]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Assemble.c
Add some comments to explain some of the bits of superswitch.
[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 */
583315d9 114 int clean = 0;
d55e3aef 115 int must_close = 0;
64c4757e 116 int old_linux = 0;
41a3b72a 117 int vers = 0; /* Keep gcc quite - it really is initialised */
64c4757e
NB
118 struct {
119 char *devname;
213ee40b
NB
120 int uptodate; /* set once we decide that this device is as
121 * recent as everything else in the array.
122 */
123 struct mdinfo i;
5787fa49 124 } *devices;
56eedc1a 125 int *best = NULL; /* indexed by raid_disk */
98c6faba
NB
126 unsigned int bestcnt = 0;
127 int devcnt = 0;
128 unsigned int okcnt, sparecnt;
129 unsigned int req_cnt;
130 unsigned int i;
64c4757e 131 int most_recent = 0;
cd29a5c8 132 int chosen_drive;
52826846 133 int change = 0;
cd29a5c8 134 int inargv = 0;
bf4fb153 135 int bitmap_done;
589395d6 136 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
98c6faba 137 unsigned int num_devs;
5787fa49 138 mddev_dev_t tmpdev;
4b1ac34b 139 struct mdinfo info;
265e0f17 140 char *avail;
308e1801 141 int nextspare = 0;
8a46fe84 142
682c7051 143 if (get_linux_version() < 2004000)
64c4757e
NB
144 old_linux = 1;
145
8a46fe84
NB
146 if (mdfd >= 0) {
147 vers = md_get_version(mdfd);
148 if (vers <= 0) {
149 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
150 return 1;
151 }
152 if (vers < 9000) {
153 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
154 " Upgrade your kernel or try --build\n");
155 return 1;
156 }
64c4757e 157
8a46fe84
NB
158 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
159 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
160 mddev);
161 return 1;
162 }
163 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
164 }
64c4757e 165 /*
52826846
NB
166 * If any subdevs are listed, then any that don't
167 * match ident are discarded. Remainder must all match and
168 * become the array.
169 * If no subdevs, then we scan all devices in the config file, but
170 * there must be something in the identity
64c4757e 171 */
64c4757e 172
cd29a5c8 173 if (!devlist &&
52826846
NB
174 ident->uuid_set == 0 &&
175 ident->super_minor < 0 &&
176 ident->devices == NULL) {
177 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
8a46fe84 178 mddev ? mddev : "further assembly");
52826846 179 return 1;
64c4757e 180 }
cd29a5c8 181 if (devlist == NULL)
8aec876d 182 devlist = conf_get_devs();
da6b5ca9
NB
183 else if (mdfd >= 0)
184 inargv = 1;
64c4757e 185
60e1bc1a
NB
186 try_again:
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;
3da92f27 214 struct supertype *tst = dup_super(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 224
8b0dabea 225 dfd = dev_open(devname, O_RDONLY|O_EXCL);
64c4757e 226 if (dfd < 0) {
6a41304b 227 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 228 fprintf(stderr, Name ": cannot open device %s: %s\n",
64c4757e 229 devname, strerror(errno));
da6b5ca9 230 tmpdev->used = 2;
52826846
NB
231 } else if (fstat(dfd, &stb)< 0) {
232 /* Impossible! */
233 fprintf(stderr, Name ": fstat failed for %s: %s\n",
234 devname, strerror(errno));
da6b5ca9 235 tmpdev->used = 2;
cd29a5c8
NB
236 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
237 fprintf(stderr, Name ": %s is not a block device.\n",
52826846 238 devname);
da6b5ca9 239 tmpdev->used = 2;
82d9eba6 240 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
6a41304b 241 if ((inargv && verbose >= 0) || verbose > 0)
da6b5ca9
NB
242 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
243 devname);
244 tmpdev->used = 2;
3da92f27 245 } else if (tst->ss->load_super(tst,dfd, NULL)) {
6a41304b 246 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 247 fprintf( stderr, Name ": no RAID superblock on %s\n",
64c4757e 248 devname);
52826846 249 } else {
3da92f27 250 tst->ss->getinfo_super(tst, &info);
64c4757e 251 }
c06487ce 252 if (dfd >= 0) close(dfd);
52826846 253
838acbc2 254 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
3da92f27
NB
255 (!tst || !tst->sb ||
256 same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
6a41304b 257 if ((inargv && verbose >= 0) || verbose > 0)
52826846
NB
258 fprintf(stderr, Name ": %s has wrong uuid.\n",
259 devname);
df37ffc0 260 goto loop;
82b27616 261 }
c4f12c13 262 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
3da92f27
NB
263 (!tst || !tst->sb ||
264 name_matches(info.name, ident->name, homehost)==0)) {
6a41304b 265 if ((inargv && verbose >= 0) || verbose > 0)
947fd4dd
NB
266 fprintf(stderr, Name ": %s has wrong name.\n",
267 devname);
df37ffc0 268 goto loop;
947fd4dd 269 }
98c6faba 270 if (ident->super_minor != UnSet &&
3da92f27
NB
271 (!tst || !tst->sb ||
272 ident->super_minor != info.array.md_minor)) {
6a41304b 273 if ((inargv && verbose >= 0) || verbose > 0)
52826846 274 fprintf(stderr, Name ": %s has wrong super-minor.\n",
64c4757e 275 devname);
df37ffc0 276 goto loop;
64c4757e 277 }
98c6faba 278 if (ident->level != UnSet &&
3da92f27
NB
279 (!tst || !tst->sb ||
280 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);
df37ffc0 284 goto loop;
cd29a5c8 285 }
98c6faba 286 if (ident->raid_disks != UnSet &&
3da92f27
NB
287 (!tst || !tst->sb ||
288 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);
df37ffc0 292 goto loop;
cd29a5c8 293 }
da6b5ca9 294 if (mdfd < 0) {
3da92f27 295 if (tst == NULL || tst->sb == NULL)
da6b5ca9 296 continue;
589395d6 297 if (update == NULL &&
3da92f27 298 tst->ss->match_home(tst, 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... */
df37ffc0 305 goto loop;
da6b5ca9
NB
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
3da92f27 315 if (!tst || !tst->sb) {
52826846
NB
316 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
317 devname);
ddd1a492
NB
318 if (st)
319 st->ss->free_super(st);
52826846
NB
320 return 1;
321 }
838acbc2 322
83b6208e 323 if (st == NULL)
3da92f27
NB
324 st = dup_super(tst);
325 if (st->minor_version == -1)
326 st->minor_version = tst->minor_version;
83b6208e
NB
327 if (st->ss != tst->ss ||
328 st->minor_version != tst->minor_version ||
64557c33 329 st->ss->compare_super(st, tst) != 0) {
83b6208e 330 /* Some mismatch. If exactly one array matches this host,
da6b5ca9
NB
331 * we can resolve on that one.
332 * Or, if we are auto assembling, we just ignore the second
333 * for now.
83b6208e 334 */
da6b5ca9 335 if (mdfd < 0)
df37ffc0 336 goto loop;
83b6208e 337 if (homehost) {
3da92f27
NB
338 int first = st->ss->match_home(st, homehost);
339 int last = tst->ss->match_home(tst, homehost);
83b6208e
NB
340 if (first+last == 1) {
341 /* We can do something */
342 if (first) {/* just ignore this one */
343 if ((inargv && verbose >= 0) || verbose > 0)
344 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
345 devname);
df37ffc0 346 goto loop;
83b6208e
NB
347 } else { /* reject all those sofar */
348 mddev_dev_t td;
349 if ((inargv && verbose >= 0) || verbose > 0)
350 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
351 devname);
352 for (td=devlist; td != tmpdev; td=td->next)
353 if (td->used == 1)
354 td->used = 0;
355 tmpdev->used = 1;
df37ffc0 356 goto loop;
83b6208e
NB
357 }
358 }
359 }
52826846
NB
360 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
361 devname);
3da92f27
NB
362 tst->ss->free_super(tst);
363 st->ss->free_super(st);
52826846 364 return 1;
64c4757e
NB
365 }
366
811e6cbe 367 tmpdev->used = 1;
df37ffc0
NB
368
369 loop:
3d2b16e7
NB
370 if (tst)
371 tst->ss->free_super(tst);
811e6cbe
NB
372 }
373
8a46fe84
NB
374 if (mdfd < 0) {
375 /* So... it is up to me to open the device.
376 * We create a name '/dev/md/XXX' based on the info in the
377 * superblock, and call open_mddev on that
378 */
da6b5ca9
NB
379 mdu_array_info_t inf;
380 char *c;
1c203a4b 381 if (!st || !st->sb) {
d55e3aef 382 return 2;
da6b5ca9 383 }
3da92f27 384 st->ss->getinfo_super(st, &info);
da6b5ca9
NB
385 c = strchr(info.name, ':');
386 if (c) c++; else c= info.name;
41a3b72a
NB
387 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
388 /* /dev/md/d0 style for partitionable */
389 asprintf(&mddev, "/dev/md/d%s", c);
390 else
391 asprintf(&mddev, "/dev/md/%s", c);
da6b5ca9 392 mdfd = open_mddev(mddev, ident->autof);
60e1bc1a 393 if (mdfd < 0) {
3da92f27 394 st->ss->free_super(st);
60e1bc1a 395 free(devices);
60e1bc1a
NB
396 goto try_again;
397 }
da6b5ca9
NB
398 vers = md_get_version(mdfd);
399 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
60e1bc1a
NB
400 for (tmpdev = devlist ;
401 tmpdev && tmpdev->used != 1;
402 tmpdev = tmpdev->next)
403 ;
da6b5ca9 404 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
60e1bc1a
NB
405 if (tmpdev)
406 fprintf(stderr, Name ": %s needed for %s...\n",
407 mddev, tmpdev->devname);
da6b5ca9 408 close(mdfd);
60e1bc1a 409 mdfd = -1;
3da92f27 410 st->ss->free_super(st);
60e1bc1a 411 free(devices);
60e1bc1a 412 goto try_again;
da6b5ca9 413 }
d55e3aef 414 must_close = 1;
8a46fe84
NB
415 }
416
811e6cbe 417 /* Ok, no bad inconsistancy, we can try updating etc */
bf4fb153 418 bitmap_done = 0;
da6b5ca9 419 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
811e6cbe
NB
420 char *devname = tmpdev->devname;
421 struct stat stb;
5787fa49 422 /* looks like a good enough match to update the super block if needed */
d1f1011b 423#ifndef MDASSEMBLE
5787fa49 424 if (update) {
811e6cbe 425 int dfd;
4b1ac34b
NB
426 /* prepare useful information in info structures */
427 struct stat stb2;
3da92f27 428 struct supertype *tst;
4b1ac34b 429 fstat(mdfd, &stb2);
7d99579f
NB
430
431 if (strcmp(update, "uuid")==0 &&
432 !ident->uuid_set) {
433 int rfd;
434 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
435 read(rfd, ident->uuid, 16) != 16) {
436 *(__u32*)(ident->uuid) = random();
437 *(__u32*)(ident->uuid+1) = random();
438 *(__u32*)(ident->uuid+2) = random();
439 *(__u32*)(ident->uuid+3) = random();
440 }
441 if (rfd >= 0) close(rfd);
7d99579f 442 }
811e6cbe
NB
443 dfd = dev_open(devname, O_RDWR|O_EXCL);
444
0430ed48
NB
445 remove_partitions(dfd);
446
3da92f27
NB
447 tst = dup_super(st);
448 tst->ss->load_super(tst, dfd, NULL);
449 tst->ss->getinfo_super(tst, &info);
811e6cbe 450
7d99579f 451 memcpy(info.uuid, ident->uuid, 16);
e5eac01f 452 strcpy(info.name, ident->name);
811e6cbe
NB
453 info.array.md_minor = minor(stb2.st_rdev);
454
3da92f27
NB
455 tst->ss->update_super(tst, &info, update,
456 devname, verbose,
457 ident->uuid_set, homehost);
e5eac01f
NB
458 if (strcmp(update, "uuid")==0 &&
459 !ident->uuid_set) {
460 ident->uuid_set = 1;
461 memcpy(ident->uuid, info.uuid, 16);
462 }
b8a8ccf9 463 if (dfd < 0)
5787fa49
NB
464 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
465 devname);
3da92f27 466 else if (tst->ss->store_super(tst, dfd))
5787fa49
NB
467 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
468 devname);
469 if (dfd >= 0)
470 close(dfd);
8131b493
NB
471
472 if (strcmp(update, "uuid")==0 &&
bf4fb153 473 ident->bitmap_fd >= 0 && !bitmap_done) {
3da92f27
NB
474 if (bitmap_update_uuid(ident->bitmap_fd,
475 info.uuid,
476 tst->ss->swapuuid) != 0)
bf4fb153
NB
477 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
478 else
479 bitmap_done = 1;
480 }
3da92f27 481 tst->ss->free_super(tst);
d1f1011b
NB
482 } else
483#endif
484 {
30e1b9a5 485 struct supertype *tst = dup_super(st);
da6b5ca9
NB
486 int dfd;
487 dfd = dev_open(devname, O_RDWR|O_EXCL);
488
0430ed48
NB
489 remove_partitions(dfd);
490
3da92f27
NB
491 tst->ss->load_super(tst, dfd, NULL);
492 tst->ss->getinfo_super(tst, &info);
493 tst->ss->free_super(tst);
da6b5ca9 494 close(dfd);
5787fa49
NB
495 }
496
811e6cbe
NB
497 stat(devname, &stb);
498
dab6685f 499 if (verbose > 0)
cd29a5c8 500 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
4b1ac34b 501 devname, mddev, info.disk.raid_disk);
64c4757e 502 devices[devcnt].devname = devname;
64c4757e 503 devices[devcnt].uptodate = 0;
213ee40b
NB
504 devices[devcnt].i = info;
505 devices[devcnt].i.disk.major = major(stb.st_rdev);
506 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
64c4757e 507 if (most_recent < devcnt) {
213ee40b
NB
508 if (devices[devcnt].i.events
509 > devices[most_recent].i.events)
64c4757e
NB
510 most_recent = devcnt;
511 }
b8a8ccf9 512 if (info.array.level == -4)
5787fa49
NB
513 /* with multipath, the raid_disk from the superblock is meaningless */
514 i = devcnt;
515 else
213ee40b 516 i = devices[devcnt].i.disk.raid_disk;
308e1801
NB
517 if (i+1 == 0) {
518 if (nextspare < info.array.raid_disks)
519 nextspare = info.array.raid_disks;
520 i = nextspare++;
08110d41
NB
521 } else {
522 if (i >= info.array.raid_disks &&
523 i >= nextspare)
524 nextspare = i+1;
308e1801 525 }
98c6faba 526 if (i < 10000) {
56eedc1a 527 if (i >= bestcnt) {
98c6faba 528 unsigned int newbestcnt = i+10;
56eedc1a 529 int *newbest = malloc(sizeof(int)*newbestcnt);
98c6faba 530 unsigned int c;
56eedc1a
NB
531 for (c=0; c < newbestcnt; c++)
532 if (c < bestcnt)
533 newbest[c] = best[c];
534 else
535 newbest[c] = -1;
536 if (best)free(best);
537 best = newbest;
538 bestcnt = newbestcnt;
539 }
6a255b69 540 if (best[i] >=0 &&
213ee40b
NB
541 devices[best[i]].i.events
542 == devices[devcnt].i.events
543 && (devices[best[i]].i.disk.minor
544 != devices[devcnt].i.disk.minor)
b8ac1967
NB
545 && st->ss == &super0
546 && info.array.level != LEVEL_MULTIPATH) {
6a255b69
NB
547 /* two different devices with identical superblock.
548 * Could be a mis-detection caused by overlapping
549 * partitions. fail-safe.
550 */
551 fprintf(stderr, Name ": WARNING %s and %s appear"
552 " to have very similar superblocks.\n"
553 " If they are really different, "
554 "please --zero the superblock on one\n"
d2df86e0
NB
555 " If they are the same or overlap,"
556 " please remove one from %s.\n",
557 devices[best[i]].devname, devname,
558 inargv ? "the list" :
559 "the\n DEVICE list in mdadm.conf"
560 );
6a255b69
NB
561 if (must_close) close(mdfd);
562 return 1;
563 }
52826846 564 if (best[i] == -1
213ee40b
NB
565 || (devices[best[i]].i.events
566 < devices[devcnt].i.events))
52826846 567 best[i] = devcnt;
56eedc1a 568 }
64c4757e 569 devcnt++;
df37ffc0 570 }
4b1ac34b 571
64c4757e 572 if (devcnt == 0) {
682c7051 573 fprintf(stderr, Name ": no devices found for %s\n",
64c4757e 574 mddev);
3d2b16e7
NB
575 if (st)
576 st->ss->free_super(st);
d55e3aef 577 if (must_close) close(mdfd);
64c4757e
NB
578 return 1;
579 }
4b1ac34b 580
3d2b16e7
NB
581 if (update && strcmp(update, "byteorder")==0)
582 st->minor_version = 90;
583
3da92f27 584 st->ss->getinfo_super(st, &info);
583315d9 585 clean = info.array.state & 1;
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)
213ee40b
NB
604 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
605 if (!(devices[j].i.disk.state
606 & (1<<MD_DISK_FAULTY)))
aa88f531 607 sparecnt++;
d013a55e 608 continue;
aa88f531 609 }
213ee40b
NB
610 if (devices[j].i.events+event_margin >=
611 devices[most_recent].i.events) {
64c4757e 612 devices[j].uptodate = 1;
265e0f17 613 if (i < info.array.raid_disks) {
52826846 614 okcnt++;
265e0f17
NB
615 avail[i]=1;
616 } else
52826846 617 sparecnt++;
64c4757e
NB
618 }
619 }
265e0f17 620 while (force && !enough(info.array.level, info.array.raid_disks,
583315d9 621 info.array.layout, 1,
265e0f17 622 avail, okcnt)) {
64c4757e
NB
623 /* Choose the newest best drive which is
624 * not up-to-date, update the superblock
625 * and add it.
626 */
52826846 627 int fd;
3da92f27 628 struct supertype *tst;
c5a6f9a6 629 long long current_events;
cd29a5c8 630 chosen_drive = -1;
4b1ac34b 631 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
52826846
NB
632 int j = best[i];
633 if (j>=0 &&
634 !devices[j].uptodate &&
213ee40b 635 devices[j].i.events > 0 &&
52826846 636 (chosen_drive < 0 ||
213ee40b
NB
637 devices[j].i.events
638 > devices[chosen_drive].i.events))
52826846
NB
639 chosen_drive = j;
640 }
641 if (chosen_drive < 0)
642 break;
213ee40b 643 current_events = devices[chosen_drive].i.events;
c5a6f9a6 644 add_another:
dab6685f
NB
645 if (verbose >= 0)
646 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
213ee40b
NB
647 devices[chosen_drive].devname,
648 devices[chosen_drive].i.disk.raid_disk,
649 (int)(devices[chosen_drive].i.events),
650 (int)(devices[most_recent].i.events));
8b0dabea 651 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846
NB
652 if (fd < 0) {
653 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
654 devices[chosen_drive].devname);
213ee40b 655 devices[chosen_drive].i.events = 0;
52826846
NB
656 continue;
657 }
3da92f27 658 tst = dup_super(st);
60b435db 659 if (tst->ss->load_super(tst,fd, NULL)) {
52826846
NB
660 close(fd);
661 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
662 devices[chosen_drive].devname);
213ee40b 663 devices[chosen_drive].i.events = 0;
52826846
NB
664 continue;
665 }
213ee40b 666 info.events = devices[most_recent].i.events;
3da92f27 667 tst->ss->update_super(tst, &info, "force-one",
67a8c82d
NB
668 devices[chosen_drive].devname, verbose,
669 0, NULL);
4b1ac34b 670
3da92f27 671 if (tst->ss->store_super(tst, fd)) {
52826846
NB
672 close(fd);
673 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
674 devices[chosen_drive].devname);
213ee40b 675 devices[chosen_drive].i.events = 0;
3da92f27 676 tst->ss->free_super(tst);
52826846
NB
677 continue;
678 }
679 close(fd);
213ee40b 680 devices[chosen_drive].i.events = devices[most_recent].i.events;
52826846 681 devices[chosen_drive].uptodate = 1;
265e0f17 682 avail[chosen_drive] = 1;
52826846 683 okcnt++;
3da92f27 684 tst->ss->free_super(tst);
c5a6f9a6
NB
685
686 /* If there are any other drives of the same vintage,
687 * add them in as well. We can't lose and we might gain
688 */
689 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
690 int j = best[i];
691 if (j >= 0 &&
692 !devices[j].uptodate &&
213ee40b
NB
693 devices[j].i.events > 0 &&
694 devices[j].i.events == current_events) {
c5a6f9a6
NB
695 chosen_drive = j;
696 goto add_another;
697 }
698 }
64c4757e 699 }
52826846
NB
700
701 /* Now we want to look at the superblock which the kernel will base things on
702 * and compare the devices that we think are working with the devices that the
703 * superblock thinks are working.
704 * If there are differences and --force is given, then update this chosen
705 * superblock.
706 */
cd29a5c8 707 chosen_drive = -1;
3da92f27 708 st->ss->free_super(st);
56eedc1a 709 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
52826846
NB
710 int j = best[i];
711 int fd;
4b1ac34b 712
52826846
NB
713 if (j<0)
714 continue;
715 if (!devices[j].uptodate)
716 continue;
717 chosen_drive = j;
8b0dabea 718 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
52826846
NB
719 fprintf(stderr, Name ": Cannot open %s: %s\n",
720 devices[j].devname, strerror(errno));
d55e3aef 721 if (must_close) close(mdfd);
52826846
NB
722 return 1;
723 }
3da92f27 724 if (st->ss->load_super(st,fd, NULL)) {
52826846
NB
725 close(fd);
726 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
727 devices[j].devname);
d55e3aef 728 if (must_close) close(mdfd);
52826846
NB
729 return 1;
730 }
731 close(fd);
732 }
3da92f27 733 if (st->sb == NULL) {
4b1ac34b 734 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
d55e3aef 735 if (must_close) close(mdfd);
4b1ac34b
NB
736 return 1;
737 }
3da92f27 738 st->ss->getinfo_super(st, &info);
56eedc1a 739 for (i=0; i<bestcnt; i++) {
52826846 740 int j = best[i];
98c6faba 741 unsigned int desired_state;
11a3e71d 742
4b1ac34b 743 if (i < info.array.raid_disks)
11a3e71d
NB
744 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
745 else
746 desired_state = 0;
747
52826846
NB
748 if (j<0)
749 continue;
750 if (!devices[j].uptodate)
751 continue;
4b1ac34b 752
213ee40b
NB
753 devices[j].i.disk.state = desired_state;
754
755 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
68c7d6d7 756 verbose, 0, NULL)) {
52826846 757 if (force) {
dab6685f
NB
758 if (verbose >= 0)
759 fprintf(stderr, Name ": "
760 "clearing FAULTY flag for device %d in %s for %s\n",
761 j, mddev, devices[j].devname);
4b1ac34b 762 change = 1;
52826846 763 } else {
dab6685f
NB
764 if (verbose >= -1)
765 fprintf(stderr, Name ": "
766 "device %d in %s has wrong state in superblock, but %s seems ok\n",
767 i, mddev, devices[j].devname);
52826846
NB
768 }
769 }
4b1ac34b 770#if 0
213ee40b 771 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
52826846
NB
772 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
773 i, mddev);
774 }
4b1ac34b 775#endif
52826846 776 }
c5a6f9a6
NB
777 if (force && !clean &&
778 !enough(info.array.level, info.array.raid_disks,
779 info.array.layout, clean,
780 avail, okcnt)) {
3da92f27 781 change += st->ss->update_super(st, &info, "force-array",
67a8c82d
NB
782 devices[chosen_drive].devname, verbose,
783 0, NULL);
583315d9 784 clean = 1;
d013a55e 785 }
52826846 786
4b1ac34b 787 if (change) {
52826846 788 int fd;
8b0dabea 789 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846 790 if (fd < 0) {
353632d9 791 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
52826846 792 devices[chosen_drive].devname);
d55e3aef 793 if (must_close) close(mdfd);
52826846
NB
794 return 1;
795 }
3da92f27 796 if (st->ss->store_super(st, fd)) {
52826846
NB
797 close(fd);
798 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
799 devices[chosen_drive].devname);
d55e3aef 800 if (must_close) close(mdfd);
52826846
NB
801 return 1;
802 }
803 close(fd);
52826846
NB
804 }
805
353632d9
NB
806 /* If we are in the middle of a reshape we may need to restore saved data
807 * that was moved aside due to the reshape overwriting live data
808 * The code of doing this lives in Grow.c
809 */
39c5a909 810#ifndef MDASSEMBLE
353632d9
NB
811 if (info.reshape_active) {
812 int err = 0;
813 int *fdlist = malloc(sizeof(int)* bestcnt);
814 for (i=0; i<bestcnt; i++) {
815 int j = best[i];
816 if (j >= 0) {
817 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
818 if (fdlist[i] < 0) {
819 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
820 devices[j].devname);
821 err = 1;
822 break;
823 }
824 } else
825 fdlist[i] = -1;
826 }
827 if (!err)
06b0d786 828 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
353632d9
NB
829 while (i>0) {
830 i--;
831 if (fdlist[i]>=0) close(fdlist[i]);
832 }
833 if (err) {
834 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
d55e3aef 835 if (must_close) close(mdfd);
353632d9
NB
836 return err;
837 }
838 }
39c5a909 839#endif
aa88f531
NB
840 /* count number of in-sync devices according to the superblock.
841 * We must have this number to start the array without -s or -R
842 */
4b1ac34b 843 req_cnt = info.array.working_disks;
aa88f531 844
64c4757e
NB
845 /* Almost ready to actually *do* something */
846 if (!old_linux) {
fbf8a0b7 847 int rv;
a19c88b8
NB
848
849#ifndef MDASSEMBLE
850 struct mdinfo *sra;
851 if (st->ss->external) {
852 char ver[100];
159c3a1a 853 strcat(strcpy(ver, "external:"), info.text_version);
a19c88b8
NB
854 sra = sysfs_read(mdfd, 0, 0);
855 if ((vers % 100) < 2 ||
856 sra == NULL ||
857 sysfs_set_str(sra, NULL, "metadata_version",
858 ver) < 0) {
859 fprintf(stderr, Name ": This kernel does not "
860 "support external metadata.\n");
861 return 1;
862 }
863 rv = sysfs_set_array(sra, &info);
864 } else
865#endif
fbf8a0b7
NB
866 if ((vers % 100) >= 1) { /* can use different versions */
867 mdu_array_info_t inf;
868 memset(&inf, 0, sizeof(inf));
b8ac1967
NB
869 inf.major_version = info.array.major_version;
870 inf.minor_version = info.array.minor_version;
fbf8a0b7 871 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
b8a8ccf9 872 } else
fbf8a0b7
NB
873 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
874
875 if (rv) {
682c7051 876 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
64c4757e 877 mddev, strerror(errno));
d55e3aef 878 if (must_close) close(mdfd);
64c4757e
NB
879 return 1;
880 }
11484a63 881 if (ident->bitmap_fd >= 0) {
c82f047c
NB
882 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
883 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
d55e3aef 884 if (must_close) close(mdfd);
c82f047c
NB
885 return 1;
886 }
7ef02d01
NB
887 } else if (ident->bitmap_file) {
888 /* From config file */
889 int bmfd = open(ident->bitmap_file, O_RDWR);
890 if (bmfd < 0) {
891 fprintf(stderr, Name ": Could not open bitmap file %s\n",
892 ident->bitmap_file);
d55e3aef 893 if (must_close) close(mdfd);
7ef02d01
NB
894 return 1;
895 }
896 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
897 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
898 close(bmfd);
d55e3aef 899 if (must_close) close(mdfd);
7ef02d01
NB
900 return 1;
901 }
902 close(bmfd);
c82f047c 903 }
7ef02d01 904
52826846 905 /* First, add the raid disks, but add the chosen one last */
56eedc1a 906 for (i=0; i<= bestcnt; i++) {
52826846 907 int j;
56eedc1a 908 if (i < bestcnt) {
52826846
NB
909 j = best[i];
910 if (j == chosen_drive)
911 continue;
912 } else
913 j = chosen_drive;
914
56eedc1a 915 if (j >= 0 /* && devices[j].uptodate */) {
a19c88b8
NB
916#ifndef MDASSEMBLE
917 if (st->ss->external) {
2318b9f0
NB
918 devices[j].i.disk.number =
919 devices[j].i.disk.raid_disk;
920 st->ss->getinfo_super_n(st,
921 &devices[j].i);
922 rv = sysfs_add_disk(sra,
923 &devices[j].i);
a19c88b8
NB
924 } else
925#endif
926 rv = ioctl(mdfd, ADD_NEW_DISK,
927 &devices[j].i.disk);
928 if (rv) {
213ee40b
NB
929 fprintf(stderr, Name ": failed to add "
930 "%s to %s: %s\n",
64c4757e
NB
931 devices[j].devname,
932 mddev,
933 strerror(errno));
213ee40b
NB
934 if (i < info.array.raid_disks
935 || i == bestcnt)
52826846
NB
936 okcnt--;
937 else
938 sparecnt--;
dab6685f 939 } else if (verbose > 0)
213ee40b
NB
940 fprintf(stderr, Name ": added %s "
941 "to %s as %d\n",
942 devices[j].devname, mddev,
943 devices[j].i.disk.raid_disk);
dab6685f 944 } else if (verbose > 0 && i < info.array.raid_disks)
213ee40b
NB
945 fprintf(stderr, Name ": no uptodate device for "
946 "slot %d of %s\n",
64c4757e
NB
947 i, mddev);
948 }
aba69144 949
598f0d58
NB
950 if (info.array.level == LEVEL_CONTAINER) {
951 if (verbose >= 0) {
952 fprintf(stderr, Name ": Container %s has been "
953 "assembled with %d drive%s",
954 mddev, okcnt, okcnt==1?"":"s");
955 if (okcnt < info.array.raid_disks)
956 fprintf(stderr, " (out of %d)",
957 info.array.raid_disks);
958 fprintf(stderr, "\n");
959 }
960 if (must_close)
961 close(mdfd);
962 return 0;
963 }
964
64c4757e 965 if (runstop == 1 ||
b8a8ccf9 966 (runstop <= 0 &&
583315d9
NB
967 ( enough(info.array.level, info.array.raid_disks,
968 info.array.layout, clean, avail, okcnt) &&
aa88f531
NB
969 (okcnt >= req_cnt || start_partial_ok)
970 ))) {
82b27616 971 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
dab6685f
NB
972 if (verbose >= 0) {
973 fprintf(stderr, Name ": %s has been started with %d drive%s",
974 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9 975 if (okcnt < info.array.raid_disks)
dab6685f
NB
976 fprintf(stderr, " (out of %d)", info.array.raid_disks);
977 if (sparecnt)
978 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
979 fprintf(stderr, ".\n");
980 }
589395d6
NB
981 if (must_close) {
982 int usecs = 1;
983 close(mdfd);
984 /* There is a nasty race with 'mdadm --monitor'.
985 * If it opens this device before we close it,
986 * it gets an incomplete open on which IO
598f0d58
NB
987 * doesn't work and the capacity is
988 * wrong.
589395d6
NB
989 * If we reopen (to check for layered devices)
990 * before --monitor closes, we loose.
991 *
992 * So: wait upto 1 second for there to be
993 * a non-zero capacity.
994 */
995 while (usecs < 1000) {
996 mdfd = open(mddev, O_RDONLY);
997 if (mdfd >= 0) {
beae1dfe
NB
998 unsigned long long size;
999 if (get_dev_size(mdfd, NULL, &size) &&
589395d6
NB
1000 size > 0)
1001 break;
1002 close(mdfd);
1003 }
1004 usleep(usecs);
1005 usecs <<= 1;
1006 }
1007 }
64c4757e 1008 return 0;
82b27616 1009 }
682c7051 1010 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
64c4757e 1011 mddev, strerror(errno));
583315d9
NB
1012
1013 if (!enough(info.array.level, info.array.raid_disks,
1014 info.array.layout, 1, avail, okcnt))
1015 fprintf(stderr, Name ": Not enough devices to "
1016 "start the array.\n");
1017 else if (!enough(info.array.level,
1018 info.array.raid_disks,
1019 info.array.layout, clean,
1020 avail, okcnt))
1021 fprintf(stderr, Name ": Not enough devices to "
1022 "start the array while not clean "
1023 "- consider --force.\n");
1024
1c203a4b
NB
1025 if (must_close) {
1026 ioctl(mdfd, STOP_ARRAY, NULL);
1027 close(mdfd);
1028 }
64c4757e
NB
1029 return 1;
1030 }
82b27616 1031 if (runstop == -1) {
b8a8ccf9 1032 fprintf(stderr, Name ": %s assembled from %d drive%s",
52826846 1033 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9
NB
1034 if (okcnt != info.array.raid_disks)
1035 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1036 fprintf(stderr, ", but not started.\n");
d55e3aef 1037 if (must_close) close(mdfd);
64c4757e 1038 return 0;
82b27616 1039 }
4855f95c 1040 if (verbose >= -1) {
dab6685f
NB
1041 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1042 if (sparecnt)
1043 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
583315d9
NB
1044 if (!enough(info.array.level, info.array.raid_disks,
1045 info.array.layout, 1, avail, okcnt))
dab6685f 1046 fprintf(stderr, " - not enough to start the array.\n");
583315d9
NB
1047 else if (!enough(info.array.level,
1048 info.array.raid_disks,
1049 info.array.layout, clean,
1050 avail, okcnt))
1051 fprintf(stderr, " - not enough to start the "
1052 "array while not clean - consider "
1053 "--force.\n");
dab6685f
NB
1054 else {
1055 if (req_cnt == info.array.raid_disks)
1056 fprintf(stderr, " - need all %d to start it", req_cnt);
1057 else
1058 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1059 fprintf(stderr, " (use --run to insist).\n");
1060 }
aa88f531 1061 }
1c203a4b
NB
1062 if (must_close) {
1063 ioctl(mdfd, STOP_ARRAY, NULL);
1064 close(mdfd);
1065 }
64c4757e 1066 return 1;
82b27616 1067 } else {
52826846
NB
1068 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1069 * been updated to point to the current locations of devices.
1070 * so we can just start the array
82b27616 1071 */
cd29a5c8 1072 unsigned long dev;
213ee40b
NB
1073 dev = makedev(devices[chosen_drive].i.disk.major,
1074 devices[chosen_drive].i.disk.minor);
82b27616
NB
1075 if (ioctl(mdfd, START_ARRAY, dev)) {
1076 fprintf(stderr, Name ": Cannot start array: %s\n",
1077 strerror(errno));
1078 }
aba69144 1079
64c4757e 1080 }
d55e3aef 1081 if (must_close) close(mdfd);
e0d19036 1082 return 0;
64c4757e 1083}