]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2001-2013 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@suse.de> | |
23 | */ | |
24 | ||
25 | #include "mdadm.h" | |
26 | #include "md_p.h" | |
27 | #include <sys/socket.h> | |
28 | #include <sys/utsname.h> | |
29 | #include <sys/wait.h> | |
30 | #include <sys/un.h> | |
31 | #include <sys/resource.h> | |
32 | #include <sys/vfs.h> | |
33 | #include <linux/magic.h> | |
34 | #include <poll.h> | |
35 | #include <ctype.h> | |
36 | #include <dirent.h> | |
37 | #include <signal.h> | |
38 | #include <dlfcn.h> | |
39 | ||
40 | ||
41 | /* | |
42 | * following taken from linux/blkpg.h because they aren't | |
43 | * anywhere else and it isn't safe to #include linux/ * stuff. | |
44 | */ | |
45 | ||
46 | #define BLKPG _IO(0x12,105) | |
47 | ||
48 | /* The argument structure */ | |
49 | struct blkpg_ioctl_arg { | |
50 | int op; | |
51 | int flags; | |
52 | int datalen; | |
53 | void *data; | |
54 | }; | |
55 | ||
56 | /* The subfunctions (for the op field) */ | |
57 | #define BLKPG_ADD_PARTITION 1 | |
58 | #define BLKPG_DEL_PARTITION 2 | |
59 | ||
60 | /* Sizes of name fields. Unused at present. */ | |
61 | #define BLKPG_DEVNAMELTH 64 | |
62 | #define BLKPG_VOLNAMELTH 64 | |
63 | ||
64 | /* The data structure for ADD_PARTITION and DEL_PARTITION */ | |
65 | struct blkpg_partition { | |
66 | long long start; /* starting offset in bytes */ | |
67 | long long length; /* length in bytes */ | |
68 | int pno; /* partition number */ | |
69 | char devname[BLKPG_DEVNAMELTH]; /* partition name, like sda5 or c0d1p2, | |
70 | to be used in kernel messages */ | |
71 | char volname[BLKPG_VOLNAMELTH]; /* volume label */ | |
72 | }; | |
73 | ||
74 | #include "part.h" | |
75 | ||
76 | /* Force a compilation error if condition is true */ | |
77 | #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition)) | |
78 | ||
79 | /* Force a compilation error if condition is true, but also produce a | |
80 | result (of value 0 and type size_t), so the expression can be used | |
81 | e.g. in a structure initializer (or where-ever else comma expressions | |
82 | aren't permitted). */ | |
83 | #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); })) | |
84 | ||
85 | static int is_dlm_hooks_ready = 0; | |
86 | ||
87 | int dlm_funs_ready(void) | |
88 | { | |
89 | return is_dlm_hooks_ready ? 1 : 0; | |
90 | } | |
91 | ||
92 | #ifndef MDASSEMBLE | |
93 | static struct dlm_hooks *dlm_hooks = NULL; | |
94 | struct dlm_lock_resource *dlm_lock_res = NULL; | |
95 | static int ast_called = 0; | |
96 | ||
97 | struct dlm_lock_resource { | |
98 | dlm_lshandle_t *ls; | |
99 | struct dlm_lksb lksb; | |
100 | }; | |
101 | ||
102 | /* Using poll(2) to wait for and dispatch ASTs */ | |
103 | static int poll_for_ast(dlm_lshandle_t ls) | |
104 | { | |
105 | struct pollfd pfd; | |
106 | ||
107 | pfd.fd = dlm_hooks->ls_get_fd(ls); | |
108 | pfd.events = POLLIN; | |
109 | ||
110 | while (!ast_called) | |
111 | { | |
112 | if (poll(&pfd, 1, 0) < 0) | |
113 | { | |
114 | perror("poll"); | |
115 | return -1; | |
116 | } | |
117 | dlm_hooks->dispatch(dlm_hooks->ls_get_fd(ls)); | |
118 | } | |
119 | ast_called = 0; | |
120 | ||
121 | return 0; | |
122 | } | |
123 | ||
124 | static void dlm_ast(void *arg) | |
125 | { | |
126 | ast_called = 1; | |
127 | } | |
128 | ||
129 | static char *cluster_name = NULL; | |
130 | /* Create the lockspace, take bitmapXXX locks on all the bitmaps. */ | |
131 | int cluster_get_dlmlock(int *lockid) | |
132 | { | |
133 | int ret = -1; | |
134 | char str[64]; | |
135 | int flags = LKF_NOQUEUE; | |
136 | ||
137 | ret = get_cluster_name(&cluster_name); | |
138 | if (ret) { | |
139 | pr_err("The md can't get cluster name\n"); | |
140 | return -1; | |
141 | } | |
142 | ||
143 | dlm_lock_res = xmalloc(sizeof(struct dlm_lock_resource)); | |
144 | dlm_lock_res->ls = dlm_hooks->create_lockspace(cluster_name, O_RDWR); | |
145 | if (!dlm_lock_res->ls) { | |
146 | pr_err("%s failed to create lockspace\n", cluster_name); | |
147 | return -ENOMEM; | |
148 | } | |
149 | ||
150 | snprintf(str, 64, "bitmap%s", cluster_name); | |
151 | ret = dlm_hooks->ls_lock(dlm_lock_res->ls, LKM_PWMODE, &dlm_lock_res->lksb, | |
152 | flags, str, strlen(str), 0, dlm_ast, | |
153 | dlm_lock_res, NULL, NULL); | |
154 | if (ret) { | |
155 | pr_err("error %d when get PW mode on lock %s\n", errno, str); | |
156 | dlm_hooks->release_lockspace(cluster_name, dlm_lock_res->ls, 1); | |
157 | return ret; | |
158 | } | |
159 | ||
160 | /* Wait for it to complete */ | |
161 | poll_for_ast(dlm_lock_res->ls); | |
162 | *lockid = dlm_lock_res->lksb.sb_lkid; | |
163 | ||
164 | return dlm_lock_res->lksb.sb_status; | |
165 | } | |
166 | ||
167 | int cluster_release_dlmlock(int lockid) | |
168 | { | |
169 | int ret = -1; | |
170 | ||
171 | if (!cluster_name) | |
172 | return -1; | |
173 | ||
174 | ret = dlm_hooks->ls_unlock(dlm_lock_res->ls, lockid, 0, | |
175 | &dlm_lock_res->lksb, dlm_lock_res); | |
176 | if (ret) { | |
177 | pr_err("error %d happened when unlock\n", errno); | |
178 | /* XXX make sure the lock is unlocked eventually */ | |
179 | goto out; | |
180 | } | |
181 | ||
182 | /* Wait for it to complete */ | |
183 | poll_for_ast(dlm_lock_res->ls); | |
184 | ||
185 | errno = dlm_lock_res->lksb.sb_status; | |
186 | if (errno != EUNLOCK) { | |
187 | pr_err("error %d happened in ast when unlock lockspace\n", errno); | |
188 | /* XXX make sure the lockspace is unlocked eventually */ | |
189 | goto out; | |
190 | } | |
191 | ||
192 | ret = dlm_hooks->release_lockspace(cluster_name, dlm_lock_res->ls, 1); | |
193 | if (ret) { | |
194 | pr_err("error %d happened when release lockspace\n", errno); | |
195 | /* XXX make sure the lockspace is released eventually */ | |
196 | goto out; | |
197 | } | |
198 | free(dlm_lock_res); | |
199 | ||
200 | out: | |
201 | return ret; | |
202 | } | |
203 | #else | |
204 | int cluster_get_dlmlock(int *lockid) | |
205 | { | |
206 | return -1; | |
207 | } | |
208 | int cluster_release_dlmlock(int lockid) | |
209 | { | |
210 | return -1; | |
211 | } | |
212 | #endif | |
213 | ||
214 | /* | |
215 | * Parse a 128 bit uuid in 4 integers | |
216 | * format is 32 hexx nibbles with options :.<space> separator | |
217 | * If not exactly 32 hex digits are found, return 0 | |
218 | * else return 1 | |
219 | */ | |
220 | int parse_uuid(char *str, int uuid[4]) | |
221 | { | |
222 | int hit = 0; /* number of Hex digIT */ | |
223 | int i; | |
224 | char c; | |
225 | for (i = 0; i < 4; i++) | |
226 | uuid[i] = 0; | |
227 | ||
228 | while ((c = *str++) != 0) { | |
229 | int n; | |
230 | if (c >= '0' && c <= '9') | |
231 | n = c-'0'; | |
232 | else if (c >= 'a' && c <= 'f') | |
233 | n = 10 + c - 'a'; | |
234 | else if (c >= 'A' && c <= 'F') | |
235 | n = 10 + c - 'A'; | |
236 | else if (strchr(":. -", c)) | |
237 | continue; | |
238 | else return 0; | |
239 | ||
240 | if (hit<32) { | |
241 | uuid[hit/8] <<= 4; | |
242 | uuid[hit/8] += n; | |
243 | } | |
244 | hit++; | |
245 | } | |
246 | if (hit == 32) | |
247 | return 1; | |
248 | return 0; | |
249 | } | |
250 | ||
251 | /* | |
252 | * Get the md version number. | |
253 | * We use the RAID_VERSION ioctl if it is supported | |
254 | * If not, but we have a block device with major '9', we assume | |
255 | * 0.36.0 | |
256 | * | |
257 | * Return version number as 24 but number - assume version parts | |
258 | * always < 255 | |
259 | */ | |
260 | ||
261 | int md_get_version(int fd) | |
262 | { | |
263 | struct stat stb; | |
264 | mdu_version_t vers; | |
265 | ||
266 | if (fstat(fd, &stb)<0) | |
267 | return -1; | |
268 | if ((S_IFMT&stb.st_mode) != S_IFBLK) | |
269 | return -1; | |
270 | ||
271 | if (ioctl(fd, RAID_VERSION, &vers) == 0) | |
272 | return (vers.major*10000) + (vers.minor*100) + vers.patchlevel; | |
273 | if (errno == EACCES) | |
274 | return -1; | |
275 | if (major(stb.st_rdev) == MD_MAJOR) | |
276 | return (3600); | |
277 | return -1; | |
278 | } | |
279 | ||
280 | int get_linux_version() | |
281 | { | |
282 | struct utsname name; | |
283 | char *cp; | |
284 | int a = 0, b = 0,c = 0; | |
285 | if (uname(&name) <0) | |
286 | return -1; | |
287 | ||
288 | cp = name.release; | |
289 | a = strtoul(cp, &cp, 10); | |
290 | if (*cp == '.') | |
291 | b = strtoul(cp+1, &cp, 10); | |
292 | if (*cp == '.') | |
293 | c = strtoul(cp+1, &cp, 10); | |
294 | ||
295 | return (a*1000000)+(b*1000)+c; | |
296 | } | |
297 | ||
298 | #ifndef MDASSEMBLE | |
299 | int mdadm_version(char *version) | |
300 | { | |
301 | int a, b, c; | |
302 | char *cp; | |
303 | ||
304 | if (!version) | |
305 | version = Version; | |
306 | ||
307 | cp = strchr(version, '-'); | |
308 | if (!cp || *(cp+1) != ' ' || *(cp+2) != 'v') | |
309 | return -1; | |
310 | cp += 3; | |
311 | a = strtoul(cp, &cp, 10); | |
312 | if (*cp != '.') | |
313 | return -1; | |
314 | b = strtoul(cp+1, &cp, 10); | |
315 | if (*cp == '.') | |
316 | c = strtoul(cp+1, &cp, 10); | |
317 | else | |
318 | c = 0; | |
319 | if (*cp != ' ' && *cp != '-') | |
320 | return -1; | |
321 | return (a*1000000)+(b*1000)+c; | |
322 | } | |
323 | ||
324 | unsigned long long parse_size(char *size) | |
325 | { | |
326 | /* parse 'size' which should be a number optionally | |
327 | * followed by 'K', 'M', or 'G'. | |
328 | * Without a suffix, K is assumed. | |
329 | * Number returned is in sectors (half-K) | |
330 | * INVALID_SECTORS returned on error. | |
331 | */ | |
332 | char *c; | |
333 | long long s = strtoll(size, &c, 10); | |
334 | if (s > 0) { | |
335 | switch (*c) { | |
336 | case 'K': | |
337 | c++; | |
338 | default: | |
339 | s *= 2; | |
340 | break; | |
341 | case 'M': | |
342 | c++; | |
343 | s *= 1024 * 2; | |
344 | break; | |
345 | case 'G': | |
346 | c++; | |
347 | s *= 1024 * 1024 * 2; | |
348 | break; | |
349 | case 's': /* sectors */ | |
350 | c++; | |
351 | break; | |
352 | } | |
353 | } else | |
354 | s = INVALID_SECTORS; | |
355 | if (*c) | |
356 | s = INVALID_SECTORS; | |
357 | return s; | |
358 | } | |
359 | ||
360 | int parse_layout_10(char *layout) | |
361 | { | |
362 | int copies, rv; | |
363 | char *cp; | |
364 | /* Parse the layout string for raid10 */ | |
365 | /* 'f', 'o' or 'n' followed by a number <= raid_disks */ | |
366 | if ((layout[0] != 'n' && layout[0] != 'f' && layout[0] != 'o') || | |
367 | (copies = strtoul(layout+1, &cp, 10)) < 1 || | |
368 | copies > 200 || | |
369 | *cp) | |
370 | return -1; | |
371 | if (layout[0] == 'n') | |
372 | rv = 256 + copies; | |
373 | else if (layout[0] == 'o') | |
374 | rv = 0x10000 + (copies<<8) + 1; | |
375 | else | |
376 | rv = 1 + (copies<<8); | |
377 | return rv; | |
378 | } | |
379 | ||
380 | int parse_layout_faulty(char *layout) | |
381 | { | |
382 | /* Parse the layout string for 'faulty' */ | |
383 | int ln = strcspn(layout, "0123456789"); | |
384 | char *m = xstrdup(layout); | |
385 | int mode; | |
386 | m[ln] = 0; | |
387 | mode = map_name(faultylayout, m); | |
388 | if (mode == UnSet) | |
389 | return -1; | |
390 | ||
391 | return mode | (atoi(layout+ln)<< ModeShift); | |
392 | } | |
393 | ||
394 | long parse_num(char *num) | |
395 | { | |
396 | /* Either return a valid number, or -1 */ | |
397 | char *c; | |
398 | long rv = strtol(num, &c, 10); | |
399 | if (rv < 0 || *c || !num[0]) | |
400 | return -1; | |
401 | else | |
402 | return rv; | |
403 | } | |
404 | #endif | |
405 | ||
406 | int parse_cluster_confirm_arg(char *input, char **devname, int *slot) | |
407 | { | |
408 | char *dev; | |
409 | *slot = strtoul(input, &dev, 10); | |
410 | if (dev == input || dev[0] != ':') | |
411 | return -1; | |
412 | *devname = dev+1; | |
413 | return 0; | |
414 | } | |
415 | ||
416 | void remove_partitions(int fd) | |
417 | { | |
418 | /* remove partitions from this block devices. | |
419 | * This is used for components added to an array | |
420 | */ | |
421 | #ifdef BLKPG_DEL_PARTITION | |
422 | struct blkpg_ioctl_arg a; | |
423 | struct blkpg_partition p; | |
424 | ||
425 | a.op = BLKPG_DEL_PARTITION; | |
426 | a.data = (void*)&p; | |
427 | a.datalen = sizeof(p); | |
428 | a.flags = 0; | |
429 | memset(a.data, 0, a.datalen); | |
430 | for (p.pno = 0; p.pno < 16; p.pno++) | |
431 | ioctl(fd, BLKPG, &a); | |
432 | #endif | |
433 | } | |
434 | ||
435 | int test_partition(int fd) | |
436 | { | |
437 | /* Check if fd is a whole-disk or a partition. | |
438 | * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION | |
439 | * will return ENXIO on an invalid partition number. | |
440 | */ | |
441 | struct blkpg_ioctl_arg a; | |
442 | struct blkpg_partition p; | |
443 | a.op = BLKPG_DEL_PARTITION; | |
444 | a.data = (void*)&p; | |
445 | a.datalen = sizeof(p); | |
446 | a.flags = 0; | |
447 | memset(a.data, 0, a.datalen); | |
448 | p.pno = 1<<30; | |
449 | if (ioctl(fd, BLKPG, &a) == 0) | |
450 | /* Very unlikely, but not a partition */ | |
451 | return 0; | |
452 | if (errno == ENXIO || errno == ENOTTY) | |
453 | /* not a partition */ | |
454 | return 0; | |
455 | ||
456 | return 1; | |
457 | } | |
458 | ||
459 | int test_partition_from_id(dev_t id) | |
460 | { | |
461 | char buf[20]; | |
462 | int fd, rv; | |
463 | ||
464 | sprintf(buf, "%d:%d", major(id), minor(id)); | |
465 | fd = dev_open(buf, O_RDONLY); | |
466 | if (fd < 0) | |
467 | return -1; | |
468 | rv = test_partition(fd); | |
469 | close(fd); | |
470 | return rv; | |
471 | } | |
472 | ||
473 | int enough(int level, int raid_disks, int layout, int clean, char *avail) | |
474 | { | |
475 | int copies, first; | |
476 | int i; | |
477 | int avail_disks = 0; | |
478 | ||
479 | for (i = 0; i < raid_disks; i++) | |
480 | avail_disks += !!avail[i]; | |
481 | ||
482 | switch (level) { | |
483 | case 10: | |
484 | /* This is the tricky one - we need to check | |
485 | * which actual disks are present. | |
486 | */ | |
487 | copies = (layout&255)* ((layout>>8) & 255); | |
488 | first = 0; | |
489 | do { | |
490 | /* there must be one of the 'copies' form 'first' */ | |
491 | int n = copies; | |
492 | int cnt = 0; | |
493 | int this = first; | |
494 | while (n--) { | |
495 | if (avail[this]) | |
496 | cnt++; | |
497 | this = (this+1) % raid_disks; | |
498 | } | |
499 | if (cnt == 0) | |
500 | return 0; | |
501 | first = (first+(layout&255)) % raid_disks; | |
502 | } while (first != 0); | |
503 | return 1; | |
504 | ||
505 | case LEVEL_MULTIPATH: | |
506 | return avail_disks>= 1; | |
507 | case LEVEL_LINEAR: | |
508 | case 0: | |
509 | return avail_disks == raid_disks; | |
510 | case 1: | |
511 | return avail_disks >= 1; | |
512 | case 4: | |
513 | if (avail_disks == raid_disks - 1 && | |
514 | !avail[raid_disks - 1]) | |
515 | /* If just the parity device is missing, then we | |
516 | * have enough, even if not clean | |
517 | */ | |
518 | return 1; | |
519 | /* FALL THROUGH */ | |
520 | case 5: | |
521 | if (clean) | |
522 | return avail_disks >= raid_disks-1; | |
523 | else | |
524 | return avail_disks >= raid_disks; | |
525 | case 6: | |
526 | if (clean) | |
527 | return avail_disks >= raid_disks-2; | |
528 | else | |
529 | return avail_disks >= raid_disks; | |
530 | default: | |
531 | return 0; | |
532 | } | |
533 | } | |
534 | ||
535 | int enough_fd(int fd) | |
536 | { | |
537 | struct mdu_array_info_s array; | |
538 | struct mdu_disk_info_s disk; | |
539 | int i, rv; | |
540 | char *avail; | |
541 | ||
542 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0 || | |
543 | array.raid_disks <= 0) | |
544 | return 0; | |
545 | avail = xcalloc(array.raid_disks, 1); | |
546 | for (i = 0; i < MAX_DISKS && array.nr_disks > 0; i++) { | |
547 | disk.number = i; | |
548 | if (ioctl(fd, GET_DISK_INFO, &disk) != 0) | |
549 | continue; | |
550 | if (disk.major == 0 && disk.minor == 0) | |
551 | continue; | |
552 | array.nr_disks--; | |
553 | ||
554 | if (! (disk.state & (1<<MD_DISK_SYNC))) | |
555 | continue; | |
556 | if (disk.raid_disk < 0 || disk.raid_disk >= array.raid_disks) | |
557 | continue; | |
558 | avail[disk.raid_disk] = 1; | |
559 | } | |
560 | /* This is used on an active array, so assume it is clean */ | |
561 | rv = enough(array.level, array.raid_disks, array.layout, | |
562 | 1, avail); | |
563 | free(avail); | |
564 | return rv; | |
565 | } | |
566 | ||
567 | const int uuid_zero[4] = { 0, 0, 0, 0 }; | |
568 | ||
569 | int same_uuid(int a[4], int b[4], int swapuuid) | |
570 | { | |
571 | if (swapuuid) { | |
572 | /* parse uuids are hostendian. | |
573 | * uuid's from some superblocks are big-ending | |
574 | * if there is a difference, we need to swap.. | |
575 | */ | |
576 | unsigned char *ac = (unsigned char *)a; | |
577 | unsigned char *bc = (unsigned char *)b; | |
578 | int i; | |
579 | for (i = 0; i < 16; i += 4) { | |
580 | if (ac[i+0] != bc[i+3] || | |
581 | ac[i+1] != bc[i+2] || | |
582 | ac[i+2] != bc[i+1] || | |
583 | ac[i+3] != bc[i+0]) | |
584 | return 0; | |
585 | } | |
586 | return 1; | |
587 | } else { | |
588 | if (a[0]==b[0] && | |
589 | a[1]==b[1] && | |
590 | a[2]==b[2] && | |
591 | a[3]==b[3]) | |
592 | return 1; | |
593 | return 0; | |
594 | } | |
595 | } | |
596 | ||
597 | void copy_uuid(void *a, int b[4], int swapuuid) | |
598 | { | |
599 | if (swapuuid) { | |
600 | /* parse uuids are hostendian. | |
601 | * uuid's from some superblocks are big-ending | |
602 | * if there is a difference, we need to swap.. | |
603 | */ | |
604 | unsigned char *ac = (unsigned char *)a; | |
605 | unsigned char *bc = (unsigned char *)b; | |
606 | int i; | |
607 | for (i = 0; i < 16; i += 4) { | |
608 | ac[i+0] = bc[i+3]; | |
609 | ac[i+1] = bc[i+2]; | |
610 | ac[i+2] = bc[i+1]; | |
611 | ac[i+3] = bc[i+0]; | |
612 | } | |
613 | } else | |
614 | memcpy(a, b, 16); | |
615 | } | |
616 | ||
617 | char *__fname_from_uuid(int id[4], int swap, char *buf, char sep) | |
618 | { | |
619 | int i, j; | |
620 | char uuid[16]; | |
621 | char *c = buf; | |
622 | strcpy(c, "UUID-"); | |
623 | c += strlen(c); | |
624 | copy_uuid(uuid, id, swap); | |
625 | for (i = 0; i < 4; i++) { | |
626 | if (i) | |
627 | *c++ = sep; | |
628 | for (j = 3; j >= 0; j--) { | |
629 | sprintf(c,"%02x", (unsigned char) uuid[j+4*i]); | |
630 | c+= 2; | |
631 | } | |
632 | } | |
633 | return buf; | |
634 | ||
635 | } | |
636 | ||
637 | char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep) | |
638 | { | |
639 | // dirty hack to work around an issue with super1 superblocks... | |
640 | // super1 superblocks need swapuuid set in order for assembly to | |
641 | // work, but can't have it set if we want this printout to match | |
642 | // all the other uuid printouts in super1.c, so we force swapuuid | |
643 | // to 1 to make our printout match the rest of super1 | |
644 | return __fname_from_uuid(info->uuid, (st->ss == &super1) ? 1 : st->ss->swapuuid, buf, sep); | |
645 | } | |
646 | ||
647 | #ifndef MDASSEMBLE | |
648 | int check_ext2(int fd, char *name) | |
649 | { | |
650 | /* | |
651 | * Check for an ext2fs file system. | |
652 | * Superblock is always 1K at 1K offset | |
653 | * | |
654 | * s_magic is le16 at 56 == 0xEF53 | |
655 | * report mtime - le32 at 44 | |
656 | * blocks - le32 at 4 | |
657 | * logblksize - le32 at 24 | |
658 | */ | |
659 | unsigned char sb[1024]; | |
660 | time_t mtime; | |
661 | unsigned long long size; | |
662 | int bsize; | |
663 | if (lseek(fd, 1024,0)!= 1024) | |
664 | return 0; | |
665 | if (read(fd, sb, 1024)!= 1024) | |
666 | return 0; | |
667 | if (sb[56] != 0x53 || sb[57] != 0xef) | |
668 | return 0; | |
669 | ||
670 | mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8; | |
671 | bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8; | |
672 | size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8; | |
673 | size <<= bsize; | |
674 | pr_err("%s appears to contain an ext2fs file system\n", | |
675 | name); | |
676 | cont_err("size=%lluK mtime=%s", size, ctime(&mtime)); | |
677 | return 1; | |
678 | } | |
679 | ||
680 | int check_reiser(int fd, char *name) | |
681 | { | |
682 | /* | |
683 | * superblock is at 64K | |
684 | * size is 1024; | |
685 | * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52 | |
686 | * | |
687 | */ | |
688 | unsigned char sb[1024]; | |
689 | unsigned long long size; | |
690 | if (lseek(fd, 64*1024, 0) != 64*1024) | |
691 | return 0; | |
692 | if (read(fd, sb, 1024) != 1024) | |
693 | return 0; | |
694 | if (strncmp((char*)sb+52, "ReIsErFs",8) != 0 && | |
695 | strncmp((char*)sb+52, "ReIsEr2Fs",9) != 0) | |
696 | return 0; | |
697 | pr_err("%s appears to contain a reiserfs file system\n",name); | |
698 | size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8; | |
699 | cont_err("size = %lluK\n", size*4); | |
700 | ||
701 | return 1; | |
702 | } | |
703 | ||
704 | int check_raid(int fd, char *name) | |
705 | { | |
706 | struct mdinfo info; | |
707 | time_t crtime; | |
708 | char *level; | |
709 | struct supertype *st = guess_super(fd); | |
710 | ||
711 | if (!st) | |
712 | return 0; | |
713 | if (st->ss->add_to_super != NULL) { | |
714 | st->ss->load_super(st, fd, name); | |
715 | /* Looks like a raid array .. */ | |
716 | pr_err("%s appears to be part of a raid array:\n", name); | |
717 | st->ss->getinfo_super(st, &info, NULL); | |
718 | st->ss->free_super(st); | |
719 | crtime = info.array.ctime; | |
720 | level = map_num(pers, info.array.level); | |
721 | if (!level) | |
722 | level = "-unknown-"; | |
723 | cont_err("level=%s devices=%d ctime=%s", | |
724 | level, info.array.raid_disks, ctime(&crtime)); | |
725 | } else { | |
726 | /* Looks like GPT or MBR */ | |
727 | pr_err("partition table exists on %s\n", name); | |
728 | } | |
729 | return 1; | |
730 | } | |
731 | ||
732 | int ask(char *mesg) | |
733 | { | |
734 | char *add = ""; | |
735 | int i; | |
736 | for (i = 0; i < 5; i++) { | |
737 | char buf[100]; | |
738 | fprintf(stderr, "%s%s", mesg, add); | |
739 | fflush(stderr); | |
740 | if (fgets(buf, 100, stdin)==NULL) | |
741 | return 0; | |
742 | if (buf[0]=='y' || buf[0]=='Y') | |
743 | return 1; | |
744 | if (buf[0]=='n' || buf[0]=='N') | |
745 | return 0; | |
746 | add = "(y/n) "; | |
747 | } | |
748 | pr_err("assuming 'no'\n"); | |
749 | return 0; | |
750 | } | |
751 | #endif /* MDASSEMBLE */ | |
752 | ||
753 | int is_standard(char *dev, int *nump) | |
754 | { | |
755 | /* tests if dev is a "standard" md dev name. | |
756 | * i.e if the last component is "/dNN" or "/mdNN", | |
757 | * where NN is a string of digits | |
758 | * Returns 1 if a partitionable standard, | |
759 | * -1 if non-partitonable, | |
760 | * 0 if not a standard name. | |
761 | */ | |
762 | char *d = strrchr(dev, '/'); | |
763 | int type = 0; | |
764 | int num; | |
765 | if (!d) | |
766 | return 0; | |
767 | if (strncmp(d, "/d",2) == 0) | |
768 | d += 2, type = 1; /* /dev/md/dN{pM} */ | |
769 | else if (strncmp(d, "/md_d", 5) == 0) | |
770 | d += 5, type = 1; /* /dev/md_dN{pM} */ | |
771 | else if (strncmp(d, "/md", 3) == 0) | |
772 | d += 3, type = -1; /* /dev/mdN */ | |
773 | else if (d-dev > 3 && strncmp(d-2, "md/", 3) == 0) | |
774 | d += 1, type = -1; /* /dev/md/N */ | |
775 | else | |
776 | return 0; | |
777 | if (!*d) | |
778 | return 0; | |
779 | num = atoi(d); | |
780 | while (isdigit(*d)) | |
781 | d++; | |
782 | if (*d) | |
783 | return 0; | |
784 | if (nump) *nump = num; | |
785 | ||
786 | return type; | |
787 | } | |
788 | ||
789 | unsigned long calc_csum(void *super, int bytes) | |
790 | { | |
791 | unsigned long long newcsum = 0; | |
792 | int i; | |
793 | unsigned int csum; | |
794 | unsigned int *superc = (unsigned int*) super; | |
795 | ||
796 | for(i = 0; i < bytes/4; i++) | |
797 | newcsum += superc[i]; | |
798 | csum = (newcsum& 0xffffffff) + (newcsum>>32); | |
799 | #ifdef __alpha__ | |
800 | /* The in-kernel checksum calculation is always 16bit on | |
801 | * the alpha, though it is 32 bit on i386... | |
802 | * I wonder what it is elsewhere... (it uses an API in | |
803 | * a way that it shouldn't). | |
804 | */ | |
805 | csum = (csum & 0xffff) + (csum >> 16); | |
806 | csum = (csum & 0xffff) + (csum >> 16); | |
807 | #endif | |
808 | return csum; | |
809 | } | |
810 | ||
811 | #ifndef MDASSEMBLE | |
812 | char *human_size(long long bytes) | |
813 | { | |
814 | static char buf[47]; | |
815 | ||
816 | /* We convert bytes to either centi-M{ega,ibi}bytes or | |
817 | * centi-G{igi,ibi}bytes, with appropriate rounding, | |
818 | * and then print 1/100th of those as a decimal. | |
819 | * We allow upto 2048Megabytes before converting to | |
820 | * gigabytes, as that shows more precision and isn't | |
821 | * too large a number. | |
822 | * Terabytes are not yet handled. | |
823 | */ | |
824 | ||
825 | if (bytes < 5000*1024) | |
826 | buf[0] = 0; | |
827 | else if (bytes < 2*1024LL*1024LL*1024LL) { | |
828 | long cMiB = (bytes * 200LL / (1LL<<20) + 1) / 2; | |
829 | long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2; | |
830 | snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)", | |
831 | cMiB/100 , cMiB % 100, | |
832 | cMB/100, cMB % 100); | |
833 | } else { | |
834 | long cGiB = (bytes * 200LL / (1LL<<30) +1) / 2; | |
835 | long cGB = (bytes / (1000000000LL/200LL ) +1) /2; | |
836 | snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)", | |
837 | cGiB/100 , cGiB % 100, | |
838 | cGB/100, cGB % 100); | |
839 | } | |
840 | return buf; | |
841 | } | |
842 | ||
843 | char *human_size_brief(long long bytes, int prefix) | |
844 | { | |
845 | static char buf[30]; | |
846 | ||
847 | /* We convert bytes to either centi-M{ega,ibi}bytes or | |
848 | * centi-G{igi,ibi}bytes, with appropriate rounding, | |
849 | * and then print 1/100th of those as a decimal. | |
850 | * We allow upto 2048Megabytes before converting to | |
851 | * gigabytes, as that shows more precision and isn't | |
852 | * too large a number. | |
853 | * Terabytes are not yet handled. | |
854 | * | |
855 | * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc. | |
856 | * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc. | |
857 | */ | |
858 | ||
859 | if (bytes < 5000*1024) | |
860 | buf[0] = 0; | |
861 | else if (prefix == IEC) { | |
862 | if (bytes < 2*1024LL*1024LL*1024LL) { | |
863 | long cMiB = (bytes * 200LL / (1LL<<20) +1) /2; | |
864 | snprintf(buf, sizeof(buf), "%ld.%02ldMiB", | |
865 | cMiB/100 , cMiB % 100); | |
866 | } else { | |
867 | long cGiB = (bytes * 200LL / (1LL<<30) +1) /2; | |
868 | snprintf(buf, sizeof(buf), "%ld.%02ldGiB", | |
869 | cGiB/100 , cGiB % 100); | |
870 | } | |
871 | } | |
872 | else if (prefix == JEDEC) { | |
873 | if (bytes < 2*1024LL*1024LL*1024LL) { | |
874 | long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2; | |
875 | snprintf(buf, sizeof(buf), "%ld.%02ldMB", | |
876 | cMB/100, cMB % 100); | |
877 | } else { | |
878 | long cGB = (bytes / (1000000000LL/200LL ) +1) /2; | |
879 | snprintf(buf, sizeof(buf), "%ld.%02ldGB", | |
880 | cGB/100 , cGB % 100); | |
881 | } | |
882 | } | |
883 | else | |
884 | buf[0] = 0; | |
885 | ||
886 | return buf; | |
887 | } | |
888 | ||
889 | void print_r10_layout(int layout) | |
890 | { | |
891 | int near = layout & 255; | |
892 | int far = (layout >> 8) & 255; | |
893 | int offset = (layout&0x10000); | |
894 | char *sep = ""; | |
895 | ||
896 | if (near != 1) { | |
897 | printf("%s near=%d", sep, near); | |
898 | sep = ","; | |
899 | } | |
900 | if (far != 1) | |
901 | printf("%s %s=%d", sep, offset?"offset":"far", far); | |
902 | if (near*far == 1) | |
903 | printf("NO REDUNDANCY"); | |
904 | } | |
905 | #endif | |
906 | ||
907 | unsigned long long calc_array_size(int level, int raid_disks, int layout, | |
908 | int chunksize, unsigned long long devsize) | |
909 | { | |
910 | if (level == 1) | |
911 | return devsize; | |
912 | devsize &= ~(unsigned long long)((chunksize>>9)-1); | |
913 | return get_data_disks(level, layout, raid_disks) * devsize; | |
914 | } | |
915 | ||
916 | int get_data_disks(int level, int layout, int raid_disks) | |
917 | { | |
918 | int data_disks = 0; | |
919 | switch (level) { | |
920 | case 0: data_disks = raid_disks; | |
921 | break; | |
922 | case 1: data_disks = 1; | |
923 | break; | |
924 | case 4: | |
925 | case 5: data_disks = raid_disks - 1; | |
926 | break; | |
927 | case 6: data_disks = raid_disks - 2; | |
928 | break; | |
929 | case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255); | |
930 | break; | |
931 | } | |
932 | ||
933 | return data_disks; | |
934 | } | |
935 | ||
936 | dev_t devnm2devid(char *devnm) | |
937 | { | |
938 | /* First look in /sys/block/$DEVNM/dev for %d:%d | |
939 | * If that fails, try parsing out a number | |
940 | */ | |
941 | char path[100]; | |
942 | char *ep; | |
943 | int fd; | |
944 | int mjr,mnr; | |
945 | ||
946 | sprintf(path, "/sys/block/%s/dev", devnm); | |
947 | fd = open(path, O_RDONLY); | |
948 | if (fd >= 0) { | |
949 | char buf[20]; | |
950 | int n = read(fd, buf, sizeof(buf)); | |
951 | close(fd); | |
952 | if (n > 0) | |
953 | buf[n] = 0; | |
954 | if (n > 0 && sscanf(buf, "%d:%d\n", &mjr, &mnr) == 2) | |
955 | return makedev(mjr, mnr); | |
956 | } | |
957 | if (strncmp(devnm, "md_d", 4) == 0 && | |
958 | isdigit(devnm[4]) && | |
959 | (mnr = strtoul(devnm+4, &ep, 10)) >= 0 && | |
960 | ep > devnm && *ep == 0) | |
961 | return makedev(get_mdp_major(), mnr << MdpMinorShift); | |
962 | ||
963 | if (strncmp(devnm, "md", 2) == 0 && | |
964 | isdigit(devnm[2]) && | |
965 | (mnr = strtoul(devnm+2, &ep, 10)) >= 0 && | |
966 | ep > devnm && *ep == 0) | |
967 | return makedev(MD_MAJOR, mnr); | |
968 | ||
969 | return 0; | |
970 | } | |
971 | ||
972 | #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) | |
973 | char *get_md_name(char *devnm) | |
974 | { | |
975 | /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */ | |
976 | /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */ | |
977 | ||
978 | static char devname[50]; | |
979 | struct stat stb; | |
980 | dev_t rdev = devnm2devid(devnm); | |
981 | char *dn; | |
982 | ||
983 | if (rdev == 0) | |
984 | return 0; | |
985 | if (strncmp(devnm, "md_", 3) == 0) { | |
986 | snprintf(devname, sizeof(devname), "/dev/md/%s", | |
987 | devnm + 3); | |
988 | if (stat(devname, &stb) == 0 | |
989 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
990 | && (stb.st_rdev == rdev)) | |
991 | return devname; | |
992 | } | |
993 | snprintf(devname, sizeof(devname), "/dev/%s", devnm); | |
994 | if (stat(devname, &stb) == 0 | |
995 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
996 | && (stb.st_rdev == rdev)) | |
997 | return devname; | |
998 | ||
999 | snprintf(devname, sizeof(devname), "/dev/md/%s", devnm+2); | |
1000 | if (stat(devname, &stb) == 0 | |
1001 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
1002 | && (stb.st_rdev == rdev)) | |
1003 | return devname; | |
1004 | ||
1005 | dn = map_dev(major(rdev), minor(rdev), 0); | |
1006 | if (dn) | |
1007 | return dn; | |
1008 | snprintf(devname, sizeof(devname), "/dev/.tmp.%s", devnm); | |
1009 | if (mknod(devname, S_IFBLK | 0600, rdev) == -1) | |
1010 | if (errno != EEXIST) | |
1011 | return NULL; | |
1012 | ||
1013 | if (stat(devname, &stb) == 0 | |
1014 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
1015 | && (stb.st_rdev == rdev)) | |
1016 | return devname; | |
1017 | unlink(devname); | |
1018 | return NULL; | |
1019 | } | |
1020 | ||
1021 | void put_md_name(char *name) | |
1022 | { | |
1023 | if (strncmp(name, "/dev/.tmp.md", 12) == 0) | |
1024 | unlink(name); | |
1025 | } | |
1026 | #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */ | |
1027 | ||
1028 | int get_maj_min(char *dev, int *major, int *minor) | |
1029 | { | |
1030 | char *e; | |
1031 | *major = strtoul(dev, &e, 0); | |
1032 | return (e > dev && *e == ':' && e[1] && | |
1033 | (*minor = strtoul(e+1, &e, 0)) >= 0 && | |
1034 | *e == 0); | |
1035 | } | |
1036 | ||
1037 | int dev_open(char *dev, int flags) | |
1038 | { | |
1039 | /* like 'open', but if 'dev' matches %d:%d, create a temp | |
1040 | * block device and open that | |
1041 | */ | |
1042 | int fd = -1; | |
1043 | char devname[32]; | |
1044 | int major; | |
1045 | int minor; | |
1046 | ||
1047 | if (!dev) | |
1048 | return -1; | |
1049 | flags |= O_DIRECT; | |
1050 | ||
1051 | if (get_maj_min(dev, &major, &minor)) { | |
1052 | snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d", | |
1053 | (int)getpid(), major, minor); | |
1054 | if (mknod(devname, S_IFBLK|0600, makedev(major, minor)) == 0) { | |
1055 | fd = open(devname, flags); | |
1056 | unlink(devname); | |
1057 | } | |
1058 | if (fd < 0) { | |
1059 | /* Try /tmp as /dev appear to be read-only */ | |
1060 | snprintf(devname, sizeof(devname), "/tmp/.tmp.md.%d:%d:%d", | |
1061 | (int)getpid(), major, minor); | |
1062 | if (mknod(devname, S_IFBLK|0600, makedev(major, minor)) == 0) { | |
1063 | fd = open(devname, flags); | |
1064 | unlink(devname); | |
1065 | } | |
1066 | } | |
1067 | } else | |
1068 | fd = open(dev, flags); | |
1069 | return fd; | |
1070 | } | |
1071 | ||
1072 | int open_dev_flags(char *devnm, int flags) | |
1073 | { | |
1074 | dev_t devid; | |
1075 | char buf[20]; | |
1076 | ||
1077 | devid = devnm2devid(devnm); | |
1078 | sprintf(buf, "%d:%d", major(devid), minor(devid)); | |
1079 | return dev_open(buf, flags); | |
1080 | } | |
1081 | ||
1082 | int open_dev(char *devnm) | |
1083 | { | |
1084 | return open_dev_flags(devnm, O_RDONLY); | |
1085 | } | |
1086 | ||
1087 | int open_dev_excl(char *devnm) | |
1088 | { | |
1089 | char buf[20]; | |
1090 | int i; | |
1091 | int flags = O_RDWR; | |
1092 | dev_t devid = devnm2devid(devnm); | |
1093 | long delay = 1000; | |
1094 | ||
1095 | sprintf(buf, "%d:%d", major(devid), minor(devid)); | |
1096 | for (i = 0 ; i < 25 ; i++) { | |
1097 | int fd = dev_open(buf, flags|O_EXCL); | |
1098 | if (fd >= 0) | |
1099 | return fd; | |
1100 | if (errno == EACCES && flags == O_RDWR) { | |
1101 | flags = O_RDONLY; | |
1102 | continue; | |
1103 | } | |
1104 | if (errno != EBUSY) | |
1105 | return fd; | |
1106 | usleep(delay); | |
1107 | if (delay < 200000) | |
1108 | delay *= 2; | |
1109 | } | |
1110 | return -1; | |
1111 | } | |
1112 | ||
1113 | int same_dev(char *one, char *two) | |
1114 | { | |
1115 | struct stat st1, st2; | |
1116 | if (stat(one, &st1) != 0) | |
1117 | return 0; | |
1118 | if (stat(two, &st2) != 0) | |
1119 | return 0; | |
1120 | if ((st1.st_mode & S_IFMT) != S_IFBLK) | |
1121 | return 0; | |
1122 | if ((st2.st_mode & S_IFMT) != S_IFBLK) | |
1123 | return 0; | |
1124 | return st1.st_rdev == st2.st_rdev; | |
1125 | } | |
1126 | ||
1127 | void wait_for(char *dev, int fd) | |
1128 | { | |
1129 | int i; | |
1130 | struct stat stb_want; | |
1131 | long delay = 1000; | |
1132 | ||
1133 | if (fstat(fd, &stb_want) != 0 || | |
1134 | (stb_want.st_mode & S_IFMT) != S_IFBLK) | |
1135 | return; | |
1136 | ||
1137 | for (i = 0 ; i < 25 ; i++) { | |
1138 | struct stat stb; | |
1139 | if (stat(dev, &stb) == 0 && | |
1140 | (stb.st_mode & S_IFMT) == S_IFBLK && | |
1141 | (stb.st_rdev == stb_want.st_rdev)) | |
1142 | return; | |
1143 | usleep(delay); | |
1144 | if (delay < 200000) | |
1145 | delay *= 2; | |
1146 | } | |
1147 | if (i == 25) | |
1148 | dprintf("timeout waiting for %s\n", dev); | |
1149 | } | |
1150 | ||
1151 | struct superswitch *superlist[] = | |
1152 | { | |
1153 | &super0, &super1, | |
1154 | &super_ddf, &super_imsm, | |
1155 | &mbr, &gpt, | |
1156 | NULL }; | |
1157 | ||
1158 | #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) | |
1159 | ||
1160 | struct supertype *super_by_fd(int fd, char **subarrayp) | |
1161 | { | |
1162 | mdu_array_info_t array; | |
1163 | int vers; | |
1164 | int minor; | |
1165 | struct supertype *st = NULL; | |
1166 | struct mdinfo *sra; | |
1167 | char *verstr; | |
1168 | char version[20]; | |
1169 | int i; | |
1170 | char *subarray = NULL; | |
1171 | char container[32] = ""; | |
1172 | ||
1173 | sra = sysfs_read(fd, NULL, GET_VERSION); | |
1174 | ||
1175 | if (sra) { | |
1176 | vers = sra->array.major_version; | |
1177 | minor = sra->array.minor_version; | |
1178 | verstr = sra->text_version; | |
1179 | } else { | |
1180 | if (ioctl(fd, GET_ARRAY_INFO, &array)) | |
1181 | array.major_version = array.minor_version = 0; | |
1182 | vers = array.major_version; | |
1183 | minor = array.minor_version; | |
1184 | verstr = ""; | |
1185 | } | |
1186 | ||
1187 | if (vers != -1) { | |
1188 | sprintf(version, "%d.%d", vers, minor); | |
1189 | verstr = version; | |
1190 | } | |
1191 | if (minor == -2 && is_subarray(verstr)) { | |
1192 | char *dev = verstr+1; | |
1193 | ||
1194 | subarray = strchr(dev, '/'); | |
1195 | if (subarray) { | |
1196 | *subarray++ = '\0'; | |
1197 | subarray = xstrdup(subarray); | |
1198 | } | |
1199 | strcpy(container, dev); | |
1200 | sysfs_free(sra); | |
1201 | sra = sysfs_read(-1, container, GET_VERSION); | |
1202 | if (sra && sra->text_version[0]) | |
1203 | verstr = sra->text_version; | |
1204 | else | |
1205 | verstr = "-no-metadata-"; | |
1206 | } | |
1207 | ||
1208 | for (i = 0; st == NULL && superlist[i] ; i++) | |
1209 | st = superlist[i]->match_metadata_desc(verstr); | |
1210 | ||
1211 | sysfs_free(sra); | |
1212 | if (st) { | |
1213 | st->sb = NULL; | |
1214 | if (subarrayp) | |
1215 | *subarrayp = subarray; | |
1216 | strcpy(st->container_devnm, container); | |
1217 | strcpy(st->devnm, fd2devnm(fd)); | |
1218 | } else | |
1219 | free(subarray); | |
1220 | ||
1221 | return st; | |
1222 | } | |
1223 | #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */ | |
1224 | ||
1225 | int dev_size_from_id(dev_t id, unsigned long long *size) | |
1226 | { | |
1227 | char buf[20]; | |
1228 | int fd; | |
1229 | ||
1230 | sprintf(buf, "%d:%d", major(id), minor(id)); | |
1231 | fd = dev_open(buf, O_RDONLY); | |
1232 | if (fd < 0) | |
1233 | return 0; | |
1234 | if (get_dev_size(fd, NULL, size)) { | |
1235 | close(fd); | |
1236 | return 1; | |
1237 | } | |
1238 | close(fd); | |
1239 | return 0; | |
1240 | } | |
1241 | ||
1242 | struct supertype *dup_super(struct supertype *orig) | |
1243 | { | |
1244 | struct supertype *st; | |
1245 | ||
1246 | if (!orig) | |
1247 | return orig; | |
1248 | st = xcalloc(1, sizeof(*st)); | |
1249 | st->ss = orig->ss; | |
1250 | st->max_devs = orig->max_devs; | |
1251 | st->minor_version = orig->minor_version; | |
1252 | st->ignore_hw_compat = orig->ignore_hw_compat; | |
1253 | st->data_offset = orig->data_offset; | |
1254 | st->sb = NULL; | |
1255 | st->info = NULL; | |
1256 | return st; | |
1257 | } | |
1258 | ||
1259 | struct supertype *guess_super_type(int fd, enum guess_types guess_type) | |
1260 | { | |
1261 | /* try each load_super to find the best match, | |
1262 | * and return the best superswitch | |
1263 | */ | |
1264 | struct superswitch *ss; | |
1265 | struct supertype *st; | |
1266 | unsigned int besttime = 0; | |
1267 | int bestsuper = -1; | |
1268 | int i; | |
1269 | ||
1270 | st = xcalloc(1, sizeof(*st)); | |
1271 | st->container_devnm[0] = 0; | |
1272 | ||
1273 | for (i = 0 ; superlist[i]; i++) { | |
1274 | int rv; | |
1275 | ss = superlist[i]; | |
1276 | if (guess_type == guess_array && ss->add_to_super == NULL) | |
1277 | continue; | |
1278 | if (guess_type == guess_partitions && ss->add_to_super != NULL) | |
1279 | continue; | |
1280 | memset(st, 0, sizeof(*st)); | |
1281 | st->ignore_hw_compat = 1; | |
1282 | rv = ss->load_super(st, fd, NULL); | |
1283 | if (rv == 0) { | |
1284 | struct mdinfo info; | |
1285 | st->ss->getinfo_super(st, &info, NULL); | |
1286 | if (bestsuper == -1 || | |
1287 | besttime < info.array.ctime) { | |
1288 | bestsuper = i; | |
1289 | besttime = info.array.ctime; | |
1290 | } | |
1291 | ss->free_super(st); | |
1292 | } | |
1293 | } | |
1294 | if (bestsuper != -1) { | |
1295 | int rv; | |
1296 | memset(st, 0, sizeof(*st)); | |
1297 | st->ignore_hw_compat = 1; | |
1298 | rv = superlist[bestsuper]->load_super(st, fd, NULL); | |
1299 | if (rv == 0) { | |
1300 | superlist[bestsuper]->free_super(st); | |
1301 | return st; | |
1302 | } | |
1303 | } | |
1304 | free(st); | |
1305 | return NULL; | |
1306 | } | |
1307 | ||
1308 | /* Return size of device in bytes */ | |
1309 | int get_dev_size(int fd, char *dname, unsigned long long *sizep) | |
1310 | { | |
1311 | unsigned long long ldsize; | |
1312 | struct stat st; | |
1313 | ||
1314 | if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode)) | |
1315 | ldsize = (unsigned long long)st.st_size; | |
1316 | else | |
1317 | #ifdef BLKGETSIZE64 | |
1318 | if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0) | |
1319 | #endif | |
1320 | { | |
1321 | unsigned long dsize; | |
1322 | if (ioctl(fd, BLKGETSIZE, &dsize) == 0) { | |
1323 | ldsize = dsize; | |
1324 | ldsize <<= 9; | |
1325 | } else { | |
1326 | if (dname) | |
1327 | pr_err("Cannot get size of %s: %s\n", | |
1328 | dname, strerror(errno)); | |
1329 | return 0; | |
1330 | } | |
1331 | } | |
1332 | *sizep = ldsize; | |
1333 | return 1; | |
1334 | } | |
1335 | ||
1336 | /* Return sector size of device in bytes */ | |
1337 | int get_dev_sector_size(int fd, char *dname, unsigned int *sectsizep) | |
1338 | { | |
1339 | unsigned int sectsize; | |
1340 | ||
1341 | if (ioctl(fd, BLKSSZGET, §size) != 0) { | |
1342 | if (dname) | |
1343 | pr_err("Cannot get sector size of %s: %s\n", | |
1344 | dname, strerror(errno)); | |
1345 | return 0; | |
1346 | } | |
1347 | ||
1348 | *sectsizep = sectsize; | |
1349 | return 1; | |
1350 | } | |
1351 | ||
1352 | /* Return true if this can only be a container, not a member device. | |
1353 | * i.e. is and md device and size is zero | |
1354 | */ | |
1355 | int must_be_container(int fd) | |
1356 | { | |
1357 | unsigned long long size; | |
1358 | if (md_get_version(fd) < 0) | |
1359 | return 0; | |
1360 | if (get_dev_size(fd, NULL, &size) == 0) | |
1361 | return 1; | |
1362 | if (size == 0) | |
1363 | return 1; | |
1364 | return 0; | |
1365 | } | |
1366 | ||
1367 | /* Sets endofpart parameter to the last block used by the last GPT partition on the device. | |
1368 | * Returns: 1 if successful | |
1369 | * -1 for unknown partition type | |
1370 | * 0 for other errors | |
1371 | */ | |
1372 | static int get_gpt_last_partition_end(int fd, unsigned long long *endofpart) | |
1373 | { | |
1374 | struct GPT gpt; | |
1375 | unsigned char empty_gpt_entry[16]= {0}; | |
1376 | struct GPT_part_entry *part; | |
1377 | char buf[512]; | |
1378 | unsigned long long curr_part_end; | |
1379 | unsigned all_partitions, entry_size; | |
1380 | unsigned part_nr; | |
1381 | unsigned int sector_size = 0; | |
1382 | ||
1383 | *endofpart = 0; | |
1384 | ||
1385 | BUILD_BUG_ON(sizeof(gpt) != 512); | |
1386 | /* skip protective MBR */ | |
1387 | if (!get_dev_sector_size(fd, NULL, §or_size)) | |
1388 | return 0; | |
1389 | lseek(fd, sector_size, SEEK_SET); | |
1390 | /* read GPT header */ | |
1391 | if (read(fd, &gpt, 512) != 512) | |
1392 | return 0; | |
1393 | ||
1394 | /* get the number of partition entries and the entry size */ | |
1395 | all_partitions = __le32_to_cpu(gpt.part_cnt); | |
1396 | entry_size = __le32_to_cpu(gpt.part_size); | |
1397 | ||
1398 | /* Check GPT signature*/ | |
1399 | if (gpt.magic != GPT_SIGNATURE_MAGIC) | |
1400 | return -1; | |
1401 | ||
1402 | /* sanity checks */ | |
1403 | if (all_partitions > 1024 || | |
1404 | entry_size > sizeof(buf)) | |
1405 | return -1; | |
1406 | ||
1407 | part = (struct GPT_part_entry *)buf; | |
1408 | ||
1409 | /* set offset to third block (GPT entries) */ | |
1410 | lseek(fd, sector_size*2, SEEK_SET); | |
1411 | for (part_nr = 0; part_nr < all_partitions; part_nr++) { | |
1412 | /* read partition entry */ | |
1413 | if (read(fd, buf, entry_size) != (ssize_t)entry_size) | |
1414 | return 0; | |
1415 | ||
1416 | /* is this valid partition? */ | |
1417 | if (memcmp(part->type_guid, empty_gpt_entry, 16) != 0) { | |
1418 | /* check the last lba for the current partition */ | |
1419 | curr_part_end = __le64_to_cpu(part->ending_lba); | |
1420 | if (curr_part_end > *endofpart) | |
1421 | *endofpart = curr_part_end; | |
1422 | } | |
1423 | ||
1424 | } | |
1425 | return 1; | |
1426 | } | |
1427 | ||
1428 | /* Sets endofpart parameter to the last block used by the last partition on the device. | |
1429 | * Returns: 1 if successful | |
1430 | * -1 for unknown partition type | |
1431 | * 0 for other errors | |
1432 | */ | |
1433 | static int get_last_partition_end(int fd, unsigned long long *endofpart) | |
1434 | { | |
1435 | struct MBR boot_sect; | |
1436 | unsigned long long curr_part_end; | |
1437 | unsigned part_nr; | |
1438 | unsigned int sector_size; | |
1439 | int retval = 0; | |
1440 | ||
1441 | *endofpart = 0; | |
1442 | ||
1443 | BUILD_BUG_ON(sizeof(boot_sect) != 512); | |
1444 | /* read MBR */ | |
1445 | lseek(fd, 0, 0); | |
1446 | if (read(fd, &boot_sect, 512) != 512) | |
1447 | goto abort; | |
1448 | ||
1449 | /* check MBP signature */ | |
1450 | if (boot_sect.magic == MBR_SIGNATURE_MAGIC) { | |
1451 | retval = 1; | |
1452 | /* found the correct signature */ | |
1453 | ||
1454 | for (part_nr = 0; part_nr < MBR_PARTITIONS; part_nr++) { | |
1455 | /* | |
1456 | * Have to make every access through boot_sect rather | |
1457 | * than using a pointer to the partition table (or an | |
1458 | * entry), since the entries are not properly aligned. | |
1459 | */ | |
1460 | ||
1461 | /* check for GPT type */ | |
1462 | if (boot_sect.parts[part_nr].part_type == | |
1463 | MBR_GPT_PARTITION_TYPE) { | |
1464 | retval = get_gpt_last_partition_end(fd, endofpart); | |
1465 | break; | |
1466 | } | |
1467 | /* check the last used lba for the current partition */ | |
1468 | curr_part_end = | |
1469 | __le32_to_cpu(boot_sect.parts[part_nr].first_sect_lba) + | |
1470 | __le32_to_cpu(boot_sect.parts[part_nr].blocks_num); | |
1471 | if (curr_part_end > *endofpart) | |
1472 | *endofpart = curr_part_end; | |
1473 | } | |
1474 | } else { | |
1475 | /* Unknown partition table */ | |
1476 | retval = -1; | |
1477 | } | |
1478 | /* calculate number of 512-byte blocks */ | |
1479 | if (get_dev_sector_size(fd, NULL, §or_size)) | |
1480 | *endofpart *= (sector_size / 512); | |
1481 | abort: | |
1482 | return retval; | |
1483 | } | |
1484 | ||
1485 | int check_partitions(int fd, char *dname, unsigned long long freesize, | |
1486 | unsigned long long size) | |
1487 | { | |
1488 | /* | |
1489 | * Check where the last partition ends | |
1490 | */ | |
1491 | unsigned long long endofpart; | |
1492 | ||
1493 | if (get_last_partition_end(fd, &endofpart) > 0) { | |
1494 | /* There appears to be a partition table here */ | |
1495 | if (freesize == 0) { | |
1496 | /* partitions will not be visible in new device */ | |
1497 | pr_err("partition table exists on %s but will be lost or\n" | |
1498 | " meaningless after creating array\n", | |
1499 | dname); | |
1500 | return 1; | |
1501 | } else if (endofpart > freesize) { | |
1502 | /* last partition overlaps metadata */ | |
1503 | pr_err("metadata will over-write last partition on %s.\n", | |
1504 | dname); | |
1505 | return 1; | |
1506 | } else if (size && endofpart > size) { | |
1507 | /* partitions will be truncated in new device */ | |
1508 | pr_err("array size is too small to cover all partitions on %s.\n", | |
1509 | dname); | |
1510 | return 1; | |
1511 | } | |
1512 | } | |
1513 | return 0; | |
1514 | } | |
1515 | ||
1516 | int open_container(int fd) | |
1517 | { | |
1518 | /* 'fd' is a block device. Find out if it is in use | |
1519 | * by a container, and return an open fd on that container. | |
1520 | */ | |
1521 | char path[256]; | |
1522 | char *e; | |
1523 | DIR *dir; | |
1524 | struct dirent *de; | |
1525 | int dfd, n; | |
1526 | char buf[200]; | |
1527 | int major, minor; | |
1528 | struct stat st; | |
1529 | ||
1530 | if (fstat(fd, &st) != 0) | |
1531 | return -1; | |
1532 | sprintf(path, "/sys/dev/block/%d:%d/holders", | |
1533 | (int)major(st.st_rdev), (int)minor(st.st_rdev)); | |
1534 | e = path + strlen(path); | |
1535 | ||
1536 | dir = opendir(path); | |
1537 | if (!dir) | |
1538 | return -1; | |
1539 | while ((de = readdir(dir))) { | |
1540 | if (de->d_ino == 0) | |
1541 | continue; | |
1542 | if (de->d_name[0] == '.') | |
1543 | continue; | |
1544 | /* Need to make sure it is a container and not a volume */ | |
1545 | sprintf(e, "/%s/md/metadata_version", de->d_name); | |
1546 | dfd = open(path, O_RDONLY); | |
1547 | if (dfd < 0) | |
1548 | continue; | |
1549 | n = read(dfd, buf, sizeof(buf)); | |
1550 | close(dfd); | |
1551 | if (n <= 0 || (unsigned)n >= sizeof(buf)) | |
1552 | continue; | |
1553 | buf[n] = 0; | |
1554 | if (strncmp(buf, "external", 8) != 0 || | |
1555 | n < 10 || | |
1556 | buf[9] == '/') | |
1557 | continue; | |
1558 | sprintf(e, "/%s/dev", de->d_name); | |
1559 | dfd = open(path, O_RDONLY); | |
1560 | if (dfd < 0) | |
1561 | continue; | |
1562 | n = read(dfd, buf, sizeof(buf)); | |
1563 | close(dfd); | |
1564 | if (n <= 0 || (unsigned)n >= sizeof(buf)) | |
1565 | continue; | |
1566 | buf[n] = 0; | |
1567 | if (sscanf(buf, "%d:%d", &major, &minor) != 2) | |
1568 | continue; | |
1569 | sprintf(buf, "%d:%d", major, minor); | |
1570 | dfd = dev_open(buf, O_RDONLY); | |
1571 | if (dfd >= 0) { | |
1572 | closedir(dir); | |
1573 | return dfd; | |
1574 | } | |
1575 | } | |
1576 | closedir(dir); | |
1577 | return -1; | |
1578 | } | |
1579 | ||
1580 | struct superswitch *version_to_superswitch(char *vers) | |
1581 | { | |
1582 | int i; | |
1583 | ||
1584 | for (i = 0; superlist[i]; i++) { | |
1585 | struct superswitch *ss = superlist[i]; | |
1586 | ||
1587 | if (strcmp(vers, ss->name) == 0) | |
1588 | return ss; | |
1589 | } | |
1590 | ||
1591 | return NULL; | |
1592 | } | |
1593 | ||
1594 | int metadata_container_matches(char *metadata, char *devnm) | |
1595 | { | |
1596 | /* Check if 'devnm' is the container named in 'metadata' | |
1597 | * which is | |
1598 | * /containername/componentname or | |
1599 | * -containername/componentname | |
1600 | */ | |
1601 | int l; | |
1602 | if (*metadata != '/' && *metadata != '-') | |
1603 | return 0; | |
1604 | l = strlen(devnm); | |
1605 | if (strncmp(metadata+1, devnm, l) != 0) | |
1606 | return 0; | |
1607 | if (metadata[l+1] != '/') | |
1608 | return 0; | |
1609 | return 1; | |
1610 | } | |
1611 | ||
1612 | int metadata_subdev_matches(char *metadata, char *devnm) | |
1613 | { | |
1614 | /* Check if 'devnm' is the subdev named in 'metadata' | |
1615 | * which is | |
1616 | * /containername/subdev or | |
1617 | * -containername/subdev | |
1618 | */ | |
1619 | char *sl; | |
1620 | if (*metadata != '/' && *metadata != '-') | |
1621 | return 0; | |
1622 | sl = strchr(metadata+1, '/'); | |
1623 | if (!sl) | |
1624 | return 0; | |
1625 | if (strcmp(sl+1, devnm) == 0) | |
1626 | return 1; | |
1627 | return 0; | |
1628 | } | |
1629 | ||
1630 | int is_container_member(struct mdstat_ent *mdstat, char *container) | |
1631 | { | |
1632 | if (mdstat->metadata_version == NULL || | |
1633 | strncmp(mdstat->metadata_version, "external:", 9) != 0 || | |
1634 | !metadata_container_matches(mdstat->metadata_version+9, container)) | |
1635 | return 0; | |
1636 | ||
1637 | return 1; | |
1638 | } | |
1639 | ||
1640 | int is_subarray_active(char *subarray, char *container) | |
1641 | { | |
1642 | struct mdstat_ent *mdstat = mdstat_read(0, 0); | |
1643 | struct mdstat_ent *ent; | |
1644 | ||
1645 | for (ent = mdstat; ent; ent = ent->next) | |
1646 | if (is_container_member(ent, container)) | |
1647 | if (strcmp(to_subarray(ent, container), subarray) == 0) | |
1648 | break; | |
1649 | ||
1650 | free_mdstat(mdstat); | |
1651 | ||
1652 | return ent != NULL; | |
1653 | } | |
1654 | ||
1655 | /* open_subarray - opens a subarray in a container | |
1656 | * @dev: container device name | |
1657 | * @st: empty supertype | |
1658 | * @quiet: block reporting errors flag | |
1659 | * | |
1660 | * On success returns an fd to a container and fills in *st | |
1661 | */ | |
1662 | int open_subarray(char *dev, char *subarray, struct supertype *st, int quiet) | |
1663 | { | |
1664 | struct mdinfo *mdi; | |
1665 | struct mdinfo *info; | |
1666 | int fd, err = 1; | |
1667 | char *_devnm; | |
1668 | ||
1669 | fd = open(dev, O_RDWR|O_EXCL); | |
1670 | if (fd < 0) { | |
1671 | if (!quiet) | |
1672 | pr_err("Couldn't open %s, aborting\n", | |
1673 | dev); | |
1674 | return -1; | |
1675 | } | |
1676 | ||
1677 | _devnm = fd2devnm(fd); | |
1678 | if (_devnm == NULL) { | |
1679 | if (!quiet) | |
1680 | pr_err("Failed to determine device number for %s\n", | |
1681 | dev); | |
1682 | goto close_fd; | |
1683 | } | |
1684 | strcpy(st->devnm, _devnm); | |
1685 | ||
1686 | mdi = sysfs_read(fd, st->devnm, GET_VERSION|GET_LEVEL); | |
1687 | if (!mdi) { | |
1688 | if (!quiet) | |
1689 | pr_err("Failed to read sysfs for %s\n", | |
1690 | dev); | |
1691 | goto close_fd; | |
1692 | } | |
1693 | ||
1694 | if (mdi->array.level != UnSet) { | |
1695 | if (!quiet) | |
1696 | pr_err("%s is not a container\n", dev); | |
1697 | goto free_sysfs; | |
1698 | } | |
1699 | ||
1700 | st->ss = version_to_superswitch(mdi->text_version); | |
1701 | if (!st->ss) { | |
1702 | if (!quiet) | |
1703 | pr_err("Operation not supported for %s metadata\n", | |
1704 | mdi->text_version); | |
1705 | goto free_sysfs; | |
1706 | } | |
1707 | ||
1708 | if (st->devnm[0] == 0) { | |
1709 | if (!quiet) | |
1710 | pr_err("Failed to allocate device name\n"); | |
1711 | goto free_sysfs; | |
1712 | } | |
1713 | ||
1714 | if (!st->ss->load_container) { | |
1715 | if (!quiet) | |
1716 | pr_err("%s is not a container\n", dev); | |
1717 | goto free_sysfs; | |
1718 | } | |
1719 | ||
1720 | if (st->ss->load_container(st, fd, NULL)) { | |
1721 | if (!quiet) | |
1722 | pr_err("Failed to load metadata for %s\n", | |
1723 | dev); | |
1724 | goto free_sysfs; | |
1725 | } | |
1726 | ||
1727 | info = st->ss->container_content(st, subarray); | |
1728 | if (!info) { | |
1729 | if (!quiet) | |
1730 | pr_err("Failed to find subarray-%s in %s\n", | |
1731 | subarray, dev); | |
1732 | goto free_super; | |
1733 | } | |
1734 | free(info); | |
1735 | ||
1736 | err = 0; | |
1737 | ||
1738 | free_super: | |
1739 | if (err) | |
1740 | st->ss->free_super(st); | |
1741 | free_sysfs: | |
1742 | sysfs_free(mdi); | |
1743 | close_fd: | |
1744 | if (err) | |
1745 | close(fd); | |
1746 | ||
1747 | if (err) | |
1748 | return -1; | |
1749 | else | |
1750 | return fd; | |
1751 | } | |
1752 | ||
1753 | int add_disk(int mdfd, struct supertype *st, | |
1754 | struct mdinfo *sra, struct mdinfo *info) | |
1755 | { | |
1756 | /* Add a device to an array, in one of 2 ways. */ | |
1757 | int rv; | |
1758 | #ifndef MDASSEMBLE | |
1759 | if (st->ss->external) { | |
1760 | if (info->disk.state & (1<<MD_DISK_SYNC)) | |
1761 | info->recovery_start = MaxSector; | |
1762 | else | |
1763 | info->recovery_start = 0; | |
1764 | rv = sysfs_add_disk(sra, info, 0); | |
1765 | if (! rv) { | |
1766 | struct mdinfo *sd2; | |
1767 | for (sd2 = sra->devs; sd2; sd2=sd2->next) | |
1768 | if (sd2 == info) | |
1769 | break; | |
1770 | if (sd2 == NULL) { | |
1771 | sd2 = xmalloc(sizeof(*sd2)); | |
1772 | *sd2 = *info; | |
1773 | sd2->next = sra->devs; | |
1774 | sra->devs = sd2; | |
1775 | } | |
1776 | } | |
1777 | } else | |
1778 | #endif | |
1779 | rv = ioctl(mdfd, ADD_NEW_DISK, &info->disk); | |
1780 | return rv; | |
1781 | } | |
1782 | ||
1783 | int remove_disk(int mdfd, struct supertype *st, | |
1784 | struct mdinfo *sra, struct mdinfo *info) | |
1785 | { | |
1786 | int rv; | |
1787 | /* Remove the disk given by 'info' from the array */ | |
1788 | #ifndef MDASSEMBLE | |
1789 | if (st->ss->external) | |
1790 | rv = sysfs_set_str(sra, info, "slot", "none"); | |
1791 | else | |
1792 | #endif | |
1793 | rv = ioctl(mdfd, HOT_REMOVE_DISK, makedev(info->disk.major, | |
1794 | info->disk.minor)); | |
1795 | return rv; | |
1796 | } | |
1797 | ||
1798 | int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info) | |
1799 | { | |
1800 | /* Initialise kernel's knowledge of array. | |
1801 | * This varies between externally managed arrays | |
1802 | * and older kernels | |
1803 | */ | |
1804 | int vers = md_get_version(mdfd); | |
1805 | int rv; | |
1806 | ||
1807 | #ifndef MDASSEMBLE | |
1808 | if (st->ss->external) | |
1809 | rv = sysfs_set_array(info, vers); | |
1810 | else | |
1811 | #endif | |
1812 | if ((vers % 100) >= 1) { /* can use different versions */ | |
1813 | mdu_array_info_t inf; | |
1814 | memset(&inf, 0, sizeof(inf)); | |
1815 | inf.major_version = info->array.major_version; | |
1816 | inf.minor_version = info->array.minor_version; | |
1817 | rv = ioctl(mdfd, SET_ARRAY_INFO, &inf); | |
1818 | } else | |
1819 | rv = ioctl(mdfd, SET_ARRAY_INFO, NULL); | |
1820 | return rv; | |
1821 | } | |
1822 | ||
1823 | unsigned long long min_recovery_start(struct mdinfo *array) | |
1824 | { | |
1825 | /* find the minimum recovery_start in an array for metadata | |
1826 | * formats that only record per-array recovery progress instead | |
1827 | * of per-device | |
1828 | */ | |
1829 | unsigned long long recovery_start = MaxSector; | |
1830 | struct mdinfo *d; | |
1831 | ||
1832 | for (d = array->devs; d; d = d->next) | |
1833 | recovery_start = min(recovery_start, d->recovery_start); | |
1834 | ||
1835 | return recovery_start; | |
1836 | } | |
1837 | ||
1838 | int mdmon_pid(char *devnm) | |
1839 | { | |
1840 | char path[100]; | |
1841 | char pid[10]; | |
1842 | int fd; | |
1843 | int n; | |
1844 | ||
1845 | sprintf(path, "%s/%s.pid", MDMON_DIR, devnm); | |
1846 | ||
1847 | fd = open(path, O_RDONLY | O_NOATIME, 0); | |
1848 | ||
1849 | if (fd < 0) | |
1850 | return -1; | |
1851 | n = read(fd, pid, 9); | |
1852 | close(fd); | |
1853 | if (n <= 0) | |
1854 | return -1; | |
1855 | return atoi(pid); | |
1856 | } | |
1857 | ||
1858 | int mdmon_running(char *devnm) | |
1859 | { | |
1860 | int pid = mdmon_pid(devnm); | |
1861 | if (pid <= 0) | |
1862 | return 0; | |
1863 | if (kill(pid, 0) == 0) | |
1864 | return 1; | |
1865 | return 0; | |
1866 | } | |
1867 | ||
1868 | int start_mdmon(char *devnm) | |
1869 | { | |
1870 | int i, skipped; | |
1871 | int len; | |
1872 | pid_t pid; | |
1873 | int status; | |
1874 | char pathbuf[1024]; | |
1875 | char *paths[4] = { | |
1876 | pathbuf, | |
1877 | BINDIR "/mdmon", | |
1878 | "./mdmon", | |
1879 | NULL | |
1880 | }; | |
1881 | ||
1882 | if (check_env("MDADM_NO_MDMON")) | |
1883 | return 0; | |
1884 | ||
1885 | len = readlink("/proc/self/exe", pathbuf, sizeof(pathbuf)-1); | |
1886 | if (len > 0) { | |
1887 | char *sl; | |
1888 | pathbuf[len] = 0; | |
1889 | sl = strrchr(pathbuf, '/'); | |
1890 | if (sl) | |
1891 | sl++; | |
1892 | else | |
1893 | sl = pathbuf; | |
1894 | strcpy(sl, "mdmon"); | |
1895 | } else | |
1896 | pathbuf[0] = '\0'; | |
1897 | ||
1898 | /* First try to run systemctl */ | |
1899 | if (!check_env("MDADM_NO_SYSTEMCTL")) | |
1900 | switch(fork()) { | |
1901 | case 0: | |
1902 | /* FIXME yuk. CLOSE_EXEC?? */ | |
1903 | skipped = 0; | |
1904 | for (i = 3; skipped < 20; i++) | |
1905 | if (close(i) < 0) | |
1906 | skipped++; | |
1907 | else | |
1908 | skipped = 0; | |
1909 | ||
1910 | /* Don't want to see error messages from | |
1911 | * systemctl. If the service doesn't exist, | |
1912 | * we start mdmon ourselves. | |
1913 | */ | |
1914 | close(2); | |
1915 | open("/dev/null", O_WRONLY); | |
1916 | snprintf(pathbuf, sizeof(pathbuf), "mdmon@%s.service", | |
1917 | devnm); | |
1918 | status = execl("/usr/bin/systemctl", "systemctl", | |
1919 | "start", | |
1920 | pathbuf, NULL); | |
1921 | status = execl("/bin/systemctl", "systemctl", "start", | |
1922 | pathbuf, NULL); | |
1923 | exit(1); | |
1924 | case -1: pr_err("cannot run mdmon. Array remains readonly\n"); | |
1925 | return -1; | |
1926 | default: /* parent - good */ | |
1927 | pid = wait(&status); | |
1928 | if (pid >= 0 && status == 0) | |
1929 | return 0; | |
1930 | } | |
1931 | ||
1932 | /* That failed, try running mdmon directly */ | |
1933 | switch(fork()) { | |
1934 | case 0: | |
1935 | /* FIXME yuk. CLOSE_EXEC?? */ | |
1936 | skipped = 0; | |
1937 | for (i = 3; skipped < 20; i++) | |
1938 | if (close(i) < 0) | |
1939 | skipped++; | |
1940 | else | |
1941 | skipped = 0; | |
1942 | ||
1943 | for (i = 0; paths[i]; i++) | |
1944 | if (paths[i][0]) { | |
1945 | execl(paths[i], paths[i], | |
1946 | devnm, NULL); | |
1947 | } | |
1948 | exit(1); | |
1949 | case -1: pr_err("cannot run mdmon. Array remains readonly\n"); | |
1950 | return -1; | |
1951 | default: /* parent - good */ | |
1952 | pid = wait(&status); | |
1953 | if (pid < 0 || status != 0) { | |
1954 | pr_err("failed to launch mdmon. Array remains readonly\n"); | |
1955 | return -1; | |
1956 | } | |
1957 | } | |
1958 | return 0; | |
1959 | } | |
1960 | ||
1961 | __u32 random32(void) | |
1962 | { | |
1963 | __u32 rv; | |
1964 | int rfd = open("/dev/urandom", O_RDONLY); | |
1965 | if (rfd < 0 || read(rfd, &rv, 4) != 4) | |
1966 | rv = random(); | |
1967 | if (rfd >= 0) | |
1968 | close(rfd); | |
1969 | return rv; | |
1970 | } | |
1971 | ||
1972 | void random_uuid(__u8 *buf) | |
1973 | { | |
1974 | int fd, i, len; | |
1975 | __u32 r[4]; | |
1976 | ||
1977 | fd = open("/dev/urandom", O_RDONLY); | |
1978 | if (fd < 0) | |
1979 | goto use_random; | |
1980 | len = read(fd, buf, 16); | |
1981 | close(fd); | |
1982 | if (len != 16) | |
1983 | goto use_random; | |
1984 | ||
1985 | return; | |
1986 | ||
1987 | use_random: | |
1988 | for (i = 0; i < 4; i++) | |
1989 | r[i] = random(); | |
1990 | memcpy(buf, r, 16); | |
1991 | } | |
1992 | ||
1993 | #ifndef MDASSEMBLE | |
1994 | int flush_metadata_updates(struct supertype *st) | |
1995 | { | |
1996 | int sfd; | |
1997 | if (!st->updates) { | |
1998 | st->update_tail = NULL; | |
1999 | return -1; | |
2000 | } | |
2001 | ||
2002 | sfd = connect_monitor(st->container_devnm); | |
2003 | if (sfd < 0) | |
2004 | return -1; | |
2005 | ||
2006 | while (st->updates) { | |
2007 | struct metadata_update *mu = st->updates; | |
2008 | st->updates = mu->next; | |
2009 | ||
2010 | send_message(sfd, mu, 0); | |
2011 | wait_reply(sfd, 0); | |
2012 | free(mu->buf); | |
2013 | free(mu); | |
2014 | } | |
2015 | ack(sfd, 0); | |
2016 | wait_reply(sfd, 0); | |
2017 | close(sfd); | |
2018 | st->update_tail = NULL; | |
2019 | return 0; | |
2020 | } | |
2021 | ||
2022 | void append_metadata_update(struct supertype *st, void *buf, int len) | |
2023 | { | |
2024 | ||
2025 | struct metadata_update *mu = xmalloc(sizeof(*mu)); | |
2026 | ||
2027 | mu->buf = buf; | |
2028 | mu->len = len; | |
2029 | mu->space = NULL; | |
2030 | mu->space_list = NULL; | |
2031 | mu->next = NULL; | |
2032 | *st->update_tail = mu; | |
2033 | st->update_tail = &mu->next; | |
2034 | } | |
2035 | #endif /* MDASSEMBLE */ | |
2036 | ||
2037 | #ifdef __TINYC__ | |
2038 | /* tinyc doesn't optimize this check in ioctl.h out ... */ | |
2039 | unsigned int __invalid_size_argument_for_IOC = 0; | |
2040 | #endif | |
2041 | ||
2042 | int experimental(void) | |
2043 | { | |
2044 | if (check_env("MDADM_EXPERIMENTAL")) | |
2045 | return 1; | |
2046 | else { | |
2047 | pr_err("To use this feature MDADM_EXPERIMENTAL environment variable has to be defined.\n"); | |
2048 | return 0; | |
2049 | } | |
2050 | } | |
2051 | ||
2052 | /* Pick all spares matching given criteria from a container | |
2053 | * if min_size == 0 do not check size | |
2054 | * if domlist == NULL do not check domains | |
2055 | * if spare_group given add it to domains of each spare | |
2056 | * metadata allows to test domains using metadata of destination array */ | |
2057 | struct mdinfo *container_choose_spares(struct supertype *st, | |
2058 | unsigned long long min_size, | |
2059 | struct domainlist *domlist, | |
2060 | char *spare_group, | |
2061 | const char *metadata, int get_one) | |
2062 | { | |
2063 | struct mdinfo *d, **dp, *disks = NULL; | |
2064 | ||
2065 | /* get list of all disks in container */ | |
2066 | if (st->ss->getinfo_super_disks) | |
2067 | disks = st->ss->getinfo_super_disks(st); | |
2068 | ||
2069 | if (!disks) | |
2070 | return disks; | |
2071 | /* find spare devices on the list */ | |
2072 | dp = &disks->devs; | |
2073 | disks->array.spare_disks = 0; | |
2074 | while (*dp) { | |
2075 | int found = 0; | |
2076 | d = *dp; | |
2077 | if (d->disk.state == 0) { | |
2078 | /* check if size is acceptable */ | |
2079 | unsigned long long dev_size; | |
2080 | dev_t dev = makedev(d->disk.major,d->disk.minor); | |
2081 | ||
2082 | if (!min_size || | |
2083 | (dev_size_from_id(dev, &dev_size) && | |
2084 | dev_size >= min_size)) | |
2085 | found = 1; | |
2086 | /* check if domain matches */ | |
2087 | if (found && domlist) { | |
2088 | struct dev_policy *pol = devid_policy(dev); | |
2089 | if (spare_group) | |
2090 | pol_add(&pol, pol_domain, | |
2091 | spare_group, NULL); | |
2092 | if (domain_test(domlist, pol, metadata) != 1) | |
2093 | found = 0; | |
2094 | dev_policy_free(pol); | |
2095 | } | |
2096 | } | |
2097 | if (found) { | |
2098 | dp = &d->next; | |
2099 | disks->array.spare_disks++; | |
2100 | if (get_one) { | |
2101 | sysfs_free(*dp); | |
2102 | d->next = NULL; | |
2103 | } | |
2104 | } else { | |
2105 | *dp = d->next; | |
2106 | d->next = NULL; | |
2107 | sysfs_free(d); | |
2108 | } | |
2109 | } | |
2110 | return disks; | |
2111 | } | |
2112 | ||
2113 | /* Checks if paths point to the same device | |
2114 | * Returns 0 if they do. | |
2115 | * Returns 1 if they don't. | |
2116 | * Returns -1 if something went wrong, | |
2117 | * e.g. paths are empty or the files | |
2118 | * they point to don't exist */ | |
2119 | int compare_paths (char* path1, char* path2) | |
2120 | { | |
2121 | struct stat st1,st2; | |
2122 | ||
2123 | if (path1 == NULL || path2 == NULL) | |
2124 | return -1; | |
2125 | if (stat(path1,&st1) != 0) | |
2126 | return -1; | |
2127 | if (stat(path2,&st2) != 0) | |
2128 | return -1; | |
2129 | if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev)) | |
2130 | return 0; | |
2131 | return 1; | |
2132 | } | |
2133 | ||
2134 | /* Make sure we can open as many devices as needed */ | |
2135 | void enable_fds(int devices) | |
2136 | { | |
2137 | unsigned int fds = 20 + devices; | |
2138 | struct rlimit lim; | |
2139 | if (getrlimit(RLIMIT_NOFILE, &lim) != 0 | |
2140 | || lim.rlim_cur >= fds) | |
2141 | return; | |
2142 | if (lim.rlim_max < fds) | |
2143 | lim.rlim_max = fds; | |
2144 | lim.rlim_cur = fds; | |
2145 | setrlimit(RLIMIT_NOFILE, &lim); | |
2146 | } | |
2147 | ||
2148 | int in_initrd(void) | |
2149 | { | |
2150 | /* This is based on similar function in systemd. */ | |
2151 | struct statfs s; | |
2152 | /* statfs.f_type is signed long on s390x and MIPS, causing all | |
2153 | sorts of sign extension problems with RAMFS_MAGIC being | |
2154 | defined as 0x858458f6 */ | |
2155 | return statfs("/", &s) >= 0 && | |
2156 | ((unsigned long)s.f_type == TMPFS_MAGIC || | |
2157 | ((unsigned long)s.f_type & 0xFFFFFFFFUL) == | |
2158 | ((unsigned long)RAMFS_MAGIC & 0xFFFFFFFFUL)); | |
2159 | } | |
2160 | ||
2161 | void reopen_mddev(int mdfd) | |
2162 | { | |
2163 | /* Re-open without any O_EXCL, but keep | |
2164 | * the same fd | |
2165 | */ | |
2166 | char *devnm; | |
2167 | int fd; | |
2168 | devnm = fd2devnm(mdfd); | |
2169 | close(mdfd); | |
2170 | fd = open_dev(devnm); | |
2171 | if (fd >= 0 && fd != mdfd) | |
2172 | dup2(fd, mdfd); | |
2173 | } | |
2174 | ||
2175 | #ifndef MDASSEMBLE | |
2176 | static struct cmap_hooks *cmap_hooks = NULL; | |
2177 | static int is_cmap_hooks_ready = 0; | |
2178 | ||
2179 | void set_cmap_hooks(void) | |
2180 | { | |
2181 | cmap_hooks = xmalloc(sizeof(struct cmap_hooks)); | |
2182 | cmap_hooks->cmap_handle = dlopen("libcmap.so.4", RTLD_NOW | RTLD_LOCAL); | |
2183 | if (!cmap_hooks->cmap_handle) | |
2184 | return; | |
2185 | ||
2186 | cmap_hooks->initialize = dlsym(cmap_hooks->cmap_handle, "cmap_initialize"); | |
2187 | cmap_hooks->get_string = dlsym(cmap_hooks->cmap_handle, "cmap_get_string"); | |
2188 | cmap_hooks->finalize = dlsym(cmap_hooks->cmap_handle, "cmap_finalize"); | |
2189 | ||
2190 | if (!cmap_hooks->initialize || !cmap_hooks->get_string || | |
2191 | !cmap_hooks->finalize) | |
2192 | dlclose(cmap_hooks->cmap_handle); | |
2193 | else | |
2194 | is_cmap_hooks_ready = 1; | |
2195 | } | |
2196 | ||
2197 | int get_cluster_name(char **cluster_name) | |
2198 | { | |
2199 | int rv = -1; | |
2200 | cmap_handle_t handle; | |
2201 | ||
2202 | if (!is_cmap_hooks_ready) | |
2203 | return rv; | |
2204 | ||
2205 | rv = cmap_hooks->initialize(&handle); | |
2206 | if (rv != CS_OK) | |
2207 | goto out; | |
2208 | ||
2209 | rv = cmap_hooks->get_string(handle, "totem.cluster_name", cluster_name); | |
2210 | if (rv != CS_OK) { | |
2211 | free(*cluster_name); | |
2212 | rv = -1; | |
2213 | goto name_err; | |
2214 | } | |
2215 | ||
2216 | rv = 0; | |
2217 | name_err: | |
2218 | cmap_hooks->finalize(handle); | |
2219 | out: | |
2220 | return rv; | |
2221 | } | |
2222 | ||
2223 | void set_dlm_hooks(void) | |
2224 | { | |
2225 | dlm_hooks = xmalloc(sizeof(struct dlm_hooks)); | |
2226 | dlm_hooks->dlm_handle = dlopen("libdlm_lt.so.3", RTLD_NOW | RTLD_LOCAL); | |
2227 | if (!dlm_hooks->dlm_handle) | |
2228 | return; | |
2229 | ||
2230 | dlm_hooks->create_lockspace = dlsym(dlm_hooks->dlm_handle, "dlm_create_lockspace"); | |
2231 | dlm_hooks->release_lockspace = dlsym(dlm_hooks->dlm_handle, "dlm_release_lockspace"); | |
2232 | dlm_hooks->ls_lock = dlsym(dlm_hooks->dlm_handle, "dlm_ls_lock"); | |
2233 | dlm_hooks->ls_unlock = dlsym(dlm_hooks->dlm_handle, "dlm_ls_unlock"); | |
2234 | dlm_hooks->ls_get_fd = dlsym(dlm_hooks->dlm_handle, "dlm_ls_get_fd"); | |
2235 | dlm_hooks->dispatch = dlsym(dlm_hooks->dlm_handle, "dlm_dispatch"); | |
2236 | ||
2237 | if (!dlm_hooks->create_lockspace || !dlm_hooks->ls_lock || | |
2238 | !dlm_hooks->ls_unlock || !dlm_hooks->release_lockspace || | |
2239 | !dlm_hooks->ls_get_fd || !dlm_hooks->dispatch) | |
2240 | dlclose(dlm_hooks->dlm_handle); | |
2241 | else | |
2242 | is_dlm_hooks_ready = 1; | |
2243 | } | |
2244 | ||
2245 | void set_hooks(void) | |
2246 | { | |
2247 | set_dlm_hooks(); | |
2248 | set_cmap_hooks(); | |
2249 | } | |
2250 | #endif |