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