]>
Commit | Line | Data |
---|---|---|
64c4757e | 1 | /* |
9a9dab36 | 2 | * mdadm - manage Linux "md" devices aka RAID arrays. |
64c4757e | 3 | * |
6f02172d | 4 | * Copyright (C) 2001-2013 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 NB |
29 | |
30 | int Manage_ro(char *devname, int fd, int readonly) | |
31 | { | |
682c7051 NB |
32 | /* switch to readonly or rw |
33 | * | |
34 | * requires >= 0.90.0 | |
35 | * first check that array is runing | |
36 | * use RESTART_ARRAY_RW or STOP_ARRAY_RO | |
37 | * | |
38 | */ | |
e9dd1598 | 39 | struct mdinfo *mdi; |
b73e45ae | 40 | int rv = 0; |
aba69144 | 41 | |
7bd04da9 | 42 | /* If this is an externally-managed array, we need to modify the |
e9dd1598 N |
43 | * metadata_version so that mdmon doesn't undo our change. |
44 | */ | |
4dd2df09 | 45 | mdi = sysfs_read(fd, NULL, GET_LEVEL|GET_VERSION); |
e9dd1598 N |
46 | if (mdi && |
47 | mdi->array.major_version == -1 && | |
e9dd1598 N |
48 | is_subarray(mdi->text_version)) { |
49 | char vers[64]; | |
50 | strcpy(vers, "external:"); | |
51 | strcat(vers, mdi->text_version); | |
52 | if (readonly > 0) { | |
53 | int rv; | |
54 | /* We set readonly ourselves. */ | |
55 | vers[9] = '-'; | |
56 | sysfs_set_str(mdi, NULL, "metadata_version", vers); | |
57 | ||
58 | close(fd); | |
59 | rv = sysfs_set_str(mdi, NULL, "array_state", "readonly"); | |
60 | ||
61 | if (rv < 0) { | |
e7b84f9d | 62 | pr_err("failed to set readonly for %s: %s\n", |
e9dd1598 N |
63 | devname, strerror(errno)); |
64 | ||
65 | vers[9] = mdi->text_version[0]; | |
66 | sysfs_set_str(mdi, NULL, "metadata_version", vers); | |
b73e45ae JS |
67 | rv = 1; |
68 | goto out; | |
e9dd1598 N |
69 | } |
70 | } else { | |
71 | char *cp; | |
72 | /* We cannot set read/write - must signal mdmon */ | |
73 | vers[9] = '/'; | |
74 | sysfs_set_str(mdi, NULL, "metadata_version", vers); | |
75 | ||
76 | cp = strchr(vers+10, '/'); | |
1471b8b1 | 77 | if (cp) |
e9dd1598 N |
78 | *cp = 0; |
79 | ping_monitor(vers+10); | |
9ea5a252 DW |
80 | if (mdi->array.level <= 0) |
81 | sysfs_set_str(mdi, NULL, "array_state", "active"); | |
e9dd1598 | 82 | } |
b73e45ae | 83 | goto out; |
e9dd1598 | 84 | } |
32141c17 | 85 | |
80223cb4 JS |
86 | if (!md_array_active(fd)) { |
87 | pr_err("%s does not appear to be active.\n", devname); | |
b73e45ae JS |
88 | rv = 1; |
89 | goto out; | |
682c7051 | 90 | } |
aba69144 | 91 | |
7bd04da9 | 92 | if (readonly > 0) { |
682c7051 | 93 | if (ioctl(fd, STOP_ARRAY_RO, NULL)) { |
e7b84f9d | 94 | pr_err("failed to set readonly for %s: %s\n", |
682c7051 | 95 | devname, strerror(errno)); |
b73e45ae JS |
96 | rv = 1; |
97 | goto out; | |
682c7051 NB |
98 | } |
99 | } else if (readonly < 0) { | |
100 | if (ioctl(fd, RESTART_ARRAY_RW, NULL)) { | |
e7b84f9d | 101 | pr_err("failed to set writable for %s: %s\n", |
682c7051 | 102 | devname, strerror(errno)); |
b73e45ae JS |
103 | rv = 1; |
104 | goto out; | |
682c7051 NB |
105 | } |
106 | } | |
b73e45ae | 107 | out: |
6e8d27e7 | 108 | sysfs_free(mdi); |
b73e45ae | 109 | return rv; |
64c4757e NB |
110 | } |
111 | ||
4dd2df09 | 112 | static void remove_devices(char *devnm, char *path) |
4ccad7b1 | 113 | { |
7bd04da9 | 114 | /* |
b1702f48 N |
115 | * Remove names at 'path' - possibly with |
116 | * partition suffixes - which link to the 'standard' | |
4dd2df09 | 117 | * name for devnm. These were probably created |
b1702f48 | 118 | * by mdadm when the array was assembled. |
4ccad7b1 N |
119 | */ |
120 | char base[40]; | |
121 | char *path2; | |
122 | char link[1024]; | |
123 | int n; | |
124 | int part; | |
125 | char *be; | |
126 | char *pe; | |
127 | ||
b1702f48 N |
128 | if (!path) |
129 | return; | |
130 | ||
4dd2df09 | 131 | sprintf(base, "/dev/%s", devnm); |
4ccad7b1 | 132 | be = base + strlen(base); |
b1702f48 | 133 | |
503975b9 | 134 | path2 = xmalloc(strlen(path)+20); |
b1702f48 N |
135 | strcpy(path2, path); |
136 | pe = path2 + strlen(path2); | |
7bd04da9 | 137 | |
4ccad7b1 N |
138 | for (part = 0; part < 16; part++) { |
139 | if (part) { | |
140 | sprintf(be, "p%d", part); | |
b1702f48 N |
141 | |
142 | if (isdigit(pe[-1])) | |
143 | sprintf(pe, "p%d", part); | |
144 | else | |
145 | sprintf(pe, "%d", part); | |
4ccad7b1 | 146 | } |
b1702f48 | 147 | n = readlink(path2, link, sizeof(link)); |
d9ca03e9 | 148 | if (n > 0 && (int)strlen(base) == n && |
b1702f48 N |
149 | strncmp(link, base, n) == 0) |
150 | unlink(path2); | |
4ccad7b1 | 151 | } |
0eb26465 | 152 | free(path2); |
4ccad7b1 | 153 | } |
4ccad7b1 | 154 | |
d5a40416 | 155 | int Manage_run(char *devname, int fd, struct context *c) |
64c4757e | 156 | { |
fe7e0e64 N |
157 | /* Run the array. Array must already be configured |
158 | * Requires >= 0.90.0 | |
682c7051 | 159 | */ |
d3786cdc | 160 | char nm[32], *nmp; |
82b27616 | 161 | |
d3786cdc N |
162 | nmp = fd2devnm(fd); |
163 | if (!nmp) { | |
164 | pr_err("Cannot find %s in sysfs!!\n", devname); | |
fe7e0e64 N |
165 | return 1; |
166 | } | |
d3786cdc | 167 | strcpy(nm, nmp); |
d5a40416 | 168 | return IncrementalScan(c, nm); |
fe7e0e64 N |
169 | } |
170 | ||
171 | int Manage_stop(char *devname, int fd, int verbose, int will_retry) | |
172 | { | |
173 | /* Stop the array. Array must already be configured | |
174 | * 'will_retry' means that error messages are not wanted. | |
175 | */ | |
176 | int rv = 0; | |
177 | struct map_ent *map = NULL; | |
178 | struct mdinfo *mdi; | |
179 | char devnm[32]; | |
180 | char container[32]; | |
181 | int err; | |
182 | int count; | |
2eba8496 N |
183 | char buf[32]; |
184 | unsigned long long rd1, rd2; | |
fe7e0e64 | 185 | |
ba728be7 N |
186 | if (will_retry && verbose == 0) |
187 | verbose = -1; | |
188 | ||
fe7e0e64 N |
189 | strcpy(devnm, fd2devnm(fd)); |
190 | /* Get EXCL access first. If this fails, then attempting | |
191 | * to stop is probably a bad idea. | |
192 | */ | |
2eba8496 | 193 | mdi = sysfs_read(fd, NULL, GET_LEVEL|GET_COMPONENT|GET_VERSION); |
fe7e0e64 N |
194 | if (mdi && is_subarray(mdi->text_version)) { |
195 | char *sl; | |
196 | strncpy(container, mdi->text_version+1, sizeof(container)); | |
197 | container[sizeof(container)-1] = 0; | |
198 | sl = strchr(container, '/'); | |
199 | if (sl) | |
200 | *sl = 0; | |
201 | } else | |
202 | container[0] = 0; | |
203 | close(fd); | |
204 | count = 5; | |
52b6ccad | 205 | while (((fd = ((devname[0] == '/') |
fe7e0e64 | 206 | ?open(devname, O_RDONLY|O_EXCL) |
fc54fe7a JS |
207 | :open_dev_flags(devnm, O_RDONLY|O_EXCL))) < 0 || |
208 | strcmp(fd2devnm(fd), devnm) != 0) && container[0] && | |
209 | mdmon_running(container) && count) { | |
932be627 N |
210 | /* Can't open, so something might be wrong. However it |
211 | * is a container, so we might be racing with mdmon, so | |
212 | * retry for a bit. | |
213 | */ | |
fe7e0e64 N |
214 | if (fd >= 0) |
215 | close(fd); | |
216 | flush_mdmon(container); | |
217 | count--; | |
682c7051 | 218 | } |
fe7e0e64 N |
219 | if (fd < 0 || strcmp(fd2devnm(fd), devnm) != 0) { |
220 | if (fd >= 0) | |
221 | close(fd); | |
ba728be7 | 222 | if (verbose >= 0) |
7a862a02 | 223 | pr_err("Cannot get exclusive access to %s:Perhaps a running process, mounted filesystem or active volume group?\n", |
fe7e0e64 N |
224 | devname); |
225 | return 1; | |
226 | } | |
932be627 N |
227 | /* If this is an mdmon managed array, just write 'inactive' |
228 | * to the array state and let mdmon clear up. | |
229 | */ | |
fe7e0e64 N |
230 | if (mdi && |
231 | mdi->array.level > 0 && | |
232 | is_subarray(mdi->text_version)) { | |
eb0af526 | 233 | int err; |
fe7e0e64 | 234 | /* This is mdmon managed. */ |
eb0af526 | 235 | close(fd); |
daf7a3ce | 236 | |
932be627 | 237 | /* As we had an O_EXCL open, any use of the device |
eb0af526 N |
238 | * which blocks STOP_ARRAY is probably a transient use, |
239 | * so it is reasonable to retry for a while - 5 seconds. | |
240 | */ | |
fe7e0e64 N |
241 | count = 25; |
242 | while (count && | |
243 | (err = sysfs_set_str(mdi, NULL, | |
244 | "array_state", | |
fc54fe7a JS |
245 | "inactive")) < 0 && |
246 | errno == EBUSY) { | |
239b3cc0 | 247 | sleep_for(0, MSEC_TO_NSEC(200), true); |
fe7e0e64 | 248 | count--; |
eb0af526 | 249 | } |
fe7e0e64 N |
250 | if (err) { |
251 | if (verbose >= 0) | |
e7b84f9d N |
252 | pr_err("failed to stop array %s: %s\n", |
253 | devname, strerror(errno)); | |
bccd8153 JS |
254 | rv = 1; |
255 | goto out; | |
682c7051 | 256 | } |
fe7e0e64 N |
257 | |
258 | /* Give monitor a chance to act */ | |
259 | ping_monitor(mdi->text_version); | |
260 | ||
261 | fd = open_dev_excl(devnm); | |
262 | if (fd < 0) { | |
263 | if (verbose >= 0) | |
7a862a02 | 264 | pr_err("failed to completely stop %s: Device is busy\n", |
fe7e0e64 N |
265 | devname); |
266 | rv = 1; | |
267 | goto out; | |
268 | } | |
269 | } else if (mdi && | |
270 | mdi->array.major_version == -1 && | |
271 | mdi->array.minor_version == -2 && | |
272 | !is_subarray(mdi->text_version)) { | |
273 | struct mdstat_ent *mds, *m; | |
274 | /* container, possibly mdmon-managed. | |
275 | * Make sure mdmon isn't opening it, which | |
276 | * would interfere with the 'stop' | |
97590376 | 277 | */ |
fe7e0e64 | 278 | ping_monitor(mdi->sys_name); |
daf7a3ce | 279 | |
fe7e0e64 N |
280 | /* now check that there are no existing arrays |
281 | * which are members of this array | |
282 | */ | |
283 | mds = mdstat_read(0, 0); | |
284 | for (m = mds; m; m = m->next) | |
285 | if (m->metadata_version && | |
286 | strncmp(m->metadata_version, "external:", 9)==0 && | |
287 | metadata_container_matches(m->metadata_version+9, | |
288 | devnm)) { | |
289 | if (verbose >= 0) | |
7a862a02 | 290 | pr_err("Cannot stop container %s: member %s still active\n", |
9581efb1 | 291 | devname, m->devnm); |
fe7e0e64 N |
292 | free_mdstat(mds); |
293 | rv = 1; | |
294 | goto out; | |
295 | } | |
296 | } | |
297 | ||
2eba8496 N |
298 | /* If the array is undergoing a reshape which changes the number |
299 | * of devices, then it would be nice to stop it at a point where | |
300 | * it has completed a full number of stripes in both old and | |
301 | * new layouts as this will allow the reshape to be reverted. | |
302 | * So if 'sync_action' is "reshape" and 'raid_disks' shows two | |
303 | * different numbers, then | |
304 | * - freeze reshape | |
305 | * - set sync_max to next multiple of both data_disks and | |
306 | * chunk sizes (or next but one) | |
307 | * - unfreeze reshape | |
308 | * - wait on 'sync_completed' for that point to be reached. | |
309 | */ | |
70f1ff42 | 310 | if (mdi && is_level456(mdi->array.level) && |
2eba8496 N |
311 | sysfs_attribute_available(mdi, NULL, "sync_action") && |
312 | sysfs_attribute_available(mdi, NULL, "reshape_direction") && | |
313 | sysfs_get_str(mdi, NULL, "sync_action", buf, 20) > 0 && | |
314 | strcmp(buf, "reshape\n") == 0 && | |
e3e0d0a8 | 315 | sysfs_get_two(mdi, NULL, "raid_disks", &rd1, &rd2) == 2) { |
2eba8496 N |
316 | unsigned long long position, curr; |
317 | unsigned long long chunk1, chunk2; | |
318 | unsigned long long rddiv, chunkdiv; | |
319 | unsigned long long sectors; | |
5509dc44 N |
320 | unsigned long long sync_max, old_sync_max; |
321 | unsigned long long completed; | |
2eba8496 N |
322 | int backwards = 0; |
323 | int delay; | |
324 | int scfd; | |
325 | ||
e3e0d0a8 N |
326 | delay = 40; |
327 | while (rd1 > rd2 && delay > 0 && | |
328 | sysfs_get_ll(mdi, NULL, "sync_max", &old_sync_max) == 0) { | |
329 | /* must be in the critical section - wait a bit */ | |
330 | delay -= 1; | |
239b3cc0 | 331 | sleep_for(0, MSEC_TO_NSEC(100), true); |
e3e0d0a8 N |
332 | } |
333 | ||
334 | if (sysfs_set_str(mdi, NULL, "sync_action", "frozen") != 0) | |
335 | goto done; | |
336 | /* Array is frozen */ | |
337 | ||
2eba8496 N |
338 | rd1 -= mdi->array.level == 6 ? 2 : 1; |
339 | rd2 -= mdi->array.level == 6 ? 2 : 1; | |
340 | sysfs_get_str(mdi, NULL, "reshape_direction", buf, sizeof(buf)); | |
341 | if (strncmp(buf, "back", 4) == 0) | |
342 | backwards = 1; | |
e3e0d0a8 N |
343 | if (sysfs_get_ll(mdi, NULL, "reshape_position", &position) != 0) { |
344 | /* reshape must have finished now */ | |
345 | sysfs_set_str(mdi, NULL, "sync_action", "idle"); | |
346 | goto done; | |
347 | } | |
2eba8496 N |
348 | sysfs_get_two(mdi, NULL, "chunk_size", &chunk1, &chunk2); |
349 | chunk1 /= 512; | |
350 | chunk2 /= 512; | |
351 | rddiv = GCD(rd1, rd2); | |
352 | chunkdiv = GCD(chunk1, chunk2); | |
353 | sectors = (chunk1/chunkdiv) * chunk2 * (rd1/rddiv) * rd2; | |
354 | ||
355 | if (backwards) { | |
356 | /* Need to subtract 'reshape_position' from | |
357 | * array size to get equivalent of sync_max. | |
358 | * Size calculation based on raid5_size in kernel. | |
359 | */ | |
360 | unsigned long long size = mdi->component_size; | |
361 | size &= ~(chunk1-1); | |
362 | size &= ~(chunk2-1); | |
363 | /* rd1 must be smaller */ | |
932be627 N |
364 | /* Reshape may have progressed further backwards than |
365 | * recorded, so target even further back (hence "-1") | |
366 | */ | |
5509dc44 | 367 | position = (position / sectors - 1) * sectors; |
932be627 N |
368 | /* rd1 is always the conversion factor between 'sync' |
369 | * position and 'reshape' position. | |
370 | * We read 1 "new" stripe worth of data from where-ever, | |
371 | * and when write out that full stripe. | |
372 | */ | |
5509dc44 | 373 | sync_max = size - position/rd1; |
2eba8496 | 374 | } else { |
932be627 N |
375 | /* Reshape will very likely be beyond position, and it may |
376 | * be too late to stop at '+1', so aim for '+2' | |
377 | */ | |
5509dc44 N |
378 | position = (position / sectors + 2) * sectors; |
379 | sync_max = position/rd1; | |
2eba8496 | 380 | } |
5509dc44 N |
381 | if (sysfs_get_ll(mdi, NULL, "sync_max", &old_sync_max) < 0) |
382 | old_sync_max = mdi->component_size; | |
383 | /* Must not advance sync_max as that could confuse | |
384 | * the reshape monitor */ | |
385 | if (sync_max < old_sync_max) | |
386 | sysfs_set_num(mdi, NULL, "sync_max", sync_max); | |
2eba8496 N |
387 | sysfs_set_str(mdi, NULL, "sync_action", "idle"); |
388 | ||
389 | /* That should have set things going again. Now we | |
3afaff93 | 390 | * wait a little while (3 second max) for sync_completed |
2eba8496 | 391 | * to reach the target. |
3afaff93 N |
392 | * The reshape process can block for 500msec if |
393 | * the sync speed limit is hit, so we need to wait | |
394 | * a lot longer than that. 1 second is usually | |
395 | * enough. 3 is safe. | |
2eba8496 | 396 | */ |
3afaff93 | 397 | delay = 3000; |
2eba8496 | 398 | scfd = sysfs_open(mdi->sys_name, NULL, "sync_completed"); |
3afaff93 | 399 | while (scfd >= 0 && delay > 0 && old_sync_max > 0) { |
30ddba7d | 400 | unsigned long long max_completed; |
5509dc44 | 401 | sysfs_get_ll(mdi, NULL, "reshape_position", &curr); |
2eba8496 | 402 | sysfs_fd_get_str(scfd, buf, sizeof(buf)); |
3afaff93 N |
403 | if (strncmp(buf, "none", 4) == 0) { |
404 | /* Either reshape has aborted, or hasn't | |
405 | * quite started yet. Wait a bit and | |
406 | * check 'sync_action' to see. | |
407 | */ | |
239b3cc0 | 408 | sleep_for(0, MSEC_TO_NSEC(10), true); |
3afaff93 N |
409 | sysfs_get_str(mdi, NULL, "sync_action", buf, sizeof(buf)); |
410 | if (strncmp(buf, "reshape", 7) != 0) | |
411 | break; | |
412 | } | |
5509dc44 | 413 | |
30ddba7d N |
414 | if (sysfs_fd_get_two(scfd, &completed, |
415 | &max_completed) == 2 && | |
416 | /* 'completed' sometimes reads as max-uulong */ | |
417 | completed < max_completed && | |
5509dc44 N |
418 | (completed > sync_max || |
419 | (completed == sync_max && curr != position))) { | |
420 | while (completed > sync_max) { | |
421 | sync_max += sectors / rd1; | |
422 | if (backwards) | |
423 | position -= sectors; | |
424 | else | |
425 | position += sectors; | |
426 | } | |
427 | if (sync_max < old_sync_max) | |
428 | sysfs_set_num(mdi, NULL, "sync_max", sync_max); | |
429 | } | |
430 | ||
2eba8496 N |
431 | if (!backwards && curr >= position) |
432 | break; | |
433 | if (backwards && curr <= position) | |
434 | break; | |
435 | sysfs_wait(scfd, &delay); | |
436 | } | |
437 | if (scfd >= 0) | |
438 | close(scfd); | |
439 | ||
440 | } | |
e3e0d0a8 | 441 | done: |
2eba8496 | 442 | |
fe7e0e64 N |
443 | /* As we have an O_EXCL open, any use of the device |
444 | * which blocks STOP_ARRAY is probably a transient use, | |
445 | * so it is reasonable to retry for a while - 5 seconds. | |
446 | */ | |
447 | count = 25; err = 0; | |
fc54fe7a JS |
448 | while (count && fd >= 0 && |
449 | (err = ioctl(fd, STOP_ARRAY, NULL)) < 0 && errno == EBUSY) { | |
239b3cc0 | 450 | sleep_for(0, MSEC_TO_NSEC(200), true); |
fe7e0e64 N |
451 | count --; |
452 | } | |
453 | if (fd >= 0 && err) { | |
454 | if (verbose >= 0) { | |
455 | pr_err("failed to stop array %s: %s\n", | |
456 | devname, strerror(errno)); | |
457 | if (errno == EBUSY) | |
7a862a02 | 458 | cont_err("Perhaps a running process, mounted filesystem or active volume group?\n"); |
4ccad7b1 | 459 | } |
fe7e0e64 N |
460 | rv = 1; |
461 | goto out; | |
462 | } | |
229e66cb JS |
463 | |
464 | if (get_linux_version() < 2006028) { | |
465 | /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array | |
466 | * was stopped, so We'll do it here just to be sure. Drop any | |
467 | * partitions as well... | |
468 | */ | |
469 | if (fd >= 0) | |
470 | ioctl(fd, BLKRRPART, 0); | |
471 | if (mdi) | |
472 | sysfs_uevent(mdi, "change"); | |
473 | } | |
4ccad7b1 | 474 | |
fe7e0e64 N |
475 | if (devnm[0] && use_udev()) { |
476 | struct map_ent *mp = map_by_devnm(&map, devnm); | |
477 | remove_devices(devnm, mp ? mp->path : NULL); | |
682c7051 | 478 | } |
fe7e0e64 N |
479 | |
480 | if (verbose >= 0) | |
481 | pr_err("stopped %s\n", devname); | |
482 | map_lock(&map); | |
483 | map_remove(&map, devnm); | |
484 | map_unlock(&map); | |
485 | out: | |
6e8d27e7 | 486 | sysfs_free(mdi); |
fe7e0e64 | 487 | |
bccd8153 | 488 | return rv; |
64c4757e NB |
489 | } |
490 | ||
5e73b024 N |
491 | static struct mddev_dev *add_one(struct mddev_dev *dv, char *name, char disp) |
492 | { | |
493 | struct mddev_dev *new; | |
494 | new = xmalloc(sizeof(*new)); | |
495 | memset(new, 0, sizeof(*new)); | |
496 | new->devname = xstrdup(name); | |
497 | new->disposition = disp; | |
498 | new->next = dv->next; | |
499 | dv->next = new; | |
500 | return new; | |
501 | } | |
502 | ||
1d997643 N |
503 | static void add_faulty(struct mddev_dev *dv, int fd, char disp) |
504 | { | |
505 | mdu_array_info_t array; | |
506 | mdu_disk_info_t disk; | |
507 | int remaining_disks; | |
508 | int i; | |
509 | ||
9cd39f01 | 510 | if (md_get_array_info(fd, &array) != 0) |
1d997643 N |
511 | return; |
512 | ||
513 | remaining_disks = array.nr_disks; | |
514 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
1d997643 N |
515 | char buf[40]; |
516 | disk.number = i; | |
d97572f5 | 517 | if (md_get_disk_info(fd, &disk) != 0) |
1d997643 N |
518 | continue; |
519 | if (disk.major == 0 && disk.minor == 0) | |
520 | continue; | |
521 | remaining_disks--; | |
522 | if ((disk.state & 1) == 0) /* not faulty */ | |
523 | continue; | |
524 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
5e73b024 | 525 | dv = add_one(dv, buf, disp); |
1d997643 N |
526 | } |
527 | } | |
528 | ||
529 | static void add_detached(struct mddev_dev *dv, int fd, char disp) | |
530 | { | |
531 | mdu_array_info_t array; | |
532 | mdu_disk_info_t disk; | |
533 | int remaining_disks; | |
534 | int i; | |
535 | ||
9cd39f01 | 536 | if (md_get_array_info(fd, &array) != 0) |
1d997643 N |
537 | return; |
538 | ||
539 | remaining_disks = array.nr_disks; | |
540 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
1d997643 N |
541 | char buf[40]; |
542 | int sfd; | |
543 | disk.number = i; | |
d97572f5 | 544 | if (md_get_disk_info(fd, &disk) != 0) |
1d997643 N |
545 | continue; |
546 | if (disk.major == 0 && disk.minor == 0) | |
547 | continue; | |
548 | remaining_disks--; | |
549 | if (disp == 'f' && (disk.state & 1) != 0) /* already faulty */ | |
550 | continue; | |
551 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
552 | sfd = dev_open(buf, O_RDONLY); | |
553 | if (sfd >= 0) { | |
554 | /* Not detached */ | |
555 | close(sfd); | |
556 | continue; | |
557 | } | |
558 | if (errno != ENXIO) | |
559 | /* Probably not detached */ | |
560 | continue; | |
5e73b024 | 561 | dv = add_one(dv, buf, disp); |
1d997643 N |
562 | } |
563 | } | |
564 | ||
64a78416 N |
565 | static void add_set(struct mddev_dev *dv, int fd, char set_char) |
566 | { | |
567 | mdu_array_info_t array; | |
568 | mdu_disk_info_t disk; | |
569 | int remaining_disks; | |
570 | int copies, set; | |
571 | int i; | |
572 | ||
9cd39f01 | 573 | if (md_get_array_info(fd, &array) != 0) |
64a78416 N |
574 | return; |
575 | if (array.level != 10) | |
576 | return; | |
577 | copies = ((array.layout & 0xff) * | |
578 | ((array.layout >> 8) & 0xff)); | |
579 | if (array.raid_disks % copies) | |
580 | return; | |
581 | ||
582 | remaining_disks = array.nr_disks; | |
583 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
584 | char buf[40]; | |
585 | disk.number = i; | |
d97572f5 | 586 | if (md_get_disk_info(fd, &disk) != 0) |
64a78416 N |
587 | continue; |
588 | if (disk.major == 0 && disk.minor == 0) | |
589 | continue; | |
590 | remaining_disks--; | |
591 | set = disk.raid_disk % copies; | |
592 | if (set_char != set + 'A') | |
593 | continue; | |
594 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
595 | dv = add_one(dv, buf, dv->disposition); | |
596 | } | |
597 | } | |
598 | ||
abe94694 N |
599 | int attempt_re_add(int fd, int tfd, struct mddev_dev *dv, |
600 | struct supertype *dev_st, struct supertype *tst, | |
f2e8393b MK |
601 | unsigned long rdev, enum update_opt update, |
602 | char *devname, int verbose, mdu_array_info_t *array) | |
abe94694 N |
603 | { |
604 | struct mdinfo mdi; | |
605 | int duuid[4]; | |
606 | int ouuid[4]; | |
607 | ||
608 | dev_st->ss->getinfo_super(dev_st, &mdi, NULL); | |
609 | dev_st->ss->uuid_from_super(dev_st, ouuid); | |
610 | if (tst->sb) | |
611 | tst->ss->uuid_from_super(tst, duuid); | |
612 | else | |
613 | /* Assume uuid matches: kernel will check */ | |
614 | memcpy(duuid, ouuid, sizeof(ouuid)); | |
615 | if ((mdi.disk.state & (1<<MD_DISK_ACTIVE)) && | |
616 | !(mdi.disk.state & (1<<MD_DISK_FAULTY)) && | |
617 | memcmp(duuid, ouuid, sizeof(ouuid))==0) { | |
618 | /* Looks like it is worth a | |
619 | * try. Need to make sure | |
620 | * kernel will accept it | |
621 | * though. | |
622 | */ | |
623 | mdu_disk_info_t disc; | |
624 | /* re-add doesn't work for version-1 superblocks | |
625 | * before 2.6.18 :-( | |
626 | */ | |
627 | if (array->major_version == 1 && | |
628 | get_linux_version() <= 2006018) | |
629 | goto skip_re_add; | |
630 | disc.number = mdi.disk.number; | |
d97572f5 JS |
631 | if (md_get_disk_info(fd, &disc) != 0 || |
632 | disc.major != 0 || disc.minor != 0) | |
abe94694 N |
633 | goto skip_re_add; |
634 | disc.major = major(rdev); | |
635 | disc.minor = minor(rdev); | |
636 | disc.number = mdi.disk.number; | |
637 | disc.raid_disk = mdi.disk.raid_disk; | |
638 | disc.state = mdi.disk.state; | |
bff96f73 GJ |
639 | if (array->state & (1 << MD_SB_CLUSTERED)) { |
640 | /* extra flags are needed when adding to a cluster as | |
641 | * there are two cases to distinguish | |
642 | */ | |
643 | if (dv->disposition == 'c') | |
644 | disc.state |= (1 << MD_DISK_CANDIDATE); | |
645 | else | |
646 | disc.state |= (1 << MD_DISK_CLUSTER_ADD); | |
647 | } | |
e22fe3ae | 648 | if (dv->writemostly == FlagSet) |
abe94694 | 649 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; |
e22fe3ae | 650 | if (dv->writemostly == FlagClear) |
abe94694 | 651 | disc.state &= ~(1 << MD_DISK_WRITEMOSTLY); |
e22fe3ae | 652 | if (dv->failfast == FlagSet) |
71574efb | 653 | disc.state |= 1 << MD_DISK_FAILFAST; |
e22fe3ae | 654 | if (dv->failfast == FlagClear) |
71574efb | 655 | disc.state &= ~(1 << MD_DISK_FAILFAST); |
abe94694 | 656 | remove_partitions(tfd); |
d7be7d87 JS |
657 | if (update || dv->writemostly != FlagDefault || |
658 | dv->failfast != FlagDefault) { | |
abe94694 N |
659 | int rv = -1; |
660 | tfd = dev_open(dv->devname, O_RDWR); | |
661 | if (tfd < 0) { | |
7a862a02 | 662 | pr_err("failed to open %s for superblock update during re-add\n", dv->devname); |
abe94694 N |
663 | return -1; |
664 | } | |
665 | ||
e22fe3ae | 666 | if (dv->writemostly == FlagSet) |
abe94694 | 667 | rv = dev_st->ss->update_super( |
03312b52 | 668 | dev_st, NULL, UOPT_SPEC_WRITEMOSTLY, |
abe94694 | 669 | devname, verbose, 0, NULL); |
e22fe3ae | 670 | if (dv->writemostly == FlagClear) |
abe94694 | 671 | rv = dev_st->ss->update_super( |
03312b52 | 672 | dev_st, NULL, UOPT_SPEC_READWRITE, |
abe94694 | 673 | devname, verbose, 0, NULL); |
e22fe3ae | 674 | if (dv->failfast == FlagSet) |
71574efb | 675 | rv = dev_st->ss->update_super( |
03312b52 | 676 | dev_st, NULL, UOPT_SPEC_FAILFAST, |
71574efb | 677 | devname, verbose, 0, NULL); |
e22fe3ae | 678 | if (dv->failfast == FlagClear) |
71574efb | 679 | rv = dev_st->ss->update_super( |
03312b52 | 680 | dev_st, NULL, UOPT_SPEC_NOFAILFAST, |
71574efb | 681 | devname, verbose, 0, NULL); |
abe94694 N |
682 | if (update) |
683 | rv = dev_st->ss->update_super( | |
f2e8393b | 684 | dev_st, NULL, update, |
abe94694 N |
685 | devname, verbose, 0, NULL); |
686 | if (rv == 0) | |
687 | rv = dev_st->ss->store_super(dev_st, tfd); | |
688 | close(tfd); | |
689 | if (rv != 0) { | |
7a862a02 | 690 | pr_err("failed to update superblock during re-add\n"); |
abe94694 N |
691 | return -1; |
692 | } | |
693 | } | |
694 | /* don't even try if disk is marked as faulty */ | |
695 | errno = 0; | |
696 | if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) { | |
697 | if (verbose >= 0) | |
698 | pr_err("re-added %s\n", dv->devname); | |
699 | return 1; | |
700 | } | |
701 | if (errno == ENOMEM || errno == EROFS) { | |
702 | pr_err("add new device failed for %s: %s\n", | |
703 | dv->devname, strerror(errno)); | |
704 | if (dv->disposition == 'M') | |
705 | return 0; | |
706 | return -1; | |
707 | } | |
708 | } | |
709 | skip_re_add: | |
710 | return 0; | |
711 | } | |
712 | ||
38aeaf3a N |
713 | int Manage_add(int fd, int tfd, struct mddev_dev *dv, |
714 | struct supertype *tst, mdu_array_info_t *array, | |
715 | int force, int verbose, char *devname, | |
f2e8393b MK |
716 | enum update_opt update, unsigned long rdev, |
717 | unsigned long long array_size, int raid_slot) | |
38aeaf3a N |
718 | { |
719 | unsigned long long ldsize; | |
fbd3e15c | 720 | struct supertype *dev_st; |
38aeaf3a N |
721 | int j; |
722 | mdu_disk_info_t disc; | |
723 | ||
724 | if (!get_dev_size(tfd, dv->devname, &ldsize)) { | |
725 | if (dv->disposition == 'M') | |
726 | return 0; | |
727 | else | |
728 | return -1; | |
729 | } | |
730 | ||
632dc30c | 731 | if (tst->ss == &super0 && ldsize > 4ULL*1024*1024*1024*1024) { |
7ccc4cc4 | 732 | /* More than 4TB is wasted on v0.90 */ |
38aeaf3a | 733 | if (!force) { |
7a862a02 N |
734 | pr_err("%s is larger than %s can effectively use.\n" |
735 | " Add --force is you really want to add this device.\n", | |
38aeaf3a N |
736 | dv->devname, devname); |
737 | return -1; | |
738 | } | |
7a862a02 N |
739 | pr_err("%s is larger than %s can effectively use.\n" |
740 | " Adding anyway as --force was given.\n", | |
38aeaf3a N |
741 | dv->devname, devname); |
742 | } | |
38aeaf3a N |
743 | |
744 | if (array->not_persistent == 0 || tst->ss->external) { | |
745 | ||
746 | /* need to find a sample superblock to copy, and | |
747 | * a spare slot to use. | |
748 | * For 'external' array (well, container based), | |
749 | * We can just load the metadata for the array-> | |
750 | */ | |
751 | int array_failed; | |
752 | if (tst->sb) | |
753 | /* already loaded */; | |
754 | else if (tst->ss->external) { | |
755 | tst->ss->load_container(tst, fd, NULL); | |
756 | } else for (j = 0; j < tst->max_devs; j++) { | |
757 | char *dev; | |
758 | int dfd; | |
759 | disc.number = j; | |
d97572f5 | 760 | if (md_get_disk_info(fd, &disc)) |
38aeaf3a N |
761 | continue; |
762 | if (disc.major==0 && disc.minor==0) | |
763 | continue; | |
764 | if ((disc.state & 4)==0) /* sync */ | |
765 | continue; | |
766 | /* Looks like a good device to try */ | |
767 | dev = map_dev(disc.major, disc.minor, 1); | |
768 | if (!dev) | |
769 | continue; | |
770 | dfd = dev_open(dev, O_RDONLY); | |
771 | if (dfd < 0) | |
772 | continue; | |
773 | if (tst->ss->load_super(tst, dfd, | |
774 | NULL)) { | |
775 | close(dfd); | |
776 | continue; | |
777 | } | |
778 | close(dfd); | |
779 | break; | |
780 | } | |
781 | /* FIXME this is a bad test to be using */ | |
fc54fe7a JS |
782 | if (!tst->sb && (dv->disposition != 'a' && |
783 | dv->disposition != 'S')) { | |
38aeaf3a N |
784 | /* we are re-adding a device to a |
785 | * completely dead array - have to depend | |
786 | * on kernel to check | |
787 | */ | |
788 | } else if (!tst->sb) { | |
789 | pr_err("cannot load array metadata from %s\n", devname); | |
790 | return -1; | |
791 | } | |
792 | ||
793 | /* Make sure device is large enough */ | |
01290056 SL |
794 | if (dv->disposition != 'j' && /* skip size check for Journal */ |
795 | tst->sb && | |
2609f339 | 796 | tst->ss->avail_size(tst, ldsize/512, INVALID_SECTORS) < |
38aeaf3a N |
797 | array_size) { |
798 | if (dv->disposition == 'M') | |
799 | return 0; | |
800 | pr_err("%s not large enough to join array\n", | |
801 | dv->devname); | |
802 | return -1; | |
803 | } | |
804 | ||
805 | /* Possibly this device was recently part of | |
806 | * the array and was temporarily removed, and | |
807 | * is now being re-added. If so, we can | |
808 | * simply re-add it. | |
809 | */ | |
810 | ||
fbd3e15c | 811 | if (array->not_persistent == 0) { |
38aeaf3a N |
812 | dev_st = dup_super(tst); |
813 | dev_st->ss->load_super(dev_st, tfd, NULL); | |
fbd3e15c JS |
814 | if (dev_st->sb && dv->disposition != 'S') { |
815 | int rv; | |
816 | ||
817 | rv = attempt_re_add(fd, tfd, dv, dev_st, tst, | |
818 | rdev, update, devname, | |
819 | verbose, array); | |
820 | dev_st->ss->free_super(dev_st); | |
821 | if (rv) | |
822 | return rv; | |
823 | } | |
38aeaf3a N |
824 | } |
825 | if (dv->disposition == 'M') { | |
826 | if (verbose > 0) | |
827 | pr_err("--re-add for %s to %s is not possible\n", | |
828 | dv->devname, devname); | |
829 | return 0; | |
830 | } | |
831 | if (dv->disposition == 'A') { | |
832 | pr_err("--re-add for %s to %s is not possible\n", | |
833 | dv->devname, devname); | |
834 | return -1; | |
835 | } | |
836 | if (array->active_disks < array->raid_disks) { | |
837 | char *avail = xcalloc(array->raid_disks, 1); | |
838 | int d; | |
839 | int found = 0; | |
840 | ||
d180d2aa | 841 | for (d = 0; d < MAX_DISKS && found < array->nr_disks; d++) { |
38aeaf3a | 842 | disc.number = d; |
d97572f5 | 843 | if (md_get_disk_info(fd, &disc)) |
38aeaf3a N |
844 | continue; |
845 | if (disc.major == 0 && disc.minor == 0) | |
846 | continue; | |
847 | if (!(disc.state & (1<<MD_DISK_SYNC))) | |
848 | continue; | |
849 | avail[disc.raid_disk] = 1; | |
d31d0f52 | 850 | found++; |
38aeaf3a N |
851 | } |
852 | array_failed = !enough(array->level, array->raid_disks, | |
853 | array->layout, 1, avail); | |
6157951f | 854 | free(avail); |
38aeaf3a N |
855 | } else |
856 | array_failed = 0; | |
857 | if (array_failed) { | |
858 | pr_err("%s has failed so using --add cannot work and might destroy\n", | |
859 | devname); | |
860 | pr_err("data on %s. You should stop the array and re-assemble it.\n", | |
861 | dv->devname); | |
862 | return -1; | |
863 | } | |
864 | } else { | |
865 | /* non-persistent. Must ensure that new drive | |
866 | * is at least array->size big. | |
867 | */ | |
868 | if (ldsize/512 < array_size) { | |
869 | pr_err("%s not large enough to join array\n", | |
870 | dv->devname); | |
871 | return -1; | |
872 | } | |
873 | } | |
874 | /* committed to really trying this device now*/ | |
875 | remove_partitions(tfd); | |
876 | ||
877 | /* in 2.6.17 and earlier, version-1 superblocks won't | |
878 | * use the number we write, but will choose a free number. | |
879 | * we must choose the same free number, which requires | |
880 | * starting at 'raid_disks' and counting up | |
881 | */ | |
882 | for (j = array->raid_disks; j < tst->max_devs; j++) { | |
883 | disc.number = j; | |
d97572f5 | 884 | if (md_get_disk_info(fd, &disc)) |
38aeaf3a N |
885 | break; |
886 | if (disc.major==0 && disc.minor==0) | |
887 | break; | |
888 | if (disc.state & 8) /* removed */ | |
889 | break; | |
890 | } | |
891 | disc.major = major(rdev); | |
892 | disc.minor = minor(rdev); | |
4de90913 GJ |
893 | if (raid_slot < 0) |
894 | disc.number = j; | |
895 | else | |
896 | disc.number = raid_slot; | |
38aeaf3a | 897 | disc.state = 0; |
01290056 SL |
898 | |
899 | /* only add journal to array that supports journaling */ | |
900 | if (dv->disposition == 'j') { | |
01290056 SL |
901 | struct mdinfo *mdp; |
902 | ||
903 | mdp = sysfs_read(fd, NULL, GET_ARRAY_STATE); | |
2a1990c0 JS |
904 | if (!mdp) { |
905 | pr_err("%s unable to read array state.\n", devname); | |
906 | return -1; | |
907 | } | |
01290056 | 908 | |
5e4ca8bb | 909 | if (mdp->array_state != ARRAY_READONLY) { |
d209181d | 910 | sysfs_free(mdp); |
01290056 SL |
911 | pr_err("%s is not readonly, cannot add journal.\n", devname); |
912 | return -1; | |
913 | } | |
914 | ||
d209181d JS |
915 | sysfs_free(mdp); |
916 | ||
38c2e05b | 917 | disc.raid_disk = 0; |
01290056 SL |
918 | } |
919 | ||
38aeaf3a N |
920 | if (array->not_persistent==0) { |
921 | int dfd; | |
01290056 SL |
922 | if (dv->disposition == 'j') |
923 | disc.state |= (1 << MD_DISK_JOURNAL) | (1 << MD_DISK_SYNC); | |
e22fe3ae | 924 | if (dv->writemostly == FlagSet) |
38aeaf3a | 925 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; |
e22fe3ae | 926 | if (dv->failfast == FlagSet) |
71574efb | 927 | disc.state |= 1 << MD_DISK_FAILFAST; |
38aeaf3a N |
928 | dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); |
929 | if (tst->ss->add_to_super(tst, &disc, dfd, | |
72ca9bcf | 930 | dv->devname, INVALID_SECTORS)) |
38aeaf3a N |
931 | return -1; |
932 | if (tst->ss->write_init_super(tst)) | |
933 | return -1; | |
934 | } else if (dv->disposition == 'A') { | |
935 | /* this had better be raid1. | |
936 | * As we are "--re-add"ing we must find a spare slot | |
937 | * to fill. | |
938 | */ | |
939 | char *used = xcalloc(array->raid_disks, 1); | |
940 | for (j = 0; j < tst->max_devs; j++) { | |
941 | mdu_disk_info_t disc2; | |
942 | disc2.number = j; | |
d97572f5 | 943 | if (md_get_disk_info(fd, &disc2)) |
38aeaf3a N |
944 | continue; |
945 | if (disc2.major==0 && disc2.minor==0) | |
946 | continue; | |
947 | if (disc2.state & 8) /* removed */ | |
948 | continue; | |
949 | if (disc2.raid_disk < 0) | |
950 | continue; | |
951 | if (disc2.raid_disk > array->raid_disks) | |
952 | continue; | |
953 | used[disc2.raid_disk] = 1; | |
954 | } | |
955 | for (j = 0 ; j < array->raid_disks; j++) | |
956 | if (!used[j]) { | |
957 | disc.raid_disk = j; | |
958 | disc.state |= (1<<MD_DISK_SYNC); | |
959 | break; | |
960 | } | |
961 | free(used); | |
962 | } | |
4de90913 GJ |
963 | |
964 | if (array->state & (1 << MD_SB_CLUSTERED)) { | |
965 | if (dv->disposition == 'c') | |
966 | disc.state |= (1 << MD_DISK_CANDIDATE); | |
967 | else | |
968 | disc.state |= (1 << MD_DISK_CLUSTER_ADD); | |
969 | } | |
970 | ||
e22fe3ae | 971 | if (dv->writemostly == FlagSet) |
38aeaf3a | 972 | disc.state |= (1 << MD_DISK_WRITEMOSTLY); |
e22fe3ae | 973 | if (dv->failfast == FlagSet) |
71574efb | 974 | disc.state |= (1 << MD_DISK_FAILFAST); |
38aeaf3a N |
975 | if (tst->ss->external) { |
976 | /* add a disk | |
977 | * to an external metadata container */ | |
978 | struct mdinfo new_mdi; | |
979 | struct mdinfo *sra; | |
980 | int container_fd; | |
4dd2df09 | 981 | char devnm[32]; |
38aeaf3a N |
982 | int dfd; |
983 | ||
4dd2df09 N |
984 | strcpy(devnm, fd2devnm(fd)); |
985 | ||
986 | container_fd = open_dev_excl(devnm); | |
38aeaf3a | 987 | if (container_fd < 0) { |
7a862a02 | 988 | pr_err("add failed for %s: could not get exclusive access to container\n", |
38aeaf3a N |
989 | dv->devname); |
990 | tst->ss->free_super(tst); | |
991 | return -1; | |
992 | } | |
993 | ||
1f5d54a0 MT |
994 | /* Check if metadata handler is able to accept the drive */ |
995 | if (!tst->ss->validate_geometry(tst, LEVEL_CONTAINER, 0, 1, NULL, | |
996 | 0, 0, dv->devname, NULL, 0, 1)) { | |
997 | close(container_fd); | |
998 | return -1; | |
999 | } | |
1000 | ||
9cf9a1de | 1001 | Kill(dv->devname, NULL, 0, -1, 0); |
38aeaf3a | 1002 | dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); |
38aeaf3a | 1003 | if (tst->ss->add_to_super(tst, &disc, dfd, |
72ca9bcf | 1004 | dv->devname, INVALID_SECTORS)) { |
38aeaf3a N |
1005 | close(dfd); |
1006 | close(container_fd); | |
1007 | return -1; | |
1008 | } | |
12724c01 | 1009 | if (!mdmon_running(tst->container_devnm)) |
38aeaf3a N |
1010 | tst->ss->sync_metadata(tst); |
1011 | ||
4dd2df09 | 1012 | sra = sysfs_read(container_fd, NULL, 0); |
38aeaf3a N |
1013 | if (!sra) { |
1014 | pr_err("add failed for %s: sysfs_read failed\n", | |
1015 | dv->devname); | |
1016 | close(container_fd); | |
1017 | tst->ss->free_super(tst); | |
1018 | return -1; | |
1019 | } | |
1020 | sra->array.level = LEVEL_CONTAINER; | |
1021 | /* Need to set data_offset and component_size */ | |
1022 | tst->ss->getinfo_super(tst, &new_mdi, NULL); | |
1023 | new_mdi.disk.major = disc.major; | |
1024 | new_mdi.disk.minor = disc.minor; | |
1025 | new_mdi.recovery_start = 0; | |
1026 | /* Make sure fds are closed as they are O_EXCL which | |
1027 | * would block add_disk */ | |
1028 | tst->ss->free_super(tst); | |
1029 | if (sysfs_add_disk(sra, &new_mdi, 0) != 0) { | |
7a862a02 | 1030 | pr_err("add new device to external metadata failed for %s\n", dv->devname); |
38aeaf3a N |
1031 | close(container_fd); |
1032 | sysfs_free(sra); | |
1033 | return -1; | |
1034 | } | |
4dd2df09 | 1035 | ping_monitor(devnm); |
38aeaf3a N |
1036 | sysfs_free(sra); |
1037 | close(container_fd); | |
1038 | } else { | |
1039 | tst->ss->free_super(tst); | |
1040 | if (ioctl(fd, ADD_NEW_DISK, &disc)) { | |
01290056 SL |
1041 | if (dv->disposition == 'j') |
1042 | pr_err("Failed to hot add %s as journal, " | |
1043 | "please try restart %s.\n", dv->devname, devname); | |
1044 | else | |
1045 | pr_err("add new device failed for %s as %d: %s\n", | |
1046 | dv->devname, j, strerror(errno)); | |
38aeaf3a N |
1047 | return -1; |
1048 | } | |
01290056 SL |
1049 | if (dv->disposition == 'j') { |
1050 | pr_err("Journal added successfully, making %s read-write\n", devname); | |
1051 | if (Manage_ro(devname, fd, -1)) | |
1052 | pr_err("Failed to make %s read-write\n", devname); | |
1053 | } | |
1054 | ||
38aeaf3a N |
1055 | } |
1056 | if (verbose >= 0) | |
1057 | pr_err("added %s\n", dv->devname); | |
1058 | return 1; | |
1059 | } | |
1060 | ||
d070235d | 1061 | int Manage_remove(struct supertype *tst, int fd, struct mddev_dev *dv, |
1ab9ed2a | 1062 | int sysfd, unsigned long rdev, int force, int verbose, char *devname) |
d070235d N |
1063 | { |
1064 | int lfd = -1; | |
1065 | int err; | |
1066 | ||
1067 | if (tst->ss->external) { | |
1068 | /* To remove a device from a container, we must | |
1069 | * check that it isn't in use in an array. | |
1070 | * This involves looking in the 'holders' | |
1071 | * directory - there must be just one entry, | |
1072 | * the container. | |
1073 | * To ensure that it doesn't get used as a | |
1074 | * hot spare while we are checking, we | |
1075 | * get an O_EXCL open on the container | |
1076 | */ | |
aab15415 | 1077 | int ret; |
4dd2df09 N |
1078 | char devnm[32]; |
1079 | strcpy(devnm, fd2devnm(fd)); | |
1080 | lfd = open_dev_excl(devnm); | |
d070235d | 1081 | if (lfd < 0) { |
7a862a02 | 1082 | pr_err("Cannot get exclusive access to container - odd\n"); |
d070235d N |
1083 | return -1; |
1084 | } | |
aab15415 N |
1085 | /* We may not be able to check on holders in |
1086 | * sysfs, either because we don't have the dev num | |
1087 | * (rdev == 0) or because the device has been detached | |
1088 | * and the 'holders' directory no longer exists | |
1089 | * (ret == -1). In that case, assume it is OK to | |
1090 | * remove. | |
d070235d | 1091 | */ |
aab15415 N |
1092 | if (rdev == 0) |
1093 | ret = -1; | |
c922221e TM |
1094 | else { |
1095 | /* | |
1096 | * The drive has already been set to 'faulty', however | |
1097 | * monitor might not have had time to process it and the | |
1098 | * drive might still have an entry in the 'holders' | |
1099 | * directory. Try a few times to avoid a false error | |
1100 | */ | |
1101 | int count = 20; | |
1102 | ||
1103 | do { | |
1104 | ret = sysfs_unique_holder(devnm, rdev); | |
1105 | if (ret < 2) | |
1106 | break; | |
239b3cc0 | 1107 | sleep_for(0, MSEC_TO_NSEC(100), true); |
c922221e TM |
1108 | } while (--count > 0); |
1109 | ||
1110 | if (ret == 0) { | |
1111 | pr_err("%s is not a member, cannot remove.\n", | |
1112 | dv->devname); | |
1113 | close(lfd); | |
1114 | return -1; | |
1115 | } | |
1116 | if (ret >= 2) { | |
1117 | pr_err("%s is still in use, cannot remove.\n", | |
1118 | dv->devname); | |
1119 | close(lfd); | |
1120 | return -1; | |
1121 | } | |
d070235d N |
1122 | } |
1123 | } | |
1124 | /* FIXME check that it is a current member */ | |
1125 | if (sysfd >= 0) { | |
1126 | /* device has been removed and we don't know | |
1127 | * the major:minor number | |
1128 | */ | |
1ab9ed2a | 1129 | err = sys_hot_remove_disk(sysfd, force); |
d070235d | 1130 | } else { |
1ab9ed2a | 1131 | err = hot_remove_disk(fd, rdev, force); |
d070235d N |
1132 | if (err && errno == ENODEV) { |
1133 | /* Old kernels rejected this if no personality | |
1134 | * is registered */ | |
4dd2df09 | 1135 | struct mdinfo *sra = sysfs_read(fd, NULL, GET_DEVS); |
d070235d N |
1136 | struct mdinfo *dv = NULL; |
1137 | if (sra) | |
1138 | dv = sra->devs; | |
1139 | for ( ; dv ; dv=dv->next) | |
1140 | if (dv->disk.major == (int)major(rdev) && | |
1141 | dv->disk.minor == (int)minor(rdev)) | |
1142 | break; | |
1143 | if (dv) | |
1144 | err = sysfs_set_str(sra, dv, | |
1145 | "state", "remove"); | |
1146 | else | |
1147 | err = -1; | |
6e8d27e7 | 1148 | sysfs_free(sra); |
d070235d N |
1149 | } |
1150 | } | |
1151 | if (err) { | |
7a862a02 | 1152 | pr_err("hot remove failed for %s: %s\n", dv->devname, |
d070235d N |
1153 | strerror(errno)); |
1154 | if (lfd >= 0) | |
1155 | close(lfd); | |
1156 | return -1; | |
1157 | } | |
1158 | if (tst->ss->external) { | |
1159 | /* | |
1160 | * Before dropping our exclusive open we make an | |
1161 | * attempt at preventing mdmon from seeing an | |
1162 | * 'add' event before reconciling this 'remove' | |
1163 | * event. | |
1164 | */ | |
4dd2df09 | 1165 | char *devnm = fd2devnm(fd); |
d070235d | 1166 | |
4dd2df09 | 1167 | if (!devnm) { |
d070235d N |
1168 | pr_err("unable to get container name\n"); |
1169 | return -1; | |
1170 | } | |
1171 | ||
4dd2df09 | 1172 | ping_manager(devnm); |
d070235d N |
1173 | } |
1174 | if (lfd >= 0) | |
1175 | close(lfd); | |
1176 | if (verbose >= 0) | |
1177 | pr_err("hot removed %s from %s\n", | |
1178 | dv->devname, devname); | |
1179 | return 1; | |
1180 | } | |
1181 | ||
70c55e36 N |
1182 | int Manage_replace(struct supertype *tst, int fd, struct mddev_dev *dv, |
1183 | unsigned long rdev, int verbose, char *devname) | |
1184 | { | |
1185 | struct mdinfo *mdi, *di; | |
1186 | if (tst->ss->external) { | |
1187 | pr_err("--replace only supported for native metadata (0.90 or 1.x)\n"); | |
1188 | return -1; | |
1189 | } | |
1190 | /* Need to find the device in sysfs and add 'want_replacement' to the | |
1191 | * status. | |
1192 | */ | |
4dd2df09 | 1193 | mdi = sysfs_read(fd, NULL, GET_DEVS); |
70c55e36 N |
1194 | if (!mdi || !mdi->devs) { |
1195 | pr_err("Cannot find status of %s to enable replacement - strange\n", | |
1196 | devname); | |
1197 | return -1; | |
1198 | } | |
1199 | for (di = mdi->devs; di; di = di->next) | |
1200 | if (di->disk.major == (int)major(rdev) && | |
1201 | di->disk.minor == (int)minor(rdev)) | |
1202 | break; | |
1203 | if (di) { | |
1204 | int rv; | |
1205 | if (di->disk.raid_disk < 0) { | |
1206 | pr_err("%s is not active and so cannot be replaced.\n", | |
1207 | dv->devname); | |
1208 | sysfs_free(mdi); | |
1209 | return -1; | |
1210 | } | |
1211 | rv = sysfs_set_str(mdi, di, | |
1212 | "state", "want_replacement"); | |
1213 | if (rv) { | |
1214 | sysfs_free(mdi); | |
1215 | pr_err("Failed to request replacement for %s\n", | |
1216 | dv->devname); | |
1217 | return -1; | |
1218 | } | |
1219 | if (verbose >= 0) | |
1220 | pr_err("Marked %s (device %d in %s) for replacement\n", | |
1221 | dv->devname, di->disk.raid_disk, devname); | |
1222 | /* If there is a matching 'with', we need to tell it which | |
1223 | * raid disk | |
1224 | */ | |
1225 | while (dv && dv->disposition != 'W') | |
1226 | dv = dv->next; | |
1227 | if (dv) { | |
1228 | dv->disposition = 'w'; | |
1229 | dv->used = di->disk.raid_disk; | |
1230 | } | |
1231 | return 1; | |
1232 | } | |
1233 | sysfs_free(mdi); | |
1234 | pr_err("%s not found in %s so cannot --replace it\n", | |
1235 | dv->devname, devname); | |
1236 | return -1; | |
1237 | } | |
1238 | ||
1239 | int Manage_with(struct supertype *tst, int fd, struct mddev_dev *dv, | |
1240 | unsigned long rdev, int verbose, char *devname) | |
1241 | { | |
1242 | struct mdinfo *mdi, *di; | |
1243 | /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */ | |
4dd2df09 | 1244 | mdi = sysfs_read(fd, NULL, GET_DEVS|GET_STATE); |
70c55e36 N |
1245 | if (!mdi || !mdi->devs) { |
1246 | pr_err("Cannot find status of %s to enable replacement - strange\n", | |
1247 | devname); | |
1248 | return -1; | |
1249 | } | |
1250 | for (di = mdi->devs; di; di = di->next) | |
1251 | if (di->disk.major == (int)major(rdev) && | |
1252 | di->disk.minor == (int)minor(rdev)) | |
1253 | break; | |
1254 | if (di) { | |
1255 | int rv; | |
1256 | if (di->disk.state & (1<<MD_DISK_FAULTY)) { | |
1257 | pr_err("%s is faulty and cannot be a replacement\n", | |
1258 | dv->devname); | |
1259 | sysfs_free(mdi); | |
1260 | return -1; | |
1261 | } | |
1262 | if (di->disk.raid_disk >= 0) { | |
1263 | pr_err("%s is active and cannot be a replacement\n", | |
1264 | dv->devname); | |
1265 | sysfs_free(mdi); | |
1266 | return -1; | |
1267 | } | |
1268 | rv = sysfs_set_num(mdi, di, | |
1269 | "slot", dv->used); | |
1270 | if (rv) { | |
1271 | sysfs_free(mdi); | |
51425978 | 1272 | pr_err("Failed to set %s as preferred replacement.\n", |
70c55e36 N |
1273 | dv->devname); |
1274 | return -1; | |
1275 | } | |
1276 | if (verbose >= 0) | |
1277 | pr_err("Marked %s in %s as replacement for device %d\n", | |
1278 | dv->devname, devname, dv->used); | |
1279 | return 1; | |
1280 | } | |
1281 | sysfs_free(mdi); | |
1282 | pr_err("%s not found in %s so cannot make it preferred replacement\n", | |
1283 | dv->devname, devname); | |
1284 | return -1; | |
1285 | } | |
1286 | ||
fc6fd406 MK |
1287 | /** |
1288 | * is_remove_safe() - Check if remove is safe. | |
1289 | * @array: Array info. | |
1290 | * @fd: Array file descriptor. | |
1291 | * @devname: Name of device to remove. | |
1292 | * @verbose: Verbose. | |
1293 | * | |
1294 | * The function determines if array will be operational | |
1295 | * after removing &devname. | |
1296 | * | |
1297 | * Return: True if array will be operational, false otherwise. | |
1298 | */ | |
1299 | bool is_remove_safe(mdu_array_info_t *array, const int fd, char *devname, const int verbose) | |
1300 | { | |
1301 | dev_t devid = devnm2devid(devname + 5); | |
1302 | struct mdinfo *mdi = sysfs_read(fd, NULL, GET_DEVS | GET_DISKS | GET_STATE); | |
1303 | ||
1304 | if (!mdi) { | |
1305 | if (verbose) | |
1306 | pr_err("Failed to read sysfs attributes for %s\n", devname); | |
1307 | return false; | |
1308 | } | |
1309 | ||
1310 | char *avail = xcalloc(array->raid_disks, sizeof(char)); | |
1311 | ||
1312 | for (mdi = mdi->devs; mdi; mdi = mdi->next) { | |
1313 | if (mdi->disk.raid_disk < 0) | |
1314 | continue; | |
1315 | if (!(mdi->disk.state & (1 << MD_DISK_SYNC))) | |
1316 | continue; | |
1317 | if (makedev(mdi->disk.major, mdi->disk.minor) == devid) | |
1318 | continue; | |
1319 | avail[mdi->disk.raid_disk] = 1; | |
1320 | } | |
1321 | sysfs_free(mdi); | |
1322 | ||
1323 | bool is_enough = enough(array->level, array->raid_disks, | |
b3e7b7eb | 1324 | array->layout, 1, avail); |
fc6fd406 MK |
1325 | |
1326 | free(avail); | |
1327 | return is_enough; | |
1328 | } | |
1329 | ||
64c4757e | 1330 | int Manage_subdevs(char *devname, int fd, |
833bb0f8 | 1331 | struct mddev_dev *devlist, int verbose, int test, |
f2e8393b | 1332 | enum update_opt update, int force) |
cd29a5c8 | 1333 | { |
7bd04da9 | 1334 | /* Do something to each dev. |
682c7051 NB |
1335 | * devmode can be |
1336 | * 'a' - add the device | |
f33a71f1 | 1337 | * 'S' - add the device as a spare - don't try re-add |
01290056 | 1338 | * 'j' - add the device as a journal device |
7bd04da9 N |
1339 | * 'A' - re-add the device |
1340 | * 'r' - remove the device: HOT_REMOVE_DISK | |
b80da661 NB |
1341 | * device can be 'faulty' or 'detached' in which case all |
1342 | * matching devices are removed. | |
682c7051 | 1343 | * 'f' - set the device faulty SET_DISK_FAULTY |
b80da661 NB |
1344 | * device can be 'detached' in which case any device that |
1345 | * is inaccessible will be marked faulty. | |
70c55e36 N |
1346 | * 'R' - mark this device as wanting replacement. |
1347 | * 'W' - this device is added if necessary and activated as | |
1348 | * a replacement for a previous 'R' device. | |
1349 | * ----- | |
1350 | * 'w' - 'W' will be changed to 'w' when it is paired with | |
1351 | * a 'R' device. If a 'W' is found while walking the list | |
1352 | * it must be unpaired, and is an error. | |
1353 | * 'M' - this is created by a 'missing' target. It is a slight | |
1354 | * variant on 'A' | |
262e3b7f N |
1355 | * 'F' - Another variant of 'A', where the device was faulty |
1356 | * so must be removed from the array first. | |
4de90913 | 1357 | * 'c' - confirm the device as found (for clustered environments) |
70c55e36 | 1358 | * |
98d27e39 N |
1359 | * For 'f' and 'r', the device can also be a kernel-internal |
1360 | * name such as 'sdb'. | |
682c7051 NB |
1361 | */ |
1362 | mdu_array_info_t array; | |
7a3be72f | 1363 | unsigned long long array_size; |
1d997643 | 1364 | struct mddev_dev *dv; |
cfad27a9 | 1365 | int tfd = -1; |
38aeaf3a | 1366 | struct supertype *tst; |
4725bc31 | 1367 | char *subarray = NULL; |
98d27e39 | 1368 | int sysfd = -1; |
7d2e6486 | 1369 | int count = 0; /* number of actions taken */ |
9f584691 | 1370 | struct mdinfo info; |
9465f170 | 1371 | struct mdinfo devinfo; |
9f584691 | 1372 | int frozen = 0; |
8af530b0 | 1373 | int busy = 0; |
4de90913 | 1374 | int raid_slot = -1; |
682c7051 | 1375 | |
dae13137 JS |
1376 | if (sysfs_init(&info, fd, NULL)) { |
1377 | pr_err("sysfs not availabile for %s\n", devname); | |
1378 | goto abort; | |
1379 | } | |
1380 | ||
9cd39f01 JS |
1381 | if (md_get_array_info(fd, &array)) { |
1382 | pr_err("Cannot get array info for %s\n", devname); | |
bcbb3112 | 1383 | goto abort; |
682c7051 | 1384 | } |
7bd04da9 | 1385 | /* array.size is only 32 bits and may be truncated. |
7a3be72f NB |
1386 | * So read from sysfs if possible, and record number of sectors |
1387 | */ | |
1388 | ||
1389 | array_size = get_component_size(fd); | |
1390 | if (array_size <= 0) | |
1391 | array_size = array.size * 2; | |
1392 | ||
4725bc31 | 1393 | tst = super_by_fd(fd, &subarray); |
3da92f27 | 1394 | if (!tst) { |
e7b84f9d | 1395 | pr_err("unsupport array - version %d.%d\n", |
3da92f27 | 1396 | array.major_version, array.minor_version); |
bcbb3112 | 1397 | goto abort; |
3da92f27 NB |
1398 | } |
1399 | ||
38aeaf3a | 1400 | for (dv = devlist; dv; dv = dv->next) { |
ffaf1a7e | 1401 | dev_t rdev = 0; /* device to add/remove etc */ |
38aeaf3a | 1402 | int rv; |
b47024f1 | 1403 | int mj,mn; |
4a39c6f2 | 1404 | |
4de90913 GJ |
1405 | raid_slot = -1; |
1406 | if (dv->disposition == 'c') { | |
1407 | rv = parse_cluster_confirm_arg(dv->devname, | |
1408 | &dv->devname, | |
1409 | &raid_slot); | |
d7a49369 | 1410 | if (rv) { |
4de90913 GJ |
1411 | pr_err("Could not get the devname of cluster\n"); |
1412 | goto abort; | |
1413 | } | |
1414 | } | |
1415 | ||
1d997643 N |
1416 | if (strcmp(dv->devname, "failed") == 0 || |
1417 | strcmp(dv->devname, "faulty") == 0) { | |
fc54fe7a | 1418 | if (dv->disposition != 'A' && dv->disposition != 'r') { |
7a862a02 | 1419 | pr_err("%s only meaningful with -r or --re-add, not -%c\n", |
b80da661 | 1420 | dv->devname, dv->disposition); |
bcbb3112 | 1421 | goto abort; |
b80da661 | 1422 | } |
262e3b7f N |
1423 | add_faulty(dv, fd, (dv->disposition == 'A' |
1424 | ? 'F' : 'r')); | |
1d997643 N |
1425 | continue; |
1426 | } | |
1427 | if (strcmp(dv->devname, "detached") == 0) { | |
b80da661 | 1428 | if (dv->disposition != 'r' && dv->disposition != 'f') { |
7a862a02 | 1429 | pr_err("%s only meaningful with -r of -f, not -%c\n", |
b80da661 | 1430 | dv->devname, dv->disposition); |
bcbb3112 | 1431 | goto abort; |
b80da661 | 1432 | } |
1d997643 N |
1433 | add_detached(dv, fd, dv->disposition); |
1434 | continue; | |
1435 | } | |
1436 | ||
1437 | if (strcmp(dv->devname, "missing") == 0) { | |
e9ddbb2b | 1438 | struct mddev_dev *add_devlist; |
1d997643 | 1439 | struct mddev_dev **dp; |
4de90913 GJ |
1440 | if (dv->disposition == 'c') { |
1441 | rv = ioctl(fd, CLUSTERED_DISK_NACK, NULL); | |
1442 | break; | |
1443 | } | |
1444 | ||
c8e1a230 | 1445 | if (dv->disposition != 'A') { |
7a862a02 | 1446 | pr_err("'missing' only meaningful with --re-add\n"); |
bcbb3112 | 1447 | goto abort; |
a4e13010 | 1448 | } |
1d997643 | 1449 | add_devlist = conf_get_devs(); |
a4e13010 | 1450 | if (add_devlist == NULL) { |
8e5b52cd | 1451 | pr_err("no devices to scan for missing members.\n"); |
a4e13010 N |
1452 | continue; |
1453 | } | |
1d997643 | 1454 | for (dp = &add_devlist; *dp; dp = & (*dp)->next) |
7bd04da9 | 1455 | /* 'M' (for 'missing') is like 'A' without errors */ |
1d997643 N |
1456 | (*dp)->disposition = 'M'; |
1457 | *dp = dv->next; | |
1458 | dv->next = add_devlist; | |
1459 | continue; | |
1460 | } | |
1461 | ||
64a78416 N |
1462 | if (strncmp(dv->devname, "set-", 4) == 0 && |
1463 | strlen(dv->devname) == 5) { | |
1464 | int copies; | |
1465 | ||
1466 | if (dv->disposition != 'r' && | |
1467 | dv->disposition != 'f') { | |
1468 | pr_err("'%s' only meaningful with -r or -f\n", | |
1469 | dv->devname); | |
1470 | goto abort; | |
1471 | } | |
1472 | if (array.level != 10) { | |
1473 | pr_err("'%s' only meaningful with RAID10 arrays\n", | |
1474 | dv->devname); | |
1475 | goto abort; | |
1476 | } | |
1477 | copies = ((array.layout & 0xff) * | |
1478 | ((array.layout >> 8) & 0xff)); | |
1479 | if (array.raid_disks % copies != 0 || | |
1480 | dv->devname[4] < 'A' || | |
1481 | dv->devname[4] >= 'A' + copies || | |
1482 | copies > 26) { | |
1483 | pr_err("'%s' not meaningful with this array\n", | |
1484 | dv->devname); | |
1485 | goto abort; | |
1486 | } | |
1487 | add_set(dv, fd, dv->devname[4]); | |
1488 | continue; | |
1489 | } | |
1490 | ||
1d997643 | 1491 | if (strchr(dv->devname, '/') == NULL && |
aab15415 N |
1492 | strchr(dv->devname, ':') == NULL && |
1493 | strlen(dv->devname) < 50) { | |
98d27e39 N |
1494 | /* Assume this is a kernel-internal name like 'sda1' */ |
1495 | int found = 0; | |
1496 | char dname[55]; | |
1497 | if (dv->disposition != 'r' && dv->disposition != 'f') { | |
7a862a02 | 1498 | pr_err("%s only meaningful with -r or -f, not -%c\n", |
98d27e39 | 1499 | dv->devname, dv->disposition); |
bcbb3112 | 1500 | goto abort; |
98d27e39 N |
1501 | } |
1502 | ||
1503 | sprintf(dname, "dev-%s", dv->devname); | |
4dd2df09 | 1504 | sysfd = sysfs_open(fd2devnm(fd), dname, "block/dev"); |
98d27e39 N |
1505 | if (sysfd >= 0) { |
1506 | char dn[20]; | |
98d27e39 N |
1507 | if (sysfs_fd_get_str(sysfd, dn, 20) > 0 && |
1508 | sscanf(dn, "%d:%d", &mj,&mn) == 2) { | |
5dffd09d | 1509 | rdev = makedev(mj,mn); |
98d27e39 N |
1510 | found = 1; |
1511 | } | |
1512 | close(sysfd); | |
1513 | sysfd = -1; | |
1514 | } | |
1515 | if (!found) { | |
4dd2df09 | 1516 | sysfd = sysfs_open(fd2devnm(fd), dname, "state"); |
98d27e39 | 1517 | if (sysfd < 0) { |
7a862a02 | 1518 | pr_err("%s does not appear to be a component of %s\n", |
98d27e39 | 1519 | dv->devname, devname); |
bcbb3112 | 1520 | goto abort; |
98d27e39 N |
1521 | } |
1522 | } | |
fc54fe7a JS |
1523 | } else if ((dv->disposition == 'r' || |
1524 | dv->disposition == 'f') && | |
1525 | get_maj_min(dv->devname, &mj, &mn)) { | |
b47024f1 N |
1526 | /* for 'fail' and 'remove', the device might |
1527 | * not exist. | |
1528 | */ | |
1529 | rdev = makedev(mj, mn); | |
b80da661 | 1530 | } else { |
c7b47447 | 1531 | tfd = dev_open(dv->devname, O_RDONLY); |
cc5083d1 | 1532 | if (tfd >= 0) { |
0a6bff09 | 1533 | fstat_is_blkdev(tfd, dv->devname, &rdev); |
cc5083d1 JS |
1534 | close(tfd); |
1535 | } else { | |
5fe7f5f7 | 1536 | int open_err = errno; |
9e04ac1c | 1537 | if (!stat_is_blkdev(dv->devname, &rdev)) { |
5fe7f5f7 N |
1538 | if (dv->disposition == 'M') |
1539 | /* non-fatal. Also improbable */ | |
1540 | continue; | |
5fe7f5f7 N |
1541 | goto abort; |
1542 | } | |
1543 | if (dv->disposition == 'r') | |
1544 | /* Be happy, the stat worked, that is | |
1545 | * enough for --remove | |
1546 | */ | |
1547 | ; | |
1548 | else { | |
1d997643 N |
1549 | if (dv->disposition == 'M') |
1550 | /* non-fatal */ | |
1551 | continue; | |
839f27a3 | 1552 | pr_err("Cannot open %s: %s\n", |
5fe7f5f7 | 1553 | dv->devname, strerror(open_err)); |
bcbb3112 | 1554 | goto abort; |
5a9de8db | 1555 | } |
b80da661 | 1556 | } |
682c7051 | 1557 | } |
cd29a5c8 | 1558 | switch(dv->disposition){ |
682c7051 | 1559 | default: |
e7b84f9d | 1560 | pr_err("internal error - devmode[%s]=%d\n", |
c913b90e | 1561 | dv->devname, dv->disposition); |
bcbb3112 | 1562 | goto abort; |
682c7051 | 1563 | case 'a': |
f33a71f1 | 1564 | case 'S': /* --add-spare */ |
01290056 | 1565 | case 'j': /* --add-journal */ |
c8e1a230 | 1566 | case 'A': |
262e3b7f N |
1567 | case 'M': /* --re-add missing */ |
1568 | case 'F': /* --re-add faulty */ | |
4de90913 | 1569 | case 'c': /* --cluster-confirm */ |
4a39c6f2 | 1570 | /* add the device */ |
4725bc31 | 1571 | if (subarray) { |
7a862a02 | 1572 | pr_err("Cannot add disks to a \'member\' array, perform this operation on the parent container\n"); |
bcbb3112 | 1573 | goto abort; |
f94d52f4 | 1574 | } |
9465f170 GJ |
1575 | |
1576 | /* Let's first try to write re-add to sysfs */ | |
1577 | if (rdev != 0 && | |
1578 | (dv->disposition == 'A' || dv->disposition == 'F')) { | |
1579 | sysfs_init_dev(&devinfo, rdev); | |
1580 | if (sysfs_set_str(&info, &devinfo, "state", "re-add") == 0) { | |
1581 | pr_err("re-add %s to %s succeed\n", | |
1582 | dv->devname, info.sys_name); | |
1583 | break; | |
1584 | } | |
1585 | } | |
1586 | ||
262e3b7f N |
1587 | if (dv->disposition == 'F') |
1588 | /* Need to remove first */ | |
1ab9ed2a | 1589 | hot_remove_disk(fd, rdev, force); |
f277ce36 | 1590 | /* Make sure it isn't in use (in 2.6 or later) */ |
1d997643 | 1591 | tfd = dev_open(dv->devname, O_RDONLY|O_EXCL); |
38aeaf3a N |
1592 | if (tfd >= 0) { |
1593 | /* We know no-one else is using it. We'll | |
1594 | * need non-exclusive access to add it, so | |
1595 | * do that now. | |
1596 | */ | |
1597 | close(tfd); | |
1598 | tfd = dev_open(dv->devname, O_RDONLY); | |
1011e834 | 1599 | } |
0fbf459d | 1600 | if (tfd < 0) { |
1d997643 N |
1601 | if (dv->disposition == 'M') |
1602 | continue; | |
e7b84f9d | 1603 | pr_err("Cannot open %s: %s\n", |
d7eaf49f | 1604 | dv->devname, strerror(errno)); |
bcbb3112 | 1605 | goto abort; |
d7eaf49f | 1606 | } |
9f584691 N |
1607 | if (!frozen) { |
1608 | if (sysfs_freeze_array(&info) == 1) | |
1609 | frozen = 1; | |
1610 | else | |
1611 | frozen = -1; | |
1612 | } | |
38aeaf3a N |
1613 | rv = Manage_add(fd, tfd, dv, tst, &array, |
1614 | force, verbose, devname, update, | |
4de90913 | 1615 | rdev, array_size, raid_slot); |
38aeaf3a N |
1616 | close(tfd); |
1617 | tfd = -1; | |
1618 | if (rv < 0) | |
bcbb3112 | 1619 | goto abort; |
38aeaf3a N |
1620 | if (rv > 0) |
1621 | count++; | |
682c7051 NB |
1622 | break; |
1623 | ||
1624 | case 'r': | |
1625 | /* hot remove */ | |
4725bc31 | 1626 | if (subarray) { |
7a862a02 | 1627 | pr_err("Cannot remove disks from a \'member\' array, perform this operation on the parent container\n"); |
d070235d N |
1628 | rv = -1; |
1629 | } else | |
1630 | rv = Manage_remove(tst, fd, dv, sysfd, | |
1ab9ed2a | 1631 | rdev, verbose, force, |
d070235d N |
1632 | devname); |
1633 | if (sysfd >= 0) | |
98d27e39 | 1634 | close(sysfd); |
d070235d N |
1635 | sysfd = -1; |
1636 | if (rv < 0) | |
bcbb3112 | 1637 | goto abort; |
d070235d N |
1638 | if (rv > 0) |
1639 | count++; | |
682c7051 NB |
1640 | break; |
1641 | ||
1642 | case 'f': /* set faulty */ | |
fc6fd406 MK |
1643 | if (!is_remove_safe(&array, fd, dv->devname, verbose)) { |
1644 | pr_err("Cannot remove %s from %s, array will be failed.\n", | |
1645 | dv->devname, devname); | |
1646 | if (sysfd >= 0) | |
1647 | close(sysfd); | |
1648 | goto abort; | |
1649 | } | |
1650 | ||
98d27e39 N |
1651 | if ((sysfd >= 0 && write(sysfd, "faulty", 6) != 6) || |
1652 | (sysfd < 0 && ioctl(fd, SET_DISK_FAULTY, | |
5dffd09d | 1653 | rdev))) { |
8af530b0 N |
1654 | if (errno == EBUSY) |
1655 | busy = 1; | |
e7b84f9d | 1656 | pr_err("set device faulty failed for %s: %s\n", |
1d997643 | 1657 | dv->devname, strerror(errno)); |
98d27e39 N |
1658 | if (sysfd >= 0) |
1659 | close(sysfd); | |
bcbb3112 | 1660 | goto abort; |
682c7051 | 1661 | } |
98d27e39 N |
1662 | if (sysfd >= 0) |
1663 | close(sysfd); | |
1664 | sysfd = -1; | |
7d2e6486 | 1665 | count++; |
dab6685f | 1666 | if (verbose >= 0) |
e7b84f9d | 1667 | pr_err("set %s faulty in %s\n", |
1d997643 | 1668 | dv->devname, devname); |
682c7051 | 1669 | break; |
70c55e36 N |
1670 | case 'R': /* Mark as replaceable */ |
1671 | if (subarray) { | |
7a862a02 | 1672 | pr_err("Cannot replace disks in a \'member\' array, perform this operation on the parent container\n"); |
70c55e36 N |
1673 | rv = -1; |
1674 | } else { | |
1675 | if (!frozen) { | |
1676 | if (sysfs_freeze_array(&info) == 1) | |
1677 | frozen = 1; | |
1678 | else | |
1679 | frozen = -1; | |
1680 | } | |
1681 | rv = Manage_replace(tst, fd, dv, | |
5dffd09d | 1682 | rdev, verbose, |
70c55e36 N |
1683 | devname); |
1684 | } | |
1685 | if (rv < 0) | |
1686 | goto abort; | |
1687 | if (rv > 0) | |
1688 | count++; | |
1689 | break; | |
1690 | case 'W': /* --with device that doesn't match */ | |
1691 | pr_err("No matching --replace device for --with %s\n", | |
1692 | dv->devname); | |
1693 | goto abort; | |
1694 | case 'w': /* --with device which was matched */ | |
1695 | rv = Manage_with(tst, fd, dv, | |
5dffd09d | 1696 | rdev, verbose, devname); |
70c55e36 N |
1697 | if (rv < 0) |
1698 | goto abort; | |
1699 | break; | |
682c7051 NB |
1700 | } |
1701 | } | |
9f584691 N |
1702 | if (frozen > 0) |
1703 | sysfs_set_str(&info, NULL, "sync_action","idle"); | |
7d2e6486 N |
1704 | if (test && count == 0) |
1705 | return 2; | |
682c7051 | 1706 | return 0; |
bcbb3112 N |
1707 | |
1708 | abort: | |
9f584691 N |
1709 | if (frozen > 0) |
1710 | sysfs_set_str(&info, NULL, "sync_action","idle"); | |
8af530b0 | 1711 | return !test && busy ? 2 : 1; |
64c4757e | 1712 | } |
1f48664b NB |
1713 | |
1714 | int autodetect(void) | |
1715 | { | |
1716 | /* Open any md device, and issue the RAID_AUTORUN ioctl */ | |
1717 | int rv = 1; | |
1718 | int fd = dev_open("9:0", O_RDONLY); | |
1719 | if (fd >= 0) { | |
1720 | if (ioctl(fd, RAID_AUTORUN, 0) == 0) | |
1721 | rv = 0; | |
1722 | close(fd); | |
1723 | } | |
1724 | return rv; | |
1725 | } | |
aa534678 | 1726 | |
f2e8393b MK |
1727 | int Update_subarray(char *dev, char *subarray, enum update_opt update, |
1728 | struct mddev_ident *ident, int verbose) | |
aa534678 DW |
1729 | { |
1730 | struct supertype supertype, *st = &supertype; | |
1731 | int fd, rv = 2; | |
70f1ff42 | 1732 | struct mdinfo *info = NULL; |
f2e8393b | 1733 | char *update_verb = map_num(update_options, update); |
aa534678 DW |
1734 | |
1735 | memset(st, 0, sizeof(*st)); | |
aa534678 | 1736 | |
ba728be7 | 1737 | fd = open_subarray(dev, subarray, st, verbose < 0); |
aa534678 DW |
1738 | if (fd < 0) |
1739 | return 2; | |
1740 | ||
1741 | if (!st->ss->update_subarray) { | |
ba728be7 | 1742 | if (verbose >= 0) |
e7b84f9d N |
1743 | pr_err("Operation not supported for %s metadata\n", |
1744 | st->ss->name); | |
aa534678 DW |
1745 | goto free_super; |
1746 | } | |
1747 | ||
db10eab6 MK |
1748 | if (is_subarray_active(subarray, st->devnm)) { |
1749 | if (verbose >= 0) | |
1750 | pr_err("Subarray %s in %s is active, cannot update %s\n", | |
f2e8393b | 1751 | subarray, dev, update_verb); |
db10eab6 MK |
1752 | goto free_super; |
1753 | } | |
1754 | ||
4dd2df09 | 1755 | if (mdmon_running(st->devnm)) |
aa534678 DW |
1756 | st->update_tail = &st->updates; |
1757 | ||
70f1ff42 LF |
1758 | info = st->ss->container_content(st, subarray); |
1759 | ||
f2e8393b | 1760 | if (update == UOPT_PPL && !is_level456(info->array.level)) { |
70f1ff42 LF |
1761 | pr_err("RWH policy ppl is supported only for raid4, raid5 and raid6.\n"); |
1762 | goto free_super; | |
1763 | } | |
1764 | ||
f2e8393b | 1765 | rv = st->ss->update_subarray(st, subarray, update, ident); |
aa534678 DW |
1766 | |
1767 | if (rv) { | |
ba728be7 | 1768 | if (verbose >= 0) |
e7b84f9d | 1769 | pr_err("Failed to update %s of subarray-%s in %s\n", |
f2e8393b | 1770 | update_verb, subarray, dev); |
aa534678 DW |
1771 | } else if (st->update_tail) |
1772 | flush_metadata_updates(st); | |
1773 | else | |
1774 | st->ss->sync_metadata(st); | |
1775 | ||
f2e8393b | 1776 | if (rv == 0 && update == UOPT_NAME && verbose >= 0) |
e7b84f9d N |
1777 | pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n", |
1778 | subarray, dev); | |
aa534678 | 1779 | |
70f1ff42 LF |
1780 | free_super: |
1781 | if (info) | |
1782 | free(info); | |
aa534678 DW |
1783 | st->ss->free_super(st); |
1784 | close(fd); | |
1785 | ||
1786 | return rv; | |
1787 | } | |
d52bb542 | 1788 | |
7bd04da9 N |
1789 | /* Move spare from one array to another If adding to destination array fails |
1790 | * add back to original array. | |
d52bb542 AC |
1791 | * Returns 1 on success, 0 on failure */ |
1792 | int move_spare(char *from_devname, char *to_devname, dev_t devid) | |
1793 | { | |
1794 | struct mddev_dev devlist; | |
1795 | char devname[20]; | |
1796 | ||
1797 | /* try to remove and add */ | |
1798 | int fd1 = open(to_devname, O_RDONLY); | |
1799 | int fd2 = open(from_devname, O_RDONLY); | |
1800 | ||
1801 | if (fd1 < 0 || fd2 < 0) { | |
9cf361f8 JS |
1802 | if (fd1 >= 0) |
1803 | close(fd1); | |
1804 | if (fd2 >= 0) | |
1805 | close(fd2); | |
d52bb542 AC |
1806 | return 0; |
1807 | } | |
1808 | ||
1809 | devlist.next = NULL; | |
1810 | devlist.used = 0; | |
e22fe3ae N |
1811 | devlist.writemostly = FlagDefault; |
1812 | devlist.failfast = FlagDefault; | |
d52bb542 AC |
1813 | devlist.devname = devname; |
1814 | sprintf(devname, "%d:%d", major(devid), minor(devid)); | |
1815 | ||
1816 | devlist.disposition = 'r'; | |
f2e8393b | 1817 | if (Manage_subdevs(from_devname, fd2, &devlist, -1, 0, UOPT_UNDEFINED, 0) == 0) { |
d52bb542 | 1818 | devlist.disposition = 'a'; |
9cf361f8 | 1819 | if (Manage_subdevs(to_devname, fd1, &devlist, -1, 0, |
f2e8393b | 1820 | UOPT_UNDEFINED, 0) == 0) { |
d52bb542 AC |
1821 | /* make sure manager is aware of changes */ |
1822 | ping_manager(to_devname); | |
1823 | ping_manager(from_devname); | |
1824 | close(fd1); | |
1825 | close(fd2); | |
1826 | return 1; | |
1827 | } | |
9cf361f8 JS |
1828 | else |
1829 | Manage_subdevs(from_devname, fd2, &devlist, | |
f2e8393b | 1830 | -1, 0, UOPT_UNDEFINED, 0); |
d52bb542 AC |
1831 | } |
1832 | close(fd1); | |
1833 | close(fd2); | |
1834 | return 0; | |
1835 | } |