]>
Commit | Line | Data |
---|---|---|
64c4757e | 1 | /* |
9a9dab36 | 2 | * mdadm - manage Linux "md" devices aka RAID arrays. |
64c4757e | 3 | * |
7bd04da9 | 4 | * Copyright (C) 2001-2012 Neil Brown <neilb@suse.de> |
64c4757e NB |
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 | |
e736b623 | 22 | * Email: <neilb@suse.de> |
64c4757e NB |
23 | */ |
24 | ||
9a9dab36 | 25 | #include "mdadm.h" |
682c7051 NB |
26 | #include "md_u.h" |
27 | #include "md_p.h" | |
4ccad7b1 | 28 | #include <ctype.h> |
64c4757e | 29 | |
82b27616 NB |
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 | ||
64c4757e NB |
34 | int Manage_ro(char *devname, int fd, int readonly) |
35 | { | |
682c7051 NB |
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; | |
0e600426 | 44 | #ifndef MDASSEMBLE |
e9dd1598 | 45 | struct mdinfo *mdi; |
0e600426 | 46 | #endif |
b73e45ae | 47 | int rv = 0; |
aba69144 | 48 | |
682c7051 | 49 | if (md_get_version(fd) < 9000) { |
e7b84f9d | 50 | pr_err("need md driver version 0.90.0 or later\n"); |
682c7051 NB |
51 | return 1; |
52 | } | |
0e600426 | 53 | #ifndef MDASSEMBLE |
7bd04da9 | 54 | /* If this is an externally-managed array, we need to modify the |
e9dd1598 N |
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 && | |
e9dd1598 N |
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) { | |
e7b84f9d | 74 | pr_err("failed to set readonly for %s: %s\n", |
e9dd1598 N |
75 | devname, strerror(errno)); |
76 | ||
77 | vers[9] = mdi->text_version[0]; | |
78 | sysfs_set_str(mdi, NULL, "metadata_version", vers); | |
b73e45ae JS |
79 | rv = 1; |
80 | goto out; | |
e9dd1598 N |
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, '/'); | |
1471b8b1 | 89 | if (cp) |
e9dd1598 N |
90 | *cp = 0; |
91 | ping_monitor(vers+10); | |
9ea5a252 DW |
92 | if (mdi->array.level <= 0) |
93 | sysfs_set_str(mdi, NULL, "array_state", "active"); | |
e9dd1598 | 94 | } |
b73e45ae | 95 | goto out; |
e9dd1598 | 96 | } |
0e600426 | 97 | #endif |
682c7051 | 98 | if (ioctl(fd, GET_ARRAY_INFO, &array)) { |
e7b84f9d | 99 | pr_err("%s does not appear to be active.\n", |
682c7051 | 100 | devname); |
b73e45ae JS |
101 | rv = 1; |
102 | goto out; | |
682c7051 | 103 | } |
aba69144 | 104 | |
7bd04da9 | 105 | if (readonly > 0) { |
682c7051 | 106 | if (ioctl(fd, STOP_ARRAY_RO, NULL)) { |
e7b84f9d | 107 | pr_err("failed to set readonly for %s: %s\n", |
682c7051 | 108 | devname, strerror(errno)); |
b73e45ae JS |
109 | rv = 1; |
110 | goto out; | |
682c7051 NB |
111 | } |
112 | } else if (readonly < 0) { | |
113 | if (ioctl(fd, RESTART_ARRAY_RW, NULL)) { | |
e7b84f9d | 114 | pr_err("failed to set writable for %s: %s\n", |
682c7051 | 115 | devname, strerror(errno)); |
b73e45ae JS |
116 | rv = 1; |
117 | goto out; | |
682c7051 NB |
118 | } |
119 | } | |
b73e45ae JS |
120 | out: |
121 | #ifndef MDASSEMBLE | |
122 | if (mdi) | |
123 | sysfs_free(mdi); | |
124 | #endif | |
125 | return rv; | |
64c4757e NB |
126 | } |
127 | ||
435d4ebb LB |
128 | #ifndef MDASSEMBLE |
129 | ||
4ccad7b1 N |
130 | static void remove_devices(int devnum, char *path) |
131 | { | |
7bd04da9 | 132 | /* |
b1702f48 N |
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. | |
4ccad7b1 N |
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 | ||
b1702f48 N |
146 | if (!path) |
147 | return; | |
148 | ||
4ccad7b1 N |
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); | |
b1702f48 | 154 | |
503975b9 | 155 | path2 = xmalloc(strlen(path)+20); |
b1702f48 N |
156 | strcpy(path2, path); |
157 | pe = path2 + strlen(path2); | |
7bd04da9 | 158 | |
4ccad7b1 N |
159 | for (part = 0; part < 16; part++) { |
160 | if (part) { | |
161 | sprintf(be, "p%d", part); | |
b1702f48 N |
162 | |
163 | if (isdigit(pe[-1])) | |
164 | sprintf(pe, "p%d", part); | |
165 | else | |
166 | sprintf(pe, "%d", part); | |
4ccad7b1 | 167 | } |
b1702f48 | 168 | n = readlink(path2, link, sizeof(link)); |
d9ca03e9 | 169 | if (n > 0 && (int)strlen(base) == n && |
b1702f48 N |
170 | strncmp(link, base, n) == 0) |
171 | unlink(path2); | |
4ccad7b1 | 172 | } |
0eb26465 | 173 | free(path2); |
4ccad7b1 | 174 | } |
4ccad7b1 | 175 | |
ba728be7 N |
176 | int Manage_runstop(char *devname, int fd, int runstop, |
177 | int verbose, int will_retry) | |
64c4757e | 178 | { |
7bd04da9 | 179 | /* Run or stop the array. Array must already be configured |
ba728be7 N |
180 | * 'Run' requires >= 0.90.0 |
181 | * 'will_retry' is only relevant for 'stop', and means | |
182 | * that error messages are not wanted. | |
682c7051 | 183 | */ |
682c7051 | 184 | mdu_param_t param; /* unused */ |
bccd8153 | 185 | int rv = 0; |
82b27616 | 186 | |
ba728be7 N |
187 | if (will_retry && verbose == 0) |
188 | verbose = -1; | |
189 | ||
82b27616 | 190 | if (runstop == -1 && md_get_version(fd) < 9000) { |
7bd04da9 N |
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; | |
82b27616 | 197 | } |
aba69144 | 198 | |
682c7051 | 199 | if (md_get_version(fd) < 9000) { |
e7b84f9d | 200 | pr_err("need md driver version 0.90.0 or later\n"); |
682c7051 NB |
201 | return 1; |
202 | } | |
ba728be7 | 203 | |
7bd04da9 | 204 | if (runstop > 0) { |
682c7051 | 205 | if (ioctl(fd, RUN_ARRAY, ¶m)) { |
ba728be7 N |
206 | if (verbose >= 0) |
207 | pr_err("failed to run array %s: %s\n", | |
208 | devname, strerror(errno)); | |
682c7051 NB |
209 | return 1; |
210 | } | |
ba728be7 | 211 | if (verbose >= 0) |
e7b84f9d | 212 | pr_err("started %s\n", devname); |
682c7051 | 213 | } else if (runstop < 0){ |
8382f19b NB |
214 | struct map_ent *map = NULL; |
215 | struct stat stb; | |
daf7a3ce | 216 | struct mdinfo *mdi; |
4ccad7b1 | 217 | int devnum; |
eb0af526 N |
218 | int err; |
219 | int count; | |
daf7a3ce NB |
220 | /* If this is an mdmon managed array, just write 'inactive' |
221 | * to the array state and let mdmon clear up. | |
222 | */ | |
4ccad7b1 | 223 | devnum = fd2devnum(fd); |
eb0af526 N |
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); | |
ba728be7 N |
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); | |
eb0af526 N |
238 | return 1; |
239 | } | |
daf7a3ce NB |
240 | mdi = sysfs_read(fd, -1, GET_LEVEL|GET_VERSION); |
241 | if (mdi && | |
242 | mdi->array.level > 0 && | |
3c558363 | 243 | is_subarray(mdi->text_version)) { |
1ae42d9d | 244 | int err; |
daf7a3ce NB |
245 | /* This is mdmon managed. */ |
246 | close(fd); | |
1ae42d9d | 247 | |
7bd04da9 N |
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 | */ | |
1ae42d9d KW |
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 | } | |
ba728be7 N |
261 | if (err) { |
262 | if (verbose >= 0) | |
263 | pr_err("failed to stop array %s: %s\n", | |
264 | devname, strerror(errno)); | |
bccd8153 JS |
265 | rv = 1; |
266 | goto out; | |
daf7a3ce NB |
267 | } |
268 | ||
269 | /* Give monitor a chance to act */ | |
c94709e8 | 270 | ping_monitor(mdi->text_version); |
daf7a3ce | 271 | |
eb0af526 N |
272 | fd = open_dev_excl(devnum); |
273 | if (fd < 0) { | |
ba728be7 N |
274 | if (verbose >= 0) |
275 | pr_err("failed to completely stop %s" | |
276 | ": Device is busy\n", | |
277 | devname); | |
bccd8153 JS |
278 | rv = 1; |
279 | goto out; | |
eb0af526 | 280 | } |
ada6c239 N |
281 | } else if (mdi && |
282 | mdi->array.major_version == -1 && | |
283 | mdi->array.minor_version == -2 && | |
3c558363 | 284 | !is_subarray(mdi->text_version)) { |
430ea469 | 285 | struct mdstat_ent *mds, *m; |
ada6c239 N |
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); | |
430ea469 N |
291 | |
292 | /* now check that there are no existing arrays | |
293 | * which are members of this array | |
294 | */ | |
295 | mds = mdstat_read(0, 0); | |
7bd04da9 | 296 | for (m = mds; m; m = m->next) |
430ea469 N |
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) { | |
ba728be7 | 301 | if (verbose >= 0) |
e7b84f9d N |
302 | pr_err("Cannot stop container %s: " |
303 | "member %s still active\n", | |
304 | devname, m->dev); | |
430ea469 | 305 | free_mdstat(mds); |
bccd8153 JS |
306 | rv = 1; |
307 | goto out; | |
430ea469 | 308 | } |
daf7a3ce | 309 | } |
daf7a3ce | 310 | |
eb0af526 N |
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 | */ | |
fb0d4b9c | 315 | count = 25; err = 0; |
eb0af526 N |
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) { | |
ba728be7 | 323 | if (verbose >= 0) { |
e7b84f9d N |
324 | pr_err("failed to stop array %s: %s\n", |
325 | devname, strerror(errno)); | |
d927f3c4 NB |
326 | if (errno == EBUSY) |
327 | fprintf(stderr, "Perhaps a running " | |
328 | "process, mounted filesystem " | |
329 | "or active volume group?\n"); | |
330 | } | |
bccd8153 JS |
331 | rv = 1; |
332 | goto out; | |
682c7051 | 333 | } |
97590376 | 334 | /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array |
d49410d3 N |
335 | * was stopped, so We'll do it here just to be sure. Drop any |
336 | * partitions as well... | |
97590376 | 337 | */ |
d49410d3 N |
338 | if (fd >= 0) |
339 | ioctl(fd, BLKRRPART, 0); | |
97590376 N |
340 | if (mdi) |
341 | sysfs_uevent(mdi, "change"); | |
daf7a3ce | 342 | |
4ccad7b1 N |
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 | ||
ba728be7 | 350 | if (verbose >= 0) |
e7b84f9d | 351 | pr_err("stopped %s\n", devname); |
4eb26970 DW |
352 | map_lock(&map); |
353 | map_remove(&map, devnum); | |
354 | map_unlock(&map); | |
bccd8153 JS |
355 | out: |
356 | if (mdi) | |
357 | sysfs_free(mdi); | |
682c7051 | 358 | } |
bccd8153 | 359 | return rv; |
64c4757e NB |
360 | } |
361 | ||
1d997643 N |
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 | ||
abe94694 N |
436 | int attempt_re_add(int fd, int tfd, struct mddev_dev *dv, |
437 | struct supertype *dev_st, struct supertype *tst, | |
438 | unsigned long rdev, | |
439 | char *update, char *devname, int verbose, | |
440 | mdu_array_info_t *array) | |
441 | { | |
442 | struct mdinfo mdi; | |
443 | int duuid[4]; | |
444 | int ouuid[4]; | |
445 | ||
446 | dev_st->ss->getinfo_super(dev_st, &mdi, NULL); | |
447 | dev_st->ss->uuid_from_super(dev_st, ouuid); | |
448 | if (tst->sb) | |
449 | tst->ss->uuid_from_super(tst, duuid); | |
450 | else | |
451 | /* Assume uuid matches: kernel will check */ | |
452 | memcpy(duuid, ouuid, sizeof(ouuid)); | |
453 | if ((mdi.disk.state & (1<<MD_DISK_ACTIVE)) && | |
454 | !(mdi.disk.state & (1<<MD_DISK_FAULTY)) && | |
455 | memcmp(duuid, ouuid, sizeof(ouuid))==0) { | |
456 | /* Looks like it is worth a | |
457 | * try. Need to make sure | |
458 | * kernel will accept it | |
459 | * though. | |
460 | */ | |
461 | mdu_disk_info_t disc; | |
462 | /* re-add doesn't work for version-1 superblocks | |
463 | * before 2.6.18 :-( | |
464 | */ | |
465 | if (array->major_version == 1 && | |
466 | get_linux_version() <= 2006018) | |
467 | goto skip_re_add; | |
468 | disc.number = mdi.disk.number; | |
469 | if (ioctl(fd, GET_DISK_INFO, &disc) != 0 | |
470 | || disc.major != 0 || disc.minor != 0 | |
471 | ) | |
472 | goto skip_re_add; | |
473 | disc.major = major(rdev); | |
474 | disc.minor = minor(rdev); | |
475 | disc.number = mdi.disk.number; | |
476 | disc.raid_disk = mdi.disk.raid_disk; | |
477 | disc.state = mdi.disk.state; | |
478 | if (dv->writemostly == 1) | |
479 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; | |
480 | if (dv->writemostly == 2) | |
481 | disc.state &= ~(1 << MD_DISK_WRITEMOSTLY); | |
482 | remove_partitions(tfd); | |
483 | if (update || dv->writemostly > 0) { | |
484 | int rv = -1; | |
485 | tfd = dev_open(dv->devname, O_RDWR); | |
486 | if (tfd < 0) { | |
487 | pr_err("failed to open %s for" | |
488 | " superblock update during re-add\n", dv->devname); | |
489 | return -1; | |
490 | } | |
491 | ||
492 | if (dv->writemostly == 1) | |
493 | rv = dev_st->ss->update_super( | |
494 | dev_st, NULL, "writemostly", | |
495 | devname, verbose, 0, NULL); | |
496 | if (dv->writemostly == 2) | |
497 | rv = dev_st->ss->update_super( | |
498 | dev_st, NULL, "readwrite", | |
499 | devname, verbose, 0, NULL); | |
500 | if (update) | |
501 | rv = dev_st->ss->update_super( | |
502 | dev_st, NULL, update, | |
503 | devname, verbose, 0, NULL); | |
504 | if (rv == 0) | |
505 | rv = dev_st->ss->store_super(dev_st, tfd); | |
506 | close(tfd); | |
507 | if (rv != 0) { | |
508 | pr_err("failed to update" | |
509 | " superblock during re-add\n"); | |
510 | return -1; | |
511 | } | |
512 | } | |
513 | /* don't even try if disk is marked as faulty */ | |
514 | errno = 0; | |
515 | if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) { | |
516 | if (verbose >= 0) | |
517 | pr_err("re-added %s\n", dv->devname); | |
518 | return 1; | |
519 | } | |
520 | if (errno == ENOMEM || errno == EROFS) { | |
521 | pr_err("add new device failed for %s: %s\n", | |
522 | dv->devname, strerror(errno)); | |
523 | if (dv->disposition == 'M') | |
524 | return 0; | |
525 | return -1; | |
526 | } | |
527 | } | |
528 | skip_re_add: | |
529 | return 0; | |
530 | } | |
531 | ||
38aeaf3a N |
532 | int Manage_add(int fd, int tfd, struct mddev_dev *dv, |
533 | struct supertype *tst, mdu_array_info_t *array, | |
534 | int force, int verbose, char *devname, | |
535 | char *update, unsigned long rdev, unsigned long long array_size) | |
536 | { | |
537 | unsigned long long ldsize; | |
6d43efb5 | 538 | struct supertype *dev_st = NULL; |
38aeaf3a N |
539 | int j; |
540 | mdu_disk_info_t disc; | |
541 | ||
542 | if (!get_dev_size(tfd, dv->devname, &ldsize)) { | |
543 | if (dv->disposition == 'M') | |
544 | return 0; | |
545 | else | |
546 | return -1; | |
547 | } | |
548 | ||
549 | if (tst->ss->validate_geometry( | |
550 | tst, array->level, array->layout, | |
551 | array->raid_disks, NULL, | |
af4348dd | 552 | ldsize >> 9, INVALID_SECTORS, NULL, NULL, 0) == 0) { |
38aeaf3a N |
553 | if (!force) { |
554 | pr_err("%s is larger than %s can " | |
555 | "effectively use.\n" | |
556 | " Add --force is you " | |
557 | "really want to add this device.\n", | |
558 | dv->devname, devname); | |
559 | return -1; | |
560 | } | |
561 | pr_err("%s is larger than %s can " | |
562 | "effectively use.\n" | |
563 | " Adding anyway as --force " | |
564 | "was given.\n", | |
565 | dv->devname, devname); | |
566 | } | |
567 | if (!tst->ss->external && | |
568 | array->major_version == 0 && | |
569 | md_get_version(fd)%100 < 2) { | |
570 | if (ioctl(fd, HOT_ADD_DISK, rdev)==0) { | |
571 | if (verbose >= 0) | |
572 | pr_err("hot added %s\n", | |
573 | dv->devname); | |
574 | return 1; | |
575 | } | |
576 | ||
577 | pr_err("hot add failed for %s: %s\n", | |
578 | dv->devname, strerror(errno)); | |
579 | return -1; | |
580 | } | |
581 | ||
582 | if (array->not_persistent == 0 || tst->ss->external) { | |
583 | ||
584 | /* need to find a sample superblock to copy, and | |
585 | * a spare slot to use. | |
586 | * For 'external' array (well, container based), | |
587 | * We can just load the metadata for the array-> | |
588 | */ | |
589 | int array_failed; | |
590 | if (tst->sb) | |
591 | /* already loaded */; | |
592 | else if (tst->ss->external) { | |
593 | tst->ss->load_container(tst, fd, NULL); | |
594 | } else for (j = 0; j < tst->max_devs; j++) { | |
595 | char *dev; | |
596 | int dfd; | |
597 | disc.number = j; | |
598 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
599 | continue; | |
600 | if (disc.major==0 && disc.minor==0) | |
601 | continue; | |
602 | if ((disc.state & 4)==0) /* sync */ | |
603 | continue; | |
604 | /* Looks like a good device to try */ | |
605 | dev = map_dev(disc.major, disc.minor, 1); | |
606 | if (!dev) | |
607 | continue; | |
608 | dfd = dev_open(dev, O_RDONLY); | |
609 | if (dfd < 0) | |
610 | continue; | |
611 | if (tst->ss->load_super(tst, dfd, | |
612 | NULL)) { | |
613 | close(dfd); | |
614 | continue; | |
615 | } | |
616 | close(dfd); | |
617 | break; | |
618 | } | |
619 | /* FIXME this is a bad test to be using */ | |
620 | if (!tst->sb && dv->disposition != 'a') { | |
621 | /* we are re-adding a device to a | |
622 | * completely dead array - have to depend | |
623 | * on kernel to check | |
624 | */ | |
625 | } else if (!tst->sb) { | |
626 | pr_err("cannot load array metadata from %s\n", devname); | |
627 | return -1; | |
628 | } | |
629 | ||
630 | /* Make sure device is large enough */ | |
387fcd59 | 631 | if (tst->ss->avail_size(tst, ldsize/512, INVALID_SECTORS) < |
38aeaf3a N |
632 | array_size) { |
633 | if (dv->disposition == 'M') | |
634 | return 0; | |
635 | pr_err("%s not large enough to join array\n", | |
636 | dv->devname); | |
637 | return -1; | |
638 | } | |
639 | ||
640 | /* Possibly this device was recently part of | |
641 | * the array and was temporarily removed, and | |
642 | * is now being re-added. If so, we can | |
643 | * simply re-add it. | |
644 | */ | |
645 | ||
646 | if (array->not_persistent==0) { | |
647 | dev_st = dup_super(tst); | |
648 | dev_st->ss->load_super(dev_st, tfd, NULL); | |
649 | } | |
650 | if (dev_st && dev_st->sb) { | |
651 | int rv = attempt_re_add(fd, tfd, dv, | |
652 | dev_st, tst, | |
653 | rdev, | |
654 | update, devname, | |
655 | verbose, | |
656 | array); | |
657 | dev_st->ss->free_super(dev_st); | |
658 | if (rv) | |
659 | return rv; | |
660 | } | |
661 | if (dv->disposition == 'M') { | |
662 | if (verbose > 0) | |
663 | pr_err("--re-add for %s to %s is not possible\n", | |
664 | dv->devname, devname); | |
665 | return 0; | |
666 | } | |
667 | if (dv->disposition == 'A') { | |
668 | pr_err("--re-add for %s to %s is not possible\n", | |
669 | dv->devname, devname); | |
670 | return -1; | |
671 | } | |
672 | if (array->active_disks < array->raid_disks) { | |
673 | char *avail = xcalloc(array->raid_disks, 1); | |
674 | int d; | |
675 | int found = 0; | |
676 | ||
677 | for (d = 0; d < MAX_DISKS && found < array->active_disks; d++) { | |
678 | disc.number = d; | |
679 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
680 | continue; | |
681 | if (disc.major == 0 && disc.minor == 0) | |
682 | continue; | |
683 | if (!(disc.state & (1<<MD_DISK_SYNC))) | |
684 | continue; | |
685 | avail[disc.raid_disk] = 1; | |
686 | found++; | |
687 | } | |
688 | array_failed = !enough(array->level, array->raid_disks, | |
689 | array->layout, 1, avail); | |
690 | } else | |
691 | array_failed = 0; | |
692 | if (array_failed) { | |
693 | pr_err("%s has failed so using --add cannot work and might destroy\n", | |
694 | devname); | |
695 | pr_err("data on %s. You should stop the array and re-assemble it.\n", | |
696 | dv->devname); | |
697 | return -1; | |
698 | } | |
699 | } else { | |
700 | /* non-persistent. Must ensure that new drive | |
701 | * is at least array->size big. | |
702 | */ | |
703 | if (ldsize/512 < array_size) { | |
704 | pr_err("%s not large enough to join array\n", | |
705 | dv->devname); | |
706 | return -1; | |
707 | } | |
708 | } | |
709 | /* committed to really trying this device now*/ | |
710 | remove_partitions(tfd); | |
711 | ||
712 | /* in 2.6.17 and earlier, version-1 superblocks won't | |
713 | * use the number we write, but will choose a free number. | |
714 | * we must choose the same free number, which requires | |
715 | * starting at 'raid_disks' and counting up | |
716 | */ | |
717 | for (j = array->raid_disks; j < tst->max_devs; j++) { | |
718 | disc.number = j; | |
719 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
720 | break; | |
721 | if (disc.major==0 && disc.minor==0) | |
722 | break; | |
723 | if (disc.state & 8) /* removed */ | |
724 | break; | |
725 | } | |
726 | disc.major = major(rdev); | |
727 | disc.minor = minor(rdev); | |
728 | disc.number =j; | |
729 | disc.state = 0; | |
730 | if (array->not_persistent==0) { | |
731 | int dfd; | |
732 | if (dv->writemostly == 1) | |
733 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; | |
734 | dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); | |
735 | if (tst->ss->add_to_super(tst, &disc, dfd, | |
72ca9bcf | 736 | dv->devname, INVALID_SECTORS)) |
38aeaf3a N |
737 | return -1; |
738 | if (tst->ss->write_init_super(tst)) | |
739 | return -1; | |
740 | } else if (dv->disposition == 'A') { | |
741 | /* this had better be raid1. | |
742 | * As we are "--re-add"ing we must find a spare slot | |
743 | * to fill. | |
744 | */ | |
745 | char *used = xcalloc(array->raid_disks, 1); | |
746 | for (j = 0; j < tst->max_devs; j++) { | |
747 | mdu_disk_info_t disc2; | |
748 | disc2.number = j; | |
749 | if (ioctl(fd, GET_DISK_INFO, &disc2)) | |
750 | continue; | |
751 | if (disc2.major==0 && disc2.minor==0) | |
752 | continue; | |
753 | if (disc2.state & 8) /* removed */ | |
754 | continue; | |
755 | if (disc2.raid_disk < 0) | |
756 | continue; | |
757 | if (disc2.raid_disk > array->raid_disks) | |
758 | continue; | |
759 | used[disc2.raid_disk] = 1; | |
760 | } | |
761 | for (j = 0 ; j < array->raid_disks; j++) | |
762 | if (!used[j]) { | |
763 | disc.raid_disk = j; | |
764 | disc.state |= (1<<MD_DISK_SYNC); | |
765 | break; | |
766 | } | |
767 | free(used); | |
768 | } | |
769 | if (dv->writemostly == 1) | |
770 | disc.state |= (1 << MD_DISK_WRITEMOSTLY); | |
771 | if (tst->ss->external) { | |
772 | /* add a disk | |
773 | * to an external metadata container */ | |
774 | struct mdinfo new_mdi; | |
775 | struct mdinfo *sra; | |
776 | int container_fd; | |
777 | int devnum = fd2devnum(fd); | |
778 | int dfd; | |
779 | ||
780 | container_fd = open_dev_excl(devnum); | |
781 | if (container_fd < 0) { | |
782 | pr_err("add failed for %s:" | |
783 | " could not get exclusive access to container\n", | |
784 | dv->devname); | |
785 | tst->ss->free_super(tst); | |
786 | return -1; | |
787 | } | |
788 | ||
9cf9a1de | 789 | Kill(dv->devname, NULL, 0, -1, 0); |
38aeaf3a N |
790 | dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); |
791 | if (mdmon_running(tst->container_dev)) | |
792 | tst->update_tail = &tst->updates; | |
793 | if (tst->ss->add_to_super(tst, &disc, dfd, | |
72ca9bcf | 794 | dv->devname, INVALID_SECTORS)) { |
38aeaf3a N |
795 | close(dfd); |
796 | close(container_fd); | |
797 | return -1; | |
798 | } | |
799 | if (tst->update_tail) | |
800 | flush_metadata_updates(tst); | |
801 | else | |
802 | tst->ss->sync_metadata(tst); | |
803 | ||
804 | sra = sysfs_read(container_fd, -1, 0); | |
805 | if (!sra) { | |
806 | pr_err("add failed for %s: sysfs_read failed\n", | |
807 | dv->devname); | |
808 | close(container_fd); | |
809 | tst->ss->free_super(tst); | |
810 | return -1; | |
811 | } | |
812 | sra->array.level = LEVEL_CONTAINER; | |
813 | /* Need to set data_offset and component_size */ | |
814 | tst->ss->getinfo_super(tst, &new_mdi, NULL); | |
815 | new_mdi.disk.major = disc.major; | |
816 | new_mdi.disk.minor = disc.minor; | |
817 | new_mdi.recovery_start = 0; | |
818 | /* Make sure fds are closed as they are O_EXCL which | |
819 | * would block add_disk */ | |
820 | tst->ss->free_super(tst); | |
821 | if (sysfs_add_disk(sra, &new_mdi, 0) != 0) { | |
822 | pr_err("add new device to external metadata" | |
823 | " failed for %s\n", dv->devname); | |
824 | close(container_fd); | |
825 | sysfs_free(sra); | |
826 | return -1; | |
827 | } | |
828 | ping_monitor_by_id(devnum); | |
829 | sysfs_free(sra); | |
830 | close(container_fd); | |
831 | } else { | |
832 | tst->ss->free_super(tst); | |
833 | if (ioctl(fd, ADD_NEW_DISK, &disc)) { | |
834 | pr_err("add new device failed for %s as %d: %s\n", | |
835 | dv->devname, j, strerror(errno)); | |
836 | return -1; | |
837 | } | |
838 | } | |
839 | if (verbose >= 0) | |
840 | pr_err("added %s\n", dv->devname); | |
841 | return 1; | |
842 | } | |
843 | ||
d070235d N |
844 | int Manage_remove(struct supertype *tst, int fd, struct mddev_dev *dv, |
845 | int sysfd, unsigned long rdev, int verbose, char *devname) | |
846 | { | |
847 | int lfd = -1; | |
848 | int err; | |
849 | ||
850 | if (tst->ss->external) { | |
851 | /* To remove a device from a container, we must | |
852 | * check that it isn't in use in an array. | |
853 | * This involves looking in the 'holders' | |
854 | * directory - there must be just one entry, | |
855 | * the container. | |
856 | * To ensure that it doesn't get used as a | |
857 | * hot spare while we are checking, we | |
858 | * get an O_EXCL open on the container | |
859 | */ | |
aab15415 | 860 | int ret; |
d070235d N |
861 | int dnum = fd2devnum(fd); |
862 | lfd = open_dev_excl(dnum); | |
863 | if (lfd < 0) { | |
864 | pr_err("Cannot get exclusive access " | |
865 | " to container - odd\n"); | |
866 | return -1; | |
867 | } | |
aab15415 N |
868 | /* We may not be able to check on holders in |
869 | * sysfs, either because we don't have the dev num | |
870 | * (rdev == 0) or because the device has been detached | |
871 | * and the 'holders' directory no longer exists | |
872 | * (ret == -1). In that case, assume it is OK to | |
873 | * remove. | |
d070235d | 874 | */ |
aab15415 N |
875 | if (rdev == 0) |
876 | ret = -1; | |
877 | else | |
878 | ret = sysfs_unique_holder(dnum, rdev); | |
879 | if (ret == 0) { | |
880 | pr_err("%s is not a member, cannot remove.\n", | |
881 | dv->devname); | |
882 | close(lfd); | |
883 | return -1; | |
884 | } | |
885 | if (ret >= 2) { | |
886 | pr_err("%s is still in use, cannot remove.\n", | |
887 | dv->devname); | |
d070235d N |
888 | close(lfd); |
889 | return -1; | |
890 | } | |
891 | } | |
892 | /* FIXME check that it is a current member */ | |
893 | if (sysfd >= 0) { | |
894 | /* device has been removed and we don't know | |
895 | * the major:minor number | |
896 | */ | |
897 | int n = write(sysfd, "remove", 6); | |
898 | if (n != 6) | |
899 | err = -1; | |
900 | else | |
901 | err = 0; | |
902 | } else { | |
903 | err = ioctl(fd, HOT_REMOVE_DISK, rdev); | |
904 | if (err && errno == ENODEV) { | |
905 | /* Old kernels rejected this if no personality | |
906 | * is registered */ | |
907 | struct mdinfo *sra = sysfs_read(fd, 0, GET_DEVS); | |
908 | struct mdinfo *dv = NULL; | |
909 | if (sra) | |
910 | dv = sra->devs; | |
911 | for ( ; dv ; dv=dv->next) | |
912 | if (dv->disk.major == (int)major(rdev) && | |
913 | dv->disk.minor == (int)minor(rdev)) | |
914 | break; | |
915 | if (dv) | |
916 | err = sysfs_set_str(sra, dv, | |
917 | "state", "remove"); | |
918 | else | |
919 | err = -1; | |
920 | if (sra) | |
921 | sysfs_free(sra); | |
922 | } | |
923 | } | |
924 | if (err) { | |
925 | pr_err("hot remove failed " | |
926 | "for %s: %s\n", dv->devname, | |
927 | strerror(errno)); | |
928 | if (lfd >= 0) | |
929 | close(lfd); | |
930 | return -1; | |
931 | } | |
932 | if (tst->ss->external) { | |
933 | /* | |
934 | * Before dropping our exclusive open we make an | |
935 | * attempt at preventing mdmon from seeing an | |
936 | * 'add' event before reconciling this 'remove' | |
937 | * event. | |
938 | */ | |
939 | char *name = devnum2devname(fd2devnum(fd)); | |
940 | ||
941 | if (!name) { | |
942 | pr_err("unable to get container name\n"); | |
943 | return -1; | |
944 | } | |
945 | ||
946 | ping_manager(name); | |
947 | free(name); | |
948 | } | |
949 | if (lfd >= 0) | |
950 | close(lfd); | |
951 | if (verbose >= 0) | |
952 | pr_err("hot removed %s from %s\n", | |
953 | dv->devname, devname); | |
954 | return 1; | |
955 | } | |
956 | ||
70c55e36 N |
957 | int Manage_replace(struct supertype *tst, int fd, struct mddev_dev *dv, |
958 | unsigned long rdev, int verbose, char *devname) | |
959 | { | |
960 | struct mdinfo *mdi, *di; | |
961 | if (tst->ss->external) { | |
962 | pr_err("--replace only supported for native metadata (0.90 or 1.x)\n"); | |
963 | return -1; | |
964 | } | |
965 | /* Need to find the device in sysfs and add 'want_replacement' to the | |
966 | * status. | |
967 | */ | |
968 | mdi = sysfs_read(fd, -1, GET_DEVS); | |
969 | if (!mdi || !mdi->devs) { | |
970 | pr_err("Cannot find status of %s to enable replacement - strange\n", | |
971 | devname); | |
972 | return -1; | |
973 | } | |
974 | for (di = mdi->devs; di; di = di->next) | |
975 | if (di->disk.major == (int)major(rdev) && | |
976 | di->disk.minor == (int)minor(rdev)) | |
977 | break; | |
978 | if (di) { | |
979 | int rv; | |
980 | if (di->disk.raid_disk < 0) { | |
981 | pr_err("%s is not active and so cannot be replaced.\n", | |
982 | dv->devname); | |
983 | sysfs_free(mdi); | |
984 | return -1; | |
985 | } | |
986 | rv = sysfs_set_str(mdi, di, | |
987 | "state", "want_replacement"); | |
988 | if (rv) { | |
989 | sysfs_free(mdi); | |
990 | pr_err("Failed to request replacement for %s\n", | |
991 | dv->devname); | |
992 | return -1; | |
993 | } | |
994 | if (verbose >= 0) | |
995 | pr_err("Marked %s (device %d in %s) for replacement\n", | |
996 | dv->devname, di->disk.raid_disk, devname); | |
997 | /* If there is a matching 'with', we need to tell it which | |
998 | * raid disk | |
999 | */ | |
1000 | while (dv && dv->disposition != 'W') | |
1001 | dv = dv->next; | |
1002 | if (dv) { | |
1003 | dv->disposition = 'w'; | |
1004 | dv->used = di->disk.raid_disk; | |
1005 | } | |
1006 | return 1; | |
1007 | } | |
1008 | sysfs_free(mdi); | |
1009 | pr_err("%s not found in %s so cannot --replace it\n", | |
1010 | dv->devname, devname); | |
1011 | return -1; | |
1012 | } | |
1013 | ||
1014 | int Manage_with(struct supertype *tst, int fd, struct mddev_dev *dv, | |
1015 | unsigned long rdev, int verbose, char *devname) | |
1016 | { | |
1017 | struct mdinfo *mdi, *di; | |
1018 | /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */ | |
1019 | mdi = sysfs_read(fd, -1, GET_DEVS|GET_STATE); | |
1020 | if (!mdi || !mdi->devs) { | |
1021 | pr_err("Cannot find status of %s to enable replacement - strange\n", | |
1022 | devname); | |
1023 | return -1; | |
1024 | } | |
1025 | for (di = mdi->devs; di; di = di->next) | |
1026 | if (di->disk.major == (int)major(rdev) && | |
1027 | di->disk.minor == (int)minor(rdev)) | |
1028 | break; | |
1029 | if (di) { | |
1030 | int rv; | |
1031 | if (di->disk.state & (1<<MD_DISK_FAULTY)) { | |
1032 | pr_err("%s is faulty and cannot be a replacement\n", | |
1033 | dv->devname); | |
1034 | sysfs_free(mdi); | |
1035 | return -1; | |
1036 | } | |
1037 | if (di->disk.raid_disk >= 0) { | |
1038 | pr_err("%s is active and cannot be a replacement\n", | |
1039 | dv->devname); | |
1040 | sysfs_free(mdi); | |
1041 | return -1; | |
1042 | } | |
1043 | rv = sysfs_set_num(mdi, di, | |
1044 | "slot", dv->used); | |
1045 | if (rv) { | |
1046 | sysfs_free(mdi); | |
1047 | pr_err("Failed to %s as preferred replacement.\n", | |
1048 | dv->devname); | |
1049 | return -1; | |
1050 | } | |
1051 | if (verbose >= 0) | |
1052 | pr_err("Marked %s in %s as replacement for device %d\n", | |
1053 | dv->devname, devname, dv->used); | |
1054 | return 1; | |
1055 | } | |
1056 | sysfs_free(mdi); | |
1057 | pr_err("%s not found in %s so cannot make it preferred replacement\n", | |
1058 | dv->devname, devname); | |
1059 | return -1; | |
1060 | } | |
1061 | ||
64c4757e | 1062 | int Manage_subdevs(char *devname, int fd, |
833bb0f8 | 1063 | struct mddev_dev *devlist, int verbose, int test, |
11b391ec | 1064 | char *update, int force) |
cd29a5c8 | 1065 | { |
7bd04da9 | 1066 | /* Do something to each dev. |
682c7051 NB |
1067 | * devmode can be |
1068 | * 'a' - add the device | |
1069 | * try HOT_ADD_DISK | |
1070 | * If that fails EINVAL, try ADD_NEW_DISK | |
7bd04da9 N |
1071 | * 'A' - re-add the device |
1072 | * 'r' - remove the device: HOT_REMOVE_DISK | |
b80da661 NB |
1073 | * device can be 'faulty' or 'detached' in which case all |
1074 | * matching devices are removed. | |
682c7051 | 1075 | * 'f' - set the device faulty SET_DISK_FAULTY |
b80da661 NB |
1076 | * device can be 'detached' in which case any device that |
1077 | * is inaccessible will be marked faulty. | |
70c55e36 N |
1078 | * 'R' - mark this device as wanting replacement. |
1079 | * 'W' - this device is added if necessary and activated as | |
1080 | * a replacement for a previous 'R' device. | |
1081 | * ----- | |
1082 | * 'w' - 'W' will be changed to 'w' when it is paired with | |
1083 | * a 'R' device. If a 'W' is found while walking the list | |
1084 | * it must be unpaired, and is an error. | |
1085 | * 'M' - this is created by a 'missing' target. It is a slight | |
1086 | * variant on 'A' | |
262e3b7f N |
1087 | * 'F' - Another variant of 'A', where the device was faulty |
1088 | * so must be removed from the array first. | |
70c55e36 | 1089 | * |
98d27e39 N |
1090 | * For 'f' and 'r', the device can also be a kernel-internal |
1091 | * name such as 'sdb'. | |
682c7051 NB |
1092 | */ |
1093 | mdu_array_info_t array; | |
7a3be72f | 1094 | unsigned long long array_size; |
1d997643 | 1095 | struct mddev_dev *dv; |
682c7051 | 1096 | struct stat stb; |
cfad27a9 | 1097 | int tfd = -1; |
38aeaf3a | 1098 | struct supertype *tst; |
4725bc31 | 1099 | char *subarray = NULL; |
98d27e39 | 1100 | int sysfd = -1; |
7d2e6486 | 1101 | int count = 0; /* number of actions taken */ |
9f584691 N |
1102 | struct mdinfo info; |
1103 | int frozen = 0; | |
682c7051 NB |
1104 | |
1105 | if (ioctl(fd, GET_ARRAY_INFO, &array)) { | |
7bd04da9 | 1106 | pr_err("Cannot get array info for %s\n", |
682c7051 | 1107 | devname); |
bcbb3112 | 1108 | goto abort; |
682c7051 | 1109 | } |
9f584691 | 1110 | sysfs_init(&info, fd, 0); |
3da92f27 | 1111 | |
7bd04da9 | 1112 | /* array.size is only 32 bits and may be truncated. |
7a3be72f NB |
1113 | * So read from sysfs if possible, and record number of sectors |
1114 | */ | |
1115 | ||
1116 | array_size = get_component_size(fd); | |
1117 | if (array_size <= 0) | |
1118 | array_size = array.size * 2; | |
1119 | ||
4725bc31 | 1120 | tst = super_by_fd(fd, &subarray); |
3da92f27 | 1121 | if (!tst) { |
e7b84f9d | 1122 | pr_err("unsupport array - version %d.%d\n", |
3da92f27 | 1123 | array.major_version, array.minor_version); |
bcbb3112 | 1124 | goto abort; |
3da92f27 NB |
1125 | } |
1126 | ||
b3b4e8a7 | 1127 | stb.st_rdev = 0; |
38aeaf3a | 1128 | for (dv = devlist; dv; dv = dv->next) { |
38aeaf3a | 1129 | int rv; |
4a39c6f2 | 1130 | |
1d997643 N |
1131 | if (strcmp(dv->devname, "failed") == 0 || |
1132 | strcmp(dv->devname, "faulty") == 0) { | |
262e3b7f N |
1133 | if (dv->disposition != 'A' |
1134 | && dv->disposition != 'r') { | |
e7b84f9d | 1135 | pr_err("%s only meaningful " |
262e3b7f | 1136 | "with -r or --re-add, not -%c\n", |
b80da661 | 1137 | dv->devname, dv->disposition); |
bcbb3112 | 1138 | goto abort; |
b80da661 | 1139 | } |
262e3b7f N |
1140 | add_faulty(dv, fd, (dv->disposition == 'A' |
1141 | ? 'F' : 'r')); | |
1d997643 N |
1142 | continue; |
1143 | } | |
1144 | if (strcmp(dv->devname, "detached") == 0) { | |
b80da661 | 1145 | if (dv->disposition != 'r' && dv->disposition != 'f') { |
e7b84f9d | 1146 | pr_err("%s only meaningful " |
b80da661 NB |
1147 | "with -r of -f, not -%c\n", |
1148 | dv->devname, dv->disposition); | |
bcbb3112 | 1149 | goto abort; |
b80da661 | 1150 | } |
1d997643 N |
1151 | add_detached(dv, fd, dv->disposition); |
1152 | continue; | |
1153 | } | |
1154 | ||
1155 | if (strcmp(dv->devname, "missing") == 0) { | |
1156 | struct mddev_dev *add_devlist = NULL; | |
1157 | struct mddev_dev **dp; | |
c8e1a230 | 1158 | if (dv->disposition != 'A') { |
e7b84f9d | 1159 | pr_err("'missing' only meaningful " |
aab15415 | 1160 | "with --re-add\n"); |
bcbb3112 | 1161 | goto abort; |
a4e13010 | 1162 | } |
1d997643 | 1163 | add_devlist = conf_get_devs(); |
a4e13010 | 1164 | if (add_devlist == NULL) { |
e7b84f9d | 1165 | pr_err("no devices to scan for missing members."); |
a4e13010 N |
1166 | continue; |
1167 | } | |
1d997643 | 1168 | for (dp = &add_devlist; *dp; dp = & (*dp)->next) |
7bd04da9 | 1169 | /* 'M' (for 'missing') is like 'A' without errors */ |
1d997643 N |
1170 | (*dp)->disposition = 'M'; |
1171 | *dp = dv->next; | |
1172 | dv->next = add_devlist; | |
1173 | continue; | |
1174 | } | |
1175 | ||
1176 | if (strchr(dv->devname, '/') == NULL && | |
aab15415 N |
1177 | strchr(dv->devname, ':') == NULL && |
1178 | strlen(dv->devname) < 50) { | |
98d27e39 N |
1179 | /* Assume this is a kernel-internal name like 'sda1' */ |
1180 | int found = 0; | |
1181 | char dname[55]; | |
1182 | if (dv->disposition != 'r' && dv->disposition != 'f') { | |
e7b84f9d | 1183 | pr_err("%s only meaningful " |
cfad27a9 | 1184 | "with -r or -f, not -%c\n", |
98d27e39 | 1185 | dv->devname, dv->disposition); |
bcbb3112 | 1186 | goto abort; |
98d27e39 N |
1187 | } |
1188 | ||
1189 | sprintf(dname, "dev-%s", dv->devname); | |
1190 | sysfd = sysfs_open(fd2devnum(fd), dname, "block/dev"); | |
1191 | if (sysfd >= 0) { | |
1192 | char dn[20]; | |
1193 | int mj,mn; | |
1194 | if (sysfs_fd_get_str(sysfd, dn, 20) > 0 && | |
1195 | sscanf(dn, "%d:%d", &mj,&mn) == 2) { | |
1196 | stb.st_rdev = makedev(mj,mn); | |
1197 | found = 1; | |
1198 | } | |
1199 | close(sysfd); | |
1200 | sysfd = -1; | |
1201 | } | |
1202 | if (!found) { | |
1203 | sysfd = sysfs_open(fd2devnum(fd), dname, "state"); | |
1204 | if (sysfd < 0) { | |
e7b84f9d | 1205 | pr_err("%s does not appear " |
98d27e39 N |
1206 | "to be a component of %s\n", |
1207 | dv->devname, devname); | |
bcbb3112 | 1208 | goto abort; |
98d27e39 N |
1209 | } |
1210 | } | |
b80da661 | 1211 | } else { |
c7b47447 | 1212 | tfd = dev_open(dv->devname, O_RDONLY); |
5fe7f5f7 N |
1213 | if (tfd >= 0) |
1214 | fstat(tfd, &stb); | |
5a9de8db | 1215 | else { |
5fe7f5f7 N |
1216 | int open_err = errno; |
1217 | if (stat(dv->devname, &stb) != 0) { | |
1218 | pr_err("Cannot find %s: %s\n", | |
1219 | dv->devname, strerror(errno)); | |
1220 | goto abort; | |
1221 | } | |
1222 | if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
1223 | if (dv->disposition == 'M') | |
1224 | /* non-fatal. Also improbable */ | |
1225 | continue; | |
1226 | pr_err("%s is not a block device.\n", | |
1227 | dv->devname); | |
1228 | goto abort; | |
1229 | } | |
1230 | if (dv->disposition == 'r') | |
1231 | /* Be happy, the stat worked, that is | |
1232 | * enough for --remove | |
1233 | */ | |
1234 | ; | |
1235 | else { | |
1d997643 N |
1236 | if (dv->disposition == 'M') |
1237 | /* non-fatal */ | |
1238 | continue; | |
839f27a3 | 1239 | pr_err("Cannot open %s: %s\n", |
5fe7f5f7 | 1240 | dv->devname, strerror(open_err)); |
bcbb3112 | 1241 | goto abort; |
5a9de8db | 1242 | } |
b80da661 | 1243 | } |
682c7051 | 1244 | } |
cd29a5c8 | 1245 | switch(dv->disposition){ |
682c7051 | 1246 | default: |
e7b84f9d | 1247 | pr_err("internal error - devmode[%s]=%d\n", |
c913b90e | 1248 | dv->devname, dv->disposition); |
bcbb3112 | 1249 | goto abort; |
682c7051 | 1250 | case 'a': |
c8e1a230 | 1251 | case 'A': |
262e3b7f N |
1252 | case 'M': /* --re-add missing */ |
1253 | case 'F': /* --re-add faulty */ | |
4a39c6f2 | 1254 | /* add the device */ |
4725bc31 | 1255 | if (subarray) { |
e7b84f9d | 1256 | pr_err("Cannot add disks to a" |
f7dd881f DW |
1257 | " \'member\' array, perform this" |
1258 | " operation on the parent container\n"); | |
bcbb3112 | 1259 | goto abort; |
f94d52f4 | 1260 | } |
262e3b7f N |
1261 | if (dv->disposition == 'F') |
1262 | /* Need to remove first */ | |
1263 | ioctl(fd, HOT_REMOVE_DISK, | |
1264 | (unsigned long)stb.st_rdev); | |
f277ce36 | 1265 | /* Make sure it isn't in use (in 2.6 or later) */ |
1d997643 | 1266 | tfd = dev_open(dv->devname, O_RDONLY|O_EXCL); |
38aeaf3a N |
1267 | if (tfd >= 0) { |
1268 | /* We know no-one else is using it. We'll | |
1269 | * need non-exclusive access to add it, so | |
1270 | * do that now. | |
1271 | */ | |
1272 | close(tfd); | |
1273 | tfd = dev_open(dv->devname, O_RDONLY); | |
1274 | } | |
0fbf459d | 1275 | if (tfd < 0) { |
1d997643 N |
1276 | if (dv->disposition == 'M') |
1277 | continue; | |
e7b84f9d | 1278 | pr_err("Cannot open %s: %s\n", |
d7eaf49f | 1279 | dv->devname, strerror(errno)); |
bcbb3112 | 1280 | goto abort; |
d7eaf49f | 1281 | } |
9f584691 N |
1282 | if (!frozen) { |
1283 | if (sysfs_freeze_array(&info) == 1) | |
1284 | frozen = 1; | |
1285 | else | |
1286 | frozen = -1; | |
1287 | } | |
38aeaf3a N |
1288 | rv = Manage_add(fd, tfd, dv, tst, &array, |
1289 | force, verbose, devname, update, | |
1290 | stb.st_rdev, array_size); | |
1291 | close(tfd); | |
1292 | tfd = -1; | |
1293 | if (rv < 0) | |
bcbb3112 | 1294 | goto abort; |
38aeaf3a N |
1295 | if (rv > 0) |
1296 | count++; | |
682c7051 NB |
1297 | break; |
1298 | ||
1299 | case 'r': | |
1300 | /* hot remove */ | |
4725bc31 | 1301 | if (subarray) { |
e7b84f9d | 1302 | pr_err("Cannot remove disks from a" |
f7dd881f DW |
1303 | " \'member\' array, perform this" |
1304 | " operation on the parent container\n"); | |
d070235d N |
1305 | rv = -1; |
1306 | } else | |
1307 | rv = Manage_remove(tst, fd, dv, sysfd, | |
1308 | stb.st_rdev, verbose, | |
1309 | devname); | |
1310 | if (sysfd >= 0) | |
98d27e39 | 1311 | close(sysfd); |
d070235d N |
1312 | sysfd = -1; |
1313 | if (rv < 0) | |
bcbb3112 | 1314 | goto abort; |
d070235d N |
1315 | if (rv > 0) |
1316 | count++; | |
682c7051 NB |
1317 | break; |
1318 | ||
1319 | case 'f': /* set faulty */ | |
1320 | /* FIXME check current member */ | |
98d27e39 N |
1321 | if ((sysfd >= 0 && write(sysfd, "faulty", 6) != 6) || |
1322 | (sysfd < 0 && ioctl(fd, SET_DISK_FAULTY, | |
1323 | (unsigned long) stb.st_rdev))) { | |
e7b84f9d | 1324 | pr_err("set device faulty failed for %s: %s\n", |
1d997643 | 1325 | dv->devname, strerror(errno)); |
98d27e39 N |
1326 | if (sysfd >= 0) |
1327 | close(sysfd); | |
bcbb3112 | 1328 | goto abort; |
682c7051 | 1329 | } |
98d27e39 N |
1330 | if (sysfd >= 0) |
1331 | close(sysfd); | |
1332 | sysfd = -1; | |
7d2e6486 | 1333 | count++; |
dab6685f | 1334 | if (verbose >= 0) |
e7b84f9d | 1335 | pr_err("set %s faulty in %s\n", |
1d997643 | 1336 | dv->devname, devname); |
682c7051 | 1337 | break; |
70c55e36 N |
1338 | case 'R': /* Mark as replaceable */ |
1339 | if (subarray) { | |
1340 | pr_err("Cannot replace disks in a" | |
1341 | " \'member\' array, perform this" | |
1342 | " operation on the parent container\n"); | |
1343 | rv = -1; | |
1344 | } else { | |
1345 | if (!frozen) { | |
1346 | if (sysfs_freeze_array(&info) == 1) | |
1347 | frozen = 1; | |
1348 | else | |
1349 | frozen = -1; | |
1350 | } | |
1351 | rv = Manage_replace(tst, fd, dv, | |
1352 | stb.st_rdev, verbose, | |
1353 | devname); | |
1354 | } | |
1355 | if (rv < 0) | |
1356 | goto abort; | |
1357 | if (rv > 0) | |
1358 | count++; | |
1359 | break; | |
1360 | case 'W': /* --with device that doesn't match */ | |
1361 | pr_err("No matching --replace device for --with %s\n", | |
1362 | dv->devname); | |
1363 | goto abort; | |
1364 | case 'w': /* --with device which was matched */ | |
1365 | rv = Manage_with(tst, fd, dv, | |
1366 | stb.st_rdev, verbose, devname); | |
1367 | if (rv < 0) | |
1368 | goto abort; | |
1369 | break; | |
682c7051 NB |
1370 | } |
1371 | } | |
9f584691 N |
1372 | if (frozen > 0) |
1373 | sysfs_set_str(&info, NULL, "sync_action","idle"); | |
7d2e6486 N |
1374 | if (test && count == 0) |
1375 | return 2; | |
682c7051 | 1376 | return 0; |
bcbb3112 N |
1377 | |
1378 | abort: | |
9f584691 N |
1379 | if (frozen > 0) |
1380 | sysfs_set_str(&info, NULL, "sync_action","idle"); | |
bcbb3112 | 1381 | return 1; |
64c4757e | 1382 | } |
1f48664b NB |
1383 | |
1384 | int autodetect(void) | |
1385 | { | |
1386 | /* Open any md device, and issue the RAID_AUTORUN ioctl */ | |
1387 | int rv = 1; | |
1388 | int fd = dev_open("9:0", O_RDONLY); | |
1389 | if (fd >= 0) { | |
1390 | if (ioctl(fd, RAID_AUTORUN, 0) == 0) | |
1391 | rv = 0; | |
1392 | close(fd); | |
1393 | } | |
1394 | return rv; | |
1395 | } | |
aa534678 | 1396 | |
ba728be7 | 1397 | int Update_subarray(char *dev, char *subarray, char *update, struct mddev_ident *ident, int verbose) |
aa534678 DW |
1398 | { |
1399 | struct supertype supertype, *st = &supertype; | |
1400 | int fd, rv = 2; | |
1401 | ||
1402 | memset(st, 0, sizeof(*st)); | |
aa534678 | 1403 | |
ba728be7 | 1404 | fd = open_subarray(dev, subarray, st, verbose < 0); |
aa534678 DW |
1405 | if (fd < 0) |
1406 | return 2; | |
1407 | ||
1408 | if (!st->ss->update_subarray) { | |
ba728be7 | 1409 | if (verbose >= 0) |
e7b84f9d N |
1410 | pr_err("Operation not supported for %s metadata\n", |
1411 | st->ss->name); | |
aa534678 DW |
1412 | goto free_super; |
1413 | } | |
1414 | ||
1415 | if (mdmon_running(st->devnum)) | |
1416 | st->update_tail = &st->updates; | |
1417 | ||
a951a4f7 | 1418 | rv = st->ss->update_subarray(st, subarray, update, ident); |
aa534678 DW |
1419 | |
1420 | if (rv) { | |
ba728be7 | 1421 | if (verbose >= 0) |
e7b84f9d | 1422 | pr_err("Failed to update %s of subarray-%s in %s\n", |
aa534678 DW |
1423 | update, subarray, dev); |
1424 | } else if (st->update_tail) | |
1425 | flush_metadata_updates(st); | |
1426 | else | |
1427 | st->ss->sync_metadata(st); | |
1428 | ||
ba728be7 | 1429 | if (rv == 0 && strcmp(update, "name") == 0 && verbose >= 0) |
e7b84f9d N |
1430 | pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n", |
1431 | subarray, dev); | |
aa534678 DW |
1432 | |
1433 | free_super: | |
1434 | st->ss->free_super(st); | |
1435 | close(fd); | |
1436 | ||
1437 | return rv; | |
1438 | } | |
d52bb542 | 1439 | |
7bd04da9 N |
1440 | /* Move spare from one array to another If adding to destination array fails |
1441 | * add back to original array. | |
d52bb542 AC |
1442 | * Returns 1 on success, 0 on failure */ |
1443 | int move_spare(char *from_devname, char *to_devname, dev_t devid) | |
1444 | { | |
1445 | struct mddev_dev devlist; | |
1446 | char devname[20]; | |
1447 | ||
1448 | /* try to remove and add */ | |
1449 | int fd1 = open(to_devname, O_RDONLY); | |
1450 | int fd2 = open(from_devname, O_RDONLY); | |
1451 | ||
1452 | if (fd1 < 0 || fd2 < 0) { | |
1453 | if (fd1>=0) close(fd1); | |
1454 | if (fd2>=0) close(fd2); | |
1455 | return 0; | |
1456 | } | |
1457 | ||
1458 | devlist.next = NULL; | |
1459 | devlist.used = 0; | |
d52bb542 AC |
1460 | devlist.writemostly = 0; |
1461 | devlist.devname = devname; | |
1462 | sprintf(devname, "%d:%d", major(devid), minor(devid)); | |
1463 | ||
1464 | devlist.disposition = 'r'; | |
11b391ec | 1465 | if (Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0) == 0) { |
d52bb542 | 1466 | devlist.disposition = 'a'; |
11b391ec | 1467 | if (Manage_subdevs(to_devname, fd1, &devlist, -1, 0, NULL, 0) == 0) { |
d52bb542 AC |
1468 | /* make sure manager is aware of changes */ |
1469 | ping_manager(to_devname); | |
1470 | ping_manager(from_devname); | |
1471 | close(fd1); | |
1472 | close(fd2); | |
1473 | return 1; | |
1474 | } | |
11b391ec | 1475 | else Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0); |
d52bb542 AC |
1476 | } |
1477 | close(fd1); | |
1478 | close(fd2); | |
1479 | return 0; | |
1480 | } | |
435d4ebb | 1481 | #endif |