]>
Commit | Line | Data |
---|---|---|
64c4757e | 1 | /* |
9a9dab36 | 2 | * mdadm - manage Linux "md" devices aka RAID arrays. |
64c4757e | 3 | * |
6f02172d | 4 | * Copyright (C) 2001-2013 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 | ||
ca3b6696 | 25 | #include "mdadm.h" |
682c7051 NB |
26 | #include "md_u.h" |
27 | #include "md_p.h" | |
60248f74 | 28 | #include <ctype.h> |
64c4757e | 29 | |
22dc741f MT |
30 | static int round_size_and_verify(unsigned long long *size, int chunk) |
31 | { | |
32 | if (*size == 0) | |
33 | return 0; | |
34 | *size &= ~(unsigned long long)(chunk - 1); | |
35 | if (*size == 0) { | |
36 | pr_err("Size cannot be smaller than chunk.\n"); | |
37 | return 1; | |
38 | } | |
39 | return 0; | |
40 | } | |
41 | ||
a18a888e DW |
42 | static int default_layout(struct supertype *st, int level, int verbose) |
43 | { | |
44 | int layout = UnSet; | |
45 | ||
30f58b22 DW |
46 | if (st && st->ss->default_geometry) |
47 | st->ss->default_geometry(st, &level, &layout, NULL); | |
a18a888e DW |
48 | |
49 | if (layout == UnSet) | |
50 | switch(level) { | |
51 | default: /* no layout */ | |
52 | layout = 0; | |
53 | break; | |
54 | case 10: | |
55 | layout = 0x102; /* near=2, far=1 */ | |
56 | if (verbose > 0) | |
e7b84f9d | 57 | pr_err("layout defaults to n2\n"); |
a18a888e DW |
58 | break; |
59 | case 5: | |
60 | case 6: | |
61 | layout = map_name(r5layout, "default"); | |
62 | if (verbose > 0) | |
e7b84f9d | 63 | pr_err("layout defaults to %s\n", map_num(r5layout, layout)); |
a18a888e DW |
64 | break; |
65 | case LEVEL_FAULTY: | |
66 | layout = map_name(faultylayout, "default"); | |
67 | ||
68 | if (verbose > 0) | |
e7b84f9d | 69 | pr_err("layout defaults to %s\n", map_num(faultylayout, layout)); |
a18a888e DW |
70 | break; |
71 | } | |
72 | ||
73 | return layout; | |
74 | } | |
75 | ||
7f91af49 | 76 | int Create(struct supertype *st, char *mddev, |
171dccc8 | 77 | char *name, int *uuid, |
a655e550 | 78 | int subdevs, struct mddev_dev *devlist, |
99cc42f4 | 79 | struct shape *s, |
40c9a66a | 80 | struct context *c, unsigned long long data_offset) |
64c4757e | 81 | { |
682c7051 NB |
82 | /* |
83 | * Create a new raid array. | |
84 | * | |
85 | * First check that necessary details are available | |
86 | * (i.e. level, raid-disks) | |
87 | * | |
88 | * Then check each disk to see what might be on it | |
89 | * and report anything interesting. | |
90 | * | |
91 | * If anything looks odd, and runstop not set, | |
92 | * abort. | |
93 | * | |
94 | * SET_ARRAY_INFO and ADD_NEW_DISK, and | |
82d9eba6 | 95 | * if runstop==run, or raiddisks disks were used, |
682c7051 NB |
96 | * RUN_ARRAY |
97 | */ | |
7f91af49 | 98 | int mdfd; |
98dbf73c | 99 | unsigned long long minsize = 0, maxsize = 0; |
cd29a5c8 NB |
100 | char *mindisc = NULL; |
101 | char *maxdisc = NULL; | |
cc1799c3 | 102 | int dnum, raid_disk_num; |
a655e550 | 103 | struct mddev_dev *dv; |
0a6bff09 | 104 | dev_t rdev; |
98dbf73c | 105 | int fail = 0, warn = 0; |
82d9eba6 | 106 | int first_missing = subdevs * 2; |
6fb79233 | 107 | int second_missing = subdevs * 2; |
52826846 | 108 | int missing_disks = 0; |
82d9eba6 | 109 | int insert_point = subdevs * 2; /* where to insert a missing drive */ |
d2ca6449 | 110 | int total_slots; |
4b1ac34b | 111 | int pass; |
82d9eba6 | 112 | int rv; |
c82f047c | 113 | int bitmap_fd; |
5f8097be | 114 | int have_container = 0; |
0e600426 | 115 | int container_fd = -1; |
a931db9e | 116 | int need_mdmon = 0; |
f9c25f1d | 117 | unsigned long long bitmapsize; |
d2ca6449 | 118 | struct mdinfo info, *infos; |
b8ac1967 | 119 | int did_default = 0; |
a18a888e | 120 | int do_default_layout = 0; |
c21e737b | 121 | int do_default_chunk = 0; |
8ed3e5e1 | 122 | unsigned long safe_mode_delay = 0; |
a04d5763 | 123 | char chosen_name[1024]; |
ad5bc697 | 124 | struct map_ent *map = NULL; |
8592f29d | 125 | unsigned long long newsize; |
5f4cc239 | 126 | mdu_array_info_t inf; |
682c7051 | 127 | |
943eafef | 128 | int major_num = BITMAP_MAJOR_HI; |
82d9485e | 129 | if (s->bitmap_file && strcmp(s->bitmap_file, "clustered") == 0) { |
6d9c7c25 | 130 | major_num = BITMAP_MAJOR_CLUSTERED; |
82d9485e GJ |
131 | if (c->nodes <= 1) { |
132 | pr_err("At least 2 nodes are needed for cluster-md\n"); | |
133 | return 1; | |
134 | } | |
135 | } | |
dcec9ee5 | 136 | |
06c7f68e | 137 | memset(&info, 0, sizeof(info)); |
99cc42f4 N |
138 | if (s->level == UnSet && st && st->ss->default_geometry) |
139 | st->ss->default_geometry(st, &s->level, NULL, NULL); | |
140 | if (s->level == UnSet) { | |
e7b84f9d | 141 | pr_err("a RAID level is needed to create an array.\n"); |
682c7051 NB |
142 | return 1; |
143 | } | |
99cc42f4 | 144 | if (s->raiddisks < 4 && s->level == 6) { |
e7b84f9d | 145 | pr_err("at least 4 raid-devices needed for level 6\n"); |
98c6faba NB |
146 | return 1; |
147 | } | |
99cc42f4 | 148 | if (s->raiddisks > 256 && s->level == 6) { |
e7b84f9d | 149 | pr_err("no more than 256 raid-devices supported for level 6\n"); |
98c6faba NB |
150 | return 1; |
151 | } | |
99cc42f4 | 152 | if (s->raiddisks < 2 && s->level >= 4) { |
e7b84f9d | 153 | pr_err("at least 2 raid-devices needed for level 4 or 5\n"); |
52826846 NB |
154 | return 1; |
155 | } | |
99cc42f4 | 156 | if (s->level <= 0 && s->sparedisks) { |
e7b84f9d | 157 | pr_err("This level does not support spare devices\n"); |
570510ba NB |
158 | return 1; |
159 | } | |
5f8097be NB |
160 | |
161 | if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) { | |
162 | /* If given a single device, it might be a container, and we can | |
163 | * extract a device list from there | |
164 | */ | |
5f8097be NB |
165 | int fd; |
166 | ||
167 | memset(&inf, 0, sizeof(inf)); | |
37ea3936 | 168 | fd = open(devlist->devname, O_RDONLY); |
5f8097be | 169 | if (fd >= 0 && |
9cd39f01 | 170 | md_get_array_info(fd, &inf) == 0 && inf.raid_disks == 0) { |
5f8097be NB |
171 | /* yep, looks like a container */ |
172 | if (st) { | |
0fb69d1d N |
173 | rv = st->ss->load_container(st, fd, |
174 | devlist->devname); | |
5f8097be NB |
175 | if (rv == 0) |
176 | have_container = 1; | |
177 | } else { | |
0fb69d1d | 178 | st = super_by_fd(fd, NULL); |
5f8097be | 179 | if (st && !(rv = st->ss-> |
0fb69d1d N |
180 | load_container(st, fd, |
181 | devlist->devname))) | |
5f8097be NB |
182 | have_container = 1; |
183 | else | |
184 | st = NULL; | |
185 | } | |
18fde300 | 186 | if (have_container) { |
99cc42f4 | 187 | subdevs = s->raiddisks; |
18fde300 DW |
188 | first_missing = subdevs * 2; |
189 | second_missing = subdevs * 2; | |
190 | insert_point = subdevs * 2; | |
191 | } | |
5f8097be NB |
192 | } |
193 | if (fd >= 0) | |
194 | close(fd); | |
5f8097be | 195 | } |
99cc42f4 | 196 | if (st && st->ss->external && s->sparedisks) { |
7a862a02 | 197 | pr_err("This metadata type does not support spare disks at create time\n"); |
ffcfc735 N |
198 | return 1; |
199 | } | |
cc1799c3 | 200 | if (subdevs > s->raiddisks+s->sparedisks+s->journaldisks) { |
99cc42f4 | 201 | pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks); |
52826846 | 202 | return 1; |
682c7051 | 203 | } |
cc1799c3 | 204 | if (!have_container && subdevs < s->raiddisks+s->sparedisks+s->journaldisks) { |
e7b84f9d | 205 | pr_err("You haven't given enough devices (real or missing) to create this array\n"); |
52826846 NB |
206 | return 1; |
207 | } | |
99cc42f4 | 208 | if (s->bitmap_file && s->level <= 0) { |
e7b84f9d | 209 | pr_err("bitmaps not meaningful with level %s\n", |
99cc42f4 | 210 | map_num(pers, s->level)?:"given"); |
5b28bd56 NB |
211 | return 1; |
212 | } | |
52826846 | 213 | |
682c7051 | 214 | /* now set some defaults */ |
b5e64645 | 215 | |
99cc42f4 | 216 | if (s->layout == UnSet) { |
a18a888e | 217 | do_default_layout = 1; |
99cc42f4 | 218 | s->layout = default_layout(st, s->level, c->verbose); |
a18a888e | 219 | } |
682c7051 | 220 | |
99cc42f4 | 221 | if (s->level == 10) |
e5329c37 | 222 | /* check layout fits in array*/ |
99cc42f4 | 223 | if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) { |
e7b84f9d | 224 | pr_err("that layout requires at least %d devices\n", |
99cc42f4 | 225 | (s->layout&255) * ((s->layout>>8)&255)); |
e5329c37 NB |
226 | return 1; |
227 | } | |
228 | ||
99cc42f4 | 229 | switch(s->level) { |
aa88f531 NB |
230 | case 4: |
231 | case 5: | |
e5329c37 | 232 | case 10: |
98c6faba | 233 | case 6: |
aa88f531 | 234 | case 0: |
99cc42f4 N |
235 | if (s->chunk == 0 || s->chunk == UnSet) { |
236 | s->chunk = UnSet; | |
c21e737b CA |
237 | do_default_chunk = 1; |
238 | /* chunk will be set later */ | |
bb7295f1 | 239 | } |
5f175898 N |
240 | break; |
241 | case LEVEL_LINEAR: | |
242 | /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */ | |
99cc42f4 N |
243 | if (get_linux_version() < 2006016 && s->chunk == 0) { |
244 | s->chunk = 64; | |
171dccc8 | 245 | if (c->verbose > 0) |
e7b84f9d | 246 | pr_err("chunk size defaults to 64K\n"); |
aa88f531 NB |
247 | } |
248 | break; | |
570510ba | 249 | case 1: |
08e43379 | 250 | case LEVEL_FAULTY: |
570510ba | 251 | case LEVEL_MULTIPATH: |
17f25ca6 | 252 | case LEVEL_CONTAINER: |
99cc42f4 N |
253 | if (s->chunk) { |
254 | s->chunk = 0; | |
171dccc8 | 255 | if (c->verbose > 0) |
e7b84f9d | 256 | pr_err("chunk size ignored for this level\n"); |
aa88f531 NB |
257 | } |
258 | break; | |
570510ba | 259 | default: |
99cc42f4 | 260 | pr_err("unknown level %d\n", s->level); |
570510ba | 261 | return 1; |
682c7051 | 262 | } |
22dc741f | 263 | |
99cc42f4 | 264 | if (s->size == MAX_SIZE) |
d04f65f4 | 265 | /* use '0' to mean 'max' now... */ |
99cc42f4 N |
266 | s->size = 0; |
267 | if (s->size && s->chunk && s->chunk != UnSet) | |
22dc741f MT |
268 | if (round_size_and_verify(&s->size, s->chunk)) |
269 | return 1; | |
270 | ||
99cc42f4 N |
271 | newsize = s->size * 2; |
272 | if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks, | |
af4348dd | 273 | &s->chunk, s->size*2, |
40c9a66a | 274 | data_offset, NULL, |
5308f117 | 275 | &newsize, s->consistency_policy, |
98dbf73c | 276 | c->verbose >= 0)) |
17f25ca6 | 277 | return 1; |
bb7295f1 | 278 | |
99cc42f4 N |
279 | if (s->chunk && s->chunk != UnSet) { |
280 | newsize &= ~(unsigned long long)(s->chunk*2 - 1); | |
64385908 CA |
281 | if (do_default_chunk) { |
282 | /* default chunk was just set */ | |
171dccc8 | 283 | if (c->verbose > 0) |
7a862a02 | 284 | pr_err("chunk size defaults to %dK\n", s->chunk); |
22dc741f MT |
285 | if (round_size_and_verify(&s->size, s->chunk)) |
286 | return 1; | |
64385908 CA |
287 | do_default_chunk = 0; |
288 | } | |
bb7295f1 | 289 | } |
64385908 | 290 | |
99cc42f4 N |
291 | if (s->size == 0) { |
292 | s->size = newsize / 2; | |
293 | if (s->level == 1) | |
ae6c05ad N |
294 | /* If this is ever reshaped to RAID5, we will |
295 | * need a chunksize. So round it off a bit | |
296 | * now just to be safe | |
297 | */ | |
99cc42f4 | 298 | s->size &= ~(64ULL-1); |
ae6c05ad | 299 | |
99cc42f4 N |
300 | if (s->size && c->verbose > 0) |
301 | pr_err("setting size to %lluK\n", s->size); | |
8592f29d | 302 | } |
17f25ca6 | 303 | |
682c7051 | 304 | /* now look at the subdevs */ |
06c7f68e NB |
305 | info.array.active_disks = 0; |
306 | info.array.working_disks = 0; | |
c913b90e | 307 | dnum = 0; |
98dbf73c | 308 | for (dv = devlist; dv; dv = dv->next) |
476066a3 N |
309 | if (data_offset == VARIABLE_OFFSET) |
310 | dv->data_offset = INVALID_SECTORS; | |
311 | else | |
312 | dv->data_offset = data_offset; | |
313 | ||
8592f29d | 314 | for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) { |
cd29a5c8 | 315 | char *dname = dv->devname; |
17f25ca6 | 316 | unsigned long long freesize; |
e9b11fee | 317 | int dfd; |
72ca9bcf | 318 | char *doff; |
e9b11fee | 319 | |
98dbf73c | 320 | if (strcasecmp(dname, "missing") == 0) { |
c913b90e NB |
321 | if (first_missing > dnum) |
322 | first_missing = dnum; | |
6fb79233 NB |
323 | if (second_missing > dnum && dnum > first_missing) |
324 | second_missing = dnum; | |
52826846 NB |
325 | missing_disks ++; |
326 | continue; | |
327 | } | |
8baab049 | 328 | if (data_offset == VARIABLE_OFFSET) { |
72ca9bcf N |
329 | doff = strchr(dname, ':'); |
330 | if (doff) { | |
331 | *doff++ = 0; | |
332 | dv->data_offset = parse_size(doff); | |
333 | } else | |
334 | dv->data_offset = INVALID_SECTORS; | |
335 | } else | |
336 | dv->data_offset = data_offset; | |
337 | ||
58b3c697 | 338 | dfd = open(dname, O_RDONLY); |
e9b11fee | 339 | if (dfd < 0) { |
e7b84f9d | 340 | pr_err("cannot open %s: %s\n", |
e9b11fee N |
341 | dname, strerror(errno)); |
342 | exit(2); | |
343 | } | |
0a6bff09 | 344 | if (!fstat_is_blkdev(dfd, dname, NULL)) { |
e9b11fee | 345 | close(dfd); |
e9b11fee N |
346 | exit(2); |
347 | } | |
348 | close(dfd); | |
06c7f68e | 349 | info.array.working_disks++; |
dfd7822c | 350 | if (dnum < s->raiddisks && dv->disposition != 'j') |
06c7f68e | 351 | info.array.active_disks++; |
058574b1 | 352 | if (st == NULL) { |
8aec876d | 353 | struct createinfo *ci = conf_get_create_info(); |
058574b1 NB |
354 | if (ci) |
355 | st = ci->supertype; | |
356 | } | |
576d6d83 NB |
357 | if (st == NULL) { |
358 | /* Need to choose a default metadata, which is different | |
17f25ca6 | 359 | * depending on geometry of array. |
576d6d83 NB |
360 | */ |
361 | int i; | |
362 | char *name = "default"; | |
98dbf73c | 363 | for(i = 0; !st && superlist[i]; i++) { |
576d6d83 | 364 | st = superlist[i]->match_metadata_desc(name); |
ecbd9e81 N |
365 | if (!st) |
366 | continue; | |
a18a888e | 367 | if (do_default_layout) |
99cc42f4 | 368 | s->layout = default_layout(st, s->level, c->verbose); |
ecbd9e81 | 369 | switch (st->ss->validate_geometry( |
99cc42f4 | 370 | st, s->level, s->layout, s->raiddisks, |
af4348dd | 371 | &s->chunk, s->size*2, |
72ca9bcf | 372 | dv->data_offset, dname, |
5308f117 AP |
373 | &freesize, s->consistency_policy, |
374 | c->verbose > 0)) { | |
ecbd9e81 N |
375 | case -1: /* Not valid, message printed, and not |
376 | * worth checking any further */ | |
377 | exit(2); | |
378 | break; | |
379 | case 0: /* Geometry not valid */ | |
55425f27 | 380 | free(st); |
17f25ca6 | 381 | st = NULL; |
99cc42f4 | 382 | s->chunk = do_default_chunk ? UnSet : s->chunk; |
ecbd9e81 N |
383 | break; |
384 | case 1: /* All happy */ | |
385 | break; | |
55425f27 | 386 | } |
17f25ca6 | 387 | } |
576d6d83 NB |
388 | |
389 | if (!st) { | |
58b3c697 N |
390 | int dfd = open(dname, O_RDONLY|O_EXCL); |
391 | if (dfd < 0) { | |
e7b84f9d | 392 | pr_err("cannot open %s: %s\n", |
58b3c697 N |
393 | dname, strerror(errno)); |
394 | exit(2); | |
395 | } | |
7a862a02 | 396 | pr_err("device %s not suitable for any style of array\n", |
17f25ca6 | 397 | dname); |
576d6d83 NB |
398 | exit(2); |
399 | } | |
b8ac1967 | 400 | if (st->ss != &super0 || |
576d6d83 | 401 | st->minor_version != 90) |
b8ac1967 | 402 | did_default = 1; |
17f25ca6 | 403 | } else { |
a18a888e | 404 | if (do_default_layout) |
99cc42f4 N |
405 | s->layout = default_layout(st, s->level, 0); |
406 | if (!st->ss->validate_geometry(st, s->level, s->layout, | |
407 | s->raiddisks, | |
af4348dd | 408 | &s->chunk, s->size*2, |
72ca9bcf | 409 | dv->data_offset, |
af4348dd | 410 | dname, &freesize, |
5308f117 | 411 | s->consistency_policy, |
171dccc8 | 412 | c->verbose >= 0)) { |
17f25ca6 | 413 | |
7a862a02 | 414 | pr_err("%s is not suitable for this array.\n", |
e7b84f9d | 415 | dname); |
17f25ca6 NB |
416 | fail = 1; |
417 | continue; | |
418 | } | |
682c7051 | 419 | } |
82d9eba6 | 420 | |
cc1799c3 | 421 | if (dv->disposition == 'j') |
dfd7822c | 422 | goto skip_size_check; /* skip write journal for size check */ |
cc1799c3 | 423 | |
82d9eba6 | 424 | freesize /= 2; /* convert to K */ |
99cc42f4 | 425 | if (s->chunk && s->chunk != UnSet) { |
d2cd3ffc | 426 | /* round to chunk size */ |
99cc42f4 | 427 | freesize = freesize & ~(s->chunk-1); |
64385908 CA |
428 | if (do_default_chunk) { |
429 | /* default chunk was just set */ | |
171dccc8 | 430 | if (c->verbose > 0) |
7a862a02 | 431 | pr_err("chunk size defaults to %dK\n", s->chunk); |
22dc741f MT |
432 | if (round_size_and_verify(&s->size, s->chunk)) |
433 | return 1; | |
64385908 CA |
434 | do_default_chunk = 0; |
435 | } | |
d2cd3ffc | 436 | } |
066e92f0 LD |
437 | if (!freesize) { |
438 | pr_err("no free space left on %s\n", dname); | |
439 | fail = 1; | |
440 | continue; | |
441 | } | |
682c7051 | 442 | |
99cc42f4 | 443 | if (s->size && freesize < s->size) { |
7a862a02 | 444 | pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n", |
99cc42f4 | 445 | dname, freesize, s->size); |
52826846 | 446 | fail = 1; |
52826846 | 447 | continue; |
682c7051 | 448 | } |
cd29a5c8 NB |
449 | if (maxdisc == NULL || (maxdisc && freesize > maxsize)) { |
450 | maxdisc = dname; | |
52826846 | 451 | maxsize = freesize; |
682c7051 | 452 | } |
cd29a5c8 NB |
453 | if (mindisc ==NULL || (mindisc && freesize < minsize)) { |
454 | mindisc = dname; | |
52826846 | 455 | minsize = freesize; |
682c7051 | 456 | } |
dfd7822c | 457 | skip_size_check: |
171dccc8 | 458 | if (c->runstop != 1 || c->verbose >= 0) { |
37ea3936 | 459 | int fd = open(dname, O_RDONLY); |
98dbf73c | 460 | if (fd < 0) { |
e7b84f9d | 461 | pr_err("Cannot open %s: %s\n", |
b6e63da4 | 462 | dname, strerror(errno)); |
98dbf73c | 463 | fail = 1; |
b6e63da4 NB |
464 | continue; |
465 | } | |
dab6685f NB |
466 | warn |= check_ext2(fd, dname); |
467 | warn |= check_reiser(fd, dname); | |
468 | warn |= check_raid(fd, dname); | |
034b203a TM |
469 | if (strcmp(st->ss->name, "1.x") == 0 && |
470 | st->minor_version >= 1) | |
471 | /* metadata at front */ | |
53ed6ac3 | 472 | warn |= check_partitions(fd, dname, 0, 0); |
cf622ec1 JS |
473 | else if (s->level == 1 || s->level == LEVEL_CONTAINER || |
474 | (s->level == 0 && s->raiddisks == 1)) | |
034b203a | 475 | /* partitions could be meaningful */ |
99cc42f4 | 476 | warn |= check_partitions(fd, dname, freesize*2, s->size*2); |
034b203a TM |
477 | else |
478 | /* partitions cannot be meaningful */ | |
53ed6ac3 | 479 | warn |= check_partitions(fd, dname, 0, 0); |
034b203a | 480 | if (strcmp(st->ss->name, "1.x") == 0 && |
a0962fe9 N |
481 | st->minor_version >= 1 && |
482 | did_default && | |
99cc42f4 | 483 | s->level == 1 && |
034b203a TM |
484 | (warn & 1024) == 0) { |
485 | warn |= 1024; | |
e7b84f9d | 486 | pr_err("Note: this array has metadata at the start and\n" |
a0962fe9 | 487 | " may not be suitable as a boot device. If you plan to\n" |
cc86f89c | 488 | " store '/boot' on this device please ensure that\n" |
a0962fe9 | 489 | " your boot-loader understands md/v1.x metadata, or use\n" |
cc86f89c | 490 | " --metadata=0.90\n"); |
a0962fe9 | 491 | } |
b6e63da4 | 492 | close(fd); |
dab6685f | 493 | } |
682c7051 | 494 | } |
b91ad097 | 495 | if (missing_disks == dnum && !have_container) { |
1db03765 XN |
496 | pr_err("Subdevs can't be all missing\n"); |
497 | return 1; | |
498 | } | |
99cc42f4 | 499 | if (s->raiddisks + s->sparedisks > st->max_devs) { |
7a862a02 | 500 | pr_err("Too many devices: %s metadata only supports %d\n", |
acab7bb1 N |
501 | st->ss->name, st->max_devs); |
502 | return 1; | |
503 | } | |
8592f29d | 504 | if (have_container) |
99cc42f4 | 505 | info.array.working_disks = s->raiddisks; |
682c7051 | 506 | if (fail) { |
e7b84f9d | 507 | pr_err("create aborted\n"); |
52826846 | 508 | return 1; |
682c7051 | 509 | } |
99cc42f4 | 510 | if (s->size == 0) { |
5f8097be | 511 | if (mindisc == NULL && !have_container) { |
e7b84f9d | 512 | pr_err("no size and no drives given - aborting create.\n"); |
52826846 NB |
513 | return 1; |
514 | } | |
cf622ec1 | 515 | if (s->level > 0 || s->level == LEVEL_MULTIPATH || |
98dbf73c | 516 | s->level == LEVEL_FAULTY || st->ss->external) { |
b5e64645 | 517 | /* size is meaningful */ |
99cc42f4 N |
518 | if (!st->ss->validate_geometry(st, s->level, s->layout, |
519 | s->raiddisks, | |
520 | &s->chunk, minsize*2, | |
40c9a66a | 521 | data_offset, |
5308f117 AP |
522 | NULL, NULL, |
523 | s->consistency_policy, 0)) { | |
99cc42f4 | 524 | pr_err("devices too large for RAID level %d\n", s->level); |
b5e64645 NB |
525 | return 1; |
526 | } | |
99cc42f4 N |
527 | s->size = minsize; |
528 | if (s->level == 1) | |
ae6c05ad N |
529 | /* If this is ever reshaped to RAID5, we will |
530 | * need a chunksize. So round it off a bit | |
531 | * now just to be safe | |
532 | */ | |
99cc42f4 | 533 | s->size &= ~(64ULL-1); |
171dccc8 | 534 | if (c->verbose > 0) |
99cc42f4 | 535 | pr_err("size set to %lluK\n", s->size); |
b5e64645 | 536 | } |
682c7051 | 537 | } |
748952f7 N |
538 | |
539 | if (!s->bitmap_file && | |
540 | s->level >= 1 && | |
39917e56 | 541 | st->ss->add_internal_bitmap && |
e97a7cd0 AP |
542 | (s->consistency_policy != CONSISTENCY_POLICY_RESYNC && |
543 | s->consistency_policy != CONSISTENCY_POLICY_PPL) && | |
748952f7 N |
544 | (s->write_behind || s->size > 100*1024*1024ULL)) { |
545 | if (c->verbose > 0) | |
546 | pr_err("automatically enabling write-intent bitmap on large array\n"); | |
547 | s->bitmap_file = "internal"; | |
548 | } | |
549 | if (s->bitmap_file && strcmp(s->bitmap_file, "none") == 0) | |
550 | s->bitmap_file = NULL; | |
551 | ||
5308f117 AP |
552 | if (s->consistency_policy == CONSISTENCY_POLICY_PPL && |
553 | !st->ss->write_init_ppl) { | |
554 | pr_err("%s metadata does not support PPL\n", st->ss->name); | |
555 | return 1; | |
556 | } | |
557 | ||
99cc42f4 | 558 | if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) { |
171dccc8 | 559 | if (c->runstop != 1 || c->verbose >= 0) |
e7b84f9d | 560 | pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n", |
99cc42f4 | 561 | maxdisc, s->size); |
52826846 | 562 | warn = 1; |
682c7051 NB |
563 | } |
564 | ||
9eafa1de | 565 | if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) { |
171dccc8 | 566 | if (c->runstop != 1 || c->verbose >= 0) |
e7b84f9d | 567 | pr_err("%s unable to enumerate platform support\n" |
5615172f DW |
568 | " array may not be compatible with hardware/firmware\n", |
569 | st->ss->name); | |
570 | warn = 1; | |
571 | } | |
529e2aa5 | 572 | st->nodes = c->nodes; |
7716570e | 573 | st->cluster_name = c->homecluster; |
5615172f | 574 | |
682c7051 | 575 | if (warn) { |
171dccc8 | 576 | if (c->runstop!= 1) { |
52826846 | 577 | if (!ask("Continue creating array? ")) { |
e7b84f9d | 578 | pr_err("create aborted.\n"); |
52826846 NB |
579 | return 1; |
580 | } | |
581 | } else { | |
171dccc8 | 582 | if (c->verbose > 0) |
e7b84f9d | 583 | pr_err("creation continuing despite oddities due to --run\n"); |
682c7051 | 584 | } |
682c7051 NB |
585 | } |
586 | ||
66f8bbbe | 587 | /* If this is raid4/5, we want to configure the last active slot |
52826846 | 588 | * as missing, so that a reconstruct happens (faster than re-parity) |
98c6faba | 589 | * FIX: Can we do this for raid6 as well? |
52826846 | 590 | */ |
98dbf73c JS |
591 | if (st->ss->external == 0 && s->assume_clean == 0 && |
592 | c->force == 0 && first_missing >= s->raiddisks) { | |
593 | switch (s->level) { | |
66f8bbbe | 594 | case 4: |
98c6faba | 595 | case 5: |
99cc42f4 N |
596 | insert_point = s->raiddisks-1; |
597 | s->sparedisks++; | |
06c7f68e | 598 | info.array.active_disks--; |
98c6faba NB |
599 | missing_disks++; |
600 | break; | |
601 | default: | |
602 | break; | |
603 | } | |
52826846 | 604 | } |
6fb79233 NB |
605 | /* For raid6, if creating with 1 missing drive, make a good drive |
606 | * into a spare, else the create will fail | |
607 | */ | |
99cc42f4 | 608 | if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks && |
ffcfc735 | 609 | st->ss->external == 0 && |
99cc42f4 N |
610 | second_missing >= s->raiddisks && s->level == 6) { |
611 | insert_point = s->raiddisks - 1; | |
6fb79233 NB |
612 | if (insert_point == first_missing) |
613 | insert_point--; | |
99cc42f4 | 614 | s->sparedisks ++; |
6fb79233 NB |
615 | info.array.active_disks--; |
616 | missing_disks++; | |
617 | } | |
570510ba | 618 | |
99cc42f4 | 619 | if (s->level <= 0 && first_missing < subdevs * 2) { |
e7b84f9d | 620 | pr_err("This level does not support missing devices\n"); |
570510ba NB |
621 | return 1; |
622 | } | |
aba69144 | 623 | |
7f91af49 | 624 | /* We need to create the device */ |
ad5bc697 | 625 | map_lock(&map); |
cd6cbb08 | 626 | mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name, 1); |
e06af9dd JS |
627 | if (mdfd < 0) { |
628 | map_unlock(&map); | |
7f91af49 | 629 | return 1; |
e06af9dd | 630 | } |
3e9df86a AK |
631 | /* verify if chosen_name is not in use, |
632 | * it could be in conflict with already existing device | |
633 | * e.g. container, array | |
634 | */ | |
cf622ec1 JS |
635 | if (strncmp(chosen_name, "/dev/md/", 8) == 0 && |
636 | map_by_name(&map, chosen_name+8) != NULL) { | |
e7b84f9d | 637 | pr_err("Array name %s is in use already.\n", |
3e9df86a AK |
638 | chosen_name); |
639 | close(mdfd); | |
640 | map_unlock(&map); | |
cd6cbb08 | 641 | udev_unblock(); |
3e9df86a AK |
642 | return 1; |
643 | } | |
f2be09f1 | 644 | mddev = chosen_name; |
7f91af49 | 645 | |
5f4cc239 JS |
646 | memset(&inf, 0, sizeof(inf)); |
647 | md_get_array_info(mdfd, &inf); | |
648 | if (inf.working_disks != 0) { | |
649 | pr_err("another array by this name is already running.\n"); | |
e06af9dd | 650 | goto abort_locked; |
7f91af49 | 651 | } |
a04d5763 | 652 | |
682c7051 NB |
653 | /* Ok, lets try some ioctls */ |
654 | ||
99cc42f4 N |
655 | info.array.level = s->level; |
656 | info.array.size = s->size; | |
657 | info.array.raid_disks = s->raiddisks; | |
82b27616 NB |
658 | /* The kernel should *know* what md_minor we are dealing |
659 | * with, but it chooses to trust me instead. Sigh | |
660 | */ | |
06c7f68e | 661 | info.array.md_minor = 0; |
0a6bff09 ZL |
662 | if (fstat_is_blkdev(mdfd, mddev, &rdev)) |
663 | info.array.md_minor = minor(rdev); | |
06c7f68e | 664 | info.array.not_persistent = 0; |
6fb79233 | 665 | |
cf622ec1 JS |
666 | if (((s->level == 4 || s->level == 5) && |
667 | (insert_point < s->raiddisks || first_missing < s->raiddisks)) || | |
668 | (s->level == 6 && (insert_point < s->raiddisks || | |
669 | second_missing < s->raiddisks)) || | |
670 | (s->level <= 0) || s->assume_clean) { | |
06c7f68e | 671 | info.array.state = 1; /* clean, but one+ drive will be missing*/ |
b7528a20 | 672 | info.resync_start = MaxSector; |
103f2410 | 673 | } else { |
06c7f68e | 674 | info.array.state = 0; /* not clean, but no errors */ |
103f2410 NB |
675 | info.resync_start = 0; |
676 | } | |
99cc42f4 | 677 | if (s->level == 10) { |
f9c25f1d NB |
678 | /* for raid10, the bitmap size is the capacity of the array, |
679 | * which is array.size * raid_disks / ncopies; | |
680 | * .. but convert to sectors. | |
681 | */ | |
99cc42f4 N |
682 | int ncopies = ((s->layout>>8) & 255) * (s->layout & 255); |
683 | bitmapsize = s->size * s->raiddisks / ncopies * 2; | |
684 | /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/ | |
f9c25f1d | 685 | } else |
99cc42f4 | 686 | bitmapsize = s->size * 2; |
f9c25f1d | 687 | |
82b27616 NB |
688 | /* There is lots of redundancy in these disk counts, |
689 | * raid_disks is the most meaningful value | |
690 | * it describes the geometry of the array | |
691 | * it is constant | |
692 | * nr_disks is total number of used slots. | |
693 | * it should be raid_disks+spare_disks | |
694 | * spare_disks is the number of extra disks present | |
695 | * see above | |
696 | * active_disks is the number of working disks in | |
697 | * active slots. (With raid_disks) | |
698 | * working_disks is the total number of working disks, | |
699 | * including spares | |
700 | * failed_disks is the number of disks marked failed | |
701 | * | |
5d500228 | 702 | * Ideally, the kernel would keep these (except raid_disks) |
82b27616 NB |
703 | * up-to-date as we ADD_NEW_DISK, but it doesn't (yet). |
704 | * So for now, we assume that all raid and spare | |
705 | * devices will be given. | |
706 | */ | |
99cc42f4 | 707 | info.array.spare_disks=s->sparedisks; |
06c7f68e NB |
708 | info.array.failed_disks=missing_disks; |
709 | info.array.nr_disks = info.array.working_disks | |
710 | + info.array.failed_disks; | |
99cc42f4 N |
711 | info.array.layout = s->layout; |
712 | info.array.chunk_size = s->chunk*1024; | |
82d9eba6 | 713 | |
b3b33eb5 NB |
714 | if (name == NULL || *name == 0) { |
715 | /* base name on mddev */ | |
d1e80164 NB |
716 | /* /dev/md0 -> 0 |
717 | * /dev/md_d0 -> d0 | |
eca944fa | 718 | * /dev/md_foo -> foo |
d1e80164 NB |
719 | * /dev/md/1 -> 1 |
720 | * /dev/md/d1 -> d1 | |
721 | * /dev/md/home -> home | |
722 | * /dev/mdhome -> home | |
723 | */ | |
69207ff6 | 724 | /* FIXME compare this with rules in create_mddev */ |
b3b33eb5 NB |
725 | name = strrchr(mddev, '/'); |
726 | if (name) { | |
727 | name++; | |
98dbf73c JS |
728 | if (strncmp(name, "md_", 3) == 0 && |
729 | strlen(name) > 3 && (name-mddev) == 5 /* /dev/ */) | |
d1e80164 | 730 | name += 3; |
98dbf73c JS |
731 | else if (strncmp(name, "md", 2) == 0 && |
732 | strlen(name) > 2 && isdigit(name[2]) && | |
60248f74 | 733 | (name-mddev) == 5 /* /dev/ */) |
b3b33eb5 NB |
734 | name += 2; |
735 | } | |
736 | } | |
5308f117 | 737 | if (!st->ss->init_super(st, &info.array, s, name, c->homehost, uuid, |
40c9a66a | 738 | data_offset)) |
e06af9dd | 739 | goto abort_locked; |
682c7051 | 740 | |
d2ca6449 | 741 | total_slots = info.array.nr_disks; |
a5d85af7 | 742 | st->ss->getinfo_super(st, &info, NULL); |
dae13137 JS |
743 | if (sysfs_init(&info, mdfd, NULL)) { |
744 | pr_err("unable to initialize sysfs\n"); | |
745 | goto abort_locked; | |
746 | } | |
159c3a1a | 747 | |
171dccc8 | 748 | if (did_default && c->verbose >= 0) { |
3c558363 | 749 | if (is_subarray(info.text_version)) { |
4dd2df09 N |
750 | char devnm[32]; |
751 | char *ep; | |
f7f1b6a1 | 752 | struct mdinfo *mdi; |
f7f1b6a1 | 753 | |
4dd2df09 N |
754 | strncpy(devnm, info.text_version+1, 32); |
755 | devnm[31] = 0; | |
756 | ep = strchr(devnm, '/'); | |
757 | if (ep) | |
758 | *ep = 0; | |
759 | ||
760 | mdi = sysfs_read(-1, devnm, GET_VERSION); | |
f7f1b6a1 | 761 | |
4dd2df09 N |
762 | pr_err("Creating array inside %s container %s\n", |
763 | mdi?mdi->text_version:"managed", devnm); | |
f7f1b6a1 NB |
764 | sysfs_free(mdi); |
765 | } else | |
7a862a02 | 766 | pr_err("Defaulting to version %s metadata\n", info.text_version); |
f7f1b6a1 | 767 | } |
b8ac1967 | 768 | |
4dd2df09 | 769 | map_update(&map, fd2devnm(mdfd), info.text_version, |
a04d5763 | 770 | info.uuid, chosen_name); |
19ad4b2c AP |
771 | /* Keep map locked until devices have been added to array |
772 | * to stop another mdadm from finding and using those devices. | |
773 | */ | |
a04d5763 | 774 | |
98dbf73c JS |
775 | if (s->bitmap_file && (strcmp(s->bitmap_file, "internal") == 0 || |
776 | strcmp(s->bitmap_file, "clustered") == 0)) { | |
ebeb3663 | 777 | if (!st->ss->add_internal_bitmap) { |
e7b84f9d | 778 | pr_err("internal bitmaps not supported with %s metadata\n", |
ebeb3663 | 779 | st->ss->name); |
19ad4b2c | 780 | goto abort_locked; |
ebeb3663 | 781 | } |
2ec2b7e9 JS |
782 | if (st->ss->add_internal_bitmap(st, &s->bitmap_chunk, |
783 | c->delay, s->write_behind, | |
784 | bitmapsize, 1, major_num)) { | |
e7b84f9d | 785 | pr_err("Given bitmap chunk size not supported.\n"); |
19ad4b2c | 786 | goto abort_locked; |
55935d51 | 787 | } |
99cc42f4 | 788 | s->bitmap_file = NULL; |
55935d51 NB |
789 | } |
790 | ||
dae13137 JS |
791 | if (sysfs_init(&info, mdfd, NULL)) { |
792 | pr_err("unable to initialize sysfs\n"); | |
793 | goto abort_locked; | |
794 | } | |
55935d51 | 795 | |
4dd2df09 | 796 | if (st->ss->external && st->container_devnm[0]) { |
f35f2525 N |
797 | /* member */ |
798 | ||
799 | /* When creating a member, we need to be careful | |
800 | * to negotiate with mdmon properly. | |
801 | * If it is already running, we cannot write to | |
802 | * the devices and must ask it to do that part. | |
803 | * If it isn't running, we write to the devices, | |
804 | * and then start it. | |
805 | * We hold an exclusive open on the container | |
806 | * device to make sure mdmon doesn't exit after | |
807 | * we checked that it is running. | |
808 | * | |
809 | * For now, fail if it is already running. | |
810 | */ | |
4dd2df09 | 811 | container_fd = open_dev_excl(st->container_devnm); |
f35f2525 | 812 | if (container_fd < 0) { |
7a862a02 | 813 | pr_err("Cannot get exclusive open on container - weird.\n"); |
19ad4b2c | 814 | goto abort_locked; |
d03373f1 | 815 | } |
4dd2df09 | 816 | if (mdmon_running(st->container_devnm)) { |
171dccc8 | 817 | if (c->verbose) |
7a862a02 | 818 | pr_err("reusing mdmon for %s.\n", |
4dd2df09 | 819 | st->container_devnm); |
f35f2525 N |
820 | st->update_tail = &st->updates; |
821 | } else | |
822 | need_mdmon = 1; | |
823 | } | |
824 | rv = set_array_info(mdfd, st, &info); | |
82d9eba6 | 825 | if (rv) { |
e7b84f9d | 826 | pr_err("failed to set array info for %s: %s\n", |
52826846 | 827 | mddev, strerror(errno)); |
19ad4b2c | 828 | goto abort_locked; |
682c7051 | 829 | } |
52826846 | 830 | |
99cc42f4 | 831 | if (s->bitmap_file) { |
c82f047c | 832 | int uuid[4]; |
55935d51 | 833 | |
3da92f27 | 834 | st->ss->uuid_from_super(st, uuid); |
99cc42f4 N |
835 | if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk, |
836 | c->delay, s->write_behind, | |
f9c25f1d | 837 | bitmapsize, |
943eafef | 838 | major_num)) { |
19ad4b2c | 839 | goto abort_locked; |
c82f047c | 840 | } |
99cc42f4 | 841 | bitmap_fd = open(s->bitmap_file, O_RDWR); |
c82f047c | 842 | if (bitmap_fd < 0) { |
ebf3be99 | 843 | pr_err("weird: %s cannot be opened\n", |
99cc42f4 | 844 | s->bitmap_file); |
19ad4b2c | 845 | goto abort_locked; |
c82f047c NB |
846 | } |
847 | if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
e7b84f9d | 848 | pr_err("Cannot set bitmap file for %s: %s\n", |
c82f047c | 849 | mddev, strerror(errno)); |
19ad4b2c | 850 | goto abort_locked; |
c82f047c NB |
851 | } |
852 | } | |
853 | ||
503975b9 | 854 | infos = xmalloc(sizeof(*infos) * total_slots); |
a7dec3fd | 855 | enable_fds(total_slots); |
98dbf73c | 856 | for (pass = 1; pass <= 2; pass++) { |
a655e550 | 857 | struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */ |
4b1ac34b | 858 | |
98dbf73c JS |
859 | for (dnum = 0, raid_disk_num = 0, dv = devlist; dv; |
860 | dv = (dv->next) ? (dv->next) : moved_disk, dnum++) { | |
4b1ac34b | 861 | int fd; |
d2ca6449 | 862 | struct mdinfo *inf = &infos[dnum]; |
4b1ac34b | 863 | |
d2ca6449 NB |
864 | if (dnum >= total_slots) |
865 | abort(); | |
4b1ac34b | 866 | if (dnum == insert_point) { |
f170a5a9 | 867 | raid_disk_num += 1; |
4b1ac34b | 868 | moved_disk = dv; |
8592f29d | 869 | continue; |
52826846 | 870 | } |
98dbf73c | 871 | if (strcasecmp(dv->devname, "missing") == 0) { |
f170a5a9 | 872 | raid_disk_num += 1; |
111d01fc | 873 | continue; |
f170a5a9 | 874 | } |
8592f29d N |
875 | if (have_container) |
876 | moved_disk = NULL; | |
877 | if (have_container && dnum < info.array.raid_disks - 1) | |
878 | /* repeatedly use the container */ | |
879 | moved_disk = dv; | |
111d01fc | 880 | |
d2ca6449 NB |
881 | switch(pass) { |
882 | case 1: | |
883 | *inf = info; | |
5f8097be | 884 | |
d2ca6449 | 885 | inf->disk.number = dnum; |
cc1799c3 SL |
886 | inf->disk.raid_disk = raid_disk_num++; |
887 | ||
888 | if (dv->disposition == 'j') { | |
889 | inf->disk.raid_disk = MD_DISK_ROLE_JOURNAL; | |
890 | inf->disk.state = (1<<MD_DISK_JOURNAL); | |
891 | raid_disk_num--; | |
892 | } else if (inf->disk.raid_disk < s->raiddisks) | |
d2ca6449 NB |
893 | inf->disk.state = (1<<MD_DISK_ACTIVE) | |
894 | (1<<MD_DISK_SYNC); | |
895 | else | |
896 | inf->disk.state = 0; | |
897 | ||
e22fe3ae | 898 | if (dv->writemostly == FlagSet) |
d2ca6449 | 899 | inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY); |
e22fe3ae | 900 | if (dv->failfast == FlagSet) |
71574efb | 901 | inf->disk.state |= (1<<MD_DISK_FAILFAST); |
d2ca6449 | 902 | |
8592f29d N |
903 | if (have_container) |
904 | fd = -1; | |
905 | else { | |
d1d599ea | 906 | if (st->ss->external && |
4dd2df09 | 907 | st->container_devnm[0]) |
8592f29d N |
908 | fd = open(dv->devname, O_RDWR); |
909 | else | |
910 | fd = open(dv->devname, O_RDWR|O_EXCL); | |
911 | ||
912 | if (fd < 0) { | |
7a862a02 | 913 | pr_err("failed to open %s after earlier success - aborting\n", |
8592f29d | 914 | dv->devname); |
19ad4b2c | 915 | goto abort_locked; |
8592f29d | 916 | } |
0a6bff09 ZL |
917 | if (!fstat_is_blkdev(fd, dv->devname, &rdev)) |
918 | return 1; | |
919 | inf->disk.major = major(rdev); | |
920 | inf->disk.minor = minor(rdev); | |
d2ca6449 | 921 | } |
8592f29d N |
922 | if (fd >= 0) |
923 | remove_partitions(fd); | |
f20c3968 | 924 | if (st->ss->add_to_super(st, &inf->disk, |
72ca9bcf N |
925 | fd, dv->devname, |
926 | dv->data_offset)) { | |
4eb26970 | 927 | ioctl(mdfd, STOP_ARRAY, NULL); |
19ad4b2c | 928 | goto abort_locked; |
4eb26970 | 929 | } |
a5d85af7 | 930 | st->ss->getinfo_super(st, inf, NULL); |
8ed3e5e1 | 931 | safe_mode_delay = inf->safe_mode_delay; |
d2ca6449 | 932 | |
171dccc8 | 933 | if (have_container && c->verbose > 0) |
e7b84f9d | 934 | pr_err("Using %s for device %d\n", |
8592f29d N |
935 | map_dev(inf->disk.major, |
936 | inf->disk.minor, | |
937 | 0), dnum); | |
938 | ||
939 | if (!have_container) { | |
940 | /* getinfo_super might have lost these ... */ | |
0a6bff09 ZL |
941 | inf->disk.major = major(rdev); |
942 | inf->disk.minor = minor(rdev); | |
8592f29d | 943 | } |
4b1ac34b NB |
944 | break; |
945 | case 2: | |
d2ca6449 | 946 | inf->errors = 0; |
4b1ac34b | 947 | |
f35f2525 | 948 | rv = add_disk(mdfd, st, &info, inf); |
d2ca6449 | 949 | |
2503d23b | 950 | if (rv) { |
7a862a02 | 951 | pr_err("ADD_NEW_DISK for %s failed: %s\n", |
e7b84f9d | 952 | dv->devname, strerror(errno)); |
19ad4b2c | 953 | goto abort_locked; |
4b1ac34b | 954 | } |
4b1ac34b NB |
955 | break; |
956 | } | |
8592f29d N |
957 | if (!have_container && |
958 | dv == moved_disk && dnum != insert_point) break; | |
52826846 | 959 | } |
edd8d13c | 960 | if (pass == 1) { |
9b1fb677 DW |
961 | struct mdinfo info_new; |
962 | struct map_ent *me = NULL; | |
963 | ||
964 | /* check to see if the uuid has changed due to these | |
965 | * metadata changes, and if so update the member array | |
966 | * and container uuid. Note ->write_init_super clears | |
967 | * the subarray cursor such that ->getinfo_super once | |
968 | * again returns container info. | |
969 | */ | |
a5d85af7 | 970 | st->ss->getinfo_super(st, &info_new, NULL); |
99cc42f4 | 971 | if (st->ss->external && s->level != LEVEL_CONTAINER && |
9b1fb677 | 972 | !same_uuid(info_new.uuid, info.uuid, 0)) { |
4dd2df09 | 973 | map_update(&map, fd2devnm(mdfd), |
9b1fb677 DW |
974 | info_new.text_version, |
975 | info_new.uuid, chosen_name); | |
4dd2df09 | 976 | me = map_by_devnm(&map, st->container_devnm); |
9b1fb677 DW |
977 | } |
978 | ||
ab80e597 | 979 | if (st->ss->write_init_super(st)) { |
ab80e597 | 980 | st->ss->free_super(st); |
e06af9dd | 981 | goto abort_locked; |
ab80e597 | 982 | } |
9b1fb677 DW |
983 | |
984 | /* update parent container uuid */ | |
985 | if (me) { | |
503975b9 | 986 | char *path = xstrdup(me->path); |
9b1fb677 | 987 | |
a5d85af7 | 988 | st->ss->getinfo_super(st, &info_new, NULL); |
4dd2df09 | 989 | map_update(&map, st->container_devnm, |
9b1fb677 DW |
990 | info_new.text_version, |
991 | info_new.uuid, path); | |
992 | free(path); | |
993 | } | |
9b1fb677 | 994 | |
edd8d13c | 995 | flush_metadata_updates(st); |
6946681d | 996 | st->ss->free_super(st); |
edd8d13c | 997 | } |
5e52ae9e | 998 | } |
19ad4b2c | 999 | map_unlock(&map); |
d2ca6449 | 1000 | free(infos); |
5e52ae9e | 1001 | |
99cc42f4 | 1002 | if (s->level == LEVEL_CONTAINER) { |
97590376 N |
1003 | /* No need to start. But we should signal udev to |
1004 | * create links */ | |
1005 | sysfs_uevent(&info, "change"); | |
171dccc8 | 1006 | if (c->verbose >= 0) |
e7b84f9d | 1007 | pr_err("container %s prepared.\n", mddev); |
a7c6e3fb | 1008 | wait_for(chosen_name, mdfd); |
99cc42f4 | 1009 | } else if (c->runstop == 1 || subdevs >= s->raiddisks) { |
598f0d58 | 1010 | if (st->ss->external) { |
4c821454 | 1011 | int err; |
99cc42f4 | 1012 | switch(s->level) { |
598f0d58 NB |
1013 | case LEVEL_LINEAR: |
1014 | case LEVEL_MULTIPATH: | |
1015 | case 0: | |
4c821454 | 1016 | err = sysfs_set_str(&info, NULL, "array_state", |
171dccc8 | 1017 | c->readonly |
72d566f6 N |
1018 | ? "readonly" |
1019 | : "active"); | |
a931db9e | 1020 | need_mdmon = 0; |
598f0d58 NB |
1021 | break; |
1022 | default: | |
4c821454 N |
1023 | err = sysfs_set_str(&info, NULL, "array_state", |
1024 | "readonly"); | |
598f0d58 NB |
1025 | break; |
1026 | } | |
f35f2525 | 1027 | sysfs_set_safemode(&info, safe_mode_delay); |
4c821454 | 1028 | if (err) { |
7a862a02 | 1029 | pr_err("failed to activate array.\n"); |
4c821454 N |
1030 | ioctl(mdfd, STOP_ARRAY, NULL); |
1031 | goto abort; | |
1032 | } | |
171dccc8 | 1033 | } else if (c->readonly && |
72d566f6 N |
1034 | sysfs_attribute_available( |
1035 | &info, NULL, "array_state")) { | |
1036 | if (sysfs_set_str(&info, NULL, | |
1037 | "array_state", "readonly") < 0) { | |
1038 | pr_err("Failed to start array: %s\n", | |
1039 | strerror(errno)); | |
1040 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1041 | goto abort; | |
1042 | } | |
598f0d58 | 1043 | } else { |
97590376 | 1044 | /* param is not actually used */ |
598f0d58 NB |
1045 | mdu_param_t param; |
1046 | if (ioctl(mdfd, RUN_ARRAY, ¶m)) { | |
e7b84f9d | 1047 | pr_err("RUN_ARRAY failed: %s\n", |
72d566f6 | 1048 | strerror(errno)); |
a252c078 | 1049 | if (info.array.chunk_size & (info.array.chunk_size-1)) { |
7a862a02 | 1050 | cont_err("Problem may be that chunk size is not a power of 2\n"); |
a252c078 | 1051 | } |
4eb26970 | 1052 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1053 | goto abort; |
598f0d58 | 1054 | } |
a21e848a N |
1055 | /* if start_ro module parameter is set, array is |
1056 | * auto-read-only, which is bad as the resync won't | |
1057 | * start. So lets make it read-write now. | |
1058 | */ | |
1059 | ioctl(mdfd, RESTART_ARRAY_RW, NULL); | |
682c7051 | 1060 | } |
171dccc8 | 1061 | if (c->verbose >= 0) |
e7b84f9d | 1062 | pr_err("array %s started.\n", mddev); |
4dd2df09 | 1063 | if (st->ss->external && st->container_devnm[0]) { |
8850ee3e | 1064 | if (need_mdmon) |
4dd2df09 | 1065 | start_mdmon(st->container_devnm); |
8850ee3e | 1066 | |
4dd2df09 | 1067 | ping_monitor(st->container_devnm); |
a931db9e NB |
1068 | close(container_fd); |
1069 | } | |
a7c6e3fb | 1070 | wait_for(chosen_name, mdfd); |
682c7051 | 1071 | } else { |
e7b84f9d | 1072 | pr_err("not starting array - not enough devices.\n"); |
682c7051 | 1073 | } |
7f91af49 | 1074 | close(mdfd); |
cd6cbb08 N |
1075 | /* Give udev a moment to process the Change event caused |
1076 | * by the close. | |
1077 | */ | |
1078 | usleep(100*1000); | |
1079 | udev_unblock(); | |
682c7051 | 1080 | return 0; |
7f91af49 N |
1081 | |
1082 | abort: | |
cd6cbb08 | 1083 | udev_unblock(); |
4eb26970 | 1084 | map_lock(&map); |
e06af9dd | 1085 | abort_locked: |
4dd2df09 | 1086 | map_remove(&map, fd2devnm(mdfd)); |
4eb26970 DW |
1087 | map_unlock(&map); |
1088 | ||
7f91af49 N |
1089 | if (mdfd >= 0) |
1090 | close(mdfd); | |
1091 | return 1; | |
64c4757e | 1092 | } |