]> 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
Manage_subdev: give 'st' a better name and narrower focus.
[thirdparty/mdadm.git] / Manage.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2012 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-managed 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 int Manage_runstop(char *devname, int fd, int runstop,
177 int verbose, int will_retry)
178 {
179 /* Run or stop the array. Array must already be configured
180 * 'Run' requires >= 0.90.0
181 * 'will_retry' is only relevant for 'stop', and means
182 * that error messages are not wanted.
183 */
184 mdu_param_t param; /* unused */
185 int rv = 0;
186
187 if (will_retry && verbose == 0)
188 verbose = -1;
189
190 if (runstop == -1 && md_get_version(fd) < 9000) {
191 if (ioctl(fd, STOP_MD, 0) == 0)
192 return 0;
193 pr_err("stopping device %s "
194 "failed: %s\n",
195 devname, strerror(errno));
196 return 1;
197 }
198
199 if (md_get_version(fd) < 9000) {
200 pr_err("need md driver version 0.90.0 or later\n");
201 return 1;
202 }
203
204 if (runstop > 0) {
205 if (ioctl(fd, RUN_ARRAY, &param)) {
206 if (verbose >= 0)
207 pr_err("failed to run array %s: %s\n",
208 devname, strerror(errno));
209 return 1;
210 }
211 if (verbose >= 0)
212 pr_err("started %s\n", devname);
213 } else if (runstop < 0){
214 struct map_ent *map = NULL;
215 struct stat stb;
216 struct mdinfo *mdi;
217 int devnum;
218 int err;
219 int count;
220 /* If this is an mdmon managed array, just write 'inactive'
221 * to the array state and let mdmon clear up.
222 */
223 devnum = fd2devnum(fd);
224 /* Get EXCL access first. If this fails, then attempting
225 * to stop is probably a bad idea.
226 */
227 close(fd);
228 fd = open(devname, O_RDONLY|O_EXCL);
229 if (fd < 0 || fd2devnum(fd) != devnum) {
230 if (fd >= 0)
231 close(fd);
232 if (verbose >= 0)
233 pr_err("Cannot get exclusive access to %s:"
234 "Perhaps a running "
235 "process, mounted filesystem "
236 "or active volume group?\n",
237 devname);
238 return 1;
239 }
240 mdi = sysfs_read(fd, -1, GET_LEVEL|GET_VERSION);
241 if (mdi &&
242 mdi->array.level > 0 &&
243 is_subarray(mdi->text_version)) {
244 int err;
245 /* This is mdmon managed. */
246 close(fd);
247
248 /* As we have an O_EXCL open, any use of the device
249 * which blocks STOP_ARRAY is probably a transient use,
250 * so it is reasonable to retry for a while - 5 seconds.
251 */
252 count = 25;
253 while (count &&
254 (err = sysfs_set_str(mdi, NULL,
255 "array_state",
256 "inactive")) < 0
257 && errno == EBUSY) {
258 usleep(200000);
259 count--;
260 }
261 if (err) {
262 if (verbose >= 0)
263 pr_err("failed to stop array %s: %s\n",
264 devname, strerror(errno));
265 rv = 1;
266 goto out;
267 }
268
269 /* Give monitor a chance to act */
270 ping_monitor(mdi->text_version);
271
272 fd = open_dev_excl(devnum);
273 if (fd < 0) {
274 if (verbose >= 0)
275 pr_err("failed to completely stop %s"
276 ": Device is busy\n",
277 devname);
278 rv = 1;
279 goto out;
280 }
281 } else if (mdi &&
282 mdi->array.major_version == -1 &&
283 mdi->array.minor_version == -2 &&
284 !is_subarray(mdi->text_version)) {
285 struct mdstat_ent *mds, *m;
286 /* container, possibly mdmon-managed.
287 * Make sure mdmon isn't opening it, which
288 * would interfere with the 'stop'
289 */
290 ping_monitor(mdi->sys_name);
291
292 /* now check that there are no existing arrays
293 * which are members of this array
294 */
295 mds = mdstat_read(0, 0);
296 for (m = mds; m; m = m->next)
297 if (m->metadata_version &&
298 strncmp(m->metadata_version, "external:", 9)==0 &&
299 is_subarray(m->metadata_version+9) &&
300 devname2devnum(m->metadata_version+10) == devnum) {
301 if (verbose >= 0)
302 pr_err("Cannot stop container %s: "
303 "member %s still active\n",
304 devname, m->dev);
305 free_mdstat(mds);
306 rv = 1;
307 goto out;
308 }
309 }
310
311 /* As we have an O_EXCL open, any use of the device
312 * which blocks STOP_ARRAY is probably a transient use,
313 * so it is reasonable to retry for a while - 5 seconds.
314 */
315 count = 25; err = 0;
316 while (count && fd >= 0
317 && (err = ioctl(fd, STOP_ARRAY, NULL)) < 0
318 && errno == EBUSY) {
319 usleep(200000);
320 count --;
321 }
322 if (fd >= 0 && err) {
323 if (verbose >= 0) {
324 pr_err("failed to stop array %s: %s\n",
325 devname, strerror(errno));
326 if (errno == EBUSY)
327 fprintf(stderr, "Perhaps a running "
328 "process, mounted filesystem "
329 "or active volume group?\n");
330 }
331 rv = 1;
332 goto out;
333 }
334 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
335 * was stopped, so We'll do it here just to be sure. Drop any
336 * partitions as well...
337 */
338 if (fd >= 0)
339 ioctl(fd, BLKRRPART, 0);
340 if (mdi)
341 sysfs_uevent(mdi, "change");
342
343 if (devnum != NoMdDev &&
344 (stat("/dev/.udev", &stb) != 0 ||
345 check_env("MDADM_NO_UDEV"))) {
346 struct map_ent *mp = map_by_devnum(&map, devnum);
347 remove_devices(devnum, mp ? mp->path : NULL);
348 }
349
350 if (verbose >= 0)
351 pr_err("stopped %s\n", devname);
352 map_lock(&map);
353 map_remove(&map, devnum);
354 map_unlock(&map);
355 out:
356 if (mdi)
357 sysfs_free(mdi);
358 }
359 return rv;
360 }
361
362 static void add_faulty(struct mddev_dev *dv, int fd, char disp)
363 {
364 mdu_array_info_t array;
365 mdu_disk_info_t disk;
366 int remaining_disks;
367 int i;
368
369 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0)
370 return;
371
372 remaining_disks = array.nr_disks;
373 for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) {
374 struct mddev_dev *new;
375 char buf[40];
376 disk.number = i;
377 if (ioctl(fd, GET_DISK_INFO, &disk) != 0)
378 continue;
379 if (disk.major == 0 && disk.minor == 0)
380 continue;
381 remaining_disks--;
382 if ((disk.state & 1) == 0) /* not faulty */
383 continue;
384 sprintf(buf, "%d:%d", disk.major, disk.minor);
385 new = xmalloc(sizeof(*new));
386 new->devname = xstrdup(buf);
387 new->disposition = disp;
388 new->next = dv->next;
389 dv->next = new;
390 dv = new;
391 }
392 }
393
394 static void add_detached(struct mddev_dev *dv, int fd, char disp)
395 {
396 mdu_array_info_t array;
397 mdu_disk_info_t disk;
398 int remaining_disks;
399 int i;
400
401 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0)
402 return;
403
404 remaining_disks = array.nr_disks;
405 for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) {
406 struct mddev_dev *new;
407 char buf[40];
408 int sfd;
409 disk.number = i;
410 if (ioctl(fd, GET_DISK_INFO, &disk) != 0)
411 continue;
412 if (disk.major == 0 && disk.minor == 0)
413 continue;
414 remaining_disks--;
415 if (disp == 'f' && (disk.state & 1) != 0) /* already faulty */
416 continue;
417 sprintf(buf, "%d:%d", disk.major, disk.minor);
418 sfd = dev_open(buf, O_RDONLY);
419 if (sfd >= 0) {
420 /* Not detached */
421 close(sfd);
422 continue;
423 }
424 if (errno != ENXIO)
425 /* Probably not detached */
426 continue;
427 new = xmalloc(sizeof(*new));
428 new->devname = xstrdup(buf);
429 new->disposition = disp;
430 new->next = dv->next;
431 dv->next = new;
432 dv = new;
433 }
434 }
435
436 int Manage_subdevs(char *devname, int fd,
437 struct mddev_dev *devlist, int verbose, int test,
438 char *update, int force)
439 {
440 /* Do something to each dev.
441 * devmode can be
442 * 'a' - add the device
443 * try HOT_ADD_DISK
444 * If that fails EINVAL, try ADD_NEW_DISK
445 * 'A' - re-add the device
446 * 'r' - remove the device: HOT_REMOVE_DISK
447 * device can be 'faulty' or 'detached' in which case all
448 * matching devices are removed.
449 * 'f' - set the device faulty SET_DISK_FAULTY
450 * device can be 'detached' in which case any device that
451 * is inaccessible will be marked faulty.
452 * For 'f' and 'r', the device can also be a kernel-internal
453 * name such as 'sdb'.
454 */
455 mdu_array_info_t array;
456 mdu_disk_info_t disc;
457 unsigned long long array_size;
458 struct mddev_dev *dv;
459 struct stat stb;
460 int j;
461 int tfd = -1;
462 struct supertype *dev_st, *tst;
463 char *subarray = NULL;
464 int duuid[4];
465 int ouuid[4];
466 int lfd = -1;
467 int sysfd = -1;
468 int count = 0; /* number of actions taken */
469 struct mdinfo info;
470 int frozen = 0;
471
472 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
473 pr_err("Cannot get array info for %s\n",
474 devname);
475 goto abort;
476 }
477 sysfs_init(&info, fd, 0);
478
479 /* array.size is only 32 bits and may be truncated.
480 * So read from sysfs if possible, and record number of sectors
481 */
482
483 array_size = get_component_size(fd);
484 if (array_size <= 0)
485 array_size = array.size * 2;
486
487 tst = super_by_fd(fd, &subarray);
488 if (!tst) {
489 pr_err("unsupport array - version %d.%d\n",
490 array.major_version, array.minor_version);
491 goto abort;
492 }
493
494 stb.st_rdev = 0;
495 for (dv = devlist, j=0 ; dv; dv = dv->next) {
496 unsigned long long ldsize;
497 int err;
498 int array_failed;
499
500 if (strcmp(dv->devname, "failed") == 0 ||
501 strcmp(dv->devname, "faulty") == 0) {
502 if (dv->disposition != 'r') {
503 pr_err("%s only meaningful "
504 "with -r, not -%c\n",
505 dv->devname, dv->disposition);
506 goto abort;
507 }
508 add_faulty(dv, fd, 'r');
509 continue;
510 }
511 if (strcmp(dv->devname, "detached") == 0) {
512 if (dv->disposition != 'r' && dv->disposition != 'f') {
513 pr_err("%s only meaningful "
514 "with -r of -f, not -%c\n",
515 dv->devname, dv->disposition);
516 goto abort;
517 }
518 add_detached(dv, fd, dv->disposition);
519 continue;
520 }
521
522 if (strcmp(dv->devname, "missing") == 0) {
523 struct mddev_dev *add_devlist = NULL;
524 struct mddev_dev **dp;
525 if (dv->disposition != 'A') {
526 pr_err("'missing' only meaningful "
527 "with --re-add\n");
528 goto abort;
529 }
530 add_devlist = conf_get_devs();
531 if (add_devlist == NULL) {
532 pr_err("no devices to scan for missing members.");
533 continue;
534 }
535 for (dp = &add_devlist; *dp; dp = & (*dp)->next)
536 /* 'M' (for 'missing') is like 'A' without errors */
537 (*dp)->disposition = 'M';
538 *dp = dv->next;
539 dv->next = add_devlist;
540 continue;
541 }
542
543 if (strchr(dv->devname, '/') == NULL &&
544 strchr(dv->devname, ':') == NULL &&
545 strlen(dv->devname) < 50) {
546 /* Assume this is a kernel-internal name like 'sda1' */
547 int found = 0;
548 char dname[55];
549 if (dv->disposition != 'r' && dv->disposition != 'f') {
550 pr_err("%s only meaningful "
551 "with -r or -f, not -%c\n",
552 dv->devname, dv->disposition);
553 goto abort;
554 }
555
556 sprintf(dname, "dev-%s", dv->devname);
557 sysfd = sysfs_open(fd2devnum(fd), dname, "block/dev");
558 if (sysfd >= 0) {
559 char dn[20];
560 int mj,mn;
561 if (sysfs_fd_get_str(sysfd, dn, 20) > 0 &&
562 sscanf(dn, "%d:%d", &mj,&mn) == 2) {
563 stb.st_rdev = makedev(mj,mn);
564 found = 1;
565 }
566 close(sysfd);
567 sysfd = -1;
568 }
569 if (!found) {
570 sysfd = sysfs_open(fd2devnum(fd), dname, "state");
571 if (sysfd < 0) {
572 pr_err("%s does not appear "
573 "to be a component of %s\n",
574 dv->devname, devname);
575 goto abort;
576 }
577 }
578 } else {
579 j = 0;
580
581 tfd = dev_open(dv->devname, O_RDONLY);
582 if (tfd < 0 && dv->disposition == 'r' &&
583 lstat(dv->devname, &stb) == 0)
584 /* Be happy, the lstat worked, that is
585 * enough for --remove
586 */
587 ;
588 else {
589 if (tfd < 0 || fstat(tfd, &stb) != 0) {
590 if (tfd >= 0)
591 close(tfd);
592 if (dv->disposition == 'M')
593 /* non-fatal */
594 continue;
595 pr_err("cannot find %s: %s\n",
596 dv->devname, strerror(errno));
597 goto abort;
598 }
599 close(tfd);
600 tfd = -1;
601 }
602 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
603 if (dv->disposition == 'M')
604 /* non-fatal. Also improbable */
605 continue;
606 pr_err("%s is not a "
607 "block device.\n",
608 dv->devname);
609 goto abort;
610 }
611 }
612 switch(dv->disposition){
613 default:
614 pr_err("internal error - devmode[%s]=%d\n",
615 dv->devname, dv->disposition);
616 goto abort;
617 case 'a':
618 case 'A':
619 case 'M':
620 /* add the device */
621 if (subarray) {
622 pr_err("Cannot add disks to a"
623 " \'member\' array, perform this"
624 " operation on the parent container\n");
625 goto abort;
626 }
627 /* Make sure it isn't in use (in 2.6 or later) */
628 tfd = dev_open(dv->devname, O_RDONLY|O_EXCL);
629 if (tfd < 0) {
630 if (dv->disposition == 'M')
631 continue;
632 pr_err("Cannot open %s: %s\n",
633 dv->devname, strerror(errno));
634 goto abort;
635 }
636 if (!frozen) {
637 if (sysfs_freeze_array(&info) == 1)
638 frozen = 1;
639 else
640 frozen = -1;
641 }
642 if (!get_dev_size(tfd, dv->devname, &ldsize)) {
643 close(tfd);
644 tfd = -1;
645 if (dv->disposition == 'M')
646 continue;
647 else
648 goto abort;
649 }
650
651 if (tst->ss->validate_geometry(
652 tst, array.level, array.layout,
653 array.raid_disks, NULL,
654 ldsize >> 9, NULL, NULL, 0) == 0) {
655 if (!force) {
656 pr_err("%s is larger than %s can "
657 "effectively use.\n"
658 " Add --force is you "
659 "really want to add this device.\n",
660 dv->devname, devname);
661 close(tfd);
662 goto abort;
663 }
664 pr_err("%s is larger than %s can "
665 "effectively use.\n"
666 " Adding anyway as --force "
667 "was given.\n",
668 dv->devname, devname);
669 }
670 if (!tst->ss->external &&
671 array.major_version == 0 &&
672 md_get_version(fd)%100 < 2) {
673 close(tfd);
674 tfd = -1;
675 if (ioctl(fd, HOT_ADD_DISK,
676 (unsigned long)stb.st_rdev)==0) {
677 if (verbose >= 0)
678 pr_err("hot added %s\n",
679 dv->devname);
680 continue;
681 }
682
683 pr_err("hot add failed for %s: %s\n",
684 dv->devname, strerror(errno));
685 goto abort;
686 }
687
688 if (array.not_persistent == 0 || tst->ss->external) {
689
690 /* need to find a sample superblock to copy, and
691 * a spare slot to use.
692 * For 'external' array (well, container based),
693 * We can just load the metadata for the array.
694 */
695 if (tst->sb)
696 /* already loaded */;
697 else if (tst->ss->external) {
698 tst->ss->load_container(tst, fd, NULL);
699 } else for (j = 0; j < tst->max_devs; j++) {
700 char *dev;
701 int dfd;
702 disc.number = j;
703 if (ioctl(fd, GET_DISK_INFO, &disc))
704 continue;
705 if (disc.major==0 && disc.minor==0)
706 continue;
707 if ((disc.state & 4)==0) /* sync */
708 continue;
709 /* Looks like a good device to try */
710 dev = map_dev(disc.major, disc.minor, 1);
711 if (!dev)
712 continue;
713 dfd = dev_open(dev, O_RDONLY);
714 if (dfd < 0)
715 continue;
716 if (tst->ss->load_super(tst, dfd,
717 NULL)) {
718 close(dfd);
719 continue;
720 }
721 close(dfd);
722 break;
723 }
724 /* FIXME this is a bad test to be using */
725 if (!tst->sb && dv->disposition != 'a') {
726 /* we are re-adding a device to a
727 * completely dead array - have to depend
728 * on kernel to check
729 */
730 } else if (!tst->sb) {
731 close(tfd);
732 pr_err("cannot load array metadata from %s\n", devname);
733 goto abort;
734 }
735
736 /* Make sure device is large enough */
737 if (tst->ss->avail_size(tst, ldsize/512) <
738 array_size) {
739 close(tfd);
740 tfd = -1;
741 if (dv->disposition == 'M')
742 continue;
743 pr_err("%s not large enough to join array\n",
744 dv->devname);
745 goto abort;
746 }
747
748 /* Possibly this device was recently part of
749 * the array and was temporarily removed, and
750 * is now being re-added. If so, we can
751 * simply re-add it.
752 */
753
754 if (array.not_persistent==0) {
755 dev_st = dup_super(tst);
756 dev_st->ss->load_super(dev_st, tfd, NULL);
757 }
758 if (dev_st && dev_st->sb) {
759 struct mdinfo mdi;
760 dev_st->ss->getinfo_super(dev_st, &mdi, NULL);
761 dev_st->ss->uuid_from_super(dev_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 /* Looks like it is worth a
771 * try. Need to make sure
772 * kernel will accept it
773 * though.
774 */
775 /* re-add doesn't work for version-1 superblocks
776 * before 2.6.18 :-(
777 */
778 if (array.major_version == 1 &&
779 get_linux_version() <= 2006018)
780 goto skip_re_add;
781 disc.number = mdi.disk.number;
782 if (ioctl(fd, GET_DISK_INFO, &disc) != 0
783 || disc.major != 0 || disc.minor != 0
784 )
785 goto skip_re_add;
786 disc.major = major(stb.st_rdev);
787 disc.minor = minor(stb.st_rdev);
788 disc.number = mdi.disk.number;
789 disc.raid_disk = mdi.disk.raid_disk;
790 disc.state = mdi.disk.state;
791 if (dv->writemostly == 1)
792 disc.state |= 1 << MD_DISK_WRITEMOSTLY;
793 if (dv->writemostly == 2)
794 disc.state &= ~(1 << MD_DISK_WRITEMOSTLY);
795 remove_partitions(tfd);
796 close(tfd);
797 tfd = -1;
798 if (update || dv->writemostly > 0) {
799 int rv = -1;
800 tfd = dev_open(dv->devname, O_RDWR);
801 if (tfd < 0) {
802 pr_err("failed to open %s for"
803 " superblock update during re-add\n", dv->devname);
804 dev_st->ss->free_super(dev_st);
805 goto abort;
806 }
807
808 if (dv->writemostly == 1)
809 rv = dev_st->ss->update_super(
810 dev_st, NULL, "writemostly",
811 devname, verbose, 0, NULL);
812 if (dv->writemostly == 2)
813 rv = dev_st->ss->update_super(
814 dev_st, NULL, "readwrite",
815 devname, verbose, 0, NULL);
816 if (update)
817 rv = dev_st->ss->update_super(
818 dev_st, NULL, update,
819 devname, verbose, 0, NULL);
820 if (rv == 0)
821 rv = dev_st->ss->store_super(dev_st, tfd);
822 close(tfd);
823 tfd = -1;
824 if (rv != 0) {
825 pr_err("failed to update"
826 " superblock during re-add\n");
827 dev_st->ss->free_super(dev_st);
828 goto abort;
829 }
830 }
831 /* don't even try if disk is marked as faulty */
832 errno = 0;
833 if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) {
834 if (verbose >= 0)
835 pr_err("re-added %s\n", dv->devname);
836 count++;
837 dev_st->ss->free_super(dev_st);
838 continue;
839 }
840 if (errno == ENOMEM || errno == EROFS) {
841 pr_err("add new device failed for %s: %s\n",
842 dv->devname, strerror(errno));
843 dev_st->ss->free_super(dev_st);
844 if (dv->disposition == 'M')
845 continue;
846 goto abort;
847 }
848 }
849 skip_re_add:
850 dev_st->ss->free_super(dev_st);
851 }
852 if (dv->disposition == 'M') {
853 if (verbose > 0)
854 pr_err("--re-add for %s to %s is not possible\n",
855 dv->devname, devname);
856 if (tfd >= 0) {
857 close(tfd);
858 tfd = -1;
859 }
860 continue;
861 }
862 if (dv->disposition == 'A') {
863 if (tfd >= 0)
864 close(tfd);
865 pr_err("--re-add for %s to %s is not possible\n",
866 dv->devname, devname);
867 goto abort;
868 }
869 if (array.active_disks < array.raid_disks) {
870 char *avail = xcalloc(array.raid_disks, 1);
871 int d;
872 int found = 0;
873
874 for (d = 0; d < MAX_DISKS && found < array.active_disks; d++) {
875 disc.number = d;
876 if (ioctl(fd, GET_DISK_INFO, &disc))
877 continue;
878 if (disc.major == 0 && disc.minor == 0)
879 continue;
880 if (!(disc.state & (1<<MD_DISK_SYNC)))
881 continue;
882 avail[disc.raid_disk] = 1;
883 found++;
884 }
885 array_failed = !enough(array.level, array.raid_disks,
886 array.layout, 1, avail);
887 } else
888 array_failed = 0;
889 if (array_failed) {
890 pr_err("%s has failed so using --add cannot work and might destroy\n",
891 devname);
892 pr_err("data on %s. You should stop the array and re-assemble it.\n",
893 dv->devname);
894 if (tfd >= 0)
895 close(tfd);
896 goto abort;
897 }
898 } else {
899 /* non-persistent. Must ensure that new drive
900 * is at least array.size big.
901 */
902 if (ldsize/512 < array_size) {
903 pr_err("%s not large enough to join array\n",
904 dv->devname);
905 if (tfd >= 0)
906 close(tfd);
907 goto abort;
908 }
909 }
910 /* committed to really trying this device now*/
911 if (tfd >= 0) {
912 remove_partitions(tfd);
913 close(tfd);
914 tfd = -1;
915 }
916 /* in 2.6.17 and earlier, version-1 superblocks won't
917 * use the number we write, but will choose a free number.
918 * we must choose the same free number, which requires
919 * starting at 'raid_disks' and counting up
920 */
921 for (j = array.raid_disks; j < tst->max_devs; j++) {
922 disc.number = j;
923 if (ioctl(fd, GET_DISK_INFO, &disc))
924 break;
925 if (disc.major==0 && disc.minor==0)
926 break;
927 if (disc.state & 8) /* removed */
928 break;
929 }
930 disc.major = major(stb.st_rdev);
931 disc.minor = minor(stb.st_rdev);
932 disc.number =j;
933 disc.state = 0;
934 if (array.not_persistent==0) {
935 int dfd;
936 if (dv->writemostly == 1)
937 disc.state |= 1 << MD_DISK_WRITEMOSTLY;
938 dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT);
939 if (tst->ss->add_to_super(tst, &disc, dfd,
940 dv->devname)) {
941 close(dfd);
942 goto abort;
943 }
944 if (tst->ss->write_init_super(tst)) {
945 close(dfd);
946 goto abort;
947 }
948 } else if (dv->disposition == 'A') {
949 /* this had better be raid1.
950 * As we are "--re-add"ing we must find a spare slot
951 * to fill.
952 */
953 char *used = xcalloc(array.raid_disks, 1);
954 for (j = 0; j < tst->max_devs; j++) {
955 mdu_disk_info_t disc2;
956 disc2.number = j;
957 if (ioctl(fd, GET_DISK_INFO, &disc2))
958 continue;
959 if (disc2.major==0 && disc2.minor==0)
960 continue;
961 if (disc2.state & 8) /* removed */
962 continue;
963 if (disc2.raid_disk < 0)
964 continue;
965 if (disc2.raid_disk > array.raid_disks)
966 continue;
967 used[disc2.raid_disk] = 1;
968 }
969 for (j = 0 ; j < array.raid_disks; j++)
970 if (!used[j]) {
971 disc.raid_disk = j;
972 disc.state |= (1<<MD_DISK_SYNC);
973 break;
974 }
975 free(used);
976 }
977 if (dv->writemostly == 1)
978 disc.state |= (1 << MD_DISK_WRITEMOSTLY);
979 if (tst->ss->external) {
980 /* add a disk
981 * to an external metadata container */
982 struct mdinfo new_mdi;
983 struct mdinfo *sra;
984 int container_fd;
985 int devnum = fd2devnum(fd);
986 int dfd;
987
988 container_fd = open_dev_excl(devnum);
989 if (container_fd < 0) {
990 pr_err("add failed for %s:"
991 " could not get exclusive access to container\n",
992 dv->devname);
993 tst->ss->free_super(tst);
994 goto abort;
995 }
996
997 dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT);
998 if (mdmon_running(tst->container_dev))
999 tst->update_tail = &tst->updates;
1000 if (tst->ss->add_to_super(tst, &disc, dfd,
1001 dv->devname)) {
1002 close(dfd);
1003 close(container_fd);
1004 goto abort;
1005 }
1006 if (tst->update_tail)
1007 flush_metadata_updates(tst);
1008 else
1009 tst->ss->sync_metadata(tst);
1010
1011 sra = sysfs_read(container_fd, -1, 0);
1012 if (!sra) {
1013 pr_err("add failed for %s: sysfs_read failed\n",
1014 dv->devname);
1015 close(container_fd);
1016 tst->ss->free_super(tst);
1017 goto abort;
1018 }
1019 sra->array.level = LEVEL_CONTAINER;
1020 /* Need to set data_offset and component_size */
1021 tst->ss->getinfo_super(tst, &new_mdi, NULL);
1022 new_mdi.disk.major = disc.major;
1023 new_mdi.disk.minor = disc.minor;
1024 new_mdi.recovery_start = 0;
1025 /* Make sure fds are closed as they are O_EXCL which
1026 * would block add_disk */
1027 tst->ss->free_super(tst);
1028 if (sysfs_add_disk(sra, &new_mdi, 0) != 0) {
1029 pr_err("add new device to external metadata"
1030 " failed for %s\n", dv->devname);
1031 close(container_fd);
1032 sysfs_free(sra);
1033 goto abort;
1034 }
1035 ping_monitor_by_id(devnum);
1036 sysfs_free(sra);
1037 close(container_fd);
1038 } else {
1039 tst->ss->free_super(tst);
1040 if (ioctl(fd, ADD_NEW_DISK, &disc)) {
1041 pr_err("add new device failed for %s as %d: %s\n",
1042 dv->devname, j, strerror(errno));
1043 goto abort;
1044 }
1045 }
1046 if (verbose >= 0)
1047 pr_err("added %s\n", dv->devname);
1048 break;
1049
1050 case 'r':
1051 /* hot remove */
1052 if (subarray) {
1053 pr_err("Cannot remove disks from a"
1054 " \'member\' array, perform this"
1055 " operation on the parent container\n");
1056 if (sysfd >= 0)
1057 close(sysfd);
1058 goto abort;
1059 }
1060 if (tst->ss->external) {
1061 /* To remove a device from a container, we must
1062 * check that it isn't in use in an array.
1063 * This involves looking in the 'holders'
1064 * directory - there must be just one entry,
1065 * the container.
1066 * To ensure that it doesn't get used as a
1067 * hot spare while we are checking, we
1068 * get an O_EXCL open on the container
1069 */
1070 int dnum = fd2devnum(fd);
1071 lfd = open_dev_excl(dnum);
1072 if (lfd < 0) {
1073 pr_err("Cannot get exclusive access "
1074 " to container - odd\n");
1075 if (sysfd >= 0)
1076 close(sysfd);
1077 goto abort;
1078 }
1079 /* In the detached case it is not possible to
1080 * check if we are the unique holder, so just
1081 * rely on the 'detached' checks
1082 */
1083 if (strcmp(dv->devname, "detached") == 0 ||
1084 sysfd >= 0 ||
1085 sysfs_unique_holder(dnum, stb.st_rdev))
1086 /* pass */;
1087 else {
1088 pr_err("%s is %s, cannot remove.\n",
1089 dv->devname,
1090 errno == EEXIST ? "still in use":
1091 "not a member");
1092 close(lfd);
1093 goto abort;
1094 }
1095 }
1096 /* FIXME check that it is a current member */
1097 if (sysfd >= 0) {
1098 /* device has been removed and we don't know
1099 * the major:minor number
1100 */
1101 int n = write(sysfd, "remove", 6);
1102 if (n != 6)
1103 err = -1;
1104 else
1105 err = 0;
1106 close(sysfd);
1107 sysfd = -1;
1108 } else {
1109 err = ioctl(fd, HOT_REMOVE_DISK, (unsigned long)stb.st_rdev);
1110 if (err && errno == ENODEV) {
1111 /* Old kernels rejected this if no personality
1112 * is registered */
1113 struct mdinfo *sra = sysfs_read(fd, 0, GET_DEVS);
1114 struct mdinfo *dv = NULL;
1115 if (sra)
1116 dv = sra->devs;
1117 for ( ; dv ; dv=dv->next)
1118 if (dv->disk.major == (int)major(stb.st_rdev) &&
1119 dv->disk.minor == (int)minor(stb.st_rdev))
1120 break;
1121 if (dv)
1122 err = sysfs_set_str(sra, dv,
1123 "state", "remove");
1124 else
1125 err = -1;
1126 if (sra)
1127 sysfs_free(sra);
1128 }
1129 }
1130 if (err) {
1131 pr_err("hot remove failed "
1132 "for %s: %s\n", dv->devname,
1133 strerror(errno));
1134 if (lfd >= 0)
1135 close(lfd);
1136 goto abort;
1137 }
1138 if (tst->ss->external) {
1139 /*
1140 * Before dropping our exclusive open we make an
1141 * attempt at preventing mdmon from seeing an
1142 * 'add' event before reconciling this 'remove'
1143 * event.
1144 */
1145 char *name = devnum2devname(fd2devnum(fd));
1146
1147 if (!name) {
1148 pr_err("unable to get container name\n");
1149 goto abort;
1150 }
1151
1152 ping_manager(name);
1153 free(name);
1154 }
1155 if (lfd >= 0)
1156 close(lfd);
1157 count++;
1158 if (verbose >= 0)
1159 pr_err("hot removed %s from %s\n",
1160 dv->devname, devname);
1161 break;
1162
1163 case 'f': /* set faulty */
1164 /* FIXME check current member */
1165 if ((sysfd >= 0 && write(sysfd, "faulty", 6) != 6) ||
1166 (sysfd < 0 && ioctl(fd, SET_DISK_FAULTY,
1167 (unsigned long) stb.st_rdev))) {
1168 pr_err("set device faulty failed for %s: %s\n",
1169 dv->devname, strerror(errno));
1170 if (sysfd >= 0)
1171 close(sysfd);
1172 goto abort;
1173 }
1174 if (sysfd >= 0)
1175 close(sysfd);
1176 sysfd = -1;
1177 count++;
1178 if (verbose >= 0)
1179 pr_err("set %s faulty in %s\n",
1180 dv->devname, devname);
1181 break;
1182 }
1183 }
1184 if (frozen > 0)
1185 sysfs_set_str(&info, NULL, "sync_action","idle");
1186 if (test && count == 0)
1187 return 2;
1188 return 0;
1189
1190 abort:
1191 if (frozen > 0)
1192 sysfs_set_str(&info, NULL, "sync_action","idle");
1193 return 1;
1194 }
1195
1196 int autodetect(void)
1197 {
1198 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1199 int rv = 1;
1200 int fd = dev_open("9:0", O_RDONLY);
1201 if (fd >= 0) {
1202 if (ioctl(fd, RAID_AUTORUN, 0) == 0)
1203 rv = 0;
1204 close(fd);
1205 }
1206 return rv;
1207 }
1208
1209 int Update_subarray(char *dev, char *subarray, char *update, struct mddev_ident *ident, int verbose)
1210 {
1211 struct supertype supertype, *st = &supertype;
1212 int fd, rv = 2;
1213
1214 memset(st, 0, sizeof(*st));
1215
1216 fd = open_subarray(dev, subarray, st, verbose < 0);
1217 if (fd < 0)
1218 return 2;
1219
1220 if (!st->ss->update_subarray) {
1221 if (verbose >= 0)
1222 pr_err("Operation not supported for %s metadata\n",
1223 st->ss->name);
1224 goto free_super;
1225 }
1226
1227 if (mdmon_running(st->devnum))
1228 st->update_tail = &st->updates;
1229
1230 rv = st->ss->update_subarray(st, subarray, update, ident);
1231
1232 if (rv) {
1233 if (verbose >= 0)
1234 pr_err("Failed to update %s of subarray-%s in %s\n",
1235 update, subarray, dev);
1236 } else if (st->update_tail)
1237 flush_metadata_updates(st);
1238 else
1239 st->ss->sync_metadata(st);
1240
1241 if (rv == 0 && strcmp(update, "name") == 0 && verbose >= 0)
1242 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1243 subarray, dev);
1244
1245 free_super:
1246 st->ss->free_super(st);
1247 close(fd);
1248
1249 return rv;
1250 }
1251
1252 /* Move spare from one array to another 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.