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