]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de> | |
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@suse.de> | |
23 | */ | |
24 | ||
25 | #include "mdadm.h" | |
26 | #include "md_u.h" | |
27 | #include "md_p.h" | |
28 | #include <ctype.h> | |
29 | ||
30 | static int default_layout(struct supertype *st, int level, int verbose) | |
31 | { | |
32 | int layout = UnSet; | |
33 | ||
34 | if (st && st->ss->default_geometry) | |
35 | st->ss->default_geometry(st, &level, &layout, NULL); | |
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) | |
45 | pr_err("layout defaults to n2\n"); | |
46 | break; | |
47 | case 5: | |
48 | case 6: | |
49 | layout = map_name(r5layout, "default"); | |
50 | if (verbose > 0) | |
51 | pr_err("layout defaults to %s\n", map_num(r5layout, layout)); | |
52 | break; | |
53 | case LEVEL_FAULTY: | |
54 | layout = map_name(faultylayout, "default"); | |
55 | ||
56 | if (verbose > 0) | |
57 | pr_err("layout defaults to %s\n", map_num(faultylayout, layout)); | |
58 | break; | |
59 | } | |
60 | ||
61 | return layout; | |
62 | } | |
63 | ||
64 | int Create(struct supertype *st, char *mddev, | |
65 | char *name, int *uuid, | |
66 | int subdevs, struct mddev_dev *devlist, | |
67 | struct shape *s, | |
68 | struct context *c, unsigned long long data_offset) | |
69 | { | |
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 | |
83 | * if runstop==run, or raiddisks disks were used, | |
84 | * RUN_ARRAY | |
85 | */ | |
86 | int mdfd; | |
87 | unsigned long long minsize=0, maxsize=0; | |
88 | char *mindisc = NULL; | |
89 | char *maxdisc = NULL; | |
90 | int dnum; | |
91 | struct mddev_dev *dv; | |
92 | int fail=0, warn=0; | |
93 | struct stat stb; | |
94 | int first_missing = subdevs * 2; | |
95 | int second_missing = subdevs * 2; | |
96 | int missing_disks = 0; | |
97 | int insert_point = subdevs * 2; /* where to insert a missing drive */ | |
98 | int total_slots; | |
99 | int pass; | |
100 | int vers; | |
101 | int rv; | |
102 | int bitmap_fd; | |
103 | int have_container = 0; | |
104 | int container_fd = -1; | |
105 | int need_mdmon = 0; | |
106 | unsigned long long bitmapsize; | |
107 | struct mdinfo info, *infos; | |
108 | int did_default = 0; | |
109 | int do_default_layout = 0; | |
110 | int do_default_chunk = 0; | |
111 | unsigned long safe_mode_delay = 0; | |
112 | char chosen_name[1024]; | |
113 | struct map_ent *map = NULL; | |
114 | unsigned long long newsize; | |
115 | ||
116 | int major_num = BITMAP_MAJOR_HI; | |
117 | ||
118 | memset(&info, 0, sizeof(info)); | |
119 | if (s->level == UnSet && st && st->ss->default_geometry) | |
120 | st->ss->default_geometry(st, &s->level, NULL, NULL); | |
121 | if (s->level == UnSet) { | |
122 | pr_err("a RAID level is needed to create an array.\n"); | |
123 | return 1; | |
124 | } | |
125 | if (s->raiddisks < 4 && s->level == 6) { | |
126 | pr_err("at least 4 raid-devices needed for level 6\n"); | |
127 | return 1; | |
128 | } | |
129 | if (s->raiddisks > 256 && s->level == 6) { | |
130 | pr_err("no more than 256 raid-devices supported for level 6\n"); | |
131 | return 1; | |
132 | } | |
133 | if (s->raiddisks < 2 && s->level >= 4) { | |
134 | pr_err("at least 2 raid-devices needed for level 4 or 5\n"); | |
135 | return 1; | |
136 | } | |
137 | if (s->level <= 0 && s->sparedisks) { | |
138 | pr_err("This level does not support spare devices\n"); | |
139 | return 1; | |
140 | } | |
141 | ||
142 | if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) { | |
143 | /* If given a single device, it might be a container, and we can | |
144 | * extract a device list from there | |
145 | */ | |
146 | mdu_array_info_t inf; | |
147 | int fd; | |
148 | ||
149 | memset(&inf, 0, sizeof(inf)); | |
150 | fd = open(devlist->devname, O_RDONLY); | |
151 | if (fd >= 0 && | |
152 | ioctl(fd, GET_ARRAY_INFO, &inf) == 0 && | |
153 | inf.raid_disks == 0) { | |
154 | /* yep, looks like a container */ | |
155 | if (st) { | |
156 | rv = st->ss->load_container(st, fd, | |
157 | devlist->devname); | |
158 | if (rv == 0) | |
159 | have_container = 1; | |
160 | } else { | |
161 | st = super_by_fd(fd, NULL); | |
162 | if (st && !(rv = st->ss-> | |
163 | load_container(st, fd, | |
164 | devlist->devname))) | |
165 | have_container = 1; | |
166 | else | |
167 | st = NULL; | |
168 | } | |
169 | if (have_container) { | |
170 | subdevs = s->raiddisks; | |
171 | first_missing = subdevs * 2; | |
172 | second_missing = subdevs * 2; | |
173 | insert_point = subdevs * 2; | |
174 | } | |
175 | } | |
176 | if (fd >= 0) | |
177 | close(fd); | |
178 | } | |
179 | if (st && st->ss->external && s->sparedisks) { | |
180 | pr_err("This metadata type does not support spare disks at create time\n"); | |
181 | return 1; | |
182 | } | |
183 | if (subdevs > s->raiddisks+s->sparedisks) { | |
184 | pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks); | |
185 | return 1; | |
186 | } | |
187 | if (!have_container && subdevs < s->raiddisks+s->sparedisks) { | |
188 | pr_err("You haven't given enough devices (real or missing) to create this array\n"); | |
189 | return 1; | |
190 | } | |
191 | if (s->bitmap_file && s->level <= 0) { | |
192 | pr_err("bitmaps not meaningful with level %s\n", | |
193 | map_num(pers, s->level)?:"given"); | |
194 | return 1; | |
195 | } | |
196 | ||
197 | /* now set some defaults */ | |
198 | ||
199 | if (s->layout == UnSet) { | |
200 | do_default_layout = 1; | |
201 | s->layout = default_layout(st, s->level, c->verbose); | |
202 | } | |
203 | ||
204 | if (s->level == 10) | |
205 | /* check layout fits in array*/ | |
206 | if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) { | |
207 | pr_err("that layout requires at least %d devices\n", | |
208 | (s->layout&255) * ((s->layout>>8)&255)); | |
209 | return 1; | |
210 | } | |
211 | ||
212 | switch(s->level) { | |
213 | case 4: | |
214 | case 5: | |
215 | case 10: | |
216 | case 6: | |
217 | case 0: | |
218 | if (s->chunk == 0 || s->chunk == UnSet) { | |
219 | s->chunk = UnSet; | |
220 | do_default_chunk = 1; | |
221 | /* chunk will be set later */ | |
222 | } | |
223 | break; | |
224 | case LEVEL_LINEAR: | |
225 | /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */ | |
226 | if (get_linux_version() < 2006016 && s->chunk == 0) { | |
227 | s->chunk = 64; | |
228 | if (c->verbose > 0) | |
229 | pr_err("chunk size defaults to 64K\n"); | |
230 | } | |
231 | break; | |
232 | case 1: | |
233 | case LEVEL_FAULTY: | |
234 | case LEVEL_MULTIPATH: | |
235 | case LEVEL_CONTAINER: | |
236 | if (s->chunk) { | |
237 | s->chunk = 0; | |
238 | if (c->verbose > 0) | |
239 | pr_err("chunk size ignored for this level\n"); | |
240 | } | |
241 | break; | |
242 | default: | |
243 | pr_err("unknown level %d\n", s->level); | |
244 | return 1; | |
245 | } | |
246 | if (s->size == MAX_SIZE) | |
247 | /* use '0' to mean 'max' now... */ | |
248 | s->size = 0; | |
249 | if (s->size && s->chunk && s->chunk != UnSet) | |
250 | s->size &= ~(unsigned long long)(s->chunk - 1); | |
251 | newsize = s->size * 2; | |
252 | if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks, | |
253 | &s->chunk, s->size*2, | |
254 | data_offset, NULL, | |
255 | &newsize, c->verbose>=0)) | |
256 | return 1; | |
257 | ||
258 | if (s->chunk && s->chunk != UnSet) { | |
259 | newsize &= ~(unsigned long long)(s->chunk*2 - 1); | |
260 | if (do_default_chunk) { | |
261 | /* default chunk was just set */ | |
262 | if (c->verbose > 0) | |
263 | pr_err("chunk size defaults to %dK\n", s->chunk); | |
264 | s->size &= ~(unsigned long long)(s->chunk - 1); | |
265 | do_default_chunk = 0; | |
266 | } | |
267 | } | |
268 | ||
269 | if (s->size == 0) { | |
270 | s->size = newsize / 2; | |
271 | if (s->level == 1) | |
272 | /* If this is ever reshaped to RAID5, we will | |
273 | * need a chunksize. So round it off a bit | |
274 | * now just to be safe | |
275 | */ | |
276 | s->size &= ~(64ULL-1); | |
277 | ||
278 | if (s->size && c->verbose > 0) | |
279 | pr_err("setting size to %lluK\n", s->size); | |
280 | } | |
281 | ||
282 | /* now look at the subdevs */ | |
283 | info.array.active_disks = 0; | |
284 | info.array.working_disks = 0; | |
285 | dnum = 0; | |
286 | for (dv = devlist; dv ; dv = dv->next) | |
287 | if (data_offset == VARIABLE_OFFSET) | |
288 | dv->data_offset = INVALID_SECTORS; | |
289 | else | |
290 | dv->data_offset = data_offset; | |
291 | ||
292 | for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) { | |
293 | char *dname = dv->devname; | |
294 | unsigned long long freesize; | |
295 | int dfd; | |
296 | char *doff; | |
297 | ||
298 | if (strcasecmp(dname, "missing")==0) { | |
299 | if (first_missing > dnum) | |
300 | first_missing = dnum; | |
301 | if (second_missing > dnum && dnum > first_missing) | |
302 | second_missing = dnum; | |
303 | missing_disks ++; | |
304 | continue; | |
305 | } | |
306 | if (data_offset == VARIABLE_OFFSET) { | |
307 | doff = strchr(dname, ':'); | |
308 | if (doff) { | |
309 | *doff++ = 0; | |
310 | dv->data_offset = parse_size(doff); | |
311 | } else | |
312 | dv->data_offset = INVALID_SECTORS; | |
313 | } else | |
314 | dv->data_offset = data_offset; | |
315 | ||
316 | dfd = open(dname, O_RDONLY); | |
317 | if (dfd < 0) { | |
318 | pr_err("cannot open %s: %s\n", | |
319 | dname, strerror(errno)); | |
320 | exit(2); | |
321 | } | |
322 | if (fstat(dfd, &stb) != 0 || | |
323 | (stb.st_mode & S_IFMT) != S_IFBLK) { | |
324 | close(dfd); | |
325 | pr_err("%s is not a block device\n", | |
326 | dname); | |
327 | exit(2); | |
328 | } | |
329 | close(dfd); | |
330 | info.array.working_disks++; | |
331 | if (dnum < s->raiddisks) | |
332 | info.array.active_disks++; | |
333 | if (st == NULL) { | |
334 | struct createinfo *ci = conf_get_create_info(); | |
335 | if (ci) | |
336 | st = ci->supertype; | |
337 | } | |
338 | if (st == NULL) { | |
339 | /* Need to choose a default metadata, which is different | |
340 | * depending on geometry of array. | |
341 | */ | |
342 | int i; | |
343 | char *name = "default"; | |
344 | for(i=0; !st && superlist[i]; i++) { | |
345 | st = superlist[i]->match_metadata_desc(name); | |
346 | if (!st) | |
347 | continue; | |
348 | if (do_default_layout) | |
349 | s->layout = default_layout(st, s->level, c->verbose); | |
350 | switch (st->ss->validate_geometry( | |
351 | st, s->level, s->layout, s->raiddisks, | |
352 | &s->chunk, s->size*2, | |
353 | dv->data_offset, dname, | |
354 | &freesize, c->verbose > 0)) { | |
355 | case -1: /* Not valid, message printed, and not | |
356 | * worth checking any further */ | |
357 | exit(2); | |
358 | break; | |
359 | case 0: /* Geometry not valid */ | |
360 | free(st); | |
361 | st = NULL; | |
362 | s->chunk = do_default_chunk ? UnSet : s->chunk; | |
363 | break; | |
364 | case 1: /* All happy */ | |
365 | break; | |
366 | } | |
367 | } | |
368 | ||
369 | if (!st) { | |
370 | int dfd = open(dname, O_RDONLY|O_EXCL); | |
371 | if (dfd < 0) { | |
372 | pr_err("cannot open %s: %s\n", | |
373 | dname, strerror(errno)); | |
374 | exit(2); | |
375 | } | |
376 | pr_err("device %s not suitable for any style of array\n", | |
377 | dname); | |
378 | exit(2); | |
379 | } | |
380 | if (st->ss != &super0 || | |
381 | st->minor_version != 90) | |
382 | did_default = 1; | |
383 | } else { | |
384 | if (do_default_layout) | |
385 | s->layout = default_layout(st, s->level, 0); | |
386 | if (!st->ss->validate_geometry(st, s->level, s->layout, | |
387 | s->raiddisks, | |
388 | &s->chunk, s->size*2, | |
389 | dv->data_offset, | |
390 | dname, &freesize, | |
391 | c->verbose >= 0)) { | |
392 | ||
393 | pr_err("%s is not suitable for this array.\n", | |
394 | dname); | |
395 | fail = 1; | |
396 | continue; | |
397 | } | |
398 | } | |
399 | ||
400 | freesize /= 2; /* convert to K */ | |
401 | if (s->chunk && s->chunk != UnSet) { | |
402 | /* round to chunk size */ | |
403 | freesize = freesize & ~(s->chunk-1); | |
404 | if (do_default_chunk) { | |
405 | /* default chunk was just set */ | |
406 | if (c->verbose > 0) | |
407 | pr_err("chunk size defaults to %dK\n", s->chunk); | |
408 | s->size &= ~(unsigned long long)(s->chunk - 1); | |
409 | do_default_chunk = 0; | |
410 | } | |
411 | } | |
412 | if (!freesize) { | |
413 | pr_err("no free space left on %s\n", dname); | |
414 | fail = 1; | |
415 | continue; | |
416 | } | |
417 | ||
418 | if (s->size && freesize < s->size) { | |
419 | pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n", | |
420 | dname, freesize, s->size); | |
421 | fail = 1; | |
422 | continue; | |
423 | } | |
424 | if (maxdisc == NULL || (maxdisc && freesize > maxsize)) { | |
425 | maxdisc = dname; | |
426 | maxsize = freesize; | |
427 | } | |
428 | if (mindisc ==NULL || (mindisc && freesize < minsize)) { | |
429 | mindisc = dname; | |
430 | minsize = freesize; | |
431 | } | |
432 | if (c->runstop != 1 || c->verbose >= 0) { | |
433 | int fd = open(dname, O_RDONLY); | |
434 | if (fd <0 ) { | |
435 | pr_err("Cannot open %s: %s\n", | |
436 | dname, strerror(errno)); | |
437 | fail=1; | |
438 | continue; | |
439 | } | |
440 | warn |= check_ext2(fd, dname); | |
441 | warn |= check_reiser(fd, dname); | |
442 | warn |= check_raid(fd, dname); | |
443 | if (strcmp(st->ss->name, "1.x") == 0 && | |
444 | st->minor_version >= 1) | |
445 | /* metadata at front */ | |
446 | warn |= check_partitions(fd, dname, 0, 0); | |
447 | else if (s->level == 1 || s->level == LEVEL_CONTAINER | |
448 | || (s->level == 0 && s->raiddisks == 1)) | |
449 | /* partitions could be meaningful */ | |
450 | warn |= check_partitions(fd, dname, freesize*2, s->size*2); | |
451 | else | |
452 | /* partitions cannot be meaningful */ | |
453 | warn |= check_partitions(fd, dname, 0, 0); | |
454 | if (strcmp(st->ss->name, "1.x") == 0 && | |
455 | st->minor_version >= 1 && | |
456 | did_default && | |
457 | s->level == 1 && | |
458 | (warn & 1024) == 0) { | |
459 | warn |= 1024; | |
460 | pr_err("Note: this array has metadata at the start and\n" | |
461 | " may not be suitable as a boot device. If you plan to\n" | |
462 | " store '/boot' on this device please ensure that\n" | |
463 | " your boot-loader understands md/v1.x metadata, or use\n" | |
464 | " --metadata=0.90\n"); | |
465 | } | |
466 | close(fd); | |
467 | } | |
468 | } | |
469 | if (s->raiddisks + s->sparedisks > st->max_devs) { | |
470 | pr_err("Too many devices: %s metadata only supports %d\n", | |
471 | st->ss->name, st->max_devs); | |
472 | return 1; | |
473 | } | |
474 | if (have_container) | |
475 | info.array.working_disks = s->raiddisks; | |
476 | if (fail) { | |
477 | pr_err("create aborted\n"); | |
478 | return 1; | |
479 | } | |
480 | if (s->size == 0) { | |
481 | if (mindisc == NULL && !have_container) { | |
482 | pr_err("no size and no drives given - aborting create.\n"); | |
483 | return 1; | |
484 | } | |
485 | if (s->level > 0 || s->level == LEVEL_MULTIPATH | |
486 | || s->level == LEVEL_FAULTY | |
487 | || st->ss->external ) { | |
488 | /* size is meaningful */ | |
489 | if (!st->ss->validate_geometry(st, s->level, s->layout, | |
490 | s->raiddisks, | |
491 | &s->chunk, minsize*2, | |
492 | data_offset, | |
493 | NULL, NULL, 0)) { | |
494 | pr_err("devices too large for RAID level %d\n", s->level); | |
495 | return 1; | |
496 | } | |
497 | s->size = minsize; | |
498 | if (s->level == 1) | |
499 | /* If this is ever reshaped to RAID5, we will | |
500 | * need a chunksize. So round it off a bit | |
501 | * now just to be safe | |
502 | */ | |
503 | s->size &= ~(64ULL-1); | |
504 | if (c->verbose > 0) | |
505 | pr_err("size set to %lluK\n", s->size); | |
506 | } | |
507 | } | |
508 | ||
509 | if (!s->bitmap_file && | |
510 | s->level >= 1 && | |
511 | st->ss->add_internal_bitmap && | |
512 | (s->write_behind || s->size > 100*1024*1024ULL)) { | |
513 | if (c->verbose > 0) | |
514 | pr_err("automatically enabling write-intent bitmap on large array\n"); | |
515 | s->bitmap_file = "internal"; | |
516 | } | |
517 | if (s->bitmap_file && strcmp(s->bitmap_file, "none") == 0) | |
518 | s->bitmap_file = NULL; | |
519 | ||
520 | if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) { | |
521 | if (c->runstop != 1 || c->verbose >= 0) | |
522 | pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n", | |
523 | maxdisc, s->size); | |
524 | warn = 1; | |
525 | } | |
526 | ||
527 | if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) { | |
528 | if (c->runstop != 1 || c->verbose >= 0) | |
529 | pr_err("%s unable to enumerate platform support\n" | |
530 | " array may not be compatible with hardware/firmware\n", | |
531 | st->ss->name); | |
532 | warn = 1; | |
533 | } | |
534 | st->nodes = c->nodes; | |
535 | st->cluster_name = c->homecluster; | |
536 | ||
537 | if (warn) { | |
538 | if (c->runstop!= 1) { | |
539 | if (!ask("Continue creating array? ")) { | |
540 | pr_err("create aborted.\n"); | |
541 | return 1; | |
542 | } | |
543 | } else { | |
544 | if (c->verbose > 0) | |
545 | pr_err("creation continuing despite oddities due to --run\n"); | |
546 | } | |
547 | } | |
548 | ||
549 | /* If this is raid4/5, we want to configure the last active slot | |
550 | * as missing, so that a reconstruct happens (faster than re-parity) | |
551 | * FIX: Can we do this for raid6 as well? | |
552 | */ | |
553 | if (st->ss->external == 0 && | |
554 | s->assume_clean==0 && c->force == 0 && first_missing >= s->raiddisks) { | |
555 | switch ( s->level ) { | |
556 | case 4: | |
557 | case 5: | |
558 | insert_point = s->raiddisks-1; | |
559 | s->sparedisks++; | |
560 | info.array.active_disks--; | |
561 | missing_disks++; | |
562 | break; | |
563 | default: | |
564 | break; | |
565 | } | |
566 | } | |
567 | /* For raid6, if creating with 1 missing drive, make a good drive | |
568 | * into a spare, else the create will fail | |
569 | */ | |
570 | if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks && | |
571 | st->ss->external == 0 && | |
572 | second_missing >= s->raiddisks && s->level == 6) { | |
573 | insert_point = s->raiddisks - 1; | |
574 | if (insert_point == first_missing) | |
575 | insert_point--; | |
576 | s->sparedisks ++; | |
577 | info.array.active_disks--; | |
578 | missing_disks++; | |
579 | } | |
580 | ||
581 | if (s->level <= 0 && first_missing < subdevs * 2) { | |
582 | pr_err("This level does not support missing devices\n"); | |
583 | return 1; | |
584 | } | |
585 | ||
586 | /* We need to create the device */ | |
587 | map_lock(&map); | |
588 | mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name); | |
589 | if (mdfd < 0) { | |
590 | map_unlock(&map); | |
591 | return 1; | |
592 | } | |
593 | /* verify if chosen_name is not in use, | |
594 | * it could be in conflict with already existing device | |
595 | * e.g. container, array | |
596 | */ | |
597 | if (strncmp(chosen_name, "/dev/md/", 8) == 0 | |
598 | && map_by_name(&map, chosen_name+8) != NULL) { | |
599 | pr_err("Array name %s is in use already.\n", | |
600 | chosen_name); | |
601 | close(mdfd); | |
602 | map_unlock(&map); | |
603 | return 1; | |
604 | } | |
605 | mddev = chosen_name; | |
606 | ||
607 | vers = md_get_version(mdfd); | |
608 | if (vers < 9000) { | |
609 | pr_err("Create requires md driver version 0.90.0 or later\n"); | |
610 | goto abort_locked; | |
611 | } else { | |
612 | mdu_array_info_t inf; | |
613 | memset(&inf, 0, sizeof(inf)); | |
614 | ioctl(mdfd, GET_ARRAY_INFO, &inf); | |
615 | if (inf.working_disks != 0) { | |
616 | pr_err("another array by this name is already running.\n"); | |
617 | goto abort_locked; | |
618 | } | |
619 | } | |
620 | ||
621 | /* Ok, lets try some ioctls */ | |
622 | ||
623 | info.array.level = s->level; | |
624 | info.array.size = s->size; | |
625 | info.array.raid_disks = s->raiddisks; | |
626 | /* The kernel should *know* what md_minor we are dealing | |
627 | * with, but it chooses to trust me instead. Sigh | |
628 | */ | |
629 | info.array.md_minor = 0; | |
630 | if (fstat(mdfd, &stb)==0) | |
631 | info.array.md_minor = minor(stb.st_rdev); | |
632 | info.array.not_persistent = 0; | |
633 | ||
634 | if ( ( (s->level == 4 || s->level == 5) && | |
635 | (insert_point < s->raiddisks || first_missing < s->raiddisks) ) | |
636 | || | |
637 | ( s->level == 6 && (insert_point < s->raiddisks | |
638 | || second_missing < s->raiddisks)) | |
639 | || | |
640 | ( s->level <= 0 ) | |
641 | || | |
642 | s->assume_clean | |
643 | ) { | |
644 | info.array.state = 1; /* clean, but one+ drive will be missing*/ | |
645 | info.resync_start = MaxSector; | |
646 | } else { | |
647 | info.array.state = 0; /* not clean, but no errors */ | |
648 | info.resync_start = 0; | |
649 | } | |
650 | if (s->level == 10) { | |
651 | /* for raid10, the bitmap size is the capacity of the array, | |
652 | * which is array.size * raid_disks / ncopies; | |
653 | * .. but convert to sectors. | |
654 | */ | |
655 | int ncopies = ((s->layout>>8) & 255) * (s->layout & 255); | |
656 | bitmapsize = s->size * s->raiddisks / ncopies * 2; | |
657 | /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/ | |
658 | } else | |
659 | bitmapsize = s->size * 2; | |
660 | ||
661 | /* There is lots of redundancy in these disk counts, | |
662 | * raid_disks is the most meaningful value | |
663 | * it describes the geometry of the array | |
664 | * it is constant | |
665 | * nr_disks is total number of used slots. | |
666 | * it should be raid_disks+spare_disks | |
667 | * spare_disks is the number of extra disks present | |
668 | * see above | |
669 | * active_disks is the number of working disks in | |
670 | * active slots. (With raid_disks) | |
671 | * working_disks is the total number of working disks, | |
672 | * including spares | |
673 | * failed_disks is the number of disks marked failed | |
674 | * | |
675 | * Ideally, the kernel would keep these (except raid_disks) | |
676 | * up-to-date as we ADD_NEW_DISK, but it doesn't (yet). | |
677 | * So for now, we assume that all raid and spare | |
678 | * devices will be given. | |
679 | */ | |
680 | info.array.spare_disks=s->sparedisks; | |
681 | info.array.failed_disks=missing_disks; | |
682 | info.array.nr_disks = info.array.working_disks | |
683 | + info.array.failed_disks; | |
684 | info.array.layout = s->layout; | |
685 | info.array.chunk_size = s->chunk*1024; | |
686 | ||
687 | if (name == NULL || *name == 0) { | |
688 | /* base name on mddev */ | |
689 | /* /dev/md0 -> 0 | |
690 | * /dev/md_d0 -> d0 | |
691 | * /dev/md_foo -> foo | |
692 | * /dev/md/1 -> 1 | |
693 | * /dev/md/d1 -> d1 | |
694 | * /dev/md/home -> home | |
695 | * /dev/mdhome -> home | |
696 | */ | |
697 | /* FIXME compare this with rules in create_mddev */ | |
698 | name = strrchr(mddev, '/'); | |
699 | if (name) { | |
700 | name++; | |
701 | if (strncmp(name, "md_", 3)==0 && | |
702 | strlen(name) > 3 && | |
703 | (name-mddev) == 5 /* /dev/ */) | |
704 | name += 3; | |
705 | else if (strncmp(name, "md", 2)==0 && | |
706 | strlen(name) > 2 && | |
707 | isdigit(name[2]) && | |
708 | (name-mddev) == 5 /* /dev/ */) | |
709 | name += 2; | |
710 | } | |
711 | } | |
712 | if (!st->ss->init_super(st, &info.array, s->size, name, c->homehost, uuid, | |
713 | data_offset)) | |
714 | goto abort_locked; | |
715 | ||
716 | total_slots = info.array.nr_disks; | |
717 | st->ss->getinfo_super(st, &info, NULL); | |
718 | sysfs_init(&info, mdfd, NULL); | |
719 | ||
720 | if (did_default && c->verbose >= 0) { | |
721 | if (is_subarray(info.text_version)) { | |
722 | char devnm[32]; | |
723 | char *ep; | |
724 | struct mdinfo *mdi; | |
725 | ||
726 | strncpy(devnm, info.text_version+1, 32); | |
727 | devnm[31] = 0; | |
728 | ep = strchr(devnm, '/'); | |
729 | if (ep) | |
730 | *ep = 0; | |
731 | ||
732 | mdi = sysfs_read(-1, devnm, GET_VERSION); | |
733 | ||
734 | pr_err("Creating array inside %s container %s\n", | |
735 | mdi?mdi->text_version:"managed", devnm); | |
736 | sysfs_free(mdi); | |
737 | } else | |
738 | pr_err("Defaulting to version %s metadata\n", info.text_version); | |
739 | } | |
740 | ||
741 | map_update(&map, fd2devnm(mdfd), info.text_version, | |
742 | info.uuid, chosen_name); | |
743 | /* Keep map locked until devices have been added to array | |
744 | * to stop another mdadm from finding and using those devices. | |
745 | */ | |
746 | ||
747 | if (s->bitmap_file && vers < 9003) { | |
748 | major_num = BITMAP_MAJOR_HOSTENDIAN; | |
749 | #ifdef __BIG_ENDIAN | |
750 | pr_err("Warning - bitmaps created on this kernel are not portable\n" | |
751 | " between different architectured. Consider upgrading the Linux kernel.\n"); | |
752 | #endif | |
753 | } | |
754 | ||
755 | if (s->bitmap_file && (strcmp(s->bitmap_file, "internal")==0 || | |
756 | strcmp(s->bitmap_file, "clustered")==0)) { | |
757 | if ((vers%100) < 2) { | |
758 | pr_err("internal bitmaps not supported by this kernel.\n"); | |
759 | goto abort_locked; | |
760 | } | |
761 | if (!st->ss->add_internal_bitmap) { | |
762 | pr_err("internal bitmaps not supported with %s metadata\n", | |
763 | st->ss->name); | |
764 | goto abort_locked; | |
765 | } | |
766 | if (!st->ss->add_internal_bitmap(st, &s->bitmap_chunk, | |
767 | c->delay, s->write_behind, | |
768 | bitmapsize, 1, major_num)) { | |
769 | pr_err("Given bitmap chunk size not supported.\n"); | |
770 | goto abort_locked; | |
771 | } | |
772 | s->bitmap_file = NULL; | |
773 | } | |
774 | ||
775 | sysfs_init(&info, mdfd, NULL); | |
776 | ||
777 | if (st->ss->external && st->container_devnm[0]) { | |
778 | /* member */ | |
779 | ||
780 | /* When creating a member, we need to be careful | |
781 | * to negotiate with mdmon properly. | |
782 | * If it is already running, we cannot write to | |
783 | * the devices and must ask it to do that part. | |
784 | * If it isn't running, we write to the devices, | |
785 | * and then start it. | |
786 | * We hold an exclusive open on the container | |
787 | * device to make sure mdmon doesn't exit after | |
788 | * we checked that it is running. | |
789 | * | |
790 | * For now, fail if it is already running. | |
791 | */ | |
792 | container_fd = open_dev_excl(st->container_devnm); | |
793 | if (container_fd < 0) { | |
794 | pr_err("Cannot get exclusive open on container - weird.\n"); | |
795 | goto abort_locked; | |
796 | } | |
797 | if (mdmon_running(st->container_devnm)) { | |
798 | if (c->verbose) | |
799 | pr_err("reusing mdmon for %s.\n", | |
800 | st->container_devnm); | |
801 | st->update_tail = &st->updates; | |
802 | } else | |
803 | need_mdmon = 1; | |
804 | } | |
805 | rv = set_array_info(mdfd, st, &info); | |
806 | if (rv) { | |
807 | pr_err("failed to set array info for %s: %s\n", | |
808 | mddev, strerror(errno)); | |
809 | goto abort_locked; | |
810 | } | |
811 | ||
812 | if (s->bitmap_file) { | |
813 | int uuid[4]; | |
814 | ||
815 | st->ss->uuid_from_super(st, uuid); | |
816 | if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk, | |
817 | c->delay, s->write_behind, | |
818 | bitmapsize, | |
819 | major_num)) { | |
820 | goto abort_locked; | |
821 | } | |
822 | bitmap_fd = open(s->bitmap_file, O_RDWR); | |
823 | if (bitmap_fd < 0) { | |
824 | pr_err("weird: %s cannot be openned\n", | |
825 | s->bitmap_file); | |
826 | goto abort_locked; | |
827 | } | |
828 | if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
829 | pr_err("Cannot set bitmap file for %s: %s\n", | |
830 | mddev, strerror(errno)); | |
831 | goto abort_locked; | |
832 | } | |
833 | } | |
834 | ||
835 | infos = xmalloc(sizeof(*infos) * total_slots); | |
836 | enable_fds(total_slots); | |
837 | for (pass=1; pass <=2 ; pass++) { | |
838 | struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */ | |
839 | ||
840 | for (dnum=0, dv = devlist ; dv ; | |
841 | dv=(dv->next)?(dv->next):moved_disk, dnum++) { | |
842 | int fd; | |
843 | struct stat stb; | |
844 | struct mdinfo *inf = &infos[dnum]; | |
845 | ||
846 | if (dnum >= total_slots) | |
847 | abort(); | |
848 | if (dnum == insert_point) { | |
849 | moved_disk = dv; | |
850 | continue; | |
851 | } | |
852 | if (strcasecmp(dv->devname, "missing")==0) | |
853 | continue; | |
854 | if (have_container) | |
855 | moved_disk = NULL; | |
856 | if (have_container && dnum < info.array.raid_disks - 1) | |
857 | /* repeatedly use the container */ | |
858 | moved_disk = dv; | |
859 | ||
860 | switch(pass) { | |
861 | case 1: | |
862 | *inf = info; | |
863 | ||
864 | inf->disk.number = dnum; | |
865 | inf->disk.raid_disk = dnum; | |
866 | if (inf->disk.raid_disk < s->raiddisks) | |
867 | inf->disk.state = (1<<MD_DISK_ACTIVE) | | |
868 | (1<<MD_DISK_SYNC); | |
869 | else | |
870 | inf->disk.state = 0; | |
871 | ||
872 | if (dv->writemostly == 1) | |
873 | inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY); | |
874 | ||
875 | if (have_container) | |
876 | fd = -1; | |
877 | else { | |
878 | if (st->ss->external && | |
879 | st->container_devnm[0]) | |
880 | fd = open(dv->devname, O_RDWR); | |
881 | else | |
882 | fd = open(dv->devname, O_RDWR|O_EXCL); | |
883 | ||
884 | if (fd < 0) { | |
885 | pr_err("failed to open %s after earlier success - aborting\n", | |
886 | dv->devname); | |
887 | goto abort_locked; | |
888 | } | |
889 | fstat(fd, &stb); | |
890 | inf->disk.major = major(stb.st_rdev); | |
891 | inf->disk.minor = minor(stb.st_rdev); | |
892 | } | |
893 | if (fd >= 0) | |
894 | remove_partitions(fd); | |
895 | if (st->ss->add_to_super(st, &inf->disk, | |
896 | fd, dv->devname, | |
897 | dv->data_offset)) { | |
898 | ioctl(mdfd, STOP_ARRAY, NULL); | |
899 | goto abort_locked; | |
900 | } | |
901 | st->ss->getinfo_super(st, inf, NULL); | |
902 | safe_mode_delay = inf->safe_mode_delay; | |
903 | ||
904 | if (have_container && c->verbose > 0) | |
905 | pr_err("Using %s for device %d\n", | |
906 | map_dev(inf->disk.major, | |
907 | inf->disk.minor, | |
908 | 0), dnum); | |
909 | ||
910 | if (!have_container) { | |
911 | /* getinfo_super might have lost these ... */ | |
912 | inf->disk.major = major(stb.st_rdev); | |
913 | inf->disk.minor = minor(stb.st_rdev); | |
914 | } | |
915 | break; | |
916 | case 2: | |
917 | inf->errors = 0; | |
918 | ||
919 | rv = add_disk(mdfd, st, &info, inf); | |
920 | ||
921 | if (rv) { | |
922 | pr_err("ADD_NEW_DISK for %s failed: %s\n", | |
923 | dv->devname, strerror(errno)); | |
924 | goto abort_locked; | |
925 | } | |
926 | break; | |
927 | } | |
928 | if (!have_container && | |
929 | dv == moved_disk && dnum != insert_point) break; | |
930 | } | |
931 | if (pass == 1) { | |
932 | struct mdinfo info_new; | |
933 | struct map_ent *me = NULL; | |
934 | ||
935 | /* check to see if the uuid has changed due to these | |
936 | * metadata changes, and if so update the member array | |
937 | * and container uuid. Note ->write_init_super clears | |
938 | * the subarray cursor such that ->getinfo_super once | |
939 | * again returns container info. | |
940 | */ | |
941 | st->ss->getinfo_super(st, &info_new, NULL); | |
942 | if (st->ss->external && s->level != LEVEL_CONTAINER && | |
943 | !same_uuid(info_new.uuid, info.uuid, 0)) { | |
944 | map_update(&map, fd2devnm(mdfd), | |
945 | info_new.text_version, | |
946 | info_new.uuid, chosen_name); | |
947 | me = map_by_devnm(&map, st->container_devnm); | |
948 | } | |
949 | ||
950 | if (st->ss->write_init_super(st)) { | |
951 | st->ss->free_super(st); | |
952 | goto abort_locked; | |
953 | } | |
954 | ||
955 | /* update parent container uuid */ | |
956 | if (me) { | |
957 | char *path = xstrdup(me->path); | |
958 | ||
959 | st->ss->getinfo_super(st, &info_new, NULL); | |
960 | map_update(&map, st->container_devnm, | |
961 | info_new.text_version, | |
962 | info_new.uuid, path); | |
963 | free(path); | |
964 | } | |
965 | ||
966 | flush_metadata_updates(st); | |
967 | st->ss->free_super(st); | |
968 | } | |
969 | } | |
970 | map_unlock(&map); | |
971 | free(infos); | |
972 | ||
973 | if (s->level == LEVEL_CONTAINER) { | |
974 | /* No need to start. But we should signal udev to | |
975 | * create links */ | |
976 | sysfs_uevent(&info, "change"); | |
977 | if (c->verbose >= 0) | |
978 | pr_err("container %s prepared.\n", mddev); | |
979 | wait_for(chosen_name, mdfd); | |
980 | } else if (c->runstop == 1 || subdevs >= s->raiddisks) { | |
981 | if (st->ss->external) { | |
982 | int err; | |
983 | switch(s->level) { | |
984 | case LEVEL_LINEAR: | |
985 | case LEVEL_MULTIPATH: | |
986 | case 0: | |
987 | err = sysfs_set_str(&info, NULL, "array_state", | |
988 | c->readonly | |
989 | ? "readonly" | |
990 | : "active"); | |
991 | need_mdmon = 0; | |
992 | break; | |
993 | default: | |
994 | err = sysfs_set_str(&info, NULL, "array_state", | |
995 | "readonly"); | |
996 | break; | |
997 | } | |
998 | sysfs_set_safemode(&info, safe_mode_delay); | |
999 | if (err) { | |
1000 | pr_err("failed to activate array.\n"); | |
1001 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1002 | goto abort; | |
1003 | } | |
1004 | } else if (c->readonly && | |
1005 | sysfs_attribute_available( | |
1006 | &info, NULL, "array_state")) { | |
1007 | if (sysfs_set_str(&info, NULL, | |
1008 | "array_state", "readonly") < 0) { | |
1009 | pr_err("Failed to start array: %s\n", | |
1010 | strerror(errno)); | |
1011 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1012 | goto abort; | |
1013 | } | |
1014 | } else { | |
1015 | /* param is not actually used */ | |
1016 | mdu_param_t param; | |
1017 | if (ioctl(mdfd, RUN_ARRAY, ¶m)) { | |
1018 | pr_err("RUN_ARRAY failed: %s\n", | |
1019 | strerror(errno)); | |
1020 | if (info.array.chunk_size & (info.array.chunk_size-1)) { | |
1021 | cont_err("Problem may be that chunk size is not a power of 2\n"); | |
1022 | } | |
1023 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1024 | goto abort; | |
1025 | } | |
1026 | /* if start_ro module parameter is set, array is | |
1027 | * auto-read-only, which is bad as the resync won't | |
1028 | * start. So lets make it read-write now. | |
1029 | */ | |
1030 | ioctl(mdfd, RESTART_ARRAY_RW, NULL); | |
1031 | } | |
1032 | if (c->verbose >= 0) | |
1033 | pr_err("array %s started.\n", mddev); | |
1034 | if (st->ss->external && st->container_devnm[0]) { | |
1035 | if (need_mdmon) | |
1036 | start_mdmon(st->container_devnm); | |
1037 | ||
1038 | ping_monitor(st->container_devnm); | |
1039 | close(container_fd); | |
1040 | } | |
1041 | wait_for(chosen_name, mdfd); | |
1042 | } else { | |
1043 | pr_err("not starting array - not enough devices.\n"); | |
1044 | } | |
1045 | close(mdfd); | |
1046 | return 0; | |
1047 | ||
1048 | abort: | |
1049 | map_lock(&map); | |
1050 | abort_locked: | |
1051 | map_remove(&map, fd2devnm(mdfd)); | |
1052 | map_unlock(&map); | |
1053 | ||
1054 | if (mdfd >= 0) | |
1055 | close(mdfd); | |
1056 | return 1; | |
1057 | } |