]>
Commit | Line | Data |
---|---|---|
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 "udev.h" | |
27 | #include "md_u.h" | |
28 | #include "md_p.h" | |
29 | #include <ctype.h> | |
30 | #include <fcntl.h> | |
31 | #include <signal.h> | |
32 | #include <sys/signalfd.h> | |
33 | #include <sys/wait.h> | |
34 | ||
35 | #ifndef FALLOC_FL_ZERO_RANGE | |
36 | #define FALLOC_FL_ZERO_RANGE 16 | |
37 | #endif | |
38 | ||
39 | static int round_size_and_verify(unsigned long long *size, int chunk) | |
40 | { | |
41 | if (*size == 0) | |
42 | return 0; | |
43 | *size &= ~(unsigned long long)(chunk - 1); | |
44 | if (*size == 0) { | |
45 | pr_err("Size cannot be smaller than chunk.\n"); | |
46 | return 1; | |
47 | } | |
48 | return 0; | |
49 | } | |
50 | ||
51 | /** | |
52 | * default_layout() - Get default layout for level. | |
53 | * @st: metadata requested, could be NULL. | |
54 | * @level: raid level requested. | |
55 | * @verbose: verbose level. | |
56 | * | |
57 | * Try to ask metadata handler first, otherwise use global defaults. | |
58 | * | |
59 | * Return: Layout or &UnSet, return value meaning depends of level used. | |
60 | */ | |
61 | int default_layout(struct supertype *st, int level, int verbose) | |
62 | { | |
63 | int layout = UnSet; | |
64 | mapping_t *layout_map = NULL; | |
65 | char *layout_name = NULL; | |
66 | ||
67 | if (st && st->ss->default_geometry) | |
68 | st->ss->default_geometry(st, &level, &layout, NULL); | |
69 | ||
70 | if (layout != UnSet) | |
71 | return layout; | |
72 | ||
73 | switch (level) { | |
74 | default: /* no layout */ | |
75 | layout = 0; | |
76 | break; | |
77 | case 0: | |
78 | layout = RAID0_ORIG_LAYOUT; | |
79 | break; | |
80 | case 10: | |
81 | layout = 0x102; /* near=2, far=1 */ | |
82 | layout_name = "n2"; | |
83 | break; | |
84 | case 5: | |
85 | case 6: | |
86 | layout_map = r5layout; | |
87 | break; | |
88 | case LEVEL_FAULTY: | |
89 | layout_map = faultylayout; | |
90 | break; | |
91 | } | |
92 | ||
93 | if (layout_map) { | |
94 | layout = map_name(layout_map, "default"); | |
95 | layout_name = map_num_s(layout_map, layout); | |
96 | } | |
97 | if (layout_name && verbose > 0) | |
98 | pr_err("layout defaults to %s\n", layout_name); | |
99 | ||
100 | return layout; | |
101 | } | |
102 | ||
103 | static pid_t write_zeroes_fork(int fd, struct shape *s, struct supertype *st, | |
104 | struct mddev_dev *dv) | |
105 | ||
106 | { | |
107 | const unsigned long long req_size = 1 << 30; | |
108 | unsigned long long offset_bytes, size_bytes, sz; | |
109 | sigset_t sigset; | |
110 | int ret = 0; | |
111 | pid_t pid; | |
112 | ||
113 | size_bytes = KIB_TO_BYTES(s->size); | |
114 | ||
115 | /* | |
116 | * If size_bytes is zero, this is a zoned raid array where | |
117 | * each disk is of a different size and uses its full | |
118 | * disk. Thus zero the entire disk. | |
119 | */ | |
120 | if (!size_bytes && !get_dev_size(fd, dv->devname, &size_bytes)) | |
121 | return -1; | |
122 | ||
123 | if (dv->data_offset != INVALID_SECTORS) | |
124 | offset_bytes = SEC_TO_BYTES(dv->data_offset); | |
125 | else | |
126 | offset_bytes = SEC_TO_BYTES(st->data_offset); | |
127 | ||
128 | pr_info("zeroing data from %lld to %lld on: %s\n", | |
129 | offset_bytes, size_bytes, dv->devname); | |
130 | ||
131 | pid = fork(); | |
132 | if (pid < 0) { | |
133 | pr_err("Could not fork to zero disks: %s\n", strerror(errno)); | |
134 | return pid; | |
135 | } else if (pid != 0) { | |
136 | return pid; | |
137 | } | |
138 | ||
139 | sigemptyset(&sigset); | |
140 | sigaddset(&sigset, SIGINT); | |
141 | sigprocmask(SIG_UNBLOCK, &sigset, NULL); | |
142 | ||
143 | while (size_bytes) { | |
144 | /* | |
145 | * Split requests to the kernel into 1GB chunks seeing the | |
146 | * fallocate() call is not interruptible and blocking a | |
147 | * ctrl-c for several minutes is not desirable. | |
148 | * | |
149 | * 1GB is chosen as a compromise: the user may still have | |
150 | * to wait several seconds if they ctrl-c on devices that | |
151 | * zero slowly, but will reduce the number of requests | |
152 | * required and thus the overhead on devices that perform | |
153 | * better. | |
154 | */ | |
155 | sz = size_bytes; | |
156 | if (sz >= req_size) | |
157 | sz = req_size; | |
158 | ||
159 | if (fallocate(fd, FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE, | |
160 | offset_bytes, sz)) { | |
161 | pr_err("zeroing %s failed: %s\n", dv->devname, | |
162 | strerror(errno)); | |
163 | ret = 1; | |
164 | break; | |
165 | } | |
166 | ||
167 | offset_bytes += sz; | |
168 | size_bytes -= sz; | |
169 | } | |
170 | ||
171 | exit(ret); | |
172 | } | |
173 | ||
174 | static int wait_for_zero_forks(int *zero_pids, int count) | |
175 | { | |
176 | int wstatus, ret = 0, i, sfd, wait_count = 0; | |
177 | struct signalfd_siginfo fdsi; | |
178 | bool interrupted = false; | |
179 | sigset_t sigset; | |
180 | ssize_t s; | |
181 | pid_t pid; | |
182 | ||
183 | for (i = 0; i < count; i++) | |
184 | if (zero_pids[i]) | |
185 | wait_count++; | |
186 | if (!wait_count) | |
187 | return 0; | |
188 | ||
189 | sigemptyset(&sigset); | |
190 | sigaddset(&sigset, SIGINT); | |
191 | sigaddset(&sigset, SIGCHLD); | |
192 | sigprocmask(SIG_BLOCK, &sigset, NULL); | |
193 | ||
194 | sfd = signalfd(-1, &sigset, 0); | |
195 | if (sfd < 0) { | |
196 | pr_err("Unable to create signalfd: %s\n", strerror(errno)); | |
197 | return 1; | |
198 | } | |
199 | ||
200 | while (wait_count) { | |
201 | s = read(sfd, &fdsi, sizeof(fdsi)); | |
202 | if (s != sizeof(fdsi)) { | |
203 | pr_err("Invalid signalfd read: %s\n", strerror(errno)); | |
204 | close(sfd); | |
205 | return 1; | |
206 | } | |
207 | ||
208 | if (fdsi.ssi_signo == SIGINT) { | |
209 | printf("\n"); | |
210 | pr_info("Interrupting zeroing processes, please wait...\n"); | |
211 | interrupted = true; | |
212 | } else if (fdsi.ssi_signo == SIGCHLD) { | |
213 | for (i = 0; i < count; i++) { | |
214 | if (!zero_pids[i]) | |
215 | continue; | |
216 | ||
217 | pid = waitpid(zero_pids[i], &wstatus, WNOHANG); | |
218 | if (pid <= 0) | |
219 | continue; | |
220 | ||
221 | zero_pids[i] = 0; | |
222 | if (!WIFEXITED(wstatus) || WEXITSTATUS(wstatus)) | |
223 | ret = 1; | |
224 | wait_count--; | |
225 | } | |
226 | } | |
227 | } | |
228 | ||
229 | close(sfd); | |
230 | ||
231 | if (interrupted) { | |
232 | pr_err("zeroing interrupted!\n"); | |
233 | return 1; | |
234 | } | |
235 | ||
236 | if (ret) | |
237 | pr_err("zeroing failed!\n"); | |
238 | else | |
239 | pr_info("zeroing finished\n"); | |
240 | ||
241 | return ret; | |
242 | } | |
243 | ||
244 | static int add_disk_to_super(int mdfd, struct shape *s, struct context *c, | |
245 | struct supertype *st, struct mddev_dev *dv, | |
246 | struct mdinfo *info, int have_container, int major_num, | |
247 | int *zero_pid) | |
248 | { | |
249 | dev_t rdev; | |
250 | int fd; | |
251 | ||
252 | if (dv->disposition == 'j') { | |
253 | info->disk.raid_disk = MD_DISK_ROLE_JOURNAL; | |
254 | info->disk.state = (1<<MD_DISK_JOURNAL); | |
255 | } else if (info->disk.raid_disk < s->raiddisks) { | |
256 | info->disk.state = (1<<MD_DISK_ACTIVE) | | |
257 | (1<<MD_DISK_SYNC); | |
258 | } else { | |
259 | info->disk.state = 0; | |
260 | } | |
261 | ||
262 | if (dv->writemostly == FlagSet) { | |
263 | if (major_num == BITMAP_MAJOR_CLUSTERED) { | |
264 | pr_err("Can not set %s --write-mostly with a clustered bitmap\n",dv->devname); | |
265 | return 1; | |
266 | } else { | |
267 | info->disk.state |= (1<<MD_DISK_WRITEMOSTLY); | |
268 | } | |
269 | ||
270 | } | |
271 | ||
272 | if (dv->failfast == FlagSet) | |
273 | info->disk.state |= (1<<MD_DISK_FAILFAST); | |
274 | ||
275 | if (have_container) { | |
276 | fd = -1; | |
277 | } else { | |
278 | if (st->ss->external && st->container_devnm[0]) | |
279 | fd = open(dv->devname, O_RDWR); | |
280 | else | |
281 | fd = open(dv->devname, O_RDWR|O_EXCL); | |
282 | ||
283 | if (fd < 0) { | |
284 | pr_err("failed to open %s after earlier success - aborting\n", | |
285 | dv->devname); | |
286 | return 1; | |
287 | } | |
288 | if (!fstat_is_blkdev(fd, dv->devname, &rdev)) { | |
289 | close(fd); | |
290 | return 1; | |
291 | } | |
292 | info->disk.major = major(rdev); | |
293 | info->disk.minor = minor(rdev); | |
294 | } | |
295 | if (fd >= 0) | |
296 | remove_partitions(fd); | |
297 | if (st->ss->add_to_super(st, &info->disk, fd, dv->devname, | |
298 | dv->data_offset)) { | |
299 | ioctl(mdfd, STOP_ARRAY, NULL); | |
300 | close(fd); | |
301 | return 1; | |
302 | } | |
303 | st->ss->getinfo_super(st, info, NULL); | |
304 | ||
305 | if (fd >= 0 && s->write_zeroes) { | |
306 | *zero_pid = write_zeroes_fork(fd, s, st, dv); | |
307 | if (*zero_pid <= 0) { | |
308 | ioctl(mdfd, STOP_ARRAY, NULL); | |
309 | close(fd); | |
310 | return 1; | |
311 | } | |
312 | } | |
313 | ||
314 | if (have_container && c->verbose > 0) | |
315 | pr_err("Using %s for device %d\n", | |
316 | map_dev(info->disk.major, info->disk.minor, 0), | |
317 | info->disk.number); | |
318 | ||
319 | if (!have_container) { | |
320 | /* getinfo_super might have lost these ... */ | |
321 | info->disk.major = major(rdev); | |
322 | info->disk.minor = minor(rdev); | |
323 | } | |
324 | ||
325 | return 0; | |
326 | } | |
327 | ||
328 | static int update_metadata(int mdfd, struct shape *s, struct supertype *st, | |
329 | struct map_ent **map, struct mdinfo *info, | |
330 | char *chosen_name) | |
331 | { | |
332 | struct mdinfo info_new; | |
333 | struct map_ent *me = NULL; | |
334 | ||
335 | /* check to see if the uuid has changed due to these | |
336 | * metadata changes, and if so update the member array | |
337 | * and container uuid. Note ->write_init_super clears | |
338 | * the subarray cursor such that ->getinfo_super once | |
339 | * again returns container info. | |
340 | */ | |
341 | st->ss->getinfo_super(st, &info_new, NULL); | |
342 | if (st->ss->external && !is_container(s->level) && | |
343 | !same_uuid(info_new.uuid, info->uuid, 0)) { | |
344 | map_update(map, fd2devnm(mdfd), | |
345 | info_new.text_version, | |
346 | info_new.uuid, chosen_name); | |
347 | me = map_by_devnm(map, st->container_devnm); | |
348 | } | |
349 | ||
350 | if (st->ss->write_init_super(st)) { | |
351 | st->ss->free_super(st); | |
352 | return 1; | |
353 | } | |
354 | ||
355 | /* | |
356 | * Before activating the array, perform extra steps | |
357 | * required to configure the internal write-intent | |
358 | * bitmap. | |
359 | */ | |
360 | if (info_new.consistency_policy == CONSISTENCY_POLICY_BITMAP && | |
361 | st->ss->set_bitmap && st->ss->set_bitmap(st, info)) { | |
362 | st->ss->free_super(st); | |
363 | return 1; | |
364 | } | |
365 | ||
366 | /* update parent container uuid */ | |
367 | if (me) { | |
368 | char *path = xstrdup(me->path); | |
369 | ||
370 | st->ss->getinfo_super(st, &info_new, NULL); | |
371 | map_update(map, st->container_devnm, info_new.text_version, | |
372 | info_new.uuid, path); | |
373 | free(path); | |
374 | } | |
375 | ||
376 | flush_metadata_updates(st); | |
377 | st->ss->free_super(st); | |
378 | ||
379 | return 0; | |
380 | } | |
381 | ||
382 | static int add_disks(int mdfd, struct mdinfo *info, struct shape *s, | |
383 | struct context *c, struct supertype *st, | |
384 | struct map_ent **map, struct mddev_dev *devlist, | |
385 | int total_slots, int have_container, int insert_point, | |
386 | int major_num, char *chosen_name) | |
387 | { | |
388 | struct mddev_dev *moved_disk = NULL; | |
389 | int pass, raid_disk_num, dnum; | |
390 | int zero_pids[total_slots]; | |
391 | struct mddev_dev *dv; | |
392 | struct mdinfo *infos; | |
393 | sigset_t sigset, orig_sigset; | |
394 | int ret = 0; | |
395 | ||
396 | /* | |
397 | * Block SIGINT so the main thread will always wait for the | |
398 | * zeroing processes when being interrupted. Otherwise the | |
399 | * zeroing processes will finish their work in the background | |
400 | * keeping the disk busy. | |
401 | */ | |
402 | sigemptyset(&sigset); | |
403 | sigaddset(&sigset, SIGINT); | |
404 | sigaddset(&sigset, SIGCHLD); | |
405 | sigprocmask(SIG_BLOCK, &sigset, &orig_sigset); | |
406 | memset(zero_pids, 0, sizeof(zero_pids)); | |
407 | ||
408 | infos = xmalloc(sizeof(*infos) * total_slots); | |
409 | enable_fds(total_slots); | |
410 | for (pass = 1; pass <= 2; pass++) { | |
411 | for (dnum = 0, raid_disk_num = 0, dv = devlist; dv; | |
412 | dv = (dv->next) ? (dv->next) : moved_disk, dnum++) { | |
413 | if (dnum >= total_slots) | |
414 | abort(); | |
415 | if (dnum == insert_point) { | |
416 | raid_disk_num += 1; | |
417 | moved_disk = dv; | |
418 | continue; | |
419 | } | |
420 | if (strcasecmp(dv->devname, "missing") == 0) { | |
421 | raid_disk_num += 1; | |
422 | continue; | |
423 | } | |
424 | if (have_container) | |
425 | moved_disk = NULL; | |
426 | if (have_container && dnum < total_slots - 1) | |
427 | /* repeatedly use the container */ | |
428 | moved_disk = dv; | |
429 | ||
430 | switch(pass) { | |
431 | case 1: | |
432 | infos[dnum] = *info; | |
433 | infos[dnum].disk.number = dnum; | |
434 | infos[dnum].disk.raid_disk = raid_disk_num++; | |
435 | ||
436 | if (dv->disposition == 'j') | |
437 | raid_disk_num--; | |
438 | ||
439 | ret = add_disk_to_super(mdfd, s, c, st, dv, | |
440 | &infos[dnum], have_container, | |
441 | major_num, &zero_pids[dnum]); | |
442 | if (ret) | |
443 | goto out; | |
444 | ||
445 | break; | |
446 | case 2: | |
447 | infos[dnum].errors = 0; | |
448 | ||
449 | ret = add_disk(mdfd, st, info, &infos[dnum]); | |
450 | if (ret) { | |
451 | pr_err("ADD_NEW_DISK for %s failed: %s\n", | |
452 | dv->devname, strerror(errno)); | |
453 | if (errno == EINVAL && | |
454 | info->array.level == 0) { | |
455 | pr_err("Possibly your kernel doesn't support RAID0 layouts.\n"); | |
456 | pr_err("Either upgrade, or use --layout=dangerous\n"); | |
457 | } | |
458 | goto out; | |
459 | } | |
460 | break; | |
461 | } | |
462 | if (!have_container && | |
463 | dv == moved_disk && dnum != insert_point) break; | |
464 | } | |
465 | ||
466 | if (pass == 1) { | |
467 | ret = wait_for_zero_forks(zero_pids, total_slots); | |
468 | if (ret) | |
469 | goto out; | |
470 | ||
471 | ret = update_metadata(mdfd, s, st, map, info, | |
472 | chosen_name); | |
473 | if (ret) | |
474 | goto out; | |
475 | } | |
476 | } | |
477 | ||
478 | out: | |
479 | if (ret) | |
480 | wait_for_zero_forks(zero_pids, total_slots); | |
481 | free(infos); | |
482 | sigprocmask(SIG_SETMASK, &orig_sigset, NULL); | |
483 | return ret; | |
484 | } | |
485 | ||
486 | int Create(struct supertype *st, struct mddev_ident *ident, int subdevs, | |
487 | struct mddev_dev *devlist, struct shape *s, struct context *c) | |
488 | { | |
489 | /* | |
490 | * Create a new raid array. | |
491 | * | |
492 | * First check that necessary details are available | |
493 | * (i.e. level, raid-disks) | |
494 | * | |
495 | * Then check each disk to see what might be on it | |
496 | * and report anything interesting. | |
497 | * | |
498 | * If anything looks odd, and runstop not set, | |
499 | * abort. | |
500 | * | |
501 | * SET_ARRAY_INFO and ADD_NEW_DISK, and | |
502 | * if runstop==run, or raiddisks disks were used, | |
503 | * RUN_ARRAY | |
504 | */ | |
505 | int mdfd; | |
506 | unsigned long long minsize = 0, maxsize = 0; | |
507 | dev_policy_t *custom_pols = NULL; | |
508 | char *mindisc = NULL; | |
509 | char *maxdisc = NULL; | |
510 | char *name = ident->name; | |
511 | int *uuid = ident->uuid_set == 1 ? ident->uuid : NULL; | |
512 | int dnum; | |
513 | struct mddev_dev *dv; | |
514 | dev_t rdev; | |
515 | int fail = 0, warn = 0; | |
516 | int first_missing = subdevs * 2; | |
517 | int second_missing = subdevs * 2; | |
518 | int missing_disks = 0; | |
519 | int insert_point = subdevs * 2; /* where to insert a missing drive */ | |
520 | int total_slots; | |
521 | int rv; | |
522 | int bitmap_fd; | |
523 | int have_container = 0; | |
524 | int container_fd = -1; | |
525 | int need_mdmon = 0; | |
526 | unsigned long long bitmapsize; | |
527 | struct mdinfo info; | |
528 | int did_default = 0; | |
529 | int do_default_layout = 0; | |
530 | int do_default_chunk = 0; | |
531 | char chosen_name[1024]; | |
532 | struct map_ent *map = NULL; | |
533 | unsigned long long newsize; | |
534 | mdu_array_info_t inf; | |
535 | ||
536 | int major_num = BITMAP_MAJOR_HI; | |
537 | if (s->bitmap_file && strcmp(s->bitmap_file, "clustered") == 0) { | |
538 | major_num = BITMAP_MAJOR_CLUSTERED; | |
539 | if (c->nodes <= 1) { | |
540 | pr_err("At least 2 nodes are needed for cluster-md\n"); | |
541 | return 1; | |
542 | } | |
543 | } | |
544 | ||
545 | memset(&info, 0, sizeof(info)); | |
546 | if (s->level == UnSet && st && st->ss->default_geometry) | |
547 | st->ss->default_geometry(st, &s->level, NULL, NULL); | |
548 | if (s->level == UnSet) { | |
549 | pr_err("a RAID level is needed to create an array.\n"); | |
550 | return 1; | |
551 | } | |
552 | if (s->raiddisks < 4 && s->level == 6) { | |
553 | pr_err("at least 4 raid-devices needed for level 6\n"); | |
554 | return 1; | |
555 | } | |
556 | if (s->raiddisks > 256 && s->level == 6) { | |
557 | pr_err("no more than 256 raid-devices supported for level 6\n"); | |
558 | return 1; | |
559 | } | |
560 | if (s->raiddisks < 2 && s->level >= 4) { | |
561 | pr_err("at least 2 raid-devices needed for level %d\n", s->level); | |
562 | return 1; | |
563 | } | |
564 | if (s->level <= 0 && s->sparedisks) { | |
565 | pr_err("This level does not support spare devices\n"); | |
566 | return 1; | |
567 | } | |
568 | ||
569 | if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) { | |
570 | /* If given a single device, it might be a container, and we can | |
571 | * extract a device list from there | |
572 | */ | |
573 | int fd; | |
574 | ||
575 | memset(&inf, 0, sizeof(inf)); | |
576 | fd = open(devlist->devname, O_RDONLY); | |
577 | if (fd >= 0 && | |
578 | md_get_array_info(fd, &inf) == 0 && inf.raid_disks == 0) { | |
579 | /* yep, looks like a container */ | |
580 | if (st) { | |
581 | rv = st->ss->load_container(st, fd, | |
582 | devlist->devname); | |
583 | if (rv == 0) | |
584 | have_container = 1; | |
585 | } else { | |
586 | st = super_by_fd(fd, NULL); | |
587 | if (st && !(rv = st->ss-> | |
588 | load_container(st, fd, | |
589 | devlist->devname))) | |
590 | have_container = 1; | |
591 | else | |
592 | st = NULL; | |
593 | } | |
594 | if (have_container) { | |
595 | subdevs = s->raiddisks; | |
596 | first_missing = subdevs * 2; | |
597 | second_missing = subdevs * 2; | |
598 | insert_point = subdevs * 2; | |
599 | ||
600 | if (mddev_test_and_add_drive_policies(st, &custom_pols, fd, 1)) | |
601 | exit(1); | |
602 | } | |
603 | } | |
604 | if (fd >= 0) | |
605 | close(fd); | |
606 | } | |
607 | if (st && st->ss->external && s->sparedisks) { | |
608 | pr_err("This metadata type does not support spare disks at create time\n"); | |
609 | return 1; | |
610 | } | |
611 | if (subdevs > s->raiddisks+s->sparedisks+s->journaldisks) { | |
612 | pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks); | |
613 | return 1; | |
614 | } | |
615 | if (!have_container && subdevs < s->raiddisks+s->sparedisks+s->journaldisks) { | |
616 | pr_err("You haven't given enough devices (real or missing) to create this array\n"); | |
617 | return 1; | |
618 | } | |
619 | if (s->bitmap_file && s->level <= 0) { | |
620 | pr_err("bitmaps not meaningful with level %s\n", | |
621 | map_num(pers, s->level)?:"given"); | |
622 | return 1; | |
623 | } | |
624 | ||
625 | /* now set some defaults */ | |
626 | ||
627 | if (s->layout == UnSet) { | |
628 | do_default_layout = 1; | |
629 | s->layout = default_layout(st, s->level, c->verbose); | |
630 | } | |
631 | ||
632 | if (s->level == 10) | |
633 | /* check layout fits in array*/ | |
634 | if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) { | |
635 | pr_err("that layout requires at least %d devices\n", | |
636 | (s->layout&255) * ((s->layout>>8)&255)); | |
637 | return 1; | |
638 | } | |
639 | ||
640 | switch(s->level) { | |
641 | case 4: | |
642 | case 5: | |
643 | case 10: | |
644 | case 6: | |
645 | case 0: | |
646 | if (s->chunk == 0 || s->chunk == UnSet) { | |
647 | s->chunk = UnSet; | |
648 | do_default_chunk = 1; | |
649 | /* chunk will be set later */ | |
650 | } | |
651 | break; | |
652 | case LEVEL_LINEAR: | |
653 | /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */ | |
654 | break; | |
655 | case 1: | |
656 | case LEVEL_FAULTY: | |
657 | case LEVEL_MULTIPATH: | |
658 | case LEVEL_CONTAINER: | |
659 | if (s->chunk) { | |
660 | pr_err("specifying chunk size is forbidden for this level\n"); | |
661 | return 1; | |
662 | } | |
663 | break; | |
664 | default: | |
665 | pr_err("unknown level %d\n", s->level); | |
666 | return 1; | |
667 | } | |
668 | ||
669 | if (s->size == MAX_SIZE) | |
670 | /* use '0' to mean 'max' now... */ | |
671 | s->size = 0; | |
672 | if (s->size && s->chunk && s->chunk != UnSet) | |
673 | if (round_size_and_verify(&s->size, s->chunk)) | |
674 | return 1; | |
675 | ||
676 | newsize = s->size * 2; | |
677 | if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks, | |
678 | &s->chunk, s->size*2, | |
679 | s->data_offset, NULL, | |
680 | &newsize, s->consistency_policy, | |
681 | c->verbose >= 0)) | |
682 | return 1; | |
683 | ||
684 | if (s->chunk && s->chunk != UnSet) { | |
685 | newsize &= ~(unsigned long long)(s->chunk*2 - 1); | |
686 | if (do_default_chunk) { | |
687 | /* default chunk was just set */ | |
688 | if (c->verbose > 0) | |
689 | pr_err("chunk size defaults to %dK\n", s->chunk); | |
690 | if (round_size_and_verify(&s->size, s->chunk)) | |
691 | return 1; | |
692 | do_default_chunk = 0; | |
693 | } | |
694 | } | |
695 | ||
696 | if (s->size == 0) { | |
697 | s->size = newsize / 2; | |
698 | if (s->level == 1) | |
699 | /* If this is ever reshaped to RAID5, we will | |
700 | * need a chunksize. So round it off a bit | |
701 | * now just to be safe | |
702 | */ | |
703 | s->size &= ~(64ULL-1); | |
704 | ||
705 | if (s->size && c->verbose > 0) | |
706 | pr_err("setting size to %lluK\n", s->size); | |
707 | } | |
708 | ||
709 | /* now look at the subdevs */ | |
710 | info.array.active_disks = 0; | |
711 | info.array.working_disks = 0; | |
712 | dnum = 0; | |
713 | for (dv = devlist; dv; dv = dv->next) | |
714 | if (s->data_offset == VARIABLE_OFFSET) | |
715 | dv->data_offset = INVALID_SECTORS; | |
716 | else | |
717 | dv->data_offset = s->data_offset; | |
718 | ||
719 | for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) { | |
720 | char *dname = dv->devname; | |
721 | unsigned long long freesize; | |
722 | int dfd; | |
723 | char *doff; | |
724 | ||
725 | if (strcasecmp(dname, "missing") == 0) { | |
726 | if (first_missing > dnum) | |
727 | first_missing = dnum; | |
728 | if (second_missing > dnum && dnum > first_missing) | |
729 | second_missing = dnum; | |
730 | missing_disks ++; | |
731 | continue; | |
732 | } | |
733 | if (s->data_offset == VARIABLE_OFFSET) { | |
734 | doff = strchr(dname, ':'); | |
735 | if (doff) { | |
736 | *doff++ = 0; | |
737 | dv->data_offset = parse_size(doff); | |
738 | } else | |
739 | dv->data_offset = INVALID_SECTORS; | |
740 | } else | |
741 | dv->data_offset = s->data_offset; | |
742 | ||
743 | dfd = open(dname, O_RDONLY); | |
744 | if (dfd < 0) { | |
745 | pr_err("cannot open %s: %s\n", | |
746 | dname, strerror(errno)); | |
747 | exit(2); | |
748 | } | |
749 | if (!fstat_is_blkdev(dfd, dname, NULL)) { | |
750 | close(dfd); | |
751 | exit(2); | |
752 | } | |
753 | ||
754 | info.array.working_disks++; | |
755 | if (dnum < s->raiddisks && dv->disposition != 'j') | |
756 | info.array.active_disks++; | |
757 | if (st == NULL) { | |
758 | struct createinfo *ci = conf_get_create_info(); | |
759 | if (ci) | |
760 | st = ci->supertype; | |
761 | } | |
762 | if (st == NULL) { | |
763 | /* Need to choose a default metadata, which is different | |
764 | * depending on geometry of array. | |
765 | */ | |
766 | int i; | |
767 | char *name = "default"; | |
768 | for(i = 0; !st && superlist[i]; i++) { | |
769 | st = superlist[i]->match_metadata_desc(name); | |
770 | if (!st) | |
771 | continue; | |
772 | if (do_default_layout) | |
773 | s->layout = default_layout(st, s->level, c->verbose); | |
774 | switch (st->ss->validate_geometry( | |
775 | st, s->level, s->layout, s->raiddisks, | |
776 | &s->chunk, s->size*2, | |
777 | dv->data_offset, dname, | |
778 | &freesize, s->consistency_policy, | |
779 | c->verbose > 0)) { | |
780 | case -1: /* Not valid, message printed, and not | |
781 | * worth checking any further */ | |
782 | exit(2); | |
783 | break; | |
784 | case 0: /* Geometry not valid */ | |
785 | free(st); | |
786 | st = NULL; | |
787 | s->chunk = do_default_chunk ? UnSet : s->chunk; | |
788 | break; | |
789 | case 1: /* All happy */ | |
790 | break; | |
791 | } | |
792 | } | |
793 | ||
794 | if (!st) { | |
795 | int dfd = open(dname, O_RDONLY|O_EXCL); | |
796 | if (dfd < 0) { | |
797 | pr_err("cannot open %s: %s\n", | |
798 | dname, strerror(errno)); | |
799 | exit(2); | |
800 | } | |
801 | pr_err("device %s not suitable for any style of array\n", | |
802 | dname); | |
803 | exit(2); | |
804 | } | |
805 | if (st->ss != &super0 || | |
806 | st->minor_version != 90) | |
807 | did_default = 1; | |
808 | } else { | |
809 | if (do_default_layout) | |
810 | s->layout = default_layout(st, s->level, 0); | |
811 | if (!st->ss->validate_geometry(st, s->level, s->layout, | |
812 | s->raiddisks, | |
813 | &s->chunk, s->size*2, | |
814 | dv->data_offset, | |
815 | dname, &freesize, | |
816 | s->consistency_policy, | |
817 | c->verbose >= 0)) { | |
818 | ||
819 | pr_err("%s is not suitable for this array.\n", | |
820 | dname); | |
821 | fail = 1; | |
822 | continue; | |
823 | } | |
824 | } | |
825 | ||
826 | if (drive_test_and_add_policies(st, &custom_pols, dfd, 1)) | |
827 | exit(1); | |
828 | ||
829 | close(dfd); | |
830 | ||
831 | if (dv->disposition == 'j') | |
832 | goto skip_size_check; /* skip write journal for size check */ | |
833 | ||
834 | freesize /= 2; /* convert to K */ | |
835 | if (s->chunk && s->chunk != UnSet) { | |
836 | /* round to chunk size */ | |
837 | freesize = freesize & ~(s->chunk-1); | |
838 | if (do_default_chunk) { | |
839 | /* default chunk was just set */ | |
840 | if (c->verbose > 0) | |
841 | pr_err("chunk size defaults to %dK\n", s->chunk); | |
842 | if (round_size_and_verify(&s->size, s->chunk)) | |
843 | return 1; | |
844 | do_default_chunk = 0; | |
845 | } | |
846 | } | |
847 | if (!freesize) { | |
848 | pr_err("no free space left on %s\n", dname); | |
849 | fail = 1; | |
850 | continue; | |
851 | } | |
852 | ||
853 | if (s->size && freesize < s->size) { | |
854 | pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n", | |
855 | dname, freesize, s->size); | |
856 | fail = 1; | |
857 | continue; | |
858 | } | |
859 | if (maxdisc == NULL || (maxdisc && freesize > maxsize)) { | |
860 | maxdisc = dname; | |
861 | maxsize = freesize; | |
862 | } | |
863 | if (mindisc ==NULL || (mindisc && freesize < minsize)) { | |
864 | mindisc = dname; | |
865 | minsize = freesize; | |
866 | } | |
867 | skip_size_check: | |
868 | if (c->runstop != 1 || c->verbose >= 0) { | |
869 | int fd = open(dname, O_RDONLY); | |
870 | if (fd < 0) { | |
871 | pr_err("Cannot open %s: %s\n", | |
872 | dname, strerror(errno)); | |
873 | fail = 1; | |
874 | continue; | |
875 | } | |
876 | warn |= check_ext2(fd, dname); | |
877 | warn |= check_reiser(fd, dname); | |
878 | warn |= check_raid(fd, dname); | |
879 | if (strcmp(st->ss->name, "1.x") == 0 && | |
880 | st->minor_version >= 1) | |
881 | /* metadata at front */ | |
882 | warn |= check_partitions(fd, dname, 0, 0); | |
883 | else if (s->level == 1 || is_container(s->level) || | |
884 | (s->level == 0 && s->raiddisks == 1)) | |
885 | /* partitions could be meaningful */ | |
886 | warn |= check_partitions(fd, dname, freesize*2, s->size*2); | |
887 | else | |
888 | /* partitions cannot be meaningful */ | |
889 | warn |= check_partitions(fd, dname, 0, 0); | |
890 | if (strcmp(st->ss->name, "1.x") == 0 && | |
891 | st->minor_version >= 1 && | |
892 | did_default && | |
893 | s->level == 1 && | |
894 | (warn & 1024) == 0) { | |
895 | warn |= 1024; | |
896 | pr_err("Note: this array has metadata at the start and\n" | |
897 | " may not be suitable as a boot device. If you plan to\n" | |
898 | " store '/boot' on this device please ensure that\n" | |
899 | " your boot-loader understands md/v1.x metadata, or use\n" | |
900 | " --metadata=0.90\n"); | |
901 | } | |
902 | close(fd); | |
903 | } | |
904 | } | |
905 | ||
906 | if (missing_disks == dnum && !have_container) { | |
907 | pr_err("Subdevs can't be all missing\n"); | |
908 | return 1; | |
909 | } | |
910 | if (s->raiddisks + s->sparedisks > st->max_devs) { | |
911 | pr_err("Too many devices: %s metadata only supports %d\n", | |
912 | st->ss->name, st->max_devs); | |
913 | return 1; | |
914 | } | |
915 | if (have_container) | |
916 | info.array.working_disks = s->raiddisks; | |
917 | if (fail) { | |
918 | pr_err("create aborted\n"); | |
919 | return 1; | |
920 | } | |
921 | if (s->size == 0) { | |
922 | if (mindisc == NULL && !have_container) { | |
923 | pr_err("no size and no drives given - aborting create.\n"); | |
924 | return 1; | |
925 | } | |
926 | if (s->level > 0 || s->level == LEVEL_MULTIPATH || | |
927 | s->level == LEVEL_FAULTY || st->ss->external) { | |
928 | /* size is meaningful */ | |
929 | if (!st->ss->validate_geometry(st, s->level, s->layout, | |
930 | s->raiddisks, | |
931 | &s->chunk, minsize*2, | |
932 | s->data_offset, | |
933 | NULL, NULL, | |
934 | s->consistency_policy, 0)) { | |
935 | pr_err("devices too large for RAID level %d\n", s->level); | |
936 | return 1; | |
937 | } | |
938 | s->size = minsize; | |
939 | if (s->level == 1) | |
940 | /* If this is ever reshaped to RAID5, we will | |
941 | * need a chunksize. So round it off a bit | |
942 | * now just to be safe | |
943 | */ | |
944 | s->size &= ~(64ULL-1); | |
945 | if (c->verbose > 0) | |
946 | pr_err("size set to %lluK\n", s->size); | |
947 | } | |
948 | } | |
949 | ||
950 | if (!s->bitmap_file && | |
951 | !st->ss->external && | |
952 | s->level >= 1 && | |
953 | st->ss->add_internal_bitmap && | |
954 | s->journaldisks == 0 && | |
955 | (s->consistency_policy != CONSISTENCY_POLICY_RESYNC && | |
956 | s->consistency_policy != CONSISTENCY_POLICY_PPL) && | |
957 | (s->write_behind || s->size > 100*1024*1024ULL)) { | |
958 | if (c->verbose > 0) | |
959 | pr_err("automatically enabling write-intent bitmap on large array\n"); | |
960 | s->bitmap_file = "internal"; | |
961 | } | |
962 | if (s->bitmap_file && str_is_none(s->bitmap_file) == true) | |
963 | s->bitmap_file = NULL; | |
964 | ||
965 | if (s->consistency_policy == CONSISTENCY_POLICY_PPL && | |
966 | !st->ss->write_init_ppl) { | |
967 | pr_err("%s metadata does not support PPL\n", st->ss->name); | |
968 | return 1; | |
969 | } | |
970 | ||
971 | if (st->ss == &super_imsm && s->level == 10 && s->raiddisks > 4) { | |
972 | /* Print no matter runstop was specifed */ | |
973 | pr_err("Warning! VROC UEFI driver does not support RAID10 in requested layout.\n"); | |
974 | pr_err("Array won't be suitable as boot device.\n"); | |
975 | warn = 1; | |
976 | } | |
977 | ||
978 | if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) { | |
979 | if (c->runstop != 1 || c->verbose >= 0) | |
980 | pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n", | |
981 | maxdisc, s->size); | |
982 | warn = 1; | |
983 | } | |
984 | ||
985 | if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) { | |
986 | if (c->runstop != 1 || c->verbose >= 0) | |
987 | pr_err("%s unable to enumerate platform support\n" | |
988 | " array may not be compatible with hardware/firmware\n", | |
989 | st->ss->name); | |
990 | warn = 1; | |
991 | } | |
992 | st->nodes = c->nodes; | |
993 | st->cluster_name = c->homecluster; | |
994 | ||
995 | if (warn) { | |
996 | if (c->runstop!= 1) { | |
997 | if (!ask("Continue creating array")) { | |
998 | pr_err("create aborted.\n"); | |
999 | return 1; | |
1000 | } | |
1001 | } else { | |
1002 | if (c->verbose > 0) | |
1003 | pr_err("creation continuing despite oddities due to --run\n"); | |
1004 | } | |
1005 | } | |
1006 | ||
1007 | /* If this is raid4/5, we want to configure the last active slot | |
1008 | * as missing, so that a reconstruct happens (faster than re-parity) | |
1009 | * FIX: Can we do this for raid6 as well? | |
1010 | */ | |
1011 | if (st->ss->external == 0 && s->assume_clean == 0 && | |
1012 | c->force == 0 && first_missing >= s->raiddisks) { | |
1013 | switch (s->level) { | |
1014 | case 4: | |
1015 | case 5: | |
1016 | insert_point = s->raiddisks-1; | |
1017 | s->sparedisks++; | |
1018 | info.array.active_disks--; | |
1019 | missing_disks++; | |
1020 | break; | |
1021 | default: | |
1022 | break; | |
1023 | } | |
1024 | } | |
1025 | /* For raid6, if creating with 1 missing drive, make a good drive | |
1026 | * into a spare, else the create will fail | |
1027 | */ | |
1028 | if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks && | |
1029 | st->ss->external == 0 && | |
1030 | second_missing >= s->raiddisks && s->level == 6) { | |
1031 | insert_point = s->raiddisks - 1; | |
1032 | if (insert_point == first_missing) | |
1033 | insert_point--; | |
1034 | s->sparedisks ++; | |
1035 | info.array.active_disks--; | |
1036 | missing_disks++; | |
1037 | } | |
1038 | ||
1039 | if (s->level <= 0 && first_missing < subdevs * 2) { | |
1040 | pr_err("This level does not support missing devices\n"); | |
1041 | return 1; | |
1042 | } | |
1043 | ||
1044 | /* We need to create the device */ | |
1045 | map_lock(&map); | |
1046 | mdfd = create_mddev(ident->devname, ident->name, c->autof, LOCAL, chosen_name, 1); | |
1047 | if (mdfd < 0) { | |
1048 | map_unlock(&map); | |
1049 | return 1; | |
1050 | } | |
1051 | /* verify if chosen_name is not in use, | |
1052 | * it could be in conflict with already existing device | |
1053 | * e.g. container, array | |
1054 | */ | |
1055 | if (strncmp(chosen_name, DEV_MD_DIR, DEV_MD_DIR_LEN) == 0 && | |
1056 | map_by_name(&map, chosen_name + DEV_MD_DIR_LEN)) { | |
1057 | pr_err("Array name %s is in use already.\n", chosen_name); | |
1058 | close(mdfd); | |
1059 | map_unlock(&map); | |
1060 | udev_unblock(); | |
1061 | return 1; | |
1062 | } | |
1063 | ||
1064 | memset(&inf, 0, sizeof(inf)); | |
1065 | md_get_array_info(mdfd, &inf); | |
1066 | if (inf.working_disks != 0) { | |
1067 | pr_err("another array by this name is already running.\n"); | |
1068 | goto abort_locked; | |
1069 | } | |
1070 | ||
1071 | /* Ok, lets try some ioctls */ | |
1072 | ||
1073 | info.array.level = s->level; | |
1074 | info.array.size = s->size; | |
1075 | info.array.raid_disks = s->raiddisks; | |
1076 | /* The kernel should *know* what md_minor we are dealing | |
1077 | * with, but it chooses to trust me instead. Sigh | |
1078 | */ | |
1079 | info.array.md_minor = 0; | |
1080 | if (fstat_is_blkdev(mdfd, chosen_name, &rdev)) | |
1081 | info.array.md_minor = minor(rdev); | |
1082 | info.array.not_persistent = 0; | |
1083 | ||
1084 | if (((s->level == 4 || s->level == 5) && | |
1085 | (insert_point < s->raiddisks || first_missing < s->raiddisks)) || | |
1086 | (s->level == 6 && (insert_point < s->raiddisks || | |
1087 | second_missing < s->raiddisks)) || | |
1088 | (s->level <= 0) || s->assume_clean) { | |
1089 | info.array.state = 1; /* clean, but one+ drive will be missing*/ | |
1090 | info.resync_start = MaxSector; | |
1091 | } else { | |
1092 | info.array.state = 0; /* not clean, but no errors */ | |
1093 | info.resync_start = 0; | |
1094 | } | |
1095 | if (s->level == 10) { | |
1096 | /* for raid10, the bitmap size is the capacity of the array, | |
1097 | * which is array.size * raid_disks / ncopies; | |
1098 | * .. but convert to sectors. | |
1099 | */ | |
1100 | int ncopies = ((s->layout>>8) & 255) * (s->layout & 255); | |
1101 | bitmapsize = s->size * s->raiddisks / ncopies * 2; | |
1102 | /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/ | |
1103 | } else | |
1104 | bitmapsize = s->size * 2; | |
1105 | ||
1106 | /* There is lots of redundancy in these disk counts, | |
1107 | * raid_disks is the most meaningful value | |
1108 | * it describes the geometry of the array | |
1109 | * it is constant | |
1110 | * nr_disks is total number of used slots. | |
1111 | * it should be raid_disks+spare_disks | |
1112 | * spare_disks is the number of extra disks present | |
1113 | * see above | |
1114 | * active_disks is the number of working disks in | |
1115 | * active slots. (With raid_disks) | |
1116 | * working_disks is the total number of working disks, | |
1117 | * including spares | |
1118 | * failed_disks is the number of disks marked failed | |
1119 | * | |
1120 | * Ideally, the kernel would keep these (except raid_disks) | |
1121 | * up-to-date as we ADD_NEW_DISK, but it doesn't (yet). | |
1122 | * So for now, we assume that all raid and spare | |
1123 | * devices will be given. | |
1124 | */ | |
1125 | info.array.spare_disks=s->sparedisks; | |
1126 | info.array.failed_disks=missing_disks; | |
1127 | info.array.nr_disks = info.array.working_disks | |
1128 | + info.array.failed_disks; | |
1129 | info.array.layout = s->layout; | |
1130 | info.array.chunk_size = s->chunk*1024; | |
1131 | ||
1132 | if (*name == 0) { | |
1133 | /* base name on devname */ | |
1134 | /* /dev/md0 -> 0 | |
1135 | * /dev/md_d0 -> d0 | |
1136 | * /dev/md_foo -> foo | |
1137 | * /dev/md/1 -> 1 | |
1138 | * /dev/md/d1 -> d1 | |
1139 | * /dev/md/home -> home | |
1140 | * /dev/mdhome -> home | |
1141 | */ | |
1142 | /* FIXME compare this with rules in create_mddev */ | |
1143 | name = strrchr(chosen_name, '/'); | |
1144 | ||
1145 | if (name) { | |
1146 | name++; | |
1147 | if (strncmp(name, "md_", 3) == 0 && | |
1148 | strlen(name) > 3 && (name - chosen_name) == 5 /* /dev/ */) | |
1149 | name += 3; | |
1150 | else if (strncmp(name, "md", 2) == 0 && | |
1151 | strlen(name) > 2 && isdigit(name[2]) && | |
1152 | (name - chosen_name) == 5 /* /dev/ */) | |
1153 | name += 2; | |
1154 | } | |
1155 | } | |
1156 | if (!st->ss->init_super(st, &info.array, s, name, c->homehost, uuid, | |
1157 | s->data_offset)) | |
1158 | goto abort_locked; | |
1159 | ||
1160 | total_slots = info.array.nr_disks; | |
1161 | st->ss->getinfo_super(st, &info, NULL); | |
1162 | if (sysfs_init(&info, mdfd, NULL)) { | |
1163 | pr_err("unable to initialize sysfs\n"); | |
1164 | goto abort_locked; | |
1165 | } | |
1166 | ||
1167 | if (did_default) { | |
1168 | if (is_subarray(info.text_version)) { | |
1169 | char devnm[MD_NAME_MAX]; | |
1170 | struct mdinfo *mdi; | |
1171 | ||
1172 | sysfs_get_container_devnm(&info, devnm); | |
1173 | ||
1174 | mdi = sysfs_read(-1, devnm, GET_VERSION | GET_DEVS); | |
1175 | if (!mdi) { | |
1176 | pr_err("Cannot open sysfs for container %s\n", devnm); | |
1177 | goto abort_locked; | |
1178 | } | |
1179 | ||
1180 | if (sysfs_test_and_add_drive_policies(st, &custom_pols, mdi, 1)) | |
1181 | goto abort_locked; | |
1182 | ||
1183 | if (c->verbose >= 0) | |
1184 | pr_info("Creating array inside %s container /dev/%s\n", | |
1185 | mdi->text_version, devnm); | |
1186 | ||
1187 | sysfs_free(mdi); | |
1188 | } else if (c->verbose >= 0) { | |
1189 | pr_info("Defaulting to version %s metadata\n", info.text_version); | |
1190 | } | |
1191 | } | |
1192 | ||
1193 | map_update(&map, fd2devnm(mdfd), info.text_version, | |
1194 | info.uuid, chosen_name); | |
1195 | /* Keep map locked until devices have been added to array | |
1196 | * to stop another mdadm from finding and using those devices. | |
1197 | */ | |
1198 | ||
1199 | if (s->bitmap_file && (strcmp(s->bitmap_file, "internal") == 0 || | |
1200 | strcmp(s->bitmap_file, "clustered") == 0)) { | |
1201 | if (!st->ss->add_internal_bitmap) { | |
1202 | pr_err("internal bitmaps not supported with %s metadata\n", | |
1203 | st->ss->name); | |
1204 | goto abort_locked; | |
1205 | } | |
1206 | if (st->ss->add_internal_bitmap(st, &s->bitmap_chunk, | |
1207 | c->delay, s->write_behind, | |
1208 | bitmapsize, 1, major_num)) { | |
1209 | pr_err("Given bitmap chunk size not supported.\n"); | |
1210 | goto abort_locked; | |
1211 | } | |
1212 | s->bitmap_file = NULL; | |
1213 | } | |
1214 | ||
1215 | if (sysfs_init(&info, mdfd, NULL)) { | |
1216 | pr_err("unable to initialize sysfs\n"); | |
1217 | goto abort_locked; | |
1218 | } | |
1219 | ||
1220 | if (st->ss->external && st->container_devnm[0]) { | |
1221 | /* member */ | |
1222 | ||
1223 | /* When creating a member, we need to be careful | |
1224 | * to negotiate with mdmon properly. | |
1225 | * If it is already running, we cannot write to | |
1226 | * the devices and must ask it to do that part. | |
1227 | * If it isn't running, we write to the devices, | |
1228 | * and then start it. | |
1229 | * We hold an exclusive open on the container | |
1230 | * device to make sure mdmon doesn't exit after | |
1231 | * we checked that it is running. | |
1232 | * | |
1233 | * For now, fail if it is already running. | |
1234 | */ | |
1235 | container_fd = open_dev_excl(st->container_devnm); | |
1236 | if (container_fd < 0) { | |
1237 | pr_err("Cannot get exclusive open on container - weird.\n"); | |
1238 | goto abort_locked; | |
1239 | } | |
1240 | if (mdmon_running(st->container_devnm)) { | |
1241 | if (c->verbose) | |
1242 | pr_err("reusing mdmon for %s.\n", | |
1243 | st->container_devnm); | |
1244 | st->update_tail = &st->updates; | |
1245 | } else | |
1246 | need_mdmon = 1; | |
1247 | } | |
1248 | rv = set_array_info(mdfd, st, &info); | |
1249 | if (rv) { | |
1250 | pr_err("failed to set array info for %s: %s\n", chosen_name, strerror(errno)); | |
1251 | goto abort_locked; | |
1252 | } | |
1253 | ||
1254 | if (s->bitmap_file) { | |
1255 | int uuid[4]; | |
1256 | ||
1257 | st->ss->uuid_from_super(st, uuid); | |
1258 | if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk, | |
1259 | c->delay, s->write_behind, | |
1260 | bitmapsize, | |
1261 | major_num)) { | |
1262 | goto abort_locked; | |
1263 | } | |
1264 | bitmap_fd = open(s->bitmap_file, O_RDWR); | |
1265 | if (bitmap_fd < 0) { | |
1266 | pr_err("weird: %s cannot be opened\n", | |
1267 | s->bitmap_file); | |
1268 | goto abort_locked; | |
1269 | } | |
1270 | if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
1271 | pr_err("Cannot set bitmap file for %s: %s\n", chosen_name, strerror(errno)); | |
1272 | goto abort_locked; | |
1273 | } | |
1274 | } | |
1275 | ||
1276 | if (add_disks(mdfd, &info, s, c, st, &map, devlist, total_slots, | |
1277 | have_container, insert_point, major_num, chosen_name)) | |
1278 | goto abort_locked; | |
1279 | ||
1280 | map_unlock(&map); | |
1281 | ||
1282 | if (is_container(s->level)) { | |
1283 | /* No need to start. But we should signal udev to | |
1284 | * create links */ | |
1285 | sysfs_uevent(&info, "change"); | |
1286 | if (c->verbose >= 0) | |
1287 | pr_err("container %s prepared.\n", chosen_name); | |
1288 | wait_for(chosen_name, mdfd); | |
1289 | } else if (c->runstop == 1 || subdevs >= s->raiddisks) { | |
1290 | if (st->ss->external) { | |
1291 | int err; | |
1292 | switch(s->level) { | |
1293 | case LEVEL_LINEAR: | |
1294 | case LEVEL_MULTIPATH: | |
1295 | case 0: | |
1296 | err = sysfs_set_str(&info, NULL, "array_state", | |
1297 | c->readonly | |
1298 | ? "readonly" | |
1299 | : "active"); | |
1300 | need_mdmon = 0; | |
1301 | break; | |
1302 | default: | |
1303 | err = sysfs_set_str(&info, NULL, "array_state", | |
1304 | "readonly"); | |
1305 | break; | |
1306 | } | |
1307 | sysfs_set_safemode(&info, info.safe_mode_delay); | |
1308 | if (err) { | |
1309 | pr_err("failed to activate array.\n"); | |
1310 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1311 | goto abort; | |
1312 | } | |
1313 | } else if (c->readonly && | |
1314 | sysfs_attribute_available( | |
1315 | &info, NULL, "array_state")) { | |
1316 | if (sysfs_set_str(&info, NULL, | |
1317 | "array_state", "readonly") < 0) { | |
1318 | pr_err("Failed to start array: %s\n", | |
1319 | strerror(errno)); | |
1320 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1321 | goto abort; | |
1322 | } | |
1323 | } else { | |
1324 | /* param is not actually used */ | |
1325 | mdu_param_t param; | |
1326 | if (ioctl(mdfd, RUN_ARRAY, ¶m)) { | |
1327 | pr_err("RUN_ARRAY failed: %s\n", | |
1328 | strerror(errno)); | |
1329 | if (errno == 524 /* ENOTSUP */ && | |
1330 | info.array.level == 0) | |
1331 | cont_err("Please use --layout=original or --layout=alternate\n"); | |
1332 | if (info.array.chunk_size & (info.array.chunk_size-1)) { | |
1333 | cont_err("Problem may be that chunk size is not a power of 2\n"); | |
1334 | } | |
1335 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1336 | goto abort; | |
1337 | } | |
1338 | /* if start_ro module parameter is set, array is | |
1339 | * auto-read-only, which is bad as the resync won't | |
1340 | * start. So lets make it read-write now. | |
1341 | */ | |
1342 | ioctl(mdfd, RESTART_ARRAY_RW, NULL); | |
1343 | } | |
1344 | if (c->verbose >= 0) | |
1345 | pr_info("array %s started.\n", chosen_name); | |
1346 | if (st->ss->external && st->container_devnm[0]) { | |
1347 | if (need_mdmon) | |
1348 | start_mdmon(st->container_devnm); | |
1349 | ||
1350 | ping_monitor(st->container_devnm); | |
1351 | close(container_fd); | |
1352 | } | |
1353 | wait_for(chosen_name, mdfd); | |
1354 | } else { | |
1355 | pr_err("not starting array - not enough devices.\n"); | |
1356 | } | |
1357 | udev_unblock(); | |
1358 | close(mdfd); | |
1359 | sysfs_uevent(&info, "change"); | |
1360 | dev_policy_free(custom_pols); | |
1361 | ||
1362 | return 0; | |
1363 | ||
1364 | abort: | |
1365 | udev_unblock(); | |
1366 | map_lock(&map); | |
1367 | abort_locked: | |
1368 | map_remove(&map, fd2devnm(mdfd)); | |
1369 | map_unlock(&map); | |
1370 | ||
1371 | if (mdfd >= 0) | |
1372 | close(mdfd); | |
1373 | ||
1374 | dev_policy_free(custom_pols); | |
1375 | return 1; | |
1376 | } |