]> 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
Make sure NOFILE resource limit is big enough.
[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 if (!freesize) {
412 pr_err("no free space left on %s\n", dname);
413 fail = 1;
414 continue;
415 }
416
417 if (s->size && freesize < s->size) {
418 pr_err("%s is smaller than given size."
419 " %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:"
471 " %s metadata only supports %d\n",
472 st->ss->name, st->max_devs);
473 return 1;
474 }
475 if (have_container)
476 info.array.working_disks = s->raiddisks;
477 if (fail) {
478 pr_err("create aborted\n");
479 return 1;
480 }
481 if (s->size == 0) {
482 if (mindisc == NULL && !have_container) {
483 pr_err("no size and no drives given - aborting create.\n");
484 return 1;
485 }
486 if (s->level > 0 || s->level == LEVEL_MULTIPATH
487 || s->level == LEVEL_FAULTY
488 || st->ss->external ) {
489 /* size is meaningful */
490 if (!st->ss->validate_geometry(st, s->level, s->layout,
491 s->raiddisks,
492 &s->chunk, minsize*2,
493 data_offset,
494 NULL, NULL, 0)) {
495 pr_err("devices too large for RAID level %d\n", s->level);
496 return 1;
497 }
498 s->size = minsize;
499 if (s->level == 1)
500 /* If this is ever reshaped to RAID5, we will
501 * need a chunksize. So round it off a bit
502 * now just to be safe
503 */
504 s->size &= ~(64ULL-1);
505 if (c->verbose > 0)
506 pr_err("size set to %lluK\n", s->size);
507 }
508 }
509
510 if (!s->bitmap_file &&
511 s->level >= 1 &&
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
535 if (warn) {
536 if (c->runstop!= 1) {
537 if (!ask("Continue creating array? ")) {
538 pr_err("create aborted.\n");
539 return 1;
540 }
541 } else {
542 if (c->verbose > 0)
543 pr_err("creation continuing despite oddities due to --run\n");
544 }
545 }
546
547 /* If this is raid4/5, we want to configure the last active slot
548 * as missing, so that a reconstruct happens (faster than re-parity)
549 * FIX: Can we do this for raid6 as well?
550 */
551 if (st->ss->external == 0 &&
552 s->assume_clean==0 && c->force == 0 && first_missing >= s->raiddisks) {
553 switch ( s->level ) {
554 case 4:
555 case 5:
556 insert_point = s->raiddisks-1;
557 s->sparedisks++;
558 info.array.active_disks--;
559 missing_disks++;
560 break;
561 default:
562 break;
563 }
564 }
565 /* For raid6, if creating with 1 missing drive, make a good drive
566 * into a spare, else the create will fail
567 */
568 if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks &&
569 st->ss->external == 0 &&
570 second_missing >= s->raiddisks && s->level == 6) {
571 insert_point = s->raiddisks - 1;
572 if (insert_point == first_missing)
573 insert_point--;
574 s->sparedisks ++;
575 info.array.active_disks--;
576 missing_disks++;
577 }
578
579 if (s->level <= 0 && first_missing < subdevs * 2) {
580 pr_err("This level does not support missing devices\n");
581 return 1;
582 }
583
584 /* We need to create the device */
585 map_lock(&map);
586 mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name);
587 if (mdfd < 0) {
588 map_unlock(&map);
589 return 1;
590 }
591 /* verify if chosen_name is not in use,
592 * it could be in conflict with already existing device
593 * e.g. container, array
594 */
595 if (strncmp(chosen_name, "/dev/md/", 8) == 0
596 && map_by_name(&map, chosen_name+8) != NULL) {
597 pr_err("Array name %s is in use already.\n",
598 chosen_name);
599 close(mdfd);
600 map_unlock(&map);
601 return 1;
602 }
603 mddev = chosen_name;
604
605 vers = md_get_version(mdfd);
606 if (vers < 9000) {
607 pr_err("Create requires md driver version 0.90.0 or later\n");
608 goto abort_locked;
609 } else {
610 mdu_array_info_t inf;
611 memset(&inf, 0, sizeof(inf));
612 ioctl(mdfd, GET_ARRAY_INFO, &inf);
613 if (inf.working_disks != 0) {
614 pr_err("another array by this name"
615 " is already running.\n");
616 goto abort_locked;
617 }
618 }
619
620 /* Ok, lets try some ioctls */
621
622 info.array.level = s->level;
623 info.array.size = s->size;
624 info.array.raid_disks = s->raiddisks;
625 /* The kernel should *know* what md_minor we are dealing
626 * with, but it chooses to trust me instead. Sigh
627 */
628 info.array.md_minor = 0;
629 if (fstat(mdfd, &stb)==0)
630 info.array.md_minor = minor(stb.st_rdev);
631 info.array.not_persistent = 0;
632
633 if ( ( (s->level == 4 || s->level == 5) &&
634 (insert_point < s->raiddisks || first_missing < s->raiddisks) )
635 ||
636 ( s->level == 6 && (insert_point < s->raiddisks
637 || second_missing < s->raiddisks))
638 ||
639 ( s->level <= 0 )
640 ||
641 s->assume_clean
642 ) {
643 info.array.state = 1; /* clean, but one+ drive will be missing*/
644 info.resync_start = MaxSector;
645 } else {
646 info.array.state = 0; /* not clean, but no errors */
647 info.resync_start = 0;
648 }
649 if (s->level == 10) {
650 /* for raid10, the bitmap size is the capacity of the array,
651 * which is array.size * raid_disks / ncopies;
652 * .. but convert to sectors.
653 */
654 int ncopies = ((s->layout>>8) & 255) * (s->layout & 255);
655 bitmapsize = s->size * s->raiddisks / ncopies * 2;
656 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
657 } else
658 bitmapsize = s->size * 2;
659
660 /* There is lots of redundancy in these disk counts,
661 * raid_disks is the most meaningful value
662 * it describes the geometry of the array
663 * it is constant
664 * nr_disks is total number of used slots.
665 * it should be raid_disks+spare_disks
666 * spare_disks is the number of extra disks present
667 * see above
668 * active_disks is the number of working disks in
669 * active slots. (With raid_disks)
670 * working_disks is the total number of working disks,
671 * including spares
672 * failed_disks is the number of disks marked failed
673 *
674 * Ideally, the kernel would keep these (except raid_disks)
675 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
676 * So for now, we assume that all raid and spare
677 * devices will be given.
678 */
679 info.array.spare_disks=s->sparedisks;
680 info.array.failed_disks=missing_disks;
681 info.array.nr_disks = info.array.working_disks
682 + info.array.failed_disks;
683 info.array.layout = s->layout;
684 info.array.chunk_size = s->chunk*1024;
685
686 if (name == NULL || *name == 0) {
687 /* base name on mddev */
688 /* /dev/md0 -> 0
689 * /dev/md_d0 -> d0
690 * /dev/md_foo -> foo
691 * /dev/md/1 -> 1
692 * /dev/md/d1 -> d1
693 * /dev/md/home -> home
694 * /dev/mdhome -> home
695 */
696 /* FIXME compare this with rules in create_mddev */
697 name = strrchr(mddev, '/');
698 if (name) {
699 name++;
700 if (strncmp(name, "md_", 3)==0 &&
701 strlen(name) > 3 &&
702 (name-mddev) == 5 /* /dev/ */)
703 name += 3;
704 else if (strncmp(name, "md", 2)==0 &&
705 strlen(name) > 2 &&
706 isdigit(name[2]) &&
707 (name-mddev) == 5 /* /dev/ */)
708 name += 2;
709 }
710 }
711 if (!st->ss->init_super(st, &info.array, s->size, name, c->homehost, uuid,
712 data_offset))
713 goto abort_locked;
714
715 total_slots = info.array.nr_disks;
716 st->ss->getinfo_super(st, &info, NULL);
717 sysfs_init(&info, mdfd, NULL);
718
719 if (did_default && c->verbose >= 0) {
720 if (is_subarray(info.text_version)) {
721 char devnm[32];
722 char *ep;
723 struct mdinfo *mdi;
724
725 strncpy(devnm, info.text_version+1, 32);
726 devnm[31] = 0;
727 ep = strchr(devnm, '/');
728 if (ep)
729 *ep = 0;
730
731 mdi = sysfs_read(-1, devnm, GET_VERSION);
732
733 pr_err("Creating array inside %s container %s\n",
734 mdi?mdi->text_version:"managed", devnm);
735 sysfs_free(mdi);
736 } else
737 pr_err("Defaulting to version"
738 " %s metadata\n", info.text_version);
739 }
740
741 map_update(&map, fd2devnm(mdfd), info.text_version,
742 info.uuid, chosen_name);
743 map_unlock(&map);
744
745 if (s->bitmap_file && vers < 9003) {
746 major_num = BITMAP_MAJOR_HOSTENDIAN;
747 #ifdef __BIG_ENDIAN
748 pr_err("Warning - bitmaps created on this kernel are not portable\n"
749 " between different architectured. Consider upgrading the Linux kernel.\n");
750 #endif
751 }
752
753 if (s->bitmap_file && strcmp(s->bitmap_file, "internal")==0) {
754 if ((vers%100) < 2) {
755 pr_err("internal bitmaps not supported by this kernel.\n");
756 goto abort;
757 }
758 if (!st->ss->add_internal_bitmap) {
759 pr_err("internal bitmaps not supported with %s metadata\n",
760 st->ss->name);
761 goto abort;
762 }
763 if (!st->ss->add_internal_bitmap(st, &s->bitmap_chunk,
764 c->delay, s->write_behind,
765 bitmapsize, 1, major_num)) {
766 pr_err("Given bitmap chunk size not supported.\n");
767 goto abort;
768 }
769 s->bitmap_file = NULL;
770 }
771
772 sysfs_init(&info, mdfd, NULL);
773
774 if (st->ss->external && st->container_devnm[0]) {
775 /* member */
776
777 /* When creating a member, we need to be careful
778 * to negotiate with mdmon properly.
779 * If it is already running, we cannot write to
780 * the devices and must ask it to do that part.
781 * If it isn't running, we write to the devices,
782 * and then start it.
783 * We hold an exclusive open on the container
784 * device to make sure mdmon doesn't exit after
785 * we checked that it is running.
786 *
787 * For now, fail if it is already running.
788 */
789 container_fd = open_dev_excl(st->container_devnm);
790 if (container_fd < 0) {
791 pr_err("Cannot get exclusive "
792 "open on container - weird.\n");
793 goto abort;
794 }
795 if (mdmon_running(st->container_devnm)) {
796 if (c->verbose)
797 pr_err("reusing mdmon "
798 "for %s.\n",
799 st->container_devnm);
800 st->update_tail = &st->updates;
801 } else
802 need_mdmon = 1;
803 }
804 rv = set_array_info(mdfd, st, &info);
805 if (rv) {
806 pr_err("failed to set array info for %s: %s\n",
807 mddev, strerror(errno));
808 goto abort;
809 }
810
811 if (s->bitmap_file) {
812 int uuid[4];
813
814 st->ss->uuid_from_super(st, uuid);
815 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
816 c->delay, s->write_behind,
817 bitmapsize,
818 major_num)) {
819 goto abort;
820 }
821 bitmap_fd = open(s->bitmap_file, O_RDWR);
822 if (bitmap_fd < 0) {
823 pr_err("weird: %s cannot be openned\n",
824 s->bitmap_file);
825 goto abort;
826 }
827 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
828 pr_err("Cannot set bitmap file for %s: %s\n",
829 mddev, strerror(errno));
830 goto abort;
831 }
832 }
833
834 infos = xmalloc(sizeof(*infos) * total_slots);
835 enable_fds(total_slots);
836 for (pass=1; pass <=2 ; pass++) {
837 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
838
839 for (dnum=0, dv = devlist ; dv ;
840 dv=(dv->next)?(dv->next):moved_disk, dnum++) {
841 int fd;
842 struct stat stb;
843 struct mdinfo *inf = &infos[dnum];
844
845 if (dnum >= total_slots)
846 abort();
847 if (dnum == insert_point) {
848 moved_disk = dv;
849 continue;
850 }
851 if (strcasecmp(dv->devname, "missing")==0)
852 continue;
853 if (have_container)
854 moved_disk = NULL;
855 if (have_container && dnum < info.array.raid_disks - 1)
856 /* repeatedly use the container */
857 moved_disk = dv;
858
859 switch(pass) {
860 case 1:
861 *inf = info;
862
863 inf->disk.number = dnum;
864 inf->disk.raid_disk = dnum;
865 if (inf->disk.raid_disk < s->raiddisks)
866 inf->disk.state = (1<<MD_DISK_ACTIVE) |
867 (1<<MD_DISK_SYNC);
868 else
869 inf->disk.state = 0;
870
871 if (dv->writemostly == 1)
872 inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
873
874 if (have_container)
875 fd = -1;
876 else {
877 if (st->ss->external &&
878 st->container_devnm[0])
879 fd = open(dv->devname, O_RDWR);
880 else
881 fd = open(dv->devname, O_RDWR|O_EXCL);
882
883 if (fd < 0) {
884 pr_err("failed to open %s "
885 "after earlier success - aborting\n",
886 dv->devname);
887 goto abort;
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;
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 "
923 "failed: %s\n",
924 dv->devname, strerror(errno));
925 goto abort;
926 }
927 break;
928 }
929 if (!have_container &&
930 dv == moved_disk && dnum != insert_point) break;
931 }
932 if (pass == 1) {
933 struct mdinfo info_new;
934 struct map_ent *me = NULL;
935
936 /* check to see if the uuid has changed due to these
937 * metadata changes, and if so update the member array
938 * and container uuid. Note ->write_init_super clears
939 * the subarray cursor such that ->getinfo_super once
940 * again returns container info.
941 */
942 map_lock(&map);
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 map_unlock(&map);
968
969 flush_metadata_updates(st);
970 st->ss->free_super(st);
971 }
972 }
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"
1003 " activate array.\n");
1004 ioctl(mdfd, STOP_ARRAY, NULL);
1005 goto abort;
1006 }
1007 } else if (c->readonly &&
1008 sysfs_attribute_available(
1009 &info, NULL, "array_state")) {
1010 if (sysfs_set_str(&info, NULL,
1011 "array_state", "readonly") < 0) {
1012 pr_err("Failed to start array: %s\n",
1013 strerror(errno));
1014 ioctl(mdfd, STOP_ARRAY, NULL);
1015 goto abort;
1016 }
1017 } else {
1018 /* param is not actually used */
1019 mdu_param_t param;
1020 if (ioctl(mdfd, RUN_ARRAY, &param)) {
1021 pr_err("RUN_ARRAY failed: %s\n",
1022 strerror(errno));
1023 if (info.array.chunk_size & (info.array.chunk_size-1)) {
1024 cont_err("Problem may be that "
1025 "chunk size is not a power of 2\n");
1026 }
1027 ioctl(mdfd, STOP_ARRAY, NULL);
1028 goto abort;
1029 }
1030 /* if start_ro module parameter is set, array is
1031 * auto-read-only, which is bad as the resync won't
1032 * start. So lets make it read-write now.
1033 */
1034 ioctl(mdfd, RESTART_ARRAY_RW, NULL);
1035 }
1036 if (c->verbose >= 0)
1037 pr_err("array %s started.\n", mddev);
1038 if (st->ss->external && st->container_devnm[0]) {
1039 if (need_mdmon)
1040 start_mdmon(st->container_devnm);
1041
1042 ping_monitor(st->container_devnm);
1043 close(container_fd);
1044 }
1045 wait_for(chosen_name, mdfd);
1046 } else {
1047 pr_err("not starting array - not enough devices.\n");
1048 }
1049 close(mdfd);
1050 return 0;
1051
1052 abort:
1053 map_lock(&map);
1054 abort_locked:
1055 map_remove(&map, fd2devnm(mdfd));
1056 map_unlock(&map);
1057
1058 if (mdfd >= 0)
1059 close(mdfd);
1060 return 1;
1061 }
Unnamed repository; edit this file 'description' to name the repository.