]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Assemble.c
Provide a mdstat_ent to subarray helper
[thirdparty/mdadm.git] / Assemble.c
CommitLineData
64c4757e 1/*
9a9dab36 2 * mdadm - manage Linux "md" devices aka RAID arrays.
64c4757e 3 *
e736b623 4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
64c4757e
NB
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
e736b623 22 * Email: <neilb@suse.de>
64c4757e
NB
23 */
24
9a9dab36 25#include "mdadm.h"
41a3b72a 26#include <ctype.h>
64c4757e 27
624920bb
NB
28static int name_matches(char *found, char *required, char *homehost)
29{
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
32 */
33 char fnd[33];
34
35 strncpy(fnd, found, 32);
36 fnd[32] = 0;
37 if (strcmp(found, required)==0)
38 return 1;
39 if (homehost) {
40 int l = strlen(homehost);
41 if (l < 32 && fnd[l] == ':' &&
42 strcmp(fnd+l+1, required)==0)
43 return 1;
44 }
45 return 0;
46}
47
9008ed1c 48static int is_member_busy(char *metadata_version)
2de8884f
DW
49{
50 /* check if the given member array is active */
51 struct mdstat_ent *mdstat = mdstat_read(1, 0);
52 struct mdstat_ent *ent;
53 int busy = 0;
54
55 for (ent = mdstat; ent; ent = ent->next) {
56 if (ent->metadata_version == NULL)
57 continue;
58 if (strncmp(ent->metadata_version, "external:", 9) != 0)
59 continue;
60 if (!is_subarray(&ent->metadata_version[9]))
61 continue;
62 /* Skip first char - it can be '/' or '-' */
63 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
64 busy = 1;
65 break;
66 }
67 }
68 free_mdstat(mdstat);
69
70 return busy;
71}
72
fa56eddb 73static int ident_matches(struct mddev_ident *ident,
08fb91a3
N
74 struct mdinfo *content,
75 struct supertype *tst,
76 char *homehost,
77 char *update, char *devname)
78{
79
80 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
81 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0) {
82 if (devname)
83 fprintf(stderr, Name ": %s has wrong uuid.\n",
84 devname);
85 return 0;
86 }
87 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
88 name_matches(content->name, ident->name, homehost)==0) {
89 if (devname)
90 fprintf(stderr, Name ": %s has wrong name.\n",
91 devname);
92 return 0;
93 }
94 if (ident->super_minor != UnSet &&
95 ident->super_minor != content->array.md_minor) {
96 if (devname)
97 fprintf(stderr, Name ": %s has wrong super-minor.\n",
98 devname);
99 return 0;
100 }
101 if (ident->level != UnSet &&
102 ident->level != content->array.level) {
103 if (devname)
104 fprintf(stderr, Name ": %s has wrong raid level.\n",
105 devname);
106 return 0;
107 }
108 if (ident->raid_disks != UnSet &&
109 ident->raid_disks!= content->array.raid_disks) {
110 if (devname)
111 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
112 devname);
113 return 0;
114 }
805d30b2
N
115 if (ident->member && ident->member[0]) {
116 /* content->text_version must match */
117 char *s = strchr(content->text_version+1, '/');
118 if (s == NULL) {
119 if (devname)
120 fprintf(stderr, Name ": %s is not a container and one is required.\n",
121 devname);
122 return 0;
123 } else if (strcmp(ident->member, s+1) != 0) {
124 if (devname)
125 fprintf(stderr, Name ": skipping wrong member %s is %s\n",
126 content->text_version, devname);
127 return 0;
128 }
129 }
08fb91a3
N
130 return 1;
131}
132
133
7f91af49 134int Assemble(struct supertype *st, char *mddev,
fa56eddb 135 struct mddev_ident *ident,
a655e550 136 struct mddev_dev *devlist, char *backup_file,
64c4757e 137 int readonly, int runstop,
0ac91628 138 char *update, char *homehost, int require_homehost,
64c4757e
NB
139 int verbose, int force)
140{
141 /*
52826846
NB
142 * The task of Assemble is to find a collection of
143 * devices that should (according to their superblocks)
144 * form an array, and to give this collection to the MD driver.
145 * In Linux-2.4 and later, this involves submitting a
64c4757e
NB
146 * SET_ARRAY_INFO ioctl with no arg - to prepare
147 * the array - and then submit a number of
148 * ADD_NEW_DISK ioctls to add disks into
149 * the array. Finally RUN_ARRAY might
150 * be submitted to start the array.
151 *
152 * Much of the work of Assemble is in finding and/or
153 * checking the disks to make sure they look right.
154 *
4b1ac34b 155 * If mddev is not set, then scan must be set and we
64c4757e
NB
156 * read through the config file for dev+uuid mapping
157 * We recurse, setting mddev, for each device that
158 * - isn't running
4b1ac34b 159 * - has a valid uuid (or any uuid if !uuidset)
64c4757e
NB
160 *
161 * If mddev is set, we try to determine state of md.
162 * check version - must be at least 0.90.0
163 * check kernel version. must be at least 2.4.
164 * If not, we can possibly fall back on START_ARRAY
165 * Try to GET_ARRAY_INFO.
166 * If possible, give up
167 * If not, try to STOP_ARRAY just to make sure
168 *
169 * If !uuidset and scan, look in conf-file for uuid
170 * If not found, give up
b8a8ccf9 171 * If !devlist and scan and uuidset, get list of devs from conf-file
64c4757e
NB
172 *
173 * For each device:
174 * Check superblock - discard if bad
175 * Check uuid (set if we don't have one) - discard if no match
5787fa49 176 * Check superblock similarity if we have a superblock - discard if different
4b1ac34b 177 * Record events, devicenum
64c4757e 178 * This should give us a list of devices for the array
4b1ac34b 179 * We should collect the most recent event number
64c4757e
NB
180 *
181 * Count disks with recent enough event count
182 * While force && !enough disks
183 * Choose newest rejected disks, update event count
184 * mark clean and rewrite superblock
185 * If recent kernel:
186 * SET_ARRAY_INFO
187 * foreach device with recent events : ADD_NEW_DISK
188 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
189 * If old kernel:
190 * Check the device numbers in superblock are right
191 * update superblock if any changes
192 * START_ARRAY
193 *
194 */
c30e5369
N
195 int mdfd;
196 int clean;
f05641cf
N
197 int auto_assem = (mddev == NULL && !ident->uuid_set &&
198 ident->super_minor == UnSet && ident->name[0] == 0
aa7c284c 199 && (ident->container == NULL || ident->member == NULL));
64c4757e 200 int old_linux = 0;
c30e5369 201 int vers = vers; /* Keep gcc quite - it really is initialised */
64c4757e
NB
202 struct {
203 char *devname;
213ee40b
NB
204 int uptodate; /* set once we decide that this device is as
205 * recent as everything else in the array.
206 */
207 struct mdinfo i;
5787fa49 208 } *devices;
02e7c5b7 209 char *devmap;
56eedc1a 210 int *best = NULL; /* indexed by raid_disk */
f21e18ca 211 int bestcnt = 0;
98c6faba 212 int devcnt = 0;
1ff98339 213 unsigned int okcnt, sparecnt, rebuilding_cnt;
98c6faba 214 unsigned int req_cnt;
f21e18ca 215 int i;
64c4757e 216 int most_recent = 0;
cd29a5c8 217 int chosen_drive;
52826846 218 int change = 0;
cd29a5c8 219 int inargv = 0;
52437b4f 220 int report_missmatch;
bf4fb153 221 int bitmap_done;
7f91af49
N
222 int start_partial_ok = (runstop >= 0) &&
223 (force || devlist==NULL || auto_assem);
98c6faba 224 unsigned int num_devs;
a655e550 225 struct mddev_dev *tmpdev;
4b1ac34b 226 struct mdinfo info;
98dbd966 227 struct mdinfo *content = NULL;
265e0f17 228 char *avail;
308e1801 229 int nextspare = 0;
197e3eb6 230 char *name = NULL;
69207ff6 231 int trustworthy;
a04d5763 232 char chosen_name[1024];
8a46fe84 233
682c7051 234 if (get_linux_version() < 2004000)
64c4757e
NB
235 old_linux = 1;
236
64c4757e 237 /*
52826846
NB
238 * If any subdevs are listed, then any that don't
239 * match ident are discarded. Remainder must all match and
240 * become the array.
241 * If no subdevs, then we scan all devices in the config file, but
242 * there must be something in the identity
64c4757e 243 */
64c4757e 244
cd29a5c8 245 if (!devlist &&
52826846 246 ident->uuid_set == 0 &&
f21e18ca 247 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
e0fe762a
N
248 ident->name[0] == 0 &&
249 (ident->container == NULL || ident->member == NULL) &&
52826846
NB
250 ident->devices == NULL) {
251 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
8a46fe84 252 mddev ? mddev : "further assembly");
52826846 253 return 1;
64c4757e 254 }
71d60c48 255
cd29a5c8 256 if (devlist == NULL)
8aec876d 257 devlist = conf_get_devs();
7f91af49 258 else if (mddev)
da6b5ca9 259 inargv = 1;
64c4757e 260
52437b4f 261 report_missmatch = ((inargv && verbose >= 0) || verbose > 0);
60e1bc1a 262 try_again:
c30e5369
N
263 /* We come back here when doing auto-assembly and attempting some
264 * set of devices failed. Those are now marked as ->used==2 and
265 * we ignore them and try again
266 */
60e1bc1a 267
5787fa49
NB
268 tmpdev = devlist; num_devs = 0;
269 while (tmpdev) {
da6b5ca9
NB
270 if (tmpdev->used)
271 tmpdev->used = 2;
272 else
273 num_devs++;
5787fa49
NB
274 tmpdev = tmpdev->next;
275 }
5787fa49 276
82d9eba6 277 if (!st && ident->st) st = ident->st;
64c4757e 278
dab6685f 279 if (verbose>0)
82b27616 280 fprintf(stderr, Name ": looking for devices for %s\n",
8a46fe84 281 mddev ? mddev : "further assembly");
82b27616 282
811e6cbe
NB
283 /* first walk the list of devices to find a consistent set
284 * that match the criterea, if that is possible.
c30e5369 285 * We flag the ones we like with 'used'.
811e6cbe
NB
286 */
287 for (tmpdev = devlist;
288 tmpdev;
5083d66b 289 tmpdev = tmpdev ? tmpdev->next : NULL) {
811e6cbe 290 char *devname = tmpdev->devname;
64c4757e
NB
291 int dfd;
292 struct stat stb;
3da92f27 293 struct supertype *tst = dup_super(st);
4e8d9f0a 294 struct dev_policy *pol = NULL;
5083d66b 295 int found_container = 0;
52826846 296
da6b5ca9
NB
297 if (tmpdev->used > 1) continue;
298
52826846 299 if (ident->devices &&
56eedc1a 300 !match_oneof(ident->devices, devname)) {
52437b4f 301 if (report_missmatch)
56eedc1a 302 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
52826846 303 continue;
56eedc1a 304 }
4b1ac34b 305
8b0dabea 306 dfd = dev_open(devname, O_RDONLY|O_EXCL);
64c4757e 307 if (dfd < 0) {
52437b4f 308 if (report_missmatch)
682c7051 309 fprintf(stderr, Name ": cannot open device %s: %s\n",
64c4757e 310 devname, strerror(errno));
da6b5ca9 311 tmpdev->used = 2;
52826846
NB
312 } else if (fstat(dfd, &stb)< 0) {
313 /* Impossible! */
314 fprintf(stderr, Name ": fstat failed for %s: %s\n",
315 devname, strerror(errno));
da6b5ca9 316 tmpdev->used = 2;
cd29a5c8
NB
317 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
318 fprintf(stderr, Name ": %s is not a block device.\n",
52826846 319 devname);
da6b5ca9 320 tmpdev->used = 2;
5083d66b
N
321 } else if (must_be_container(dfd)) {
322 if (st) {
323 /* already found some components, this cannot
324 * be another one.
325 */
326 if (report_missmatch)
327 fprintf(stderr, Name ": %s is a container, but we are looking for components\n",
328 devname);
329 tmpdev->used = 2;
330 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
331 if (report_missmatch)
332 fprintf(stderr, Name ": not a recognisable container: %s\n",
333 devname);
334 tmpdev->used = 2;
335 } else if (tst->ss->load_container(tst, dfd, NULL)) {
336 if (report_missmatch)
337 fprintf(stderr, Name ": no correct container type: %s\n",
338 devname);
339 tmpdev->used = 2;
340 } else if (auto_assem &&
341 !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)),
342 tst->ss->match_home(tst, homehost) == 1)) {
343 if (report_missmatch)
344 fprintf(stderr, Name ": %s has metadata type %s for which "
345 "auto-assembly is disabled\n",
346 devname, tst->ss->name);
347 tmpdev->used = 2;
348 } else
349 found_container = 1;
52826846 350 } else {
5083d66b
N
351 if (!tst && (tst = guess_super(dfd)) == NULL) {
352 if (report_missmatch)
353 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
354 devname);
355 tmpdev->used = 2;
356 } else if (tst->ss->load_super(tst,dfd, NULL)) {
357 if (report_missmatch)
358 fprintf(stderr, Name ": no RAID superblock on %s\n",
359 devname);
360 tmpdev->used = 2;
361 } else if (tst->ss->compare_super == NULL) {
362 if (report_missmatch)
363 fprintf(stderr, Name ": Cannot assemble %s metadata on %s\n",
364 tst->ss->name, devname);
365 tmpdev->used = 2;
366 } else if (auto_assem && st == NULL &&
367 !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)),
368 tst->ss->match_home(tst, homehost) == 1)) {
369 if (report_missmatch)
370 fprintf(stderr, Name ": %s has metadata type %s for which "
371 "auto-assembly is disabled\n",
372 devname, tst->ss->name);
373 tmpdev->used = 2;
374 }
64c4757e 375 }
c06487ce 376 if (dfd >= 0) close(dfd);
d68ea4d7
N
377 if (tmpdev->used == 2) {
378 if (auto_assem)
379 /* Ignore unrecognised devices during auto-assembly */
380 goto loop;
381 if (ident->uuid_set || ident->name[0] ||
382 ident->super_minor != UnSet)
383 /* Ignore unrecognised device if looking for
384 * specific array */
385 goto loop;
386
387
388 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
389 devname);
390 if (st)
391 st->ss->free_super(st);
392 dev_policy_free(pol);
393 return 1;
394 }
52826846 395
5083d66b 396 if (found_container) {
9008ed1c
N
397 /* tmpdev is a container. We need to be either
398 * looking for a member, or auto-assembling
399 */
9008ed1c
N
400
401 if (ident->container) {
402 if (ident->container[0] == '/' &&
403 !same_dev(ident->container, devname)) {
404 if (report_missmatch)
405 fprintf(stderr, Name ": %s is not the container required (%s)\n",
406 devname, ident->container);
407 goto loop;
408 }
409 if (ident->container[0] != '/') {
410 /* we have a uuid */
411 int uuid[4];
87477e6d
N
412
413 content = &info;
414 memset(content, 0, sizeof(*content));
415 tst->ss->getinfo_super(tst, content, NULL);
416
9008ed1c
N
417 if (!parse_uuid(ident->container, uuid) ||
418 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
419 if (report_missmatch)
420 fprintf(stderr, Name ": %s has wrong UUID to be required container\n",
421 devname);
422 goto loop;
423 }
424 }
425 }
426 /* It is worth looking inside this container.
427 */
7636b5a8
N
428 if (verbose > 0)
429 fprintf(stderr, Name ": looking in container %s\n",
430 devname);
1415fe4b 431
88cef9b3
N
432 for (content = tst->ss->container_content(tst, NULL);
433 content;
434 content = content->next) {
1415fe4b 435
88cef9b3
N
436 if (!ident_matches(ident, content, tst,
437 homehost, update,
438 report_missmatch ? devname : NULL))
439 /* message already printed */;
440 else if (is_member_busy(content->text_version)) {
441 if (report_missmatch)
442 fprintf(stderr, Name ": member %s in %s is already assembled\n",
443 content->text_version,
444 devname);
445 } else
446 break;
447 }
448 if (!content) {
f7ad3ccc 449 tmpdev->used = 2;
88cef9b3 450 goto loop; /* empty container */
fa031239 451 }
88cef9b3 452
9008ed1c 453 st = tst; tst = NULL;
4c1c3ad8 454 if (!auto_assem && inargv && tmpdev->next != NULL) {
9008ed1c
N
455 fprintf(stderr, Name ": %s is a container, but is not "
456 "only device given: confused and aborting\n",
457 devname);
458 st->ss->free_super(st);
4e8d9f0a 459 dev_policy_free(pol);
9008ed1c
N
460 return 1;
461 }
7636b5a8
N
462 if (verbose > 0)
463 fprintf(stderr, Name ": found match on member %s in %s\n",
464 content->text_version, devname);
bac0d92e 465
5083d66b
N
466 /* make sure we finished the loop */
467 tmpdev = NULL;
bac0d92e 468 goto loop;
5083d66b 469 } else {
bac0d92e 470
5083d66b
N
471 content = &info;
472 memset(content, 0, sizeof(*content));
473 tst->ss->getinfo_super(tst, content, NULL);
474
475 if (!ident_matches(ident, content, tst,
476 homehost, update,
477 report_missmatch ? devname : NULL))
df37ffc0 478 goto loop;
5083d66b
N
479
480 if (st == NULL)
481 st = dup_super(tst);
482 if (st->minor_version == -1)
483 st->minor_version = tst->minor_version;
484 if (st->ss != tst->ss ||
485 st->minor_version != tst->minor_version ||
486 st->ss->compare_super(st, tst) != 0) {
487 /* Some mismatch. If exactly one array matches this host,
488 * we can resolve on that one.
489 * Or, if we are auto assembling, we just ignore the second
490 * for now.
491 */
492 if (auto_assem)
493 goto loop;
494 if (homehost) {
495 int first = st->ss->match_home(st, homehost);
496 int last = tst->ss->match_home(tst, homehost);
497 if (first != last &&
498 (first == 1 || last == 1)) {
499 /* We can do something */
500 if (first) {/* just ignore this one */
501 if (report_missmatch)
502 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
503 devname);
504 goto loop;
505 } else { /* reject all those sofar */
506 struct mddev_dev *td;
507 if (report_missmatch)
508 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
509 devname);
510 for (td=devlist; td != tmpdev; td=td->next)
511 if (td->used == 1)
512 td->used = 0;
513 tmpdev->used = 1;
514 goto loop;
515 }
83b6208e
NB
516 }
517 }
5083d66b
N
518 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
519 devname);
520 tst->ss->free_super(tst);
521 st->ss->free_super(st);
522 dev_policy_free(pol);
523 return 1;
83b6208e 524 }
5083d66b 525 tmpdev->used = 1;
64c4757e 526 }
df37ffc0 527 loop:
4e8d9f0a
N
528 dev_policy_free(pol);
529 pol = NULL;
3d2b16e7
NB
530 if (tst)
531 tst->ss->free_super(tst);
811e6cbe
NB
532 }
533
98dbd966 534 if (!st || !st->sb || !content)
c30e5369
N
535 return 2;
536
05833051 537 /* Now need to open the array device. Use create_mddev */
98dbd966 538 if (content == &info)
a5d85af7 539 st->ss->getinfo_super(st, content, NULL);
3d2c4fc7 540
69207ff6 541 trustworthy = FOREIGN;
05833051 542 name = content->name;
745f72f6
N
543 switch (st->ss->match_home(st, homehost)
544 ?: st->ss->match_home(st, "any")) {
69207ff6
N
545 case 1:
546 trustworthy = LOCAL;
98dbd966 547 name = strchr(content->name, ':');
69207ff6
N
548 if (name)
549 name++;
3d2c4fc7 550 else
98dbd966 551 name = content->name;
69207ff6 552 break;
c30e5369 553 }
05833051 554 if (!auto_assem)
aa7c284c 555 /* If the array is listed in mdadm.conf or on
ac2ecf55
N
556 * command line, then we trust the name
557 * even if the array doesn't look local
558 */
559 trustworthy = LOCAL;
560
05833051 561 if (name[0] == 0 &&
98dbd966
DW
562 content->array.level == LEVEL_CONTAINER) {
563 name = content->text_version;
a4bc1720
N
564 trustworthy = METADATA;
565 }
0ac91628
N
566
567 if (name[0] && trustworthy != LOCAL &&
568 ! require_homehost &&
569 conf_name_is_free(name))
570 trustworthy = LOCAL;
571
7cdc0872
N
572 if (trustworthy == LOCAL &&
573 strchr(name, ':'))
574 /* Ignore 'host:' prefix of name */
575 name = strchr(name, ':')+1;
576
a04d5763
N
577 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
578 chosen_name);
c30e5369
N
579 if (mdfd < 0) {
580 st->ss->free_super(st);
c30e5369 581 if (auto_assem)
60e1bc1a 582 goto try_again;
c30e5369
N
583 return 1;
584 }
a04d5763 585 mddev = chosen_name;
c30e5369
N
586 vers = md_get_version(mdfd);
587 if (vers < 9000) {
588 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
589 " Upgrade your kernel or try --build\n");
590 close(mdfd);
591 return 1;
592 }
66afdfa9 593 if (mddev_busy(fd2devnum(mdfd))) {
c30e5369
N
594 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
595 mddev);
596 for (tmpdev = devlist ;
597 tmpdev && tmpdev->used != 1;
598 tmpdev = tmpdev->next)
599 ;
600 if (tmpdev && auto_assem)
601 fprintf(stderr, Name ": %s needed for %s...\n",
602 mddev, tmpdev->devname);
603 close(mdfd);
604 mdfd = -3;
605 st->ss->free_super(st);
c30e5369
N
606 if (auto_assem)
607 goto try_again;
608 return 1;
8a46fe84 609 }
c30e5369 610 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
8a46fe84 611
9008ed1c
N
612#ifndef MDASSEMBLE
613 if (content != &info) {
614 /* This is a member of a container. Try starting the array. */
615 return assemble_container_content(st, mdfd, content, runstop,
616 chosen_name, verbose);
617 }
618#endif
811e6cbe 619 /* Ok, no bad inconsistancy, we can try updating etc */
bf4fb153 620 bitmap_done = 0;
6e46bf34 621 content->update_private = NULL;
d7f7ebb7 622 devices = malloc(num_devs * sizeof(*devices));
02e7c5b7 623 devmap = calloc(num_devs * content->array.raid_disks, 1);
da6b5ca9 624 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
811e6cbe
NB
625 char *devname = tmpdev->devname;
626 struct stat stb;
5787fa49 627 /* looks like a good enough match to update the super block if needed */
d1f1011b 628#ifndef MDASSEMBLE
5787fa49 629 if (update) {
811e6cbe 630 int dfd;
4b1ac34b
NB
631 /* prepare useful information in info structures */
632 struct stat stb2;
3da92f27 633 struct supertype *tst;
1e2b2765 634 int err;
4b1ac34b 635 fstat(mdfd, &stb2);
7d99579f
NB
636
637 if (strcmp(update, "uuid")==0 &&
638 !ident->uuid_set) {
639 int rfd;
640 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
641 read(rfd, ident->uuid, 16) != 16) {
642 *(__u32*)(ident->uuid) = random();
643 *(__u32*)(ident->uuid+1) = random();
644 *(__u32*)(ident->uuid+2) = random();
645 *(__u32*)(ident->uuid+3) = random();
646 }
647 if (rfd >= 0) close(rfd);
7d99579f 648 }
811e6cbe
NB
649 dfd = dev_open(devname, O_RDWR|O_EXCL);
650
0430ed48
NB
651 remove_partitions(dfd);
652
3da92f27 653 tst = dup_super(st);
9f22b13f
N
654 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
655 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
656 devname);
657 if (dfd >= 0)
658 close(dfd);
659 close(mdfd);
d7f7ebb7 660 free(devices);
02e7c5b7 661 free(devmap);
9f22b13f
N
662 return 1;
663 }
02e7c5b7 664 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
811e6cbe 665
98dbd966
DW
666 memcpy(content->uuid, ident->uuid, 16);
667 strcpy(content->name, ident->name);
668 content->array.md_minor = minor(stb2.st_rdev);
811e6cbe 669
1e2b2765
N
670 if (strcmp(update, "byteorder") == 0)
671 err = 0;
672 else
673 err = tst->ss->update_super(tst, content, update,
674 devname, verbose,
675 ident->uuid_set,
676 homehost);
677 if (err < 0) {
678 fprintf(stderr,
679 Name ": --update=%s not understood"
680 " for %s metadata\n",
681 update, tst->ss->name);
682 tst->ss->free_super(tst);
683 free(tst);
684 close(mdfd);
685 close(dfd);
d7f7ebb7 686 free(devices);
02e7c5b7 687 free(devmap);
1e2b2765
N
688 return 1;
689 }
e5eac01f
NB
690 if (strcmp(update, "uuid")==0 &&
691 !ident->uuid_set) {
692 ident->uuid_set = 1;
98dbd966 693 memcpy(ident->uuid, content->uuid, 16);
e5eac01f 694 }
1e2b2765 695 if (tst->ss->store_super(tst, dfd))
5787fa49
NB
696 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
697 devname);
1e2b2765 698 close(dfd);
8131b493
NB
699
700 if (strcmp(update, "uuid")==0 &&
bf4fb153 701 ident->bitmap_fd >= 0 && !bitmap_done) {
3da92f27 702 if (bitmap_update_uuid(ident->bitmap_fd,
98dbd966 703 content->uuid,
3da92f27 704 tst->ss->swapuuid) != 0)
bf4fb153
NB
705 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
706 else
707 bitmap_done = 1;
708 }
3da92f27 709 tst->ss->free_super(tst);
d1f1011b
NB
710 } else
711#endif
712 {
30e1b9a5 713 struct supertype *tst = dup_super(st);
da6b5ca9
NB
714 int dfd;
715 dfd = dev_open(devname, O_RDWR|O_EXCL);
716
0430ed48
NB
717 remove_partitions(dfd);
718
9f22b13f
N
719 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
720 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
721 devname);
722 if (dfd >= 0)
723 close(dfd);
724 close(mdfd);
d7f7ebb7 725 free(devices);
02e7c5b7 726 free(devmap);
9f22b13f
N
727 return 1;
728 }
02e7c5b7 729 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
3da92f27 730 tst->ss->free_super(tst);
da6b5ca9 731 close(dfd);
5787fa49
NB
732 }
733
811e6cbe
NB
734 stat(devname, &stb);
735
dab6685f 736 if (verbose > 0)
cd29a5c8 737 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
98dbd966 738 devname, mddev, content->disk.raid_disk);
64c4757e 739 devices[devcnt].devname = devname;
64c4757e 740 devices[devcnt].uptodate = 0;
98dbd966 741 devices[devcnt].i = *content;
213ee40b
NB
742 devices[devcnt].i.disk.major = major(stb.st_rdev);
743 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
64c4757e 744 if (most_recent < devcnt) {
213ee40b
NB
745 if (devices[devcnt].i.events
746 > devices[most_recent].i.events)
64c4757e
NB
747 most_recent = devcnt;
748 }
df0d4ea0 749 if (content->array.level == LEVEL_MULTIPATH)
5787fa49
NB
750 /* with multipath, the raid_disk from the superblock is meaningless */
751 i = devcnt;
752 else
213ee40b 753 i = devices[devcnt].i.disk.raid_disk;
308e1801 754 if (i+1 == 0) {
98dbd966
DW
755 if (nextspare < content->array.raid_disks)
756 nextspare = content->array.raid_disks;
308e1801 757 i = nextspare++;
08110d41 758 } else {
98dbd966 759 if (i >= content->array.raid_disks &&
08110d41
NB
760 i >= nextspare)
761 nextspare = i+1;
308e1801 762 }
98c6faba 763 if (i < 10000) {
56eedc1a 764 if (i >= bestcnt) {
f21e18ca 765 int newbestcnt = i+10;
56eedc1a 766 int *newbest = malloc(sizeof(int)*newbestcnt);
f21e18ca 767 int c;
56eedc1a
NB
768 for (c=0; c < newbestcnt; c++)
769 if (c < bestcnt)
770 newbest[c] = best[c];
771 else
772 newbest[c] = -1;
773 if (best)free(best);
774 best = newbest;
775 bestcnt = newbestcnt;
776 }
6a255b69 777 if (best[i] >=0 &&
213ee40b
NB
778 devices[best[i]].i.events
779 == devices[devcnt].i.events
780 && (devices[best[i]].i.disk.minor
781 != devices[devcnt].i.disk.minor)
b8ac1967 782 && st->ss == &super0
98dbd966 783 && content->array.level != LEVEL_MULTIPATH) {
6a255b69
NB
784 /* two different devices with identical superblock.
785 * Could be a mis-detection caused by overlapping
786 * partitions. fail-safe.
787 */
788 fprintf(stderr, Name ": WARNING %s and %s appear"
789 " to have very similar superblocks.\n"
790 " If they are really different, "
791 "please --zero the superblock on one\n"
d2df86e0
NB
792 " If they are the same or overlap,"
793 " please remove one from %s.\n",
794 devices[best[i]].devname, devname,
795 inargv ? "the list" :
796 "the\n DEVICE list in mdadm.conf"
797 );
7f91af49 798 close(mdfd);
d7f7ebb7 799 free(devices);
02e7c5b7 800 free(devmap);
6a255b69
NB
801 return 1;
802 }
52826846 803 if (best[i] == -1
213ee40b
NB
804 || (devices[best[i]].i.events
805 < devices[devcnt].i.events))
52826846 806 best[i] = devcnt;
56eedc1a 807 }
64c4757e 808 devcnt++;
df37ffc0 809 }
6e46bf34
DW
810 free(content->update_private);
811 content->update_private = NULL;
4b1ac34b 812
64c4757e 813 if (devcnt == 0) {
682c7051 814 fprintf(stderr, Name ": no devices found for %s\n",
64c4757e 815 mddev);
3d2b16e7
NB
816 if (st)
817 st->ss->free_super(st);
7f91af49 818 close(mdfd);
d7f7ebb7 819 free(devices);
02e7c5b7 820 free(devmap);
64c4757e
NB
821 return 1;
822 }
4b1ac34b 823
3d2b16e7
NB
824 if (update && strcmp(update, "byteorder")==0)
825 st->minor_version = 90;
826
a5d85af7 827 st->ss->getinfo_super(st, content, NULL);
98dbd966 828 clean = content->array.state & 1;
4b1ac34b 829
64c4757e
NB
830 /* now we have some devices that might be suitable.
831 * I wonder how many
832 */
98dbd966
DW
833 avail = malloc(content->array.raid_disks);
834 memset(avail, 0, content->array.raid_disks);
64c4757e 835 okcnt = 0;
52826846 836 sparecnt=0;
1ff98339 837 rebuilding_cnt=0;
f21e18ca 838 for (i=0; i< bestcnt; i++) {
64c4757e 839 int j = best[i];
ee04451c
NB
840 int event_margin = 1; /* always allow a difference of '1'
841 * like the kernel does
842 */
64c4757e 843 if (j < 0) continue;
d013a55e
NB
844 /* note: we ignore error flags in multipath arrays
845 * as they don't make sense
846 */
df0d4ea0 847 if (content->array.level != LEVEL_MULTIPATH)
f22385f9 848 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
213ee40b
NB
849 if (!(devices[j].i.disk.state
850 & (1<<MD_DISK_FAULTY)))
aa88f531 851 sparecnt++;
d013a55e 852 continue;
aa88f531 853 }
02e7c5b7
N
854 /* If this devices thinks that 'most_recent' has failed, then
855 * we must reject this device.
856 */
857 if (j != most_recent &&
858 content->array.raid_disks > 0 &&
859 devices[most_recent].i.disk.raid_disk >= 0 &&
860 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
861 if (verbose > -1)
862 fprintf(stderr, Name ": ignoring %s as it reports %s as failed\n",
863 devices[j].devname, devices[most_recent].devname);
864 best[i] = -1;
865 continue;
866 }
213ee40b
NB
867 if (devices[j].i.events+event_margin >=
868 devices[most_recent].i.events) {
64c4757e 869 devices[j].uptodate = 1;
1ff98339
N
870 if (i < content->array.raid_disks) {
871 if (devices[j].i.recovery_start == MaxSector) {
872 okcnt++;
873 avail[i]=1;
874 } else
875 rebuilding_cnt++;
265e0f17 876 } else
52826846 877 sparecnt++;
64c4757e
NB
878 }
879 }
02e7c5b7 880 free(devmap);
98dbd966
DW
881 while (force && !enough(content->array.level, content->array.raid_disks,
882 content->array.layout, 1,
265e0f17 883 avail, okcnt)) {
64c4757e
NB
884 /* Choose the newest best drive which is
885 * not up-to-date, update the superblock
886 * and add it.
887 */
52826846 888 int fd;
3da92f27 889 struct supertype *tst;
f21e18ca 890 unsigned long long current_events;
cd29a5c8 891 chosen_drive = -1;
f21e18ca 892 for (i = 0; i < content->array.raid_disks && i < bestcnt; i++) {
52826846
NB
893 int j = best[i];
894 if (j>=0 &&
895 !devices[j].uptodate &&
921d9e16 896 devices[j].i.recovery_start == MaxSector &&
52826846 897 (chosen_drive < 0 ||
213ee40b
NB
898 devices[j].i.events
899 > devices[chosen_drive].i.events))
52826846
NB
900 chosen_drive = j;
901 }
902 if (chosen_drive < 0)
903 break;
213ee40b 904 current_events = devices[chosen_drive].i.events;
c5a6f9a6 905 add_another:
dab6685f
NB
906 if (verbose >= 0)
907 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
213ee40b
NB
908 devices[chosen_drive].devname,
909 devices[chosen_drive].i.disk.raid_disk,
910 (int)(devices[chosen_drive].i.events),
911 (int)(devices[most_recent].i.events));
8b0dabea 912 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846
NB
913 if (fd < 0) {
914 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
915 devices[chosen_drive].devname);
213ee40b 916 devices[chosen_drive].i.events = 0;
52826846
NB
917 continue;
918 }
3da92f27 919 tst = dup_super(st);
60b435db 920 if (tst->ss->load_super(tst,fd, NULL)) {
52826846
NB
921 close(fd);
922 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
923 devices[chosen_drive].devname);
213ee40b 924 devices[chosen_drive].i.events = 0;
52826846
NB
925 continue;
926 }
98dbd966
DW
927 content->events = devices[most_recent].i.events;
928 tst->ss->update_super(tst, content, "force-one",
67a8c82d
NB
929 devices[chosen_drive].devname, verbose,
930 0, NULL);
4b1ac34b 931
3da92f27 932 if (tst->ss->store_super(tst, fd)) {
52826846
NB
933 close(fd);
934 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
935 devices[chosen_drive].devname);
213ee40b 936 devices[chosen_drive].i.events = 0;
3da92f27 937 tst->ss->free_super(tst);
52826846
NB
938 continue;
939 }
940 close(fd);
213ee40b 941 devices[chosen_drive].i.events = devices[most_recent].i.events;
52826846 942 devices[chosen_drive].uptodate = 1;
265e0f17 943 avail[chosen_drive] = 1;
52826846 944 okcnt++;
3da92f27 945 tst->ss->free_super(tst);
c5a6f9a6
NB
946
947 /* If there are any other drives of the same vintage,
948 * add them in as well. We can't lose and we might gain
949 */
f21e18ca 950 for (i = 0; i < content->array.raid_disks && i < bestcnt ; i++) {
c5a6f9a6
NB
951 int j = best[i];
952 if (j >= 0 &&
953 !devices[j].uptodate &&
213ee40b 954 devices[j].i.events == current_events) {
c5a6f9a6
NB
955 chosen_drive = j;
956 goto add_another;
957 }
958 }
64c4757e 959 }
52826846
NB
960
961 /* Now we want to look at the superblock which the kernel will base things on
962 * and compare the devices that we think are working with the devices that the
963 * superblock thinks are working.
964 * If there are differences and --force is given, then update this chosen
965 * superblock.
966 */
cd29a5c8 967 chosen_drive = -1;
3da92f27 968 st->ss->free_super(st);
56eedc1a 969 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
52826846
NB
970 int j = best[i];
971 int fd;
4b1ac34b 972
52826846
NB
973 if (j<0)
974 continue;
975 if (!devices[j].uptodate)
976 continue;
977 chosen_drive = j;
8b0dabea 978 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
52826846
NB
979 fprintf(stderr, Name ": Cannot open %s: %s\n",
980 devices[j].devname, strerror(errno));
7f91af49 981 close(mdfd);
d7f7ebb7 982 free(devices);
52826846
NB
983 return 1;
984 }
3da92f27 985 if (st->ss->load_super(st,fd, NULL)) {
52826846
NB
986 close(fd);
987 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
988 devices[j].devname);
7f91af49 989 close(mdfd);
d7f7ebb7 990 free(devices);
52826846
NB
991 return 1;
992 }
993 close(fd);
994 }
3da92f27 995 if (st->sb == NULL) {
4b1ac34b 996 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
7f91af49 997 close(mdfd);
d7f7ebb7 998 free(devices);
4b1ac34b
NB
999 return 1;
1000 }
a5d85af7 1001 st->ss->getinfo_super(st, content, NULL);
f35f2525 1002#ifndef MDASSEMBLE
98dbd966 1003 sysfs_init(content, mdfd, 0);
f35f2525 1004#endif
56eedc1a 1005 for (i=0; i<bestcnt; i++) {
52826846 1006 int j = best[i];
98c6faba 1007 unsigned int desired_state;
11a3e71d 1008
98dbd966 1009 if (i < content->array.raid_disks)
11a3e71d
NB
1010 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1011 else
1012 desired_state = 0;
1013
52826846
NB
1014 if (j<0)
1015 continue;
1016 if (!devices[j].uptodate)
1017 continue;
4b1ac34b 1018
213ee40b 1019 devices[j].i.disk.state = desired_state;
4e9a6ff7
N
1020 if (!(devices[j].i.array.state & 1))
1021 clean = 0;
213ee40b
NB
1022
1023 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
68c7d6d7 1024 verbose, 0, NULL)) {
52826846 1025 if (force) {
dab6685f
NB
1026 if (verbose >= 0)
1027 fprintf(stderr, Name ": "
1028 "clearing FAULTY flag for device %d in %s for %s\n",
1029 j, mddev, devices[j].devname);
4b1ac34b 1030 change = 1;
52826846 1031 } else {
dab6685f
NB
1032 if (verbose >= -1)
1033 fprintf(stderr, Name ": "
1034 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1035 i, mddev, devices[j].devname);
52826846
NB
1036 }
1037 }
4b1ac34b 1038#if 0
213ee40b 1039 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
52826846
NB
1040 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1041 i, mddev);
1042 }
4b1ac34b 1043#endif
52826846 1044 }
c5a6f9a6 1045 if (force && !clean &&
98dbd966
DW
1046 !enough(content->array.level, content->array.raid_disks,
1047 content->array.layout, clean,
c5a6f9a6 1048 avail, okcnt)) {
98dbd966 1049 change += st->ss->update_super(st, content, "force-array",
67a8c82d
NB
1050 devices[chosen_drive].devname, verbose,
1051 0, NULL);
583315d9 1052 clean = 1;
d013a55e 1053 }
52826846 1054
4b1ac34b 1055 if (change) {
52826846 1056 int fd;
8b0dabea 1057 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846 1058 if (fd < 0) {
353632d9 1059 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
52826846 1060 devices[chosen_drive].devname);
7f91af49 1061 close(mdfd);
d7f7ebb7 1062 free(devices);
52826846
NB
1063 return 1;
1064 }
3da92f27 1065 if (st->ss->store_super(st, fd)) {
52826846
NB
1066 close(fd);
1067 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
1068 devices[chosen_drive].devname);
7f91af49 1069 close(mdfd);
d7f7ebb7 1070 free(devices);
52826846
NB
1071 return 1;
1072 }
1073 close(fd);
52826846
NB
1074 }
1075
353632d9
NB
1076 /* If we are in the middle of a reshape we may need to restore saved data
1077 * that was moved aside due to the reshape overwriting live data
1078 * The code of doing this lives in Grow.c
1079 */
39c5a909 1080#ifndef MDASSEMBLE
98dbd966 1081 if (content->reshape_active) {
353632d9
NB
1082 int err = 0;
1083 int *fdlist = malloc(sizeof(int)* bestcnt);
cd77ac4e 1084 if (verbose > 0)
ea0ebe96
N
1085 fprintf(stderr, Name ":%s has an active reshape - checking "
1086 "if critical section needs to be restored\n",
1087 chosen_name);
353632d9
NB
1088 for (i=0; i<bestcnt; i++) {
1089 int j = best[i];
1090 if (j >= 0) {
1091 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1092 if (fdlist[i] < 0) {
1093 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1094 devices[j].devname);
1095 err = 1;
1096 break;
1097 }
1098 } else
1099 fdlist[i] = -1;
1100 }
1101 if (!err)
cd77ac4e 1102 err = Grow_restart(st, content, fdlist, bestcnt, backup_file, verbose > 0);
353632d9
NB
1103 while (i>0) {
1104 i--;
1105 if (fdlist[i]>=0) close(fdlist[i]);
1106 }
1107 if (err) {
1108 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
e9e43ec3
N
1109 if (backup_file == NULL)
1110 fprintf(stderr," Possibly you needed to specify the --backup-file\n");
7f91af49 1111 close(mdfd);
d7f7ebb7 1112 free(devices);
353632d9
NB
1113 return err;
1114 }
1115 }
39c5a909 1116#endif
aa88f531
NB
1117 /* count number of in-sync devices according to the superblock.
1118 * We must have this number to start the array without -s or -R
1119 */
98dbd966 1120 req_cnt = content->array.working_disks;
aa88f531 1121
64c4757e
NB
1122 /* Almost ready to actually *do* something */
1123 if (!old_linux) {
fbf8a0b7 1124 int rv;
a19c88b8 1125
a04d5763
N
1126 /* First, fill in the map, so that udev can find our name
1127 * as soon as we become active.
1128 */
98dbd966
DW
1129 map_update(NULL, fd2devnum(mdfd), content->text_version,
1130 content->uuid, chosen_name);
a04d5763 1131
98dbd966 1132 rv = set_array_info(mdfd, st, content);
fbf8a0b7 1133 if (rv) {
f35f2525 1134 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
64c4757e 1135 mddev, strerror(errno));
24af7a87 1136 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1137 close(mdfd);
d7f7ebb7 1138 free(devices);
64c4757e
NB
1139 return 1;
1140 }
11484a63 1141 if (ident->bitmap_fd >= 0) {
c82f047c
NB
1142 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1143 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
24af7a87 1144 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1145 close(mdfd);
d7f7ebb7 1146 free(devices);
c82f047c
NB
1147 return 1;
1148 }
7ef02d01
NB
1149 } else if (ident->bitmap_file) {
1150 /* From config file */
1151 int bmfd = open(ident->bitmap_file, O_RDWR);
1152 if (bmfd < 0) {
1153 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1154 ident->bitmap_file);
24af7a87 1155 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1156 close(mdfd);
d7f7ebb7 1157 free(devices);
7ef02d01
NB
1158 return 1;
1159 }
1160 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1161 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1162 close(bmfd);
24af7a87 1163 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1164 close(mdfd);
d7f7ebb7 1165 free(devices);
7ef02d01
NB
1166 return 1;
1167 }
1168 close(bmfd);
c82f047c 1169 }
7ef02d01 1170
52826846 1171 /* First, add the raid disks, but add the chosen one last */
56eedc1a 1172 for (i=0; i<= bestcnt; i++) {
52826846 1173 int j;
56eedc1a 1174 if (i < bestcnt) {
52826846
NB
1175 j = best[i];
1176 if (j == chosen_drive)
1177 continue;
1178 } else
1179 j = chosen_drive;
1180
56eedc1a 1181 if (j >= 0 /* && devices[j].uptodate */) {
98dbd966 1182 rv = add_disk(mdfd, st, content, &devices[j].i);
7801ac20 1183
a19c88b8 1184 if (rv) {
213ee40b
NB
1185 fprintf(stderr, Name ": failed to add "
1186 "%s to %s: %s\n",
64c4757e
NB
1187 devices[j].devname,
1188 mddev,
1189 strerror(errno));
98dbd966 1190 if (i < content->array.raid_disks
213ee40b 1191 || i == bestcnt)
52826846
NB
1192 okcnt--;
1193 else
1194 sparecnt--;
dab6685f 1195 } else if (verbose > 0)
213ee40b
NB
1196 fprintf(stderr, Name ": added %s "
1197 "to %s as %d\n",
1198 devices[j].devname, mddev,
1199 devices[j].i.disk.raid_disk);
98dbd966 1200 } else if (verbose > 0 && i < content->array.raid_disks)
213ee40b
NB
1201 fprintf(stderr, Name ": no uptodate device for "
1202 "slot %d of %s\n",
64c4757e
NB
1203 i, mddev);
1204 }
aba69144 1205
98dbd966 1206 if (content->array.level == LEVEL_CONTAINER) {
598f0d58
NB
1207 if (verbose >= 0) {
1208 fprintf(stderr, Name ": Container %s has been "
1209 "assembled with %d drive%s",
57ed8c91 1210 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
f21e18ca 1211 if (okcnt < (unsigned)content->array.raid_disks)
598f0d58 1212 fprintf(stderr, " (out of %d)",
98dbd966 1213 content->array.raid_disks);
598f0d58
NB
1214 fprintf(stderr, "\n");
1215 }
98dbd966 1216 sysfs_uevent(content, "change");
a7c6e3fb 1217 wait_for(chosen_name, mdfd);
7f91af49 1218 close(mdfd);
d7f7ebb7 1219 free(devices);
598f0d58
NB
1220 return 0;
1221 }
1222
64c4757e 1223 if (runstop == 1 ||
b8a8ccf9 1224 (runstop <= 0 &&
98dbd966
DW
1225 ( enough(content->array.level, content->array.raid_disks,
1226 content->array.layout, clean, avail, okcnt) &&
1ff98339 1227 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
aa88f531 1228 ))) {
e9e43ec3
N
1229 /* This array is good-to-go.
1230 * If a reshape is in progress then we might need to
1231 * continue monitoring it. In that case we start
1232 * it read-only and let the grow code make it writable.
1233 */
1234 int rv;
eb3929a4 1235#ifndef MDASSEMBLE
e9e43ec3
N
1236 if (content->reshape_active &&
1237 content->delta_disks <= 0)
1238 rv = Grow_continue(mdfd, st, content, backup_file);
1239 else
eb3929a4 1240#endif
e9e43ec3
N
1241 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1242 if (rv == 0) {
dab6685f
NB
1243 if (verbose >= 0) {
1244 fprintf(stderr, Name ": %s has been started with %d drive%s",
1245 mddev, okcnt, okcnt==1?"":"s");
f21e18ca 1246 if (okcnt < (unsigned)content->array.raid_disks)
98dbd966 1247 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1ff98339
N
1248 if (rebuilding_cnt)
1249 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
dab6685f
NB
1250 if (sparecnt)
1251 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1252 fprintf(stderr, ".\n");
1253 }
8a659c33
N
1254 if (content->reshape_active &&
1255 content->array.level >= 4 &&
1256 content->array.level <= 6) {
acee8e89
N
1257 /* might need to increase the size
1258 * of the stripe cache - default is 256
1259 */
8a659c33 1260 if (256 < 4 * (content->array.chunk_size/4096)) {
acee8e89
N
1261 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1262 if (sra)
1263 sysfs_set_num(sra, NULL,
1264 "stripe_cache_size",
8a659c33 1265 (4 * content->array.chunk_size / 4096) + 1);
acee8e89
N
1266 }
1267 }
7e83544b
N
1268 if (okcnt < (unsigned)content->array.raid_disks) {
1269 /* If any devices did not get added
1270 * because the kernel rejected them based
1271 * on event count, try adding them
1272 * again providing the action policy is
1273 * 're-add' or greater. The bitmap
1274 * might allow them to be included, or
1275 * they will become spares.
1276 */
1277 for (i = 0; i <= bestcnt; i++) {
1278 int j = best[i];
1279 if (j >= 0 && !devices[j].uptodate) {
1280 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1281 continue;
1282 rv = add_disk(mdfd, st, content,
1283 &devices[j].i);
1284 if (rv == 0 && verbose >= 0)
1285 fprintf(stderr,
1286 Name ": %s has been re-added.\n",
1287 devices[j].devname);
1288 }
1289 }
1290 }
a7c6e3fb 1291 wait_for(mddev, mdfd);
7f91af49
N
1292 close(mdfd);
1293 if (auto_assem) {
589395d6 1294 int usecs = 1;
589395d6
NB
1295 /* There is a nasty race with 'mdadm --monitor'.
1296 * If it opens this device before we close it,
1297 * it gets an incomplete open on which IO
598f0d58
NB
1298 * doesn't work and the capacity is
1299 * wrong.
589395d6
NB
1300 * If we reopen (to check for layered devices)
1301 * before --monitor closes, we loose.
1302 *
1303 * So: wait upto 1 second for there to be
1304 * a non-zero capacity.
1305 */
1306 while (usecs < 1000) {
1307 mdfd = open(mddev, O_RDONLY);
1308 if (mdfd >= 0) {
beae1dfe
NB
1309 unsigned long long size;
1310 if (get_dev_size(mdfd, NULL, &size) &&
589395d6
NB
1311 size > 0)
1312 break;
1313 close(mdfd);
1314 }
1315 usleep(usecs);
1316 usecs <<= 1;
1317 }
1318 }
d7f7ebb7 1319 free(devices);
64c4757e 1320 return 0;
82b27616 1321 }
682c7051 1322 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
64c4757e 1323 mddev, strerror(errno));
583315d9 1324
98dbd966
DW
1325 if (!enough(content->array.level, content->array.raid_disks,
1326 content->array.layout, 1, avail, okcnt))
583315d9
NB
1327 fprintf(stderr, Name ": Not enough devices to "
1328 "start the array.\n");
98dbd966
DW
1329 else if (!enough(content->array.level,
1330 content->array.raid_disks,
1331 content->array.layout, clean,
583315d9
NB
1332 avail, okcnt))
1333 fprintf(stderr, Name ": Not enough devices to "
1334 "start the array while not clean "
1335 "- consider --force.\n");
1336
7f91af49 1337 if (auto_assem)
1c203a4b 1338 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1339 close(mdfd);
d7f7ebb7 1340 free(devices);
64c4757e
NB
1341 return 1;
1342 }
82b27616 1343 if (runstop == -1) {
b8a8ccf9 1344 fprintf(stderr, Name ": %s assembled from %d drive%s",
52826846 1345 mddev, okcnt, okcnt==1?"":"s");
f21e18ca 1346 if (okcnt != (unsigned)content->array.raid_disks)
98dbd966 1347 fprintf(stderr, " (out of %d)", content->array.raid_disks);
b8a8ccf9 1348 fprintf(stderr, ", but not started.\n");
7f91af49 1349 close(mdfd);
d7f7ebb7 1350 free(devices);
64c4757e 1351 return 0;
82b27616 1352 }
4855f95c 1353 if (verbose >= -1) {
dab6685f 1354 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1ff98339
N
1355 if (rebuilding_cnt)
1356 fprintf(stderr, "%s %d rebuilding", sparecnt?", ":" and ", rebuilding_cnt);
dab6685f
NB
1357 if (sparecnt)
1358 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
98dbd966
DW
1359 if (!enough(content->array.level, content->array.raid_disks,
1360 content->array.layout, 1, avail, okcnt))
dab6685f 1361 fprintf(stderr, " - not enough to start the array.\n");
98dbd966
DW
1362 else if (!enough(content->array.level,
1363 content->array.raid_disks,
1364 content->array.layout, clean,
583315d9
NB
1365 avail, okcnt))
1366 fprintf(stderr, " - not enough to start the "
1367 "array while not clean - consider "
1368 "--force.\n");
dab6685f 1369 else {
f21e18ca 1370 if (req_cnt == (unsigned)content->array.raid_disks)
dab6685f
NB
1371 fprintf(stderr, " - need all %d to start it", req_cnt);
1372 else
98dbd966 1373 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
dab6685f
NB
1374 fprintf(stderr, " (use --run to insist).\n");
1375 }
aa88f531 1376 }
7f91af49 1377 if (auto_assem)
1c203a4b 1378 ioctl(mdfd, STOP_ARRAY, NULL);
56a8da69 1379 close(mdfd);
d7f7ebb7 1380 free(devices);
64c4757e 1381 return 1;
82b27616 1382 } else {
52826846
NB
1383 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1384 * been updated to point to the current locations of devices.
1385 * so we can just start the array
82b27616 1386 */
cd29a5c8 1387 unsigned long dev;
213ee40b
NB
1388 dev = makedev(devices[chosen_drive].i.disk.major,
1389 devices[chosen_drive].i.disk.minor);
82b27616
NB
1390 if (ioctl(mdfd, START_ARRAY, dev)) {
1391 fprintf(stderr, Name ": Cannot start array: %s\n",
1392 strerror(errno));
1393 }
aba69144 1394
64c4757e 1395 }
7f91af49 1396 close(mdfd);
d7f7ebb7 1397 free(devices);
e0d19036 1398 return 0;
64c4757e 1399}
6234c63c
DW
1400
1401#ifndef MDASSEMBLE
1402int assemble_container_content(struct supertype *st, int mdfd,
1403 struct mdinfo *content, int runstop,
1404 char *chosen_name, int verbose)
1405{
1406 struct mdinfo *dev, *sra;
1407 int working = 0, preexist = 0;
1408 struct map_ent *map = NULL;
1409
1410 sysfs_init(content, mdfd, 0);
1411
1412 sra = sysfs_read(mdfd, 0, GET_VERSION);
1413 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
4408ee76
N
1414 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1415 close(mdfd);
6234c63c 1416 return 1;
4408ee76 1417 }
6234c63c
DW
1418 if (sra)
1419 sysfs_free(sra);
1420
1421 for (dev = content->devs; dev; dev = dev->next)
462906cd 1422 if (sysfs_add_disk(content, dev, 1) == 0)
6234c63c
DW
1423 working++;
1424 else if (errno == EEXIST)
1425 preexist++;
4408ee76
N
1426 if (working == 0) {
1427 close(mdfd);
7cb2aa33 1428 return 1;/* Nothing new, don't try to start */
8b4e5ea9
N
1429 }
1430
1431 map_update(&map, fd2devnum(mdfd),
1432 content->text_version,
1433 content->uuid, chosen_name);
1434
1435 if (runstop > 0 ||
6234c63c 1436 (working + preexist) >= content->array.working_disks) {
bb50e5d3 1437 int err;
a714580e 1438
6234c63c
DW
1439 switch(content->array.level) {
1440 case LEVEL_LINEAR:
1441 case LEVEL_MULTIPATH:
1442 case 0:
bb50e5d3
N
1443 err = sysfs_set_str(content, NULL, "array_state",
1444 "active");
6234c63c
DW
1445 break;
1446 default:
bb50e5d3 1447 err = sysfs_set_str(content, NULL, "array_state",
6234c63c
DW
1448 "readonly");
1449 /* start mdmon if needed. */
bb50e5d3
N
1450 if (!err) {
1451 if (!mdmon_running(st->container_dev))
1452 start_mdmon(st->container_dev);
1453 ping_monitor(devnum2devname(st->container_dev));
1454 }
6234c63c
DW
1455 break;
1456 }
bb50e5d3
N
1457 if (!err)
1458 sysfs_set_safemode(content, content->safe_mode_delay);
6234c63c 1459 if (verbose >= 0) {
bb50e5d3
N
1460 if (err)
1461 fprintf(stderr, Name
1462 ": array %s now has %d devices",
1463 chosen_name, working + preexist);
1464 else
1465 fprintf(stderr, Name
1466 ": Started %s with %d devices",
1467 chosen_name, working + preexist);
6234c63c
DW
1468 if (preexist)
1469 fprintf(stderr, " (%d new)", working);
1470 fprintf(stderr, "\n");
1471 }
bb50e5d3 1472 if (!err)
a7c6e3fb 1473 wait_for(chosen_name, mdfd);
4408ee76 1474 close(mdfd);
7cb2aa33 1475 return 0;
6234c63c 1476 /* FIXME should have an O_EXCL and wait for read-auto */
7cb2aa33 1477 } else {
6234c63c
DW
1478 if (verbose >= 0)
1479 fprintf(stderr, Name
1480 ": %s assembled with %d devices but "
1481 "not started\n",
1482 chosen_name, working);
4408ee76 1483 close(mdfd);
7cb2aa33
N
1484 return 1;
1485 }
6234c63c
DW
1486}
1487#endif
1488