]>
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 | |
a18a888e DW |
30 | static int default_layout(struct supertype *st, int level, int verbose) |
31 | { | |
32 | int layout = UnSet; | |
33 | ||
30f58b22 DW |
34 | if (st && st->ss->default_geometry) |
35 | st->ss->default_geometry(st, &level, &layout, NULL); | |
a18a888e DW |
36 | |
37 | if (layout == UnSet) | |
38 | switch(level) { | |
39 | default: /* no layout */ | |
40 | layout = 0; | |
41 | break; | |
42 | case 10: | |
43 | layout = 0x102; /* near=2, far=1 */ | |
44 | if (verbose > 0) | |
e7b84f9d | 45 | pr_err("layout defaults to n2\n"); |
a18a888e DW |
46 | break; |
47 | case 5: | |
48 | case 6: | |
49 | layout = map_name(r5layout, "default"); | |
50 | if (verbose > 0) | |
e7b84f9d | 51 | pr_err("layout defaults to %s\n", map_num(r5layout, layout)); |
a18a888e DW |
52 | break; |
53 | case LEVEL_FAULTY: | |
54 | layout = map_name(faultylayout, "default"); | |
55 | ||
56 | if (verbose > 0) | |
e7b84f9d | 57 | pr_err("layout defaults to %s\n", map_num(faultylayout, layout)); |
a18a888e DW |
58 | break; |
59 | } | |
60 | ||
61 | return layout; | |
62 | } | |
63 | ||
7f91af49 | 64 | int Create(struct supertype *st, char *mddev, |
171dccc8 | 65 | char *name, int *uuid, |
a655e550 | 66 | int subdevs, struct mddev_dev *devlist, |
99cc42f4 | 67 | struct shape *s, |
40c9a66a | 68 | struct context *c, unsigned long long data_offset) |
64c4757e | 69 | { |
682c7051 NB |
70 | /* |
71 | * Create a new raid array. | |
72 | * | |
73 | * First check that necessary details are available | |
74 | * (i.e. level, raid-disks) | |
75 | * | |
76 | * Then check each disk to see what might be on it | |
77 | * and report anything interesting. | |
78 | * | |
79 | * If anything looks odd, and runstop not set, | |
80 | * abort. | |
81 | * | |
82 | * SET_ARRAY_INFO and ADD_NEW_DISK, and | |
82d9eba6 | 83 | * if runstop==run, or raiddisks disks were used, |
682c7051 NB |
84 | * RUN_ARRAY |
85 | */ | |
7f91af49 | 86 | int mdfd; |
b5e64645 | 87 | unsigned long long minsize=0, maxsize=0; |
cd29a5c8 NB |
88 | char *mindisc = NULL; |
89 | char *maxdisc = NULL; | |
cc1799c3 | 90 | int dnum, raid_disk_num; |
a655e550 | 91 | struct mddev_dev *dv; |
682c7051 | 92 | int fail=0, warn=0; |
82b27616 | 93 | struct stat stb; |
82d9eba6 | 94 | int first_missing = subdevs * 2; |
6fb79233 | 95 | int second_missing = subdevs * 2; |
52826846 | 96 | int missing_disks = 0; |
82d9eba6 | 97 | int insert_point = subdevs * 2; /* where to insert a missing drive */ |
d2ca6449 | 98 | int total_slots; |
4b1ac34b | 99 | int pass; |
82d9eba6 NB |
100 | int vers; |
101 | int rv; | |
c82f047c | 102 | int bitmap_fd; |
5f8097be | 103 | int have_container = 0; |
0e600426 | 104 | int container_fd = -1; |
a931db9e | 105 | int need_mdmon = 0; |
f9c25f1d | 106 | unsigned long long bitmapsize; |
d2ca6449 | 107 | struct mdinfo info, *infos; |
b8ac1967 | 108 | int did_default = 0; |
a18a888e | 109 | int do_default_layout = 0; |
c21e737b | 110 | int do_default_chunk = 0; |
8ed3e5e1 | 111 | unsigned long safe_mode_delay = 0; |
a04d5763 | 112 | char chosen_name[1024]; |
ad5bc697 | 113 | struct map_ent *map = NULL; |
8592f29d | 114 | unsigned long long newsize; |
682c7051 | 115 | |
943eafef | 116 | int major_num = BITMAP_MAJOR_HI; |
82d9485e | 117 | if (s->bitmap_file && strcmp(s->bitmap_file, "clustered") == 0) { |
6d9c7c25 | 118 | major_num = BITMAP_MAJOR_CLUSTERED; |
82d9485e GJ |
119 | if (c->nodes <= 1) { |
120 | pr_err("At least 2 nodes are needed for cluster-md\n"); | |
121 | return 1; | |
122 | } | |
123 | } | |
dcec9ee5 | 124 | |
06c7f68e | 125 | memset(&info, 0, sizeof(info)); |
99cc42f4 N |
126 | if (s->level == UnSet && st && st->ss->default_geometry) |
127 | st->ss->default_geometry(st, &s->level, NULL, NULL); | |
128 | if (s->level == UnSet) { | |
e7b84f9d | 129 | pr_err("a RAID level is needed to create an array.\n"); |
682c7051 NB |
130 | return 1; |
131 | } | |
99cc42f4 | 132 | if (s->raiddisks < 4 && s->level == 6) { |
e7b84f9d | 133 | pr_err("at least 4 raid-devices needed for level 6\n"); |
98c6faba NB |
134 | return 1; |
135 | } | |
99cc42f4 | 136 | if (s->raiddisks > 256 && s->level == 6) { |
e7b84f9d | 137 | pr_err("no more than 256 raid-devices supported for level 6\n"); |
98c6faba NB |
138 | return 1; |
139 | } | |
99cc42f4 | 140 | if (s->raiddisks < 2 && s->level >= 4) { |
e7b84f9d | 141 | pr_err("at least 2 raid-devices needed for level 4 or 5\n"); |
52826846 NB |
142 | return 1; |
143 | } | |
99cc42f4 | 144 | if (s->level <= 0 && s->sparedisks) { |
e7b84f9d | 145 | pr_err("This level does not support spare devices\n"); |
570510ba NB |
146 | return 1; |
147 | } | |
5f8097be NB |
148 | |
149 | if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) { | |
150 | /* If given a single device, it might be a container, and we can | |
151 | * extract a device list from there | |
152 | */ | |
153 | mdu_array_info_t inf; | |
154 | int fd; | |
155 | ||
156 | memset(&inf, 0, sizeof(inf)); | |
37ea3936 | 157 | fd = open(devlist->devname, O_RDONLY); |
5f8097be NB |
158 | if (fd >= 0 && |
159 | ioctl(fd, GET_ARRAY_INFO, &inf) == 0 && | |
160 | inf.raid_disks == 0) { | |
161 | /* yep, looks like a container */ | |
162 | if (st) { | |
0fb69d1d N |
163 | rv = st->ss->load_container(st, fd, |
164 | devlist->devname); | |
5f8097be NB |
165 | if (rv == 0) |
166 | have_container = 1; | |
167 | } else { | |
0fb69d1d | 168 | st = super_by_fd(fd, NULL); |
5f8097be | 169 | if (st && !(rv = st->ss-> |
0fb69d1d N |
170 | load_container(st, fd, |
171 | devlist->devname))) | |
5f8097be NB |
172 | have_container = 1; |
173 | else | |
174 | st = NULL; | |
175 | } | |
18fde300 | 176 | if (have_container) { |
99cc42f4 | 177 | subdevs = s->raiddisks; |
18fde300 DW |
178 | first_missing = subdevs * 2; |
179 | second_missing = subdevs * 2; | |
180 | insert_point = subdevs * 2; | |
181 | } | |
5f8097be NB |
182 | } |
183 | if (fd >= 0) | |
184 | close(fd); | |
5f8097be | 185 | } |
99cc42f4 | 186 | if (st && st->ss->external && s->sparedisks) { |
7a862a02 | 187 | pr_err("This metadata type does not support spare disks at create time\n"); |
ffcfc735 N |
188 | return 1; |
189 | } | |
cc1799c3 | 190 | if (subdevs > s->raiddisks+s->sparedisks+s->journaldisks) { |
99cc42f4 | 191 | pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks); |
52826846 | 192 | return 1; |
682c7051 | 193 | } |
cc1799c3 | 194 | if (!have_container && subdevs < s->raiddisks+s->sparedisks+s->journaldisks) { |
e7b84f9d | 195 | pr_err("You haven't given enough devices (real or missing) to create this array\n"); |
52826846 NB |
196 | return 1; |
197 | } | |
99cc42f4 | 198 | if (s->bitmap_file && s->level <= 0) { |
e7b84f9d | 199 | pr_err("bitmaps not meaningful with level %s\n", |
99cc42f4 | 200 | map_num(pers, s->level)?:"given"); |
5b28bd56 NB |
201 | return 1; |
202 | } | |
52826846 | 203 | |
682c7051 | 204 | /* now set some defaults */ |
b5e64645 | 205 | |
99cc42f4 | 206 | if (s->layout == UnSet) { |
a18a888e | 207 | do_default_layout = 1; |
99cc42f4 | 208 | s->layout = default_layout(st, s->level, c->verbose); |
a18a888e | 209 | } |
682c7051 | 210 | |
99cc42f4 | 211 | if (s->level == 10) |
e5329c37 | 212 | /* check layout fits in array*/ |
99cc42f4 | 213 | if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) { |
e7b84f9d | 214 | pr_err("that layout requires at least %d devices\n", |
99cc42f4 | 215 | (s->layout&255) * ((s->layout>>8)&255)); |
e5329c37 NB |
216 | return 1; |
217 | } | |
218 | ||
99cc42f4 | 219 | switch(s->level) { |
aa88f531 NB |
220 | case 4: |
221 | case 5: | |
e5329c37 | 222 | case 10: |
98c6faba | 223 | case 6: |
aa88f531 | 224 | case 0: |
99cc42f4 N |
225 | if (s->chunk == 0 || s->chunk == UnSet) { |
226 | s->chunk = UnSet; | |
c21e737b CA |
227 | do_default_chunk = 1; |
228 | /* chunk will be set later */ | |
bb7295f1 | 229 | } |
5f175898 N |
230 | break; |
231 | case LEVEL_LINEAR: | |
232 | /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */ | |
99cc42f4 N |
233 | if (get_linux_version() < 2006016 && s->chunk == 0) { |
234 | s->chunk = 64; | |
171dccc8 | 235 | if (c->verbose > 0) |
e7b84f9d | 236 | pr_err("chunk size defaults to 64K\n"); |
aa88f531 NB |
237 | } |
238 | break; | |
570510ba | 239 | case 1: |
08e43379 | 240 | case LEVEL_FAULTY: |
570510ba | 241 | case LEVEL_MULTIPATH: |
17f25ca6 | 242 | case LEVEL_CONTAINER: |
99cc42f4 N |
243 | if (s->chunk) { |
244 | s->chunk = 0; | |
171dccc8 | 245 | if (c->verbose > 0) |
e7b84f9d | 246 | pr_err("chunk size ignored for this level\n"); |
aa88f531 NB |
247 | } |
248 | break; | |
570510ba | 249 | default: |
99cc42f4 | 250 | pr_err("unknown level %d\n", s->level); |
570510ba | 251 | return 1; |
682c7051 | 252 | } |
99cc42f4 | 253 | if (s->size == MAX_SIZE) |
d04f65f4 | 254 | /* use '0' to mean 'max' now... */ |
99cc42f4 N |
255 | s->size = 0; |
256 | if (s->size && s->chunk && s->chunk != UnSet) | |
257 | s->size &= ~(unsigned long long)(s->chunk - 1); | |
258 | newsize = s->size * 2; | |
259 | if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks, | |
af4348dd | 260 | &s->chunk, s->size*2, |
40c9a66a | 261 | data_offset, NULL, |
af4348dd | 262 | &newsize, c->verbose>=0)) |
17f25ca6 | 263 | return 1; |
bb7295f1 | 264 | |
99cc42f4 N |
265 | if (s->chunk && s->chunk != UnSet) { |
266 | newsize &= ~(unsigned long long)(s->chunk*2 - 1); | |
64385908 CA |
267 | if (do_default_chunk) { |
268 | /* default chunk was just set */ | |
171dccc8 | 269 | if (c->verbose > 0) |
7a862a02 | 270 | pr_err("chunk size defaults to %dK\n", s->chunk); |
99cc42f4 | 271 | s->size &= ~(unsigned long long)(s->chunk - 1); |
64385908 CA |
272 | do_default_chunk = 0; |
273 | } | |
bb7295f1 | 274 | } |
64385908 | 275 | |
99cc42f4 N |
276 | if (s->size == 0) { |
277 | s->size = newsize / 2; | |
278 | if (s->level == 1) | |
ae6c05ad N |
279 | /* If this is ever reshaped to RAID5, we will |
280 | * need a chunksize. So round it off a bit | |
281 | * now just to be safe | |
282 | */ | |
99cc42f4 | 283 | s->size &= ~(64ULL-1); |
ae6c05ad | 284 | |
99cc42f4 N |
285 | if (s->size && c->verbose > 0) |
286 | pr_err("setting size to %lluK\n", s->size); | |
8592f29d | 287 | } |
17f25ca6 | 288 | |
682c7051 | 289 | /* now look at the subdevs */ |
06c7f68e NB |
290 | info.array.active_disks = 0; |
291 | info.array.working_disks = 0; | |
c913b90e | 292 | dnum = 0; |
476066a3 N |
293 | for (dv = devlist; dv ; dv = dv->next) |
294 | if (data_offset == VARIABLE_OFFSET) | |
295 | dv->data_offset = INVALID_SECTORS; | |
296 | else | |
297 | dv->data_offset = data_offset; | |
298 | ||
8592f29d | 299 | for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) { |
cd29a5c8 | 300 | char *dname = dv->devname; |
17f25ca6 | 301 | unsigned long long freesize; |
e9b11fee | 302 | int dfd; |
72ca9bcf | 303 | char *doff; |
e9b11fee | 304 | |
cd29a5c8 | 305 | if (strcasecmp(dname, "missing")==0) { |
c913b90e NB |
306 | if (first_missing > dnum) |
307 | first_missing = dnum; | |
6fb79233 NB |
308 | if (second_missing > dnum && dnum > first_missing) |
309 | second_missing = dnum; | |
52826846 NB |
310 | missing_disks ++; |
311 | continue; | |
312 | } | |
8baab049 | 313 | if (data_offset == VARIABLE_OFFSET) { |
72ca9bcf N |
314 | doff = strchr(dname, ':'); |
315 | if (doff) { | |
316 | *doff++ = 0; | |
317 | dv->data_offset = parse_size(doff); | |
318 | } else | |
319 | dv->data_offset = INVALID_SECTORS; | |
320 | } else | |
321 | dv->data_offset = data_offset; | |
322 | ||
58b3c697 | 323 | dfd = open(dname, O_RDONLY); |
e9b11fee | 324 | if (dfd < 0) { |
e7b84f9d | 325 | pr_err("cannot open %s: %s\n", |
e9b11fee N |
326 | dname, strerror(errno)); |
327 | exit(2); | |
328 | } | |
329 | if (fstat(dfd, &stb) != 0 || | |
330 | (stb.st_mode & S_IFMT) != S_IFBLK) { | |
331 | close(dfd); | |
e7b84f9d | 332 | pr_err("%s is not a block device\n", |
e9b11fee N |
333 | dname); |
334 | exit(2); | |
335 | } | |
336 | close(dfd); | |
06c7f68e | 337 | info.array.working_disks++; |
dfd7822c | 338 | if (dnum < s->raiddisks && dv->disposition != 'j') |
06c7f68e | 339 | info.array.active_disks++; |
058574b1 | 340 | if (st == NULL) { |
8aec876d | 341 | struct createinfo *ci = conf_get_create_info(); |
058574b1 NB |
342 | if (ci) |
343 | st = ci->supertype; | |
344 | } | |
576d6d83 NB |
345 | if (st == NULL) { |
346 | /* Need to choose a default metadata, which is different | |
17f25ca6 | 347 | * depending on geometry of array. |
576d6d83 NB |
348 | */ |
349 | int i; | |
350 | char *name = "default"; | |
17f25ca6 | 351 | for(i=0; !st && superlist[i]; i++) { |
576d6d83 | 352 | st = superlist[i]->match_metadata_desc(name); |
ecbd9e81 N |
353 | if (!st) |
354 | continue; | |
a18a888e | 355 | if (do_default_layout) |
99cc42f4 | 356 | s->layout = default_layout(st, s->level, c->verbose); |
ecbd9e81 | 357 | switch (st->ss->validate_geometry( |
99cc42f4 | 358 | st, s->level, s->layout, s->raiddisks, |
af4348dd | 359 | &s->chunk, s->size*2, |
72ca9bcf | 360 | dv->data_offset, dname, |
af4348dd | 361 | &freesize, c->verbose > 0)) { |
ecbd9e81 N |
362 | case -1: /* Not valid, message printed, and not |
363 | * worth checking any further */ | |
364 | exit(2); | |
365 | break; | |
366 | case 0: /* Geometry not valid */ | |
55425f27 | 367 | free(st); |
17f25ca6 | 368 | st = NULL; |
99cc42f4 | 369 | s->chunk = do_default_chunk ? UnSet : s->chunk; |
ecbd9e81 N |
370 | break; |
371 | case 1: /* All happy */ | |
372 | break; | |
55425f27 | 373 | } |
17f25ca6 | 374 | } |
576d6d83 NB |
375 | |
376 | if (!st) { | |
58b3c697 N |
377 | int dfd = open(dname, O_RDONLY|O_EXCL); |
378 | if (dfd < 0) { | |
e7b84f9d | 379 | pr_err("cannot open %s: %s\n", |
58b3c697 N |
380 | dname, strerror(errno)); |
381 | exit(2); | |
382 | } | |
7a862a02 | 383 | pr_err("device %s not suitable for any style of array\n", |
17f25ca6 | 384 | dname); |
576d6d83 NB |
385 | exit(2); |
386 | } | |
b8ac1967 | 387 | if (st->ss != &super0 || |
576d6d83 | 388 | st->minor_version != 90) |
b8ac1967 | 389 | did_default = 1; |
17f25ca6 | 390 | } else { |
a18a888e | 391 | if (do_default_layout) |
99cc42f4 N |
392 | s->layout = default_layout(st, s->level, 0); |
393 | if (!st->ss->validate_geometry(st, s->level, s->layout, | |
394 | s->raiddisks, | |
af4348dd | 395 | &s->chunk, s->size*2, |
72ca9bcf | 396 | dv->data_offset, |
af4348dd | 397 | dname, &freesize, |
171dccc8 | 398 | c->verbose >= 0)) { |
17f25ca6 | 399 | |
7a862a02 | 400 | pr_err("%s is not suitable for this array.\n", |
e7b84f9d | 401 | dname); |
17f25ca6 NB |
402 | fail = 1; |
403 | continue; | |
404 | } | |
682c7051 | 405 | } |
82d9eba6 | 406 | |
cc1799c3 | 407 | if (dv->disposition == 'j') |
dfd7822c | 408 | goto skip_size_check; /* skip write journal for size check */ |
cc1799c3 | 409 | |
82d9eba6 | 410 | freesize /= 2; /* convert to K */ |
99cc42f4 | 411 | if (s->chunk && s->chunk != UnSet) { |
d2cd3ffc | 412 | /* round to chunk size */ |
99cc42f4 | 413 | freesize = freesize & ~(s->chunk-1); |
64385908 CA |
414 | if (do_default_chunk) { |
415 | /* default chunk was just set */ | |
171dccc8 | 416 | if (c->verbose > 0) |
7a862a02 | 417 | pr_err("chunk size defaults to %dK\n", s->chunk); |
99cc42f4 | 418 | s->size &= ~(unsigned long long)(s->chunk - 1); |
64385908 CA |
419 | do_default_chunk = 0; |
420 | } | |
d2cd3ffc | 421 | } |
066e92f0 LD |
422 | if (!freesize) { |
423 | pr_err("no free space left on %s\n", dname); | |
424 | fail = 1; | |
425 | continue; | |
426 | } | |
682c7051 | 427 | |
99cc42f4 | 428 | if (s->size && freesize < s->size) { |
7a862a02 | 429 | pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n", |
99cc42f4 | 430 | dname, freesize, s->size); |
52826846 | 431 | fail = 1; |
52826846 | 432 | continue; |
682c7051 | 433 | } |
cd29a5c8 NB |
434 | if (maxdisc == NULL || (maxdisc && freesize > maxsize)) { |
435 | maxdisc = dname; | |
52826846 | 436 | maxsize = freesize; |
682c7051 | 437 | } |
cd29a5c8 NB |
438 | if (mindisc ==NULL || (mindisc && freesize < minsize)) { |
439 | mindisc = dname; | |
52826846 | 440 | minsize = freesize; |
682c7051 | 441 | } |
dfd7822c | 442 | skip_size_check: |
171dccc8 | 443 | if (c->runstop != 1 || c->verbose >= 0) { |
37ea3936 | 444 | int fd = open(dname, O_RDONLY); |
b6e63da4 | 445 | if (fd <0 ) { |
e7b84f9d | 446 | pr_err("Cannot open %s: %s\n", |
b6e63da4 NB |
447 | dname, strerror(errno)); |
448 | fail=1; | |
449 | continue; | |
450 | } | |
dab6685f NB |
451 | warn |= check_ext2(fd, dname); |
452 | warn |= check_reiser(fd, dname); | |
453 | warn |= check_raid(fd, dname); | |
034b203a TM |
454 | if (strcmp(st->ss->name, "1.x") == 0 && |
455 | st->minor_version >= 1) | |
456 | /* metadata at front */ | |
53ed6ac3 | 457 | warn |= check_partitions(fd, dname, 0, 0); |
99cc42f4 N |
458 | else if (s->level == 1 || s->level == LEVEL_CONTAINER |
459 | || (s->level == 0 && s->raiddisks == 1)) | |
034b203a | 460 | /* partitions could be meaningful */ |
99cc42f4 | 461 | warn |= check_partitions(fd, dname, freesize*2, s->size*2); |
034b203a TM |
462 | else |
463 | /* partitions cannot be meaningful */ | |
53ed6ac3 | 464 | warn |= check_partitions(fd, dname, 0, 0); |
034b203a | 465 | if (strcmp(st->ss->name, "1.x") == 0 && |
a0962fe9 N |
466 | st->minor_version >= 1 && |
467 | did_default && | |
99cc42f4 | 468 | s->level == 1 && |
034b203a TM |
469 | (warn & 1024) == 0) { |
470 | warn |= 1024; | |
e7b84f9d | 471 | pr_err("Note: this array has metadata at the start and\n" |
a0962fe9 | 472 | " may not be suitable as a boot device. If you plan to\n" |
cc86f89c | 473 | " store '/boot' on this device please ensure that\n" |
a0962fe9 | 474 | " your boot-loader understands md/v1.x metadata, or use\n" |
cc86f89c | 475 | " --metadata=0.90\n"); |
a0962fe9 | 476 | } |
b6e63da4 | 477 | close(fd); |
dab6685f | 478 | } |
682c7051 | 479 | } |
99cc42f4 | 480 | if (s->raiddisks + s->sparedisks > st->max_devs) { |
7a862a02 | 481 | pr_err("Too many devices: %s metadata only supports %d\n", |
acab7bb1 N |
482 | st->ss->name, st->max_devs); |
483 | return 1; | |
484 | } | |
8592f29d | 485 | if (have_container) |
99cc42f4 | 486 | info.array.working_disks = s->raiddisks; |
682c7051 | 487 | if (fail) { |
e7b84f9d | 488 | pr_err("create aborted\n"); |
52826846 | 489 | return 1; |
682c7051 | 490 | } |
99cc42f4 | 491 | if (s->size == 0) { |
5f8097be | 492 | if (mindisc == NULL && !have_container) { |
e7b84f9d | 493 | pr_err("no size and no drives given - aborting create.\n"); |
52826846 NB |
494 | return 1; |
495 | } | |
99cc42f4 N |
496 | if (s->level > 0 || s->level == LEVEL_MULTIPATH |
497 | || s->level == LEVEL_FAULTY | |
b8ac1967 | 498 | || st->ss->external ) { |
b5e64645 | 499 | /* size is meaningful */ |
99cc42f4 N |
500 | if (!st->ss->validate_geometry(st, s->level, s->layout, |
501 | s->raiddisks, | |
502 | &s->chunk, minsize*2, | |
40c9a66a | 503 | data_offset, |
2c514b71 | 504 | NULL, NULL, 0)) { |
99cc42f4 | 505 | pr_err("devices too large for RAID level %d\n", s->level); |
b5e64645 NB |
506 | return 1; |
507 | } | |
99cc42f4 N |
508 | s->size = minsize; |
509 | if (s->level == 1) | |
ae6c05ad N |
510 | /* If this is ever reshaped to RAID5, we will |
511 | * need a chunksize. So round it off a bit | |
512 | * now just to be safe | |
513 | */ | |
99cc42f4 | 514 | s->size &= ~(64ULL-1); |
171dccc8 | 515 | if (c->verbose > 0) |
99cc42f4 | 516 | pr_err("size set to %lluK\n", s->size); |
b5e64645 | 517 | } |
682c7051 | 518 | } |
748952f7 N |
519 | |
520 | if (!s->bitmap_file && | |
521 | s->level >= 1 && | |
39917e56 | 522 | st->ss->add_internal_bitmap && |
748952f7 N |
523 | (s->write_behind || s->size > 100*1024*1024ULL)) { |
524 | if (c->verbose > 0) | |
525 | pr_err("automatically enabling write-intent bitmap on large array\n"); | |
526 | s->bitmap_file = "internal"; | |
527 | } | |
528 | if (s->bitmap_file && strcmp(s->bitmap_file, "none") == 0) | |
529 | s->bitmap_file = NULL; | |
530 | ||
99cc42f4 | 531 | if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) { |
171dccc8 | 532 | if (c->runstop != 1 || c->verbose >= 0) |
e7b84f9d | 533 | pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n", |
99cc42f4 | 534 | maxdisc, s->size); |
52826846 | 535 | warn = 1; |
682c7051 NB |
536 | } |
537 | ||
9eafa1de | 538 | if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) { |
171dccc8 | 539 | if (c->runstop != 1 || c->verbose >= 0) |
e7b84f9d | 540 | pr_err("%s unable to enumerate platform support\n" |
5615172f DW |
541 | " array may not be compatible with hardware/firmware\n", |
542 | st->ss->name); | |
543 | warn = 1; | |
544 | } | |
529e2aa5 | 545 | st->nodes = c->nodes; |
7716570e | 546 | st->cluster_name = c->homecluster; |
5615172f | 547 | |
682c7051 | 548 | if (warn) { |
171dccc8 | 549 | if (c->runstop!= 1) { |
52826846 | 550 | if (!ask("Continue creating array? ")) { |
e7b84f9d | 551 | pr_err("create aborted.\n"); |
52826846 NB |
552 | return 1; |
553 | } | |
554 | } else { | |
171dccc8 | 555 | if (c->verbose > 0) |
e7b84f9d | 556 | pr_err("creation continuing despite oddities due to --run\n"); |
682c7051 | 557 | } |
682c7051 NB |
558 | } |
559 | ||
66f8bbbe | 560 | /* If this is raid4/5, we want to configure the last active slot |
52826846 | 561 | * as missing, so that a reconstruct happens (faster than re-parity) |
98c6faba | 562 | * FIX: Can we do this for raid6 as well? |
52826846 | 563 | */ |
87b47257 | 564 | if (st->ss->external == 0 && |
99cc42f4 N |
565 | s->assume_clean==0 && c->force == 0 && first_missing >= s->raiddisks) { |
566 | switch ( s->level ) { | |
66f8bbbe | 567 | case 4: |
98c6faba | 568 | case 5: |
99cc42f4 N |
569 | insert_point = s->raiddisks-1; |
570 | s->sparedisks++; | |
06c7f68e | 571 | info.array.active_disks--; |
98c6faba NB |
572 | missing_disks++; |
573 | break; | |
574 | default: | |
575 | break; | |
576 | } | |
52826846 | 577 | } |
6fb79233 NB |
578 | /* For raid6, if creating with 1 missing drive, make a good drive |
579 | * into a spare, else the create will fail | |
580 | */ | |
99cc42f4 | 581 | if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks && |
ffcfc735 | 582 | st->ss->external == 0 && |
99cc42f4 N |
583 | second_missing >= s->raiddisks && s->level == 6) { |
584 | insert_point = s->raiddisks - 1; | |
6fb79233 NB |
585 | if (insert_point == first_missing) |
586 | insert_point--; | |
99cc42f4 | 587 | s->sparedisks ++; |
6fb79233 NB |
588 | info.array.active_disks--; |
589 | missing_disks++; | |
590 | } | |
570510ba | 591 | |
99cc42f4 | 592 | if (s->level <= 0 && first_missing < subdevs * 2) { |
e7b84f9d | 593 | pr_err("This level does not support missing devices\n"); |
570510ba NB |
594 | return 1; |
595 | } | |
aba69144 | 596 | |
7f91af49 | 597 | /* We need to create the device */ |
ad5bc697 | 598 | map_lock(&map); |
171dccc8 | 599 | mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name); |
e06af9dd JS |
600 | if (mdfd < 0) { |
601 | map_unlock(&map); | |
7f91af49 | 602 | return 1; |
e06af9dd | 603 | } |
3e9df86a AK |
604 | /* verify if chosen_name is not in use, |
605 | * it could be in conflict with already existing device | |
606 | * e.g. container, array | |
607 | */ | |
608 | if (strncmp(chosen_name, "/dev/md/", 8) == 0 | |
609 | && map_by_name(&map, chosen_name+8) != NULL) { | |
e7b84f9d | 610 | pr_err("Array name %s is in use already.\n", |
3e9df86a AK |
611 | chosen_name); |
612 | close(mdfd); | |
613 | map_unlock(&map); | |
614 | return 1; | |
615 | } | |
f2be09f1 | 616 | mddev = chosen_name; |
7f91af49 N |
617 | |
618 | vers = md_get_version(mdfd); | |
619 | if (vers < 9000) { | |
e7b84f9d | 620 | pr_err("Create requires md driver version 0.90.0 or later\n"); |
e06af9dd | 621 | goto abort_locked; |
7f91af49 N |
622 | } else { |
623 | mdu_array_info_t inf; | |
624 | memset(&inf, 0, sizeof(inf)); | |
625 | ioctl(mdfd, GET_ARRAY_INFO, &inf); | |
626 | if (inf.working_disks != 0) { | |
7a862a02 | 627 | pr_err("another array by this name is already running.\n"); |
e06af9dd | 628 | goto abort_locked; |
7f91af49 N |
629 | } |
630 | } | |
a04d5763 | 631 | |
682c7051 NB |
632 | /* Ok, lets try some ioctls */ |
633 | ||
99cc42f4 N |
634 | info.array.level = s->level; |
635 | info.array.size = s->size; | |
636 | info.array.raid_disks = s->raiddisks; | |
82b27616 NB |
637 | /* The kernel should *know* what md_minor we are dealing |
638 | * with, but it chooses to trust me instead. Sigh | |
639 | */ | |
06c7f68e | 640 | info.array.md_minor = 0; |
82b27616 | 641 | if (fstat(mdfd, &stb)==0) |
06c7f68e NB |
642 | info.array.md_minor = minor(stb.st_rdev); |
643 | info.array.not_persistent = 0; | |
6fb79233 | 644 | |
99cc42f4 N |
645 | if ( ( (s->level == 4 || s->level == 5) && |
646 | (insert_point < s->raiddisks || first_missing < s->raiddisks) ) | |
b5e64645 | 647 | || |
99cc42f4 N |
648 | ( s->level == 6 && (insert_point < s->raiddisks |
649 | || second_missing < s->raiddisks)) | |
47d79ef8 | 650 | || |
99cc42f4 | 651 | ( s->level <= 0 ) |
d024b0a7 | 652 | || |
99cc42f4 | 653 | s->assume_clean |
103f2410 | 654 | ) { |
06c7f68e | 655 | info.array.state = 1; /* clean, but one+ drive will be missing*/ |
b7528a20 | 656 | info.resync_start = MaxSector; |
103f2410 | 657 | } else { |
06c7f68e | 658 | info.array.state = 0; /* not clean, but no errors */ |
103f2410 NB |
659 | info.resync_start = 0; |
660 | } | |
99cc42f4 | 661 | if (s->level == 10) { |
f9c25f1d NB |
662 | /* for raid10, the bitmap size is the capacity of the array, |
663 | * which is array.size * raid_disks / ncopies; | |
664 | * .. but convert to sectors. | |
665 | */ | |
99cc42f4 N |
666 | int ncopies = ((s->layout>>8) & 255) * (s->layout & 255); |
667 | bitmapsize = s->size * s->raiddisks / ncopies * 2; | |
668 | /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/ | |
f9c25f1d | 669 | } else |
99cc42f4 | 670 | bitmapsize = s->size * 2; |
f9c25f1d | 671 | |
82b27616 NB |
672 | /* There is lots of redundancy in these disk counts, |
673 | * raid_disks is the most meaningful value | |
674 | * it describes the geometry of the array | |
675 | * it is constant | |
676 | * nr_disks is total number of used slots. | |
677 | * it should be raid_disks+spare_disks | |
678 | * spare_disks is the number of extra disks present | |
679 | * see above | |
680 | * active_disks is the number of working disks in | |
681 | * active slots. (With raid_disks) | |
682 | * working_disks is the total number of working disks, | |
683 | * including spares | |
684 | * failed_disks is the number of disks marked failed | |
685 | * | |
5d500228 | 686 | * Ideally, the kernel would keep these (except raid_disks) |
82b27616 NB |
687 | * up-to-date as we ADD_NEW_DISK, but it doesn't (yet). |
688 | * So for now, we assume that all raid and spare | |
689 | * devices will be given. | |
690 | */ | |
99cc42f4 | 691 | info.array.spare_disks=s->sparedisks; |
06c7f68e NB |
692 | info.array.failed_disks=missing_disks; |
693 | info.array.nr_disks = info.array.working_disks | |
694 | + info.array.failed_disks; | |
99cc42f4 N |
695 | info.array.layout = s->layout; |
696 | info.array.chunk_size = s->chunk*1024; | |
82d9eba6 | 697 | |
b3b33eb5 NB |
698 | if (name == NULL || *name == 0) { |
699 | /* base name on mddev */ | |
d1e80164 NB |
700 | /* /dev/md0 -> 0 |
701 | * /dev/md_d0 -> d0 | |
eca944fa | 702 | * /dev/md_foo -> foo |
d1e80164 NB |
703 | * /dev/md/1 -> 1 |
704 | * /dev/md/d1 -> d1 | |
705 | * /dev/md/home -> home | |
706 | * /dev/mdhome -> home | |
707 | */ | |
69207ff6 | 708 | /* FIXME compare this with rules in create_mddev */ |
b3b33eb5 NB |
709 | name = strrchr(mddev, '/'); |
710 | if (name) { | |
711 | name++; | |
eca944fa N |
712 | if (strncmp(name, "md_", 3)==0 && |
713 | strlen(name) > 3 && | |
b3b33eb5 | 714 | (name-mddev) == 5 /* /dev/ */) |
d1e80164 | 715 | name += 3; |
60248f74 NB |
716 | else if (strncmp(name, "md", 2)==0 && |
717 | strlen(name) > 2 && | |
718 | isdigit(name[2]) && | |
719 | (name-mddev) == 5 /* /dev/ */) | |
b3b33eb5 NB |
720 | name += 2; |
721 | } | |
722 | } | |
83cd1e97 | 723 | if (!st->ss->init_super(st, &info.array, s->size, name, c->homehost, uuid, |
40c9a66a | 724 | data_offset)) |
e06af9dd | 725 | goto abort_locked; |
682c7051 | 726 | |
d2ca6449 | 727 | total_slots = info.array.nr_disks; |
a5d85af7 | 728 | st->ss->getinfo_super(st, &info, NULL); |
4dd2df09 | 729 | sysfs_init(&info, mdfd, NULL); |
159c3a1a | 730 | |
171dccc8 | 731 | if (did_default && c->verbose >= 0) { |
3c558363 | 732 | if (is_subarray(info.text_version)) { |
4dd2df09 N |
733 | char devnm[32]; |
734 | char *ep; | |
f7f1b6a1 | 735 | struct mdinfo *mdi; |
f7f1b6a1 | 736 | |
4dd2df09 N |
737 | strncpy(devnm, info.text_version+1, 32); |
738 | devnm[31] = 0; | |
739 | ep = strchr(devnm, '/'); | |
740 | if (ep) | |
741 | *ep = 0; | |
742 | ||
743 | mdi = sysfs_read(-1, devnm, GET_VERSION); | |
f7f1b6a1 | 744 | |
4dd2df09 N |
745 | pr_err("Creating array inside %s container %s\n", |
746 | mdi?mdi->text_version:"managed", devnm); | |
f7f1b6a1 NB |
747 | sysfs_free(mdi); |
748 | } else | |
7a862a02 | 749 | pr_err("Defaulting to version %s metadata\n", info.text_version); |
f7f1b6a1 | 750 | } |
b8ac1967 | 751 | |
4dd2df09 | 752 | map_update(&map, fd2devnm(mdfd), info.text_version, |
a04d5763 | 753 | info.uuid, chosen_name); |
19ad4b2c AP |
754 | /* Keep map locked until devices have been added to array |
755 | * to stop another mdadm from finding and using those devices. | |
756 | */ | |
a04d5763 | 757 | |
99cc42f4 | 758 | if (s->bitmap_file && vers < 9003) { |
943eafef | 759 | major_num = BITMAP_MAJOR_HOSTENDIAN; |
dcec9ee5 | 760 | #ifdef __BIG_ENDIAN |
e7b84f9d | 761 | pr_err("Warning - bitmaps created on this kernel are not portable\n" |
dcec9ee5 NB |
762 | " between different architectured. Consider upgrading the Linux kernel.\n"); |
763 | #endif | |
764 | } | |
765 | ||
95a05b37 GJ |
766 | if (s->bitmap_file && (strcmp(s->bitmap_file, "internal")==0 || |
767 | strcmp(s->bitmap_file, "clustered")==0)) { | |
55935d51 | 768 | if ((vers%100) < 2) { |
e7b84f9d | 769 | pr_err("internal bitmaps not supported by this kernel.\n"); |
19ad4b2c | 770 | goto abort_locked; |
55935d51 | 771 | } |
ebeb3663 | 772 | if (!st->ss->add_internal_bitmap) { |
e7b84f9d | 773 | pr_err("internal bitmaps not supported with %s metadata\n", |
ebeb3663 | 774 | st->ss->name); |
19ad4b2c | 775 | goto abort_locked; |
ebeb3663 | 776 | } |
2ec2b7e9 JS |
777 | if (st->ss->add_internal_bitmap(st, &s->bitmap_chunk, |
778 | c->delay, s->write_behind, | |
779 | bitmapsize, 1, major_num)) { | |
e7b84f9d | 780 | pr_err("Given bitmap chunk size not supported.\n"); |
19ad4b2c | 781 | goto abort_locked; |
55935d51 | 782 | } |
99cc42f4 | 783 | s->bitmap_file = NULL; |
55935d51 NB |
784 | } |
785 | ||
4dd2df09 | 786 | sysfs_init(&info, mdfd, NULL); |
55935d51 | 787 | |
4dd2df09 | 788 | if (st->ss->external && st->container_devnm[0]) { |
f35f2525 N |
789 | /* member */ |
790 | ||
791 | /* When creating a member, we need to be careful | |
792 | * to negotiate with mdmon properly. | |
793 | * If it is already running, we cannot write to | |
794 | * the devices and must ask it to do that part. | |
795 | * If it isn't running, we write to the devices, | |
796 | * and then start it. | |
797 | * We hold an exclusive open on the container | |
798 | * device to make sure mdmon doesn't exit after | |
799 | * we checked that it is running. | |
800 | * | |
801 | * For now, fail if it is already running. | |
802 | */ | |
4dd2df09 | 803 | container_fd = open_dev_excl(st->container_devnm); |
f35f2525 | 804 | if (container_fd < 0) { |
7a862a02 | 805 | pr_err("Cannot get exclusive open on container - weird.\n"); |
19ad4b2c | 806 | goto abort_locked; |
d03373f1 | 807 | } |
4dd2df09 | 808 | if (mdmon_running(st->container_devnm)) { |
171dccc8 | 809 | if (c->verbose) |
7a862a02 | 810 | pr_err("reusing mdmon for %s.\n", |
4dd2df09 | 811 | st->container_devnm); |
f35f2525 N |
812 | st->update_tail = &st->updates; |
813 | } else | |
814 | need_mdmon = 1; | |
815 | } | |
816 | rv = set_array_info(mdfd, st, &info); | |
82d9eba6 | 817 | if (rv) { |
e7b84f9d | 818 | pr_err("failed to set array info for %s: %s\n", |
52826846 | 819 | mddev, strerror(errno)); |
19ad4b2c | 820 | goto abort_locked; |
682c7051 | 821 | } |
52826846 | 822 | |
99cc42f4 | 823 | if (s->bitmap_file) { |
c82f047c | 824 | int uuid[4]; |
55935d51 | 825 | |
3da92f27 | 826 | st->ss->uuid_from_super(st, uuid); |
99cc42f4 N |
827 | if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk, |
828 | c->delay, s->write_behind, | |
f9c25f1d | 829 | bitmapsize, |
943eafef | 830 | major_num)) { |
19ad4b2c | 831 | goto abort_locked; |
c82f047c | 832 | } |
99cc42f4 | 833 | bitmap_fd = open(s->bitmap_file, O_RDWR); |
c82f047c | 834 | if (bitmap_fd < 0) { |
e7b84f9d | 835 | pr_err("weird: %s cannot be openned\n", |
99cc42f4 | 836 | s->bitmap_file); |
19ad4b2c | 837 | goto abort_locked; |
c82f047c NB |
838 | } |
839 | if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
e7b84f9d | 840 | pr_err("Cannot set bitmap file for %s: %s\n", |
c82f047c | 841 | mddev, strerror(errno)); |
19ad4b2c | 842 | goto abort_locked; |
c82f047c NB |
843 | } |
844 | } | |
845 | ||
503975b9 | 846 | infos = xmalloc(sizeof(*infos) * total_slots); |
a7dec3fd | 847 | enable_fds(total_slots); |
4b1ac34b | 848 | for (pass=1; pass <=2 ; pass++) { |
a655e550 | 849 | struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */ |
4b1ac34b | 850 | |
cc1799c3 | 851 | for (dnum=0, raid_disk_num=0, dv = devlist ; dv ; |
4b1ac34b NB |
852 | dv=(dv->next)?(dv->next):moved_disk, dnum++) { |
853 | int fd; | |
854 | struct stat stb; | |
d2ca6449 | 855 | struct mdinfo *inf = &infos[dnum]; |
4b1ac34b | 856 | |
d2ca6449 NB |
857 | if (dnum >= total_slots) |
858 | abort(); | |
4b1ac34b | 859 | if (dnum == insert_point) { |
f170a5a9 | 860 | raid_disk_num += 1; |
4b1ac34b | 861 | moved_disk = dv; |
8592f29d | 862 | continue; |
52826846 | 863 | } |
f170a5a9 N |
864 | if (strcasecmp(dv->devname, "missing")==0) { |
865 | raid_disk_num += 1; | |
111d01fc | 866 | continue; |
f170a5a9 | 867 | } |
8592f29d N |
868 | if (have_container) |
869 | moved_disk = NULL; | |
870 | if (have_container && dnum < info.array.raid_disks - 1) | |
871 | /* repeatedly use the container */ | |
872 | moved_disk = dv; | |
111d01fc | 873 | |
d2ca6449 NB |
874 | switch(pass) { |
875 | case 1: | |
876 | *inf = info; | |
5f8097be | 877 | |
d2ca6449 | 878 | inf->disk.number = dnum; |
cc1799c3 SL |
879 | inf->disk.raid_disk = raid_disk_num++; |
880 | ||
881 | if (dv->disposition == 'j') { | |
882 | inf->disk.raid_disk = MD_DISK_ROLE_JOURNAL; | |
883 | inf->disk.state = (1<<MD_DISK_JOURNAL); | |
884 | raid_disk_num--; | |
885 | } else if (inf->disk.raid_disk < s->raiddisks) | |
d2ca6449 NB |
886 | inf->disk.state = (1<<MD_DISK_ACTIVE) | |
887 | (1<<MD_DISK_SYNC); | |
888 | else | |
889 | inf->disk.state = 0; | |
890 | ||
b01b06bd | 891 | if (dv->writemostly == 1) |
d2ca6449 NB |
892 | inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY); |
893 | ||
8592f29d N |
894 | if (have_container) |
895 | fd = -1; | |
896 | else { | |
d1d599ea | 897 | if (st->ss->external && |
4dd2df09 | 898 | st->container_devnm[0]) |
8592f29d N |
899 | fd = open(dv->devname, O_RDWR); |
900 | else | |
901 | fd = open(dv->devname, O_RDWR|O_EXCL); | |
902 | ||
903 | if (fd < 0) { | |
7a862a02 | 904 | pr_err("failed to open %s after earlier success - aborting\n", |
8592f29d | 905 | dv->devname); |
19ad4b2c | 906 | goto abort_locked; |
8592f29d N |
907 | } |
908 | fstat(fd, &stb); | |
909 | inf->disk.major = major(stb.st_rdev); | |
910 | inf->disk.minor = minor(stb.st_rdev); | |
d2ca6449 | 911 | } |
8592f29d N |
912 | if (fd >= 0) |
913 | remove_partitions(fd); | |
f20c3968 | 914 | if (st->ss->add_to_super(st, &inf->disk, |
72ca9bcf N |
915 | fd, dv->devname, |
916 | dv->data_offset)) { | |
4eb26970 | 917 | ioctl(mdfd, STOP_ARRAY, NULL); |
19ad4b2c | 918 | goto abort_locked; |
4eb26970 | 919 | } |
a5d85af7 | 920 | st->ss->getinfo_super(st, inf, NULL); |
8ed3e5e1 | 921 | safe_mode_delay = inf->safe_mode_delay; |
d2ca6449 | 922 | |
171dccc8 | 923 | if (have_container && c->verbose > 0) |
e7b84f9d | 924 | pr_err("Using %s for device %d\n", |
8592f29d N |
925 | map_dev(inf->disk.major, |
926 | inf->disk.minor, | |
927 | 0), dnum); | |
928 | ||
929 | if (!have_container) { | |
930 | /* getinfo_super might have lost these ... */ | |
931 | inf->disk.major = major(stb.st_rdev); | |
932 | inf->disk.minor = minor(stb.st_rdev); | |
933 | } | |
4b1ac34b NB |
934 | break; |
935 | case 2: | |
d2ca6449 | 936 | inf->errors = 0; |
4b1ac34b | 937 | |
f35f2525 | 938 | rv = add_disk(mdfd, st, &info, inf); |
d2ca6449 | 939 | |
2503d23b | 940 | if (rv) { |
7a862a02 | 941 | pr_err("ADD_NEW_DISK for %s failed: %s\n", |
e7b84f9d | 942 | dv->devname, strerror(errno)); |
19ad4b2c | 943 | goto abort_locked; |
4b1ac34b | 944 | } |
4b1ac34b NB |
945 | break; |
946 | } | |
8592f29d N |
947 | if (!have_container && |
948 | dv == moved_disk && dnum != insert_point) break; | |
52826846 | 949 | } |
edd8d13c | 950 | if (pass == 1) { |
9b1fb677 DW |
951 | struct mdinfo info_new; |
952 | struct map_ent *me = NULL; | |
953 | ||
954 | /* check to see if the uuid has changed due to these | |
955 | * metadata changes, and if so update the member array | |
956 | * and container uuid. Note ->write_init_super clears | |
957 | * the subarray cursor such that ->getinfo_super once | |
958 | * again returns container info. | |
959 | */ | |
a5d85af7 | 960 | st->ss->getinfo_super(st, &info_new, NULL); |
99cc42f4 | 961 | if (st->ss->external && s->level != LEVEL_CONTAINER && |
9b1fb677 | 962 | !same_uuid(info_new.uuid, info.uuid, 0)) { |
4dd2df09 | 963 | map_update(&map, fd2devnm(mdfd), |
9b1fb677 DW |
964 | info_new.text_version, |
965 | info_new.uuid, chosen_name); | |
4dd2df09 | 966 | me = map_by_devnm(&map, st->container_devnm); |
9b1fb677 DW |
967 | } |
968 | ||
ab80e597 | 969 | if (st->ss->write_init_super(st)) { |
ab80e597 | 970 | st->ss->free_super(st); |
e06af9dd | 971 | goto abort_locked; |
ab80e597 | 972 | } |
9b1fb677 DW |
973 | |
974 | /* update parent container uuid */ | |
975 | if (me) { | |
503975b9 | 976 | char *path = xstrdup(me->path); |
9b1fb677 | 977 | |
a5d85af7 | 978 | st->ss->getinfo_super(st, &info_new, NULL); |
4dd2df09 | 979 | map_update(&map, st->container_devnm, |
9b1fb677 DW |
980 | info_new.text_version, |
981 | info_new.uuid, path); | |
982 | free(path); | |
983 | } | |
9b1fb677 | 984 | |
edd8d13c | 985 | flush_metadata_updates(st); |
6946681d | 986 | st->ss->free_super(st); |
edd8d13c | 987 | } |
5e52ae9e | 988 | } |
19ad4b2c | 989 | map_unlock(&map); |
d2ca6449 | 990 | free(infos); |
5e52ae9e | 991 | |
99cc42f4 | 992 | if (s->level == LEVEL_CONTAINER) { |
97590376 N |
993 | /* No need to start. But we should signal udev to |
994 | * create links */ | |
995 | sysfs_uevent(&info, "change"); | |
171dccc8 | 996 | if (c->verbose >= 0) |
e7b84f9d | 997 | pr_err("container %s prepared.\n", mddev); |
a7c6e3fb | 998 | wait_for(chosen_name, mdfd); |
99cc42f4 | 999 | } else if (c->runstop == 1 || subdevs >= s->raiddisks) { |
598f0d58 | 1000 | if (st->ss->external) { |
4c821454 | 1001 | int err; |
99cc42f4 | 1002 | switch(s->level) { |
598f0d58 NB |
1003 | case LEVEL_LINEAR: |
1004 | case LEVEL_MULTIPATH: | |
1005 | case 0: | |
4c821454 | 1006 | err = sysfs_set_str(&info, NULL, "array_state", |
171dccc8 | 1007 | c->readonly |
72d566f6 N |
1008 | ? "readonly" |
1009 | : "active"); | |
a931db9e | 1010 | need_mdmon = 0; |
598f0d58 NB |
1011 | break; |
1012 | default: | |
4c821454 N |
1013 | err = sysfs_set_str(&info, NULL, "array_state", |
1014 | "readonly"); | |
598f0d58 NB |
1015 | break; |
1016 | } | |
f35f2525 | 1017 | sysfs_set_safemode(&info, safe_mode_delay); |
4c821454 | 1018 | if (err) { |
7a862a02 | 1019 | pr_err("failed to activate array.\n"); |
4c821454 N |
1020 | ioctl(mdfd, STOP_ARRAY, NULL); |
1021 | goto abort; | |
1022 | } | |
171dccc8 | 1023 | } else if (c->readonly && |
72d566f6 N |
1024 | sysfs_attribute_available( |
1025 | &info, NULL, "array_state")) { | |
1026 | if (sysfs_set_str(&info, NULL, | |
1027 | "array_state", "readonly") < 0) { | |
1028 | pr_err("Failed to start array: %s\n", | |
1029 | strerror(errno)); | |
1030 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1031 | goto abort; | |
1032 | } | |
598f0d58 | 1033 | } else { |
97590376 | 1034 | /* param is not actually used */ |
598f0d58 NB |
1035 | mdu_param_t param; |
1036 | if (ioctl(mdfd, RUN_ARRAY, ¶m)) { | |
e7b84f9d | 1037 | pr_err("RUN_ARRAY failed: %s\n", |
72d566f6 | 1038 | strerror(errno)); |
a252c078 | 1039 | if (info.array.chunk_size & (info.array.chunk_size-1)) { |
7a862a02 | 1040 | cont_err("Problem may be that chunk size is not a power of 2\n"); |
a252c078 | 1041 | } |
4eb26970 | 1042 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1043 | goto abort; |
598f0d58 | 1044 | } |
a21e848a N |
1045 | /* if start_ro module parameter is set, array is |
1046 | * auto-read-only, which is bad as the resync won't | |
1047 | * start. So lets make it read-write now. | |
1048 | */ | |
1049 | ioctl(mdfd, RESTART_ARRAY_RW, NULL); | |
682c7051 | 1050 | } |
171dccc8 | 1051 | if (c->verbose >= 0) |
e7b84f9d | 1052 | pr_err("array %s started.\n", mddev); |
4dd2df09 | 1053 | if (st->ss->external && st->container_devnm[0]) { |
8850ee3e | 1054 | if (need_mdmon) |
4dd2df09 | 1055 | start_mdmon(st->container_devnm); |
8850ee3e | 1056 | |
4dd2df09 | 1057 | ping_monitor(st->container_devnm); |
a931db9e NB |
1058 | close(container_fd); |
1059 | } | |
a7c6e3fb | 1060 | wait_for(chosen_name, mdfd); |
682c7051 | 1061 | } else { |
e7b84f9d | 1062 | pr_err("not starting array - not enough devices.\n"); |
682c7051 | 1063 | } |
7f91af49 | 1064 | close(mdfd); |
682c7051 | 1065 | return 0; |
7f91af49 N |
1066 | |
1067 | abort: | |
4eb26970 | 1068 | map_lock(&map); |
e06af9dd | 1069 | abort_locked: |
4dd2df09 | 1070 | map_remove(&map, fd2devnm(mdfd)); |
4eb26970 DW |
1071 | map_unlock(&map); |
1072 | ||
7f91af49 N |
1073 | if (mdfd >= 0) |
1074 | close(mdfd); | |
1075 | return 1; | |
64c4757e | 1076 | } |