]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Manage.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Introduce pr_err for printing error messages.
[thirdparty/mdadm.git] / Manage.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 #define REGISTER_DEV _IO (MD_MAJOR, 1)
31 #define START_MD _IO (MD_MAJOR, 2)
32 #define STOP_MD _IO (MD_MAJOR, 3)
33
34 int Manage_ro(char *devname, int fd, int readonly)
35 {
36 /* switch to readonly or rw
37 *
38 * requires >= 0.90.0
39 * first check that array is runing
40 * use RESTART_ARRAY_RW or STOP_ARRAY_RO
41 *
42 */
43 mdu_array_info_t array;
44 #ifndef MDASSEMBLE
45 struct mdinfo *mdi;
46 #endif
47 int rv = 0;
48
49 if (md_get_version(fd) < 9000) {
50 pr_err("need md driver version 0.90.0 or later\n");
51 return 1;
52 }
53 #ifndef MDASSEMBLE
54 /* If this is an externally-manage array, we need to modify the
55 * metadata_version so that mdmon doesn't undo our change.
56 */
57 mdi = sysfs_read(fd, -1, GET_LEVEL|GET_VERSION);
58 if (mdi &&
59 mdi->array.major_version == -1 &&
60 is_subarray(mdi->text_version)) {
61 char vers[64];
62 strcpy(vers, "external:");
63 strcat(vers, mdi->text_version);
64 if (readonly > 0) {
65 int rv;
66 /* We set readonly ourselves. */
67 vers[9] = '-';
68 sysfs_set_str(mdi, NULL, "metadata_version", vers);
69
70 close(fd);
71 rv = sysfs_set_str(mdi, NULL, "array_state", "readonly");
72
73 if (rv < 0) {
74 pr_err("failed to set readonly for %s: %s\n",
75 devname, strerror(errno));
76
77 vers[9] = mdi->text_version[0];
78 sysfs_set_str(mdi, NULL, "metadata_version", vers);
79 rv = 1;
80 goto out;
81 }
82 } else {
83 char *cp;
84 /* We cannot set read/write - must signal mdmon */
85 vers[9] = '/';
86 sysfs_set_str(mdi, NULL, "metadata_version", vers);
87
88 cp = strchr(vers+10, '/');
89 if (cp)
90 *cp = 0;
91 ping_monitor(vers+10);
92 if (mdi->array.level <= 0)
93 sysfs_set_str(mdi, NULL, "array_state", "active");
94 }
95 goto out;
96 }
97 #endif
98 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
99 pr_err("%s does not appear to be active.\n",
100 devname);
101 rv = 1;
102 goto out;
103 }
104
105 if (readonly>0) {
106 if (ioctl(fd, STOP_ARRAY_RO, NULL)) {
107 pr_err("failed to set readonly for %s: %s\n",
108 devname, strerror(errno));
109 rv = 1;
110 goto out;
111 }
112 } else if (readonly < 0) {
113 if (ioctl(fd, RESTART_ARRAY_RW, NULL)) {
114 pr_err("failed to set writable for %s: %s\n",
115 devname, strerror(errno));
116 rv = 1;
117 goto out;
118 }
119 }
120 out:
121 #ifndef MDASSEMBLE
122 if (mdi)
123 sysfs_free(mdi);
124 #endif
125 return rv;
126 }
127
128 #ifndef MDASSEMBLE
129
130 static void remove_devices(int devnum, char *path)
131 {
132 /*
133 * Remove names at 'path' - possibly with
134 * partition suffixes - which link to the 'standard'
135 * name for devnum. These were probably created
136 * by mdadm when the array was assembled.
137 */
138 char base[40];
139 char *path2;
140 char link[1024];
141 int n;
142 int part;
143 char *be;
144 char *pe;
145
146 if (!path)
147 return;
148
149 if (devnum >= 0)
150 sprintf(base, "/dev/md%d", devnum);
151 else
152 sprintf(base, "/dev/md_d%d", -1-devnum);
153 be = base + strlen(base);
154
155 path2 = malloc(strlen(path)+20);
156 strcpy(path2, path);
157 pe = path2 + strlen(path2);
158
159 for (part = 0; part < 16; part++) {
160 if (part) {
161 sprintf(be, "p%d", part);
162
163 if (isdigit(pe[-1]))
164 sprintf(pe, "p%d", part);
165 else
166 sprintf(pe, "%d", part);
167 }
168 n = readlink(path2, link, sizeof(link));
169 if (n > 0 && (int)strlen(base) == n &&
170 strncmp(link, base, n) == 0)
171 unlink(path2);
172 }
173 free(path2);
174 }
175
176
177 int Manage_runstop(char *devname, int fd, int runstop, int quiet)
178 {
179 /* Run or stop the array. array must already be configured
180 * required >= 0.90.0
181 * Only print failure messages if quiet == 0;
182 * quiet > 0 means really be quiet
183 * quiet < 0 means we will try again if it fails.
184 */
185 mdu_param_t param; /* unused */
186 int rv = 0;
187
188 if (runstop == -1 && md_get_version(fd) < 9000) {
189 if (ioctl(fd, STOP_MD, 0)) {
190 pr_err("stopping device %s "
191 "failed: %s\n",
192 devname, strerror(errno));
193 return 1;
194 }
195 }
196
197 if (md_get_version(fd) < 9000) {
198 pr_err("need md driver version 0.90.0 or later\n");
199 return 1;
200 }
201 /*
202 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
203 pr_err("%s does not appear to be active.\n",
204 devname);
205 return 1;
206 }
207 */
208 if (runstop>0) {
209 if (ioctl(fd, RUN_ARRAY, &param)) {
210 pr_err("failed to run array %s: %s\n",
211 devname, strerror(errno));
212 return 1;
213 }
214 if (quiet <= 0)
215 pr_err("started %s\n", devname);
216 } else if (runstop < 0){
217 struct map_ent *map = NULL;
218 struct stat stb;
219 struct mdinfo *mdi;
220 int devnum;
221 int err;
222 int count;
223 /* If this is an mdmon managed array, just write 'inactive'
224 * to the array state and let mdmon clear up.
225 */
226 devnum = fd2devnum(fd);
227 /* Get EXCL access first. If this fails, then attempting
228 * to stop is probably a bad idea.
229 */
230 close(fd);
231 fd = open(devname, O_RDONLY|O_EXCL);
232 if (fd < 0 || fd2devnum(fd) != devnum) {
233 if (fd >= 0)
234 close(fd);
235 pr_err("Cannot get exclusive access to %s:"
236 "Perhaps a running "
237 "process, mounted filesystem "
238 "or active volume group?\n",
239 devname);
240 return 1;
241 }
242 mdi = sysfs_read(fd, -1, GET_LEVEL|GET_VERSION);
243 if (mdi &&
244 mdi->array.level > 0 &&
245 is_subarray(mdi->text_version)) {
246 int err;
247 /* This is mdmon managed. */
248 close(fd);
249
250 count = 25;
251 while (count &&
252 (err = sysfs_set_str(mdi, NULL,
253 "array_state",
254 "inactive")) < 0
255 && errno == EBUSY) {
256 usleep(200000);
257 count--;
258 }
259 if (err && !quiet) {
260 pr_err("failed to stop array %s: %s\n",
261 devname, strerror(errno));
262 rv = 1;
263 goto out;
264 }
265
266 /* Give monitor a chance to act */
267 ping_monitor(mdi->text_version);
268
269 fd = open_dev_excl(devnum);
270 if (fd < 0) {
271 pr_err("failed to completely stop %s"
272 ": Device is busy\n",
273 devname);
274 rv = 1;
275 goto out;
276 }
277 } else if (mdi &&
278 mdi->array.major_version == -1 &&
279 mdi->array.minor_version == -2 &&
280 !is_subarray(mdi->text_version)) {
281 struct mdstat_ent *mds, *m;
282 /* container, possibly mdmon-managed.
283 * Make sure mdmon isn't opening it, which
284 * would interfere with the 'stop'
285 */
286 ping_monitor(mdi->sys_name);
287
288 /* now check that there are no existing arrays
289 * which are members of this array
290 */
291 mds = mdstat_read(0, 0);
292 for (m=mds; m; m=m->next)
293 if (m->metadata_version &&
294 strncmp(m->metadata_version, "external:", 9)==0 &&
295 is_subarray(m->metadata_version+9) &&
296 devname2devnum(m->metadata_version+10) == devnum) {
297 if (!quiet)
298 pr_err("Cannot stop container %s: "
299 "member %s still active\n",
300 devname, m->dev);
301 free_mdstat(mds);
302 rv = 1;
303 goto out;
304 }
305 }
306
307 /* As we have an O_EXCL open, any use of the device
308 * which blocks STOP_ARRAY is probably a transient use,
309 * so it is reasonable to retry for a while - 5 seconds.
310 */
311 count = 25; err = 0;
312 while (count && fd >= 0
313 && (err = ioctl(fd, STOP_ARRAY, NULL)) < 0
314 && errno == EBUSY) {
315 usleep(200000);
316 count --;
317 }
318 if (fd >= 0 && err) {
319 if (quiet == 0) {
320 pr_err("failed to stop array %s: %s\n",
321 devname, strerror(errno));
322 if (errno == EBUSY)
323 fprintf(stderr, "Perhaps a running "
324 "process, mounted filesystem "
325 "or active volume group?\n");
326 }
327 rv = 1;
328 goto out;
329 }
330 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
331 * was stopped, so We'll do it here just to be sure. Drop any
332 * partitions as well...
333 */
334 if (fd >= 0)
335 ioctl(fd, BLKRRPART, 0);
336 if (mdi)
337 sysfs_uevent(mdi, "change");
338
339
340 if (devnum != NoMdDev &&
341 (stat("/dev/.udev", &stb) != 0 ||
342 check_env("MDADM_NO_UDEV"))) {
343 struct map_ent *mp = map_by_devnum(&map, devnum);
344 remove_devices(devnum, mp ? mp->path : NULL);
345 }
346
347
348 if (quiet <= 0)
349 pr_err("stopped %s\n", devname);
350 map_lock(&map);
351 map_remove(&map, devnum);
352 map_unlock(&map);
353 out:
354 if (mdi)
355 sysfs_free(mdi);
356 }
357 return rv;
358 }
359
360 int Manage_resize(char *devname, int fd, long long size, int raid_disks)
361 {
362 mdu_array_info_t info;
363 if (ioctl(fd, GET_ARRAY_INFO, &info) != 0) {
364 pr_err("Cannot get array information for %s: %s\n",
365 devname, strerror(errno));
366 return 1;
367 }
368 if (size >= 0)
369 info.size = size;
370 if (raid_disks > 0)
371 info.raid_disks = raid_disks;
372 if (ioctl(fd, SET_ARRAY_INFO, &info) != 0) {
373 pr_err("Cannot set device size/shape for %s: %s\n",
374 devname, strerror(errno));
375 return 1;
376 }
377 return 0;
378 }
379
380 int Manage_subdevs(char *devname, int fd,
381 struct mddev_dev *devlist, int verbose, int test,
382 char *update, int force)
383 {
384 /* do something to each dev.
385 * devmode can be
386 * 'a' - add the device
387 * try HOT_ADD_DISK
388 * If that fails EINVAL, try ADD_NEW_DISK
389 * 'r' - remove the device HOT_REMOVE_DISK
390 * device can be 'faulty' or 'detached' in which case all
391 * matching devices are removed.
392 * 'f' - set the device faulty SET_DISK_FAULTY
393 * device can be 'detached' in which case any device that
394 * is inaccessible will be marked faulty.
395 * For 'f' and 'r', the device can also be a kernel-internal
396 * name such as 'sdb'.
397 */
398 struct mddev_dev *add_devlist = NULL;
399 mdu_array_info_t array;
400 mdu_disk_info_t disc;
401 unsigned long long array_size;
402 struct mddev_dev *dv, *next = NULL;
403 struct stat stb;
404 int j, jnext = 0;
405 int tfd = -1;
406 struct supertype *st, *tst;
407 char *subarray = NULL;
408 int duuid[4];
409 int ouuid[4];
410 int lfd = -1;
411 int sysfd = -1;
412 int count = 0; /* number of actions taken */
413 struct mdinfo info;
414 int frozen = 0;
415
416 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
417 pr_err("cannot get array info for %s\n",
418 devname);
419 goto abort;
420 }
421 sysfs_init(&info, fd, 0);
422
423 /* array.size is only 32 bit and may be truncated.
424 * So read from sysfs if possible, and record number of sectors
425 */
426
427 array_size = get_component_size(fd);
428 if (array_size <= 0)
429 array_size = array.size * 2;
430
431 tst = super_by_fd(fd, &subarray);
432 if (!tst) {
433 pr_err("unsupport array - version %d.%d\n",
434 array.major_version, array.minor_version);
435 goto abort;
436 }
437
438 stb.st_rdev = 0;
439 for (dv = devlist, j=0 ; dv; dv = next, j = jnext) {
440 unsigned long long ldsize;
441 char dvname[20];
442 char *dnprintable = dv->devname;
443 char *add_dev = dv->devname;
444 int err;
445 int array_failed;
446
447 next = dv->next;
448 jnext = 0;
449
450 if (strcmp(dv->devname, "failed")==0 ||
451 strcmp(dv->devname, "faulty")==0) {
452 int remaining_disks = array.nr_disks;
453 if (dv->disposition != 'r') {
454 pr_err("%s only meaningful "
455 "with -r, not -%c\n",
456 dv->devname, dv->disposition);
457 goto abort;
458 }
459 for (; j < MAX_DISKS && remaining_disks > 0; j++) {
460 unsigned dev;
461 disc.number = j;
462 if (ioctl(fd, GET_DISK_INFO, &disc))
463 continue;
464 if (disc.major == 0 && disc.minor == 0)
465 continue;
466 remaining_disks --;
467 if ((disc.state & 1) == 0) /* faulty */
468 continue;
469 dev = makedev(disc.major, disc.minor);
470 if (stb.st_rdev == dev)
471 /* already did that one */
472 continue;
473 stb.st_rdev = dev;
474 next = dv;
475 /* same slot again next time - things might
476 * have reshuffled */
477 jnext = j;
478 sprintf(dvname,"%d:%d", disc.major, disc.minor);
479 dnprintable = dvname;
480 break;
481 }
482 if (next != dv)
483 continue;
484 } else if (strcmp(dv->devname, "detached") == 0) {
485 int remaining_disks = array.nr_disks;
486 if (dv->disposition != 'r' && dv->disposition != 'f') {
487 pr_err("%s only meaningful "
488 "with -r of -f, not -%c\n",
489 dv->devname, dv->disposition);
490 goto abort;
491 }
492 for (; j < MAX_DISKS && remaining_disks > 0; j++) {
493 int sfd;
494 unsigned dev;
495 disc.number = j;
496 if (ioctl(fd, GET_DISK_INFO, &disc))
497 continue;
498 if (disc.major == 0 && disc.minor == 0)
499 continue;
500 remaining_disks --;
501 sprintf(dvname,"%d:%d", disc.major, disc.minor);
502 sfd = dev_open(dvname, O_RDONLY);
503 if (sfd >= 0) {
504 close(sfd);
505 continue;
506 }
507 if (dv->disposition == 'f' &&
508 (disc.state & 1) == 1) /* already faulty */
509 continue;
510 if (errno != ENXIO)
511 continue;
512 dev = makedev(disc.major, disc.minor);
513 if (stb.st_rdev == dev)
514 /* already did that one */
515 continue;
516 stb.st_rdev = dev;
517 next = dv;
518 /* same slot again next time - things might
519 * have reshuffled */
520 jnext = j;
521 dnprintable = dvname;
522 break;
523 }
524 if (next != dv)
525 continue;
526 } else if (strcmp(dv->devname, "missing") == 0) {
527 if (dv->disposition != 'a' || dv->re_add == 0) {
528 pr_err("'missing' only meaningful "
529 "with --re-add\n");
530 goto abort;
531 }
532 if (add_devlist == NULL)
533 add_devlist = conf_get_devs();
534 if (add_devlist == NULL) {
535 pr_err("no devices to scan for missing members.");
536 continue;
537 }
538 add_dev = add_devlist->devname;
539 add_devlist = add_devlist->next;
540 if (add_devlist != NULL)
541 next = dv;
542 if (stat(add_dev, &stb) < 0)
543 continue;
544 } else if (strchr(dv->devname, '/') == NULL &&
545 strchr(dv->devname, ':') == NULL &&
546 strlen(dv->devname) < 50) {
547 /* Assume this is a kernel-internal name like 'sda1' */
548 int found = 0;
549 char dname[55];
550 if (dv->disposition != 'r' && dv->disposition != 'f') {
551 pr_err("%s only meaningful "
552 "with -r or -f, not -%c\n",
553 dv->devname, dv->disposition);
554 goto abort;
555 }
556
557 sprintf(dname, "dev-%s", dv->devname);
558 sysfd = sysfs_open(fd2devnum(fd), dname, "block/dev");
559 if (sysfd >= 0) {
560 char dn[20];
561 int mj,mn;
562 if (sysfs_fd_get_str(sysfd, dn, 20) > 0 &&
563 sscanf(dn, "%d:%d", &mj,&mn) == 2) {
564 stb.st_rdev = makedev(mj,mn);
565 found = 1;
566 }
567 close(sysfd);
568 sysfd = -1;
569 }
570 if (!found) {
571 sysfd = sysfs_open(fd2devnum(fd), dname, "state");
572 if (sysfd < 0) {
573 pr_err("%s does not appear "
574 "to be a component of %s\n",
575 dv->devname, devname);
576 goto abort;
577 }
578 }
579 } else {
580 j = 0;
581
582 tfd = dev_open(dv->devname, O_RDONLY);
583 if (tfd < 0 && dv->disposition == 'r' &&
584 lstat(dv->devname, &stb) == 0)
585 /* Be happy, the lstat worked, that is
586 * enough for --remove
587 */
588 ;
589 else {
590 if (tfd < 0 || fstat(tfd, &stb) != 0) {
591 pr_err("cannot find %s: %s\n",
592 dv->devname, strerror(errno));
593 if (tfd >= 0)
594 close(tfd);
595 goto abort;
596 }
597 close(tfd);
598 tfd = -1;
599 }
600 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
601 pr_err("%s is not a "
602 "block device.\n",
603 dv->devname);
604 goto abort;
605 }
606 }
607 switch(dv->disposition){
608 default:
609 pr_err("internal error - devmode[%s]=%d\n",
610 dv->devname, dv->disposition);
611 goto abort;
612 case 'a':
613 /* add the device */
614 if (subarray) {
615 pr_err("Cannot add disks to a"
616 " \'member\' array, perform this"
617 " operation on the parent container\n");
618 goto abort;
619 }
620 /* Make sure it isn't in use (in 2.6 or later) */
621 tfd = dev_open(add_dev, O_RDONLY|O_EXCL|O_DIRECT);
622 if (tfd < 0 && add_dev != dv->devname)
623 continue;
624 if (tfd < 0) {
625 pr_err("Cannot open %s: %s\n",
626 dv->devname, strerror(errno));
627 goto abort;
628 }
629 if (!frozen) {
630 if (sysfs_freeze_array(&info) == 1)
631 frozen = 1;
632 else
633 frozen = -1;
634 }
635
636 st = dup_super(tst);
637
638 if (array.not_persistent==0)
639 st->ss->load_super(st, tfd, NULL);
640
641 if (add_dev == dv->devname) {
642 if (!get_dev_size(tfd, dv->devname, &ldsize)) {
643 st->ss->free_super(st);
644 close(tfd);
645 goto abort;
646 }
647 } else if (!get_dev_size(tfd, NULL, &ldsize)) {
648 st->ss->free_super(st);
649 close(tfd);
650 tfd = -1;
651 continue;
652 }
653
654 if (tst->ss->validate_geometry(
655 tst, array.level, array.layout,
656 array.raid_disks, NULL,
657 ldsize >> 9, NULL, NULL, 0) == 0) {
658 if (!force) {
659 pr_err("%s is larger than %s can "
660 "effectively use.\n"
661 " Add --force is you "
662 "really want to add this device.\n",
663 add_dev, devname);
664 st->ss->free_super(st);
665 close(tfd);
666 goto abort;
667 }
668 pr_err("%s is larger than %s can "
669 "effectively use.\n"
670 " Adding anyway as --force "
671 "was given.\n",
672 add_dev, devname);
673 }
674 if (!tst->ss->external &&
675 array.major_version == 0 &&
676 md_get_version(fd)%100 < 2) {
677 close(tfd);
678 st->ss->free_super(st);
679 tfd = -1;
680 if (ioctl(fd, HOT_ADD_DISK,
681 (unsigned long)stb.st_rdev)==0) {
682 if (verbose >= 0)
683 pr_err("hot added %s\n",
684 add_dev);
685 continue;
686 }
687
688 pr_err("hot add failed for %s: %s\n",
689 add_dev, strerror(errno));
690 goto abort;
691 }
692
693 if (array.not_persistent == 0 || tst->ss->external) {
694
695 /* need to find a sample superblock to copy, and
696 * a spare slot to use.
697 * For 'external' array (well, container based),
698 * We can just load the metadata for the array.
699 */
700 if (tst->sb)
701 /* already loaded */;
702 else if (tst->ss->external) {
703 tst->ss->load_container(tst, fd, NULL);
704 } else for (j = 0; j < tst->max_devs; j++) {
705 char *dev;
706 int dfd;
707 disc.number = j;
708 if (ioctl(fd, GET_DISK_INFO, &disc))
709 continue;
710 if (disc.major==0 && disc.minor==0)
711 continue;
712 if ((disc.state & 4)==0) continue; /* sync */
713 /* Looks like a good device to try */
714 dev = map_dev(disc.major, disc.minor, 1);
715 if (!dev) continue;
716 dfd = dev_open(dev, O_RDONLY);
717 if (dfd < 0) continue;
718 if (tst->ss->load_super(tst, dfd,
719 NULL)) {
720 close(dfd);
721 continue;
722 }
723 close(dfd);
724 break;
725 }
726 /* FIXME this is a bad test to be using */
727 if (!tst->sb &&
728 dv->re_add) {
729 /* we are re-adding a device to a
730 * completely dead array - have to depend
731 * on kernel to check
732 */
733 } else if (!tst->sb) {
734 close(tfd);
735 st->ss->free_super(st);
736 pr_err("cannot load array metadata from %s\n", devname);
737 goto abort;
738 }
739
740 /* Make sure device is large enough */
741 if (tst->ss->avail_size(tst, ldsize/512) <
742 array_size) {
743 close(tfd);
744 tfd = -1;
745 st->ss->free_super(st);
746 if (add_dev != dv->devname)
747 continue;
748 pr_err("%s not large enough to join array\n",
749 dv->devname);
750 goto abort;
751 }
752
753 /* Possibly this device was recently part of the array
754 * and was temporarily removed, and is now being re-added.
755 * If so, we can simply re-add it.
756 */
757
758 if (st->sb) {
759 struct mdinfo mdi;
760 st->ss->getinfo_super(st, &mdi, NULL);
761 st->ss->uuid_from_super(st, ouuid);
762 if (tst->sb)
763 tst->ss->uuid_from_super(tst, duuid);
764 else
765 /* Assume uuid matches: kernel will check */
766 memcpy(duuid, ouuid, sizeof(ouuid));
767 if ((mdi.disk.state & (1<<MD_DISK_ACTIVE)) &&
768 !(mdi.disk.state & (1<<MD_DISK_FAULTY)) &&
769 memcmp(duuid, ouuid, sizeof(ouuid))==0) {
770 /* look like it is worth a try. Need to
771 * make sure kernel will accept it though.
772 */
773 /* re-add doesn't work for version-1 superblocks
774 * before 2.6.18 :-(
775 */
776 if (array.major_version == 1 &&
777 get_linux_version() <= 2006018)
778 goto skip_re_add;
779 disc.number = mdi.disk.number;
780 if (ioctl(fd, GET_DISK_INFO, &disc) != 0
781 || disc.major != 0 || disc.minor != 0
782 )
783 goto skip_re_add;
784 disc.major = major(stb.st_rdev);
785 disc.minor = minor(stb.st_rdev);
786 disc.number = mdi.disk.number;
787 disc.raid_disk = mdi.disk.raid_disk;
788 disc.state = mdi.disk.state;
789 if (dv->writemostly == 1)
790 disc.state |= 1 << MD_DISK_WRITEMOSTLY;
791 if (dv->writemostly == 2)
792 disc.state &= ~(1 << MD_DISK_WRITEMOSTLY);
793 remove_partitions(tfd);
794 close(tfd);
795 tfd = -1;
796 if (update || dv->writemostly > 0) {
797 int rv = -1;
798 tfd = dev_open(dv->devname, O_RDWR);
799 if (tfd < 0) {
800 pr_err("failed to open %s for"
801 " superblock update during re-add\n", dv->devname);
802 st->ss->free_super(st);
803 goto abort;
804 }
805
806 if (dv->writemostly == 1)
807 rv = st->ss->update_super(
808 st, NULL, "writemostly",
809 devname, verbose, 0, NULL);
810 if (dv->writemostly == 2)
811 rv = st->ss->update_super(
812 st, NULL, "readwrite",
813 devname, verbose, 0, NULL);
814 if (update)
815 rv = st->ss->update_super(
816 st, NULL, update,
817 devname, verbose, 0, NULL);
818 if (rv == 0)
819 rv = st->ss->store_super(st, tfd);
820 close(tfd);
821 tfd = -1;
822 if (rv != 0) {
823 pr_err("failed to update"
824 " superblock during re-add\n");
825 st->ss->free_super(st);
826 goto abort;
827 }
828 }
829 /* don't even try if disk is marked as faulty */
830 errno = 0;
831 if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) {
832 if (verbose >= 0)
833 pr_err("re-added %s\n", add_dev);
834 count++;
835 st->ss->free_super(st);
836 continue;
837 }
838 if (errno == ENOMEM || errno == EROFS) {
839 pr_err("add new device failed for %s: %s\n",
840 add_dev, strerror(errno));
841 st->ss->free_super(st);
842 if (add_dev != dv->devname)
843 continue;
844 goto abort;
845 }
846 }
847 skip_re_add:
848 st->ss->free_super(st);
849 }
850 if (add_dev != dv->devname) {
851 if (verbose > 0)
852 pr_err("--re-add for %s to %s is not possible\n",
853 add_dev, devname);
854 if (tfd >= 0) {
855 close(tfd);
856 tfd = -1;
857 }
858 continue;
859 }
860 if (dv->re_add) {
861 if (tfd >= 0)
862 close(tfd);
863 pr_err("--re-add for %s to %s is not possible\n",
864 dv->devname, devname);
865 goto abort;
866 }
867 if (array.active_disks < array.raid_disks) {
868 char *avail = calloc(array.raid_disks, 1);
869 int d;
870 int found = 0;
871
872 for (d = 0; d < MAX_DISKS && found < array.active_disks; d++) {
873 disc.number = d;
874 if (ioctl(fd, GET_DISK_INFO, &disc))
875 continue;
876 if (disc.major == 0 && disc.minor == 0)
877 continue;
878 if (!(disc.state & (1<<MD_DISK_SYNC)))
879 continue;
880 avail[disc.raid_disk] = 1;
881 found++;
882 }
883 array_failed = !enough(array.level, array.raid_disks,
884 array.layout, 1, avail);
885 } else
886 array_failed = 0;
887 if (array_failed) {
888 pr_err("%s has failed so using --add cannot work and might destroy\n",
889 devname);
890 pr_err("data on %s. You should stop the array and re-assemble it.\n",
891 dv->devname);
892 if (tfd >= 0)
893 close(tfd);
894 goto abort;
895 }
896 } else {
897 /* non-persistent. Must ensure that new drive
898 * is at least array.size big.
899 */
900 if (ldsize/512 < array_size) {
901 pr_err("%s not large enough to join array\n",
902 dv->devname);
903 if (tfd >= 0)
904 close(tfd);
905 goto abort;
906 }
907 }
908 /* committed to really trying this device now*/
909 if (tfd >= 0) {
910 remove_partitions(tfd);
911 close(tfd);
912 tfd = -1;
913 }
914 /* in 2.6.17 and earlier, version-1 superblocks won't
915 * use the number we write, but will choose a free number.
916 * we must choose the same free number, which requires
917 * starting at 'raid_disks' and counting up
918 */
919 for (j = array.raid_disks; j< tst->max_devs; j++) {
920 disc.number = j;
921 if (ioctl(fd, GET_DISK_INFO, &disc))
922 break;
923 if (disc.major==0 && disc.minor==0)
924 break;
925 if (disc.state & 8) /* removed */
926 break;
927 }
928 disc.major = major(stb.st_rdev);
929 disc.minor = minor(stb.st_rdev);
930 disc.number =j;
931 disc.state = 0;
932 if (array.not_persistent==0) {
933 int dfd;
934 if (dv->writemostly == 1)
935 disc.state |= 1 << MD_DISK_WRITEMOSTLY;
936 dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT);
937 if (tst->ss->add_to_super(tst, &disc, dfd,
938 dv->devname)) {
939 close(dfd);
940 goto abort;
941 }
942 if (tst->ss->write_init_super(tst)) {
943 close(dfd);
944 goto abort;
945 }
946 } else if (dv->re_add) {
947 /* this had better be raid1.
948 * As we are "--re-add"ing we must find a spare slot
949 * to fill.
950 */
951 char *used = malloc(array.raid_disks);
952 memset(used, 0, array.raid_disks);
953 for (j=0; j< tst->max_devs; j++) {
954 mdu_disk_info_t disc2;
955 disc2.number = j;
956 if (ioctl(fd, GET_DISK_INFO, &disc2))
957 continue;
958 if (disc2.major==0 && disc2.minor==0)
959 continue;
960 if (disc2.state & 8) /* removed */
961 continue;
962 if (disc2.raid_disk < 0)
963 continue;
964 if (disc2.raid_disk > array.raid_disks)
965 continue;
966 used[disc2.raid_disk] = 1;
967 }
968 for (j=0 ; j<array.raid_disks; j++)
969 if (!used[j]) {
970 disc.raid_disk = j;
971 disc.state |= (1<<MD_DISK_SYNC);
972 break;
973 }
974 free(used);
975 }
976 if (dv->writemostly == 1)
977 disc.state |= (1 << MD_DISK_WRITEMOSTLY);
978 if (tst->ss->external) {
979 /* add a disk
980 * to an external metadata container */
981 struct mdinfo new_mdi;
982 struct mdinfo *sra;
983 int container_fd;
984 int devnum = fd2devnum(fd);
985 int dfd;
986
987 container_fd = open_dev_excl(devnum);
988 if (container_fd < 0) {
989 pr_err("add failed for %s:"
990 " could not get exclusive access to container\n",
991 dv->devname);
992 tst->ss->free_super(tst);
993 goto abort;
994 }
995
996 dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT);
997 if (mdmon_running(tst->container_dev))
998 tst->update_tail = &tst->updates;
999 if (tst->ss->add_to_super(tst, &disc, dfd,
1000 dv->devname)) {
1001 close(dfd);
1002 close(container_fd);
1003 goto abort;
1004 }
1005 if (tst->update_tail)
1006 flush_metadata_updates(tst);
1007 else
1008 tst->ss->sync_metadata(tst);
1009
1010 sra = sysfs_read(container_fd, -1, 0);
1011 if (!sra) {
1012 pr_err("add failed for %s: sysfs_read failed\n",
1013 dv->devname);
1014 close(container_fd);
1015 tst->ss->free_super(tst);
1016 goto abort;
1017 }
1018 sra->array.level = LEVEL_CONTAINER;
1019 /* Need to set data_offset and component_size */
1020 tst->ss->getinfo_super(tst, &new_mdi, NULL);
1021 new_mdi.disk.major = disc.major;
1022 new_mdi.disk.minor = disc.minor;
1023 new_mdi.recovery_start = 0;
1024 /* Make sure fds are closed as they are O_EXCL which
1025 * would block add_disk */
1026 tst->ss->free_super(tst);
1027 if (sysfs_add_disk(sra, &new_mdi, 0) != 0) {
1028 pr_err("add new device to external metadata"
1029 " failed for %s\n", dv->devname);
1030 close(container_fd);
1031 sysfs_free(sra);
1032 goto abort;
1033 }
1034 ping_monitor_by_id(devnum);
1035 sysfs_free(sra);
1036 close(container_fd);
1037 } else {
1038 tst->ss->free_super(tst);
1039 if (ioctl(fd, ADD_NEW_DISK, &disc)) {
1040 pr_err("add new device failed for %s as %d: %s\n",
1041 dv->devname, j, strerror(errno));
1042 goto abort;
1043 }
1044 }
1045 if (verbose >= 0)
1046 pr_err("added %s\n", dv->devname);
1047 break;
1048
1049 case 'r':
1050 /* hot remove */
1051 if (subarray) {
1052 pr_err("Cannot remove disks from a"
1053 " \'member\' array, perform this"
1054 " operation on the parent container\n");
1055 if (sysfd >= 0)
1056 close(sysfd);
1057 goto abort;
1058 }
1059 if (tst->ss->external) {
1060 /* To remove a device from a container, we must
1061 * check that it isn't in use in an array.
1062 * This involves looking in the 'holders'
1063 * directory - there must be just one entry,
1064 * the container.
1065 * To ensure that it doesn't get used as a
1066 * hold spare while we are checking, we
1067 * get an O_EXCL open on the container
1068 */
1069 int dnum = fd2devnum(fd);
1070 lfd = open_dev_excl(dnum);
1071 if (lfd < 0) {
1072 pr_err("Cannot get exclusive access "
1073 " to container - odd\n");
1074 if (sysfd >= 0)
1075 close(sysfd);
1076 goto abort;
1077 }
1078 /* in the detached case it is not possible to
1079 * check if we are the unique holder, so just
1080 * rely on the 'detached' checks
1081 */
1082 if (strcmp(dv->devname, "detached") == 0 ||
1083 sysfd >= 0 ||
1084 sysfs_unique_holder(dnum, stb.st_rdev))
1085 /* pass */;
1086 else {
1087 pr_err("%s is %s, cannot remove.\n",
1088 dnprintable,
1089 errno == EEXIST ? "still in use":
1090 "not a member");
1091 close(lfd);
1092 goto abort;
1093 }
1094 }
1095 /* FIXME check that it is a current member */
1096 if (sysfd >= 0) {
1097 /* device has been removed and we don't know
1098 * the major:minor number
1099 */
1100 int n = write(sysfd, "remove", 6);
1101 if (n != 6)
1102 err = -1;
1103 else
1104 err = 0;
1105 close(sysfd);
1106 sysfd = -1;
1107 } else {
1108 err = ioctl(fd, HOT_REMOVE_DISK, (unsigned long)stb.st_rdev);
1109 if (err && errno == ENODEV) {
1110 /* Old kernels rejected this if no personality
1111 * registered */
1112 struct mdinfo *sra = sysfs_read(fd, 0, GET_DEVS);
1113 struct mdinfo *dv = NULL;
1114 if (sra)
1115 dv = sra->devs;
1116 for ( ; dv ; dv=dv->next)
1117 if (dv->disk.major == (int)major(stb.st_rdev) &&
1118 dv->disk.minor == (int)minor(stb.st_rdev))
1119 break;
1120 if (dv)
1121 err = sysfs_set_str(sra, dv,
1122 "state", "remove");
1123 else
1124 err = -1;
1125 if (sra)
1126 sysfs_free(sra);
1127 }
1128 }
1129 if (err) {
1130 pr_err("hot remove failed "
1131 "for %s: %s\n", dnprintable,
1132 strerror(errno));
1133 if (lfd >= 0)
1134 close(lfd);
1135 goto abort;
1136 }
1137 if (tst->ss->external) {
1138 /*
1139 * Before dropping our exclusive open we make an
1140 * attempt at preventing mdmon from seeing an
1141 * 'add' event before reconciling this 'remove'
1142 * event.
1143 */
1144 char *name = devnum2devname(fd2devnum(fd));
1145
1146 if (!name) {
1147 pr_err("unable to get container name\n");
1148 goto abort;
1149 }
1150
1151 ping_manager(name);
1152 free(name);
1153 }
1154 if (lfd >= 0)
1155 close(lfd);
1156 count++;
1157 if (verbose >= 0)
1158 pr_err("hot removed %s from %s\n",
1159 dnprintable, devname);
1160 break;
1161
1162 case 'f': /* set faulty */
1163 /* FIXME check current member */
1164 if ((sysfd >= 0 && write(sysfd, "faulty", 6) != 6) ||
1165 (sysfd < 0 && ioctl(fd, SET_DISK_FAULTY,
1166 (unsigned long) stb.st_rdev))) {
1167 pr_err("set device faulty failed for %s: %s\n",
1168 dnprintable, strerror(errno));
1169 if (sysfd >= 0)
1170 close(sysfd);
1171 goto abort;
1172 }
1173 if (sysfd >= 0)
1174 close(sysfd);
1175 sysfd = -1;
1176 count++;
1177 if (verbose >= 0)
1178 pr_err("set %s faulty in %s\n",
1179 dnprintable, devname);
1180 break;
1181 }
1182 }
1183 if (frozen > 0)
1184 sysfs_set_str(&info, NULL, "sync_action","idle");
1185 if (test && count == 0)
1186 return 2;
1187 return 0;
1188
1189 abort:
1190 if (frozen > 0)
1191 sysfs_set_str(&info, NULL, "sync_action","idle");
1192 return 1;
1193 }
1194
1195 int autodetect(void)
1196 {
1197 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1198 int rv = 1;
1199 int fd = dev_open("9:0", O_RDONLY);
1200 if (fd >= 0) {
1201 if (ioctl(fd, RAID_AUTORUN, 0) == 0)
1202 rv = 0;
1203 close(fd);
1204 }
1205 return rv;
1206 }
1207
1208 int Update_subarray(char *dev, char *subarray, char *update, struct mddev_ident *ident, int quiet)
1209 {
1210 struct supertype supertype, *st = &supertype;
1211 int fd, rv = 2;
1212
1213 memset(st, 0, sizeof(*st));
1214
1215 fd = open_subarray(dev, subarray, st, quiet);
1216 if (fd < 0)
1217 return 2;
1218
1219 if (!st->ss->update_subarray) {
1220 if (!quiet)
1221 pr_err("Operation not supported for %s metadata\n",
1222 st->ss->name);
1223 goto free_super;
1224 }
1225
1226 if (mdmon_running(st->devnum))
1227 st->update_tail = &st->updates;
1228
1229 rv = st->ss->update_subarray(st, subarray, update, ident);
1230
1231 if (rv) {
1232 if (!quiet)
1233 pr_err("Failed to update %s of subarray-%s in %s\n",
1234 update, subarray, dev);
1235 } else if (st->update_tail)
1236 flush_metadata_updates(st);
1237 else
1238 st->ss->sync_metadata(st);
1239
1240 if (rv == 0 && strcmp(update, "name") == 0 && !quiet)
1241 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1242 subarray, dev);
1243
1244 free_super:
1245 st->ss->free_super(st);
1246 close(fd);
1247
1248 return rv;
1249 }
1250
1251 /* Move spare from one array to another
1252 * If adding to destination array fails
1253 * add back to original array
1254 * Returns 1 on success, 0 on failure */
1255 int move_spare(char *from_devname, char *to_devname, dev_t devid)
1256 {
1257 struct mddev_dev devlist;
1258 char devname[20];
1259
1260 /* try to remove and add */
1261 int fd1 = open(to_devname, O_RDONLY);
1262 int fd2 = open(from_devname, O_RDONLY);
1263
1264 if (fd1 < 0 || fd2 < 0) {
1265 if (fd1>=0) close(fd1);
1266 if (fd2>=0) close(fd2);
1267 return 0;
1268 }
1269
1270 devlist.next = NULL;
1271 devlist.used = 0;
1272 devlist.re_add = 0;
1273 devlist.writemostly = 0;
1274 devlist.devname = devname;
1275 sprintf(devname, "%d:%d", major(devid), minor(devid));
1276
1277 devlist.disposition = 'r';
1278 if (Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0) == 0) {
1279 devlist.disposition = 'a';
1280 if (Manage_subdevs(to_devname, fd1, &devlist, -1, 0, NULL, 0) == 0) {
1281 /* make sure manager is aware of changes */
1282 ping_manager(to_devname);
1283 ping_manager(from_devname);
1284 close(fd1);
1285 close(fd2);
1286 return 1;
1287 }
1288 else Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0);
1289 }
1290 close(fd1);
1291 close(fd2);
1292 return 0;
1293 }
1294 #endif
Unnamed repository; edit this file 'description' to name the repository.