]> 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
super1: PPL support
[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 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 mdu_array_info_t inf;
154 int fd;
155
156 memset(&inf, 0, sizeof(inf));
157 fd = open(devlist->devname, O_RDONLY);
158 if (fd >= 0 &&
159 ioctl(fd, GET_ARRAY_INFO, &inf) == 0 &&
160 inf.raid_disks == 0) {
161 /* yep, looks like a container */
162 if (st) {
163 rv = st->ss->load_container(st, fd,
164 devlist->devname);
165 if (rv == 0)
166 have_container = 1;
167 } else {
168 st = super_by_fd(fd, NULL);
169 if (st && !(rv = st->ss->
170 load_container(st, fd,
171 devlist->devname)))
172 have_container = 1;
173 else
174 st = NULL;
175 }
176 if (have_container) {
177 subdevs = s->raiddisks;
178 first_missing = subdevs * 2;
179 second_missing = subdevs * 2;
180 insert_point = subdevs * 2;
181 }
182 }
183 if (fd >= 0)
184 close(fd);
185 }
186 if (st && st->ss->external && s->sparedisks) {
187 pr_err("This metadata type does not support spare disks at create time\n");
188 return 1;
189 }
190 if (subdevs > s->raiddisks+s->sparedisks+s->journaldisks) {
191 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks);
192 return 1;
193 }
194 if (!have_container && subdevs < s->raiddisks+s->sparedisks+s->journaldisks) {
195 pr_err("You haven't given enough devices (real or missing) to create this array\n");
196 return 1;
197 }
198 if (s->bitmap_file && s->level <= 0) {
199 pr_err("bitmaps not meaningful with level %s\n",
200 map_num(pers, s->level)?:"given");
201 return 1;
202 }
203
204 /* now set some defaults */
205
206 if (s->layout == UnSet) {
207 do_default_layout = 1;
208 s->layout = default_layout(st, s->level, c->verbose);
209 }
210
211 if (s->level == 10)
212 /* check layout fits in array*/
213 if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) {
214 pr_err("that layout requires at least %d devices\n",
215 (s->layout&255) * ((s->layout>>8)&255));
216 return 1;
217 }
218
219 switch(s->level) {
220 case 4:
221 case 5:
222 case 10:
223 case 6:
224 case 0:
225 if (s->chunk == 0 || s->chunk == UnSet) {
226 s->chunk = UnSet;
227 do_default_chunk = 1;
228 /* chunk will be set later */
229 }
230 break;
231 case LEVEL_LINEAR:
232 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
233 if (get_linux_version() < 2006016 && s->chunk == 0) {
234 s->chunk = 64;
235 if (c->verbose > 0)
236 pr_err("chunk size defaults to 64K\n");
237 }
238 break;
239 case 1:
240 case LEVEL_FAULTY:
241 case LEVEL_MULTIPATH:
242 case LEVEL_CONTAINER:
243 if (s->chunk) {
244 s->chunk = 0;
245 if (c->verbose > 0)
246 pr_err("chunk size ignored for this level\n");
247 }
248 break;
249 default:
250 pr_err("unknown level %d\n", s->level);
251 return 1;
252 }
253 if (s->size == MAX_SIZE)
254 /* use '0' to mean 'max' now... */
255 s->size = 0;
256 if (s->size && s->chunk && s->chunk != UnSet)
257 s->size &= ~(unsigned long long)(s->chunk - 1);
258 newsize = s->size * 2;
259 if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks,
260 &s->chunk, s->size*2,
261 data_offset, NULL,
262 &newsize, s->consistency_policy,
263 c->verbose>=0))
264 return 1;
265
266 if (s->chunk && s->chunk != UnSet) {
267 newsize &= ~(unsigned long long)(s->chunk*2 - 1);
268 if (do_default_chunk) {
269 /* default chunk was just set */
270 if (c->verbose > 0)
271 pr_err("chunk size defaults to %dK\n", s->chunk);
272 s->size &= ~(unsigned long long)(s->chunk - 1);
273 do_default_chunk = 0;
274 }
275 }
276
277 if (s->size == 0) {
278 s->size = newsize / 2;
279 if (s->level == 1)
280 /* If this is ever reshaped to RAID5, we will
281 * need a chunksize. So round it off a bit
282 * now just to be safe
283 */
284 s->size &= ~(64ULL-1);
285
286 if (s->size && c->verbose > 0)
287 pr_err("setting size to %lluK\n", s->size);
288 }
289
290 /* now look at the subdevs */
291 info.array.active_disks = 0;
292 info.array.working_disks = 0;
293 dnum = 0;
294 for (dv = devlist; dv ; dv = dv->next)
295 if (data_offset == VARIABLE_OFFSET)
296 dv->data_offset = INVALID_SECTORS;
297 else
298 dv->data_offset = data_offset;
299
300 for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) {
301 char *dname = dv->devname;
302 unsigned long long freesize;
303 int dfd;
304 char *doff;
305
306 if (strcasecmp(dname, "missing")==0) {
307 if (first_missing > dnum)
308 first_missing = dnum;
309 if (second_missing > dnum && dnum > first_missing)
310 second_missing = dnum;
311 missing_disks ++;
312 continue;
313 }
314 if (data_offset == VARIABLE_OFFSET) {
315 doff = strchr(dname, ':');
316 if (doff) {
317 *doff++ = 0;
318 dv->data_offset = parse_size(doff);
319 } else
320 dv->data_offset = INVALID_SECTORS;
321 } else
322 dv->data_offset = data_offset;
323
324 dfd = open(dname, O_RDONLY);
325 if (dfd < 0) {
326 pr_err("cannot open %s: %s\n",
327 dname, strerror(errno));
328 exit(2);
329 }
330 if (fstat(dfd, &stb) != 0 ||
331 (stb.st_mode & S_IFMT) != S_IFBLK) {
332 close(dfd);
333 pr_err("%s is not a block device\n",
334 dname);
335 exit(2);
336 }
337 close(dfd);
338 info.array.working_disks++;
339 if (dnum < s->raiddisks && dv->disposition != 'j')
340 info.array.active_disks++;
341 if (st == NULL) {
342 struct createinfo *ci = conf_get_create_info();
343 if (ci)
344 st = ci->supertype;
345 }
346 if (st == NULL) {
347 /* Need to choose a default metadata, which is different
348 * depending on geometry of array.
349 */
350 int i;
351 char *name = "default";
352 for(i=0; !st && superlist[i]; i++) {
353 st = superlist[i]->match_metadata_desc(name);
354 if (!st)
355 continue;
356 if (do_default_layout)
357 s->layout = default_layout(st, s->level, c->verbose);
358 switch (st->ss->validate_geometry(
359 st, s->level, s->layout, s->raiddisks,
360 &s->chunk, s->size*2,
361 dv->data_offset, dname,
362 &freesize, s->consistency_policy,
363 c->verbose > 0)) {
364 case -1: /* Not valid, message printed, and not
365 * worth checking any further */
366 exit(2);
367 break;
368 case 0: /* Geometry not valid */
369 free(st);
370 st = NULL;
371 s->chunk = do_default_chunk ? UnSet : s->chunk;
372 break;
373 case 1: /* All happy */
374 break;
375 }
376 }
377
378 if (!st) {
379 int dfd = open(dname, O_RDONLY|O_EXCL);
380 if (dfd < 0) {
381 pr_err("cannot open %s: %s\n",
382 dname, strerror(errno));
383 exit(2);
384 }
385 pr_err("device %s not suitable for any style of array\n",
386 dname);
387 exit(2);
388 }
389 if (st->ss != &super0 ||
390 st->minor_version != 90)
391 did_default = 1;
392 } else {
393 if (do_default_layout)
394 s->layout = default_layout(st, s->level, 0);
395 if (!st->ss->validate_geometry(st, s->level, s->layout,
396 s->raiddisks,
397 &s->chunk, s->size*2,
398 dv->data_offset,
399 dname, &freesize,
400 s->consistency_policy,
401 c->verbose >= 0)) {
402
403 pr_err("%s is not suitable for this array.\n",
404 dname);
405 fail = 1;
406 continue;
407 }
408 }
409
410 if (dv->disposition == 'j')
411 goto skip_size_check; /* skip write journal for size check */
412
413 freesize /= 2; /* convert to K */
414 if (s->chunk && s->chunk != UnSet) {
415 /* round to chunk size */
416 freesize = freesize & ~(s->chunk-1);
417 if (do_default_chunk) {
418 /* default chunk was just set */
419 if (c->verbose > 0)
420 pr_err("chunk size defaults to %dK\n", s->chunk);
421 s->size &= ~(unsigned long long)(s->chunk - 1);
422 do_default_chunk = 0;
423 }
424 }
425 if (!freesize) {
426 pr_err("no free space left on %s\n", dname);
427 fail = 1;
428 continue;
429 }
430
431 if (s->size && freesize < s->size) {
432 pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
433 dname, freesize, s->size);
434 fail = 1;
435 continue;
436 }
437 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
438 maxdisc = dname;
439 maxsize = freesize;
440 }
441 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
442 mindisc = dname;
443 minsize = freesize;
444 }
445 skip_size_check:
446 if (c->runstop != 1 || c->verbose >= 0) {
447 int fd = open(dname, O_RDONLY);
448 if (fd <0 ) {
449 pr_err("Cannot open %s: %s\n",
450 dname, strerror(errno));
451 fail=1;
452 continue;
453 }
454 warn |= check_ext2(fd, dname);
455 warn |= check_reiser(fd, dname);
456 warn |= check_raid(fd, dname);
457 if (strcmp(st->ss->name, "1.x") == 0 &&
458 st->minor_version >= 1)
459 /* metadata at front */
460 warn |= check_partitions(fd, dname, 0, 0);
461 else if (s->level == 1 || s->level == LEVEL_CONTAINER
462 || (s->level == 0 && s->raiddisks == 1))
463 /* partitions could be meaningful */
464 warn |= check_partitions(fd, dname, freesize*2, s->size*2);
465 else
466 /* partitions cannot be meaningful */
467 warn |= check_partitions(fd, dname, 0, 0);
468 if (strcmp(st->ss->name, "1.x") == 0 &&
469 st->minor_version >= 1 &&
470 did_default &&
471 s->level == 1 &&
472 (warn & 1024) == 0) {
473 warn |= 1024;
474 pr_err("Note: this array has metadata at the start and\n"
475 " may not be suitable as a boot device. If you plan to\n"
476 " store '/boot' on this device please ensure that\n"
477 " your boot-loader understands md/v1.x metadata, or use\n"
478 " --metadata=0.90\n");
479 }
480 close(fd);
481 }
482 }
483 if (s->raiddisks + s->sparedisks > st->max_devs) {
484 pr_err("Too many devices: %s metadata only supports %d\n",
485 st->ss->name, st->max_devs);
486 return 1;
487 }
488 if (have_container)
489 info.array.working_disks = s->raiddisks;
490 if (fail) {
491 pr_err("create aborted\n");
492 return 1;
493 }
494 if (s->size == 0) {
495 if (mindisc == NULL && !have_container) {
496 pr_err("no size and no drives given - aborting create.\n");
497 return 1;
498 }
499 if (s->level > 0 || s->level == LEVEL_MULTIPATH
500 || s->level == LEVEL_FAULTY
501 || st->ss->external ) {
502 /* size is meaningful */
503 if (!st->ss->validate_geometry(st, s->level, s->layout,
504 s->raiddisks,
505 &s->chunk, minsize*2,
506 data_offset,
507 NULL, NULL,
508 s->consistency_policy, 0)) {
509 pr_err("devices too large for RAID level %d\n", s->level);
510 return 1;
511 }
512 s->size = minsize;
513 if (s->level == 1)
514 /* If this is ever reshaped to RAID5, we will
515 * need a chunksize. So round it off a bit
516 * now just to be safe
517 */
518 s->size &= ~(64ULL-1);
519 if (c->verbose > 0)
520 pr_err("size set to %lluK\n", s->size);
521 }
522 }
523
524 if (!s->bitmap_file &&
525 s->level >= 1 &&
526 st->ss->add_internal_bitmap &&
527 (s->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
528 s->consistency_policy != CONSISTENCY_POLICY_PPL) &&
529 (s->write_behind || s->size > 100*1024*1024ULL)) {
530 if (c->verbose > 0)
531 pr_err("automatically enabling write-intent bitmap on large array\n");
532 s->bitmap_file = "internal";
533 }
534 if (s->bitmap_file && strcmp(s->bitmap_file, "none") == 0)
535 s->bitmap_file = NULL;
536
537 if (s->consistency_policy == CONSISTENCY_POLICY_PPL &&
538 !st->ss->write_init_ppl) {
539 pr_err("%s metadata does not support PPL\n", st->ss->name);
540 return 1;
541 }
542
543 if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) {
544 if (c->runstop != 1 || c->verbose >= 0)
545 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
546 maxdisc, s->size);
547 warn = 1;
548 }
549
550 if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) {
551 if (c->runstop != 1 || c->verbose >= 0)
552 pr_err("%s unable to enumerate platform support\n"
553 " array may not be compatible with hardware/firmware\n",
554 st->ss->name);
555 warn = 1;
556 }
557 st->nodes = c->nodes;
558 st->cluster_name = c->homecluster;
559
560 if (warn) {
561 if (c->runstop!= 1) {
562 if (!ask("Continue creating array? ")) {
563 pr_err("create aborted.\n");
564 return 1;
565 }
566 } else {
567 if (c->verbose > 0)
568 pr_err("creation continuing despite oddities due to --run\n");
569 }
570 }
571
572 /* If this is raid4/5, we want to configure the last active slot
573 * as missing, so that a reconstruct happens (faster than re-parity)
574 * FIX: Can we do this for raid6 as well?
575 */
576 if (st->ss->external == 0 &&
577 s->assume_clean==0 && c->force == 0 && first_missing >= s->raiddisks) {
578 switch ( s->level ) {
579 case 4:
580 case 5:
581 insert_point = s->raiddisks-1;
582 s->sparedisks++;
583 info.array.active_disks--;
584 missing_disks++;
585 break;
586 default:
587 break;
588 }
589 }
590 /* For raid6, if creating with 1 missing drive, make a good drive
591 * into a spare, else the create will fail
592 */
593 if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks &&
594 st->ss->external == 0 &&
595 second_missing >= s->raiddisks && s->level == 6) {
596 insert_point = s->raiddisks - 1;
597 if (insert_point == first_missing)
598 insert_point--;
599 s->sparedisks ++;
600 info.array.active_disks--;
601 missing_disks++;
602 }
603
604 if (s->level <= 0 && first_missing < subdevs * 2) {
605 pr_err("This level does not support missing devices\n");
606 return 1;
607 }
608
609 /* We need to create the device */
610 map_lock(&map);
611 mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name);
612 if (mdfd < 0) {
613 map_unlock(&map);
614 return 1;
615 }
616 /* verify if chosen_name is not in use,
617 * it could be in conflict with already existing device
618 * e.g. container, array
619 */
620 if (strncmp(chosen_name, "/dev/md/", 8) == 0
621 && map_by_name(&map, chosen_name+8) != NULL) {
622 pr_err("Array name %s is in use already.\n",
623 chosen_name);
624 close(mdfd);
625 map_unlock(&map);
626 return 1;
627 }
628 mddev = chosen_name;
629
630 vers = md_get_version(mdfd);
631 if (vers < 9000) {
632 pr_err("Create requires md driver version 0.90.0 or later\n");
633 goto abort_locked;
634 } else {
635 mdu_array_info_t inf;
636 memset(&inf, 0, sizeof(inf));
637 ioctl(mdfd, GET_ARRAY_INFO, &inf);
638 if (inf.working_disks != 0) {
639 pr_err("another array by this name is already running.\n");
640 goto abort_locked;
641 }
642 }
643
644 /* Ok, lets try some ioctls */
645
646 info.array.level = s->level;
647 info.array.size = s->size;
648 info.array.raid_disks = s->raiddisks;
649 /* The kernel should *know* what md_minor we are dealing
650 * with, but it chooses to trust me instead. Sigh
651 */
652 info.array.md_minor = 0;
653 if (fstat(mdfd, &stb)==0)
654 info.array.md_minor = minor(stb.st_rdev);
655 info.array.not_persistent = 0;
656
657 if ( ( (s->level == 4 || s->level == 5) &&
658 (insert_point < s->raiddisks || first_missing < s->raiddisks) )
659 ||
660 ( s->level == 6 && (insert_point < s->raiddisks
661 || second_missing < s->raiddisks))
662 ||
663 ( s->level <= 0 )
664 ||
665 s->assume_clean
666 ) {
667 info.array.state = 1; /* clean, but one+ drive will be missing*/
668 info.resync_start = MaxSector;
669 } else {
670 info.array.state = 0; /* not clean, but no errors */
671 info.resync_start = 0;
672 }
673 if (s->level == 10) {
674 /* for raid10, the bitmap size is the capacity of the array,
675 * which is array.size * raid_disks / ncopies;
676 * .. but convert to sectors.
677 */
678 int ncopies = ((s->layout>>8) & 255) * (s->layout & 255);
679 bitmapsize = s->size * s->raiddisks / ncopies * 2;
680 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
681 } else
682 bitmapsize = s->size * 2;
683
684 /* There is lots of redundancy in these disk counts,
685 * raid_disks is the most meaningful value
686 * it describes the geometry of the array
687 * it is constant
688 * nr_disks is total number of used slots.
689 * it should be raid_disks+spare_disks
690 * spare_disks is the number of extra disks present
691 * see above
692 * active_disks is the number of working disks in
693 * active slots. (With raid_disks)
694 * working_disks is the total number of working disks,
695 * including spares
696 * failed_disks is the number of disks marked failed
697 *
698 * Ideally, the kernel would keep these (except raid_disks)
699 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
700 * So for now, we assume that all raid and spare
701 * devices will be given.
702 */
703 info.array.spare_disks=s->sparedisks;
704 info.array.failed_disks=missing_disks;
705 info.array.nr_disks = info.array.working_disks
706 + info.array.failed_disks;
707 info.array.layout = s->layout;
708 info.array.chunk_size = s->chunk*1024;
709
710 if (name == NULL || *name == 0) {
711 /* base name on mddev */
712 /* /dev/md0 -> 0
713 * /dev/md_d0 -> d0
714 * /dev/md_foo -> foo
715 * /dev/md/1 -> 1
716 * /dev/md/d1 -> d1
717 * /dev/md/home -> home
718 * /dev/mdhome -> home
719 */
720 /* FIXME compare this with rules in create_mddev */
721 name = strrchr(mddev, '/');
722 if (name) {
723 name++;
724 if (strncmp(name, "md_", 3)==0 &&
725 strlen(name) > 3 &&
726 (name-mddev) == 5 /* /dev/ */)
727 name += 3;
728 else if (strncmp(name, "md", 2)==0 &&
729 strlen(name) > 2 &&
730 isdigit(name[2]) &&
731 (name-mddev) == 5 /* /dev/ */)
732 name += 2;
733 }
734 }
735 if (!st->ss->init_super(st, &info.array, s, name, c->homehost, uuid,
736 data_offset))
737 goto abort_locked;
738
739 total_slots = info.array.nr_disks;
740 st->ss->getinfo_super(st, &info, NULL);
741 sysfs_init(&info, mdfd, NULL);
742
743 if (did_default && c->verbose >= 0) {
744 if (is_subarray(info.text_version)) {
745 char devnm[32];
746 char *ep;
747 struct mdinfo *mdi;
748
749 strncpy(devnm, info.text_version+1, 32);
750 devnm[31] = 0;
751 ep = strchr(devnm, '/');
752 if (ep)
753 *ep = 0;
754
755 mdi = sysfs_read(-1, devnm, GET_VERSION);
756
757 pr_err("Creating array inside %s container %s\n",
758 mdi?mdi->text_version:"managed", devnm);
759 sysfs_free(mdi);
760 } else
761 pr_err("Defaulting to version %s metadata\n", info.text_version);
762 }
763
764 map_update(&map, fd2devnm(mdfd), info.text_version,
765 info.uuid, chosen_name);
766 /* Keep map locked until devices have been added to array
767 * to stop another mdadm from finding and using those devices.
768 */
769
770 if (s->bitmap_file && vers < 9003) {
771 major_num = BITMAP_MAJOR_HOSTENDIAN;
772 #ifdef __BIG_ENDIAN
773 pr_err("Warning - bitmaps created on this kernel are not portable\n"
774 " between different architectured. Consider upgrading the Linux kernel.\n");
775 #endif
776 }
777
778 if (s->bitmap_file && (strcmp(s->bitmap_file, "internal")==0 ||
779 strcmp(s->bitmap_file, "clustered")==0)) {
780 if ((vers%100) < 2) {
781 pr_err("internal bitmaps not supported by this kernel.\n");
782 goto abort_locked;
783 }
784 if (!st->ss->add_internal_bitmap) {
785 pr_err("internal bitmaps not supported with %s metadata\n",
786 st->ss->name);
787 goto abort_locked;
788 }
789 if (st->ss->add_internal_bitmap(st, &s->bitmap_chunk,
790 c->delay, s->write_behind,
791 bitmapsize, 1, major_num)) {
792 pr_err("Given bitmap chunk size not supported.\n");
793 goto abort_locked;
794 }
795 s->bitmap_file = NULL;
796 }
797
798 sysfs_init(&info, mdfd, NULL);
799
800 if (st->ss->external && st->container_devnm[0]) {
801 /* member */
802
803 /* When creating a member, we need to be careful
804 * to negotiate with mdmon properly.
805 * If it is already running, we cannot write to
806 * the devices and must ask it to do that part.
807 * If it isn't running, we write to the devices,
808 * and then start it.
809 * We hold an exclusive open on the container
810 * device to make sure mdmon doesn't exit after
811 * we checked that it is running.
812 *
813 * For now, fail if it is already running.
814 */
815 container_fd = open_dev_excl(st->container_devnm);
816 if (container_fd < 0) {
817 pr_err("Cannot get exclusive open on container - weird.\n");
818 goto abort_locked;
819 }
820 if (mdmon_running(st->container_devnm)) {
821 if (c->verbose)
822 pr_err("reusing mdmon for %s.\n",
823 st->container_devnm);
824 st->update_tail = &st->updates;
825 } else
826 need_mdmon = 1;
827 }
828 rv = set_array_info(mdfd, st, &info);
829 if (rv) {
830 pr_err("failed to set array info for %s: %s\n",
831 mddev, strerror(errno));
832 goto abort_locked;
833 }
834
835 if (s->bitmap_file) {
836 int uuid[4];
837
838 st->ss->uuid_from_super(st, uuid);
839 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
840 c->delay, s->write_behind,
841 bitmapsize,
842 major_num)) {
843 goto abort_locked;
844 }
845 bitmap_fd = open(s->bitmap_file, O_RDWR);
846 if (bitmap_fd < 0) {
847 pr_err("weird: %s cannot be openned\n",
848 s->bitmap_file);
849 goto abort_locked;
850 }
851 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
852 pr_err("Cannot set bitmap file for %s: %s\n",
853 mddev, strerror(errno));
854 goto abort_locked;
855 }
856 }
857
858 infos = xmalloc(sizeof(*infos) * total_slots);
859 enable_fds(total_slots);
860 for (pass=1; pass <=2 ; pass++) {
861 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
862
863 for (dnum=0, raid_disk_num=0, dv = devlist ; dv ;
864 dv=(dv->next)?(dv->next):moved_disk, dnum++) {
865 int fd;
866 struct stat stb;
867 struct mdinfo *inf = &infos[dnum];
868
869 if (dnum >= total_slots)
870 abort();
871 if (dnum == insert_point) {
872 raid_disk_num += 1;
873 moved_disk = dv;
874 continue;
875 }
876 if (strcasecmp(dv->devname, "missing")==0) {
877 raid_disk_num += 1;
878 continue;
879 }
880 if (have_container)
881 moved_disk = NULL;
882 if (have_container && dnum < info.array.raid_disks - 1)
883 /* repeatedly use the container */
884 moved_disk = dv;
885
886 switch(pass) {
887 case 1:
888 *inf = info;
889
890 inf->disk.number = dnum;
891 inf->disk.raid_disk = raid_disk_num++;
892
893 if (dv->disposition == 'j') {
894 inf->disk.raid_disk = MD_DISK_ROLE_JOURNAL;
895 inf->disk.state = (1<<MD_DISK_JOURNAL);
896 raid_disk_num--;
897 } else if (inf->disk.raid_disk < s->raiddisks)
898 inf->disk.state = (1<<MD_DISK_ACTIVE) |
899 (1<<MD_DISK_SYNC);
900 else
901 inf->disk.state = 0;
902
903 if (dv->writemostly == FlagSet)
904 inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
905 if (dv->failfast == FlagSet)
906 inf->disk.state |= (1<<MD_DISK_FAILFAST);
907
908 if (have_container)
909 fd = -1;
910 else {
911 if (st->ss->external &&
912 st->container_devnm[0])
913 fd = open(dv->devname, O_RDWR);
914 else
915 fd = open(dv->devname, O_RDWR|O_EXCL);
916
917 if (fd < 0) {
918 pr_err("failed to open %s after earlier success - aborting\n",
919 dv->devname);
920 goto abort_locked;
921 }
922 fstat(fd, &stb);
923 inf->disk.major = major(stb.st_rdev);
924 inf->disk.minor = minor(stb.st_rdev);
925 }
926 if (fd >= 0)
927 remove_partitions(fd);
928 if (st->ss->add_to_super(st, &inf->disk,
929 fd, dv->devname,
930 dv->data_offset)) {
931 ioctl(mdfd, STOP_ARRAY, NULL);
932 goto abort_locked;
933 }
934 st->ss->getinfo_super(st, inf, NULL);
935 safe_mode_delay = inf->safe_mode_delay;
936
937 if (have_container && c->verbose > 0)
938 pr_err("Using %s for device %d\n",
939 map_dev(inf->disk.major,
940 inf->disk.minor,
941 0), dnum);
942
943 if (!have_container) {
944 /* getinfo_super might have lost these ... */
945 inf->disk.major = major(stb.st_rdev);
946 inf->disk.minor = minor(stb.st_rdev);
947 }
948 break;
949 case 2:
950 inf->errors = 0;
951
952 rv = add_disk(mdfd, st, &info, inf);
953
954 if (rv) {
955 pr_err("ADD_NEW_DISK for %s failed: %s\n",
956 dv->devname, strerror(errno));
957 goto abort_locked;
958 }
959 break;
960 }
961 if (!have_container &&
962 dv == moved_disk && dnum != insert_point) break;
963 }
964 if (pass == 1) {
965 struct mdinfo info_new;
966 struct map_ent *me = NULL;
967
968 /* check to see if the uuid has changed due to these
969 * metadata changes, and if so update the member array
970 * and container uuid. Note ->write_init_super clears
971 * the subarray cursor such that ->getinfo_super once
972 * again returns container info.
973 */
974 st->ss->getinfo_super(st, &info_new, NULL);
975 if (st->ss->external && s->level != LEVEL_CONTAINER &&
976 !same_uuid(info_new.uuid, info.uuid, 0)) {
977 map_update(&map, fd2devnm(mdfd),
978 info_new.text_version,
979 info_new.uuid, chosen_name);
980 me = map_by_devnm(&map, st->container_devnm);
981 }
982
983 if (st->ss->write_init_super(st)) {
984 st->ss->free_super(st);
985 goto abort_locked;
986 }
987
988 /* update parent container uuid */
989 if (me) {
990 char *path = xstrdup(me->path);
991
992 st->ss->getinfo_super(st, &info_new, NULL);
993 map_update(&map, st->container_devnm,
994 info_new.text_version,
995 info_new.uuid, path);
996 free(path);
997 }
998
999 flush_metadata_updates(st);
1000 st->ss->free_super(st);
1001 }
1002 }
1003 map_unlock(&map);
1004 free(infos);
1005
1006 if (s->level == LEVEL_CONTAINER) {
1007 /* No need to start. But we should signal udev to
1008 * create links */
1009 sysfs_uevent(&info, "change");
1010 if (c->verbose >= 0)
1011 pr_err("container %s prepared.\n", mddev);
1012 wait_for(chosen_name, mdfd);
1013 } else if (c->runstop == 1 || subdevs >= s->raiddisks) {
1014 if (st->ss->external) {
1015 int err;
1016 switch(s->level) {
1017 case LEVEL_LINEAR:
1018 case LEVEL_MULTIPATH:
1019 case 0:
1020 err = sysfs_set_str(&info, NULL, "array_state",
1021 c->readonly
1022 ? "readonly"
1023 : "active");
1024 need_mdmon = 0;
1025 break;
1026 default:
1027 err = sysfs_set_str(&info, NULL, "array_state",
1028 "readonly");
1029 break;
1030 }
1031 sysfs_set_safemode(&info, safe_mode_delay);
1032 if (err) {
1033 pr_err("failed to activate array.\n");
1034 ioctl(mdfd, STOP_ARRAY, NULL);
1035 goto abort;
1036 }
1037 } else if (c->readonly &&
1038 sysfs_attribute_available(
1039 &info, NULL, "array_state")) {
1040 if (sysfs_set_str(&info, NULL,
1041 "array_state", "readonly") < 0) {
1042 pr_err("Failed to start array: %s\n",
1043 strerror(errno));
1044 ioctl(mdfd, STOP_ARRAY, NULL);
1045 goto abort;
1046 }
1047 } else {
1048 /* param is not actually used */
1049 mdu_param_t param;
1050 if (ioctl(mdfd, RUN_ARRAY, &param)) {
1051 pr_err("RUN_ARRAY failed: %s\n",
1052 strerror(errno));
1053 if (info.array.chunk_size & (info.array.chunk_size-1)) {
1054 cont_err("Problem may be that chunk size is not a power of 2\n");
1055 }
1056 ioctl(mdfd, STOP_ARRAY, NULL);
1057 goto abort;
1058 }
1059 /* if start_ro module parameter is set, array is
1060 * auto-read-only, which is bad as the resync won't
1061 * start. So lets make it read-write now.
1062 */
1063 ioctl(mdfd, RESTART_ARRAY_RW, NULL);
1064 }
1065 if (c->verbose >= 0)
1066 pr_err("array %s started.\n", mddev);
1067 if (st->ss->external && st->container_devnm[0]) {
1068 if (need_mdmon)
1069 start_mdmon(st->container_devnm);
1070
1071 ping_monitor(st->container_devnm);
1072 close(container_fd);
1073 }
1074 wait_for(chosen_name, mdfd);
1075 } else {
1076 pr_err("not starting array - not enough devices.\n");
1077 }
1078 close(mdfd);
1079 return 0;
1080
1081 abort:
1082 map_lock(&map);
1083 abort_locked:
1084 map_remove(&map, fd2devnm(mdfd));
1085 map_unlock(&map);
1086
1087 if (mdfd >= 0)
1088 close(mdfd);
1089 return 1;
1090 }
Unnamed repository; edit this file 'description' to name the repository.