]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Assemble.c
A recent patch broke --create --level=faulty - fix it.
[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 */
4b1ac34b 118 void *first_super = NULL, *super = NULL;
64c4757e
NB
119 struct {
120 char *devname;
98c6faba
NB
121 unsigned int major, minor;
122 unsigned int oldmajor, oldminor;
64c4757e 123 long long events;
64c4757e 124 int uptodate;
d013a55e 125 int state;
52826846 126 int raid_disk;
6a41304b 127 int disk_nr;
5787fa49 128 } *devices;
56eedc1a 129 int *best = NULL; /* indexed by raid_disk */
98c6faba
NB
130 unsigned int bestcnt = 0;
131 int devcnt = 0;
132 unsigned int okcnt, sparecnt;
133 unsigned int req_cnt;
134 unsigned int i;
64c4757e 135 int most_recent = 0;
cd29a5c8 136 int chosen_drive;
52826846 137 int change = 0;
cd29a5c8 138 int inargv = 0;
bf4fb153 139 int bitmap_done;
589395d6 140 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
98c6faba 141 unsigned int num_devs;
5787fa49 142 mddev_dev_t tmpdev;
4b1ac34b 143 struct mdinfo info;
265e0f17 144 char *avail;
308e1801 145 int nextspare = 0;
8a46fe84 146
682c7051 147 if (get_linux_version() < 2004000)
64c4757e
NB
148 old_linux = 1;
149
8a46fe84
NB
150 if (mdfd >= 0) {
151 vers = md_get_version(mdfd);
152 if (vers <= 0) {
153 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
154 return 1;
155 }
156 if (vers < 9000) {
157 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
158 " Upgrade your kernel or try --build\n");
159 return 1;
160 }
64c4757e 161
8a46fe84
NB
162 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
163 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
164 mddev);
165 return 1;
166 }
167 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
168 }
64c4757e 169 /*
52826846
NB
170 * If any subdevs are listed, then any that don't
171 * match ident are discarded. Remainder must all match and
172 * become the array.
173 * If no subdevs, then we scan all devices in the config file, but
174 * there must be something in the identity
64c4757e 175 */
64c4757e 176
cd29a5c8 177 if (!devlist &&
52826846
NB
178 ident->uuid_set == 0 &&
179 ident->super_minor < 0 &&
180 ident->devices == NULL) {
181 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
8a46fe84 182 mddev ? mddev : "further assembly");
52826846 183 return 1;
64c4757e 184 }
cd29a5c8 185 if (devlist == NULL)
8aec876d 186 devlist = conf_get_devs();
da6b5ca9
NB
187 else if (mdfd >= 0)
188 inargv = 1;
64c4757e 189
60e1bc1a
NB
190 try_again:
191
5787fa49
NB
192 tmpdev = devlist; num_devs = 0;
193 while (tmpdev) {
da6b5ca9
NB
194 if (tmpdev->used)
195 tmpdev->used = 2;
196 else
197 num_devs++;
5787fa49
NB
198 tmpdev = tmpdev->next;
199 }
5787fa49
NB
200 devices = malloc(num_devs * sizeof(*devices));
201
82d9eba6 202 if (!st && ident->st) st = ident->st;
64c4757e 203
dab6685f 204 if (verbose>0)
82b27616 205 fprintf(stderr, Name ": looking for devices for %s\n",
8a46fe84 206 mddev ? mddev : "further assembly");
82b27616 207
811e6cbe
NB
208 /* first walk the list of devices to find a consistent set
209 * that match the criterea, if that is possible.
210 * We flag the one we like with 'used'.
211 */
212 for (tmpdev = devlist;
213 tmpdev;
214 tmpdev = tmpdev->next) {
215 char *devname = tmpdev->devname;
64c4757e
NB
216 int dfd;
217 struct stat stb;
82d9eba6 218 struct supertype *tst = st;
52826846 219
da6b5ca9
NB
220 if (tmpdev->used > 1) continue;
221
52826846 222 if (ident->devices &&
56eedc1a 223 !match_oneof(ident->devices, devname)) {
6a41304b 224 if ((inargv && verbose>=0) || verbose > 0)
56eedc1a 225 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
52826846 226 continue;
56eedc1a 227 }
4b1ac34b
NB
228
229 if (super) {
230 free(super);
231 super = NULL;
232 }
52826846 233
8b0dabea 234 dfd = dev_open(devname, O_RDONLY|O_EXCL);
64c4757e 235 if (dfd < 0) {
6a41304b 236 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 237 fprintf(stderr, Name ": cannot open device %s: %s\n",
64c4757e 238 devname, strerror(errno));
da6b5ca9 239 tmpdev->used = 2;
52826846
NB
240 } else if (fstat(dfd, &stb)< 0) {
241 /* Impossible! */
242 fprintf(stderr, Name ": fstat failed for %s: %s\n",
243 devname, strerror(errno));
da6b5ca9 244 tmpdev->used = 2;
cd29a5c8
NB
245 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
246 fprintf(stderr, Name ": %s is not a block device.\n",
52826846 247 devname);
da6b5ca9 248 tmpdev->used = 2;
82d9eba6 249 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
6a41304b 250 if ((inargv && verbose >= 0) || verbose > 0)
da6b5ca9
NB
251 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
252 devname);
253 tmpdev->used = 2;
82d9eba6 254 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
6a41304b 255 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 256 fprintf( stderr, Name ": no RAID superblock on %s\n",
64c4757e 257 devname);
52826846 258 } else {
31317663 259 tst->ss->getinfo_super(&info, super);
64c4757e 260 }
c06487ce 261 if (dfd >= 0) close(dfd);
52826846 262
838acbc2 263 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
f277ce36 264 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
6a41304b 265 if ((inargv && verbose >= 0) || verbose > 0)
52826846
NB
266 fprintf(stderr, Name ": %s has wrong uuid.\n",
267 devname);
268 continue;
82b27616 269 }
c4f12c13 270 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
624920bb 271 (!super || name_matches(info.name, ident->name, homehost)==0)) {
6a41304b 272 if ((inargv && verbose >= 0) || verbose > 0)
947fd4dd
NB
273 fprintf(stderr, Name ": %s has wrong name.\n",
274 devname);
275 continue;
276 }
98c6faba 277 if (ident->super_minor != UnSet &&
4b1ac34b 278 (!super || ident->super_minor != info.array.md_minor)) {
6a41304b 279 if ((inargv && verbose >= 0) || verbose > 0)
52826846 280 fprintf(stderr, Name ": %s has wrong super-minor.\n",
64c4757e
NB
281 devname);
282 continue;
283 }
98c6faba 284 if (ident->level != UnSet &&
4b1ac34b 285 (!super|| ident->level != info.array.level)) {
6a41304b 286 if ((inargv && verbose >= 0) || verbose > 0)
cd29a5c8
NB
287 fprintf(stderr, Name ": %s has wrong raid level.\n",
288 devname);
289 continue;
290 }
98c6faba 291 if (ident->raid_disks != UnSet &&
4b1ac34b 292 (!super || ident->raid_disks!= info.array.raid_disks)) {
6a41304b 293 if ((inargv && verbose >= 0) || verbose > 0)
cd29a5c8
NB
294 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
295 devname);
296 continue;
297 }
da6b5ca9
NB
298 if (mdfd < 0) {
299 if (tst == NULL || super == NULL)
300 continue;
589395d6
NB
301 if (update == NULL &&
302 tst->ss->match_home(super, homehost)==0) {
da6b5ca9
NB
303 if ((inargv && verbose >= 0) || verbose > 0)
304 fprintf(stderr, Name ": %s is not built for host %s.\n",
305 devname, homehost);
306 /* Auto-assemble, and this is not a usable host */
589395d6
NB
307 /* if update != NULL, we are updating the host
308 * name... */
da6b5ca9
NB
309 continue;
310 }
311 }
83b6208e 312 /* If we are this far, then we are nearly commited to this device.
52826846 313 * If the super_block doesn't exist, or doesn't match others,
83b6208e
NB
314 * then we probably cannot continue
315 * However if one of the arrays is for the homehost, and
316 * the other isn't that can disambiguate.
52826846 317 */
52826846 318
4b1ac34b 319 if (!super) {
52826846
NB
320 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
321 devname);
4b1ac34b 322 free(first_super);
52826846
NB
323 return 1;
324 }
838acbc2 325
83b6208e
NB
326 if (st == NULL)
327 st = tst;
328 if (st->ss != tst->ss ||
329 st->minor_version != tst->minor_version ||
330 st->ss->compare_super(&first_super, super) != 0) {
331 /* Some mismatch. If exactly one array matches this host,
da6b5ca9
NB
332 * we can resolve on that one.
333 * Or, if we are auto assembling, we just ignore the second
334 * for now.
83b6208e 335 */
da6b5ca9
NB
336 if (mdfd < 0)
337 continue;
83b6208e
NB
338 if (homehost) {
339 int first = st->ss->match_home(first_super, homehost);
340 int last = tst->ss->match_home(super, homehost);
341 if (first+last == 1) {
342 /* We can do something */
343 if (first) {/* just ignore this one */
344 if ((inargv && verbose >= 0) || verbose > 0)
345 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
346 devname);
347 continue;
348 } else { /* reject all those sofar */
349 mddev_dev_t td;
350 if ((inargv && verbose >= 0) || verbose > 0)
351 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
352 devname);
353 for (td=devlist; td != tmpdev; td=td->next)
354 if (td->used == 1)
355 td->used = 0;
356 tmpdev->used = 1;
357 continue;
358 }
359 }
360 }
52826846
NB
361 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
362 devname);
4b1ac34b
NB
363 free(super);
364 free(first_super);
52826846 365 return 1;
64c4757e
NB
366 }
367
811e6cbe
NB
368 tmpdev->used = 1;
369 }
370
8a46fe84
NB
371 if (mdfd < 0) {
372 /* So... it is up to me to open the device.
373 * We create a name '/dev/md/XXX' based on the info in the
374 * superblock, and call open_mddev on that
375 */
da6b5ca9
NB
376 mdu_array_info_t inf;
377 char *c;
378 if (!first_super) {
d55e3aef 379 return 2;
da6b5ca9
NB
380 }
381 st->ss->getinfo_super(&info, first_super);
382 c = strchr(info.name, ':');
383 if (c) c++; else c= info.name;
41a3b72a
NB
384 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
385 /* /dev/md/d0 style for partitionable */
386 asprintf(&mddev, "/dev/md/d%s", c);
387 else
388 asprintf(&mddev, "/dev/md/%s", c);
da6b5ca9 389 mdfd = open_mddev(mddev, ident->autof);
60e1bc1a
NB
390 if (mdfd < 0) {
391 free(first_super);
392 free(devices);
393 first_super = NULL;
394 goto try_again;
395 }
da6b5ca9
NB
396 vers = md_get_version(mdfd);
397 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
60e1bc1a
NB
398 for (tmpdev = devlist ;
399 tmpdev && tmpdev->used != 1;
400 tmpdev = tmpdev->next)
401 ;
da6b5ca9 402 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
60e1bc1a
NB
403 if (tmpdev)
404 fprintf(stderr, Name ": %s needed for %s...\n",
405 mddev, tmpdev->devname);
da6b5ca9 406 close(mdfd);
60e1bc1a 407 mdfd = -1;
da6b5ca9 408 free(first_super);
60e1bc1a
NB
409 free(devices);
410 first_super = NULL;
411 goto try_again;
da6b5ca9 412 }
d55e3aef 413 must_close = 1;
8a46fe84
NB
414 }
415
811e6cbe 416 /* Ok, no bad inconsistancy, we can try updating etc */
bf4fb153 417 bitmap_done = 0;
da6b5ca9 418 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
811e6cbe
NB
419 char *devname = tmpdev->devname;
420 struct stat stb;
5787fa49 421 /* looks like a good enough match to update the super block if needed */
d1f1011b 422#ifndef MDASSEMBLE
5787fa49 423 if (update) {
811e6cbe 424 int dfd;
4b1ac34b
NB
425 /* prepare useful information in info structures */
426 struct stat stb2;
427 fstat(mdfd, &stb2);
7d99579f
NB
428
429 if (strcmp(update, "uuid")==0 &&
430 !ident->uuid_set) {
431 int rfd;
432 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
433 read(rfd, ident->uuid, 16) != 16) {
434 *(__u32*)(ident->uuid) = random();
435 *(__u32*)(ident->uuid+1) = random();
436 *(__u32*)(ident->uuid+2) = random();
437 *(__u32*)(ident->uuid+3) = random();
438 }
439 if (rfd >= 0) close(rfd);
7d99579f 440 }
811e6cbe
NB
441 dfd = dev_open(devname, O_RDWR|O_EXCL);
442
0430ed48
NB
443 remove_partitions(dfd);
444
811e6cbe
NB
445 if (super) {
446 free(super);
447 super = NULL;
448 }
449
450 st->ss->load_super(st, dfd, &super, NULL);
451 st->ss->getinfo_super(&info, super);
452
7d99579f 453 memcpy(info.uuid, ident->uuid, 16);
e5eac01f 454 strcpy(info.name, ident->name);
811e6cbe
NB
455 info.array.md_minor = minor(stb2.st_rdev);
456
e5eac01f
NB
457 st->ss->update_super(&info, super, update, devname, verbose,
458 ident->uuid_set, homehost);
459 if (strcmp(update, "uuid")==0 &&
460 !ident->uuid_set) {
461 ident->uuid_set = 1;
462 memcpy(ident->uuid, info.uuid, 16);
463 }
b8a8ccf9 464 if (dfd < 0)
5787fa49
NB
465 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
466 devname);
96395475 467 else if (st->ss->store_super(st, dfd, super))
5787fa49
NB
468 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
469 devname);
470 if (dfd >= 0)
471 close(dfd);
8131b493
NB
472
473 if (strcmp(update, "uuid")==0 &&
bf4fb153
NB
474 ident->bitmap_fd >= 0 && !bitmap_done) {
475 if (bitmap_update_uuid(ident->bitmap_fd, info.uuid, st->ss->swapuuid) != 0)
476 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
477 else
478 bitmap_done = 1;
479 }
d1f1011b
NB
480 } else
481#endif
482 {
da6b5ca9
NB
483 int dfd;
484 dfd = dev_open(devname, O_RDWR|O_EXCL);
485
0430ed48
NB
486 remove_partitions(dfd);
487
da6b5ca9
NB
488 if (super) {
489 free(super);
490 super = NULL;
491 }
492
493 st->ss->load_super(st, dfd, &super, NULL);
494 st->ss->getinfo_super(&info, super);
495 close(dfd);
5787fa49
NB
496 }
497
811e6cbe
NB
498 stat(devname, &stb);
499
dab6685f 500 if (verbose > 0)
cd29a5c8 501 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
4b1ac34b 502 devname, mddev, info.disk.raid_disk);
64c4757e 503 devices[devcnt].devname = devname;
0df46c2a
NB
504 devices[devcnt].major = major(stb.st_rdev);
505 devices[devcnt].minor = minor(stb.st_rdev);
4b1ac34b
NB
506 devices[devcnt].oldmajor = info.disk.major;
507 devices[devcnt].oldminor = info.disk.minor;
508 devices[devcnt].events = info.events;
509 devices[devcnt].raid_disk = info.disk.raid_disk;
6a41304b 510 devices[devcnt].disk_nr = info.disk.number;
64c4757e 511 devices[devcnt].uptodate = 0;
4b1ac34b 512 devices[devcnt].state = info.disk.state;
64c4757e
NB
513 if (most_recent < devcnt) {
514 if (devices[devcnt].events
515 > devices[most_recent].events)
516 most_recent = devcnt;
517 }
b8a8ccf9 518 if (info.array.level == -4)
5787fa49
NB
519 /* with multipath, the raid_disk from the superblock is meaningless */
520 i = devcnt;
521 else
522 i = devices[devcnt].raid_disk;
308e1801
NB
523 if (i+1 == 0) {
524 if (nextspare < info.array.raid_disks)
525 nextspare = info.array.raid_disks;
526 i = nextspare++;
08110d41
NB
527 } else {
528 if (i >= info.array.raid_disks &&
529 i >= nextspare)
530 nextspare = i+1;
308e1801 531 }
98c6faba 532 if (i < 10000) {
56eedc1a 533 if (i >= bestcnt) {
98c6faba 534 unsigned int newbestcnt = i+10;
56eedc1a 535 int *newbest = malloc(sizeof(int)*newbestcnt);
98c6faba 536 unsigned int c;
56eedc1a
NB
537 for (c=0; c < newbestcnt; c++)
538 if (c < bestcnt)
539 newbest[c] = best[c];
540 else
541 newbest[c] = -1;
542 if (best)free(best);
543 best = newbest;
544 bestcnt = newbestcnt;
545 }
6a255b69
NB
546 if (best[i] >=0 &&
547 devices[best[i]].events == devices[devcnt].events &&
548 devices[best[i]].minor != devices[devcnt].minor &&
549 st->ss->major == 0 &&
550 info.array.level != -4) {
551 /* two different devices with identical superblock.
552 * Could be a mis-detection caused by overlapping
553 * partitions. fail-safe.
554 */
555 fprintf(stderr, Name ": WARNING %s and %s appear"
556 " to have very similar superblocks.\n"
557 " If they are really different, "
558 "please --zero the superblock on one\n"
d2df86e0
NB
559 " If they are the same or overlap,"
560 " please remove one from %s.\n",
561 devices[best[i]].devname, devname,
562 inargv ? "the list" :
563 "the\n DEVICE list in mdadm.conf"
564 );
6a255b69
NB
565 if (must_close) close(mdfd);
566 return 1;
567 }
52826846
NB
568 if (best[i] == -1
569 || devices[best[i]].events < devices[devcnt].events)
570 best[i] = devcnt;
56eedc1a 571 }
64c4757e
NB
572 devcnt++;
573 }
574
4b1ac34b
NB
575 if (super)
576 free(super);
577 super = NULL;
578
586ed405
NB
579 if (update && strcmp(update, "byteorder")==0)
580 st->minor_version = 90;
581
64c4757e 582 if (devcnt == 0) {
682c7051 583 fprintf(stderr, Name ": no devices found for %s\n",
64c4757e 584 mddev);
4b1ac34b 585 free(first_super);
d55e3aef 586 if (must_close) close(mdfd);
64c4757e
NB
587 return 1;
588 }
4b1ac34b 589
31317663 590 st->ss->getinfo_super(&info, first_super);
583315d9 591 clean = info.array.state & 1;
4b1ac34b 592
64c4757e
NB
593 /* now we have some devices that might be suitable.
594 * I wonder how many
595 */
265e0f17
NB
596 avail = malloc(info.array.raid_disks);
597 memset(avail, 0, info.array.raid_disks);
64c4757e 598 okcnt = 0;
52826846 599 sparecnt=0;
56eedc1a 600 for (i=0; i< bestcnt ;i++) {
64c4757e 601 int j = best[i];
ee04451c
NB
602 int event_margin = 1; /* always allow a difference of '1'
603 * like the kernel does
604 */
64c4757e 605 if (j < 0) continue;
d013a55e
NB
606 /* note: we ignore error flags in multipath arrays
607 * as they don't make sense
608 */
4b1ac34b 609 if (info.array.level != -4)
aa88f531
NB
610 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
611 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
612 sparecnt++;
d013a55e 613 continue;
aa88f531 614 }
cd29a5c8 615 if (devices[j].events+event_margin >=
64c4757e
NB
616 devices[most_recent].events) {
617 devices[j].uptodate = 1;
265e0f17 618 if (i < info.array.raid_disks) {
52826846 619 okcnt++;
265e0f17
NB
620 avail[i]=1;
621 } else
52826846 622 sparecnt++;
64c4757e
NB
623 }
624 }
265e0f17 625 while (force && !enough(info.array.level, info.array.raid_disks,
583315d9 626 info.array.layout, 1,
265e0f17 627 avail, okcnt)) {
64c4757e
NB
628 /* Choose the newest best drive which is
629 * not up-to-date, update the superblock
630 * and add it.
631 */
52826846 632 int fd;
c5a6f9a6 633 long long current_events;
cd29a5c8 634 chosen_drive = -1;
4b1ac34b 635 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
52826846
NB
636 int j = best[i];
637 if (j>=0 &&
638 !devices[j].uptodate &&
639 devices[j].events > 0 &&
640 (chosen_drive < 0 ||
641 devices[j].events > devices[chosen_drive].events))
642 chosen_drive = j;
643 }
644 if (chosen_drive < 0)
645 break;
c5a6f9a6
NB
646 current_events = devices[chosen_drive].events;
647 add_another:
dab6685f
NB
648 if (verbose >= 0)
649 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
650 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
651 (int)(devices[chosen_drive].events),
652 (int)(devices[most_recent].events));
8b0dabea 653 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846
NB
654 if (fd < 0) {
655 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
656 devices[chosen_drive].devname);
657 devices[chosen_drive].events = 0;
658 continue;
659 }
82d9eba6 660 if (st->ss->load_super(st,fd, &super, NULL)) {
52826846
NB
661 close(fd);
662 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
663 devices[chosen_drive].devname);
664 devices[chosen_drive].events = 0;
665 continue;
666 }
4b1ac34b 667 info.events = devices[most_recent].events;
67a8c82d
NB
668 st->ss->update_super(&info, super, "force-one",
669 devices[chosen_drive].devname, verbose,
670 0, NULL);
4b1ac34b 671
96395475 672 if (st->ss->store_super(st, fd, super)) {
52826846
NB
673 close(fd);
674 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
675 devices[chosen_drive].devname);
676 devices[chosen_drive].events = 0;
4b1ac34b 677 free(super);
52826846
NB
678 continue;
679 }
680 close(fd);
681 devices[chosen_drive].events = devices[most_recent].events;
682 devices[chosen_drive].uptodate = 1;
265e0f17 683 avail[chosen_drive] = 1;
52826846 684 okcnt++;
4b1ac34b 685 free(super);
c5a6f9a6
NB
686
687 /* If there are any other drives of the same vintage,
688 * add them in as well. We can't lose and we might gain
689 */
690 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
691 int j = best[i];
692 if (j >= 0 &&
693 !devices[j].uptodate &&
694 devices[j].events > 0 &&
695 devices[j].events == current_events) {
696 chosen_drive = j;
697 goto add_another;
698 }
699 }
64c4757e 700 }
52826846
NB
701
702 /* Now we want to look at the superblock which the kernel will base things on
703 * and compare the devices that we think are working with the devices that the
704 * superblock thinks are working.
705 * If there are differences and --force is given, then update this chosen
706 * superblock.
707 */
cd29a5c8 708 chosen_drive = -1;
4b1ac34b 709 super = NULL;
56eedc1a 710 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
52826846
NB
711 int j = best[i];
712 int fd;
4b1ac34b 713
52826846
NB
714 if (j<0)
715 continue;
716 if (!devices[j].uptodate)
717 continue;
718 chosen_drive = j;
8b0dabea 719 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
52826846
NB
720 fprintf(stderr, Name ": Cannot open %s: %s\n",
721 devices[j].devname, strerror(errno));
d55e3aef 722 if (must_close) close(mdfd);
52826846
NB
723 return 1;
724 }
82d9eba6 725 if (st->ss->load_super(st,fd, &super, NULL)) {
52826846
NB
726 close(fd);
727 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
728 devices[j].devname);
d55e3aef 729 if (must_close) close(mdfd);
52826846
NB
730 return 1;
731 }
732 close(fd);
733 }
4b1ac34b
NB
734 if (super == NULL) {
735 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
d55e3aef 736 if (must_close) close(mdfd);
4b1ac34b
NB
737 return 1;
738 }
31317663 739 st->ss->getinfo_super(&info, super);
56eedc1a 740 for (i=0; i<bestcnt; i++) {
52826846 741 int j = best[i];
98c6faba 742 unsigned int desired_state;
11a3e71d 743
4b1ac34b 744 if (i < info.array.raid_disks)
11a3e71d
NB
745 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
746 else
747 desired_state = 0;
748
52826846
NB
749 if (j<0)
750 continue;
751 if (!devices[j].uptodate)
752 continue;
6a41304b 753 info.disk.number = devices[j].disk_nr;
fbf8a0b7 754 info.disk.raid_disk = i;
4b1ac34b
NB
755 info.disk.state = desired_state;
756
52826846 757 if (devices[j].uptodate &&
e5eac01f 758 st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) {
52826846 759 if (force) {
dab6685f
NB
760 if (verbose >= 0)
761 fprintf(stderr, Name ": "
762 "clearing FAULTY flag for device %d in %s for %s\n",
763 j, mddev, devices[j].devname);
4b1ac34b 764 change = 1;
52826846 765 } else {
dab6685f
NB
766 if (verbose >= -1)
767 fprintf(stderr, Name ": "
768 "device %d in %s has wrong state in superblock, but %s seems ok\n",
769 i, mddev, devices[j].devname);
52826846
NB
770 }
771 }
4b1ac34b 772#if 0
52826846
NB
773 if (!devices[j].uptodate &&
774 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
775 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
776 i, mddev);
777 }
4b1ac34b 778#endif
52826846 779 }
c5a6f9a6
NB
780 if (force && !clean &&
781 !enough(info.array.level, info.array.raid_disks,
782 info.array.layout, clean,
783 avail, okcnt)) {
67a8c82d
NB
784 change += st->ss->update_super(&info, super, "force-array",
785 devices[chosen_drive].devname, verbose,
786 0, NULL);
583315d9 787 clean = 1;
d013a55e 788 }
52826846 789
4b1ac34b 790 if (change) {
52826846 791 int fd;
8b0dabea 792 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846 793 if (fd < 0) {
353632d9 794 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
52826846 795 devices[chosen_drive].devname);
d55e3aef 796 if (must_close) close(mdfd);
52826846
NB
797 return 1;
798 }
96395475 799 if (st->ss->store_super(st, fd, super)) {
52826846
NB
800 close(fd);
801 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
802 devices[chosen_drive].devname);
d55e3aef 803 if (must_close) close(mdfd);
52826846
NB
804 return 1;
805 }
806 close(fd);
52826846
NB
807 }
808
353632d9
NB
809 /* If we are in the middle of a reshape we may need to restore saved data
810 * that was moved aside due to the reshape overwriting live data
811 * The code of doing this lives in Grow.c
812 */
39c5a909 813#ifndef MDASSEMBLE
353632d9
NB
814 if (info.reshape_active) {
815 int err = 0;
816 int *fdlist = malloc(sizeof(int)* bestcnt);
817 for (i=0; i<bestcnt; i++) {
818 int j = best[i];
819 if (j >= 0) {
820 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
821 if (fdlist[i] < 0) {
822 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
823 devices[j].devname);
824 err = 1;
825 break;
826 }
827 } else
828 fdlist[i] = -1;
829 }
830 if (!err)
06b0d786 831 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
353632d9
NB
832 while (i>0) {
833 i--;
834 if (fdlist[i]>=0) close(fdlist[i]);
835 }
836 if (err) {
837 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
d55e3aef 838 if (must_close) close(mdfd);
353632d9
NB
839 return err;
840 }
841 }
39c5a909 842#endif
aa88f531
NB
843 /* count number of in-sync devices according to the superblock.
844 * We must have this number to start the array without -s or -R
845 */
4b1ac34b 846 req_cnt = info.array.working_disks;
aa88f531 847
64c4757e
NB
848 /* Almost ready to actually *do* something */
849 if (!old_linux) {
fbf8a0b7
NB
850 int rv;
851 if ((vers % 100) >= 1) { /* can use different versions */
852 mdu_array_info_t inf;
853 memset(&inf, 0, sizeof(inf));
854 inf.major_version = st->ss->major;
855 inf.minor_version = st->minor_version;
856 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
b8a8ccf9 857 } else
fbf8a0b7
NB
858 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
859
860 if (rv) {
682c7051 861 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
64c4757e 862 mddev, strerror(errno));
d55e3aef 863 if (must_close) close(mdfd);
64c4757e
NB
864 return 1;
865 }
11484a63 866 if (ident->bitmap_fd >= 0) {
c82f047c
NB
867 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
868 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
d55e3aef 869 if (must_close) close(mdfd);
c82f047c
NB
870 return 1;
871 }
7ef02d01
NB
872 } else if (ident->bitmap_file) {
873 /* From config file */
874 int bmfd = open(ident->bitmap_file, O_RDWR);
875 if (bmfd < 0) {
876 fprintf(stderr, Name ": Could not open bitmap file %s\n",
877 ident->bitmap_file);
d55e3aef 878 if (must_close) close(mdfd);
7ef02d01
NB
879 return 1;
880 }
881 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
882 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
883 close(bmfd);
d55e3aef 884 if (must_close) close(mdfd);
7ef02d01
NB
885 return 1;
886 }
887 close(bmfd);
c82f047c 888 }
7ef02d01 889
52826846 890 /* First, add the raid disks, but add the chosen one last */
56eedc1a 891 for (i=0; i<= bestcnt; i++) {
52826846 892 int j;
56eedc1a 893 if (i < bestcnt) {
52826846
NB
894 j = best[i];
895 if (j == chosen_drive)
896 continue;
897 } else
898 j = chosen_drive;
899
56eedc1a 900 if (j >= 0 /* && devices[j].uptodate */) {
64c4757e
NB
901 mdu_disk_info_t disk;
902 memset(&disk, 0, sizeof(disk));
903 disk.major = devices[j].major;
904 disk.minor = devices[j].minor;
905 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
682c7051 906 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
64c4757e
NB
907 devices[j].devname,
908 mddev,
909 strerror(errno));
4b1ac34b 910 if (i < info.array.raid_disks || i == bestcnt)
52826846
NB
911 okcnt--;
912 else
913 sparecnt--;
dab6685f 914 } else if (verbose > 0)
52826846
NB
915 fprintf(stderr, Name ": added %s to %s as %d\n",
916 devices[j].devname, mddev, devices[j].raid_disk);
dab6685f 917 } else if (verbose > 0 && i < info.array.raid_disks)
682c7051 918 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
64c4757e
NB
919 i, mddev);
920 }
52826846 921
64c4757e 922 if (runstop == 1 ||
b8a8ccf9 923 (runstop <= 0 &&
583315d9
NB
924 ( enough(info.array.level, info.array.raid_disks,
925 info.array.layout, clean, avail, okcnt) &&
aa88f531
NB
926 (okcnt >= req_cnt || start_partial_ok)
927 ))) {
82b27616 928 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
dab6685f
NB
929 if (verbose >= 0) {
930 fprintf(stderr, Name ": %s has been started with %d drive%s",
931 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9 932 if (okcnt < info.array.raid_disks)
dab6685f
NB
933 fprintf(stderr, " (out of %d)", info.array.raid_disks);
934 if (sparecnt)
935 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
936 fprintf(stderr, ".\n");
937 }
589395d6
NB
938 if (must_close) {
939 int usecs = 1;
940 close(mdfd);
941 /* There is a nasty race with 'mdadm --monitor'.
942 * If it opens this device before we close it,
943 * it gets an incomplete open on which IO
944 * doesn't work and the capacity if wrong.
945 * If we reopen (to check for layered devices)
946 * before --monitor closes, we loose.
947 *
948 * So: wait upto 1 second for there to be
949 * a non-zero capacity.
950 */
951 while (usecs < 1000) {
952 mdfd = open(mddev, O_RDONLY);
953 if (mdfd >= 0) {
beae1dfe
NB
954 unsigned long long size;
955 if (get_dev_size(mdfd, NULL, &size) &&
589395d6
NB
956 size > 0)
957 break;
958 close(mdfd);
959 }
960 usleep(usecs);
961 usecs <<= 1;
962 }
963 }
64c4757e 964 return 0;
82b27616 965 }
682c7051 966 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
64c4757e 967 mddev, strerror(errno));
583315d9
NB
968
969 if (!enough(info.array.level, info.array.raid_disks,
970 info.array.layout, 1, avail, okcnt))
971 fprintf(stderr, Name ": Not enough devices to "
972 "start the array.\n");
973 else if (!enough(info.array.level,
974 info.array.raid_disks,
975 info.array.layout, clean,
976 avail, okcnt))
977 fprintf(stderr, Name ": Not enough devices to "
978 "start the array while not clean "
979 "- consider --force.\n");
980
d55e3aef 981 if (must_close) close(mdfd);
64c4757e
NB
982 return 1;
983 }
82b27616 984 if (runstop == -1) {
b8a8ccf9 985 fprintf(stderr, Name ": %s assembled from %d drive%s",
52826846 986 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9
NB
987 if (okcnt != info.array.raid_disks)
988 fprintf(stderr, " (out of %d)", info.array.raid_disks);
989 fprintf(stderr, ", but not started.\n");
d55e3aef 990 if (must_close) close(mdfd);
64c4757e 991 return 0;
82b27616 992 }
4855f95c 993 if (verbose >= -1) {
dab6685f
NB
994 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
995 if (sparecnt)
996 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
583315d9
NB
997 if (!enough(info.array.level, info.array.raid_disks,
998 info.array.layout, 1, avail, okcnt))
dab6685f 999 fprintf(stderr, " - not enough to start the array.\n");
583315d9
NB
1000 else if (!enough(info.array.level,
1001 info.array.raid_disks,
1002 info.array.layout, clean,
1003 avail, okcnt))
1004 fprintf(stderr, " - not enough to start the "
1005 "array while not clean - consider "
1006 "--force.\n");
dab6685f
NB
1007 else {
1008 if (req_cnt == info.array.raid_disks)
1009 fprintf(stderr, " - need all %d to start it", req_cnt);
1010 else
1011 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1012 fprintf(stderr, " (use --run to insist).\n");
1013 }
aa88f531 1014 }
d55e3aef 1015 if (must_close) close(mdfd);
64c4757e 1016 return 1;
82b27616 1017 } else {
52826846
NB
1018 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1019 * been updated to point to the current locations of devices.
1020 * so we can just start the array
82b27616 1021 */
cd29a5c8 1022 unsigned long dev;
0df46c2a 1023 dev = makedev(devices[chosen_drive].major,
82b27616
NB
1024 devices[chosen_drive].minor);
1025 if (ioctl(mdfd, START_ARRAY, dev)) {
1026 fprintf(stderr, Name ": Cannot start array: %s\n",
1027 strerror(errno));
1028 }
1029
64c4757e 1030 }
d55e3aef 1031 if (must_close) close(mdfd);
e0d19036 1032 return 0;
64c4757e 1033}