2 * Incremental.c - support --incremental. Part of:
3 * mdadm - manage Linux "md" devices aka RAID arrays.
5 * Copyright (C) 2006-2013 Neil Brown <neilb@suse.de>
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Email: <neilb@suse.de>
27 * QVB Post Office, NSW 1230
36 static int count_active(struct supertype *st, struct mdinfo *sra,
37 int mdfd, char **availp,
39 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
40 int number, __u64 events, int verbose,
42 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
43 struct map_ent *target,
44 struct supertype *st, int verbose);
46 static int Incremental_container(struct supertype *st, char *devname,
47 struct context *c, char *only);
49 int Incremental(struct mddev_dev *devlist, struct context *c,
52 /* Add this device to an array, creating the array if necessary
53 * and starting the array if sensible or - if runstop>0 - if possible.
55 * This has several steps:
57 * 1/ Check if device is permitted by mdadm.conf, reject if not.
58 * 2/ Find metadata, reject if none appropriate (check
59 * version/name from args)
60 * 3/ Check if there is a match in mdadm.conf
61 * 3a/ if not, check for homehost match. If no match, assemble as
63 * 4/ Determine device number.
64 * - If in mdadm.conf with std name, use that
65 * - UUID in /var/run/mdadm.map use that
66 * - If name is suggestive, use that. unless in use with different uuid.
67 * - Choose a free, high number.
68 * - Use a partitioned device unless strong suggestion not to.
70 * Don't choose partitioned for containers.
71 * 5/ Find out if array already exists
73 * - choose a name, from mdadm.conf or 'name' field in array.
77 * - check one drive in array to make sure metadata is a reasonably
78 * close match. Reject if not (e.g. different type)
80 * 6/ Make sure /var/run/mdadm.map contains this array.
81 * 7/ Is there enough devices to possibly start the array?
82 * For a container, this means running Incremental_container.
83 * 7a/ if not, finish with success.
85 * - read all metadata and arrange devices like -A does
86 * - if number of OK devices match expected, or -R and there are enough,
87 * start the array (auto-readonly).
90 struct mdinfo info, dinfo;
91 struct mdinfo *sra = NULL, *d;
92 struct mddev_ident *match;
93 char chosen_name[1024];
96 struct map_ent *mp, *map = NULL;
97 int dfd = -1, mdfd = -1;
102 mdu_array_info_t ainf;
103 struct dev_policy *policy = NULL;
104 struct map_ent target_array;
106 char *devname = devlist->devname;
107 int journal_device_missing = 0;
109 struct createinfo *ci = conf_get_create_info();
111 if (stat(devname, &stb) < 0) {
113 pr_err("stat failed for %s: %s.\n",
114 devname, strerror(errno));
117 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
119 pr_err("%s is not a block device.\n",
123 dfd = dev_open(devname, O_RDONLY);
126 pr_err("cannot open %s: %s.\n",
127 devname, strerror(errno));
130 /* If the device is a container, we do something very different */
131 if (must_be_container(dfd)) {
133 st = super_by_fd(dfd, NULL);
134 if (st && st->ss->load_container)
135 rv = st->ss->load_container(st, dfd, NULL);
138 if (!rv && st->ss->container_content) {
140 pr_err("failed to get exclusive lock on mapfile\n");
142 printf("MD_DEVNAME=%s\n", devname);
143 rv = Incremental_container(st, devname, c, NULL);
148 pr_err("%s is not part of an md array.\n",
153 /* 1/ Check if device is permitted by mdadm.conf */
155 for (;devlist; devlist = devlist->next)
156 if (conf_test_dev(devlist->devname))
159 devlist = conf_get_devs();
160 for (;devlist; devlist = devlist->next) {
162 if (stat(devlist->devname, &st2) == 0 &&
163 (st2.st_mode & S_IFMT) == S_IFBLK &&
164 st2.st_rdev == stb.st_rdev)
170 pr_err("%s not permitted by mdadm.conf.\n",
175 /* 2/ Find metadata, reject if none appropriate (check
176 * version/name from args) */
178 if (fstat(dfd, &stb) < 0) {
180 pr_err("fstat failed for %s: %s.\n",
181 devname, strerror(errno));
184 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
186 pr_err("%s is not a block device.\n",
191 dinfo.disk.major = major(stb.st_rdev);
192 dinfo.disk.minor = minor(stb.st_rdev);
194 policy = disk_policy(&dinfo);
195 have_target = policy_check_path(&dinfo, &target_array);
197 if (st == NULL && (st = guess_super_type(dfd, guess_array)) == NULL) {
199 pr_err("no recognisable superblock on %s.\n",
201 rv = try_spare(devname, &dfd, policy,
202 have_target ? &target_array : NULL,
206 st->ignore_hw_compat = 0;
208 if (st->ss->compare_super == NULL ||
209 st->ss->load_super(st, dfd, c->verbose >= 0 ? devname : NULL)) {
211 pr_err("no RAID superblock on %s.\n",
213 rv = try_spare(devname, &dfd, policy,
214 have_target ? &target_array : NULL,
219 close (dfd); dfd = -1;
221 st->ss->getinfo_super(st, &info, NULL);
223 /* 3/ Check if there is a match in mdadm.conf */
224 match = conf_match(st, &info, devname, c->verbose, &rv);
225 if (!match && rv == 2)
228 if (match && match->devname
229 && strcasecmp(match->devname, "<ignore>") == 0) {
231 pr_err("array containing %s is explicitly ignored by mdadm.conf\n",
236 /* 3a/ if not, check for homehost match. If no match, continue
237 * but don't trust the 'name' in the array. Thus a 'random' minor
238 * number will be assigned, and the device name will be based
242 else if (st->ss->match_home(st, c->homehost) == 1)
244 else if (st->ss->match_home(st, "any") == 1)
245 trustworthy = LOCAL_ANY;
247 trustworthy = FOREIGN;
249 if (!match && !conf_test_metadata(st->ss->name, policy,
250 (trustworthy == LOCAL))) {
252 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
253 devname, st->ss->name);
256 if (trustworthy == LOCAL_ANY)
259 /* There are three possible sources for 'autof': command line,
260 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
261 * ARRAY takes precedence, then command line, then
264 if (match && match->autof)
265 c->autof = match->autof;
267 c->autof = ci->autof;
269 name_to_use = info.name;
270 if (name_to_use[0] == 0 &&
271 info.array.level == LEVEL_CONTAINER) {
272 name_to_use = info.text_version;
273 trustworthy = METADATA;
275 if (name_to_use[0] && trustworthy != LOCAL &&
276 ! c->require_homehost &&
277 conf_name_is_free(name_to_use))
280 /* strip "hostname:" prefix from name if we have decided
281 * to treat it as LOCAL
283 if (trustworthy == LOCAL && strchr(name_to_use, ':') != NULL)
284 name_to_use = strchr(name_to_use, ':')+1;
286 /* 4/ Check if array exists.
289 pr_err("failed to get exclusive lock on mapfile\n");
290 /* Now check we can get O_EXCL. If not, probably "mdadm -A" has
293 dfd = dev_open(devname, O_RDONLY|O_EXCL);
296 pr_err("cannot reopen %s: %s.\n",
297 devname, strerror(errno));
300 /* Cannot hold it open while we add the device to the array,
301 * so we must release the O_EXCL and depend on the map_lock()
302 * So now is the best time to remove any partitions.
304 remove_partitions(dfd);
308 mp = map_by_uuid(&map, info.uuid);
310 mdfd = open_dev(mp->devnm);
316 /* Skip the clustered ones. This should be started by
317 * clustering resource agents
319 if (info.array.state & (1 << MD_SB_CLUSTERED))
322 /* Couldn't find an existing array, maybe make a new one */
323 mdfd = create_mddev(match ? match->devname : NULL,
324 name_to_use, c->autof, trustworthy, chosen_name);
329 sysfs_init(&info, mdfd, NULL);
331 if (set_array_info(mdfd, st, &info) != 0) {
332 pr_err("failed to set array info for %s: %s\n",
333 chosen_name, strerror(errno));
339 dinfo.disk.major = major(stb.st_rdev);
340 dinfo.disk.minor = minor(stb.st_rdev);
341 if (add_disk(mdfd, st, &info, &dinfo) != 0) {
342 pr_err("failed to add %s to new array %s: %s.\n",
343 devname, chosen_name, strerror(errno));
344 ioctl(mdfd, STOP_ARRAY, 0);
348 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
349 GET_OFFSET | GET_SIZE));
351 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
352 /* It really should be 'none' - must be old buggy
353 * kernel, and mdadm -I may not be able to complete.
356 ioctl(mdfd, STOP_ARRAY, NULL);
357 pr_err("You have an old buggy kernel which cannot support\n --incremental reliably. Aborting.\n");
361 info.array.working_disks = 1;
362 /* 6/ Make sure /var/run/mdadm.map contains this array. */
363 map_update(&map, fd2devnm(mdfd),
365 info.uuid, chosen_name);
368 /* - check one drive in array to make sure metadata is a reasonably */
369 /* close match. Reject if not (e.g. different type) */
370 /* - add the device */
374 struct supertype *st2;
375 struct mdinfo info2, *d;
377 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
378 GET_OFFSET | GET_SIZE));
381 strcpy(chosen_name, mp->path);
383 strcpy(chosen_name, mp->devnm);
385 /* It is generally not OK to add non-spare drives to a
386 * running array as they are probably missing because
387 * they failed. However if runstop is 1, then the
388 * array was possibly started early and our best bet is
389 * to add this anyway.
390 * Also if action policy is re-add or better we allow
392 * This doesn't apply to containers as the 'non-spare'
393 * flag has a different meaning. The test has to happen
394 * at the device level there
396 if (!st->ss->external
397 && (info.disk.state & (1<<MD_DISK_SYNC)) != 0
398 && ! policy_action_allows(policy, st->ss->name,
401 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
402 pr_err("not adding %s to active array (without --run) %s\n",
403 devname, chosen_name);
413 sprintf(dn, "%d:%d", sra->devs->disk.major,
414 sra->devs->disk.minor);
415 dfd2 = dev_open(dn, O_RDONLY);
417 pr_err("unable to open %s\n", devname);
422 if (st2->ss->load_super(st2, dfd2, NULL) ||
423 st->ss->compare_super(st, st2) != 0) {
424 pr_err("metadata mismatch between %s and chosen array %s\n",
425 devname, chosen_name);
431 st2->ss->getinfo_super(st2, &info2, NULL);
432 st2->ss->free_super(st2);
433 if (info.array.level != info2.array.level ||
434 memcmp(info.uuid, info2.uuid, 16) != 0 ||
435 info.array.raid_disks != info2.array.raid_disks) {
436 pr_err("unexpected difference between %s and %s.\n",
437 chosen_name, devname);
442 info.disk.major = major(stb.st_rdev);
443 info.disk.minor = minor(stb.st_rdev);
444 /* add disk needs to know about containers */
445 if (st->ss->external)
446 sra->array.level = LEVEL_CONTAINER;
448 if (info.array.state & (1 << MD_SB_CLUSTERED))
449 info.disk.state |= (1 << MD_DISK_CLUSTER_ADD);
451 err = add_disk(mdfd, st, sra, &info);
452 if (err < 0 && errno == EBUSY) {
453 /* could be another device present with the same
454 * disk.number. Find and reject any such
456 find_reject(mdfd, st, sra, info.disk.number,
457 info.events, c->verbose, chosen_name);
458 err = add_disk(mdfd, st, sra, &info);
460 if (err < 0 && errno == EINVAL &&
461 info.disk.state & (1<<MD_DISK_SYNC)) {
462 /* Maybe it needs to be added as a spare */
463 if (policy_action_allows(policy, st->ss->name,
465 info.disk.state &= ~(1<<MD_DISK_SYNC);
466 err = add_disk(mdfd, st, sra, &info);
469 pr_err("can only add %s to %s as a spare, and force-spare is not set.\n",
470 devname, chosen_name);
473 pr_err("failed to add %s to existing array %s: %s.\n",
474 devname, chosen_name, strerror(errno));
478 info.array.working_disks = 0;
479 for (d = sra->devs; d; d=d->next)
480 info.array.working_disks ++;
483 if (strncmp(chosen_name, "/dev/md/", 8) == 0)
484 md_devname = chosen_name+8;
486 md_devname = chosen_name;
488 printf("MD_DEVICE=%s\n", fd2devnm(mdfd));
489 printf("MD_DEVNAME=%s\n", md_devname);
490 printf("MD_FOREIGN=%s\n", trustworthy == FOREIGN ? "yes" : "no");
493 /* 7/ Is there enough devices to possibly start the array? */
494 /* 7a/ if not, finish with success. */
495 if (info.array.level == LEVEL_CONTAINER) {
497 /* Try to assemble within the container */
498 sysfs_uevent(sra, "change");
499 if (!c->export && c->verbose >= 0)
500 pr_err("container %s now has %d device%s\n",
501 chosen_name, info.array.working_disks,
502 info.array.working_disks == 1?"":"s");
503 wait_for(chosen_name, mdfd);
504 if (st->ss->external)
505 strcpy(devnm, fd2devnm(mdfd));
506 if (st->ss->load_container)
507 rv = st->ss->load_container(st, mdfd, NULL);
511 rv = Incremental_container(st, chosen_name, c, NULL);
513 /* after spare is added, ping monitor for external metadata
514 * so that it can eg. try to rebuild degraded array */
515 if (st->ss->external)
520 /* We have added something to the array, so need to re-read the
521 * state. Eventually this state should be kept up-to-date as
525 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
526 GET_OFFSET | GET_SIZE));
527 active_disks = count_active(st, sra, mdfd, &avail, &info);
529 journal_device_missing = (info.journal_device_required) && (info.journal_clean == 0);
531 if (enough(info.array.level, info.array.raid_disks,
532 info.array.layout, info.array.state & 1,
535 printf("MD_STARTED=no\n");
536 } else if (c->verbose >= 0)
537 pr_err("%s attached to %s, not enough to start (%d).\n",
538 devname, chosen_name, active_disks);
544 /* - if number of OK devices match expected, or -R and there */
546 /* + add any bitmap file */
547 /* + start the array (auto-readonly). */
549 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
551 printf("MD_STARTED=already\n");
552 } else if (c->verbose >= 0)
553 pr_err("%s attached to %s which is already active.\n",
554 devname, chosen_name);
560 if (c->runstop > 0 || (!journal_device_missing && active_disks >= info.array.working_disks)) {
562 /* Let's try to start it */
564 if (journal_device_missing)
565 pr_err("Trying to run with missing journal device\n");
566 if (info.reshape_active && !(info.reshape_active & RESHAPE_NO_BACKUP)) {
567 pr_err("%s: This array is being reshaped and cannot be started\n",
569 cont_err("by --incremental. Please use --assemble\n");
572 if (match && match->bitmap_file) {
573 int bmfd = open(match->bitmap_file, O_RDWR);
575 pr_err("Could not open bitmap file %s.\n",
579 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
581 pr_err("Failed to set bitmapfile for %s.\n",
587 /* Need to remove from the array any devices which
588 * 'count_active' discerned were too old or inappropriate
590 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
591 if (d->disk.state & (1<<MD_DISK_REMOVED))
592 remove_disk(mdfd, st, sra, d);
594 if ((sra == NULL || active_disks >= info.array.working_disks)
595 && trustworthy != FOREIGN)
596 rv = ioctl(mdfd, RUN_ARRAY, NULL);
598 rv = sysfs_set_str(sra, NULL,
599 "array_state", "read-auto");
600 /* Array might be O_EXCL which will interfere with
601 * fsck and mount. So re-open without O_EXCL.
606 printf("MD_STARTED=yes\n");
607 } else if (c->verbose >= 0)
608 pr_err("%s attached to %s, which has been started.\n",
609 devname, chosen_name);
611 wait_for(chosen_name, mdfd);
612 /* We just started the array, so some devices
613 * might have been evicted from the array
614 * because their event counts were too old.
615 * If the action=re-add policy is in-force for
616 * those devices we should re-add them now.
618 for (dsk = sra->devs; dsk ; dsk = dsk->next) {
619 if (disk_action_allows(dsk, st->ss->name, act_re_add) &&
620 add_disk(mdfd, st, sra, dsk) == 0)
621 pr_err("%s re-added to %s\n",
622 dsk->sys_name, chosen_name);
625 pr_err("%s attached to %s, but failed to start: %s.\n",
626 devname, chosen_name, strerror(errno));
631 printf("MD_STARTED=unsafe\n");
632 } else if (journal_device_missing) {
633 pr_err("Journal device is missing, not safe to start yet.\n");
634 } else if (c->verbose >= 0)
635 pr_err("%s attached to %s, not enough to start safely.\n",
636 devname, chosen_name);
646 dev_policy_free(policy);
654 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
655 int number, __u64 events, int verbose,
658 /* Find a device attached to this array with a disk.number of number
659 * and events less than the passed events, and remove the device.
664 if (ioctl(mdfd, GET_ARRAY_INFO, &ra) == 0)
665 return; /* not safe to remove from active arrays
666 * without thinking more */
668 for (d = sra->devs; d ; d = d->next) {
669 char dn[24]; // 2*11 bytes for ints (including sign) + colon + null byte
672 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
673 dfd = dev_open(dn, O_RDONLY);
676 if (st->ss->load_super(st, dfd, NULL)) {
680 st->ss->getinfo_super(st, &info, NULL);
681 st->ss->free_super(st);
684 if (info.disk.number != number ||
685 info.events >= events)
688 if (d->disk.raid_disk > -1)
689 sysfs_set_str(sra, d, "slot", "none");
690 if (sysfs_set_str(sra, d, "state", "remove") == 0)
692 pr_err("removing old device %s from %s\n",
693 d->sys_name+4, array_name);
697 static int count_active(struct supertype *st, struct mdinfo *sra,
698 int mdfd, char **availp,
699 struct mdinfo *bestinfo)
701 /* count how many devices in sra think they are active */
705 __u64 max_events = 0;
717 for (d = sra->devs ; d ; d = d->next)
719 for (d = sra->devs, devnum = 0 ; d ; d = d->next, devnum++) {
725 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
726 dfd = dev_open(dn, O_RDONLY);
729 ok = st->ss->load_super(st, dfd, NULL);
734 info.array.raid_disks = raid_disks;
735 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
736 if (info.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
737 bestinfo->journal_clean = 1;
739 raid_disks = info.array.raid_disks;
740 avail = xcalloc(raid_disks, 1);
743 best = xcalloc(raid_disks, sizeof(int));
744 devmap = xcalloc(raid_disks, numdevs);
746 st->ss->getinfo_super(st, &info, devmap);
749 if (info.disk.state & (1<<MD_DISK_SYNC))
753 max_events = info.events;
754 avail[info.disk.raid_disk] = 2;
755 best[info.disk.raid_disk] = devnum;
756 st->ss->getinfo_super(st, bestinfo, NULL);
757 } else if (info.events == max_events) {
758 avail[info.disk.raid_disk] = 2;
759 best[info.disk.raid_disk] = devnum;
760 } else if (info.events == max_events-1) {
761 if (avail[info.disk.raid_disk] == 0) {
762 avail[info.disk.raid_disk] = 1;
763 best[info.disk.raid_disk] = devnum;
765 } else if (info.events < max_events - 1)
767 else if (info.events == max_events+1) {
769 max_events = info.events;
770 for (i = 0; i < raid_disks; i++)
773 avail[info.disk.raid_disk] = 2;
774 best[info.disk.raid_disk] = devnum;
775 st->ss->getinfo_super(st, bestinfo, NULL);
776 } else { /* info.events much bigger */
777 memset(avail, 0, raid_disks);
778 max_events = info.events;
779 avail[info.disk.raid_disk] = 2;
780 best[info.disk.raid_disk] = devnum;
781 st->ss->getinfo_super(st, bestinfo, NULL);
783 } else if (info.disk.state & (1<<MD_DISK_REPLACEMENT))
785 st->ss->free_super(st);
790 /* We need to reject any device that thinks the best device is
791 * failed or missing */
792 for (b = 0; b < raid_disks; b++)
796 for (i = 0 ; i < raid_disks ; i++) {
797 if (i != b && avail[i])
798 if (devmap[raid_disks * best[i] + b] == 0) {
799 /* This device thinks 'b' is failed -
802 for (d=sra->devs ; devnum; d = d->next)
804 d->disk.state |= (1 << MD_DISK_REMOVED);
810 /* Also need to reject any spare device with an event count that
813 for (d = sra->devs; d; d = d->next) {
814 if (!(d->disk.state & (1<<MD_DISK_SYNC)) &&
815 d->events > max_events)
816 d->disk.state |= (1 << MD_DISK_REMOVED);
820 return cnt + replcnt;
823 /* test if container has degraded member(s) */
824 static int container_members_max_degradation(struct map_ent *map, struct map_ent *me)
826 mdu_array_info_t array;
828 int max_degraded = 0;
830 for(; map; map = map->next) {
831 if (!metadata_container_matches(map->metadata, me->devnm))
833 afd = open_dev(map->devnm);
836 /* most accurate information regarding array degradation */
837 if (ioctl(afd, GET_ARRAY_INFO, &array) >= 0) {
838 int degraded = array.raid_disks - array.active_disks -
840 if (degraded > max_degraded)
841 max_degraded = degraded;
845 return (max_degraded);
848 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
849 struct map_ent *target, int bare,
850 struct supertype *st, int verbose)
852 /* This device doesn't have any md metadata
853 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
854 * If 'st' is not set, then we only know that some metadata allows this,
855 * others possibly don't.
856 * So look for a container or array to attach the device to.
857 * Prefer 'target' if that is set and the array is found.
859 * If st is set, then only arrays of that type are considered
860 * Return 0 on success, or some exit code on failure, probably 1.
864 struct map_ent *mp, *map = NULL;
865 struct mdinfo *chosen = NULL;
868 if (fstat(dfd, &stb) != 0)
872 * Now we need to find a suitable array to add this to.
873 * We only accept arrays that:
875 * - are in the same domains as the device
876 * - are of an size for which the device will be useful
877 * and we choose the one that is the most degraded
880 if (map_lock(&map)) {
881 pr_err("failed to get exclusive lock on mapfile\n");
884 for (mp = map ; mp ; mp = mp->next) {
885 struct supertype *st2;
886 struct domainlist *dl = NULL;
888 unsigned long long devsize;
889 unsigned long long component_size = 0;
891 if (is_subarray(mp->metadata))
894 st2 = st->ss->match_metadata_desc(mp->metadata);
896 (st->minor_version >= 0 &&
897 st->minor_version != st2->minor_version)) {
899 pr_err("not adding %s to %s as metadata type doesn't match\n",
906 sra = sysfs_read(-1, mp->devnm,
907 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
908 GET_DEGRADED|GET_COMPONENT|GET_VERSION);
910 /* Probably a container - no degraded info */
911 sra = sysfs_read(-1, mp->devnm,
912 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
913 GET_COMPONENT|GET_VERSION);
915 sra->array.failed_disks = -1;
922 for(i = 0; !st2 && superlist[i]; i++)
923 st2 = superlist[i]->match_metadata_desc(
927 pr_err("not adding %s to %s as metadata not recognised.\n",
931 /* Need to double check the 'act_spare' permissions applies
934 if (!policy_action_allows(pol, st2->ss->name, act_spare))
936 if (!bare && !policy_action_allows(pol, st2->ss->name,
937 act_spare_same_slot))
941 /* update number of failed disks for mostly degraded
942 * container member */
943 if (sra->array.failed_disks == -1)
944 sra->array.failed_disks = container_members_max_degradation(map, mp);
946 get_dev_size(dfd, NULL, &devsize);
947 if (sra->component_size == 0) {
948 /* true for containers, here we must read superblock
949 * to obtain minimum spare size */
950 struct supertype *st3 = dup_super(st2);
951 int mdfd = open_dev(mp->devnm);
956 if (st3->ss->load_container &&
957 !st3->ss->load_container(st3, mdfd, mp->path)) {
958 component_size = st3->ss->min_acceptable_spare_size(st3);
959 st3->ss->free_super(st3);
964 if ((sra->component_size > 0 &&
965 st2->ss->avail_size(st2, devsize,
967 ? sra->devs->data_offset
969 < sra->component_size)
971 (sra->component_size == 0 && devsize < component_size)) {
973 pr_err("not adding %s to %s as it is too small\n",
977 /* test against target.
978 * If 'target' is set and 'bare' is false, we only accept
979 * arrays/containers that match 'target'.
980 * If 'target' is set and 'bare' is true, we prefer the
981 * array which matches 'target'.
982 * target is considered only if we deal with degraded array
984 if (target && policy_action_allows(pol, st2->ss->name,
985 act_spare_same_slot)) {
986 if (strcmp(target->metadata, mp->metadata) == 0 &&
987 memcmp(target->uuid, mp->uuid,
988 sizeof(target->uuid)) == 0 &&
989 sra->array.failed_disks > 0) {
990 /* This is our target!! */
994 /* skip to end so we don't check any more */
1004 dl = domain_from_array(sra, st2->ss->name);
1005 if (domain_test(dl, pol, st2->ss->name) != 1) {
1006 /* domain test fails */
1008 pr_err("not adding %s to %s as it is not in a compatible domain\n",
1013 /* all tests passed, OK to add to this array */
1017 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
1030 /* add current device to chosen array as a spare */
1031 int mdfd = open_dev(chosen->sys_name);
1033 struct mddev_dev devlist;
1034 char chosen_devname[24]; // 2*11 for int (including signs) + colon + null
1035 devlist.next = NULL;
1037 devlist.writemostly = FlagDefault;
1038 devlist.failfast = FlagDefault;
1039 devlist.devname = chosen_devname;
1040 sprintf(chosen_devname, "%d:%d", major(stb.st_rdev),
1041 minor(stb.st_rdev));
1042 devlist.disposition = 'a';
1045 rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
1051 pr_err("added %s as spare for %s\n",
1052 devname, chosen->sys_name);
1054 pr_err("failed to add %s as spare for %s\n",
1055 devname, chosen->sys_name);
1063 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1064 struct supertype *st, int verbose)
1066 /* we know that at least one partition virtual-metadata is
1067 * allowed to incorporate spares like this device. We need to
1068 * find a suitable device to copy partition information from.
1070 * Getting a list of all disk (not partition) devices is
1071 * slightly non-trivial. We could look at /sys/block, but
1072 * that is theoretically due to be removed. Maybe best to use
1073 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1074 * we depend on this directory of 'path' info. But that fails
1075 * to find loop devices and probably others. Maybe don't
1076 * worry about that, they aren't the real target.
1078 * So: check things in /dev/disk/by-path to see if they are in
1079 * a compatible domain, then load the partition table and see
1080 * if it is OK for the new device, and choose the largest
1081 * partition table that fits.
1085 char *chosen = NULL;
1086 unsigned long long chosen_size = 0;
1087 struct supertype *chosen_st = NULL;
1090 dir = opendir("/dev/disk/by-path");
1093 while ((de = readdir(dir)) != NULL) {
1095 struct dev_policy *pol2 = NULL;
1096 struct domainlist *domlist = NULL;
1099 struct supertype *st2 = NULL;
1100 char *devname = NULL;
1101 unsigned long long devsectors;
1103 if (de->d_ino == 0 ||
1104 de->d_name[0] == '.' ||
1105 (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1108 ep = de->d_name + strlen(de->d_name);
1109 while (ep > de->d_name &&
1112 if (ep > de->d_name + 5 &&
1113 strncmp(ep-5, "-part", 5) == 0)
1114 /* This is a partition - skip it */
1117 pol2 = path_policy(de->d_name, type_disk);
1119 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1120 if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
1121 /* new device is incompatible with this device. */
1124 domain_free(domlist);
1127 if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
1131 fd = open(devname, O_RDONLY);
1134 if (get_dev_size(fd, devname, &devsectors) == 0)
1139 st2 = dup_super(st);
1141 st2 = guess_super_type(fd, guess_partitions);
1143 st2->ss->load_super(st2, fd, NULL) < 0)
1145 st2->ignore_hw_compat = 0;
1148 /* Check domain policy again, this time referring to metadata */
1149 domain_merge(&domlist, pol2, st2->ss->name);
1150 if (domain_test(domlist, pol, st2->ss->name) != 1)
1151 /* Incompatible devices for this metadata type */
1153 if (!policy_action_allows(pol, st2->ss->name, act_spare))
1154 /* Some partition types allow sparing, but not
1160 st2->ss->getinfo_super(st2, &info, NULL);
1161 if (info.component_size > devsectors)
1162 /* This partitioning doesn't fit in the device */
1165 /* This is an acceptable device to copy partition
1166 * metadata from. We could just stop here, but I
1167 * think I want to keep looking incase a larger
1168 * metadata which makes better use of the device can
1171 if (chosen == NULL ||
1172 chosen_size < info.component_size) {
1173 chosen_size = info.component_size;
1178 chosen_st->ss->free_super(chosen_st);
1187 domain_free(domlist);
1188 dev_policy_free(pol2);
1190 st2->ss->free_super(st2);
1202 /* 'chosen' is the best device we can find. Let's write its
1203 * metadata to devname dfd is read-only so don't use that
1205 fd = open(devname, O_RDWR);
1207 chosen_st->ss->store_super(chosen_st, fd);
1211 chosen_st->ss->free_super(chosen_st);
1216 static int is_bare(int dfd)
1218 unsigned long long size = 0;
1219 char bufpad[4096 + 4096];
1220 char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1222 if (lseek(dfd, 0, SEEK_SET) != 0 ||
1223 read(dfd, buf, 4096) != 4096)
1226 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1228 if (memcmp(buf, buf+1, 4095) != 0)
1231 /* OK, first 4K appear blank, try the end. */
1232 get_dev_size(dfd, NULL, &size);
1233 if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1234 read(dfd, buf, 4096) != 4096)
1237 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1239 if (memcmp(buf, buf+1, 4095) != 0)
1245 /* adding a spare to a regular array is quite different from adding one to
1246 * a set-of-partitions virtual array.
1247 * This function determines which is worth trying and tries as appropriate.
1248 * Arrays are given priority over partitions.
1250 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1251 struct map_ent *target,
1252 struct supertype *st, int verbose)
1257 int partitions_ok = 0;
1261 /* Can only add a spare if device has at least one domain */
1262 if (pol_find(pol, pol_domain) == NULL)
1264 /* And only if some action allows spares */
1265 if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1268 /* Now check if the device is bare.
1269 * bare devices can always be added as a spare
1270 * non-bare devices can only be added if spare-same-slot is permitted,
1271 * and this device is replacing a previous device - in which case 'target'
1274 if (!is_bare(dfd)) {
1275 /* Must have a target and allow same_slot */
1276 /* Later - may allow force_spare without target */
1278 !policy_action_allows(pol, st?st->ss->name:NULL,
1279 act_spare_same_slot)) {
1281 pr_err("%s is not bare, so not considering as a spare\n",
1289 /* It might be OK to add this device to an array - need to see
1290 * what arrays might be candidates.
1293 /* just try try 'array' or 'partition' based on this metadata */
1294 if (st->ss->add_to_super)
1295 return array_try_spare(devname, dfdp, pol, target, bare,
1298 return partition_try_spare(devname, dfdp, pol,
1301 /* No metadata was specified or found so options are open.
1302 * Check for whether any array metadata, or any partition metadata
1303 * might allow adding the spare. This check is just help to avoid
1304 * a more costly scan of all arrays when we can be sure that will
1307 for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1308 if (superlist[i]->add_to_super && !arrays_ok &&
1309 policy_action_allows(pol, superlist[i]->name, act_spare))
1311 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1312 policy_action_allows(pol, superlist[i]->name, act_spare))
1317 rv = array_try_spare(devname, dfdp, pol, target, bare,
1319 if (rv != 0 && partitions_ok)
1320 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1324 int IncrementalScan(struct context *c, char *devnm)
1326 /* look at every device listed in the 'map' file.
1327 * If one is found that is not running then:
1328 * look in mdadm.conf for bitmap file.
1329 * if one exists, but array has none, add it.
1330 * try to start array in auto-readonly mode
1332 struct map_ent *mapl = NULL;
1334 struct mddev_ident *devs, *mddev;
1340 devs = conf_get_ident(NULL);
1343 for (me = mapl ; me ; me = me->next) {
1344 mdu_array_info_t array;
1345 mdu_bitmap_file_t bmf;
1349 if (devnm && strcmp(devnm, me->devnm) != 0)
1351 if (me->metadata[0] == '/') {
1357 /* member array, need to work on container */
1358 strncpy(container, me->metadata+1, 32);
1360 sl = strchr(container, '/');
1367 mdfd = open_dev(me->devnm);
1371 if (!isdigit(me->metadata[0])) {
1372 /* must be a container */
1373 struct supertype *st = super_by_fd(mdfd, NULL);
1375 struct map_ent *map = NULL;
1377 if (st && st->ss->load_container)
1378 ret = st->ss->load_container(st, mdfd, NULL);
1380 if (!ret && st && st->ss->container_content) {
1382 pr_err("failed to get exclusive lock on mapfile\n");
1383 ret = Incremental_container(st, me->path, c, only);
1390 if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 ||
1395 /* Ok, we can try this one. Maybe it needs a bitmap */
1396 for (mddev = devs ; mddev ; mddev = mddev->next)
1397 if (mddev->devname && me->path
1398 && devname_matches(mddev->devname, me->path))
1400 if (mddev && mddev->bitmap_file) {
1402 * Note: early kernels will wrongly fail this, so it
1406 if (ioctl(mdfd, GET_ARRAY_INFO, &bmf) < 0) {
1407 int bmfd = open(mddev->bitmap_file, O_RDWR);
1409 added = ioctl(mdfd, SET_BITMAP_FILE,
1414 if (c->verbose >= 0) {
1416 pr_err("Added bitmap %s to %s\n",
1417 mddev->bitmap_file, me->path);
1418 else if (errno != EEXIST)
1419 pr_err("Failed to add bitmap to %s: %s\n",
1420 me->path, strerror(errno));
1423 /* FIXME check for reshape_active and consider not
1426 sra = sysfs_read(mdfd, NULL, 0);
1428 if (sysfs_set_str(sra, NULL,
1429 "array_state", "read-auto") == 0) {
1430 if (c->verbose >= 0)
1431 pr_err("started array %s\n",
1432 me->path ?: me->devnm);
1434 pr_err("failed to start array %s: %s\n",
1435 me->path ?: me->devnm,
1445 static char *container2devname(char *devname)
1447 char *mdname = NULL;
1449 if (devname[0] == '/') {
1450 int fd = open(devname, O_RDONLY);
1452 mdname = xstrdup(fd2devnm(fd));
1457 struct map_ent *mp, *map = NULL;
1459 if (!parse_uuid(devname, uuid))
1461 mp = map_by_uuid(&map, uuid);
1463 mdname = xstrdup(mp->devnm);
1470 static int Incremental_container(struct supertype *st, char *devname,
1471 struct context *c, char *only)
1473 /* Collect the contents of this container and for each
1474 * array, choose a device name and assemble the array.
1477 struct mdinfo *list;
1479 struct map_ent *map = NULL;
1482 struct mddev_ident *match;
1484 struct domainlist *domains;
1485 struct map_ent *smp;
1492 st->ss->getinfo_super(st, &info, NULL);
1494 if ((c->runstop > 0 && info.container_enough >= 0) ||
1495 info.container_enough > 0)
1499 printf("MD_STARTED=no\n");
1500 } else if (c->verbose)
1501 pr_err("not enough devices to start the container\n");
1505 match = conf_match(st, &info, devname, c->verbose, &rv);
1506 if (match == NULL && rv == 2)
1509 /* Need to compute 'trustworthy' */
1511 trustworthy = LOCAL;
1512 else if (st->ss->match_home(st, c->homehost) == 1)
1513 trustworthy = LOCAL;
1514 else if (st->ss->match_home(st, "any") == 1)
1515 trustworthy = LOCAL;
1517 trustworthy = FOREIGN;
1519 list = st->ss->container_content(st, NULL);
1520 /* when nothing to activate - quit */
1523 printf("MD_STARTED=nothing\n");
1527 for (ra = list ; ra ; ra = ra->next) {
1529 char chosen_name[1024];
1531 struct mddev_ident *match = NULL;
1534 /* do not activate arrays blocked by metadata handler */
1535 if (ra->array.state & (1 << MD_SB_BLOCK_VOLUME)) {
1536 pr_err("Cannot activate array %s in %s.\n",
1537 ra->text_version, devname);
1541 mp = map_by_uuid(&map, ra->uuid);
1544 mdfd = open_dev(mp->devnm);
1546 strcpy(chosen_name, mp->path);
1548 strcpy(chosen_name, mp->devnm);
1551 /* Check in mdadm.conf for container == devname and
1552 * member == ra->text_version after second slash.
1554 char *sub = strchr(ra->text_version+1, '/');
1555 struct mddev_ident *array_list;
1558 array_list = conf_get_ident(NULL);
1561 for(; array_list ; array_list = array_list->next) {
1563 if (array_list->member == NULL ||
1564 array_list->container == NULL)
1566 if (strcmp(array_list->member, sub) != 0)
1568 if (array_list->uuid_set &&
1569 !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1571 dn = container2devname(array_list->container);
1574 if (strncmp(dn, ra->text_version+1,
1576 ra->text_version[strlen(dn)+1] != '/') {
1581 /* we have a match */
1584 pr_err("match found for member %s\n",
1585 array_list->member);
1589 if (match && match->devname &&
1590 strcasecmp(match->devname, "<ignore>") == 0) {
1592 pr_err("array %s/%s is explicitly ignored by mdadm.conf\n",
1593 match->container, match->member);
1597 trustworthy = LOCAL;
1599 mdfd = create_mddev(match ? match->devname : NULL,
1605 if (only && (!mp || strcmp(mp->devnm, only) != 0))
1609 pr_err("failed to open %s: %s.\n",
1610 chosen_name, strerror(errno));
1614 assemble_container_content(st, mdfd, ra, c,
1615 chosen_name, &result);
1618 if (c->export && result) {
1620 printf("MD_STARTED");
1621 if (result & INCR_NO) {
1622 printf("%cno", sep);
1625 if (result & INCR_UNSAFE) {
1626 printf("%cunsafe", sep);
1629 if (result & INCR_ALREADY) {
1630 printf("%calready", sep);
1633 if (result & INCR_YES) {
1634 printf("%cyes", sep);
1640 /* don't move spares to container with volume being activated
1641 when all volumes are blocked */
1642 if (ra_all == ra_blocked)
1645 /* Now move all suitable spares from spare container */
1646 domains = domain_from_array(list, st->ss->name);
1647 memcpy(suuid, uuid_zero, sizeof(int[4]));
1649 (smp = map_by_uuid(&map, suuid)) != NULL &&
1650 (sfd = open(smp->path, O_RDONLY)) >= 0) {
1651 /* spare container found */
1652 struct supertype *sst =
1653 super_imsm.match_metadata_desc("imsm");
1654 struct mdinfo *sinfo;
1655 unsigned long long min_size = 0;
1656 if (st->ss->min_acceptable_spare_size)
1657 min_size = st->ss->min_acceptable_spare_size(st);
1658 if (!sst->ss->load_container(sst, sfd, NULL)) {
1660 sinfo = container_choose_spares(sst, min_size,
1663 sst->ss->free_super(sst);
1666 struct mdinfo *disks = sinfo->devs;
1668 /* move spare from spare
1669 * container to currently
1675 makedev(disks->disk.major,
1676 disks->disk.minor)))
1678 disks = disks->next;
1681 pr_err("Added %d spare%s to %s\n",
1682 count, count>1?"s":"", devname);
1688 domain_free(domains);
1692 static void run_udisks(char *arg1, char *arg2)
1697 execl("/usr/bin/udisks", "udisks", arg1, arg2, NULL);
1698 execl("/bin/udisks", "udisks", arg1, arg2, NULL);
1701 while (pid > 0 && wait(&status) != pid)
1706 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1707 * raid arrays, and if so first fail (if needed) and then remove the device.
1709 * @devname - The device we want to remove
1710 * @id_path - name as found in /dev/disk/by-path for this device
1712 * Note: the device name must be a kernel name like "sda", so
1713 * that we can find it in /proc/mdstat
1715 int IncrementalRemove(char *devname, char *id_path, int verbose)
1719 struct mdstat_ent *ent;
1720 struct mddev_dev devlist;
1725 dprintf("incremental removal without --path <id_path> lacks the possibility to re-add new device in this port\n");
1727 if (strchr(devname, '/')) {
1728 pr_err("incremental removal requires a kernel device name, not a file: %s\n", devname);
1731 ent = mdstat_by_component(devname);
1734 pr_err("%s does not appear to be a component of any array\n", devname);
1737 sysfs_init(&mdi, -1, ent->devnm);
1738 mdfd = open_dev_excl(ent->devnm);
1741 if (sysfs_get_str(&mdi, NULL, "array_state",
1742 buf, sizeof(buf)) > 0) {
1743 if (strncmp(buf, "active", 6) == 0 ||
1744 strncmp(buf, "clean", 5) == 0)
1745 sysfs_set_str(&mdi, NULL,
1746 "array_state", "read-auto");
1749 mdfd = open_dev(ent->devnm);
1752 pr_err("Cannot open array %s!!\n", ent->devnm);
1758 struct map_ent *map = NULL, *me;
1759 me = map_by_devnm(&map, ent->devnm);
1761 policy_save_path(id_path, me);
1765 memset(&devlist, 0, sizeof(devlist));
1766 devlist.devname = devname;
1767 devlist.disposition = 'f';
1768 /* for a container, we must fail each member array */
1769 if (ent->metadata_version &&
1770 strncmp(ent->metadata_version, "external:", 9) == 0) {
1771 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1772 struct mdstat_ent *memb;
1773 for (memb = mdstat ; memb ; memb = memb->next)
1774 if (is_container_member(memb, ent->devnm)) {
1775 int subfd = open_dev(memb->devnm);
1777 rv |= Manage_subdevs(
1779 &devlist, verbose, 0,
1784 free_mdstat(mdstat);
1786 rv |= Manage_subdevs(ent->devnm, mdfd, &devlist,
1787 verbose, 0, NULL, 0);
1789 /* Failed due to EBUSY, try to stop the array.
1790 * Give udisks a chance to unmount it first.
1792 int devid = devnm2devid(ent->devnm);
1793 run_udisks("--unmount", map_dev(major(devid),minor(devid), 0));
1794 rv = Manage_stop(ent->devnm, mdfd, verbose, 1);
1796 /* At least we can try to trigger a 'remove' */
1797 sysfs_uevent(&mdi, "remove");
1800 pr_err("Fail to stop %s too.\n", ent->devnm);
1803 devlist.disposition = 'r';
1804 rv = Manage_subdevs(ent->devnm, mdfd, &devlist,
1805 verbose, 0, NULL, 0);