]> 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
Minor cosmetic fixes in various files.
[thirdparty/mdadm.git] / Create.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2012 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24
25 #include "mdadm.h"
26 #include "md_u.h"
27 #include "md_p.h"
28 #include <ctype.h>
29
30 static int default_layout(struct supertype *st, int level, int verbose)
31 {
32 int layout = UnSet;
33
34 if (st && st->ss->default_geometry)
35 st->ss->default_geometry(st, &level, &layout, NULL);
36
37 if (layout == UnSet)
38 switch(level) {
39 default: /* no layout */
40 layout = 0;
41 break;
42 case 10:
43 layout = 0x102; /* near=2, far=1 */
44 if (verbose > 0)
45 pr_err("layout defaults to n2\n");
46 break;
47 case 5:
48 case 6:
49 layout = map_name(r5layout, "default");
50 if (verbose > 0)
51 pr_err("layout defaults to %s\n", map_num(r5layout, layout));
52 break;
53 case LEVEL_FAULTY:
54 layout = map_name(faultylayout, "default");
55
56 if (verbose > 0)
57 pr_err("layout defaults to %s\n", map_num(faultylayout, layout));
58 break;
59 }
60
61 return layout;
62 }
63
64 int Create(struct supertype *st, char *mddev,
65 char *name, int *uuid,
66 int subdevs, struct mddev_dev *devlist,
67 struct shape *s,
68 struct context *c)
69 {
70 /*
71 * Create a new raid array.
72 *
73 * First check that necessary details are available
74 * (i.e. level, raid-disks)
75 *
76 * Then check each disk to see what might be on it
77 * and report anything interesting.
78 *
79 * If anything looks odd, and runstop not set,
80 * abort.
81 *
82 * SET_ARRAY_INFO and ADD_NEW_DISK, and
83 * if runstop==run, or raiddisks disks were used,
84 * RUN_ARRAY
85 */
86 int mdfd;
87 unsigned long long minsize=0, maxsize=0;
88 char *mindisc = NULL;
89 char *maxdisc = NULL;
90 int dnum;
91 struct mddev_dev *dv;
92 int fail=0, warn=0;
93 struct stat stb;
94 int first_missing = subdevs * 2;
95 int second_missing = subdevs * 2;
96 int missing_disks = 0;
97 int insert_point = subdevs * 2; /* where to insert a missing drive */
98 int total_slots;
99 int pass;
100 int vers;
101 int rv;
102 int bitmap_fd;
103 int have_container = 0;
104 int container_fd = -1;
105 int need_mdmon = 0;
106 unsigned long long bitmapsize;
107 struct mdinfo info, *infos;
108 int did_default = 0;
109 int do_default_layout = 0;
110 int do_default_chunk = 0;
111 unsigned long safe_mode_delay = 0;
112 char chosen_name[1024];
113 struct map_ent *map = NULL;
114 unsigned long long newsize;
115
116 int major_num = BITMAP_MAJOR_HI;
117
118 memset(&info, 0, sizeof(info));
119 if (s->level == UnSet && st && st->ss->default_geometry)
120 st->ss->default_geometry(st, &s->level, NULL, NULL);
121 if (s->level == UnSet) {
122 pr_err("a RAID level is needed to create an array.\n");
123 return 1;
124 }
125 if (s->raiddisks < 4 && s->level == 6) {
126 pr_err("at least 4 raid-devices needed for level 6\n");
127 return 1;
128 }
129 if (s->raiddisks > 256 && s->level == 6) {
130 pr_err("no more than 256 raid-devices supported for level 6\n");
131 return 1;
132 }
133 if (s->raiddisks < 2 && s->level >= 4) {
134 pr_err("at least 2 raid-devices needed for level 4 or 5\n");
135 return 1;
136 }
137 if (s->level <= 0 && s->sparedisks) {
138 pr_err("This level does not support spare devices\n");
139 return 1;
140 }
141
142 if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
143 /* If given a single device, it might be a container, and we can
144 * extract a device list from there
145 */
146 mdu_array_info_t inf;
147 int fd;
148
149 memset(&inf, 0, sizeof(inf));
150 fd = open(devlist->devname, O_RDONLY);
151 if (fd >= 0 &&
152 ioctl(fd, GET_ARRAY_INFO, &inf) == 0 &&
153 inf.raid_disks == 0) {
154 /* yep, looks like a container */
155 if (st) {
156 rv = st->ss->load_container(st, fd,
157 devlist->devname);
158 if (rv == 0)
159 have_container = 1;
160 } else {
161 st = super_by_fd(fd, NULL);
162 if (st && !(rv = st->ss->
163 load_container(st, fd,
164 devlist->devname)))
165 have_container = 1;
166 else
167 st = NULL;
168 }
169 if (have_container) {
170 subdevs = s->raiddisks;
171 first_missing = subdevs * 2;
172 second_missing = subdevs * 2;
173 insert_point = subdevs * 2;
174 }
175 }
176 if (fd >= 0)
177 close(fd);
178 }
179 if (st && st->ss->external && s->sparedisks) {
180 pr_err("This metadata type does not support "
181 "spare disks at create time\n");
182 return 1;
183 }
184 if (subdevs > s->raiddisks+s->sparedisks) {
185 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks);
186 return 1;
187 }
188 if (!have_container && subdevs < s->raiddisks+s->sparedisks) {
189 pr_err("You haven't given enough devices (real or missing) to create this array\n");
190 return 1;
191 }
192 if (s->bitmap_file && s->level <= 0) {
193 pr_err("bitmaps not meaningful with level %s\n",
194 map_num(pers, s->level)?:"given");
195 return 1;
196 }
197
198 /* now set some defaults */
199
200 if (s->layout == UnSet) {
201 do_default_layout = 1;
202 s->layout = default_layout(st, s->level, c->verbose);
203 }
204
205 if (s->level == 10)
206 /* check layout fits in array*/
207 if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) {
208 pr_err("that layout requires at least %d devices\n",
209 (s->layout&255) * ((s->layout>>8)&255));
210 return 1;
211 }
212
213 switch(s->level) {
214 case 4:
215 case 5:
216 case 10:
217 case 6:
218 case 0:
219 if (s->chunk == 0 || s->chunk == UnSet) {
220 s->chunk = UnSet;
221 do_default_chunk = 1;
222 /* chunk will be set later */
223 }
224 break;
225 case LEVEL_LINEAR:
226 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
227 if (get_linux_version() < 2006016 && s->chunk == 0) {
228 s->chunk = 64;
229 if (c->verbose > 0)
230 pr_err("chunk size defaults to 64K\n");
231 }
232 break;
233 case 1:
234 case LEVEL_FAULTY:
235 case LEVEL_MULTIPATH:
236 case LEVEL_CONTAINER:
237 if (s->chunk) {
238 s->chunk = 0;
239 if (c->verbose > 0)
240 pr_err("chunk size ignored for this level\n");
241 }
242 break;
243 default:
244 pr_err("unknown level %d\n", s->level);
245 return 1;
246 }
247 if (s->size == MAX_SIZE)
248 /* use '0' to mean 'max' now... */
249 s->size = 0;
250 if (s->size && s->chunk && s->chunk != UnSet)
251 s->size &= ~(unsigned long long)(s->chunk - 1);
252 newsize = s->size * 2;
253 if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks,
254 &s->chunk, s->size*2, NULL, &newsize, c->verbose>=0))
255 return 1;
256
257 if (s->chunk && s->chunk != UnSet) {
258 newsize &= ~(unsigned long long)(s->chunk*2 - 1);
259 if (do_default_chunk) {
260 /* default chunk was just set */
261 if (c->verbose > 0)
262 pr_err("chunk size "
263 "defaults to %dK\n", s->chunk);
264 s->size &= ~(unsigned long long)(s->chunk - 1);
265 do_default_chunk = 0;
266 }
267 }
268
269 if (s->size == 0) {
270 s->size = newsize / 2;
271 if (s->level == 1)
272 /* If this is ever reshaped to RAID5, we will
273 * need a chunksize. So round it off a bit
274 * now just to be safe
275 */
276 s->size &= ~(64ULL-1);
277
278 if (s->size && c->verbose > 0)
279 pr_err("setting size to %lluK\n", s->size);
280 }
281
282 /* now look at the subdevs */
283 info.array.active_disks = 0;
284 info.array.working_disks = 0;
285 dnum = 0;
286 for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) {
287 char *dname = dv->devname;
288 unsigned long long freesize;
289 int dfd;
290
291 if (strcasecmp(dname, "missing")==0) {
292 if (first_missing > dnum)
293 first_missing = dnum;
294 if (second_missing > dnum && dnum > first_missing)
295 second_missing = dnum;
296 missing_disks ++;
297 continue;
298 }
299 dfd = open(dname, O_RDONLY);
300 if (dfd < 0) {
301 pr_err("cannot open %s: %s\n",
302 dname, strerror(errno));
303 exit(2);
304 }
305 if (fstat(dfd, &stb) != 0 ||
306 (stb.st_mode & S_IFMT) != S_IFBLK) {
307 close(dfd);
308 pr_err("%s is not a block device\n",
309 dname);
310 exit(2);
311 }
312 close(dfd);
313 info.array.working_disks++;
314 if (dnum < s->raiddisks)
315 info.array.active_disks++;
316 if (st == NULL) {
317 struct createinfo *ci = conf_get_create_info();
318 if (ci)
319 st = ci->supertype;
320 }
321 if (st == NULL) {
322 /* Need to choose a default metadata, which is different
323 * depending on geometry of array.
324 */
325 int i;
326 char *name = "default";
327 for(i=0; !st && superlist[i]; i++) {
328 st = superlist[i]->match_metadata_desc(name);
329 if (!st)
330 continue;
331 if (do_default_layout)
332 s->layout = default_layout(st, s->level, c->verbose);
333 switch (st->ss->validate_geometry(
334 st, s->level, s->layout, s->raiddisks,
335 &s->chunk, s->size*2, dname, &freesize,
336 c->verbose > 0)) {
337 case -1: /* Not valid, message printed, and not
338 * worth checking any further */
339 exit(2);
340 break;
341 case 0: /* Geometry not valid */
342 free(st);
343 st = NULL;
344 s->chunk = do_default_chunk ? UnSet : s->chunk;
345 break;
346 case 1: /* All happy */
347 break;
348 }
349 }
350
351 if (!st) {
352 int dfd = open(dname, O_RDONLY|O_EXCL);
353 if (dfd < 0) {
354 pr_err("cannot open %s: %s\n",
355 dname, strerror(errno));
356 exit(2);
357 }
358 pr_err("device %s not suitable "
359 "for any style of array\n",
360 dname);
361 exit(2);
362 }
363 if (st->ss != &super0 ||
364 st->minor_version != 90)
365 did_default = 1;
366 } else {
367 if (do_default_layout)
368 s->layout = default_layout(st, s->level, 0);
369 if (!st->ss->validate_geometry(st, s->level, s->layout,
370 s->raiddisks,
371 &s->chunk, s->size*2, dname,
372 &freesize,
373 c->verbose >= 0)) {
374
375 pr_err("%s is not suitable for "
376 "this array.\n",
377 dname);
378 fail = 1;
379 continue;
380 }
381 }
382
383 freesize /= 2; /* convert to K */
384 if (s->chunk && s->chunk != UnSet) {
385 /* round to chunk size */
386 freesize = freesize & ~(s->chunk-1);
387 if (do_default_chunk) {
388 /* default chunk was just set */
389 if (c->verbose > 0)
390 pr_err("chunk size "
391 "defaults to %dK\n", s->chunk);
392 s->size &= ~(unsigned long long)(s->chunk - 1);
393 do_default_chunk = 0;
394 }
395 }
396
397 if (s->size && freesize < s->size) {
398 pr_err("%s is smaller than given size."
399 " %lluK < %lluK + metadata\n",
400 dname, freesize, s->size);
401 fail = 1;
402 continue;
403 }
404 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
405 maxdisc = dname;
406 maxsize = freesize;
407 }
408 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
409 mindisc = dname;
410 minsize = freesize;
411 }
412 if (c->runstop != 1 || c->verbose >= 0) {
413 int fd = open(dname, O_RDONLY);
414 if (fd <0 ) {
415 pr_err("Cannot open %s: %s\n",
416 dname, strerror(errno));
417 fail=1;
418 continue;
419 }
420 warn |= check_ext2(fd, dname);
421 warn |= check_reiser(fd, dname);
422 warn |= check_raid(fd, dname);
423 if (strcmp(st->ss->name, "1.x") == 0 &&
424 st->minor_version >= 1)
425 /* metadata at front */
426 warn |= check_partitions(fd, dname, 0, 0);
427 else if (s->level == 1 || s->level == LEVEL_CONTAINER
428 || (s->level == 0 && s->raiddisks == 1))
429 /* partitions could be meaningful */
430 warn |= check_partitions(fd, dname, freesize*2, s->size*2);
431 else
432 /* partitions cannot be meaningful */
433 warn |= check_partitions(fd, dname, 0, 0);
434 if (strcmp(st->ss->name, "1.x") == 0 &&
435 st->minor_version >= 1 &&
436 did_default &&
437 s->level == 1 &&
438 (warn & 1024) == 0) {
439 warn |= 1024;
440 pr_err("Note: this array has metadata at the start and\n"
441 " may not be suitable as a boot device. If you plan to\n"
442 " store '/boot' on this device please ensure that\n"
443 " your boot-loader understands md/v1.x metadata, or use\n"
444 " --metadata=0.90\n");
445 }
446 close(fd);
447 }
448 }
449 if (s->raiddisks + s->sparedisks > st->max_devs) {
450 pr_err("Too many devices:"
451 " %s metadata only supports %d\n",
452 st->ss->name, st->max_devs);
453 return 1;
454 }
455 if (have_container)
456 info.array.working_disks = s->raiddisks;
457 if (fail) {
458 pr_err("create aborted\n");
459 return 1;
460 }
461 if (s->size == 0) {
462 if (mindisc == NULL && !have_container) {
463 pr_err("no size and no drives given - aborting create.\n");
464 return 1;
465 }
466 if (s->level > 0 || s->level == LEVEL_MULTIPATH
467 || s->level == LEVEL_FAULTY
468 || st->ss->external ) {
469 /* size is meaningful */
470 if (!st->ss->validate_geometry(st, s->level, s->layout,
471 s->raiddisks,
472 &s->chunk, minsize*2,
473 NULL, NULL, 0)) {
474 pr_err("devices too large for RAID level %d\n", s->level);
475 return 1;
476 }
477 s->size = minsize;
478 if (s->level == 1)
479 /* If this is ever reshaped to RAID5, we will
480 * need a chunksize. So round it off a bit
481 * now just to be safe
482 */
483 s->size &= ~(64ULL-1);
484 if (c->verbose > 0)
485 pr_err("size set to %lluK\n", s->size);
486 }
487 }
488 if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) {
489 if (c->runstop != 1 || c->verbose >= 0)
490 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
491 maxdisc, s->size);
492 warn = 1;
493 }
494
495 if (st->ss->detail_platform && st->ss->detail_platform(0, 1) != 0) {
496 if (c->runstop != 1 || c->verbose >= 0)
497 pr_err("%s unable to enumerate platform support\n"
498 " array may not be compatible with hardware/firmware\n",
499 st->ss->name);
500 warn = 1;
501 }
502
503 if (warn) {
504 if (c->runstop!= 1) {
505 if (!ask("Continue creating array? ")) {
506 pr_err("create aborted.\n");
507 return 1;
508 }
509 } else {
510 if (c->verbose > 0)
511 pr_err("creation continuing despite oddities due to --run\n");
512 }
513 }
514
515 /* If this is raid4/5, we want to configure the last active slot
516 * as missing, so that a reconstruct happens (faster than re-parity)
517 * FIX: Can we do this for raid6 as well?
518 */
519 if (st->ss->external == 0 &&
520 s->assume_clean==0 && c->force == 0 && first_missing >= s->raiddisks) {
521 switch ( s->level ) {
522 case 4:
523 case 5:
524 insert_point = s->raiddisks-1;
525 s->sparedisks++;
526 info.array.active_disks--;
527 missing_disks++;
528 break;
529 default:
530 break;
531 }
532 }
533 /* For raid6, if creating with 1 missing drive, make a good drive
534 * into a spare, else the create will fail
535 */
536 if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks &&
537 st->ss->external == 0 &&
538 second_missing >= s->raiddisks && s->level == 6) {
539 insert_point = s->raiddisks - 1;
540 if (insert_point == first_missing)
541 insert_point--;
542 s->sparedisks ++;
543 info.array.active_disks--;
544 missing_disks++;
545 }
546
547 if (s->level <= 0 && first_missing < subdevs * 2) {
548 pr_err("This level does not support missing devices\n");
549 return 1;
550 }
551
552 /* We need to create the device */
553 map_lock(&map);
554 mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name);
555 if (mdfd < 0) {
556 map_unlock(&map);
557 return 1;
558 }
559 /* verify if chosen_name is not in use,
560 * it could be in conflict with already existing device
561 * e.g. container, array
562 */
563 if (strncmp(chosen_name, "/dev/md/", 8) == 0
564 && map_by_name(&map, chosen_name+8) != NULL) {
565 pr_err("Array name %s is in use already.\n",
566 chosen_name);
567 close(mdfd);
568 map_unlock(&map);
569 return 1;
570 }
571 mddev = chosen_name;
572
573 vers = md_get_version(mdfd);
574 if (vers < 9000) {
575 pr_err("Create requires md driver version 0.90.0 or later\n");
576 goto abort_locked;
577 } else {
578 mdu_array_info_t inf;
579 memset(&inf, 0, sizeof(inf));
580 ioctl(mdfd, GET_ARRAY_INFO, &inf);
581 if (inf.working_disks != 0) {
582 pr_err("another array by this name"
583 " is already running.\n");
584 goto abort_locked;
585 }
586 }
587
588 /* Ok, lets try some ioctls */
589
590 info.array.level = s->level;
591 info.array.size = s->size;
592 info.array.raid_disks = s->raiddisks;
593 /* The kernel should *know* what md_minor we are dealing
594 * with, but it chooses to trust me instead. Sigh
595 */
596 info.array.md_minor = 0;
597 if (fstat(mdfd, &stb)==0)
598 info.array.md_minor = minor(stb.st_rdev);
599 info.array.not_persistent = 0;
600
601 if ( ( (s->level == 4 || s->level == 5) &&
602 (insert_point < s->raiddisks || first_missing < s->raiddisks) )
603 ||
604 ( s->level == 6 && (insert_point < s->raiddisks
605 || second_missing < s->raiddisks))
606 ||
607 ( s->level <= 0 )
608 ||
609 s->assume_clean
610 ) {
611 info.array.state = 1; /* clean, but one+ drive will be missing*/
612 info.resync_start = MaxSector;
613 } else {
614 info.array.state = 0; /* not clean, but no errors */
615 info.resync_start = 0;
616 }
617 if (s->level == 10) {
618 /* for raid10, the bitmap size is the capacity of the array,
619 * which is array.size * raid_disks / ncopies;
620 * .. but convert to sectors.
621 */
622 int ncopies = ((s->layout>>8) & 255) * (s->layout & 255);
623 bitmapsize = s->size * s->raiddisks / ncopies * 2;
624 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
625 } else
626 bitmapsize = s->size * 2;
627
628 /* There is lots of redundancy in these disk counts,
629 * raid_disks is the most meaningful value
630 * it describes the geometry of the array
631 * it is constant
632 * nr_disks is total number of used slots.
633 * it should be raid_disks+spare_disks
634 * spare_disks is the number of extra disks present
635 * see above
636 * active_disks is the number of working disks in
637 * active slots. (With raid_disks)
638 * working_disks is the total number of working disks,
639 * including spares
640 * failed_disks is the number of disks marked failed
641 *
642 * Ideally, the kernel would keep these (except raid_disks)
643 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
644 * So for now, we assume that all raid and spare
645 * devices will be given.
646 */
647 info.array.spare_disks=s->sparedisks;
648 info.array.failed_disks=missing_disks;
649 info.array.nr_disks = info.array.working_disks
650 + info.array.failed_disks;
651 info.array.layout = s->layout;
652 info.array.chunk_size = s->chunk*1024;
653
654 if (name == NULL || *name == 0) {
655 /* base name on mddev */
656 /* /dev/md0 -> 0
657 * /dev/md_d0 -> d0
658 * /dev/md/1 -> 1
659 * /dev/md/d1 -> d1
660 * /dev/md/home -> home
661 * /dev/mdhome -> home
662 */
663 /* FIXME compare this with rules in create_mddev */
664 name = strrchr(mddev, '/');
665 if (name) {
666 name++;
667 if (strncmp(name, "md_d", 4)==0 &&
668 strlen(name) > 4 &&
669 isdigit(name[4]) &&
670 (name-mddev) == 5 /* /dev/ */)
671 name += 3;
672 else if (strncmp(name, "md", 2)==0 &&
673 strlen(name) > 2 &&
674 isdigit(name[2]) &&
675 (name-mddev) == 5 /* /dev/ */)
676 name += 2;
677 }
678 }
679 if (!st->ss->init_super(st, &info.array, s->size, name, c->homehost, uuid))
680 goto abort_locked;
681
682 total_slots = info.array.nr_disks;
683 st->ss->getinfo_super(st, &info, NULL);
684 sysfs_init(&info, mdfd, 0);
685
686 if (did_default && c->verbose >= 0) {
687 if (is_subarray(info.text_version)) {
688 int dnum = devname2devnum(info.text_version+1);
689 char *path;
690 int mdp = get_mdp_major();
691 struct mdinfo *mdi;
692 if (dnum > 0)
693 path = map_dev(MD_MAJOR, dnum, 1);
694 else
695 path = map_dev(mdp, (-1-dnum)<< 6, 1);
696
697 mdi = sysfs_read(-1, dnum, GET_VERSION);
698
699 pr_err("Creating array inside "
700 "%s container %s\n",
701 mdi?mdi->text_version:"managed", path);
702 sysfs_free(mdi);
703 } else
704 pr_err("Defaulting to version"
705 " %s metadata\n", info.text_version);
706 }
707
708 map_update(&map, fd2devnum(mdfd), info.text_version,
709 info.uuid, chosen_name);
710 map_unlock(&map);
711
712 if (s->bitmap_file && vers < 9003) {
713 major_num = BITMAP_MAJOR_HOSTENDIAN;
714 #ifdef __BIG_ENDIAN
715 pr_err("Warning - bitmaps created on this kernel are not portable\n"
716 " between different architectured. Consider upgrading the Linux kernel.\n");
717 #endif
718 }
719
720 if (s->bitmap_file && strcmp(s->bitmap_file, "internal")==0) {
721 if ((vers%100) < 2) {
722 pr_err("internal bitmaps not supported by this kernel.\n");
723 goto abort;
724 }
725 if (!st->ss->add_internal_bitmap) {
726 pr_err("internal bitmaps not supported with %s metadata\n",
727 st->ss->name);
728 goto abort;
729 }
730 if (!st->ss->add_internal_bitmap(st, &s->bitmap_chunk,
731 c->delay, s->write_behind,
732 bitmapsize, 1, major_num)) {
733 pr_err("Given bitmap chunk size not supported.\n");
734 goto abort;
735 }
736 s->bitmap_file = NULL;
737 }
738
739 sysfs_init(&info, mdfd, 0);
740
741 if (st->ss->external && st->container_dev != NoMdDev) {
742 /* member */
743
744 /* When creating a member, we need to be careful
745 * to negotiate with mdmon properly.
746 * If it is already running, we cannot write to
747 * the devices and must ask it to do that part.
748 * If it isn't running, we write to the devices,
749 * and then start it.
750 * We hold an exclusive open on the container
751 * device to make sure mdmon doesn't exit after
752 * we checked that it is running.
753 *
754 * For now, fail if it is already running.
755 */
756 container_fd = open_dev_excl(st->container_dev);
757 if (container_fd < 0) {
758 pr_err("Cannot get exclusive "
759 "open on container - weird.\n");
760 goto abort;
761 }
762 if (mdmon_running(st->container_dev)) {
763 if (c->verbose)
764 pr_err("reusing mdmon "
765 "for %s.\n",
766 devnum2devname(st->container_dev));
767 st->update_tail = &st->updates;
768 } else
769 need_mdmon = 1;
770 }
771 rv = set_array_info(mdfd, st, &info);
772 if (rv) {
773 pr_err("failed to set array info for %s: %s\n",
774 mddev, strerror(errno));
775 goto abort;
776 }
777
778 if (s->bitmap_file) {
779 int uuid[4];
780
781 st->ss->uuid_from_super(st, uuid);
782 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
783 c->delay, s->write_behind,
784 bitmapsize,
785 major_num)) {
786 goto abort;
787 }
788 bitmap_fd = open(s->bitmap_file, O_RDWR);
789 if (bitmap_fd < 0) {
790 pr_err("weird: %s cannot be openned\n",
791 s->bitmap_file);
792 goto abort;
793 }
794 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
795 pr_err("Cannot set bitmap file for %s: %s\n",
796 mddev, strerror(errno));
797 goto abort;
798 }
799 }
800
801 infos = xmalloc(sizeof(*infos) * total_slots);
802
803 for (pass=1; pass <=2 ; pass++) {
804 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
805
806 for (dnum=0, dv = devlist ; dv ;
807 dv=(dv->next)?(dv->next):moved_disk, dnum++) {
808 int fd;
809 struct stat stb;
810 struct mdinfo *inf = &infos[dnum];
811
812 if (dnum >= total_slots)
813 abort();
814 if (dnum == insert_point) {
815 moved_disk = dv;
816 continue;
817 }
818 if (strcasecmp(dv->devname, "missing")==0)
819 continue;
820 if (have_container)
821 moved_disk = NULL;
822 if (have_container && dnum < info.array.raid_disks - 1)
823 /* repeatedly use the container */
824 moved_disk = dv;
825
826 switch(pass) {
827 case 1:
828 *inf = info;
829
830 inf->disk.number = dnum;
831 inf->disk.raid_disk = dnum;
832 if (inf->disk.raid_disk < s->raiddisks)
833 inf->disk.state = (1<<MD_DISK_ACTIVE) |
834 (1<<MD_DISK_SYNC);
835 else
836 inf->disk.state = 0;
837
838 if (dv->writemostly == 1)
839 inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
840
841 if (have_container)
842 fd = -1;
843 else {
844 if (st->ss->external &&
845 st->container_dev != NoMdDev)
846 fd = open(dv->devname, O_RDWR);
847 else
848 fd = open(dv->devname, O_RDWR|O_EXCL);
849
850 if (fd < 0) {
851 pr_err("failed to open %s "
852 "after earlier success - aborting\n",
853 dv->devname);
854 goto abort;
855 }
856 fstat(fd, &stb);
857 inf->disk.major = major(stb.st_rdev);
858 inf->disk.minor = minor(stb.st_rdev);
859 }
860 if (fd >= 0)
861 remove_partitions(fd);
862 if (st->ss->add_to_super(st, &inf->disk,
863 fd, dv->devname)) {
864 ioctl(mdfd, STOP_ARRAY, NULL);
865 goto abort;
866 }
867 st->ss->getinfo_super(st, inf, NULL);
868 safe_mode_delay = inf->safe_mode_delay;
869
870 if (have_container && c->verbose > 0)
871 pr_err("Using %s for device %d\n",
872 map_dev(inf->disk.major,
873 inf->disk.minor,
874 0), dnum);
875
876 if (!have_container) {
877 /* getinfo_super might have lost these ... */
878 inf->disk.major = major(stb.st_rdev);
879 inf->disk.minor = minor(stb.st_rdev);
880 }
881 break;
882 case 2:
883 inf->errors = 0;
884
885 rv = add_disk(mdfd, st, &info, inf);
886
887 if (rv) {
888 pr_err("ADD_NEW_DISK for %s "
889 "failed: %s\n",
890 dv->devname, strerror(errno));
891 goto abort;
892 }
893 break;
894 }
895 if (!have_container &&
896 dv == moved_disk && dnum != insert_point) break;
897 }
898 if (pass == 1) {
899 struct mdinfo info_new;
900 struct map_ent *me = NULL;
901
902 /* check to see if the uuid has changed due to these
903 * metadata changes, and if so update the member array
904 * and container uuid. Note ->write_init_super clears
905 * the subarray cursor such that ->getinfo_super once
906 * again returns container info.
907 */
908 map_lock(&map);
909 st->ss->getinfo_super(st, &info_new, NULL);
910 if (st->ss->external && s->level != LEVEL_CONTAINER &&
911 !same_uuid(info_new.uuid, info.uuid, 0)) {
912 map_update(&map, fd2devnum(mdfd),
913 info_new.text_version,
914 info_new.uuid, chosen_name);
915 me = map_by_devnum(&map, st->container_dev);
916 }
917
918 if (st->ss->write_init_super(st)) {
919 st->ss->free_super(st);
920 goto abort_locked;
921 }
922
923 /* update parent container uuid */
924 if (me) {
925 char *path = xstrdup(me->path);
926
927 st->ss->getinfo_super(st, &info_new, NULL);
928 map_update(&map, st->container_dev,
929 info_new.text_version,
930 info_new.uuid, path);
931 free(path);
932 }
933 map_unlock(&map);
934
935 flush_metadata_updates(st);
936 st->ss->free_super(st);
937 }
938 }
939 free(infos);
940
941 if (s->level == LEVEL_CONTAINER) {
942 /* No need to start. But we should signal udev to
943 * create links */
944 sysfs_uevent(&info, "change");
945 if (c->verbose >= 0)
946 pr_err("container %s prepared.\n", mddev);
947 wait_for(chosen_name, mdfd);
948 } else if (c->runstop == 1 || subdevs >= s->raiddisks) {
949 if (st->ss->external) {
950 int err;
951 switch(s->level) {
952 case LEVEL_LINEAR:
953 case LEVEL_MULTIPATH:
954 case 0:
955 err = sysfs_set_str(&info, NULL, "array_state",
956 c->readonly
957 ? "readonly"
958 : "active");
959 need_mdmon = 0;
960 break;
961 default:
962 err = sysfs_set_str(&info, NULL, "array_state",
963 "readonly");
964 break;
965 }
966 sysfs_set_safemode(&info, safe_mode_delay);
967 if (err) {
968 pr_err("failed to"
969 " activate array.\n");
970 ioctl(mdfd, STOP_ARRAY, NULL);
971 goto abort;
972 }
973 } else if (c->readonly &&
974 sysfs_attribute_available(
975 &info, NULL, "array_state")) {
976 if (sysfs_set_str(&info, NULL,
977 "array_state", "readonly") < 0) {
978 pr_err("Failed to start array: %s\n",
979 strerror(errno));
980 ioctl(mdfd, STOP_ARRAY, NULL);
981 goto abort;
982 }
983 } else {
984 /* param is not actually used */
985 mdu_param_t param;
986 if (ioctl(mdfd, RUN_ARRAY, &param)) {
987 pr_err("RUN_ARRAY failed: %s\n",
988 strerror(errno));
989 if (info.array.chunk_size & (info.array.chunk_size-1)) {
990 cont_err("Problem may be that "
991 "chunk size is not a power of 2\n");
992 }
993 ioctl(mdfd, STOP_ARRAY, NULL);
994 goto abort;
995 }
996 }
997 if (c->verbose >= 0)
998 pr_err("array %s started.\n", mddev);
999 if (st->ss->external && st->container_dev != NoMdDev) {
1000 if (need_mdmon)
1001 start_mdmon(st->container_dev);
1002
1003 ping_monitor_by_id(st->container_dev);
1004 close(container_fd);
1005 }
1006 wait_for(chosen_name, mdfd);
1007 } else {
1008 pr_err("not starting array - not enough devices.\n");
1009 }
1010 close(mdfd);
1011 return 0;
1012
1013 abort:
1014 map_lock(&map);
1015 abort_locked:
1016 map_remove(&map, fd2devnum(mdfd));
1017 map_unlock(&map);
1018
1019 if (mdfd >= 0)
1020 close(mdfd);
1021 return 1;
1022 }
Unnamed repository; edit this file 'description' to name the repository.