]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Create.c
projects / thirdparty / mdadm.git / blob
? search:


b200d9767bb616aeb44693ae322ef7dd77e76728
[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;
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 if (s->bitmap_file && strcmp(s->bitmap_file, "clustered") == 0)
118 major_num = BITMAP_MAJOR_CLUSTERED;
119
120 memset(&info, 0, sizeof(info));
121 if (s->level == UnSet && st && st->ss->default_geometry)
122 st->ss->default_geometry(st, &s->level, NULL, NULL);
123 if (s->level == UnSet) {
124 pr_err("a RAID level is needed to create an array.\n");
125 return 1;
126 }
127 if (s->raiddisks < 4 && s->level == 6) {
128 pr_err("at least 4 raid-devices needed for level 6\n");
129 return 1;
130 }
131 if (s->raiddisks > 256 && s->level == 6) {
132 pr_err("no more than 256 raid-devices supported for level 6\n");
133 return 1;
134 }
135 if (s->raiddisks < 2 && s->level >= 4) {
136 pr_err("at least 2 raid-devices needed for level 4 or 5\n");
137 return 1;
138 }
139 if (s->level <= 0 && s->sparedisks) {
140 pr_err("This level does not support spare devices\n");
141 return 1;
142 }
143
144 if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
145 /* If given a single device, it might be a container, and we can
146 * extract a device list from there
147 */
148 mdu_array_info_t inf;
149 int fd;
150
151 memset(&inf, 0, sizeof(inf));
152 fd = open(devlist->devname, O_RDONLY);
153 if (fd >= 0 &&
154 ioctl(fd, GET_ARRAY_INFO, &inf) == 0 &&
155 inf.raid_disks == 0) {
156 /* yep, looks like a container */
157 if (st) {
158 rv = st->ss->load_container(st, fd,
159 devlist->devname);
160 if (rv == 0)
161 have_container = 1;
162 } else {
163 st = super_by_fd(fd, NULL);
164 if (st && !(rv = st->ss->
165 load_container(st, fd,
166 devlist->devname)))
167 have_container = 1;
168 else
169 st = NULL;
170 }
171 if (have_container) {
172 subdevs = s->raiddisks;
173 first_missing = subdevs * 2;
174 second_missing = subdevs * 2;
175 insert_point = subdevs * 2;
176 }
177 }
178 if (fd >= 0)
179 close(fd);
180 }
181 if (st && st->ss->external && s->sparedisks) {
182 pr_err("This metadata type does not support spare disks at create time\n");
183 return 1;
184 }
185 if (subdevs > s->raiddisks+s->sparedisks) {
186 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks);
187 return 1;
188 }
189 if (!have_container && subdevs < s->raiddisks+s->sparedisks) {
190 pr_err("You haven't given enough devices (real or missing) to create this array\n");
191 return 1;
192 }
193 if (s->bitmap_file && s->level <= 0) {
194 pr_err("bitmaps not meaningful with level %s\n",
195 map_num(pers, s->level)?:"given");
196 return 1;
197 }
198
199 /* now set some defaults */
200
201 if (s->layout == UnSet) {
202 do_default_layout = 1;
203 s->layout = default_layout(st, s->level, c->verbose);
204 }
205
206 if (s->level == 10)
207 /* check layout fits in array*/
208 if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) {
209 pr_err("that layout requires at least %d devices\n",
210 (s->layout&255) * ((s->layout>>8)&255));
211 return 1;
212 }
213
214 switch(s->level) {
215 case 4:
216 case 5:
217 case 10:
218 case 6:
219 case 0:
220 if (s->chunk == 0 || s->chunk == UnSet) {
221 s->chunk = UnSet;
222 do_default_chunk = 1;
223 /* chunk will be set later */
224 }
225 break;
226 case LEVEL_LINEAR:
227 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
228 if (get_linux_version() < 2006016 && s->chunk == 0) {
229 s->chunk = 64;
230 if (c->verbose > 0)
231 pr_err("chunk size defaults to 64K\n");
232 }
233 break;
234 case 1:
235 case LEVEL_FAULTY:
236 case LEVEL_MULTIPATH:
237 case LEVEL_CONTAINER:
238 if (s->chunk) {
239 s->chunk = 0;
240 if (c->verbose > 0)
241 pr_err("chunk size ignored for this level\n");
242 }
243 break;
244 default:
245 pr_err("unknown level %d\n", s->level);
246 return 1;
247 }
248 if (s->size == MAX_SIZE)
249 /* use '0' to mean 'max' now... */
250 s->size = 0;
251 if (s->size && s->chunk && s->chunk != UnSet)
252 s->size &= ~(unsigned long long)(s->chunk - 1);
253 newsize = s->size * 2;
254 if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks,
255 &s->chunk, s->size*2,
256 data_offset, NULL,
257 &newsize, c->verbose>=0))
258 return 1;
259
260 if (s->chunk && s->chunk != UnSet) {
261 newsize &= ~(unsigned long long)(s->chunk*2 - 1);
262 if (do_default_chunk) {
263 /* default chunk was just set */
264 if (c->verbose > 0)
265 pr_err("chunk size defaults to %dK\n", s->chunk);
266 s->size &= ~(unsigned long long)(s->chunk - 1);
267 do_default_chunk = 0;
268 }
269 }
270
271 if (s->size == 0) {
272 s->size = newsize / 2;
273 if (s->level == 1)
274 /* If this is ever reshaped to RAID5, we will
275 * need a chunksize. So round it off a bit
276 * now just to be safe
277 */
278 s->size &= ~(64ULL-1);
279
280 if (s->size && c->verbose > 0)
281 pr_err("setting size to %lluK\n", s->size);
282 }
283
284 /* now look at the subdevs */
285 info.array.active_disks = 0;
286 info.array.working_disks = 0;
287 dnum = 0;
288 for (dv = devlist; dv ; dv = dv->next)
289 if (data_offset == VARIABLE_OFFSET)
290 dv->data_offset = INVALID_SECTORS;
291 else
292 dv->data_offset = data_offset;
293
294 for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) {
295 char *dname = dv->devname;
296 unsigned long long freesize;
297 int dfd;
298 char *doff;
299
300 if (strcasecmp(dname, "missing")==0) {
301 if (first_missing > dnum)
302 first_missing = dnum;
303 if (second_missing > dnum && dnum > first_missing)
304 second_missing = dnum;
305 missing_disks ++;
306 continue;
307 }
308 if (data_offset == VARIABLE_OFFSET) {
309 doff = strchr(dname, ':');
310 if (doff) {
311 *doff++ = 0;
312 dv->data_offset = parse_size(doff);
313 } else
314 dv->data_offset = INVALID_SECTORS;
315 } else
316 dv->data_offset = data_offset;
317
318 dfd = open(dname, O_RDONLY);
319 if (dfd < 0) {
320 pr_err("cannot open %s: %s\n",
321 dname, strerror(errno));
322 exit(2);
323 }
324 if (fstat(dfd, &stb) != 0 ||
325 (stb.st_mode & S_IFMT) != S_IFBLK) {
326 close(dfd);
327 pr_err("%s is not a block device\n",
328 dname);
329 exit(2);
330 }
331 close(dfd);
332 info.array.working_disks++;
333 if (dnum < s->raiddisks)
334 info.array.active_disks++;
335 if (st == NULL) {
336 struct createinfo *ci = conf_get_create_info();
337 if (ci)
338 st = ci->supertype;
339 }
340 if (st == NULL) {
341 /* Need to choose a default metadata, which is different
342 * depending on geometry of array.
343 */
344 int i;
345 char *name = "default";
346 for(i=0; !st && superlist[i]; i++) {
347 st = superlist[i]->match_metadata_desc(name);
348 if (!st)
349 continue;
350 if (do_default_layout)
351 s->layout = default_layout(st, s->level, c->verbose);
352 switch (st->ss->validate_geometry(
353 st, s->level, s->layout, s->raiddisks,
354 &s->chunk, s->size*2,
355 dv->data_offset, dname,
356 &freesize, c->verbose > 0)) {
357 case -1: /* Not valid, message printed, and not
358 * worth checking any further */
359 exit(2);
360 break;
361 case 0: /* Geometry not valid */
362 free(st);
363 st = NULL;
364 s->chunk = do_default_chunk ? UnSet : s->chunk;
365 break;
366 case 1: /* All happy */
367 break;
368 }
369 }
370
371 if (!st) {
372 int dfd = open(dname, O_RDONLY|O_EXCL);
373 if (dfd < 0) {
374 pr_err("cannot open %s: %s\n",
375 dname, strerror(errno));
376 exit(2);
377 }
378 pr_err("device %s not suitable for any style of array\n",
379 dname);
380 exit(2);
381 }
382 if (st->ss != &super0 ||
383 st->minor_version != 90)
384 did_default = 1;
385 } else {
386 if (do_default_layout)
387 s->layout = default_layout(st, s->level, 0);
388 if (!st->ss->validate_geometry(st, s->level, s->layout,
389 s->raiddisks,
390 &s->chunk, s->size*2,
391 dv->data_offset,
392 dname, &freesize,
393 c->verbose >= 0)) {
394
395 pr_err("%s is not suitable for this array.\n",
396 dname);
397 fail = 1;
398 continue;
399 }
400 }
401
402 freesize /= 2; /* convert to K */
403 if (s->chunk && s->chunk != UnSet) {
404 /* round to chunk size */
405 freesize = freesize & ~(s->chunk-1);
406 if (do_default_chunk) {
407 /* default chunk was just set */
408 if (c->verbose > 0)
409 pr_err("chunk size defaults to %dK\n", s->chunk);
410 s->size &= ~(unsigned long long)(s->chunk - 1);
411 do_default_chunk = 0;
412 }
413 }
414 if (!freesize) {
415 pr_err("no free space left on %s\n", dname);
416 fail = 1;
417 continue;
418 }
419
420 if (s->size && freesize < s->size) {
421 pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
422 dname, freesize, s->size);
423 fail = 1;
424 continue;
425 }
426 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
427 maxdisc = dname;
428 maxsize = freesize;
429 }
430 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
431 mindisc = dname;
432 minsize = freesize;
433 }
434 if (c->runstop != 1 || c->verbose >= 0) {
435 int fd = open(dname, O_RDONLY);
436 if (fd <0 ) {
437 pr_err("Cannot open %s: %s\n",
438 dname, strerror(errno));
439 fail=1;
440 continue;
441 }
442 warn |= check_ext2(fd, dname);
443 warn |= check_reiser(fd, dname);
444 warn |= check_raid(fd, dname);
445 if (strcmp(st->ss->name, "1.x") == 0 &&
446 st->minor_version >= 1)
447 /* metadata at front */
448 warn |= check_partitions(fd, dname, 0, 0);
449 else if (s->level == 1 || s->level == LEVEL_CONTAINER
450 || (s->level == 0 && s->raiddisks == 1))
451 /* partitions could be meaningful */
452 warn |= check_partitions(fd, dname, freesize*2, s->size*2);
453 else
454 /* partitions cannot be meaningful */
455 warn |= check_partitions(fd, dname, 0, 0);
456 if (strcmp(st->ss->name, "1.x") == 0 &&
457 st->minor_version >= 1 &&
458 did_default &&
459 s->level == 1 &&
460 (warn & 1024) == 0) {
461 warn |= 1024;
462 pr_err("Note: this array has metadata at the start and\n"
463 " may not be suitable as a boot device. If you plan to\n"
464 " store '/boot' on this device please ensure that\n"
465 " your boot-loader understands md/v1.x metadata, or use\n"
466 " --metadata=0.90\n");
467 }
468 close(fd);
469 }
470 }
471 if (s->raiddisks + s->sparedisks > st->max_devs) {
472 pr_err("Too many devices: %s metadata only supports %d\n",
473 st->ss->name, st->max_devs);
474 return 1;
475 }
476 if (have_container)
477 info.array.working_disks = s->raiddisks;
478 if (fail) {
479 pr_err("create aborted\n");
480 return 1;
481 }
482 if (s->size == 0) {
483 if (mindisc == NULL && !have_container) {
484 pr_err("no size and no drives given - aborting create.\n");
485 return 1;
486 }
487 if (s->level > 0 || s->level == LEVEL_MULTIPATH
488 || s->level == LEVEL_FAULTY
489 || st->ss->external ) {
490 /* size is meaningful */
491 if (!st->ss->validate_geometry(st, s->level, s->layout,
492 s->raiddisks,
493 &s->chunk, minsize*2,
494 data_offset,
495 NULL, NULL, 0)) {
496 pr_err("devices too large for RAID level %d\n", s->level);
497 return 1;
498 }
499 s->size = minsize;
500 if (s->level == 1)
501 /* If this is ever reshaped to RAID5, we will
502 * need a chunksize. So round it off a bit
503 * now just to be safe
504 */
505 s->size &= ~(64ULL-1);
506 if (c->verbose > 0)
507 pr_err("size set to %lluK\n", s->size);
508 }
509 }
510
511 if (!s->bitmap_file &&
512 s->level >= 1 &&
513 st->ss->add_internal_bitmap &&
514 (s->write_behind || s->size > 100*1024*1024ULL)) {
515 if (c->verbose > 0)
516 pr_err("automatically enabling write-intent bitmap on large array\n");
517 s->bitmap_file = "internal";
518 }
519 if (s->bitmap_file && strcmp(s->bitmap_file, "none") == 0)
520 s->bitmap_file = NULL;
521
522 if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) {
523 if (c->runstop != 1 || c->verbose >= 0)
524 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
525 maxdisc, s->size);
526 warn = 1;
527 }
528
529 if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) {
530 if (c->runstop != 1 || c->verbose >= 0)
531 pr_err("%s unable to enumerate platform support\n"
532 " array may not be compatible with hardware/firmware\n",
533 st->ss->name);
534 warn = 1;
535 }
536 st->nodes = c->nodes;
537 st->cluster_name = c->homecluster;
538
539 if (warn) {
540 if (c->runstop!= 1) {
541 if (!ask("Continue creating array? ")) {
542 pr_err("create aborted.\n");
543 return 1;
544 }
545 } else {
546 if (c->verbose > 0)
547 pr_err("creation continuing despite oddities due to --run\n");
548 }
549 }
550
551 /* If this is raid4/5, we want to configure the last active slot
552 * as missing, so that a reconstruct happens (faster than re-parity)
553 * FIX: Can we do this for raid6 as well?
554 */
555 if (st->ss->external == 0 &&
556 s->assume_clean==0 && c->force == 0 && first_missing >= s->raiddisks) {
557 switch ( s->level ) {
558 case 4:
559 case 5:
560 insert_point = s->raiddisks-1;
561 s->sparedisks++;
562 info.array.active_disks--;
563 missing_disks++;
564 break;
565 default:
566 break;
567 }
568 }
569 /* For raid6, if creating with 1 missing drive, make a good drive
570 * into a spare, else the create will fail
571 */
572 if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks &&
573 st->ss->external == 0 &&
574 second_missing >= s->raiddisks && s->level == 6) {
575 insert_point = s->raiddisks - 1;
576 if (insert_point == first_missing)
577 insert_point--;
578 s->sparedisks ++;
579 info.array.active_disks--;
580 missing_disks++;
581 }
582
583 if (s->level <= 0 && first_missing < subdevs * 2) {
584 pr_err("This level does not support missing devices\n");
585 return 1;
586 }
587
588 /* We need to create the device */
589 map_lock(&map);
590 mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name);
591 if (mdfd < 0) {
592 map_unlock(&map);
593 return 1;
594 }
595 /* verify if chosen_name is not in use,
596 * it could be in conflict with already existing device
597 * e.g. container, array
598 */
599 if (strncmp(chosen_name, "/dev/md/", 8) == 0
600 && map_by_name(&map, chosen_name+8) != NULL) {
601 pr_err("Array name %s is in use already.\n",
602 chosen_name);
603 close(mdfd);
604 map_unlock(&map);
605 return 1;
606 }
607 mddev = chosen_name;
608
609 vers = md_get_version(mdfd);
610 if (vers < 9000) {
611 pr_err("Create requires md driver version 0.90.0 or later\n");
612 goto abort_locked;
613 } else {
614 mdu_array_info_t inf;
615 memset(&inf, 0, sizeof(inf));
616 ioctl(mdfd, GET_ARRAY_INFO, &inf);
617 if (inf.working_disks != 0) {
618 pr_err("another array by this name is already running.\n");
619 goto abort_locked;
620 }
621 }
622
623 /* Ok, lets try some ioctls */
624
625 info.array.level = s->level;
626 info.array.size = s->size;
627 info.array.raid_disks = s->raiddisks;
628 /* The kernel should *know* what md_minor we are dealing
629 * with, but it chooses to trust me instead. Sigh
630 */
631 info.array.md_minor = 0;
632 if (fstat(mdfd, &stb)==0)
633 info.array.md_minor = minor(stb.st_rdev);
634 info.array.not_persistent = 0;
635
636 if ( ( (s->level == 4 || s->level == 5) &&
637 (insert_point < s->raiddisks || first_missing < s->raiddisks) )
638 ||
639 ( s->level == 6 && (insert_point < s->raiddisks
640 || second_missing < s->raiddisks))
641 ||
642 ( s->level <= 0 )
643 ||
644 s->assume_clean
645 ) {
646 info.array.state = 1; /* clean, but one+ drive will be missing*/
647 info.resync_start = MaxSector;
648 } else {
649 info.array.state = 0; /* not clean, but no errors */
650 info.resync_start = 0;
651 }
652 if (s->level == 10) {
653 /* for raid10, the bitmap size is the capacity of the array,
654 * which is array.size * raid_disks / ncopies;
655 * .. but convert to sectors.
656 */
657 int ncopies = ((s->layout>>8) & 255) * (s->layout & 255);
658 bitmapsize = s->size * s->raiddisks / ncopies * 2;
659 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
660 } else
661 bitmapsize = s->size * 2;
662
663 /* There is lots of redundancy in these disk counts,
664 * raid_disks is the most meaningful value
665 * it describes the geometry of the array
666 * it is constant
667 * nr_disks is total number of used slots.
668 * it should be raid_disks+spare_disks
669 * spare_disks is the number of extra disks present
670 * see above
671 * active_disks is the number of working disks in
672 * active slots. (With raid_disks)
673 * working_disks is the total number of working disks,
674 * including spares
675 * failed_disks is the number of disks marked failed
676 *
677 * Ideally, the kernel would keep these (except raid_disks)
678 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
679 * So for now, we assume that all raid and spare
680 * devices will be given.
681 */
682 info.array.spare_disks=s->sparedisks;
683 info.array.failed_disks=missing_disks;
684 info.array.nr_disks = info.array.working_disks
685 + info.array.failed_disks;
686 info.array.layout = s->layout;
687 info.array.chunk_size = s->chunk*1024;
688
689 if (name == NULL || *name == 0) {
690 /* base name on mddev */
691 /* /dev/md0 -> 0
692 * /dev/md_d0 -> d0
693 * /dev/md_foo -> foo
694 * /dev/md/1 -> 1
695 * /dev/md/d1 -> d1
696 * /dev/md/home -> home
697 * /dev/mdhome -> home
698 */
699 /* FIXME compare this with rules in create_mddev */
700 name = strrchr(mddev, '/');
701 if (name) {
702 name++;
703 if (strncmp(name, "md_", 3)==0 &&
704 strlen(name) > 3 &&
705 (name-mddev) == 5 /* /dev/ */)
706 name += 3;
707 else if (strncmp(name, "md", 2)==0 &&
708 strlen(name) > 2 &&
709 isdigit(name[2]) &&
710 (name-mddev) == 5 /* /dev/ */)
711 name += 2;
712 }
713 }
714 if (!st->ss->init_super(st, &info.array, s->size, name, c->homehost, uuid,
715 data_offset))
716 goto abort_locked;
717
718 total_slots = info.array.nr_disks;
719 st->ss->getinfo_super(st, &info, NULL);
720 sysfs_init(&info, mdfd, NULL);
721
722 if (did_default && c->verbose >= 0) {
723 if (is_subarray(info.text_version)) {
724 char devnm[32];
725 char *ep;
726 struct mdinfo *mdi;
727
728 strncpy(devnm, info.text_version+1, 32);
729 devnm[31] = 0;
730 ep = strchr(devnm, '/');
731 if (ep)
732 *ep = 0;
733
734 mdi = sysfs_read(-1, devnm, GET_VERSION);
735
736 pr_err("Creating array inside %s container %s\n",
737 mdi?mdi->text_version:"managed", devnm);
738 sysfs_free(mdi);
739 } else
740 pr_err("Defaulting to version %s metadata\n", info.text_version);
741 }
742
743 map_update(&map, fd2devnm(mdfd), info.text_version,
744 info.uuid, chosen_name);
745 /* Keep map locked until devices have been added to array
746 * to stop another mdadm from finding and using those devices.
747 */
748
749 if (s->bitmap_file && vers < 9003) {
750 major_num = BITMAP_MAJOR_HOSTENDIAN;
751 #ifdef __BIG_ENDIAN
752 pr_err("Warning - bitmaps created on this kernel are not portable\n"
753 " between different architectured. Consider upgrading the Linux kernel.\n");
754 #endif
755 }
756
757 if (s->bitmap_file && (strcmp(s->bitmap_file, "internal")==0 ||
758 strcmp(s->bitmap_file, "clustered")==0)) {
759 if ((vers%100) < 2) {
760 pr_err("internal bitmaps not supported by this kernel.\n");
761 goto abort_locked;
762 }
763 if (!st->ss->add_internal_bitmap) {
764 pr_err("internal bitmaps not supported with %s metadata\n",
765 st->ss->name);
766 goto abort_locked;
767 }
768 if (!st->ss->add_internal_bitmap(st, &s->bitmap_chunk,
769 c->delay, s->write_behind,
770 bitmapsize, 1, major_num)) {
771 pr_err("Given bitmap chunk size not supported.\n");
772 goto abort_locked;
773 }
774 s->bitmap_file = NULL;
775 }
776
777 sysfs_init(&info, mdfd, NULL);
778
779 if (st->ss->external && st->container_devnm[0]) {
780 /* member */
781
782 /* When creating a member, we need to be careful
783 * to negotiate with mdmon properly.
784 * If it is already running, we cannot write to
785 * the devices and must ask it to do that part.
786 * If it isn't running, we write to the devices,
787 * and then start it.
788 * We hold an exclusive open on the container
789 * device to make sure mdmon doesn't exit after
790 * we checked that it is running.
791 *
792 * For now, fail if it is already running.
793 */
794 container_fd = open_dev_excl(st->container_devnm);
795 if (container_fd < 0) {
796 pr_err("Cannot get exclusive open on container - weird.\n");
797 goto abort_locked;
798 }
799 if (mdmon_running(st->container_devnm)) {
800 if (c->verbose)
801 pr_err("reusing mdmon for %s.\n",
802 st->container_devnm);
803 st->update_tail = &st->updates;
804 } else
805 need_mdmon = 1;
806 }
807 rv = set_array_info(mdfd, st, &info);
808 if (rv) {
809 pr_err("failed to set array info for %s: %s\n",
810 mddev, strerror(errno));
811 goto abort_locked;
812 }
813
814 if (s->bitmap_file) {
815 int uuid[4];
816
817 st->ss->uuid_from_super(st, uuid);
818 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
819 c->delay, s->write_behind,
820 bitmapsize,
821 major_num)) {
822 goto abort_locked;
823 }
824 bitmap_fd = open(s->bitmap_file, O_RDWR);
825 if (bitmap_fd < 0) {
826 pr_err("weird: %s cannot be openned\n",
827 s->bitmap_file);
828 goto abort_locked;
829 }
830 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
831 pr_err("Cannot set bitmap file for %s: %s\n",
832 mddev, strerror(errno));
833 goto abort_locked;
834 }
835 }
836
837 infos = xmalloc(sizeof(*infos) * total_slots);
838 enable_fds(total_slots);
839 for (pass=1; pass <=2 ; pass++) {
840 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
841
842 for (dnum=0, dv = devlist ; dv ;
843 dv=(dv->next)?(dv->next):moved_disk, dnum++) {
844 int fd;
845 struct stat stb;
846 struct mdinfo *inf = &infos[dnum];
847
848 if (dnum >= total_slots)
849 abort();
850 if (dnum == insert_point) {
851 moved_disk = dv;
852 continue;
853 }
854 if (strcasecmp(dv->devname, "missing")==0)
855 continue;
856 if (have_container)
857 moved_disk = NULL;
858 if (have_container && dnum < info.array.raid_disks - 1)
859 /* repeatedly use the container */
860 moved_disk = dv;
861
862 switch(pass) {
863 case 1:
864 *inf = info;
865
866 inf->disk.number = dnum;
867 inf->disk.raid_disk = dnum;
868 if (inf->disk.raid_disk < s->raiddisks)
869 inf->disk.state = (1<<MD_DISK_ACTIVE) |
870 (1<<MD_DISK_SYNC);
871 else
872 inf->disk.state = 0;
873
874 if (dv->writemostly == 1)
875 inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
876
877 if (have_container)
878 fd = -1;
879 else {
880 if (st->ss->external &&
881 st->container_devnm[0])
882 fd = open(dv->devname, O_RDWR);
883 else
884 fd = open(dv->devname, O_RDWR|O_EXCL);
885
886 if (fd < 0) {
887 pr_err("failed to open %s after earlier success - aborting\n",
888 dv->devname);
889 goto abort_locked;
890 }
891 fstat(fd, &stb);
892 inf->disk.major = major(stb.st_rdev);
893 inf->disk.minor = minor(stb.st_rdev);
894 }
895 if (fd >= 0)
896 remove_partitions(fd);
897 if (st->ss->add_to_super(st, &inf->disk,
898 fd, dv->devname,
899 dv->data_offset)) {
900 ioctl(mdfd, STOP_ARRAY, NULL);
901 goto abort_locked;
902 }
903 st->ss->getinfo_super(st, inf, NULL);
904 safe_mode_delay = inf->safe_mode_delay;
905
906 if (have_container && c->verbose > 0)
907 pr_err("Using %s for device %d\n",
908 map_dev(inf->disk.major,
909 inf->disk.minor,
910 0), dnum);
911
912 if (!have_container) {
913 /* getinfo_super might have lost these ... */
914 inf->disk.major = major(stb.st_rdev);
915 inf->disk.minor = minor(stb.st_rdev);
916 }
917 break;
918 case 2:
919 inf->errors = 0;
920
921 rv = add_disk(mdfd, st, &info, inf);
922
923 if (rv) {
924 pr_err("ADD_NEW_DISK for %s failed: %s\n",
925 dv->devname, strerror(errno));
926 goto abort_locked;
927 }
928 break;
929 }
930 if (!have_container &&
931 dv == moved_disk && dnum != insert_point) break;
932 }
933 if (pass == 1) {
934 struct mdinfo info_new;
935 struct map_ent *me = NULL;
936
937 /* check to see if the uuid has changed due to these
938 * metadata changes, and if so update the member array
939 * and container uuid. Note ->write_init_super clears
940 * the subarray cursor such that ->getinfo_super once
941 * again returns container info.
942 */
943 st->ss->getinfo_super(st, &info_new, NULL);
944 if (st->ss->external && s->level != LEVEL_CONTAINER &&
945 !same_uuid(info_new.uuid, info.uuid, 0)) {
946 map_update(&map, fd2devnm(mdfd),
947 info_new.text_version,
948 info_new.uuid, chosen_name);
949 me = map_by_devnm(&map, st->container_devnm);
950 }
951
952 if (st->ss->write_init_super(st)) {
953 st->ss->free_super(st);
954 goto abort_locked;
955 }
956
957 /* update parent container uuid */
958 if (me) {
959 char *path = xstrdup(me->path);
960
961 st->ss->getinfo_super(st, &info_new, NULL);
962 map_update(&map, st->container_devnm,
963 info_new.text_version,
964 info_new.uuid, path);
965 free(path);
966 }
967
968 flush_metadata_updates(st);
969 st->ss->free_super(st);
970 }
971 }
972 map_unlock(&map);
973 free(infos);
974
975 if (s->level == LEVEL_CONTAINER) {
976 /* No need to start. But we should signal udev to
977 * create links */
978 sysfs_uevent(&info, "change");
979 if (c->verbose >= 0)
980 pr_err("container %s prepared.\n", mddev);
981 wait_for(chosen_name, mdfd);
982 } else if (c->runstop == 1 || subdevs >= s->raiddisks) {
983 if (st->ss->external) {
984 int err;
985 switch(s->level) {
986 case LEVEL_LINEAR:
987 case LEVEL_MULTIPATH:
988 case 0:
989 err = sysfs_set_str(&info, NULL, "array_state",
990 c->readonly
991 ? "readonly"
992 : "active");
993 need_mdmon = 0;
994 break;
995 default:
996 err = sysfs_set_str(&info, NULL, "array_state",
997 "readonly");
998 break;
999 }
1000 sysfs_set_safemode(&info, safe_mode_delay);
1001 if (err) {
1002 pr_err("failed to activate array.\n");
1003 ioctl(mdfd, STOP_ARRAY, NULL);
1004 goto abort;
1005 }
1006 } else if (c->readonly &&
1007 sysfs_attribute_available(
1008 &info, NULL, "array_state")) {
1009 if (sysfs_set_str(&info, NULL,
1010 "array_state", "readonly") < 0) {
1011 pr_err("Failed to start array: %s\n",
1012 strerror(errno));
1013 ioctl(mdfd, STOP_ARRAY, NULL);
1014 goto abort;
1015 }
1016 } else {
1017 /* param is not actually used */
1018 mdu_param_t param;
1019 if (ioctl(mdfd, RUN_ARRAY, &param)) {
1020 pr_err("RUN_ARRAY failed: %s\n",
1021 strerror(errno));
1022 if (info.array.chunk_size & (info.array.chunk_size-1)) {
1023 cont_err("Problem may be that chunk size is not a power of 2\n");
1024 }
1025 ioctl(mdfd, STOP_ARRAY, NULL);
1026 goto abort;
1027 }
1028 /* if start_ro module parameter is set, array is
1029 * auto-read-only, which is bad as the resync won't
1030 * start. So lets make it read-write now.
1031 */
1032 ioctl(mdfd, RESTART_ARRAY_RW, NULL);
1033 }
1034 if (c->verbose >= 0)
1035 pr_err("array %s started.\n", mddev);
1036 if (st->ss->external && st->container_devnm[0]) {
1037 if (need_mdmon)
1038 start_mdmon(st->container_devnm);
1039
1040 ping_monitor(st->container_devnm);
1041 close(container_fd);
1042 }
1043 wait_for(chosen_name, mdfd);
1044 } else {
1045 pr_err("not starting array - not enough devices.\n");
1046 }
1047 close(mdfd);
1048 return 0;
1049
1050 abort:
1051 map_lock(&map);
1052 abort_locked:
1053 map_remove(&map, fd2devnm(mdfd));
1054 map_unlock(&map);
1055
1056 if (mdfd >= 0)
1057 close(mdfd);
1058 return 1;
1059 }
Unnamed repository; edit this file 'description' to name the repository.