]> 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
Enable create array with write journal (--write-journal DEVICE).
[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 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+s->journaldisks) {
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+s->journaldisks) {
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 if (dv->disposition == 'j')
403 continue; /* skip write journal for size check */
404
405 freesize /= 2; /* convert to K */
406 if (s->chunk && s->chunk != UnSet) {
407 /* round to chunk size */
408 freesize = freesize & ~(s->chunk-1);
409 if (do_default_chunk) {
410 /* default chunk was just set */
411 if (c->verbose > 0)
412 pr_err("chunk size defaults to %dK\n", s->chunk);
413 s->size &= ~(unsigned long long)(s->chunk - 1);
414 do_default_chunk = 0;
415 }
416 }
417 if (!freesize) {
418 pr_err("no free space left on %s\n", dname);
419 fail = 1;
420 continue;
421 }
422
423 if (s->size && freesize < s->size) {
424 pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
425 dname, freesize, s->size);
426 fail = 1;
427 continue;
428 }
429 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
430 maxdisc = dname;
431 maxsize = freesize;
432 }
433 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
434 mindisc = dname;
435 minsize = freesize;
436 }
437 if (c->runstop != 1 || c->verbose >= 0) {
438 int fd = open(dname, O_RDONLY);
439 if (fd <0 ) {
440 pr_err("Cannot open %s: %s\n",
441 dname, strerror(errno));
442 fail=1;
443 continue;
444 }
445 warn |= check_ext2(fd, dname);
446 warn |= check_reiser(fd, dname);
447 warn |= check_raid(fd, dname);
448 if (strcmp(st->ss->name, "1.x") == 0 &&
449 st->minor_version >= 1)
450 /* metadata at front */
451 warn |= check_partitions(fd, dname, 0, 0);
452 else if (s->level == 1 || s->level == LEVEL_CONTAINER
453 || (s->level == 0 && s->raiddisks == 1))
454 /* partitions could be meaningful */
455 warn |= check_partitions(fd, dname, freesize*2, s->size*2);
456 else
457 /* partitions cannot be meaningful */
458 warn |= check_partitions(fd, dname, 0, 0);
459 if (strcmp(st->ss->name, "1.x") == 0 &&
460 st->minor_version >= 1 &&
461 did_default &&
462 s->level == 1 &&
463 (warn & 1024) == 0) {
464 warn |= 1024;
465 pr_err("Note: this array has metadata at the start and\n"
466 " may not be suitable as a boot device. If you plan to\n"
467 " store '/boot' on this device please ensure that\n"
468 " your boot-loader understands md/v1.x metadata, or use\n"
469 " --metadata=0.90\n");
470 }
471 close(fd);
472 }
473 }
474 if (s->raiddisks + s->sparedisks > st->max_devs) {
475 pr_err("Too many devices: %s metadata only supports %d\n",
476 st->ss->name, st->max_devs);
477 return 1;
478 }
479 if (have_container)
480 info.array.working_disks = s->raiddisks;
481 if (fail) {
482 pr_err("create aborted\n");
483 return 1;
484 }
485 if (s->size == 0) {
486 if (mindisc == NULL && !have_container) {
487 pr_err("no size and no drives given - aborting create.\n");
488 return 1;
489 }
490 if (s->level > 0 || s->level == LEVEL_MULTIPATH
491 || s->level == LEVEL_FAULTY
492 || st->ss->external ) {
493 /* size is meaningful */
494 if (!st->ss->validate_geometry(st, s->level, s->layout,
495 s->raiddisks,
496 &s->chunk, minsize*2,
497 data_offset,
498 NULL, NULL, 0)) {
499 pr_err("devices too large for RAID level %d\n", s->level);
500 return 1;
501 }
502 s->size = minsize;
503 if (s->level == 1)
504 /* If this is ever reshaped to RAID5, we will
505 * need a chunksize. So round it off a bit
506 * now just to be safe
507 */
508 s->size &= ~(64ULL-1);
509 if (c->verbose > 0)
510 pr_err("size set to %lluK\n", s->size);
511 }
512 }
513
514 if (!s->bitmap_file &&
515 s->level >= 1 &&
516 st->ss->add_internal_bitmap &&
517 (s->write_behind || s->size > 100*1024*1024ULL)) {
518 if (c->verbose > 0)
519 pr_err("automatically enabling write-intent bitmap on large array\n");
520 s->bitmap_file = "internal";
521 }
522 if (s->bitmap_file && strcmp(s->bitmap_file, "none") == 0)
523 s->bitmap_file = NULL;
524
525 if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) {
526 if (c->runstop != 1 || c->verbose >= 0)
527 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
528 maxdisc, s->size);
529 warn = 1;
530 }
531
532 if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) {
533 if (c->runstop != 1 || c->verbose >= 0)
534 pr_err("%s unable to enumerate platform support\n"
535 " array may not be compatible with hardware/firmware\n",
536 st->ss->name);
537 warn = 1;
538 }
539 st->nodes = c->nodes;
540 st->cluster_name = c->homecluster;
541
542 if (warn) {
543 if (c->runstop!= 1) {
544 if (!ask("Continue creating array? ")) {
545 pr_err("create aborted.\n");
546 return 1;
547 }
548 } else {
549 if (c->verbose > 0)
550 pr_err("creation continuing despite oddities due to --run\n");
551 }
552 }
553
554 /* If this is raid4/5, we want to configure the last active slot
555 * as missing, so that a reconstruct happens (faster than re-parity)
556 * FIX: Can we do this for raid6 as well?
557 */
558 if (st->ss->external == 0 &&
559 s->assume_clean==0 && c->force == 0 && first_missing >= s->raiddisks) {
560 switch ( s->level ) {
561 case 4:
562 case 5:
563 insert_point = s->raiddisks-1;
564 s->sparedisks++;
565 info.array.active_disks--;
566 missing_disks++;
567 break;
568 default:
569 break;
570 }
571 }
572 /* For raid6, if creating with 1 missing drive, make a good drive
573 * into a spare, else the create will fail
574 */
575 if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks &&
576 st->ss->external == 0 &&
577 second_missing >= s->raiddisks && s->level == 6) {
578 insert_point = s->raiddisks - 1;
579 if (insert_point == first_missing)
580 insert_point--;
581 s->sparedisks ++;
582 info.array.active_disks--;
583 missing_disks++;
584 }
585
586 if (s->level <= 0 && first_missing < subdevs * 2) {
587 pr_err("This level does not support missing devices\n");
588 return 1;
589 }
590
591 /* We need to create the device */
592 map_lock(&map);
593 mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name);
594 if (mdfd < 0) {
595 map_unlock(&map);
596 return 1;
597 }
598 /* verify if chosen_name is not in use,
599 * it could be in conflict with already existing device
600 * e.g. container, array
601 */
602 if (strncmp(chosen_name, "/dev/md/", 8) == 0
603 && map_by_name(&map, chosen_name+8) != NULL) {
604 pr_err("Array name %s is in use already.\n",
605 chosen_name);
606 close(mdfd);
607 map_unlock(&map);
608 return 1;
609 }
610 mddev = chosen_name;
611
612 vers = md_get_version(mdfd);
613 if (vers < 9000) {
614 pr_err("Create requires md driver version 0.90.0 or later\n");
615 goto abort_locked;
616 } else {
617 mdu_array_info_t inf;
618 memset(&inf, 0, sizeof(inf));
619 ioctl(mdfd, GET_ARRAY_INFO, &inf);
620 if (inf.working_disks != 0) {
621 pr_err("another array by this name is already running.\n");
622 goto abort_locked;
623 }
624 }
625
626 /* Ok, lets try some ioctls */
627
628 info.array.level = s->level;
629 info.array.size = s->size;
630 info.array.raid_disks = s->raiddisks;
631 /* The kernel should *know* what md_minor we are dealing
632 * with, but it chooses to trust me instead. Sigh
633 */
634 info.array.md_minor = 0;
635 if (fstat(mdfd, &stb)==0)
636 info.array.md_minor = minor(stb.st_rdev);
637 info.array.not_persistent = 0;
638
639 if ( ( (s->level == 4 || s->level == 5) &&
640 (insert_point < s->raiddisks || first_missing < s->raiddisks) )
641 ||
642 ( s->level == 6 && (insert_point < s->raiddisks
643 || second_missing < s->raiddisks))
644 ||
645 ( s->level <= 0 )
646 ||
647 s->assume_clean
648 ) {
649 info.array.state = 1; /* clean, but one+ drive will be missing*/
650 info.resync_start = MaxSector;
651 } else {
652 info.array.state = 0; /* not clean, but no errors */
653 info.resync_start = 0;
654 }
655 if (s->level == 10) {
656 /* for raid10, the bitmap size is the capacity of the array,
657 * which is array.size * raid_disks / ncopies;
658 * .. but convert to sectors.
659 */
660 int ncopies = ((s->layout>>8) & 255) * (s->layout & 255);
661 bitmapsize = s->size * s->raiddisks / ncopies * 2;
662 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
663 } else
664 bitmapsize = s->size * 2;
665
666 /* There is lots of redundancy in these disk counts,
667 * raid_disks is the most meaningful value
668 * it describes the geometry of the array
669 * it is constant
670 * nr_disks is total number of used slots.
671 * it should be raid_disks+spare_disks
672 * spare_disks is the number of extra disks present
673 * see above
674 * active_disks is the number of working disks in
675 * active slots. (With raid_disks)
676 * working_disks is the total number of working disks,
677 * including spares
678 * failed_disks is the number of disks marked failed
679 *
680 * Ideally, the kernel would keep these (except raid_disks)
681 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
682 * So for now, we assume that all raid and spare
683 * devices will be given.
684 */
685 info.array.spare_disks=s->sparedisks;
686 info.array.failed_disks=missing_disks;
687 info.array.nr_disks = info.array.working_disks
688 + info.array.failed_disks;
689 info.array.layout = s->layout;
690 info.array.chunk_size = s->chunk*1024;
691
692 if (name == NULL || *name == 0) {
693 /* base name on mddev */
694 /* /dev/md0 -> 0
695 * /dev/md_d0 -> d0
696 * /dev/md_foo -> foo
697 * /dev/md/1 -> 1
698 * /dev/md/d1 -> d1
699 * /dev/md/home -> home
700 * /dev/mdhome -> home
701 */
702 /* FIXME compare this with rules in create_mddev */
703 name = strrchr(mddev, '/');
704 if (name) {
705 name++;
706 if (strncmp(name, "md_", 3)==0 &&
707 strlen(name) > 3 &&
708 (name-mddev) == 5 /* /dev/ */)
709 name += 3;
710 else if (strncmp(name, "md", 2)==0 &&
711 strlen(name) > 2 &&
712 isdigit(name[2]) &&
713 (name-mddev) == 5 /* /dev/ */)
714 name += 2;
715 }
716 }
717 if (!st->ss->init_super(st, &info.array, s->size, name, c->homehost, uuid,
718 data_offset))
719 goto abort_locked;
720
721 total_slots = info.array.nr_disks;
722 st->ss->getinfo_super(st, &info, NULL);
723 sysfs_init(&info, mdfd, NULL);
724
725 if (did_default && c->verbose >= 0) {
726 if (is_subarray(info.text_version)) {
727 char devnm[32];
728 char *ep;
729 struct mdinfo *mdi;
730
731 strncpy(devnm, info.text_version+1, 32);
732 devnm[31] = 0;
733 ep = strchr(devnm, '/');
734 if (ep)
735 *ep = 0;
736
737 mdi = sysfs_read(-1, devnm, GET_VERSION);
738
739 pr_err("Creating array inside %s container %s\n",
740 mdi?mdi->text_version:"managed", devnm);
741 sysfs_free(mdi);
742 } else
743 pr_err("Defaulting to version %s metadata\n", info.text_version);
744 }
745
746 map_update(&map, fd2devnm(mdfd), info.text_version,
747 info.uuid, chosen_name);
748 /* Keep map locked until devices have been added to array
749 * to stop another mdadm from finding and using those devices.
750 */
751
752 if (s->bitmap_file && vers < 9003) {
753 major_num = BITMAP_MAJOR_HOSTENDIAN;
754 #ifdef __BIG_ENDIAN
755 pr_err("Warning - bitmaps created on this kernel are not portable\n"
756 " between different architectured. Consider upgrading the Linux kernel.\n");
757 #endif
758 }
759
760 if (s->bitmap_file && (strcmp(s->bitmap_file, "internal")==0 ||
761 strcmp(s->bitmap_file, "clustered")==0)) {
762 if ((vers%100) < 2) {
763 pr_err("internal bitmaps not supported by this kernel.\n");
764 goto abort_locked;
765 }
766 if (!st->ss->add_internal_bitmap) {
767 pr_err("internal bitmaps not supported with %s metadata\n",
768 st->ss->name);
769 goto abort_locked;
770 }
771 if (!st->ss->add_internal_bitmap(st, &s->bitmap_chunk,
772 c->delay, s->write_behind,
773 bitmapsize, 1, major_num)) {
774 pr_err("Given bitmap chunk size not supported.\n");
775 goto abort_locked;
776 }
777 s->bitmap_file = NULL;
778 }
779
780 sysfs_init(&info, mdfd, NULL);
781
782 if (st->ss->external && st->container_devnm[0]) {
783 /* member */
784
785 /* When creating a member, we need to be careful
786 * to negotiate with mdmon properly.
787 * If it is already running, we cannot write to
788 * the devices and must ask it to do that part.
789 * If it isn't running, we write to the devices,
790 * and then start it.
791 * We hold an exclusive open on the container
792 * device to make sure mdmon doesn't exit after
793 * we checked that it is running.
794 *
795 * For now, fail if it is already running.
796 */
797 container_fd = open_dev_excl(st->container_devnm);
798 if (container_fd < 0) {
799 pr_err("Cannot get exclusive open on container - weird.\n");
800 goto abort_locked;
801 }
802 if (mdmon_running(st->container_devnm)) {
803 if (c->verbose)
804 pr_err("reusing mdmon for %s.\n",
805 st->container_devnm);
806 st->update_tail = &st->updates;
807 } else
808 need_mdmon = 1;
809 }
810 rv = set_array_info(mdfd, st, &info);
811 if (rv) {
812 pr_err("failed to set array info for %s: %s\n",
813 mddev, strerror(errno));
814 goto abort_locked;
815 }
816
817 if (s->bitmap_file) {
818 int uuid[4];
819
820 st->ss->uuid_from_super(st, uuid);
821 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
822 c->delay, s->write_behind,
823 bitmapsize,
824 major_num)) {
825 goto abort_locked;
826 }
827 bitmap_fd = open(s->bitmap_file, O_RDWR);
828 if (bitmap_fd < 0) {
829 pr_err("weird: %s cannot be openned\n",
830 s->bitmap_file);
831 goto abort_locked;
832 }
833 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
834 pr_err("Cannot set bitmap file for %s: %s\n",
835 mddev, strerror(errno));
836 goto abort_locked;
837 }
838 }
839
840 infos = xmalloc(sizeof(*infos) * total_slots);
841 enable_fds(total_slots);
842 for (pass=1; pass <=2 ; pass++) {
843 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
844
845 for (dnum=0, raid_disk_num=0, dv = devlist ; dv ;
846 dv=(dv->next)?(dv->next):moved_disk, dnum++) {
847 int fd;
848 struct stat stb;
849 struct mdinfo *inf = &infos[dnum];
850
851 if (dnum >= total_slots)
852 abort();
853 if (dnum == insert_point) {
854 moved_disk = dv;
855 continue;
856 }
857 if (strcasecmp(dv->devname, "missing")==0)
858 continue;
859 if (have_container)
860 moved_disk = NULL;
861 if (have_container && dnum < info.array.raid_disks - 1)
862 /* repeatedly use the container */
863 moved_disk = dv;
864
865 switch(pass) {
866 case 1:
867 *inf = info;
868
869 inf->disk.number = dnum;
870 inf->disk.raid_disk = raid_disk_num++;
871
872 if (dv->disposition == 'j') {
873 inf->disk.raid_disk = MD_DISK_ROLE_JOURNAL;
874 inf->disk.state = (1<<MD_DISK_JOURNAL);
875 raid_disk_num--;
876 } else if (inf->disk.raid_disk < s->raiddisks)
877 inf->disk.state = (1<<MD_DISK_ACTIVE) |
878 (1<<MD_DISK_SYNC);
879 else
880 inf->disk.state = 0;
881
882 if (dv->writemostly == 1)
883 inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
884
885 if (have_container)
886 fd = -1;
887 else {
888 if (st->ss->external &&
889 st->container_devnm[0])
890 fd = open(dv->devname, O_RDWR);
891 else
892 fd = open(dv->devname, O_RDWR|O_EXCL);
893
894 if (fd < 0) {
895 pr_err("failed to open %s after earlier success - aborting\n",
896 dv->devname);
897 goto abort_locked;
898 }
899 fstat(fd, &stb);
900 inf->disk.major = major(stb.st_rdev);
901 inf->disk.minor = minor(stb.st_rdev);
902 }
903 if (fd >= 0)
904 remove_partitions(fd);
905 if (st->ss->add_to_super(st, &inf->disk,
906 fd, dv->devname,
907 dv->data_offset)) {
908 ioctl(mdfd, STOP_ARRAY, NULL);
909 goto abort_locked;
910 }
911 st->ss->getinfo_super(st, inf, NULL);
912 safe_mode_delay = inf->safe_mode_delay;
913
914 if (have_container && c->verbose > 0)
915 pr_err("Using %s for device %d\n",
916 map_dev(inf->disk.major,
917 inf->disk.minor,
918 0), dnum);
919
920 if (!have_container) {
921 /* getinfo_super might have lost these ... */
922 inf->disk.major = major(stb.st_rdev);
923 inf->disk.minor = minor(stb.st_rdev);
924 }
925 break;
926 case 2:
927 inf->errors = 0;
928
929 rv = add_disk(mdfd, st, &info, inf);
930
931 if (rv) {
932 pr_err("ADD_NEW_DISK for %s failed: %s\n",
933 dv->devname, strerror(errno));
934 goto abort_locked;
935 }
936 break;
937 }
938 if (!have_container &&
939 dv == moved_disk && dnum != insert_point) break;
940 }
941 if (pass == 1) {
942 struct mdinfo info_new;
943 struct map_ent *me = NULL;
944
945 /* check to see if the uuid has changed due to these
946 * metadata changes, and if so update the member array
947 * and container uuid. Note ->write_init_super clears
948 * the subarray cursor such that ->getinfo_super once
949 * again returns container info.
950 */
951 st->ss->getinfo_super(st, &info_new, NULL);
952 if (st->ss->external && s->level != LEVEL_CONTAINER &&
953 !same_uuid(info_new.uuid, info.uuid, 0)) {
954 map_update(&map, fd2devnm(mdfd),
955 info_new.text_version,
956 info_new.uuid, chosen_name);
957 me = map_by_devnm(&map, st->container_devnm);
958 }
959
960 if (st->ss->write_init_super(st)) {
961 st->ss->free_super(st);
962 goto abort_locked;
963 }
964
965 /* update parent container uuid */
966 if (me) {
967 char *path = xstrdup(me->path);
968
969 st->ss->getinfo_super(st, &info_new, NULL);
970 map_update(&map, st->container_devnm,
971 info_new.text_version,
972 info_new.uuid, path);
973 free(path);
974 }
975
976 flush_metadata_updates(st);
977 st->ss->free_super(st);
978 }
979 }
980 map_unlock(&map);
981 free(infos);
982
983 if (s->level == LEVEL_CONTAINER) {
984 /* No need to start. But we should signal udev to
985 * create links */
986 sysfs_uevent(&info, "change");
987 if (c->verbose >= 0)
988 pr_err("container %s prepared.\n", mddev);
989 wait_for(chosen_name, mdfd);
990 } else if (c->runstop == 1 || subdevs >= s->raiddisks) {
991 if (st->ss->external) {
992 int err;
993 switch(s->level) {
994 case LEVEL_LINEAR:
995 case LEVEL_MULTIPATH:
996 case 0:
997 err = sysfs_set_str(&info, NULL, "array_state",
998 c->readonly
999 ? "readonly"
1000 : "active");
1001 need_mdmon = 0;
1002 break;
1003 default:
1004 err = sysfs_set_str(&info, NULL, "array_state",
1005 "readonly");
1006 break;
1007 }
1008 sysfs_set_safemode(&info, safe_mode_delay);
1009 if (err) {
1010 pr_err("failed to activate array.\n");
1011 ioctl(mdfd, STOP_ARRAY, NULL);
1012 goto abort;
1013 }
1014 } else if (c->readonly &&
1015 sysfs_attribute_available(
1016 &info, NULL, "array_state")) {
1017 if (sysfs_set_str(&info, NULL,
1018 "array_state", "readonly") < 0) {
1019 pr_err("Failed to start array: %s\n",
1020 strerror(errno));
1021 ioctl(mdfd, STOP_ARRAY, NULL);
1022 goto abort;
1023 }
1024 } else {
1025 /* param is not actually used */
1026 mdu_param_t param;
1027 if (ioctl(mdfd, RUN_ARRAY, &param)) {
1028 pr_err("RUN_ARRAY failed: %s\n",
1029 strerror(errno));
1030 if (info.array.chunk_size & (info.array.chunk_size-1)) {
1031 cont_err("Problem may be that chunk size is not a power of 2\n");
1032 }
1033 ioctl(mdfd, STOP_ARRAY, NULL);
1034 goto abort;
1035 }
1036 /* if start_ro module parameter is set, array is
1037 * auto-read-only, which is bad as the resync won't
1038 * start. So lets make it read-write now.
1039 */
1040 ioctl(mdfd, RESTART_ARRAY_RW, NULL);
1041 }
1042 if (c->verbose >= 0)
1043 pr_err("array %s started.\n", mddev);
1044 if (st->ss->external && st->container_devnm[0]) {
1045 if (need_mdmon)
1046 start_mdmon(st->container_devnm);
1047
1048 ping_monitor(st->container_devnm);
1049 close(container_fd);
1050 }
1051 wait_for(chosen_name, mdfd);
1052 } else {
1053 pr_err("not starting array - not enough devices.\n");
1054 }
1055 close(mdfd);
1056 return 0;
1057
1058 abort:
1059 map_lock(&map);
1060 abort_locked:
1061 map_remove(&map, fd2devnm(mdfd));
1062 map_unlock(&map);
1063
1064 if (mdfd >= 0)
1065 close(mdfd);
1066 return 1;
1067 }
Unnamed repository; edit this file 'description' to name the repository.