]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Assemble.c
Remove partitions from components of an md array
[thirdparty/mdadm.git] / Assemble.c
CommitLineData
64c4757e 1/*
9a9dab36 2 * mdadm - manage Linux "md" devices aka RAID arrays.
64c4757e 3 *
4f589ad0 4 * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de>
64c4757e
NB
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@cse.unsw.edu.au>
23 * Paper: Neil Brown
24 * School of Computer Science and Engineering
25 * The University of New South Wales
26 * Sydney, 2052
27 * Australia
28 */
29
9a9dab36 30#include "mdadm.h"
41a3b72a 31#include <ctype.h>
64c4757e 32
624920bb
NB
33static int name_matches(char *found, char *required, char *homehost)
34{
35 /* See if the name found matches the required name, possibly
36 * prefixed with 'homehost'
37 */
38 char fnd[33];
39
40 strncpy(fnd, found, 32);
41 fnd[32] = 0;
42 if (strcmp(found, required)==0)
43 return 1;
44 if (homehost) {
45 int l = strlen(homehost);
46 if (l < 32 && fnd[l] == ':' &&
47 strcmp(fnd+l+1, required)==0)
48 return 1;
49 }
50 return 0;
51}
52
82d9eba6 53int Assemble(struct supertype *st, char *mddev, int mdfd,
8aec876d 54 mddev_ident_t ident,
06b0d786 55 mddev_dev_t devlist, char *backup_file,
64c4757e 56 int readonly, int runstop,
e5eac01f 57 char *update, char *homehost,
64c4757e
NB
58 int verbose, int force)
59{
60 /*
52826846
NB
61 * The task of Assemble is to find a collection of
62 * devices that should (according to their superblocks)
63 * form an array, and to give this collection to the MD driver.
64 * In Linux-2.4 and later, this involves submitting a
64c4757e
NB
65 * SET_ARRAY_INFO ioctl with no arg - to prepare
66 * the array - and then submit a number of
67 * ADD_NEW_DISK ioctls to add disks into
68 * the array. Finally RUN_ARRAY might
69 * be submitted to start the array.
70 *
71 * Much of the work of Assemble is in finding and/or
72 * checking the disks to make sure they look right.
73 *
4b1ac34b 74 * If mddev is not set, then scan must be set and we
64c4757e
NB
75 * read through the config file for dev+uuid mapping
76 * We recurse, setting mddev, for each device that
77 * - isn't running
4b1ac34b 78 * - has a valid uuid (or any uuid if !uuidset)
64c4757e
NB
79 *
80 * If mddev is set, we try to determine state of md.
81 * check version - must be at least 0.90.0
82 * check kernel version. must be at least 2.4.
83 * If not, we can possibly fall back on START_ARRAY
84 * Try to GET_ARRAY_INFO.
85 * If possible, give up
86 * If not, try to STOP_ARRAY just to make sure
87 *
88 * If !uuidset and scan, look in conf-file for uuid
89 * If not found, give up
b8a8ccf9 90 * If !devlist and scan and uuidset, get list of devs from conf-file
64c4757e
NB
91 *
92 * For each device:
93 * Check superblock - discard if bad
94 * Check uuid (set if we don't have one) - discard if no match
5787fa49 95 * Check superblock similarity if we have a superblock - discard if different
4b1ac34b 96 * Record events, devicenum
64c4757e 97 * This should give us a list of devices for the array
4b1ac34b 98 * We should collect the most recent event number
64c4757e
NB
99 *
100 * Count disks with recent enough event count
101 * While force && !enough disks
102 * Choose newest rejected disks, update event count
103 * mark clean and rewrite superblock
104 * If recent kernel:
105 * SET_ARRAY_INFO
106 * foreach device with recent events : ADD_NEW_DISK
107 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
108 * If old kernel:
109 * Check the device numbers in superblock are right
110 * update superblock if any changes
111 * START_ARRAY
112 *
113 */
d55e3aef 114 int must_close = 0;
64c4757e 115 int old_linux = 0;
41a3b72a 116 int vers = 0; /* Keep gcc quite - it really is initialised */
4b1ac34b 117 void *first_super = NULL, *super = NULL;
64c4757e
NB
118 struct {
119 char *devname;
98c6faba
NB
120 unsigned int major, minor;
121 unsigned int oldmajor, oldminor;
64c4757e 122 long long events;
64c4757e 123 int uptodate;
d013a55e 124 int state;
52826846 125 int raid_disk;
6a41304b 126 int disk_nr;
5787fa49 127 } *devices;
56eedc1a 128 int *best = NULL; /* indexed by raid_disk */
98c6faba
NB
129 unsigned int bestcnt = 0;
130 int devcnt = 0;
131 unsigned int okcnt, sparecnt;
132 unsigned int req_cnt;
133 unsigned int i;
64c4757e 134 int most_recent = 0;
cd29a5c8 135 int chosen_drive;
52826846 136 int change = 0;
cd29a5c8 137 int inargv = 0;
589395d6 138 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
98c6faba 139 unsigned int num_devs;
5787fa49 140 mddev_dev_t tmpdev;
4b1ac34b 141 struct mdinfo info;
265e0f17 142 char *avail;
308e1801 143 int nextspare = 0;
8a46fe84 144
682c7051 145 if (get_linux_version() < 2004000)
64c4757e
NB
146 old_linux = 1;
147
8a46fe84
NB
148 if (mdfd >= 0) {
149 vers = md_get_version(mdfd);
150 if (vers <= 0) {
151 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
152 return 1;
153 }
154 if (vers < 9000) {
155 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
156 " Upgrade your kernel or try --build\n");
157 return 1;
158 }
64c4757e 159
8a46fe84
NB
160 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
161 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
162 mddev);
163 return 1;
164 }
165 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
166 }
64c4757e 167 /*
52826846
NB
168 * If any subdevs are listed, then any that don't
169 * match ident are discarded. Remainder must all match and
170 * become the array.
171 * If no subdevs, then we scan all devices in the config file, but
172 * there must be something in the identity
64c4757e 173 */
64c4757e 174
cd29a5c8 175 if (!devlist &&
52826846
NB
176 ident->uuid_set == 0 &&
177 ident->super_minor < 0 &&
178 ident->devices == NULL) {
179 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
8a46fe84 180 mddev ? mddev : "further assembly");
52826846 181 return 1;
64c4757e 182 }
cd29a5c8 183 if (devlist == NULL)
8aec876d 184 devlist = conf_get_devs();
da6b5ca9
NB
185 else if (mdfd >= 0)
186 inargv = 1;
64c4757e 187
5787fa49
NB
188 tmpdev = devlist; num_devs = 0;
189 while (tmpdev) {
da6b5ca9
NB
190 if (tmpdev->used)
191 tmpdev->used = 2;
192 else
193 num_devs++;
5787fa49
NB
194 tmpdev = tmpdev->next;
195 }
5787fa49
NB
196 devices = malloc(num_devs * sizeof(*devices));
197
82d9eba6 198 if (!st && ident->st) st = ident->st;
64c4757e 199
dab6685f 200 if (verbose>0)
82b27616 201 fprintf(stderr, Name ": looking for devices for %s\n",
8a46fe84 202 mddev ? mddev : "further assembly");
82b27616 203
811e6cbe
NB
204 /* first walk the list of devices to find a consistent set
205 * that match the criterea, if that is possible.
206 * We flag the one we like with 'used'.
207 */
208 for (tmpdev = devlist;
209 tmpdev;
210 tmpdev = tmpdev->next) {
211 char *devname = tmpdev->devname;
64c4757e
NB
212 int dfd;
213 struct stat stb;
82d9eba6 214 struct supertype *tst = st;
52826846 215
da6b5ca9
NB
216 if (tmpdev->used > 1) continue;
217
52826846 218 if (ident->devices &&
56eedc1a 219 !match_oneof(ident->devices, devname)) {
6a41304b 220 if ((inargv && verbose>=0) || verbose > 0)
56eedc1a 221 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
52826846 222 continue;
56eedc1a 223 }
4b1ac34b
NB
224
225 if (super) {
226 free(super);
227 super = NULL;
228 }
52826846 229
8b0dabea 230 dfd = dev_open(devname, O_RDONLY|O_EXCL);
64c4757e 231 if (dfd < 0) {
6a41304b 232 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 233 fprintf(stderr, Name ": cannot open device %s: %s\n",
64c4757e 234 devname, strerror(errno));
da6b5ca9 235 tmpdev->used = 2;
52826846
NB
236 } else if (fstat(dfd, &stb)< 0) {
237 /* Impossible! */
238 fprintf(stderr, Name ": fstat failed for %s: %s\n",
239 devname, strerror(errno));
da6b5ca9 240 tmpdev->used = 2;
cd29a5c8
NB
241 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
242 fprintf(stderr, Name ": %s is not a block device.\n",
52826846 243 devname);
da6b5ca9 244 tmpdev->used = 2;
82d9eba6 245 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
6a41304b 246 if ((inargv && verbose >= 0) || verbose > 0)
da6b5ca9
NB
247 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
248 devname);
249 tmpdev->used = 2;
82d9eba6 250 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
6a41304b 251 if ((inargv && verbose >= 0) || verbose > 0)
682c7051 252 fprintf( stderr, Name ": no RAID superblock on %s\n",
64c4757e 253 devname);
52826846 254 } else {
31317663 255 tst->ss->getinfo_super(&info, super);
64c4757e 256 }
c06487ce 257 if (dfd >= 0) close(dfd);
52826846 258
838acbc2 259 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
f277ce36 260 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
6a41304b 261 if ((inargv && verbose >= 0) || verbose > 0)
52826846
NB
262 fprintf(stderr, Name ": %s has wrong uuid.\n",
263 devname);
264 continue;
82b27616 265 }
c4f12c13 266 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
624920bb 267 (!super || name_matches(info.name, ident->name, homehost)==0)) {
6a41304b 268 if ((inargv && verbose >= 0) || verbose > 0)
947fd4dd
NB
269 fprintf(stderr, Name ": %s has wrong name.\n",
270 devname);
271 continue;
272 }
98c6faba 273 if (ident->super_minor != UnSet &&
4b1ac34b 274 (!super || ident->super_minor != info.array.md_minor)) {
6a41304b 275 if ((inargv && verbose >= 0) || verbose > 0)
52826846 276 fprintf(stderr, Name ": %s has wrong super-minor.\n",
64c4757e
NB
277 devname);
278 continue;
279 }
98c6faba 280 if (ident->level != UnSet &&
4b1ac34b 281 (!super|| ident->level != info.array.level)) {
6a41304b 282 if ((inargv && verbose >= 0) || verbose > 0)
cd29a5c8
NB
283 fprintf(stderr, Name ": %s has wrong raid level.\n",
284 devname);
285 continue;
286 }
98c6faba 287 if (ident->raid_disks != UnSet &&
4b1ac34b 288 (!super || ident->raid_disks!= info.array.raid_disks)) {
6a41304b 289 if ((inargv && verbose >= 0) || verbose > 0)
cd29a5c8
NB
290 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
291 devname);
292 continue;
293 }
da6b5ca9
NB
294 if (mdfd < 0) {
295 if (tst == NULL || super == NULL)
296 continue;
589395d6
NB
297 if (update == NULL &&
298 tst->ss->match_home(super, homehost)==0) {
da6b5ca9
NB
299 if ((inargv && verbose >= 0) || verbose > 0)
300 fprintf(stderr, Name ": %s is not built for host %s.\n",
301 devname, homehost);
302 /* Auto-assemble, and this is not a usable host */
589395d6
NB
303 /* if update != NULL, we are updating the host
304 * name... */
da6b5ca9
NB
305 continue;
306 }
307 }
83b6208e 308 /* If we are this far, then we are nearly commited to this device.
52826846 309 * If the super_block doesn't exist, or doesn't match others,
83b6208e
NB
310 * then we probably cannot continue
311 * However if one of the arrays is for the homehost, and
312 * the other isn't that can disambiguate.
52826846 313 */
52826846 314
4b1ac34b 315 if (!super) {
52826846
NB
316 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
317 devname);
4b1ac34b 318 free(first_super);
52826846
NB
319 return 1;
320 }
838acbc2 321
83b6208e
NB
322 if (st == NULL)
323 st = tst;
324 if (st->ss != tst->ss ||
325 st->minor_version != tst->minor_version ||
326 st->ss->compare_super(&first_super, super) != 0) {
327 /* Some mismatch. If exactly one array matches this host,
da6b5ca9
NB
328 * we can resolve on that one.
329 * Or, if we are auto assembling, we just ignore the second
330 * for now.
83b6208e 331 */
da6b5ca9
NB
332 if (mdfd < 0)
333 continue;
83b6208e
NB
334 if (homehost) {
335 int first = st->ss->match_home(first_super, homehost);
336 int last = tst->ss->match_home(super, homehost);
337 if (first+last == 1) {
338 /* We can do something */
339 if (first) {/* just ignore this one */
340 if ((inargv && verbose >= 0) || verbose > 0)
341 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
342 devname);
343 continue;
344 } else { /* reject all those sofar */
345 mddev_dev_t td;
346 if ((inargv && verbose >= 0) || verbose > 0)
347 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
348 devname);
349 for (td=devlist; td != tmpdev; td=td->next)
350 if (td->used == 1)
351 td->used = 0;
352 tmpdev->used = 1;
353 continue;
354 }
355 }
356 }
52826846
NB
357 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
358 devname);
4b1ac34b
NB
359 free(super);
360 free(first_super);
52826846 361 return 1;
64c4757e
NB
362 }
363
811e6cbe
NB
364 tmpdev->used = 1;
365 }
366
8a46fe84
NB
367 if (mdfd < 0) {
368 /* So... it is up to me to open the device.
369 * We create a name '/dev/md/XXX' based on the info in the
370 * superblock, and call open_mddev on that
371 */
da6b5ca9
NB
372 mdu_array_info_t inf;
373 char *c;
374 if (!first_super) {
d55e3aef 375 return 2;
da6b5ca9
NB
376 }
377 st->ss->getinfo_super(&info, first_super);
378 c = strchr(info.name, ':');
379 if (c) c++; else c= info.name;
41a3b72a
NB
380 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
381 /* /dev/md/d0 style for partitionable */
382 asprintf(&mddev, "/dev/md/d%s", c);
383 else
384 asprintf(&mddev, "/dev/md/%s", c);
da6b5ca9 385 mdfd = open_mddev(mddev, ident->autof);
8a46fe84
NB
386 if (mdfd < 0)
387 return mdfd;
da6b5ca9
NB
388 vers = md_get_version(mdfd);
389 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
390 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
391 close(mdfd);
392 free(first_super);
393 return 1;
394 }
d55e3aef 395 must_close = 1;
8a46fe84
NB
396 }
397
811e6cbe 398 /* Ok, no bad inconsistancy, we can try updating etc */
da6b5ca9 399 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
811e6cbe
NB
400 char *devname = tmpdev->devname;
401 struct stat stb;
5787fa49 402 /* looks like a good enough match to update the super block if needed */
d1f1011b 403#ifndef MDASSEMBLE
5787fa49 404 if (update) {
811e6cbe 405 int dfd;
4b1ac34b
NB
406 /* prepare useful information in info structures */
407 struct stat stb2;
408 fstat(mdfd, &stb2);
7d99579f
NB
409
410 if (strcmp(update, "uuid")==0 &&
411 !ident->uuid_set) {
412 int rfd;
413 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
414 read(rfd, ident->uuid, 16) != 16) {
415 *(__u32*)(ident->uuid) = random();
416 *(__u32*)(ident->uuid+1) = random();
417 *(__u32*)(ident->uuid+2) = random();
418 *(__u32*)(ident->uuid+3) = random();
419 }
420 if (rfd >= 0) close(rfd);
7d99579f 421 }
811e6cbe
NB
422 dfd = dev_open(devname, O_RDWR|O_EXCL);
423
0430ed48
NB
424 remove_partitions(dfd);
425
811e6cbe
NB
426 if (super) {
427 free(super);
428 super = NULL;
429 }
430
431 st->ss->load_super(st, dfd, &super, NULL);
432 st->ss->getinfo_super(&info, super);
433
7d99579f 434 memcpy(info.uuid, ident->uuid, 16);
e5eac01f 435 strcpy(info.name, ident->name);
811e6cbe
NB
436 info.array.md_minor = minor(stb2.st_rdev);
437
e5eac01f
NB
438 st->ss->update_super(&info, super, update, devname, verbose,
439 ident->uuid_set, homehost);
440 if (strcmp(update, "uuid")==0 &&
441 !ident->uuid_set) {
442 ident->uuid_set = 1;
443 memcpy(ident->uuid, info.uuid, 16);
444 }
b8a8ccf9 445 if (dfd < 0)
5787fa49
NB
446 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
447 devname);
96395475 448 else if (st->ss->store_super(st, dfd, super))
5787fa49
NB
449 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
450 devname);
451 if (dfd >= 0)
452 close(dfd);
8131b493
NB
453
454 if (strcmp(update, "uuid")==0 &&
455 ident->bitmap_fd)
9fca7d62
NB
456 if (bitmap_update_uuid(ident->bitmap_fd, info.uuid) != 0)
457 fprintf(stderr, Name ": Could not update uuid on %s.\n",
458 devname);
d1f1011b
NB
459 } else
460#endif
461 {
da6b5ca9
NB
462 int dfd;
463 dfd = dev_open(devname, O_RDWR|O_EXCL);
464
0430ed48
NB
465 remove_partitions(dfd);
466
da6b5ca9
NB
467 if (super) {
468 free(super);
469 super = NULL;
470 }
471
472 st->ss->load_super(st, dfd, &super, NULL);
473 st->ss->getinfo_super(&info, super);
474 close(dfd);
5787fa49
NB
475 }
476
811e6cbe
NB
477 stat(devname, &stb);
478
dab6685f 479 if (verbose > 0)
cd29a5c8 480 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
4b1ac34b 481 devname, mddev, info.disk.raid_disk);
64c4757e 482 devices[devcnt].devname = devname;
0df46c2a
NB
483 devices[devcnt].major = major(stb.st_rdev);
484 devices[devcnt].minor = minor(stb.st_rdev);
4b1ac34b
NB
485 devices[devcnt].oldmajor = info.disk.major;
486 devices[devcnt].oldminor = info.disk.minor;
487 devices[devcnt].events = info.events;
488 devices[devcnt].raid_disk = info.disk.raid_disk;
6a41304b 489 devices[devcnt].disk_nr = info.disk.number;
64c4757e 490 devices[devcnt].uptodate = 0;
4b1ac34b 491 devices[devcnt].state = info.disk.state;
64c4757e
NB
492 if (most_recent < devcnt) {
493 if (devices[devcnt].events
494 > devices[most_recent].events)
495 most_recent = devcnt;
496 }
b8a8ccf9 497 if (info.array.level == -4)
5787fa49
NB
498 /* with multipath, the raid_disk from the superblock is meaningless */
499 i = devcnt;
500 else
501 i = devices[devcnt].raid_disk;
308e1801
NB
502 if (i+1 == 0) {
503 if (nextspare < info.array.raid_disks)
504 nextspare = info.array.raid_disks;
505 i = nextspare++;
08110d41
NB
506 } else {
507 if (i >= info.array.raid_disks &&
508 i >= nextspare)
509 nextspare = i+1;
308e1801 510 }
98c6faba 511 if (i < 10000) {
56eedc1a 512 if (i >= bestcnt) {
98c6faba 513 unsigned int newbestcnt = i+10;
56eedc1a 514 int *newbest = malloc(sizeof(int)*newbestcnt);
98c6faba 515 unsigned int c;
56eedc1a
NB
516 for (c=0; c < newbestcnt; c++)
517 if (c < bestcnt)
518 newbest[c] = best[c];
519 else
520 newbest[c] = -1;
521 if (best)free(best);
522 best = newbest;
523 bestcnt = newbestcnt;
524 }
6a255b69
NB
525 if (best[i] >=0 &&
526 devices[best[i]].events == devices[devcnt].events &&
527 devices[best[i]].minor != devices[devcnt].minor &&
528 st->ss->major == 0 &&
529 info.array.level != -4) {
530 /* two different devices with identical superblock.
531 * Could be a mis-detection caused by overlapping
532 * partitions. fail-safe.
533 */
534 fprintf(stderr, Name ": WARNING %s and %s appear"
535 " to have very similar superblocks.\n"
536 " If they are really different, "
537 "please --zero the superblock on one\n"
d2df86e0
NB
538 " If they are the same or overlap,"
539 " please remove one from %s.\n",
540 devices[best[i]].devname, devname,
541 inargv ? "the list" :
542 "the\n DEVICE list in mdadm.conf"
543 );
6a255b69
NB
544 if (must_close) close(mdfd);
545 return 1;
546 }
52826846
NB
547 if (best[i] == -1
548 || devices[best[i]].events < devices[devcnt].events)
549 best[i] = devcnt;
56eedc1a 550 }
64c4757e
NB
551 devcnt++;
552 }
553
4b1ac34b
NB
554 if (super)
555 free(super);
556 super = NULL;
557
586ed405
NB
558 if (update && strcmp(update, "byteorder")==0)
559 st->minor_version = 90;
560
64c4757e 561 if (devcnt == 0) {
682c7051 562 fprintf(stderr, Name ": no devices found for %s\n",
64c4757e 563 mddev);
4b1ac34b 564 free(first_super);
d55e3aef 565 if (must_close) close(mdfd);
64c4757e
NB
566 return 1;
567 }
4b1ac34b 568
31317663 569 st->ss->getinfo_super(&info, first_super);
4b1ac34b 570
64c4757e
NB
571 /* now we have some devices that might be suitable.
572 * I wonder how many
573 */
265e0f17
NB
574 avail = malloc(info.array.raid_disks);
575 memset(avail, 0, info.array.raid_disks);
64c4757e 576 okcnt = 0;
52826846 577 sparecnt=0;
56eedc1a 578 for (i=0; i< bestcnt ;i++) {
64c4757e 579 int j = best[i];
ee04451c
NB
580 int event_margin = 1; /* always allow a difference of '1'
581 * like the kernel does
582 */
64c4757e 583 if (j < 0) continue;
d013a55e
NB
584 /* note: we ignore error flags in multipath arrays
585 * as they don't make sense
586 */
4b1ac34b 587 if (info.array.level != -4)
aa88f531
NB
588 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
589 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
590 sparecnt++;
d013a55e 591 continue;
aa88f531 592 }
cd29a5c8 593 if (devices[j].events+event_margin >=
64c4757e
NB
594 devices[most_recent].events) {
595 devices[j].uptodate = 1;
265e0f17 596 if (i < info.array.raid_disks) {
52826846 597 okcnt++;
265e0f17
NB
598 avail[i]=1;
599 } else
52826846 600 sparecnt++;
64c4757e
NB
601 }
602 }
265e0f17
NB
603 while (force && !enough(info.array.level, info.array.raid_disks,
604 info.array.layout,
605 avail, okcnt)) {
64c4757e
NB
606 /* Choose the newest best drive which is
607 * not up-to-date, update the superblock
608 * and add it.
609 */
52826846 610 int fd;
cd29a5c8 611 chosen_drive = -1;
4b1ac34b 612 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
52826846
NB
613 int j = best[i];
614 if (j>=0 &&
615 !devices[j].uptodate &&
616 devices[j].events > 0 &&
617 (chosen_drive < 0 ||
618 devices[j].events > devices[chosen_drive].events))
619 chosen_drive = j;
620 }
621 if (chosen_drive < 0)
622 break;
dab6685f
NB
623 if (verbose >= 0)
624 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
625 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
626 (int)(devices[chosen_drive].events),
627 (int)(devices[most_recent].events));
8b0dabea 628 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846
NB
629 if (fd < 0) {
630 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
631 devices[chosen_drive].devname);
632 devices[chosen_drive].events = 0;
633 continue;
634 }
82d9eba6 635 if (st->ss->load_super(st,fd, &super, NULL)) {
52826846
NB
636 close(fd);
637 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
638 devices[chosen_drive].devname);
639 devices[chosen_drive].events = 0;
640 continue;
641 }
4b1ac34b 642 info.events = devices[most_recent].events;
e5eac01f 643 st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose, 0, NULL);
4b1ac34b 644
96395475 645 if (st->ss->store_super(st, fd, super)) {
52826846
NB
646 close(fd);
647 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
648 devices[chosen_drive].devname);
649 devices[chosen_drive].events = 0;
4b1ac34b 650 free(super);
52826846
NB
651 continue;
652 }
653 close(fd);
654 devices[chosen_drive].events = devices[most_recent].events;
655 devices[chosen_drive].uptodate = 1;
265e0f17 656 avail[chosen_drive] = 1;
52826846 657 okcnt++;
4b1ac34b 658 free(super);
64c4757e 659 }
52826846
NB
660
661 /* Now we want to look at the superblock which the kernel will base things on
662 * and compare the devices that we think are working with the devices that the
663 * superblock thinks are working.
664 * If there are differences and --force is given, then update this chosen
665 * superblock.
666 */
cd29a5c8 667 chosen_drive = -1;
4b1ac34b 668 super = NULL;
56eedc1a 669 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
52826846
NB
670 int j = best[i];
671 int fd;
4b1ac34b 672
52826846
NB
673 if (j<0)
674 continue;
675 if (!devices[j].uptodate)
676 continue;
677 chosen_drive = j;
8b0dabea 678 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
52826846
NB
679 fprintf(stderr, Name ": Cannot open %s: %s\n",
680 devices[j].devname, strerror(errno));
d55e3aef 681 if (must_close) close(mdfd);
52826846
NB
682 return 1;
683 }
82d9eba6 684 if (st->ss->load_super(st,fd, &super, NULL)) {
52826846
NB
685 close(fd);
686 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
687 devices[j].devname);
d55e3aef 688 if (must_close) close(mdfd);
52826846
NB
689 return 1;
690 }
691 close(fd);
692 }
4b1ac34b
NB
693 if (super == NULL) {
694 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
d55e3aef 695 if (must_close) close(mdfd);
4b1ac34b
NB
696 return 1;
697 }
31317663 698 st->ss->getinfo_super(&info, super);
56eedc1a 699 for (i=0; i<bestcnt; i++) {
52826846 700 int j = best[i];
98c6faba 701 unsigned int desired_state;
11a3e71d 702
4b1ac34b 703 if (i < info.array.raid_disks)
11a3e71d
NB
704 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
705 else
706 desired_state = 0;
707
52826846
NB
708 if (j<0)
709 continue;
710 if (!devices[j].uptodate)
711 continue;
6a41304b 712 info.disk.number = devices[j].disk_nr;
fbf8a0b7 713 info.disk.raid_disk = i;
4b1ac34b
NB
714 info.disk.state = desired_state;
715
52826846 716 if (devices[j].uptodate &&
e5eac01f 717 st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) {
52826846 718 if (force) {
dab6685f
NB
719 if (verbose >= 0)
720 fprintf(stderr, Name ": "
721 "clearing FAULTY flag for device %d in %s for %s\n",
722 j, mddev, devices[j].devname);
4b1ac34b 723 change = 1;
52826846 724 } else {
dab6685f
NB
725 if (verbose >= -1)
726 fprintf(stderr, Name ": "
727 "device %d in %s has wrong state in superblock, but %s seems ok\n",
728 i, mddev, devices[j].devname);
52826846
NB
729 }
730 }
4b1ac34b 731#if 0
52826846
NB
732 if (!devices[j].uptodate &&
733 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
734 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
735 i, mddev);
736 }
4b1ac34b 737#endif
52826846 738 }
4b1ac34b
NB
739 if (force && okcnt == info.array.raid_disks-1) {
740 /* FIXME check event count */
b8a8ccf9 741 change += st->ss->update_super(&info, super, "force",
e5eac01f 742 devices[chosen_drive].devname, verbose, 0, NULL);
d013a55e 743 }
52826846 744
4b1ac34b 745 if (change) {
52826846 746 int fd;
8b0dabea 747 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846 748 if (fd < 0) {
353632d9 749 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
52826846 750 devices[chosen_drive].devname);
d55e3aef 751 if (must_close) close(mdfd);
52826846
NB
752 return 1;
753 }
96395475 754 if (st->ss->store_super(st, fd, super)) {
52826846
NB
755 close(fd);
756 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
757 devices[chosen_drive].devname);
d55e3aef 758 if (must_close) close(mdfd);
52826846
NB
759 return 1;
760 }
761 close(fd);
52826846
NB
762 }
763
353632d9
NB
764 /* If we are in the middle of a reshape we may need to restore saved data
765 * that was moved aside due to the reshape overwriting live data
766 * The code of doing this lives in Grow.c
767 */
39c5a909 768#ifndef MDASSEMBLE
353632d9
NB
769 if (info.reshape_active) {
770 int err = 0;
771 int *fdlist = malloc(sizeof(int)* bestcnt);
772 for (i=0; i<bestcnt; i++) {
773 int j = best[i];
774 if (j >= 0) {
775 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
776 if (fdlist[i] < 0) {
777 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
778 devices[j].devname);
779 err = 1;
780 break;
781 }
782 } else
783 fdlist[i] = -1;
784 }
785 if (!err)
06b0d786 786 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
353632d9
NB
787 while (i>0) {
788 i--;
789 if (fdlist[i]>=0) close(fdlist[i]);
790 }
791 if (err) {
792 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
d55e3aef 793 if (must_close) close(mdfd);
353632d9
NB
794 return err;
795 }
796 }
39c5a909 797#endif
aa88f531
NB
798 /* count number of in-sync devices according to the superblock.
799 * We must have this number to start the array without -s or -R
800 */
4b1ac34b 801 req_cnt = info.array.working_disks;
aa88f531 802
64c4757e
NB
803 /* Almost ready to actually *do* something */
804 if (!old_linux) {
fbf8a0b7
NB
805 int rv;
806 if ((vers % 100) >= 1) { /* can use different versions */
807 mdu_array_info_t inf;
808 memset(&inf, 0, sizeof(inf));
809 inf.major_version = st->ss->major;
810 inf.minor_version = st->minor_version;
811 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
b8a8ccf9 812 } else
fbf8a0b7
NB
813 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
814
815 if (rv) {
682c7051 816 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
64c4757e 817 mddev, strerror(errno));
d55e3aef 818 if (must_close) close(mdfd);
64c4757e
NB
819 return 1;
820 }
11484a63 821 if (ident->bitmap_fd >= 0) {
c82f047c
NB
822 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
823 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
d55e3aef 824 if (must_close) close(mdfd);
c82f047c
NB
825 return 1;
826 }
7ef02d01
NB
827 } else if (ident->bitmap_file) {
828 /* From config file */
829 int bmfd = open(ident->bitmap_file, O_RDWR);
830 if (bmfd < 0) {
831 fprintf(stderr, Name ": Could not open bitmap file %s\n",
832 ident->bitmap_file);
d55e3aef 833 if (must_close) close(mdfd);
7ef02d01
NB
834 return 1;
835 }
836 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
837 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
838 close(bmfd);
d55e3aef 839 if (must_close) close(mdfd);
7ef02d01
NB
840 return 1;
841 }
842 close(bmfd);
c82f047c 843 }
7ef02d01 844
52826846 845 /* First, add the raid disks, but add the chosen one last */
56eedc1a 846 for (i=0; i<= bestcnt; i++) {
52826846 847 int j;
56eedc1a 848 if (i < bestcnt) {
52826846
NB
849 j = best[i];
850 if (j == chosen_drive)
851 continue;
852 } else
853 j = chosen_drive;
854
56eedc1a 855 if (j >= 0 /* && devices[j].uptodate */) {
64c4757e
NB
856 mdu_disk_info_t disk;
857 memset(&disk, 0, sizeof(disk));
858 disk.major = devices[j].major;
859 disk.minor = devices[j].minor;
860 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
682c7051 861 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
64c4757e
NB
862 devices[j].devname,
863 mddev,
864 strerror(errno));
4b1ac34b 865 if (i < info.array.raid_disks || i == bestcnt)
52826846
NB
866 okcnt--;
867 else
868 sparecnt--;
dab6685f 869 } else if (verbose > 0)
52826846
NB
870 fprintf(stderr, Name ": added %s to %s as %d\n",
871 devices[j].devname, mddev, devices[j].raid_disk);
dab6685f 872 } else if (verbose > 0 && i < info.array.raid_disks)
682c7051 873 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
64c4757e
NB
874 i, mddev);
875 }
52826846 876
64c4757e 877 if (runstop == 1 ||
b8a8ccf9 878 (runstop <= 0 &&
265e0f17 879 ( enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt) &&
aa88f531
NB
880 (okcnt >= req_cnt || start_partial_ok)
881 ))) {
82b27616 882 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
dab6685f
NB
883 if (verbose >= 0) {
884 fprintf(stderr, Name ": %s has been started with %d drive%s",
885 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9 886 if (okcnt < info.array.raid_disks)
dab6685f
NB
887 fprintf(stderr, " (out of %d)", info.array.raid_disks);
888 if (sparecnt)
889 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
890 fprintf(stderr, ".\n");
891 }
589395d6
NB
892 if (must_close) {
893 int usecs = 1;
894 close(mdfd);
895 /* There is a nasty race with 'mdadm --monitor'.
896 * If it opens this device before we close it,
897 * it gets an incomplete open on which IO
898 * doesn't work and the capacity if wrong.
899 * If we reopen (to check for layered devices)
900 * before --monitor closes, we loose.
901 *
902 * So: wait upto 1 second for there to be
903 * a non-zero capacity.
904 */
905 while (usecs < 1000) {
906 mdfd = open(mddev, O_RDONLY);
907 if (mdfd >= 0) {
908 unsigned long size;
909 if (ioctl(mdfd, BLKGETSIZE, &size) == 0 &&
910 size > 0)
911 break;
912 close(mdfd);
913 }
914 usleep(usecs);
915 usecs <<= 1;
916 }
917 }
64c4757e 918 return 0;
82b27616 919 }
682c7051 920 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
64c4757e 921 mddev, strerror(errno));
d55e3aef 922 if (must_close) close(mdfd);
64c4757e
NB
923 return 1;
924 }
82b27616 925 if (runstop == -1) {
b8a8ccf9 926 fprintf(stderr, Name ": %s assembled from %d drive%s",
52826846 927 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9
NB
928 if (okcnt != info.array.raid_disks)
929 fprintf(stderr, " (out of %d)", info.array.raid_disks);
930 fprintf(stderr, ", but not started.\n");
d55e3aef 931 if (must_close) close(mdfd);
64c4757e 932 return 0;
82b27616 933 }
dab6685f
NB
934 if (verbose >= 0) {
935 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
936 if (sparecnt)
937 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
265e0f17 938 if (!enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt))
dab6685f
NB
939 fprintf(stderr, " - not enough to start the array.\n");
940 else {
941 if (req_cnt == info.array.raid_disks)
942 fprintf(stderr, " - need all %d to start it", req_cnt);
943 else
944 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
945 fprintf(stderr, " (use --run to insist).\n");
946 }
aa88f531 947 }
d55e3aef 948 if (must_close) close(mdfd);
64c4757e 949 return 1;
82b27616 950 } else {
52826846
NB
951 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
952 * been updated to point to the current locations of devices.
953 * so we can just start the array
82b27616 954 */
cd29a5c8 955 unsigned long dev;
0df46c2a 956 dev = makedev(devices[chosen_drive].major,
82b27616
NB
957 devices[chosen_drive].minor);
958 if (ioctl(mdfd, START_ARRAY, dev)) {
959 fprintf(stderr, Name ": Cannot start array: %s\n",
960 strerror(errno));
961 }
962
64c4757e 963 }
d55e3aef 964 if (must_close) close(mdfd);
e0d19036 965 return 0;
64c4757e 966}