]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Assemble.c
Make sure everything compiles...
[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"
64c4757e 31
624920bb
NB
32static int name_matches(char *found, char *required, char *homehost)
33{
34 /* See if the name found matches the required name, possibly
35 * prefixed with 'homehost'
36 */
37 char fnd[33];
38
39 strncpy(fnd, found, 32);
40 fnd[32] = 0;
41 if (strcmp(found, required)==0)
42 return 1;
43 if (homehost) {
44 int l = strlen(homehost);
45 if (l < 32 && fnd[l] == ':' &&
46 strcmp(fnd+l+1, required)==0)
47 return 1;
48 }
49 return 0;
50}
51
82d9eba6 52int Assemble(struct supertype *st, char *mddev, int mdfd,
52826846 53 mddev_ident_t ident, char *conffile,
06b0d786 54 mddev_dev_t devlist, char *backup_file,
64c4757e 55 int readonly, int runstop,
e5eac01f 56 char *update, char *homehost,
64c4757e
NB
57 int verbose, int force)
58{
59 /*
52826846
NB
60 * The task of Assemble is to find a collection of
61 * devices that should (according to their superblocks)
62 * form an array, and to give this collection to the MD driver.
63 * In Linux-2.4 and later, this involves submitting a
64c4757e
NB
64 * SET_ARRAY_INFO ioctl with no arg - to prepare
65 * the array - and then submit a number of
66 * ADD_NEW_DISK ioctls to add disks into
67 * the array. Finally RUN_ARRAY might
68 * be submitted to start the array.
69 *
70 * Much of the work of Assemble is in finding and/or
71 * checking the disks to make sure they look right.
72 *
4b1ac34b 73 * If mddev is not set, then scan must be set and we
64c4757e
NB
74 * read through the config file for dev+uuid mapping
75 * We recurse, setting mddev, for each device that
76 * - isn't running
4b1ac34b 77 * - has a valid uuid (or any uuid if !uuidset)
64c4757e
NB
78 *
79 * If mddev is set, we try to determine state of md.
80 * check version - must be at least 0.90.0
81 * check kernel version. must be at least 2.4.
82 * If not, we can possibly fall back on START_ARRAY
83 * Try to GET_ARRAY_INFO.
84 * If possible, give up
85 * If not, try to STOP_ARRAY just to make sure
86 *
87 * If !uuidset and scan, look in conf-file for uuid
88 * If not found, give up
b8a8ccf9 89 * If !devlist and scan and uuidset, get list of devs from conf-file
64c4757e
NB
90 *
91 * For each device:
92 * Check superblock - discard if bad
93 * Check uuid (set if we don't have one) - discard if no match
5787fa49 94 * Check superblock similarity if we have a superblock - discard if different
4b1ac34b 95 * Record events, devicenum
64c4757e 96 * This should give us a list of devices for the array
4b1ac34b 97 * We should collect the most recent event number
64c4757e
NB
98 *
99 * Count disks with recent enough event count
100 * While force && !enough disks
101 * Choose newest rejected disks, update event count
102 * mark clean and rewrite superblock
103 * If recent kernel:
104 * SET_ARRAY_INFO
105 * foreach device with recent events : ADD_NEW_DISK
106 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
107 * If old kernel:
108 * Check the device numbers in superblock are right
109 * update superblock if any changes
110 * START_ARRAY
111 *
112 */
d55e3aef 113 int must_close = 0;
64c4757e
NB
114 int old_linux = 0;
115 int vers;
4b1ac34b 116 void *first_super = NULL, *super = NULL;
64c4757e
NB
117 struct {
118 char *devname;
98c6faba
NB
119 unsigned int major, minor;
120 unsigned int oldmajor, oldminor;
64c4757e 121 long long events;
64c4757e 122 int uptodate;
d013a55e 123 int state;
52826846 124 int raid_disk;
6a41304b 125 int disk_nr;
5787fa49 126 } *devices;
56eedc1a 127 int *best = NULL; /* indexed by raid_disk */
98c6faba
NB
128 unsigned int bestcnt = 0;
129 int devcnt = 0;
130 unsigned int okcnt, sparecnt;
131 unsigned int req_cnt;
132 unsigned int i;
64c4757e 133 int most_recent = 0;
cd29a5c8 134 int chosen_drive;
52826846 135 int change = 0;
cd29a5c8 136 int inargv = 0;
589395d6 137 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
98c6faba 138 unsigned int num_devs;
5787fa49 139 mddev_dev_t tmpdev;
4b1ac34b 140 struct mdinfo info;
265e0f17 141 char *avail;
308e1801 142 int nextspare = 0;
8a46fe84 143
682c7051 144 if (get_linux_version() < 2004000)
64c4757e
NB
145 old_linux = 1;
146
8a46fe84
NB
147 if (mdfd >= 0) {
148 vers = md_get_version(mdfd);
149 if (vers <= 0) {
150 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
151 return 1;
152 }
153 if (vers < 9000) {
154 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
155 " Upgrade your kernel or try --build\n");
156 return 1;
157 }
64c4757e 158
8a46fe84
NB
159 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
160 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
161 mddev);
162 return 1;
163 }
164 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
165 }
64c4757e 166 /*
52826846
NB
167 * If any subdevs are listed, then any that don't
168 * match ident are discarded. Remainder must all match and
169 * become the array.
170 * If no subdevs, then we scan all devices in the config file, but
171 * there must be something in the identity
64c4757e 172 */
64c4757e 173
cd29a5c8 174 if (!devlist &&
52826846
NB
175 ident->uuid_set == 0 &&
176 ident->super_minor < 0 &&
177 ident->devices == NULL) {
178 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
8a46fe84 179 mddev ? mddev : "further assembly");
52826846 180 return 1;
64c4757e 181 }
cd29a5c8 182 if (devlist == NULL)
64c4757e 183 devlist = conf_get_devs(conffile);
da6b5ca9
NB
184 else if (mdfd >= 0)
185 inargv = 1;
64c4757e 186
5787fa49
NB
187 tmpdev = devlist; num_devs = 0;
188 while (tmpdev) {
da6b5ca9
NB
189 if (tmpdev->used)
190 tmpdev->used = 2;
191 else
192 num_devs++;
5787fa49
NB
193 tmpdev = tmpdev->next;
194 }
5787fa49
NB
195 devices = malloc(num_devs * sizeof(*devices));
196
82d9eba6 197 if (!st && ident->st) st = ident->st;
64c4757e 198
dab6685f 199 if (verbose>0)
82b27616 200 fprintf(stderr, Name ": looking for devices for %s\n",
8a46fe84 201 mddev ? mddev : "further assembly");
82b27616 202
811e6cbe
NB
203 /* first walk the list of devices to find a consistent set
204 * that match the criterea, if that is possible.
205 * We flag the one we like with 'used'.
206 */
207 for (tmpdev = devlist;
208 tmpdev;
209 tmpdev = tmpdev->next) {
210 char *devname = tmpdev->devname;
64c4757e
NB
211 int dfd;
212 struct stat stb;
82d9eba6 213 struct supertype *tst = st;
52826846 214
da6b5ca9
NB
215 if (tmpdev->used > 1) continue;
216
52826846 217 if (ident->devices &&
56eedc1a 218 !match_oneof(ident->devices, devname)) {
6a41304b 219 if ((inargv && verbose>=0) || verbose > 0)
56eedc1a 220 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
52826846 221 continue;
56eedc1a 222 }
4b1ac34b
NB
223
224 if (super) {
225 free(super);
226 super = NULL;
227 }
52826846 228
8b0dabea 229 dfd = dev_open(devname, O_RDONLY|O_EXCL);
64c4757e 230 if (dfd < 0) {
6a41304b 231 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 232 fprintf(stderr, Name ": cannot open device %s: %s\n",
64c4757e 233 devname, strerror(errno));
da6b5ca9 234 tmpdev->used = 2;
52826846
NB
235 } else if (fstat(dfd, &stb)< 0) {
236 /* Impossible! */
237 fprintf(stderr, Name ": fstat failed for %s: %s\n",
238 devname, strerror(errno));
da6b5ca9 239 tmpdev->used = 2;
cd29a5c8
NB
240 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
241 fprintf(stderr, Name ": %s is not a block device.\n",
52826846 242 devname);
da6b5ca9 243 tmpdev->used = 2;
82d9eba6 244 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
6a41304b 245 if ((inargv && verbose >= 0) || verbose > 0)
da6b5ca9
NB
246 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
247 devname);
248 tmpdev->used = 2;
82d9eba6 249 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
6a41304b 250 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 251 fprintf( stderr, Name ": no RAID superblock on %s\n",
64c4757e 252 devname);
52826846 253 } else {
31317663 254 tst->ss->getinfo_super(&info, super);
64c4757e 255 }
c06487ce 256 if (dfd >= 0) close(dfd);
52826846 257
838acbc2 258 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
f277ce36 259 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
6a41304b 260 if ((inargv && verbose >= 0) || verbose > 0)
52826846
NB
261 fprintf(stderr, Name ": %s has wrong uuid.\n",
262 devname);
263 continue;
82b27616 264 }
c4f12c13 265 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
624920bb 266 (!super || name_matches(info.name, ident->name, homehost)==0)) {
6a41304b 267 if ((inargv && verbose >= 0) || verbose > 0)
947fd4dd
NB
268 fprintf(stderr, Name ": %s has wrong name.\n",
269 devname);
270 continue;
271 }
98c6faba 272 if (ident->super_minor != UnSet &&
4b1ac34b 273 (!super || ident->super_minor != info.array.md_minor)) {
6a41304b 274 if ((inargv && verbose >= 0) || verbose > 0)
52826846 275 fprintf(stderr, Name ": %s has wrong super-minor.\n",
64c4757e
NB
276 devname);
277 continue;
278 }
98c6faba 279 if (ident->level != UnSet &&
4b1ac34b 280 (!super|| ident->level != info.array.level)) {
6a41304b 281 if ((inargv && verbose >= 0) || verbose > 0)
cd29a5c8
NB
282 fprintf(stderr, Name ": %s has wrong raid level.\n",
283 devname);
284 continue;
285 }
98c6faba 286 if (ident->raid_disks != UnSet &&
4b1ac34b 287 (!super || ident->raid_disks!= info.array.raid_disks)) {
6a41304b 288 if ((inargv && verbose >= 0) || verbose > 0)
cd29a5c8
NB
289 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
290 devname);
291 continue;
292 }
da6b5ca9
NB
293 if (mdfd < 0) {
294 if (tst == NULL || super == NULL)
295 continue;
589395d6
NB
296 if (update == NULL &&
297 tst->ss->match_home(super, homehost)==0) {
da6b5ca9
NB
298 if ((inargv && verbose >= 0) || verbose > 0)
299 fprintf(stderr, Name ": %s is not built for host %s.\n",
300 devname, homehost);
301 /* Auto-assemble, and this is not a usable host */
589395d6
NB
302 /* if update != NULL, we are updating the host
303 * name... */
da6b5ca9
NB
304 continue;
305 }
306 }
83b6208e 307 /* If we are this far, then we are nearly commited to this device.
52826846 308 * If the super_block doesn't exist, or doesn't match others,
83b6208e
NB
309 * then we probably cannot continue
310 * However if one of the arrays is for the homehost, and
311 * the other isn't that can disambiguate.
52826846 312 */
52826846 313
4b1ac34b 314 if (!super) {
52826846
NB
315 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
316 devname);
4b1ac34b 317 free(first_super);
52826846
NB
318 return 1;
319 }
838acbc2 320
83b6208e
NB
321 if (st == NULL)
322 st = tst;
323 if (st->ss != tst->ss ||
324 st->minor_version != tst->minor_version ||
325 st->ss->compare_super(&first_super, super) != 0) {
326 /* Some mismatch. If exactly one array matches this host,
da6b5ca9
NB
327 * we can resolve on that one.
328 * Or, if we are auto assembling, we just ignore the second
329 * for now.
83b6208e 330 */
da6b5ca9
NB
331 if (mdfd < 0)
332 continue;
83b6208e
NB
333 if (homehost) {
334 int first = st->ss->match_home(first_super, homehost);
335 int last = tst->ss->match_home(super, homehost);
336 if (first+last == 1) {
337 /* We can do something */
338 if (first) {/* just ignore this one */
339 if ((inargv && verbose >= 0) || verbose > 0)
340 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
341 devname);
342 continue;
343 } else { /* reject all those sofar */
344 mddev_dev_t td;
345 if ((inargv && verbose >= 0) || verbose > 0)
346 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
347 devname);
348 for (td=devlist; td != tmpdev; td=td->next)
349 if (td->used == 1)
350 td->used = 0;
351 tmpdev->used = 1;
352 continue;
353 }
354 }
355 }
52826846
NB
356 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
357 devname);
4b1ac34b
NB
358 free(super);
359 free(first_super);
52826846 360 return 1;
64c4757e
NB
361 }
362
811e6cbe
NB
363 tmpdev->used = 1;
364 }
365
8a46fe84
NB
366 if (mdfd < 0) {
367 /* So... it is up to me to open the device.
368 * We create a name '/dev/md/XXX' based on the info in the
369 * superblock, and call open_mddev on that
370 */
da6b5ca9
NB
371 mdu_array_info_t inf;
372 char *c;
373 if (!first_super) {
d55e3aef 374 return 2;
da6b5ca9
NB
375 }
376 st->ss->getinfo_super(&info, first_super);
377 c = strchr(info.name, ':');
378 if (c) c++; else c= info.name;
379 asprintf(&mddev, "/dev/md/%s", c);
380 mdfd = open_mddev(mddev, ident->autof);
8a46fe84
NB
381 if (mdfd < 0)
382 return mdfd;
da6b5ca9
NB
383 vers = md_get_version(mdfd);
384 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
385 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
386 close(mdfd);
387 free(first_super);
388 return 1;
389 }
d55e3aef 390 must_close = 1;
8a46fe84
NB
391 }
392
811e6cbe 393 /* Ok, no bad inconsistancy, we can try updating etc */
da6b5ca9 394 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
811e6cbe
NB
395 char *devname = tmpdev->devname;
396 struct stat stb;
5787fa49 397 /* looks like a good enough match to update the super block if needed */
d1f1011b 398#ifndef MDASSEMBLE
5787fa49 399 if (update) {
811e6cbe 400 int dfd;
4b1ac34b
NB
401 /* prepare useful information in info structures */
402 struct stat stb2;
403 fstat(mdfd, &stb2);
7d99579f
NB
404
405 if (strcmp(update, "uuid")==0 &&
406 !ident->uuid_set) {
407 int rfd;
408 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
409 read(rfd, ident->uuid, 16) != 16) {
410 *(__u32*)(ident->uuid) = random();
411 *(__u32*)(ident->uuid+1) = random();
412 *(__u32*)(ident->uuid+2) = random();
413 *(__u32*)(ident->uuid+3) = random();
414 }
415 if (rfd >= 0) close(rfd);
7d99579f 416 }
811e6cbe
NB
417 dfd = dev_open(devname, O_RDWR|O_EXCL);
418
419 if (super) {
420 free(super);
421 super = NULL;
422 }
423
424 st->ss->load_super(st, dfd, &super, NULL);
425 st->ss->getinfo_super(&info, super);
426
7d99579f 427 memcpy(info.uuid, ident->uuid, 16);
e5eac01f 428 strcpy(info.name, ident->name);
811e6cbe
NB
429 info.array.md_minor = minor(stb2.st_rdev);
430
e5eac01f
NB
431 st->ss->update_super(&info, super, update, devname, verbose,
432 ident->uuid_set, homehost);
433 if (strcmp(update, "uuid")==0 &&
434 !ident->uuid_set) {
435 ident->uuid_set = 1;
436 memcpy(ident->uuid, info.uuid, 16);
437 }
b8a8ccf9 438 if (dfd < 0)
5787fa49
NB
439 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
440 devname);
96395475 441 else if (st->ss->store_super(st, dfd, super))
5787fa49
NB
442 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
443 devname);
444 if (dfd >= 0)
445 close(dfd);
8131b493
NB
446
447 if (strcmp(update, "uuid")==0 &&
448 ident->bitmap_fd)
449 bitmap_update_uuid(ident->bitmap_fd, info.uuid);
d1f1011b
NB
450 } else
451#endif
452 {
da6b5ca9
NB
453 int dfd;
454 dfd = dev_open(devname, O_RDWR|O_EXCL);
455
456 if (super) {
457 free(super);
458 super = NULL;
459 }
460
461 st->ss->load_super(st, dfd, &super, NULL);
462 st->ss->getinfo_super(&info, super);
463 close(dfd);
5787fa49
NB
464 }
465
811e6cbe
NB
466 stat(devname, &stb);
467
dab6685f 468 if (verbose > 0)
cd29a5c8 469 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
4b1ac34b 470 devname, mddev, info.disk.raid_disk);
64c4757e 471 devices[devcnt].devname = devname;
0df46c2a
NB
472 devices[devcnt].major = major(stb.st_rdev);
473 devices[devcnt].minor = minor(stb.st_rdev);
4b1ac34b
NB
474 devices[devcnt].oldmajor = info.disk.major;
475 devices[devcnt].oldminor = info.disk.minor;
476 devices[devcnt].events = info.events;
477 devices[devcnt].raid_disk = info.disk.raid_disk;
6a41304b 478 devices[devcnt].disk_nr = info.disk.number;
64c4757e 479 devices[devcnt].uptodate = 0;
4b1ac34b 480 devices[devcnt].state = info.disk.state;
64c4757e
NB
481 if (most_recent < devcnt) {
482 if (devices[devcnt].events
483 > devices[most_recent].events)
484 most_recent = devcnt;
485 }
b8a8ccf9 486 if (info.array.level == -4)
5787fa49
NB
487 /* with multipath, the raid_disk from the superblock is meaningless */
488 i = devcnt;
489 else
490 i = devices[devcnt].raid_disk;
308e1801
NB
491 if (i+1 == 0) {
492 if (nextspare < info.array.raid_disks)
493 nextspare = info.array.raid_disks;
494 i = nextspare++;
495 }
98c6faba 496 if (i < 10000) {
56eedc1a 497 if (i >= bestcnt) {
98c6faba 498 unsigned int newbestcnt = i+10;
56eedc1a 499 int *newbest = malloc(sizeof(int)*newbestcnt);
98c6faba 500 unsigned int c;
56eedc1a
NB
501 for (c=0; c < newbestcnt; c++)
502 if (c < bestcnt)
503 newbest[c] = best[c];
504 else
505 newbest[c] = -1;
506 if (best)free(best);
507 best = newbest;
508 bestcnt = newbestcnt;
509 }
52826846
NB
510 if (best[i] == -1
511 || devices[best[i]].events < devices[devcnt].events)
512 best[i] = devcnt;
56eedc1a 513 }
64c4757e
NB
514 devcnt++;
515 }
516
4b1ac34b
NB
517 if (super)
518 free(super);
519 super = NULL;
520
586ed405
NB
521 if (update && strcmp(update, "byteorder")==0)
522 st->minor_version = 90;
523
64c4757e 524 if (devcnt == 0) {
682c7051 525 fprintf(stderr, Name ": no devices found for %s\n",
64c4757e 526 mddev);
4b1ac34b 527 free(first_super);
d55e3aef 528 if (must_close) close(mdfd);
64c4757e
NB
529 return 1;
530 }
4b1ac34b 531
31317663 532 st->ss->getinfo_super(&info, first_super);
4b1ac34b 533
64c4757e
NB
534 /* now we have some devices that might be suitable.
535 * I wonder how many
536 */
265e0f17
NB
537 avail = malloc(info.array.raid_disks);
538 memset(avail, 0, info.array.raid_disks);
64c4757e 539 okcnt = 0;
52826846 540 sparecnt=0;
56eedc1a 541 for (i=0; i< bestcnt ;i++) {
64c4757e 542 int j = best[i];
ee04451c
NB
543 int event_margin = 1; /* always allow a difference of '1'
544 * like the kernel does
545 */
64c4757e 546 if (j < 0) continue;
d013a55e
NB
547 /* note: we ignore error flags in multipath arrays
548 * as they don't make sense
549 */
4b1ac34b 550 if (info.array.level != -4)
aa88f531
NB
551 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
552 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
553 sparecnt++;
d013a55e 554 continue;
aa88f531 555 }
cd29a5c8 556 if (devices[j].events+event_margin >=
64c4757e
NB
557 devices[most_recent].events) {
558 devices[j].uptodate = 1;
265e0f17 559 if (i < info.array.raid_disks) {
52826846 560 okcnt++;
265e0f17
NB
561 avail[i]=1;
562 } else
52826846 563 sparecnt++;
64c4757e
NB
564 }
565 }
265e0f17
NB
566 while (force && !enough(info.array.level, info.array.raid_disks,
567 info.array.layout,
568 avail, okcnt)) {
64c4757e
NB
569 /* Choose the newest best drive which is
570 * not up-to-date, update the superblock
571 * and add it.
572 */
52826846 573 int fd;
cd29a5c8 574 chosen_drive = -1;
4b1ac34b 575 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
52826846
NB
576 int j = best[i];
577 if (j>=0 &&
578 !devices[j].uptodate &&
579 devices[j].events > 0 &&
580 (chosen_drive < 0 ||
581 devices[j].events > devices[chosen_drive].events))
582 chosen_drive = j;
583 }
584 if (chosen_drive < 0)
585 break;
dab6685f
NB
586 if (verbose >= 0)
587 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
588 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
589 (int)(devices[chosen_drive].events),
590 (int)(devices[most_recent].events));
8b0dabea 591 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846
NB
592 if (fd < 0) {
593 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
594 devices[chosen_drive].devname);
595 devices[chosen_drive].events = 0;
596 continue;
597 }
82d9eba6 598 if (st->ss->load_super(st,fd, &super, NULL)) {
52826846
NB
599 close(fd);
600 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
601 devices[chosen_drive].devname);
602 devices[chosen_drive].events = 0;
603 continue;
604 }
4b1ac34b 605 info.events = devices[most_recent].events;
e5eac01f 606 st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose, 0, NULL);
4b1ac34b 607
96395475 608 if (st->ss->store_super(st, fd, super)) {
52826846
NB
609 close(fd);
610 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
611 devices[chosen_drive].devname);
612 devices[chosen_drive].events = 0;
4b1ac34b 613 free(super);
52826846
NB
614 continue;
615 }
616 close(fd);
617 devices[chosen_drive].events = devices[most_recent].events;
618 devices[chosen_drive].uptodate = 1;
265e0f17 619 avail[chosen_drive] = 1;
52826846 620 okcnt++;
4b1ac34b 621 free(super);
64c4757e 622 }
52826846
NB
623
624 /* Now we want to look at the superblock which the kernel will base things on
625 * and compare the devices that we think are working with the devices that the
626 * superblock thinks are working.
627 * If there are differences and --force is given, then update this chosen
628 * superblock.
629 */
cd29a5c8 630 chosen_drive = -1;
4b1ac34b 631 super = NULL;
56eedc1a 632 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
52826846
NB
633 int j = best[i];
634 int fd;
4b1ac34b 635
52826846
NB
636 if (j<0)
637 continue;
638 if (!devices[j].uptodate)
639 continue;
640 chosen_drive = j;
8b0dabea 641 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
52826846
NB
642 fprintf(stderr, Name ": Cannot open %s: %s\n",
643 devices[j].devname, strerror(errno));
d55e3aef 644 if (must_close) close(mdfd);
52826846
NB
645 return 1;
646 }
82d9eba6 647 if (st->ss->load_super(st,fd, &super, NULL)) {
52826846
NB
648 close(fd);
649 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
650 devices[j].devname);
d55e3aef 651 if (must_close) close(mdfd);
52826846
NB
652 return 1;
653 }
654 close(fd);
655 }
4b1ac34b
NB
656 if (super == NULL) {
657 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
d55e3aef 658 if (must_close) close(mdfd);
4b1ac34b
NB
659 return 1;
660 }
31317663 661 st->ss->getinfo_super(&info, super);
56eedc1a 662 for (i=0; i<bestcnt; i++) {
52826846 663 int j = best[i];
98c6faba 664 unsigned int desired_state;
11a3e71d 665
4b1ac34b 666 if (i < info.array.raid_disks)
11a3e71d
NB
667 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
668 else
669 desired_state = 0;
670
52826846
NB
671 if (j<0)
672 continue;
673 if (!devices[j].uptodate)
674 continue;
6a41304b 675 info.disk.number = devices[j].disk_nr;
fbf8a0b7 676 info.disk.raid_disk = i;
4b1ac34b
NB
677 info.disk.state = desired_state;
678
52826846 679 if (devices[j].uptodate &&
e5eac01f 680 st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) {
52826846 681 if (force) {
dab6685f
NB
682 if (verbose >= 0)
683 fprintf(stderr, Name ": "
684 "clearing FAULTY flag for device %d in %s for %s\n",
685 j, mddev, devices[j].devname);
4b1ac34b 686 change = 1;
52826846 687 } else {
dab6685f
NB
688 if (verbose >= -1)
689 fprintf(stderr, Name ": "
690 "device %d in %s has wrong state in superblock, but %s seems ok\n",
691 i, mddev, devices[j].devname);
52826846
NB
692 }
693 }
4b1ac34b 694#if 0
52826846
NB
695 if (!devices[j].uptodate &&
696 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
697 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
698 i, mddev);
699 }
4b1ac34b 700#endif
52826846 701 }
4b1ac34b
NB
702 if (force && okcnt == info.array.raid_disks-1) {
703 /* FIXME check event count */
b8a8ccf9 704 change += st->ss->update_super(&info, super, "force",
e5eac01f 705 devices[chosen_drive].devname, verbose, 0, NULL);
d013a55e 706 }
52826846 707
4b1ac34b 708 if (change) {
52826846 709 int fd;
8b0dabea 710 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846 711 if (fd < 0) {
353632d9 712 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
52826846 713 devices[chosen_drive].devname);
d55e3aef 714 if (must_close) close(mdfd);
52826846
NB
715 return 1;
716 }
96395475 717 if (st->ss->store_super(st, fd, super)) {
52826846
NB
718 close(fd);
719 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
720 devices[chosen_drive].devname);
d55e3aef 721 if (must_close) close(mdfd);
52826846
NB
722 return 1;
723 }
724 close(fd);
52826846
NB
725 }
726
353632d9
NB
727 /* If we are in the middle of a reshape we may need to restore saved data
728 * that was moved aside due to the reshape overwriting live data
729 * The code of doing this lives in Grow.c
730 */
39c5a909 731#ifndef MDASSEMBLE
353632d9
NB
732 if (info.reshape_active) {
733 int err = 0;
734 int *fdlist = malloc(sizeof(int)* bestcnt);
735 for (i=0; i<bestcnt; i++) {
736 int j = best[i];
737 if (j >= 0) {
738 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
739 if (fdlist[i] < 0) {
740 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
741 devices[j].devname);
742 err = 1;
743 break;
744 }
745 } else
746 fdlist[i] = -1;
747 }
748 if (!err)
06b0d786 749 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
353632d9
NB
750 while (i>0) {
751 i--;
752 if (fdlist[i]>=0) close(fdlist[i]);
753 }
754 if (err) {
755 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
d55e3aef 756 if (must_close) close(mdfd);
353632d9
NB
757 return err;
758 }
759 }
39c5a909 760#endif
aa88f531
NB
761 /* count number of in-sync devices according to the superblock.
762 * We must have this number to start the array without -s or -R
763 */
4b1ac34b 764 req_cnt = info.array.working_disks;
aa88f531 765
64c4757e
NB
766 /* Almost ready to actually *do* something */
767 if (!old_linux) {
fbf8a0b7
NB
768 int rv;
769 if ((vers % 100) >= 1) { /* can use different versions */
770 mdu_array_info_t inf;
771 memset(&inf, 0, sizeof(inf));
772 inf.major_version = st->ss->major;
773 inf.minor_version = st->minor_version;
774 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
b8a8ccf9 775 } else
fbf8a0b7
NB
776 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
777
778 if (rv) {
682c7051 779 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
64c4757e 780 mddev, strerror(errno));
d55e3aef 781 if (must_close) close(mdfd);
64c4757e
NB
782 return 1;
783 }
11484a63 784 if (ident->bitmap_fd >= 0) {
c82f047c
NB
785 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
786 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
d55e3aef 787 if (must_close) close(mdfd);
c82f047c
NB
788 return 1;
789 }
7ef02d01
NB
790 } else if (ident->bitmap_file) {
791 /* From config file */
792 int bmfd = open(ident->bitmap_file, O_RDWR);
793 if (bmfd < 0) {
794 fprintf(stderr, Name ": Could not open bitmap file %s\n",
795 ident->bitmap_file);
d55e3aef 796 if (must_close) close(mdfd);
7ef02d01
NB
797 return 1;
798 }
799 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
800 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
801 close(bmfd);
d55e3aef 802 if (must_close) close(mdfd);
7ef02d01
NB
803 return 1;
804 }
805 close(bmfd);
c82f047c 806 }
7ef02d01 807
52826846 808 /* First, add the raid disks, but add the chosen one last */
56eedc1a 809 for (i=0; i<= bestcnt; i++) {
52826846 810 int j;
56eedc1a 811 if (i < bestcnt) {
52826846
NB
812 j = best[i];
813 if (j == chosen_drive)
814 continue;
815 } else
816 j = chosen_drive;
817
56eedc1a 818 if (j >= 0 /* && devices[j].uptodate */) {
64c4757e
NB
819 mdu_disk_info_t disk;
820 memset(&disk, 0, sizeof(disk));
821 disk.major = devices[j].major;
822 disk.minor = devices[j].minor;
823 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
682c7051 824 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
64c4757e
NB
825 devices[j].devname,
826 mddev,
827 strerror(errno));
4b1ac34b 828 if (i < info.array.raid_disks || i == bestcnt)
52826846
NB
829 okcnt--;
830 else
831 sparecnt--;
dab6685f 832 } else if (verbose > 0)
52826846
NB
833 fprintf(stderr, Name ": added %s to %s as %d\n",
834 devices[j].devname, mddev, devices[j].raid_disk);
dab6685f 835 } else if (verbose > 0 && i < info.array.raid_disks)
682c7051 836 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
64c4757e
NB
837 i, mddev);
838 }
52826846 839
64c4757e 840 if (runstop == 1 ||
b8a8ccf9 841 (runstop <= 0 &&
265e0f17 842 ( enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt) &&
aa88f531
NB
843 (okcnt >= req_cnt || start_partial_ok)
844 ))) {
82b27616 845 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
dab6685f
NB
846 if (verbose >= 0) {
847 fprintf(stderr, Name ": %s has been started with %d drive%s",
848 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9 849 if (okcnt < info.array.raid_disks)
dab6685f
NB
850 fprintf(stderr, " (out of %d)", info.array.raid_disks);
851 if (sparecnt)
852 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
853 fprintf(stderr, ".\n");
854 }
589395d6
NB
855 if (must_close) {
856 int usecs = 1;
857 close(mdfd);
858 /* There is a nasty race with 'mdadm --monitor'.
859 * If it opens this device before we close it,
860 * it gets an incomplete open on which IO
861 * doesn't work and the capacity if wrong.
862 * If we reopen (to check for layered devices)
863 * before --monitor closes, we loose.
864 *
865 * So: wait upto 1 second for there to be
866 * a non-zero capacity.
867 */
868 while (usecs < 1000) {
869 mdfd = open(mddev, O_RDONLY);
870 if (mdfd >= 0) {
871 unsigned long size;
872 if (ioctl(mdfd, BLKGETSIZE, &size) == 0 &&
873 size > 0)
874 break;
875 close(mdfd);
876 }
877 usleep(usecs);
878 usecs <<= 1;
879 }
880 }
64c4757e 881 return 0;
82b27616 882 }
682c7051 883 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
64c4757e 884 mddev, strerror(errno));
d55e3aef 885 if (must_close) close(mdfd);
64c4757e
NB
886 return 1;
887 }
82b27616 888 if (runstop == -1) {
b8a8ccf9 889 fprintf(stderr, Name ": %s assembled from %d drive%s",
52826846 890 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9
NB
891 if (okcnt != info.array.raid_disks)
892 fprintf(stderr, " (out of %d)", info.array.raid_disks);
893 fprintf(stderr, ", but not started.\n");
d55e3aef 894 if (must_close) close(mdfd);
64c4757e 895 return 0;
82b27616 896 }
dab6685f
NB
897 if (verbose >= 0) {
898 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
899 if (sparecnt)
900 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
265e0f17 901 if (!enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt))
dab6685f
NB
902 fprintf(stderr, " - not enough to start the array.\n");
903 else {
904 if (req_cnt == info.array.raid_disks)
905 fprintf(stderr, " - need all %d to start it", req_cnt);
906 else
907 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
908 fprintf(stderr, " (use --run to insist).\n");
909 }
aa88f531 910 }
d55e3aef 911 if (must_close) close(mdfd);
64c4757e 912 return 1;
82b27616 913 } else {
52826846
NB
914 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
915 * been updated to point to the current locations of devices.
916 * so we can just start the array
82b27616 917 */
cd29a5c8 918 unsigned long dev;
0df46c2a 919 dev = makedev(devices[chosen_drive].major,
82b27616
NB
920 devices[chosen_drive].minor);
921 if (ioctl(mdfd, START_ARRAY, dev)) {
922 fprintf(stderr, Name ": Cannot start array: %s\n",
923 strerror(errno));
924 }
925
64c4757e 926 }
d55e3aef 927 if (must_close) close(mdfd);
e0d19036 928 return 0;
64c4757e 929}