]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Assemble.c
Always set 'homehost' if not specified.
[thirdparty/mdadm.git] / Assemble.c
CommitLineData
64c4757e 1/*
9a9dab36 2 * mdadm - manage Linux "md" devices aka RAID arrays.
64c4757e 3 *
4f589ad0 4 * Copyright (C) 2001-2006 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
22 * Email: <neilb@cse.unsw.edu.au>
23 * Paper: Neil Brown
24 * School of Computer Science and Engineering
25 * The University of New South Wales
26 * Sydney, 2052
27 * Australia
28 */
29
9a9dab36 30#include "mdadm.h"
41a3b72a 31#include <ctype.h>
64c4757e 32
624920bb
NB
33static int name_matches(char *found, char *required, char *homehost)
34{
35 /* See if the name found matches the required name, possibly
36 * prefixed with 'homehost'
37 */
38 char fnd[33];
39
40 strncpy(fnd, found, 32);
41 fnd[32] = 0;
42 if (strcmp(found, required)==0)
43 return 1;
44 if (homehost) {
45 int l = strlen(homehost);
46 if (l < 32 && fnd[l] == ':' &&
47 strcmp(fnd+l+1, required)==0)
48 return 1;
49 }
50 return 0;
51}
52
7f91af49 53int Assemble(struct supertype *st, char *mddev,
8aec876d 54 mddev_ident_t ident,
06b0d786 55 mddev_dev_t devlist, char *backup_file,
64c4757e 56 int readonly, int runstop,
e5eac01f 57 char *update, char *homehost,
64c4757e
NB
58 int verbose, int force)
59{
60 /*
52826846
NB
61 * The task of Assemble is to find a collection of
62 * devices that should (according to their superblocks)
63 * form an array, and to give this collection to the MD driver.
64 * In Linux-2.4 and later, this involves submitting a
64c4757e
NB
65 * SET_ARRAY_INFO ioctl with no arg - to prepare
66 * the array - and then submit a number of
67 * ADD_NEW_DISK ioctls to add disks into
68 * the array. Finally RUN_ARRAY might
69 * be submitted to start the array.
70 *
71 * Much of the work of Assemble is in finding and/or
72 * checking the disks to make sure they look right.
73 *
4b1ac34b 74 * If mddev is not set, then scan must be set and we
64c4757e
NB
75 * read through the config file for dev+uuid mapping
76 * We recurse, setting mddev, for each device that
77 * - isn't running
4b1ac34b 78 * - has a valid uuid (or any uuid if !uuidset)
64c4757e
NB
79 *
80 * If mddev is set, we try to determine state of md.
81 * check version - must be at least 0.90.0
82 * check kernel version. must be at least 2.4.
83 * If not, we can possibly fall back on START_ARRAY
84 * Try to GET_ARRAY_INFO.
85 * If possible, give up
86 * If not, try to STOP_ARRAY just to make sure
87 *
88 * If !uuidset and scan, look in conf-file for uuid
89 * If not found, give up
b8a8ccf9 90 * If !devlist and scan and uuidset, get list of devs from conf-file
64c4757e
NB
91 *
92 * For each device:
93 * Check superblock - discard if bad
94 * Check uuid (set if we don't have one) - discard if no match
5787fa49 95 * Check superblock similarity if we have a superblock - discard if different
4b1ac34b 96 * Record events, devicenum
64c4757e 97 * This should give us a list of devices for the array
4b1ac34b 98 * We should collect the most recent event number
64c4757e
NB
99 *
100 * Count disks with recent enough event count
101 * While force && !enough disks
102 * Choose newest rejected disks, update event count
103 * mark clean and rewrite superblock
104 * If recent kernel:
105 * SET_ARRAY_INFO
106 * foreach device with recent events : ADD_NEW_DISK
107 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
108 * If old kernel:
109 * Check the device numbers in superblock are right
110 * update superblock if any changes
111 * START_ARRAY
112 *
113 */
c30e5369
N
114 int mdfd;
115 int clean;
7f91af49 116 int auto_assem = (mddev == NULL);
64c4757e 117 int old_linux = 0;
c30e5369 118 int vers = vers; /* Keep gcc quite - it really is initialised */
64c4757e
NB
119 struct {
120 char *devname;
213ee40b
NB
121 int uptodate; /* set once we decide that this device is as
122 * recent as everything else in the array.
123 */
124 struct mdinfo i;
5787fa49 125 } *devices;
56eedc1a 126 int *best = NULL; /* indexed by raid_disk */
98c6faba
NB
127 unsigned int bestcnt = 0;
128 int devcnt = 0;
129 unsigned int okcnt, sparecnt;
130 unsigned int req_cnt;
131 unsigned int i;
64c4757e 132 int most_recent = 0;
cd29a5c8 133 int chosen_drive;
52826846 134 int change = 0;
cd29a5c8 135 int inargv = 0;
bf4fb153 136 int bitmap_done;
7f91af49
N
137 int start_partial_ok = (runstop >= 0) &&
138 (force || devlist==NULL || auto_assem);
98c6faba 139 unsigned int num_devs;
5787fa49 140 mddev_dev_t tmpdev;
4b1ac34b 141 struct mdinfo info;
c30e5369 142 mdu_array_info_t tmp_inf;
265e0f17 143 char *avail;
308e1801 144 int nextspare = 0;
69207ff6
N
145 char *name;
146 int trustworthy;
8a46fe84 147
f35f2525
N
148 memset(&info, 0, sizeof(info));
149
682c7051 150 if (get_linux_version() < 2004000)
64c4757e
NB
151 old_linux = 1;
152
64c4757e 153 /*
52826846
NB
154 * If any subdevs are listed, then any that don't
155 * match ident are discarded. Remainder must all match and
156 * become the array.
157 * If no subdevs, then we scan all devices in the config file, but
158 * there must be something in the identity
64c4757e 159 */
64c4757e 160
cd29a5c8 161 if (!devlist &&
52826846
NB
162 ident->uuid_set == 0 &&
163 ident->super_minor < 0 &&
164 ident->devices == NULL) {
165 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
8a46fe84 166 mddev ? mddev : "further assembly");
52826846 167 return 1;
64c4757e 168 }
71d60c48
DW
169
170 /* if the configuration specifies a container then we use that to
171 * determine the devices and retrieve the array configuration
172 */
173#ifndef MDASSEMBLE
174 if (ident->container && ident->member) {
175 int cfd = open(ident->container, O_RDWR);
176 struct mdinfo *mdi;
177 struct supertype container;
178
179 if (verbose>0)
180 fprintf(stderr, Name ": looking to assemble member array %s"
181 " inside container %s\n", ident->member, ident->container);
182 if (cfd < 0) {
183 if (verbose>0)
184 fprintf(stderr, Name ": unable to open container %s: %s\n",
185 ident->container, strerror(errno));
186 return 1;
187 }
188
189 mdi = sysfs_read(cfd, fd2devnum(cfd), GET_VERSION);
190 if (!mdi) {
191 close(cfd);
192 if (verbose>0)
193 fprintf(stderr, Name ": unable to read container %s\n",
194 ident->container);
195 return 1;
196 }
197 container.ss = find_metadata_methods(mdi->text_version);
198 sysfs_free(mdi);
199 if (!container.ss) {
200 close(cfd);
201 fprintf(stderr, Name ": %s uses unknown metadata: %s\n",
202 ident->container, mdi->text_version);
203 return 1;
204 }
205 if (container.ss->load_super(&container, cfd, ident->container)) {
206 fprintf(stderr, Name ": Cannot load metadata for container %s\n",
207 ident->container);
208 return 1;
209 }
210
211 return Incremental_container(&container, ident->container,
215bb3f7
N
212 verbose, runstop, ident->autof,
213 LOCAL);
71d60c48
DW
214 }
215#endif
cd29a5c8 216 if (devlist == NULL)
8aec876d 217 devlist = conf_get_devs();
7f91af49 218 else if (mddev)
da6b5ca9 219 inargv = 1;
64c4757e 220
60e1bc1a 221 try_again:
c30e5369
N
222 /* We come back here when doing auto-assembly and attempting some
223 * set of devices failed. Those are now marked as ->used==2 and
224 * we ignore them and try again
225 */
60e1bc1a 226
5787fa49
NB
227 tmpdev = devlist; num_devs = 0;
228 while (tmpdev) {
da6b5ca9
NB
229 if (tmpdev->used)
230 tmpdev->used = 2;
231 else
232 num_devs++;
5787fa49
NB
233 tmpdev = tmpdev->next;
234 }
5787fa49
NB
235 devices = malloc(num_devs * sizeof(*devices));
236
82d9eba6 237 if (!st && ident->st) st = ident->st;
64c4757e 238
dab6685f 239 if (verbose>0)
82b27616 240 fprintf(stderr, Name ": looking for devices for %s\n",
8a46fe84 241 mddev ? mddev : "further assembly");
82b27616 242
811e6cbe
NB
243 /* first walk the list of devices to find a consistent set
244 * that match the criterea, if that is possible.
c30e5369 245 * We flag the ones we like with 'used'.
811e6cbe
NB
246 */
247 for (tmpdev = devlist;
248 tmpdev;
249 tmpdev = tmpdev->next) {
250 char *devname = tmpdev->devname;
64c4757e
NB
251 int dfd;
252 struct stat stb;
3da92f27 253 struct supertype *tst = dup_super(st);
52826846 254
da6b5ca9
NB
255 if (tmpdev->used > 1) continue;
256
52826846 257 if (ident->devices &&
56eedc1a 258 !match_oneof(ident->devices, devname)) {
6a41304b 259 if ((inargv && verbose>=0) || verbose > 0)
56eedc1a 260 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
52826846 261 continue;
56eedc1a 262 }
4b1ac34b 263
8b0dabea 264 dfd = dev_open(devname, O_RDONLY|O_EXCL);
64c4757e 265 if (dfd < 0) {
6a41304b 266 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 267 fprintf(stderr, Name ": cannot open device %s: %s\n",
64c4757e 268 devname, strerror(errno));
da6b5ca9 269 tmpdev->used = 2;
52826846
NB
270 } else if (fstat(dfd, &stb)< 0) {
271 /* Impossible! */
272 fprintf(stderr, Name ": fstat failed for %s: %s\n",
273 devname, strerror(errno));
da6b5ca9 274 tmpdev->used = 2;
cd29a5c8
NB
275 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
276 fprintf(stderr, Name ": %s is not a block device.\n",
52826846 277 devname);
da6b5ca9 278 tmpdev->used = 2;
82d9eba6 279 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
6a41304b 280 if ((inargv && verbose >= 0) || verbose > 0)
da6b5ca9
NB
281 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
282 devname);
283 tmpdev->used = 2;
3da92f27 284 } else if (tst->ss->load_super(tst,dfd, NULL)) {
6a41304b 285 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 286 fprintf( stderr, Name ": no RAID superblock on %s\n",
64c4757e 287 devname);
52826846 288 } else {
3da92f27 289 tst->ss->getinfo_super(tst, &info);
64c4757e 290 }
c06487ce 291 if (dfd >= 0) close(dfd);
52826846 292
838acbc2 293 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
3da92f27
NB
294 (!tst || !tst->sb ||
295 same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
6a41304b 296 if ((inargv && verbose >= 0) || verbose > 0)
52826846
NB
297 fprintf(stderr, Name ": %s has wrong uuid.\n",
298 devname);
df37ffc0 299 goto loop;
82b27616 300 }
c4f12c13 301 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
3da92f27
NB
302 (!tst || !tst->sb ||
303 name_matches(info.name, ident->name, homehost)==0)) {
6a41304b 304 if ((inargv && verbose >= 0) || verbose > 0)
947fd4dd
NB
305 fprintf(stderr, Name ": %s has wrong name.\n",
306 devname);
df37ffc0 307 goto loop;
947fd4dd 308 }
98c6faba 309 if (ident->super_minor != UnSet &&
3da92f27
NB
310 (!tst || !tst->sb ||
311 ident->super_minor != info.array.md_minor)) {
6a41304b 312 if ((inargv && verbose >= 0) || verbose > 0)
52826846 313 fprintf(stderr, Name ": %s has wrong super-minor.\n",
64c4757e 314 devname);
df37ffc0 315 goto loop;
64c4757e 316 }
98c6faba 317 if (ident->level != UnSet &&
3da92f27
NB
318 (!tst || !tst->sb ||
319 ident->level != info.array.level)) {
6a41304b 320 if ((inargv && verbose >= 0) || verbose > 0)
cd29a5c8
NB
321 fprintf(stderr, Name ": %s has wrong raid level.\n",
322 devname);
df37ffc0 323 goto loop;
cd29a5c8 324 }
98c6faba 325 if (ident->raid_disks != UnSet &&
3da92f27
NB
326 (!tst || !tst->sb ||
327 ident->raid_disks!= info.array.raid_disks)) {
6a41304b 328 if ((inargv && verbose >= 0) || verbose > 0)
cd29a5c8
NB
329 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
330 devname);
df37ffc0 331 goto loop;
cd29a5c8 332 }
c30e5369 333 if (auto_assem) {
3da92f27 334 if (tst == NULL || tst->sb == NULL)
da6b5ca9 335 continue;
da6b5ca9 336 }
83b6208e 337 /* If we are this far, then we are nearly commited to this device.
52826846 338 * If the super_block doesn't exist, or doesn't match others,
83b6208e
NB
339 * then we probably cannot continue
340 * However if one of the arrays is for the homehost, and
341 * the other isn't that can disambiguate.
52826846 342 */
52826846 343
3da92f27 344 if (!tst || !tst->sb) {
52826846
NB
345 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
346 devname);
ddd1a492
NB
347 if (st)
348 st->ss->free_super(st);
52826846
NB
349 return 1;
350 }
838acbc2 351
83b6208e 352 if (st == NULL)
3da92f27
NB
353 st = dup_super(tst);
354 if (st->minor_version == -1)
355 st->minor_version = tst->minor_version;
83b6208e
NB
356 if (st->ss != tst->ss ||
357 st->minor_version != tst->minor_version ||
64557c33 358 st->ss->compare_super(st, tst) != 0) {
83b6208e 359 /* Some mismatch. If exactly one array matches this host,
da6b5ca9
NB
360 * we can resolve on that one.
361 * Or, if we are auto assembling, we just ignore the second
362 * for now.
83b6208e 363 */
c30e5369 364 if (auto_assem)
df37ffc0 365 goto loop;
83b6208e 366 if (homehost) {
3da92f27
NB
367 int first = st->ss->match_home(st, homehost);
368 int last = tst->ss->match_home(tst, homehost);
9362c1c8
N
369 if (first != last &&
370 (first == 1 || last == 1)) {
83b6208e
NB
371 /* We can do something */
372 if (first) {/* just ignore this one */
373 if ((inargv && verbose >= 0) || verbose > 0)
374 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
375 devname);
df37ffc0 376 goto loop;
83b6208e
NB
377 } else { /* reject all those sofar */
378 mddev_dev_t td;
379 if ((inargv && verbose >= 0) || verbose > 0)
380 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
381 devname);
382 for (td=devlist; td != tmpdev; td=td->next)
383 if (td->used == 1)
384 td->used = 0;
385 tmpdev->used = 1;
df37ffc0 386 goto loop;
83b6208e
NB
387 }
388 }
389 }
52826846
NB
390 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
391 devname);
3da92f27
NB
392 tst->ss->free_super(tst);
393 st->ss->free_super(st);
52826846 394 return 1;
64c4757e
NB
395 }
396
811e6cbe 397 tmpdev->used = 1;
df37ffc0
NB
398
399 loop:
3d2b16e7
NB
400 if (tst)
401 tst->ss->free_super(tst);
811e6cbe
NB
402 }
403
c30e5369
N
404 if (!st || !st->sb)
405 return 2;
406
69207ff6
N
407 /* Now need to open array the device. Use create_mddev */
408 st->ss->getinfo_super(st, &info);
3d2c4fc7 409
69207ff6
N
410 trustworthy = FOREIGN;
411 switch (st->ss->match_home(st, homehost)) {
412 case 0:
413 trustworthy = FOREIGN;
414 name = info.name;
415 break;
416 case 1:
417 trustworthy = LOCAL;
418 name = strchr(info.name, ':');
419 if (name)
420 name++;
3d2c4fc7 421 else
69207ff6
N
422 name = info.name;
423 break;
424 case -1:
425 if (info.name[0] == 0 && info.array.level == LEVEL_CONTAINER) {
426 name = info.text_version;
427 trustworthy = METADATA;
428 } else
429 trustworthy = FOREIGN;
430 break;
c30e5369 431 }
69207ff6 432 mdfd = create_mddev(mddev, name, ident->autof, trustworthy, NULL);
c30e5369
N
433 if (mdfd < 0) {
434 st->ss->free_super(st);
435 free(devices);
436 if (auto_assem)
60e1bc1a 437 goto try_again;
c30e5369
N
438 return 1;
439 }
440 vers = md_get_version(mdfd);
441 if (vers < 9000) {
442 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
443 " Upgrade your kernel or try --build\n");
444 close(mdfd);
445 return 1;
446 }
447 if (ioctl(mdfd, GET_ARRAY_INFO, &tmp_inf)==0) {
448 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
449 mddev);
450 for (tmpdev = devlist ;
451 tmpdev && tmpdev->used != 1;
452 tmpdev = tmpdev->next)
453 ;
454 if (tmpdev && auto_assem)
455 fprintf(stderr, Name ": %s needed for %s...\n",
456 mddev, tmpdev->devname);
457 close(mdfd);
458 mdfd = -3;
459 st->ss->free_super(st);
460 free(devices);
461 if (auto_assem)
462 goto try_again;
463 return 1;
8a46fe84 464 }
c30e5369 465 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
8a46fe84 466
811e6cbe 467 /* Ok, no bad inconsistancy, we can try updating etc */
bf4fb153 468 bitmap_done = 0;
da6b5ca9 469 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
811e6cbe
NB
470 char *devname = tmpdev->devname;
471 struct stat stb;
5787fa49 472 /* looks like a good enough match to update the super block if needed */
d1f1011b 473#ifndef MDASSEMBLE
5787fa49 474 if (update) {
811e6cbe 475 int dfd;
4b1ac34b
NB
476 /* prepare useful information in info structures */
477 struct stat stb2;
3da92f27 478 struct supertype *tst;
4b1ac34b 479 fstat(mdfd, &stb2);
7d99579f
NB
480
481 if (strcmp(update, "uuid")==0 &&
482 !ident->uuid_set) {
483 int rfd;
484 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
485 read(rfd, ident->uuid, 16) != 16) {
486 *(__u32*)(ident->uuid) = random();
487 *(__u32*)(ident->uuid+1) = random();
488 *(__u32*)(ident->uuid+2) = random();
489 *(__u32*)(ident->uuid+3) = random();
490 }
491 if (rfd >= 0) close(rfd);
7d99579f 492 }
811e6cbe
NB
493 dfd = dev_open(devname, O_RDWR|O_EXCL);
494
0430ed48
NB
495 remove_partitions(dfd);
496
3da92f27
NB
497 tst = dup_super(st);
498 tst->ss->load_super(tst, dfd, NULL);
499 tst->ss->getinfo_super(tst, &info);
811e6cbe 500
7d99579f 501 memcpy(info.uuid, ident->uuid, 16);
e5eac01f 502 strcpy(info.name, ident->name);
811e6cbe
NB
503 info.array.md_minor = minor(stb2.st_rdev);
504
3da92f27
NB
505 tst->ss->update_super(tst, &info, update,
506 devname, verbose,
507 ident->uuid_set, homehost);
e5eac01f
NB
508 if (strcmp(update, "uuid")==0 &&
509 !ident->uuid_set) {
510 ident->uuid_set = 1;
511 memcpy(ident->uuid, info.uuid, 16);
512 }
b8a8ccf9 513 if (dfd < 0)
5787fa49
NB
514 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
515 devname);
3da92f27 516 else if (tst->ss->store_super(tst, dfd))
5787fa49
NB
517 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
518 devname);
519 if (dfd >= 0)
520 close(dfd);
8131b493
NB
521
522 if (strcmp(update, "uuid")==0 &&
bf4fb153 523 ident->bitmap_fd >= 0 && !bitmap_done) {
3da92f27
NB
524 if (bitmap_update_uuid(ident->bitmap_fd,
525 info.uuid,
526 tst->ss->swapuuid) != 0)
bf4fb153
NB
527 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
528 else
529 bitmap_done = 1;
530 }
3da92f27 531 tst->ss->free_super(tst);
d1f1011b
NB
532 } else
533#endif
534 {
30e1b9a5 535 struct supertype *tst = dup_super(st);
da6b5ca9
NB
536 int dfd;
537 dfd = dev_open(devname, O_RDWR|O_EXCL);
538
0430ed48
NB
539 remove_partitions(dfd);
540
3da92f27
NB
541 tst->ss->load_super(tst, dfd, NULL);
542 tst->ss->getinfo_super(tst, &info);
543 tst->ss->free_super(tst);
da6b5ca9 544 close(dfd);
5787fa49
NB
545 }
546
811e6cbe
NB
547 stat(devname, &stb);
548
dab6685f 549 if (verbose > 0)
cd29a5c8 550 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
4b1ac34b 551 devname, mddev, info.disk.raid_disk);
64c4757e 552 devices[devcnt].devname = devname;
64c4757e 553 devices[devcnt].uptodate = 0;
213ee40b
NB
554 devices[devcnt].i = info;
555 devices[devcnt].i.disk.major = major(stb.st_rdev);
556 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
64c4757e 557 if (most_recent < devcnt) {
213ee40b
NB
558 if (devices[devcnt].i.events
559 > devices[most_recent].i.events)
64c4757e
NB
560 most_recent = devcnt;
561 }
b8a8ccf9 562 if (info.array.level == -4)
5787fa49
NB
563 /* with multipath, the raid_disk from the superblock is meaningless */
564 i = devcnt;
565 else
213ee40b 566 i = devices[devcnt].i.disk.raid_disk;
308e1801
NB
567 if (i+1 == 0) {
568 if (nextspare < info.array.raid_disks)
569 nextspare = info.array.raid_disks;
570 i = nextspare++;
08110d41
NB
571 } else {
572 if (i >= info.array.raid_disks &&
573 i >= nextspare)
574 nextspare = i+1;
308e1801 575 }
98c6faba 576 if (i < 10000) {
56eedc1a 577 if (i >= bestcnt) {
98c6faba 578 unsigned int newbestcnt = i+10;
56eedc1a 579 int *newbest = malloc(sizeof(int)*newbestcnt);
98c6faba 580 unsigned int c;
56eedc1a
NB
581 for (c=0; c < newbestcnt; c++)
582 if (c < bestcnt)
583 newbest[c] = best[c];
584 else
585 newbest[c] = -1;
586 if (best)free(best);
587 best = newbest;
588 bestcnt = newbestcnt;
589 }
6a255b69 590 if (best[i] >=0 &&
213ee40b
NB
591 devices[best[i]].i.events
592 == devices[devcnt].i.events
593 && (devices[best[i]].i.disk.minor
594 != devices[devcnt].i.disk.minor)
b8ac1967
NB
595 && st->ss == &super0
596 && info.array.level != LEVEL_MULTIPATH) {
6a255b69
NB
597 /* two different devices with identical superblock.
598 * Could be a mis-detection caused by overlapping
599 * partitions. fail-safe.
600 */
601 fprintf(stderr, Name ": WARNING %s and %s appear"
602 " to have very similar superblocks.\n"
603 " If they are really different, "
604 "please --zero the superblock on one\n"
d2df86e0
NB
605 " If they are the same or overlap,"
606 " please remove one from %s.\n",
607 devices[best[i]].devname, devname,
608 inargv ? "the list" :
609 "the\n DEVICE list in mdadm.conf"
610 );
7f91af49 611 close(mdfd);
6a255b69
NB
612 return 1;
613 }
52826846 614 if (best[i] == -1
213ee40b
NB
615 || (devices[best[i]].i.events
616 < devices[devcnt].i.events))
52826846 617 best[i] = devcnt;
56eedc1a 618 }
64c4757e 619 devcnt++;
df37ffc0 620 }
4b1ac34b 621
64c4757e 622 if (devcnt == 0) {
682c7051 623 fprintf(stderr, Name ": no devices found for %s\n",
64c4757e 624 mddev);
3d2b16e7
NB
625 if (st)
626 st->ss->free_super(st);
7f91af49 627 close(mdfd);
64c4757e
NB
628 return 1;
629 }
4b1ac34b 630
3d2b16e7
NB
631 if (update && strcmp(update, "byteorder")==0)
632 st->minor_version = 90;
633
3da92f27 634 st->ss->getinfo_super(st, &info);
583315d9 635 clean = info.array.state & 1;
4b1ac34b 636
64c4757e
NB
637 /* now we have some devices that might be suitable.
638 * I wonder how many
639 */
265e0f17
NB
640 avail = malloc(info.array.raid_disks);
641 memset(avail, 0, info.array.raid_disks);
64c4757e 642 okcnt = 0;
52826846 643 sparecnt=0;
56eedc1a 644 for (i=0; i< bestcnt ;i++) {
64c4757e 645 int j = best[i];
ee04451c
NB
646 int event_margin = 1; /* always allow a difference of '1'
647 * like the kernel does
648 */
64c4757e 649 if (j < 0) continue;
d013a55e
NB
650 /* note: we ignore error flags in multipath arrays
651 * as they don't make sense
652 */
4b1ac34b 653 if (info.array.level != -4)
213ee40b
NB
654 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
655 if (!(devices[j].i.disk.state
656 & (1<<MD_DISK_FAULTY)))
aa88f531 657 sparecnt++;
d013a55e 658 continue;
aa88f531 659 }
213ee40b
NB
660 if (devices[j].i.events+event_margin >=
661 devices[most_recent].i.events) {
64c4757e 662 devices[j].uptodate = 1;
265e0f17 663 if (i < info.array.raid_disks) {
52826846 664 okcnt++;
265e0f17
NB
665 avail[i]=1;
666 } else
52826846 667 sparecnt++;
64c4757e
NB
668 }
669 }
265e0f17 670 while (force && !enough(info.array.level, info.array.raid_disks,
583315d9 671 info.array.layout, 1,
265e0f17 672 avail, okcnt)) {
64c4757e
NB
673 /* Choose the newest best drive which is
674 * not up-to-date, update the superblock
675 * and add it.
676 */
52826846 677 int fd;
3da92f27 678 struct supertype *tst;
c5a6f9a6 679 long long current_events;
cd29a5c8 680 chosen_drive = -1;
4b1ac34b 681 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
52826846
NB
682 int j = best[i];
683 if (j>=0 &&
684 !devices[j].uptodate &&
213ee40b 685 devices[j].i.events > 0 &&
52826846 686 (chosen_drive < 0 ||
213ee40b
NB
687 devices[j].i.events
688 > devices[chosen_drive].i.events))
52826846
NB
689 chosen_drive = j;
690 }
691 if (chosen_drive < 0)
692 break;
213ee40b 693 current_events = devices[chosen_drive].i.events;
c5a6f9a6 694 add_another:
dab6685f
NB
695 if (verbose >= 0)
696 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
213ee40b
NB
697 devices[chosen_drive].devname,
698 devices[chosen_drive].i.disk.raid_disk,
699 (int)(devices[chosen_drive].i.events),
700 (int)(devices[most_recent].i.events));
8b0dabea 701 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846
NB
702 if (fd < 0) {
703 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
704 devices[chosen_drive].devname);
213ee40b 705 devices[chosen_drive].i.events = 0;
52826846
NB
706 continue;
707 }
3da92f27 708 tst = dup_super(st);
60b435db 709 if (tst->ss->load_super(tst,fd, NULL)) {
52826846
NB
710 close(fd);
711 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
712 devices[chosen_drive].devname);
213ee40b 713 devices[chosen_drive].i.events = 0;
52826846
NB
714 continue;
715 }
213ee40b 716 info.events = devices[most_recent].i.events;
3da92f27 717 tst->ss->update_super(tst, &info, "force-one",
67a8c82d
NB
718 devices[chosen_drive].devname, verbose,
719 0, NULL);
4b1ac34b 720
3da92f27 721 if (tst->ss->store_super(tst, fd)) {
52826846
NB
722 close(fd);
723 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
724 devices[chosen_drive].devname);
213ee40b 725 devices[chosen_drive].i.events = 0;
3da92f27 726 tst->ss->free_super(tst);
52826846
NB
727 continue;
728 }
729 close(fd);
213ee40b 730 devices[chosen_drive].i.events = devices[most_recent].i.events;
52826846 731 devices[chosen_drive].uptodate = 1;
265e0f17 732 avail[chosen_drive] = 1;
52826846 733 okcnt++;
3da92f27 734 tst->ss->free_super(tst);
c5a6f9a6
NB
735
736 /* If there are any other drives of the same vintage,
737 * add them in as well. We can't lose and we might gain
738 */
739 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
740 int j = best[i];
741 if (j >= 0 &&
742 !devices[j].uptodate &&
213ee40b
NB
743 devices[j].i.events > 0 &&
744 devices[j].i.events == current_events) {
c5a6f9a6
NB
745 chosen_drive = j;
746 goto add_another;
747 }
748 }
64c4757e 749 }
52826846
NB
750
751 /* Now we want to look at the superblock which the kernel will base things on
752 * and compare the devices that we think are working with the devices that the
753 * superblock thinks are working.
754 * If there are differences and --force is given, then update this chosen
755 * superblock.
756 */
cd29a5c8 757 chosen_drive = -1;
3da92f27 758 st->ss->free_super(st);
56eedc1a 759 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
52826846
NB
760 int j = best[i];
761 int fd;
4b1ac34b 762
52826846
NB
763 if (j<0)
764 continue;
765 if (!devices[j].uptodate)
766 continue;
767 chosen_drive = j;
8b0dabea 768 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
52826846
NB
769 fprintf(stderr, Name ": Cannot open %s: %s\n",
770 devices[j].devname, strerror(errno));
7f91af49 771 close(mdfd);
52826846
NB
772 return 1;
773 }
3da92f27 774 if (st->ss->load_super(st,fd, NULL)) {
52826846
NB
775 close(fd);
776 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
777 devices[j].devname);
7f91af49 778 close(mdfd);
52826846
NB
779 return 1;
780 }
781 close(fd);
782 }
3da92f27 783 if (st->sb == NULL) {
4b1ac34b 784 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
7f91af49 785 close(mdfd);
4b1ac34b
NB
786 return 1;
787 }
3da92f27 788 st->ss->getinfo_super(st, &info);
f35f2525
N
789#ifndef MDASSEMBLE
790 sysfs_init(&info, mdfd, 0);
791#endif
56eedc1a 792 for (i=0; i<bestcnt; i++) {
52826846 793 int j = best[i];
98c6faba 794 unsigned int desired_state;
11a3e71d 795
4b1ac34b 796 if (i < info.array.raid_disks)
11a3e71d
NB
797 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
798 else
799 desired_state = 0;
800
52826846
NB
801 if (j<0)
802 continue;
803 if (!devices[j].uptodate)
804 continue;
4b1ac34b 805
213ee40b
NB
806 devices[j].i.disk.state = desired_state;
807
808 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
68c7d6d7 809 verbose, 0, NULL)) {
52826846 810 if (force) {
dab6685f
NB
811 if (verbose >= 0)
812 fprintf(stderr, Name ": "
813 "clearing FAULTY flag for device %d in %s for %s\n",
814 j, mddev, devices[j].devname);
4b1ac34b 815 change = 1;
52826846 816 } else {
dab6685f
NB
817 if (verbose >= -1)
818 fprintf(stderr, Name ": "
819 "device %d in %s has wrong state in superblock, but %s seems ok\n",
820 i, mddev, devices[j].devname);
52826846
NB
821 }
822 }
4b1ac34b 823#if 0
213ee40b 824 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
52826846
NB
825 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
826 i, mddev);
827 }
4b1ac34b 828#endif
52826846 829 }
c5a6f9a6
NB
830 if (force && !clean &&
831 !enough(info.array.level, info.array.raid_disks,
832 info.array.layout, clean,
833 avail, okcnt)) {
3da92f27 834 change += st->ss->update_super(st, &info, "force-array",
67a8c82d
NB
835 devices[chosen_drive].devname, verbose,
836 0, NULL);
583315d9 837 clean = 1;
d013a55e 838 }
52826846 839
4b1ac34b 840 if (change) {
52826846 841 int fd;
8b0dabea 842 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846 843 if (fd < 0) {
353632d9 844 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
52826846 845 devices[chosen_drive].devname);
7f91af49 846 close(mdfd);
52826846
NB
847 return 1;
848 }
3da92f27 849 if (st->ss->store_super(st, fd)) {
52826846
NB
850 close(fd);
851 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
852 devices[chosen_drive].devname);
7f91af49 853 close(mdfd);
52826846
NB
854 return 1;
855 }
856 close(fd);
52826846
NB
857 }
858
353632d9
NB
859 /* If we are in the middle of a reshape we may need to restore saved data
860 * that was moved aside due to the reshape overwriting live data
861 * The code of doing this lives in Grow.c
862 */
39c5a909 863#ifndef MDASSEMBLE
353632d9
NB
864 if (info.reshape_active) {
865 int err = 0;
866 int *fdlist = malloc(sizeof(int)* bestcnt);
867 for (i=0; i<bestcnt; i++) {
868 int j = best[i];
869 if (j >= 0) {
870 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
871 if (fdlist[i] < 0) {
872 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
873 devices[j].devname);
874 err = 1;
875 break;
876 }
877 } else
878 fdlist[i] = -1;
879 }
880 if (!err)
06b0d786 881 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
353632d9
NB
882 while (i>0) {
883 i--;
884 if (fdlist[i]>=0) close(fdlist[i]);
885 }
886 if (err) {
887 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
7f91af49 888 close(mdfd);
353632d9
NB
889 return err;
890 }
891 }
39c5a909 892#endif
aa88f531
NB
893 /* count number of in-sync devices according to the superblock.
894 * We must have this number to start the array without -s or -R
895 */
4b1ac34b 896 req_cnt = info.array.working_disks;
aa88f531 897
64c4757e
NB
898 /* Almost ready to actually *do* something */
899 if (!old_linux) {
fbf8a0b7 900 int rv;
a19c88b8 901
f35f2525 902 rv = set_array_info(mdfd, st, &info);
fbf8a0b7 903 if (rv) {
f35f2525 904 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
64c4757e 905 mddev, strerror(errno));
7f91af49 906 close(mdfd);
64c4757e
NB
907 return 1;
908 }
11484a63 909 if (ident->bitmap_fd >= 0) {
c82f047c
NB
910 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
911 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
7f91af49 912 close(mdfd);
c82f047c
NB
913 return 1;
914 }
7ef02d01
NB
915 } else if (ident->bitmap_file) {
916 /* From config file */
917 int bmfd = open(ident->bitmap_file, O_RDWR);
918 if (bmfd < 0) {
919 fprintf(stderr, Name ": Could not open bitmap file %s\n",
920 ident->bitmap_file);
7f91af49 921 close(mdfd);
7ef02d01
NB
922 return 1;
923 }
924 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
925 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
926 close(bmfd);
7f91af49 927 close(mdfd);
7ef02d01
NB
928 return 1;
929 }
930 close(bmfd);
c82f047c 931 }
7ef02d01 932
52826846 933 /* First, add the raid disks, but add the chosen one last */
56eedc1a 934 for (i=0; i<= bestcnt; i++) {
52826846 935 int j;
56eedc1a 936 if (i < bestcnt) {
52826846
NB
937 j = best[i];
938 if (j == chosen_drive)
939 continue;
940 } else
941 j = chosen_drive;
942
56eedc1a 943 if (j >= 0 /* && devices[j].uptodate */) {
f35f2525 944 rv = add_disk(mdfd, st, &info, &devices[j].i);
7801ac20 945
a19c88b8 946 if (rv) {
213ee40b
NB
947 fprintf(stderr, Name ": failed to add "
948 "%s to %s: %s\n",
64c4757e
NB
949 devices[j].devname,
950 mddev,
951 strerror(errno));
213ee40b
NB
952 if (i < info.array.raid_disks
953 || i == bestcnt)
52826846
NB
954 okcnt--;
955 else
956 sparecnt--;
dab6685f 957 } else if (verbose > 0)
213ee40b
NB
958 fprintf(stderr, Name ": added %s "
959 "to %s as %d\n",
960 devices[j].devname, mddev,
961 devices[j].i.disk.raid_disk);
dab6685f 962 } else if (verbose > 0 && i < info.array.raid_disks)
213ee40b
NB
963 fprintf(stderr, Name ": no uptodate device for "
964 "slot %d of %s\n",
64c4757e
NB
965 i, mddev);
966 }
aba69144 967
598f0d58
NB
968 if (info.array.level == LEVEL_CONTAINER) {
969 if (verbose >= 0) {
970 fprintf(stderr, Name ": Container %s has been "
971 "assembled with %d drive%s",
57ed8c91 972 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
598f0d58
NB
973 if (okcnt < info.array.raid_disks)
974 fprintf(stderr, " (out of %d)",
975 info.array.raid_disks);
976 fprintf(stderr, "\n");
977 }
7f91af49 978 close(mdfd);
598f0d58
NB
979 return 0;
980 }
981
64c4757e 982 if (runstop == 1 ||
b8a8ccf9 983 (runstop <= 0 &&
583315d9
NB
984 ( enough(info.array.level, info.array.raid_disks,
985 info.array.layout, clean, avail, okcnt) &&
aa88f531
NB
986 (okcnt >= req_cnt || start_partial_ok)
987 ))) {
82b27616 988 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
dab6685f
NB
989 if (verbose >= 0) {
990 fprintf(stderr, Name ": %s has been started with %d drive%s",
991 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9 992 if (okcnt < info.array.raid_disks)
dab6685f
NB
993 fprintf(stderr, " (out of %d)", info.array.raid_disks);
994 if (sparecnt)
995 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
996 fprintf(stderr, ".\n");
997 }
7f91af49
N
998 close(mdfd);
999 if (auto_assem) {
589395d6 1000 int usecs = 1;
589395d6
NB
1001 /* There is a nasty race with 'mdadm --monitor'.
1002 * If it opens this device before we close it,
1003 * it gets an incomplete open on which IO
598f0d58
NB
1004 * doesn't work and the capacity is
1005 * wrong.
589395d6
NB
1006 * If we reopen (to check for layered devices)
1007 * before --monitor closes, we loose.
1008 *
1009 * So: wait upto 1 second for there to be
1010 * a non-zero capacity.
1011 */
1012 while (usecs < 1000) {
1013 mdfd = open(mddev, O_RDONLY);
1014 if (mdfd >= 0) {
beae1dfe
NB
1015 unsigned long long size;
1016 if (get_dev_size(mdfd, NULL, &size) &&
589395d6
NB
1017 size > 0)
1018 break;
1019 close(mdfd);
1020 }
1021 usleep(usecs);
1022 usecs <<= 1;
1023 }
1024 }
64c4757e 1025 return 0;
82b27616 1026 }
682c7051 1027 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
64c4757e 1028 mddev, strerror(errno));
583315d9
NB
1029
1030 if (!enough(info.array.level, info.array.raid_disks,
1031 info.array.layout, 1, avail, okcnt))
1032 fprintf(stderr, Name ": Not enough devices to "
1033 "start the array.\n");
1034 else if (!enough(info.array.level,
1035 info.array.raid_disks,
1036 info.array.layout, clean,
1037 avail, okcnt))
1038 fprintf(stderr, Name ": Not enough devices to "
1039 "start the array while not clean "
1040 "- consider --force.\n");
1041
7f91af49 1042 if (auto_assem)
1c203a4b 1043 ioctl(mdfd, STOP_ARRAY, NULL);
7f91af49 1044 close(mdfd);
64c4757e
NB
1045 return 1;
1046 }
82b27616 1047 if (runstop == -1) {
b8a8ccf9 1048 fprintf(stderr, Name ": %s assembled from %d drive%s",
52826846 1049 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9
NB
1050 if (okcnt != info.array.raid_disks)
1051 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1052 fprintf(stderr, ", but not started.\n");
7f91af49 1053 close(mdfd);
64c4757e 1054 return 0;
82b27616 1055 }
4855f95c 1056 if (verbose >= -1) {
dab6685f
NB
1057 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1058 if (sparecnt)
1059 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
583315d9
NB
1060 if (!enough(info.array.level, info.array.raid_disks,
1061 info.array.layout, 1, avail, okcnt))
dab6685f 1062 fprintf(stderr, " - not enough to start the array.\n");
583315d9
NB
1063 else if (!enough(info.array.level,
1064 info.array.raid_disks,
1065 info.array.layout, clean,
1066 avail, okcnt))
1067 fprintf(stderr, " - not enough to start the "
1068 "array while not clean - consider "
1069 "--force.\n");
dab6685f
NB
1070 else {
1071 if (req_cnt == info.array.raid_disks)
1072 fprintf(stderr, " - need all %d to start it", req_cnt);
1073 else
1074 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1075 fprintf(stderr, " (use --run to insist).\n");
1076 }
aa88f531 1077 }
7f91af49 1078 if (auto_assem)
1c203a4b 1079 ioctl(mdfd, STOP_ARRAY, NULL);
64c4757e 1080 return 1;
82b27616 1081 } else {
52826846
NB
1082 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1083 * been updated to point to the current locations of devices.
1084 * so we can just start the array
82b27616 1085 */
cd29a5c8 1086 unsigned long dev;
213ee40b
NB
1087 dev = makedev(devices[chosen_drive].i.disk.major,
1088 devices[chosen_drive].i.disk.minor);
82b27616
NB
1089 if (ioctl(mdfd, START_ARRAY, dev)) {
1090 fprintf(stderr, Name ": Cannot start array: %s\n",
1091 strerror(errno));
1092 }
aba69144 1093
64c4757e 1094 }
7f91af49 1095 close(mdfd);
e0d19036 1096 return 0;
64c4757e 1097}