]> 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
Fix memory leak of 'st3' in array_try_spare()
[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
48 if (md_get_version(fd) < 9000) {
49 fprintf(stderr, Name ": need md driver version 0.90.0 or later\n");
50 return 1;
51 }
52 #ifndef MDASSEMBLE
53 /* If this is an externally-manage array, we need to modify the
54 * metadata_version so that mdmon doesn't undo our change.
55 */
56 mdi = sysfs_read(fd, -1, GET_LEVEL|GET_VERSION);
57 if (mdi &&
58 mdi->array.major_version == -1 &&
59 is_subarray(mdi->text_version)) {
60 char vers[64];
61 strcpy(vers, "external:");
62 strcat(vers, mdi->text_version);
63 if (readonly > 0) {
64 int rv;
65 /* We set readonly ourselves. */
66 vers[9] = '-';
67 sysfs_set_str(mdi, NULL, "metadata_version", vers);
68
69 close(fd);
70 rv = sysfs_set_str(mdi, NULL, "array_state", "readonly");
71
72 if (rv < 0) {
73 fprintf(stderr, Name ": failed to set readonly for %s: %s\n",
74 devname, strerror(errno));
75
76 vers[9] = mdi->text_version[0];
77 sysfs_set_str(mdi, NULL, "metadata_version", vers);
78 return 1;
79 }
80 } else {
81 char *cp;
82 /* We cannot set read/write - must signal mdmon */
83 vers[9] = '/';
84 sysfs_set_str(mdi, NULL, "metadata_version", vers);
85
86 cp = strchr(vers+10, '/');
87 if (*cp)
88 *cp = 0;
89 ping_monitor(vers+10);
90 if (mdi->array.level <= 0)
91 sysfs_set_str(mdi, NULL, "array_state", "active");
92 }
93 return 0;
94 }
95 #endif
96 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
97 fprintf(stderr, Name ": %s does not appear to be active.\n",
98 devname);
99 return 1;
100 }
101
102 if (readonly>0) {
103 if (ioctl(fd, STOP_ARRAY_RO, NULL)) {
104 fprintf(stderr, Name ": failed to set readonly for %s: %s\n",
105 devname, strerror(errno));
106 return 1;
107 }
108 } else if (readonly < 0) {
109 if (ioctl(fd, RESTART_ARRAY_RW, NULL)) {
110 fprintf(stderr, Name ": failed to set writable for %s: %s\n",
111 devname, strerror(errno));
112 return 1;
113 }
114 }
115 return 0;
116 }
117
118 #ifndef MDASSEMBLE
119
120 static void remove_devices(int devnum, char *path)
121 {
122 /*
123 * Remove names at 'path' - possibly with
124 * partition suffixes - which link to the 'standard'
125 * name for devnum. These were probably created
126 * by mdadm when the array was assembled.
127 */
128 char base[40];
129 char *path2;
130 char link[1024];
131 int n;
132 int part;
133 char *be;
134 char *pe;
135
136 if (!path)
137 return;
138
139 if (devnum >= 0)
140 sprintf(base, "/dev/md%d", devnum);
141 else
142 sprintf(base, "/dev/md_d%d", -1-devnum);
143 be = base + strlen(base);
144
145 path2 = malloc(strlen(path)+20);
146 strcpy(path2, path);
147 pe = path2 + strlen(path2);
148
149 for (part = 0; part < 16; part++) {
150 if (part) {
151 sprintf(be, "p%d", part);
152
153 if (isdigit(pe[-1]))
154 sprintf(pe, "p%d", part);
155 else
156 sprintf(pe, "%d", part);
157 }
158 n = readlink(path2, link, sizeof(link));
159 if (n && (int)strlen(base) == n &&
160 strncmp(link, base, n) == 0)
161 unlink(path2);
162 }
163 free(path2);
164 }
165
166
167 int Manage_runstop(char *devname, int fd, int runstop, int quiet)
168 {
169 /* Run or stop the array. array must already be configured
170 * required >= 0.90.0
171 * Only print failure messages if quiet == 0;
172 * quiet > 0 means really be quiet
173 * quiet < 0 means we will try again if it fails.
174 */
175 mdu_param_t param; /* unused */
176
177 if (runstop == -1 && md_get_version(fd) < 9000) {
178 if (ioctl(fd, STOP_MD, 0)) {
179 if (quiet == 0) fprintf(stderr,
180 Name ": stopping device %s "
181 "failed: %s\n",
182 devname, strerror(errno));
183 return 1;
184 }
185 }
186
187 if (md_get_version(fd) < 9000) {
188 fprintf(stderr, Name ": need md driver version 0.90.0 or later\n");
189 return 1;
190 }
191 /*
192 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
193 fprintf(stderr, Name ": %s does not appear to be active.\n",
194 devname);
195 return 1;
196 }
197 */
198 if (runstop>0) {
199 if (ioctl(fd, RUN_ARRAY, &param)) {
200 fprintf(stderr, Name ": failed to run array %s: %s\n",
201 devname, strerror(errno));
202 return 1;
203 }
204 if (quiet <= 0)
205 fprintf(stderr, Name ": started %s\n", devname);
206 } else if (runstop < 0){
207 struct map_ent *map = NULL;
208 struct stat stb;
209 struct mdinfo *mdi;
210 int devnum;
211 int err;
212 int count;
213 /* If this is an mdmon managed array, just write 'inactive'
214 * to the array state and let mdmon clear up.
215 */
216 devnum = fd2devnum(fd);
217 /* Get EXCL access first. If this fails, then attempting
218 * to stop is probably a bad idea.
219 */
220 close(fd);
221 fd = open(devname, O_RDONLY|O_EXCL);
222 if (fd < 0 || fd2devnum(fd) != devnum) {
223 if (fd >= 0)
224 close(fd);
225 fprintf(stderr,
226 Name ": Cannot get exclusive access to %s:"
227 "Perhaps a running "
228 "process, mounted filesystem "
229 "or active volume group?\n",
230 devname);
231 return 1;
232 }
233 mdi = sysfs_read(fd, -1, GET_LEVEL|GET_VERSION);
234 if (mdi &&
235 mdi->array.level > 0 &&
236 is_subarray(mdi->text_version)) {
237 int err;
238 /* This is mdmon managed. */
239 close(fd);
240
241 count = 25;
242 while (count &&
243 (err = sysfs_set_str(mdi, NULL,
244 "array_state",
245 "inactive")) < 0
246 && errno == EBUSY) {
247 usleep(200000);
248 count--;
249 }
250 if (err && !quiet) {
251 fprintf(stderr, Name
252 ": failed to stop array %s: %s\n",
253 devname, strerror(errno));
254 return 1;
255 }
256
257 /* Give monitor a chance to act */
258 ping_monitor(mdi->text_version);
259
260 fd = open_dev_excl(devnum);
261 if (fd < 0) {
262 fprintf(stderr, Name
263 ": failed to completely stop %s"
264 ": Device is busy\n",
265 devname);
266 return 1;
267 }
268 } else if (mdi &&
269 mdi->array.major_version == -1 &&
270 mdi->array.minor_version == -2 &&
271 !is_subarray(mdi->text_version)) {
272 struct mdstat_ent *mds, *m;
273 /* container, possibly mdmon-managed.
274 * Make sure mdmon isn't opening it, which
275 * would interfere with the 'stop'
276 */
277 ping_monitor(mdi->sys_name);
278
279 /* now check that there are no existing arrays
280 * which are members of this array
281 */
282 mds = mdstat_read(0, 0);
283 for (m=mds; m; m=m->next)
284 if (m->metadata_version &&
285 strncmp(m->metadata_version, "external:", 9)==0 &&
286 is_subarray(m->metadata_version+9) &&
287 devname2devnum(m->metadata_version+10) == devnum) {
288 if (!quiet)
289 fprintf(stderr, Name
290 ": Cannot stop container %s: "
291 "member %s still active\n",
292 devname, m->dev);
293 free_mdstat(mds);
294 if (mdi)
295 sysfs_free(mdi);
296 return 1;
297 }
298 }
299
300 /* As we have an O_EXCL open, any use of the device
301 * which blocks STOP_ARRAY is probably a transient use,
302 * so it is reasonable to retry for a while - 5 seconds.
303 */
304 count = 25; err = 0;
305 while (count && fd >= 0
306 && (err = ioctl(fd, STOP_ARRAY, NULL)) < 0
307 && errno == EBUSY) {
308 usleep(200000);
309 count --;
310 }
311 if (fd >= 0 && err) {
312 if (quiet == 0) {
313 fprintf(stderr, Name
314 ": failed to stop array %s: %s\n",
315 devname, strerror(errno));
316 if (errno == EBUSY)
317 fprintf(stderr, "Perhaps a running "
318 "process, mounted filesystem "
319 "or active volume group?\n");
320 }
321 if (mdi)
322 sysfs_free(mdi);
323 return 1;
324 }
325 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
326 * was stopped, so We'll do it here just to be sure. Drop any
327 * partitions as well...
328 */
329 if (fd >= 0)
330 ioctl(fd, BLKRRPART, 0);
331 if (mdi)
332 sysfs_uevent(mdi, "change");
333
334
335 if (devnum != NoMdDev &&
336 (stat("/dev/.udev", &stb) != 0 ||
337 check_env("MDADM_NO_UDEV"))) {
338 struct map_ent *mp = map_by_devnum(&map, devnum);
339 remove_devices(devnum, mp ? mp->path : NULL);
340 }
341
342
343 if (quiet <= 0)
344 fprintf(stderr, Name ": stopped %s\n", devname);
345 map_lock(&map);
346 map_remove(&map, devnum);
347 map_unlock(&map);
348 }
349 return 0;
350 }
351
352 int Manage_resize(char *devname, int fd, long long size, int raid_disks)
353 {
354 mdu_array_info_t info;
355 if (ioctl(fd, GET_ARRAY_INFO, &info) != 0) {
356 fprintf(stderr, Name ": Cannot get array information for %s: %s\n",
357 devname, strerror(errno));
358 return 1;
359 }
360 if (size >= 0)
361 info.size = size;
362 if (raid_disks > 0)
363 info.raid_disks = raid_disks;
364 if (ioctl(fd, SET_ARRAY_INFO, &info) != 0) {
365 fprintf(stderr, Name ": Cannot set device size/shape for %s: %s\n",
366 devname, strerror(errno));
367 return 1;
368 }
369 return 0;
370 }
371
372 int Manage_subdevs(char *devname, int fd,
373 struct mddev_dev *devlist, int verbose, int test,
374 char *update, int force)
375 {
376 /* do something to each dev.
377 * devmode can be
378 * 'a' - add the device
379 * try HOT_ADD_DISK
380 * If that fails EINVAL, try ADD_NEW_DISK
381 * 'r' - remove the device HOT_REMOVE_DISK
382 * device can be 'faulty' or 'detached' in which case all
383 * matching devices are removed.
384 * 'f' - set the device faulty SET_DISK_FAULTY
385 * device can be 'detached' in which case any device that
386 * is inaccessible will be marked faulty.
387 * For 'f' and 'r', the device can also be a kernel-internal
388 * name such as 'sdb'.
389 */
390 struct mddev_dev *add_devlist = NULL;
391 mdu_array_info_t array;
392 mdu_disk_info_t disc;
393 unsigned long long array_size;
394 struct mddev_dev *dv, *next = NULL;
395 struct stat stb;
396 int j, jnext = 0;
397 int tfd = -1;
398 struct supertype *st, *tst;
399 char *subarray = NULL;
400 int duuid[4];
401 int ouuid[4];
402 int lfd = -1;
403 int sysfd = -1;
404 int count = 0; /* number of actions taken */
405
406 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
407 fprintf(stderr, Name ": cannot get array info for %s\n",
408 devname);
409 return 1;
410 }
411
412 /* array.size is only 32 bit and may be truncated.
413 * So read from sysfs if possible, and record number of sectors
414 */
415
416 array_size = get_component_size(fd);
417 if (array_size <= 0)
418 array_size = array.size * 2;
419
420 tst = super_by_fd(fd, &subarray);
421 if (!tst) {
422 fprintf(stderr, Name ": unsupport array - version %d.%d\n",
423 array.major_version, array.minor_version);
424 return 1;
425 }
426
427 stb.st_rdev = 0;
428 for (dv = devlist, j=0 ; dv; dv = next, j = jnext) {
429 unsigned long long ldsize;
430 char dvname[20];
431 char *dnprintable = dv->devname;
432 char *add_dev = dv->devname;
433 int err;
434 int re_add_failed = 0;
435
436 next = dv->next;
437 jnext = 0;
438
439 if (strcmp(dv->devname, "failed")==0 ||
440 strcmp(dv->devname, "faulty")==0) {
441 int remaining_disks = array.nr_disks;
442 if (dv->disposition != 'r') {
443 fprintf(stderr, Name ": %s only meaningful "
444 "with -r, not -%c\n",
445 dv->devname, dv->disposition);
446 return 1;
447 }
448 for (; j < 1024 && remaining_disks > 0; j++) {
449 unsigned dev;
450 disc.number = j;
451 if (ioctl(fd, GET_DISK_INFO, &disc))
452 continue;
453 if (disc.major == 0 && disc.minor == 0)
454 continue;
455 remaining_disks --;
456 if ((disc.state & 1) == 0) /* faulty */
457 continue;
458 dev = makedev(disc.major, disc.minor);
459 if (stb.st_rdev == dev)
460 /* already did that one */
461 continue;
462 stb.st_rdev = dev;
463 next = dv;
464 /* same slot again next time - things might
465 * have reshuffled */
466 jnext = j;
467 sprintf(dvname,"%d:%d", disc.major, disc.minor);
468 dnprintable = dvname;
469 break;
470 }
471 if (next != dv)
472 continue;
473 } else if (strcmp(dv->devname, "detached") == 0) {
474 int remaining_disks = array.nr_disks;
475 if (dv->disposition != 'r' && dv->disposition != 'f') {
476 fprintf(stderr, Name ": %s only meaningful "
477 "with -r of -f, not -%c\n",
478 dv->devname, dv->disposition);
479 return 1;
480 }
481 for (; j < 1024 && remaining_disks > 0; j++) {
482 int sfd;
483 unsigned dev;
484 disc.number = j;
485 if (ioctl(fd, GET_DISK_INFO, &disc))
486 continue;
487 if (disc.major == 0 && disc.minor == 0)
488 continue;
489 remaining_disks --;
490 sprintf(dvname,"%d:%d", disc.major, disc.minor);
491 sfd = dev_open(dvname, O_RDONLY);
492 if (sfd >= 0) {
493 close(sfd);
494 continue;
495 }
496 if (dv->disposition == 'f' &&
497 (disc.state & 1) == 1) /* already faulty */
498 continue;
499 if (errno != ENXIO)
500 continue;
501 dev = makedev(disc.major, disc.minor);
502 if (stb.st_rdev == dev)
503 /* already did that one */
504 continue;
505 stb.st_rdev = dev;
506 next = dv;
507 /* same slot again next time - things might
508 * have reshuffled */
509 jnext = j;
510 dnprintable = dvname;
511 break;
512 }
513 if (next != dv)
514 continue;
515 } else if (strcmp(dv->devname, "missing") == 0) {
516 if (dv->disposition != 'a' || dv->re_add == 0) {
517 fprintf(stderr, Name ": 'missing' only meaningful "
518 "with --re-add\n");
519 return 1;
520 }
521 if (add_devlist == NULL)
522 add_devlist = conf_get_devs();
523 if (add_devlist == NULL) {
524 fprintf(stderr, Name ": no devices to scan for missing members.");
525 continue;
526 }
527 add_dev = add_devlist->devname;
528 add_devlist = add_devlist->next;
529 if (add_devlist != NULL)
530 next = dv;
531 if (stat(add_dev, &stb) < 0)
532 continue;
533 } else if (strchr(dv->devname, '/') == NULL &&
534 strchr(dv->devname, ':') == NULL &&
535 strlen(dv->devname) < 50) {
536 /* Assume this is a kernel-internal name like 'sda1' */
537 int found = 0;
538 char dname[55];
539 if (dv->disposition != 'r' && dv->disposition != 'f') {
540 fprintf(stderr, Name ": %s only meaningful "
541 "with -r or -f, not -%c\n",
542 dv->devname, dv->disposition);
543 return 1;
544 }
545
546 sprintf(dname, "dev-%s", dv->devname);
547 sysfd = sysfs_open(fd2devnum(fd), dname, "block/dev");
548 if (sysfd >= 0) {
549 char dn[20];
550 int mj,mn;
551 if (sysfs_fd_get_str(sysfd, dn, 20) > 0 &&
552 sscanf(dn, "%d:%d", &mj,&mn) == 2) {
553 stb.st_rdev = makedev(mj,mn);
554 found = 1;
555 }
556 close(sysfd);
557 sysfd = -1;
558 }
559 if (!found) {
560 sysfd = sysfs_open(fd2devnum(fd), dname, "state");
561 if (sysfd < 0) {
562 fprintf(stderr, Name ": %s does not appear "
563 "to be a component of %s\n",
564 dv->devname, devname);
565 return 1;
566 }
567 }
568 } else {
569 j = 0;
570
571 tfd = dev_open(dv->devname, O_RDONLY);
572 if (tfd < 0 && dv->disposition == 'r' &&
573 lstat(dv->devname, &stb) == 0)
574 /* Be happy, the lstat worked, that is
575 * enough for --remove
576 */
577 ;
578 else {
579 if (tfd < 0 || fstat(tfd, &stb) != 0) {
580 fprintf(stderr, Name ": cannot find %s: %s\n",
581 dv->devname, strerror(errno));
582 if (tfd >= 0)
583 close(tfd);
584 return 1;
585 }
586 close(tfd);
587 tfd = -1;
588 }
589 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
590 fprintf(stderr, Name ": %s is not a "
591 "block device.\n",
592 dv->devname);
593 return 1;
594 }
595 }
596 switch(dv->disposition){
597 default:
598 fprintf(stderr, Name ": internal error - devmode[%s]=%d\n",
599 dv->devname, dv->disposition);
600 return 1;
601 case 'a':
602 /* add the device */
603 if (subarray) {
604 fprintf(stderr, Name ": Cannot add disks to a"
605 " \'member\' array, perform this"
606 " operation on the parent container\n");
607 return 1;
608 }
609 /* Make sure it isn't in use (in 2.6 or later) */
610 tfd = dev_open(add_dev, O_RDONLY|O_EXCL|O_DIRECT);
611 if (tfd < 0 && add_dev != dv->devname)
612 continue;
613 if (tfd < 0) {
614 fprintf(stderr, Name ": Cannot open %s: %s\n",
615 dv->devname, strerror(errno));
616 return 1;
617 }
618
619 st = dup_super(tst);
620
621 if (array.not_persistent==0)
622 st->ss->load_super(st, tfd, NULL);
623
624 if (add_dev == dv->devname) {
625 if (!get_dev_size(tfd, dv->devname, &ldsize)) {
626 close(tfd);
627 return 1;
628 }
629 } else if (!get_dev_size(tfd, NULL, &ldsize)) {
630 close(tfd);
631 tfd = -1;
632 continue;
633 }
634
635 if (tst->ss->validate_geometry(
636 tst, array.level, array.layout,
637 array.raid_disks, NULL,
638 ldsize >> 9, NULL, NULL, 0) == 0) {
639 if (!force) {
640 fprintf(stderr, Name
641 ": %s is larger than %s can "
642 "effectively use.\n"
643 " Add --force is you "
644 "really wan to add this device.\n",
645 add_dev, devname);
646 close(tfd);
647 return 1;
648 }
649 fprintf(stderr, Name
650 ": %s is larger than %s can "
651 "effectively use.\n"
652 " Adding anyway as --force "
653 "was given.\n",
654 add_dev, devname);
655 }
656 if (!tst->ss->external &&
657 array.major_version == 0 &&
658 md_get_version(fd)%100 < 2) {
659 close(tfd);
660 tfd = -1;
661 if (ioctl(fd, HOT_ADD_DISK,
662 (unsigned long)stb.st_rdev)==0) {
663 if (verbose >= 0)
664 fprintf(stderr, Name ": hot added %s\n",
665 add_dev);
666 continue;
667 }
668
669 fprintf(stderr, Name ": hot add failed for %s: %s\n",
670 add_dev, strerror(errno));
671 return 1;
672 }
673
674 if (array.not_persistent == 0 || tst->ss->external) {
675
676 /* need to find a sample superblock to copy, and
677 * a spare slot to use.
678 * For 'external' array (well, container based),
679 * We can just load the metadata for the array.
680 */
681 if (tst->sb)
682 /* already loaded */;
683 else if (tst->ss->external) {
684 tst->ss->load_container(tst, fd, NULL);
685 } else for (j = 0; j < tst->max_devs; j++) {
686 char *dev;
687 int dfd;
688 disc.number = j;
689 if (ioctl(fd, GET_DISK_INFO, &disc))
690 continue;
691 if (disc.major==0 && disc.minor==0)
692 continue;
693 if ((disc.state & 4)==0) continue; /* sync */
694 /* Looks like a good device to try */
695 dev = map_dev(disc.major, disc.minor, 1);
696 if (!dev) continue;
697 dfd = dev_open(dev, O_RDONLY);
698 if (dfd < 0) continue;
699 if (tst->ss->load_super(tst, dfd,
700 NULL)) {
701 close(dfd);
702 continue;
703 }
704 close(dfd);
705 break;
706 }
707 /* FIXME this is a bad test to be using */
708 if (!tst->sb) {
709 close(tfd);
710 fprintf(stderr, Name ": cannot load array metadata from %s\n", devname);
711 return 1;
712 }
713
714 /* Make sure device is large enough */
715 if (tst->ss->avail_size(tst, ldsize/512) <
716 array_size) {
717 close(tfd);
718 tfd = -1;
719 if (add_dev != dv->devname)
720 continue;
721 fprintf(stderr, Name ": %s not large enough to join array\n",
722 dv->devname);
723 return 1;
724 }
725
726 /* Possibly this device was recently part of the array
727 * and was temporarily removed, and is now being re-added.
728 * If so, we can simply re-add it.
729 */
730 tst->ss->uuid_from_super(tst, duuid);
731
732 if (st->sb) {
733 struct mdinfo mdi;
734 st->ss->getinfo_super(st, &mdi, NULL);
735 st->ss->uuid_from_super(st, ouuid);
736 if ((mdi.disk.state & (1<<MD_DISK_ACTIVE)) &&
737 !(mdi.disk.state & (1<<MD_DISK_FAULTY)) &&
738 memcmp(duuid, ouuid, sizeof(ouuid))==0) {
739 /* look like it is worth a try. Need to
740 * make sure kernel will accept it though.
741 */
742 /* re-add doesn't work for version-1 superblocks
743 * before 2.6.18 :-(
744 */
745 if (array.major_version == 1 &&
746 get_linux_version() <= 2006018)
747 goto skip_re_add;
748 disc.number = mdi.disk.number;
749 if (ioctl(fd, GET_DISK_INFO, &disc) != 0
750 || disc.major != 0 || disc.minor != 0
751 || !enough_fd(fd))
752 goto skip_re_add;
753 disc.major = major(stb.st_rdev);
754 disc.minor = minor(stb.st_rdev);
755 disc.number = mdi.disk.number;
756 disc.raid_disk = mdi.disk.raid_disk;
757 disc.state = mdi.disk.state;
758 if (dv->writemostly == 1)
759 disc.state |= 1 << MD_DISK_WRITEMOSTLY;
760 if (dv->writemostly == 2)
761 disc.state &= ~(1 << MD_DISK_WRITEMOSTLY);
762 remove_partitions(tfd);
763 close(tfd);
764 tfd = -1;
765 if (update || dv->writemostly > 0) {
766 int rv = -1;
767 tfd = dev_open(dv->devname, O_RDWR);
768 if (tfd < 0) {
769 fprintf(stderr, Name ": failed to open %s for"
770 " superblock update during re-add\n", dv->devname);
771 return 1;
772 }
773
774 if (dv->writemostly == 1)
775 rv = st->ss->update_super(
776 st, NULL, "writemostly",
777 devname, verbose, 0, NULL);
778 if (dv->writemostly == 2)
779 rv = st->ss->update_super(
780 st, NULL, "readwrite",
781 devname, verbose, 0, NULL);
782 if (update)
783 rv = st->ss->update_super(
784 st, NULL, update,
785 devname, verbose, 0, NULL);
786 if (rv == 0)
787 rv = st->ss->store_super(st, tfd);
788 close(tfd);
789 tfd = -1;
790 if (rv != 0) {
791 fprintf(stderr, Name ": failed to update"
792 " superblock during re-add\n");
793 return 1;
794 }
795 }
796 /* don't even try if disk is marked as faulty */
797 errno = 0;
798 if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) {
799 if (verbose >= 0)
800 fprintf(stderr, Name ": re-added %s\n", add_dev);
801 count++;
802 continue;
803 }
804 if (errno == ENOMEM || errno == EROFS) {
805 fprintf(stderr, Name ": add new device failed for %s: %s\n",
806 add_dev, strerror(errno));
807 if (add_dev != dv->devname)
808 continue;
809 return 1;
810 }
811 skip_re_add:
812 re_add_failed = 1;
813 }
814 st->ss->free_super(st);
815 }
816 if (add_dev != dv->devname) {
817 if (verbose > 0)
818 fprintf(stderr, Name
819 ": --re-add for %s to %s is not possible\n",
820 add_dev, devname);
821 if (tfd >= 0) {
822 close(tfd);
823 tfd = -1;
824 }
825 continue;
826 }
827 if (dv->re_add) {
828 if (tfd >= 0)
829 close(tfd);
830 fprintf(stderr, Name
831 ": --re-add for %s to %s is not possible\n",
832 dv->devname, devname);
833 return 1;
834 }
835 if (re_add_failed) {
836 fprintf(stderr, Name ": %s reports being an active member for %s, but a --re-add fails.\n",
837 dv->devname, devname);
838 fprintf(stderr, Name ": not performing --add as that would convert %s in to a spare.\n",
839 dv->devname);
840 fprintf(stderr, Name ": To make this a spare, use \"mdadm --zero-superblock %s\" first.\n",
841 dv->devname);
842 if (tfd >= 0)
843 close(tfd);
844 return 1;
845 }
846 } else {
847 /* non-persistent. Must ensure that new drive
848 * is at least array.size big.
849 */
850 if (ldsize/512 < array_size) {
851 fprintf(stderr, Name ": %s not large enough to join array\n",
852 dv->devname);
853 if (tfd >= 0)
854 close(tfd);
855 return 1;
856 }
857 }
858 /* committed to really trying this device now*/
859 if (tfd >= 0) {
860 remove_partitions(tfd);
861 close(tfd);
862 tfd = -1;
863 }
864 /* in 2.6.17 and earlier, version-1 superblocks won't
865 * use the number we write, but will choose a free number.
866 * we must choose the same free number, which requires
867 * starting at 'raid_disks' and counting up
868 */
869 for (j = array.raid_disks; j< tst->max_devs; j++) {
870 disc.number = j;
871 if (ioctl(fd, GET_DISK_INFO, &disc))
872 break;
873 if (disc.major==0 && disc.minor==0)
874 break;
875 if (disc.state & 8) /* removed */
876 break;
877 }
878 disc.major = major(stb.st_rdev);
879 disc.minor = minor(stb.st_rdev);
880 disc.number =j;
881 disc.state = 0;
882 if (array.not_persistent==0) {
883 int dfd;
884 if (dv->writemostly == 1)
885 disc.state |= 1 << MD_DISK_WRITEMOSTLY;
886 dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT);
887 if (tst->ss->add_to_super(tst, &disc, dfd,
888 dv->devname)) {
889 close(dfd);
890 return 1;
891 }
892 if (tst->ss->write_init_super(tst)) {
893 close(dfd);
894 return 1;
895 }
896 } else if (dv->re_add) {
897 /* this had better be raid1.
898 * As we are "--re-add"ing we must find a spare slot
899 * to fill.
900 */
901 char *used = malloc(array.raid_disks);
902 memset(used, 0, array.raid_disks);
903 for (j=0; j< tst->max_devs; j++) {
904 mdu_disk_info_t disc2;
905 disc2.number = j;
906 if (ioctl(fd, GET_DISK_INFO, &disc2))
907 continue;
908 if (disc2.major==0 && disc2.minor==0)
909 continue;
910 if (disc2.state & 8) /* removed */
911 continue;
912 if (disc2.raid_disk < 0)
913 continue;
914 if (disc2.raid_disk > array.raid_disks)
915 continue;
916 used[disc2.raid_disk] = 1;
917 }
918 for (j=0 ; j<array.raid_disks; j++)
919 if (!used[j]) {
920 disc.raid_disk = j;
921 disc.state |= (1<<MD_DISK_SYNC);
922 break;
923 }
924 free(used);
925 }
926 if (dv->writemostly == 1)
927 disc.state |= (1 << MD_DISK_WRITEMOSTLY);
928 if (tst->ss->external) {
929 /* add a disk
930 * to an external metadata container */
931 struct mdinfo new_mdi;
932 struct mdinfo *sra;
933 int container_fd;
934 int devnum = fd2devnum(fd);
935 int dfd;
936
937 container_fd = open_dev_excl(devnum);
938 if (container_fd < 0) {
939 fprintf(stderr, Name ": add failed for %s:"
940 " could not get exclusive access to container\n",
941 dv->devname);
942 tst->ss->free_super(tst);
943 return 1;
944 }
945
946 dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT);
947 if (mdmon_running(tst->container_dev))
948 tst->update_tail = &tst->updates;
949 if (tst->ss->add_to_super(tst, &disc, dfd,
950 dv->devname)) {
951 close(dfd);
952 close(container_fd);
953 return 1;
954 }
955 if (tst->update_tail)
956 flush_metadata_updates(tst);
957 else
958 tst->ss->sync_metadata(tst);
959
960 sra = sysfs_read(container_fd, -1, 0);
961 if (!sra) {
962 fprintf(stderr, Name ": add failed for %s: sysfs_read failed\n",
963 dv->devname);
964 close(container_fd);
965 tst->ss->free_super(tst);
966 return 1;
967 }
968 sra->array.level = LEVEL_CONTAINER;
969 /* Need to set data_offset and component_size */
970 tst->ss->getinfo_super(tst, &new_mdi, NULL);
971 new_mdi.disk.major = disc.major;
972 new_mdi.disk.minor = disc.minor;
973 new_mdi.recovery_start = 0;
974 /* Make sure fds are closed as they are O_EXCL which
975 * would block add_disk */
976 tst->ss->free_super(tst);
977 if (sysfs_add_disk(sra, &new_mdi, 0) != 0) {
978 fprintf(stderr, Name ": add new device to external metadata"
979 " failed for %s\n", dv->devname);
980 close(container_fd);
981 sysfs_free(sra);
982 return 1;
983 }
984 ping_monitor_by_id(devnum);
985 sysfs_free(sra);
986 close(container_fd);
987 } else {
988 tst->ss->free_super(tst);
989 if (ioctl(fd, ADD_NEW_DISK, &disc)) {
990 fprintf(stderr, Name ": add new device failed for %s as %d: %s\n",
991 dv->devname, j, strerror(errno));
992 return 1;
993 }
994 }
995 if (verbose >= 0)
996 fprintf(stderr, Name ": added %s\n", dv->devname);
997 break;
998
999 case 'r':
1000 /* hot remove */
1001 if (subarray) {
1002 fprintf(stderr, Name ": Cannot remove disks from a"
1003 " \'member\' array, perform this"
1004 " operation on the parent container\n");
1005 if (sysfd >= 0)
1006 close(sysfd);
1007 return 1;
1008 }
1009 if (tst->ss->external) {
1010 /* To remove a device from a container, we must
1011 * check that it isn't in use in an array.
1012 * This involves looking in the 'holders'
1013 * directory - there must be just one entry,
1014 * the container.
1015 * To ensure that it doesn't get used as a
1016 * hold spare while we are checking, we
1017 * get an O_EXCL open on the container
1018 */
1019 int dnum = fd2devnum(fd);
1020 lfd = open_dev_excl(dnum);
1021 if (lfd < 0) {
1022 fprintf(stderr, Name
1023 ": Cannot get exclusive access "
1024 " to container - odd\n");
1025 if (sysfd >= 0)
1026 close(sysfd);
1027 return 1;
1028 }
1029 /* in the detached case it is not possible to
1030 * check if we are the unique holder, so just
1031 * rely on the 'detached' checks
1032 */
1033 if (strcmp(dv->devname, "detached") == 0 ||
1034 sysfd >= 0 ||
1035 sysfs_unique_holder(dnum, stb.st_rdev))
1036 /* pass */;
1037 else {
1038 fprintf(stderr, Name
1039 ": %s is %s, cannot remove.\n",
1040 dnprintable,
1041 errno == EEXIST ? "still in use":
1042 "not a member");
1043 close(lfd);
1044 return 1;
1045 }
1046 }
1047 /* FIXME check that it is a current member */
1048 if (sysfd >= 0) {
1049 /* device has been removed and we don't know
1050 * the major:minor number
1051 */
1052 int n = write(sysfd, "remove", 6);
1053 if (n != 6)
1054 err = -1;
1055 else
1056 err = 0;
1057 close(sysfd);
1058 sysfd = -1;
1059 } else {
1060 err = ioctl(fd, HOT_REMOVE_DISK, (unsigned long)stb.st_rdev);
1061 if (err && errno == ENODEV) {
1062 /* Old kernels rejected this if no personality
1063 * registered */
1064 struct mdinfo *sra = sysfs_read(fd, 0, GET_DEVS);
1065 struct mdinfo *dv = NULL;
1066 if (sra)
1067 dv = sra->devs;
1068 for ( ; dv ; dv=dv->next)
1069 if (dv->disk.major == (int)major(stb.st_rdev) &&
1070 dv->disk.minor == (int)minor(stb.st_rdev))
1071 break;
1072 if (dv)
1073 err = sysfs_set_str(sra, dv,
1074 "state", "remove");
1075 else
1076 err = -1;
1077 if (sra)
1078 sysfs_free(sra);
1079 }
1080 }
1081 if (err) {
1082 fprintf(stderr, Name ": hot remove failed "
1083 "for %s: %s\n", dnprintable,
1084 strerror(errno));
1085 if (lfd >= 0)
1086 close(lfd);
1087 return 1;
1088 }
1089 if (tst->ss->external) {
1090 /*
1091 * Before dropping our exclusive open we make an
1092 * attempt at preventing mdmon from seeing an
1093 * 'add' event before reconciling this 'remove'
1094 * event.
1095 */
1096 char *name = devnum2devname(fd2devnum(fd));
1097
1098 if (!name) {
1099 fprintf(stderr, Name ": unable to get container name\n");
1100 return 1;
1101 }
1102
1103 ping_manager(name);
1104 free(name);
1105 }
1106 if (lfd >= 0)
1107 close(lfd);
1108 count++;
1109 if (verbose >= 0)
1110 fprintf(stderr, Name ": hot removed %s from %s\n",
1111 dnprintable, devname);
1112 break;
1113
1114 case 'f': /* set faulty */
1115 /* FIXME check current member */
1116 if ((sysfd >= 0 && write(sysfd, "faulty", 6) != 6) ||
1117 (sysfd < 0 && ioctl(fd, SET_DISK_FAULTY,
1118 (unsigned long) stb.st_rdev))) {
1119 fprintf(stderr, Name ": set device faulty failed for %s: %s\n",
1120 dnprintable, strerror(errno));
1121 if (sysfd >= 0)
1122 close(sysfd);
1123 return 1;
1124 }
1125 if (sysfd >= 0)
1126 close(sysfd);
1127 sysfd = -1;
1128 count++;
1129 if (verbose >= 0)
1130 fprintf(stderr, Name ": set %s faulty in %s\n",
1131 dnprintable, devname);
1132 break;
1133 }
1134 }
1135 if (test && count == 0)
1136 return 2;
1137 return 0;
1138 }
1139
1140 int autodetect(void)
1141 {
1142 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1143 int rv = 1;
1144 int fd = dev_open("9:0", O_RDONLY);
1145 if (fd >= 0) {
1146 if (ioctl(fd, RAID_AUTORUN, 0) == 0)
1147 rv = 0;
1148 close(fd);
1149 }
1150 return rv;
1151 }
1152
1153 int Update_subarray(char *dev, char *subarray, char *update, struct mddev_ident *ident, int quiet)
1154 {
1155 struct supertype supertype, *st = &supertype;
1156 int fd, rv = 2;
1157
1158 memset(st, 0, sizeof(*st));
1159
1160 fd = open_subarray(dev, subarray, st, quiet);
1161 if (fd < 0)
1162 return 2;
1163
1164 if (!st->ss->update_subarray) {
1165 if (!quiet)
1166 fprintf(stderr,
1167 Name ": Operation not supported for %s metadata\n",
1168 st->ss->name);
1169 goto free_super;
1170 }
1171
1172 if (mdmon_running(st->devnum))
1173 st->update_tail = &st->updates;
1174
1175 rv = st->ss->update_subarray(st, subarray, update, ident);
1176
1177 if (rv) {
1178 if (!quiet)
1179 fprintf(stderr, Name ": Failed to update %s of subarray-%s in %s\n",
1180 update, subarray, dev);
1181 } else if (st->update_tail)
1182 flush_metadata_updates(st);
1183 else
1184 st->ss->sync_metadata(st);
1185
1186 if (rv == 0 && strcmp(update, "name") == 0 && !quiet)
1187 fprintf(stderr,
1188 Name ": Updated subarray-%s name from %s, UUIDs may have changed\n",
1189 subarray, dev);
1190
1191 free_super:
1192 st->ss->free_super(st);
1193 close(fd);
1194
1195 return rv;
1196 }
1197
1198 /* Move spare from one array to another
1199 * If adding to destination array fails
1200 * add back to original array
1201 * Returns 1 on success, 0 on failure */
1202 int move_spare(char *from_devname, char *to_devname, dev_t devid)
1203 {
1204 struct mddev_dev devlist;
1205 char devname[20];
1206
1207 /* try to remove and add */
1208 int fd1 = open(to_devname, O_RDONLY);
1209 int fd2 = open(from_devname, O_RDONLY);
1210
1211 if (fd1 < 0 || fd2 < 0) {
1212 if (fd1>=0) close(fd1);
1213 if (fd2>=0) close(fd2);
1214 return 0;
1215 }
1216
1217 devlist.next = NULL;
1218 devlist.used = 0;
1219 devlist.re_add = 0;
1220 devlist.writemostly = 0;
1221 devlist.devname = devname;
1222 sprintf(devname, "%d:%d", major(devid), minor(devid));
1223
1224 devlist.disposition = 'r';
1225 if (Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0) == 0) {
1226 devlist.disposition = 'a';
1227 if (Manage_subdevs(to_devname, fd1, &devlist, -1, 0, NULL, 0) == 0) {
1228 /* make sure manager is aware of changes */
1229 ping_manager(to_devname);
1230 ping_manager(from_devname);
1231 close(fd1);
1232 close(fd2);
1233 return 1;
1234 }
1235 else Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0);
1236 }
1237 close(fd1);
1238 close(fd2);
1239 return 0;
1240 }
1241 #endif
Unnamed repository; edit this file 'description' to name the repository.