+ if (!avail)
+ return 0;
+ /* We need to reject any device that thinks the best device is
+ * failed or missing */
+ for (b = 0; b < raid_disks; b++)
+ if (avail[b] == 2)
+ break;
+ cnt = 0;
+ for (i = 0 ; i < raid_disks ; i++) {
+ if (i != b && avail[i])
+ if (devmap[raid_disks * best[i] + b] == 0) {
+ /* This device thinks 'b' is failed -
+ * don't use it */
+ devnum = best[i];
+ for (d=sra->devs ; devnum; d = d->next)
+ devnum--;
+ d->disk.state |= (1 << MD_DISK_REMOVED);
+ avail[i] = 0;
+ }
+ if (avail[i])
+ cnt++;
+ }
+ free(best);
+ free(devmap);
+ return cnt;
+}
+
+/* test if container has degraded member(s) */
+static int container_members_max_degradation(struct map_ent *map, struct map_ent *me)
+{
+ mdu_array_info_t array;
+ int afd;
+ int max_degraded = 0;
+
+ for(; map; map = map->next) {
+ if (!is_subarray(map->metadata) ||
+ devname2devnum(map->metadata+1) != me->devnum)
+ continue;
+ afd = open_dev(map->devnum);
+ if (afd < 0)
+ continue;
+ /* most accurate information regarding array degradation */
+ if (ioctl(afd, GET_ARRAY_INFO, &array) >= 0) {
+ int degraded = array.raid_disks - array.active_disks -
+ array.spare_disks;
+ if (degraded > max_degraded)
+ max_degraded = degraded;
+ }
+ close(afd);
+ }
+ return (max_degraded);
+}
+
+static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
+ struct map_ent *target, int bare,
+ struct supertype *st, int verbose)
+{
+ /* This device doesn't have any md metadata
+ * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
+ * If 'st' is not set, then we only know that some metadata allows this,
+ * others possibly don't.
+ * So look for a container or array to attach the device to.
+ * Prefer 'target' if that is set and the array is found.
+ *
+ * If st is set, then only arrays of that type are considered
+ * Return 0 on success, or some exit code on failure, probably 1.
+ */
+ int rv = 1;
+ struct stat stb;
+ struct map_ent *mp, *map = NULL;
+ struct mdinfo *chosen = NULL;
+ int dfd = *dfdp;
+
+ if (fstat(dfd, &stb) != 0)
+ return 1;
+
+ /*
+ * Now we need to find a suitable array to add this to.
+ * We only accept arrays that:
+ * - match 'st'
+ * - are in the same domains as the device
+ * - are of an size for which the device will be useful
+ * and we choose the one that is the most degraded
+ */
+
+ if (map_lock(&map)) {
+ fprintf(stderr, Name ": failed to get exclusive lock on "
+ "mapfile\n");
+ return 1;
+ }
+ for (mp = map ; mp ; mp = mp->next) {
+ struct supertype *st2;
+ struct domainlist *dl = NULL;
+ struct mdinfo *sra;
+ unsigned long long devsize;
+ unsigned long long component_size = 0;
+
+ if (is_subarray(mp->metadata))
+ continue;
+ if (st) {
+ st2 = st->ss->match_metadata_desc(mp->metadata);
+ if (!st2 ||
+ (st->minor_version >= 0 &&
+ st->minor_version != st2->minor_version)) {
+ if (verbose > 1)
+ fprintf(stderr, Name ": not adding %s to %s as metadata type doesn't match\n",
+ devname, mp->path);
+ free(st2);
+ continue;
+ }
+ free(st2);
+ }
+ sra = sysfs_read(-1, mp->devnum,
+ GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
+ GET_DEGRADED|GET_COMPONENT|GET_VERSION);
+ if (!sra) {
+ /* Probably a container - no degraded info */
+ sra = sysfs_read(-1, mp->devnum,
+ GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
+ GET_COMPONENT|GET_VERSION);
+ if (sra)
+ sra->array.failed_disks = -1;
+ }
+ if (!sra)
+ continue;
+ if (st == NULL) {
+ int i;
+ st2 = NULL;
+ for(i=0; !st2 && superlist[i]; i++)
+ st2 = superlist[i]->match_metadata_desc(
+ sra->text_version);
+ if (!st2) {
+ if (verbose > 1)
+ fprintf(stderr, Name ": not adding %s to %s"
+ " as metadata not recognised.\n",
+ devname, mp->path);
+ goto next;
+ }
+ /* Need to double check the 'act_spare' permissions applies
+ * to this metadata.
+ */
+ if (!policy_action_allows(pol, st2->ss->name, act_spare))
+ goto next;
+ if (!bare && !policy_action_allows(pol, st2->ss->name,
+ act_spare_same_slot))
+ goto next;
+ } else
+ st2 = st;
+ /* update number of failed disks for mostly degraded
+ * container member */
+ if (sra->array.failed_disks == -1)
+ sra->array.failed_disks = container_members_max_degradation(map, mp);
+
+ get_dev_size(dfd, NULL, &devsize);
+ if (sra->component_size == 0) {
+ /* true for containers, here we must read superblock
+ * to obtain minimum spare size */
+ struct supertype *st3 = dup_super(st2);
+ int mdfd = open_dev(mp->devnum);
+ if (!mdfd)
+ goto next;
+ if (st3->ss->load_container &&
+ !st3->ss->load_container(st3, mdfd, mp->path)) {
+ component_size = st3->ss->min_acceptable_spare_size(st3);
+ st3->ss->free_super(st3);
+ }
+ free(st3);
+ close(mdfd);
+ }
+ if ((sra->component_size > 0 &&
+ st2->ss->avail_size(st2, devsize) < sra->component_size)
+ ||
+ (sra->component_size == 0 && devsize < component_size)) {
+ if (verbose > 1)
+ fprintf(stderr, Name ": not adding %s to %s as it is too small\n",
+ devname, mp->path);
+ goto next;
+ }
+ /* test against target.
+ * If 'target' is set and 'bare' is false, we only accept
+ * arrays/containers that match 'target'.
+ * If 'target' is set and 'bare' is true, we prefer the
+ * array which matches 'target'.
+ * target is considered only if we deal with degraded array
+ */
+ if (target && policy_action_allows(pol, st2->ss->name,
+ act_spare_same_slot)) {
+ if (strcmp(target->metadata, mp->metadata) == 0 &&
+ memcmp(target->uuid, mp->uuid,
+ sizeof(target->uuid)) == 0 &&
+ sra->array.failed_disks > 0) {
+ /* This is our target!! */
+ if (chosen)
+ sysfs_free(chosen);
+ chosen = sra;
+ sra = NULL;
+ /* skip to end so we don't check any more */
+ while (mp->next)
+ mp = mp->next;
+ goto next;
+ }
+ /* not our target */
+ if (!bare)
+ goto next;
+ }
+
+ dl = domain_from_array(sra, st2->ss->name);
+ if (domain_test(dl, pol, st2->ss->name) != 1) {
+ /* domain test fails */
+ if (verbose > 1)
+ fprintf(stderr, Name ": not adding %s to %s as"
+ " it is not in a compatible domain\n",
+ devname, mp->path);
+
+ goto next;
+ }
+ /* all tests passed, OK to add to this array */
+ if (!chosen) {
+ chosen = sra;
+ sra = NULL;
+ } else if (chosen->array.failed_disks < sra->array.failed_disks) {
+ sysfs_free(chosen);
+ chosen = sra;
+ sra = NULL;
+ }
+ next:
+ if (sra)
+ sysfs_free(sra);
+ if (st != st2)
+ free(st2);
+ if (dl)
+ domain_free(dl);
+ }
+ if (chosen) {
+ /* add current device to chosen array as a spare */
+ int mdfd = open_dev(devname2devnum(chosen->sys_name));
+ if (mdfd >= 0) {
+ struct mddev_dev devlist;
+ char devname[20];
+ devlist.next = NULL;
+ devlist.used = 0;
+ devlist.re_add = 0;
+ devlist.writemostly = 0;
+ devlist.devname = devname;
+ sprintf(devname, "%d:%d", major(stb.st_rdev),
+ minor(stb.st_rdev));
+ devlist.disposition = 'a';
+ close(dfd);
+ *dfdp = -1;
+ rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
+ -1, 0, NULL);
+ close(mdfd);
+ }
+ if (verbose > 0) {
+ if (rv == 0)
+ fprintf(stderr, Name ": added %s as spare for %s\n",
+ devname, chosen->sys_name);
+ else
+ fprintf(stderr, Name ": failed to add %s as spare for %s\n",
+ devname, chosen->sys_name);
+ }
+ sysfs_free(chosen);
+ }
+ return rv;
+}
+
+static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
+ struct supertype *st, int verbose)
+{
+ /* we know that at least one partition virtual-metadata is
+ * allowed to incorporate spares like this device. We need to
+ * find a suitable device to copy partition information from.
+ *
+ * Getting a list of all disk (not partition) devices is
+ * slightly non-trivial. We could look at /sys/block, but
+ * that is theoretically due to be removed. Maybe best to use
+ * /dev/disk/by-path/?* and ignore names ending '-partNN' as
+ * we depend on this directory of 'path' info. But that fails
+ * to find loop devices and probably others. Maybe don't
+ * worry about that, they aren't the real target.
+ *
+ * So: check things in /dev/disk/by-path to see if they are in
+ * a compatible domain, then load the partition table and see
+ * if it is OK for the new device, and choose the largest
+ * partition table that fits.
+ */
+ DIR *dir;
+ struct dirent *de;
+ char *chosen = NULL;
+ unsigned long long chosen_size = 0;
+ struct supertype *chosen_st = NULL;
+ int fd;
+
+ dir = opendir("/dev/disk/by-path");
+ if (!dir)
+ return 1;
+ while ((de = readdir(dir)) != NULL) {
+ char *ep;
+ struct dev_policy *pol2 = NULL;
+ struct domainlist *domlist = NULL;
+ int fd = -1;
+ struct mdinfo info;
+ struct supertype *st2 = NULL;
+ char *devname = NULL;
+ unsigned long long devsectors;
+
+ if (de->d_ino == 0 ||
+ de->d_name[0] == '.' ||
+ (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
+ goto next;
+
+ ep = de->d_name + strlen(de->d_name);
+ while (ep > de->d_name &&
+ isdigit(ep[-1]))
+ ep--;
+ if (ep > de->d_name + 5 &&
+ strncmp(ep-5, "-part", 5) == 0)
+ /* This is a partition - skip it */
+ goto next;
+
+ pol2 = path_policy(de->d_name, type_disk);
+
+ domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
+ if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
+ /* new device is incompatible with this device. */
+ goto next;
+
+ domain_free(domlist);
+ domlist = NULL;
+
+ if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
+ devname = NULL;
+ goto next;
+ }
+ fd = open(devname, O_RDONLY);
+ if (fd < 0)
+ goto next;
+ if (get_dev_size(fd, devname, &devsectors) == 0)
+ goto next;
+ devsectors >>= 9;
+
+ if (st)
+ st2 = dup_super(st);
+ else
+ st2 = guess_super_type(fd, guess_partitions);
+ if (st2 == NULL ||
+ st2->ss->load_super(st2, fd, NULL) < 0)
+ goto next;
+
+ if (!st) {
+ /* Check domain policy again, this time referring to metadata */
+ domain_merge(&domlist, pol2, st2->ss->name);
+ if (domain_test(domlist, pol, st2->ss->name) != 1)
+ /* Incompatible devices for this metadata type */
+ goto next;
+ if (!policy_action_allows(pol, st2->ss->name, act_spare))
+ /* Some partition types allow sparing, but not
+ * this one.
+ */
+ goto next;
+ }
+
+ st2->ss->getinfo_super(st2, &info, NULL);
+ if (info.component_size > devsectors)
+ /* This partitioning doesn't fit in the device */
+ goto next;
+
+ /* This is an acceptable device to copy partition
+ * metadata from. We could just stop here, but I
+ * think I want to keep looking incase a larger
+ * metadata which makes better use of the device can
+ * be found.
+ */
+ if (chosen == NULL ||
+ chosen_size < info.component_size) {
+ chosen_size = info.component_size;
+ free(chosen);
+ chosen = devname;
+ devname = NULL;
+ if (chosen_st) {
+ chosen_st->ss->free_super(chosen_st);
+ free(chosen_st);
+ }
+ chosen_st = st2;
+ st2 = NULL;
+ }
+
+ next:
+ free(devname);
+ domain_free(domlist);
+ dev_policy_free(pol2);
+ if (st2)
+ st2->ss->free_super(st2);
+ free(st2);
+
+ if (fd >= 0)
+ close(fd);
+ }
+
+ if (!chosen)
+ return 1;
+
+ /* 'chosen' is the best device we can find. Let's write its
+ * metadata to devname dfd is read-only so don't use that
+ */
+ fd = open(devname, O_RDWR);
+ if (fd >= 0) {
+ chosen_st->ss->store_super(chosen_st, fd);
+ close(fd);
+ }
+ free(chosen);
+ chosen_st->ss->free_super(chosen_st);
+ free(chosen_st);
+ return 0;
+}
+
+static int is_bare(int dfd)
+{
+ unsigned long long size = 0;
+ char bufpad[4096 + 4096];
+ char *buf = (char*)(((long)bufpad + 4096) & ~4095);
+
+ if (lseek(dfd, 0, SEEK_SET) != 0 ||
+ read(dfd, buf, 4096) != 4096)
+ return 0;
+
+ if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
+ return 0;
+ if (memcmp(buf, buf+1, 4095) != 0)
+ return 0;
+
+ /* OK, first 4K appear blank, try the end. */
+ get_dev_size(dfd, NULL, &size);
+ if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
+ read(dfd, buf, 4096) != 4096)
+ return 0;
+
+ if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
+ return 0;
+ if (memcmp(buf, buf+1, 4095) != 0)
+ return 0;
+
+ return 1;
+}
+
+/* adding a spare to a regular array is quite different from adding one to
+ * a set-of-partitions virtual array.
+ * This function determines which is worth trying and tries as appropriate.
+ * Arrays are given priority over partitions.
+ */
+static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
+ struct map_ent *target,
+ struct supertype *st, int verbose)
+{
+ int i;
+ int rv;
+ int arrays_ok = 0;
+ int partitions_ok = 0;
+ int dfd = *dfdp;
+ int bare;
+
+ /* Can only add a spare if device has at least one domain */
+ if (pol_find(pol, pol_domain) == NULL)
+ return 1;
+ /* And only if some action allows spares */
+ if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
+ return 1;
+
+ /* Now check if the device is bare.
+ * bare devices can always be added as a spare
+ * non-bare devices can only be added if spare-same-slot is permitted,
+ * and this device is replacing a previous device - in which case 'target'
+ * will be set.
+ */
+ if (!is_bare(dfd)) {
+ /* Must have a target and allow same_slot */
+ /* Later - may allow force_spare without target */
+ if (!target ||
+ !policy_action_allows(pol, st?st->ss->name:NULL,
+ act_spare_same_slot)) {
+ if (verbose > 1)
+ fprintf(stderr, Name ": %s is not bare, so not "
+ "considering as a spare\n",
+ devname);
+ return 1;
+ }
+ bare = 0;
+ } else
+ bare = 1;
+
+ /* It might be OK to add this device to an array - need to see
+ * what arrays might be candidates.
+ */
+ if (st) {
+ /* just try try 'array' or 'partition' based on this metadata */
+ if (st->ss->add_to_super)
+ return array_try_spare(devname, dfdp, pol, target, bare,
+ st, verbose);
+ else
+ return partition_try_spare(devname, dfdp, pol,
+ st, verbose);
+ }
+ /* No metadata was specified or found so options are open.
+ * Check for whether any array metadata, or any partition metadata
+ * might allow adding the spare. This check is just help to avoid
+ * a more costly scan of all arrays when we can be sure that will
+ * fail.
+ */
+ for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
+ if (superlist[i]->add_to_super && !arrays_ok &&
+ policy_action_allows(pol, superlist[i]->name, act_spare))
+ arrays_ok = 1;
+ if (superlist[i]->add_to_super == NULL && !partitions_ok &&
+ policy_action_allows(pol, superlist[i]->name, act_spare))
+ partitions_ok = 1;
+ }
+ rv = 1;
+ if (arrays_ok)
+ rv = array_try_spare(devname, dfdp, pol, target, bare,
+ st, verbose);
+ if (rv != 0 && partitions_ok)
+ rv = partition_try_spare(devname, dfdp, pol, st, verbose);
+ return rv;