]> 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
Use new 'struct shape' to pass args to Create
[thirdparty/mdadm.git] / Create.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2009 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
65 int Create(struct supertype *st, char *mddev,
66 char *name, int *uuid,
67 int subdevs, struct mddev_dev *devlist,
68 struct shape *s,
69 struct context *c)
70 {
71 /*
72 * Create a new raid array.
73 *
74 * First check that necessary details are available
75 * (i.e. level, raid-disks)
76 *
77 * Then check each disk to see what might be on it
78 * and report anything interesting.
79 *
80 * If anything looks odd, and runstop not set,
81 * abort.
82 *
83 * SET_ARRAY_INFO and ADD_NEW_DISK, and
84 * if runstop==run, or raiddisks disks were used,
85 * RUN_ARRAY
86 */
87 int mdfd;
88 unsigned long long minsize=0, maxsize=0;
89 char *mindisc = NULL;
90 char *maxdisc = NULL;
91 int dnum;
92 struct mddev_dev *dv;
93 int fail=0, warn=0;
94 struct stat stb;
95 int first_missing = subdevs * 2;
96 int second_missing = subdevs * 2;
97 int missing_disks = 0;
98 int insert_point = subdevs * 2; /* where to insert a missing drive */
99 int total_slots;
100 int pass;
101 int vers;
102 int rv;
103 int bitmap_fd;
104 int have_container = 0;
105 int container_fd = -1;
106 int need_mdmon = 0;
107 unsigned long long bitmapsize;
108 struct mdinfo info, *infos;
109 int did_default = 0;
110 int do_default_layout = 0;
111 int do_default_chunk = 0;
112 unsigned long safe_mode_delay = 0;
113 char chosen_name[1024];
114 struct map_ent *map = NULL;
115 unsigned long long newsize;
116
117 int major_num = BITMAP_MAJOR_HI;
118
119 memset(&info, 0, sizeof(info));
120 if (s->level == UnSet && st && st->ss->default_geometry)
121 st->ss->default_geometry(st, &s->level, NULL, NULL);
122 if (s->level == UnSet) {
123 pr_err("a RAID level is needed to create an array.\n");
124 return 1;
125 }
126 if (s->raiddisks < 4 && s->level == 6) {
127 pr_err("at least 4 raid-devices needed for level 6\n");
128 return 1;
129 }
130 if (s->raiddisks > 256 && s->level == 6) {
131 pr_err("no more than 256 raid-devices supported for level 6\n");
132 return 1;
133 }
134 if (s->raiddisks < 2 && s->level >= 4) {
135 pr_err("at least 2 raid-devices needed for level 4 or 5\n");
136 return 1;
137 }
138 if (s->level <= 0 && s->sparedisks) {
139 pr_err("This level does not support spare devices\n");
140 return 1;
141 }
142
143 if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
144 /* If given a single device, it might be a container, and we can
145 * extract a device list from there
146 */
147 mdu_array_info_t inf;
148 int fd;
149
150 memset(&inf, 0, sizeof(inf));
151 fd = open(devlist->devname, O_RDONLY);
152 if (fd >= 0 &&
153 ioctl(fd, GET_ARRAY_INFO, &inf) == 0 &&
154 inf.raid_disks == 0) {
155 /* yep, looks like a container */
156 if (st) {
157 rv = st->ss->load_container(st, fd,
158 devlist->devname);
159 if (rv == 0)
160 have_container = 1;
161 } else {
162 st = super_by_fd(fd, NULL);
163 if (st && !(rv = st->ss->
164 load_container(st, fd,
165 devlist->devname)))
166 have_container = 1;
167 else
168 st = NULL;
169 }
170 if (have_container) {
171 subdevs = s->raiddisks;
172 first_missing = subdevs * 2;
173 second_missing = subdevs * 2;
174 insert_point = subdevs * 2;
175 }
176 }
177 if (fd >= 0)
178 close(fd);
179 }
180 if (st && st->ss->external && s->sparedisks) {
181 pr_err("This metadata type does not support "
182 "spare disks at create time\n");
183 return 1;
184 }
185 if (subdevs > s->raiddisks+s->sparedisks) {
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) {
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
202 if (s->layout == UnSet) {
203 do_default_layout = 1;
204 s->layout = default_layout(st, s->level, c->verbose);
205 }
206
207 if (s->level == 10)
208 /* check layout fits in array*/
209 if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) {
210 pr_err("that layout requires at least %d devices\n",
211 (s->layout&255) * ((s->layout>>8)&255));
212 return 1;
213 }
214
215 switch(s->level) {
216 case 4:
217 case 5:
218 case 10:
219 case 6:
220 case 0:
221 if (s->chunk == 0 || s->chunk == UnSet) {
222 s->chunk = UnSet;
223 do_default_chunk = 1;
224 /* chunk will be set later */
225 }
226 break;
227 case LEVEL_LINEAR:
228 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
229 if (get_linux_version() < 2006016 && s->chunk == 0) {
230 s->chunk = 64;
231 if (c->verbose > 0)
232 pr_err("chunk size defaults to 64K\n");
233 }
234 break;
235 case 1:
236 case LEVEL_FAULTY:
237 case LEVEL_MULTIPATH:
238 case LEVEL_CONTAINER:
239 if (s->chunk) {
240 s->chunk = 0;
241 if (c->verbose > 0)
242 pr_err("chunk size ignored for this level\n");
243 }
244 break;
245 default:
246 pr_err("unknown level %d\n", s->level);
247 return 1;
248 }
249 if (s->size == MAX_SIZE)
250 /* use '0' to mean 'max' now... */
251 s->size = 0;
252 if (s->size && s->chunk && s->chunk != UnSet)
253 s->size &= ~(unsigned long long)(s->chunk - 1);
254 newsize = s->size * 2;
255 if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks,
256 &s->chunk, s->size*2, NULL, &newsize, c->verbose>=0))
257 return 1;
258
259 if (s->chunk && s->chunk != UnSet) {
260 newsize &= ~(unsigned long long)(s->chunk*2 - 1);
261 if (do_default_chunk) {
262 /* default chunk was just set */
263 if (c->verbose > 0)
264 pr_err("chunk size "
265 "defaults to %dK\n", s->chunk);
266 s->size &= ~(unsigned long long)(s->chunk - 1);
267 do_default_chunk = 0;
268 }
269 }
270
271 if (s->size == 0) {
272 s->size = newsize / 2;
273 if (s->level == 1)
274 /* If this is ever reshaped to RAID5, we will
275 * need a chunksize. So round it off a bit
276 * now just to be safe
277 */
278 s->size &= ~(64ULL-1);
279
280 if (s->size && c->verbose > 0)
281 pr_err("setting size to %lluK\n", s->size);
282 }
283
284 /* now look at the subdevs */
285 info.array.active_disks = 0;
286 info.array.working_disks = 0;
287 dnum = 0;
288 for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) {
289 char *dname = dv->devname;
290 unsigned long long freesize;
291 int dfd;
292
293 if (strcasecmp(dname, "missing")==0) {
294 if (first_missing > dnum)
295 first_missing = dnum;
296 if (second_missing > dnum && dnum > first_missing)
297 second_missing = dnum;
298 missing_disks ++;
299 continue;
300 }
301 dfd = open(dname, O_RDONLY);
302 if (dfd < 0) {
303 pr_err("cannot open %s: %s\n",
304 dname, strerror(errno));
305 exit(2);
306 }
307 if (fstat(dfd, &stb) != 0 ||
308 (stb.st_mode & S_IFMT) != S_IFBLK) {
309 close(dfd);
310 pr_err("%s is not a block device\n",
311 dname);
312 exit(2);
313 }
314 close(dfd);
315 info.array.working_disks++;
316 if (dnum < s->raiddisks)
317 info.array.active_disks++;
318 if (st == NULL) {
319 struct createinfo *ci = conf_get_create_info();
320 if (ci)
321 st = ci->supertype;
322 }
323 if (st == NULL) {
324 /* Need to choose a default metadata, which is different
325 * depending on geometry of array.
326 */
327 int i;
328 char *name = "default";
329 for(i=0; !st && superlist[i]; i++) {
330 st = superlist[i]->match_metadata_desc(name);
331 if (!st)
332 continue;
333 if (do_default_layout)
334 s->layout = default_layout(st, s->level, c->verbose);
335 switch (st->ss->validate_geometry(
336 st, s->level, s->layout, s->raiddisks,
337 &s->chunk, s->size*2, dname, &freesize,
338 c->verbose > 0)) {
339 case -1: /* Not valid, message printed, and not
340 * worth checking any further */
341 exit(2);
342 break;
343 case 0: /* Geometry not valid */
344 free(st);
345 st = NULL;
346 s->chunk = do_default_chunk ? UnSet : s->chunk;
347 break;
348 case 1: /* All happy */
349 break;
350 }
351 }
352
353 if (!st) {
354 int dfd = open(dname, O_RDONLY|O_EXCL);
355 if (dfd < 0) {
356 pr_err("cannot open %s: %s\n",
357 dname, strerror(errno));
358 exit(2);
359 }
360 pr_err("device %s not suitable "
361 "for any style of array\n",
362 dname);
363 exit(2);
364 }
365 if (st->ss != &super0 ||
366 st->minor_version != 90)
367 did_default = 1;
368 } else {
369 if (do_default_layout)
370 s->layout = default_layout(st, s->level, 0);
371 if (!st->ss->validate_geometry(st, s->level, s->layout,
372 s->raiddisks,
373 &s->chunk, s->size*2, dname,
374 &freesize,
375 c->verbose >= 0)) {
376
377 pr_err("%s is not suitable for "
378 "this array.\n",
379 dname);
380 fail = 1;
381 continue;
382 }
383 }
384
385 freesize /= 2; /* convert to K */
386 if (s->chunk && s->chunk != UnSet) {
387 /* round to chunk size */
388 freesize = freesize & ~(s->chunk-1);
389 if (do_default_chunk) {
390 /* default chunk was just set */
391 if (c->verbose > 0)
392 pr_err("chunk size "
393 "defaults to %dK\n", s->chunk);
394 s->size &= ~(unsigned long long)(s->chunk - 1);
395 do_default_chunk = 0;
396 }
397 }
398
399 if (s->size && freesize < s->size) {
400 pr_err("%s is smaller than given size."
401 " %lluK < %lluK + metadata\n",
402 dname, freesize, s->size);
403 fail = 1;
404 continue;
405 }
406 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
407 maxdisc = dname;
408 maxsize = freesize;
409 }
410 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
411 mindisc = dname;
412 minsize = freesize;
413 }
414 if (c->runstop != 1 || c->verbose >= 0) {
415 int fd = open(dname, O_RDONLY);
416 if (fd <0 ) {
417 pr_err("Cannot open %s: %s\n",
418 dname, strerror(errno));
419 fail=1;
420 continue;
421 }
422 warn |= check_ext2(fd, dname);
423 warn |= check_reiser(fd, dname);
424 warn |= check_raid(fd, dname);
425 if (strcmp(st->ss->name, "1.x") == 0 &&
426 st->minor_version >= 1)
427 /* metadata at front */
428 warn |= check_partitions(fd, dname, 0, 0);
429 else if (s->level == 1 || s->level == LEVEL_CONTAINER
430 || (s->level == 0 && s->raiddisks == 1))
431 /* partitions could be meaningful */
432 warn |= check_partitions(fd, dname, freesize*2, s->size*2);
433 else
434 /* partitions cannot be meaningful */
435 warn |= check_partitions(fd, dname, 0, 0);
436 if (strcmp(st->ss->name, "1.x") == 0 &&
437 st->minor_version >= 1 &&
438 did_default &&
439 s->level == 1 &&
440 (warn & 1024) == 0) {
441 warn |= 1024;
442 pr_err("Note: this array has metadata at the start and\n"
443 " may not be suitable as a boot device. If you plan to\n"
444 " store '/boot' on this device please ensure that\n"
445 " your boot-loader understands md/v1.x metadata, or use\n"
446 " --metadata=0.90\n");
447 }
448 close(fd);
449 }
450 }
451 if (s->raiddisks + s->sparedisks > st->max_devs) {
452 pr_err("Too many devices:"
453 " %s metadata only supports %d\n",
454 st->ss->name, st->max_devs);
455 return 1;
456 }
457 if (have_container)
458 info.array.working_disks = s->raiddisks;
459 if (fail) {
460 pr_err("create aborted\n");
461 return 1;
462 }
463 if (s->size == 0) {
464 if (mindisc == NULL && !have_container) {
465 pr_err("no size and no drives given - aborting create.\n");
466 return 1;
467 }
468 if (s->level > 0 || s->level == LEVEL_MULTIPATH
469 || s->level == LEVEL_FAULTY
470 || st->ss->external ) {
471 /* size is meaningful */
472 if (!st->ss->validate_geometry(st, s->level, s->layout,
473 s->raiddisks,
474 &s->chunk, minsize*2,
475 NULL, NULL, 0)) {
476 pr_err("devices too large for RAID level %d\n", s->level);
477 return 1;
478 }
479 s->size = minsize;
480 if (s->level == 1)
481 /* If this is ever reshaped to RAID5, we will
482 * need a chunksize. So round it off a bit
483 * now just to be safe
484 */
485 s->size &= ~(64ULL-1);
486 if (c->verbose > 0)
487 pr_err("size set to %lluK\n", s->size);
488 }
489 }
490 if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) {
491 if (c->runstop != 1 || c->verbose >= 0)
492 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
493 maxdisc, s->size);
494 warn = 1;
495 }
496
497 if (st->ss->detail_platform && st->ss->detail_platform(0, 1) != 0) {
498 if (c->runstop != 1 || c->verbose >= 0)
499 pr_err("%s unable to enumerate platform support\n"
500 " array may not be compatible with hardware/firmware\n",
501 st->ss->name);
502 warn = 1;
503 }
504
505 if (warn) {
506 if (c->runstop!= 1) {
507 if (!ask("Continue creating array? ")) {
508 pr_err("create aborted.\n");
509 return 1;
510 }
511 } else {
512 if (c->verbose > 0)
513 pr_err("creation continuing despite oddities due to --run\n");
514 }
515 }
516
517 /* If this is raid4/5, we want to configure the last active slot
518 * as missing, so that a reconstruct happens (faster than re-parity)
519 * FIX: Can we do this for raid6 as well?
520 */
521 if (st->ss->external == 0 &&
522 s->assume_clean==0 && c->force == 0 && first_missing >= s->raiddisks) {
523 switch ( s->level ) {
524 case 4:
525 case 5:
526 insert_point = s->raiddisks-1;
527 s->sparedisks++;
528 info.array.active_disks--;
529 missing_disks++;
530 break;
531 default:
532 break;
533 }
534 }
535 /* For raid6, if creating with 1 missing drive, make a good drive
536 * into a spare, else the create will fail
537 */
538 if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks &&
539 st->ss->external == 0 &&
540 second_missing >= s->raiddisks && s->level == 6) {
541 insert_point = s->raiddisks - 1;
542 if (insert_point == first_missing)
543 insert_point--;
544 s->sparedisks ++;
545 info.array.active_disks--;
546 missing_disks++;
547 }
548
549 if (s->level <= 0 && first_missing < subdevs * 2) {
550 pr_err("This level does not support missing devices\n");
551 return 1;
552 }
553
554 /* We need to create the device */
555 map_lock(&map);
556 mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name);
557 if (mdfd < 0) {
558 map_unlock(&map);
559 return 1;
560 }
561 /* verify if chosen_name is not in use,
562 * it could be in conflict with already existing device
563 * e.g. container, array
564 */
565 if (strncmp(chosen_name, "/dev/md/", 8) == 0
566 && map_by_name(&map, chosen_name+8) != NULL) {
567 pr_err("Array name %s is in use already.\n",
568 chosen_name);
569 close(mdfd);
570 map_unlock(&map);
571 return 1;
572 }
573 mddev = chosen_name;
574
575 vers = md_get_version(mdfd);
576 if (vers < 9000) {
577 pr_err("Create requires md driver version 0.90.0 or later\n");
578 goto abort_locked;
579 } else {
580 mdu_array_info_t inf;
581 memset(&inf, 0, sizeof(inf));
582 ioctl(mdfd, GET_ARRAY_INFO, &inf);
583 if (inf.working_disks != 0) {
584 pr_err("another array by this name"
585 " is already running.\n");
586 goto abort_locked;
587 }
588 }
589
590 /* Ok, lets try some ioctls */
591
592 info.array.level = s->level;
593 info.array.size = s->size;
594 info.array.raid_disks = s->raiddisks;
595 /* The kernel should *know* what md_minor we are dealing
596 * with, but it chooses to trust me instead. Sigh
597 */
598 info.array.md_minor = 0;
599 if (fstat(mdfd, &stb)==0)
600 info.array.md_minor = minor(stb.st_rdev);
601 info.array.not_persistent = 0;
602
603 if ( ( (s->level == 4 || s->level == 5) &&
604 (insert_point < s->raiddisks || first_missing < s->raiddisks) )
605 ||
606 ( s->level == 6 && (insert_point < s->raiddisks
607 || second_missing < s->raiddisks))
608 ||
609 ( s->level <= 0 )
610 ||
611 s->assume_clean
612 ) {
613 info.array.state = 1; /* clean, but one+ drive will be missing*/
614 info.resync_start = MaxSector;
615 } else {
616 info.array.state = 0; /* not clean, but no errors */
617 info.resync_start = 0;
618 }
619 if (s->level == 10) {
620 /* for raid10, the bitmap size is the capacity of the array,
621 * which is array.size * raid_disks / ncopies;
622 * .. but convert to sectors.
623 */
624 int ncopies = ((s->layout>>8) & 255) * (s->layout & 255);
625 bitmapsize = s->size * s->raiddisks / ncopies * 2;
626 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
627 } else
628 bitmapsize = s->size * 2;
629
630 /* There is lots of redundancy in these disk counts,
631 * raid_disks is the most meaningful value
632 * it describes the geometry of the array
633 * it is constant
634 * nr_disks is total number of used slots.
635 * it should be raid_disks+spare_disks
636 * spare_disks is the number of extra disks present
637 * see above
638 * active_disks is the number of working disks in
639 * active slots. (With raid_disks)
640 * working_disks is the total number of working disks,
641 * including spares
642 * failed_disks is the number of disks marked failed
643 *
644 * Ideally, the kernel would keep these (except raid_disks)
645 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
646 * So for now, we assume that all raid and spare
647 * devices will be given.
648 */
649 info.array.spare_disks=s->sparedisks;
650 info.array.failed_disks=missing_disks;
651 info.array.nr_disks = info.array.working_disks
652 + info.array.failed_disks;
653 info.array.layout = s->layout;
654 info.array.chunk_size = s->chunk*1024;
655
656 if (name == NULL || *name == 0) {
657 /* base name on mddev */
658 /* /dev/md0 -> 0
659 * /dev/md_d0 -> d0
660 * /dev/md/1 -> 1
661 * /dev/md/d1 -> d1
662 * /dev/md/home -> home
663 * /dev/mdhome -> home
664 */
665 /* FIXME compare this with rules in create_mddev */
666 name = strrchr(mddev, '/');
667 if (name) {
668 name++;
669 if (strncmp(name, "md_d", 4)==0 &&
670 strlen(name) > 4 &&
671 isdigit(name[4]) &&
672 (name-mddev) == 5 /* /dev/ */)
673 name += 3;
674 else if (strncmp(name, "md", 2)==0 &&
675 strlen(name) > 2 &&
676 isdigit(name[2]) &&
677 (name-mddev) == 5 /* /dev/ */)
678 name += 2;
679 }
680 }
681 if (!st->ss->init_super(st, &info.array, s->size, name, c->homehost, uuid))
682 goto abort_locked;
683
684 total_slots = info.array.nr_disks;
685 st->ss->getinfo_super(st, &info, NULL);
686 sysfs_init(&info, mdfd, 0);
687
688 if (did_default && c->verbose >= 0) {
689 if (is_subarray(info.text_version)) {
690 int dnum = devname2devnum(info.text_version+1);
691 char *path;
692 int mdp = get_mdp_major();
693 struct mdinfo *mdi;
694 if (dnum > 0)
695 path = map_dev(MD_MAJOR, dnum, 1);
696 else
697 path = map_dev(mdp, (-1-dnum)<< 6, 1);
698
699 mdi = sysfs_read(-1, dnum, GET_VERSION);
700
701 pr_err("Creating array inside "
702 "%s container %s\n",
703 mdi?mdi->text_version:"managed", path);
704 sysfs_free(mdi);
705 } else
706 pr_err("Defaulting to version"
707 " %s metadata\n", info.text_version);
708 }
709
710 map_update(&map, fd2devnum(mdfd), info.text_version,
711 info.uuid, chosen_name);
712 map_unlock(&map);
713
714 if (s->bitmap_file && vers < 9003) {
715 major_num = BITMAP_MAJOR_HOSTENDIAN;
716 #ifdef __BIG_ENDIAN
717 pr_err("Warning - bitmaps created on this kernel are not portable\n"
718 " between different architectured. Consider upgrading the Linux kernel.\n");
719 #endif
720 }
721
722 if (s->bitmap_file && strcmp(s->bitmap_file, "internal")==0) {
723 if ((vers%100) < 2) {
724 pr_err("internal bitmaps not supported by this kernel.\n");
725 goto abort;
726 }
727 if (!st->ss->add_internal_bitmap) {
728 pr_err("internal bitmaps not supported with %s metadata\n",
729 st->ss->name);
730 goto abort;
731 }
732 if (!st->ss->add_internal_bitmap(st, &s->bitmap_chunk,
733 c->delay, s->write_behind,
734 bitmapsize, 1, major_num)) {
735 pr_err("Given bitmap chunk size not supported.\n");
736 goto abort;
737 }
738 s->bitmap_file = NULL;
739 }
740
741
742 sysfs_init(&info, mdfd, 0);
743
744 if (st->ss->external && st->container_dev != NoMdDev) {
745 /* member */
746
747 /* When creating a member, we need to be careful
748 * to negotiate with mdmon properly.
749 * If it is already running, we cannot write to
750 * the devices and must ask it to do that part.
751 * If it isn't running, we write to the devices,
752 * and then start it.
753 * We hold an exclusive open on the container
754 * device to make sure mdmon doesn't exit after
755 * we checked that it is running.
756 *
757 * For now, fail if it is already running.
758 */
759 container_fd = open_dev_excl(st->container_dev);
760 if (container_fd < 0) {
761 pr_err("Cannot get exclusive "
762 "open on container - weird.\n");
763 goto abort;
764 }
765 if (mdmon_running(st->container_dev)) {
766 if (c->verbose)
767 pr_err("reusing mdmon "
768 "for %s.\n",
769 devnum2devname(st->container_dev));
770 st->update_tail = &st->updates;
771 } else
772 need_mdmon = 1;
773 }
774 rv = set_array_info(mdfd, st, &info);
775 if (rv) {
776 pr_err("failed to set array info for %s: %s\n",
777 mddev, strerror(errno));
778 goto abort;
779 }
780
781 if (s->bitmap_file) {
782 int uuid[4];
783
784 st->ss->uuid_from_super(st, uuid);
785 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
786 c->delay, s->write_behind,
787 bitmapsize,
788 major_num)) {
789 goto abort;
790 }
791 bitmap_fd = open(s->bitmap_file, O_RDWR);
792 if (bitmap_fd < 0) {
793 pr_err("weird: %s cannot be openned\n",
794 s->bitmap_file);
795 goto abort;
796 }
797 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
798 pr_err("Cannot set bitmap file for %s: %s\n",
799 mddev, strerror(errno));
800 goto abort;
801 }
802 }
803
804 infos = xmalloc(sizeof(*infos) * total_slots);
805
806 for (pass=1; pass <=2 ; pass++) {
807 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
808
809 for (dnum=0, dv = devlist ; dv ;
810 dv=(dv->next)?(dv->next):moved_disk, dnum++) {
811 int fd;
812 struct stat stb;
813 struct mdinfo *inf = &infos[dnum];
814
815 if (dnum >= total_slots)
816 abort();
817 if (dnum == insert_point) {
818 moved_disk = dv;
819 continue;
820 }
821 if (strcasecmp(dv->devname, "missing")==0)
822 continue;
823 if (have_container)
824 moved_disk = NULL;
825 if (have_container && dnum < info.array.raid_disks - 1)
826 /* repeatedly use the container */
827 moved_disk = dv;
828
829 switch(pass) {
830 case 1:
831 *inf = info;
832
833 inf->disk.number = dnum;
834 inf->disk.raid_disk = dnum;
835 if (inf->disk.raid_disk < s->raiddisks)
836 inf->disk.state = (1<<MD_DISK_ACTIVE) |
837 (1<<MD_DISK_SYNC);
838 else
839 inf->disk.state = 0;
840
841 if (dv->writemostly == 1)
842 inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
843
844 if (have_container)
845 fd = -1;
846 else {
847 if (st->ss->external &&
848 st->container_dev != NoMdDev)
849 fd = open(dv->devname, O_RDWR);
850 else
851 fd = open(dv->devname, O_RDWR|O_EXCL);
852
853 if (fd < 0) {
854 pr_err("failed to open %s "
855 "after earlier success - aborting\n",
856 dv->devname);
857 goto abort;
858 }
859 fstat(fd, &stb);
860 inf->disk.major = major(stb.st_rdev);
861 inf->disk.minor = minor(stb.st_rdev);
862 }
863 if (fd >= 0)
864 remove_partitions(fd);
865 if (st->ss->add_to_super(st, &inf->disk,
866 fd, dv->devname)) {
867 ioctl(mdfd, STOP_ARRAY, NULL);
868 goto abort;
869 }
870 st->ss->getinfo_super(st, inf, NULL);
871 safe_mode_delay = inf->safe_mode_delay;
872
873 if (have_container && c->verbose > 0)
874 pr_err("Using %s for device %d\n",
875 map_dev(inf->disk.major,
876 inf->disk.minor,
877 0), dnum);
878
879 if (!have_container) {
880 /* getinfo_super might have lost these ... */
881 inf->disk.major = major(stb.st_rdev);
882 inf->disk.minor = minor(stb.st_rdev);
883 }
884 break;
885 case 2:
886 inf->errors = 0;
887
888 rv = add_disk(mdfd, st, &info, inf);
889
890 if (rv) {
891 pr_err("ADD_NEW_DISK for %s "
892 "failed: %s\n",
893 dv->devname, strerror(errno));
894 goto abort;
895 }
896 break;
897 }
898 if (!have_container &&
899 dv == moved_disk && dnum != insert_point) break;
900 }
901 if (pass == 1) {
902 struct mdinfo info_new;
903 struct map_ent *me = NULL;
904
905 /* check to see if the uuid has changed due to these
906 * metadata changes, and if so update the member array
907 * and container uuid. Note ->write_init_super clears
908 * the subarray cursor such that ->getinfo_super once
909 * again returns container info.
910 */
911 map_lock(&map);
912 st->ss->getinfo_super(st, &info_new, NULL);
913 if (st->ss->external && s->level != LEVEL_CONTAINER &&
914 !same_uuid(info_new.uuid, info.uuid, 0)) {
915 map_update(&map, fd2devnum(mdfd),
916 info_new.text_version,
917 info_new.uuid, chosen_name);
918 me = map_by_devnum(&map, st->container_dev);
919 }
920
921 if (st->ss->write_init_super(st)) {
922 st->ss->free_super(st);
923 goto abort_locked;
924 }
925
926 /* update parent container uuid */
927 if (me) {
928 char *path = xstrdup(me->path);
929
930 st->ss->getinfo_super(st, &info_new, NULL);
931 map_update(&map, st->container_dev,
932 info_new.text_version,
933 info_new.uuid, path);
934 free(path);
935 }
936 map_unlock(&map);
937
938 flush_metadata_updates(st);
939 st->ss->free_super(st);
940 }
941 }
942 free(infos);
943
944 if (s->level == LEVEL_CONTAINER) {
945 /* No need to start. But we should signal udev to
946 * create links */
947 sysfs_uevent(&info, "change");
948 if (c->verbose >= 0)
949 pr_err("container %s prepared.\n", mddev);
950 wait_for(chosen_name, mdfd);
951 } else if (c->runstop == 1 || subdevs >= s->raiddisks) {
952 if (st->ss->external) {
953 int err;
954 switch(s->level) {
955 case LEVEL_LINEAR:
956 case LEVEL_MULTIPATH:
957 case 0:
958 err = sysfs_set_str(&info, NULL, "array_state",
959 c->readonly
960 ? "readonly"
961 : "active");
962 need_mdmon = 0;
963 break;
964 default:
965 err = sysfs_set_str(&info, NULL, "array_state",
966 "readonly");
967 break;
968 }
969 sysfs_set_safemode(&info, safe_mode_delay);
970 if (err) {
971 pr_err("failed to"
972 " activate array.\n");
973 ioctl(mdfd, STOP_ARRAY, NULL);
974 goto abort;
975 }
976 } else if (c->readonly &&
977 sysfs_attribute_available(
978 &info, NULL, "array_state")) {
979 if (sysfs_set_str(&info, NULL,
980 "array_state", "readonly") < 0) {
981 pr_err("Failed to start array: %s\n",
982 strerror(errno));
983 ioctl(mdfd, STOP_ARRAY, NULL);
984 goto abort;
985 }
986 } else {
987 /* param is not actually used */
988 mdu_param_t param;
989 if (ioctl(mdfd, RUN_ARRAY, &param)) {
990 pr_err("RUN_ARRAY failed: %s\n",
991 strerror(errno));
992 if (info.array.chunk_size & (info.array.chunk_size-1)) {
993 cont_err("Problem may be that "
994 "chunk size is not a power of 2\n");
995 }
996 ioctl(mdfd, STOP_ARRAY, NULL);
997 goto abort;
998 }
999 }
1000 if (c->verbose >= 0)
1001 pr_err("array %s started.\n", mddev);
1002 if (st->ss->external && st->container_dev != NoMdDev) {
1003 if (need_mdmon)
1004 start_mdmon(st->container_dev);
1005
1006 ping_monitor_by_id(st->container_dev);
1007 close(container_fd);
1008 }
1009 wait_for(chosen_name, mdfd);
1010 } else {
1011 pr_err("not starting array - not enough devices.\n");
1012 }
1013 close(mdfd);
1014 return 0;
1015
1016 abort:
1017 map_lock(&map);
1018 abort_locked:
1019 map_remove(&map, fd2devnum(mdfd));
1020 map_unlock(&map);
1021
1022 if (mdfd >= 0)
1023 close(mdfd);
1024 return 1;
1025 }
Unnamed repository; edit this file 'description' to name the repository.