]>
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 | 224 | devname); |
f6feb3fb | 225 | sysfs_free(mdi); |
fe7e0e64 N |
226 | return 1; |
227 | } | |
932be627 N |
228 | /* If this is an mdmon managed array, just write 'inactive' |
229 | * to the array state and let mdmon clear up. | |
230 | */ | |
fe7e0e64 N |
231 | if (mdi && |
232 | mdi->array.level > 0 && | |
233 | is_subarray(mdi->text_version)) { | |
eb0af526 | 234 | int err; |
fe7e0e64 | 235 | /* This is mdmon managed. */ |
eb0af526 | 236 | close(fd); |
daf7a3ce | 237 | |
932be627 | 238 | /* As we had an O_EXCL open, any use of the device |
eb0af526 N |
239 | * which blocks STOP_ARRAY is probably a transient use, |
240 | * so it is reasonable to retry for a while - 5 seconds. | |
241 | */ | |
fe7e0e64 N |
242 | count = 25; |
243 | while (count && | |
244 | (err = sysfs_set_str(mdi, NULL, | |
245 | "array_state", | |
fc54fe7a JS |
246 | "inactive")) < 0 && |
247 | errno == EBUSY) { | |
239b3cc0 | 248 | sleep_for(0, MSEC_TO_NSEC(200), true); |
fe7e0e64 | 249 | count--; |
eb0af526 | 250 | } |
fe7e0e64 N |
251 | if (err) { |
252 | if (verbose >= 0) | |
e7b84f9d N |
253 | pr_err("failed to stop array %s: %s\n", |
254 | devname, strerror(errno)); | |
bccd8153 JS |
255 | rv = 1; |
256 | goto out; | |
682c7051 | 257 | } |
fe7e0e64 N |
258 | |
259 | /* Give monitor a chance to act */ | |
260 | ping_monitor(mdi->text_version); | |
261 | ||
262 | fd = open_dev_excl(devnm); | |
263 | if (fd < 0) { | |
264 | if (verbose >= 0) | |
7a862a02 | 265 | pr_err("failed to completely stop %s: Device is busy\n", |
fe7e0e64 N |
266 | devname); |
267 | rv = 1; | |
268 | goto out; | |
269 | } | |
270 | } else if (mdi && | |
271 | mdi->array.major_version == -1 && | |
272 | mdi->array.minor_version == -2 && | |
273 | !is_subarray(mdi->text_version)) { | |
274 | struct mdstat_ent *mds, *m; | |
275 | /* container, possibly mdmon-managed. | |
276 | * Make sure mdmon isn't opening it, which | |
277 | * would interfere with the 'stop' | |
97590376 | 278 | */ |
fe7e0e64 | 279 | ping_monitor(mdi->sys_name); |
daf7a3ce | 280 | |
fe7e0e64 N |
281 | /* now check that there are no existing arrays |
282 | * which are members of this array | |
283 | */ | |
284 | mds = mdstat_read(0, 0); | |
285 | for (m = mds; m; m = m->next) | |
286 | if (m->metadata_version && | |
287 | strncmp(m->metadata_version, "external:", 9)==0 && | |
288 | metadata_container_matches(m->metadata_version+9, | |
289 | devnm)) { | |
290 | if (verbose >= 0) | |
7a862a02 | 291 | pr_err("Cannot stop container %s: member %s still active\n", |
9581efb1 | 292 | devname, m->devnm); |
fe7e0e64 N |
293 | free_mdstat(mds); |
294 | rv = 1; | |
295 | goto out; | |
296 | } | |
297 | } | |
298 | ||
2eba8496 N |
299 | /* If the array is undergoing a reshape which changes the number |
300 | * of devices, then it would be nice to stop it at a point where | |
301 | * it has completed a full number of stripes in both old and | |
302 | * new layouts as this will allow the reshape to be reverted. | |
303 | * So if 'sync_action' is "reshape" and 'raid_disks' shows two | |
304 | * different numbers, then | |
305 | * - freeze reshape | |
306 | * - set sync_max to next multiple of both data_disks and | |
307 | * chunk sizes (or next but one) | |
308 | * - unfreeze reshape | |
309 | * - wait on 'sync_completed' for that point to be reached. | |
310 | */ | |
70f1ff42 | 311 | if (mdi && is_level456(mdi->array.level) && |
2eba8496 N |
312 | sysfs_attribute_available(mdi, NULL, "sync_action") && |
313 | sysfs_attribute_available(mdi, NULL, "reshape_direction") && | |
314 | sysfs_get_str(mdi, NULL, "sync_action", buf, 20) > 0 && | |
315 | strcmp(buf, "reshape\n") == 0 && | |
e3e0d0a8 | 316 | sysfs_get_two(mdi, NULL, "raid_disks", &rd1, &rd2) == 2) { |
2eba8496 N |
317 | unsigned long long position, curr; |
318 | unsigned long long chunk1, chunk2; | |
319 | unsigned long long rddiv, chunkdiv; | |
320 | unsigned long long sectors; | |
5509dc44 N |
321 | unsigned long long sync_max, old_sync_max; |
322 | unsigned long long completed; | |
2eba8496 N |
323 | int backwards = 0; |
324 | int delay; | |
325 | int scfd; | |
326 | ||
e3e0d0a8 N |
327 | delay = 40; |
328 | while (rd1 > rd2 && delay > 0 && | |
329 | sysfs_get_ll(mdi, NULL, "sync_max", &old_sync_max) == 0) { | |
330 | /* must be in the critical section - wait a bit */ | |
331 | delay -= 1; | |
239b3cc0 | 332 | sleep_for(0, MSEC_TO_NSEC(100), true); |
e3e0d0a8 N |
333 | } |
334 | ||
335 | if (sysfs_set_str(mdi, NULL, "sync_action", "frozen") != 0) | |
336 | goto done; | |
337 | /* Array is frozen */ | |
338 | ||
2eba8496 N |
339 | rd1 -= mdi->array.level == 6 ? 2 : 1; |
340 | rd2 -= mdi->array.level == 6 ? 2 : 1; | |
341 | sysfs_get_str(mdi, NULL, "reshape_direction", buf, sizeof(buf)); | |
342 | if (strncmp(buf, "back", 4) == 0) | |
343 | backwards = 1; | |
e3e0d0a8 N |
344 | if (sysfs_get_ll(mdi, NULL, "reshape_position", &position) != 0) { |
345 | /* reshape must have finished now */ | |
346 | sysfs_set_str(mdi, NULL, "sync_action", "idle"); | |
347 | goto done; | |
348 | } | |
2eba8496 N |
349 | sysfs_get_two(mdi, NULL, "chunk_size", &chunk1, &chunk2); |
350 | chunk1 /= 512; | |
351 | chunk2 /= 512; | |
352 | rddiv = GCD(rd1, rd2); | |
353 | chunkdiv = GCD(chunk1, chunk2); | |
354 | sectors = (chunk1/chunkdiv) * chunk2 * (rd1/rddiv) * rd2; | |
355 | ||
356 | if (backwards) { | |
357 | /* Need to subtract 'reshape_position' from | |
358 | * array size to get equivalent of sync_max. | |
359 | * Size calculation based on raid5_size in kernel. | |
360 | */ | |
361 | unsigned long long size = mdi->component_size; | |
362 | size &= ~(chunk1-1); | |
363 | size &= ~(chunk2-1); | |
364 | /* rd1 must be smaller */ | |
932be627 N |
365 | /* Reshape may have progressed further backwards than |
366 | * recorded, so target even further back (hence "-1") | |
367 | */ | |
5509dc44 | 368 | position = (position / sectors - 1) * sectors; |
932be627 N |
369 | /* rd1 is always the conversion factor between 'sync' |
370 | * position and 'reshape' position. | |
371 | * We read 1 "new" stripe worth of data from where-ever, | |
372 | * and when write out that full stripe. | |
373 | */ | |
5509dc44 | 374 | sync_max = size - position/rd1; |
2eba8496 | 375 | } else { |
932be627 N |
376 | /* Reshape will very likely be beyond position, and it may |
377 | * be too late to stop at '+1', so aim for '+2' | |
378 | */ | |
5509dc44 N |
379 | position = (position / sectors + 2) * sectors; |
380 | sync_max = position/rd1; | |
2eba8496 | 381 | } |
5509dc44 N |
382 | if (sysfs_get_ll(mdi, NULL, "sync_max", &old_sync_max) < 0) |
383 | old_sync_max = mdi->component_size; | |
384 | /* Must not advance sync_max as that could confuse | |
385 | * the reshape monitor */ | |
386 | if (sync_max < old_sync_max) | |
387 | sysfs_set_num(mdi, NULL, "sync_max", sync_max); | |
2eba8496 N |
388 | sysfs_set_str(mdi, NULL, "sync_action", "idle"); |
389 | ||
390 | /* That should have set things going again. Now we | |
3afaff93 | 391 | * wait a little while (3 second max) for sync_completed |
2eba8496 | 392 | * to reach the target. |
3afaff93 N |
393 | * The reshape process can block for 500msec if |
394 | * the sync speed limit is hit, so we need to wait | |
395 | * a lot longer than that. 1 second is usually | |
396 | * enough. 3 is safe. | |
2eba8496 | 397 | */ |
3afaff93 | 398 | delay = 3000; |
2eba8496 | 399 | scfd = sysfs_open(mdi->sys_name, NULL, "sync_completed"); |
3afaff93 | 400 | while (scfd >= 0 && delay > 0 && old_sync_max > 0) { |
30ddba7d | 401 | unsigned long long max_completed; |
5509dc44 | 402 | sysfs_get_ll(mdi, NULL, "reshape_position", &curr); |
2eba8496 | 403 | sysfs_fd_get_str(scfd, buf, sizeof(buf)); |
3afaff93 N |
404 | if (strncmp(buf, "none", 4) == 0) { |
405 | /* Either reshape has aborted, or hasn't | |
406 | * quite started yet. Wait a bit and | |
407 | * check 'sync_action' to see. | |
408 | */ | |
239b3cc0 | 409 | sleep_for(0, MSEC_TO_NSEC(10), true); |
3afaff93 N |
410 | sysfs_get_str(mdi, NULL, "sync_action", buf, sizeof(buf)); |
411 | if (strncmp(buf, "reshape", 7) != 0) | |
412 | break; | |
413 | } | |
5509dc44 | 414 | |
30ddba7d N |
415 | if (sysfs_fd_get_two(scfd, &completed, |
416 | &max_completed) == 2 && | |
417 | /* 'completed' sometimes reads as max-uulong */ | |
418 | completed < max_completed && | |
5509dc44 N |
419 | (completed > sync_max || |
420 | (completed == sync_max && curr != position))) { | |
421 | while (completed > sync_max) { | |
422 | sync_max += sectors / rd1; | |
423 | if (backwards) | |
424 | position -= sectors; | |
425 | else | |
426 | position += sectors; | |
427 | } | |
428 | if (sync_max < old_sync_max) | |
429 | sysfs_set_num(mdi, NULL, "sync_max", sync_max); | |
430 | } | |
431 | ||
2eba8496 N |
432 | if (!backwards && curr >= position) |
433 | break; | |
434 | if (backwards && curr <= position) | |
435 | break; | |
436 | sysfs_wait(scfd, &delay); | |
437 | } | |
438 | if (scfd >= 0) | |
439 | close(scfd); | |
440 | ||
441 | } | |
e3e0d0a8 | 442 | done: |
2eba8496 | 443 | |
fe7e0e64 N |
444 | /* As we have an O_EXCL open, any use of the device |
445 | * which blocks STOP_ARRAY is probably a transient use, | |
446 | * so it is reasonable to retry for a while - 5 seconds. | |
447 | */ | |
448 | count = 25; err = 0; | |
fc54fe7a JS |
449 | while (count && fd >= 0 && |
450 | (err = ioctl(fd, STOP_ARRAY, NULL)) < 0 && errno == EBUSY) { | |
239b3cc0 | 451 | sleep_for(0, MSEC_TO_NSEC(200), true); |
fe7e0e64 N |
452 | count --; |
453 | } | |
454 | if (fd >= 0 && err) { | |
455 | if (verbose >= 0) { | |
456 | pr_err("failed to stop array %s: %s\n", | |
457 | devname, strerror(errno)); | |
458 | if (errno == EBUSY) | |
7a862a02 | 459 | cont_err("Perhaps a running process, mounted filesystem or active volume group?\n"); |
4ccad7b1 | 460 | } |
fe7e0e64 N |
461 | rv = 1; |
462 | goto out; | |
463 | } | |
229e66cb | 464 | |
fe7e0e64 N |
465 | if (devnm[0] && use_udev()) { |
466 | struct map_ent *mp = map_by_devnm(&map, devnm); | |
467 | remove_devices(devnm, mp ? mp->path : NULL); | |
682c7051 | 468 | } |
fe7e0e64 N |
469 | |
470 | if (verbose >= 0) | |
471 | pr_err("stopped %s\n", devname); | |
472 | map_lock(&map); | |
473 | map_remove(&map, devnm); | |
474 | map_unlock(&map); | |
475 | out: | |
6e8d27e7 | 476 | sysfs_free(mdi); |
fe7e0e64 | 477 | |
bccd8153 | 478 | return rv; |
64c4757e NB |
479 | } |
480 | ||
5e73b024 N |
481 | static struct mddev_dev *add_one(struct mddev_dev *dv, char *name, char disp) |
482 | { | |
483 | struct mddev_dev *new; | |
484 | new = xmalloc(sizeof(*new)); | |
485 | memset(new, 0, sizeof(*new)); | |
486 | new->devname = xstrdup(name); | |
487 | new->disposition = disp; | |
488 | new->next = dv->next; | |
489 | dv->next = new; | |
490 | return new; | |
491 | } | |
492 | ||
1d997643 N |
493 | static void add_faulty(struct mddev_dev *dv, int fd, char disp) |
494 | { | |
495 | mdu_array_info_t array; | |
496 | mdu_disk_info_t disk; | |
497 | int remaining_disks; | |
498 | int i; | |
499 | ||
9cd39f01 | 500 | if (md_get_array_info(fd, &array) != 0) |
1d997643 N |
501 | return; |
502 | ||
503 | remaining_disks = array.nr_disks; | |
504 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
1d997643 N |
505 | char buf[40]; |
506 | disk.number = i; | |
d97572f5 | 507 | if (md_get_disk_info(fd, &disk) != 0) |
1d997643 N |
508 | continue; |
509 | if (disk.major == 0 && disk.minor == 0) | |
510 | continue; | |
511 | remaining_disks--; | |
512 | if ((disk.state & 1) == 0) /* not faulty */ | |
513 | continue; | |
514 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
5e73b024 | 515 | dv = add_one(dv, buf, disp); |
1d997643 N |
516 | } |
517 | } | |
518 | ||
519 | static void add_detached(struct mddev_dev *dv, int fd, char disp) | |
520 | { | |
521 | mdu_array_info_t array; | |
522 | mdu_disk_info_t disk; | |
523 | int remaining_disks; | |
524 | int i; | |
525 | ||
9cd39f01 | 526 | if (md_get_array_info(fd, &array) != 0) |
1d997643 N |
527 | return; |
528 | ||
529 | remaining_disks = array.nr_disks; | |
530 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
1d997643 N |
531 | char buf[40]; |
532 | int sfd; | |
533 | disk.number = i; | |
d97572f5 | 534 | if (md_get_disk_info(fd, &disk) != 0) |
1d997643 N |
535 | continue; |
536 | if (disk.major == 0 && disk.minor == 0) | |
537 | continue; | |
538 | remaining_disks--; | |
539 | if (disp == 'f' && (disk.state & 1) != 0) /* already faulty */ | |
540 | continue; | |
541 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
542 | sfd = dev_open(buf, O_RDONLY); | |
543 | if (sfd >= 0) { | |
544 | /* Not detached */ | |
545 | close(sfd); | |
546 | continue; | |
547 | } | |
548 | if (errno != ENXIO) | |
549 | /* Probably not detached */ | |
550 | continue; | |
5e73b024 | 551 | dv = add_one(dv, buf, disp); |
1d997643 N |
552 | } |
553 | } | |
554 | ||
64a78416 N |
555 | static void add_set(struct mddev_dev *dv, int fd, char set_char) |
556 | { | |
557 | mdu_array_info_t array; | |
558 | mdu_disk_info_t disk; | |
559 | int remaining_disks; | |
560 | int copies, set; | |
561 | int i; | |
562 | ||
9cd39f01 | 563 | if (md_get_array_info(fd, &array) != 0) |
64a78416 N |
564 | return; |
565 | if (array.level != 10) | |
566 | return; | |
567 | copies = ((array.layout & 0xff) * | |
568 | ((array.layout >> 8) & 0xff)); | |
569 | if (array.raid_disks % copies) | |
570 | return; | |
571 | ||
572 | remaining_disks = array.nr_disks; | |
573 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
574 | char buf[40]; | |
575 | disk.number = i; | |
d97572f5 | 576 | if (md_get_disk_info(fd, &disk) != 0) |
64a78416 N |
577 | continue; |
578 | if (disk.major == 0 && disk.minor == 0) | |
579 | continue; | |
580 | remaining_disks--; | |
581 | set = disk.raid_disk % copies; | |
582 | if (set_char != set + 'A') | |
583 | continue; | |
584 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
585 | dv = add_one(dv, buf, dv->disposition); | |
586 | } | |
587 | } | |
588 | ||
abe94694 N |
589 | int attempt_re_add(int fd, int tfd, struct mddev_dev *dv, |
590 | struct supertype *dev_st, struct supertype *tst, | |
f2e8393b MK |
591 | unsigned long rdev, enum update_opt update, |
592 | char *devname, int verbose, mdu_array_info_t *array) | |
abe94694 N |
593 | { |
594 | struct mdinfo mdi; | |
595 | int duuid[4]; | |
596 | int ouuid[4]; | |
597 | ||
598 | dev_st->ss->getinfo_super(dev_st, &mdi, NULL); | |
599 | dev_st->ss->uuid_from_super(dev_st, ouuid); | |
600 | if (tst->sb) | |
601 | tst->ss->uuid_from_super(tst, duuid); | |
602 | else | |
603 | /* Assume uuid matches: kernel will check */ | |
604 | memcpy(duuid, ouuid, sizeof(ouuid)); | |
605 | if ((mdi.disk.state & (1<<MD_DISK_ACTIVE)) && | |
606 | !(mdi.disk.state & (1<<MD_DISK_FAULTY)) && | |
607 | memcmp(duuid, ouuid, sizeof(ouuid))==0) { | |
608 | /* Looks like it is worth a | |
609 | * try. Need to make sure | |
610 | * kernel will accept it | |
611 | * though. | |
612 | */ | |
613 | mdu_disk_info_t disc; | |
abe94694 | 614 | disc.number = mdi.disk.number; |
d97572f5 JS |
615 | if (md_get_disk_info(fd, &disc) != 0 || |
616 | disc.major != 0 || disc.minor != 0) | |
abe94694 N |
617 | goto skip_re_add; |
618 | disc.major = major(rdev); | |
619 | disc.minor = minor(rdev); | |
620 | disc.number = mdi.disk.number; | |
621 | disc.raid_disk = mdi.disk.raid_disk; | |
622 | disc.state = mdi.disk.state; | |
bff96f73 GJ |
623 | if (array->state & (1 << MD_SB_CLUSTERED)) { |
624 | /* extra flags are needed when adding to a cluster as | |
625 | * there are two cases to distinguish | |
626 | */ | |
627 | if (dv->disposition == 'c') | |
628 | disc.state |= (1 << MD_DISK_CANDIDATE); | |
629 | else | |
630 | disc.state |= (1 << MD_DISK_CLUSTER_ADD); | |
631 | } | |
e22fe3ae | 632 | if (dv->writemostly == FlagSet) |
abe94694 | 633 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; |
e22fe3ae | 634 | if (dv->writemostly == FlagClear) |
abe94694 | 635 | disc.state &= ~(1 << MD_DISK_WRITEMOSTLY); |
e22fe3ae | 636 | if (dv->failfast == FlagSet) |
71574efb | 637 | disc.state |= 1 << MD_DISK_FAILFAST; |
e22fe3ae | 638 | if (dv->failfast == FlagClear) |
71574efb | 639 | disc.state &= ~(1 << MD_DISK_FAILFAST); |
abe94694 | 640 | remove_partitions(tfd); |
d7be7d87 JS |
641 | if (update || dv->writemostly != FlagDefault || |
642 | dv->failfast != FlagDefault) { | |
abe94694 N |
643 | int rv = -1; |
644 | tfd = dev_open(dv->devname, O_RDWR); | |
645 | if (tfd < 0) { | |
7a862a02 | 646 | pr_err("failed to open %s for superblock update during re-add\n", dv->devname); |
abe94694 N |
647 | return -1; |
648 | } | |
649 | ||
e22fe3ae | 650 | if (dv->writemostly == FlagSet) |
abe94694 | 651 | rv = dev_st->ss->update_super( |
03312b52 | 652 | dev_st, NULL, UOPT_SPEC_WRITEMOSTLY, |
abe94694 | 653 | devname, verbose, 0, NULL); |
e22fe3ae | 654 | if (dv->writemostly == FlagClear) |
abe94694 | 655 | rv = dev_st->ss->update_super( |
03312b52 | 656 | dev_st, NULL, UOPT_SPEC_READWRITE, |
abe94694 | 657 | devname, verbose, 0, NULL); |
e22fe3ae | 658 | if (dv->failfast == FlagSet) |
71574efb | 659 | rv = dev_st->ss->update_super( |
03312b52 | 660 | dev_st, NULL, UOPT_SPEC_FAILFAST, |
71574efb | 661 | devname, verbose, 0, NULL); |
e22fe3ae | 662 | if (dv->failfast == FlagClear) |
71574efb | 663 | rv = dev_st->ss->update_super( |
03312b52 | 664 | dev_st, NULL, UOPT_SPEC_NOFAILFAST, |
71574efb | 665 | devname, verbose, 0, NULL); |
abe94694 N |
666 | if (update) |
667 | rv = dev_st->ss->update_super( | |
f2e8393b | 668 | dev_st, NULL, update, |
abe94694 N |
669 | devname, verbose, 0, NULL); |
670 | if (rv == 0) | |
671 | rv = dev_st->ss->store_super(dev_st, tfd); | |
672 | close(tfd); | |
673 | if (rv != 0) { | |
7a862a02 | 674 | pr_err("failed to update superblock during re-add\n"); |
abe94694 N |
675 | return -1; |
676 | } | |
677 | } | |
678 | /* don't even try if disk is marked as faulty */ | |
679 | errno = 0; | |
680 | if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) { | |
681 | if (verbose >= 0) | |
682 | pr_err("re-added %s\n", dv->devname); | |
683 | return 1; | |
684 | } | |
685 | if (errno == ENOMEM || errno == EROFS) { | |
686 | pr_err("add new device failed for %s: %s\n", | |
687 | dv->devname, strerror(errno)); | |
688 | if (dv->disposition == 'M') | |
689 | return 0; | |
690 | return -1; | |
691 | } | |
692 | } | |
693 | skip_re_add: | |
694 | return 0; | |
695 | } | |
696 | ||
38aeaf3a N |
697 | int Manage_add(int fd, int tfd, struct mddev_dev *dv, |
698 | struct supertype *tst, mdu_array_info_t *array, | |
699 | int force, int verbose, char *devname, | |
f2e8393b MK |
700 | enum update_opt update, unsigned long rdev, |
701 | unsigned long long array_size, int raid_slot) | |
38aeaf3a N |
702 | { |
703 | unsigned long long ldsize; | |
fbd3e15c | 704 | struct supertype *dev_st; |
38aeaf3a N |
705 | int j; |
706 | mdu_disk_info_t disc; | |
707 | ||
708 | if (!get_dev_size(tfd, dv->devname, &ldsize)) { | |
709 | if (dv->disposition == 'M') | |
710 | return 0; | |
711 | else | |
712 | return -1; | |
713 | } | |
714 | ||
632dc30c | 715 | if (tst->ss == &super0 && ldsize > 4ULL*1024*1024*1024*1024) { |
7ccc4cc4 | 716 | /* More than 4TB is wasted on v0.90 */ |
38aeaf3a | 717 | if (!force) { |
7a862a02 N |
718 | pr_err("%s is larger than %s can effectively use.\n" |
719 | " Add --force is you really want to add this device.\n", | |
38aeaf3a N |
720 | dv->devname, devname); |
721 | return -1; | |
722 | } | |
7a862a02 N |
723 | pr_err("%s is larger than %s can effectively use.\n" |
724 | " Adding anyway as --force was given.\n", | |
38aeaf3a N |
725 | dv->devname, devname); |
726 | } | |
38aeaf3a N |
727 | |
728 | if (array->not_persistent == 0 || tst->ss->external) { | |
729 | ||
730 | /* need to find a sample superblock to copy, and | |
731 | * a spare slot to use. | |
732 | * For 'external' array (well, container based), | |
733 | * We can just load the metadata for the array-> | |
734 | */ | |
735 | int array_failed; | |
736 | if (tst->sb) | |
737 | /* already loaded */; | |
738 | else if (tst->ss->external) { | |
739 | tst->ss->load_container(tst, fd, NULL); | |
740 | } else for (j = 0; j < tst->max_devs; j++) { | |
741 | char *dev; | |
742 | int dfd; | |
743 | disc.number = j; | |
d97572f5 | 744 | if (md_get_disk_info(fd, &disc)) |
38aeaf3a N |
745 | continue; |
746 | if (disc.major==0 && disc.minor==0) | |
747 | continue; | |
748 | if ((disc.state & 4)==0) /* sync */ | |
749 | continue; | |
750 | /* Looks like a good device to try */ | |
751 | dev = map_dev(disc.major, disc.minor, 1); | |
752 | if (!dev) | |
753 | continue; | |
754 | dfd = dev_open(dev, O_RDONLY); | |
755 | if (dfd < 0) | |
756 | continue; | |
757 | if (tst->ss->load_super(tst, dfd, | |
758 | NULL)) { | |
759 | close(dfd); | |
760 | continue; | |
761 | } | |
762 | close(dfd); | |
763 | break; | |
764 | } | |
765 | /* FIXME this is a bad test to be using */ | |
fc54fe7a JS |
766 | if (!tst->sb && (dv->disposition != 'a' && |
767 | dv->disposition != 'S')) { | |
38aeaf3a N |
768 | /* we are re-adding a device to a |
769 | * completely dead array - have to depend | |
770 | * on kernel to check | |
771 | */ | |
772 | } else if (!tst->sb) { | |
773 | pr_err("cannot load array metadata from %s\n", devname); | |
774 | return -1; | |
775 | } | |
776 | ||
777 | /* Make sure device is large enough */ | |
01290056 SL |
778 | if (dv->disposition != 'j' && /* skip size check for Journal */ |
779 | tst->sb && | |
2609f339 | 780 | tst->ss->avail_size(tst, ldsize/512, INVALID_SECTORS) < |
38aeaf3a N |
781 | array_size) { |
782 | if (dv->disposition == 'M') | |
783 | return 0; | |
784 | pr_err("%s not large enough to join array\n", | |
785 | dv->devname); | |
786 | return -1; | |
787 | } | |
788 | ||
789 | /* Possibly this device was recently part of | |
790 | * the array and was temporarily removed, and | |
791 | * is now being re-added. If so, we can | |
792 | * simply re-add it. | |
793 | */ | |
794 | ||
fbd3e15c | 795 | if (array->not_persistent == 0) { |
38aeaf3a N |
796 | dev_st = dup_super(tst); |
797 | dev_st->ss->load_super(dev_st, tfd, NULL); | |
fbd3e15c JS |
798 | if (dev_st->sb && dv->disposition != 'S') { |
799 | int rv; | |
800 | ||
801 | rv = attempt_re_add(fd, tfd, dv, dev_st, tst, | |
802 | rdev, update, devname, | |
803 | verbose, array); | |
804 | dev_st->ss->free_super(dev_st); | |
f6feb3fb GM |
805 | if (rv) { |
806 | free(dev_st); | |
fbd3e15c | 807 | return rv; |
f6feb3fb GM |
808 | } |
809 | } | |
810 | if (dev_st) { | |
811 | dev_st->ss->free_super(dev_st); | |
812 | free(dev_st); | |
fbd3e15c | 813 | } |
38aeaf3a N |
814 | } |
815 | if (dv->disposition == 'M') { | |
816 | if (verbose > 0) | |
817 | pr_err("--re-add for %s to %s is not possible\n", | |
818 | dv->devname, devname); | |
819 | return 0; | |
820 | } | |
821 | if (dv->disposition == 'A') { | |
822 | pr_err("--re-add for %s to %s is not possible\n", | |
823 | dv->devname, devname); | |
824 | return -1; | |
825 | } | |
826 | if (array->active_disks < array->raid_disks) { | |
827 | char *avail = xcalloc(array->raid_disks, 1); | |
828 | int d; | |
829 | int found = 0; | |
830 | ||
d180d2aa | 831 | for (d = 0; d < MAX_DISKS && found < array->nr_disks; d++) { |
38aeaf3a | 832 | disc.number = d; |
d97572f5 | 833 | if (md_get_disk_info(fd, &disc)) |
38aeaf3a N |
834 | continue; |
835 | if (disc.major == 0 && disc.minor == 0) | |
836 | continue; | |
837 | if (!(disc.state & (1<<MD_DISK_SYNC))) | |
838 | continue; | |
839 | avail[disc.raid_disk] = 1; | |
d31d0f52 | 840 | found++; |
38aeaf3a N |
841 | } |
842 | array_failed = !enough(array->level, array->raid_disks, | |
843 | array->layout, 1, avail); | |
6157951f | 844 | free(avail); |
38aeaf3a N |
845 | } else |
846 | array_failed = 0; | |
847 | if (array_failed) { | |
848 | pr_err("%s has failed so using --add cannot work and might destroy\n", | |
849 | devname); | |
850 | pr_err("data on %s. You should stop the array and re-assemble it.\n", | |
851 | dv->devname); | |
852 | return -1; | |
853 | } | |
854 | } else { | |
855 | /* non-persistent. Must ensure that new drive | |
856 | * is at least array->size big. | |
857 | */ | |
858 | if (ldsize/512 < array_size) { | |
859 | pr_err("%s not large enough to join array\n", | |
860 | dv->devname); | |
861 | return -1; | |
862 | } | |
863 | } | |
864 | /* committed to really trying this device now*/ | |
865 | remove_partitions(tfd); | |
866 | ||
867 | /* in 2.6.17 and earlier, version-1 superblocks won't | |
868 | * use the number we write, but will choose a free number. | |
869 | * we must choose the same free number, which requires | |
870 | * starting at 'raid_disks' and counting up | |
871 | */ | |
872 | for (j = array->raid_disks; j < tst->max_devs; j++) { | |
873 | disc.number = j; | |
d97572f5 | 874 | if (md_get_disk_info(fd, &disc)) |
38aeaf3a N |
875 | break; |
876 | if (disc.major==0 && disc.minor==0) | |
877 | break; | |
878 | if (disc.state & 8) /* removed */ | |
879 | break; | |
880 | } | |
881 | disc.major = major(rdev); | |
882 | disc.minor = minor(rdev); | |
4de90913 GJ |
883 | if (raid_slot < 0) |
884 | disc.number = j; | |
885 | else | |
886 | disc.number = raid_slot; | |
38aeaf3a | 887 | disc.state = 0; |
01290056 SL |
888 | |
889 | /* only add journal to array that supports journaling */ | |
890 | if (dv->disposition == 'j') { | |
01290056 SL |
891 | struct mdinfo *mdp; |
892 | ||
893 | mdp = sysfs_read(fd, NULL, GET_ARRAY_STATE); | |
2a1990c0 JS |
894 | if (!mdp) { |
895 | pr_err("%s unable to read array state.\n", devname); | |
896 | return -1; | |
897 | } | |
01290056 | 898 | |
5e4ca8bb | 899 | if (mdp->array_state != ARRAY_READONLY) { |
d209181d | 900 | sysfs_free(mdp); |
01290056 SL |
901 | pr_err("%s is not readonly, cannot add journal.\n", devname); |
902 | return -1; | |
903 | } | |
904 | ||
d209181d JS |
905 | sysfs_free(mdp); |
906 | ||
38c2e05b | 907 | disc.raid_disk = 0; |
01290056 SL |
908 | } |
909 | ||
38aeaf3a N |
910 | if (array->not_persistent==0) { |
911 | int dfd; | |
01290056 SL |
912 | if (dv->disposition == 'j') |
913 | disc.state |= (1 << MD_DISK_JOURNAL) | (1 << MD_DISK_SYNC); | |
e22fe3ae | 914 | if (dv->writemostly == FlagSet) |
38aeaf3a | 915 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; |
e22fe3ae | 916 | if (dv->failfast == FlagSet) |
71574efb | 917 | disc.state |= 1 << MD_DISK_FAILFAST; |
38aeaf3a N |
918 | dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); |
919 | if (tst->ss->add_to_super(tst, &disc, dfd, | |
72ca9bcf | 920 | dv->devname, INVALID_SECTORS)) |
38aeaf3a N |
921 | return -1; |
922 | if (tst->ss->write_init_super(tst)) | |
923 | return -1; | |
924 | } else if (dv->disposition == 'A') { | |
925 | /* this had better be raid1. | |
926 | * As we are "--re-add"ing we must find a spare slot | |
927 | * to fill. | |
928 | */ | |
929 | char *used = xcalloc(array->raid_disks, 1); | |
930 | for (j = 0; j < tst->max_devs; j++) { | |
931 | mdu_disk_info_t disc2; | |
932 | disc2.number = j; | |
d97572f5 | 933 | if (md_get_disk_info(fd, &disc2)) |
38aeaf3a N |
934 | continue; |
935 | if (disc2.major==0 && disc2.minor==0) | |
936 | continue; | |
937 | if (disc2.state & 8) /* removed */ | |
938 | continue; | |
939 | if (disc2.raid_disk < 0) | |
940 | continue; | |
941 | if (disc2.raid_disk > array->raid_disks) | |
942 | continue; | |
943 | used[disc2.raid_disk] = 1; | |
944 | } | |
945 | for (j = 0 ; j < array->raid_disks; j++) | |
946 | if (!used[j]) { | |
947 | disc.raid_disk = j; | |
948 | disc.state |= (1<<MD_DISK_SYNC); | |
949 | break; | |
950 | } | |
951 | free(used); | |
952 | } | |
4de90913 GJ |
953 | |
954 | if (array->state & (1 << MD_SB_CLUSTERED)) { | |
955 | if (dv->disposition == 'c') | |
956 | disc.state |= (1 << MD_DISK_CANDIDATE); | |
957 | else | |
958 | disc.state |= (1 << MD_DISK_CLUSTER_ADD); | |
959 | } | |
960 | ||
e22fe3ae | 961 | if (dv->writemostly == FlagSet) |
38aeaf3a | 962 | disc.state |= (1 << MD_DISK_WRITEMOSTLY); |
e22fe3ae | 963 | if (dv->failfast == FlagSet) |
71574efb | 964 | disc.state |= (1 << MD_DISK_FAILFAST); |
38aeaf3a N |
965 | if (tst->ss->external) { |
966 | /* add a disk | |
967 | * to an external metadata container */ | |
968 | struct mdinfo new_mdi; | |
969 | struct mdinfo *sra; | |
970 | int container_fd; | |
4dd2df09 | 971 | char devnm[32]; |
38aeaf3a N |
972 | int dfd; |
973 | ||
4dd2df09 N |
974 | strcpy(devnm, fd2devnm(fd)); |
975 | ||
976 | container_fd = open_dev_excl(devnm); | |
38aeaf3a | 977 | if (container_fd < 0) { |
7a862a02 | 978 | pr_err("add failed for %s: could not get exclusive access to container\n", |
38aeaf3a N |
979 | dv->devname); |
980 | tst->ss->free_super(tst); | |
981 | return -1; | |
982 | } | |
983 | ||
1f5d54a0 MT |
984 | /* Check if metadata handler is able to accept the drive */ |
985 | if (!tst->ss->validate_geometry(tst, LEVEL_CONTAINER, 0, 1, NULL, | |
986 | 0, 0, dv->devname, NULL, 0, 1)) { | |
987 | close(container_fd); | |
988 | return -1; | |
989 | } | |
990 | ||
9cf9a1de | 991 | Kill(dv->devname, NULL, 0, -1, 0); |
38aeaf3a | 992 | dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); |
38aeaf3a | 993 | if (tst->ss->add_to_super(tst, &disc, dfd, |
72ca9bcf | 994 | dv->devname, INVALID_SECTORS)) { |
38aeaf3a N |
995 | close(dfd); |
996 | close(container_fd); | |
997 | return -1; | |
998 | } | |
12724c01 | 999 | if (!mdmon_running(tst->container_devnm)) |
38aeaf3a N |
1000 | tst->ss->sync_metadata(tst); |
1001 | ||
4dd2df09 | 1002 | sra = sysfs_read(container_fd, NULL, 0); |
38aeaf3a N |
1003 | if (!sra) { |
1004 | pr_err("add failed for %s: sysfs_read failed\n", | |
1005 | dv->devname); | |
1006 | close(container_fd); | |
1007 | tst->ss->free_super(tst); | |
1008 | return -1; | |
1009 | } | |
1010 | sra->array.level = LEVEL_CONTAINER; | |
1011 | /* Need to set data_offset and component_size */ | |
1012 | tst->ss->getinfo_super(tst, &new_mdi, NULL); | |
1013 | new_mdi.disk.major = disc.major; | |
1014 | new_mdi.disk.minor = disc.minor; | |
1015 | new_mdi.recovery_start = 0; | |
1016 | /* Make sure fds are closed as they are O_EXCL which | |
1017 | * would block add_disk */ | |
1018 | tst->ss->free_super(tst); | |
1019 | if (sysfs_add_disk(sra, &new_mdi, 0) != 0) { | |
7a862a02 | 1020 | pr_err("add new device to external metadata failed for %s\n", dv->devname); |
38aeaf3a N |
1021 | close(container_fd); |
1022 | sysfs_free(sra); | |
1023 | return -1; | |
1024 | } | |
4dd2df09 | 1025 | ping_monitor(devnm); |
38aeaf3a N |
1026 | sysfs_free(sra); |
1027 | close(container_fd); | |
1028 | } else { | |
1029 | tst->ss->free_super(tst); | |
1030 | if (ioctl(fd, ADD_NEW_DISK, &disc)) { | |
01290056 SL |
1031 | if (dv->disposition == 'j') |
1032 | pr_err("Failed to hot add %s as journal, " | |
1033 | "please try restart %s.\n", dv->devname, devname); | |
1034 | else | |
1035 | pr_err("add new device failed for %s as %d: %s\n", | |
1036 | dv->devname, j, strerror(errno)); | |
38aeaf3a N |
1037 | return -1; |
1038 | } | |
01290056 SL |
1039 | if (dv->disposition == 'j') { |
1040 | pr_err("Journal added successfully, making %s read-write\n", devname); | |
1041 | if (Manage_ro(devname, fd, -1)) | |
1042 | pr_err("Failed to make %s read-write\n", devname); | |
1043 | } | |
1044 | ||
38aeaf3a N |
1045 | } |
1046 | if (verbose >= 0) | |
1047 | pr_err("added %s\n", dv->devname); | |
1048 | return 1; | |
1049 | } | |
1050 | ||
d070235d | 1051 | int Manage_remove(struct supertype *tst, int fd, struct mddev_dev *dv, |
1ab9ed2a | 1052 | int sysfd, unsigned long rdev, int force, int verbose, char *devname) |
d070235d N |
1053 | { |
1054 | int lfd = -1; | |
1055 | int err; | |
1056 | ||
1057 | if (tst->ss->external) { | |
1058 | /* To remove a device from a container, we must | |
1059 | * check that it isn't in use in an array. | |
1060 | * This involves looking in the 'holders' | |
1061 | * directory - there must be just one entry, | |
1062 | * the container. | |
1063 | * To ensure that it doesn't get used as a | |
1064 | * hot spare while we are checking, we | |
1065 | * get an O_EXCL open on the container | |
1066 | */ | |
aab15415 | 1067 | int ret; |
4dd2df09 N |
1068 | char devnm[32]; |
1069 | strcpy(devnm, fd2devnm(fd)); | |
1070 | lfd = open_dev_excl(devnm); | |
d070235d | 1071 | if (lfd < 0) { |
7a862a02 | 1072 | pr_err("Cannot get exclusive access to container - odd\n"); |
d070235d N |
1073 | return -1; |
1074 | } | |
aab15415 N |
1075 | /* We may not be able to check on holders in |
1076 | * sysfs, either because we don't have the dev num | |
1077 | * (rdev == 0) or because the device has been detached | |
1078 | * and the 'holders' directory no longer exists | |
1079 | * (ret == -1). In that case, assume it is OK to | |
1080 | * remove. | |
d070235d | 1081 | */ |
aab15415 N |
1082 | if (rdev == 0) |
1083 | ret = -1; | |
c922221e TM |
1084 | else { |
1085 | /* | |
1086 | * The drive has already been set to 'faulty', however | |
1087 | * monitor might not have had time to process it and the | |
1088 | * drive might still have an entry in the 'holders' | |
1089 | * directory. Try a few times to avoid a false error | |
1090 | */ | |
1091 | int count = 20; | |
1092 | ||
1093 | do { | |
1094 | ret = sysfs_unique_holder(devnm, rdev); | |
1095 | if (ret < 2) | |
1096 | break; | |
239b3cc0 | 1097 | sleep_for(0, MSEC_TO_NSEC(100), true); |
c922221e TM |
1098 | } while (--count > 0); |
1099 | ||
1100 | if (ret == 0) { | |
1101 | pr_err("%s is not a member, cannot remove.\n", | |
1102 | dv->devname); | |
1103 | close(lfd); | |
1104 | return -1; | |
1105 | } | |
1106 | if (ret >= 2) { | |
1107 | pr_err("%s is still in use, cannot remove.\n", | |
1108 | dv->devname); | |
1109 | close(lfd); | |
1110 | return -1; | |
1111 | } | |
d070235d N |
1112 | } |
1113 | } | |
1114 | /* FIXME check that it is a current member */ | |
1115 | if (sysfd >= 0) { | |
1116 | /* device has been removed and we don't know | |
1117 | * the major:minor number | |
1118 | */ | |
1ab9ed2a | 1119 | err = sys_hot_remove_disk(sysfd, force); |
d070235d | 1120 | } else { |
1ab9ed2a | 1121 | err = hot_remove_disk(fd, rdev, force); |
d070235d N |
1122 | if (err && errno == ENODEV) { |
1123 | /* Old kernels rejected this if no personality | |
1124 | * is registered */ | |
4dd2df09 | 1125 | struct mdinfo *sra = sysfs_read(fd, NULL, GET_DEVS); |
d070235d N |
1126 | struct mdinfo *dv = NULL; |
1127 | if (sra) | |
1128 | dv = sra->devs; | |
1129 | for ( ; dv ; dv=dv->next) | |
1130 | if (dv->disk.major == (int)major(rdev) && | |
1131 | dv->disk.minor == (int)minor(rdev)) | |
1132 | break; | |
1133 | if (dv) | |
1134 | err = sysfs_set_str(sra, dv, | |
1135 | "state", "remove"); | |
1136 | else | |
1137 | err = -1; | |
6e8d27e7 | 1138 | sysfs_free(sra); |
d070235d N |
1139 | } |
1140 | } | |
1141 | if (err) { | |
7a862a02 | 1142 | pr_err("hot remove failed for %s: %s\n", dv->devname, |
d070235d N |
1143 | strerror(errno)); |
1144 | if (lfd >= 0) | |
1145 | close(lfd); | |
1146 | return -1; | |
1147 | } | |
1148 | if (tst->ss->external) { | |
1149 | /* | |
1150 | * Before dropping our exclusive open we make an | |
1151 | * attempt at preventing mdmon from seeing an | |
1152 | * 'add' event before reconciling this 'remove' | |
1153 | * event. | |
1154 | */ | |
4dd2df09 | 1155 | char *devnm = fd2devnm(fd); |
d070235d | 1156 | |
4dd2df09 | 1157 | if (!devnm) { |
d070235d N |
1158 | pr_err("unable to get container name\n"); |
1159 | return -1; | |
1160 | } | |
1161 | ||
4dd2df09 | 1162 | ping_manager(devnm); |
d070235d N |
1163 | } |
1164 | if (lfd >= 0) | |
1165 | close(lfd); | |
1166 | if (verbose >= 0) | |
1167 | pr_err("hot removed %s from %s\n", | |
1168 | dv->devname, devname); | |
1169 | return 1; | |
1170 | } | |
1171 | ||
70c55e36 N |
1172 | int Manage_replace(struct supertype *tst, int fd, struct mddev_dev *dv, |
1173 | unsigned long rdev, int verbose, char *devname) | |
1174 | { | |
1175 | struct mdinfo *mdi, *di; | |
1176 | if (tst->ss->external) { | |
1177 | pr_err("--replace only supported for native metadata (0.90 or 1.x)\n"); | |
1178 | return -1; | |
1179 | } | |
1180 | /* Need to find the device in sysfs and add 'want_replacement' to the | |
1181 | * status. | |
1182 | */ | |
4dd2df09 | 1183 | mdi = sysfs_read(fd, NULL, GET_DEVS); |
70c55e36 N |
1184 | if (!mdi || !mdi->devs) { |
1185 | pr_err("Cannot find status of %s to enable replacement - strange\n", | |
1186 | devname); | |
1187 | return -1; | |
1188 | } | |
1189 | for (di = mdi->devs; di; di = di->next) | |
1190 | if (di->disk.major == (int)major(rdev) && | |
1191 | di->disk.minor == (int)minor(rdev)) | |
1192 | break; | |
1193 | if (di) { | |
1194 | int rv; | |
1195 | if (di->disk.raid_disk < 0) { | |
1196 | pr_err("%s is not active and so cannot be replaced.\n", | |
1197 | dv->devname); | |
1198 | sysfs_free(mdi); | |
1199 | return -1; | |
1200 | } | |
1201 | rv = sysfs_set_str(mdi, di, | |
1202 | "state", "want_replacement"); | |
1203 | if (rv) { | |
1204 | sysfs_free(mdi); | |
1205 | pr_err("Failed to request replacement for %s\n", | |
1206 | dv->devname); | |
1207 | return -1; | |
1208 | } | |
1209 | if (verbose >= 0) | |
1210 | pr_err("Marked %s (device %d in %s) for replacement\n", | |
1211 | dv->devname, di->disk.raid_disk, devname); | |
1212 | /* If there is a matching 'with', we need to tell it which | |
1213 | * raid disk | |
1214 | */ | |
1215 | while (dv && dv->disposition != 'W') | |
1216 | dv = dv->next; | |
1217 | if (dv) { | |
1218 | dv->disposition = 'w'; | |
1219 | dv->used = di->disk.raid_disk; | |
1220 | } | |
1221 | return 1; | |
1222 | } | |
1223 | sysfs_free(mdi); | |
1224 | pr_err("%s not found in %s so cannot --replace it\n", | |
1225 | dv->devname, devname); | |
1226 | return -1; | |
1227 | } | |
1228 | ||
1229 | int Manage_with(struct supertype *tst, int fd, struct mddev_dev *dv, | |
1230 | unsigned long rdev, int verbose, char *devname) | |
1231 | { | |
1232 | struct mdinfo *mdi, *di; | |
1233 | /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */ | |
4dd2df09 | 1234 | mdi = sysfs_read(fd, NULL, GET_DEVS|GET_STATE); |
70c55e36 N |
1235 | if (!mdi || !mdi->devs) { |
1236 | pr_err("Cannot find status of %s to enable replacement - strange\n", | |
1237 | devname); | |
1238 | return -1; | |
1239 | } | |
1240 | for (di = mdi->devs; di; di = di->next) | |
1241 | if (di->disk.major == (int)major(rdev) && | |
1242 | di->disk.minor == (int)minor(rdev)) | |
1243 | break; | |
1244 | if (di) { | |
1245 | int rv; | |
1246 | if (di->disk.state & (1<<MD_DISK_FAULTY)) { | |
1247 | pr_err("%s is faulty and cannot be a replacement\n", | |
1248 | dv->devname); | |
1249 | sysfs_free(mdi); | |
1250 | return -1; | |
1251 | } | |
1252 | if (di->disk.raid_disk >= 0) { | |
1253 | pr_err("%s is active and cannot be a replacement\n", | |
1254 | dv->devname); | |
1255 | sysfs_free(mdi); | |
1256 | return -1; | |
1257 | } | |
1258 | rv = sysfs_set_num(mdi, di, | |
1259 | "slot", dv->used); | |
1260 | if (rv) { | |
1261 | sysfs_free(mdi); | |
51425978 | 1262 | pr_err("Failed to set %s as preferred replacement.\n", |
70c55e36 N |
1263 | dv->devname); |
1264 | return -1; | |
1265 | } | |
1266 | if (verbose >= 0) | |
1267 | pr_err("Marked %s in %s as replacement for device %d\n", | |
1268 | dv->devname, devname, dv->used); | |
1269 | return 1; | |
1270 | } | |
1271 | sysfs_free(mdi); | |
1272 | pr_err("%s not found in %s so cannot make it preferred replacement\n", | |
1273 | dv->devname, devname); | |
1274 | return -1; | |
1275 | } | |
1276 | ||
fc6fd406 MK |
1277 | /** |
1278 | * is_remove_safe() - Check if remove is safe. | |
1279 | * @array: Array info. | |
1280 | * @fd: Array file descriptor. | |
1281 | * @devname: Name of device to remove. | |
1282 | * @verbose: Verbose. | |
1283 | * | |
1284 | * The function determines if array will be operational | |
1285 | * after removing &devname. | |
1286 | * | |
1287 | * Return: True if array will be operational, false otherwise. | |
1288 | */ | |
1289 | bool is_remove_safe(mdu_array_info_t *array, const int fd, char *devname, const int verbose) | |
1290 | { | |
1291 | dev_t devid = devnm2devid(devname + 5); | |
1292 | struct mdinfo *mdi = sysfs_read(fd, NULL, GET_DEVS | GET_DISKS | GET_STATE); | |
1293 | ||
1294 | if (!mdi) { | |
1295 | if (verbose) | |
1296 | pr_err("Failed to read sysfs attributes for %s\n", devname); | |
1297 | return false; | |
1298 | } | |
1299 | ||
1300 | char *avail = xcalloc(array->raid_disks, sizeof(char)); | |
1301 | ||
1302 | for (mdi = mdi->devs; mdi; mdi = mdi->next) { | |
1303 | if (mdi->disk.raid_disk < 0) | |
1304 | continue; | |
1305 | if (!(mdi->disk.state & (1 << MD_DISK_SYNC))) | |
1306 | continue; | |
1307 | if (makedev(mdi->disk.major, mdi->disk.minor) == devid) | |
1308 | continue; | |
1309 | avail[mdi->disk.raid_disk] = 1; | |
1310 | } | |
1311 | sysfs_free(mdi); | |
1312 | ||
1313 | bool is_enough = enough(array->level, array->raid_disks, | |
b3e7b7eb | 1314 | array->layout, 1, avail); |
fc6fd406 MK |
1315 | |
1316 | free(avail); | |
1317 | return is_enough; | |
1318 | } | |
1319 | ||
36a70782 KT |
1320 | /** |
1321 | * Manage_subdevs() - Execute operation depending on devmode. | |
1322 | * | |
1323 | * @devname: name of the device. | |
1324 | * @fd: file descriptor. | |
1325 | * @devlist: list of sub-devices to manage. | |
1326 | * @verbose: verbose level. | |
1327 | * @test: test flag. | |
1328 | * @update: type of update. | |
1329 | * @force: force flag. | |
1330 | * | |
1331 | * This function executes operation defined by devmode | |
1332 | * for each dev from devlist. | |
1333 | * Devmode can be: | |
1334 | * 'a' - add the device | |
1335 | * 'S' - add the device as a spare - don't try re-add | |
1336 | * 'j' - add the device as a journal device | |
1337 | * 'A' - re-add the device | |
1338 | * 'r' - remove the device: HOT_REMOVE_DISK | |
1339 | * device can be 'faulty' or 'detached' in which case all | |
1340 | * matching devices are removed. | |
1341 | * 'f' - set the device faulty SET_DISK_FAULTY | |
1342 | * device can be 'detached' in which case any device that | |
1343 | * is inaccessible will be marked faulty. | |
1344 | * 'I' - remove device by using incremental fail | |
1345 | * which is executed when device is removed surprisingly. | |
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' | |
1355 | * 'F' - Another variant of 'A', where the device was faulty | |
1356 | * so must be removed from the array first. | |
1357 | * 'c' - confirm the device as found (for clustered environments) | |
1358 | * | |
1359 | * For 'f' and 'r', the device can also be a kernel-internal | |
1360 | * name such as 'sdb'. | |
1361 | * | |
1362 | * Return: 0 on success, otherwise 1 or 2. | |
1363 | */ | |
64c4757e | 1364 | int Manage_subdevs(char *devname, int fd, |
833bb0f8 | 1365 | struct mddev_dev *devlist, int verbose, int test, |
f2e8393b | 1366 | enum update_opt update, int force) |
cd29a5c8 | 1367 | { |
682c7051 | 1368 | mdu_array_info_t array; |
7a3be72f | 1369 | unsigned long long array_size; |
1d997643 | 1370 | struct mddev_dev *dv; |
cfad27a9 | 1371 | int tfd = -1; |
f6feb3fb | 1372 | struct supertype *tst = NULL; |
4725bc31 | 1373 | char *subarray = NULL; |
98d27e39 | 1374 | int sysfd = -1; |
7d2e6486 | 1375 | int count = 0; /* number of actions taken */ |
9f584691 | 1376 | struct mdinfo info; |
9465f170 | 1377 | struct mdinfo devinfo; |
9f584691 | 1378 | int frozen = 0; |
8af530b0 | 1379 | int busy = 0; |
4de90913 | 1380 | int raid_slot = -1; |
682c7051 | 1381 | |
dae13137 JS |
1382 | if (sysfs_init(&info, fd, NULL)) { |
1383 | pr_err("sysfs not availabile for %s\n", devname); | |
1384 | goto abort; | |
1385 | } | |
1386 | ||
9cd39f01 JS |
1387 | if (md_get_array_info(fd, &array)) { |
1388 | pr_err("Cannot get array info for %s\n", devname); | |
bcbb3112 | 1389 | goto abort; |
682c7051 | 1390 | } |
7bd04da9 | 1391 | /* array.size is only 32 bits and may be truncated. |
7a3be72f NB |
1392 | * So read from sysfs if possible, and record number of sectors |
1393 | */ | |
1394 | ||
1395 | array_size = get_component_size(fd); | |
1396 | if (array_size <= 0) | |
1397 | array_size = array.size * 2; | |
1398 | ||
4725bc31 | 1399 | tst = super_by_fd(fd, &subarray); |
3da92f27 | 1400 | if (!tst) { |
e7b84f9d | 1401 | pr_err("unsupport array - version %d.%d\n", |
3da92f27 | 1402 | array.major_version, array.minor_version); |
bcbb3112 | 1403 | goto abort; |
3da92f27 NB |
1404 | } |
1405 | ||
38aeaf3a | 1406 | for (dv = devlist; dv; dv = dv->next) { |
ffaf1a7e | 1407 | dev_t rdev = 0; /* device to add/remove etc */ |
38aeaf3a | 1408 | int rv; |
b47024f1 | 1409 | int mj,mn; |
4a39c6f2 | 1410 | |
4de90913 GJ |
1411 | raid_slot = -1; |
1412 | if (dv->disposition == 'c') { | |
1413 | rv = parse_cluster_confirm_arg(dv->devname, | |
1414 | &dv->devname, | |
1415 | &raid_slot); | |
d7a49369 | 1416 | if (rv) { |
4de90913 GJ |
1417 | pr_err("Could not get the devname of cluster\n"); |
1418 | goto abort; | |
1419 | } | |
1420 | } | |
1421 | ||
1d997643 N |
1422 | if (strcmp(dv->devname, "failed") == 0 || |
1423 | strcmp(dv->devname, "faulty") == 0) { | |
fc54fe7a | 1424 | if (dv->disposition != 'A' && dv->disposition != 'r') { |
7a862a02 | 1425 | pr_err("%s only meaningful with -r or --re-add, not -%c\n", |
b80da661 | 1426 | dv->devname, dv->disposition); |
bcbb3112 | 1427 | goto abort; |
b80da661 | 1428 | } |
262e3b7f N |
1429 | add_faulty(dv, fd, (dv->disposition == 'A' |
1430 | ? 'F' : 'r')); | |
1d997643 N |
1431 | continue; |
1432 | } | |
1433 | if (strcmp(dv->devname, "detached") == 0) { | |
b80da661 | 1434 | if (dv->disposition != 'r' && dv->disposition != 'f') { |
7a862a02 | 1435 | pr_err("%s only meaningful with -r of -f, not -%c\n", |
b80da661 | 1436 | dv->devname, dv->disposition); |
bcbb3112 | 1437 | goto abort; |
b80da661 | 1438 | } |
1d997643 N |
1439 | add_detached(dv, fd, dv->disposition); |
1440 | continue; | |
1441 | } | |
1442 | ||
1443 | if (strcmp(dv->devname, "missing") == 0) { | |
e9ddbb2b | 1444 | struct mddev_dev *add_devlist; |
1d997643 | 1445 | struct mddev_dev **dp; |
4de90913 GJ |
1446 | if (dv->disposition == 'c') { |
1447 | rv = ioctl(fd, CLUSTERED_DISK_NACK, NULL); | |
1448 | break; | |
1449 | } | |
1450 | ||
c8e1a230 | 1451 | if (dv->disposition != 'A') { |
7a862a02 | 1452 | pr_err("'missing' only meaningful with --re-add\n"); |
bcbb3112 | 1453 | goto abort; |
a4e13010 | 1454 | } |
1d997643 | 1455 | add_devlist = conf_get_devs(); |
a4e13010 | 1456 | if (add_devlist == NULL) { |
8e5b52cd | 1457 | pr_err("no devices to scan for missing members.\n"); |
a4e13010 N |
1458 | continue; |
1459 | } | |
1d997643 | 1460 | for (dp = &add_devlist; *dp; dp = & (*dp)->next) |
7bd04da9 | 1461 | /* 'M' (for 'missing') is like 'A' without errors */ |
1d997643 N |
1462 | (*dp)->disposition = 'M'; |
1463 | *dp = dv->next; | |
1464 | dv->next = add_devlist; | |
1465 | continue; | |
1466 | } | |
1467 | ||
64a78416 N |
1468 | if (strncmp(dv->devname, "set-", 4) == 0 && |
1469 | strlen(dv->devname) == 5) { | |
1470 | int copies; | |
1471 | ||
1472 | if (dv->disposition != 'r' && | |
1473 | dv->disposition != 'f') { | |
1474 | pr_err("'%s' only meaningful with -r or -f\n", | |
1475 | dv->devname); | |
1476 | goto abort; | |
1477 | } | |
1478 | if (array.level != 10) { | |
1479 | pr_err("'%s' only meaningful with RAID10 arrays\n", | |
1480 | dv->devname); | |
1481 | goto abort; | |
1482 | } | |
1483 | copies = ((array.layout & 0xff) * | |
1484 | ((array.layout >> 8) & 0xff)); | |
1485 | if (array.raid_disks % copies != 0 || | |
1486 | dv->devname[4] < 'A' || | |
1487 | dv->devname[4] >= 'A' + copies || | |
1488 | copies > 26) { | |
1489 | pr_err("'%s' not meaningful with this array\n", | |
1490 | dv->devname); | |
1491 | goto abort; | |
1492 | } | |
1493 | add_set(dv, fd, dv->devname[4]); | |
1494 | continue; | |
1495 | } | |
1496 | ||
1d997643 | 1497 | if (strchr(dv->devname, '/') == NULL && |
aab15415 N |
1498 | strchr(dv->devname, ':') == NULL && |
1499 | strlen(dv->devname) < 50) { | |
98d27e39 N |
1500 | /* Assume this is a kernel-internal name like 'sda1' */ |
1501 | int found = 0; | |
1502 | char dname[55]; | |
461fae7e KT |
1503 | if (dv->disposition != 'r' && dv->disposition != 'f' && |
1504 | dv->disposition != 'I') { | |
1505 | pr_err("%s only meaningful with -r, -f or -I, not -%c\n", | |
98d27e39 | 1506 | dv->devname, dv->disposition); |
bcbb3112 | 1507 | goto abort; |
98d27e39 N |
1508 | } |
1509 | ||
1510 | sprintf(dname, "dev-%s", dv->devname); | |
4dd2df09 | 1511 | sysfd = sysfs_open(fd2devnm(fd), dname, "block/dev"); |
98d27e39 N |
1512 | if (sysfd >= 0) { |
1513 | char dn[20]; | |
98d27e39 N |
1514 | if (sysfs_fd_get_str(sysfd, dn, 20) > 0 && |
1515 | sscanf(dn, "%d:%d", &mj,&mn) == 2) { | |
5dffd09d | 1516 | rdev = makedev(mj,mn); |
98d27e39 N |
1517 | found = 1; |
1518 | } | |
1519 | close(sysfd); | |
1520 | sysfd = -1; | |
1521 | } | |
1522 | if (!found) { | |
4dd2df09 | 1523 | sysfd = sysfs_open(fd2devnm(fd), dname, "state"); |
98d27e39 | 1524 | if (sysfd < 0) { |
7a862a02 | 1525 | pr_err("%s does not appear to be a component of %s\n", |
98d27e39 | 1526 | dv->devname, devname); |
bcbb3112 | 1527 | goto abort; |
98d27e39 N |
1528 | } |
1529 | } | |
fc54fe7a JS |
1530 | } else if ((dv->disposition == 'r' || |
1531 | dv->disposition == 'f') && | |
1532 | get_maj_min(dv->devname, &mj, &mn)) { | |
b47024f1 N |
1533 | /* for 'fail' and 'remove', the device might |
1534 | * not exist. | |
1535 | */ | |
1536 | rdev = makedev(mj, mn); | |
b80da661 | 1537 | } else { |
c7b47447 | 1538 | tfd = dev_open(dv->devname, O_RDONLY); |
cc5083d1 | 1539 | if (tfd >= 0) { |
0a6bff09 | 1540 | fstat_is_blkdev(tfd, dv->devname, &rdev); |
cc5083d1 JS |
1541 | close(tfd); |
1542 | } else { | |
5fe7f5f7 | 1543 | int open_err = errno; |
9e04ac1c | 1544 | if (!stat_is_blkdev(dv->devname, &rdev)) { |
5fe7f5f7 N |
1545 | if (dv->disposition == 'M') |
1546 | /* non-fatal. Also improbable */ | |
1547 | continue; | |
5fe7f5f7 N |
1548 | goto abort; |
1549 | } | |
1550 | if (dv->disposition == 'r') | |
1551 | /* Be happy, the stat worked, that is | |
1552 | * enough for --remove | |
1553 | */ | |
1554 | ; | |
1555 | else { | |
1d997643 N |
1556 | if (dv->disposition == 'M') |
1557 | /* non-fatal */ | |
1558 | continue; | |
839f27a3 | 1559 | pr_err("Cannot open %s: %s\n", |
5fe7f5f7 | 1560 | dv->devname, strerror(open_err)); |
bcbb3112 | 1561 | goto abort; |
5a9de8db | 1562 | } |
b80da661 | 1563 | } |
682c7051 | 1564 | } |
cd29a5c8 | 1565 | switch(dv->disposition){ |
682c7051 | 1566 | default: |
e7b84f9d | 1567 | pr_err("internal error - devmode[%s]=%d\n", |
c913b90e | 1568 | dv->devname, dv->disposition); |
bcbb3112 | 1569 | goto abort; |
682c7051 | 1570 | case 'a': |
f33a71f1 | 1571 | case 'S': /* --add-spare */ |
01290056 | 1572 | case 'j': /* --add-journal */ |
c8e1a230 | 1573 | case 'A': |
262e3b7f N |
1574 | case 'M': /* --re-add missing */ |
1575 | case 'F': /* --re-add faulty */ | |
4de90913 | 1576 | case 'c': /* --cluster-confirm */ |
4a39c6f2 | 1577 | /* add the device */ |
4725bc31 | 1578 | if (subarray) { |
7a862a02 | 1579 | pr_err("Cannot add disks to a \'member\' array, perform this operation on the parent container\n"); |
bcbb3112 | 1580 | goto abort; |
f94d52f4 | 1581 | } |
9465f170 GJ |
1582 | |
1583 | /* Let's first try to write re-add to sysfs */ | |
1584 | if (rdev != 0 && | |
1585 | (dv->disposition == 'A' || dv->disposition == 'F')) { | |
1586 | sysfs_init_dev(&devinfo, rdev); | |
1587 | if (sysfs_set_str(&info, &devinfo, "state", "re-add") == 0) { | |
1588 | pr_err("re-add %s to %s succeed\n", | |
1589 | dv->devname, info.sys_name); | |
1590 | break; | |
1591 | } | |
1592 | } | |
1593 | ||
262e3b7f N |
1594 | if (dv->disposition == 'F') |
1595 | /* Need to remove first */ | |
1ab9ed2a | 1596 | hot_remove_disk(fd, rdev, force); |
f277ce36 | 1597 | /* Make sure it isn't in use (in 2.6 or later) */ |
1d997643 | 1598 | tfd = dev_open(dv->devname, O_RDONLY|O_EXCL); |
38aeaf3a N |
1599 | if (tfd >= 0) { |
1600 | /* We know no-one else is using it. We'll | |
1601 | * need non-exclusive access to add it, so | |
1602 | * do that now. | |
1603 | */ | |
1604 | close(tfd); | |
1605 | tfd = dev_open(dv->devname, O_RDONLY); | |
1011e834 | 1606 | } |
0fbf459d | 1607 | if (tfd < 0) { |
1d997643 N |
1608 | if (dv->disposition == 'M') |
1609 | continue; | |
e7b84f9d | 1610 | pr_err("Cannot open %s: %s\n", |
d7eaf49f | 1611 | dv->devname, strerror(errno)); |
bcbb3112 | 1612 | goto abort; |
d7eaf49f | 1613 | } |
9f584691 N |
1614 | if (!frozen) { |
1615 | if (sysfs_freeze_array(&info) == 1) | |
1616 | frozen = 1; | |
1617 | else | |
1618 | frozen = -1; | |
1619 | } | |
38aeaf3a N |
1620 | rv = Manage_add(fd, tfd, dv, tst, &array, |
1621 | force, verbose, devname, update, | |
4de90913 | 1622 | rdev, array_size, raid_slot); |
38aeaf3a N |
1623 | close(tfd); |
1624 | tfd = -1; | |
1625 | if (rv < 0) | |
bcbb3112 | 1626 | goto abort; |
38aeaf3a N |
1627 | if (rv > 0) |
1628 | count++; | |
682c7051 NB |
1629 | break; |
1630 | ||
1631 | case 'r': | |
1632 | /* hot remove */ | |
4725bc31 | 1633 | if (subarray) { |
7a862a02 | 1634 | pr_err("Cannot remove disks from a \'member\' array, perform this operation on the parent container\n"); |
d070235d N |
1635 | rv = -1; |
1636 | } else | |
1637 | rv = Manage_remove(tst, fd, dv, sysfd, | |
1ab9ed2a | 1638 | rdev, verbose, force, |
d070235d N |
1639 | devname); |
1640 | if (sysfd >= 0) | |
98d27e39 | 1641 | close(sysfd); |
d070235d N |
1642 | sysfd = -1; |
1643 | if (rv < 0) | |
bcbb3112 | 1644 | goto abort; |
d070235d N |
1645 | if (rv > 0) |
1646 | count++; | |
682c7051 NB |
1647 | break; |
1648 | ||
1649 | case 'f': /* set faulty */ | |
fc6fd406 MK |
1650 | if (!is_remove_safe(&array, fd, dv->devname, verbose)) { |
1651 | pr_err("Cannot remove %s from %s, array will be failed.\n", | |
1652 | dv->devname, devname); | |
1653 | if (sysfd >= 0) | |
1654 | close(sysfd); | |
1655 | goto abort; | |
1656 | } | |
461fae7e | 1657 | case 'I': /* incremental fail */ |
98d27e39 N |
1658 | if ((sysfd >= 0 && write(sysfd, "faulty", 6) != 6) || |
1659 | (sysfd < 0 && ioctl(fd, SET_DISK_FAULTY, | |
5dffd09d | 1660 | rdev))) { |
8af530b0 N |
1661 | if (errno == EBUSY) |
1662 | busy = 1; | |
e7b84f9d | 1663 | pr_err("set device faulty failed for %s: %s\n", |
1d997643 | 1664 | dv->devname, strerror(errno)); |
98d27e39 N |
1665 | if (sysfd >= 0) |
1666 | close(sysfd); | |
bcbb3112 | 1667 | goto abort; |
682c7051 | 1668 | } |
98d27e39 N |
1669 | if (sysfd >= 0) |
1670 | close(sysfd); | |
1671 | sysfd = -1; | |
7d2e6486 | 1672 | count++; |
dab6685f | 1673 | if (verbose >= 0) |
e7b84f9d | 1674 | pr_err("set %s faulty in %s\n", |
1d997643 | 1675 | dv->devname, devname); |
682c7051 | 1676 | break; |
70c55e36 N |
1677 | case 'R': /* Mark as replaceable */ |
1678 | if (subarray) { | |
7a862a02 | 1679 | pr_err("Cannot replace disks in a \'member\' array, perform this operation on the parent container\n"); |
70c55e36 N |
1680 | rv = -1; |
1681 | } else { | |
1682 | if (!frozen) { | |
1683 | if (sysfs_freeze_array(&info) == 1) | |
1684 | frozen = 1; | |
1685 | else | |
1686 | frozen = -1; | |
1687 | } | |
1688 | rv = Manage_replace(tst, fd, dv, | |
5dffd09d | 1689 | rdev, verbose, |
70c55e36 N |
1690 | devname); |
1691 | } | |
1692 | if (rv < 0) | |
1693 | goto abort; | |
1694 | if (rv > 0) | |
1695 | count++; | |
1696 | break; | |
1697 | case 'W': /* --with device that doesn't match */ | |
1698 | pr_err("No matching --replace device for --with %s\n", | |
1699 | dv->devname); | |
1700 | goto abort; | |
1701 | case 'w': /* --with device which was matched */ | |
1702 | rv = Manage_with(tst, fd, dv, | |
5dffd09d | 1703 | rdev, verbose, devname); |
70c55e36 N |
1704 | if (rv < 0) |
1705 | goto abort; | |
1706 | break; | |
682c7051 NB |
1707 | } |
1708 | } | |
f6feb3fb | 1709 | free(tst); |
9f584691 N |
1710 | if (frozen > 0) |
1711 | sysfs_set_str(&info, NULL, "sync_action","idle"); | |
7d2e6486 N |
1712 | if (test && count == 0) |
1713 | return 2; | |
682c7051 | 1714 | return 0; |
bcbb3112 N |
1715 | |
1716 | abort: | |
f6feb3fb | 1717 | free(tst); |
9f584691 N |
1718 | if (frozen > 0) |
1719 | sysfs_set_str(&info, NULL, "sync_action","idle"); | |
8af530b0 | 1720 | return !test && busy ? 2 : 1; |
64c4757e | 1721 | } |
1f48664b NB |
1722 | |
1723 | int autodetect(void) | |
1724 | { | |
1725 | /* Open any md device, and issue the RAID_AUTORUN ioctl */ | |
1726 | int rv = 1; | |
1727 | int fd = dev_open("9:0", O_RDONLY); | |
1728 | if (fd >= 0) { | |
1729 | if (ioctl(fd, RAID_AUTORUN, 0) == 0) | |
1730 | rv = 0; | |
1731 | close(fd); | |
1732 | } | |
1733 | return rv; | |
1734 | } | |
aa534678 | 1735 | |
f2e8393b MK |
1736 | int Update_subarray(char *dev, char *subarray, enum update_opt update, |
1737 | struct mddev_ident *ident, int verbose) | |
aa534678 DW |
1738 | { |
1739 | struct supertype supertype, *st = &supertype; | |
1740 | int fd, rv = 2; | |
70f1ff42 | 1741 | struct mdinfo *info = NULL; |
f2e8393b | 1742 | char *update_verb = map_num(update_options, update); |
aa534678 DW |
1743 | |
1744 | memset(st, 0, sizeof(*st)); | |
aa534678 | 1745 | |
ba728be7 | 1746 | fd = open_subarray(dev, subarray, st, verbose < 0); |
aa534678 DW |
1747 | if (fd < 0) |
1748 | return 2; | |
1749 | ||
1750 | if (!st->ss->update_subarray) { | |
ba728be7 | 1751 | if (verbose >= 0) |
e7b84f9d N |
1752 | pr_err("Operation not supported for %s metadata\n", |
1753 | st->ss->name); | |
aa534678 DW |
1754 | goto free_super; |
1755 | } | |
1756 | ||
db10eab6 MK |
1757 | if (is_subarray_active(subarray, st->devnm)) { |
1758 | if (verbose >= 0) | |
1759 | pr_err("Subarray %s in %s is active, cannot update %s\n", | |
f2e8393b | 1760 | subarray, dev, update_verb); |
db10eab6 MK |
1761 | goto free_super; |
1762 | } | |
1763 | ||
4dd2df09 | 1764 | if (mdmon_running(st->devnm)) |
aa534678 DW |
1765 | st->update_tail = &st->updates; |
1766 | ||
70f1ff42 LF |
1767 | info = st->ss->container_content(st, subarray); |
1768 | ||
f2e8393b | 1769 | if (update == UOPT_PPL && !is_level456(info->array.level)) { |
70f1ff42 LF |
1770 | pr_err("RWH policy ppl is supported only for raid4, raid5 and raid6.\n"); |
1771 | goto free_super; | |
1772 | } | |
1773 | ||
f2e8393b | 1774 | rv = st->ss->update_subarray(st, subarray, update, ident); |
aa534678 DW |
1775 | |
1776 | if (rv) { | |
ba728be7 | 1777 | if (verbose >= 0) |
e7b84f9d | 1778 | pr_err("Failed to update %s of subarray-%s in %s\n", |
f2e8393b | 1779 | update_verb, subarray, dev); |
aa534678 DW |
1780 | } else if (st->update_tail) |
1781 | flush_metadata_updates(st); | |
1782 | else | |
1783 | st->ss->sync_metadata(st); | |
1784 | ||
f2e8393b | 1785 | if (rv == 0 && update == UOPT_NAME && verbose >= 0) |
e7b84f9d N |
1786 | pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n", |
1787 | subarray, dev); | |
aa534678 | 1788 | |
70f1ff42 LF |
1789 | free_super: |
1790 | if (info) | |
1791 | free(info); | |
aa534678 DW |
1792 | st->ss->free_super(st); |
1793 | close(fd); | |
1794 | ||
1795 | return rv; | |
1796 | } | |
d52bb542 | 1797 | |
7bd04da9 N |
1798 | /* Move spare from one array to another If adding to destination array fails |
1799 | * add back to original array. | |
d52bb542 AC |
1800 | * Returns 1 on success, 0 on failure */ |
1801 | int move_spare(char *from_devname, char *to_devname, dev_t devid) | |
1802 | { | |
1803 | struct mddev_dev devlist; | |
1804 | char devname[20]; | |
1805 | ||
1806 | /* try to remove and add */ | |
1807 | int fd1 = open(to_devname, O_RDONLY); | |
1808 | int fd2 = open(from_devname, O_RDONLY); | |
1809 | ||
1810 | if (fd1 < 0 || fd2 < 0) { | |
9cf361f8 JS |
1811 | if (fd1 >= 0) |
1812 | close(fd1); | |
1813 | if (fd2 >= 0) | |
1814 | close(fd2); | |
d52bb542 AC |
1815 | return 0; |
1816 | } | |
1817 | ||
1818 | devlist.next = NULL; | |
1819 | devlist.used = 0; | |
e22fe3ae N |
1820 | devlist.writemostly = FlagDefault; |
1821 | devlist.failfast = FlagDefault; | |
d52bb542 AC |
1822 | devlist.devname = devname; |
1823 | sprintf(devname, "%d:%d", major(devid), minor(devid)); | |
1824 | ||
1825 | devlist.disposition = 'r'; | |
f2e8393b | 1826 | if (Manage_subdevs(from_devname, fd2, &devlist, -1, 0, UOPT_UNDEFINED, 0) == 0) { |
d52bb542 | 1827 | devlist.disposition = 'a'; |
9cf361f8 | 1828 | if (Manage_subdevs(to_devname, fd1, &devlist, -1, 0, |
f2e8393b | 1829 | UOPT_UNDEFINED, 0) == 0) { |
d52bb542 AC |
1830 | /* make sure manager is aware of changes */ |
1831 | ping_manager(to_devname); | |
1832 | ping_manager(from_devname); | |
1833 | close(fd1); | |
1834 | close(fd2); | |
1835 | return 1; | |
1836 | } | |
9cf361f8 JS |
1837 | else |
1838 | Manage_subdevs(from_devname, fd2, &devlist, | |
f2e8393b | 1839 | -1, 0, UOPT_UNDEFINED, 0); |
d52bb542 AC |
1840 | } |
1841 | close(fd1); | |
1842 | close(fd2); | |
1843 | return 0; | |
1844 | } |