]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Assemble.c
Fixed problems that could cause infinitel loop with auto assemble.
[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
82d9eba6 53int Assemble(struct supertype *st, char *mddev, int mdfd,
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 */
d55e3aef 114 int must_close = 0;
64c4757e 115 int old_linux = 0;
41a3b72a 116 int vers = 0; /* Keep gcc quite - it really is initialised */
4b1ac34b 117 void *first_super = NULL, *super = NULL;
64c4757e
NB
118 struct {
119 char *devname;
98c6faba
NB
120 unsigned int major, minor;
121 unsigned int oldmajor, oldminor;
64c4757e 122 long long events;
64c4757e 123 int uptodate;
d013a55e 124 int state;
52826846 125 int raid_disk;
6a41304b 126 int disk_nr;
5787fa49 127 } *devices;
56eedc1a 128 int *best = NULL; /* indexed by raid_disk */
98c6faba
NB
129 unsigned int bestcnt = 0;
130 int devcnt = 0;
131 unsigned int okcnt, sparecnt;
132 unsigned int req_cnt;
133 unsigned int i;
64c4757e 134 int most_recent = 0;
cd29a5c8 135 int chosen_drive;
52826846 136 int change = 0;
cd29a5c8 137 int inargv = 0;
589395d6 138 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
98c6faba 139 unsigned int num_devs;
5787fa49 140 mddev_dev_t tmpdev;
4b1ac34b 141 struct mdinfo info;
265e0f17 142 char *avail;
308e1801 143 int nextspare = 0;
8a46fe84 144
682c7051 145 if (get_linux_version() < 2004000)
64c4757e
NB
146 old_linux = 1;
147
8a46fe84
NB
148 if (mdfd >= 0) {
149 vers = md_get_version(mdfd);
150 if (vers <= 0) {
151 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
152 return 1;
153 }
154 if (vers < 9000) {
155 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
156 " Upgrade your kernel or try --build\n");
157 return 1;
158 }
64c4757e 159
8a46fe84
NB
160 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
161 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
162 mddev);
163 return 1;
164 }
165 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
166 }
64c4757e 167 /*
52826846
NB
168 * If any subdevs are listed, then any that don't
169 * match ident are discarded. Remainder must all match and
170 * become the array.
171 * If no subdevs, then we scan all devices in the config file, but
172 * there must be something in the identity
64c4757e 173 */
64c4757e 174
cd29a5c8 175 if (!devlist &&
52826846
NB
176 ident->uuid_set == 0 &&
177 ident->super_minor < 0 &&
178 ident->devices == NULL) {
179 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
8a46fe84 180 mddev ? mddev : "further assembly");
52826846 181 return 1;
64c4757e 182 }
cd29a5c8 183 if (devlist == NULL)
8aec876d 184 devlist = conf_get_devs();
da6b5ca9
NB
185 else if (mdfd >= 0)
186 inargv = 1;
64c4757e 187
60e1bc1a
NB
188 try_again:
189
5787fa49
NB
190 tmpdev = devlist; num_devs = 0;
191 while (tmpdev) {
da6b5ca9
NB
192 if (tmpdev->used)
193 tmpdev->used = 2;
194 else
195 num_devs++;
5787fa49
NB
196 tmpdev = tmpdev->next;
197 }
5787fa49
NB
198 devices = malloc(num_devs * sizeof(*devices));
199
82d9eba6 200 if (!st && ident->st) st = ident->st;
64c4757e 201
dab6685f 202 if (verbose>0)
82b27616 203 fprintf(stderr, Name ": looking for devices for %s\n",
8a46fe84 204 mddev ? mddev : "further assembly");
82b27616 205
811e6cbe
NB
206 /* first walk the list of devices to find a consistent set
207 * that match the criterea, if that is possible.
208 * We flag the one we like with 'used'.
209 */
210 for (tmpdev = devlist;
211 tmpdev;
212 tmpdev = tmpdev->next) {
213 char *devname = tmpdev->devname;
64c4757e
NB
214 int dfd;
215 struct stat stb;
82d9eba6 216 struct supertype *tst = st;
52826846 217
da6b5ca9
NB
218 if (tmpdev->used > 1) continue;
219
52826846 220 if (ident->devices &&
56eedc1a 221 !match_oneof(ident->devices, devname)) {
6a41304b 222 if ((inargv && verbose>=0) || verbose > 0)
56eedc1a 223 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
52826846 224 continue;
56eedc1a 225 }
4b1ac34b
NB
226
227 if (super) {
228 free(super);
229 super = NULL;
230 }
52826846 231
8b0dabea 232 dfd = dev_open(devname, O_RDONLY|O_EXCL);
64c4757e 233 if (dfd < 0) {
6a41304b 234 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 235 fprintf(stderr, Name ": cannot open device %s: %s\n",
64c4757e 236 devname, strerror(errno));
da6b5ca9 237 tmpdev->used = 2;
52826846
NB
238 } else if (fstat(dfd, &stb)< 0) {
239 /* Impossible! */
240 fprintf(stderr, Name ": fstat failed for %s: %s\n",
241 devname, strerror(errno));
da6b5ca9 242 tmpdev->used = 2;
cd29a5c8
NB
243 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
244 fprintf(stderr, Name ": %s is not a block device.\n",
52826846 245 devname);
da6b5ca9 246 tmpdev->used = 2;
82d9eba6 247 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
6a41304b 248 if ((inargv && verbose >= 0) || verbose > 0)
da6b5ca9
NB
249 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
250 devname);
251 tmpdev->used = 2;
82d9eba6 252 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
6a41304b 253 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 254 fprintf( stderr, Name ": no RAID superblock on %s\n",
64c4757e 255 devname);
52826846 256 } else {
31317663 257 tst->ss->getinfo_super(&info, super);
64c4757e 258 }
c06487ce 259 if (dfd >= 0) close(dfd);
52826846 260
838acbc2 261 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
f277ce36 262 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
6a41304b 263 if ((inargv && verbose >= 0) || verbose > 0)
52826846
NB
264 fprintf(stderr, Name ": %s has wrong uuid.\n",
265 devname);
266 continue;
82b27616 267 }
c4f12c13 268 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
624920bb 269 (!super || name_matches(info.name, ident->name, homehost)==0)) {
6a41304b 270 if ((inargv && verbose >= 0) || verbose > 0)
947fd4dd
NB
271 fprintf(stderr, Name ": %s has wrong name.\n",
272 devname);
273 continue;
274 }
98c6faba 275 if (ident->super_minor != UnSet &&
4b1ac34b 276 (!super || ident->super_minor != info.array.md_minor)) {
6a41304b 277 if ((inargv && verbose >= 0) || verbose > 0)
52826846 278 fprintf(stderr, Name ": %s has wrong super-minor.\n",
64c4757e
NB
279 devname);
280 continue;
281 }
98c6faba 282 if (ident->level != UnSet &&
4b1ac34b 283 (!super|| ident->level != info.array.level)) {
6a41304b 284 if ((inargv && verbose >= 0) || verbose > 0)
cd29a5c8
NB
285 fprintf(stderr, Name ": %s has wrong raid level.\n",
286 devname);
287 continue;
288 }
98c6faba 289 if (ident->raid_disks != UnSet &&
4b1ac34b 290 (!super || ident->raid_disks!= info.array.raid_disks)) {
6a41304b 291 if ((inargv && verbose >= 0) || verbose > 0)
cd29a5c8
NB
292 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
293 devname);
294 continue;
295 }
da6b5ca9
NB
296 if (mdfd < 0) {
297 if (tst == NULL || super == NULL)
298 continue;
589395d6
NB
299 if (update == NULL &&
300 tst->ss->match_home(super, homehost)==0) {
da6b5ca9
NB
301 if ((inargv && verbose >= 0) || verbose > 0)
302 fprintf(stderr, Name ": %s is not built for host %s.\n",
303 devname, homehost);
304 /* Auto-assemble, and this is not a usable host */
589395d6
NB
305 /* if update != NULL, we are updating the host
306 * name... */
da6b5ca9
NB
307 continue;
308 }
309 }
83b6208e 310 /* If we are this far, then we are nearly commited to this device.
52826846 311 * If the super_block doesn't exist, or doesn't match others,
83b6208e
NB
312 * then we probably cannot continue
313 * However if one of the arrays is for the homehost, and
314 * the other isn't that can disambiguate.
52826846 315 */
52826846 316
4b1ac34b 317 if (!super) {
52826846
NB
318 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
319 devname);
4b1ac34b 320 free(first_super);
52826846
NB
321 return 1;
322 }
838acbc2 323
83b6208e
NB
324 if (st == NULL)
325 st = tst;
326 if (st->ss != tst->ss ||
327 st->minor_version != tst->minor_version ||
328 st->ss->compare_super(&first_super, super) != 0) {
329 /* Some mismatch. If exactly one array matches this host,
da6b5ca9
NB
330 * we can resolve on that one.
331 * Or, if we are auto assembling, we just ignore the second
332 * for now.
83b6208e 333 */
da6b5ca9
NB
334 if (mdfd < 0)
335 continue;
83b6208e
NB
336 if (homehost) {
337 int first = st->ss->match_home(first_super, homehost);
338 int last = tst->ss->match_home(super, homehost);
339 if (first+last == 1) {
340 /* We can do something */
341 if (first) {/* just ignore this one */
342 if ((inargv && verbose >= 0) || verbose > 0)
343 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
344 devname);
345 continue;
346 } else { /* reject all those sofar */
347 mddev_dev_t td;
348 if ((inargv && verbose >= 0) || verbose > 0)
349 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
350 devname);
351 for (td=devlist; td != tmpdev; td=td->next)
352 if (td->used == 1)
353 td->used = 0;
354 tmpdev->used = 1;
355 continue;
356 }
357 }
358 }
52826846
NB
359 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
360 devname);
4b1ac34b
NB
361 free(super);
362 free(first_super);
52826846 363 return 1;
64c4757e
NB
364 }
365
811e6cbe
NB
366 tmpdev->used = 1;
367 }
368
8a46fe84
NB
369 if (mdfd < 0) {
370 /* So... it is up to me to open the device.
371 * We create a name '/dev/md/XXX' based on the info in the
372 * superblock, and call open_mddev on that
373 */
da6b5ca9
NB
374 mdu_array_info_t inf;
375 char *c;
376 if (!first_super) {
d55e3aef 377 return 2;
da6b5ca9
NB
378 }
379 st->ss->getinfo_super(&info, first_super);
380 c = strchr(info.name, ':');
381 if (c) c++; else c= info.name;
41a3b72a
NB
382 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
383 /* /dev/md/d0 style for partitionable */
384 asprintf(&mddev, "/dev/md/d%s", c);
385 else
386 asprintf(&mddev, "/dev/md/%s", c);
da6b5ca9 387 mdfd = open_mddev(mddev, ident->autof);
60e1bc1a
NB
388 if (mdfd < 0) {
389 free(first_super);
390 free(devices);
391 first_super = NULL;
392 goto try_again;
393 }
da6b5ca9
NB
394 vers = md_get_version(mdfd);
395 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
60e1bc1a
NB
396 for (tmpdev = devlist ;
397 tmpdev && tmpdev->used != 1;
398 tmpdev = tmpdev->next)
399 ;
da6b5ca9 400 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
60e1bc1a
NB
401 if (tmpdev)
402 fprintf(stderr, Name ": %s needed for %s...\n",
403 mddev, tmpdev->devname);
da6b5ca9 404 close(mdfd);
60e1bc1a 405 mdfd = -1;
da6b5ca9 406 free(first_super);
60e1bc1a
NB
407 free(devices);
408 first_super = NULL;
409 goto try_again;
da6b5ca9 410 }
d55e3aef 411 must_close = 1;
8a46fe84
NB
412 }
413
811e6cbe 414 /* Ok, no bad inconsistancy, we can try updating etc */
da6b5ca9 415 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
811e6cbe
NB
416 char *devname = tmpdev->devname;
417 struct stat stb;
5787fa49 418 /* looks like a good enough match to update the super block if needed */
d1f1011b 419#ifndef MDASSEMBLE
5787fa49 420 if (update) {
811e6cbe 421 int dfd;
4b1ac34b
NB
422 /* prepare useful information in info structures */
423 struct stat stb2;
424 fstat(mdfd, &stb2);
7d99579f
NB
425
426 if (strcmp(update, "uuid")==0 &&
427 !ident->uuid_set) {
428 int rfd;
429 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
430 read(rfd, ident->uuid, 16) != 16) {
431 *(__u32*)(ident->uuid) = random();
432 *(__u32*)(ident->uuid+1) = random();
433 *(__u32*)(ident->uuid+2) = random();
434 *(__u32*)(ident->uuid+3) = random();
435 }
436 if (rfd >= 0) close(rfd);
7d99579f 437 }
811e6cbe
NB
438 dfd = dev_open(devname, O_RDWR|O_EXCL);
439
0430ed48
NB
440 remove_partitions(dfd);
441
811e6cbe
NB
442 if (super) {
443 free(super);
444 super = NULL;
445 }
446
447 st->ss->load_super(st, dfd, &super, NULL);
448 st->ss->getinfo_super(&info, super);
449
7d99579f 450 memcpy(info.uuid, ident->uuid, 16);
e5eac01f 451 strcpy(info.name, ident->name);
811e6cbe
NB
452 info.array.md_minor = minor(stb2.st_rdev);
453
e5eac01f
NB
454 st->ss->update_super(&info, super, update, devname, verbose,
455 ident->uuid_set, homehost);
456 if (strcmp(update, "uuid")==0 &&
457 !ident->uuid_set) {
458 ident->uuid_set = 1;
459 memcpy(ident->uuid, info.uuid, 16);
460 }
b8a8ccf9 461 if (dfd < 0)
5787fa49
NB
462 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
463 devname);
96395475 464 else if (st->ss->store_super(st, dfd, super))
5787fa49
NB
465 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
466 devname);
467 if (dfd >= 0)
468 close(dfd);
8131b493
NB
469
470 if (strcmp(update, "uuid")==0 &&
471 ident->bitmap_fd)
9fca7d62
NB
472 if (bitmap_update_uuid(ident->bitmap_fd, info.uuid) != 0)
473 fprintf(stderr, Name ": Could not update uuid on %s.\n",
474 devname);
d1f1011b
NB
475 } else
476#endif
477 {
da6b5ca9
NB
478 int dfd;
479 dfd = dev_open(devname, O_RDWR|O_EXCL);
480
0430ed48
NB
481 remove_partitions(dfd);
482
da6b5ca9
NB
483 if (super) {
484 free(super);
485 super = NULL;
486 }
487
488 st->ss->load_super(st, dfd, &super, NULL);
489 st->ss->getinfo_super(&info, super);
490 close(dfd);
5787fa49
NB
491 }
492
811e6cbe
NB
493 stat(devname, &stb);
494
dab6685f 495 if (verbose > 0)
cd29a5c8 496 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
4b1ac34b 497 devname, mddev, info.disk.raid_disk);
64c4757e 498 devices[devcnt].devname = devname;
0df46c2a
NB
499 devices[devcnt].major = major(stb.st_rdev);
500 devices[devcnt].minor = minor(stb.st_rdev);
4b1ac34b
NB
501 devices[devcnt].oldmajor = info.disk.major;
502 devices[devcnt].oldminor = info.disk.minor;
503 devices[devcnt].events = info.events;
504 devices[devcnt].raid_disk = info.disk.raid_disk;
6a41304b 505 devices[devcnt].disk_nr = info.disk.number;
64c4757e 506 devices[devcnt].uptodate = 0;
4b1ac34b 507 devices[devcnt].state = info.disk.state;
64c4757e
NB
508 if (most_recent < devcnt) {
509 if (devices[devcnt].events
510 > devices[most_recent].events)
511 most_recent = devcnt;
512 }
b8a8ccf9 513 if (info.array.level == -4)
5787fa49
NB
514 /* with multipath, the raid_disk from the superblock is meaningless */
515 i = devcnt;
516 else
517 i = devices[devcnt].raid_disk;
308e1801
NB
518 if (i+1 == 0) {
519 if (nextspare < info.array.raid_disks)
520 nextspare = info.array.raid_disks;
521 i = nextspare++;
08110d41
NB
522 } else {
523 if (i >= info.array.raid_disks &&
524 i >= nextspare)
525 nextspare = i+1;
308e1801 526 }
98c6faba 527 if (i < 10000) {
56eedc1a 528 if (i >= bestcnt) {
98c6faba 529 unsigned int newbestcnt = i+10;
56eedc1a 530 int *newbest = malloc(sizeof(int)*newbestcnt);
98c6faba 531 unsigned int c;
56eedc1a
NB
532 for (c=0; c < newbestcnt; c++)
533 if (c < bestcnt)
534 newbest[c] = best[c];
535 else
536 newbest[c] = -1;
537 if (best)free(best);
538 best = newbest;
539 bestcnt = newbestcnt;
540 }
6a255b69
NB
541 if (best[i] >=0 &&
542 devices[best[i]].events == devices[devcnt].events &&
543 devices[best[i]].minor != devices[devcnt].minor &&
544 st->ss->major == 0 &&
545 info.array.level != -4) {
546 /* two different devices with identical superblock.
547 * Could be a mis-detection caused by overlapping
548 * partitions. fail-safe.
549 */
550 fprintf(stderr, Name ": WARNING %s and %s appear"
551 " to have very similar superblocks.\n"
552 " If they are really different, "
553 "please --zero the superblock on one\n"
d2df86e0
NB
554 " If they are the same or overlap,"
555 " please remove one from %s.\n",
556 devices[best[i]].devname, devname,
557 inargv ? "the list" :
558 "the\n DEVICE list in mdadm.conf"
559 );
6a255b69
NB
560 if (must_close) close(mdfd);
561 return 1;
562 }
52826846
NB
563 if (best[i] == -1
564 || devices[best[i]].events < devices[devcnt].events)
565 best[i] = devcnt;
56eedc1a 566 }
64c4757e
NB
567 devcnt++;
568 }
569
4b1ac34b
NB
570 if (super)
571 free(super);
572 super = NULL;
573
586ed405
NB
574 if (update && strcmp(update, "byteorder")==0)
575 st->minor_version = 90;
576
64c4757e 577 if (devcnt == 0) {
682c7051 578 fprintf(stderr, Name ": no devices found for %s\n",
64c4757e 579 mddev);
4b1ac34b 580 free(first_super);
d55e3aef 581 if (must_close) close(mdfd);
64c4757e
NB
582 return 1;
583 }
4b1ac34b 584
31317663 585 st->ss->getinfo_super(&info, first_super);
4b1ac34b 586
64c4757e
NB
587 /* now we have some devices that might be suitable.
588 * I wonder how many
589 */
265e0f17
NB
590 avail = malloc(info.array.raid_disks);
591 memset(avail, 0, info.array.raid_disks);
64c4757e 592 okcnt = 0;
52826846 593 sparecnt=0;
56eedc1a 594 for (i=0; i< bestcnt ;i++) {
64c4757e 595 int j = best[i];
ee04451c
NB
596 int event_margin = 1; /* always allow a difference of '1'
597 * like the kernel does
598 */
64c4757e 599 if (j < 0) continue;
d013a55e
NB
600 /* note: we ignore error flags in multipath arrays
601 * as they don't make sense
602 */
4b1ac34b 603 if (info.array.level != -4)
aa88f531
NB
604 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
605 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
606 sparecnt++;
d013a55e 607 continue;
aa88f531 608 }
cd29a5c8 609 if (devices[j].events+event_margin >=
64c4757e
NB
610 devices[most_recent].events) {
611 devices[j].uptodate = 1;
265e0f17 612 if (i < info.array.raid_disks) {
52826846 613 okcnt++;
265e0f17
NB
614 avail[i]=1;
615 } else
52826846 616 sparecnt++;
64c4757e
NB
617 }
618 }
265e0f17
NB
619 while (force && !enough(info.array.level, info.array.raid_disks,
620 info.array.layout,
621 avail, okcnt)) {
64c4757e
NB
622 /* Choose the newest best drive which is
623 * not up-to-date, update the superblock
624 * and add it.
625 */
52826846 626 int fd;
cd29a5c8 627 chosen_drive = -1;
4b1ac34b 628 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
52826846
NB
629 int j = best[i];
630 if (j>=0 &&
631 !devices[j].uptodate &&
632 devices[j].events > 0 &&
633 (chosen_drive < 0 ||
634 devices[j].events > devices[chosen_drive].events))
635 chosen_drive = j;
636 }
637 if (chosen_drive < 0)
638 break;
dab6685f
NB
639 if (verbose >= 0)
640 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
641 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
642 (int)(devices[chosen_drive].events),
643 (int)(devices[most_recent].events));
8b0dabea 644 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846
NB
645 if (fd < 0) {
646 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
647 devices[chosen_drive].devname);
648 devices[chosen_drive].events = 0;
649 continue;
650 }
82d9eba6 651 if (st->ss->load_super(st,fd, &super, NULL)) {
52826846
NB
652 close(fd);
653 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
654 devices[chosen_drive].devname);
655 devices[chosen_drive].events = 0;
656 continue;
657 }
4b1ac34b 658 info.events = devices[most_recent].events;
e5eac01f 659 st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose, 0, NULL);
4b1ac34b 660
96395475 661 if (st->ss->store_super(st, fd, super)) {
52826846
NB
662 close(fd);
663 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
664 devices[chosen_drive].devname);
665 devices[chosen_drive].events = 0;
4b1ac34b 666 free(super);
52826846
NB
667 continue;
668 }
669 close(fd);
670 devices[chosen_drive].events = devices[most_recent].events;
671 devices[chosen_drive].uptodate = 1;
265e0f17 672 avail[chosen_drive] = 1;
52826846 673 okcnt++;
4b1ac34b 674 free(super);
64c4757e 675 }
52826846
NB
676
677 /* Now we want to look at the superblock which the kernel will base things on
678 * and compare the devices that we think are working with the devices that the
679 * superblock thinks are working.
680 * If there are differences and --force is given, then update this chosen
681 * superblock.
682 */
cd29a5c8 683 chosen_drive = -1;
4b1ac34b 684 super = NULL;
56eedc1a 685 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
52826846
NB
686 int j = best[i];
687 int fd;
4b1ac34b 688
52826846
NB
689 if (j<0)
690 continue;
691 if (!devices[j].uptodate)
692 continue;
693 chosen_drive = j;
8b0dabea 694 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
52826846
NB
695 fprintf(stderr, Name ": Cannot open %s: %s\n",
696 devices[j].devname, strerror(errno));
d55e3aef 697 if (must_close) close(mdfd);
52826846
NB
698 return 1;
699 }
82d9eba6 700 if (st->ss->load_super(st,fd, &super, NULL)) {
52826846
NB
701 close(fd);
702 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
703 devices[j].devname);
d55e3aef 704 if (must_close) close(mdfd);
52826846
NB
705 return 1;
706 }
707 close(fd);
708 }
4b1ac34b
NB
709 if (super == NULL) {
710 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
d55e3aef 711 if (must_close) close(mdfd);
4b1ac34b
NB
712 return 1;
713 }
31317663 714 st->ss->getinfo_super(&info, super);
56eedc1a 715 for (i=0; i<bestcnt; i++) {
52826846 716 int j = best[i];
98c6faba 717 unsigned int desired_state;
11a3e71d 718
4b1ac34b 719 if (i < info.array.raid_disks)
11a3e71d
NB
720 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
721 else
722 desired_state = 0;
723
52826846
NB
724 if (j<0)
725 continue;
726 if (!devices[j].uptodate)
727 continue;
6a41304b 728 info.disk.number = devices[j].disk_nr;
fbf8a0b7 729 info.disk.raid_disk = i;
4b1ac34b
NB
730 info.disk.state = desired_state;
731
52826846 732 if (devices[j].uptodate &&
e5eac01f 733 st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) {
52826846 734 if (force) {
dab6685f
NB
735 if (verbose >= 0)
736 fprintf(stderr, Name ": "
737 "clearing FAULTY flag for device %d in %s for %s\n",
738 j, mddev, devices[j].devname);
4b1ac34b 739 change = 1;
52826846 740 } else {
dab6685f
NB
741 if (verbose >= -1)
742 fprintf(stderr, Name ": "
743 "device %d in %s has wrong state in superblock, but %s seems ok\n",
744 i, mddev, devices[j].devname);
52826846
NB
745 }
746 }
4b1ac34b 747#if 0
52826846
NB
748 if (!devices[j].uptodate &&
749 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
750 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
751 i, mddev);
752 }
4b1ac34b 753#endif
52826846 754 }
4b1ac34b
NB
755 if (force && okcnt == info.array.raid_disks-1) {
756 /* FIXME check event count */
b8a8ccf9 757 change += st->ss->update_super(&info, super, "force",
e5eac01f 758 devices[chosen_drive].devname, verbose, 0, NULL);
d013a55e 759 }
52826846 760
4b1ac34b 761 if (change) {
52826846 762 int fd;
8b0dabea 763 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846 764 if (fd < 0) {
353632d9 765 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
52826846 766 devices[chosen_drive].devname);
d55e3aef 767 if (must_close) close(mdfd);
52826846
NB
768 return 1;
769 }
96395475 770 if (st->ss->store_super(st, fd, super)) {
52826846
NB
771 close(fd);
772 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
773 devices[chosen_drive].devname);
d55e3aef 774 if (must_close) close(mdfd);
52826846
NB
775 return 1;
776 }
777 close(fd);
52826846
NB
778 }
779
353632d9
NB
780 /* If we are in the middle of a reshape we may need to restore saved data
781 * that was moved aside due to the reshape overwriting live data
782 * The code of doing this lives in Grow.c
783 */
39c5a909 784#ifndef MDASSEMBLE
353632d9
NB
785 if (info.reshape_active) {
786 int err = 0;
787 int *fdlist = malloc(sizeof(int)* bestcnt);
788 for (i=0; i<bestcnt; i++) {
789 int j = best[i];
790 if (j >= 0) {
791 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
792 if (fdlist[i] < 0) {
793 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
794 devices[j].devname);
795 err = 1;
796 break;
797 }
798 } else
799 fdlist[i] = -1;
800 }
801 if (!err)
06b0d786 802 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
353632d9
NB
803 while (i>0) {
804 i--;
805 if (fdlist[i]>=0) close(fdlist[i]);
806 }
807 if (err) {
808 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
d55e3aef 809 if (must_close) close(mdfd);
353632d9
NB
810 return err;
811 }
812 }
39c5a909 813#endif
aa88f531
NB
814 /* count number of in-sync devices according to the superblock.
815 * We must have this number to start the array without -s or -R
816 */
4b1ac34b 817 req_cnt = info.array.working_disks;
aa88f531 818
64c4757e
NB
819 /* Almost ready to actually *do* something */
820 if (!old_linux) {
fbf8a0b7
NB
821 int rv;
822 if ((vers % 100) >= 1) { /* can use different versions */
823 mdu_array_info_t inf;
824 memset(&inf, 0, sizeof(inf));
825 inf.major_version = st->ss->major;
826 inf.minor_version = st->minor_version;
827 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
b8a8ccf9 828 } else
fbf8a0b7
NB
829 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
830
831 if (rv) {
682c7051 832 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
64c4757e 833 mddev, strerror(errno));
d55e3aef 834 if (must_close) close(mdfd);
64c4757e
NB
835 return 1;
836 }
11484a63 837 if (ident->bitmap_fd >= 0) {
c82f047c
NB
838 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
839 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
d55e3aef 840 if (must_close) close(mdfd);
c82f047c
NB
841 return 1;
842 }
7ef02d01
NB
843 } else if (ident->bitmap_file) {
844 /* From config file */
845 int bmfd = open(ident->bitmap_file, O_RDWR);
846 if (bmfd < 0) {
847 fprintf(stderr, Name ": Could not open bitmap file %s\n",
848 ident->bitmap_file);
d55e3aef 849 if (must_close) close(mdfd);
7ef02d01
NB
850 return 1;
851 }
852 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
853 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
854 close(bmfd);
d55e3aef 855 if (must_close) close(mdfd);
7ef02d01
NB
856 return 1;
857 }
858 close(bmfd);
c82f047c 859 }
7ef02d01 860
52826846 861 /* First, add the raid disks, but add the chosen one last */
56eedc1a 862 for (i=0; i<= bestcnt; i++) {
52826846 863 int j;
56eedc1a 864 if (i < bestcnt) {
52826846
NB
865 j = best[i];
866 if (j == chosen_drive)
867 continue;
868 } else
869 j = chosen_drive;
870
56eedc1a 871 if (j >= 0 /* && devices[j].uptodate */) {
64c4757e
NB
872 mdu_disk_info_t disk;
873 memset(&disk, 0, sizeof(disk));
874 disk.major = devices[j].major;
875 disk.minor = devices[j].minor;
876 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
682c7051 877 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
64c4757e
NB
878 devices[j].devname,
879 mddev,
880 strerror(errno));
4b1ac34b 881 if (i < info.array.raid_disks || i == bestcnt)
52826846
NB
882 okcnt--;
883 else
884 sparecnt--;
dab6685f 885 } else if (verbose > 0)
52826846
NB
886 fprintf(stderr, Name ": added %s to %s as %d\n",
887 devices[j].devname, mddev, devices[j].raid_disk);
dab6685f 888 } else if (verbose > 0 && i < info.array.raid_disks)
682c7051 889 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
64c4757e
NB
890 i, mddev);
891 }
52826846 892
64c4757e 893 if (runstop == 1 ||
b8a8ccf9 894 (runstop <= 0 &&
265e0f17 895 ( enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt) &&
aa88f531
NB
896 (okcnt >= req_cnt || start_partial_ok)
897 ))) {
82b27616 898 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
dab6685f
NB
899 if (verbose >= 0) {
900 fprintf(stderr, Name ": %s has been started with %d drive%s",
901 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9 902 if (okcnt < info.array.raid_disks)
dab6685f
NB
903 fprintf(stderr, " (out of %d)", info.array.raid_disks);
904 if (sparecnt)
905 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
906 fprintf(stderr, ".\n");
907 }
589395d6
NB
908 if (must_close) {
909 int usecs = 1;
910 close(mdfd);
911 /* There is a nasty race with 'mdadm --monitor'.
912 * If it opens this device before we close it,
913 * it gets an incomplete open on which IO
914 * doesn't work and the capacity if wrong.
915 * If we reopen (to check for layered devices)
916 * before --monitor closes, we loose.
917 *
918 * So: wait upto 1 second for there to be
919 * a non-zero capacity.
920 */
921 while (usecs < 1000) {
922 mdfd = open(mddev, O_RDONLY);
923 if (mdfd >= 0) {
924 unsigned long size;
925 if (ioctl(mdfd, BLKGETSIZE, &size) == 0 &&
926 size > 0)
927 break;
928 close(mdfd);
929 }
930 usleep(usecs);
931 usecs <<= 1;
932 }
933 }
64c4757e 934 return 0;
82b27616 935 }
682c7051 936 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
64c4757e 937 mddev, strerror(errno));
d55e3aef 938 if (must_close) close(mdfd);
64c4757e
NB
939 return 1;
940 }
82b27616 941 if (runstop == -1) {
b8a8ccf9 942 fprintf(stderr, Name ": %s assembled from %d drive%s",
52826846 943 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9
NB
944 if (okcnt != info.array.raid_disks)
945 fprintf(stderr, " (out of %d)", info.array.raid_disks);
946 fprintf(stderr, ", but not started.\n");
d55e3aef 947 if (must_close) close(mdfd);
64c4757e 948 return 0;
82b27616 949 }
dab6685f
NB
950 if (verbose >= 0) {
951 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
952 if (sparecnt)
953 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
265e0f17 954 if (!enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt))
dab6685f
NB
955 fprintf(stderr, " - not enough to start the array.\n");
956 else {
957 if (req_cnt == info.array.raid_disks)
958 fprintf(stderr, " - need all %d to start it", req_cnt);
959 else
960 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
961 fprintf(stderr, " (use --run to insist).\n");
962 }
aa88f531 963 }
d55e3aef 964 if (must_close) close(mdfd);
64c4757e 965 return 1;
82b27616 966 } else {
52826846
NB
967 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
968 * been updated to point to the current locations of devices.
969 * so we can just start the array
82b27616 970 */
cd29a5c8 971 unsigned long dev;
0df46c2a 972 dev = makedev(devices[chosen_drive].major,
82b27616
NB
973 devices[chosen_drive].minor);
974 if (ioctl(mdfd, START_ARRAY, dev)) {
975 fprintf(stderr, Name ": Cannot start array: %s\n",
976 strerror(errno));
977 }
978
64c4757e 979 }
d55e3aef 980 if (must_close) close(mdfd);
e0d19036 981 return 0;
64c4757e 982}