]>
Commit | Line | Data |
---|---|---|
64c4757e | 1 | /* |
9a9dab36 | 2 | * mdadm - manage Linux "md" devices aka RAID arrays. |
64c4757e | 3 | * |
e736b623 | 4 | * Copyright (C) 2001-2009 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 | |
e736b623 | 22 | * Email: <neilb@suse.de> |
64c4757e NB |
23 | */ |
24 | ||
9a9dab36 | 25 | #include "mdadm.h" |
41a3b72a | 26 | #include <ctype.h> |
64c4757e | 27 | |
624920bb NB |
28 | static int name_matches(char *found, char *required, char *homehost) |
29 | { | |
30 | /* See if the name found matches the required name, possibly | |
31 | * prefixed with 'homehost' | |
32 | */ | |
33 | char fnd[33]; | |
34 | ||
35 | strncpy(fnd, found, 32); | |
36 | fnd[32] = 0; | |
37 | if (strcmp(found, required)==0) | |
38 | return 1; | |
39 | if (homehost) { | |
40 | int l = strlen(homehost); | |
41 | if (l < 32 && fnd[l] == ':' && | |
42 | strcmp(fnd+l+1, required)==0) | |
43 | return 1; | |
44 | } | |
45 | return 0; | |
46 | } | |
47 | ||
9008ed1c | 48 | static int is_member_busy(char *metadata_version) |
2de8884f DW |
49 | { |
50 | /* check if the given member array is active */ | |
51 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
52 | struct mdstat_ent *ent; | |
53 | int busy = 0; | |
54 | ||
55 | for (ent = mdstat; ent; ent = ent->next) { | |
56 | if (ent->metadata_version == NULL) | |
57 | continue; | |
58 | if (strncmp(ent->metadata_version, "external:", 9) != 0) | |
59 | continue; | |
60 | if (!is_subarray(&ent->metadata_version[9])) | |
61 | continue; | |
62 | /* Skip first char - it can be '/' or '-' */ | |
63 | if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) { | |
64 | busy = 1; | |
65 | break; | |
66 | } | |
67 | } | |
68 | free_mdstat(mdstat); | |
69 | ||
70 | return busy; | |
71 | } | |
72 | ||
fa56eddb | 73 | static int ident_matches(struct mddev_ident *ident, |
08fb91a3 N |
74 | struct mdinfo *content, |
75 | struct supertype *tst, | |
76 | char *homehost, | |
77 | char *update, char *devname) | |
78 | { | |
79 | ||
80 | if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) && | |
cbeeb0e5 AC |
81 | same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0 && |
82 | memcmp(content->uuid, uuid_zero, sizeof(int[4])) != 0) { | |
08fb91a3 | 83 | if (devname) |
e7b84f9d | 84 | pr_err("%s has wrong uuid.\n", devname); |
08fb91a3 N |
85 | return 0; |
86 | } | |
87 | if (ident->name[0] && (!update || strcmp(update, "name")!= 0) && | |
88 | name_matches(content->name, ident->name, homehost)==0) { | |
89 | if (devname) | |
e7b84f9d | 90 | pr_err("%s has wrong name.\n", devname); |
08fb91a3 N |
91 | return 0; |
92 | } | |
93 | if (ident->super_minor != UnSet && | |
94 | ident->super_minor != content->array.md_minor) { | |
95 | if (devname) | |
e7b84f9d N |
96 | pr_err("%s has wrong super-minor.\n", |
97 | devname); | |
08fb91a3 N |
98 | return 0; |
99 | } | |
100 | if (ident->level != UnSet && | |
101 | ident->level != content->array.level) { | |
102 | if (devname) | |
e7b84f9d N |
103 | pr_err("%s has wrong raid level.\n", |
104 | devname); | |
08fb91a3 N |
105 | return 0; |
106 | } | |
107 | if (ident->raid_disks != UnSet && | |
108 | ident->raid_disks!= content->array.raid_disks) { | |
109 | if (devname) | |
e7b84f9d N |
110 | pr_err("%s requires wrong number of drives.\n", |
111 | devname); | |
08fb91a3 N |
112 | return 0; |
113 | } | |
805d30b2 N |
114 | if (ident->member && ident->member[0]) { |
115 | /* content->text_version must match */ | |
116 | char *s = strchr(content->text_version+1, '/'); | |
117 | if (s == NULL) { | |
118 | if (devname) | |
e7b84f9d N |
119 | pr_err("%s is not a container and one is required.\n", |
120 | devname); | |
805d30b2 N |
121 | return 0; |
122 | } else if (strcmp(ident->member, s+1) != 0) { | |
123 | if (devname) | |
e7b84f9d N |
124 | pr_err("skipping wrong member %s is %s\n", |
125 | content->text_version, devname); | |
805d30b2 N |
126 | return 0; |
127 | } | |
128 | } | |
08fb91a3 N |
129 | return 1; |
130 | } | |
131 | ||
132 | ||
7f91af49 | 133 | int Assemble(struct supertype *st, char *mddev, |
fa56eddb | 134 | struct mddev_ident *ident, |
87f26d14 N |
135 | struct mddev_dev *devlist, |
136 | char *backup_file, int invalid_backup, | |
64c4757e | 137 | int readonly, int runstop, |
0ac91628 | 138 | char *update, char *homehost, int require_homehost, |
b76b30e0 | 139 | int verbose, int force, int freeze_reshape) |
64c4757e NB |
140 | { |
141 | /* | |
52826846 NB |
142 | * The task of Assemble is to find a collection of |
143 | * devices that should (according to their superblocks) | |
144 | * form an array, and to give this collection to the MD driver. | |
145 | * In Linux-2.4 and later, this involves submitting a | |
64c4757e NB |
146 | * SET_ARRAY_INFO ioctl with no arg - to prepare |
147 | * the array - and then submit a number of | |
148 | * ADD_NEW_DISK ioctls to add disks into | |
149 | * the array. Finally RUN_ARRAY might | |
150 | * be submitted to start the array. | |
151 | * | |
152 | * Much of the work of Assemble is in finding and/or | |
153 | * checking the disks to make sure they look right. | |
154 | * | |
4b1ac34b | 155 | * If mddev is not set, then scan must be set and we |
64c4757e NB |
156 | * read through the config file for dev+uuid mapping |
157 | * We recurse, setting mddev, for each device that | |
158 | * - isn't running | |
4b1ac34b | 159 | * - has a valid uuid (or any uuid if !uuidset) |
64c4757e NB |
160 | * |
161 | * If mddev is set, we try to determine state of md. | |
162 | * check version - must be at least 0.90.0 | |
163 | * check kernel version. must be at least 2.4. | |
164 | * If not, we can possibly fall back on START_ARRAY | |
165 | * Try to GET_ARRAY_INFO. | |
166 | * If possible, give up | |
167 | * If not, try to STOP_ARRAY just to make sure | |
168 | * | |
169 | * If !uuidset and scan, look in conf-file for uuid | |
170 | * If not found, give up | |
b8a8ccf9 | 171 | * If !devlist and scan and uuidset, get list of devs from conf-file |
64c4757e NB |
172 | * |
173 | * For each device: | |
174 | * Check superblock - discard if bad | |
175 | * Check uuid (set if we don't have one) - discard if no match | |
5787fa49 | 176 | * Check superblock similarity if we have a superblock - discard if different |
4b1ac34b | 177 | * Record events, devicenum |
64c4757e | 178 | * This should give us a list of devices for the array |
4b1ac34b | 179 | * We should collect the most recent event number |
64c4757e NB |
180 | * |
181 | * Count disks with recent enough event count | |
182 | * While force && !enough disks | |
183 | * Choose newest rejected disks, update event count | |
184 | * mark clean and rewrite superblock | |
185 | * If recent kernel: | |
186 | * SET_ARRAY_INFO | |
187 | * foreach device with recent events : ADD_NEW_DISK | |
188 | * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY | |
189 | * If old kernel: | |
190 | * Check the device numbers in superblock are right | |
191 | * update superblock if any changes | |
192 | * START_ARRAY | |
193 | * | |
194 | */ | |
c30e5369 N |
195 | int mdfd; |
196 | int clean; | |
f05641cf N |
197 | int auto_assem = (mddev == NULL && !ident->uuid_set && |
198 | ident->super_minor == UnSet && ident->name[0] == 0 | |
aa7c284c | 199 | && (ident->container == NULL || ident->member == NULL)); |
64c4757e | 200 | int old_linux = 0; |
c30e5369 | 201 | int vers = vers; /* Keep gcc quite - it really is initialised */ |
64c4757e NB |
202 | struct { |
203 | char *devname; | |
213ee40b NB |
204 | int uptodate; /* set once we decide that this device is as |
205 | * recent as everything else in the array. | |
206 | */ | |
207 | struct mdinfo i; | |
5787fa49 | 208 | } *devices; |
02e7c5b7 | 209 | char *devmap; |
56eedc1a | 210 | int *best = NULL; /* indexed by raid_disk */ |
f21e18ca | 211 | int bestcnt = 0; |
98c6faba | 212 | int devcnt = 0; |
1ff98339 | 213 | unsigned int okcnt, sparecnt, rebuilding_cnt; |
98c6faba | 214 | unsigned int req_cnt; |
f21e18ca | 215 | int i; |
64c4757e | 216 | int most_recent = 0; |
cd29a5c8 | 217 | int chosen_drive; |
52826846 | 218 | int change = 0; |
cd29a5c8 | 219 | int inargv = 0; |
52437b4f | 220 | int report_missmatch; |
a6482415 | 221 | #ifndef MDASSEMBLE |
bf4fb153 | 222 | int bitmap_done; |
a6482415 | 223 | #endif |
7f91af49 N |
224 | int start_partial_ok = (runstop >= 0) && |
225 | (force || devlist==NULL || auto_assem); | |
98c6faba | 226 | unsigned int num_devs; |
a655e550 | 227 | struct mddev_dev *tmpdev; |
4b1ac34b | 228 | struct mdinfo info; |
98dbd966 | 229 | struct mdinfo *content = NULL; |
265e0f17 | 230 | char *avail; |
308e1801 | 231 | int nextspare = 0; |
197e3eb6 | 232 | char *name = NULL; |
69207ff6 | 233 | int trustworthy; |
a04d5763 | 234 | char chosen_name[1024]; |
cbeeb0e5 | 235 | struct domainlist *domains = NULL; |
8a46fe84 | 236 | |
682c7051 | 237 | if (get_linux_version() < 2004000) |
64c4757e NB |
238 | old_linux = 1; |
239 | ||
64c4757e | 240 | /* |
52826846 NB |
241 | * If any subdevs are listed, then any that don't |
242 | * match ident are discarded. Remainder must all match and | |
243 | * become the array. | |
244 | * If no subdevs, then we scan all devices in the config file, but | |
245 | * there must be something in the identity | |
64c4757e | 246 | */ |
64c4757e | 247 | |
cd29a5c8 | 248 | if (!devlist && |
52826846 | 249 | ident->uuid_set == 0 && |
f21e18ca | 250 | (ident->super_minor < 0 || ident->super_minor == UnSet) && |
e0fe762a N |
251 | ident->name[0] == 0 && |
252 | (ident->container == NULL || ident->member == NULL) && | |
52826846 | 253 | ident->devices == NULL) { |
e7b84f9d N |
254 | pr_err("No identity information available for %s - cannot assemble.\n", |
255 | mddev ? mddev : "further assembly"); | |
52826846 | 256 | return 1; |
64c4757e | 257 | } |
71d60c48 | 258 | |
cd29a5c8 | 259 | if (devlist == NULL) |
8aec876d | 260 | devlist = conf_get_devs(); |
7f91af49 | 261 | else if (mddev) |
da6b5ca9 | 262 | inargv = 1; |
64c4757e | 263 | |
52437b4f | 264 | report_missmatch = ((inargv && verbose >= 0) || verbose > 0); |
60e1bc1a | 265 | try_again: |
c30e5369 N |
266 | /* We come back here when doing auto-assembly and attempting some |
267 | * set of devices failed. Those are now marked as ->used==2 and | |
268 | * we ignore them and try again | |
269 | */ | |
60e1bc1a | 270 | |
5787fa49 NB |
271 | tmpdev = devlist; num_devs = 0; |
272 | while (tmpdev) { | |
da6b5ca9 NB |
273 | if (tmpdev->used) |
274 | tmpdev->used = 2; | |
275 | else | |
276 | num_devs++; | |
5787fa49 NB |
277 | tmpdev = tmpdev->next; |
278 | } | |
5787fa49 | 279 | |
82d9eba6 | 280 | if (!st && ident->st) st = ident->st; |
64c4757e | 281 | |
dab6685f | 282 | if (verbose>0) |
e7b84f9d N |
283 | pr_err("looking for devices for %s\n", |
284 | mddev ? mddev : "further assembly"); | |
82b27616 | 285 | |
811e6cbe NB |
286 | /* first walk the list of devices to find a consistent set |
287 | * that match the criterea, if that is possible. | |
c30e5369 | 288 | * We flag the ones we like with 'used'. |
811e6cbe NB |
289 | */ |
290 | for (tmpdev = devlist; | |
291 | tmpdev; | |
5083d66b | 292 | tmpdev = tmpdev ? tmpdev->next : NULL) { |
811e6cbe | 293 | char *devname = tmpdev->devname; |
64c4757e NB |
294 | int dfd; |
295 | struct stat stb; | |
518a60f3 | 296 | struct supertype *tst; |
4e8d9f0a | 297 | struct dev_policy *pol = NULL; |
5083d66b | 298 | int found_container = 0; |
52826846 | 299 | |
da6b5ca9 NB |
300 | if (tmpdev->used > 1) continue; |
301 | ||
52826846 | 302 | if (ident->devices && |
56eedc1a | 303 | !match_oneof(ident->devices, devname)) { |
52437b4f | 304 | if (report_missmatch) |
e7b84f9d | 305 | pr_err("%s is not one of %s\n", devname, ident->devices); |
52826846 | 306 | continue; |
56eedc1a | 307 | } |
4b1ac34b | 308 | |
518a60f3 JS |
309 | tst = dup_super(st); |
310 | ||
56d18859 | 311 | dfd = dev_open(devname, O_RDONLY); |
64c4757e | 312 | if (dfd < 0) { |
52437b4f | 313 | if (report_missmatch) |
e7b84f9d N |
314 | pr_err("cannot open device %s: %s\n", |
315 | devname, strerror(errno)); | |
da6b5ca9 | 316 | tmpdev->used = 2; |
52826846 NB |
317 | } else if (fstat(dfd, &stb)< 0) { |
318 | /* Impossible! */ | |
e7b84f9d N |
319 | pr_err("fstat failed for %s: %s\n", |
320 | devname, strerror(errno)); | |
da6b5ca9 | 321 | tmpdev->used = 2; |
cd29a5c8 | 322 | } else if ((stb.st_mode & S_IFMT) != S_IFBLK) { |
e7b84f9d N |
323 | pr_err("%s is not a block device.\n", |
324 | devname); | |
da6b5ca9 | 325 | tmpdev->used = 2; |
5083d66b N |
326 | } else if (must_be_container(dfd)) { |
327 | if (st) { | |
328 | /* already found some components, this cannot | |
329 | * be another one. | |
330 | */ | |
331 | if (report_missmatch) | |
e7b84f9d N |
332 | pr_err("%s is a container, but we are looking for components\n", |
333 | devname); | |
5083d66b | 334 | tmpdev->used = 2; |
71204a50 | 335 | #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) |
5083d66b N |
336 | } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) { |
337 | if (report_missmatch) | |
e7b84f9d N |
338 | pr_err("not a recognisable container: %s\n", |
339 | devname); | |
5083d66b | 340 | tmpdev->used = 2; |
71204a50 | 341 | #endif |
417f346e HCP |
342 | } else if (!tst->ss->load_container |
343 | || tst->ss->load_container(tst, dfd, NULL)) { | |
5083d66b | 344 | if (report_missmatch) |
e7b84f9d N |
345 | pr_err("no correct container type: %s\n", |
346 | devname); | |
5083d66b N |
347 | tmpdev->used = 2; |
348 | } else if (auto_assem && | |
349 | !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)), | |
350 | tst->ss->match_home(tst, homehost) == 1)) { | |
351 | if (report_missmatch) | |
e7b84f9d N |
352 | pr_err("%s has metadata type %s for which " |
353 | "auto-assembly is disabled\n", | |
354 | devname, tst->ss->name); | |
5083d66b N |
355 | tmpdev->used = 2; |
356 | } else | |
357 | found_container = 1; | |
52826846 | 358 | } else { |
5083d66b N |
359 | if (!tst && (tst = guess_super(dfd)) == NULL) { |
360 | if (report_missmatch) | |
e7b84f9d N |
361 | pr_err("no recogniseable superblock on %s\n", |
362 | devname); | |
5083d66b N |
363 | tmpdev->used = 2; |
364 | } else if (tst->ss->load_super(tst,dfd, NULL)) { | |
365 | if (report_missmatch) | |
e7b84f9d N |
366 | pr_err("no RAID superblock on %s\n", |
367 | devname); | |
5083d66b N |
368 | tmpdev->used = 2; |
369 | } else if (tst->ss->compare_super == NULL) { | |
370 | if (report_missmatch) | |
e7b84f9d N |
371 | pr_err("Cannot assemble %s metadata on %s\n", |
372 | tst->ss->name, devname); | |
5083d66b N |
373 | tmpdev->used = 2; |
374 | } else if (auto_assem && st == NULL && | |
375 | !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)), | |
376 | tst->ss->match_home(tst, homehost) == 1)) { | |
377 | if (report_missmatch) | |
e7b84f9d N |
378 | pr_err("%s has metadata type %s for which " |
379 | "auto-assembly is disabled\n", | |
380 | devname, tst->ss->name); | |
5083d66b N |
381 | tmpdev->used = 2; |
382 | } | |
64c4757e | 383 | } |
c06487ce | 384 | if (dfd >= 0) close(dfd); |
d68ea4d7 | 385 | if (tmpdev->used == 2) { |
d4386799 | 386 | if (auto_assem || !inargv) |
d68ea4d7 N |
387 | /* Ignore unrecognised devices during auto-assembly */ |
388 | goto loop; | |
389 | if (ident->uuid_set || ident->name[0] || | |
390 | ident->super_minor != UnSet) | |
391 | /* Ignore unrecognised device if looking for | |
392 | * specific array */ | |
393 | goto loop; | |
394 | ||
395 | ||
e7b84f9d | 396 | pr_err("%s has no superblock - assembly aborted\n", |
d68ea4d7 N |
397 | devname); |
398 | if (st) | |
399 | st->ss->free_super(st); | |
400 | dev_policy_free(pol); | |
cbeeb0e5 | 401 | domain_free(domains); |
d68ea4d7 N |
402 | return 1; |
403 | } | |
52826846 | 404 | |
5083d66b | 405 | if (found_container) { |
9008ed1c N |
406 | /* tmpdev is a container. We need to be either |
407 | * looking for a member, or auto-assembling | |
408 | */ | |
56d18859 N |
409 | /* should be safe to try an exclusive open now, we |
410 | * have rejected anything that some other mdadm might | |
411 | * be looking at | |
412 | */ | |
413 | dfd = dev_open(devname, O_RDONLY | O_EXCL); | |
414 | if (dfd < 0) { | |
415 | if (report_missmatch) | |
e7b84f9d | 416 | pr_err("%s is busy - skipping\n", devname); |
56d18859 N |
417 | goto loop; |
418 | } | |
419 | close(dfd); | |
9008ed1c N |
420 | |
421 | if (ident->container) { | |
422 | if (ident->container[0] == '/' && | |
423 | !same_dev(ident->container, devname)) { | |
424 | if (report_missmatch) | |
e7b84f9d | 425 | pr_err("%s is not the container required (%s)\n", |
9008ed1c N |
426 | devname, ident->container); |
427 | goto loop; | |
428 | } | |
429 | if (ident->container[0] != '/') { | |
430 | /* we have a uuid */ | |
431 | int uuid[4]; | |
87477e6d N |
432 | |
433 | content = &info; | |
87477e6d N |
434 | tst->ss->getinfo_super(tst, content, NULL); |
435 | ||
9008ed1c N |
436 | if (!parse_uuid(ident->container, uuid) || |
437 | !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) { | |
438 | if (report_missmatch) | |
e7b84f9d | 439 | pr_err("%s has wrong UUID to be required container\n", |
9008ed1c N |
440 | devname); |
441 | goto loop; | |
442 | } | |
443 | } | |
444 | } | |
445 | /* It is worth looking inside this container. | |
446 | */ | |
7636b5a8 | 447 | if (verbose > 0) |
e7b84f9d | 448 | pr_err("looking in container %s\n", |
7636b5a8 | 449 | devname); |
1415fe4b | 450 | |
88cef9b3 N |
451 | for (content = tst->ss->container_content(tst, NULL); |
452 | content; | |
453 | content = content->next) { | |
1415fe4b | 454 | |
88cef9b3 N |
455 | if (!ident_matches(ident, content, tst, |
456 | homehost, update, | |
457 | report_missmatch ? devname : NULL)) | |
458 | /* message already printed */; | |
459 | else if (is_member_busy(content->text_version)) { | |
460 | if (report_missmatch) | |
e7b84f9d | 461 | pr_err("member %s in %s is already assembled\n", |
88cef9b3 N |
462 | content->text_version, |
463 | devname); | |
81219e70 LM |
464 | } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) { |
465 | /* do not assemble arrays with unsupported configurations */ | |
e7b84f9d | 466 | pr_err("Cannot activate member %s in %s.\n", |
81219e70 LM |
467 | content->text_version, |
468 | devname); | |
88cef9b3 N |
469 | } else |
470 | break; | |
471 | } | |
472 | if (!content) { | |
f7ad3ccc | 473 | tmpdev->used = 2; |
88cef9b3 | 474 | goto loop; /* empty container */ |
fa031239 | 475 | } |
88cef9b3 | 476 | |
9008ed1c | 477 | st = tst; tst = NULL; |
4c1c3ad8 | 478 | if (!auto_assem && inargv && tmpdev->next != NULL) { |
e7b84f9d | 479 | pr_err("%s is a container, but is not " |
9008ed1c N |
480 | "only device given: confused and aborting\n", |
481 | devname); | |
482 | st->ss->free_super(st); | |
4e8d9f0a | 483 | dev_policy_free(pol); |
cbeeb0e5 | 484 | domain_free(domains); |
9008ed1c N |
485 | return 1; |
486 | } | |
7636b5a8 | 487 | if (verbose > 0) |
e7b84f9d | 488 | pr_err("found match on member %s in %s\n", |
7636b5a8 | 489 | content->text_version, devname); |
bac0d92e | 490 | |
5083d66b N |
491 | /* make sure we finished the loop */ |
492 | tmpdev = NULL; | |
bac0d92e | 493 | goto loop; |
5083d66b | 494 | } else { |
bac0d92e | 495 | |
5083d66b | 496 | content = &info; |
5083d66b N |
497 | tst->ss->getinfo_super(tst, content, NULL); |
498 | ||
499 | if (!ident_matches(ident, content, tst, | |
500 | homehost, update, | |
501 | report_missmatch ? devname : NULL)) | |
df37ffc0 | 502 | goto loop; |
cbeeb0e5 | 503 | |
56d18859 N |
504 | /* should be safe to try an exclusive open now, we |
505 | * have rejected anything that some other mdadm might | |
506 | * be looking at | |
507 | */ | |
508 | dfd = dev_open(devname, O_RDONLY | O_EXCL); | |
509 | if (dfd < 0) { | |
510 | if (report_missmatch) | |
e7b84f9d | 511 | pr_err("%s is busy - skipping\n", devname); |
56d18859 N |
512 | goto loop; |
513 | } | |
514 | close(dfd); | |
515 | ||
5083d66b N |
516 | if (st == NULL) |
517 | st = dup_super(tst); | |
518 | if (st->minor_version == -1) | |
519 | st->minor_version = tst->minor_version; | |
ed7fc6b4 AC |
520 | |
521 | if (memcmp(content->uuid, uuid_zero, | |
522 | sizeof(int[4])) == 0) { | |
523 | /* this is a floating spare. It cannot define | |
524 | * an array unless there are no more arrays of | |
525 | * this type to be found. It can be included | |
526 | * in an array of this type though. | |
527 | */ | |
528 | tmpdev->used = 3; | |
529 | goto loop; | |
530 | } | |
531 | ||
5083d66b N |
532 | if (st->ss != tst->ss || |
533 | st->minor_version != tst->minor_version || | |
534 | st->ss->compare_super(st, tst) != 0) { | |
535 | /* Some mismatch. If exactly one array matches this host, | |
536 | * we can resolve on that one. | |
537 | * Or, if we are auto assembling, we just ignore the second | |
538 | * for now. | |
539 | */ | |
540 | if (auto_assem) | |
541 | goto loop; | |
542 | if (homehost) { | |
543 | int first = st->ss->match_home(st, homehost); | |
544 | int last = tst->ss->match_home(tst, homehost); | |
545 | if (first != last && | |
546 | (first == 1 || last == 1)) { | |
547 | /* We can do something */ | |
548 | if (first) {/* just ignore this one */ | |
549 | if (report_missmatch) | |
e7b84f9d | 550 | pr_err("%s misses out due to wrong homehost\n", |
5083d66b N |
551 | devname); |
552 | goto loop; | |
553 | } else { /* reject all those sofar */ | |
554 | struct mddev_dev *td; | |
555 | if (report_missmatch) | |
e7b84f9d | 556 | pr_err("%s overrides previous devices due to good homehost\n", |
5083d66b N |
557 | devname); |
558 | for (td=devlist; td != tmpdev; td=td->next) | |
559 | if (td->used == 1) | |
560 | td->used = 0; | |
561 | tmpdev->used = 1; | |
562 | goto loop; | |
563 | } | |
83b6208e NB |
564 | } |
565 | } | |
e7b84f9d | 566 | pr_err("superblock on %s doesn't match others - assembly aborted\n", |
5083d66b N |
567 | devname); |
568 | tst->ss->free_super(tst); | |
569 | st->ss->free_super(st); | |
570 | dev_policy_free(pol); | |
cbeeb0e5 | 571 | domain_free(domains); |
5083d66b | 572 | return 1; |
83b6208e | 573 | } |
5083d66b | 574 | tmpdev->used = 1; |
64c4757e | 575 | } |
df37ffc0 | 576 | loop: |
cbeeb0e5 | 577 | /* Collect domain information from members only */ |
26b05aea AC |
578 | if (tmpdev && tmpdev->used == 1) { |
579 | if (!pol) | |
580 | pol = devnum_policy(stb.st_rdev); | |
cbeeb0e5 | 581 | domain_merge(&domains, pol, tst?tst->ss->name:NULL); |
26b05aea | 582 | } |
4e8d9f0a N |
583 | dev_policy_free(pol); |
584 | pol = NULL; | |
3d2b16e7 NB |
585 | if (tst) |
586 | tst->ss->free_super(tst); | |
811e6cbe NB |
587 | } |
588 | ||
ed7fc6b4 | 589 | /* Check if we found some imsm spares but no members */ |
3c7b4a25 CA |
590 | if ((auto_assem || |
591 | (ident->uuid_set && | |
592 | memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) && | |
593 | (!st || !st->sb)) | |
ed7fc6b4 AC |
594 | for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) { |
595 | if (tmpdev->used != 3) | |
596 | continue; | |
597 | tmpdev->used = 1; | |
598 | content = &info; | |
599 | ||
600 | if (!st->sb) { | |
601 | /* we need sb from one of the spares */ | |
602 | int dfd = dev_open(tmpdev->devname, O_RDONLY); | |
603 | if (dfd < 0 || | |
604 | st->ss->load_super(st, dfd, NULL)) | |
605 | tmpdev->used = 2; | |
606 | if (dfd > 0) | |
607 | close(dfd); | |
608 | } | |
609 | } | |
610 | ||
cbeeb0e5 AC |
611 | /* Now reject spares that don't match domains of identified members */ |
612 | for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) { | |
613 | struct stat stb; | |
614 | if (tmpdev->used != 3) | |
615 | continue; | |
616 | if (stat(tmpdev->devname, &stb)< 0) { | |
e7b84f9d | 617 | pr_err("fstat failed for %s: %s\n", |
cbeeb0e5 AC |
618 | tmpdev->devname, strerror(errno)); |
619 | tmpdev->used = 2; | |
620 | } else { | |
a5d10dce N |
621 | struct dev_policy *pol = devnum_policy(stb.st_rdev); |
622 | int dt = domain_test(domains, pol, NULL); | |
623 | if (inargv && dt != 0) | |
624 | /* take this spare as domains match | |
625 | * if there are any */ | |
626 | tmpdev->used = 1; | |
627 | else if (!inargv && dt == 1) | |
628 | /* device wasn't explicitly listed, so need | |
629 | * explicit domain match - which we have */ | |
cbeeb0e5 AC |
630 | tmpdev->used = 1; |
631 | else | |
632 | /* if domains don't match mark as unused */ | |
633 | tmpdev->used = 0; | |
634 | dev_policy_free(pol); | |
635 | } | |
636 | } | |
637 | domain_free(domains); | |
638 | ||
98dbd966 | 639 | if (!st || !st->sb || !content) |
c30e5369 N |
640 | return 2; |
641 | ||
05833051 | 642 | /* Now need to open the array device. Use create_mddev */ |
98dbd966 | 643 | if (content == &info) |
a5d85af7 | 644 | st->ss->getinfo_super(st, content, NULL); |
3d2c4fc7 | 645 | |
69207ff6 | 646 | trustworthy = FOREIGN; |
05833051 | 647 | name = content->name; |
745f72f6 N |
648 | switch (st->ss->match_home(st, homehost) |
649 | ?: st->ss->match_home(st, "any")) { | |
69207ff6 N |
650 | case 1: |
651 | trustworthy = LOCAL; | |
98dbd966 | 652 | name = strchr(content->name, ':'); |
69207ff6 N |
653 | if (name) |
654 | name++; | |
3d2c4fc7 | 655 | else |
98dbd966 | 656 | name = content->name; |
69207ff6 | 657 | break; |
c30e5369 | 658 | } |
05833051 | 659 | if (!auto_assem) |
aa7c284c | 660 | /* If the array is listed in mdadm.conf or on |
ac2ecf55 N |
661 | * command line, then we trust the name |
662 | * even if the array doesn't look local | |
663 | */ | |
664 | trustworthy = LOCAL; | |
665 | ||
05833051 | 666 | if (name[0] == 0 && |
98dbd966 DW |
667 | content->array.level == LEVEL_CONTAINER) { |
668 | name = content->text_version; | |
a4bc1720 N |
669 | trustworthy = METADATA; |
670 | } | |
0ac91628 N |
671 | |
672 | if (name[0] && trustworthy != LOCAL && | |
673 | ! require_homehost && | |
674 | conf_name_is_free(name)) | |
675 | trustworthy = LOCAL; | |
676 | ||
7cdc0872 N |
677 | if (trustworthy == LOCAL && |
678 | strchr(name, ':')) | |
679 | /* Ignore 'host:' prefix of name */ | |
680 | name = strchr(name, ':')+1; | |
681 | ||
a04d5763 N |
682 | mdfd = create_mddev(mddev, name, ident->autof, trustworthy, |
683 | chosen_name); | |
c30e5369 N |
684 | if (mdfd < 0) { |
685 | st->ss->free_super(st); | |
c30e5369 | 686 | if (auto_assem) |
60e1bc1a | 687 | goto try_again; |
c30e5369 N |
688 | return 1; |
689 | } | |
a04d5763 | 690 | mddev = chosen_name; |
c30e5369 N |
691 | vers = md_get_version(mdfd); |
692 | if (vers < 9000) { | |
e7b84f9d | 693 | pr_err("Assemble requires driver version 0.90.0 or later.\n" |
c30e5369 N |
694 | " Upgrade your kernel or try --build\n"); |
695 | close(mdfd); | |
696 | return 1; | |
697 | } | |
66afdfa9 | 698 | if (mddev_busy(fd2devnum(mdfd))) { |
e7b84f9d | 699 | pr_err("%s already active, cannot restart it!\n", |
c30e5369 N |
700 | mddev); |
701 | for (tmpdev = devlist ; | |
702 | tmpdev && tmpdev->used != 1; | |
703 | tmpdev = tmpdev->next) | |
704 | ; | |
705 | if (tmpdev && auto_assem) | |
e7b84f9d | 706 | pr_err("%s needed for %s...\n", |
c30e5369 N |
707 | mddev, tmpdev->devname); |
708 | close(mdfd); | |
709 | mdfd = -3; | |
710 | st->ss->free_super(st); | |
c30e5369 N |
711 | if (auto_assem) |
712 | goto try_again; | |
713 | return 1; | |
8a46fe84 | 714 | } |
c30e5369 | 715 | ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */ |
8a46fe84 | 716 | |
9008ed1c N |
717 | #ifndef MDASSEMBLE |
718 | if (content != &info) { | |
719 | /* This is a member of a container. Try starting the array. */ | |
588bebfc AK |
720 | int err; |
721 | err = assemble_container_content(st, mdfd, content, runstop, | |
49680258 | 722 | chosen_name, verbose, |
b76b30e0 | 723 | backup_file, freeze_reshape); |
588bebfc | 724 | close(mdfd); |
588bebfc | 725 | return err; |
9008ed1c | 726 | } |
a6482415 | 727 | bitmap_done = 0; |
9008ed1c | 728 | #endif |
811e6cbe | 729 | /* Ok, no bad inconsistancy, we can try updating etc */ |
503975b9 N |
730 | devices = xcalloc(num_devs, sizeof(*devices)); |
731 | devmap = xcalloc(num_devs, content->array.raid_disks); | |
da6b5ca9 | 732 | for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) { |
811e6cbe NB |
733 | char *devname = tmpdev->devname; |
734 | struct stat stb; | |
5787fa49 | 735 | /* looks like a good enough match to update the super block if needed */ |
d1f1011b | 736 | #ifndef MDASSEMBLE |
5787fa49 | 737 | if (update) { |
811e6cbe | 738 | int dfd; |
4b1ac34b NB |
739 | /* prepare useful information in info structures */ |
740 | struct stat stb2; | |
3da92f27 | 741 | struct supertype *tst; |
1e2b2765 | 742 | int err; |
4b1ac34b | 743 | fstat(mdfd, &stb2); |
7d99579f NB |
744 | |
745 | if (strcmp(update, "uuid")==0 && | |
746 | !ident->uuid_set) { | |
747 | int rfd; | |
748 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
749 | read(rfd, ident->uuid, 16) != 16) { | |
750 | *(__u32*)(ident->uuid) = random(); | |
751 | *(__u32*)(ident->uuid+1) = random(); | |
752 | *(__u32*)(ident->uuid+2) = random(); | |
753 | *(__u32*)(ident->uuid+3) = random(); | |
754 | } | |
755 | if (rfd >= 0) close(rfd); | |
7d99579f | 756 | } |
811e6cbe NB |
757 | dfd = dev_open(devname, O_RDWR|O_EXCL); |
758 | ||
3da92f27 | 759 | tst = dup_super(st); |
9f22b13f | 760 | if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) { |
e7b84f9d | 761 | pr_err("cannot re-read metadata from %s - aborting\n", |
9f22b13f N |
762 | devname); |
763 | if (dfd >= 0) | |
764 | close(dfd); | |
765 | close(mdfd); | |
d7f7ebb7 | 766 | free(devices); |
02e7c5b7 | 767 | free(devmap); |
9f22b13f N |
768 | return 1; |
769 | } | |
02e7c5b7 | 770 | tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks); |
811e6cbe | 771 | |
98dbd966 DW |
772 | memcpy(content->uuid, ident->uuid, 16); |
773 | strcpy(content->name, ident->name); | |
774 | content->array.md_minor = minor(stb2.st_rdev); | |
811e6cbe | 775 | |
1e2b2765 N |
776 | if (strcmp(update, "byteorder") == 0) |
777 | err = 0; | |
778 | else | |
779 | err = tst->ss->update_super(tst, content, update, | |
780 | devname, verbose, | |
781 | ident->uuid_set, | |
782 | homehost); | |
783 | if (err < 0) { | |
e7b84f9d N |
784 | pr_err("--update=%s not understood" |
785 | " for %s metadata\n", | |
786 | update, tst->ss->name); | |
1e2b2765 N |
787 | tst->ss->free_super(tst); |
788 | free(tst); | |
789 | close(mdfd); | |
790 | close(dfd); | |
d7f7ebb7 | 791 | free(devices); |
02e7c5b7 | 792 | free(devmap); |
1e2b2765 N |
793 | return 1; |
794 | } | |
e5eac01f NB |
795 | if (strcmp(update, "uuid")==0 && |
796 | !ident->uuid_set) { | |
797 | ident->uuid_set = 1; | |
98dbd966 | 798 | memcpy(ident->uuid, content->uuid, 16); |
e5eac01f | 799 | } |
1e2b2765 | 800 | if (tst->ss->store_super(tst, dfd)) |
e7b84f9d | 801 | pr_err("Could not re-write superblock on %s.\n", |
5787fa49 | 802 | devname); |
1e2b2765 | 803 | close(dfd); |
8131b493 NB |
804 | |
805 | if (strcmp(update, "uuid")==0 && | |
bf4fb153 | 806 | ident->bitmap_fd >= 0 && !bitmap_done) { |
3da92f27 | 807 | if (bitmap_update_uuid(ident->bitmap_fd, |
98dbd966 | 808 | content->uuid, |
3da92f27 | 809 | tst->ss->swapuuid) != 0) |
e7b84f9d | 810 | pr_err("Could not update uuid on external bitmap.\n"); |
bf4fb153 NB |
811 | else |
812 | bitmap_done = 1; | |
813 | } | |
3da92f27 | 814 | tst->ss->free_super(tst); |
d1f1011b NB |
815 | } else |
816 | #endif | |
817 | { | |
30e1b9a5 | 818 | struct supertype *tst = dup_super(st); |
da6b5ca9 NB |
819 | int dfd; |
820 | dfd = dev_open(devname, O_RDWR|O_EXCL); | |
821 | ||
9f22b13f | 822 | if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) { |
e7b84f9d | 823 | pr_err("cannot re-read metadata from %s - aborting\n", |
9f22b13f N |
824 | devname); |
825 | if (dfd >= 0) | |
826 | close(dfd); | |
827 | close(mdfd); | |
d7f7ebb7 | 828 | free(devices); |
02e7c5b7 | 829 | free(devmap); |
9f22b13f N |
830 | return 1; |
831 | } | |
02e7c5b7 | 832 | tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks); |
3da92f27 | 833 | tst->ss->free_super(tst); |
da6b5ca9 | 834 | close(dfd); |
5787fa49 NB |
835 | } |
836 | ||
811e6cbe NB |
837 | stat(devname, &stb); |
838 | ||
dab6685f | 839 | if (verbose > 0) |
e7b84f9d | 840 | pr_err("%s is identified as a member of %s, slot %d.\n", |
98dbd966 | 841 | devname, mddev, content->disk.raid_disk); |
64c4757e | 842 | devices[devcnt].devname = devname; |
64c4757e | 843 | devices[devcnt].uptodate = 0; |
98dbd966 | 844 | devices[devcnt].i = *content; |
213ee40b NB |
845 | devices[devcnt].i.disk.major = major(stb.st_rdev); |
846 | devices[devcnt].i.disk.minor = minor(stb.st_rdev); | |
64c4757e | 847 | if (most_recent < devcnt) { |
213ee40b NB |
848 | if (devices[devcnt].i.events |
849 | > devices[most_recent].i.events) | |
64c4757e NB |
850 | most_recent = devcnt; |
851 | } | |
df0d4ea0 | 852 | if (content->array.level == LEVEL_MULTIPATH) |
5787fa49 NB |
853 | /* with multipath, the raid_disk from the superblock is meaningless */ |
854 | i = devcnt; | |
855 | else | |
213ee40b | 856 | i = devices[devcnt].i.disk.raid_disk; |
308e1801 | 857 | if (i+1 == 0) { |
98dbd966 DW |
858 | if (nextspare < content->array.raid_disks) |
859 | nextspare = content->array.raid_disks; | |
308e1801 | 860 | i = nextspare++; |
08110d41 | 861 | } else { |
98dbd966 | 862 | if (i >= content->array.raid_disks && |
08110d41 NB |
863 | i >= nextspare) |
864 | nextspare = i+1; | |
308e1801 | 865 | } |
98c6faba | 866 | if (i < 10000) { |
56eedc1a | 867 | if (i >= bestcnt) { |
f21e18ca | 868 | int newbestcnt = i+10; |
503975b9 | 869 | int *newbest = xmalloc(sizeof(int)*newbestcnt); |
f21e18ca | 870 | int c; |
56eedc1a NB |
871 | for (c=0; c < newbestcnt; c++) |
872 | if (c < bestcnt) | |
873 | newbest[c] = best[c]; | |
874 | else | |
875 | newbest[c] = -1; | |
876 | if (best)free(best); | |
877 | best = newbest; | |
878 | bestcnt = newbestcnt; | |
879 | } | |
6a255b69 | 880 | if (best[i] >=0 && |
213ee40b NB |
881 | devices[best[i]].i.events |
882 | == devices[devcnt].i.events | |
883 | && (devices[best[i]].i.disk.minor | |
884 | != devices[devcnt].i.disk.minor) | |
b8ac1967 | 885 | && st->ss == &super0 |
98dbd966 | 886 | && content->array.level != LEVEL_MULTIPATH) { |
6a255b69 NB |
887 | /* two different devices with identical superblock. |
888 | * Could be a mis-detection caused by overlapping | |
889 | * partitions. fail-safe. | |
890 | */ | |
e7b84f9d | 891 | pr_err("WARNING %s and %s appear" |
6a255b69 NB |
892 | " to have very similar superblocks.\n" |
893 | " If they are really different, " | |
894 | "please --zero the superblock on one\n" | |
d2df86e0 NB |
895 | " If they are the same or overlap," |
896 | " please remove one from %s.\n", | |
897 | devices[best[i]].devname, devname, | |
898 | inargv ? "the list" : | |
899 | "the\n DEVICE list in mdadm.conf" | |
900 | ); | |
7f91af49 | 901 | close(mdfd); |
d7f7ebb7 | 902 | free(devices); |
02e7c5b7 | 903 | free(devmap); |
6a255b69 NB |
904 | return 1; |
905 | } | |
52826846 | 906 | if (best[i] == -1 |
213ee40b NB |
907 | || (devices[best[i]].i.events |
908 | < devices[devcnt].i.events)) | |
52826846 | 909 | best[i] = devcnt; |
56eedc1a | 910 | } |
64c4757e | 911 | devcnt++; |
df37ffc0 | 912 | } |
4b1ac34b | 913 | |
64c4757e | 914 | if (devcnt == 0) { |
e7b84f9d | 915 | pr_err("no devices found for %s\n", |
64c4757e | 916 | mddev); |
3d2b16e7 NB |
917 | if (st) |
918 | st->ss->free_super(st); | |
7f91af49 | 919 | close(mdfd); |
d7f7ebb7 | 920 | free(devices); |
02e7c5b7 | 921 | free(devmap); |
64c4757e NB |
922 | return 1; |
923 | } | |
4b1ac34b | 924 | |
3d2b16e7 NB |
925 | if (update && strcmp(update, "byteorder")==0) |
926 | st->minor_version = 90; | |
927 | ||
a5d85af7 | 928 | st->ss->getinfo_super(st, content, NULL); |
98dbd966 | 929 | clean = content->array.state & 1; |
4b1ac34b | 930 | |
64c4757e NB |
931 | /* now we have some devices that might be suitable. |
932 | * I wonder how many | |
933 | */ | |
503975b9 | 934 | avail = xcalloc(content->array.raid_disks, 1); |
64c4757e | 935 | okcnt = 0; |
52826846 | 936 | sparecnt=0; |
1ff98339 | 937 | rebuilding_cnt=0; |
f21e18ca | 938 | for (i=0; i< bestcnt; i++) { |
64c4757e | 939 | int j = best[i]; |
ee04451c NB |
940 | int event_margin = 1; /* always allow a difference of '1' |
941 | * like the kernel does | |
942 | */ | |
64c4757e | 943 | if (j < 0) continue; |
d013a55e NB |
944 | /* note: we ignore error flags in multipath arrays |
945 | * as they don't make sense | |
946 | */ | |
df0d4ea0 | 947 | if (content->array.level != LEVEL_MULTIPATH) |
f22385f9 | 948 | if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) { |
213ee40b | 949 | if (!(devices[j].i.disk.state |
ed7fc6b4 AC |
950 | & (1<<MD_DISK_FAULTY))) { |
951 | devices[j].uptodate = 1; | |
aa88f531 | 952 | sparecnt++; |
ed7fc6b4 | 953 | } |
d013a55e | 954 | continue; |
aa88f531 | 955 | } |
de5a472e | 956 | /* If this device thinks that 'most_recent' has failed, then |
02e7c5b7 N |
957 | * we must reject this device. |
958 | */ | |
959 | if (j != most_recent && | |
960 | content->array.raid_disks > 0 && | |
961 | devices[most_recent].i.disk.raid_disk >= 0 && | |
962 | devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) { | |
963 | if (verbose > -1) | |
e7b84f9d | 964 | pr_err("ignoring %s as it reports %s as failed\n", |
02e7c5b7 N |
965 | devices[j].devname, devices[most_recent].devname); |
966 | best[i] = -1; | |
967 | continue; | |
968 | } | |
213ee40b NB |
969 | if (devices[j].i.events+event_margin >= |
970 | devices[most_recent].i.events) { | |
64c4757e | 971 | devices[j].uptodate = 1; |
1ff98339 | 972 | if (i < content->array.raid_disks) { |
dcc4210f DW |
973 | if (devices[j].i.recovery_start == MaxSector || |
974 | (content->reshape_active && | |
b720636a N |
975 | ((i >= content->array.raid_disks - content->delta_disks) || |
976 | (i >= content->array.raid_disks - content->delta_disks - 1 | |
977 | && content->array.level == 4)))) { | |
1ff98339 N |
978 | okcnt++; |
979 | avail[i]=1; | |
980 | } else | |
981 | rebuilding_cnt++; | |
265e0f17 | 982 | } else |
52826846 | 983 | sparecnt++; |
64c4757e NB |
984 | } |
985 | } | |
02e7c5b7 | 986 | free(devmap); |
da8fe5aa N |
987 | while (force && |
988 | (!enough(content->array.level, content->array.raid_disks, | |
989 | content->array.layout, 1, | |
990 | avail) | |
991 | || | |
992 | (content->reshape_active && content->delta_disks > 0 && | |
993 | !enough(content->array.level, (content->array.raid_disks | |
994 | - content->delta_disks), | |
995 | content->new_layout, 1, | |
996 | avail) | |
997 | ))) { | |
64c4757e NB |
998 | /* Choose the newest best drive which is |
999 | * not up-to-date, update the superblock | |
1000 | * and add it. | |
1001 | */ | |
52826846 | 1002 | int fd; |
3da92f27 | 1003 | struct supertype *tst; |
f21e18ca | 1004 | unsigned long long current_events; |
cd29a5c8 | 1005 | chosen_drive = -1; |
f21e18ca | 1006 | for (i = 0; i < content->array.raid_disks && i < bestcnt; i++) { |
52826846 NB |
1007 | int j = best[i]; |
1008 | if (j>=0 && | |
1009 | !devices[j].uptodate && | |
921d9e16 | 1010 | devices[j].i.recovery_start == MaxSector && |
52826846 | 1011 | (chosen_drive < 0 || |
213ee40b NB |
1012 | devices[j].i.events |
1013 | > devices[chosen_drive].i.events)) | |
52826846 NB |
1014 | chosen_drive = j; |
1015 | } | |
1016 | if (chosen_drive < 0) | |
1017 | break; | |
213ee40b | 1018 | current_events = devices[chosen_drive].i.events; |
c5a6f9a6 | 1019 | add_another: |
dab6685f | 1020 | if (verbose >= 0) |
e7b84f9d | 1021 | pr_err("forcing event count in %s(%d) from %d upto %d\n", |
213ee40b NB |
1022 | devices[chosen_drive].devname, |
1023 | devices[chosen_drive].i.disk.raid_disk, | |
1024 | (int)(devices[chosen_drive].i.events), | |
1025 | (int)(devices[most_recent].i.events)); | |
8b0dabea | 1026 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 | 1027 | if (fd < 0) { |
e7b84f9d | 1028 | pr_err("Couldn't open %s for write - not updating\n", |
52826846 | 1029 | devices[chosen_drive].devname); |
213ee40b | 1030 | devices[chosen_drive].i.events = 0; |
52826846 NB |
1031 | continue; |
1032 | } | |
3da92f27 | 1033 | tst = dup_super(st); |
60b435db | 1034 | if (tst->ss->load_super(tst,fd, NULL)) { |
52826846 | 1035 | close(fd); |
e7b84f9d | 1036 | pr_err("RAID superblock disappeared from %s - not updating.\n", |
52826846 | 1037 | devices[chosen_drive].devname); |
213ee40b | 1038 | devices[chosen_drive].i.events = 0; |
52826846 NB |
1039 | continue; |
1040 | } | |
98dbd966 DW |
1041 | content->events = devices[most_recent].i.events; |
1042 | tst->ss->update_super(tst, content, "force-one", | |
67a8c82d NB |
1043 | devices[chosen_drive].devname, verbose, |
1044 | 0, NULL); | |
4b1ac34b | 1045 | |
3da92f27 | 1046 | if (tst->ss->store_super(tst, fd)) { |
52826846 | 1047 | close(fd); |
e7b84f9d | 1048 | pr_err("Could not re-write superblock on %s\n", |
52826846 | 1049 | devices[chosen_drive].devname); |
213ee40b | 1050 | devices[chosen_drive].i.events = 0; |
3da92f27 | 1051 | tst->ss->free_super(tst); |
52826846 NB |
1052 | continue; |
1053 | } | |
1054 | close(fd); | |
213ee40b | 1055 | devices[chosen_drive].i.events = devices[most_recent].i.events; |
52826846 | 1056 | devices[chosen_drive].uptodate = 1; |
265e0f17 | 1057 | avail[chosen_drive] = 1; |
52826846 | 1058 | okcnt++; |
3da92f27 | 1059 | tst->ss->free_super(tst); |
c5a6f9a6 NB |
1060 | |
1061 | /* If there are any other drives of the same vintage, | |
1062 | * add them in as well. We can't lose and we might gain | |
1063 | */ | |
f21e18ca | 1064 | for (i = 0; i < content->array.raid_disks && i < bestcnt ; i++) { |
c5a6f9a6 NB |
1065 | int j = best[i]; |
1066 | if (j >= 0 && | |
1067 | !devices[j].uptodate && | |
135a31f5 | 1068 | devices[j].i.recovery_start == MaxSector && |
213ee40b | 1069 | devices[j].i.events == current_events) { |
c5a6f9a6 NB |
1070 | chosen_drive = j; |
1071 | goto add_another; | |
1072 | } | |
1073 | } | |
64c4757e | 1074 | } |
52826846 NB |
1075 | |
1076 | /* Now we want to look at the superblock which the kernel will base things on | |
1077 | * and compare the devices that we think are working with the devices that the | |
1078 | * superblock thinks are working. | |
1079 | * If there are differences and --force is given, then update this chosen | |
1080 | * superblock. | |
1081 | */ | |
cd29a5c8 | 1082 | chosen_drive = -1; |
3da92f27 | 1083 | st->ss->free_super(st); |
56eedc1a | 1084 | for (i=0; chosen_drive < 0 && i<bestcnt; i++) { |
52826846 NB |
1085 | int j = best[i]; |
1086 | int fd; | |
4b1ac34b | 1087 | |
52826846 NB |
1088 | if (j<0) |
1089 | continue; | |
1090 | if (!devices[j].uptodate) | |
1091 | continue; | |
a28232b8 N |
1092 | if (devices[j].i.events < devices[most_recent].i.events) |
1093 | continue; | |
52826846 | 1094 | chosen_drive = j; |
8b0dabea | 1095 | if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) { |
e7b84f9d | 1096 | pr_err("Cannot open %s: %s\n", |
52826846 | 1097 | devices[j].devname, strerror(errno)); |
7f91af49 | 1098 | close(mdfd); |
d7f7ebb7 | 1099 | free(devices); |
52826846 NB |
1100 | return 1; |
1101 | } | |
3da92f27 | 1102 | if (st->ss->load_super(st,fd, NULL)) { |
52826846 | 1103 | close(fd); |
e7b84f9d | 1104 | pr_err("RAID superblock has disappeared from %s\n", |
52826846 | 1105 | devices[j].devname); |
7f91af49 | 1106 | close(mdfd); |
d7f7ebb7 | 1107 | free(devices); |
52826846 NB |
1108 | return 1; |
1109 | } | |
1110 | close(fd); | |
1111 | } | |
3da92f27 | 1112 | if (st->sb == NULL) { |
e7b84f9d | 1113 | pr_err("No suitable drives found for %s\n", mddev); |
7f91af49 | 1114 | close(mdfd); |
d7f7ebb7 | 1115 | free(devices); |
4b1ac34b NB |
1116 | return 1; |
1117 | } | |
a5d85af7 | 1118 | st->ss->getinfo_super(st, content, NULL); |
f35f2525 | 1119 | #ifndef MDASSEMBLE |
98dbd966 | 1120 | sysfs_init(content, mdfd, 0); |
f35f2525 | 1121 | #endif |
56eedc1a | 1122 | for (i=0; i<bestcnt; i++) { |
52826846 | 1123 | int j = best[i]; |
98c6faba | 1124 | unsigned int desired_state; |
11a3e71d | 1125 | |
98dbd966 | 1126 | if (i < content->array.raid_disks) |
11a3e71d NB |
1127 | desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC); |
1128 | else | |
1129 | desired_state = 0; | |
1130 | ||
52826846 NB |
1131 | if (j<0) |
1132 | continue; | |
1133 | if (!devices[j].uptodate) | |
1134 | continue; | |
4b1ac34b | 1135 | |
213ee40b | 1136 | devices[j].i.disk.state = desired_state; |
4e9a6ff7 N |
1137 | if (!(devices[j].i.array.state & 1)) |
1138 | clean = 0; | |
213ee40b NB |
1139 | |
1140 | if (st->ss->update_super(st, &devices[j].i, "assemble", NULL, | |
68c7d6d7 | 1141 | verbose, 0, NULL)) { |
52826846 | 1142 | if (force) { |
dab6685f | 1143 | if (verbose >= 0) |
e7b84f9d | 1144 | pr_err("clearing FAULTY flag for device %d in %s for %s\n", |
dab6685f | 1145 | j, mddev, devices[j].devname); |
4b1ac34b | 1146 | change = 1; |
52826846 | 1147 | } else { |
dab6685f | 1148 | if (verbose >= -1) |
e7b84f9d | 1149 | pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n", |
dab6685f | 1150 | i, mddev, devices[j].devname); |
52826846 NB |
1151 | } |
1152 | } | |
4b1ac34b | 1153 | #if 0 |
213ee40b | 1154 | if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) { |
e7b84f9d | 1155 | pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n", |
52826846 NB |
1156 | i, mddev); |
1157 | } | |
4b1ac34b | 1158 | #endif |
52826846 | 1159 | } |
c5a6f9a6 | 1160 | if (force && !clean && |
98dbd966 DW |
1161 | !enough(content->array.level, content->array.raid_disks, |
1162 | content->array.layout, clean, | |
de5a472e | 1163 | avail)) { |
98dbd966 | 1164 | change += st->ss->update_super(st, content, "force-array", |
67a8c82d NB |
1165 | devices[chosen_drive].devname, verbose, |
1166 | 0, NULL); | |
583315d9 | 1167 | clean = 1; |
d013a55e | 1168 | } |
52826846 | 1169 | |
4b1ac34b | 1170 | if (change) { |
52826846 | 1171 | int fd; |
8b0dabea | 1172 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 | 1173 | if (fd < 0) { |
e7b84f9d | 1174 | pr_err("Could not open %s for write - cannot Assemble array.\n", |
52826846 | 1175 | devices[chosen_drive].devname); |
7f91af49 | 1176 | close(mdfd); |
d7f7ebb7 | 1177 | free(devices); |
52826846 NB |
1178 | return 1; |
1179 | } | |
3da92f27 | 1180 | if (st->ss->store_super(st, fd)) { |
52826846 | 1181 | close(fd); |
e7b84f9d | 1182 | pr_err("Could not re-write superblock on %s\n", |
52826846 | 1183 | devices[chosen_drive].devname); |
7f91af49 | 1184 | close(mdfd); |
d7f7ebb7 | 1185 | free(devices); |
52826846 NB |
1186 | return 1; |
1187 | } | |
a28232b8 | 1188 | if (verbose >= 0) |
e7b84f9d | 1189 | pr_err("Marking array %s as 'clean'\n", |
a28232b8 | 1190 | mddev); |
52826846 | 1191 | close(fd); |
52826846 NB |
1192 | } |
1193 | ||
353632d9 NB |
1194 | /* If we are in the middle of a reshape we may need to restore saved data |
1195 | * that was moved aside due to the reshape overwriting live data | |
1196 | * The code of doing this lives in Grow.c | |
1197 | */ | |
39c5a909 | 1198 | #ifndef MDASSEMBLE |
98dbd966 | 1199 | if (content->reshape_active) { |
353632d9 | 1200 | int err = 0; |
503975b9 | 1201 | int *fdlist = xmalloc(sizeof(int)* bestcnt); |
cd77ac4e | 1202 | if (verbose > 0) |
e7b84f9d | 1203 | pr_err(":%s has an active reshape - checking " |
ea0ebe96 N |
1204 | "if critical section needs to be restored\n", |
1205 | chosen_name); | |
353632d9 NB |
1206 | for (i=0; i<bestcnt; i++) { |
1207 | int j = best[i]; | |
1208 | if (j >= 0) { | |
1209 | fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL); | |
1210 | if (fdlist[i] < 0) { | |
e7b84f9d | 1211 | pr_err("Could not open %s for write - cannot Assemble array.\n", |
353632d9 NB |
1212 | devices[j].devname); |
1213 | err = 1; | |
1214 | break; | |
1215 | } | |
1216 | } else | |
1217 | fdlist[i] = -1; | |
1218 | } | |
87f26d14 | 1219 | if (!err) { |
ba53ea59 AK |
1220 | if (st->ss->external && st->ss->recover_backup) |
1221 | err = st->ss->recover_backup(st, content); | |
1222 | else | |
1223 | err = Grow_restart(st, content, fdlist, bestcnt, | |
1224 | backup_file, verbose > 0); | |
87f26d14 N |
1225 | if (err && invalid_backup) { |
1226 | if (verbose > 0) | |
e7b84f9d | 1227 | pr_err("continuing" |
87f26d14 N |
1228 | " without restoring backup\n"); |
1229 | err = 0; | |
1230 | } | |
1231 | } | |
353632d9 NB |
1232 | while (i>0) { |
1233 | i--; | |
1234 | if (fdlist[i]>=0) close(fdlist[i]); | |
1235 | } | |
1236 | if (err) { | |
e7b84f9d | 1237 | pr_err("Failed to restore critical section for reshape, sorry.\n"); |
e9e43ec3 | 1238 | if (backup_file == NULL) |
e7b84f9d | 1239 | cont_err("Possibly you needed to specify the --backup-file\n"); |
7f91af49 | 1240 | close(mdfd); |
d7f7ebb7 | 1241 | free(devices); |
353632d9 NB |
1242 | return err; |
1243 | } | |
1244 | } | |
39c5a909 | 1245 | #endif |
aa88f531 NB |
1246 | /* count number of in-sync devices according to the superblock. |
1247 | * We must have this number to start the array without -s or -R | |
1248 | */ | |
98dbd966 | 1249 | req_cnt = content->array.working_disks; |
aa88f531 | 1250 | |
64c4757e NB |
1251 | /* Almost ready to actually *do* something */ |
1252 | if (!old_linux) { | |
fbf8a0b7 | 1253 | int rv; |
a19c88b8 | 1254 | |
a04d5763 N |
1255 | /* First, fill in the map, so that udev can find our name |
1256 | * as soon as we become active. | |
1257 | */ | |
98dbd966 DW |
1258 | map_update(NULL, fd2devnum(mdfd), content->text_version, |
1259 | content->uuid, chosen_name); | |
a04d5763 | 1260 | |
98dbd966 | 1261 | rv = set_array_info(mdfd, st, content); |
fbf8a0b7 | 1262 | if (rv) { |
e7b84f9d | 1263 | pr_err("failed to set array info for %s: %s\n", |
64c4757e | 1264 | mddev, strerror(errno)); |
24af7a87 | 1265 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1266 | close(mdfd); |
d7f7ebb7 | 1267 | free(devices); |
64c4757e NB |
1268 | return 1; |
1269 | } | |
11484a63 | 1270 | if (ident->bitmap_fd >= 0) { |
c82f047c | 1271 | if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) { |
e7b84f9d | 1272 | pr_err("SET_BITMAP_FILE failed.\n"); |
24af7a87 | 1273 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1274 | close(mdfd); |
d7f7ebb7 | 1275 | free(devices); |
c82f047c NB |
1276 | return 1; |
1277 | } | |
7ef02d01 NB |
1278 | } else if (ident->bitmap_file) { |
1279 | /* From config file */ | |
1280 | int bmfd = open(ident->bitmap_file, O_RDWR); | |
1281 | if (bmfd < 0) { | |
e7b84f9d | 1282 | pr_err("Could not open bitmap file %s\n", |
7ef02d01 | 1283 | ident->bitmap_file); |
24af7a87 | 1284 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1285 | close(mdfd); |
d7f7ebb7 | 1286 | free(devices); |
7ef02d01 NB |
1287 | return 1; |
1288 | } | |
1289 | if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) { | |
e7b84f9d | 1290 | pr_err("Failed to set bitmapfile for %s\n", mddev); |
7ef02d01 | 1291 | close(bmfd); |
24af7a87 | 1292 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1293 | close(mdfd); |
d7f7ebb7 | 1294 | free(devices); |
7ef02d01 NB |
1295 | return 1; |
1296 | } | |
1297 | close(bmfd); | |
c82f047c | 1298 | } |
7ef02d01 | 1299 | |
52826846 | 1300 | /* First, add the raid disks, but add the chosen one last */ |
56eedc1a | 1301 | for (i=0; i<= bestcnt; i++) { |
52826846 | 1302 | int j; |
56eedc1a | 1303 | if (i < bestcnt) { |
52826846 NB |
1304 | j = best[i]; |
1305 | if (j == chosen_drive) | |
1306 | continue; | |
1307 | } else | |
1308 | j = chosen_drive; | |
1309 | ||
56eedc1a | 1310 | if (j >= 0 /* && devices[j].uptodate */) { |
484ae54d N |
1311 | int dfd = dev_open(devices[j].devname, |
1312 | O_RDWR|O_EXCL); | |
1313 | if (dfd >= 0) { | |
1314 | remove_partitions(dfd); | |
1315 | close(dfd); | |
1316 | } | |
98dbd966 | 1317 | rv = add_disk(mdfd, st, content, &devices[j].i); |
7801ac20 | 1318 | |
a19c88b8 | 1319 | if (rv) { |
e7b84f9d N |
1320 | pr_err("failed to add " |
1321 | "%s to %s: %s\n", | |
1322 | devices[j].devname, | |
1323 | mddev, | |
1324 | strerror(errno)); | |
98dbd966 | 1325 | if (i < content->array.raid_disks |
213ee40b | 1326 | || i == bestcnt) |
52826846 NB |
1327 | okcnt--; |
1328 | else | |
1329 | sparecnt--; | |
dab6685f | 1330 | } else if (verbose > 0) |
e7b84f9d N |
1331 | pr_err("added %s to %s as %d%s\n", |
1332 | devices[j].devname, mddev, | |
1333 | devices[j].i.disk.raid_disk, | |
1334 | devices[j].uptodate?"": | |
1335 | " (possibly out of date)"); | |
98dbd966 | 1336 | } else if (verbose > 0 && i < content->array.raid_disks) |
e7b84f9d | 1337 | pr_err("no uptodate device for " |
213ee40b | 1338 | "slot %d of %s\n", |
64c4757e NB |
1339 | i, mddev); |
1340 | } | |
aba69144 | 1341 | |
98dbd966 | 1342 | if (content->array.level == LEVEL_CONTAINER) { |
598f0d58 | 1343 | if (verbose >= 0) { |
e7b84f9d | 1344 | pr_err("Container %s has been " |
598f0d58 | 1345 | "assembled with %d drive%s", |
57ed8c91 | 1346 | mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s"); |
f21e18ca | 1347 | if (okcnt < (unsigned)content->array.raid_disks) |
598f0d58 | 1348 | fprintf(stderr, " (out of %d)", |
98dbd966 | 1349 | content->array.raid_disks); |
598f0d58 NB |
1350 | fprintf(stderr, "\n"); |
1351 | } | |
ac597b1c | 1352 | st->ss->free_super(st); |
98dbd966 | 1353 | sysfs_uevent(content, "change"); |
a7c6e3fb | 1354 | wait_for(chosen_name, mdfd); |
7f91af49 | 1355 | close(mdfd); |
d7f7ebb7 | 1356 | free(devices); |
598f0d58 NB |
1357 | return 0; |
1358 | } | |
1359 | ||
64c4757e | 1360 | if (runstop == 1 || |
b8a8ccf9 | 1361 | (runstop <= 0 && |
98dbd966 | 1362 | ( enough(content->array.level, content->array.raid_disks, |
de5a472e | 1363 | content->array.layout, clean, avail) && |
1ff98339 | 1364 | (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok) |
aa88f531 | 1365 | ))) { |
e9e43ec3 N |
1366 | /* This array is good-to-go. |
1367 | * If a reshape is in progress then we might need to | |
1368 | * continue monitoring it. In that case we start | |
1369 | * it read-only and let the grow code make it writable. | |
1370 | */ | |
1371 | int rv; | |
eb3929a4 | 1372 | #ifndef MDASSEMBLE |
e9e43ec3 | 1373 | if (content->reshape_active && |
3bd58dc6 AK |
1374 | content->delta_disks <= 0) { |
1375 | rv = sysfs_set_str(content, NULL, | |
1376 | "array_state", "readonly"); | |
1377 | if (rv == 0) | |
1378 | rv = Grow_continue(mdfd, st, content, | |
1379 | backup_file, | |
1380 | freeze_reshape); | |
1381 | } else | |
eb3929a4 | 1382 | #endif |
e9e43ec3 N |
1383 | rv = ioctl(mdfd, RUN_ARRAY, NULL); |
1384 | if (rv == 0) { | |
dab6685f | 1385 | if (verbose >= 0) { |
e7b84f9d | 1386 | pr_err("%s has been started with %d drive%s", |
dab6685f | 1387 | mddev, okcnt, okcnt==1?"":"s"); |
f21e18ca | 1388 | if (okcnt < (unsigned)content->array.raid_disks) |
98dbd966 | 1389 | fprintf(stderr, " (out of %d)", content->array.raid_disks); |
1ff98339 N |
1390 | if (rebuilding_cnt) |
1391 | fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt); | |
dab6685f NB |
1392 | if (sparecnt) |
1393 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
1394 | fprintf(stderr, ".\n"); | |
1395 | } | |
8a659c33 N |
1396 | if (content->reshape_active && |
1397 | content->array.level >= 4 && | |
1398 | content->array.level <= 6) { | |
acee8e89 N |
1399 | /* might need to increase the size |
1400 | * of the stripe cache - default is 256 | |
1401 | */ | |
8a659c33 | 1402 | if (256 < 4 * (content->array.chunk_size/4096)) { |
acee8e89 N |
1403 | struct mdinfo *sra = sysfs_read(mdfd, 0, 0); |
1404 | if (sra) | |
1405 | sysfs_set_num(sra, NULL, | |
1406 | "stripe_cache_size", | |
8a659c33 | 1407 | (4 * content->array.chunk_size / 4096) + 1); |
83366b33 | 1408 | sysfs_free(sra); |
acee8e89 N |
1409 | } |
1410 | } | |
7e83544b N |
1411 | if (okcnt < (unsigned)content->array.raid_disks) { |
1412 | /* If any devices did not get added | |
1413 | * because the kernel rejected them based | |
1414 | * on event count, try adding them | |
1415 | * again providing the action policy is | |
1416 | * 're-add' or greater. The bitmap | |
1417 | * might allow them to be included, or | |
1418 | * they will become spares. | |
1419 | */ | |
b787bec6 | 1420 | for (i = 0; i < bestcnt; i++) { |
7e83544b N |
1421 | int j = best[i]; |
1422 | if (j >= 0 && !devices[j].uptodate) { | |
1423 | if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add)) | |
1424 | continue; | |
1425 | rv = add_disk(mdfd, st, content, | |
1426 | &devices[j].i); | |
1427 | if (rv == 0 && verbose >= 0) | |
e7b84f9d N |
1428 | pr_err("%s has been re-added.\n", |
1429 | devices[j].devname); | |
7e83544b N |
1430 | } |
1431 | } | |
1432 | } | |
a7c6e3fb | 1433 | wait_for(mddev, mdfd); |
7f91af49 N |
1434 | close(mdfd); |
1435 | if (auto_assem) { | |
589395d6 | 1436 | int usecs = 1; |
589395d6 NB |
1437 | /* There is a nasty race with 'mdadm --monitor'. |
1438 | * If it opens this device before we close it, | |
1439 | * it gets an incomplete open on which IO | |
598f0d58 NB |
1440 | * doesn't work and the capacity is |
1441 | * wrong. | |
589395d6 NB |
1442 | * If we reopen (to check for layered devices) |
1443 | * before --monitor closes, we loose. | |
1444 | * | |
1445 | * So: wait upto 1 second for there to be | |
1446 | * a non-zero capacity. | |
1447 | */ | |
1448 | while (usecs < 1000) { | |
1449 | mdfd = open(mddev, O_RDONLY); | |
1450 | if (mdfd >= 0) { | |
beae1dfe NB |
1451 | unsigned long long size; |
1452 | if (get_dev_size(mdfd, NULL, &size) && | |
589395d6 NB |
1453 | size > 0) |
1454 | break; | |
1455 | close(mdfd); | |
1456 | } | |
1457 | usleep(usecs); | |
1458 | usecs <<= 1; | |
1459 | } | |
1460 | } | |
d7f7ebb7 | 1461 | free(devices); |
64c4757e | 1462 | return 0; |
82b27616 | 1463 | } |
e7b84f9d | 1464 | pr_err("failed to RUN_ARRAY %s: %s\n", |
64c4757e | 1465 | mddev, strerror(errno)); |
583315d9 | 1466 | |
98dbd966 | 1467 | if (!enough(content->array.level, content->array.raid_disks, |
de5a472e | 1468 | content->array.layout, 1, avail)) |
e7b84f9d | 1469 | pr_err("Not enough devices to " |
583315d9 | 1470 | "start the array.\n"); |
98dbd966 DW |
1471 | else if (!enough(content->array.level, |
1472 | content->array.raid_disks, | |
1473 | content->array.layout, clean, | |
de5a472e | 1474 | avail)) |
e7b84f9d | 1475 | pr_err("Not enough devices to " |
583315d9 NB |
1476 | "start the array while not clean " |
1477 | "- consider --force.\n"); | |
1478 | ||
7f91af49 | 1479 | if (auto_assem) |
1c203a4b | 1480 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1481 | close(mdfd); |
d7f7ebb7 | 1482 | free(devices); |
64c4757e NB |
1483 | return 1; |
1484 | } | |
82b27616 | 1485 | if (runstop == -1) { |
e7b84f9d N |
1486 | pr_err("%s assembled from %d drive%s", |
1487 | mddev, okcnt, okcnt==1?"":"s"); | |
f21e18ca | 1488 | if (okcnt != (unsigned)content->array.raid_disks) |
98dbd966 | 1489 | fprintf(stderr, " (out of %d)", content->array.raid_disks); |
b8a8ccf9 | 1490 | fprintf(stderr, ", but not started.\n"); |
7f91af49 | 1491 | close(mdfd); |
d7f7ebb7 | 1492 | free(devices); |
64c4757e | 1493 | return 0; |
82b27616 | 1494 | } |
4855f95c | 1495 | if (verbose >= -1) { |
e7b84f9d | 1496 | pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s"); |
1ff98339 N |
1497 | if (rebuilding_cnt) |
1498 | fprintf(stderr, "%s %d rebuilding", sparecnt?", ":" and ", rebuilding_cnt); | |
dab6685f NB |
1499 | if (sparecnt) |
1500 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
98dbd966 | 1501 | if (!enough(content->array.level, content->array.raid_disks, |
de5a472e | 1502 | content->array.layout, 1, avail)) |
dab6685f | 1503 | fprintf(stderr, " - not enough to start the array.\n"); |
98dbd966 DW |
1504 | else if (!enough(content->array.level, |
1505 | content->array.raid_disks, | |
1506 | content->array.layout, clean, | |
de5a472e | 1507 | avail)) |
583315d9 NB |
1508 | fprintf(stderr, " - not enough to start the " |
1509 | "array while not clean - consider " | |
1510 | "--force.\n"); | |
dab6685f | 1511 | else { |
f21e18ca | 1512 | if (req_cnt == (unsigned)content->array.raid_disks) |
dab6685f NB |
1513 | fprintf(stderr, " - need all %d to start it", req_cnt); |
1514 | else | |
98dbd966 | 1515 | fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks); |
dab6685f NB |
1516 | fprintf(stderr, " (use --run to insist).\n"); |
1517 | } | |
aa88f531 | 1518 | } |
7f91af49 | 1519 | if (auto_assem) |
1c203a4b | 1520 | ioctl(mdfd, STOP_ARRAY, NULL); |
56a8da69 | 1521 | close(mdfd); |
d7f7ebb7 | 1522 | free(devices); |
64c4757e | 1523 | return 1; |
82b27616 | 1524 | } else { |
52826846 NB |
1525 | /* The "chosen_drive" is a good choice, and if necessary, the superblock has |
1526 | * been updated to point to the current locations of devices. | |
1527 | * so we can just start the array | |
82b27616 | 1528 | */ |
cd29a5c8 | 1529 | unsigned long dev; |
213ee40b NB |
1530 | dev = makedev(devices[chosen_drive].i.disk.major, |
1531 | devices[chosen_drive].i.disk.minor); | |
82b27616 | 1532 | if (ioctl(mdfd, START_ARRAY, dev)) { |
e7b84f9d | 1533 | pr_err("Cannot start array: %s\n", |
82b27616 NB |
1534 | strerror(errno)); |
1535 | } | |
aba69144 | 1536 | |
64c4757e | 1537 | } |
7f91af49 | 1538 | close(mdfd); |
d7f7ebb7 | 1539 | free(devices); |
e0d19036 | 1540 | return 0; |
64c4757e | 1541 | } |
6234c63c DW |
1542 | |
1543 | #ifndef MDASSEMBLE | |
1544 | int assemble_container_content(struct supertype *st, int mdfd, | |
1545 | struct mdinfo *content, int runstop, | |
49680258 | 1546 | char *chosen_name, int verbose, |
b76b30e0 | 1547 | char *backup_file, int freeze_reshape) |
6234c63c DW |
1548 | { |
1549 | struct mdinfo *dev, *sra; | |
1550 | int working = 0, preexist = 0; | |
882029c8 | 1551 | int expansion = 0; |
6234c63c | 1552 | struct map_ent *map = NULL; |
7af03341 | 1553 | int old_raid_disks; |
4aecb54a | 1554 | int start_reshape; |
6234c63c DW |
1555 | |
1556 | sysfs_init(content, mdfd, 0); | |
1557 | ||
1558 | sra = sysfs_read(mdfd, 0, GET_VERSION); | |
1559 | if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) | |
22472ee1 JS |
1560 | if (sysfs_set_array(content, md_get_version(mdfd)) != 0) { |
1561 | if (sra) | |
1562 | sysfs_free(sra); | |
6234c63c | 1563 | return 1; |
22472ee1 | 1564 | } |
588bebfc | 1565 | |
4aecb54a AK |
1566 | /* There are two types of reshape: container wide or sub-array specific |
1567 | * Check if metadata requests blocking container wide reshapes | |
1568 | */ | |
1569 | start_reshape = (content->reshape_active && | |
1570 | !((content->reshape_active == CONTAINER_RESHAPE) && | |
1571 | (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))); | |
1572 | ||
1573 | /* Block subarray here if it is under reshape now | |
1574 | * Do not allow for any changes in this array | |
1575 | */ | |
1576 | if (st->ss->external && content->recovery_blocked && start_reshape) | |
b8063f07 AK |
1577 | block_subarray(content); |
1578 | ||
6234c63c DW |
1579 | if (sra) |
1580 | sysfs_free(sra); | |
7af03341 | 1581 | old_raid_disks = content->array.raid_disks - content->delta_disks; |
6234c63c | 1582 | for (dev = content->devs; dev; dev = dev->next) |
14032016 | 1583 | if (sysfs_add_disk(content, dev, 1) == 0) { |
7af03341 | 1584 | if (dev->disk.raid_disk >= old_raid_disks && |
14032016 AK |
1585 | content->reshape_active) |
1586 | expansion++; | |
1587 | else | |
1588 | working++; | |
1589 | } else if (errno == EEXIST) | |
6234c63c | 1590 | preexist++; |
111e9fda | 1591 | if (working + expansion == 0) |
7cb2aa33 | 1592 | return 1;/* Nothing new, don't try to start */ |
588bebfc | 1593 | |
8b4e5ea9 N |
1594 | map_update(&map, fd2devnum(mdfd), |
1595 | content->text_version, | |
1596 | content->uuid, chosen_name); | |
1597 | ||
1598 | if (runstop > 0 || | |
882029c8 AK |
1599 | (working + preexist + expansion) >= |
1600 | content->array.working_disks) { | |
bb50e5d3 | 1601 | int err; |
a714580e | 1602 | |
81219e70 | 1603 | if (start_reshape) { |
49680258 | 1604 | int spare = content->array.raid_disks + expansion; |
3f54bd62 AK |
1605 | if (restore_backup(st, content, |
1606 | working, | |
1607 | spare, backup_file, verbose) == 1) | |
49680258 | 1608 | return 1; |
49680258 | 1609 | |
3bd58dc6 AK |
1610 | err = sysfs_set_str(content, NULL, |
1611 | "array_state", "readonly"); | |
1612 | if (err) | |
1613 | return 1; | |
1614 | ||
a93ada3b AK |
1615 | if (st->ss->external) { |
1616 | if (!mdmon_running(st->container_dev)) | |
1617 | start_mdmon(st->container_dev); | |
1618 | ping_monitor_by_id(st->container_dev); | |
7728e1c6 LD |
1619 | if (mdmon_running(st->container_dev) && |
1620 | st->update_tail == NULL) | |
1621 | st->update_tail = &st->updates; | |
a93ada3b AK |
1622 | } |
1623 | ||
b76b30e0 AK |
1624 | err = Grow_continue(mdfd, st, content, backup_file, |
1625 | freeze_reshape); | |
49680258 | 1626 | } else switch(content->array.level) { |
6234c63c DW |
1627 | case LEVEL_LINEAR: |
1628 | case LEVEL_MULTIPATH: | |
1629 | case 0: | |
bb50e5d3 N |
1630 | err = sysfs_set_str(content, NULL, "array_state", |
1631 | "active"); | |
6234c63c DW |
1632 | break; |
1633 | default: | |
bb50e5d3 | 1634 | err = sysfs_set_str(content, NULL, "array_state", |
6234c63c DW |
1635 | "readonly"); |
1636 | /* start mdmon if needed. */ | |
bb50e5d3 N |
1637 | if (!err) { |
1638 | if (!mdmon_running(st->container_dev)) | |
1639 | start_mdmon(st->container_dev); | |
983fff45 | 1640 | ping_monitor_by_id(st->container_dev); |
bb50e5d3 | 1641 | } |
6234c63c DW |
1642 | break; |
1643 | } | |
bb50e5d3 N |
1644 | if (!err) |
1645 | sysfs_set_safemode(content, content->safe_mode_delay); | |
4aecb54a AK |
1646 | |
1647 | /* Block subarray here if it is not reshaped now | |
1648 | * It has be blocked a little later to allow mdmon to switch in | |
1649 | * in to R/W state | |
1650 | */ | |
1651 | if (st->ss->external && content->recovery_blocked && | |
1652 | !start_reshape) | |
1653 | block_subarray(content); | |
1654 | ||
6234c63c | 1655 | if (verbose >= 0) { |
bb50e5d3 | 1656 | if (err) |
e7b84f9d N |
1657 | pr_err("array %s now has %d device%s", |
1658 | chosen_name, working + preexist, | |
1659 | working + preexist == 1 ? "":"s"); | |
bb50e5d3 | 1660 | else |
e7b84f9d N |
1661 | pr_err("Started %s with %d device%s", |
1662 | chosen_name, working + preexist, | |
1663 | working + preexist == 1 ? "":"s"); | |
6234c63c DW |
1664 | if (preexist) |
1665 | fprintf(stderr, " (%d new)", working); | |
882029c8 AK |
1666 | if (expansion) |
1667 | fprintf(stderr, " ( + %d for expansion)", | |
1668 | expansion); | |
6234c63c DW |
1669 | fprintf(stderr, "\n"); |
1670 | } | |
bb50e5d3 | 1671 | if (!err) |
a7c6e3fb | 1672 | wait_for(chosen_name, mdfd); |
588bebfc | 1673 | return err; |
6234c63c | 1674 | /* FIXME should have an O_EXCL and wait for read-auto */ |
7cb2aa33 | 1675 | } else { |
88716263 | 1676 | if (verbose >= 0) { |
e7b84f9d N |
1677 | pr_err("%s assembled with %d device%s", |
1678 | chosen_name, preexist + working, | |
1679 | preexist + working == 1 ? "":"s"); | |
88716263 N |
1680 | if (preexist) |
1681 | fprintf(stderr, " (%d new)", working); | |
1682 | fprintf(stderr, " but not started\n"); | |
1683 | } | |
7cb2aa33 N |
1684 | return 1; |
1685 | } | |
6234c63c DW |
1686 | } |
1687 | #endif | |
1688 |