]>
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 | 29 | |
1011e834 N |
30 | #define REGISTER_DEV _IO (MD_MAJOR, 1) |
31 | #define START_MD _IO (MD_MAJOR, 2) | |
32 | #define STOP_MD _IO (MD_MAJOR, 3) | |
82b27616 | 33 | |
64c4757e NB |
34 | int Manage_ro(char *devname, int fd, int readonly) |
35 | { | |
682c7051 NB |
36 | /* switch to readonly or rw |
37 | * | |
38 | * requires >= 0.90.0 | |
39 | * first check that array is runing | |
40 | * use RESTART_ARRAY_RW or STOP_ARRAY_RO | |
41 | * | |
42 | */ | |
43 | mdu_array_info_t array; | |
0e600426 | 44 | #ifndef MDASSEMBLE |
e9dd1598 | 45 | struct mdinfo *mdi; |
0e600426 | 46 | #endif |
b73e45ae | 47 | int rv = 0; |
aba69144 | 48 | |
682c7051 | 49 | if (md_get_version(fd) < 9000) { |
e7b84f9d | 50 | pr_err("need md driver version 0.90.0 or later\n"); |
682c7051 NB |
51 | return 1; |
52 | } | |
0e600426 | 53 | #ifndef MDASSEMBLE |
7bd04da9 | 54 | /* If this is an externally-managed array, we need to modify the |
e9dd1598 N |
55 | * metadata_version so that mdmon doesn't undo our change. |
56 | */ | |
4dd2df09 | 57 | mdi = sysfs_read(fd, NULL, GET_LEVEL|GET_VERSION); |
e9dd1598 N |
58 | if (mdi && |
59 | mdi->array.major_version == -1 && | |
e9dd1598 N |
60 | is_subarray(mdi->text_version)) { |
61 | char vers[64]; | |
62 | strcpy(vers, "external:"); | |
63 | strcat(vers, mdi->text_version); | |
64 | if (readonly > 0) { | |
65 | int rv; | |
66 | /* We set readonly ourselves. */ | |
67 | vers[9] = '-'; | |
68 | sysfs_set_str(mdi, NULL, "metadata_version", vers); | |
69 | ||
70 | close(fd); | |
71 | rv = sysfs_set_str(mdi, NULL, "array_state", "readonly"); | |
72 | ||
73 | if (rv < 0) { | |
e7b84f9d | 74 | pr_err("failed to set readonly for %s: %s\n", |
e9dd1598 N |
75 | devname, strerror(errno)); |
76 | ||
77 | vers[9] = mdi->text_version[0]; | |
78 | sysfs_set_str(mdi, NULL, "metadata_version", vers); | |
b73e45ae JS |
79 | rv = 1; |
80 | goto out; | |
e9dd1598 N |
81 | } |
82 | } else { | |
83 | char *cp; | |
84 | /* We cannot set read/write - must signal mdmon */ | |
85 | vers[9] = '/'; | |
86 | sysfs_set_str(mdi, NULL, "metadata_version", vers); | |
87 | ||
88 | cp = strchr(vers+10, '/'); | |
1471b8b1 | 89 | if (cp) |
e9dd1598 N |
90 | *cp = 0; |
91 | ping_monitor(vers+10); | |
9ea5a252 DW |
92 | if (mdi->array.level <= 0) |
93 | sysfs_set_str(mdi, NULL, "array_state", "active"); | |
e9dd1598 | 94 | } |
b73e45ae | 95 | goto out; |
e9dd1598 | 96 | } |
0e600426 | 97 | #endif |
682c7051 | 98 | if (ioctl(fd, GET_ARRAY_INFO, &array)) { |
e7b84f9d | 99 | pr_err("%s does not appear to be active.\n", |
682c7051 | 100 | devname); |
b73e45ae JS |
101 | rv = 1; |
102 | goto out; | |
682c7051 | 103 | } |
aba69144 | 104 | |
7bd04da9 | 105 | if (readonly > 0) { |
682c7051 | 106 | if (ioctl(fd, STOP_ARRAY_RO, NULL)) { |
e7b84f9d | 107 | pr_err("failed to set readonly for %s: %s\n", |
682c7051 | 108 | devname, strerror(errno)); |
b73e45ae JS |
109 | rv = 1; |
110 | goto out; | |
682c7051 NB |
111 | } |
112 | } else if (readonly < 0) { | |
113 | if (ioctl(fd, RESTART_ARRAY_RW, NULL)) { | |
e7b84f9d | 114 | pr_err("failed to set writable for %s: %s\n", |
682c7051 | 115 | devname, strerror(errno)); |
b73e45ae JS |
116 | rv = 1; |
117 | goto out; | |
682c7051 NB |
118 | } |
119 | } | |
b73e45ae JS |
120 | out: |
121 | #ifndef MDASSEMBLE | |
6e8d27e7 | 122 | sysfs_free(mdi); |
b73e45ae JS |
123 | #endif |
124 | return rv; | |
64c4757e NB |
125 | } |
126 | ||
435d4ebb LB |
127 | #ifndef MDASSEMBLE |
128 | ||
4dd2df09 | 129 | static void remove_devices(char *devnm, char *path) |
4ccad7b1 | 130 | { |
7bd04da9 | 131 | /* |
b1702f48 N |
132 | * Remove names at 'path' - possibly with |
133 | * partition suffixes - which link to the 'standard' | |
4dd2df09 | 134 | * name for devnm. These were probably created |
b1702f48 | 135 | * by mdadm when the array was assembled. |
4ccad7b1 N |
136 | */ |
137 | char base[40]; | |
138 | char *path2; | |
139 | char link[1024]; | |
140 | int n; | |
141 | int part; | |
142 | char *be; | |
143 | char *pe; | |
144 | ||
b1702f48 N |
145 | if (!path) |
146 | return; | |
147 | ||
4dd2df09 | 148 | sprintf(base, "/dev/%s", devnm); |
4ccad7b1 | 149 | be = base + strlen(base); |
b1702f48 | 150 | |
503975b9 | 151 | path2 = xmalloc(strlen(path)+20); |
b1702f48 N |
152 | strcpy(path2, path); |
153 | pe = path2 + strlen(path2); | |
7bd04da9 | 154 | |
4ccad7b1 N |
155 | for (part = 0; part < 16; part++) { |
156 | if (part) { | |
157 | sprintf(be, "p%d", part); | |
b1702f48 N |
158 | |
159 | if (isdigit(pe[-1])) | |
160 | sprintf(pe, "p%d", part); | |
161 | else | |
162 | sprintf(pe, "%d", part); | |
4ccad7b1 | 163 | } |
b1702f48 | 164 | n = readlink(path2, link, sizeof(link)); |
d9ca03e9 | 165 | if (n > 0 && (int)strlen(base) == n && |
b1702f48 N |
166 | strncmp(link, base, n) == 0) |
167 | unlink(path2); | |
4ccad7b1 | 168 | } |
0eb26465 | 169 | free(path2); |
4ccad7b1 | 170 | } |
4ccad7b1 | 171 | |
d5a40416 | 172 | int Manage_run(char *devname, int fd, struct context *c) |
64c4757e | 173 | { |
fe7e0e64 N |
174 | /* Run the array. Array must already be configured |
175 | * Requires >= 0.90.0 | |
682c7051 | 176 | */ |
d3786cdc | 177 | char nm[32], *nmp; |
82b27616 | 178 | |
fe7e0e64 N |
179 | if (md_get_version(fd) < 9000) { |
180 | pr_err("need md driver version 0.90.0 or later\n"); | |
181 | return 1; | |
182 | } | |
d3786cdc N |
183 | nmp = fd2devnm(fd); |
184 | if (!nmp) { | |
185 | pr_err("Cannot find %s in sysfs!!\n", devname); | |
fe7e0e64 N |
186 | return 1; |
187 | } | |
d3786cdc | 188 | strcpy(nm, nmp); |
d5a40416 | 189 | return IncrementalScan(c, nm); |
fe7e0e64 N |
190 | } |
191 | ||
192 | int Manage_stop(char *devname, int fd, int verbose, int will_retry) | |
193 | { | |
194 | /* Stop the array. Array must already be configured | |
195 | * 'will_retry' means that error messages are not wanted. | |
196 | */ | |
197 | int rv = 0; | |
198 | struct map_ent *map = NULL; | |
199 | struct mdinfo *mdi; | |
200 | char devnm[32]; | |
201 | char container[32]; | |
202 | int err; | |
203 | int count; | |
2eba8496 N |
204 | char buf[32]; |
205 | unsigned long long rd1, rd2; | |
fe7e0e64 | 206 | |
ba728be7 N |
207 | if (will_retry && verbose == 0) |
208 | verbose = -1; | |
209 | ||
fe7e0e64 | 210 | if (md_get_version(fd) < 9000) { |
7bd04da9 N |
211 | if (ioctl(fd, STOP_MD, 0) == 0) |
212 | return 0; | |
7a862a02 | 213 | pr_err("stopping device %s failed: %s\n", |
7bd04da9 N |
214 | devname, strerror(errno)); |
215 | return 1; | |
82b27616 | 216 | } |
aba69144 | 217 | |
fe7e0e64 N |
218 | strcpy(devnm, fd2devnm(fd)); |
219 | /* Get EXCL access first. If this fails, then attempting | |
220 | * to stop is probably a bad idea. | |
221 | */ | |
2eba8496 | 222 | mdi = sysfs_read(fd, NULL, GET_LEVEL|GET_COMPONENT|GET_VERSION); |
fe7e0e64 N |
223 | if (mdi && is_subarray(mdi->text_version)) { |
224 | char *sl; | |
225 | strncpy(container, mdi->text_version+1, sizeof(container)); | |
226 | container[sizeof(container)-1] = 0; | |
227 | sl = strchr(container, '/'); | |
228 | if (sl) | |
229 | *sl = 0; | |
230 | } else | |
231 | container[0] = 0; | |
232 | close(fd); | |
233 | count = 5; | |
52b6ccad | 234 | while (((fd = ((devname[0] == '/') |
fe7e0e64 N |
235 | ?open(devname, O_RDONLY|O_EXCL) |
236 | :open_dev_flags(devnm, O_RDONLY|O_EXCL))) < 0 | |
237 | || strcmp(fd2devnm(fd), devnm) != 0) | |
238 | && container[0] | |
239 | && mdmon_running(container) | |
240 | && count) { | |
932be627 N |
241 | /* Can't open, so something might be wrong. However it |
242 | * is a container, so we might be racing with mdmon, so | |
243 | * retry for a bit. | |
244 | */ | |
fe7e0e64 N |
245 | if (fd >= 0) |
246 | close(fd); | |
247 | flush_mdmon(container); | |
248 | count--; | |
682c7051 | 249 | } |
fe7e0e64 N |
250 | if (fd < 0 || strcmp(fd2devnm(fd), devnm) != 0) { |
251 | if (fd >= 0) | |
252 | close(fd); | |
ba728be7 | 253 | if (verbose >= 0) |
7a862a02 | 254 | pr_err("Cannot get exclusive access to %s:Perhaps a running process, mounted filesystem or active volume group?\n", |
fe7e0e64 N |
255 | devname); |
256 | return 1; | |
257 | } | |
932be627 N |
258 | /* If this is an mdmon managed array, just write 'inactive' |
259 | * to the array state and let mdmon clear up. | |
260 | */ | |
fe7e0e64 N |
261 | if (mdi && |
262 | mdi->array.level > 0 && | |
263 | is_subarray(mdi->text_version)) { | |
eb0af526 | 264 | int err; |
fe7e0e64 | 265 | /* This is mdmon managed. */ |
eb0af526 | 266 | close(fd); |
daf7a3ce | 267 | |
932be627 | 268 | /* As we had an O_EXCL open, any use of the device |
eb0af526 N |
269 | * which blocks STOP_ARRAY is probably a transient use, |
270 | * so it is reasonable to retry for a while - 5 seconds. | |
271 | */ | |
fe7e0e64 N |
272 | count = 25; |
273 | while (count && | |
274 | (err = sysfs_set_str(mdi, NULL, | |
275 | "array_state", | |
276 | "inactive")) < 0 | |
eb0af526 N |
277 | && errno == EBUSY) { |
278 | usleep(200000); | |
fe7e0e64 | 279 | count--; |
eb0af526 | 280 | } |
fe7e0e64 N |
281 | if (err) { |
282 | if (verbose >= 0) | |
e7b84f9d N |
283 | pr_err("failed to stop array %s: %s\n", |
284 | devname, strerror(errno)); | |
bccd8153 JS |
285 | rv = 1; |
286 | goto out; | |
682c7051 | 287 | } |
fe7e0e64 N |
288 | |
289 | /* Give monitor a chance to act */ | |
290 | ping_monitor(mdi->text_version); | |
291 | ||
292 | fd = open_dev_excl(devnm); | |
293 | if (fd < 0) { | |
294 | if (verbose >= 0) | |
7a862a02 | 295 | pr_err("failed to completely stop %s: Device is busy\n", |
fe7e0e64 N |
296 | devname); |
297 | rv = 1; | |
298 | goto out; | |
299 | } | |
300 | } else if (mdi && | |
301 | mdi->array.major_version == -1 && | |
302 | mdi->array.minor_version == -2 && | |
303 | !is_subarray(mdi->text_version)) { | |
304 | struct mdstat_ent *mds, *m; | |
305 | /* container, possibly mdmon-managed. | |
306 | * Make sure mdmon isn't opening it, which | |
307 | * would interfere with the 'stop' | |
97590376 | 308 | */ |
fe7e0e64 | 309 | ping_monitor(mdi->sys_name); |
daf7a3ce | 310 | |
fe7e0e64 N |
311 | /* now check that there are no existing arrays |
312 | * which are members of this array | |
313 | */ | |
314 | mds = mdstat_read(0, 0); | |
315 | for (m = mds; m; m = m->next) | |
316 | if (m->metadata_version && | |
317 | strncmp(m->metadata_version, "external:", 9)==0 && | |
318 | metadata_container_matches(m->metadata_version+9, | |
319 | devnm)) { | |
320 | if (verbose >= 0) | |
7a862a02 | 321 | pr_err("Cannot stop container %s: member %s still active\n", |
9581efb1 | 322 | devname, m->devnm); |
fe7e0e64 N |
323 | free_mdstat(mds); |
324 | rv = 1; | |
325 | goto out; | |
326 | } | |
327 | } | |
328 | ||
2eba8496 N |
329 | /* If the array is undergoing a reshape which changes the number |
330 | * of devices, then it would be nice to stop it at a point where | |
331 | * it has completed a full number of stripes in both old and | |
332 | * new layouts as this will allow the reshape to be reverted. | |
333 | * So if 'sync_action' is "reshape" and 'raid_disks' shows two | |
334 | * different numbers, then | |
335 | * - freeze reshape | |
336 | * - set sync_max to next multiple of both data_disks and | |
337 | * chunk sizes (or next but one) | |
338 | * - unfreeze reshape | |
339 | * - wait on 'sync_completed' for that point to be reached. | |
340 | */ | |
341 | if (mdi && (mdi->array.level >= 4 && mdi->array.level <= 6) && | |
342 | sysfs_attribute_available(mdi, NULL, "sync_action") && | |
343 | sysfs_attribute_available(mdi, NULL, "reshape_direction") && | |
344 | sysfs_get_str(mdi, NULL, "sync_action", buf, 20) > 0 && | |
345 | strcmp(buf, "reshape\n") == 0 && | |
e3e0d0a8 | 346 | sysfs_get_two(mdi, NULL, "raid_disks", &rd1, &rd2) == 2) { |
2eba8496 N |
347 | unsigned long long position, curr; |
348 | unsigned long long chunk1, chunk2; | |
349 | unsigned long long rddiv, chunkdiv; | |
350 | unsigned long long sectors; | |
5509dc44 N |
351 | unsigned long long sync_max, old_sync_max; |
352 | unsigned long long completed; | |
2eba8496 N |
353 | int backwards = 0; |
354 | int delay; | |
355 | int scfd; | |
356 | ||
e3e0d0a8 N |
357 | delay = 40; |
358 | while (rd1 > rd2 && delay > 0 && | |
359 | sysfs_get_ll(mdi, NULL, "sync_max", &old_sync_max) == 0) { | |
360 | /* must be in the critical section - wait a bit */ | |
361 | delay -= 1; | |
362 | usleep(100000); | |
363 | } | |
364 | ||
365 | if (sysfs_set_str(mdi, NULL, "sync_action", "frozen") != 0) | |
366 | goto done; | |
367 | /* Array is frozen */ | |
368 | ||
2eba8496 N |
369 | rd1 -= mdi->array.level == 6 ? 2 : 1; |
370 | rd2 -= mdi->array.level == 6 ? 2 : 1; | |
371 | sysfs_get_str(mdi, NULL, "reshape_direction", buf, sizeof(buf)); | |
372 | if (strncmp(buf, "back", 4) == 0) | |
373 | backwards = 1; | |
e3e0d0a8 N |
374 | if (sysfs_get_ll(mdi, NULL, "reshape_position", &position) != 0) { |
375 | /* reshape must have finished now */ | |
376 | sysfs_set_str(mdi, NULL, "sync_action", "idle"); | |
377 | goto done; | |
378 | } | |
2eba8496 N |
379 | sysfs_get_two(mdi, NULL, "chunk_size", &chunk1, &chunk2); |
380 | chunk1 /= 512; | |
381 | chunk2 /= 512; | |
382 | rddiv = GCD(rd1, rd2); | |
383 | chunkdiv = GCD(chunk1, chunk2); | |
384 | sectors = (chunk1/chunkdiv) * chunk2 * (rd1/rddiv) * rd2; | |
385 | ||
386 | if (backwards) { | |
387 | /* Need to subtract 'reshape_position' from | |
388 | * array size to get equivalent of sync_max. | |
389 | * Size calculation based on raid5_size in kernel. | |
390 | */ | |
391 | unsigned long long size = mdi->component_size; | |
392 | size &= ~(chunk1-1); | |
393 | size &= ~(chunk2-1); | |
394 | /* rd1 must be smaller */ | |
932be627 N |
395 | /* Reshape may have progressed further backwards than |
396 | * recorded, so target even further back (hence "-1") | |
397 | */ | |
5509dc44 | 398 | position = (position / sectors - 1) * sectors; |
932be627 N |
399 | /* rd1 is always the conversion factor between 'sync' |
400 | * position and 'reshape' position. | |
401 | * We read 1 "new" stripe worth of data from where-ever, | |
402 | * and when write out that full stripe. | |
403 | */ | |
5509dc44 | 404 | sync_max = size - position/rd1; |
2eba8496 | 405 | } else { |
932be627 N |
406 | /* Reshape will very likely be beyond position, and it may |
407 | * be too late to stop at '+1', so aim for '+2' | |
408 | */ | |
5509dc44 N |
409 | position = (position / sectors + 2) * sectors; |
410 | sync_max = position/rd1; | |
2eba8496 | 411 | } |
5509dc44 N |
412 | if (sysfs_get_ll(mdi, NULL, "sync_max", &old_sync_max) < 0) |
413 | old_sync_max = mdi->component_size; | |
414 | /* Must not advance sync_max as that could confuse | |
415 | * the reshape monitor */ | |
416 | if (sync_max < old_sync_max) | |
417 | sysfs_set_num(mdi, NULL, "sync_max", sync_max); | |
2eba8496 N |
418 | sysfs_set_str(mdi, NULL, "sync_action", "idle"); |
419 | ||
420 | /* That should have set things going again. Now we | |
3afaff93 | 421 | * wait a little while (3 second max) for sync_completed |
2eba8496 | 422 | * to reach the target. |
3afaff93 N |
423 | * The reshape process can block for 500msec if |
424 | * the sync speed limit is hit, so we need to wait | |
425 | * a lot longer than that. 1 second is usually | |
426 | * enough. 3 is safe. | |
2eba8496 | 427 | */ |
3afaff93 | 428 | delay = 3000; |
2eba8496 | 429 | scfd = sysfs_open(mdi->sys_name, NULL, "sync_completed"); |
3afaff93 | 430 | while (scfd >= 0 && delay > 0 && old_sync_max > 0) { |
30ddba7d | 431 | unsigned long long max_completed; |
5509dc44 | 432 | sysfs_get_ll(mdi, NULL, "reshape_position", &curr); |
2eba8496 | 433 | sysfs_fd_get_str(scfd, buf, sizeof(buf)); |
3afaff93 N |
434 | if (strncmp(buf, "none", 4) == 0) { |
435 | /* Either reshape has aborted, or hasn't | |
436 | * quite started yet. Wait a bit and | |
437 | * check 'sync_action' to see. | |
438 | */ | |
439 | usleep(10000); | |
440 | sysfs_get_str(mdi, NULL, "sync_action", buf, sizeof(buf)); | |
441 | if (strncmp(buf, "reshape", 7) != 0) | |
442 | break; | |
443 | } | |
5509dc44 | 444 | |
30ddba7d N |
445 | if (sysfs_fd_get_two(scfd, &completed, |
446 | &max_completed) == 2 && | |
447 | /* 'completed' sometimes reads as max-uulong */ | |
448 | completed < max_completed && | |
5509dc44 N |
449 | (completed > sync_max || |
450 | (completed == sync_max && curr != position))) { | |
451 | while (completed > sync_max) { | |
452 | sync_max += sectors / rd1; | |
453 | if (backwards) | |
454 | position -= sectors; | |
455 | else | |
456 | position += sectors; | |
457 | } | |
458 | if (sync_max < old_sync_max) | |
459 | sysfs_set_num(mdi, NULL, "sync_max", sync_max); | |
460 | } | |
461 | ||
2eba8496 N |
462 | if (!backwards && curr >= position) |
463 | break; | |
464 | if (backwards && curr <= position) | |
465 | break; | |
466 | sysfs_wait(scfd, &delay); | |
467 | } | |
468 | if (scfd >= 0) | |
469 | close(scfd); | |
470 | ||
471 | } | |
e3e0d0a8 | 472 | done: |
2eba8496 | 473 | |
fe7e0e64 N |
474 | /* As we have an O_EXCL open, any use of the device |
475 | * which blocks STOP_ARRAY is probably a transient use, | |
476 | * so it is reasonable to retry for a while - 5 seconds. | |
477 | */ | |
478 | count = 25; err = 0; | |
479 | while (count && fd >= 0 | |
480 | && (err = ioctl(fd, STOP_ARRAY, NULL)) < 0 | |
481 | && errno == EBUSY) { | |
482 | usleep(200000); | |
483 | count --; | |
484 | } | |
485 | if (fd >= 0 && err) { | |
486 | if (verbose >= 0) { | |
487 | pr_err("failed to stop array %s: %s\n", | |
488 | devname, strerror(errno)); | |
489 | if (errno == EBUSY) | |
7a862a02 | 490 | cont_err("Perhaps a running process, mounted filesystem or active volume group?\n"); |
4ccad7b1 | 491 | } |
fe7e0e64 N |
492 | rv = 1; |
493 | goto out; | |
494 | } | |
229e66cb JS |
495 | |
496 | if (get_linux_version() < 2006028) { | |
497 | /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array | |
498 | * was stopped, so We'll do it here just to be sure. Drop any | |
499 | * partitions as well... | |
500 | */ | |
501 | if (fd >= 0) | |
502 | ioctl(fd, BLKRRPART, 0); | |
503 | if (mdi) | |
504 | sysfs_uevent(mdi, "change"); | |
505 | } | |
4ccad7b1 | 506 | |
fe7e0e64 N |
507 | if (devnm[0] && use_udev()) { |
508 | struct map_ent *mp = map_by_devnm(&map, devnm); | |
509 | remove_devices(devnm, mp ? mp->path : NULL); | |
682c7051 | 510 | } |
fe7e0e64 N |
511 | |
512 | if (verbose >= 0) | |
513 | pr_err("stopped %s\n", devname); | |
514 | map_lock(&map); | |
515 | map_remove(&map, devnm); | |
516 | map_unlock(&map); | |
517 | out: | |
6e8d27e7 | 518 | sysfs_free(mdi); |
fe7e0e64 | 519 | |
bccd8153 | 520 | return rv; |
64c4757e NB |
521 | } |
522 | ||
5e73b024 N |
523 | static struct mddev_dev *add_one(struct mddev_dev *dv, char *name, char disp) |
524 | { | |
525 | struct mddev_dev *new; | |
526 | new = xmalloc(sizeof(*new)); | |
527 | memset(new, 0, sizeof(*new)); | |
528 | new->devname = xstrdup(name); | |
529 | new->disposition = disp; | |
530 | new->next = dv->next; | |
531 | dv->next = new; | |
532 | return new; | |
533 | } | |
534 | ||
1d997643 N |
535 | static void add_faulty(struct mddev_dev *dv, int fd, char disp) |
536 | { | |
537 | mdu_array_info_t array; | |
538 | mdu_disk_info_t disk; | |
539 | int remaining_disks; | |
540 | int i; | |
541 | ||
542 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) | |
543 | return; | |
544 | ||
545 | remaining_disks = array.nr_disks; | |
546 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
1d997643 N |
547 | char buf[40]; |
548 | disk.number = i; | |
549 | if (ioctl(fd, GET_DISK_INFO, &disk) != 0) | |
550 | continue; | |
551 | if (disk.major == 0 && disk.minor == 0) | |
552 | continue; | |
553 | remaining_disks--; | |
554 | if ((disk.state & 1) == 0) /* not faulty */ | |
555 | continue; | |
556 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
5e73b024 | 557 | dv = add_one(dv, buf, disp); |
1d997643 N |
558 | } |
559 | } | |
560 | ||
561 | static void add_detached(struct mddev_dev *dv, int fd, char disp) | |
562 | { | |
563 | mdu_array_info_t array; | |
564 | mdu_disk_info_t disk; | |
565 | int remaining_disks; | |
566 | int i; | |
567 | ||
568 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) | |
569 | return; | |
570 | ||
571 | remaining_disks = array.nr_disks; | |
572 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
1d997643 N |
573 | char buf[40]; |
574 | int sfd; | |
575 | disk.number = i; | |
576 | if (ioctl(fd, GET_DISK_INFO, &disk) != 0) | |
577 | continue; | |
578 | if (disk.major == 0 && disk.minor == 0) | |
579 | continue; | |
580 | remaining_disks--; | |
581 | if (disp == 'f' && (disk.state & 1) != 0) /* already faulty */ | |
582 | continue; | |
583 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
584 | sfd = dev_open(buf, O_RDONLY); | |
585 | if (sfd >= 0) { | |
586 | /* Not detached */ | |
587 | close(sfd); | |
588 | continue; | |
589 | } | |
590 | if (errno != ENXIO) | |
591 | /* Probably not detached */ | |
592 | continue; | |
5e73b024 | 593 | dv = add_one(dv, buf, disp); |
1d997643 N |
594 | } |
595 | } | |
596 | ||
64a78416 N |
597 | static void add_set(struct mddev_dev *dv, int fd, char set_char) |
598 | { | |
599 | mdu_array_info_t array; | |
600 | mdu_disk_info_t disk; | |
601 | int remaining_disks; | |
602 | int copies, set; | |
603 | int i; | |
604 | ||
605 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) | |
606 | return; | |
607 | if (array.level != 10) | |
608 | return; | |
609 | copies = ((array.layout & 0xff) * | |
610 | ((array.layout >> 8) & 0xff)); | |
611 | if (array.raid_disks % copies) | |
612 | return; | |
613 | ||
614 | remaining_disks = array.nr_disks; | |
615 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
616 | char buf[40]; | |
617 | disk.number = i; | |
618 | if (ioctl(fd, GET_DISK_INFO, &disk) != 0) | |
619 | continue; | |
620 | if (disk.major == 0 && disk.minor == 0) | |
621 | continue; | |
622 | remaining_disks--; | |
623 | set = disk.raid_disk % copies; | |
624 | if (set_char != set + 'A') | |
625 | continue; | |
626 | sprintf(buf, "%d:%d", disk.major, disk.minor); | |
627 | dv = add_one(dv, buf, dv->disposition); | |
628 | } | |
629 | } | |
630 | ||
abe94694 N |
631 | int attempt_re_add(int fd, int tfd, struct mddev_dev *dv, |
632 | struct supertype *dev_st, struct supertype *tst, | |
633 | unsigned long rdev, | |
634 | char *update, char *devname, int verbose, | |
635 | mdu_array_info_t *array) | |
636 | { | |
637 | struct mdinfo mdi; | |
638 | int duuid[4]; | |
639 | int ouuid[4]; | |
640 | ||
641 | dev_st->ss->getinfo_super(dev_st, &mdi, NULL); | |
642 | dev_st->ss->uuid_from_super(dev_st, ouuid); | |
643 | if (tst->sb) | |
644 | tst->ss->uuid_from_super(tst, duuid); | |
645 | else | |
646 | /* Assume uuid matches: kernel will check */ | |
647 | memcpy(duuid, ouuid, sizeof(ouuid)); | |
648 | if ((mdi.disk.state & (1<<MD_DISK_ACTIVE)) && | |
649 | !(mdi.disk.state & (1<<MD_DISK_FAULTY)) && | |
650 | memcmp(duuid, ouuid, sizeof(ouuid))==0) { | |
651 | /* Looks like it is worth a | |
652 | * try. Need to make sure | |
653 | * kernel will accept it | |
654 | * though. | |
655 | */ | |
656 | mdu_disk_info_t disc; | |
657 | /* re-add doesn't work for version-1 superblocks | |
658 | * before 2.6.18 :-( | |
659 | */ | |
660 | if (array->major_version == 1 && | |
661 | get_linux_version() <= 2006018) | |
662 | goto skip_re_add; | |
663 | disc.number = mdi.disk.number; | |
664 | if (ioctl(fd, GET_DISK_INFO, &disc) != 0 | |
665 | || disc.major != 0 || disc.minor != 0 | |
666 | ) | |
667 | goto skip_re_add; | |
668 | disc.major = major(rdev); | |
669 | disc.minor = minor(rdev); | |
670 | disc.number = mdi.disk.number; | |
671 | disc.raid_disk = mdi.disk.raid_disk; | |
672 | disc.state = mdi.disk.state; | |
bff96f73 GJ |
673 | if (array->state & (1 << MD_SB_CLUSTERED)) { |
674 | /* extra flags are needed when adding to a cluster as | |
675 | * there are two cases to distinguish | |
676 | */ | |
677 | if (dv->disposition == 'c') | |
678 | disc.state |= (1 << MD_DISK_CANDIDATE); | |
679 | else | |
680 | disc.state |= (1 << MD_DISK_CLUSTER_ADD); | |
681 | } | |
abe94694 N |
682 | if (dv->writemostly == 1) |
683 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; | |
684 | if (dv->writemostly == 2) | |
685 | disc.state &= ~(1 << MD_DISK_WRITEMOSTLY); | |
686 | remove_partitions(tfd); | |
687 | if (update || dv->writemostly > 0) { | |
688 | int rv = -1; | |
689 | tfd = dev_open(dv->devname, O_RDWR); | |
690 | if (tfd < 0) { | |
7a862a02 | 691 | pr_err("failed to open %s for superblock update during re-add\n", dv->devname); |
abe94694 N |
692 | return -1; |
693 | } | |
694 | ||
695 | if (dv->writemostly == 1) | |
696 | rv = dev_st->ss->update_super( | |
697 | dev_st, NULL, "writemostly", | |
698 | devname, verbose, 0, NULL); | |
699 | if (dv->writemostly == 2) | |
700 | rv = dev_st->ss->update_super( | |
701 | dev_st, NULL, "readwrite", | |
702 | devname, verbose, 0, NULL); | |
703 | if (update) | |
704 | rv = dev_st->ss->update_super( | |
705 | dev_st, NULL, update, | |
706 | devname, verbose, 0, NULL); | |
707 | if (rv == 0) | |
708 | rv = dev_st->ss->store_super(dev_st, tfd); | |
709 | close(tfd); | |
710 | if (rv != 0) { | |
7a862a02 | 711 | pr_err("failed to update superblock during re-add\n"); |
abe94694 N |
712 | return -1; |
713 | } | |
714 | } | |
715 | /* don't even try if disk is marked as faulty */ | |
716 | errno = 0; | |
717 | if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) { | |
718 | if (verbose >= 0) | |
719 | pr_err("re-added %s\n", dv->devname); | |
720 | return 1; | |
721 | } | |
722 | if (errno == ENOMEM || errno == EROFS) { | |
723 | pr_err("add new device failed for %s: %s\n", | |
724 | dv->devname, strerror(errno)); | |
725 | if (dv->disposition == 'M') | |
726 | return 0; | |
727 | return -1; | |
728 | } | |
729 | } | |
730 | skip_re_add: | |
731 | return 0; | |
732 | } | |
733 | ||
38aeaf3a N |
734 | int Manage_add(int fd, int tfd, struct mddev_dev *dv, |
735 | struct supertype *tst, mdu_array_info_t *array, | |
736 | int force, int verbose, char *devname, | |
4de90913 GJ |
737 | char *update, unsigned long rdev, unsigned long long array_size, |
738 | int raid_slot) | |
38aeaf3a N |
739 | { |
740 | unsigned long long ldsize; | |
fbd3e15c | 741 | struct supertype *dev_st; |
38aeaf3a N |
742 | int j; |
743 | mdu_disk_info_t disc; | |
744 | ||
745 | if (!get_dev_size(tfd, dv->devname, &ldsize)) { | |
746 | if (dv->disposition == 'M') | |
747 | return 0; | |
748 | else | |
749 | return -1; | |
750 | } | |
751 | ||
632dc30c | 752 | if (tst->ss == &super0 && ldsize > 4ULL*1024*1024*1024*1024) { |
7ccc4cc4 | 753 | /* More than 4TB is wasted on v0.90 */ |
38aeaf3a | 754 | if (!force) { |
7a862a02 N |
755 | pr_err("%s is larger than %s can effectively use.\n" |
756 | " Add --force is you really want to add this device.\n", | |
38aeaf3a N |
757 | dv->devname, devname); |
758 | return -1; | |
759 | } | |
7a862a02 N |
760 | pr_err("%s is larger than %s can effectively use.\n" |
761 | " Adding anyway as --force was given.\n", | |
38aeaf3a N |
762 | dv->devname, devname); |
763 | } | |
764 | if (!tst->ss->external && | |
765 | array->major_version == 0 && | |
766 | md_get_version(fd)%100 < 2) { | |
767 | if (ioctl(fd, HOT_ADD_DISK, rdev)==0) { | |
768 | if (verbose >= 0) | |
769 | pr_err("hot added %s\n", | |
770 | dv->devname); | |
771 | return 1; | |
772 | } | |
773 | ||
774 | pr_err("hot add failed for %s: %s\n", | |
775 | dv->devname, strerror(errno)); | |
776 | return -1; | |
777 | } | |
778 | ||
779 | if (array->not_persistent == 0 || tst->ss->external) { | |
780 | ||
781 | /* need to find a sample superblock to copy, and | |
782 | * a spare slot to use. | |
783 | * For 'external' array (well, container based), | |
784 | * We can just load the metadata for the array-> | |
785 | */ | |
786 | int array_failed; | |
787 | if (tst->sb) | |
788 | /* already loaded */; | |
789 | else if (tst->ss->external) { | |
790 | tst->ss->load_container(tst, fd, NULL); | |
791 | } else for (j = 0; j < tst->max_devs; j++) { | |
792 | char *dev; | |
793 | int dfd; | |
794 | disc.number = j; | |
795 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
796 | continue; | |
797 | if (disc.major==0 && disc.minor==0) | |
798 | continue; | |
799 | if ((disc.state & 4)==0) /* sync */ | |
800 | continue; | |
801 | /* Looks like a good device to try */ | |
802 | dev = map_dev(disc.major, disc.minor, 1); | |
803 | if (!dev) | |
804 | continue; | |
805 | dfd = dev_open(dev, O_RDONLY); | |
806 | if (dfd < 0) | |
807 | continue; | |
808 | if (tst->ss->load_super(tst, dfd, | |
809 | NULL)) { | |
810 | close(dfd); | |
811 | continue; | |
812 | } | |
813 | close(dfd); | |
814 | break; | |
815 | } | |
816 | /* FIXME this is a bad test to be using */ | |
f33a71f1 N |
817 | if (!tst->sb && (dv->disposition != 'a' |
818 | && dv->disposition != 'S')) { | |
38aeaf3a N |
819 | /* we are re-adding a device to a |
820 | * completely dead array - have to depend | |
821 | * on kernel to check | |
822 | */ | |
823 | } else if (!tst->sb) { | |
824 | pr_err("cannot load array metadata from %s\n", devname); | |
825 | return -1; | |
826 | } | |
827 | ||
828 | /* Make sure device is large enough */ | |
01290056 SL |
829 | if (dv->disposition != 'j' && /* skip size check for Journal */ |
830 | tst->sb && | |
2609f339 | 831 | tst->ss->avail_size(tst, ldsize/512, INVALID_SECTORS) < |
38aeaf3a N |
832 | array_size) { |
833 | if (dv->disposition == 'M') | |
834 | return 0; | |
835 | pr_err("%s not large enough to join array\n", | |
836 | dv->devname); | |
837 | return -1; | |
838 | } | |
839 | ||
840 | /* Possibly this device was recently part of | |
841 | * the array and was temporarily removed, and | |
842 | * is now being re-added. If so, we can | |
843 | * simply re-add it. | |
844 | */ | |
845 | ||
fbd3e15c | 846 | if (array->not_persistent == 0) { |
38aeaf3a N |
847 | dev_st = dup_super(tst); |
848 | dev_st->ss->load_super(dev_st, tfd, NULL); | |
fbd3e15c JS |
849 | if (dev_st->sb && dv->disposition != 'S') { |
850 | int rv; | |
851 | ||
852 | rv = attempt_re_add(fd, tfd, dv, dev_st, tst, | |
853 | rdev, update, devname, | |
854 | verbose, array); | |
855 | dev_st->ss->free_super(dev_st); | |
856 | if (rv) | |
857 | return rv; | |
858 | } | |
38aeaf3a N |
859 | } |
860 | if (dv->disposition == 'M') { | |
861 | if (verbose > 0) | |
862 | pr_err("--re-add for %s to %s is not possible\n", | |
863 | dv->devname, devname); | |
864 | return 0; | |
865 | } | |
866 | if (dv->disposition == 'A') { | |
867 | pr_err("--re-add for %s to %s is not possible\n", | |
868 | dv->devname, devname); | |
869 | return -1; | |
870 | } | |
871 | if (array->active_disks < array->raid_disks) { | |
872 | char *avail = xcalloc(array->raid_disks, 1); | |
873 | int d; | |
874 | int found = 0; | |
875 | ||
d180d2aa | 876 | for (d = 0; d < MAX_DISKS && found < array->nr_disks; d++) { |
38aeaf3a N |
877 | disc.number = d; |
878 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
879 | continue; | |
880 | if (disc.major == 0 && disc.minor == 0) | |
881 | continue; | |
882 | if (!(disc.state & (1<<MD_DISK_SYNC))) | |
883 | continue; | |
884 | avail[disc.raid_disk] = 1; | |
d31d0f52 | 885 | found++; |
38aeaf3a N |
886 | } |
887 | array_failed = !enough(array->level, array->raid_disks, | |
888 | array->layout, 1, avail); | |
6157951f | 889 | free(avail); |
38aeaf3a N |
890 | } else |
891 | array_failed = 0; | |
892 | if (array_failed) { | |
893 | pr_err("%s has failed so using --add cannot work and might destroy\n", | |
894 | devname); | |
895 | pr_err("data on %s. You should stop the array and re-assemble it.\n", | |
896 | dv->devname); | |
897 | return -1; | |
898 | } | |
899 | } else { | |
900 | /* non-persistent. Must ensure that new drive | |
901 | * is at least array->size big. | |
902 | */ | |
903 | if (ldsize/512 < array_size) { | |
904 | pr_err("%s not large enough to join array\n", | |
905 | dv->devname); | |
906 | return -1; | |
907 | } | |
908 | } | |
909 | /* committed to really trying this device now*/ | |
910 | remove_partitions(tfd); | |
911 | ||
912 | /* in 2.6.17 and earlier, version-1 superblocks won't | |
913 | * use the number we write, but will choose a free number. | |
914 | * we must choose the same free number, which requires | |
915 | * starting at 'raid_disks' and counting up | |
916 | */ | |
917 | for (j = array->raid_disks; j < tst->max_devs; j++) { | |
918 | disc.number = j; | |
919 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
920 | break; | |
921 | if (disc.major==0 && disc.minor==0) | |
922 | break; | |
923 | if (disc.state & 8) /* removed */ | |
924 | break; | |
925 | } | |
926 | disc.major = major(rdev); | |
927 | disc.minor = minor(rdev); | |
4de90913 GJ |
928 | if (raid_slot < 0) |
929 | disc.number = j; | |
930 | else | |
931 | disc.number = raid_slot; | |
38aeaf3a | 932 | disc.state = 0; |
01290056 SL |
933 | |
934 | /* only add journal to array that supports journaling */ | |
935 | if (dv->disposition == 'j') { | |
936 | struct mdinfo mdi; | |
937 | struct mdinfo *mdp; | |
938 | ||
939 | mdp = sysfs_read(fd, NULL, GET_ARRAY_STATE); | |
2a1990c0 JS |
940 | if (!mdp) { |
941 | pr_err("%s unable to read array state.\n", devname); | |
942 | return -1; | |
943 | } | |
01290056 SL |
944 | |
945 | if (strncmp(mdp->sysfs_array_state, "readonly", 8) != 0) { | |
d209181d | 946 | sysfs_free(mdp); |
01290056 SL |
947 | pr_err("%s is not readonly, cannot add journal.\n", devname); |
948 | return -1; | |
949 | } | |
950 | ||
d209181d JS |
951 | sysfs_free(mdp); |
952 | ||
01290056 SL |
953 | tst->ss->getinfo_super(tst, &mdi, NULL); |
954 | if (mdi.journal_device_required == 0) { | |
955 | pr_err("%s does not support journal device.\n", devname); | |
956 | return -1; | |
957 | } | |
38c2e05b | 958 | disc.raid_disk = 0; |
01290056 SL |
959 | } |
960 | ||
38aeaf3a N |
961 | if (array->not_persistent==0) { |
962 | int dfd; | |
01290056 SL |
963 | if (dv->disposition == 'j') |
964 | disc.state |= (1 << MD_DISK_JOURNAL) | (1 << MD_DISK_SYNC); | |
38aeaf3a N |
965 | if (dv->writemostly == 1) |
966 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; | |
967 | dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); | |
968 | if (tst->ss->add_to_super(tst, &disc, dfd, | |
72ca9bcf | 969 | dv->devname, INVALID_SECTORS)) |
38aeaf3a N |
970 | return -1; |
971 | if (tst->ss->write_init_super(tst)) | |
972 | return -1; | |
973 | } else if (dv->disposition == 'A') { | |
974 | /* this had better be raid1. | |
975 | * As we are "--re-add"ing we must find a spare slot | |
976 | * to fill. | |
977 | */ | |
978 | char *used = xcalloc(array->raid_disks, 1); | |
979 | for (j = 0; j < tst->max_devs; j++) { | |
980 | mdu_disk_info_t disc2; | |
981 | disc2.number = j; | |
982 | if (ioctl(fd, GET_DISK_INFO, &disc2)) | |
983 | continue; | |
984 | if (disc2.major==0 && disc2.minor==0) | |
985 | continue; | |
986 | if (disc2.state & 8) /* removed */ | |
987 | continue; | |
988 | if (disc2.raid_disk < 0) | |
989 | continue; | |
990 | if (disc2.raid_disk > array->raid_disks) | |
991 | continue; | |
992 | used[disc2.raid_disk] = 1; | |
993 | } | |
994 | for (j = 0 ; j < array->raid_disks; j++) | |
995 | if (!used[j]) { | |
996 | disc.raid_disk = j; | |
997 | disc.state |= (1<<MD_DISK_SYNC); | |
998 | break; | |
999 | } | |
1000 | free(used); | |
1001 | } | |
4de90913 GJ |
1002 | |
1003 | if (array->state & (1 << MD_SB_CLUSTERED)) { | |
1004 | if (dv->disposition == 'c') | |
1005 | disc.state |= (1 << MD_DISK_CANDIDATE); | |
1006 | else | |
1007 | disc.state |= (1 << MD_DISK_CLUSTER_ADD); | |
1008 | } | |
1009 | ||
38aeaf3a N |
1010 | if (dv->writemostly == 1) |
1011 | disc.state |= (1 << MD_DISK_WRITEMOSTLY); | |
1012 | if (tst->ss->external) { | |
1013 | /* add a disk | |
1014 | * to an external metadata container */ | |
1015 | struct mdinfo new_mdi; | |
1016 | struct mdinfo *sra; | |
1017 | int container_fd; | |
4dd2df09 | 1018 | char devnm[32]; |
38aeaf3a N |
1019 | int dfd; |
1020 | ||
4dd2df09 N |
1021 | strcpy(devnm, fd2devnm(fd)); |
1022 | ||
1023 | container_fd = open_dev_excl(devnm); | |
38aeaf3a | 1024 | if (container_fd < 0) { |
7a862a02 | 1025 | pr_err("add failed for %s: could not get exclusive access to container\n", |
38aeaf3a N |
1026 | dv->devname); |
1027 | tst->ss->free_super(tst); | |
1028 | return -1; | |
1029 | } | |
1030 | ||
9cf9a1de | 1031 | Kill(dv->devname, NULL, 0, -1, 0); |
38aeaf3a | 1032 | dfd = dev_open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); |
4dd2df09 | 1033 | if (mdmon_running(tst->container_devnm)) |
38aeaf3a N |
1034 | tst->update_tail = &tst->updates; |
1035 | if (tst->ss->add_to_super(tst, &disc, dfd, | |
72ca9bcf | 1036 | dv->devname, INVALID_SECTORS)) { |
38aeaf3a N |
1037 | close(dfd); |
1038 | close(container_fd); | |
1039 | return -1; | |
1040 | } | |
1041 | if (tst->update_tail) | |
1042 | flush_metadata_updates(tst); | |
1043 | else | |
1044 | tst->ss->sync_metadata(tst); | |
1045 | ||
4dd2df09 | 1046 | sra = sysfs_read(container_fd, NULL, 0); |
38aeaf3a N |
1047 | if (!sra) { |
1048 | pr_err("add failed for %s: sysfs_read failed\n", | |
1049 | dv->devname); | |
1050 | close(container_fd); | |
1051 | tst->ss->free_super(tst); | |
1052 | return -1; | |
1053 | } | |
1054 | sra->array.level = LEVEL_CONTAINER; | |
1055 | /* Need to set data_offset and component_size */ | |
1056 | tst->ss->getinfo_super(tst, &new_mdi, NULL); | |
1057 | new_mdi.disk.major = disc.major; | |
1058 | new_mdi.disk.minor = disc.minor; | |
1059 | new_mdi.recovery_start = 0; | |
1060 | /* Make sure fds are closed as they are O_EXCL which | |
1061 | * would block add_disk */ | |
1062 | tst->ss->free_super(tst); | |
1063 | if (sysfs_add_disk(sra, &new_mdi, 0) != 0) { | |
7a862a02 | 1064 | pr_err("add new device to external metadata failed for %s\n", dv->devname); |
38aeaf3a N |
1065 | close(container_fd); |
1066 | sysfs_free(sra); | |
1067 | return -1; | |
1068 | } | |
4dd2df09 | 1069 | ping_monitor(devnm); |
38aeaf3a N |
1070 | sysfs_free(sra); |
1071 | close(container_fd); | |
1072 | } else { | |
1073 | tst->ss->free_super(tst); | |
1074 | if (ioctl(fd, ADD_NEW_DISK, &disc)) { | |
01290056 SL |
1075 | if (dv->disposition == 'j') |
1076 | pr_err("Failed to hot add %s as journal, " | |
1077 | "please try restart %s.\n", dv->devname, devname); | |
1078 | else | |
1079 | pr_err("add new device failed for %s as %d: %s\n", | |
1080 | dv->devname, j, strerror(errno)); | |
38aeaf3a N |
1081 | return -1; |
1082 | } | |
01290056 SL |
1083 | if (dv->disposition == 'j') { |
1084 | pr_err("Journal added successfully, making %s read-write\n", devname); | |
1085 | if (Manage_ro(devname, fd, -1)) | |
1086 | pr_err("Failed to make %s read-write\n", devname); | |
1087 | } | |
1088 | ||
38aeaf3a N |
1089 | } |
1090 | if (verbose >= 0) | |
1091 | pr_err("added %s\n", dv->devname); | |
1092 | return 1; | |
1093 | } | |
1094 | ||
d070235d N |
1095 | int Manage_remove(struct supertype *tst, int fd, struct mddev_dev *dv, |
1096 | int sysfd, unsigned long rdev, int verbose, char *devname) | |
1097 | { | |
1098 | int lfd = -1; | |
1099 | int err; | |
1100 | ||
1101 | if (tst->ss->external) { | |
1102 | /* To remove a device from a container, we must | |
1103 | * check that it isn't in use in an array. | |
1104 | * This involves looking in the 'holders' | |
1105 | * directory - there must be just one entry, | |
1106 | * the container. | |
1107 | * To ensure that it doesn't get used as a | |
1108 | * hot spare while we are checking, we | |
1109 | * get an O_EXCL open on the container | |
1110 | */ | |
aab15415 | 1111 | int ret; |
4dd2df09 N |
1112 | char devnm[32]; |
1113 | strcpy(devnm, fd2devnm(fd)); | |
1114 | lfd = open_dev_excl(devnm); | |
d070235d | 1115 | if (lfd < 0) { |
7a862a02 | 1116 | pr_err("Cannot get exclusive access to container - odd\n"); |
d070235d N |
1117 | return -1; |
1118 | } | |
aab15415 N |
1119 | /* We may not be able to check on holders in |
1120 | * sysfs, either because we don't have the dev num | |
1121 | * (rdev == 0) or because the device has been detached | |
1122 | * and the 'holders' directory no longer exists | |
1123 | * (ret == -1). In that case, assume it is OK to | |
1124 | * remove. | |
d070235d | 1125 | */ |
aab15415 N |
1126 | if (rdev == 0) |
1127 | ret = -1; | |
c922221e TM |
1128 | else { |
1129 | /* | |
1130 | * The drive has already been set to 'faulty', however | |
1131 | * monitor might not have had time to process it and the | |
1132 | * drive might still have an entry in the 'holders' | |
1133 | * directory. Try a few times to avoid a false error | |
1134 | */ | |
1135 | int count = 20; | |
1136 | ||
1137 | do { | |
1138 | ret = sysfs_unique_holder(devnm, rdev); | |
1139 | if (ret < 2) | |
1140 | break; | |
1141 | usleep(100 * 1000); /* 100ms */ | |
1142 | } while (--count > 0); | |
1143 | ||
1144 | if (ret == 0) { | |
1145 | pr_err("%s is not a member, cannot remove.\n", | |
1146 | dv->devname); | |
1147 | close(lfd); | |
1148 | return -1; | |
1149 | } | |
1150 | if (ret >= 2) { | |
1151 | pr_err("%s is still in use, cannot remove.\n", | |
1152 | dv->devname); | |
1153 | close(lfd); | |
1154 | return -1; | |
1155 | } | |
d070235d N |
1156 | } |
1157 | } | |
1158 | /* FIXME check that it is a current member */ | |
1159 | if (sysfd >= 0) { | |
1160 | /* device has been removed and we don't know | |
1161 | * the major:minor number | |
1162 | */ | |
1163 | int n = write(sysfd, "remove", 6); | |
1164 | if (n != 6) | |
1165 | err = -1; | |
1166 | else | |
1167 | err = 0; | |
1168 | } else { | |
1169 | err = ioctl(fd, HOT_REMOVE_DISK, rdev); | |
1170 | if (err && errno == ENODEV) { | |
1171 | /* Old kernels rejected this if no personality | |
1172 | * is registered */ | |
4dd2df09 | 1173 | struct mdinfo *sra = sysfs_read(fd, NULL, GET_DEVS); |
d070235d N |
1174 | struct mdinfo *dv = NULL; |
1175 | if (sra) | |
1176 | dv = sra->devs; | |
1177 | for ( ; dv ; dv=dv->next) | |
1178 | if (dv->disk.major == (int)major(rdev) && | |
1179 | dv->disk.minor == (int)minor(rdev)) | |
1180 | break; | |
1181 | if (dv) | |
1182 | err = sysfs_set_str(sra, dv, | |
1183 | "state", "remove"); | |
1184 | else | |
1185 | err = -1; | |
6e8d27e7 | 1186 | sysfs_free(sra); |
d070235d N |
1187 | } |
1188 | } | |
1189 | if (err) { | |
7a862a02 | 1190 | pr_err("hot remove failed for %s: %s\n", dv->devname, |
d070235d N |
1191 | strerror(errno)); |
1192 | if (lfd >= 0) | |
1193 | close(lfd); | |
1194 | return -1; | |
1195 | } | |
1196 | if (tst->ss->external) { | |
1197 | /* | |
1198 | * Before dropping our exclusive open we make an | |
1199 | * attempt at preventing mdmon from seeing an | |
1200 | * 'add' event before reconciling this 'remove' | |
1201 | * event. | |
1202 | */ | |
4dd2df09 | 1203 | char *devnm = fd2devnm(fd); |
d070235d | 1204 | |
4dd2df09 | 1205 | if (!devnm) { |
d070235d N |
1206 | pr_err("unable to get container name\n"); |
1207 | return -1; | |
1208 | } | |
1209 | ||
4dd2df09 | 1210 | ping_manager(devnm); |
d070235d N |
1211 | } |
1212 | if (lfd >= 0) | |
1213 | close(lfd); | |
1214 | if (verbose >= 0) | |
1215 | pr_err("hot removed %s from %s\n", | |
1216 | dv->devname, devname); | |
1217 | return 1; | |
1218 | } | |
1219 | ||
70c55e36 N |
1220 | int Manage_replace(struct supertype *tst, int fd, struct mddev_dev *dv, |
1221 | unsigned long rdev, int verbose, char *devname) | |
1222 | { | |
1223 | struct mdinfo *mdi, *di; | |
1224 | if (tst->ss->external) { | |
1225 | pr_err("--replace only supported for native metadata (0.90 or 1.x)\n"); | |
1226 | return -1; | |
1227 | } | |
1228 | /* Need to find the device in sysfs and add 'want_replacement' to the | |
1229 | * status. | |
1230 | */ | |
4dd2df09 | 1231 | mdi = sysfs_read(fd, NULL, GET_DEVS); |
70c55e36 N |
1232 | if (!mdi || !mdi->devs) { |
1233 | pr_err("Cannot find status of %s to enable replacement - strange\n", | |
1234 | devname); | |
1235 | return -1; | |
1236 | } | |
1237 | for (di = mdi->devs; di; di = di->next) | |
1238 | if (di->disk.major == (int)major(rdev) && | |
1239 | di->disk.minor == (int)minor(rdev)) | |
1240 | break; | |
1241 | if (di) { | |
1242 | int rv; | |
1243 | if (di->disk.raid_disk < 0) { | |
1244 | pr_err("%s is not active and so cannot be replaced.\n", | |
1245 | dv->devname); | |
1246 | sysfs_free(mdi); | |
1247 | return -1; | |
1248 | } | |
1249 | rv = sysfs_set_str(mdi, di, | |
1250 | "state", "want_replacement"); | |
1251 | if (rv) { | |
1252 | sysfs_free(mdi); | |
1253 | pr_err("Failed to request replacement for %s\n", | |
1254 | dv->devname); | |
1255 | return -1; | |
1256 | } | |
1257 | if (verbose >= 0) | |
1258 | pr_err("Marked %s (device %d in %s) for replacement\n", | |
1259 | dv->devname, di->disk.raid_disk, devname); | |
1260 | /* If there is a matching 'with', we need to tell it which | |
1261 | * raid disk | |
1262 | */ | |
1263 | while (dv && dv->disposition != 'W') | |
1264 | dv = dv->next; | |
1265 | if (dv) { | |
1266 | dv->disposition = 'w'; | |
1267 | dv->used = di->disk.raid_disk; | |
1268 | } | |
1269 | return 1; | |
1270 | } | |
1271 | sysfs_free(mdi); | |
1272 | pr_err("%s not found in %s so cannot --replace it\n", | |
1273 | dv->devname, devname); | |
1274 | return -1; | |
1275 | } | |
1276 | ||
1277 | int Manage_with(struct supertype *tst, int fd, struct mddev_dev *dv, | |
1278 | unsigned long rdev, int verbose, char *devname) | |
1279 | { | |
1280 | struct mdinfo *mdi, *di; | |
1281 | /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */ | |
4dd2df09 | 1282 | mdi = sysfs_read(fd, NULL, GET_DEVS|GET_STATE); |
70c55e36 N |
1283 | if (!mdi || !mdi->devs) { |
1284 | pr_err("Cannot find status of %s to enable replacement - strange\n", | |
1285 | devname); | |
1286 | return -1; | |
1287 | } | |
1288 | for (di = mdi->devs; di; di = di->next) | |
1289 | if (di->disk.major == (int)major(rdev) && | |
1290 | di->disk.minor == (int)minor(rdev)) | |
1291 | break; | |
1292 | if (di) { | |
1293 | int rv; | |
1294 | if (di->disk.state & (1<<MD_DISK_FAULTY)) { | |
1295 | pr_err("%s is faulty and cannot be a replacement\n", | |
1296 | dv->devname); | |
1297 | sysfs_free(mdi); | |
1298 | return -1; | |
1299 | } | |
1300 | if (di->disk.raid_disk >= 0) { | |
1301 | pr_err("%s is active and cannot be a replacement\n", | |
1302 | dv->devname); | |
1303 | sysfs_free(mdi); | |
1304 | return -1; | |
1305 | } | |
1306 | rv = sysfs_set_num(mdi, di, | |
1307 | "slot", dv->used); | |
1308 | if (rv) { | |
1309 | sysfs_free(mdi); | |
51425978 | 1310 | pr_err("Failed to set %s as preferred replacement.\n", |
70c55e36 N |
1311 | dv->devname); |
1312 | return -1; | |
1313 | } | |
1314 | if (verbose >= 0) | |
1315 | pr_err("Marked %s in %s as replacement for device %d\n", | |
1316 | dv->devname, devname, dv->used); | |
1317 | return 1; | |
1318 | } | |
1319 | sysfs_free(mdi); | |
1320 | pr_err("%s not found in %s so cannot make it preferred replacement\n", | |
1321 | dv->devname, devname); | |
1322 | return -1; | |
1323 | } | |
1324 | ||
64c4757e | 1325 | int Manage_subdevs(char *devname, int fd, |
833bb0f8 | 1326 | struct mddev_dev *devlist, int verbose, int test, |
11b391ec | 1327 | char *update, int force) |
cd29a5c8 | 1328 | { |
7bd04da9 | 1329 | /* Do something to each dev. |
682c7051 NB |
1330 | * devmode can be |
1331 | * 'a' - add the device | |
1332 | * try HOT_ADD_DISK | |
1333 | * If that fails EINVAL, try ADD_NEW_DISK | |
f33a71f1 | 1334 | * 'S' - add the device as a spare - don't try re-add |
01290056 | 1335 | * 'j' - add the device as a journal device |
7bd04da9 N |
1336 | * 'A' - re-add the device |
1337 | * 'r' - remove the device: HOT_REMOVE_DISK | |
b80da661 NB |
1338 | * device can be 'faulty' or 'detached' in which case all |
1339 | * matching devices are removed. | |
682c7051 | 1340 | * 'f' - set the device faulty SET_DISK_FAULTY |
b80da661 NB |
1341 | * device can be 'detached' in which case any device that |
1342 | * is inaccessible will be marked faulty. | |
70c55e36 N |
1343 | * 'R' - mark this device as wanting replacement. |
1344 | * 'W' - this device is added if necessary and activated as | |
1345 | * a replacement for a previous 'R' device. | |
1346 | * ----- | |
1347 | * 'w' - 'W' will be changed to 'w' when it is paired with | |
1348 | * a 'R' device. If a 'W' is found while walking the list | |
1349 | * it must be unpaired, and is an error. | |
1350 | * 'M' - this is created by a 'missing' target. It is a slight | |
1351 | * variant on 'A' | |
262e3b7f N |
1352 | * 'F' - Another variant of 'A', where the device was faulty |
1353 | * so must be removed from the array first. | |
4de90913 | 1354 | * 'c' - confirm the device as found (for clustered environments) |
70c55e36 | 1355 | * |
98d27e39 N |
1356 | * For 'f' and 'r', the device can also be a kernel-internal |
1357 | * name such as 'sdb'. | |
682c7051 NB |
1358 | */ |
1359 | mdu_array_info_t array; | |
7a3be72f | 1360 | unsigned long long array_size; |
1d997643 | 1361 | struct mddev_dev *dv; |
cfad27a9 | 1362 | int tfd = -1; |
38aeaf3a | 1363 | struct supertype *tst; |
4725bc31 | 1364 | char *subarray = NULL; |
98d27e39 | 1365 | int sysfd = -1; |
7d2e6486 | 1366 | int count = 0; /* number of actions taken */ |
9f584691 | 1367 | struct mdinfo info; |
9465f170 | 1368 | struct mdinfo devinfo; |
9f584691 | 1369 | int frozen = 0; |
8af530b0 | 1370 | int busy = 0; |
4de90913 | 1371 | int raid_slot = -1; |
682c7051 NB |
1372 | |
1373 | if (ioctl(fd, GET_ARRAY_INFO, &array)) { | |
7bd04da9 | 1374 | pr_err("Cannot get array info for %s\n", |
682c7051 | 1375 | devname); |
bcbb3112 | 1376 | goto abort; |
682c7051 | 1377 | } |
4dd2df09 | 1378 | sysfs_init(&info, fd, NULL); |
3da92f27 | 1379 | |
7bd04da9 | 1380 | /* array.size is only 32 bits and may be truncated. |
7a3be72f NB |
1381 | * So read from sysfs if possible, and record number of sectors |
1382 | */ | |
1383 | ||
1384 | array_size = get_component_size(fd); | |
1385 | if (array_size <= 0) | |
1386 | array_size = array.size * 2; | |
1387 | ||
4725bc31 | 1388 | tst = super_by_fd(fd, &subarray); |
3da92f27 | 1389 | if (!tst) { |
e7b84f9d | 1390 | pr_err("unsupport array - version %d.%d\n", |
3da92f27 | 1391 | array.major_version, array.minor_version); |
bcbb3112 | 1392 | goto abort; |
3da92f27 NB |
1393 | } |
1394 | ||
38aeaf3a | 1395 | for (dv = devlist; dv; dv = dv->next) { |
5dffd09d | 1396 | unsigned long rdev = 0; /* device to add/remove etc */ |
38aeaf3a | 1397 | int rv; |
b47024f1 | 1398 | int mj,mn; |
4a39c6f2 | 1399 | |
4de90913 GJ |
1400 | raid_slot = -1; |
1401 | if (dv->disposition == 'c') { | |
1402 | rv = parse_cluster_confirm_arg(dv->devname, | |
1403 | &dv->devname, | |
1404 | &raid_slot); | |
d7a49369 | 1405 | if (rv) { |
4de90913 GJ |
1406 | pr_err("Could not get the devname of cluster\n"); |
1407 | goto abort; | |
1408 | } | |
1409 | } | |
1410 | ||
1d997643 N |
1411 | if (strcmp(dv->devname, "failed") == 0 || |
1412 | strcmp(dv->devname, "faulty") == 0) { | |
262e3b7f N |
1413 | if (dv->disposition != 'A' |
1414 | && dv->disposition != 'r') { | |
7a862a02 | 1415 | pr_err("%s only meaningful with -r or --re-add, not -%c\n", |
b80da661 | 1416 | dv->devname, dv->disposition); |
bcbb3112 | 1417 | goto abort; |
b80da661 | 1418 | } |
262e3b7f N |
1419 | add_faulty(dv, fd, (dv->disposition == 'A' |
1420 | ? 'F' : 'r')); | |
1d997643 N |
1421 | continue; |
1422 | } | |
1423 | if (strcmp(dv->devname, "detached") == 0) { | |
b80da661 | 1424 | if (dv->disposition != 'r' && dv->disposition != 'f') { |
7a862a02 | 1425 | pr_err("%s only meaningful with -r of -f, not -%c\n", |
b80da661 | 1426 | dv->devname, dv->disposition); |
bcbb3112 | 1427 | goto abort; |
b80da661 | 1428 | } |
1d997643 N |
1429 | add_detached(dv, fd, dv->disposition); |
1430 | continue; | |
1431 | } | |
1432 | ||
1433 | if (strcmp(dv->devname, "missing") == 0) { | |
e9ddbb2b | 1434 | struct mddev_dev *add_devlist; |
1d997643 | 1435 | struct mddev_dev **dp; |
4de90913 GJ |
1436 | if (dv->disposition == 'c') { |
1437 | rv = ioctl(fd, CLUSTERED_DISK_NACK, NULL); | |
1438 | break; | |
1439 | } | |
1440 | ||
c8e1a230 | 1441 | if (dv->disposition != 'A') { |
7a862a02 | 1442 | pr_err("'missing' only meaningful with --re-add\n"); |
bcbb3112 | 1443 | goto abort; |
a4e13010 | 1444 | } |
1d997643 | 1445 | add_devlist = conf_get_devs(); |
a4e13010 | 1446 | if (add_devlist == NULL) { |
e7b84f9d | 1447 | pr_err("no devices to scan for missing members."); |
a4e13010 N |
1448 | continue; |
1449 | } | |
1d997643 | 1450 | for (dp = &add_devlist; *dp; dp = & (*dp)->next) |
7bd04da9 | 1451 | /* 'M' (for 'missing') is like 'A' without errors */ |
1d997643 N |
1452 | (*dp)->disposition = 'M'; |
1453 | *dp = dv->next; | |
1454 | dv->next = add_devlist; | |
1455 | continue; | |
1456 | } | |
1457 | ||
64a78416 N |
1458 | if (strncmp(dv->devname, "set-", 4) == 0 && |
1459 | strlen(dv->devname) == 5) { | |
1460 | int copies; | |
1461 | ||
1462 | if (dv->disposition != 'r' && | |
1463 | dv->disposition != 'f') { | |
1464 | pr_err("'%s' only meaningful with -r or -f\n", | |
1465 | dv->devname); | |
1466 | goto abort; | |
1467 | } | |
1468 | if (array.level != 10) { | |
1469 | pr_err("'%s' only meaningful with RAID10 arrays\n", | |
1470 | dv->devname); | |
1471 | goto abort; | |
1472 | } | |
1473 | copies = ((array.layout & 0xff) * | |
1474 | ((array.layout >> 8) & 0xff)); | |
1475 | if (array.raid_disks % copies != 0 || | |
1476 | dv->devname[4] < 'A' || | |
1477 | dv->devname[4] >= 'A' + copies || | |
1478 | copies > 26) { | |
1479 | pr_err("'%s' not meaningful with this array\n", | |
1480 | dv->devname); | |
1481 | goto abort; | |
1482 | } | |
1483 | add_set(dv, fd, dv->devname[4]); | |
1484 | continue; | |
1485 | } | |
1486 | ||
1d997643 | 1487 | if (strchr(dv->devname, '/') == NULL && |
aab15415 N |
1488 | strchr(dv->devname, ':') == NULL && |
1489 | strlen(dv->devname) < 50) { | |
98d27e39 N |
1490 | /* Assume this is a kernel-internal name like 'sda1' */ |
1491 | int found = 0; | |
1492 | char dname[55]; | |
1493 | if (dv->disposition != 'r' && dv->disposition != 'f') { | |
7a862a02 | 1494 | pr_err("%s only meaningful with -r or -f, not -%c\n", |
98d27e39 | 1495 | dv->devname, dv->disposition); |
bcbb3112 | 1496 | goto abort; |
98d27e39 N |
1497 | } |
1498 | ||
1499 | sprintf(dname, "dev-%s", dv->devname); | |
4dd2df09 | 1500 | sysfd = sysfs_open(fd2devnm(fd), dname, "block/dev"); |
98d27e39 N |
1501 | if (sysfd >= 0) { |
1502 | char dn[20]; | |
98d27e39 N |
1503 | if (sysfs_fd_get_str(sysfd, dn, 20) > 0 && |
1504 | sscanf(dn, "%d:%d", &mj,&mn) == 2) { | |
5dffd09d | 1505 | rdev = makedev(mj,mn); |
98d27e39 N |
1506 | found = 1; |
1507 | } | |
1508 | close(sysfd); | |
1509 | sysfd = -1; | |
1510 | } | |
1511 | if (!found) { | |
4dd2df09 | 1512 | sysfd = sysfs_open(fd2devnm(fd), dname, "state"); |
98d27e39 | 1513 | if (sysfd < 0) { |
7a862a02 | 1514 | pr_err("%s does not appear to be a component of %s\n", |
98d27e39 | 1515 | dv->devname, devname); |
bcbb3112 | 1516 | goto abort; |
98d27e39 N |
1517 | } |
1518 | } | |
b47024f1 N |
1519 | } else if ((dv->disposition == 'r' || dv->disposition == 'f') |
1520 | && get_maj_min(dv->devname, &mj, &mn)) { | |
1521 | /* for 'fail' and 'remove', the device might | |
1522 | * not exist. | |
1523 | */ | |
1524 | rdev = makedev(mj, mn); | |
b80da661 | 1525 | } else { |
5dffd09d | 1526 | struct stat stb; |
c7b47447 | 1527 | tfd = dev_open(dv->devname, O_RDONLY); |
cc5083d1 | 1528 | if (tfd >= 0) { |
5fe7f5f7 | 1529 | fstat(tfd, &stb); |
cc5083d1 JS |
1530 | close(tfd); |
1531 | } else { | |
5fe7f5f7 N |
1532 | int open_err = errno; |
1533 | if (stat(dv->devname, &stb) != 0) { | |
1534 | pr_err("Cannot find %s: %s\n", | |
1535 | dv->devname, strerror(errno)); | |
1536 | goto abort; | |
1537 | } | |
1538 | if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
1539 | if (dv->disposition == 'M') | |
1540 | /* non-fatal. Also improbable */ | |
1541 | continue; | |
1542 | pr_err("%s is not a block device.\n", | |
1543 | dv->devname); | |
1544 | goto abort; | |
1545 | } | |
1546 | if (dv->disposition == 'r') | |
1547 | /* Be happy, the stat worked, that is | |
1548 | * enough for --remove | |
1549 | */ | |
1550 | ; | |
1551 | else { | |
1d997643 N |
1552 | if (dv->disposition == 'M') |
1553 | /* non-fatal */ | |
1554 | continue; | |
839f27a3 | 1555 | pr_err("Cannot open %s: %s\n", |
5fe7f5f7 | 1556 | dv->devname, strerror(open_err)); |
bcbb3112 | 1557 | goto abort; |
5a9de8db | 1558 | } |
b80da661 | 1559 | } |
5dffd09d | 1560 | rdev = stb.st_rdev; |
682c7051 | 1561 | } |
cd29a5c8 | 1562 | switch(dv->disposition){ |
682c7051 | 1563 | default: |
e7b84f9d | 1564 | pr_err("internal error - devmode[%s]=%d\n", |
c913b90e | 1565 | dv->devname, dv->disposition); |
bcbb3112 | 1566 | goto abort; |
682c7051 | 1567 | case 'a': |
f33a71f1 | 1568 | case 'S': /* --add-spare */ |
01290056 | 1569 | case 'j': /* --add-journal */ |
c8e1a230 | 1570 | case 'A': |
262e3b7f N |
1571 | case 'M': /* --re-add missing */ |
1572 | case 'F': /* --re-add faulty */ | |
4de90913 | 1573 | case 'c': /* --cluster-confirm */ |
4a39c6f2 | 1574 | /* add the device */ |
4725bc31 | 1575 | if (subarray) { |
7a862a02 | 1576 | pr_err("Cannot add disks to a \'member\' array, perform this operation on the parent container\n"); |
bcbb3112 | 1577 | goto abort; |
f94d52f4 | 1578 | } |
9465f170 GJ |
1579 | |
1580 | /* Let's first try to write re-add to sysfs */ | |
1581 | if (rdev != 0 && | |
1582 | (dv->disposition == 'A' || dv->disposition == 'F')) { | |
1583 | sysfs_init_dev(&devinfo, rdev); | |
1584 | if (sysfs_set_str(&info, &devinfo, "state", "re-add") == 0) { | |
1585 | pr_err("re-add %s to %s succeed\n", | |
1586 | dv->devname, info.sys_name); | |
1587 | break; | |
1588 | } | |
1589 | } | |
1590 | ||
262e3b7f N |
1591 | if (dv->disposition == 'F') |
1592 | /* Need to remove first */ | |
5dffd09d | 1593 | ioctl(fd, HOT_REMOVE_DISK, rdev); |
f277ce36 | 1594 | /* Make sure it isn't in use (in 2.6 or later) */ |
1d997643 | 1595 | tfd = dev_open(dv->devname, O_RDONLY|O_EXCL); |
38aeaf3a N |
1596 | if (tfd >= 0) { |
1597 | /* We know no-one else is using it. We'll | |
1598 | * need non-exclusive access to add it, so | |
1599 | * do that now. | |
1600 | */ | |
1601 | close(tfd); | |
1602 | tfd = dev_open(dv->devname, O_RDONLY); | |
1011e834 | 1603 | } |
0fbf459d | 1604 | if (tfd < 0) { |
1d997643 N |
1605 | if (dv->disposition == 'M') |
1606 | continue; | |
e7b84f9d | 1607 | pr_err("Cannot open %s: %s\n", |
d7eaf49f | 1608 | dv->devname, strerror(errno)); |
bcbb3112 | 1609 | goto abort; |
d7eaf49f | 1610 | } |
9f584691 N |
1611 | if (!frozen) { |
1612 | if (sysfs_freeze_array(&info) == 1) | |
1613 | frozen = 1; | |
1614 | else | |
1615 | frozen = -1; | |
1616 | } | |
38aeaf3a N |
1617 | rv = Manage_add(fd, tfd, dv, tst, &array, |
1618 | force, verbose, devname, update, | |
4de90913 | 1619 | rdev, array_size, raid_slot); |
38aeaf3a N |
1620 | close(tfd); |
1621 | tfd = -1; | |
1622 | if (rv < 0) | |
bcbb3112 | 1623 | goto abort; |
38aeaf3a N |
1624 | if (rv > 0) |
1625 | count++; | |
682c7051 NB |
1626 | break; |
1627 | ||
1628 | case 'r': | |
1629 | /* hot remove */ | |
4725bc31 | 1630 | if (subarray) { |
7a862a02 | 1631 | pr_err("Cannot remove disks from a \'member\' array, perform this operation on the parent container\n"); |
d070235d N |
1632 | rv = -1; |
1633 | } else | |
1634 | rv = Manage_remove(tst, fd, dv, sysfd, | |
5dffd09d | 1635 | rdev, verbose, |
d070235d N |
1636 | devname); |
1637 | if (sysfd >= 0) | |
98d27e39 | 1638 | close(sysfd); |
d070235d N |
1639 | sysfd = -1; |
1640 | if (rv < 0) | |
bcbb3112 | 1641 | goto abort; |
d070235d N |
1642 | if (rv > 0) |
1643 | count++; | |
682c7051 NB |
1644 | break; |
1645 | ||
1646 | case 'f': /* set faulty */ | |
1647 | /* FIXME check current member */ | |
98d27e39 N |
1648 | if ((sysfd >= 0 && write(sysfd, "faulty", 6) != 6) || |
1649 | (sysfd < 0 && ioctl(fd, SET_DISK_FAULTY, | |
5dffd09d | 1650 | rdev))) { |
8af530b0 N |
1651 | if (errno == EBUSY) |
1652 | busy = 1; | |
e7b84f9d | 1653 | pr_err("set device faulty failed for %s: %s\n", |
1d997643 | 1654 | dv->devname, strerror(errno)); |
98d27e39 N |
1655 | if (sysfd >= 0) |
1656 | close(sysfd); | |
bcbb3112 | 1657 | goto abort; |
682c7051 | 1658 | } |
98d27e39 N |
1659 | if (sysfd >= 0) |
1660 | close(sysfd); | |
1661 | sysfd = -1; | |
7d2e6486 | 1662 | count++; |
dab6685f | 1663 | if (verbose >= 0) |
e7b84f9d | 1664 | pr_err("set %s faulty in %s\n", |
1d997643 | 1665 | dv->devname, devname); |
682c7051 | 1666 | break; |
70c55e36 N |
1667 | case 'R': /* Mark as replaceable */ |
1668 | if (subarray) { | |
7a862a02 | 1669 | pr_err("Cannot replace disks in a \'member\' array, perform this operation on the parent container\n"); |
70c55e36 N |
1670 | rv = -1; |
1671 | } else { | |
1672 | if (!frozen) { | |
1673 | if (sysfs_freeze_array(&info) == 1) | |
1674 | frozen = 1; | |
1675 | else | |
1676 | frozen = -1; | |
1677 | } | |
1678 | rv = Manage_replace(tst, fd, dv, | |
5dffd09d | 1679 | rdev, verbose, |
70c55e36 N |
1680 | devname); |
1681 | } | |
1682 | if (rv < 0) | |
1683 | goto abort; | |
1684 | if (rv > 0) | |
1685 | count++; | |
1686 | break; | |
1687 | case 'W': /* --with device that doesn't match */ | |
1688 | pr_err("No matching --replace device for --with %s\n", | |
1689 | dv->devname); | |
1690 | goto abort; | |
1691 | case 'w': /* --with device which was matched */ | |
1692 | rv = Manage_with(tst, fd, dv, | |
5dffd09d | 1693 | rdev, verbose, devname); |
70c55e36 N |
1694 | if (rv < 0) |
1695 | goto abort; | |
1696 | break; | |
682c7051 NB |
1697 | } |
1698 | } | |
9f584691 N |
1699 | if (frozen > 0) |
1700 | sysfs_set_str(&info, NULL, "sync_action","idle"); | |
7d2e6486 N |
1701 | if (test && count == 0) |
1702 | return 2; | |
682c7051 | 1703 | return 0; |
bcbb3112 N |
1704 | |
1705 | abort: | |
9f584691 N |
1706 | if (frozen > 0) |
1707 | sysfs_set_str(&info, NULL, "sync_action","idle"); | |
8af530b0 | 1708 | return !test && busy ? 2 : 1; |
64c4757e | 1709 | } |
1f48664b NB |
1710 | |
1711 | int autodetect(void) | |
1712 | { | |
1713 | /* Open any md device, and issue the RAID_AUTORUN ioctl */ | |
1714 | int rv = 1; | |
1715 | int fd = dev_open("9:0", O_RDONLY); | |
1716 | if (fd >= 0) { | |
1717 | if (ioctl(fd, RAID_AUTORUN, 0) == 0) | |
1718 | rv = 0; | |
1719 | close(fd); | |
1720 | } | |
1721 | return rv; | |
1722 | } | |
aa534678 | 1723 | |
ba728be7 | 1724 | int Update_subarray(char *dev, char *subarray, char *update, struct mddev_ident *ident, int verbose) |
aa534678 DW |
1725 | { |
1726 | struct supertype supertype, *st = &supertype; | |
1727 | int fd, rv = 2; | |
1728 | ||
1729 | memset(st, 0, sizeof(*st)); | |
aa534678 | 1730 | |
ba728be7 | 1731 | fd = open_subarray(dev, subarray, st, verbose < 0); |
aa534678 DW |
1732 | if (fd < 0) |
1733 | return 2; | |
1734 | ||
1735 | if (!st->ss->update_subarray) { | |
ba728be7 | 1736 | if (verbose >= 0) |
e7b84f9d N |
1737 | pr_err("Operation not supported for %s metadata\n", |
1738 | st->ss->name); | |
aa534678 DW |
1739 | goto free_super; |
1740 | } | |
1741 | ||
4dd2df09 | 1742 | if (mdmon_running(st->devnm)) |
aa534678 DW |
1743 | st->update_tail = &st->updates; |
1744 | ||
a951a4f7 | 1745 | rv = st->ss->update_subarray(st, subarray, update, ident); |
aa534678 DW |
1746 | |
1747 | if (rv) { | |
ba728be7 | 1748 | if (verbose >= 0) |
e7b84f9d | 1749 | pr_err("Failed to update %s of subarray-%s in %s\n", |
aa534678 DW |
1750 | update, subarray, dev); |
1751 | } else if (st->update_tail) | |
1752 | flush_metadata_updates(st); | |
1753 | else | |
1754 | st->ss->sync_metadata(st); | |
1755 | ||
ba728be7 | 1756 | if (rv == 0 && strcmp(update, "name") == 0 && verbose >= 0) |
e7b84f9d N |
1757 | pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n", |
1758 | subarray, dev); | |
aa534678 DW |
1759 | |
1760 | free_super: | |
1761 | st->ss->free_super(st); | |
1762 | close(fd); | |
1763 | ||
1764 | return rv; | |
1765 | } | |
d52bb542 | 1766 | |
7bd04da9 N |
1767 | /* Move spare from one array to another If adding to destination array fails |
1768 | * add back to original array. | |
d52bb542 AC |
1769 | * Returns 1 on success, 0 on failure */ |
1770 | int move_spare(char *from_devname, char *to_devname, dev_t devid) | |
1771 | { | |
1772 | struct mddev_dev devlist; | |
1773 | char devname[20]; | |
1774 | ||
1775 | /* try to remove and add */ | |
1776 | int fd1 = open(to_devname, O_RDONLY); | |
1777 | int fd2 = open(from_devname, O_RDONLY); | |
1778 | ||
1779 | if (fd1 < 0 || fd2 < 0) { | |
1780 | if (fd1>=0) close(fd1); | |
1781 | if (fd2>=0) close(fd2); | |
1782 | return 0; | |
1783 | } | |
1784 | ||
1785 | devlist.next = NULL; | |
1786 | devlist.used = 0; | |
d52bb542 AC |
1787 | devlist.writemostly = 0; |
1788 | devlist.devname = devname; | |
1789 | sprintf(devname, "%d:%d", major(devid), minor(devid)); | |
1790 | ||
1791 | devlist.disposition = 'r'; | |
11b391ec | 1792 | if (Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0) == 0) { |
d52bb542 | 1793 | devlist.disposition = 'a'; |
11b391ec | 1794 | if (Manage_subdevs(to_devname, fd1, &devlist, -1, 0, NULL, 0) == 0) { |
d52bb542 AC |
1795 | /* make sure manager is aware of changes */ |
1796 | ping_manager(to_devname); | |
1797 | ping_manager(from_devname); | |
1798 | close(fd1); | |
1799 | close(fd2); | |
1800 | return 1; | |
1801 | } | |
11b391ec | 1802 | else Manage_subdevs(from_devname, fd2, &devlist, -1, 0, NULL, 0); |
d52bb542 AC |
1803 | } |
1804 | close(fd1); | |
1805 | close(fd2); | |
1806 | return 0; | |
1807 | } | |
435d4ebb | 1808 | #endif |