]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de> | |
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 | |
22 | * Email: <neilb@cse.unsw.edu.au> | |
23 | * Paper: Neil Brown | |
24 | * School of Computer Science and Engineering | |
25 | * The University of New South Wales | |
26 | * Sydney, 2052 | |
27 | * Australia | |
28 | */ | |
29 | ||
30 | #include "mdadm.h" | |
31 | #include "md_u.h" | |
32 | #include "md_p.h" | |
33 | #include <ctype.h> | |
34 | ||
35 | #define REGISTER_DEV _IO (MD_MAJOR, 1) | |
36 | #define START_MD _IO (MD_MAJOR, 2) | |
37 | #define STOP_MD _IO (MD_MAJOR, 3) | |
38 | ||
39 | int Manage_ro(char *devname, int fd, int readonly) | |
40 | { | |
41 | /* switch to readonly or rw | |
42 | * | |
43 | * requires >= 0.90.0 | |
44 | * first check that array is runing | |
45 | * use RESTART_ARRAY_RW or STOP_ARRAY_RO | |
46 | * | |
47 | */ | |
48 | mdu_array_info_t array; | |
49 | #ifndef MDASSEMBLE | |
50 | struct mdinfo *mdi; | |
51 | #endif | |
52 | ||
53 | if (md_get_version(fd) < 9000) { | |
54 | fprintf(stderr, Name ": need md driver version 0.90.0 or later\n"); | |
55 | return 1; | |
56 | } | |
57 | #ifndef MDASSEMBLE | |
58 | /* If this is an externally-manage array, we need to modify the | |
59 | * metadata_version so that mdmon doesn't undo our change. | |
60 | */ | |
61 | mdi = sysfs_read(fd, -1, GET_LEVEL|GET_VERSION); | |
62 | if (mdi && | |
63 | mdi->array.major_version == -1 && | |
64 | mdi->array.level > 0 && | |
65 | is_subarray(mdi->text_version)) { | |
66 | char vers[64]; | |
67 | strcpy(vers, "external:"); | |
68 | strcat(vers, mdi->text_version); | |
69 | if (readonly > 0) { | |
70 | int rv; | |
71 | /* We set readonly ourselves. */ | |
72 | vers[9] = '-'; | |
73 | sysfs_set_str(mdi, NULL, "metadata_version", vers); | |
74 | ||
75 | close(fd); | |
76 | rv = sysfs_set_str(mdi, NULL, "array_state", "readonly"); | |
77 | ||
78 | if (rv < 0) { | |
79 | fprintf(stderr, Name ": failed to set readonly for %s: %s\n", | |
80 | devname, strerror(errno)); | |
81 | ||
82 | vers[9] = mdi->text_version[0]; | |
83 | sysfs_set_str(mdi, NULL, "metadata_version", vers); | |
84 | return 1; | |
85 | } | |
86 | } else { | |
87 | char *cp; | |
88 | /* We cannot set read/write - must signal mdmon */ | |
89 | vers[9] = '/'; | |
90 | sysfs_set_str(mdi, NULL, "metadata_version", vers); | |
91 | ||
92 | cp = strchr(vers+10, '/'); | |
93 | if (*cp) | |
94 | *cp = 0; | |
95 | ping_monitor(vers+10); | |
96 | } | |
97 | return 0; | |
98 | } | |
99 | #endif | |
100 | if (ioctl(fd, GET_ARRAY_INFO, &array)) { | |
101 | fprintf(stderr, Name ": %s does not appear to be active.\n", | |
102 | devname); | |
103 | return 1; | |
104 | } | |
105 | ||
106 | if (readonly>0) { | |
107 | if (ioctl(fd, STOP_ARRAY_RO, NULL)) { | |
108 | fprintf(stderr, Name ": failed to set readonly for %s: %s\n", | |
109 | devname, strerror(errno)); | |
110 | return 1; | |
111 | } | |
112 | } else if (readonly < 0) { | |
113 | if (ioctl(fd, RESTART_ARRAY_RW, NULL)) { | |
114 | fprintf(stderr, Name ": failed to set writable for %s: %s\n", | |
115 | devname, strerror(errno)); | |
116 | return 1; | |
117 | } | |
118 | } | |
119 | return 0; | |
120 | } | |
121 | ||
122 | #ifndef MDASSEMBLE | |
123 | ||
124 | static void remove_devices(int devnum, char *path) | |
125 | { | |
126 | /* Remove all 'standard' devices for 'devnum', including | |
127 | * partitions. Also remove names at 'path' - possibly with | |
128 | * partition suffixes - which link to those names. | |
129 | */ | |
130 | char base[40]; | |
131 | char *path2; | |
132 | char link[1024]; | |
133 | int n; | |
134 | int part; | |
135 | char *be; | |
136 | char *pe; | |
137 | ||
138 | if (devnum >= 0) | |
139 | sprintf(base, "/dev/md%d", devnum); | |
140 | else | |
141 | sprintf(base, "/dev/md_d%d", -1-devnum); | |
142 | be = base + strlen(base); | |
143 | if (path) { | |
144 | path2 = malloc(strlen(path)+20); | |
145 | strcpy(path2, path); | |
146 | pe = path2 + strlen(path2); | |
147 | } else | |
148 | path = NULL; | |
149 | ||
150 | for (part = 0; part < 16; part++) { | |
151 | if (part) { | |
152 | sprintf(be, "p%d", part); | |
153 | if (path) { | |
154 | if (isdigit(pe[-1])) | |
155 | sprintf(pe, "p%d", part); | |
156 | else | |
157 | sprintf(pe, "%d", part); | |
158 | } | |
159 | } | |
160 | /* FIXME test if really is md device ?? */ | |
161 | unlink(base); | |
162 | if (path) { | |
163 | n = readlink(path2, link, sizeof(link)); | |
164 | if (n && strlen(base) == n && | |
165 | strncmp(link, base, n) == 0) | |
166 | unlink(path2); | |
167 | } | |
168 | } | |
169 | } | |
170 | ||
171 | ||
172 | int Manage_runstop(char *devname, int fd, int runstop, int quiet) | |
173 | { | |
174 | /* Run or stop the array. array must already be configured | |
175 | * required >= 0.90.0 | |
176 | * Only print failure messages if quiet == 0; | |
177 | * quiet > 0 means really be quiet | |
178 | * quiet < 0 means we will try again if it fails. | |
179 | */ | |
180 | mdu_param_t param; /* unused */ | |
181 | ||
182 | if (runstop == -1 && md_get_version(fd) < 9000) { | |
183 | if (ioctl(fd, STOP_MD, 0)) { | |
184 | if (quiet == 0) fprintf(stderr, | |
185 | Name ": stopping device %s " | |
186 | "failed: %s\n", | |
187 | devname, strerror(errno)); | |
188 | return 1; | |
189 | } | |
190 | } | |
191 | ||
192 | if (md_get_version(fd) < 9000) { | |
193 | fprintf(stderr, Name ": need md driver version 0.90.0 or later\n"); | |
194 | return 1; | |
195 | } | |
196 | /* | |
197 | if (ioctl(fd, GET_ARRAY_INFO, &array)) { | |
198 | fprintf(stderr, Name ": %s does not appear to be active.\n", | |
199 | devname); | |
200 | return 1; | |
201 | } | |
202 | */ | |
203 | if (runstop>0) { | |
204 | if (ioctl(fd, RUN_ARRAY, ¶m)) { | |
205 | fprintf(stderr, Name ": failed to run array %s: %s\n", | |
206 | devname, strerror(errno)); | |
207 | return 1; | |
208 | } | |
209 | if (quiet <= 0) | |
210 | fprintf(stderr, Name ": started %s\n", devname); | |
211 | } else if (runstop < 0){ | |
212 | struct map_ent *map = NULL; | |
213 | struct stat stb; | |
214 | struct mdinfo *mdi; | |
215 | int devnum; | |
216 | /* If this is an mdmon managed array, just write 'inactive' | |
217 | * to the array state and let mdmon clear up. | |
218 | */ | |
219 | devnum = fd2devnum(fd); | |
220 | mdi = sysfs_read(fd, -1, GET_LEVEL|GET_VERSION); | |
221 | if (mdi && | |
222 | mdi->array.level > 0 && | |
223 | is_subarray(mdi->text_version)) { | |
224 | /* This is mdmon managed. */ | |
225 | close(fd); | |
226 | if (sysfs_set_str(mdi, NULL, | |
227 | "array_state", "inactive") < 0) { | |
228 | if (quiet == 0) | |
229 | fprintf(stderr, Name | |
230 | ": failed to stop array %s: %s\n", | |
231 | devname, strerror(errno)); | |
232 | return 1; | |
233 | } | |
234 | ||
235 | /* Give monitor a chance to act */ | |
236 | ping_monitor(mdi->text_version); | |
237 | ||
238 | fd = open(devname, O_RDONLY); | |
239 | } else if (mdi && | |
240 | mdi->array.major_version == -1 && | |
241 | mdi->array.minor_version == -2 && | |
242 | !is_subarray(mdi->text_version)) { | |
243 | /* container, possibly mdmon-managed. | |
244 | * Make sure mdmon isn't opening it, which | |
245 | * would interfere with the 'stop' | |
246 | */ | |
247 | ping_monitor(mdi->sys_name); | |
248 | } | |
249 | ||
250 | if (fd >= 0 && ioctl(fd, STOP_ARRAY, NULL)) { | |
251 | if (quiet == 0) { | |
252 | fprintf(stderr, Name | |
253 | ": failed to stop array %s: %s\n", | |
254 | devname, strerror(errno)); | |
255 | if (errno == EBUSY) | |
256 | fprintf(stderr, "Perhaps a running " | |
257 | "process, mounted filesystem " | |
258 | "or active volume group?\n"); | |
259 | } | |
260 | if (mdi) | |
261 | sysfs_free(mdi); | |
262 | return 1; | |
263 | } | |
264 | /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array | |
265 | * was stopped, so We'll do it here just to be sure. Drop any | |
266 | * partitions as well... | |
267 | */ | |
268 | if (fd >= 0) | |
269 | ioctl(fd, BLKRRPART, 0); | |
270 | if (mdi) | |
271 | sysfs_uevent(mdi, "change"); | |
272 | ||
273 | ||
274 | if (devnum != NoMdDev && | |
275 | (stat("/dev/.udev", &stb) != 0 || | |
276 | check_env("MDADM_NO_UDEV"))) { | |
277 | struct map_ent *mp = map_by_devnum(&map, devnum); | |
278 | remove_devices(devnum, mp ? mp->path : NULL); | |
279 | } | |
280 | ||
281 | ||
282 | if (quiet <= 0) | |
283 | fprintf(stderr, Name ": stopped %s\n", devname); | |
284 | if (devnum != NoMdDev) { | |
285 | map_delete(&map, devnum); | |
286 | map_write(map); | |
287 | map_free(map); | |
288 | } | |
289 | } | |
290 | return 0; | |
291 | } | |
292 | ||
293 | int Manage_resize(char *devname, int fd, long long size, int raid_disks) | |
294 | { | |
295 | mdu_array_info_t info; | |
296 | if (ioctl(fd, GET_ARRAY_INFO, &info) != 0) { | |
297 | fprintf(stderr, Name ": Cannot get array information for %s: %s\n", | |
298 | devname, strerror(errno)); | |
299 | return 1; | |
300 | } | |
301 | if (size >= 0) | |
302 | info.size = size; | |
303 | if (raid_disks > 0) | |
304 | info.raid_disks = raid_disks; | |
305 | if (ioctl(fd, SET_ARRAY_INFO, &info) != 0) { | |
306 | fprintf(stderr, Name ": Cannot set device size/shape for %s: %s\n", | |
307 | devname, strerror(errno)); | |
308 | return 1; | |
309 | } | |
310 | return 0; | |
311 | } | |
312 | ||
313 | int Manage_reconfig(char *devname, int fd, int layout) | |
314 | { | |
315 | mdu_array_info_t info; | |
316 | if (ioctl(fd, GET_ARRAY_INFO, &info) != 0) { | |
317 | fprintf(stderr, Name ": Cannot get array information for %s: %s\n", | |
318 | devname, strerror(errno)); | |
319 | return 1; | |
320 | } | |
321 | info.layout = layout; | |
322 | printf("layout set to %d\n", info.layout); | |
323 | if (ioctl(fd, SET_ARRAY_INFO, &info) != 0) { | |
324 | fprintf(stderr, Name ": Cannot set layout for %s: %s\n", | |
325 | devname, strerror(errno)); | |
326 | return 1; | |
327 | } | |
328 | return 0; | |
329 | } | |
330 | ||
331 | int Manage_subdevs(char *devname, int fd, | |
332 | mddev_dev_t devlist, int verbose) | |
333 | { | |
334 | /* do something to each dev. | |
335 | * devmode can be | |
336 | * 'a' - add the device | |
337 | * try HOT_ADD_DISK | |
338 | * If that fails EINVAL, try ADD_NEW_DISK | |
339 | * 'r' - remove the device HOT_REMOVE_DISK | |
340 | * device can be 'faulty' or 'detached' in which case all | |
341 | * matching devices are removed. | |
342 | * 'f' - set the device faulty SET_DISK_FAULTY | |
343 | * device can be 'detached' in which case any device that | |
344 | * is inaccessible will be marked faulty. | |
345 | */ | |
346 | mdu_array_info_t array; | |
347 | mdu_disk_info_t disc; | |
348 | unsigned long long array_size; | |
349 | mddev_dev_t dv, next = NULL; | |
350 | struct stat stb; | |
351 | int j, jnext = 0; | |
352 | int tfd; | |
353 | struct supertype *st, *tst; | |
354 | int duuid[4]; | |
355 | int ouuid[4]; | |
356 | int lfd = -1; | |
357 | ||
358 | if (ioctl(fd, GET_ARRAY_INFO, &array)) { | |
359 | fprintf(stderr, Name ": cannot get array info for %s\n", | |
360 | devname); | |
361 | return 1; | |
362 | } | |
363 | ||
364 | /* array.size is only 32 bit and may be truncated. | |
365 | * So read from sysfs if possible, and record number of sectors | |
366 | */ | |
367 | ||
368 | array_size = get_component_size(fd); | |
369 | if (array_size <= 0) | |
370 | array_size = array.size * 2; | |
371 | ||
372 | tst = super_by_fd(fd); | |
373 | if (!tst) { | |
374 | fprintf(stderr, Name ": unsupport array - version %d.%d\n", | |
375 | array.major_version, array.minor_version); | |
376 | return 1; | |
377 | } | |
378 | ||
379 | for (dv = devlist, j=0 ; dv; dv = next, j = jnext) { | |
380 | unsigned long long ldsize; | |
381 | char dvname[20]; | |
382 | char *dnprintable = dv->devname; | |
383 | int err; | |
384 | ||
385 | next = dv->next; | |
386 | jnext = 0; | |
387 | ||
388 | if (strcmp(dv->devname, "failed")==0 || | |
389 | strcmp(dv->devname, "faulty")==0) { | |
390 | if (dv->disposition != 'r') { | |
391 | fprintf(stderr, Name ": %s only meaningful " | |
392 | "with -r, not -%c\n", | |
393 | dv->devname, dv->disposition); | |
394 | return 1; | |
395 | } | |
396 | for (; j < array.raid_disks + array.nr_disks ; j++) { | |
397 | disc.number = j; | |
398 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
399 | continue; | |
400 | if (disc.major == 0 && disc.minor == 0) | |
401 | continue; | |
402 | if ((disc.state & 1) == 0) /* faulty */ | |
403 | continue; | |
404 | stb.st_rdev = makedev(disc.major, disc.minor); | |
405 | next = dv; | |
406 | jnext = j+1; | |
407 | sprintf(dvname,"%d:%d", disc.major, disc.minor); | |
408 | dnprintable = dvname; | |
409 | break; | |
410 | } | |
411 | if (jnext == 0) | |
412 | continue; | |
413 | } else if (strcmp(dv->devname, "detached") == 0) { | |
414 | if (dv->disposition != 'r' && dv->disposition != 'f') { | |
415 | fprintf(stderr, Name ": %s only meaningful " | |
416 | "with -r of -f, not -%c\n", | |
417 | dv->devname, dv->disposition); | |
418 | return 1; | |
419 | } | |
420 | for (; j < array.raid_disks + array.nr_disks; j++) { | |
421 | int sfd; | |
422 | disc.number = j; | |
423 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
424 | continue; | |
425 | if (disc.major == 0 && disc.minor == 0) | |
426 | continue; | |
427 | sprintf(dvname,"%d:%d", disc.major, disc.minor); | |
428 | sfd = dev_open(dvname, O_RDONLY); | |
429 | if (sfd >= 0) { | |
430 | close(sfd); | |
431 | continue; | |
432 | } | |
433 | if (dv->disposition == 'f' && | |
434 | (disc.state & 1) == 1) /* already faulty */ | |
435 | continue; | |
436 | if (errno != ENXIO) | |
437 | continue; | |
438 | stb.st_rdev = makedev(disc.major, disc.minor); | |
439 | next = dv; | |
440 | jnext = j+1; | |
441 | dnprintable = dvname; | |
442 | break; | |
443 | } | |
444 | if (jnext == 0) | |
445 | continue; | |
446 | } else { | |
447 | j = 0; | |
448 | ||
449 | if (stat(dv->devname, &stb)) { | |
450 | fprintf(stderr, Name ": cannot find %s: %s\n", | |
451 | dv->devname, strerror(errno)); | |
452 | return 1; | |
453 | } | |
454 | if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
455 | fprintf(stderr, Name ": %s is not a " | |
456 | "block device.\n", | |
457 | dv->devname); | |
458 | return 1; | |
459 | } | |
460 | } | |
461 | switch(dv->disposition){ | |
462 | default: | |
463 | fprintf(stderr, Name ": internal error - devmode[%s]=%d\n", | |
464 | dv->devname, dv->disposition); | |
465 | return 1; | |
466 | case 'a': | |
467 | /* add the device */ | |
468 | if (tst->subarray[0]) { | |
469 | fprintf(stderr, Name ": Cannot add disks to a" | |
470 | " \'member\' array, perform this" | |
471 | " operation on the parent container\n"); | |
472 | return 1; | |
473 | } | |
474 | /* Make sure it isn't in use (in 2.6 or later) */ | |
475 | tfd = open(dv->devname, O_RDONLY|O_EXCL|O_DIRECT); | |
476 | if (tfd < 0) { | |
477 | fprintf(stderr, Name ": Cannot open %s: %s\n", | |
478 | dv->devname, strerror(errno)); | |
479 | return 1; | |
480 | } | |
481 | remove_partitions(tfd); | |
482 | ||
483 | st = dup_super(tst); | |
484 | ||
485 | if (array.not_persistent==0) | |
486 | st->ss->load_super(st, tfd, NULL); | |
487 | ||
488 | if (!get_dev_size(tfd, dv->devname, &ldsize)) { | |
489 | close(tfd); | |
490 | return 1; | |
491 | } | |
492 | close(tfd); | |
493 | ||
494 | ||
495 | if (!tst->ss->external && | |
496 | array.major_version == 0 && | |
497 | md_get_version(fd)%100 < 2) { | |
498 | if (ioctl(fd, HOT_ADD_DISK, | |
499 | (unsigned long)stb.st_rdev)==0) { | |
500 | if (verbose >= 0) | |
501 | fprintf(stderr, Name ": hot added %s\n", | |
502 | dv->devname); | |
503 | continue; | |
504 | } | |
505 | ||
506 | fprintf(stderr, Name ": hot add failed for %s: %s\n", | |
507 | dv->devname, strerror(errno)); | |
508 | return 1; | |
509 | } | |
510 | ||
511 | if (array.not_persistent == 0 || tst->ss->external) { | |
512 | ||
513 | /* need to find a sample superblock to copy, and | |
514 | * a spare slot to use. | |
515 | * For 'external' array (well, container based), | |
516 | * We can just load the metadata for the array. | |
517 | */ | |
518 | if (tst->ss->external) { | |
519 | tst->ss->load_super(tst, fd, NULL); | |
520 | } else for (j = 0; j < tst->max_devs; j++) { | |
521 | char *dev; | |
522 | int dfd; | |
523 | disc.number = j; | |
524 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
525 | continue; | |
526 | if (disc.major==0 && disc.minor==0) | |
527 | continue; | |
528 | if ((disc.state & 4)==0) continue; /* sync */ | |
529 | /* Looks like a good device to try */ | |
530 | dev = map_dev(disc.major, disc.minor, 1); | |
531 | if (!dev) continue; | |
532 | dfd = dev_open(dev, O_RDONLY); | |
533 | if (dfd < 0) continue; | |
534 | if (tst->ss->load_super(tst, dfd, | |
535 | NULL)) { | |
536 | close(dfd); | |
537 | continue; | |
538 | } | |
539 | close(dfd); | |
540 | break; | |
541 | } | |
542 | /* FIXME this is a bad test to be using */ | |
543 | if (!tst->sb) { | |
544 | fprintf(stderr, Name ": cannot find valid superblock in this array - HELP\n"); | |
545 | return 1; | |
546 | } | |
547 | ||
548 | /* Make sure device is large enough */ | |
549 | if (tst->ss->avail_size(tst, ldsize/512) < | |
550 | array_size) { | |
551 | fprintf(stderr, Name ": %s not large enough to join array\n", | |
552 | dv->devname); | |
553 | return 1; | |
554 | } | |
555 | ||
556 | /* Possibly this device was recently part of the array | |
557 | * and was temporarily removed, and is now being re-added. | |
558 | * If so, we can simply re-add it. | |
559 | */ | |
560 | tst->ss->uuid_from_super(tst, duuid); | |
561 | ||
562 | /* re-add doesn't work for version-1 superblocks | |
563 | * before 2.6.18 :-( | |
564 | */ | |
565 | if (array.major_version == 1 && | |
566 | get_linux_version() <= 2006018) | |
567 | ; | |
568 | else if (st->sb) { | |
569 | st->ss->uuid_from_super(st, ouuid); | |
570 | if (memcmp(duuid, ouuid, sizeof(ouuid))==0) { | |
571 | /* looks close enough for now. Kernel | |
572 | * will worry about whether a bitmap | |
573 | * based reconstruction is possible. | |
574 | */ | |
575 | struct mdinfo mdi; | |
576 | st->ss->getinfo_super(st, &mdi); | |
577 | disc.major = major(stb.st_rdev); | |
578 | disc.minor = minor(stb.st_rdev); | |
579 | disc.number = mdi.disk.number; | |
580 | disc.raid_disk = mdi.disk.raid_disk; | |
581 | disc.state = mdi.disk.state; | |
582 | if (dv->writemostly == 1) | |
583 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; | |
584 | if (dv->writemostly == 2) | |
585 | disc.state &= ~(1 << MD_DISK_WRITEMOSTLY); | |
586 | if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) { | |
587 | if (verbose >= 0) | |
588 | fprintf(stderr, Name ": re-added %s\n", dv->devname); | |
589 | continue; | |
590 | } | |
591 | if (errno == ENOMEM || errno == EROFS) { | |
592 | fprintf(stderr, Name ": add new device failed for %s: %s\n", | |
593 | dv->devname, strerror(errno)); | |
594 | return 1; | |
595 | } | |
596 | /* fall back on normal-add */ | |
597 | } | |
598 | } | |
599 | } else { | |
600 | /* non-persistent. Must ensure that new drive | |
601 | * is at least array.size big. | |
602 | */ | |
603 | if (ldsize/512 < array_size) { | |
604 | fprintf(stderr, Name ": %s not large enough to join array\n", | |
605 | dv->devname); | |
606 | return 1; | |
607 | } | |
608 | } | |
609 | /* in 2.6.17 and earlier, version-1 superblocks won't | |
610 | * use the number we write, but will choose a free number. | |
611 | * we must choose the same free number, which requires | |
612 | * starting at 'raid_disks' and counting up | |
613 | */ | |
614 | for (j = array.raid_disks; j< tst->max_devs; j++) { | |
615 | disc.number = j; | |
616 | if (ioctl(fd, GET_DISK_INFO, &disc)) | |
617 | break; | |
618 | if (disc.major==0 && disc.minor==0) | |
619 | break; | |
620 | if (disc.state & 8) /* removed */ | |
621 | break; | |
622 | } | |
623 | disc.major = major(stb.st_rdev); | |
624 | disc.minor = minor(stb.st_rdev); | |
625 | disc.number =j; | |
626 | disc.state = 0; | |
627 | if (array.not_persistent==0 || tst->ss->external) { | |
628 | int dfd; | |
629 | if (dv->writemostly == 1) | |
630 | disc.state |= 1 << MD_DISK_WRITEMOSTLY; | |
631 | dfd = open(dv->devname, O_RDWR | O_EXCL|O_DIRECT); | |
632 | if (tst->ss->add_to_super(tst, &disc, dfd, | |
633 | dv->devname)) { | |
634 | close(dfd); | |
635 | return 1; | |
636 | } | |
637 | /* write_init_super will close 'dfd' */ | |
638 | if (tst->ss->external) | |
639 | /* mdmon will write the metadata */ | |
640 | close(dfd); | |
641 | else if (tst->ss->write_init_super(tst)) | |
642 | return 1; | |
643 | } else if (dv->re_add) { | |
644 | /* this had better be raid1. | |
645 | * As we are "--re-add"ing we must find a spare slot | |
646 | * to fill. | |
647 | */ | |
648 | char *used = malloc(array.raid_disks); | |
649 | memset(used, 0, array.raid_disks); | |
650 | for (j=0; j< tst->max_devs; j++) { | |
651 | mdu_disk_info_t disc2; | |
652 | disc2.number = j; | |
653 | if (ioctl(fd, GET_DISK_INFO, &disc2)) | |
654 | continue; | |
655 | if (disc2.major==0 && disc2.minor==0) | |
656 | continue; | |
657 | if (disc2.state & 8) /* removed */ | |
658 | continue; | |
659 | if (disc2.raid_disk < 0) | |
660 | continue; | |
661 | if (disc2.raid_disk > array.raid_disks) | |
662 | continue; | |
663 | used[disc2.raid_disk] = 1; | |
664 | } | |
665 | for (j=0 ; j<array.raid_disks; j++) | |
666 | if (!used[j]) { | |
667 | disc.raid_disk = j; | |
668 | disc.state |= (1<<MD_DISK_SYNC); | |
669 | break; | |
670 | } | |
671 | } | |
672 | if (dv->writemostly == 1) | |
673 | disc.state |= (1 << MD_DISK_WRITEMOSTLY); | |
674 | if (tst->ss->external) { | |
675 | /* add a disk to an external metadata container | |
676 | * only if mdmon is around to see it | |
677 | */ | |
678 | struct mdinfo new_mdi; | |
679 | struct mdinfo *sra; | |
680 | int container_fd; | |
681 | int devnum = fd2devnum(fd); | |
682 | ||
683 | container_fd = open_dev_excl(devnum); | |
684 | if (container_fd < 0) { | |
685 | fprintf(stderr, Name ": add failed for %s:" | |
686 | " could not get exclusive access to container\n", | |
687 | dv->devname); | |
688 | return 1; | |
689 | } | |
690 | ||
691 | if (!mdmon_running(devnum)) { | |
692 | fprintf(stderr, Name ": add failed for %s: mdmon not running\n", | |
693 | dv->devname); | |
694 | close(container_fd); | |
695 | return 1; | |
696 | } | |
697 | ||
698 | sra = sysfs_read(container_fd, -1, 0); | |
699 | if (!sra) { | |
700 | fprintf(stderr, Name ": add failed for %s: sysfs_read failed\n", | |
701 | dv->devname); | |
702 | close(container_fd); | |
703 | return 1; | |
704 | } | |
705 | sra->array.level = LEVEL_CONTAINER; | |
706 | /* Need to set data_offset and component_size */ | |
707 | tst->ss->getinfo_super(tst, &new_mdi); | |
708 | new_mdi.disk.major = disc.major; | |
709 | new_mdi.disk.minor = disc.minor; | |
710 | if (sysfs_add_disk(sra, &new_mdi, 0) != 0) { | |
711 | fprintf(stderr, Name ": add new device to external metadata" | |
712 | " failed for %s\n", dv->devname); | |
713 | close(container_fd); | |
714 | return 1; | |
715 | } | |
716 | ping_monitor(devnum2devname(devnum)); | |
717 | sysfs_free(sra); | |
718 | close(container_fd); | |
719 | } else if (ioctl(fd, ADD_NEW_DISK, &disc)) { | |
720 | fprintf(stderr, Name ": add new device failed for %s as %d: %s\n", | |
721 | dv->devname, j, strerror(errno)); | |
722 | return 1; | |
723 | } | |
724 | if (verbose >= 0) | |
725 | fprintf(stderr, Name ": added %s\n", dv->devname); | |
726 | break; | |
727 | ||
728 | case 'r': | |
729 | /* hot remove */ | |
730 | if (tst->subarray[0]) { | |
731 | fprintf(stderr, Name ": Cannot remove disks from a" | |
732 | " \'member\' array, perform this" | |
733 | " operation on the parent container\n"); | |
734 | return 1; | |
735 | } | |
736 | if (tst->ss->external) { | |
737 | /* To remove a device from a container, we must | |
738 | * check that it isn't in use in an array. | |
739 | * This involves looking in the 'holders' | |
740 | * directory - there must be just one entry, | |
741 | * the container. | |
742 | * To ensure that it doesn't get used as a | |
743 | * hold spare while we are checking, we | |
744 | * get an O_EXCL open on the container | |
745 | */ | |
746 | int dnum = fd2devnum(fd); | |
747 | lfd = open_dev_excl(dnum); | |
748 | if (lfd < 0) { | |
749 | fprintf(stderr, Name | |
750 | ": Cannot get exclusive access " | |
751 | " to container - odd\n"); | |
752 | return 1; | |
753 | } | |
754 | /* in the detached case it is not possible to | |
755 | * check if we are the unique holder, so just | |
756 | * rely on the 'detached' checks | |
757 | */ | |
758 | if (strcmp(dv->devname, "detached") == 0 || | |
759 | sysfs_unique_holder(dnum, stb.st_rdev)) | |
760 | /* pass */; | |
761 | else { | |
762 | fprintf(stderr, Name | |
763 | ": %s is %s, cannot remove.\n", | |
764 | dnprintable, | |
765 | errno == EEXIST ? "still in use": | |
766 | "not a member"); | |
767 | close(lfd); | |
768 | return 1; | |
769 | } | |
770 | } | |
771 | /* FIXME check that it is a current member */ | |
772 | err = ioctl(fd, HOT_REMOVE_DISK, (unsigned long)stb.st_rdev); | |
773 | if (err && errno == ENODEV) { | |
774 | /* Old kernels rejected this if no personality | |
775 | * registered */ | |
776 | struct mdinfo *sra = sysfs_read(fd, 0, GET_DEVS); | |
777 | struct mdinfo *dv = NULL; | |
778 | if (sra) | |
779 | dv = sra->devs; | |
780 | for ( ; dv ; dv=dv->next) | |
781 | if (dv->disk.major == major(stb.st_rdev) && | |
782 | dv->disk.minor == minor(stb.st_rdev)) | |
783 | break; | |
784 | if (dv) | |
785 | err = sysfs_set_str(sra, dv, | |
786 | "state", "remove"); | |
787 | else | |
788 | err = -1; | |
789 | if (sra) | |
790 | sysfs_free(sra); | |
791 | } | |
792 | if (err) { | |
793 | fprintf(stderr, Name ": hot remove failed " | |
794 | "for %s: %s\n", dnprintable, | |
795 | strerror(errno)); | |
796 | if (lfd >= 0) | |
797 | close(lfd); | |
798 | return 1; | |
799 | } | |
800 | if (tst->ss->external) { | |
801 | /* | |
802 | * Before dropping our exclusive open we make an | |
803 | * attempt at preventing mdmon from seeing an | |
804 | * 'add' event before reconciling this 'remove' | |
805 | * event. | |
806 | */ | |
807 | char *name = devnum2devname(fd2devnum(fd)); | |
808 | ||
809 | if (!name) { | |
810 | fprintf(stderr, Name ": unable to get container name\n"); | |
811 | return 1; | |
812 | } | |
813 | ||
814 | ping_manager(name); | |
815 | free(name); | |
816 | } | |
817 | close(lfd); | |
818 | if (verbose >= 0) | |
819 | fprintf(stderr, Name ": hot removed %s\n", | |
820 | dnprintable); | |
821 | break; | |
822 | ||
823 | case 'f': /* set faulty */ | |
824 | /* FIXME check current member */ | |
825 | if (ioctl(fd, SET_DISK_FAULTY, (unsigned long) stb.st_rdev)) { | |
826 | fprintf(stderr, Name ": set device faulty failed for %s: %s\n", | |
827 | dnprintable, strerror(errno)); | |
828 | return 1; | |
829 | } | |
830 | if (verbose >= 0) | |
831 | fprintf(stderr, Name ": set %s faulty in %s\n", | |
832 | dnprintable, devname); | |
833 | break; | |
834 | } | |
835 | } | |
836 | return 0; | |
837 | ||
838 | } | |
839 | ||
840 | int autodetect(void) | |
841 | { | |
842 | /* Open any md device, and issue the RAID_AUTORUN ioctl */ | |
843 | int rv = 1; | |
844 | int fd = dev_open("9:0", O_RDONLY); | |
845 | if (fd >= 0) { | |
846 | if (ioctl(fd, RAID_AUTORUN, 0) == 0) | |
847 | rv = 0; | |
848 | close(fd); | |
849 | } | |
850 | return rv; | |
851 | } | |
852 | #endif |