]> 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
Grow: fix a couple of typos with --assume-clean usage
[thirdparty/mdadm.git] / Create.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2012 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 "
181 "spare disks at create time\n");
182 return 1;
183 }
184 if (subdevs > s->raiddisks+s->sparedisks) {
185 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks);
186 return 1;
187 }
188 if (!have_container && subdevs < s->raiddisks+s->sparedisks) {
189 pr_err("You haven't given enough devices (real or missing) to create this array\n");
190 return 1;
191 }
192 if (s->bitmap_file && s->level <= 0) {
193 pr_err("bitmaps not meaningful with level %s\n",
194 map_num(pers, s->level)?:"given");
195 return 1;
196 }
197
198 /* now set some defaults */
199
200 if (s->layout == UnSet) {
201 do_default_layout = 1;
202 s->layout = default_layout(st, s->level, c->verbose);
203 }
204
205 if (s->level == 10)
206 /* check layout fits in array*/
207 if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) {
208 pr_err("that layout requires at least %d devices\n",
209 (s->layout&255) * ((s->layout>>8)&255));
210 return 1;
211 }
212
213 switch(s->level) {
214 case 4:
215 case 5:
216 case 10:
217 case 6:
218 case 0:
219 if (s->chunk == 0 || s->chunk == UnSet) {
220 s->chunk = UnSet;
221 do_default_chunk = 1;
222 /* chunk will be set later */
223 }
224 break;
225 case LEVEL_LINEAR:
226 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
227 if (get_linux_version() < 2006016 && s->chunk == 0) {
228 s->chunk = 64;
229 if (c->verbose > 0)
230 pr_err("chunk size defaults to 64K\n");
231 }
232 break;
233 case 1:
234 case LEVEL_FAULTY:
235 case LEVEL_MULTIPATH:
236 case LEVEL_CONTAINER:
237 if (s->chunk) {
238 s->chunk = 0;
239 if (c->verbose > 0)
240 pr_err("chunk size ignored for this level\n");
241 }
242 break;
243 default:
244 pr_err("unknown level %d\n", s->level);
245 return 1;
246 }
247 if (s->size == MAX_SIZE)
248 /* use '0' to mean 'max' now... */
249 s->size = 0;
250 if (s->size && s->chunk && s->chunk != UnSet)
251 s->size &= ~(unsigned long long)(s->chunk - 1);
252 newsize = s->size * 2;
253 if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks,
254 &s->chunk, s->size*2,
255 data_offset, NULL,
256 &newsize, c->verbose>=0))
257 return 1;
258
259 if (s->chunk && s->chunk != UnSet) {
260 newsize &= ~(unsigned long long)(s->chunk*2 - 1);
261 if (do_default_chunk) {
262 /* default chunk was just set */
263 if (c->verbose > 0)
264 pr_err("chunk size "
265 "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 && !have_container; dv=dv->next, dnum++) {
289 char *dname = dv->devname;
290 unsigned long long freesize;
291 int dfd;
292 char *doff;
293
294 if (strcasecmp(dname, "missing")==0) {
295 if (first_missing > dnum)
296 first_missing = dnum;
297 if (second_missing > dnum && dnum > first_missing)
298 second_missing = dnum;
299 missing_disks ++;
300 continue;
301 }
302 if (data_offset != VARIABLE_OFFSET) {
303 doff = strchr(dname, ':');
304 if (doff) {
305 *doff++ = 0;
306 dv->data_offset = parse_size(doff);
307 } else
308 dv->data_offset = INVALID_SECTORS;
309 } else
310 dv->data_offset = data_offset;
311
312 dfd = open(dname, O_RDONLY);
313 if (dfd < 0) {
314 pr_err("cannot open %s: %s\n",
315 dname, strerror(errno));
316 exit(2);
317 }
318 if (fstat(dfd, &stb) != 0 ||
319 (stb.st_mode & S_IFMT) != S_IFBLK) {
320 close(dfd);
321 pr_err("%s is not a block device\n",
322 dname);
323 exit(2);
324 }
325 close(dfd);
326 info.array.working_disks++;
327 if (dnum < s->raiddisks)
328 info.array.active_disks++;
329 if (st == NULL) {
330 struct createinfo *ci = conf_get_create_info();
331 if (ci)
332 st = ci->supertype;
333 }
334 if (st == NULL) {
335 /* Need to choose a default metadata, which is different
336 * depending on geometry of array.
337 */
338 int i;
339 char *name = "default";
340 for(i=0; !st && superlist[i]; i++) {
341 st = superlist[i]->match_metadata_desc(name);
342 if (!st)
343 continue;
344 if (do_default_layout)
345 s->layout = default_layout(st, s->level, c->verbose);
346 switch (st->ss->validate_geometry(
347 st, s->level, s->layout, s->raiddisks,
348 &s->chunk, s->size*2,
349 dv->data_offset, dname,
350 &freesize, c->verbose > 0)) {
351 case -1: /* Not valid, message printed, and not
352 * worth checking any further */
353 exit(2);
354 break;
355 case 0: /* Geometry not valid */
356 free(st);
357 st = NULL;
358 s->chunk = do_default_chunk ? UnSet : s->chunk;
359 break;
360 case 1: /* All happy */
361 break;
362 }
363 }
364
365 if (!st) {
366 int dfd = open(dname, O_RDONLY|O_EXCL);
367 if (dfd < 0) {
368 pr_err("cannot open %s: %s\n",
369 dname, strerror(errno));
370 exit(2);
371 }
372 pr_err("device %s not suitable "
373 "for any style of array\n",
374 dname);
375 exit(2);
376 }
377 if (st->ss != &super0 ||
378 st->minor_version != 90)
379 did_default = 1;
380 } else {
381 if (do_default_layout)
382 s->layout = default_layout(st, s->level, 0);
383 if (!st->ss->validate_geometry(st, s->level, s->layout,
384 s->raiddisks,
385 &s->chunk, s->size*2,
386 dv->data_offset,
387 dname, &freesize,
388 c->verbose >= 0)) {
389
390 pr_err("%s is not suitable for "
391 "this array.\n",
392 dname);
393 fail = 1;
394 continue;
395 }
396 }
397
398 freesize /= 2; /* convert to K */
399 if (s->chunk && s->chunk != UnSet) {
400 /* round to chunk size */
401 freesize = freesize & ~(s->chunk-1);
402 if (do_default_chunk) {
403 /* default chunk was just set */
404 if (c->verbose > 0)
405 pr_err("chunk size "
406 "defaults to %dK\n", s->chunk);
407 s->size &= ~(unsigned long long)(s->chunk - 1);
408 do_default_chunk = 0;
409 }
410 }
411
412 if (s->size && freesize < s->size) {
413 pr_err("%s is smaller than given size."
414 " %lluK < %lluK + metadata\n",
415 dname, freesize, s->size);
416 fail = 1;
417 continue;
418 }
419 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
420 maxdisc = dname;
421 maxsize = freesize;
422 }
423 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
424 mindisc = dname;
425 minsize = freesize;
426 }
427 if (c->runstop != 1 || c->verbose >= 0) {
428 int fd = open(dname, O_RDONLY);
429 if (fd <0 ) {
430 pr_err("Cannot open %s: %s\n",
431 dname, strerror(errno));
432 fail=1;
433 continue;
434 }
435 warn |= check_ext2(fd, dname);
436 warn |= check_reiser(fd, dname);
437 warn |= check_raid(fd, dname);
438 if (strcmp(st->ss->name, "1.x") == 0 &&
439 st->minor_version >= 1)
440 /* metadata at front */
441 warn |= check_partitions(fd, dname, 0, 0);
442 else if (s->level == 1 || s->level == LEVEL_CONTAINER
443 || (s->level == 0 && s->raiddisks == 1))
444 /* partitions could be meaningful */
445 warn |= check_partitions(fd, dname, freesize*2, s->size*2);
446 else
447 /* partitions cannot be meaningful */
448 warn |= check_partitions(fd, dname, 0, 0);
449 if (strcmp(st->ss->name, "1.x") == 0 &&
450 st->minor_version >= 1 &&
451 did_default &&
452 s->level == 1 &&
453 (warn & 1024) == 0) {
454 warn |= 1024;
455 pr_err("Note: this array has metadata at the start and\n"
456 " may not be suitable as a boot device. If you plan to\n"
457 " store '/boot' on this device please ensure that\n"
458 " your boot-loader understands md/v1.x metadata, or use\n"
459 " --metadata=0.90\n");
460 }
461 close(fd);
462 }
463 }
464 if (s->raiddisks + s->sparedisks > st->max_devs) {
465 pr_err("Too many devices:"
466 " %s metadata only supports %d\n",
467 st->ss->name, st->max_devs);
468 return 1;
469 }
470 if (have_container)
471 info.array.working_disks = s->raiddisks;
472 if (fail) {
473 pr_err("create aborted\n");
474 return 1;
475 }
476 if (s->size == 0) {
477 if (mindisc == NULL && !have_container) {
478 pr_err("no size and no drives given - aborting create.\n");
479 return 1;
480 }
481 if (s->level > 0 || s->level == LEVEL_MULTIPATH
482 || s->level == LEVEL_FAULTY
483 || st->ss->external ) {
484 /* size is meaningful */
485 if (!st->ss->validate_geometry(st, s->level, s->layout,
486 s->raiddisks,
487 &s->chunk, minsize*2,
488 data_offset,
489 NULL, NULL, 0)) {
490 pr_err("devices too large for RAID level %d\n", s->level);
491 return 1;
492 }
493 s->size = minsize;
494 if (s->level == 1)
495 /* If this is ever reshaped to RAID5, we will
496 * need a chunksize. So round it off a bit
497 * now just to be safe
498 */
499 s->size &= ~(64ULL-1);
500 if (c->verbose > 0)
501 pr_err("size set to %lluK\n", s->size);
502 }
503 }
504 if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) {
505 if (c->runstop != 1 || c->verbose >= 0)
506 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
507 maxdisc, s->size);
508 warn = 1;
509 }
510
511 if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) {
512 if (c->runstop != 1 || c->verbose >= 0)
513 pr_err("%s unable to enumerate platform support\n"
514 " array may not be compatible with hardware/firmware\n",
515 st->ss->name);
516 warn = 1;
517 }
518
519 if (warn) {
520 if (c->runstop!= 1) {
521 if (!ask("Continue creating array? ")) {
522 pr_err("create aborted.\n");
523 return 1;
524 }
525 } else {
526 if (c->verbose > 0)
527 pr_err("creation continuing despite oddities due to --run\n");
528 }
529 }
530
531 /* If this is raid4/5, we want to configure the last active slot
532 * as missing, so that a reconstruct happens (faster than re-parity)
533 * FIX: Can we do this for raid6 as well?
534 */
535 if (st->ss->external == 0 &&
536 s->assume_clean==0 && c->force == 0 && first_missing >= s->raiddisks) {
537 switch ( s->level ) {
538 case 4:
539 case 5:
540 insert_point = s->raiddisks-1;
541 s->sparedisks++;
542 info.array.active_disks--;
543 missing_disks++;
544 break;
545 default:
546 break;
547 }
548 }
549 /* For raid6, if creating with 1 missing drive, make a good drive
550 * into a spare, else the create will fail
551 */
552 if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks &&
553 st->ss->external == 0 &&
554 second_missing >= s->raiddisks && s->level == 6) {
555 insert_point = s->raiddisks - 1;
556 if (insert_point == first_missing)
557 insert_point--;
558 s->sparedisks ++;
559 info.array.active_disks--;
560 missing_disks++;
561 }
562
563 if (s->level <= 0 && first_missing < subdevs * 2) {
564 pr_err("This level does not support missing devices\n");
565 return 1;
566 }
567
568 /* We need to create the device */
569 map_lock(&map);
570 mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name);
571 if (mdfd < 0) {
572 map_unlock(&map);
573 return 1;
574 }
575 /* verify if chosen_name is not in use,
576 * it could be in conflict with already existing device
577 * e.g. container, array
578 */
579 if (strncmp(chosen_name, "/dev/md/", 8) == 0
580 && map_by_name(&map, chosen_name+8) != NULL) {
581 pr_err("Array name %s is in use already.\n",
582 chosen_name);
583 close(mdfd);
584 map_unlock(&map);
585 return 1;
586 }
587 mddev = chosen_name;
588
589 vers = md_get_version(mdfd);
590 if (vers < 9000) {
591 pr_err("Create requires md driver version 0.90.0 or later\n");
592 goto abort_locked;
593 } else {
594 mdu_array_info_t inf;
595 memset(&inf, 0, sizeof(inf));
596 ioctl(mdfd, GET_ARRAY_INFO, &inf);
597 if (inf.working_disks != 0) {
598 pr_err("another array by this name"
599 " is already running.\n");
600 goto abort_locked;
601 }
602 }
603
604 /* Ok, lets try some ioctls */
605
606 info.array.level = s->level;
607 info.array.size = s->size;
608 info.array.raid_disks = s->raiddisks;
609 /* The kernel should *know* what md_minor we are dealing
610 * with, but it chooses to trust me instead. Sigh
611 */
612 info.array.md_minor = 0;
613 if (fstat(mdfd, &stb)==0)
614 info.array.md_minor = minor(stb.st_rdev);
615 info.array.not_persistent = 0;
616
617 if ( ( (s->level == 4 || s->level == 5) &&
618 (insert_point < s->raiddisks || first_missing < s->raiddisks) )
619 ||
620 ( s->level == 6 && (insert_point < s->raiddisks
621 || second_missing < s->raiddisks))
622 ||
623 ( s->level <= 0 )
624 ||
625 s->assume_clean
626 ) {
627 info.array.state = 1; /* clean, but one+ drive will be missing*/
628 info.resync_start = MaxSector;
629 } else {
630 info.array.state = 0; /* not clean, but no errors */
631 info.resync_start = 0;
632 }
633 if (s->level == 10) {
634 /* for raid10, the bitmap size is the capacity of the array,
635 * which is array.size * raid_disks / ncopies;
636 * .. but convert to sectors.
637 */
638 int ncopies = ((s->layout>>8) & 255) * (s->layout & 255);
639 bitmapsize = s->size * s->raiddisks / ncopies * 2;
640 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
641 } else
642 bitmapsize = s->size * 2;
643
644 /* There is lots of redundancy in these disk counts,
645 * raid_disks is the most meaningful value
646 * it describes the geometry of the array
647 * it is constant
648 * nr_disks is total number of used slots.
649 * it should be raid_disks+spare_disks
650 * spare_disks is the number of extra disks present
651 * see above
652 * active_disks is the number of working disks in
653 * active slots. (With raid_disks)
654 * working_disks is the total number of working disks,
655 * including spares
656 * failed_disks is the number of disks marked failed
657 *
658 * Ideally, the kernel would keep these (except raid_disks)
659 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
660 * So for now, we assume that all raid and spare
661 * devices will be given.
662 */
663 info.array.spare_disks=s->sparedisks;
664 info.array.failed_disks=missing_disks;
665 info.array.nr_disks = info.array.working_disks
666 + info.array.failed_disks;
667 info.array.layout = s->layout;
668 info.array.chunk_size = s->chunk*1024;
669
670 if (name == NULL || *name == 0) {
671 /* base name on mddev */
672 /* /dev/md0 -> 0
673 * /dev/md_d0 -> d0
674 * /dev/md/1 -> 1
675 * /dev/md/d1 -> d1
676 * /dev/md/home -> home
677 * /dev/mdhome -> home
678 */
679 /* FIXME compare this with rules in create_mddev */
680 name = strrchr(mddev, '/');
681 if (name) {
682 name++;
683 if (strncmp(name, "md_d", 4)==0 &&
684 strlen(name) > 4 &&
685 isdigit(name[4]) &&
686 (name-mddev) == 5 /* /dev/ */)
687 name += 3;
688 else if (strncmp(name, "md", 2)==0 &&
689 strlen(name) > 2 &&
690 isdigit(name[2]) &&
691 (name-mddev) == 5 /* /dev/ */)
692 name += 2;
693 }
694 }
695 if (!st->ss->init_super(st, &info.array, s->size, name, c->homehost, uuid,
696 data_offset))
697 goto abort_locked;
698
699 total_slots = info.array.nr_disks;
700 st->ss->getinfo_super(st, &info, NULL);
701 sysfs_init(&info, mdfd, 0);
702
703 if (did_default && c->verbose >= 0) {
704 if (is_subarray(info.text_version)) {
705 int dnum = devname2devnum(info.text_version+1);
706 char *path;
707 int mdp = get_mdp_major();
708 struct mdinfo *mdi;
709 if (dnum > 0)
710 path = map_dev(MD_MAJOR, dnum, 1);
711 else
712 path = map_dev(mdp, (-1-dnum)<< 6, 1);
713
714 mdi = sysfs_read(-1, dnum, GET_VERSION);
715
716 pr_err("Creating array inside "
717 "%s container %s\n",
718 mdi?mdi->text_version:"managed", path);
719 sysfs_free(mdi);
720 } else
721 pr_err("Defaulting to version"
722 " %s metadata\n", info.text_version);
723 }
724
725 map_update(&map, fd2devnum(mdfd), info.text_version,
726 info.uuid, chosen_name);
727 map_unlock(&map);
728
729 if (s->bitmap_file && vers < 9003) {
730 major_num = BITMAP_MAJOR_HOSTENDIAN;
731 #ifdef __BIG_ENDIAN
732 pr_err("Warning - bitmaps created on this kernel are not portable\n"
733 " between different architectured. Consider upgrading the Linux kernel.\n");
734 #endif
735 }
736
737 if (s->bitmap_file && strcmp(s->bitmap_file, "internal")==0) {
738 if ((vers%100) < 2) {
739 pr_err("internal bitmaps not supported by this kernel.\n");
740 goto abort;
741 }
742 if (!st->ss->add_internal_bitmap) {
743 pr_err("internal bitmaps not supported with %s metadata\n",
744 st->ss->name);
745 goto abort;
746 }
747 if (!st->ss->add_internal_bitmap(st, &s->bitmap_chunk,
748 c->delay, s->write_behind,
749 bitmapsize, 1, major_num)) {
750 pr_err("Given bitmap chunk size not supported.\n");
751 goto abort;
752 }
753 s->bitmap_file = NULL;
754 }
755
756 sysfs_init(&info, mdfd, 0);
757
758 if (st->ss->external && st->container_dev != NoMdDev) {
759 /* member */
760
761 /* When creating a member, we need to be careful
762 * to negotiate with mdmon properly.
763 * If it is already running, we cannot write to
764 * the devices and must ask it to do that part.
765 * If it isn't running, we write to the devices,
766 * and then start it.
767 * We hold an exclusive open on the container
768 * device to make sure mdmon doesn't exit after
769 * we checked that it is running.
770 *
771 * For now, fail if it is already running.
772 */
773 container_fd = open_dev_excl(st->container_dev);
774 if (container_fd < 0) {
775 pr_err("Cannot get exclusive "
776 "open on container - weird.\n");
777 goto abort;
778 }
779 if (mdmon_running(st->container_dev)) {
780 if (c->verbose)
781 pr_err("reusing mdmon "
782 "for %s.\n",
783 devnum2devname(st->container_dev));
784 st->update_tail = &st->updates;
785 } else
786 need_mdmon = 1;
787 }
788 rv = set_array_info(mdfd, st, &info);
789 if (rv) {
790 pr_err("failed to set array info for %s: %s\n",
791 mddev, strerror(errno));
792 goto abort;
793 }
794
795 if (s->bitmap_file) {
796 int uuid[4];
797
798 st->ss->uuid_from_super(st, uuid);
799 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
800 c->delay, s->write_behind,
801 bitmapsize,
802 major_num)) {
803 goto abort;
804 }
805 bitmap_fd = open(s->bitmap_file, O_RDWR);
806 if (bitmap_fd < 0) {
807 pr_err("weird: %s cannot be openned\n",
808 s->bitmap_file);
809 goto abort;
810 }
811 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
812 pr_err("Cannot set bitmap file for %s: %s\n",
813 mddev, strerror(errno));
814 goto abort;
815 }
816 }
817
818 infos = xmalloc(sizeof(*infos) * total_slots);
819
820 for (pass=1; pass <=2 ; pass++) {
821 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
822
823 for (dnum=0, dv = devlist ; dv ;
824 dv=(dv->next)?(dv->next):moved_disk, dnum++) {
825 int fd;
826 struct stat stb;
827 struct mdinfo *inf = &infos[dnum];
828
829 if (dnum >= total_slots)
830 abort();
831 if (dnum == insert_point) {
832 moved_disk = dv;
833 continue;
834 }
835 if (strcasecmp(dv->devname, "missing")==0)
836 continue;
837 if (have_container)
838 moved_disk = NULL;
839 if (have_container && dnum < info.array.raid_disks - 1)
840 /* repeatedly use the container */
841 moved_disk = dv;
842
843 switch(pass) {
844 case 1:
845 *inf = info;
846
847 inf->disk.number = dnum;
848 inf->disk.raid_disk = dnum;
849 if (inf->disk.raid_disk < s->raiddisks)
850 inf->disk.state = (1<<MD_DISK_ACTIVE) |
851 (1<<MD_DISK_SYNC);
852 else
853 inf->disk.state = 0;
854
855 if (dv->writemostly == 1)
856 inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
857
858 if (have_container)
859 fd = -1;
860 else {
861 if (st->ss->external &&
862 st->container_dev != NoMdDev)
863 fd = open(dv->devname, O_RDWR);
864 else
865 fd = open(dv->devname, O_RDWR|O_EXCL);
866
867 if (fd < 0) {
868 pr_err("failed to open %s "
869 "after earlier success - aborting\n",
870 dv->devname);
871 goto abort;
872 }
873 fstat(fd, &stb);
874 inf->disk.major = major(stb.st_rdev);
875 inf->disk.minor = minor(stb.st_rdev);
876 }
877 if (fd >= 0)
878 remove_partitions(fd);
879 if (st->ss->add_to_super(st, &inf->disk,
880 fd, dv->devname,
881 dv->data_offset)) {
882 ioctl(mdfd, STOP_ARRAY, NULL);
883 goto abort;
884 }
885 st->ss->getinfo_super(st, inf, NULL);
886 safe_mode_delay = inf->safe_mode_delay;
887
888 if (have_container && c->verbose > 0)
889 pr_err("Using %s for device %d\n",
890 map_dev(inf->disk.major,
891 inf->disk.minor,
892 0), dnum);
893
894 if (!have_container) {
895 /* getinfo_super might have lost these ... */
896 inf->disk.major = major(stb.st_rdev);
897 inf->disk.minor = minor(stb.st_rdev);
898 }
899 break;
900 case 2:
901 inf->errors = 0;
902
903 rv = add_disk(mdfd, st, &info, inf);
904
905 if (rv) {
906 pr_err("ADD_NEW_DISK for %s "
907 "failed: %s\n",
908 dv->devname, strerror(errno));
909 goto abort;
910 }
911 break;
912 }
913 if (!have_container &&
914 dv == moved_disk && dnum != insert_point) break;
915 }
916 if (pass == 1) {
917 struct mdinfo info_new;
918 struct map_ent *me = NULL;
919
920 /* check to see if the uuid has changed due to these
921 * metadata changes, and if so update the member array
922 * and container uuid. Note ->write_init_super clears
923 * the subarray cursor such that ->getinfo_super once
924 * again returns container info.
925 */
926 map_lock(&map);
927 st->ss->getinfo_super(st, &info_new, NULL);
928 if (st->ss->external && s->level != LEVEL_CONTAINER &&
929 !same_uuid(info_new.uuid, info.uuid, 0)) {
930 map_update(&map, fd2devnum(mdfd),
931 info_new.text_version,
932 info_new.uuid, chosen_name);
933 me = map_by_devnum(&map, st->container_dev);
934 }
935
936 if (st->ss->write_init_super(st)) {
937 st->ss->free_super(st);
938 goto abort_locked;
939 }
940
941 /* update parent container uuid */
942 if (me) {
943 char *path = xstrdup(me->path);
944
945 st->ss->getinfo_super(st, &info_new, NULL);
946 map_update(&map, st->container_dev,
947 info_new.text_version,
948 info_new.uuid, path);
949 free(path);
950 }
951 map_unlock(&map);
952
953 flush_metadata_updates(st);
954 st->ss->free_super(st);
955 }
956 }
957 free(infos);
958
959 if (s->level == LEVEL_CONTAINER) {
960 /* No need to start. But we should signal udev to
961 * create links */
962 sysfs_uevent(&info, "change");
963 if (c->verbose >= 0)
964 pr_err("container %s prepared.\n", mddev);
965 wait_for(chosen_name, mdfd);
966 } else if (c->runstop == 1 || subdevs >= s->raiddisks) {
967 if (st->ss->external) {
968 int err;
969 switch(s->level) {
970 case LEVEL_LINEAR:
971 case LEVEL_MULTIPATH:
972 case 0:
973 err = sysfs_set_str(&info, NULL, "array_state",
974 c->readonly
975 ? "readonly"
976 : "active");
977 need_mdmon = 0;
978 break;
979 default:
980 err = sysfs_set_str(&info, NULL, "array_state",
981 "readonly");
982 break;
983 }
984 sysfs_set_safemode(&info, safe_mode_delay);
985 if (err) {
986 pr_err("failed to"
987 " activate array.\n");
988 ioctl(mdfd, STOP_ARRAY, NULL);
989 goto abort;
990 }
991 } else if (c->readonly &&
992 sysfs_attribute_available(
993 &info, NULL, "array_state")) {
994 if (sysfs_set_str(&info, NULL,
995 "array_state", "readonly") < 0) {
996 pr_err("Failed to start array: %s\n",
997 strerror(errno));
998 ioctl(mdfd, STOP_ARRAY, NULL);
999 goto abort;
1000 }
1001 } else {
1002 /* param is not actually used */
1003 mdu_param_t param;
1004 if (ioctl(mdfd, RUN_ARRAY, &param)) {
1005 pr_err("RUN_ARRAY failed: %s\n",
1006 strerror(errno));
1007 if (info.array.chunk_size & (info.array.chunk_size-1)) {
1008 cont_err("Problem may be that "
1009 "chunk size is not a power of 2\n");
1010 }
1011 ioctl(mdfd, STOP_ARRAY, NULL);
1012 goto abort;
1013 }
1014 }
1015 if (c->verbose >= 0)
1016 pr_err("array %s started.\n", mddev);
1017 if (st->ss->external && st->container_dev != NoMdDev) {
1018 if (need_mdmon)
1019 start_mdmon(st->container_dev);
1020
1021 ping_monitor_by_id(st->container_dev);
1022 close(container_fd);
1023 }
1024 wait_for(chosen_name, mdfd);
1025 } else {
1026 pr_err("not starting array - not enough devices.\n");
1027 }
1028 close(mdfd);
1029 return 0;
1030
1031 abort:
1032 map_lock(&map);
1033 abort_locked:
1034 map_remove(&map, fd2devnum(mdfd));
1035 map_unlock(&map);
1036
1037 if (mdfd >= 0)
1038 close(mdfd);
1039 return 1;
1040 }
Unnamed repository; edit this file 'description' to name the repository.