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