]>
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; | |
329dfc28 N |
54 | case 0: |
55 | layout = RAID0_ORIG_LAYOUT; | |
56 | break; | |
a18a888e DW |
57 | case 10: |
58 | layout = 0x102; /* near=2, far=1 */ | |
59 | if (verbose > 0) | |
e7b84f9d | 60 | pr_err("layout defaults to n2\n"); |
a18a888e DW |
61 | break; |
62 | case 5: | |
63 | case 6: | |
64 | layout = map_name(r5layout, "default"); | |
65 | if (verbose > 0) | |
e7b84f9d | 66 | pr_err("layout defaults to %s\n", map_num(r5layout, layout)); |
a18a888e DW |
67 | break; |
68 | case LEVEL_FAULTY: | |
69 | layout = map_name(faultylayout, "default"); | |
70 | ||
71 | if (verbose > 0) | |
e7b84f9d | 72 | pr_err("layout defaults to %s\n", map_num(faultylayout, layout)); |
a18a888e DW |
73 | break; |
74 | } | |
75 | ||
76 | return layout; | |
77 | } | |
78 | ||
7f91af49 | 79 | int Create(struct supertype *st, char *mddev, |
171dccc8 | 80 | char *name, int *uuid, |
a655e550 | 81 | int subdevs, struct mddev_dev *devlist, |
99cc42f4 | 82 | struct shape *s, |
40c9a66a | 83 | struct context *c, unsigned long long data_offset) |
64c4757e | 84 | { |
682c7051 NB |
85 | /* |
86 | * Create a new raid array. | |
87 | * | |
88 | * First check that necessary details are available | |
89 | * (i.e. level, raid-disks) | |
90 | * | |
91 | * Then check each disk to see what might be on it | |
92 | * and report anything interesting. | |
93 | * | |
94 | * If anything looks odd, and runstop not set, | |
95 | * abort. | |
96 | * | |
97 | * SET_ARRAY_INFO and ADD_NEW_DISK, and | |
82d9eba6 | 98 | * if runstop==run, or raiddisks disks were used, |
682c7051 NB |
99 | * RUN_ARRAY |
100 | */ | |
7f91af49 | 101 | int mdfd; |
98dbf73c | 102 | unsigned long long minsize = 0, maxsize = 0; |
cd29a5c8 NB |
103 | char *mindisc = NULL; |
104 | char *maxdisc = NULL; | |
cc1799c3 | 105 | int dnum, raid_disk_num; |
a655e550 | 106 | struct mddev_dev *dv; |
0a6bff09 | 107 | dev_t rdev; |
98dbf73c | 108 | int fail = 0, warn = 0; |
82d9eba6 | 109 | int first_missing = subdevs * 2; |
6fb79233 | 110 | int second_missing = subdevs * 2; |
52826846 | 111 | int missing_disks = 0; |
82d9eba6 | 112 | int insert_point = subdevs * 2; /* where to insert a missing drive */ |
d2ca6449 | 113 | int total_slots; |
4b1ac34b | 114 | int pass; |
82d9eba6 | 115 | int rv; |
c82f047c | 116 | int bitmap_fd; |
5f8097be | 117 | int have_container = 0; |
0e600426 | 118 | int container_fd = -1; |
a931db9e | 119 | int need_mdmon = 0; |
f9c25f1d | 120 | unsigned long long bitmapsize; |
d2ca6449 | 121 | struct mdinfo info, *infos; |
b8ac1967 | 122 | int did_default = 0; |
a18a888e | 123 | int do_default_layout = 0; |
c21e737b | 124 | int do_default_chunk = 0; |
8ed3e5e1 | 125 | unsigned long safe_mode_delay = 0; |
a04d5763 | 126 | char chosen_name[1024]; |
ad5bc697 | 127 | struct map_ent *map = NULL; |
8592f29d | 128 | unsigned long long newsize; |
5f4cc239 | 129 | mdu_array_info_t inf; |
682c7051 | 130 | |
943eafef | 131 | int major_num = BITMAP_MAJOR_HI; |
82d9485e | 132 | if (s->bitmap_file && strcmp(s->bitmap_file, "clustered") == 0) { |
6d9c7c25 | 133 | major_num = BITMAP_MAJOR_CLUSTERED; |
82d9485e GJ |
134 | if (c->nodes <= 1) { |
135 | pr_err("At least 2 nodes are needed for cluster-md\n"); | |
136 | return 1; | |
137 | } | |
138 | } | |
dcec9ee5 | 139 | |
06c7f68e | 140 | memset(&info, 0, sizeof(info)); |
99cc42f4 N |
141 | if (s->level == UnSet && st && st->ss->default_geometry) |
142 | st->ss->default_geometry(st, &s->level, NULL, NULL); | |
143 | if (s->level == UnSet) { | |
e7b84f9d | 144 | pr_err("a RAID level is needed to create an array.\n"); |
682c7051 NB |
145 | return 1; |
146 | } | |
99cc42f4 | 147 | if (s->raiddisks < 4 && s->level == 6) { |
e7b84f9d | 148 | pr_err("at least 4 raid-devices needed for level 6\n"); |
98c6faba NB |
149 | return 1; |
150 | } | |
99cc42f4 | 151 | if (s->raiddisks > 256 && s->level == 6) { |
e7b84f9d | 152 | pr_err("no more than 256 raid-devices supported for level 6\n"); |
98c6faba NB |
153 | return 1; |
154 | } | |
99cc42f4 | 155 | if (s->raiddisks < 2 && s->level >= 4) { |
e7b84f9d | 156 | pr_err("at least 2 raid-devices needed for level 4 or 5\n"); |
52826846 NB |
157 | return 1; |
158 | } | |
99cc42f4 | 159 | if (s->level <= 0 && s->sparedisks) { |
e7b84f9d | 160 | pr_err("This level does not support spare devices\n"); |
570510ba NB |
161 | return 1; |
162 | } | |
5f8097be NB |
163 | |
164 | if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) { | |
165 | /* If given a single device, it might be a container, and we can | |
166 | * extract a device list from there | |
167 | */ | |
5f8097be NB |
168 | int fd; |
169 | ||
170 | memset(&inf, 0, sizeof(inf)); | |
37ea3936 | 171 | fd = open(devlist->devname, O_RDONLY); |
5f8097be | 172 | if (fd >= 0 && |
9cd39f01 | 173 | md_get_array_info(fd, &inf) == 0 && inf.raid_disks == 0) { |
5f8097be NB |
174 | /* yep, looks like a container */ |
175 | if (st) { | |
0fb69d1d N |
176 | rv = st->ss->load_container(st, fd, |
177 | devlist->devname); | |
5f8097be NB |
178 | if (rv == 0) |
179 | have_container = 1; | |
180 | } else { | |
0fb69d1d | 181 | st = super_by_fd(fd, NULL); |
5f8097be | 182 | if (st && !(rv = st->ss-> |
0fb69d1d N |
183 | load_container(st, fd, |
184 | devlist->devname))) | |
5f8097be NB |
185 | have_container = 1; |
186 | else | |
187 | st = NULL; | |
188 | } | |
18fde300 | 189 | if (have_container) { |
99cc42f4 | 190 | subdevs = s->raiddisks; |
18fde300 DW |
191 | first_missing = subdevs * 2; |
192 | second_missing = subdevs * 2; | |
193 | insert_point = subdevs * 2; | |
194 | } | |
5f8097be NB |
195 | } |
196 | if (fd >= 0) | |
197 | close(fd); | |
5f8097be | 198 | } |
99cc42f4 | 199 | if (st && st->ss->external && s->sparedisks) { |
7a862a02 | 200 | pr_err("This metadata type does not support spare disks at create time\n"); |
ffcfc735 N |
201 | return 1; |
202 | } | |
cc1799c3 | 203 | if (subdevs > s->raiddisks+s->sparedisks+s->journaldisks) { |
99cc42f4 | 204 | pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks); |
52826846 | 205 | return 1; |
682c7051 | 206 | } |
cc1799c3 | 207 | if (!have_container && subdevs < s->raiddisks+s->sparedisks+s->journaldisks) { |
e7b84f9d | 208 | pr_err("You haven't given enough devices (real or missing) to create this array\n"); |
52826846 NB |
209 | return 1; |
210 | } | |
99cc42f4 | 211 | if (s->bitmap_file && s->level <= 0) { |
e7b84f9d | 212 | pr_err("bitmaps not meaningful with level %s\n", |
99cc42f4 | 213 | map_num(pers, s->level)?:"given"); |
5b28bd56 NB |
214 | return 1; |
215 | } | |
52826846 | 216 | |
682c7051 | 217 | /* now set some defaults */ |
b5e64645 | 218 | |
99cc42f4 | 219 | if (s->layout == UnSet) { |
a18a888e | 220 | do_default_layout = 1; |
99cc42f4 | 221 | s->layout = default_layout(st, s->level, c->verbose); |
a18a888e | 222 | } |
682c7051 | 223 | |
99cc42f4 | 224 | if (s->level == 10) |
e5329c37 | 225 | /* check layout fits in array*/ |
99cc42f4 | 226 | if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) { |
e7b84f9d | 227 | pr_err("that layout requires at least %d devices\n", |
99cc42f4 | 228 | (s->layout&255) * ((s->layout>>8)&255)); |
e5329c37 NB |
229 | return 1; |
230 | } | |
231 | ||
99cc42f4 | 232 | switch(s->level) { |
aa88f531 NB |
233 | case 4: |
234 | case 5: | |
e5329c37 | 235 | case 10: |
98c6faba | 236 | case 6: |
aa88f531 | 237 | case 0: |
99cc42f4 N |
238 | if (s->chunk == 0 || s->chunk == UnSet) { |
239 | s->chunk = UnSet; | |
c21e737b CA |
240 | do_default_chunk = 1; |
241 | /* chunk will be set later */ | |
bb7295f1 | 242 | } |
5f175898 N |
243 | break; |
244 | case LEVEL_LINEAR: | |
245 | /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */ | |
99cc42f4 N |
246 | if (get_linux_version() < 2006016 && s->chunk == 0) { |
247 | s->chunk = 64; | |
171dccc8 | 248 | if (c->verbose > 0) |
e7b84f9d | 249 | pr_err("chunk size defaults to 64K\n"); |
aa88f531 NB |
250 | } |
251 | break; | |
570510ba | 252 | case 1: |
08e43379 | 253 | case LEVEL_FAULTY: |
570510ba | 254 | case LEVEL_MULTIPATH: |
17f25ca6 | 255 | case LEVEL_CONTAINER: |
99cc42f4 N |
256 | if (s->chunk) { |
257 | s->chunk = 0; | |
171dccc8 | 258 | if (c->verbose > 0) |
e7b84f9d | 259 | pr_err("chunk size ignored for this level\n"); |
aa88f531 NB |
260 | } |
261 | break; | |
570510ba | 262 | default: |
99cc42f4 | 263 | pr_err("unknown level %d\n", s->level); |
570510ba | 264 | return 1; |
682c7051 | 265 | } |
22dc741f | 266 | |
99cc42f4 | 267 | if (s->size == MAX_SIZE) |
d04f65f4 | 268 | /* use '0' to mean 'max' now... */ |
99cc42f4 N |
269 | s->size = 0; |
270 | if (s->size && s->chunk && s->chunk != UnSet) | |
22dc741f MT |
271 | if (round_size_and_verify(&s->size, s->chunk)) |
272 | return 1; | |
273 | ||
99cc42f4 N |
274 | newsize = s->size * 2; |
275 | if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks, | |
af4348dd | 276 | &s->chunk, s->size*2, |
40c9a66a | 277 | data_offset, NULL, |
5308f117 | 278 | &newsize, s->consistency_policy, |
98dbf73c | 279 | c->verbose >= 0)) |
17f25ca6 | 280 | return 1; |
bb7295f1 | 281 | |
99cc42f4 N |
282 | if (s->chunk && s->chunk != UnSet) { |
283 | newsize &= ~(unsigned long long)(s->chunk*2 - 1); | |
64385908 CA |
284 | if (do_default_chunk) { |
285 | /* default chunk was just set */ | |
171dccc8 | 286 | if (c->verbose > 0) |
7a862a02 | 287 | pr_err("chunk size defaults to %dK\n", s->chunk); |
22dc741f MT |
288 | if (round_size_and_verify(&s->size, s->chunk)) |
289 | return 1; | |
64385908 CA |
290 | do_default_chunk = 0; |
291 | } | |
bb7295f1 | 292 | } |
64385908 | 293 | |
99cc42f4 N |
294 | if (s->size == 0) { |
295 | s->size = newsize / 2; | |
296 | if (s->level == 1) | |
ae6c05ad N |
297 | /* If this is ever reshaped to RAID5, we will |
298 | * need a chunksize. So round it off a bit | |
299 | * now just to be safe | |
300 | */ | |
99cc42f4 | 301 | s->size &= ~(64ULL-1); |
ae6c05ad | 302 | |
99cc42f4 N |
303 | if (s->size && c->verbose > 0) |
304 | pr_err("setting size to %lluK\n", s->size); | |
8592f29d | 305 | } |
17f25ca6 | 306 | |
682c7051 | 307 | /* now look at the subdevs */ |
06c7f68e NB |
308 | info.array.active_disks = 0; |
309 | info.array.working_disks = 0; | |
c913b90e | 310 | dnum = 0; |
98dbf73c | 311 | for (dv = devlist; dv; dv = dv->next) |
476066a3 N |
312 | if (data_offset == VARIABLE_OFFSET) |
313 | dv->data_offset = INVALID_SECTORS; | |
314 | else | |
315 | dv->data_offset = data_offset; | |
316 | ||
8592f29d | 317 | for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) { |
cd29a5c8 | 318 | char *dname = dv->devname; |
17f25ca6 | 319 | unsigned long long freesize; |
e9b11fee | 320 | int dfd; |
72ca9bcf | 321 | char *doff; |
e9b11fee | 322 | |
98dbf73c | 323 | if (strcasecmp(dname, "missing") == 0) { |
c913b90e NB |
324 | if (first_missing > dnum) |
325 | first_missing = dnum; | |
6fb79233 NB |
326 | if (second_missing > dnum && dnum > first_missing) |
327 | second_missing = dnum; | |
52826846 NB |
328 | missing_disks ++; |
329 | continue; | |
330 | } | |
8baab049 | 331 | if (data_offset == VARIABLE_OFFSET) { |
72ca9bcf N |
332 | doff = strchr(dname, ':'); |
333 | if (doff) { | |
334 | *doff++ = 0; | |
335 | dv->data_offset = parse_size(doff); | |
336 | } else | |
337 | dv->data_offset = INVALID_SECTORS; | |
338 | } else | |
339 | dv->data_offset = data_offset; | |
340 | ||
58b3c697 | 341 | dfd = open(dname, O_RDONLY); |
e9b11fee | 342 | if (dfd < 0) { |
e7b84f9d | 343 | pr_err("cannot open %s: %s\n", |
e9b11fee N |
344 | dname, strerror(errno)); |
345 | exit(2); | |
346 | } | |
0a6bff09 | 347 | if (!fstat_is_blkdev(dfd, dname, NULL)) { |
e9b11fee | 348 | close(dfd); |
e9b11fee N |
349 | exit(2); |
350 | } | |
351 | close(dfd); | |
06c7f68e | 352 | info.array.working_disks++; |
dfd7822c | 353 | if (dnum < s->raiddisks && dv->disposition != 'j') |
06c7f68e | 354 | info.array.active_disks++; |
058574b1 | 355 | if (st == NULL) { |
8aec876d | 356 | struct createinfo *ci = conf_get_create_info(); |
058574b1 NB |
357 | if (ci) |
358 | st = ci->supertype; | |
359 | } | |
576d6d83 NB |
360 | if (st == NULL) { |
361 | /* Need to choose a default metadata, which is different | |
17f25ca6 | 362 | * depending on geometry of array. |
576d6d83 NB |
363 | */ |
364 | int i; | |
365 | char *name = "default"; | |
98dbf73c | 366 | for(i = 0; !st && superlist[i]; i++) { |
576d6d83 | 367 | st = superlist[i]->match_metadata_desc(name); |
ecbd9e81 N |
368 | if (!st) |
369 | continue; | |
a18a888e | 370 | if (do_default_layout) |
99cc42f4 | 371 | s->layout = default_layout(st, s->level, c->verbose); |
ecbd9e81 | 372 | switch (st->ss->validate_geometry( |
99cc42f4 | 373 | st, s->level, s->layout, s->raiddisks, |
af4348dd | 374 | &s->chunk, s->size*2, |
72ca9bcf | 375 | dv->data_offset, dname, |
5308f117 AP |
376 | &freesize, s->consistency_policy, |
377 | c->verbose > 0)) { | |
ecbd9e81 N |
378 | case -1: /* Not valid, message printed, and not |
379 | * worth checking any further */ | |
380 | exit(2); | |
381 | break; | |
382 | case 0: /* Geometry not valid */ | |
55425f27 | 383 | free(st); |
17f25ca6 | 384 | st = NULL; |
99cc42f4 | 385 | s->chunk = do_default_chunk ? UnSet : s->chunk; |
ecbd9e81 N |
386 | break; |
387 | case 1: /* All happy */ | |
388 | break; | |
55425f27 | 389 | } |
17f25ca6 | 390 | } |
576d6d83 NB |
391 | |
392 | if (!st) { | |
58b3c697 N |
393 | int dfd = open(dname, O_RDONLY|O_EXCL); |
394 | if (dfd < 0) { | |
e7b84f9d | 395 | pr_err("cannot open %s: %s\n", |
58b3c697 N |
396 | dname, strerror(errno)); |
397 | exit(2); | |
398 | } | |
7a862a02 | 399 | pr_err("device %s not suitable for any style of array\n", |
17f25ca6 | 400 | dname); |
576d6d83 NB |
401 | exit(2); |
402 | } | |
b8ac1967 | 403 | if (st->ss != &super0 || |
576d6d83 | 404 | st->minor_version != 90) |
b8ac1967 | 405 | did_default = 1; |
17f25ca6 | 406 | } else { |
a18a888e | 407 | if (do_default_layout) |
99cc42f4 N |
408 | s->layout = default_layout(st, s->level, 0); |
409 | if (!st->ss->validate_geometry(st, s->level, s->layout, | |
410 | s->raiddisks, | |
af4348dd | 411 | &s->chunk, s->size*2, |
72ca9bcf | 412 | dv->data_offset, |
af4348dd | 413 | dname, &freesize, |
5308f117 | 414 | s->consistency_policy, |
171dccc8 | 415 | c->verbose >= 0)) { |
17f25ca6 | 416 | |
7a862a02 | 417 | pr_err("%s is not suitable for this array.\n", |
e7b84f9d | 418 | dname); |
17f25ca6 NB |
419 | fail = 1; |
420 | continue; | |
421 | } | |
682c7051 | 422 | } |
82d9eba6 | 423 | |
cc1799c3 | 424 | if (dv->disposition == 'j') |
dfd7822c | 425 | goto skip_size_check; /* skip write journal for size check */ |
cc1799c3 | 426 | |
82d9eba6 | 427 | freesize /= 2; /* convert to K */ |
99cc42f4 | 428 | if (s->chunk && s->chunk != UnSet) { |
d2cd3ffc | 429 | /* round to chunk size */ |
99cc42f4 | 430 | freesize = freesize & ~(s->chunk-1); |
64385908 CA |
431 | if (do_default_chunk) { |
432 | /* default chunk was just set */ | |
171dccc8 | 433 | if (c->verbose > 0) |
7a862a02 | 434 | pr_err("chunk size defaults to %dK\n", s->chunk); |
22dc741f MT |
435 | if (round_size_and_verify(&s->size, s->chunk)) |
436 | return 1; | |
64385908 CA |
437 | do_default_chunk = 0; |
438 | } | |
d2cd3ffc | 439 | } |
066e92f0 LD |
440 | if (!freesize) { |
441 | pr_err("no free space left on %s\n", dname); | |
442 | fail = 1; | |
443 | continue; | |
444 | } | |
682c7051 | 445 | |
99cc42f4 | 446 | if (s->size && freesize < s->size) { |
7a862a02 | 447 | pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n", |
99cc42f4 | 448 | dname, freesize, s->size); |
52826846 | 449 | fail = 1; |
52826846 | 450 | continue; |
682c7051 | 451 | } |
cd29a5c8 NB |
452 | if (maxdisc == NULL || (maxdisc && freesize > maxsize)) { |
453 | maxdisc = dname; | |
52826846 | 454 | maxsize = freesize; |
682c7051 | 455 | } |
cd29a5c8 NB |
456 | if (mindisc ==NULL || (mindisc && freesize < minsize)) { |
457 | mindisc = dname; | |
52826846 | 458 | minsize = freesize; |
682c7051 | 459 | } |
dfd7822c | 460 | skip_size_check: |
171dccc8 | 461 | if (c->runstop != 1 || c->verbose >= 0) { |
37ea3936 | 462 | int fd = open(dname, O_RDONLY); |
98dbf73c | 463 | if (fd < 0) { |
e7b84f9d | 464 | pr_err("Cannot open %s: %s\n", |
b6e63da4 | 465 | dname, strerror(errno)); |
98dbf73c | 466 | fail = 1; |
b6e63da4 NB |
467 | continue; |
468 | } | |
dab6685f NB |
469 | warn |= check_ext2(fd, dname); |
470 | warn |= check_reiser(fd, dname); | |
471 | warn |= check_raid(fd, dname); | |
034b203a TM |
472 | if (strcmp(st->ss->name, "1.x") == 0 && |
473 | st->minor_version >= 1) | |
474 | /* metadata at front */ | |
53ed6ac3 | 475 | warn |= check_partitions(fd, dname, 0, 0); |
cf622ec1 JS |
476 | else if (s->level == 1 || s->level == LEVEL_CONTAINER || |
477 | (s->level == 0 && s->raiddisks == 1)) | |
034b203a | 478 | /* partitions could be meaningful */ |
99cc42f4 | 479 | warn |= check_partitions(fd, dname, freesize*2, s->size*2); |
034b203a TM |
480 | else |
481 | /* partitions cannot be meaningful */ | |
53ed6ac3 | 482 | warn |= check_partitions(fd, dname, 0, 0); |
034b203a | 483 | if (strcmp(st->ss->name, "1.x") == 0 && |
a0962fe9 N |
484 | st->minor_version >= 1 && |
485 | did_default && | |
99cc42f4 | 486 | s->level == 1 && |
034b203a TM |
487 | (warn & 1024) == 0) { |
488 | warn |= 1024; | |
e7b84f9d | 489 | pr_err("Note: this array has metadata at the start and\n" |
a0962fe9 | 490 | " may not be suitable as a boot device. If you plan to\n" |
cc86f89c | 491 | " store '/boot' on this device please ensure that\n" |
a0962fe9 | 492 | " your boot-loader understands md/v1.x metadata, or use\n" |
cc86f89c | 493 | " --metadata=0.90\n"); |
a0962fe9 | 494 | } |
b6e63da4 | 495 | close(fd); |
dab6685f | 496 | } |
682c7051 | 497 | } |
b91ad097 | 498 | if (missing_disks == dnum && !have_container) { |
1db03765 XN |
499 | pr_err("Subdevs can't be all missing\n"); |
500 | return 1; | |
501 | } | |
99cc42f4 | 502 | if (s->raiddisks + s->sparedisks > st->max_devs) { |
7a862a02 | 503 | pr_err("Too many devices: %s metadata only supports %d\n", |
acab7bb1 N |
504 | st->ss->name, st->max_devs); |
505 | return 1; | |
506 | } | |
8592f29d | 507 | if (have_container) |
99cc42f4 | 508 | info.array.working_disks = s->raiddisks; |
682c7051 | 509 | if (fail) { |
e7b84f9d | 510 | pr_err("create aborted\n"); |
52826846 | 511 | return 1; |
682c7051 | 512 | } |
99cc42f4 | 513 | if (s->size == 0) { |
5f8097be | 514 | if (mindisc == NULL && !have_container) { |
e7b84f9d | 515 | pr_err("no size and no drives given - aborting create.\n"); |
52826846 NB |
516 | return 1; |
517 | } | |
cf622ec1 | 518 | if (s->level > 0 || s->level == LEVEL_MULTIPATH || |
98dbf73c | 519 | s->level == LEVEL_FAULTY || st->ss->external) { |
b5e64645 | 520 | /* size is meaningful */ |
99cc42f4 N |
521 | if (!st->ss->validate_geometry(st, s->level, s->layout, |
522 | s->raiddisks, | |
523 | &s->chunk, minsize*2, | |
40c9a66a | 524 | data_offset, |
5308f117 AP |
525 | NULL, NULL, |
526 | s->consistency_policy, 0)) { | |
99cc42f4 | 527 | pr_err("devices too large for RAID level %d\n", s->level); |
b5e64645 NB |
528 | return 1; |
529 | } | |
99cc42f4 N |
530 | s->size = minsize; |
531 | if (s->level == 1) | |
ae6c05ad N |
532 | /* If this is ever reshaped to RAID5, we will |
533 | * need a chunksize. So round it off a bit | |
534 | * now just to be safe | |
535 | */ | |
99cc42f4 | 536 | s->size &= ~(64ULL-1); |
171dccc8 | 537 | if (c->verbose > 0) |
99cc42f4 | 538 | pr_err("size set to %lluK\n", s->size); |
b5e64645 | 539 | } |
682c7051 | 540 | } |
748952f7 N |
541 | |
542 | if (!s->bitmap_file && | |
543 | s->level >= 1 && | |
39917e56 | 544 | st->ss->add_internal_bitmap && |
2ce09172 | 545 | s->journaldisks == 0 && |
e97a7cd0 AP |
546 | (s->consistency_policy != CONSISTENCY_POLICY_RESYNC && |
547 | s->consistency_policy != CONSISTENCY_POLICY_PPL) && | |
748952f7 N |
548 | (s->write_behind || s->size > 100*1024*1024ULL)) { |
549 | if (c->verbose > 0) | |
550 | pr_err("automatically enabling write-intent bitmap on large array\n"); | |
551 | s->bitmap_file = "internal"; | |
552 | } | |
553 | if (s->bitmap_file && strcmp(s->bitmap_file, "none") == 0) | |
554 | s->bitmap_file = NULL; | |
555 | ||
5308f117 AP |
556 | if (s->consistency_policy == CONSISTENCY_POLICY_PPL && |
557 | !st->ss->write_init_ppl) { | |
558 | pr_err("%s metadata does not support PPL\n", st->ss->name); | |
559 | return 1; | |
560 | } | |
561 | ||
99cc42f4 | 562 | if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) { |
171dccc8 | 563 | if (c->runstop != 1 || c->verbose >= 0) |
e7b84f9d | 564 | pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n", |
99cc42f4 | 565 | maxdisc, s->size); |
52826846 | 566 | warn = 1; |
682c7051 NB |
567 | } |
568 | ||
9eafa1de | 569 | if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) { |
171dccc8 | 570 | if (c->runstop != 1 || c->verbose >= 0) |
e7b84f9d | 571 | pr_err("%s unable to enumerate platform support\n" |
5615172f DW |
572 | " array may not be compatible with hardware/firmware\n", |
573 | st->ss->name); | |
574 | warn = 1; | |
575 | } | |
529e2aa5 | 576 | st->nodes = c->nodes; |
7716570e | 577 | st->cluster_name = c->homecluster; |
5615172f | 578 | |
682c7051 | 579 | if (warn) { |
171dccc8 | 580 | if (c->runstop!= 1) { |
52826846 | 581 | if (!ask("Continue creating array? ")) { |
e7b84f9d | 582 | pr_err("create aborted.\n"); |
52826846 NB |
583 | return 1; |
584 | } | |
585 | } else { | |
171dccc8 | 586 | if (c->verbose > 0) |
e7b84f9d | 587 | pr_err("creation continuing despite oddities due to --run\n"); |
682c7051 | 588 | } |
682c7051 NB |
589 | } |
590 | ||
66f8bbbe | 591 | /* If this is raid4/5, we want to configure the last active slot |
52826846 | 592 | * as missing, so that a reconstruct happens (faster than re-parity) |
98c6faba | 593 | * FIX: Can we do this for raid6 as well? |
52826846 | 594 | */ |
98dbf73c JS |
595 | if (st->ss->external == 0 && s->assume_clean == 0 && |
596 | c->force == 0 && first_missing >= s->raiddisks) { | |
597 | switch (s->level) { | |
66f8bbbe | 598 | case 4: |
98c6faba | 599 | case 5: |
99cc42f4 N |
600 | insert_point = s->raiddisks-1; |
601 | s->sparedisks++; | |
06c7f68e | 602 | info.array.active_disks--; |
98c6faba NB |
603 | missing_disks++; |
604 | break; | |
605 | default: | |
606 | break; | |
607 | } | |
52826846 | 608 | } |
6fb79233 NB |
609 | /* For raid6, if creating with 1 missing drive, make a good drive |
610 | * into a spare, else the create will fail | |
611 | */ | |
99cc42f4 | 612 | if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks && |
ffcfc735 | 613 | st->ss->external == 0 && |
99cc42f4 N |
614 | second_missing >= s->raiddisks && s->level == 6) { |
615 | insert_point = s->raiddisks - 1; | |
6fb79233 NB |
616 | if (insert_point == first_missing) |
617 | insert_point--; | |
99cc42f4 | 618 | s->sparedisks ++; |
6fb79233 NB |
619 | info.array.active_disks--; |
620 | missing_disks++; | |
621 | } | |
570510ba | 622 | |
99cc42f4 | 623 | if (s->level <= 0 && first_missing < subdevs * 2) { |
e7b84f9d | 624 | pr_err("This level does not support missing devices\n"); |
570510ba NB |
625 | return 1; |
626 | } | |
aba69144 | 627 | |
7f91af49 | 628 | /* We need to create the device */ |
ad5bc697 | 629 | map_lock(&map); |
cd6cbb08 | 630 | mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name, 1); |
e06af9dd JS |
631 | if (mdfd < 0) { |
632 | map_unlock(&map); | |
7f91af49 | 633 | return 1; |
e06af9dd | 634 | } |
3e9df86a AK |
635 | /* verify if chosen_name is not in use, |
636 | * it could be in conflict with already existing device | |
637 | * e.g. container, array | |
638 | */ | |
cf622ec1 JS |
639 | if (strncmp(chosen_name, "/dev/md/", 8) == 0 && |
640 | map_by_name(&map, chosen_name+8) != NULL) { | |
e7b84f9d | 641 | pr_err("Array name %s is in use already.\n", |
3e9df86a AK |
642 | chosen_name); |
643 | close(mdfd); | |
644 | map_unlock(&map); | |
cd6cbb08 | 645 | udev_unblock(); |
3e9df86a AK |
646 | return 1; |
647 | } | |
f2be09f1 | 648 | mddev = chosen_name; |
7f91af49 | 649 | |
5f4cc239 JS |
650 | memset(&inf, 0, sizeof(inf)); |
651 | md_get_array_info(mdfd, &inf); | |
652 | if (inf.working_disks != 0) { | |
653 | pr_err("another array by this name is already running.\n"); | |
e06af9dd | 654 | goto abort_locked; |
7f91af49 | 655 | } |
a04d5763 | 656 | |
682c7051 NB |
657 | /* Ok, lets try some ioctls */ |
658 | ||
99cc42f4 N |
659 | info.array.level = s->level; |
660 | info.array.size = s->size; | |
661 | info.array.raid_disks = s->raiddisks; | |
82b27616 NB |
662 | /* The kernel should *know* what md_minor we are dealing |
663 | * with, but it chooses to trust me instead. Sigh | |
664 | */ | |
06c7f68e | 665 | info.array.md_minor = 0; |
0a6bff09 ZL |
666 | if (fstat_is_blkdev(mdfd, mddev, &rdev)) |
667 | info.array.md_minor = minor(rdev); | |
06c7f68e | 668 | info.array.not_persistent = 0; |
6fb79233 | 669 | |
cf622ec1 JS |
670 | if (((s->level == 4 || s->level == 5) && |
671 | (insert_point < s->raiddisks || first_missing < s->raiddisks)) || | |
672 | (s->level == 6 && (insert_point < s->raiddisks || | |
673 | second_missing < s->raiddisks)) || | |
674 | (s->level <= 0) || s->assume_clean) { | |
06c7f68e | 675 | info.array.state = 1; /* clean, but one+ drive will be missing*/ |
b7528a20 | 676 | info.resync_start = MaxSector; |
103f2410 | 677 | } else { |
06c7f68e | 678 | info.array.state = 0; /* not clean, but no errors */ |
103f2410 NB |
679 | info.resync_start = 0; |
680 | } | |
99cc42f4 | 681 | if (s->level == 10) { |
f9c25f1d NB |
682 | /* for raid10, the bitmap size is the capacity of the array, |
683 | * which is array.size * raid_disks / ncopies; | |
684 | * .. but convert to sectors. | |
685 | */ | |
99cc42f4 N |
686 | int ncopies = ((s->layout>>8) & 255) * (s->layout & 255); |
687 | bitmapsize = s->size * s->raiddisks / ncopies * 2; | |
688 | /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/ | |
f9c25f1d | 689 | } else |
99cc42f4 | 690 | bitmapsize = s->size * 2; |
f9c25f1d | 691 | |
82b27616 NB |
692 | /* There is lots of redundancy in these disk counts, |
693 | * raid_disks is the most meaningful value | |
694 | * it describes the geometry of the array | |
695 | * it is constant | |
696 | * nr_disks is total number of used slots. | |
697 | * it should be raid_disks+spare_disks | |
698 | * spare_disks is the number of extra disks present | |
699 | * see above | |
700 | * active_disks is the number of working disks in | |
701 | * active slots. (With raid_disks) | |
702 | * working_disks is the total number of working disks, | |
703 | * including spares | |
704 | * failed_disks is the number of disks marked failed | |
705 | * | |
5d500228 | 706 | * Ideally, the kernel would keep these (except raid_disks) |
82b27616 NB |
707 | * up-to-date as we ADD_NEW_DISK, but it doesn't (yet). |
708 | * So for now, we assume that all raid and spare | |
709 | * devices will be given. | |
710 | */ | |
99cc42f4 | 711 | info.array.spare_disks=s->sparedisks; |
06c7f68e NB |
712 | info.array.failed_disks=missing_disks; |
713 | info.array.nr_disks = info.array.working_disks | |
714 | + info.array.failed_disks; | |
99cc42f4 N |
715 | info.array.layout = s->layout; |
716 | info.array.chunk_size = s->chunk*1024; | |
82d9eba6 | 717 | |
b3b33eb5 NB |
718 | if (name == NULL || *name == 0) { |
719 | /* base name on mddev */ | |
d1e80164 NB |
720 | /* /dev/md0 -> 0 |
721 | * /dev/md_d0 -> d0 | |
eca944fa | 722 | * /dev/md_foo -> foo |
d1e80164 NB |
723 | * /dev/md/1 -> 1 |
724 | * /dev/md/d1 -> d1 | |
725 | * /dev/md/home -> home | |
726 | * /dev/mdhome -> home | |
727 | */ | |
69207ff6 | 728 | /* FIXME compare this with rules in create_mddev */ |
b3b33eb5 NB |
729 | name = strrchr(mddev, '/'); |
730 | if (name) { | |
731 | name++; | |
98dbf73c JS |
732 | if (strncmp(name, "md_", 3) == 0 && |
733 | strlen(name) > 3 && (name-mddev) == 5 /* /dev/ */) | |
d1e80164 | 734 | name += 3; |
98dbf73c JS |
735 | else if (strncmp(name, "md", 2) == 0 && |
736 | strlen(name) > 2 && isdigit(name[2]) && | |
60248f74 | 737 | (name-mddev) == 5 /* /dev/ */) |
b3b33eb5 NB |
738 | name += 2; |
739 | } | |
740 | } | |
5308f117 | 741 | if (!st->ss->init_super(st, &info.array, s, name, c->homehost, uuid, |
40c9a66a | 742 | data_offset)) |
e06af9dd | 743 | goto abort_locked; |
682c7051 | 744 | |
d2ca6449 | 745 | total_slots = info.array.nr_disks; |
a5d85af7 | 746 | st->ss->getinfo_super(st, &info, NULL); |
dae13137 JS |
747 | if (sysfs_init(&info, mdfd, NULL)) { |
748 | pr_err("unable to initialize sysfs\n"); | |
749 | goto abort_locked; | |
750 | } | |
159c3a1a | 751 | |
171dccc8 | 752 | if (did_default && c->verbose >= 0) { |
3c558363 | 753 | if (is_subarray(info.text_version)) { |
4dd2df09 N |
754 | char devnm[32]; |
755 | char *ep; | |
f7f1b6a1 | 756 | struct mdinfo *mdi; |
f7f1b6a1 | 757 | |
4dd2df09 N |
758 | strncpy(devnm, info.text_version+1, 32); |
759 | devnm[31] = 0; | |
760 | ep = strchr(devnm, '/'); | |
761 | if (ep) | |
762 | *ep = 0; | |
763 | ||
764 | mdi = sysfs_read(-1, devnm, GET_VERSION); | |
f7f1b6a1 | 765 | |
4dd2df09 N |
766 | pr_err("Creating array inside %s container %s\n", |
767 | mdi?mdi->text_version:"managed", devnm); | |
f7f1b6a1 NB |
768 | sysfs_free(mdi); |
769 | } else | |
7a862a02 | 770 | pr_err("Defaulting to version %s metadata\n", info.text_version); |
f7f1b6a1 | 771 | } |
b8ac1967 | 772 | |
4dd2df09 | 773 | map_update(&map, fd2devnm(mdfd), info.text_version, |
a04d5763 | 774 | info.uuid, chosen_name); |
19ad4b2c AP |
775 | /* Keep map locked until devices have been added to array |
776 | * to stop another mdadm from finding and using those devices. | |
777 | */ | |
a04d5763 | 778 | |
98dbf73c JS |
779 | if (s->bitmap_file && (strcmp(s->bitmap_file, "internal") == 0 || |
780 | strcmp(s->bitmap_file, "clustered") == 0)) { | |
ebeb3663 | 781 | if (!st->ss->add_internal_bitmap) { |
e7b84f9d | 782 | pr_err("internal bitmaps not supported with %s metadata\n", |
ebeb3663 | 783 | st->ss->name); |
19ad4b2c | 784 | goto abort_locked; |
ebeb3663 | 785 | } |
2ec2b7e9 JS |
786 | if (st->ss->add_internal_bitmap(st, &s->bitmap_chunk, |
787 | c->delay, s->write_behind, | |
788 | bitmapsize, 1, major_num)) { | |
e7b84f9d | 789 | pr_err("Given bitmap chunk size not supported.\n"); |
19ad4b2c | 790 | goto abort_locked; |
55935d51 | 791 | } |
99cc42f4 | 792 | s->bitmap_file = NULL; |
55935d51 NB |
793 | } |
794 | ||
dae13137 JS |
795 | if (sysfs_init(&info, mdfd, NULL)) { |
796 | pr_err("unable to initialize sysfs\n"); | |
797 | goto abort_locked; | |
798 | } | |
55935d51 | 799 | |
4dd2df09 | 800 | if (st->ss->external && st->container_devnm[0]) { |
f35f2525 N |
801 | /* member */ |
802 | ||
803 | /* When creating a member, we need to be careful | |
804 | * to negotiate with mdmon properly. | |
805 | * If it is already running, we cannot write to | |
806 | * the devices and must ask it to do that part. | |
807 | * If it isn't running, we write to the devices, | |
808 | * and then start it. | |
809 | * We hold an exclusive open on the container | |
810 | * device to make sure mdmon doesn't exit after | |
811 | * we checked that it is running. | |
812 | * | |
813 | * For now, fail if it is already running. | |
814 | */ | |
4dd2df09 | 815 | container_fd = open_dev_excl(st->container_devnm); |
f35f2525 | 816 | if (container_fd < 0) { |
7a862a02 | 817 | pr_err("Cannot get exclusive open on container - weird.\n"); |
19ad4b2c | 818 | goto abort_locked; |
d03373f1 | 819 | } |
4dd2df09 | 820 | if (mdmon_running(st->container_devnm)) { |
171dccc8 | 821 | if (c->verbose) |
7a862a02 | 822 | pr_err("reusing mdmon for %s.\n", |
4dd2df09 | 823 | st->container_devnm); |
f35f2525 N |
824 | st->update_tail = &st->updates; |
825 | } else | |
826 | need_mdmon = 1; | |
827 | } | |
828 | rv = set_array_info(mdfd, st, &info); | |
82d9eba6 | 829 | if (rv) { |
e7b84f9d | 830 | pr_err("failed to set array info for %s: %s\n", |
52826846 | 831 | mddev, strerror(errno)); |
19ad4b2c | 832 | goto abort_locked; |
682c7051 | 833 | } |
52826846 | 834 | |
99cc42f4 | 835 | if (s->bitmap_file) { |
c82f047c | 836 | int uuid[4]; |
55935d51 | 837 | |
3da92f27 | 838 | st->ss->uuid_from_super(st, uuid); |
99cc42f4 N |
839 | if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk, |
840 | c->delay, s->write_behind, | |
f9c25f1d | 841 | bitmapsize, |
943eafef | 842 | major_num)) { |
19ad4b2c | 843 | goto abort_locked; |
c82f047c | 844 | } |
99cc42f4 | 845 | bitmap_fd = open(s->bitmap_file, O_RDWR); |
c82f047c | 846 | if (bitmap_fd < 0) { |
ebf3be99 | 847 | pr_err("weird: %s cannot be opened\n", |
99cc42f4 | 848 | s->bitmap_file); |
19ad4b2c | 849 | goto abort_locked; |
c82f047c NB |
850 | } |
851 | if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
e7b84f9d | 852 | pr_err("Cannot set bitmap file for %s: %s\n", |
c82f047c | 853 | mddev, strerror(errno)); |
19ad4b2c | 854 | goto abort_locked; |
c82f047c NB |
855 | } |
856 | } | |
857 | ||
503975b9 | 858 | infos = xmalloc(sizeof(*infos) * total_slots); |
a7dec3fd | 859 | enable_fds(total_slots); |
98dbf73c | 860 | for (pass = 1; pass <= 2; pass++) { |
a655e550 | 861 | struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */ |
4b1ac34b | 862 | |
98dbf73c JS |
863 | for (dnum = 0, raid_disk_num = 0, dv = devlist; dv; |
864 | dv = (dv->next) ? (dv->next) : moved_disk, dnum++) { | |
4b1ac34b | 865 | int fd; |
d2ca6449 | 866 | struct mdinfo *inf = &infos[dnum]; |
4b1ac34b | 867 | |
d2ca6449 NB |
868 | if (dnum >= total_slots) |
869 | abort(); | |
4b1ac34b | 870 | if (dnum == insert_point) { |
f170a5a9 | 871 | raid_disk_num += 1; |
4b1ac34b | 872 | moved_disk = dv; |
8592f29d | 873 | continue; |
52826846 | 874 | } |
98dbf73c | 875 | if (strcasecmp(dv->devname, "missing") == 0) { |
f170a5a9 | 876 | raid_disk_num += 1; |
111d01fc | 877 | continue; |
f170a5a9 | 878 | } |
8592f29d N |
879 | if (have_container) |
880 | moved_disk = NULL; | |
881 | if (have_container && dnum < info.array.raid_disks - 1) | |
882 | /* repeatedly use the container */ | |
883 | moved_disk = dv; | |
111d01fc | 884 | |
d2ca6449 NB |
885 | switch(pass) { |
886 | case 1: | |
887 | *inf = info; | |
5f8097be | 888 | |
d2ca6449 | 889 | inf->disk.number = dnum; |
cc1799c3 SL |
890 | inf->disk.raid_disk = raid_disk_num++; |
891 | ||
892 | if (dv->disposition == 'j') { | |
893 | inf->disk.raid_disk = MD_DISK_ROLE_JOURNAL; | |
894 | inf->disk.state = (1<<MD_DISK_JOURNAL); | |
895 | raid_disk_num--; | |
896 | } else if (inf->disk.raid_disk < s->raiddisks) | |
d2ca6449 NB |
897 | inf->disk.state = (1<<MD_DISK_ACTIVE) | |
898 | (1<<MD_DISK_SYNC); | |
899 | else | |
900 | inf->disk.state = 0; | |
901 | ||
e22fe3ae | 902 | if (dv->writemostly == FlagSet) |
d2ca6449 | 903 | inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY); |
e22fe3ae | 904 | if (dv->failfast == FlagSet) |
71574efb | 905 | inf->disk.state |= (1<<MD_DISK_FAILFAST); |
d2ca6449 | 906 | |
8592f29d N |
907 | if (have_container) |
908 | fd = -1; | |
909 | else { | |
d1d599ea | 910 | if (st->ss->external && |
4dd2df09 | 911 | st->container_devnm[0]) |
8592f29d N |
912 | fd = open(dv->devname, O_RDWR); |
913 | else | |
914 | fd = open(dv->devname, O_RDWR|O_EXCL); | |
915 | ||
916 | if (fd < 0) { | |
7a862a02 | 917 | pr_err("failed to open %s after earlier success - aborting\n", |
8592f29d | 918 | dv->devname); |
19ad4b2c | 919 | goto abort_locked; |
8592f29d | 920 | } |
0a6bff09 ZL |
921 | if (!fstat_is_blkdev(fd, dv->devname, &rdev)) |
922 | return 1; | |
923 | inf->disk.major = major(rdev); | |
924 | inf->disk.minor = minor(rdev); | |
d2ca6449 | 925 | } |
8592f29d N |
926 | if (fd >= 0) |
927 | remove_partitions(fd); | |
f20c3968 | 928 | if (st->ss->add_to_super(st, &inf->disk, |
72ca9bcf N |
929 | fd, dv->devname, |
930 | dv->data_offset)) { | |
4eb26970 | 931 | ioctl(mdfd, STOP_ARRAY, NULL); |
19ad4b2c | 932 | goto abort_locked; |
4eb26970 | 933 | } |
a5d85af7 | 934 | st->ss->getinfo_super(st, inf, NULL); |
8ed3e5e1 | 935 | safe_mode_delay = inf->safe_mode_delay; |
d2ca6449 | 936 | |
171dccc8 | 937 | if (have_container && c->verbose > 0) |
e7b84f9d | 938 | pr_err("Using %s for device %d\n", |
8592f29d N |
939 | map_dev(inf->disk.major, |
940 | inf->disk.minor, | |
941 | 0), dnum); | |
942 | ||
943 | if (!have_container) { | |
944 | /* getinfo_super might have lost these ... */ | |
0a6bff09 ZL |
945 | inf->disk.major = major(rdev); |
946 | inf->disk.minor = minor(rdev); | |
8592f29d | 947 | } |
4b1ac34b NB |
948 | break; |
949 | case 2: | |
d2ca6449 | 950 | inf->errors = 0; |
4b1ac34b | 951 | |
f35f2525 | 952 | rv = add_disk(mdfd, st, &info, inf); |
d2ca6449 | 953 | |
2503d23b | 954 | if (rv) { |
7a862a02 | 955 | pr_err("ADD_NEW_DISK for %s failed: %s\n", |
e7b84f9d | 956 | dv->devname, strerror(errno)); |
329dfc28 N |
957 | if (errno == EINVAL && |
958 | info.array.level == 0) { | |
959 | pr_err("Possibly your kernel doesn't support RAID0 layouts.\n"); | |
960 | pr_err("Either upgrade, or use --layout=dangerous\n"); | |
961 | } | |
19ad4b2c | 962 | goto abort_locked; |
4b1ac34b | 963 | } |
4b1ac34b NB |
964 | break; |
965 | } | |
8592f29d N |
966 | if (!have_container && |
967 | dv == moved_disk && dnum != insert_point) break; | |
52826846 | 968 | } |
edd8d13c | 969 | if (pass == 1) { |
9b1fb677 DW |
970 | struct mdinfo info_new; |
971 | struct map_ent *me = NULL; | |
972 | ||
973 | /* check to see if the uuid has changed due to these | |
974 | * metadata changes, and if so update the member array | |
975 | * and container uuid. Note ->write_init_super clears | |
976 | * the subarray cursor such that ->getinfo_super once | |
977 | * again returns container info. | |
978 | */ | |
a5d85af7 | 979 | st->ss->getinfo_super(st, &info_new, NULL); |
99cc42f4 | 980 | if (st->ss->external && s->level != LEVEL_CONTAINER && |
9b1fb677 | 981 | !same_uuid(info_new.uuid, info.uuid, 0)) { |
4dd2df09 | 982 | map_update(&map, fd2devnm(mdfd), |
9b1fb677 DW |
983 | info_new.text_version, |
984 | info_new.uuid, chosen_name); | |
4dd2df09 | 985 | me = map_by_devnm(&map, st->container_devnm); |
9b1fb677 DW |
986 | } |
987 | ||
ab80e597 | 988 | if (st->ss->write_init_super(st)) { |
ab80e597 | 989 | st->ss->free_super(st); |
e06af9dd | 990 | goto abort_locked; |
ab80e597 | 991 | } |
9b1fb677 DW |
992 | |
993 | /* update parent container uuid */ | |
994 | if (me) { | |
503975b9 | 995 | char *path = xstrdup(me->path); |
9b1fb677 | 996 | |
a5d85af7 | 997 | st->ss->getinfo_super(st, &info_new, NULL); |
4dd2df09 | 998 | map_update(&map, st->container_devnm, |
9b1fb677 DW |
999 | info_new.text_version, |
1000 | info_new.uuid, path); | |
1001 | free(path); | |
1002 | } | |
9b1fb677 | 1003 | |
edd8d13c | 1004 | flush_metadata_updates(st); |
6946681d | 1005 | st->ss->free_super(st); |
edd8d13c | 1006 | } |
5e52ae9e | 1007 | } |
19ad4b2c | 1008 | map_unlock(&map); |
d2ca6449 | 1009 | free(infos); |
5e52ae9e | 1010 | |
99cc42f4 | 1011 | if (s->level == LEVEL_CONTAINER) { |
97590376 N |
1012 | /* No need to start. But we should signal udev to |
1013 | * create links */ | |
1014 | sysfs_uevent(&info, "change"); | |
171dccc8 | 1015 | if (c->verbose >= 0) |
e7b84f9d | 1016 | pr_err("container %s prepared.\n", mddev); |
a7c6e3fb | 1017 | wait_for(chosen_name, mdfd); |
99cc42f4 | 1018 | } else if (c->runstop == 1 || subdevs >= s->raiddisks) { |
598f0d58 | 1019 | if (st->ss->external) { |
4c821454 | 1020 | int err; |
99cc42f4 | 1021 | switch(s->level) { |
598f0d58 NB |
1022 | case LEVEL_LINEAR: |
1023 | case LEVEL_MULTIPATH: | |
1024 | case 0: | |
4c821454 | 1025 | err = sysfs_set_str(&info, NULL, "array_state", |
171dccc8 | 1026 | c->readonly |
72d566f6 N |
1027 | ? "readonly" |
1028 | : "active"); | |
a931db9e | 1029 | need_mdmon = 0; |
598f0d58 NB |
1030 | break; |
1031 | default: | |
4c821454 N |
1032 | err = sysfs_set_str(&info, NULL, "array_state", |
1033 | "readonly"); | |
598f0d58 NB |
1034 | break; |
1035 | } | |
f35f2525 | 1036 | sysfs_set_safemode(&info, safe_mode_delay); |
4c821454 | 1037 | if (err) { |
7a862a02 | 1038 | pr_err("failed to activate array.\n"); |
4c821454 N |
1039 | ioctl(mdfd, STOP_ARRAY, NULL); |
1040 | goto abort; | |
1041 | } | |
171dccc8 | 1042 | } else if (c->readonly && |
72d566f6 N |
1043 | sysfs_attribute_available( |
1044 | &info, NULL, "array_state")) { | |
1045 | if (sysfs_set_str(&info, NULL, | |
1046 | "array_state", "readonly") < 0) { | |
1047 | pr_err("Failed to start array: %s\n", | |
1048 | strerror(errno)); | |
1049 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1050 | goto abort; | |
1051 | } | |
598f0d58 | 1052 | } else { |
97590376 | 1053 | /* param is not actually used */ |
598f0d58 NB |
1054 | mdu_param_t param; |
1055 | if (ioctl(mdfd, RUN_ARRAY, ¶m)) { | |
e7b84f9d | 1056 | pr_err("RUN_ARRAY failed: %s\n", |
72d566f6 | 1057 | strerror(errno)); |
329dfc28 N |
1058 | if (errno == 524 /* ENOTSUP */ && |
1059 | info.array.level == 0) | |
1060 | cont_err("Please use --layout=original or --layout=alternate\n"); | |
a252c078 | 1061 | if (info.array.chunk_size & (info.array.chunk_size-1)) { |
7a862a02 | 1062 | cont_err("Problem may be that chunk size is not a power of 2\n"); |
a252c078 | 1063 | } |
4eb26970 | 1064 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1065 | goto abort; |
598f0d58 | 1066 | } |
a21e848a N |
1067 | /* if start_ro module parameter is set, array is |
1068 | * auto-read-only, which is bad as the resync won't | |
1069 | * start. So lets make it read-write now. | |
1070 | */ | |
1071 | ioctl(mdfd, RESTART_ARRAY_RW, NULL); | |
682c7051 | 1072 | } |
171dccc8 | 1073 | if (c->verbose >= 0) |
e7b84f9d | 1074 | pr_err("array %s started.\n", mddev); |
4dd2df09 | 1075 | if (st->ss->external && st->container_devnm[0]) { |
8850ee3e | 1076 | if (need_mdmon) |
4dd2df09 | 1077 | start_mdmon(st->container_devnm); |
8850ee3e | 1078 | |
4dd2df09 | 1079 | ping_monitor(st->container_devnm); |
a931db9e NB |
1080 | close(container_fd); |
1081 | } | |
a7c6e3fb | 1082 | wait_for(chosen_name, mdfd); |
682c7051 | 1083 | } else { |
e7b84f9d | 1084 | pr_err("not starting array - not enough devices.\n"); |
682c7051 | 1085 | } |
7f91af49 | 1086 | close(mdfd); |
cd6cbb08 N |
1087 | /* Give udev a moment to process the Change event caused |
1088 | * by the close. | |
1089 | */ | |
1090 | usleep(100*1000); | |
1091 | udev_unblock(); | |
682c7051 | 1092 | return 0; |
7f91af49 N |
1093 | |
1094 | abort: | |
cd6cbb08 | 1095 | udev_unblock(); |
4eb26970 | 1096 | map_lock(&map); |
e06af9dd | 1097 | abort_locked: |
4dd2df09 | 1098 | map_remove(&map, fd2devnm(mdfd)); |
4eb26970 DW |
1099 | map_unlock(&map); |
1100 | ||
7f91af49 N |
1101 | if (mdfd >= 0) |
1102 | close(mdfd); | |
1103 | return 1; | |
64c4757e | 1104 | } |