+                */
+               used_disks = imsm_num_data_members(dev, 0);
+               if (used_disks) {
+                       unsigned long long array_blocks;
+
+                       array_blocks =
+                               map->blocks_per_member
+                               * used_disks;
+                       /* round array size down to closest MB
+                        */
+                       array_blocks = (array_blocks
+                                       >> SECT_PER_MB_SHIFT)
+                               << SECT_PER_MB_SHIFT;
+                       dev->size_low =
+                               __cpu_to_le32((__u32)array_blocks);
+                       dev->size_high =
+                               __cpu_to_le32(
+                                       (__u32)(array_blocks >> 32));
+               }
+               super->updates_pending++;
+       }
+}
+
 /* Handle dirty -> clean transititions, resync and reshape.  Degraded and rebuild
  * states are handled in imsm_set_disk() with one exception, when a
  * resync is stopped due to a new failure this routine will set the
@@ -4918,6 +5183,7 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
                                                       * array size
                                                       */
                                super->updates_pending++;
+                               imsm_progress_container_reshape(super);
                        }                               
                }
        }
@@ -5335,6 +5601,12 @@ static struct mdinfo *imsm_activate_spare(struct active_array *a,
                /* No repair during migration */
                return NULL;
 
+       if (a->info.array.level == 4)
+               /* No repair for takeovered array
+                * imsm doesn't support raid4
+                */
+               return NULL;
+
        if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
                return NULL;
 
@@ -5563,6 +5835,226 @@ static int add_remove_disk_update(struct intel_super *super)
        return check_degraded;
 }
 
+static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
+                                               struct intel_super *super,
+                                               void ***space_list)
+{
+       struct dl *new_disk;
+       struct intel_dev *id;
+       int i;
+       int delta_disks = u->new_raid_disks - u->old_raid_disks;
+       int disk_count = u->old_raid_disks;
+       void **tofree = NULL;
+       int devices_to_reshape = 1;
+       struct imsm_super *mpb = super->anchor;
+       int ret_val = 0;
+
+       dprintf("imsm: imsm_process_update() for update_reshape\n");
+
+       /* enable spares to use in array */
+       for (i = 0; i < delta_disks; i++) {
+               new_disk = get_disk_super(super,
+                                         major(u->new_disks[i]),
+                                         minor(u->new_disks[i]));
+               dprintf("imsm: imsm_process_update(): new disk "
+                       "for reshape is: %i:%i (%p, index = %i)\n",
+                       major(u->new_disks[i]), minor(u->new_disks[i]),
+                       new_disk, new_disk->index);
+               if ((new_disk == NULL) ||
+                   ((new_disk->index >= 0) &&
+                    (new_disk->index < u->old_raid_disks)))
+                       goto update_reshape_exit;
+               new_disk->index = disk_count++;
+               /* slot to fill in autolayout
+                */
+               new_disk->raiddisk = new_disk->index;
+               new_disk->disk.status |=
+                       CONFIGURED_DISK;
+               new_disk->disk.status &= ~SPARE_DISK;
+       }
+
+       dprintf("imsm: process_update(): update_reshape: volume set"
+               " mpb->num_raid_devs = %i\n", mpb->num_raid_devs);
+       /* manage changes in volume
+        */
+       for (id = super->devlist ; id; id = id->next) {
+               void **sp = *space_list;
+               struct imsm_dev *newdev;
+               struct imsm_map *newmap, *oldmap;
+
+               if (!sp)
+                       continue;
+               *space_list = *sp;
+               newdev = (void*)sp;
+               /* Copy the dev, but not (all of) the map */
+               memcpy(newdev, id->dev, sizeof(*newdev));
+               oldmap = get_imsm_map(id->dev, 0);
+               newmap = get_imsm_map(newdev, 0);
+               /* Copy the current map */
+               memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
+               /* update one device only
+                */
+               if (devices_to_reshape) {
+                       int used_disks;
+
+                       dprintf("process_update(): modifying "
+                               "subdev: %i\n", id->index);
+                       devices_to_reshape--;
+                       newdev->vol.migr_state = 1;
+                       newdev->vol.curr_migr_unit = 0;
+                       newdev->vol.migr_type = MIGR_GEN_MIGR;
+                       newmap->num_members = u->new_raid_disks;
+                       for (i = 0; i < delta_disks; i++) {
+                               set_imsm_ord_tbl_ent(newmap,
+                                                    u->old_raid_disks + i,
+                                                    u->old_raid_disks + i);
+                       }
+                       /* New map is correct, now need to save old map
+                        */
+                       newmap = get_imsm_map(newdev, 1);
+                       memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
+
+                       /* calculate new size
+                        */
+                       used_disks = imsm_num_data_members(newdev, 0);
+                       if (used_disks) {
+                               unsigned long long array_blocks;
+
+                               array_blocks =
+                                       newmap->blocks_per_member * used_disks;
+                               /* round array size down to closest MB
+                                */
+                               array_blocks = (array_blocks
+                                               >> SECT_PER_MB_SHIFT)
+                                       << SECT_PER_MB_SHIFT;
+                               newdev->size_low =
+                                       __cpu_to_le32((__u32)array_blocks);
+                               newdev->size_high =
+                                       __cpu_to_le32((__u32)(array_blocks >> 32));
+                       }
+               }
+
+               sp = (void **)id->dev;
+               id->dev = newdev;
+               *sp = tofree;
+               tofree = sp;
+       }
+       if (tofree)
+               *space_list = tofree;
+       ret_val = 1;
+
+update_reshape_exit:
+
+       return ret_val;
+}
+
+static 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->disk.status = FAILED_DISK;
+                       du->disk.scsi_id = 0;
+                       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->failed_disk_num = map->num_members;
+               map->num_members = map->num_members * 2;
+               map->map_state = IMSM_T_STATE_NORMAL;
+               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 | IMSM_ORD_REBUILD);
+
+       return 1;
+}
+
 static void imsm_process_update(struct supertype *st,
                                struct metadata_update *update)
 {
@@ -5605,88 +6097,18 @@ 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))
+                       super->updates_pending++;
+               break;
+       }
+
        case update_reshape_container_disks: {
                struct imsm_update_reshape *u = (void *)update->buf;
-               struct dl *new_disk;
-               struct intel_dev *id;
-               int i;
-               int delta_disks = u->new_raid_disks - u->old_raid_disks;
-               void **tofree = NULL;
-               int devices_to_reshape = 1;
-
-               dprintf("imsm: imsm_process_update() for update_reshape\n");
-
-               /* enable spares to use in array */
-               for (i = 0; i < delta_disks; i++) {
-                       new_disk = get_disk_super(super,
-                                                 major(u->new_disks[i]),
-                                                 minor(u->new_disks[i]));
-                       dprintf("imsm: imsm_process_update(): new disk "\
-                               "for reshape is: %i:%i (%p, index = %i)\n",
-                               major(u->new_disks[i]), minor(u->new_disks[i]),
-                               new_disk, new_disk->index);
-                       if ((new_disk == NULL) ||
-                           ((new_disk->index >= 0) &&
-                            (new_disk->index < u->old_raid_disks)))
-                               goto update_reshape_exit;
-
-                       new_disk->index = mpb->num_disks++;
-                       /* slot to fill in autolayout */
-                       new_disk->raiddisk = new_disk->index;
-                       new_disk->disk.status |=
-                               CONFIGURED_DISK;
-                       new_disk->disk.status &= ~SPARE_DISK;
-               }
-
-               dprintf("imsm: process_update(): update_reshape: volume set"
-                       " mpb->num_raid_devs = %i\n", mpb->num_raid_devs);
-               /* manage changes in volume
-                */
-               for (id = super->devlist ; id; id = id->next) {
-                       void **sp = update->space_list;
-                       struct imsm_dev *newdev;
-                       struct imsm_map *newmap, *oldmap;
-
-                       if (!sp)
-                               continue;
-                       update->space_list = *sp;
-                       newdev = (void*)sp;
-                       /* Copy the dev, but not (all of) the map */
-                       memcpy(newdev, id->dev, sizeof(*newdev));
-                       oldmap = get_imsm_map(id->dev, 0);
-                       newmap = get_imsm_map(newdev, 0);
-                       /* Copy the current map */
-                       memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
-                       /* update one device only
-                        */
-                       if (devices_to_reshape) {
-                               dprintf("process_update(): modifying "\
-                                       "subdev: %i\n", id->index);
-                               devices_to_reshape--;
-                               newdev->vol.migr_state = 1;
-                               newdev->vol.curr_migr_unit = 0;
-                               newdev->vol.migr_type = MIGR_GEN_MIGR;
-                               newmap->num_members = u->new_raid_disks;
-                               for (i = 0; i < delta_disks; i++) {
-                                       set_imsm_ord_tbl_ent(newmap,
-                                                       u->old_raid_disks + i,
-                                                       u->old_raid_disks + i);
-                               }
-                               /* New map is correct, now need to save old map
-                                */
-                               newmap = get_imsm_map(newdev, 1);
-                               memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
-                       }
-
-                       sp = (void **)id->dev;
-                       id->dev = newdev;
-                       *sp = tofree;
-                       tofree = sp;
-               }
-
-               update->space_list = tofree;
-               super->updates_pending++;
-update_reshape_exit:
+               if (apply_reshape_container_disks_update(
+                           u, super, &update->space_list))
+                       super->updates_pending++;
                break;
        }
        case update_activate_spare: {
@@ -6002,6 +6424,53 @@ 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
@@ -6158,34 +6627,33 @@ 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)
@@ -6207,6 +6675,10 @@ 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;
@@ -6233,7 +6705,7 @@ 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
                         */
@@ -6329,6 +6801,7 @@ static int imsm_create_metadata_update_for_reshape(
        struct mdinfo *spares = NULL;
        int i;
        int delta_disks = 0;
+       struct mdinfo *dev;
 
        dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
                geo->raid_disks);
@@ -6367,27 +6840,18 @@ static int imsm_create_metadata_update_for_reshape(
        dprintf("imsm: %i spares are available.\n\n",
                spares->array.spare_disks);
 
+       dev = spares->devs;
        for (i = 0; i < delta_disks; i++) {
-               struct mdinfo *dev = spares->devs;
                struct dl *dl;
 
+               if (dev == NULL)
+                       break;
                u->new_disks[i] = makedev(dev->disk.major,
                                          dev->disk.minor);
                dl = get_disk_super(super, dev->disk.major, dev->disk.minor);
-               dl->index = mpb->num_disks++;
-       }
-       /* Now update the metadata so that container_content will find
-        * the new devices
-        */
-       for (i = 0; i < mpb->num_raid_devs; i++) {
-               int d;
-               struct imsm_dev *dev = get_imsm_dev(super, i);
-               struct imsm_map *map = get_imsm_map(dev, 0);
-               map->num_members = geo->raid_disks;
-               for (d = 0; d < delta_disks; d++) {
-                       set_imsm_ord_tbl_ent(map, old_raid_disks + d,
-                                            mpb->num_disks - delta_disks + d);
-               }
+               dl->index = mpb->num_disks;
+               mpb->num_disks++;
+               dev = dev->next;
        }
 
 abort:
@@ -6407,16 +6871,178 @@ abort:
        return 0;
 }
 
+static void imsm_update_metadata_locally(struct supertype *st,
+                                        void *buf, int len)
+{
+       struct metadata_update mu;
+
+       mu.buf = buf;
+       mu.len = len;
+       mu.space = NULL;
+       mu.space_list = NULL;
+       mu.next = NULL;
+       imsm_prepare_update(st, &mu);
+       imsm_process_update(st, &mu);
+
+       while (mu.space_list) {
+               void **space = mu.space_list;
+               mu.space_list = *space;
+               free(space);
+       }
+}
+
+/***************************************************************************
+* Function:    imsm_analyze_change
+* Description: Function analyze change for single volume
+*              and validate if transition is supported
+* Parameters:  Geometry parameters, supertype structure
+* Returns:     Operation type code on success, -1 if fail
+****************************************************************************/
+enum imsm_reshape_type imsm_analyze_change(struct supertype *st,
+                                          struct geo_params *geo)
+{
+       struct mdinfo info;
+       int change = -1;
+       int check_devs = 0;
+
+       getinfo_super_imsm_volume(st, &info, NULL);
+
+       if ((geo->level != info.array.level) &&
+           (geo->level >= 0) &&
+           (geo->level != UnSet)) {
+               switch (info.array.level) {
+               case 0:
+                       if (geo->level == 5) {
+                               change = CH_LEVEL_MIGRATION;
+                               check_devs = 1;
+                       }
+                       if (geo->level == 10) {
+                               change = CH_TAKEOVER;
+                               check_devs = 1;
+                       }
+                       break;
+               case 1:
+                       if (geo->level == 0) {
+                               change = CH_TAKEOVER;
+                               check_devs = 1;
+                       }
+                       break;
+               case 5:
+                       if (geo->level != 0)
+                               change = CH_LEVEL_MIGRATION;
+                       break;
+               case 10:
+                       if (geo->level == 0) {
+                               change = CH_TAKEOVER;
+                               check_devs = 1;
+                       }
+                       break;
+               }
+               if (change == -1) {
+                       fprintf(stderr,
+                               Name " Error. Level Migration from %d to %d "
+                               "not supported!\n",
+                               info.array.level, geo->level);
+                       goto analyse_change_exit;
+               }
+       } else
+               geo->level = info.array.level;
+
+       if ((geo->layout != info.array.layout)
+           && ((geo->layout != UnSet) && (geo->layout != -1))) {
+               change = CH_LEVEL_MIGRATION;
+               if ((info.array.layout == 0)
+                   && (info.array.level == 5)
+                   && (geo->layout == 5)) {
+                       /* reshape 5 -> 4 */
+               } else if ((info.array.layout == 5)
+                          && (info.array.level == 5)
+                          && (geo->layout == 0)) {
+                       /* reshape 4 -> 5 */
+                       geo->layout = 0;
+                       geo->level = 5;
+               } else {
+                       fprintf(stderr,
+                               Name " Error. Layout Migration from %d to %d "
+                               "not supported!\n",
+                               info.array.layout, geo->layout);
+                       change = -1;
+                       goto analyse_change_exit;
+               }
+       } else
+               geo->layout = info.array.layout;
+
+       if ((geo->chunksize > 0) && (geo->chunksize != UnSet)
+           && (geo->chunksize != info.array.chunk_size))
+               change = CH_CHUNK_MIGR;
+       else
+               geo->chunksize = info.array.chunk_size;
+
+       if (!validate_geometry_imsm(st,
+                                   geo->level,
+                                   geo->layout,
+                                   geo->raid_disks,
+                                   (geo->chunksize / 1024),
+                                   geo->size,
+                                   0, 0, 1))
+               change = -1;
+
+       if (check_devs) {
+               struct intel_super *super = st->sb;
+               struct imsm_super *mpb = super->anchor;
+
+               if (mpb->num_raid_devs > 1) {
+                       fprintf(stderr,
+                               Name " Error. Cannot perform operation on %s"
+                               "- for this operation it MUST be single "
+                               "array in container\n",
+                               geo->dev_name);
+                       change = -1;
+               }
+       }
+
+analyse_change_exit:
+
+       return change;
+}
+
+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)
 {
-       /* currently we only support increasing the number of devices
-        * for a container.  This increases the number of device for each
-        * member array.  They must all be RAID0 or RAID5.
-        */
-
        int ret_val = 1;
        struct geo_params geo;
 
@@ -6425,6 +7051,7 @@ static int imsm_reshape_super(struct supertype *st, long long size, int level,
        memset(&geo, sizeof(struct geo_params), 0);
 
        geo.dev_name = dev;
+       geo.dev_id = st->devnum;
        geo.size = size;
        geo.level = level;
        geo.layout = layout;
@@ -6437,11 +7064,9 @@ static int imsm_reshape_super(struct supertype *st, long long size, int level,
        if (experimental() == 0)
                return ret_val;
 
-       /* verify reshape conditions
-        * on container level we can only increase number of devices. */
        if (st->container_dev == st->devnum) {
-               /* check for delta_disks > 0
-                *and supported raid levels 0 and 5 only in container */
+               /* On container level we can only increase number of devices. */
+               dprintf("imsm: info: Container operation\n");
                int old_raid_disks = 0;
                if (imsm_reshape_is_allowed_on_container(
                            st, &geo, &old_raid_disks)) {
@@ -6451,22 +7076,80 @@ static int imsm_reshape_super(struct supertype *st, long long size, int level,
                        len = imsm_create_metadata_update_for_reshape(
                                st, &geo, old_raid_disks, &u);
 
-                       if (len) {
-                               ret_val = 0;
+                       if (len <= 0) {
+                               dprintf("imsm: Cannot prepare update\n");
+                               goto exit_imsm_reshape_super;
+                       }
+
+                       ret_val = 0;
+                       /* update metadata locally */
+                       imsm_update_metadata_locally(st, u, len);
+                       /* and possibly remotely */
+                       if (st->update_tail)
                                append_metadata_update(st, u, len);
-                       else
-                               fprintf(stderr, Name "imsm: Cannot prepare "
-                                       "update\n");
-               } else
+                       else
+                               free(u);
+
+               } else {
                        fprintf(stderr, Name "imsm: Operation is not allowed "
                                "on this container\n");
-       } else
-               fprintf(stderr, Name "imsm: not a container operation\n");
+               }
+       } else {
+               /* On volume level we support following operations
+                * - takeover: raid10 -> raid0; raid0 -> raid10
+                * - chunk size migration
+                * - migration: raid5 -> raid0; raid0 -> raid5
+                */
+               struct intel_super *super = st->sb;
+               struct intel_dev *dev = super->devlist;
+               int change, devnum;
+               dprintf("imsm: info: Volume operation\n");
+               /* find requested device */
+               while (dev) {
+                       imsm_find_array_minor_by_subdev(dev->index, st->container_dev, &devnum);
+                       if (devnum == geo.dev_id)
+                               break;
+                       dev = dev->next;
+               }
+               if (dev == NULL) {
+                       fprintf(stderr, Name " Cannot find %s (%i) subarray\n",
+                               geo.dev_name, geo.dev_id);
+                       goto exit_imsm_reshape_super;
+               }
+               super->current_vol = dev->index;
+               change = imsm_analyze_change(st, &geo);
+               switch (change) {
+               case CH_TAKEOVER:
+                       ret_val = imsm_takeover(st, &geo);
+                       break;
+               case CH_CHUNK_MIGR:
+                       ret_val = 0;
+                       break;
+               case CH_LEVEL_MIGRATION:
+                       ret_val = 0;
+                       break;
+               default:
+                       ret_val = 1;
+               }
+       }
 
+exit_imsm_reshape_super:
        dprintf("imsm: reshape_super Exit code = %i\n", ret_val);
        return ret_val;
 }
 
+static int imsm_manage_reshape(
+       int afd, struct mdinfo *sra, struct reshape *reshape,
+       struct supertype *st, unsigned long stripes,
+       int *fds, unsigned long long *offsets,
+       int dests, int *destfd, unsigned long long *destoffsets)
+{
+       /* Just use child_monitor for now */
+       return child_monitor(
+               afd, sra, reshape, st, stripes,
+               fds, offsets, dests, destfd, destoffsets);
+}
+
 struct superswitch super_imsm = {
 #ifndef        MDASSEMBLE
        .examine_super  = examine_super_imsm,
@@ -6504,6 +7187,7 @@ struct superswitch super_imsm = {
        .default_geometry = default_geometry_imsm,
        .get_disk_controller_domain = imsm_get_disk_controller_domain,
        .reshape_super  = imsm_reshape_super,
+       .manage_reshape = imsm_manage_reshape,
 
        .external       = 1,
        .name = "imsm",