]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Assemble.c
fix load_super/free_super mismatch in util.c
[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);
3da92f27 318 st->ss->free_super(st);
52826846
NB
319 return 1;
320 }
838acbc2 321
83b6208e 322 if (st == NULL)
3da92f27
NB
323 st = dup_super(tst);
324 if (st->minor_version == -1)
325 st->minor_version = tst->minor_version;
83b6208e
NB
326 if (st->ss != tst->ss ||
327 st->minor_version != tst->minor_version ||
64557c33 328 st->ss->compare_super(st, tst) != 0) {
83b6208e 329 /* Some mismatch. If exactly one array matches this host,
da6b5ca9
NB
330 * we can resolve on that one.
331 * Or, if we are auto assembling, we just ignore the second
332 * for now.
83b6208e 333 */
da6b5ca9 334 if (mdfd < 0)
df37ffc0 335 goto loop;
83b6208e 336 if (homehost) {
3da92f27
NB
337 int first = st->ss->match_home(st, homehost);
338 int last = tst->ss->match_home(tst, homehost);
83b6208e
NB
339 if (first+last == 1) {
340 /* We can do something */
341 if (first) {/* just ignore this one */
342 if ((inargv && verbose >= 0) || verbose > 0)
343 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
344 devname);
df37ffc0 345 goto loop;
83b6208e
NB
346 } else { /* reject all those sofar */
347 mddev_dev_t td;
348 if ((inargv && verbose >= 0) || verbose > 0)
349 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
350 devname);
351 for (td=devlist; td != tmpdev; td=td->next)
352 if (td->used == 1)
353 td->used = 0;
354 tmpdev->used = 1;
df37ffc0 355 goto loop;
83b6208e
NB
356 }
357 }
358 }
52826846
NB
359 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
360 devname);
3da92f27
NB
361 tst->ss->free_super(tst);
362 st->ss->free_super(st);
52826846 363 return 1;
64c4757e
NB
364 }
365
811e6cbe 366 tmpdev->used = 1;
df37ffc0
NB
367
368 loop:
3da92f27 369 tst->ss->free_super(tst);
811e6cbe
NB
370 }
371
8a46fe84
NB
372 if (mdfd < 0) {
373 /* So... it is up to me to open the device.
374 * We create a name '/dev/md/XXX' based on the info in the
375 * superblock, and call open_mddev on that
376 */
da6b5ca9
NB
377 mdu_array_info_t inf;
378 char *c;
3da92f27 379 if (!st->sb) {
d55e3aef 380 return 2;
da6b5ca9 381 }
3da92f27 382 st->ss->getinfo_super(st, &info);
da6b5ca9
NB
383 c = strchr(info.name, ':');
384 if (c) c++; else c= info.name;
41a3b72a
NB
385 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
386 /* /dev/md/d0 style for partitionable */
387 asprintf(&mddev, "/dev/md/d%s", c);
388 else
389 asprintf(&mddev, "/dev/md/%s", c);
da6b5ca9 390 mdfd = open_mddev(mddev, ident->autof);
60e1bc1a 391 if (mdfd < 0) {
3da92f27 392 st->ss->free_super(st);
60e1bc1a 393 free(devices);
60e1bc1a
NB
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;
3da92f27 408 st->ss->free_super(st);
60e1bc1a 409 free(devices);
60e1bc1a 410 goto try_again;
da6b5ca9 411 }
d55e3aef 412 must_close = 1;
8a46fe84
NB
413 }
414
811e6cbe 415 /* Ok, no bad inconsistancy, we can try updating etc */
bf4fb153 416 bitmap_done = 0;
da6b5ca9 417 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
811e6cbe
NB
418 char *devname = tmpdev->devname;
419 struct stat stb;
5787fa49 420 /* looks like a good enough match to update the super block if needed */
d1f1011b 421#ifndef MDASSEMBLE
5787fa49 422 if (update) {
811e6cbe 423 int dfd;
4b1ac34b
NB
424 /* prepare useful information in info structures */
425 struct stat stb2;
3da92f27 426 struct supertype *tst;
4b1ac34b 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
3da92f27
NB
445 tst = dup_super(st);
446 tst->ss->load_super(tst, dfd, NULL);
447 tst->ss->getinfo_super(tst, &info);
811e6cbe 448
7d99579f 449 memcpy(info.uuid, ident->uuid, 16);
e5eac01f 450 strcpy(info.name, ident->name);
811e6cbe
NB
451 info.array.md_minor = minor(stb2.st_rdev);
452
3da92f27
NB
453 tst->ss->update_super(tst, &info, update,
454 devname, verbose,
455 ident->uuid_set, homehost);
e5eac01f
NB
456 if (strcmp(update, "uuid")==0 &&
457 !ident->uuid_set) {
458 ident->uuid_set = 1;
459 memcpy(ident->uuid, info.uuid, 16);
460 }
b8a8ccf9 461 if (dfd < 0)
5787fa49
NB
462 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
463 devname);
3da92f27 464 else if (tst->ss->store_super(tst, dfd))
5787fa49
NB
465 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
466 devname);
467 if (dfd >= 0)
468 close(dfd);
8131b493
NB
469
470 if (strcmp(update, "uuid")==0 &&
bf4fb153 471 ident->bitmap_fd >= 0 && !bitmap_done) {
3da92f27
NB
472 if (bitmap_update_uuid(ident->bitmap_fd,
473 info.uuid,
474 tst->ss->swapuuid) != 0)
bf4fb153
NB
475 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
476 else
477 bitmap_done = 1;
478 }
3da92f27 479 tst->ss->free_super(tst);
d1f1011b
NB
480 } else
481#endif
482 {
3da92f27 483 struct supertype *tst = dup_super(st);;
da6b5ca9
NB
484 int dfd;
485 dfd = dev_open(devname, O_RDWR|O_EXCL);
486
0430ed48
NB
487 remove_partitions(dfd);
488
3da92f27
NB
489 tst->ss->load_super(tst, dfd, NULL);
490 tst->ss->getinfo_super(tst, &info);
491 tst->ss->free_super(tst);
da6b5ca9 492 close(dfd);
5787fa49
NB
493 }
494
811e6cbe
NB
495 stat(devname, &stb);
496
dab6685f 497 if (verbose > 0)
cd29a5c8 498 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
4b1ac34b 499 devname, mddev, info.disk.raid_disk);
64c4757e 500 devices[devcnt].devname = devname;
64c4757e 501 devices[devcnt].uptodate = 0;
213ee40b
NB
502 devices[devcnt].i = info;
503 devices[devcnt].i.disk.major = major(stb.st_rdev);
504 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
64c4757e 505 if (most_recent < devcnt) {
213ee40b
NB
506 if (devices[devcnt].i.events
507 > devices[most_recent].i.events)
64c4757e
NB
508 most_recent = devcnt;
509 }
b8a8ccf9 510 if (info.array.level == -4)
5787fa49
NB
511 /* with multipath, the raid_disk from the superblock is meaningless */
512 i = devcnt;
513 else
213ee40b 514 i = devices[devcnt].i.disk.raid_disk;
308e1801
NB
515 if (i+1 == 0) {
516 if (nextspare < info.array.raid_disks)
517 nextspare = info.array.raid_disks;
518 i = nextspare++;
08110d41
NB
519 } else {
520 if (i >= info.array.raid_disks &&
521 i >= nextspare)
522 nextspare = i+1;
308e1801 523 }
98c6faba 524 if (i < 10000) {
56eedc1a 525 if (i >= bestcnt) {
98c6faba 526 unsigned int newbestcnt = i+10;
56eedc1a 527 int *newbest = malloc(sizeof(int)*newbestcnt);
98c6faba 528 unsigned int c;
56eedc1a
NB
529 for (c=0; c < newbestcnt; c++)
530 if (c < bestcnt)
531 newbest[c] = best[c];
532 else
533 newbest[c] = -1;
534 if (best)free(best);
535 best = newbest;
536 bestcnt = newbestcnt;
537 }
6a255b69 538 if (best[i] >=0 &&
213ee40b
NB
539 devices[best[i]].i.events
540 == devices[devcnt].i.events
541 && (devices[best[i]].i.disk.minor
542 != devices[devcnt].i.disk.minor)
543 && st->ss->major == 0
544 && info.array.level != -4) {
6a255b69
NB
545 /* two different devices with identical superblock.
546 * Could be a mis-detection caused by overlapping
547 * partitions. fail-safe.
548 */
549 fprintf(stderr, Name ": WARNING %s and %s appear"
550 " to have very similar superblocks.\n"
551 " If they are really different, "
552 "please --zero the superblock on one\n"
d2df86e0
NB
553 " If they are the same or overlap,"
554 " please remove one from %s.\n",
555 devices[best[i]].devname, devname,
556 inargv ? "the list" :
557 "the\n DEVICE list in mdadm.conf"
558 );
6a255b69
NB
559 if (must_close) close(mdfd);
560 return 1;
561 }
52826846 562 if (best[i] == -1
213ee40b
NB
563 || (devices[best[i]].i.events
564 < devices[devcnt].i.events))
52826846 565 best[i] = devcnt;
56eedc1a 566 }
64c4757e 567 devcnt++;
df37ffc0 568 }
4b1ac34b 569
586ed405
NB
570 if (update && strcmp(update, "byteorder")==0)
571 st->minor_version = 90;
572
64c4757e 573 if (devcnt == 0) {
682c7051 574 fprintf(stderr, Name ": no devices found for %s\n",
64c4757e 575 mddev);
3da92f27 576 st->ss->free_super(st);
d55e3aef 577 if (must_close) close(mdfd);
64c4757e
NB
578 return 1;
579 }
4b1ac34b 580
3da92f27 581 st->ss->getinfo_super(st, &info);
583315d9 582 clean = info.array.state & 1;
4b1ac34b 583
64c4757e
NB
584 /* now we have some devices that might be suitable.
585 * I wonder how many
586 */
265e0f17
NB
587 avail = malloc(info.array.raid_disks);
588 memset(avail, 0, info.array.raid_disks);
64c4757e 589 okcnt = 0;
52826846 590 sparecnt=0;
56eedc1a 591 for (i=0; i< bestcnt ;i++) {
64c4757e 592 int j = best[i];
ee04451c
NB
593 int event_margin = 1; /* always allow a difference of '1'
594 * like the kernel does
595 */
64c4757e 596 if (j < 0) continue;
d013a55e
NB
597 /* note: we ignore error flags in multipath arrays
598 * as they don't make sense
599 */
4b1ac34b 600 if (info.array.level != -4)
213ee40b
NB
601 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
602 if (!(devices[j].i.disk.state
603 & (1<<MD_DISK_FAULTY)))
aa88f531 604 sparecnt++;
d013a55e 605 continue;
aa88f531 606 }
213ee40b
NB
607 if (devices[j].i.events+event_margin >=
608 devices[most_recent].i.events) {
64c4757e 609 devices[j].uptodate = 1;
265e0f17 610 if (i < info.array.raid_disks) {
52826846 611 okcnt++;
265e0f17
NB
612 avail[i]=1;
613 } else
52826846 614 sparecnt++;
64c4757e
NB
615 }
616 }
265e0f17 617 while (force && !enough(info.array.level, info.array.raid_disks,
583315d9 618 info.array.layout, 1,
265e0f17 619 avail, okcnt)) {
64c4757e
NB
620 /* Choose the newest best drive which is
621 * not up-to-date, update the superblock
622 * and add it.
623 */
52826846 624 int fd;
3da92f27 625 struct supertype *tst;
c5a6f9a6 626 long long current_events;
cd29a5c8 627 chosen_drive = -1;
4b1ac34b 628 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
52826846
NB
629 int j = best[i];
630 if (j>=0 &&
631 !devices[j].uptodate &&
213ee40b 632 devices[j].i.events > 0 &&
52826846 633 (chosen_drive < 0 ||
213ee40b
NB
634 devices[j].i.events
635 > devices[chosen_drive].i.events))
52826846
NB
636 chosen_drive = j;
637 }
638 if (chosen_drive < 0)
639 break;
213ee40b 640 current_events = devices[chosen_drive].i.events;
c5a6f9a6 641 add_another:
dab6685f
NB
642 if (verbose >= 0)
643 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
213ee40b
NB
644 devices[chosen_drive].devname,
645 devices[chosen_drive].i.disk.raid_disk,
646 (int)(devices[chosen_drive].i.events),
647 (int)(devices[most_recent].i.events));
8b0dabea 648 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846
NB
649 if (fd < 0) {
650 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
651 devices[chosen_drive].devname);
213ee40b 652 devices[chosen_drive].i.events = 0;
52826846
NB
653 continue;
654 }
3da92f27
NB
655 tst = dup_super(st);
656 if (tst->ss->load_super(st,fd, NULL)) {
52826846
NB
657 close(fd);
658 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
659 devices[chosen_drive].devname);
213ee40b 660 devices[chosen_drive].i.events = 0;
52826846
NB
661 continue;
662 }
213ee40b 663 info.events = devices[most_recent].i.events;
3da92f27 664 tst->ss->update_super(tst, &info, "force-one",
67a8c82d
NB
665 devices[chosen_drive].devname, verbose,
666 0, NULL);
4b1ac34b 667
3da92f27 668 if (tst->ss->store_super(tst, fd)) {
52826846
NB
669 close(fd);
670 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
671 devices[chosen_drive].devname);
213ee40b 672 devices[chosen_drive].i.events = 0;
3da92f27 673 tst->ss->free_super(tst);
52826846
NB
674 continue;
675 }
676 close(fd);
213ee40b 677 devices[chosen_drive].i.events = devices[most_recent].i.events;
52826846 678 devices[chosen_drive].uptodate = 1;
265e0f17 679 avail[chosen_drive] = 1;
52826846 680 okcnt++;
3da92f27 681 tst->ss->free_super(tst);
c5a6f9a6
NB
682
683 /* If there are any other drives of the same vintage,
684 * add them in as well. We can't lose and we might gain
685 */
686 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
687 int j = best[i];
688 if (j >= 0 &&
689 !devices[j].uptodate &&
213ee40b
NB
690 devices[j].i.events > 0 &&
691 devices[j].i.events == current_events) {
c5a6f9a6
NB
692 chosen_drive = j;
693 goto add_another;
694 }
695 }
64c4757e 696 }
52826846
NB
697
698 /* Now we want to look at the superblock which the kernel will base things on
699 * and compare the devices that we think are working with the devices that the
700 * superblock thinks are working.
701 * If there are differences and --force is given, then update this chosen
702 * superblock.
703 */
cd29a5c8 704 chosen_drive = -1;
3da92f27 705 st->ss->free_super(st);
56eedc1a 706 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
52826846
NB
707 int j = best[i];
708 int fd;
4b1ac34b 709
52826846
NB
710 if (j<0)
711 continue;
712 if (!devices[j].uptodate)
713 continue;
714 chosen_drive = j;
8b0dabea 715 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
52826846
NB
716 fprintf(stderr, Name ": Cannot open %s: %s\n",
717 devices[j].devname, strerror(errno));
d55e3aef 718 if (must_close) close(mdfd);
52826846
NB
719 return 1;
720 }
3da92f27 721 if (st->ss->load_super(st,fd, NULL)) {
52826846
NB
722 close(fd);
723 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
724 devices[j].devname);
d55e3aef 725 if (must_close) close(mdfd);
52826846
NB
726 return 1;
727 }
728 close(fd);
729 }
3da92f27 730 if (st->sb == NULL) {
4b1ac34b 731 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
d55e3aef 732 if (must_close) close(mdfd);
4b1ac34b
NB
733 return 1;
734 }
3da92f27 735 st->ss->getinfo_super(st, &info);
56eedc1a 736 for (i=0; i<bestcnt; i++) {
52826846 737 int j = best[i];
98c6faba 738 unsigned int desired_state;
11a3e71d 739
4b1ac34b 740 if (i < info.array.raid_disks)
11a3e71d
NB
741 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
742 else
743 desired_state = 0;
744
52826846
NB
745 if (j<0)
746 continue;
747 if (!devices[j].uptodate)
748 continue;
4b1ac34b 749
213ee40b
NB
750 devices[j].i.disk.state = desired_state;
751
752 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
68c7d6d7 753 verbose, 0, NULL)) {
52826846 754 if (force) {
dab6685f
NB
755 if (verbose >= 0)
756 fprintf(stderr, Name ": "
757 "clearing FAULTY flag for device %d in %s for %s\n",
758 j, mddev, devices[j].devname);
4b1ac34b 759 change = 1;
52826846 760 } else {
dab6685f
NB
761 if (verbose >= -1)
762 fprintf(stderr, Name ": "
763 "device %d in %s has wrong state in superblock, but %s seems ok\n",
764 i, mddev, devices[j].devname);
52826846
NB
765 }
766 }
4b1ac34b 767#if 0
213ee40b 768 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
52826846
NB
769 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
770 i, mddev);
771 }
4b1ac34b 772#endif
52826846 773 }
c5a6f9a6
NB
774 if (force && !clean &&
775 !enough(info.array.level, info.array.raid_disks,
776 info.array.layout, clean,
777 avail, okcnt)) {
3da92f27 778 change += st->ss->update_super(st, &info, "force-array",
67a8c82d
NB
779 devices[chosen_drive].devname, verbose,
780 0, NULL);
583315d9 781 clean = 1;
d013a55e 782 }
52826846 783
4b1ac34b 784 if (change) {
52826846 785 int fd;
8b0dabea 786 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
52826846 787 if (fd < 0) {
353632d9 788 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
52826846 789 devices[chosen_drive].devname);
d55e3aef 790 if (must_close) close(mdfd);
52826846
NB
791 return 1;
792 }
3da92f27 793 if (st->ss->store_super(st, fd)) {
52826846
NB
794 close(fd);
795 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
796 devices[chosen_drive].devname);
d55e3aef 797 if (must_close) close(mdfd);
52826846
NB
798 return 1;
799 }
800 close(fd);
52826846
NB
801 }
802
353632d9
NB
803 /* If we are in the middle of a reshape we may need to restore saved data
804 * that was moved aside due to the reshape overwriting live data
805 * The code of doing this lives in Grow.c
806 */
39c5a909 807#ifndef MDASSEMBLE
353632d9
NB
808 if (info.reshape_active) {
809 int err = 0;
810 int *fdlist = malloc(sizeof(int)* bestcnt);
811 for (i=0; i<bestcnt; i++) {
812 int j = best[i];
813 if (j >= 0) {
814 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
815 if (fdlist[i] < 0) {
816 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
817 devices[j].devname);
818 err = 1;
819 break;
820 }
821 } else
822 fdlist[i] = -1;
823 }
824 if (!err)
06b0d786 825 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
353632d9
NB
826 while (i>0) {
827 i--;
828 if (fdlist[i]>=0) close(fdlist[i]);
829 }
830 if (err) {
831 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
d55e3aef 832 if (must_close) close(mdfd);
353632d9
NB
833 return err;
834 }
835 }
39c5a909 836#endif
aa88f531
NB
837 /* count number of in-sync devices according to the superblock.
838 * We must have this number to start the array without -s or -R
839 */
4b1ac34b 840 req_cnt = info.array.working_disks;
aa88f531 841
64c4757e
NB
842 /* Almost ready to actually *do* something */
843 if (!old_linux) {
fbf8a0b7
NB
844 int rv;
845 if ((vers % 100) >= 1) { /* can use different versions */
846 mdu_array_info_t inf;
847 memset(&inf, 0, sizeof(inf));
848 inf.major_version = st->ss->major;
849 inf.minor_version = st->minor_version;
850 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
b8a8ccf9 851 } else
fbf8a0b7
NB
852 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
853
854 if (rv) {
682c7051 855 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
64c4757e 856 mddev, strerror(errno));
d55e3aef 857 if (must_close) close(mdfd);
64c4757e
NB
858 return 1;
859 }
11484a63 860 if (ident->bitmap_fd >= 0) {
c82f047c
NB
861 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
862 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
d55e3aef 863 if (must_close) close(mdfd);
c82f047c
NB
864 return 1;
865 }
7ef02d01
NB
866 } else if (ident->bitmap_file) {
867 /* From config file */
868 int bmfd = open(ident->bitmap_file, O_RDWR);
869 if (bmfd < 0) {
870 fprintf(stderr, Name ": Could not open bitmap file %s\n",
871 ident->bitmap_file);
d55e3aef 872 if (must_close) close(mdfd);
7ef02d01
NB
873 return 1;
874 }
875 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
876 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
877 close(bmfd);
d55e3aef 878 if (must_close) close(mdfd);
7ef02d01
NB
879 return 1;
880 }
881 close(bmfd);
c82f047c 882 }
7ef02d01 883
52826846 884 /* First, add the raid disks, but add the chosen one last */
56eedc1a 885 for (i=0; i<= bestcnt; i++) {
52826846 886 int j;
56eedc1a 887 if (i < bestcnt) {
52826846
NB
888 j = best[i];
889 if (j == chosen_drive)
890 continue;
891 } else
892 j = chosen_drive;
893
56eedc1a 894 if (j >= 0 /* && devices[j].uptodate */) {
213ee40b
NB
895 if (ioctl(mdfd, ADD_NEW_DISK,
896 &devices[j].i.disk)!=0) {
897 fprintf(stderr, Name ": failed to add "
898 "%s to %s: %s\n",
64c4757e
NB
899 devices[j].devname,
900 mddev,
901 strerror(errno));
213ee40b
NB
902 if (i < info.array.raid_disks
903 || i == bestcnt)
52826846
NB
904 okcnt--;
905 else
906 sparecnt--;
dab6685f 907 } else if (verbose > 0)
213ee40b
NB
908 fprintf(stderr, Name ": added %s "
909 "to %s as %d\n",
910 devices[j].devname, mddev,
911 devices[j].i.disk.raid_disk);
dab6685f 912 } else if (verbose > 0 && i < info.array.raid_disks)
213ee40b
NB
913 fprintf(stderr, Name ": no uptodate device for "
914 "slot %d of %s\n",
64c4757e
NB
915 i, mddev);
916 }
aba69144 917
64c4757e 918 if (runstop == 1 ||
b8a8ccf9 919 (runstop <= 0 &&
583315d9
NB
920 ( enough(info.array.level, info.array.raid_disks,
921 info.array.layout, clean, avail, okcnt) &&
aa88f531
NB
922 (okcnt >= req_cnt || start_partial_ok)
923 ))) {
82b27616 924 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
dab6685f
NB
925 if (verbose >= 0) {
926 fprintf(stderr, Name ": %s has been started with %d drive%s",
927 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9 928 if (okcnt < info.array.raid_disks)
dab6685f
NB
929 fprintf(stderr, " (out of %d)", info.array.raid_disks);
930 if (sparecnt)
931 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
932 fprintf(stderr, ".\n");
933 }
589395d6
NB
934 if (must_close) {
935 int usecs = 1;
936 close(mdfd);
937 /* There is a nasty race with 'mdadm --monitor'.
938 * If it opens this device before we close it,
939 * it gets an incomplete open on which IO
940 * doesn't work and the capacity if wrong.
941 * If we reopen (to check for layered devices)
942 * before --monitor closes, we loose.
943 *
944 * So: wait upto 1 second for there to be
945 * a non-zero capacity.
946 */
947 while (usecs < 1000) {
948 mdfd = open(mddev, O_RDONLY);
949 if (mdfd >= 0) {
beae1dfe
NB
950 unsigned long long size;
951 if (get_dev_size(mdfd, NULL, &size) &&
589395d6
NB
952 size > 0)
953 break;
954 close(mdfd);
955 }
956 usleep(usecs);
957 usecs <<= 1;
958 }
959 }
64c4757e 960 return 0;
82b27616 961 }
682c7051 962 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
64c4757e 963 mddev, strerror(errno));
583315d9
NB
964
965 if (!enough(info.array.level, info.array.raid_disks,
966 info.array.layout, 1, avail, okcnt))
967 fprintf(stderr, Name ": Not enough devices to "
968 "start the array.\n");
969 else if (!enough(info.array.level,
970 info.array.raid_disks,
971 info.array.layout, clean,
972 avail, okcnt))
973 fprintf(stderr, Name ": Not enough devices to "
974 "start the array while not clean "
975 "- consider --force.\n");
976
d55e3aef 977 if (must_close) close(mdfd);
64c4757e
NB
978 return 1;
979 }
82b27616 980 if (runstop == -1) {
b8a8ccf9 981 fprintf(stderr, Name ": %s assembled from %d drive%s",
52826846 982 mddev, okcnt, okcnt==1?"":"s");
b8a8ccf9
NB
983 if (okcnt != info.array.raid_disks)
984 fprintf(stderr, " (out of %d)", info.array.raid_disks);
985 fprintf(stderr, ", but not started.\n");
d55e3aef 986 if (must_close) close(mdfd);
64c4757e 987 return 0;
82b27616 988 }
4855f95c 989 if (verbose >= -1) {
dab6685f
NB
990 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
991 if (sparecnt)
992 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
583315d9
NB
993 if (!enough(info.array.level, info.array.raid_disks,
994 info.array.layout, 1, avail, okcnt))
dab6685f 995 fprintf(stderr, " - not enough to start the array.\n");
583315d9
NB
996 else if (!enough(info.array.level,
997 info.array.raid_disks,
998 info.array.layout, clean,
999 avail, okcnt))
1000 fprintf(stderr, " - not enough to start the "
1001 "array while not clean - consider "
1002 "--force.\n");
dab6685f
NB
1003 else {
1004 if (req_cnt == info.array.raid_disks)
1005 fprintf(stderr, " - need all %d to start it", req_cnt);
1006 else
1007 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1008 fprintf(stderr, " (use --run to insist).\n");
1009 }
aa88f531 1010 }
d55e3aef 1011 if (must_close) close(mdfd);
64c4757e 1012 return 1;
82b27616 1013 } else {
52826846
NB
1014 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1015 * been updated to point to the current locations of devices.
1016 * so we can just start the array
82b27616 1017 */
cd29a5c8 1018 unsigned long dev;
213ee40b
NB
1019 dev = makedev(devices[chosen_drive].i.disk.major,
1020 devices[chosen_drive].i.disk.minor);
82b27616
NB
1021 if (ioctl(mdfd, START_ARRAY, dev)) {
1022 fprintf(stderr, Name ": Cannot start array: %s\n",
1023 strerror(errno));
1024 }
aba69144 1025
64c4757e 1026 }
d55e3aef 1027 if (must_close) close(mdfd);
e0d19036 1028 return 0;
64c4757e 1029}