]>
Commit | Line | Data |
---|---|---|
64c4757e | 1 | /* |
9a9dab36 | 2 | * mdadm - manage Linux "md" devices aka RAID arrays. |
64c4757e | 3 | * |
4f589ad0 | 4 | * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de> |
64c4757e NB |
5 | * |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | * Author: Neil Brown | |
22 | * Email: <neilb@cse.unsw.edu.au> | |
23 | * Paper: Neil Brown | |
24 | * School of Computer Science and Engineering | |
25 | * The University of New South Wales | |
26 | * Sydney, 2052 | |
27 | * Australia | |
28 | */ | |
29 | ||
9a9dab36 | 30 | #include "mdadm.h" |
682c7051 NB |
31 | #include "md_u.h" |
32 | #include "md_p.h" | |
60248f74 | 33 | #include <ctype.h> |
64c4757e | 34 | |
a18a888e DW |
35 | static int default_layout(struct supertype *st, int level, int verbose) |
36 | { | |
37 | int layout = UnSet; | |
38 | ||
39 | if (st && st->ss->default_layout) | |
40 | layout = st->ss->default_layout(level); | |
41 | ||
42 | if (layout == UnSet) | |
43 | switch(level) { | |
44 | default: /* no layout */ | |
45 | layout = 0; | |
46 | break; | |
47 | case 10: | |
48 | layout = 0x102; /* near=2, far=1 */ | |
49 | if (verbose > 0) | |
50 | fprintf(stderr, | |
51 | Name ": layout defaults to n1\n"); | |
52 | break; | |
53 | case 5: | |
54 | case 6: | |
55 | layout = map_name(r5layout, "default"); | |
56 | if (verbose > 0) | |
57 | fprintf(stderr, | |
58 | Name ": layout defaults to %s\n", map_num(r5layout, layout)); | |
59 | break; | |
60 | case LEVEL_FAULTY: | |
61 | layout = map_name(faultylayout, "default"); | |
62 | ||
63 | if (verbose > 0) | |
64 | fprintf(stderr, | |
65 | Name ": layout defaults to %s\n", map_num(faultylayout, layout)); | |
66 | break; | |
67 | } | |
68 | ||
69 | return layout; | |
70 | } | |
71 | ||
72 | ||
7f91af49 | 73 | int Create(struct supertype *st, char *mddev, |
5dd497ee | 74 | int chunk, int level, int layout, unsigned long long size, int raiddisks, int sparedisks, |
3d3dd91e | 75 | char *name, char *homehost, int *uuid, |
cd29a5c8 | 76 | int subdevs, mddev_dev_t devlist, |
47d79ef8 | 77 | int runstop, int verbose, int force, int assume_clean, |
7f91af49 | 78 | char *bitmap_file, int bitmap_chunk, int write_behind, int delay, int autof) |
64c4757e | 79 | { |
682c7051 NB |
80 | /* |
81 | * Create a new raid array. | |
82 | * | |
83 | * First check that necessary details are available | |
84 | * (i.e. level, raid-disks) | |
85 | * | |
86 | * Then check each disk to see what might be on it | |
87 | * and report anything interesting. | |
88 | * | |
89 | * If anything looks odd, and runstop not set, | |
90 | * abort. | |
91 | * | |
92 | * SET_ARRAY_INFO and ADD_NEW_DISK, and | |
82d9eba6 | 93 | * if runstop==run, or raiddisks disks were used, |
682c7051 NB |
94 | * RUN_ARRAY |
95 | */ | |
7f91af49 | 96 | int mdfd; |
b5e64645 | 97 | unsigned long long minsize=0, maxsize=0; |
cd29a5c8 NB |
98 | char *mindisc = NULL; |
99 | char *maxdisc = NULL; | |
c913b90e | 100 | int dnum; |
cd29a5c8 | 101 | mddev_dev_t dv; |
682c7051 | 102 | int fail=0, warn=0; |
82b27616 | 103 | struct stat stb; |
82d9eba6 | 104 | int first_missing = subdevs * 2; |
6fb79233 | 105 | int second_missing = subdevs * 2; |
52826846 | 106 | int missing_disks = 0; |
82d9eba6 | 107 | int insert_point = subdevs * 2; /* where to insert a missing drive */ |
d2ca6449 | 108 | int total_slots; |
4b1ac34b | 109 | int pass; |
82d9eba6 NB |
110 | int vers; |
111 | int rv; | |
c82f047c | 112 | int bitmap_fd; |
5f8097be | 113 | int have_container = 0; |
0e600426 | 114 | int container_fd = -1; |
a931db9e | 115 | int need_mdmon = 0; |
f9c25f1d | 116 | unsigned long long bitmapsize; |
d2ca6449 | 117 | struct mdinfo info, *infos; |
b8ac1967 | 118 | int did_default = 0; |
a18a888e | 119 | int do_default_layout = 0; |
8ed3e5e1 | 120 | unsigned long safe_mode_delay = 0; |
a04d5763 | 121 | char chosen_name[1024]; |
ad5bc697 | 122 | struct map_ent *map = NULL; |
8592f29d | 123 | unsigned long long newsize; |
682c7051 | 124 | |
943eafef | 125 | int major_num = BITMAP_MAJOR_HI; |
dcec9ee5 | 126 | |
06c7f68e | 127 | memset(&info, 0, sizeof(info)); |
29e766a5 | 128 | |
a322f70c | 129 | if (level == UnSet) { |
3cfe6247 | 130 | /* "ddf" and "imsm" metadata only supports one level - should possibly |
a322f70c DW |
131 | * push this into metadata handler?? |
132 | */ | |
3cfe6247 | 133 | if (st && (st->ss == &super_ddf || st->ss == &super_imsm)) |
a322f70c DW |
134 | level = LEVEL_CONTAINER; |
135 | } | |
136 | ||
98c6faba | 137 | if (level == UnSet) { |
682c7051 NB |
138 | fprintf(stderr, |
139 | Name ": a RAID level is needed to create an array.\n"); | |
140 | return 1; | |
141 | } | |
98c6faba NB |
142 | if (raiddisks < 4 && level == 6) { |
143 | fprintf(stderr, | |
144 | Name ": at least 4 raid-devices needed for level 6\n"); | |
145 | return 1; | |
146 | } | |
147 | if (raiddisks > 256 && level == 6) { | |
148 | fprintf(stderr, | |
149 | Name ": no more than 256 raid-devices supported for level 6\n"); | |
150 | return 1; | |
151 | } | |
52826846 NB |
152 | if (raiddisks < 2 && level >= 4) { |
153 | fprintf(stderr, | |
98c6faba | 154 | Name ": at least 2 raid-devices needed for level 4 or 5\n"); |
52826846 NB |
155 | return 1; |
156 | } | |
570510ba NB |
157 | if (level <= 0 && sparedisks) { |
158 | fprintf(stderr, | |
159 | Name ": This level does not support spare devices\n"); | |
160 | return 1; | |
161 | } | |
5f8097be NB |
162 | |
163 | if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) { | |
164 | /* If given a single device, it might be a container, and we can | |
165 | * extract a device list from there | |
166 | */ | |
167 | mdu_array_info_t inf; | |
168 | int fd; | |
169 | ||
170 | memset(&inf, 0, sizeof(inf)); | |
37ea3936 | 171 | fd = open(devlist->devname, O_RDONLY); |
5f8097be NB |
172 | if (fd >= 0 && |
173 | ioctl(fd, GET_ARRAY_INFO, &inf) == 0 && | |
174 | inf.raid_disks == 0) { | |
175 | /* yep, looks like a container */ | |
176 | if (st) { | |
177 | rv = st->ss->load_super(st, fd, | |
178 | devlist->devname); | |
179 | if (rv == 0) | |
180 | have_container = 1; | |
181 | } else { | |
182 | st = guess_super(fd); | |
183 | if (st && !(rv = st->ss-> | |
184 | load_super(st, fd, | |
185 | devlist->devname))) | |
186 | have_container = 1; | |
187 | else | |
188 | st = NULL; | |
189 | } | |
8592f29d N |
190 | if (have_container) |
191 | subdevs = raiddisks; | |
5f8097be NB |
192 | } |
193 | if (fd >= 0) | |
194 | close(fd); | |
5f8097be | 195 | } |
ffcfc735 N |
196 | if (st && st->ss->external && sparedisks) { |
197 | fprintf(stderr, | |
198 | Name ": This metadata type does not support " | |
199 | "spare disks are create time\n"); | |
200 | return 1; | |
201 | } | |
682c7051 | 202 | if (subdevs > raiddisks+sparedisks) { |
b83d95f3 | 203 | fprintf(stderr, Name ": You have listed more devices (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks); |
52826846 | 204 | return 1; |
682c7051 | 205 | } |
5f8097be | 206 | if (!have_container && subdevs < raiddisks+sparedisks) { |
52826846 NB |
207 | fprintf(stderr, Name ": You haven't given enough devices (real or missing) to create this array\n"); |
208 | return 1; | |
209 | } | |
5b28bd56 NB |
210 | if (bitmap_file && level <= 0) { |
211 | fprintf(stderr, Name ": bitmaps not meaningful with level %s\n", | |
212 | map_num(pers, level)?:"given"); | |
213 | return 1; | |
214 | } | |
52826846 | 215 | |
682c7051 | 216 | /* now set some defaults */ |
b5e64645 | 217 | |
a18a888e DW |
218 | |
219 | if (layout == UnSet) { | |
220 | do_default_layout = 1; | |
221 | layout = default_layout(st, level, verbose); | |
222 | } | |
682c7051 | 223 | |
e5329c37 NB |
224 | if (level == 10) |
225 | /* check layout fits in array*/ | |
226 | if ((layout&255) * ((layout>>8)&255) > raiddisks) { | |
227 | fprintf(stderr, Name ": that layout requires at least %d devices\n", | |
228 | (layout&255) * ((layout>>8)&255)); | |
229 | return 1; | |
230 | } | |
231 | ||
aa88f531 NB |
232 | switch(level) { |
233 | case 4: | |
234 | case 5: | |
e5329c37 | 235 | case 10: |
98c6faba | 236 | case 6: |
aa88f531 | 237 | case 0: |
570510ba | 238 | case LEVEL_LINEAR: /* linear */ |
aa88f531 NB |
239 | if (chunk == 0) { |
240 | chunk = 64; | |
dab6685f | 241 | if (verbose > 0) |
aa88f531 NB |
242 | fprintf(stderr, Name ": chunk size defaults to 64K\n"); |
243 | } | |
244 | break; | |
570510ba | 245 | case 1: |
08e43379 | 246 | case LEVEL_FAULTY: |
570510ba | 247 | case LEVEL_MULTIPATH: |
17f25ca6 | 248 | case LEVEL_CONTAINER: |
aa88f531 NB |
249 | if (chunk) { |
250 | chunk = 0; | |
dab6685f | 251 | if (verbose > 0) |
aa88f531 NB |
252 | fprintf(stderr, Name ": chunk size ignored for this level\n"); |
253 | } | |
254 | break; | |
570510ba NB |
255 | default: |
256 | fprintf(stderr, Name ": unknown level %d\n", level); | |
257 | return 1; | |
682c7051 | 258 | } |
8592f29d N |
259 | |
260 | newsize = size * 2; | |
17f25ca6 | 261 | if (st && ! st->ss->validate_geometry(st, level, layout, raiddisks, |
8592f29d | 262 | chunk, size*2, NULL, &newsize, verbose>=0)) |
17f25ca6 | 263 | return 1; |
8592f29d N |
264 | if (size == 0) { |
265 | size = newsize / 2; | |
266 | if (size && verbose > 0) | |
267 | fprintf(stderr, Name ": setting size to %lluK\n", | |
268 | (unsigned long long)size); | |
269 | } | |
17f25ca6 | 270 | |
682c7051 | 271 | /* now look at the subdevs */ |
06c7f68e NB |
272 | info.array.active_disks = 0; |
273 | info.array.working_disks = 0; | |
c913b90e | 274 | dnum = 0; |
8592f29d | 275 | for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) { |
cd29a5c8 | 276 | char *dname = dv->devname; |
17f25ca6 | 277 | unsigned long long freesize; |
cd29a5c8 | 278 | if (strcasecmp(dname, "missing")==0) { |
c913b90e NB |
279 | if (first_missing > dnum) |
280 | first_missing = dnum; | |
6fb79233 NB |
281 | if (second_missing > dnum && dnum > first_missing) |
282 | second_missing = dnum; | |
52826846 NB |
283 | missing_disks ++; |
284 | continue; | |
285 | } | |
06c7f68e | 286 | info.array.working_disks++; |
c913b90e | 287 | if (dnum < raiddisks) |
06c7f68e | 288 | info.array.active_disks++; |
058574b1 | 289 | if (st == NULL) { |
8aec876d | 290 | struct createinfo *ci = conf_get_create_info(); |
058574b1 NB |
291 | if (ci) |
292 | st = ci->supertype; | |
293 | } | |
576d6d83 NB |
294 | if (st == NULL) { |
295 | /* Need to choose a default metadata, which is different | |
17f25ca6 | 296 | * depending on geometry of array. |
576d6d83 NB |
297 | */ |
298 | int i; | |
299 | char *name = "default"; | |
17f25ca6 | 300 | for(i=0; !st && superlist[i]; i++) { |
576d6d83 | 301 | st = superlist[i]->match_metadata_desc(name); |
a18a888e DW |
302 | if (do_default_layout) |
303 | layout = default_layout(st, level, verbose); | |
17f25ca6 NB |
304 | if (st && !st->ss->validate_geometry |
305 | (st, level, layout, raiddisks, | |
e46273eb | 306 | chunk, size*2, dname, &freesize, |
2c514b71 | 307 | verbose > 0)) |
17f25ca6 NB |
308 | st = NULL; |
309 | } | |
576d6d83 NB |
310 | |
311 | if (!st) { | |
17f25ca6 NB |
312 | fprintf(stderr, Name ": device %s not suitable " |
313 | "for any style of array\n", | |
314 | dname); | |
576d6d83 NB |
315 | exit(2); |
316 | } | |
b8ac1967 | 317 | if (st->ss != &super0 || |
576d6d83 | 318 | st->minor_version != 90) |
b8ac1967 | 319 | did_default = 1; |
17f25ca6 | 320 | } else { |
a18a888e DW |
321 | if (do_default_layout) |
322 | layout = default_layout(st, level, verbose); | |
17f25ca6 NB |
323 | if (!st->ss->validate_geometry(st, level, layout, |
324 | raiddisks, | |
e46273eb | 325 | chunk, size*2, dname, |
2c514b71 NB |
326 | &freesize, |
327 | verbose > 0)) { | |
17f25ca6 NB |
328 | |
329 | fprintf(stderr, | |
330 | Name ": %s is not suitable for " | |
331 | "this array.\n", | |
332 | dname); | |
333 | fail = 1; | |
334 | continue; | |
335 | } | |
682c7051 | 336 | } |
82d9eba6 NB |
337 | |
338 | freesize /= 2; /* convert to K */ | |
d2cd3ffc NB |
339 | if (chunk) { |
340 | /* round to chunk size */ | |
341 | freesize = freesize & ~(chunk-1); | |
342 | } | |
682c7051 NB |
343 | |
344 | if (size && freesize < size) { | |
42dc2744 | 345 | fprintf(stderr, Name ": %s is smaller than given size." |
17f25ca6 NB |
346 | " %lluK < %lluK + metadata\n", |
347 | dname, freesize, size); | |
52826846 | 348 | fail = 1; |
52826846 | 349 | continue; |
682c7051 | 350 | } |
cd29a5c8 NB |
351 | if (maxdisc == NULL || (maxdisc && freesize > maxsize)) { |
352 | maxdisc = dname; | |
52826846 | 353 | maxsize = freesize; |
682c7051 | 354 | } |
cd29a5c8 NB |
355 | if (mindisc ==NULL || (mindisc && freesize < minsize)) { |
356 | mindisc = dname; | |
52826846 | 357 | minsize = freesize; |
682c7051 | 358 | } |
dab6685f | 359 | if (runstop != 1 || verbose >= 0) { |
37ea3936 | 360 | int fd = open(dname, O_RDONLY); |
b6e63da4 NB |
361 | if (fd <0 ) { |
362 | fprintf(stderr, Name ": Cannot open %s: %s\n", | |
363 | dname, strerror(errno)); | |
364 | fail=1; | |
365 | continue; | |
366 | } | |
dab6685f NB |
367 | warn |= check_ext2(fd, dname); |
368 | warn |= check_reiser(fd, dname); | |
369 | warn |= check_raid(fd, dname); | |
b6e63da4 | 370 | close(fd); |
dab6685f | 371 | } |
682c7051 | 372 | } |
8592f29d N |
373 | if (have_container) |
374 | info.array.working_disks = raiddisks; | |
682c7051 | 375 | if (fail) { |
52826846 NB |
376 | fprintf(stderr, Name ": create aborted\n"); |
377 | return 1; | |
682c7051 NB |
378 | } |
379 | if (size == 0) { | |
5f8097be | 380 | if (mindisc == NULL && !have_container) { |
52826846 NB |
381 | fprintf(stderr, Name ": no size and no drives given - aborting create.\n"); |
382 | return 1; | |
383 | } | |
5f8097be NB |
384 | if (level > 0 || level == LEVEL_MULTIPATH |
385 | || level == LEVEL_FAULTY | |
b8ac1967 | 386 | || st->ss->external ) { |
b5e64645 | 387 | /* size is meaningful */ |
b8ac1967 NB |
388 | if (!st->ss->validate_geometry(st, level, layout, |
389 | raiddisks, | |
e46273eb | 390 | chunk, minsize*2, |
2c514b71 | 391 | NULL, NULL, 0)) { |
aba69144 | 392 | fprintf(stderr, Name ": devices too large for RAID level %d\n", level); |
b5e64645 NB |
393 | return 1; |
394 | } | |
395 | size = minsize; | |
dab6685f | 396 | if (verbose > 0) |
5dd497ee | 397 | fprintf(stderr, Name ": size set to %lluK\n", size); |
b5e64645 | 398 | } |
682c7051 | 399 | } |
8592f29d | 400 | if (!have_container && level > 0 && ((maxsize-size)*100 > maxsize)) { |
dab6685f | 401 | if (runstop != 1 || verbose >= 0) |
8592f29d | 402 | fprintf(stderr, Name ": largest drive (%s) exceeds size (%lluK) by more than 1%%\n", |
dab6685f | 403 | maxdisc, size); |
52826846 | 404 | warn = 1; |
682c7051 NB |
405 | } |
406 | ||
5615172f DW |
407 | if (st->ss->detail_platform && st->ss->detail_platform(0, 1) != 0) { |
408 | if (runstop != 1 || verbose >= 0) | |
409 | fprintf(stderr, Name ": %s unable to enumerate platform support\n" | |
410 | " array may not be compatible with hardware/firmware\n", | |
411 | st->ss->name); | |
412 | warn = 1; | |
413 | } | |
414 | ||
682c7051 | 415 | if (warn) { |
52826846 NB |
416 | if (runstop!= 1) { |
417 | if (!ask("Continue creating array? ")) { | |
418 | fprintf(stderr, Name ": create aborted.\n"); | |
419 | return 1; | |
420 | } | |
421 | } else { | |
dab6685f | 422 | if (verbose > 0) |
52826846 | 423 | fprintf(stderr, Name ": creation continuing despite oddities due to --run\n"); |
682c7051 | 424 | } |
682c7051 NB |
425 | } |
426 | ||
66f8bbbe | 427 | /* If this is raid4/5, we want to configure the last active slot |
52826846 | 428 | * as missing, so that a reconstruct happens (faster than re-parity) |
98c6faba | 429 | * FIX: Can we do this for raid6 as well? |
52826846 | 430 | */ |
87b47257 N |
431 | if (st->ss->external == 0 && |
432 | assume_clean==0 && force == 0 && first_missing >= raiddisks) { | |
98c6faba | 433 | switch ( level ) { |
66f8bbbe | 434 | case 4: |
98c6faba NB |
435 | case 5: |
436 | insert_point = raiddisks-1; | |
437 | sparedisks++; | |
06c7f68e | 438 | info.array.active_disks--; |
98c6faba NB |
439 | missing_disks++; |
440 | break; | |
441 | default: | |
442 | break; | |
443 | } | |
52826846 | 444 | } |
6fb79233 NB |
445 | /* For raid6, if creating with 1 missing drive, make a good drive |
446 | * into a spare, else the create will fail | |
447 | */ | |
448 | if (assume_clean == 0 && force == 0 && first_missing < raiddisks && | |
ffcfc735 | 449 | st->ss->external == 0 && |
6fb79233 NB |
450 | second_missing >= raiddisks && level == 6) { |
451 | insert_point = raiddisks - 1; | |
452 | if (insert_point == first_missing) | |
453 | insert_point--; | |
454 | sparedisks ++; | |
455 | info.array.active_disks--; | |
456 | missing_disks++; | |
457 | } | |
570510ba | 458 | |
5f8097be | 459 | if (level <= 0 && first_missing < subdevs * 2) { |
570510ba NB |
460 | fprintf(stderr, |
461 | Name ": This level does not support missing devices\n"); | |
462 | return 1; | |
463 | } | |
aba69144 | 464 | |
7f91af49 | 465 | /* We need to create the device */ |
ad5bc697 | 466 | map_lock(&map); |
a04d5763 | 467 | mdfd = create_mddev(mddev, name, autof, LOCAL, chosen_name); |
7f91af49 N |
468 | if (mdfd < 0) |
469 | return 1; | |
f2be09f1 | 470 | mddev = chosen_name; |
7f91af49 N |
471 | |
472 | vers = md_get_version(mdfd); | |
473 | if (vers < 9000) { | |
474 | fprintf(stderr, Name ": Create requires md driver version 0.90.0 or later\n"); | |
475 | goto abort; | |
476 | } else { | |
477 | mdu_array_info_t inf; | |
478 | memset(&inf, 0, sizeof(inf)); | |
479 | ioctl(mdfd, GET_ARRAY_INFO, &inf); | |
480 | if (inf.working_disks != 0) { | |
481 | fprintf(stderr, Name ": another array by this name" | |
482 | " is already running.\n"); | |
483 | goto abort; | |
484 | } | |
485 | } | |
a04d5763 | 486 | |
682c7051 NB |
487 | /* Ok, lets try some ioctls */ |
488 | ||
06c7f68e NB |
489 | info.array.level = level; |
490 | info.array.size = size; | |
491 | info.array.raid_disks = raiddisks; | |
82b27616 NB |
492 | /* The kernel should *know* what md_minor we are dealing |
493 | * with, but it chooses to trust me instead. Sigh | |
494 | */ | |
06c7f68e | 495 | info.array.md_minor = 0; |
82b27616 | 496 | if (fstat(mdfd, &stb)==0) |
06c7f68e NB |
497 | info.array.md_minor = minor(stb.st_rdev); |
498 | info.array.not_persistent = 0; | |
6fb79233 | 499 | |
66f8bbbe | 500 | if ( ( (level == 4 || level == 5) && |
b5e64645 NB |
501 | (insert_point < raiddisks || first_missing < raiddisks) ) |
502 | || | |
6fb79233 NB |
503 | ( level == 6 && (insert_point < raiddisks |
504 | || second_missing < raiddisks)) | |
47d79ef8 | 505 | || |
d024b0a7 NB |
506 | ( level <= 0 ) |
507 | || | |
47d79ef8 | 508 | assume_clean |
103f2410 | 509 | ) { |
06c7f68e | 510 | info.array.state = 1; /* clean, but one+ drive will be missing*/ |
103f2410 NB |
511 | info.resync_start = ~0ULL; |
512 | } else { | |
06c7f68e | 513 | info.array.state = 0; /* not clean, but no errors */ |
103f2410 NB |
514 | info.resync_start = 0; |
515 | } | |
f9c25f1d NB |
516 | if (level == 10) { |
517 | /* for raid10, the bitmap size is the capacity of the array, | |
518 | * which is array.size * raid_disks / ncopies; | |
519 | * .. but convert to sectors. | |
520 | */ | |
702b557b | 521 | int ncopies = ((layout>>8) & 255) * (layout & 255); |
5dd497ee NB |
522 | bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2; |
523 | /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/ | |
f9c25f1d | 524 | } else |
5dd497ee | 525 | bitmapsize = (unsigned long long)size * 2; |
f9c25f1d | 526 | |
82b27616 NB |
527 | /* There is lots of redundancy in these disk counts, |
528 | * raid_disks is the most meaningful value | |
529 | * it describes the geometry of the array | |
530 | * it is constant | |
531 | * nr_disks is total number of used slots. | |
532 | * it should be raid_disks+spare_disks | |
533 | * spare_disks is the number of extra disks present | |
534 | * see above | |
535 | * active_disks is the number of working disks in | |
536 | * active slots. (With raid_disks) | |
537 | * working_disks is the total number of working disks, | |
538 | * including spares | |
539 | * failed_disks is the number of disks marked failed | |
540 | * | |
541 | * Ideally, the kernel would keep these (except raid_disks) | |
542 | * up-to-date as we ADD_NEW_DISK, but it doesn't (yet). | |
543 | * So for now, we assume that all raid and spare | |
544 | * devices will be given. | |
545 | */ | |
06c7f68e NB |
546 | info.array.spare_disks=sparedisks; |
547 | info.array.failed_disks=missing_disks; | |
548 | info.array.nr_disks = info.array.working_disks | |
549 | + info.array.failed_disks; | |
550 | info.array.layout = layout; | |
551 | info.array.chunk_size = chunk*1024; | |
82d9eba6 | 552 | |
b3b33eb5 NB |
553 | if (name == NULL || *name == 0) { |
554 | /* base name on mddev */ | |
d1e80164 NB |
555 | /* /dev/md0 -> 0 |
556 | * /dev/md_d0 -> d0 | |
557 | * /dev/md/1 -> 1 | |
558 | * /dev/md/d1 -> d1 | |
559 | * /dev/md/home -> home | |
560 | * /dev/mdhome -> home | |
561 | */ | |
69207ff6 | 562 | /* FIXME compare this with rules in create_mddev */ |
b3b33eb5 NB |
563 | name = strrchr(mddev, '/'); |
564 | if (name) { | |
565 | name++; | |
60248f74 NB |
566 | if (strncmp(name, "md_d", 4)==0 && |
567 | strlen(name) > 4 && | |
568 | isdigit(name[4]) && | |
b3b33eb5 | 569 | (name-mddev) == 5 /* /dev/ */) |
d1e80164 | 570 | name += 3; |
60248f74 NB |
571 | else if (strncmp(name, "md", 2)==0 && |
572 | strlen(name) > 2 && | |
573 | isdigit(name[2]) && | |
574 | (name-mddev) == 5 /* /dev/ */) | |
b3b33eb5 NB |
575 | name += 2; |
576 | } | |
577 | } | |
06c7f68e | 578 | if (!st->ss->init_super(st, &info.array, size, name, homehost, uuid)) |
7f91af49 | 579 | goto abort; |
682c7051 | 580 | |
d2ca6449 | 581 | total_slots = info.array.nr_disks; |
f35f2525 | 582 | sysfs_init(&info, mdfd, 0); |
159c3a1a NB |
583 | st->ss->getinfo_super(st, &info); |
584 | ||
f7f1b6a1 | 585 | if (did_default && verbose >= 0) { |
3c558363 | 586 | if (is_subarray(info.text_version)) { |
f7f1b6a1 NB |
587 | int dnum = devname2devnum(info.text_version+1); |
588 | char *path; | |
589 | int mdp = get_mdp_major(); | |
590 | struct mdinfo *mdi; | |
591 | if (dnum > 0) | |
592 | path = map_dev(MD_MAJOR, dnum, 1); | |
593 | else | |
594 | path = map_dev(mdp, (-1-dnum)<< 6, 1); | |
595 | ||
596 | mdi = sysfs_read(-1, dnum, GET_VERSION); | |
597 | ||
598 | fprintf(stderr, Name ": Creating array inside " | |
599 | "%s container %s\n", | |
600 | mdi?mdi->text_version:"managed", path); | |
601 | sysfs_free(mdi); | |
602 | } else | |
603 | fprintf(stderr, Name ": Defaulting to version" | |
604 | " %s metadata\n", info.text_version); | |
605 | } | |
b8ac1967 | 606 | |
ad5bc697 | 607 | map_update(&map, fd2devnum(mdfd), info.text_version, |
a04d5763 | 608 | info.uuid, chosen_name); |
ad5bc697 | 609 | map_unlock(&map); |
a04d5763 | 610 | |
dcec9ee5 | 611 | if (bitmap_file && vers < 9003) { |
943eafef | 612 | major_num = BITMAP_MAJOR_HOSTENDIAN; |
dcec9ee5 NB |
613 | #ifdef __BIG_ENDIAN |
614 | fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n" | |
615 | " between different architectured. Consider upgrading the Linux kernel.\n"); | |
616 | #endif | |
617 | } | |
618 | ||
55935d51 NB |
619 | if (bitmap_file && strcmp(bitmap_file, "internal")==0) { |
620 | if ((vers%100) < 2) { | |
621 | fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n"); | |
7f91af49 | 622 | goto abort; |
55935d51 | 623 | } |
3da92f27 | 624 | if (!st->ss->add_internal_bitmap(st, &bitmap_chunk, |
199171a2 | 625 | delay, write_behind, |
943eafef | 626 | bitmapsize, 1, major_num)) { |
55935d51 | 627 | fprintf(stderr, Name ": Given bitmap chunk size not supported.\n"); |
7f91af49 | 628 | goto abort; |
55935d51 NB |
629 | } |
630 | bitmap_file = NULL; | |
631 | } | |
632 | ||
633 | ||
f35f2525 | 634 | sysfs_init(&info, mdfd, 0); |
55935d51 | 635 | |
f35f2525 N |
636 | if (st->ss->external && st->subarray[0]) { |
637 | /* member */ | |
638 | ||
639 | /* When creating a member, we need to be careful | |
640 | * to negotiate with mdmon properly. | |
641 | * If it is already running, we cannot write to | |
642 | * the devices and must ask it to do that part. | |
643 | * If it isn't running, we write to the devices, | |
644 | * and then start it. | |
645 | * We hold an exclusive open on the container | |
646 | * device to make sure mdmon doesn't exit after | |
647 | * we checked that it is running. | |
648 | * | |
649 | * For now, fail if it is already running. | |
650 | */ | |
651 | container_fd = open_dev_excl(st->container_dev); | |
652 | if (container_fd < 0) { | |
653 | fprintf(stderr, Name ": Cannot get exclusive " | |
654 | "open on container - weird.\n"); | |
7f91af49 | 655 | goto abort; |
d03373f1 | 656 | } |
f35f2525 N |
657 | if (mdmon_running(st->container_dev)) { |
658 | if (verbose) | |
659 | fprintf(stderr, Name ": reusing mdmon " | |
660 | "for %s.\n", | |
661 | devnum2devname(st->container_dev)); | |
662 | st->update_tail = &st->updates; | |
663 | } else | |
664 | need_mdmon = 1; | |
665 | } | |
666 | rv = set_array_info(mdfd, st, &info); | |
82d9eba6 | 667 | if (rv) { |
f35f2525 | 668 | fprintf(stderr, Name ": failed to set array info for %s: %s\n", |
52826846 | 669 | mddev, strerror(errno)); |
7f91af49 | 670 | goto abort; |
682c7051 | 671 | } |
52826846 | 672 | |
c82f047c NB |
673 | if (bitmap_file) { |
674 | int uuid[4]; | |
55935d51 | 675 | |
3da92f27 | 676 | st->ss->uuid_from_super(st, uuid); |
dfd4d8ee | 677 | if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk, |
dcec9ee5 | 678 | delay, write_behind, |
f9c25f1d | 679 | bitmapsize, |
943eafef | 680 | major_num)) { |
7f91af49 | 681 | goto abort; |
c82f047c NB |
682 | } |
683 | bitmap_fd = open(bitmap_file, O_RDWR); | |
684 | if (bitmap_fd < 0) { | |
685 | fprintf(stderr, Name ": weird: %s cannot be openned\n", | |
dcec9ee5 | 686 | bitmap_file); |
7f91af49 | 687 | goto abort; |
c82f047c NB |
688 | } |
689 | if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
690 | fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n", | |
691 | mddev, strerror(errno)); | |
7f91af49 | 692 | goto abort; |
c82f047c NB |
693 | } |
694 | } | |
695 | ||
d2ca6449 | 696 | infos = malloc(sizeof(*infos) * total_slots); |
4b1ac34b | 697 | |
4b1ac34b NB |
698 | for (pass=1; pass <=2 ; pass++) { |
699 | mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */ | |
700 | ||
701 | for (dnum=0, dv = devlist ; dv ; | |
702 | dv=(dv->next)?(dv->next):moved_disk, dnum++) { | |
703 | int fd; | |
704 | struct stat stb; | |
d2ca6449 | 705 | struct mdinfo *inf = &infos[dnum]; |
4b1ac34b | 706 | |
d2ca6449 NB |
707 | if (dnum >= total_slots) |
708 | abort(); | |
4b1ac34b NB |
709 | if (dnum == insert_point) { |
710 | moved_disk = dv; | |
8592f29d | 711 | continue; |
52826846 | 712 | } |
8592f29d | 713 | if (strcasecmp(dv->devname, "missing")==0) |
111d01fc | 714 | continue; |
8592f29d N |
715 | if (have_container) |
716 | moved_disk = NULL; | |
717 | if (have_container && dnum < info.array.raid_disks - 1) | |
718 | /* repeatedly use the container */ | |
719 | moved_disk = dv; | |
111d01fc | 720 | |
d2ca6449 NB |
721 | switch(pass) { |
722 | case 1: | |
723 | *inf = info; | |
5f8097be | 724 | |
d2ca6449 NB |
725 | inf->disk.number = dnum; |
726 | inf->disk.raid_disk = dnum; | |
727 | if (inf->disk.raid_disk < raiddisks) | |
728 | inf->disk.state = (1<<MD_DISK_ACTIVE) | | |
729 | (1<<MD_DISK_SYNC); | |
730 | else | |
731 | inf->disk.state = 0; | |
732 | ||
b01b06bd | 733 | if (dv->writemostly == 1) |
d2ca6449 NB |
734 | inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY); |
735 | ||
8592f29d N |
736 | if (have_container) |
737 | fd = -1; | |
738 | else { | |
739 | if (st->ss->external && st->subarray[0]) | |
740 | fd = open(dv->devname, O_RDWR); | |
741 | else | |
742 | fd = open(dv->devname, O_RDWR|O_EXCL); | |
743 | ||
744 | if (fd < 0) { | |
745 | fprintf(stderr, Name ": failed to open %s " | |
746 | "after earlier success - aborting\n", | |
747 | dv->devname); | |
748 | goto abort; | |
749 | } | |
750 | fstat(fd, &stb); | |
751 | inf->disk.major = major(stb.st_rdev); | |
752 | inf->disk.minor = minor(stb.st_rdev); | |
d2ca6449 | 753 | } |
8592f29d N |
754 | if (fd >= 0) |
755 | remove_partitions(fd); | |
f20c3968 | 756 | if (st->ss->add_to_super(st, &inf->disk, |
208933a7 | 757 | fd, dv->devname)) |
f20c3968 | 758 | goto abort; |
d2ca6449 | 759 | st->ss->getinfo_super(st, inf); |
8ed3e5e1 | 760 | safe_mode_delay = inf->safe_mode_delay; |
d2ca6449 | 761 | |
8592f29d N |
762 | if (have_container && verbose > 0) |
763 | fprintf(stderr, Name ": Using %s for device %d\n", | |
764 | map_dev(inf->disk.major, | |
765 | inf->disk.minor, | |
766 | 0), dnum); | |
767 | ||
768 | if (!have_container) { | |
769 | /* getinfo_super might have lost these ... */ | |
770 | inf->disk.major = major(stb.st_rdev); | |
771 | inf->disk.minor = minor(stb.st_rdev); | |
772 | } | |
4b1ac34b NB |
773 | break; |
774 | case 2: | |
d2ca6449 | 775 | inf->errors = 0; |
2503d23b | 776 | rv = 0; |
4b1ac34b | 777 | |
f35f2525 | 778 | rv = add_disk(mdfd, st, &info, inf); |
d2ca6449 | 779 | |
2503d23b | 780 | if (rv) { |
d03373f1 NB |
781 | fprintf(stderr, |
782 | Name ": ADD_NEW_DISK for %s " | |
783 | "failed: %s\n", | |
4b1ac34b | 784 | dv->devname, strerror(errno)); |
3da92f27 | 785 | st->ss->free_super(st); |
7f91af49 | 786 | goto abort; |
4b1ac34b | 787 | } |
4b1ac34b NB |
788 | break; |
789 | } | |
8592f29d N |
790 | if (!have_container && |
791 | dv == moved_disk && dnum != insert_point) break; | |
52826846 | 792 | } |
edd8d13c | 793 | if (pass == 1) { |
111d01fc | 794 | st->ss->write_init_super(st); |
edd8d13c NB |
795 | flush_metadata_updates(st); |
796 | } | |
5e52ae9e | 797 | } |
d2ca6449 | 798 | free(infos); |
3da92f27 | 799 | st->ss->free_super(st); |
5e52ae9e | 800 | |
97590376 N |
801 | if (level == LEVEL_CONTAINER) { |
802 | /* No need to start. But we should signal udev to | |
803 | * create links */ | |
804 | sysfs_uevent(&info, "change"); | |
ed034b11 N |
805 | if (verbose >= 0) |
806 | fprintf(stderr, Name ": container %s prepared.\n", mddev); | |
97590376 | 807 | } else if (runstop == 1 || subdevs >= raiddisks) { |
598f0d58 NB |
808 | if (st->ss->external) { |
809 | switch(level) { | |
810 | case LEVEL_LINEAR: | |
811 | case LEVEL_MULTIPATH: | |
812 | case 0: | |
f35f2525 | 813 | sysfs_set_str(&info, NULL, "array_state", |
598f0d58 | 814 | "active"); |
a931db9e | 815 | need_mdmon = 0; |
598f0d58 NB |
816 | break; |
817 | default: | |
f35f2525 | 818 | sysfs_set_str(&info, NULL, "array_state", |
598f0d58 NB |
819 | "readonly"); |
820 | break; | |
821 | } | |
f35f2525 | 822 | sysfs_set_safemode(&info, safe_mode_delay); |
598f0d58 | 823 | } else { |
97590376 | 824 | /* param is not actually used */ |
598f0d58 NB |
825 | mdu_param_t param; |
826 | if (ioctl(mdfd, RUN_ARRAY, ¶m)) { | |
827 | fprintf(stderr, Name ": RUN_ARRAY failed: %s\n", | |
828 | strerror(errno)); | |
829 | Manage_runstop(mddev, mdfd, -1, 0); | |
7f91af49 | 830 | goto abort; |
598f0d58 | 831 | } |
682c7051 | 832 | } |
dab6685f NB |
833 | if (verbose >= 0) |
834 | fprintf(stderr, Name ": array %s started.\n", mddev); | |
f7e7067b | 835 | if (st->ss->external && st->subarray[0]) { |
8850ee3e N |
836 | if (need_mdmon) |
837 | start_mdmon(st->container_dev); | |
838 | ||
839 | ping_monitor(devnum2devname(st->container_dev)); | |
a931db9e NB |
840 | close(container_fd); |
841 | } | |
01cbfcc8 | 842 | wait_for(chosen_name); |
682c7051 | 843 | } else { |
b83d95f3 | 844 | fprintf(stderr, Name ": not starting array - not enough devices.\n"); |
682c7051 | 845 | } |
7f91af49 | 846 | close(mdfd); |
682c7051 | 847 | return 0; |
7f91af49 N |
848 | |
849 | abort: | |
850 | if (mdfd >= 0) | |
851 | close(mdfd); | |
852 | return 1; | |
64c4757e | 853 | } |