]> 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
Create: Remove all attemps to handle md driver older than 0.90.03
[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 rv;
101 int bitmap_fd;
102 int have_container = 0;
103 int container_fd = -1;
104 int need_mdmon = 0;
105 unsigned long long bitmapsize;
106 struct mdinfo info, *infos;
107 int did_default = 0;
108 int do_default_layout = 0;
109 int do_default_chunk = 0;
110 unsigned long safe_mode_delay = 0;
111 char chosen_name[1024];
112 struct map_ent *map = NULL;
113 unsigned long long newsize;
114 mdu_array_info_t inf;
115
116 int major_num = BITMAP_MAJOR_HI;
117 if (s->bitmap_file && strcmp(s->bitmap_file, "clustered") == 0) {
118 major_num = BITMAP_MAJOR_CLUSTERED;
119 if (c->nodes <= 1) {
120 pr_err("At least 2 nodes are needed for cluster-md\n");
121 return 1;
122 }
123 }
124
125 memset(&info, 0, sizeof(info));
126 if (s->level == UnSet && st && st->ss->default_geometry)
127 st->ss->default_geometry(st, &s->level, NULL, NULL);
128 if (s->level == UnSet) {
129 pr_err("a RAID level is needed to create an array.\n");
130 return 1;
131 }
132 if (s->raiddisks < 4 && s->level == 6) {
133 pr_err("at least 4 raid-devices needed for level 6\n");
134 return 1;
135 }
136 if (s->raiddisks > 256 && s->level == 6) {
137 pr_err("no more than 256 raid-devices supported for level 6\n");
138 return 1;
139 }
140 if (s->raiddisks < 2 && s->level >= 4) {
141 pr_err("at least 2 raid-devices needed for level 4 or 5\n");
142 return 1;
143 }
144 if (s->level <= 0 && s->sparedisks) {
145 pr_err("This level does not support spare devices\n");
146 return 1;
147 }
148
149 if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
150 /* If given a single device, it might be a container, and we can
151 * extract a device list from there
152 */
153 int fd;
154
155 memset(&inf, 0, sizeof(inf));
156 fd = open(devlist->devname, O_RDONLY);
157 if (fd >= 0 &&
158 md_get_array_info(fd, &inf) == 0 && inf.raid_disks == 0) {
159 /* yep, looks like a container */
160 if (st) {
161 rv = st->ss->load_container(st, fd,
162 devlist->devname);
163 if (rv == 0)
164 have_container = 1;
165 } else {
166 st = super_by_fd(fd, NULL);
167 if (st && !(rv = st->ss->
168 load_container(st, fd,
169 devlist->devname)))
170 have_container = 1;
171 else
172 st = NULL;
173 }
174 if (have_container) {
175 subdevs = s->raiddisks;
176 first_missing = subdevs * 2;
177 second_missing = subdevs * 2;
178 insert_point = subdevs * 2;
179 }
180 }
181 if (fd >= 0)
182 close(fd);
183 }
184 if (st && st->ss->external && s->sparedisks) {
185 pr_err("This metadata type does not support spare disks at create time\n");
186 return 1;
187 }
188 if (subdevs > s->raiddisks+s->sparedisks+s->journaldisks) {
189 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks);
190 return 1;
191 }
192 if (!have_container && subdevs < s->raiddisks+s->sparedisks+s->journaldisks) {
193 pr_err("You haven't given enough devices (real or missing) to create this array\n");
194 return 1;
195 }
196 if (s->bitmap_file && s->level <= 0) {
197 pr_err("bitmaps not meaningful with level %s\n",
198 map_num(pers, s->level)?:"given");
199 return 1;
200 }
201
202 /* now set some defaults */
203
204 if (s->layout == UnSet) {
205 do_default_layout = 1;
206 s->layout = default_layout(st, s->level, c->verbose);
207 }
208
209 if (s->level == 10)
210 /* check layout fits in array*/
211 if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) {
212 pr_err("that layout requires at least %d devices\n",
213 (s->layout&255) * ((s->layout>>8)&255));
214 return 1;
215 }
216
217 switch(s->level) {
218 case 4:
219 case 5:
220 case 10:
221 case 6:
222 case 0:
223 if (s->chunk == 0 || s->chunk == UnSet) {
224 s->chunk = UnSet;
225 do_default_chunk = 1;
226 /* chunk will be set later */
227 }
228 break;
229 case LEVEL_LINEAR:
230 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
231 if (get_linux_version() < 2006016 && s->chunk == 0) {
232 s->chunk = 64;
233 if (c->verbose > 0)
234 pr_err("chunk size defaults to 64K\n");
235 }
236 break;
237 case 1:
238 case LEVEL_FAULTY:
239 case LEVEL_MULTIPATH:
240 case LEVEL_CONTAINER:
241 if (s->chunk) {
242 s->chunk = 0;
243 if (c->verbose > 0)
244 pr_err("chunk size ignored for this level\n");
245 }
246 break;
247 default:
248 pr_err("unknown level %d\n", s->level);
249 return 1;
250 }
251 if (s->size == MAX_SIZE)
252 /* use '0' to mean 'max' now... */
253 s->size = 0;
254 if (s->size && s->chunk && s->chunk != UnSet)
255 s->size &= ~(unsigned long long)(s->chunk - 1);
256 newsize = s->size * 2;
257 if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks,
258 &s->chunk, s->size*2,
259 data_offset, NULL,
260 &newsize, s->consistency_policy,
261 c->verbose >= 0))
262 return 1;
263
264 if (s->chunk && s->chunk != UnSet) {
265 newsize &= ~(unsigned long long)(s->chunk*2 - 1);
266 if (do_default_chunk) {
267 /* default chunk was just set */
268 if (c->verbose > 0)
269 pr_err("chunk size defaults to %dK\n", s->chunk);
270 s->size &= ~(unsigned long long)(s->chunk - 1);
271 do_default_chunk = 0;
272 }
273 }
274
275 if (s->size == 0) {
276 s->size = newsize / 2;
277 if (s->level == 1)
278 /* If this is ever reshaped to RAID5, we will
279 * need a chunksize. So round it off a bit
280 * now just to be safe
281 */
282 s->size &= ~(64ULL-1);
283
284 if (s->size && c->verbose > 0)
285 pr_err("setting size to %lluK\n", s->size);
286 }
287
288 /* now look at the subdevs */
289 info.array.active_disks = 0;
290 info.array.working_disks = 0;
291 dnum = 0;
292 for (dv = devlist; dv; dv = dv->next)
293 if (data_offset == VARIABLE_OFFSET)
294 dv->data_offset = INVALID_SECTORS;
295 else
296 dv->data_offset = data_offset;
297
298 for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) {
299 char *dname = dv->devname;
300 unsigned long long freesize;
301 int dfd;
302 char *doff;
303
304 if (strcasecmp(dname, "missing") == 0) {
305 if (first_missing > dnum)
306 first_missing = dnum;
307 if (second_missing > dnum && dnum > first_missing)
308 second_missing = dnum;
309 missing_disks ++;
310 continue;
311 }
312 if (data_offset == VARIABLE_OFFSET) {
313 doff = strchr(dname, ':');
314 if (doff) {
315 *doff++ = 0;
316 dv->data_offset = parse_size(doff);
317 } else
318 dv->data_offset = INVALID_SECTORS;
319 } else
320 dv->data_offset = data_offset;
321
322 dfd = open(dname, O_RDONLY);
323 if (dfd < 0) {
324 pr_err("cannot open %s: %s\n",
325 dname, strerror(errno));
326 exit(2);
327 }
328 if (fstat(dfd, &stb) != 0 ||
329 (stb.st_mode & S_IFMT) != S_IFBLK) {
330 close(dfd);
331 pr_err("%s is not a block device\n",
332 dname);
333 exit(2);
334 }
335 close(dfd);
336 info.array.working_disks++;
337 if (dnum < s->raiddisks && dv->disposition != 'j')
338 info.array.active_disks++;
339 if (st == NULL) {
340 struct createinfo *ci = conf_get_create_info();
341 if (ci)
342 st = ci->supertype;
343 }
344 if (st == NULL) {
345 /* Need to choose a default metadata, which is different
346 * depending on geometry of array.
347 */
348 int i;
349 char *name = "default";
350 for(i = 0; !st && superlist[i]; i++) {
351 st = superlist[i]->match_metadata_desc(name);
352 if (!st)
353 continue;
354 if (do_default_layout)
355 s->layout = default_layout(st, s->level, c->verbose);
356 switch (st->ss->validate_geometry(
357 st, s->level, s->layout, s->raiddisks,
358 &s->chunk, s->size*2,
359 dv->data_offset, dname,
360 &freesize, s->consistency_policy,
361 c->verbose > 0)) {
362 case -1: /* Not valid, message printed, and not
363 * worth checking any further */
364 exit(2);
365 break;
366 case 0: /* Geometry not valid */
367 free(st);
368 st = NULL;
369 s->chunk = do_default_chunk ? UnSet : s->chunk;
370 break;
371 case 1: /* All happy */
372 break;
373 }
374 }
375
376 if (!st) {
377 int dfd = open(dname, O_RDONLY|O_EXCL);
378 if (dfd < 0) {
379 pr_err("cannot open %s: %s\n",
380 dname, strerror(errno));
381 exit(2);
382 }
383 pr_err("device %s not suitable for any style of array\n",
384 dname);
385 exit(2);
386 }
387 if (st->ss != &super0 ||
388 st->minor_version != 90)
389 did_default = 1;
390 } else {
391 if (do_default_layout)
392 s->layout = default_layout(st, s->level, 0);
393 if (!st->ss->validate_geometry(st, s->level, s->layout,
394 s->raiddisks,
395 &s->chunk, s->size*2,
396 dv->data_offset,
397 dname, &freesize,
398 s->consistency_policy,
399 c->verbose >= 0)) {
400
401 pr_err("%s is not suitable for this array.\n",
402 dname);
403 fail = 1;
404 continue;
405 }
406 }
407
408 if (dv->disposition == 'j')
409 goto skip_size_check; /* skip write journal for size check */
410
411 freesize /= 2; /* convert to K */
412 if (s->chunk && s->chunk != UnSet) {
413 /* round to chunk size */
414 freesize = freesize & ~(s->chunk-1);
415 if (do_default_chunk) {
416 /* default chunk was just set */
417 if (c->verbose > 0)
418 pr_err("chunk size defaults to %dK\n", s->chunk);
419 s->size &= ~(unsigned long long)(s->chunk - 1);
420 do_default_chunk = 0;
421 }
422 }
423 if (!freesize) {
424 pr_err("no free space left on %s\n", dname);
425 fail = 1;
426 continue;
427 }
428
429 if (s->size && freesize < s->size) {
430 pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
431 dname, freesize, s->size);
432 fail = 1;
433 continue;
434 }
435 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
436 maxdisc = dname;
437 maxsize = freesize;
438 }
439 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
440 mindisc = dname;
441 minsize = freesize;
442 }
443 skip_size_check:
444 if (c->runstop != 1 || c->verbose >= 0) {
445 int fd = open(dname, O_RDONLY);
446 if (fd < 0) {
447 pr_err("Cannot open %s: %s\n",
448 dname, strerror(errno));
449 fail = 1;
450 continue;
451 }
452 warn |= check_ext2(fd, dname);
453 warn |= check_reiser(fd, dname);
454 warn |= check_raid(fd, dname);
455 if (strcmp(st->ss->name, "1.x") == 0 &&
456 st->minor_version >= 1)
457 /* metadata at front */
458 warn |= check_partitions(fd, dname, 0, 0);
459 else if (s->level == 1 || s->level == LEVEL_CONTAINER ||
460 (s->level == 0 && s->raiddisks == 1))
461 /* partitions could be meaningful */
462 warn |= check_partitions(fd, dname, freesize*2, s->size*2);
463 else
464 /* partitions cannot be meaningful */
465 warn |= check_partitions(fd, dname, 0, 0);
466 if (strcmp(st->ss->name, "1.x") == 0 &&
467 st->minor_version >= 1 &&
468 did_default &&
469 s->level == 1 &&
470 (warn & 1024) == 0) {
471 warn |= 1024;
472 pr_err("Note: this array has metadata at the start and\n"
473 " may not be suitable as a boot device. If you plan to\n"
474 " store '/boot' on this device please ensure that\n"
475 " your boot-loader understands md/v1.x metadata, or use\n"
476 " --metadata=0.90\n");
477 }
478 close(fd);
479 }
480 }
481 if (s->raiddisks + s->sparedisks > st->max_devs) {
482 pr_err("Too many devices: %s metadata only supports %d\n",
483 st->ss->name, st->max_devs);
484 return 1;
485 }
486 if (have_container)
487 info.array.working_disks = s->raiddisks;
488 if (fail) {
489 pr_err("create aborted\n");
490 return 1;
491 }
492 if (s->size == 0) {
493 if (mindisc == NULL && !have_container) {
494 pr_err("no size and no drives given - aborting create.\n");
495 return 1;
496 }
497 if (s->level > 0 || s->level == LEVEL_MULTIPATH ||
498 s->level == LEVEL_FAULTY || st->ss->external) {
499 /* size is meaningful */
500 if (!st->ss->validate_geometry(st, s->level, s->layout,
501 s->raiddisks,
502 &s->chunk, minsize*2,
503 data_offset,
504 NULL, NULL,
505 s->consistency_policy, 0)) {
506 pr_err("devices too large for RAID level %d\n", s->level);
507 return 1;
508 }
509 s->size = minsize;
510 if (s->level == 1)
511 /* If this is ever reshaped to RAID5, we will
512 * need a chunksize. So round it off a bit
513 * now just to be safe
514 */
515 s->size &= ~(64ULL-1);
516 if (c->verbose > 0)
517 pr_err("size set to %lluK\n", s->size);
518 }
519 }
520
521 if (!s->bitmap_file &&
522 s->level >= 1 &&
523 st->ss->add_internal_bitmap &&
524 (s->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
525 s->consistency_policy != CONSISTENCY_POLICY_PPL) &&
526 (s->write_behind || s->size > 100*1024*1024ULL)) {
527 if (c->verbose > 0)
528 pr_err("automatically enabling write-intent bitmap on large array\n");
529 s->bitmap_file = "internal";
530 }
531 if (s->bitmap_file && strcmp(s->bitmap_file, "none") == 0)
532 s->bitmap_file = NULL;
533
534 if (s->consistency_policy == CONSISTENCY_POLICY_PPL &&
535 !st->ss->write_init_ppl) {
536 pr_err("%s metadata does not support PPL\n", st->ss->name);
537 return 1;
538 }
539
540 if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) {
541 if (c->runstop != 1 || c->verbose >= 0)
542 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
543 maxdisc, s->size);
544 warn = 1;
545 }
546
547 if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) {
548 if (c->runstop != 1 || c->verbose >= 0)
549 pr_err("%s unable to enumerate platform support\n"
550 " array may not be compatible with hardware/firmware\n",
551 st->ss->name);
552 warn = 1;
553 }
554 st->nodes = c->nodes;
555 st->cluster_name = c->homecluster;
556
557 if (warn) {
558 if (c->runstop!= 1) {
559 if (!ask("Continue creating array? ")) {
560 pr_err("create aborted.\n");
561 return 1;
562 }
563 } else {
564 if (c->verbose > 0)
565 pr_err("creation continuing despite oddities due to --run\n");
566 }
567 }
568
569 /* If this is raid4/5, we want to configure the last active slot
570 * as missing, so that a reconstruct happens (faster than re-parity)
571 * FIX: Can we do this for raid6 as well?
572 */
573 if (st->ss->external == 0 && s->assume_clean == 0 &&
574 c->force == 0 && first_missing >= s->raiddisks) {
575 switch (s->level) {
576 case 4:
577 case 5:
578 insert_point = s->raiddisks-1;
579 s->sparedisks++;
580 info.array.active_disks--;
581 missing_disks++;
582 break;
583 default:
584 break;
585 }
586 }
587 /* For raid6, if creating with 1 missing drive, make a good drive
588 * into a spare, else the create will fail
589 */
590 if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks &&
591 st->ss->external == 0 &&
592 second_missing >= s->raiddisks && s->level == 6) {
593 insert_point = s->raiddisks - 1;
594 if (insert_point == first_missing)
595 insert_point--;
596 s->sparedisks ++;
597 info.array.active_disks--;
598 missing_disks++;
599 }
600
601 if (s->level <= 0 && first_missing < subdevs * 2) {
602 pr_err("This level does not support missing devices\n");
603 return 1;
604 }
605
606 /* We need to create the device */
607 map_lock(&map);
608 mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name);
609 if (mdfd < 0) {
610 map_unlock(&map);
611 return 1;
612 }
613 /* verify if chosen_name is not in use,
614 * it could be in conflict with already existing device
615 * e.g. container, array
616 */
617 if (strncmp(chosen_name, "/dev/md/", 8) == 0 &&
618 map_by_name(&map, chosen_name+8) != NULL) {
619 pr_err("Array name %s is in use already.\n",
620 chosen_name);
621 close(mdfd);
622 map_unlock(&map);
623 return 1;
624 }
625 mddev = chosen_name;
626
627 memset(&inf, 0, sizeof(inf));
628 md_get_array_info(mdfd, &inf);
629 if (inf.working_disks != 0) {
630 pr_err("another array by this name is already running.\n");
631 goto abort_locked;
632 }
633
634 /* Ok, lets try some ioctls */
635
636 info.array.level = s->level;
637 info.array.size = s->size;
638 info.array.raid_disks = s->raiddisks;
639 /* The kernel should *know* what md_minor we are dealing
640 * with, but it chooses to trust me instead. Sigh
641 */
642 info.array.md_minor = 0;
643 if (fstat(mdfd, &stb) == 0)
644 info.array.md_minor = minor(stb.st_rdev);
645 info.array.not_persistent = 0;
646
647 if (((s->level == 4 || s->level == 5) &&
648 (insert_point < s->raiddisks || first_missing < s->raiddisks)) ||
649 (s->level == 6 && (insert_point < s->raiddisks ||
650 second_missing < s->raiddisks)) ||
651 (s->level <= 0) || s->assume_clean) {
652 info.array.state = 1; /* clean, but one+ drive will be missing*/
653 info.resync_start = MaxSector;
654 } else {
655 info.array.state = 0; /* not clean, but no errors */
656 info.resync_start = 0;
657 }
658 if (s->level == 10) {
659 /* for raid10, the bitmap size is the capacity of the array,
660 * which is array.size * raid_disks / ncopies;
661 * .. but convert to sectors.
662 */
663 int ncopies = ((s->layout>>8) & 255) * (s->layout & 255);
664 bitmapsize = s->size * s->raiddisks / ncopies * 2;
665 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
666 } else
667 bitmapsize = s->size * 2;
668
669 /* There is lots of redundancy in these disk counts,
670 * raid_disks is the most meaningful value
671 * it describes the geometry of the array
672 * it is constant
673 * nr_disks is total number of used slots.
674 * it should be raid_disks+spare_disks
675 * spare_disks is the number of extra disks present
676 * see above
677 * active_disks is the number of working disks in
678 * active slots. (With raid_disks)
679 * working_disks is the total number of working disks,
680 * including spares
681 * failed_disks is the number of disks marked failed
682 *
683 * Ideally, the kernel would keep these (except raid_disks)
684 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
685 * So for now, we assume that all raid and spare
686 * devices will be given.
687 */
688 info.array.spare_disks=s->sparedisks;
689 info.array.failed_disks=missing_disks;
690 info.array.nr_disks = info.array.working_disks
691 + info.array.failed_disks;
692 info.array.layout = s->layout;
693 info.array.chunk_size = s->chunk*1024;
694
695 if (name == NULL || *name == 0) {
696 /* base name on mddev */
697 /* /dev/md0 -> 0
698 * /dev/md_d0 -> d0
699 * /dev/md_foo -> foo
700 * /dev/md/1 -> 1
701 * /dev/md/d1 -> d1
702 * /dev/md/home -> home
703 * /dev/mdhome -> home
704 */
705 /* FIXME compare this with rules in create_mddev */
706 name = strrchr(mddev, '/');
707 if (name) {
708 name++;
709 if (strncmp(name, "md_", 3) == 0 &&
710 strlen(name) > 3 && (name-mddev) == 5 /* /dev/ */)
711 name += 3;
712 else if (strncmp(name, "md", 2) == 0 &&
713 strlen(name) > 2 && isdigit(name[2]) &&
714 (name-mddev) == 5 /* /dev/ */)
715 name += 2;
716 }
717 }
718 if (!st->ss->init_super(st, &info.array, s, name, c->homehost, uuid,
719 data_offset))
720 goto abort_locked;
721
722 total_slots = info.array.nr_disks;
723 st->ss->getinfo_super(st, &info, NULL);
724 if (sysfs_init(&info, mdfd, NULL)) {
725 pr_err("unable to initialize sysfs\n");
726 goto abort_locked;
727 }
728
729 if (did_default && c->verbose >= 0) {
730 if (is_subarray(info.text_version)) {
731 char devnm[32];
732 char *ep;
733 struct mdinfo *mdi;
734
735 strncpy(devnm, info.text_version+1, 32);
736 devnm[31] = 0;
737 ep = strchr(devnm, '/');
738 if (ep)
739 *ep = 0;
740
741 mdi = sysfs_read(-1, devnm, GET_VERSION);
742
743 pr_err("Creating array inside %s container %s\n",
744 mdi?mdi->text_version:"managed", devnm);
745 sysfs_free(mdi);
746 } else
747 pr_err("Defaulting to version %s metadata\n", info.text_version);
748 }
749
750 map_update(&map, fd2devnm(mdfd), info.text_version,
751 info.uuid, chosen_name);
752 /* Keep map locked until devices have been added to array
753 * to stop another mdadm from finding and using those devices.
754 */
755
756 if (s->bitmap_file && (strcmp(s->bitmap_file, "internal") == 0 ||
757 strcmp(s->bitmap_file, "clustered") == 0)) {
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_locked;
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_locked;
768 }
769 s->bitmap_file = NULL;
770 }
771
772 if (sysfs_init(&info, mdfd, NULL)) {
773 pr_err("unable to initialize sysfs\n");
774 goto abort_locked;
775 }
776
777 if (st->ss->external && st->container_devnm[0]) {
778 /* member */
779
780 /* When creating a member, we need to be careful
781 * to negotiate with mdmon properly.
782 * If it is already running, we cannot write to
783 * the devices and must ask it to do that part.
784 * If it isn't running, we write to the devices,
785 * and then start it.
786 * We hold an exclusive open on the container
787 * device to make sure mdmon doesn't exit after
788 * we checked that it is running.
789 *
790 * For now, fail if it is already running.
791 */
792 container_fd = open_dev_excl(st->container_devnm);
793 if (container_fd < 0) {
794 pr_err("Cannot get exclusive open on container - weird.\n");
795 goto abort_locked;
796 }
797 if (mdmon_running(st->container_devnm)) {
798 if (c->verbose)
799 pr_err("reusing mdmon for %s.\n",
800 st->container_devnm);
801 st->update_tail = &st->updates;
802 } else
803 need_mdmon = 1;
804 }
805 rv = set_array_info(mdfd, st, &info);
806 if (rv) {
807 pr_err("failed to set array info for %s: %s\n",
808 mddev, strerror(errno));
809 goto abort_locked;
810 }
811
812 if (s->bitmap_file) {
813 int uuid[4];
814
815 st->ss->uuid_from_super(st, uuid);
816 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
817 c->delay, s->write_behind,
818 bitmapsize,
819 major_num)) {
820 goto abort_locked;
821 }
822 bitmap_fd = open(s->bitmap_file, O_RDWR);
823 if (bitmap_fd < 0) {
824 pr_err("weird: %s cannot be openned\n",
825 s->bitmap_file);
826 goto abort_locked;
827 }
828 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
829 pr_err("Cannot set bitmap file for %s: %s\n",
830 mddev, strerror(errno));
831 goto abort_locked;
832 }
833 }
834
835 infos = xmalloc(sizeof(*infos) * total_slots);
836 enable_fds(total_slots);
837 for (pass = 1; pass <= 2; pass++) {
838 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
839
840 for (dnum = 0, raid_disk_num = 0, dv = devlist; dv;
841 dv = (dv->next) ? (dv->next) : moved_disk, dnum++) {
842 int fd;
843 struct stat stb2;
844 struct mdinfo *inf = &infos[dnum];
845
846 if (dnum >= total_slots)
847 abort();
848 if (dnum == insert_point) {
849 raid_disk_num += 1;
850 moved_disk = dv;
851 continue;
852 }
853 if (strcasecmp(dv->devname, "missing") == 0) {
854 raid_disk_num += 1;
855 continue;
856 }
857 if (have_container)
858 moved_disk = NULL;
859 if (have_container && dnum < info.array.raid_disks - 1)
860 /* repeatedly use the container */
861 moved_disk = dv;
862
863 switch(pass) {
864 case 1:
865 *inf = info;
866
867 inf->disk.number = dnum;
868 inf->disk.raid_disk = raid_disk_num++;
869
870 if (dv->disposition == 'j') {
871 inf->disk.raid_disk = MD_DISK_ROLE_JOURNAL;
872 inf->disk.state = (1<<MD_DISK_JOURNAL);
873 raid_disk_num--;
874 } else if (inf->disk.raid_disk < s->raiddisks)
875 inf->disk.state = (1<<MD_DISK_ACTIVE) |
876 (1<<MD_DISK_SYNC);
877 else
878 inf->disk.state = 0;
879
880 if (dv->writemostly == FlagSet)
881 inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
882 if (dv->failfast == FlagSet)
883 inf->disk.state |= (1<<MD_DISK_FAILFAST);
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, &stb2);
900 inf->disk.major = major(stb2.st_rdev);
901 inf->disk.minor = minor(stb2.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(stb2.st_rdev);
923 inf->disk.minor = minor(stb2.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.