]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Assemble.c
Spare migration tests
[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 870 if (i < content->array.raid_disks) {
dcc4210f
DW
871 if (devices[j].i.recovery_start == MaxSector ||
872 (content->reshape_active &&
873 j >= content->array.raid_disks - content->delta_disks)) {
1ff98339
N
874 okcnt++;
875 avail[i]=1;
876 } else
877 rebuilding_cnt++;
265e0f17 878 } else
52826846 879 sparecnt++;
64c4757e
NB
880 }
881 }
02e7c5b7 882 free(devmap);
98dbd966
DW
883 while (force && !enough(content->array.level, content->array.raid_disks,
884 content->array.layout, 1,
265e0f17 885 avail, okcnt)) {
64c4757e
NB
886 /* Choose the newest best drive which is
887 * not up-to-date, update the superblock
888 * and add it.
889 */
52826846 890 int fd;
3da92f27 891 struct supertype *tst;
f21e18ca 892 unsigned long long current_events;
cd29a5c8 893 chosen_drive = -1;
f21e18ca 894 for (i = 0; i < content->array.raid_disks && i < bestcnt; i++) {
52826846
NB
895 int j = best[i];
896 if (j>=0 &&
897 !devices[j].uptodate &&
921d9e16 898 devices[j].i.recovery_start == MaxSector &&
52826846 899 (chosen_drive < 0 ||
213ee40b
NB
900 devices[j].i.events
901 > devices[chosen_drive].i.events))
52826846
NB
902 chosen_drive = j;
903 }
904 if (chosen_drive < 0)
905 break;
213ee40b 906 current_events = devices[chosen_drive].i.events;
c5a6f9a6 907 add_another:
dab6685f
NB
908 if (verbose >= 0)
909 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
213ee40b
NB
910 devices[chosen_drive].devname,
911 devices[chosen_drive].i.disk.raid_disk,
912 (int)(devices[chosen_drive].i.events),
913 (int)(devices[most_recent].i.events));
8b0dabea 914 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846
NB
915 if (fd < 0) {
916 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
917 devices[chosen_drive].devname);
213ee40b 918 devices[chosen_drive].i.events = 0;
52826846
NB
919 continue;
920 }
3da92f27 921 tst = dup_super(st);
60b435db 922 if (tst->ss->load_super(tst,fd, NULL)) {
52826846
NB
923 close(fd);
924 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
925 devices[chosen_drive].devname);
213ee40b 926 devices[chosen_drive].i.events = 0;
52826846
NB
927 continue;
928 }
98dbd966
DW
929 content->events = devices[most_recent].i.events;
930 tst->ss->update_super(tst, content, "force-one",
67a8c82d
NB
931 devices[chosen_drive].devname, verbose,
932 0, NULL);
4b1ac34b 933
3da92f27 934 if (tst->ss->store_super(tst, fd)) {
52826846
NB
935 close(fd);
936 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
937 devices[chosen_drive].devname);
213ee40b 938 devices[chosen_drive].i.events = 0;
3da92f27 939 tst->ss->free_super(tst);
52826846
NB
940 continue;
941 }
942 close(fd);
213ee40b 943 devices[chosen_drive].i.events = devices[most_recent].i.events;
52826846 944 devices[chosen_drive].uptodate = 1;
265e0f17 945 avail[chosen_drive] = 1;
52826846 946 okcnt++;
3da92f27 947 tst->ss->free_super(tst);
c5a6f9a6
NB
948
949 /* If there are any other drives of the same vintage,
950 * add them in as well. We can't lose and we might gain
951 */
f21e18ca 952 for (i = 0; i < content->array.raid_disks && i < bestcnt ; i++) {
c5a6f9a6
NB
953 int j = best[i];
954 if (j >= 0 &&
955 !devices[j].uptodate &&
213ee40b 956 devices[j].i.events == current_events) {
c5a6f9a6
NB
957 chosen_drive = j;
958 goto add_another;
959 }
960 }
64c4757e 961 }
52826846
NB
962
963 /* Now we want to look at the superblock which the kernel will base things on
964 * and compare the devices that we think are working with the devices that the
965 * superblock thinks are working.
966 * If there are differences and --force is given, then update this chosen
967 * superblock.
968 */
cd29a5c8 969 chosen_drive = -1;
3da92f27 970 st->ss->free_super(st);
56eedc1a 971 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
52826846
NB
972 int j = best[i];
973 int fd;
4b1ac34b 974
52826846
NB
975 if (j<0)
976 continue;
977 if (!devices[j].uptodate)
978 continue;
979 chosen_drive = j;
8b0dabea 980 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
52826846
NB
981 fprintf(stderr, Name ": Cannot open %s: %s\n",
982 devices[j].devname, strerror(errno));
7f91af49 983 close(mdfd);
d7f7ebb7 984 free(devices);
52826846
NB
985 return 1;
986 }
3da92f27 987 if (st->ss->load_super(st,fd, NULL)) {
52826846
NB
988 close(fd);
989 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
990 devices[j].devname);
7f91af49 991 close(mdfd);
d7f7ebb7 992 free(devices);
52826846
NB
993 return 1;
994 }
995 close(fd);
996 }
3da92f27 997 if (st->sb == NULL) {
4b1ac34b 998 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
7f91af49 999 close(mdfd);
d7f7ebb7 1000 free(devices);
4b1ac34b
NB
1001 return 1;
1002 }
a5d85af7 1003 st->ss->getinfo_super(st, content, NULL);
f35f2525 1004#ifndef MDASSEMBLE
98dbd966 1005 sysfs_init(content, mdfd, 0);
f35f2525 1006#endif
56eedc1a 1007 for (i=0; i<bestcnt; i++) {
52826846 1008 int j = best[i];
98c6faba 1009 unsigned int desired_state;
11a3e71d 1010
98dbd966 1011 if (i < content->array.raid_disks)
11a3e71d
NB
1012 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1013 else
1014 desired_state = 0;
1015
52826846
NB
1016 if (j<0)
1017 continue;
1018 if (!devices[j].uptodate)
1019 continue;
4b1ac34b 1020
213ee40b 1021 devices[j].i.disk.state = desired_state;
4e9a6ff7
N
1022 if (!(devices[j].i.array.state & 1))
1023 clean = 0;
213ee40b
NB
1024
1025 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
68c7d6d7 1026 verbose, 0, NULL)) {
52826846 1027 if (force) {
dab6685f
NB
1028 if (verbose >= 0)
1029 fprintf(stderr, Name ": "
1030 "clearing FAULTY flag for device %d in %s for %s\n",
1031 j, mddev, devices[j].devname);
4b1ac34b 1032 change = 1;
52826846 1033 } else {
dab6685f
NB
1034 if (verbose >= -1)
1035 fprintf(stderr, Name ": "
1036 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1037 i, mddev, devices[j].devname);
52826846
NB
1038 }
1039 }
4b1ac34b 1040#if 0
213ee40b 1041 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
52826846
NB
1042 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1043 i, mddev);
1044 }
4b1ac34b 1045#endif
52826846 1046 }
c5a6f9a6 1047 if (force && !clean &&
98dbd966
DW
1048 !enough(content->array.level, content->array.raid_disks,
1049 content->array.layout, clean,
c5a6f9a6 1050 avail, okcnt)) {
98dbd966 1051 change += st->ss->update_super(st, content, "force-array",
67a8c82d
NB
1052 devices[chosen_drive].devname, verbose,
1053 0, NULL);
583315d9 1054 clean = 1;
d013a55e 1055 }
52826846 1056
4b1ac34b 1057 if (change) {
52826846 1058 int fd;
8b0dabea 1059 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846 1060 if (fd < 0) {
353632d9 1061 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
52826846 1062 devices[chosen_drive].devname);
7f91af49 1063 close(mdfd);
d7f7ebb7 1064 free(devices);
52826846
NB
1065 return 1;
1066 }
3da92f27 1067 if (st->ss->store_super(st, fd)) {
52826846
NB
1068 close(fd);
1069 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
1070 devices[chosen_drive].devname);
7f91af49 1071 close(mdfd);
d7f7ebb7 1072 free(devices);
52826846
NB
1073 return 1;
1074 }
1075 close(fd);
52826846
NB
1076 }
1077
353632d9
NB
1078 /* If we are in the middle of a reshape we may need to restore saved data
1079 * that was moved aside due to the reshape overwriting live data
1080 * The code of doing this lives in Grow.c
1081 */
39c5a909 1082#ifndef MDASSEMBLE
98dbd966 1083 if (content->reshape_active) {
353632d9
NB
1084 int err = 0;
1085 int *fdlist = malloc(sizeof(int)* bestcnt);
cd77ac4e 1086 if (verbose > 0)
ea0ebe96
N
1087 fprintf(stderr, Name ":%s has an active reshape - checking "
1088 "if critical section needs to be restored\n",
1089 chosen_name);
353632d9
NB
1090 for (i=0; i<bestcnt; i++) {
1091 int j = best[i];
1092 if (j >= 0) {
1093 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1094 if (fdlist[i] < 0) {
1095 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1096 devices[j].devname);
1097 err = 1;
1098 break;
1099 }
1100 } else
1101 fdlist[i] = -1;
1102 }
1103 if (!err)
cd77ac4e 1104 err = Grow_restart(st, content, fdlist, bestcnt, backup_file, verbose > 0);
353632d9
NB
1105 while (i>0) {
1106 i--;
1107 if (fdlist[i]>=0) close(fdlist[i]);
1108 }
1109 if (err) {
1110 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
e9e43ec3
N
1111 if (backup_file == NULL)
1112 fprintf(stderr," Possibly you needed to specify the --backup-file\n");
7f91af49 1113 close(mdfd);
d7f7ebb7 1114 free(devices);
353632d9
NB
1115 return err;
1116 }
1117 }
39c5a909 1118#endif
aa88f531
NB
1119 /* count number of in-sync devices according to the superblock.
1120 * We must have this number to start the array without -s or -R
1121 */
98dbd966 1122 req_cnt = content->array.working_disks;
aa88f531 1123
64c4757e
NB
1124 /* Almost ready to actually *do* something */
1125 if (!old_linux) {
fbf8a0b7 1126 int rv;
a19c88b8 1127
a04d5763
N
1128 /* First, fill in the map, so that udev can find our name
1129 * as soon as we become active.
1130 */
98dbd966
DW
1131 map_update(NULL, fd2devnum(mdfd), content->text_version,
1132 content->uuid, chosen_name);
a04d5763 1133
98dbd966 1134 rv = set_array_info(mdfd, st, content);
fbf8a0b7 1135 if (rv) {
f35f2525 1136 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
64c4757e 1137 mddev, strerror(errno));
24af7a87 1138 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1139 close(mdfd);
d7f7ebb7 1140 free(devices);
64c4757e
NB
1141 return 1;
1142 }
11484a63 1143 if (ident->bitmap_fd >= 0) {
c82f047c
NB
1144 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1145 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
24af7a87 1146 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1147 close(mdfd);
d7f7ebb7 1148 free(devices);
c82f047c
NB
1149 return 1;
1150 }
7ef02d01
NB
1151 } else if (ident->bitmap_file) {
1152 /* From config file */
1153 int bmfd = open(ident->bitmap_file, O_RDWR);
1154 if (bmfd < 0) {
1155 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1156 ident->bitmap_file);
24af7a87 1157 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1158 close(mdfd);
d7f7ebb7 1159 free(devices);
7ef02d01
NB
1160 return 1;
1161 }
1162 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1163 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1164 close(bmfd);
24af7a87 1165 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1166 close(mdfd);
d7f7ebb7 1167 free(devices);
7ef02d01
NB
1168 return 1;
1169 }
1170 close(bmfd);
c82f047c 1171 }
7ef02d01 1172
52826846 1173 /* First, add the raid disks, but add the chosen one last */
56eedc1a 1174 for (i=0; i<= bestcnt; i++) {
52826846 1175 int j;
56eedc1a 1176 if (i < bestcnt) {
52826846
NB
1177 j = best[i];
1178 if (j == chosen_drive)
1179 continue;
1180 } else
1181 j = chosen_drive;
1182
56eedc1a 1183 if (j >= 0 /* && devices[j].uptodate */) {
98dbd966 1184 rv = add_disk(mdfd, st, content, &devices[j].i);
7801ac20 1185
a19c88b8 1186 if (rv) {
213ee40b
NB
1187 fprintf(stderr, Name ": failed to add "
1188 "%s to %s: %s\n",
64c4757e
NB
1189 devices[j].devname,
1190 mddev,
1191 strerror(errno));
98dbd966 1192 if (i < content->array.raid_disks
213ee40b 1193 || i == bestcnt)
52826846
NB
1194 okcnt--;
1195 else
1196 sparecnt--;
dab6685f 1197 } else if (verbose > 0)
213ee40b
NB
1198 fprintf(stderr, Name ": added %s "
1199 "to %s as %d\n",
1200 devices[j].devname, mddev,
1201 devices[j].i.disk.raid_disk);
98dbd966 1202 } else if (verbose > 0 && i < content->array.raid_disks)
213ee40b
NB
1203 fprintf(stderr, Name ": no uptodate device for "
1204 "slot %d of %s\n",
64c4757e
NB
1205 i, mddev);
1206 }
aba69144 1207
98dbd966 1208 if (content->array.level == LEVEL_CONTAINER) {
598f0d58
NB
1209 if (verbose >= 0) {
1210 fprintf(stderr, Name ": Container %s has been "
1211 "assembled with %d drive%s",
57ed8c91 1212 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
f21e18ca 1213 if (okcnt < (unsigned)content->array.raid_disks)
598f0d58 1214 fprintf(stderr, " (out of %d)",
98dbd966 1215 content->array.raid_disks);
598f0d58
NB
1216 fprintf(stderr, "\n");
1217 }
98dbd966 1218 sysfs_uevent(content, "change");
a7c6e3fb 1219 wait_for(chosen_name, mdfd);
7f91af49 1220 close(mdfd);
d7f7ebb7 1221 free(devices);
598f0d58
NB
1222 return 0;
1223 }
1224
64c4757e 1225 if (runstop == 1 ||
b8a8ccf9 1226 (runstop <= 0 &&
98dbd966
DW
1227 ( enough(content->array.level, content->array.raid_disks,
1228 content->array.layout, clean, avail, okcnt) &&
1ff98339 1229 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
aa88f531 1230 ))) {
e9e43ec3
N
1231 /* This array is good-to-go.
1232 * If a reshape is in progress then we might need to
1233 * continue monitoring it. In that case we start
1234 * it read-only and let the grow code make it writable.
1235 */
1236 int rv;
eb3929a4 1237#ifndef MDASSEMBLE
e9e43ec3
N
1238 if (content->reshape_active &&
1239 content->delta_disks <= 0)
1240 rv = Grow_continue(mdfd, st, content, backup_file);
1241 else
eb3929a4 1242#endif
e9e43ec3
N
1243 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1244 if (rv == 0) {
dab6685f
NB
1245 if (verbose >= 0) {
1246 fprintf(stderr, Name ": %s has been started with %d drive%s",
1247 mddev, okcnt, okcnt==1?"":"s");
f21e18ca 1248 if (okcnt < (unsigned)content->array.raid_disks)
98dbd966 1249 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1ff98339
N
1250 if (rebuilding_cnt)
1251 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
dab6685f
NB
1252 if (sparecnt)
1253 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1254 fprintf(stderr, ".\n");
1255 }
8a659c33
N
1256 if (content->reshape_active &&
1257 content->array.level >= 4 &&
1258 content->array.level <= 6) {
acee8e89
N
1259 /* might need to increase the size
1260 * of the stripe cache - default is 256
1261 */
8a659c33 1262 if (256 < 4 * (content->array.chunk_size/4096)) {
acee8e89
N
1263 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1264 if (sra)
1265 sysfs_set_num(sra, NULL,
1266 "stripe_cache_size",
8a659c33 1267 (4 * content->array.chunk_size / 4096) + 1);
acee8e89
N
1268 }
1269 }
7e83544b
N
1270 if (okcnt < (unsigned)content->array.raid_disks) {
1271 /* If any devices did not get added
1272 * because the kernel rejected them based
1273 * on event count, try adding them
1274 * again providing the action policy is
1275 * 're-add' or greater. The bitmap
1276 * might allow them to be included, or
1277 * they will become spares.
1278 */
1279 for (i = 0; i <= bestcnt; i++) {
1280 int j = best[i];
1281 if (j >= 0 && !devices[j].uptodate) {
1282 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1283 continue;
1284 rv = add_disk(mdfd, st, content,
1285 &devices[j].i);
1286 if (rv == 0 && verbose >= 0)
1287 fprintf(stderr,
1288 Name ": %s has been re-added.\n",
1289 devices[j].devname);
1290 }
1291 }
1292 }
a7c6e3fb 1293 wait_for(mddev, mdfd);
7f91af49
N
1294 close(mdfd);
1295 if (auto_assem) {
589395d6 1296 int usecs = 1;
589395d6
NB
1297 /* There is a nasty race with 'mdadm --monitor'.
1298 * If it opens this device before we close it,
1299 * it gets an incomplete open on which IO
598f0d58
NB
1300 * doesn't work and the capacity is
1301 * wrong.
589395d6
NB
1302 * If we reopen (to check for layered devices)
1303 * before --monitor closes, we loose.
1304 *
1305 * So: wait upto 1 second for there to be
1306 * a non-zero capacity.
1307 */
1308 while (usecs < 1000) {
1309 mdfd = open(mddev, O_RDONLY);
1310 if (mdfd >= 0) {
beae1dfe
NB
1311 unsigned long long size;
1312 if (get_dev_size(mdfd, NULL, &size) &&
589395d6
NB
1313 size > 0)
1314 break;
1315 close(mdfd);
1316 }
1317 usleep(usecs);
1318 usecs <<= 1;
1319 }
1320 }
d7f7ebb7 1321 free(devices);
64c4757e 1322 return 0;
82b27616 1323 }
682c7051 1324 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
64c4757e 1325 mddev, strerror(errno));
583315d9 1326
98dbd966
DW
1327 if (!enough(content->array.level, content->array.raid_disks,
1328 content->array.layout, 1, avail, okcnt))
583315d9
NB
1329 fprintf(stderr, Name ": Not enough devices to "
1330 "start the array.\n");
98dbd966
DW
1331 else if (!enough(content->array.level,
1332 content->array.raid_disks,
1333 content->array.layout, clean,
583315d9
NB
1334 avail, okcnt))
1335 fprintf(stderr, Name ": Not enough devices to "
1336 "start the array while not clean "
1337 "- consider --force.\n");
1338
7f91af49 1339 if (auto_assem)
1c203a4b 1340 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1341 close(mdfd);
d7f7ebb7 1342 free(devices);
64c4757e
NB
1343 return 1;
1344 }
82b27616 1345 if (runstop == -1) {
b8a8ccf9 1346 fprintf(stderr, Name ": %s assembled from %d drive%s",
52826846 1347 mddev, okcnt, okcnt==1?"":"s");
f21e18ca 1348 if (okcnt != (unsigned)content->array.raid_disks)
98dbd966 1349 fprintf(stderr, " (out of %d)", content->array.raid_disks);
b8a8ccf9 1350 fprintf(stderr, ", but not started.\n");
7f91af49 1351 close(mdfd);
d7f7ebb7 1352 free(devices);
64c4757e 1353 return 0;
82b27616 1354 }
4855f95c 1355 if (verbose >= -1) {
dab6685f 1356 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1ff98339
N
1357 if (rebuilding_cnt)
1358 fprintf(stderr, "%s %d rebuilding", sparecnt?", ":" and ", rebuilding_cnt);
dab6685f
NB
1359 if (sparecnt)
1360 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
98dbd966
DW
1361 if (!enough(content->array.level, content->array.raid_disks,
1362 content->array.layout, 1, avail, okcnt))
dab6685f 1363 fprintf(stderr, " - not enough to start the array.\n");
98dbd966
DW
1364 else if (!enough(content->array.level,
1365 content->array.raid_disks,
1366 content->array.layout, clean,
583315d9
NB
1367 avail, okcnt))
1368 fprintf(stderr, " - not enough to start the "
1369 "array while not clean - consider "
1370 "--force.\n");
dab6685f 1371 else {
f21e18ca 1372 if (req_cnt == (unsigned)content->array.raid_disks)
dab6685f
NB
1373 fprintf(stderr, " - need all %d to start it", req_cnt);
1374 else
98dbd966 1375 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
dab6685f
NB
1376 fprintf(stderr, " (use --run to insist).\n");
1377 }
aa88f531 1378 }
7f91af49 1379 if (auto_assem)
1c203a4b 1380 ioctl(mdfd, STOP_ARRAY, NULL);
56a8da69 1381 close(mdfd);
d7f7ebb7 1382 free(devices);
64c4757e 1383 return 1;
82b27616 1384 } else {
52826846
NB
1385 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1386 * been updated to point to the current locations of devices.
1387 * so we can just start the array
82b27616 1388 */
cd29a5c8 1389 unsigned long dev;
213ee40b
NB
1390 dev = makedev(devices[chosen_drive].i.disk.major,
1391 devices[chosen_drive].i.disk.minor);
82b27616
NB
1392 if (ioctl(mdfd, START_ARRAY, dev)) {
1393 fprintf(stderr, Name ": Cannot start array: %s\n",
1394 strerror(errno));
1395 }
aba69144 1396
64c4757e 1397 }
7f91af49 1398 close(mdfd);
d7f7ebb7 1399 free(devices);
e0d19036 1400 return 0;
64c4757e 1401}
6234c63c
DW
1402
1403#ifndef MDASSEMBLE
1404int assemble_container_content(struct supertype *st, int mdfd,
1405 struct mdinfo *content, int runstop,
1406 char *chosen_name, int verbose)
1407{
1408 struct mdinfo *dev, *sra;
1409 int working = 0, preexist = 0;
1410 struct map_ent *map = NULL;
1411
1412 sysfs_init(content, mdfd, 0);
1413
1414 sra = sysfs_read(mdfd, 0, GET_VERSION);
1415 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
4408ee76
N
1416 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1417 close(mdfd);
6234c63c 1418 return 1;
4408ee76 1419 }
6234c63c
DW
1420 if (sra)
1421 sysfs_free(sra);
1422
1423 for (dev = content->devs; dev; dev = dev->next)
462906cd 1424 if (sysfs_add_disk(content, dev, 1) == 0)
6234c63c
DW
1425 working++;
1426 else if (errno == EEXIST)
1427 preexist++;
4408ee76
N
1428 if (working == 0) {
1429 close(mdfd);
7cb2aa33 1430 return 1;/* Nothing new, don't try to start */
8b4e5ea9
N
1431 }
1432
1433 map_update(&map, fd2devnum(mdfd),
1434 content->text_version,
1435 content->uuid, chosen_name);
1436
1437 if (runstop > 0 ||
6234c63c 1438 (working + preexist) >= content->array.working_disks) {
bb50e5d3 1439 int err;
a714580e 1440
6234c63c
DW
1441 switch(content->array.level) {
1442 case LEVEL_LINEAR:
1443 case LEVEL_MULTIPATH:
1444 case 0:
bb50e5d3
N
1445 err = sysfs_set_str(content, NULL, "array_state",
1446 "active");
6234c63c
DW
1447 break;
1448 default:
bb50e5d3 1449 err = sysfs_set_str(content, NULL, "array_state",
6234c63c
DW
1450 "readonly");
1451 /* start mdmon if needed. */
bb50e5d3
N
1452 if (!err) {
1453 if (!mdmon_running(st->container_dev))
1454 start_mdmon(st->container_dev);
1455 ping_monitor(devnum2devname(st->container_dev));
1456 }
6234c63c
DW
1457 break;
1458 }
bb50e5d3
N
1459 if (!err)
1460 sysfs_set_safemode(content, content->safe_mode_delay);
6234c63c 1461 if (verbose >= 0) {
bb50e5d3
N
1462 if (err)
1463 fprintf(stderr, Name
1464 ": array %s now has %d devices",
1465 chosen_name, working + preexist);
1466 else
1467 fprintf(stderr, Name
1468 ": Started %s with %d devices",
1469 chosen_name, working + preexist);
6234c63c
DW
1470 if (preexist)
1471 fprintf(stderr, " (%d new)", working);
1472 fprintf(stderr, "\n");
1473 }
bb50e5d3 1474 if (!err)
a7c6e3fb 1475 wait_for(chosen_name, mdfd);
4408ee76 1476 close(mdfd);
7cb2aa33 1477 return 0;
6234c63c 1478 /* FIXME should have an O_EXCL and wait for read-auto */
7cb2aa33 1479 } else {
6234c63c
DW
1480 if (verbose >= 0)
1481 fprintf(stderr, Name
1482 ": %s assembled with %d devices but "
1483 "not started\n",
1484 chosen_name, working);
4408ee76 1485 close(mdfd);
7cb2aa33
N
1486 return 1;
1487 }
6234c63c
DW
1488}
1489#endif
1490