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