]>
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 | */ | |
4dd2df09 | 57 | mdi = sysfs_read(fd, NULL, GET_LEVEL|GET_VERSION); |
e9dd1598 N |
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 | ||
4dd2df09 | 130 | static void remove_devices(char *devnm, char *path) |
4ccad7b1 | 131 | { |
7bd04da9 | 132 | /* |
b1702f48 N |
133 | * Remove names at 'path' - possibly with |
134 | * partition suffixes - which link to the 'standard' | |
4dd2df09 | 135 | * name for devnm. These were probably created |
b1702f48 | 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 | ||
4dd2df09 | 149 | sprintf(base, "/dev/%s", devnm); |
4ccad7b1 | 150 | be = base + strlen(base); |
b1702f48 | 151 | |
503975b9 | 152 | path2 = xmalloc(strlen(path)+20); |
b1702f48 N |
153 | strcpy(path2, path); |
154 | pe = path2 + strlen(path2); | |
7bd04da9 | 155 | |
4ccad7b1 N |
156 | for (part = 0; part < 16; part++) { |
157 | if (part) { | |
158 | sprintf(be, "p%d", part); | |
b1702f48 N |
159 | |
160 | if (isdigit(pe[-1])) | |
161 | sprintf(pe, "p%d", part); | |
162 | else | |
163 | sprintf(pe, "%d", part); | |
4ccad7b1 | 164 | } |
b1702f48 | 165 | n = readlink(path2, link, sizeof(link)); |
d9ca03e9 | 166 | if (n > 0 && (int)strlen(base) == n && |
b1702f48 N |
167 | strncmp(link, base, n) == 0) |
168 | unlink(path2); | |
4ccad7b1 | 169 | } |
0eb26465 | 170 | free(path2); |
4ccad7b1 | 171 | } |
4ccad7b1 | 172 | |
ba728be7 N |
173 | int Manage_runstop(char *devname, int fd, int runstop, |
174 | int verbose, int will_retry) | |
64c4757e | 175 | { |
7bd04da9 | 176 | /* Run or stop the array. Array must already be configured |
ba728be7 N |
177 | * 'Run' requires >= 0.90.0 |
178 | * 'will_retry' is only relevant for 'stop', and means | |
179 | * that error messages are not wanted. | |
682c7051 | 180 | */ |
682c7051 | 181 | mdu_param_t param; /* unused */ |
bccd8153 | 182 | int rv = 0; |
82b27616 | 183 | |
ba728be7 N |
184 | if (will_retry && verbose == 0) |
185 | verbose = -1; | |
186 | ||
82b27616 | 187 | if (runstop == -1 && md_get_version(fd) < 9000) { |
7bd04da9 N |
188 | if (ioctl(fd, STOP_MD, 0) == 0) |
189 | return 0; | |
190 | pr_err("stopping device %s " | |
191 | "failed: %s\n", | |
192 | devname, strerror(errno)); | |
193 | return 1; | |
82b27616 | 194 | } |
aba69144 | 195 | |
682c7051 | 196 | if (md_get_version(fd) < 9000) { |
e7b84f9d | 197 | pr_err("need md driver version 0.90.0 or later\n"); |
682c7051 NB |
198 | return 1; |
199 | } | |
ba728be7 | 200 | |
7bd04da9 | 201 | if (runstop > 0) { |
682c7051 | 202 | if (ioctl(fd, RUN_ARRAY, ¶m)) { |
ba728be7 N |
203 | if (verbose >= 0) |
204 | pr_err("failed to run array %s: %s\n", | |
205 | devname, strerror(errno)); | |
682c7051 NB |
206 | return 1; |
207 | } | |
ba728be7 | 208 | if (verbose >= 0) |
e7b84f9d | 209 | pr_err("started %s\n", devname); |
682c7051 | 210 | } else if (runstop < 0){ |
8382f19b | 211 | struct map_ent *map = NULL; |
daf7a3ce | 212 | struct mdinfo *mdi; |
4dd2df09 | 213 | char devnm[32]; |
2fdf559d | 214 | char container[32]; |
eb0af526 N |
215 | int err; |
216 | int count; | |
daf7a3ce NB |
217 | /* If this is an mdmon managed array, just write 'inactive' |
218 | * to the array state and let mdmon clear up. | |
219 | */ | |
4dd2df09 | 220 | strcpy(devnm, fd2devnm(fd)); |
eb0af526 N |
221 | /* Get EXCL access first. If this fails, then attempting |
222 | * to stop is probably a bad idea. | |
223 | */ | |
2fdf559d N |
224 | mdi = sysfs_read(fd, NULL, GET_LEVEL|GET_VERSION); |
225 | if (mdi && is_subarray(mdi->text_version)) { | |
226 | char *sl; | |
227 | strncpy(container, mdi->text_version+1, sizeof(container)); | |
228 | container[sizeof(container)-1] = 0; | |
229 | sl = strchr(container, '/'); | |
230 | if (sl) | |
231 | *sl = 0; | |
232 | } else | |
233 | container[0] = 0; | |
eb0af526 | 234 | close(fd); |
2fdf559d N |
235 | count = 5; |
236 | while (((fd = ((devnm[0] == '/') | |
237 | ?open(devname, O_RDONLY|O_EXCL) | |
238 | :open_dev_flags(devnm, O_RDONLY|O_EXCL))) < 0 | |
239 | || strcmp(fd2devnm(fd), devnm) != 0) | |
240 | && container[0] | |
241 | && mdmon_running(container) | |
242 | && count) { | |
243 | if (fd >= 0) | |
244 | close(fd); | |
245 | flush_mdmon(container); | |
246 | count--; | |
247 | } | |
4dd2df09 | 248 | if (fd < 0 || strcmp(fd2devnm(fd), devnm) != 0) { |
eb0af526 N |
249 | if (fd >= 0) |
250 | close(fd); | |
ba728be7 N |
251 | if (verbose >= 0) |
252 | pr_err("Cannot get exclusive access to %s:" | |
253 | "Perhaps a running " | |
254 | "process, mounted filesystem " | |
255 | "or active volume group?\n", | |
256 | devname); | |
eb0af526 N |
257 | return 1; |
258 | } | |
daf7a3ce NB |
259 | if (mdi && |
260 | mdi->array.level > 0 && | |
3c558363 | 261 | is_subarray(mdi->text_version)) { |
1ae42d9d | 262 | int err; |
daf7a3ce NB |
263 | /* This is mdmon managed. */ |
264 | close(fd); | |
1ae42d9d | 265 | |
7bd04da9 N |
266 | /* As we have an O_EXCL open, any use of the device |
267 | * which blocks STOP_ARRAY is probably a transient use, | |
268 | * so it is reasonable to retry for a while - 5 seconds. | |
269 | */ | |
1ae42d9d KW |
270 | count = 25; |
271 | while (count && | |
272 | (err = sysfs_set_str(mdi, NULL, | |
273 | "array_state", | |
274 | "inactive")) < 0 | |
275 | && errno == EBUSY) { | |
276 | usleep(200000); | |
277 | count--; | |
278 | } | |
ba728be7 N |
279 | if (err) { |
280 | if (verbose >= 0) | |
281 | pr_err("failed to stop array %s: %s\n", | |
282 | devname, strerror(errno)); | |
bccd8153 JS |
283 | rv = 1; |
284 | goto out; | |
daf7a3ce NB |
285 | } |
286 | ||
287 | /* Give monitor a chance to act */ | |
c94709e8 | 288 | ping_monitor(mdi->text_version); |
daf7a3ce | 289 | |
4dd2df09 | 290 | fd = open_dev_excl(devnm); |
eb0af526 | 291 | if (fd < 0) { |
ba728be7 N |
292 | if (verbose >= 0) |
293 | pr_err("failed to completely stop %s" | |
294 | ": Device is busy\n", | |
295 | devname); | |
bccd8153 JS |
296 | rv = 1; |
297 | goto out; | |
eb0af526 | 298 | } |
ada6c239 N |
299 | } else if (mdi && |
300 | mdi->array.major_version == -1 && | |
301 | mdi->array.minor_version == -2 && | |
3c558363 | 302 | !is_subarray(mdi->text_version)) { |
430ea469 | 303 | struct mdstat_ent *mds, *m; |
ada6c239 N |
304 | /* container, possibly mdmon-managed. |
305 | * Make sure mdmon isn't opening it, which | |
306 | * would interfere with the 'stop' | |
307 | */ | |
308 | ping_monitor(mdi->sys_name); | |
430ea469 N |
309 | |
310 | /* now check that there are no existing arrays | |
311 | * which are members of this array | |
312 | */ | |
313 | mds = mdstat_read(0, 0); | |
7bd04da9 | 314 | for (m = mds; m; m = m->next) |
430ea469 N |
315 | if (m->metadata_version && |
316 | strncmp(m->metadata_version, "external:", 9)==0 && | |
4dd2df09 N |
317 | metadata_container_matches(m->metadata_version+9, |
318 | devnm)) { | |
ba728be7 | 319 | if (verbose >= 0) |
e7b84f9d N |
320 | pr_err("Cannot stop container %s: " |
321 | "member %s still active\n", | |
322 | devname, m->dev); | |
430ea469 | 323 | free_mdstat(mds); |
bccd8153 JS |
324 | rv = 1; |
325 | goto out; | |
430ea469 | 326 | } |
daf7a3ce | 327 | } |
daf7a3ce | 328 | |
eb0af526 N |
329 | /* As we have an O_EXCL open, any use of the device |
330 | * which blocks STOP_ARRAY is probably a transient use, | |
331 | * so it is reasonable to retry for a while - 5 seconds. | |
332 | */ | |
fb0d4b9c | 333 | count = 25; err = 0; |
eb0af526 N |
334 | while (count && fd >= 0 |
335 | && (err = ioctl(fd, STOP_ARRAY, NULL)) < 0 | |
336 | && errno == EBUSY) { | |
337 | usleep(200000); | |
338 | count --; | |
339 | } | |
340 | if (fd >= 0 && err) { | |
ba728be7 | 341 | if (verbose >= 0) { |
e7b84f9d N |
342 | pr_err("failed to stop array %s: %s\n", |
343 | devname, strerror(errno)); | |
d927f3c4 | 344 | if (errno == EBUSY) |
d33f1518 | 345 | cont_err("Perhaps a running " |
d927f3c4 NB |
346 | "process, mounted filesystem " |
347 | "or active volume group?\n"); | |
348 | } | |
bccd8153 JS |
349 | rv = 1; |
350 | goto out; | |
682c7051 | 351 | } |
97590376 | 352 | /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array |
d49410d3 N |
353 | * was stopped, so We'll do it here just to be sure. Drop any |
354 | * partitions as well... | |
97590376 | 355 | */ |
d49410d3 N |
356 | if (fd >= 0) |
357 | ioctl(fd, BLKRRPART, 0); | |
97590376 N |
358 | if (mdi) |
359 | sysfs_uevent(mdi, "change"); | |
daf7a3ce | 360 | |
8adabef5 | 361 | if (devnm[0] && use_udev()) { |
4dd2df09 N |
362 | struct map_ent *mp = map_by_devnm(&map, devnm); |
363 | remove_devices(devnm, mp ? mp->path : NULL); | |
4ccad7b1 N |
364 | } |
365 | ||
ba728be7 | 366 | if (verbose >= 0) |
e7b84f9d | 367 | pr_err("stopped %s\n", devname); |
4eb26970 | 368 | map_lock(&map); |
4dd2df09 | 369 | map_remove(&map, devnm); |
4eb26970 | 370 | map_unlock(&map); |
bccd8153 JS |
371 | out: |
372 | if (mdi) | |
373 | sysfs_free(mdi); | |
682c7051 | 374 | } |
bccd8153 | 375 | return rv; |
64c4757e NB |
376 | } |
377 | ||
5e73b024 N |
378 | static struct mddev_dev *add_one(struct mddev_dev *dv, char *name, char disp) |
379 | { | |
380 | struct mddev_dev *new; | |
381 | new = xmalloc(sizeof(*new)); | |
382 | memset(new, 0, sizeof(*new)); | |
383 | new->devname = xstrdup(name); | |
384 | new->disposition = disp; | |
385 | new->next = dv->next; | |
386 | dv->next = new; | |
387 | return new; | |
388 | } | |
389 | ||
1d997643 N |
390 | static void add_faulty(struct mddev_dev *dv, int fd, char disp) |
391 | { | |
392 | mdu_array_info_t array; | |
393 | mdu_disk_info_t disk; | |
394 | int remaining_disks; | |
395 | int i; | |
396 | ||
397 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) | |
398 | return; | |
399 | ||
400 | remaining_disks = array.nr_disks; | |
401 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
1d997643 N |
402 | char buf[40]; |
403 | disk.number = i; | |
404 | if (ioctl(fd, GET_DISK_INFO, &disk) != 0) | |
405 | continue; | |
406 | if (disk.major == 0 && disk.minor == 0) | |
407 | continue; | |
408 | remaining_disks--; | |
409 | if ((disk.state & 1) == 0) /* not faulty */ | |
410 | continue; | |
411 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
5e73b024 | 412 | dv = add_one(dv, buf, disp); |
1d997643 N |
413 | } |
414 | } | |
415 | ||
416 | static void add_detached(struct mddev_dev *dv, int fd, char disp) | |
417 | { | |
418 | mdu_array_info_t array; | |
419 | mdu_disk_info_t disk; | |
420 | int remaining_disks; | |
421 | int i; | |
422 | ||
423 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) | |
424 | return; | |
425 | ||
426 | remaining_disks = array.nr_disks; | |
427 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
1d997643 N |
428 | char buf[40]; |
429 | int sfd; | |
430 | disk.number = i; | |
431 | if (ioctl(fd, GET_DISK_INFO, &disk) != 0) | |
432 | continue; | |
433 | if (disk.major == 0 && disk.minor == 0) | |
434 | continue; | |
435 | remaining_disks--; | |
436 | if (disp == 'f' && (disk.state & 1) != 0) /* already faulty */ | |
437 | continue; | |
438 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
439 | sfd = dev_open(buf, O_RDONLY); | |
440 | if (sfd >= 0) { | |
441 | /* Not detached */ | |
442 | close(sfd); | |
443 | continue; | |
444 | } | |
445 | if (errno != ENXIO) | |
446 | /* Probably not detached */ | |
447 | continue; | |
5e73b024 | 448 | dv = add_one(dv, buf, disp); |
1d997643 N |
449 | } |
450 | } | |
451 | ||
64a78416 N |
452 | static void add_set(struct mddev_dev *dv, int fd, char set_char) |
453 | { | |
454 | mdu_array_info_t array; | |
455 | mdu_disk_info_t disk; | |
456 | int remaining_disks; | |
457 | int copies, set; | |
458 | int i; | |
459 | ||
460 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) | |
461 | return; | |
462 | if (array.level != 10) | |
463 | return; | |
464 | copies = ((array.layout & 0xff) * | |
465 | ((array.layout >> 8) & 0xff)); | |
466 | if (array.raid_disks % copies) | |
467 | return; | |
468 | ||
469 | remaining_disks = array.nr_disks; | |
470 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
471 | char buf[40]; | |
472 | disk.number = i; | |
473 | if (ioctl(fd, GET_DISK_INFO, &disk) != 0) | |
474 | continue; | |
475 | if (disk.major == 0 && disk.minor == 0) | |
476 | continue; | |
477 | remaining_disks--; | |
478 | set = disk.raid_disk % copies; | |
479 | if (set_char != set + 'A') | |
480 | continue; | |
481 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
482 | dv = add_one(dv, buf, dv->disposition); | |
483 | } | |
484 | } | |
485 | ||
abe94694 N |
486 | int attempt_re_add(int fd, int tfd, struct mddev_dev *dv, |
487 | struct supertype *dev_st, struct supertype *tst, | |
488 | unsigned long rdev, | |
489 | char *update, char *devname, int verbose, | |
490 | mdu_array_info_t *array) | |
491 | { | |
492 | struct mdinfo mdi; | |
493 | int duuid[4]; | |
494 | int ouuid[4]; | |
495 | ||
496 | dev_st->ss->getinfo_super(dev_st, &mdi, NULL); | |
497 | dev_st->ss->uuid_from_super(dev_st, ouuid); | |
498 | if (tst->sb) | |
499 | tst->ss->uuid_from_super(tst, duuid); | |
500 | else | |
501 | /* Assume uuid matches: kernel will check */ | |
502 | memcpy(duuid, ouuid, sizeof(ouuid)); | |
503 | if ((mdi.disk.state & (1<<MD_DISK_ACTIVE)) && | |
504 | !(mdi.disk.state & (1<<MD_DISK_FAULTY)) && | |
505 | memcmp(duuid, ouuid, sizeof(ouuid))==0) { | |
506 | /* Looks like it is worth a | |
507 | * try. Need to make sure | |
508 | * kernel will accept it | |
509 | * though. | |
510 | */ | |
511 | mdu_disk_info_t disc; | |
512 | /* re-add doesn't work for version-1 superblocks | |
513 | * before 2.6.18 :-( | |
514 | */ | |
515 | if (array->major_version == 1 && | |
516 | get_linux_version() <= 2006018) | |
517 | goto skip_re_add; | |
518 | disc.number = mdi.disk.number; | |
519 | if (ioctl(fd, GET_DISK_INFO, &disc) != 0 | |
520 | || disc.major != 0 || disc.minor != 0 | |
521 | ) | |
522 | goto skip_re_add; | |
523 | disc.major = major(rdev); | |
524 | disc.minor = minor(rdev); | |
525 | disc.number = mdi.disk.number; | |
526 | disc.raid_disk = mdi.disk.raid_disk; | |
527 | disc.state = mdi.disk.state; | |
528 | if (dv->writemostly == 1) | |
529 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; | |
530 | if (dv->writemostly == 2) | |
531 | disc.state &= ~(1 << MD_DISK_WRITEMOSTLY); | |
532 | remove_partitions(tfd); | |
533 | if (update || dv->writemostly > 0) { | |
534 | int rv = -1; | |
535 | tfd = dev_open(dv->devname, O_RDWR); | |
536 | if (tfd < 0) { | |
537 | pr_err("failed to open %s for" | |
538 | " superblock update during re-add\n", dv->devname); | |
539 | return -1; | |
540 | } | |
541 | ||
542 | if (dv->writemostly == 1) | |
543 | rv = dev_st->ss->update_super( | |
544 | dev_st, NULL, "writemostly", | |
545 | devname, verbose, 0, NULL); | |
546 | if (dv->writemostly == 2) | |
547 | rv = dev_st->ss->update_super( | |
548 | dev_st, NULL, "readwrite", | |
549 | devname, verbose, 0, NULL); | |
550 | if (update) | |
551 | rv = dev_st->ss->update_super( | |
552 | dev_st, NULL, update, | |
553 | devname, verbose, 0, NULL); | |
554 | if (rv == 0) | |
555 | rv = dev_st->ss->store_super(dev_st, tfd); | |
556 | close(tfd); | |
557 | if (rv != 0) { | |
558 | pr_err("failed to update" | |
559 | " superblock during re-add\n"); | |
560 | return -1; | |
561 | } | |
562 | } | |
563 | /* don't even try if disk is marked as faulty */ | |
564 | errno = 0; | |
565 | if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) { | |
566 | if (verbose >= 0) | |
567 | pr_err("re-added %s\n", dv->devname); | |
568 | return 1; | |
569 | } | |
570 | if (errno == ENOMEM || errno == EROFS) { | |
571 | pr_err("add new device failed for %s: %s\n", | |
572 | dv->devname, strerror(errno)); | |
573 | if (dv->disposition == 'M') | |
574 | return 0; | |
575 | return -1; | |
576 | } | |
577 | } | |
578 | skip_re_add: | |
579 | return 0; | |
580 | } | |
581 | ||
38aeaf3a N |
582 | int Manage_add(int fd, int tfd, struct mddev_dev *dv, |
583 | struct supertype *tst, mdu_array_info_t *array, | |
584 | int force, int verbose, char *devname, | |
585 | char *update, unsigned long rdev, unsigned long long array_size) | |
586 | { | |
587 | unsigned long long ldsize; | |
6d43efb5 | 588 | struct supertype *dev_st = NULL; |
38aeaf3a N |
589 | int j; |
590 | mdu_disk_info_t disc; | |
591 | ||
592 | if (!get_dev_size(tfd, dv->devname, &ldsize)) { | |
593 | if (dv->disposition == 'M') | |
594 | return 0; | |
595 | else | |
596 | return -1; | |
597 | } | |
598 | ||
599 | if (tst->ss->validate_geometry( | |
600 | tst, array->level, array->layout, | |
601 | array->raid_disks, NULL, | |
af4348dd | 602 | ldsize >> 9, INVALID_SECTORS, NULL, NULL, 0) == 0) { |
38aeaf3a N |
603 | if (!force) { |
604 | pr_err("%s is larger than %s can " | |
605 | "effectively use.\n" | |
606 | " Add --force is you " | |
607 | "really want to add this device.\n", | |
608 | dv->devname, devname); | |
609 | return -1; | |
610 | } | |
611 | pr_err("%s is larger than %s can " | |
612 | "effectively use.\n" | |
613 | " Adding anyway as --force " | |
614 | "was given.\n", | |
615 | dv->devname, devname); | |
616 | } | |
617 | if (!tst->ss->external && | |
618 | array->major_version == 0 && | |
619 | md_get_version(fd)%100 < 2) { | |
620 | if (ioctl(fd, HOT_ADD_DISK, rdev)==0) { | |
621 | if (verbose >= 0) | |
622 | pr_err("hot added %s\n", | |
623 | dv->devname); | |
624 | return 1; | |
625 | } | |
626 | ||
627 | pr_err("hot add failed for %s: %s\n", | |
628 | dv->devname, strerror(errno)); | |
629 | return -1; | |
630 | } | |
631 | ||
632 | if (array->not_persistent == 0 || tst->ss->external) { | |
633 | ||
634 | /* need to find a sample superblock to copy, and | |
635 | * a spare slot to use. | |
636 | * For 'external' array (well, container based), | |
637 | * We can just load the metadata for the array-> | |
638 | */ | |
639 | int array_failed; | |
640 | if (tst->sb) | |
641 | /* already loaded */; | |
642 | else if (tst->ss->external) { | |
643 | tst->ss->load_container(tst, fd, NULL); | |
644 | } else for (j = 0; j < tst->max_devs; j++) { | |
645 | char *dev; | |
646 | int dfd; | |
647 | disc.number = j; | |
648 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
649 | continue; | |
650 | if (disc.major==0 && disc.minor==0) | |
651 | continue; | |
652 | if ((disc.state & 4)==0) /* sync */ | |
653 | continue; | |
654 | /* Looks like a good device to try */ | |
655 | dev = map_dev(disc.major, disc.minor, 1); | |
656 | if (!dev) | |
657 | continue; | |
658 | dfd = dev_open(dev, O_RDONLY); | |
659 | if (dfd < 0) | |
660 | continue; | |
661 | if (tst->ss->load_super(tst, dfd, | |
662 | NULL)) { | |
663 | close(dfd); | |
664 | continue; | |
665 | } | |
666 | close(dfd); | |
667 | break; | |
668 | } | |
669 | /* FIXME this is a bad test to be using */ | |
670 | if (!tst->sb && dv->disposition != 'a') { | |
671 | /* we are re-adding a device to a | |
672 | * completely dead array - have to depend | |
673 | * on kernel to check | |
674 | */ | |
675 | } else if (!tst->sb) { | |
676 | pr_err("cannot load array metadata from %s\n", devname); | |
677 | return -1; | |
678 | } | |
679 | ||
680 | /* Make sure device is large enough */ | |
387fcd59 | 681 | if (tst->ss->avail_size(tst, ldsize/512, INVALID_SECTORS) < |
38aeaf3a N |
682 | array_size) { |
683 | if (dv->disposition == 'M') | |
684 | return 0; | |
685 | pr_err("%s not large enough to join array\n", | |
686 | dv->devname); | |
687 | return -1; | |
688 | } | |
689 | ||
690 | /* Possibly this device was recently part of | |
691 | * the array and was temporarily removed, and | |
692 | * is now being re-added. If so, we can | |
693 | * simply re-add it. | |
694 | */ | |
695 | ||
696 | if (array->not_persistent==0) { | |
697 | dev_st = dup_super(tst); | |
698 | dev_st->ss->load_super(dev_st, tfd, NULL); | |
699 | } | |
700 | if (dev_st && dev_st->sb) { | |
701 | int rv = attempt_re_add(fd, tfd, dv, | |
702 | dev_st, tst, | |
703 | rdev, | |
704 | update, devname, | |
705 | verbose, | |
706 | array); | |
707 | dev_st->ss->free_super(dev_st); | |
708 | if (rv) | |
709 | return rv; | |
710 | } | |
711 | if (dv->disposition == 'M') { | |
712 | if (verbose > 0) | |
713 | pr_err("--re-add for %s to %s is not possible\n", | |
714 | dv->devname, devname); | |
715 | return 0; | |
716 | } | |
717 | if (dv->disposition == 'A') { | |
718 | pr_err("--re-add for %s to %s is not possible\n", | |
719 | dv->devname, devname); | |
720 | return -1; | |
721 | } | |
722 | if (array->active_disks < array->raid_disks) { | |
723 | char *avail = xcalloc(array->raid_disks, 1); | |
724 | int d; | |
725 | int found = 0; | |
726 | ||
727 | for (d = 0; d < MAX_DISKS && found < array->active_disks; d++) { | |
728 | disc.number = d; | |
729 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
730 | continue; | |
731 | if (disc.major == 0 && disc.minor == 0) | |
732 | continue; | |
733 | if (!(disc.state & (1<<MD_DISK_SYNC))) | |
734 | continue; | |
735 | avail[disc.raid_disk] = 1; | |
736 | found++; | |
737 | } | |
738 | array_failed = !enough(array->level, array->raid_disks, | |
739 | array->layout, 1, avail); | |
740 | } else | |
741 | array_failed = 0; | |
742 | if (array_failed) { | |
743 | pr_err("%s has failed so using --add cannot work and might destroy\n", | |
744 | devname); | |
745 | pr_err("data on %s. You should stop the array and re-assemble it.\n", | |
746 | dv->devname); | |
747 | return -1; | |
748 | } | |
749 | } else { | |
750 | /* non-persistent. Must ensure that new drive | |
751 | * is at least array->size big. | |
752 | */ | |
753 | if (ldsize/512 < array_size) { | |
754 | pr_err("%s not large enough to join array\n", | |
755 | dv->devname); | |
756 | return -1; | |
757 | } | |
758 | } | |
759 | /* committed to really trying this device now*/ | |
760 | remove_partitions(tfd); | |
761 | ||
762 | /* in 2.6.17 and earlier, version-1 superblocks won't | |
763 | * use the number we write, but will choose a free number. | |
764 | * we must choose the same free number, which requires | |
765 | * starting at 'raid_disks' and counting up | |
766 | */ | |
767 | for (j = array->raid_disks; j < tst->max_devs; j++) { | |
768 | disc.number = j; | |
769 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
770 | break; | |
771 | if (disc.major==0 && disc.minor==0) | |
772 | break; | |
773 | if (disc.state & 8) /* removed */ | |
774 | break; | |
775 | } | |
776 | disc.major = major(rdev); | |
777 | disc.minor = minor(rdev); | |
778 | disc.number =j; | |
779 | disc.state = 0; | |
780 | if (array->not_persistent==0) { | |
781 | int dfd; | |
782 | if (dv->writemostly == 1) | |
783 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; | |
784 | dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); | |
785 | if (tst->ss->add_to_super(tst, &disc, dfd, | |
72ca9bcf | 786 | dv->devname, INVALID_SECTORS)) |
38aeaf3a N |
787 | return -1; |
788 | if (tst->ss->write_init_super(tst)) | |
789 | return -1; | |
790 | } else if (dv->disposition == 'A') { | |
791 | /* this had better be raid1. | |
792 | * As we are "--re-add"ing we must find a spare slot | |
793 | * to fill. | |
794 | */ | |
795 | char *used = xcalloc(array->raid_disks, 1); | |
796 | for (j = 0; j < tst->max_devs; j++) { | |
797 | mdu_disk_info_t disc2; | |
798 | disc2.number = j; | |
799 | if (ioctl(fd, GET_DISK_INFO, &disc2)) | |
800 | continue; | |
801 | if (disc2.major==0 && disc2.minor==0) | |
802 | continue; | |
803 | if (disc2.state & 8) /* removed */ | |
804 | continue; | |
805 | if (disc2.raid_disk < 0) | |
806 | continue; | |
807 | if (disc2.raid_disk > array->raid_disks) | |
808 | continue; | |
809 | used[disc2.raid_disk] = 1; | |
810 | } | |
811 | for (j = 0 ; j < array->raid_disks; j++) | |
812 | if (!used[j]) { | |
813 | disc.raid_disk = j; | |
814 | disc.state |= (1<<MD_DISK_SYNC); | |
815 | break; | |
816 | } | |
817 | free(used); | |
818 | } | |
819 | if (dv->writemostly == 1) | |
820 | disc.state |= (1 << MD_DISK_WRITEMOSTLY); | |
821 | if (tst->ss->external) { | |
822 | /* add a disk | |
823 | * to an external metadata container */ | |
824 | struct mdinfo new_mdi; | |
825 | struct mdinfo *sra; | |
826 | int container_fd; | |
4dd2df09 | 827 | char devnm[32]; |
38aeaf3a N |
828 | int dfd; |
829 | ||
4dd2df09 N |
830 | strcpy(devnm, fd2devnm(fd)); |
831 | ||
832 | container_fd = open_dev_excl(devnm); | |
38aeaf3a N |
833 | if (container_fd < 0) { |
834 | pr_err("add failed for %s:" | |
835 | " could not get exclusive access to container\n", | |
836 | dv->devname); | |
837 | tst->ss->free_super(tst); | |
838 | return -1; | |
839 | } | |
840 | ||
9cf9a1de | 841 | Kill(dv->devname, NULL, 0, -1, 0); |
38aeaf3a | 842 | dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); |
4dd2df09 | 843 | if (mdmon_running(tst->container_devnm)) |
38aeaf3a N |
844 | tst->update_tail = &tst->updates; |
845 | if (tst->ss->add_to_super(tst, &disc, dfd, | |
72ca9bcf | 846 | dv->devname, INVALID_SECTORS)) { |
38aeaf3a N |
847 | close(dfd); |
848 | close(container_fd); | |
849 | return -1; | |
850 | } | |
851 | if (tst->update_tail) | |
852 | flush_metadata_updates(tst); | |
853 | else | |
854 | tst->ss->sync_metadata(tst); | |
855 | ||
4dd2df09 | 856 | sra = sysfs_read(container_fd, NULL, 0); |
38aeaf3a N |
857 | if (!sra) { |
858 | pr_err("add failed for %s: sysfs_read failed\n", | |
859 | dv->devname); | |
860 | close(container_fd); | |
861 | tst->ss->free_super(tst); | |
862 | return -1; | |
863 | } | |
864 | sra->array.level = LEVEL_CONTAINER; | |
865 | /* Need to set data_offset and component_size */ | |
866 | tst->ss->getinfo_super(tst, &new_mdi, NULL); | |
867 | new_mdi.disk.major = disc.major; | |
868 | new_mdi.disk.minor = disc.minor; | |
869 | new_mdi.recovery_start = 0; | |
870 | /* Make sure fds are closed as they are O_EXCL which | |
871 | * would block add_disk */ | |
872 | tst->ss->free_super(tst); | |
873 | if (sysfs_add_disk(sra, &new_mdi, 0) != 0) { | |
874 | pr_err("add new device to external metadata" | |
875 | " failed for %s\n", dv->devname); | |
876 | close(container_fd); | |
877 | sysfs_free(sra); | |
878 | return -1; | |
879 | } | |
4dd2df09 | 880 | ping_monitor(devnm); |
38aeaf3a N |
881 | sysfs_free(sra); |
882 | close(container_fd); | |
883 | } else { | |
884 | tst->ss->free_super(tst); | |
885 | if (ioctl(fd, ADD_NEW_DISK, &disc)) { | |
886 | pr_err("add new device failed for %s as %d: %s\n", | |
887 | dv->devname, j, strerror(errno)); | |
888 | return -1; | |
889 | } | |
890 | } | |
891 | if (verbose >= 0) | |
892 | pr_err("added %s\n", dv->devname); | |
893 | return 1; | |
894 | } | |
895 | ||
d070235d N |
896 | int Manage_remove(struct supertype *tst, int fd, struct mddev_dev *dv, |
897 | int sysfd, unsigned long rdev, int verbose, char *devname) | |
898 | { | |
899 | int lfd = -1; | |
900 | int err; | |
901 | ||
902 | if (tst->ss->external) { | |
903 | /* To remove a device from a container, we must | |
904 | * check that it isn't in use in an array. | |
905 | * This involves looking in the 'holders' | |
906 | * directory - there must be just one entry, | |
907 | * the container. | |
908 | * To ensure that it doesn't get used as a | |
909 | * hot spare while we are checking, we | |
910 | * get an O_EXCL open on the container | |
911 | */ | |
aab15415 | 912 | int ret; |
4dd2df09 N |
913 | char devnm[32]; |
914 | strcpy(devnm, fd2devnm(fd)); | |
915 | lfd = open_dev_excl(devnm); | |
d070235d N |
916 | if (lfd < 0) { |
917 | pr_err("Cannot get exclusive access " | |
918 | " to container - odd\n"); | |
919 | return -1; | |
920 | } | |
aab15415 N |
921 | /* We may not be able to check on holders in |
922 | * sysfs, either because we don't have the dev num | |
923 | * (rdev == 0) or because the device has been detached | |
924 | * and the 'holders' directory no longer exists | |
925 | * (ret == -1). In that case, assume it is OK to | |
926 | * remove. | |
d070235d | 927 | */ |
aab15415 N |
928 | if (rdev == 0) |
929 | ret = -1; | |
930 | else | |
4dd2df09 | 931 | ret = sysfs_unique_holder(devnm, rdev); |
aab15415 N |
932 | if (ret == 0) { |
933 | pr_err("%s is not a member, cannot remove.\n", | |
934 | dv->devname); | |
935 | close(lfd); | |
936 | return -1; | |
937 | } | |
938 | if (ret >= 2) { | |
939 | pr_err("%s is still in use, cannot remove.\n", | |
940 | dv->devname); | |
d070235d N |
941 | close(lfd); |
942 | return -1; | |
943 | } | |
944 | } | |
945 | /* FIXME check that it is a current member */ | |
946 | if (sysfd >= 0) { | |
947 | /* device has been removed and we don't know | |
948 | * the major:minor number | |
949 | */ | |
950 | int n = write(sysfd, "remove", 6); | |
951 | if (n != 6) | |
952 | err = -1; | |
953 | else | |
954 | err = 0; | |
955 | } else { | |
956 | err = ioctl(fd, HOT_REMOVE_DISK, rdev); | |
957 | if (err && errno == ENODEV) { | |
958 | /* Old kernels rejected this if no personality | |
959 | * is registered */ | |
4dd2df09 | 960 | struct mdinfo *sra = sysfs_read(fd, NULL, GET_DEVS); |
d070235d N |
961 | struct mdinfo *dv = NULL; |
962 | if (sra) | |
963 | dv = sra->devs; | |
964 | for ( ; dv ; dv=dv->next) | |
965 | if (dv->disk.major == (int)major(rdev) && | |
966 | dv->disk.minor == (int)minor(rdev)) | |
967 | break; | |
968 | if (dv) | |
969 | err = sysfs_set_str(sra, dv, | |
970 | "state", "remove"); | |
971 | else | |
972 | err = -1; | |
973 | if (sra) | |
974 | sysfs_free(sra); | |
975 | } | |
976 | } | |
977 | if (err) { | |
978 | pr_err("hot remove failed " | |
979 | "for %s: %s\n", dv->devname, | |
980 | strerror(errno)); | |
981 | if (lfd >= 0) | |
982 | close(lfd); | |
983 | return -1; | |
984 | } | |
985 | if (tst->ss->external) { | |
986 | /* | |
987 | * Before dropping our exclusive open we make an | |
988 | * attempt at preventing mdmon from seeing an | |
989 | * 'add' event before reconciling this 'remove' | |
990 | * event. | |
991 | */ | |
4dd2df09 | 992 | char *devnm = fd2devnm(fd); |
d070235d | 993 | |
4dd2df09 | 994 | if (!devnm) { |
d070235d N |
995 | pr_err("unable to get container name\n"); |
996 | return -1; | |
997 | } | |
998 | ||
4dd2df09 | 999 | ping_manager(devnm); |
d070235d N |
1000 | } |
1001 | if (lfd >= 0) | |
1002 | close(lfd); | |
1003 | if (verbose >= 0) | |
1004 | pr_err("hot removed %s from %s\n", | |
1005 | dv->devname, devname); | |
1006 | return 1; | |
1007 | } | |
1008 | ||
70c55e36 N |
1009 | int Manage_replace(struct supertype *tst, int fd, struct mddev_dev *dv, |
1010 | unsigned long rdev, int verbose, char *devname) | |
1011 | { | |
1012 | struct mdinfo *mdi, *di; | |
1013 | if (tst->ss->external) { | |
1014 | pr_err("--replace only supported for native metadata (0.90 or 1.x)\n"); | |
1015 | return -1; | |
1016 | } | |
1017 | /* Need to find the device in sysfs and add 'want_replacement' to the | |
1018 | * status. | |
1019 | */ | |
4dd2df09 | 1020 | mdi = sysfs_read(fd, NULL, GET_DEVS); |
70c55e36 N |
1021 | if (!mdi || !mdi->devs) { |
1022 | pr_err("Cannot find status of %s to enable replacement - strange\n", | |
1023 | devname); | |
1024 | return -1; | |
1025 | } | |
1026 | for (di = mdi->devs; di; di = di->next) | |
1027 | if (di->disk.major == (int)major(rdev) && | |
1028 | di->disk.minor == (int)minor(rdev)) | |
1029 | break; | |
1030 | if (di) { | |
1031 | int rv; | |
1032 | if (di->disk.raid_disk < 0) { | |
1033 | pr_err("%s is not active and so cannot be replaced.\n", | |
1034 | dv->devname); | |
1035 | sysfs_free(mdi); | |
1036 | return -1; | |
1037 | } | |
1038 | rv = sysfs_set_str(mdi, di, | |
1039 | "state", "want_replacement"); | |
1040 | if (rv) { | |
1041 | sysfs_free(mdi); | |
1042 | pr_err("Failed to request replacement for %s\n", | |
1043 | dv->devname); | |
1044 | return -1; | |
1045 | } | |
1046 | if (verbose >= 0) | |
1047 | pr_err("Marked %s (device %d in %s) for replacement\n", | |
1048 | dv->devname, di->disk.raid_disk, devname); | |
1049 | /* If there is a matching 'with', we need to tell it which | |
1050 | * raid disk | |
1051 | */ | |
1052 | while (dv && dv->disposition != 'W') | |
1053 | dv = dv->next; | |
1054 | if (dv) { | |
1055 | dv->disposition = 'w'; | |
1056 | dv->used = di->disk.raid_disk; | |
1057 | } | |
1058 | return 1; | |
1059 | } | |
1060 | sysfs_free(mdi); | |
1061 | pr_err("%s not found in %s so cannot --replace it\n", | |
1062 | dv->devname, devname); | |
1063 | return -1; | |
1064 | } | |
1065 | ||
1066 | int Manage_with(struct supertype *tst, int fd, struct mddev_dev *dv, | |
1067 | unsigned long rdev, int verbose, char *devname) | |
1068 | { | |
1069 | struct mdinfo *mdi, *di; | |
1070 | /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */ | |
4dd2df09 | 1071 | mdi = sysfs_read(fd, NULL, GET_DEVS|GET_STATE); |
70c55e36 N |
1072 | if (!mdi || !mdi->devs) { |
1073 | pr_err("Cannot find status of %s to enable replacement - strange\n", | |
1074 | devname); | |
1075 | return -1; | |
1076 | } | |
1077 | for (di = mdi->devs; di; di = di->next) | |
1078 | if (di->disk.major == (int)major(rdev) && | |
1079 | di->disk.minor == (int)minor(rdev)) | |
1080 | break; | |
1081 | if (di) { | |
1082 | int rv; | |
1083 | if (di->disk.state & (1<<MD_DISK_FAULTY)) { | |
1084 | pr_err("%s is faulty and cannot be a replacement\n", | |
1085 | dv->devname); | |
1086 | sysfs_free(mdi); | |
1087 | return -1; | |
1088 | } | |
1089 | if (di->disk.raid_disk >= 0) { | |
1090 | pr_err("%s is active and cannot be a replacement\n", | |
1091 | dv->devname); | |
1092 | sysfs_free(mdi); | |
1093 | return -1; | |
1094 | } | |
1095 | rv = sysfs_set_num(mdi, di, | |
1096 | "slot", dv->used); | |
1097 | if (rv) { | |
1098 | sysfs_free(mdi); | |
1099 | pr_err("Failed to %s as preferred replacement.\n", | |
1100 | dv->devname); | |
1101 | return -1; | |
1102 | } | |
1103 | if (verbose >= 0) | |
1104 | pr_err("Marked %s in %s as replacement for device %d\n", | |
1105 | dv->devname, devname, dv->used); | |
1106 | return 1; | |
1107 | } | |
1108 | sysfs_free(mdi); | |
1109 | pr_err("%s not found in %s so cannot make it preferred replacement\n", | |
1110 | dv->devname, devname); | |
1111 | return -1; | |
1112 | } | |
1113 | ||
64c4757e | 1114 | int Manage_subdevs(char *devname, int fd, |
833bb0f8 | 1115 | struct mddev_dev *devlist, int verbose, int test, |
11b391ec | 1116 | char *update, int force) |
cd29a5c8 | 1117 | { |
7bd04da9 | 1118 | /* Do something to each dev. |
682c7051 NB |
1119 | * devmode can be |
1120 | * 'a' - add the device | |
1121 | * try HOT_ADD_DISK | |
1122 | * If that fails EINVAL, try ADD_NEW_DISK | |
7bd04da9 N |
1123 | * 'A' - re-add the device |
1124 | * 'r' - remove the device: HOT_REMOVE_DISK | |
b80da661 NB |
1125 | * device can be 'faulty' or 'detached' in which case all |
1126 | * matching devices are removed. | |
682c7051 | 1127 | * 'f' - set the device faulty SET_DISK_FAULTY |
b80da661 NB |
1128 | * device can be 'detached' in which case any device that |
1129 | * is inaccessible will be marked faulty. | |
70c55e36 N |
1130 | * 'R' - mark this device as wanting replacement. |
1131 | * 'W' - this device is added if necessary and activated as | |
1132 | * a replacement for a previous 'R' device. | |
1133 | * ----- | |
1134 | * 'w' - 'W' will be changed to 'w' when it is paired with | |
1135 | * a 'R' device. If a 'W' is found while walking the list | |
1136 | * it must be unpaired, and is an error. | |
1137 | * 'M' - this is created by a 'missing' target. It is a slight | |
1138 | * variant on 'A' | |
262e3b7f N |
1139 | * 'F' - Another variant of 'A', where the device was faulty |
1140 | * so must be removed from the array first. | |
70c55e36 | 1141 | * |
98d27e39 N |
1142 | * For 'f' and 'r', the device can also be a kernel-internal |
1143 | * name such as 'sdb'. | |
682c7051 NB |
1144 | */ |
1145 | mdu_array_info_t array; | |
7a3be72f | 1146 | unsigned long long array_size; |
1d997643 | 1147 | struct mddev_dev *dv; |
682c7051 | 1148 | struct stat stb; |
cfad27a9 | 1149 | int tfd = -1; |
38aeaf3a | 1150 | struct supertype *tst; |
4725bc31 | 1151 | char *subarray = NULL; |
98d27e39 | 1152 | int sysfd = -1; |
7d2e6486 | 1153 | int count = 0; /* number of actions taken */ |
9f584691 N |
1154 | struct mdinfo info; |
1155 | int frozen = 0; | |
8af530b0 | 1156 | int busy = 0; |
682c7051 NB |
1157 | |
1158 | if (ioctl(fd, GET_ARRAY_INFO, &array)) { | |
7bd04da9 | 1159 | pr_err("Cannot get array info for %s\n", |
682c7051 | 1160 | devname); |
bcbb3112 | 1161 | goto abort; |
682c7051 | 1162 | } |
4dd2df09 | 1163 | sysfs_init(&info, fd, NULL); |
3da92f27 | 1164 | |
7bd04da9 | 1165 | /* array.size is only 32 bits and may be truncated. |
7a3be72f NB |
1166 | * So read from sysfs if possible, and record number of sectors |
1167 | */ | |
1168 | ||
1169 | array_size = get_component_size(fd); | |
1170 | if (array_size <= 0) | |
1171 | array_size = array.size * 2; | |
1172 | ||
4725bc31 | 1173 | tst = super_by_fd(fd, &subarray); |
3da92f27 | 1174 | if (!tst) { |
e7b84f9d | 1175 | pr_err("unsupport array - version %d.%d\n", |
3da92f27 | 1176 | array.major_version, array.minor_version); |
bcbb3112 | 1177 | goto abort; |
3da92f27 NB |
1178 | } |
1179 | ||
b3b4e8a7 | 1180 | stb.st_rdev = 0; |
38aeaf3a | 1181 | for (dv = devlist; dv; dv = dv->next) { |
38aeaf3a | 1182 | int rv; |
4a39c6f2 | 1183 | |
1d997643 N |
1184 | if (strcmp(dv->devname, "failed") == 0 || |
1185 | strcmp(dv->devname, "faulty") == 0) { | |
262e3b7f N |
1186 | if (dv->disposition != 'A' |
1187 | && dv->disposition != 'r') { | |
e7b84f9d | 1188 | pr_err("%s only meaningful " |
262e3b7f | 1189 | "with -r or --re-add, not -%c\n", |
b80da661 | 1190 | dv->devname, dv->disposition); |
bcbb3112 | 1191 | goto abort; |
b80da661 | 1192 | } |
262e3b7f N |
1193 | add_faulty(dv, fd, (dv->disposition == 'A' |
1194 | ? 'F' : 'r')); | |
1d997643 N |
1195 | continue; |
1196 | } | |
1197 | if (strcmp(dv->devname, "detached") == 0) { | |
b80da661 | 1198 | if (dv->disposition != 'r' && dv->disposition != 'f') { |
e7b84f9d | 1199 | pr_err("%s only meaningful " |
b80da661 NB |
1200 | "with -r of -f, not -%c\n", |
1201 | dv->devname, dv->disposition); | |
bcbb3112 | 1202 | goto abort; |
b80da661 | 1203 | } |
1d997643 N |
1204 | add_detached(dv, fd, dv->disposition); |
1205 | continue; | |
1206 | } | |
1207 | ||
1208 | if (strcmp(dv->devname, "missing") == 0) { | |
1209 | struct mddev_dev *add_devlist = NULL; | |
1210 | struct mddev_dev **dp; | |
c8e1a230 | 1211 | if (dv->disposition != 'A') { |
e7b84f9d | 1212 | pr_err("'missing' only meaningful " |
aab15415 | 1213 | "with --re-add\n"); |
bcbb3112 | 1214 | goto abort; |
a4e13010 | 1215 | } |
1d997643 | 1216 | add_devlist = conf_get_devs(); |
a4e13010 | 1217 | if (add_devlist == NULL) { |
e7b84f9d | 1218 | pr_err("no devices to scan for missing members."); |
a4e13010 N |
1219 | continue; |
1220 | } | |
1d997643 | 1221 | for (dp = &add_devlist; *dp; dp = & (*dp)->next) |
7bd04da9 | 1222 | /* 'M' (for 'missing') is like 'A' without errors */ |
1d997643 N |
1223 | (*dp)->disposition = 'M'; |
1224 | *dp = dv->next; | |
1225 | dv->next = add_devlist; | |
1226 | continue; | |
1227 | } | |
1228 | ||
64a78416 N |
1229 | if (strncmp(dv->devname, "set-", 4) == 0 && |
1230 | strlen(dv->devname) == 5) { | |
1231 | int copies; | |
1232 | ||
1233 | if (dv->disposition != 'r' && | |
1234 | dv->disposition != 'f') { | |
1235 | pr_err("'%s' only meaningful with -r or -f\n", | |
1236 | dv->devname); | |
1237 | goto abort; | |
1238 | } | |
1239 | if (array.level != 10) { | |
1240 | pr_err("'%s' only meaningful with RAID10 arrays\n", | |
1241 | dv->devname); | |
1242 | goto abort; | |
1243 | } | |
1244 | copies = ((array.layout & 0xff) * | |
1245 | ((array.layout >> 8) & 0xff)); | |
1246 | if (array.raid_disks % copies != 0 || | |
1247 | dv->devname[4] < 'A' || | |
1248 | dv->devname[4] >= 'A' + copies || | |
1249 | copies > 26) { | |
1250 | pr_err("'%s' not meaningful with this array\n", | |
1251 | dv->devname); | |
1252 | goto abort; | |
1253 | } | |
1254 | add_set(dv, fd, dv->devname[4]); | |
1255 | continue; | |
1256 | } | |
1257 | ||
1d997643 | 1258 | if (strchr(dv->devname, '/') == NULL && |
aab15415 N |
1259 | strchr(dv->devname, ':') == NULL && |
1260 | strlen(dv->devname) < 50) { | |
98d27e39 N |
1261 | /* Assume this is a kernel-internal name like 'sda1' */ |
1262 | int found = 0; | |
1263 | char dname[55]; | |
1264 | if (dv->disposition != 'r' && dv->disposition != 'f') { | |
e7b84f9d | 1265 | pr_err("%s only meaningful " |
cfad27a9 | 1266 | "with -r or -f, not -%c\n", |
98d27e39 | 1267 | dv->devname, dv->disposition); |
bcbb3112 | 1268 | goto abort; |
98d27e39 N |
1269 | } |
1270 | ||
1271 | sprintf(dname, "dev-%s", dv->devname); | |
4dd2df09 | 1272 | sysfd = sysfs_open(fd2devnm(fd), dname, "block/dev"); |
98d27e39 N |
1273 | if (sysfd >= 0) { |
1274 | char dn[20]; | |
1275 | int mj,mn; | |
1276 | if (sysfs_fd_get_str(sysfd, dn, 20) > 0 && | |
1277 | sscanf(dn, "%d:%d", &mj,&mn) == 2) { | |
1278 | stb.st_rdev = makedev(mj,mn); | |
1279 | found = 1; | |
1280 | } | |
1281 | close(sysfd); | |
1282 | sysfd = -1; | |
1283 | } | |
1284 | if (!found) { | |
4dd2df09 | 1285 | sysfd = sysfs_open(fd2devnm(fd), dname, "state"); |
98d27e39 | 1286 | if (sysfd < 0) { |
e7b84f9d | 1287 | pr_err("%s does not appear " |
98d27e39 N |
1288 | "to be a component of %s\n", |
1289 | dv->devname, devname); | |
bcbb3112 | 1290 | goto abort; |
98d27e39 N |
1291 | } |
1292 | } | |
b80da661 | 1293 | } else { |
c7b47447 | 1294 | tfd = dev_open(dv->devname, O_RDONLY); |
5fe7f5f7 N |
1295 | if (tfd >= 0) |
1296 | fstat(tfd, &stb); | |
5a9de8db | 1297 | else { |
5fe7f5f7 N |
1298 | int open_err = errno; |
1299 | if (stat(dv->devname, &stb) != 0) { | |
1300 | pr_err("Cannot find %s: %s\n", | |
1301 | dv->devname, strerror(errno)); | |
1302 | goto abort; | |
1303 | } | |
1304 | if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
1305 | if (dv->disposition == 'M') | |
1306 | /* non-fatal. Also improbable */ | |
1307 | continue; | |
1308 | pr_err("%s is not a block device.\n", | |
1309 | dv->devname); | |
1310 | goto abort; | |
1311 | } | |
1312 | if (dv->disposition == 'r') | |
1313 | /* Be happy, the stat worked, that is | |
1314 | * enough for --remove | |
1315 | */ | |
1316 | ; | |
1317 | else { | |
1d997643 N |
1318 | if (dv->disposition == 'M') |
1319 | /* non-fatal */ | |
1320 | continue; | |
839f27a3 | 1321 | pr_err("Cannot open %s: %s\n", |
5fe7f5f7 | 1322 | dv->devname, strerror(open_err)); |
bcbb3112 | 1323 | goto abort; |
5a9de8db | 1324 | } |
b80da661 | 1325 | } |
682c7051 | 1326 | } |
cd29a5c8 | 1327 | switch(dv->disposition){ |
682c7051 | 1328 | default: |
e7b84f9d | 1329 | pr_err("internal error - devmode[%s]=%d\n", |
c913b90e | 1330 | dv->devname, dv->disposition); |
bcbb3112 | 1331 | goto abort; |
682c7051 | 1332 | case 'a': |
c8e1a230 | 1333 | case 'A': |
262e3b7f N |
1334 | case 'M': /* --re-add missing */ |
1335 | case 'F': /* --re-add faulty */ | |
4a39c6f2 | 1336 | /* add the device */ |
4725bc31 | 1337 | if (subarray) { |
e7b84f9d | 1338 | pr_err("Cannot add disks to a" |
f7dd881f DW |
1339 | " \'member\' array, perform this" |
1340 | " operation on the parent container\n"); | |
bcbb3112 | 1341 | goto abort; |
f94d52f4 | 1342 | } |
262e3b7f N |
1343 | if (dv->disposition == 'F') |
1344 | /* Need to remove first */ | |
1345 | ioctl(fd, HOT_REMOVE_DISK, | |
1346 | (unsigned long)stb.st_rdev); | |
f277ce36 | 1347 | /* Make sure it isn't in use (in 2.6 or later) */ |
1d997643 | 1348 | tfd = dev_open(dv->devname, O_RDONLY|O_EXCL); |
38aeaf3a N |
1349 | if (tfd >= 0) { |
1350 | /* We know no-one else is using it. We'll | |
1351 | * need non-exclusive access to add it, so | |
1352 | * do that now. | |
1353 | */ | |
1354 | close(tfd); | |
1355 | tfd = dev_open(dv->devname, O_RDONLY); | |
1356 | } | |
0fbf459d | 1357 | if (tfd < 0) { |
1d997643 N |
1358 | if (dv->disposition == 'M') |
1359 | continue; | |
e7b84f9d | 1360 | pr_err("Cannot open %s: %s\n", |
d7eaf49f | 1361 | dv->devname, strerror(errno)); |
bcbb3112 | 1362 | goto abort; |
d7eaf49f | 1363 | } |
9f584691 N |
1364 | if (!frozen) { |
1365 | if (sysfs_freeze_array(&info) == 1) | |
1366 | frozen = 1; | |
1367 | else | |
1368 | frozen = -1; | |
1369 | } | |
38aeaf3a N |
1370 | rv = Manage_add(fd, tfd, dv, tst, &array, |
1371 | force, verbose, devname, update, | |
1372 | stb.st_rdev, array_size); | |
1373 | close(tfd); | |
1374 | tfd = -1; | |
1375 | if (rv < 0) | |
bcbb3112 | 1376 | goto abort; |
38aeaf3a N |
1377 | if (rv > 0) |
1378 | count++; | |
682c7051 NB |
1379 | break; |
1380 | ||
1381 | case 'r': | |
1382 | /* hot remove */ | |
4725bc31 | 1383 | if (subarray) { |
e7b84f9d | 1384 | pr_err("Cannot remove disks from a" |
f7dd881f DW |
1385 | " \'member\' array, perform this" |
1386 | " operation on the parent container\n"); | |
d070235d N |
1387 | rv = -1; |
1388 | } else | |
1389 | rv = Manage_remove(tst, fd, dv, sysfd, | |
1390 | stb.st_rdev, verbose, | |
1391 | devname); | |
1392 | if (sysfd >= 0) | |
98d27e39 | 1393 | close(sysfd); |
d070235d N |
1394 | sysfd = -1; |
1395 | if (rv < 0) | |
bcbb3112 | 1396 | goto abort; |
d070235d N |
1397 | if (rv > 0) |
1398 | count++; | |
682c7051 NB |
1399 | break; |
1400 | ||
1401 | case 'f': /* set faulty */ | |
1402 | /* FIXME check current member */ | |
98d27e39 N |
1403 | if ((sysfd >= 0 && write(sysfd, "faulty", 6) != 6) || |
1404 | (sysfd < 0 && ioctl(fd, SET_DISK_FAULTY, | |
1405 | (unsigned long) stb.st_rdev))) { | |
8af530b0 N |
1406 | if (errno == EBUSY) |
1407 | busy = 1; | |
e7b84f9d | 1408 | pr_err("set device faulty failed for %s: %s\n", |
1d997643 | 1409 | dv->devname, strerror(errno)); |
98d27e39 N |
1410 | if (sysfd >= 0) |
1411 | close(sysfd); | |
bcbb3112 | 1412 | goto abort; |
682c7051 | 1413 | } |
98d27e39 N |
1414 | if (sysfd >= 0) |
1415 | close(sysfd); | |
1416 | sysfd = -1; | |
7d2e6486 | 1417 | count++; |
dab6685f | 1418 | if (verbose >= 0) |
e7b84f9d | 1419 | pr_err("set %s faulty in %s\n", |
1d997643 | 1420 | dv->devname, devname); |
682c7051 | 1421 | break; |
70c55e36 N |
1422 | case 'R': /* Mark as replaceable */ |
1423 | if (subarray) { | |
1424 | pr_err("Cannot replace disks in a" | |
1425 | " \'member\' array, perform this" | |
1426 | " operation on the parent container\n"); | |
1427 | rv = -1; | |
1428 | } else { | |
1429 | if (!frozen) { | |
1430 | if (sysfs_freeze_array(&info) == 1) | |
1431 | frozen = 1; | |
1432 | else | |
1433 | frozen = -1; | |
1434 | } | |
1435 | rv = Manage_replace(tst, fd, dv, | |
1436 | stb.st_rdev, verbose, | |
1437 | devname); | |
1438 | } | |
1439 | if (rv < 0) | |
1440 | goto abort; | |
1441 | if (rv > 0) | |
1442 | count++; | |
1443 | break; | |
1444 | case 'W': /* --with device that doesn't match */ | |
1445 | pr_err("No matching --replace device for --with %s\n", | |
1446 | dv->devname); | |
1447 | goto abort; | |
1448 | case 'w': /* --with device which was matched */ | |
1449 | rv = Manage_with(tst, fd, dv, | |
1450 | stb.st_rdev, verbose, devname); | |
1451 | if (rv < 0) | |
1452 | goto abort; | |
1453 | break; | |
682c7051 NB |
1454 | } |
1455 | } | |
9f584691 N |
1456 | if (frozen > 0) |
1457 | sysfs_set_str(&info, NULL, "sync_action","idle"); | |
7d2e6486 N |
1458 | if (test && count == 0) |
1459 | return 2; | |
682c7051 | 1460 | return 0; |
bcbb3112 N |
1461 | |
1462 | abort: | |
9f584691 N |
1463 | if (frozen > 0) |
1464 | sysfs_set_str(&info, NULL, "sync_action","idle"); | |
8af530b0 | 1465 | return !test && busy ? 2 : 1; |
64c4757e | 1466 | } |
1f48664b NB |
1467 | |
1468 | int autodetect(void) | |
1469 | { | |
1470 | /* Open any md device, and issue the RAID_AUTORUN ioctl */ | |
1471 | int rv = 1; | |
1472 | int fd = dev_open("9:0", O_RDONLY); | |
1473 | if (fd >= 0) { | |
1474 | if (ioctl(fd, RAID_AUTORUN, 0) == 0) | |
1475 | rv = 0; | |
1476 | close(fd); | |
1477 | } | |
1478 | return rv; | |
1479 | } | |
aa534678 | 1480 | |
ba728be7 | 1481 | int Update_subarray(char *dev, char *subarray, char *update, struct mddev_ident *ident, int verbose) |
aa534678 DW |
1482 | { |
1483 | struct supertype supertype, *st = &supertype; | |
1484 | int fd, rv = 2; | |
1485 | ||
1486 | memset(st, 0, sizeof(*st)); | |
aa534678 | 1487 | |
ba728be7 | 1488 | fd = open_subarray(dev, subarray, st, verbose < 0); |
aa534678 DW |
1489 | if (fd < 0) |
1490 | return 2; | |
1491 | ||
1492 | if (!st->ss->update_subarray) { | |
ba728be7 | 1493 | if (verbose >= 0) |
e7b84f9d N |
1494 | pr_err("Operation not supported for %s metadata\n", |
1495 | st->ss->name); | |
aa534678 DW |
1496 | goto free_super; |
1497 | } | |
1498 | ||
4dd2df09 | 1499 | if (mdmon_running(st->devnm)) |
aa534678 DW |
1500 | st->update_tail = &st->updates; |
1501 | ||
a951a4f7 | 1502 | rv = st->ss->update_subarray(st, subarray, update, ident); |
aa534678 DW |
1503 | |
1504 | if (rv) { | |
ba728be7 | 1505 | if (verbose >= 0) |
e7b84f9d | 1506 | pr_err("Failed to update %s of subarray-%s in %s\n", |
aa534678 DW |
1507 | update, subarray, dev); |
1508 | } else if (st->update_tail) | |
1509 | flush_metadata_updates(st); | |
1510 | else | |
1511 | st->ss->sync_metadata(st); | |
1512 | ||
ba728be7 | 1513 | if (rv == 0 && strcmp(update, "name") == 0 && verbose >= 0) |
e7b84f9d N |
1514 | pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n", |
1515 | subarray, dev); | |
aa534678 DW |
1516 | |
1517 | free_super: | |
1518 | st->ss->free_super(st); | |
1519 | close(fd); | |
1520 | ||
1521 | return rv; | |
1522 | } | |
d52bb542 | 1523 | |
7bd04da9 N |
1524 | /* Move spare from one array to another If adding to destination array fails |
1525 | * add back to original array. | |
d52bb542 AC |
1526 | * Returns 1 on success, 0 on failure */ |
1527 | int move_spare(char *from_devname, char *to_devname, dev_t devid) | |
1528 | { | |
1529 | struct mddev_dev devlist; | |
1530 | char devname[20]; | |
1531 | ||
1532 | /* try to remove and add */ | |
1533 | int fd1 = open(to_devname, O_RDONLY); | |
1534 | int fd2 = open(from_devname, O_RDONLY); | |
1535 | ||
1536 | if (fd1 < 0 || fd2 < 0) { | |
1537 | if (fd1>=0) close(fd1); | |
1538 | if (fd2>=0) close(fd2); | |
1539 | return 0; | |
1540 | } | |
1541 | ||
1542 | devlist.next = NULL; | |
1543 | devlist.used = 0; | |
d52bb542 AC |
1544 | devlist.writemostly = 0; |
1545 | devlist.devname = devname; | |
1546 | sprintf(devname, "%d:%d", major(devid), minor(devid)); | |
1547 | ||
1548 | devlist.disposition = 'r'; | |
11b391ec | 1549 | if (Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0) == 0) { |
d52bb542 | 1550 | devlist.disposition = 'a'; |
11b391ec | 1551 | if (Manage_subdevs(to_devname, fd1, &devlist, -1, 0, NULL, 0) == 0) { |
d52bb542 AC |
1552 | /* make sure manager is aware of changes */ |
1553 | ping_manager(to_devname); | |
1554 | ping_manager(from_devname); | |
1555 | close(fd1); | |
1556 | close(fd2); | |
1557 | return 1; | |
1558 | } | |
11b391ec | 1559 | else Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0); |
d52bb542 AC |
1560 | } |
1561 | close(fd1); | |
1562 | close(fd2); | |
1563 | return 0; | |
1564 | } | |
435d4ebb | 1565 | #endif |