]> 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
Convert 'quiet' to 'not verbose' in various places.
[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 = xmalloc(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,
178 int verbose, int will_retry)
179 {
180 /* Run or stop the array. array must already be configured
181 * 'Run' requires >= 0.90.0
182 * 'will_retry' is only relevant for 'stop', and means
183 * that error messages are not wanted.
184 */
185 mdu_param_t param; /* unused */
186 int rv = 0;
187
188 if (will_retry && verbose == 0)
189 verbose = -1;
190
191 if (runstop == -1 && md_get_version(fd) < 9000) {
192 if (ioctl(fd, STOP_MD, 0)) {
193 pr_err("stopping device %s "
194 "failed: %s\n",
195 devname, strerror(errno));
196 return 1;
197 }
198 }
199
200 if (md_get_version(fd) < 9000) {
201 pr_err("need md driver version 0.90.0 or later\n");
202 return 1;
203 }
204
205 if (runstop>0) {
206 if (ioctl(fd, RUN_ARRAY, &param)) {
207 if (verbose >= 0)
208 pr_err("failed to run array %s: %s\n",
209 devname, strerror(errno));
210 return 1;
211 }
212 if (verbose >= 0)
213 pr_err("started %s\n", devname);
214 } else if (runstop < 0){
215 struct map_ent *map = NULL;
216 struct stat stb;
217 struct mdinfo *mdi;
218 int devnum;
219 int err;
220 int count;
221 /* If this is an mdmon managed array, just write 'inactive'
222 * to the array state and let mdmon clear up.
223 */
224 devnum = fd2devnum(fd);
225 /* Get EXCL access first. If this fails, then attempting
226 * to stop is probably a bad idea.
227 */
228 close(fd);
229 fd = open(devname, O_RDONLY|O_EXCL);
230 if (fd < 0 || fd2devnum(fd) != devnum) {
231 if (fd >= 0)
232 close(fd);
233 if (verbose >= 0)
234 pr_err("Cannot get exclusive access to %s:"
235 "Perhaps a running "
236 "process, mounted filesystem "
237 "or active volume group?\n",
238 devname);
239 return 1;
240 }
241 mdi = sysfs_read(fd, -1, GET_LEVEL|GET_VERSION);
242 if (mdi &&
243 mdi->array.level > 0 &&
244 is_subarray(mdi->text_version)) {
245 int err;
246 /* This is mdmon managed. */
247 close(fd);
248
249 count = 25;
250 while (count &&
251 (err = sysfs_set_str(mdi, NULL,
252 "array_state",
253 "inactive")) < 0
254 && errno == EBUSY) {
255 usleep(200000);
256 count--;
257 }
258 if (err) {
259 if (verbose >= 0)
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 if (verbose >= 0)
272 pr_err("failed to completely stop %s"
273 ": Device is busy\n",
274 devname);
275 rv = 1;
276 goto out;
277 }
278 } else if (mdi &&
279 mdi->array.major_version == -1 &&
280 mdi->array.minor_version == -2 &&
281 !is_subarray(mdi->text_version)) {
282 struct mdstat_ent *mds, *m;
283 /* container, possibly mdmon-managed.
284 * Make sure mdmon isn't opening it, which
285 * would interfere with the 'stop'
286 */
287 ping_monitor(mdi->sys_name);
288
289 /* now check that there are no existing arrays
290 * which are members of this array
291 */
292 mds = mdstat_read(0, 0);
293 for (m=mds; m; m=m->next)
294 if (m->metadata_version &&
295 strncmp(m->metadata_version, "external:", 9)==0 &&
296 is_subarray(m->metadata_version+9) &&
297 devname2devnum(m->metadata_version+10) == devnum) {
298 if (verbose >= 0)
299 pr_err("Cannot stop container %s: "
300 "member %s still active\n",
301 devname, m->dev);
302 free_mdstat(mds);
303 rv = 1;
304 goto out;
305 }
306 }
307
308 /* As we have an O_EXCL open, any use of the device
309 * which blocks STOP_ARRAY is probably a transient use,
310 * so it is reasonable to retry for a while - 5 seconds.
311 */
312 count = 25; err = 0;
313 while (count && fd >= 0
314 && (err = ioctl(fd, STOP_ARRAY, NULL)) < 0
315 && errno == EBUSY) {
316 usleep(200000);
317 count --;
318 }
319 if (fd >= 0 && err) {
320 if (verbose >= 0) {
321 pr_err("failed to stop array %s: %s\n",
322 devname, strerror(errno));
323 if (errno == EBUSY)
324 fprintf(stderr, "Perhaps a running "
325 "process, mounted filesystem "
326 "or active volume group?\n");
327 }
328 rv = 1;
329 goto out;
330 }
331 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
332 * was stopped, so We'll do it here just to be sure. Drop any
333 * partitions as well...
334 */
335 if (fd >= 0)
336 ioctl(fd, BLKRRPART, 0);
337 if (mdi)
338 sysfs_uevent(mdi, "change");
339
340
341 if (devnum != NoMdDev &&
342 (stat("/dev/.udev", &stb) != 0 ||
343 check_env("MDADM_NO_UDEV"))) {
344 struct map_ent *mp = map_by_devnum(&map, devnum);
345 remove_devices(devnum, mp ? mp->path : NULL);
346 }
347
348
349 if (verbose >= 0)
350 pr_err("stopped %s\n", devname);
351 map_lock(&map);
352 map_remove(&map, devnum);
353 map_unlock(&map);
354 out:
355 if (mdi)
356 sysfs_free(mdi);
357 }
358 return rv;
359 }
360
361 int Manage_resize(char *devname, int fd, long long size, int raid_disks)
362 {
363 mdu_array_info_t info;
364 if (ioctl(fd, GET_ARRAY_INFO, &info) != 0) {
365 pr_err("Cannot get array information for %s: %s\n",
366 devname, strerror(errno));
367 return 1;
368 }
369 if (size >= 0)
370 info.size = size;
371 if (raid_disks > 0)
372 info.raid_disks = raid_disks;
373 if (ioctl(fd, SET_ARRAY_INFO, &info) != 0) {
374 pr_err("Cannot set device size/shape for %s: %s\n",
375 devname, strerror(errno));
376 return 1;
377 }
378 return 0;
379 }
380
381 int Manage_subdevs(char *devname, int fd,
382 struct mddev_dev *devlist, int verbose, int test,
383 char *update, int force)
384 {
385 /* do something to each dev.
386 * devmode can be
387 * 'a' - add the device
388 * try HOT_ADD_DISK
389 * If that fails EINVAL, try ADD_NEW_DISK
390 * 'r' - remove the device HOT_REMOVE_DISK
391 * device can be 'faulty' or 'detached' in which case all
392 * matching devices are removed.
393 * 'f' - set the device faulty SET_DISK_FAULTY
394 * device can be 'detached' in which case any device that
395 * is inaccessible will be marked faulty.
396 * For 'f' and 'r', the device can also be a kernel-internal
397 * name such as 'sdb'.
398 */
399 struct mddev_dev *add_devlist = NULL;
400 mdu_array_info_t array;
401 mdu_disk_info_t disc;
402 unsigned long long array_size;
403 struct mddev_dev *dv, *next = NULL;
404 struct stat stb;
405 int j, jnext = 0;
406 int tfd = -1;
407 struct supertype *st, *tst;
408 char *subarray = NULL;
409 int duuid[4];
410 int ouuid[4];
411 int lfd = -1;
412 int sysfd = -1;
413 int count = 0; /* number of actions taken */
414 struct mdinfo info;
415 int frozen = 0;
416
417 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
418 pr_err("cannot get array info for %s\n",
419 devname);
420 goto abort;
421 }
422 sysfs_init(&info, fd, 0);
423
424 /* array.size is only 32 bit and may be truncated.
425 * So read from sysfs if possible, and record number of sectors
426 */
427
428 array_size = get_component_size(fd);
429 if (array_size <= 0)
430 array_size = array.size * 2;
431
432 tst = super_by_fd(fd, &subarray);
433 if (!tst) {
434 pr_err("unsupport array - version %d.%d\n",
435 array.major_version, array.minor_version);
436 goto abort;
437 }
438
439 stb.st_rdev = 0;
440 for (dv = devlist, j=0 ; dv; dv = next, j = jnext) {
441 unsigned long long ldsize;
442 char dvname[20];
443 char *dnprintable = dv->devname;
444 char *add_dev = dv->devname;
445 int err;
446 int array_failed;
447
448 next = dv->next;
449 jnext = 0;
450
451 if (strcmp(dv->devname, "failed")==0 ||
452 strcmp(dv->devname, "faulty")==0) {
453 int remaining_disks = array.nr_disks;
454 if (dv->disposition != 'r') {
455 pr_err("%s only meaningful "
456 "with -r, not -%c\n",
457 dv->devname, dv->disposition);
458 goto abort;
459 }
460 for (; j < MAX_DISKS && remaining_disks > 0; j++) {
461 unsigned dev;
462 disc.number = j;
463 if (ioctl(fd, GET_DISK_INFO, &disc))
464 continue;
465 if (disc.major == 0 && disc.minor == 0)
466 continue;
467 remaining_disks --;
468 if ((disc.state & 1) == 0) /* faulty */
469 continue;
470 dev = makedev(disc.major, disc.minor);
471 if (stb.st_rdev == dev)
472 /* already did that one */
473 continue;
474 stb.st_rdev = dev;
475 next = dv;
476 /* same slot again next time - things might
477 * have reshuffled */
478 jnext = j;
479 sprintf(dvname,"%d:%d", disc.major, disc.minor);
480 dnprintable = dvname;
481 break;
482 }
483 if (next != dv)
484 continue;
485 } else if (strcmp(dv->devname, "detached") == 0) {
486 int remaining_disks = array.nr_disks;
487 if (dv->disposition != 'r' && dv->disposition != 'f') {
488 pr_err("%s only meaningful "
489 "with -r of -f, not -%c\n",
490 dv->devname, dv->disposition);
491 goto abort;
492 }
493 for (; j < MAX_DISKS && remaining_disks > 0; j++) {
494 int sfd;
495 unsigned dev;
496 disc.number = j;
497 if (ioctl(fd, GET_DISK_INFO, &disc))
498 continue;
499 if (disc.major == 0 && disc.minor == 0)
500 continue;
501 remaining_disks --;
502 sprintf(dvname,"%d:%d", disc.major, disc.minor);
503 sfd = dev_open(dvname, O_RDONLY);
504 if (sfd >= 0) {
505 close(sfd);
506 continue;
507 }
508 if (dv->disposition == 'f' &&
509 (disc.state & 1) == 1) /* already faulty */
510 continue;
511 if (errno != ENXIO)
512 continue;
513 dev = makedev(disc.major, disc.minor);
514 if (stb.st_rdev == dev)
515 /* already did that one */
516 continue;
517 stb.st_rdev = dev;
518 next = dv;
519 /* same slot again next time - things might
520 * have reshuffled */
521 jnext = j;
522 dnprintable = dvname;
523 break;
524 }
525 if (next != dv)
526 continue;
527 } else if (strcmp(dv->devname, "missing") == 0) {
528 if (dv->disposition != 'A') {
529 pr_err("'missing' only meaningful "
530 "with --re-add\n");
531 goto abort;
532 }
533 if (add_devlist == NULL)
534 add_devlist = conf_get_devs();
535 if (add_devlist == NULL) {
536 pr_err("no devices to scan for missing members.");
537 continue;
538 }
539 add_dev = add_devlist->devname;
540 add_devlist = add_devlist->next;
541 if (add_devlist != NULL)
542 next = dv;
543 if (stat(add_dev, &stb) < 0)
544 continue;
545 } else if (strchr(dv->devname, '/') == NULL &&
546 strchr(dv->devname, ':') == NULL &&
547 strlen(dv->devname) < 50) {
548 /* Assume this is a kernel-internal name like 'sda1' */
549 int found = 0;
550 char dname[55];
551 if (dv->disposition != 'r' && dv->disposition != 'f') {
552 pr_err("%s only meaningful "
553 "with -r or -f, not -%c\n",
554 dv->devname, dv->disposition);
555 goto abort;
556 }
557
558 sprintf(dname, "dev-%s", dv->devname);
559 sysfd = sysfs_open(fd2devnum(fd), dname, "block/dev");
560 if (sysfd >= 0) {
561 char dn[20];
562 int mj,mn;
563 if (sysfs_fd_get_str(sysfd, dn, 20) > 0 &&
564 sscanf(dn, "%d:%d", &mj,&mn) == 2) {
565 stb.st_rdev = makedev(mj,mn);
566 found = 1;
567 }
568 close(sysfd);
569 sysfd = -1;
570 }
571 if (!found) {
572 sysfd = sysfs_open(fd2devnum(fd), dname, "state");
573 if (sysfd < 0) {
574 pr_err("%s does not appear "
575 "to be a component of %s\n",
576 dv->devname, devname);
577 goto abort;
578 }
579 }
580 } else {
581 j = 0;
582
583 tfd = dev_open(dv->devname, O_RDONLY);
584 if (tfd < 0 && dv->disposition == 'r' &&
585 lstat(dv->devname, &stb) == 0)
586 /* Be happy, the lstat worked, that is
587 * enough for --remove
588 */
589 ;
590 else {
591 if (tfd < 0 || fstat(tfd, &stb) != 0) {
592 pr_err("cannot find %s: %s\n",
593 dv->devname, strerror(errno));
594 if (tfd >= 0)
595 close(tfd);
596 goto abort;
597 }
598 close(tfd);
599 tfd = -1;
600 }
601 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
602 pr_err("%s is not a "
603 "block device.\n",
604 dv->devname);
605 goto abort;
606 }
607 }
608 switch(dv->disposition){
609 default:
610 pr_err("internal error - devmode[%s]=%d\n",
611 dv->devname, dv->disposition);
612 goto abort;
613 case 'a':
614 case 'A':
615 /* add the device */
616 if (subarray) {
617 pr_err("Cannot add disks to a"
618 " \'member\' array, perform this"
619 " operation on the parent container\n");
620 goto abort;
621 }
622 /* Make sure it isn't in use (in 2.6 or later) */
623 tfd = dev_open(add_dev, O_RDONLY|O_EXCL|O_DIRECT);
624 if (tfd < 0 && add_dev != dv->devname)
625 continue;
626 if (tfd < 0) {
627 pr_err("Cannot open %s: %s\n",
628 dv->devname, strerror(errno));
629 goto abort;
630 }
631 if (!frozen) {
632 if (sysfs_freeze_array(&info) == 1)
633 frozen = 1;
634 else
635 frozen = -1;
636 }
637
638 st = dup_super(tst);
639
640 if (array.not_persistent==0)
641 st->ss->load_super(st, tfd, NULL);
642
643 if (add_dev == dv->devname) {
644 if (!get_dev_size(tfd, dv->devname, &ldsize)) {
645 st->ss->free_super(st);
646 close(tfd);
647 goto abort;
648 }
649 } else if (!get_dev_size(tfd, NULL, &ldsize)) {
650 st->ss->free_super(st);
651 close(tfd);
652 tfd = -1;
653 continue;
654 }
655
656 if (tst->ss->validate_geometry(
657 tst, array.level, array.layout,
658 array.raid_disks, NULL,
659 ldsize >> 9, NULL, NULL, 0) == 0) {
660 if (!force) {
661 pr_err("%s is larger than %s can "
662 "effectively use.\n"
663 " Add --force is you "
664 "really want to add this device.\n",
665 add_dev, devname);
666 st->ss->free_super(st);
667 close(tfd);
668 goto abort;
669 }
670 pr_err("%s is larger than %s can "
671 "effectively use.\n"
672 " Adding anyway as --force "
673 "was given.\n",
674 add_dev, devname);
675 }
676 if (!tst->ss->external &&
677 array.major_version == 0 &&
678 md_get_version(fd)%100 < 2) {
679 close(tfd);
680 st->ss->free_super(st);
681 tfd = -1;
682 if (ioctl(fd, HOT_ADD_DISK,
683 (unsigned long)stb.st_rdev)==0) {
684 if (verbose >= 0)
685 pr_err("hot added %s\n",
686 add_dev);
687 continue;
688 }
689
690 pr_err("hot add failed for %s: %s\n",
691 add_dev, strerror(errno));
692 goto abort;
693 }
694
695 if (array.not_persistent == 0 || tst->ss->external) {
696
697 /* need to find a sample superblock to copy, and
698 * a spare slot to use.
699 * For 'external' array (well, container based),
700 * We can just load the metadata for the array.
701 */
702 if (tst->sb)
703 /* already loaded */;
704 else if (tst->ss->external) {
705 tst->ss->load_container(tst, fd, NULL);
706 } else for (j = 0; j < tst->max_devs; j++) {
707 char *dev;
708 int dfd;
709 disc.number = j;
710 if (ioctl(fd, GET_DISK_INFO, &disc))
711 continue;
712 if (disc.major==0 && disc.minor==0)
713 continue;
714 if ((disc.state & 4)==0) continue; /* sync */
715 /* Looks like a good device to try */
716 dev = map_dev(disc.major, disc.minor, 1);
717 if (!dev) continue;
718 dfd = dev_open(dev, O_RDONLY);
719 if (dfd < 0) continue;
720 if (tst->ss->load_super(tst, dfd,
721 NULL)) {
722 close(dfd);
723 continue;
724 }
725 close(dfd);
726 break;
727 }
728 /* FIXME this is a bad test to be using */
729 if (!tst->sb && dv->disposition == 'A') {
730 /* we are re-adding a device to a
731 * completely dead array - have to depend
732 * on kernel to check
733 */
734 } else if (!tst->sb) {
735 close(tfd);
736 st->ss->free_super(st);
737 pr_err("cannot load array metadata from %s\n", devname);
738 goto abort;
739 }
740
741 /* Make sure device is large enough */
742 if (tst->ss->avail_size(tst, ldsize/512) <
743 array_size) {
744 close(tfd);
745 tfd = -1;
746 st->ss->free_super(st);
747 if (add_dev != dv->devname)
748 continue;
749 pr_err("%s not large enough to join array\n",
750 dv->devname);
751 goto abort;
752 }
753
754 /* Possibly this device was recently part of the array
755 * and was temporarily removed, and is now being re-added.
756 * If so, we can simply re-add it.
757 */
758
759 if (st->sb) {
760 struct mdinfo mdi;
761 st->ss->getinfo_super(st, &mdi, NULL);
762 st->ss->uuid_from_super(st, ouuid);
763 if (tst->sb)
764 tst->ss->uuid_from_super(tst, duuid);
765 else
766 /* Assume uuid matches: kernel will check */
767 memcpy(duuid, ouuid, sizeof(ouuid));
768 if ((mdi.disk.state & (1<<MD_DISK_ACTIVE)) &&
769 !(mdi.disk.state & (1<<MD_DISK_FAULTY)) &&
770 memcmp(duuid, ouuid, sizeof(ouuid))==0) {
771 /* look like it is worth a try. Need to
772 * make sure kernel will accept it though.
773 */
774 /* re-add doesn't work for version-1 superblocks
775 * before 2.6.18 :-(
776 */
777 if (array.major_version == 1 &&
778 get_linux_version() <= 2006018)
779 goto skip_re_add;
780 disc.number = mdi.disk.number;
781 if (ioctl(fd, GET_DISK_INFO, &disc) != 0
782 || disc.major != 0 || disc.minor != 0
783 )
784 goto skip_re_add;
785 disc.major = major(stb.st_rdev);
786 disc.minor = minor(stb.st_rdev);
787 disc.number = mdi.disk.number;
788 disc.raid_disk = mdi.disk.raid_disk;
789 disc.state = mdi.disk.state;
790 if (dv->writemostly == 1)
791 disc.state |= 1 << MD_DISK_WRITEMOSTLY;
792 if (dv->writemostly == 2)
793 disc.state &= ~(1 << MD_DISK_WRITEMOSTLY);
794 remove_partitions(tfd);
795 close(tfd);
796 tfd = -1;
797 if (update || dv->writemostly > 0) {
798 int rv = -1;
799 tfd = dev_open(dv->devname, O_RDWR);
800 if (tfd < 0) {
801 pr_err("failed to open %s for"
802 " superblock update during re-add\n", dv->devname);
803 st->ss->free_super(st);
804 goto abort;
805 }
806
807 if (dv->writemostly == 1)
808 rv = st->ss->update_super(
809 st, NULL, "writemostly",
810 devname, verbose, 0, NULL);
811 if (dv->writemostly == 2)
812 rv = st->ss->update_super(
813 st, NULL, "readwrite",
814 devname, verbose, 0, NULL);
815 if (update)
816 rv = st->ss->update_super(
817 st, NULL, update,
818 devname, verbose, 0, NULL);
819 if (rv == 0)
820 rv = st->ss->store_super(st, tfd);
821 close(tfd);
822 tfd = -1;
823 if (rv != 0) {
824 pr_err("failed to update"
825 " superblock during re-add\n");
826 st->ss->free_super(st);
827 goto abort;
828 }
829 }
830 /* don't even try if disk is marked as faulty */
831 errno = 0;
832 if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) {
833 if (verbose >= 0)
834 pr_err("re-added %s\n", add_dev);
835 count++;
836 st->ss->free_super(st);
837 continue;
838 }
839 if (errno == ENOMEM || errno == EROFS) {
840 pr_err("add new device failed for %s: %s\n",
841 add_dev, strerror(errno));
842 st->ss->free_super(st);
843 if (add_dev != dv->devname)
844 continue;
845 goto abort;
846 }
847 }
848 skip_re_add:
849 st->ss->free_super(st);
850 }
851 if (add_dev != dv->devname) {
852 if (verbose > 0)
853 pr_err("--re-add for %s to %s is not possible\n",
854 add_dev, devname);
855 if (tfd >= 0) {
856 close(tfd);
857 tfd = -1;
858 }
859 continue;
860 }
861 if (dv->disposition == 'A') {
862 if (tfd >= 0)
863 close(tfd);
864 pr_err("--re-add for %s to %s is not possible\n",
865 dv->devname, devname);
866 goto abort;
867 }
868 if (array.active_disks < array.raid_disks) {
869 char *avail = xcalloc(array.raid_disks, 1);
870 int d;
871 int found = 0;
872
873 for (d = 0; d < MAX_DISKS && found < array.active_disks; d++) {
874 disc.number = d;
875 if (ioctl(fd, GET_DISK_INFO, &disc))
876 continue;
877 if (disc.major == 0 && disc.minor == 0)
878 continue;
879 if (!(disc.state & (1<<MD_DISK_SYNC)))
880 continue;
881 avail[disc.raid_disk] = 1;
882 found++;
883 }
884 array_failed = !enough(array.level, array.raid_disks,
885 array.layout, 1, avail);
886 } else
887 array_failed = 0;
888 if (array_failed) {
889 pr_err("%s has failed so using --add cannot work and might destroy\n",
890 devname);
891 pr_err("data on %s. You should stop the array and re-assemble it.\n",
892 dv->devname);
893 if (tfd >= 0)
894 close(tfd);
895 goto abort;
896 }
897 } else {
898 /* non-persistent. Must ensure that new drive
899 * is at least array.size big.
900 */
901 if (ldsize/512 < array_size) {
902 pr_err("%s not large enough to join array\n",
903 dv->devname);
904 if (tfd >= 0)
905 close(tfd);
906 goto abort;
907 }
908 }
909 /* committed to really trying this device now*/
910 if (tfd >= 0) {
911 remove_partitions(tfd);
912 close(tfd);
913 tfd = -1;
914 }
915 /* in 2.6.17 and earlier, version-1 superblocks won't
916 * use the number we write, but will choose a free number.
917 * we must choose the same free number, which requires
918 * starting at 'raid_disks' and counting up
919 */
920 for (j = array.raid_disks; j< tst->max_devs; j++) {
921 disc.number = j;
922 if (ioctl(fd, GET_DISK_INFO, &disc))
923 break;
924 if (disc.major==0 && disc.minor==0)
925 break;
926 if (disc.state & 8) /* removed */
927 break;
928 }
929 disc.major = major(stb.st_rdev);
930 disc.minor = minor(stb.st_rdev);
931 disc.number =j;
932 disc.state = 0;
933 if (array.not_persistent==0) {
934 int dfd;
935 if (dv->writemostly == 1)
936 disc.state |= 1 << MD_DISK_WRITEMOSTLY;
937 dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT);
938 if (tst->ss->add_to_super(tst, &disc, dfd,
939 dv->devname)) {
940 close(dfd);
941 goto abort;
942 }
943 if (tst->ss->write_init_super(tst)) {
944 close(dfd);
945 goto abort;
946 }
947 } else if (dv->disposition == 'A') {
948 /* this had better be raid1.
949 * As we are "--re-add"ing we must find a spare slot
950 * to fill.
951 */
952 char *used = xcalloc(array.raid_disks, 1);
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 verbose)
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, verbose < 0);
1216 if (fd < 0)
1217 return 2;
1218
1219 if (!st->ss->update_subarray) {
1220 if (verbose >= 0)
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 (verbose >= 0)
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 && verbose >= 0)
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.writemostly = 0;
1273 devlist.devname = devname;
1274 sprintf(devname, "%d:%d", major(devid), minor(devid));
1275
1276 devlist.disposition = 'r';
1277 if (Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0) == 0) {
1278 devlist.disposition = 'a';
1279 if (Manage_subdevs(to_devname, fd1, &devlist, -1, 0, NULL, 0) == 0) {
1280 /* make sure manager is aware of changes */
1281 ping_manager(to_devname);
1282 ping_manager(from_devname);
1283 close(fd1);
1284 close(fd2);
1285 return 1;
1286 }
1287 else Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0);
1288 }
1289 close(fd1);
1290 close(fd2);
1291 return 0;
1292 }
1293 #endif
Unnamed repository; edit this file 'description' to name the repository.