]>
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 | static struct dlm_hooks *dlm_hooks = NULL; | |
93 | struct dlm_lock_resource *dlm_lock_res = NULL; | |
94 | static int ast_called = 0; | |
95 | ||
96 | struct dlm_lock_resource { | |
97 | dlm_lshandle_t *ls; | |
98 | struct dlm_lksb lksb; | |
99 | }; | |
100 | ||
101 | /* Using poll(2) to wait for and dispatch ASTs */ | |
102 | static int poll_for_ast(dlm_lshandle_t ls) | |
103 | { | |
104 | struct pollfd pfd; | |
105 | ||
106 | pfd.fd = dlm_hooks->ls_get_fd(ls); | |
107 | pfd.events = POLLIN; | |
108 | ||
109 | while (!ast_called) | |
110 | { | |
111 | if (poll(&pfd, 1, 0) < 0) | |
112 | { | |
113 | perror("poll"); | |
114 | return -1; | |
115 | } | |
116 | dlm_hooks->dispatch(dlm_hooks->ls_get_fd(ls)); | |
117 | } | |
118 | ast_called = 0; | |
119 | ||
120 | return 0; | |
121 | } | |
122 | ||
123 | static void dlm_ast(void *arg) | |
124 | { | |
125 | ast_called = 1; | |
126 | } | |
127 | ||
128 | static char *cluster_name = NULL; | |
129 | /* Create the lockspace, take bitmapXXX locks on all the bitmaps. */ | |
130 | int cluster_get_dlmlock(int *lockid) | |
131 | { | |
132 | int ret = -1; | |
133 | char str[64]; | |
134 | int flags = LKF_NOQUEUE; | |
135 | ||
136 | ret = get_cluster_name(&cluster_name); | |
137 | if (ret) { | |
138 | pr_err("The md can't get cluster name\n"); | |
139 | return -1; | |
140 | } | |
141 | ||
142 | dlm_lock_res = xmalloc(sizeof(struct dlm_lock_resource)); | |
143 | dlm_lock_res->ls = dlm_hooks->create_lockspace(cluster_name, O_RDWR); | |
144 | if (!dlm_lock_res->ls) { | |
145 | pr_err("%s failed to create lockspace\n", cluster_name); | |
146 | return -ENOMEM; | |
147 | } | |
148 | ||
149 | snprintf(str, 64, "bitmap%s", cluster_name); | |
150 | ret = dlm_hooks->ls_lock(dlm_lock_res->ls, LKM_PWMODE, &dlm_lock_res->lksb, | |
151 | flags, str, strlen(str), 0, dlm_ast, | |
152 | dlm_lock_res, NULL, NULL); | |
153 | if (ret) { | |
154 | pr_err("error %d when get PW mode on lock %s\n", errno, str); | |
155 | dlm_hooks->release_lockspace(cluster_name, dlm_lock_res->ls, 1); | |
156 | return ret; | |
157 | } | |
158 | ||
159 | /* Wait for it to complete */ | |
160 | poll_for_ast(dlm_lock_res->ls); | |
161 | *lockid = dlm_lock_res->lksb.sb_lkid; | |
162 | ||
163 | return dlm_lock_res->lksb.sb_status; | |
164 | } | |
165 | ||
166 | int cluster_release_dlmlock(int lockid) | |
167 | { | |
168 | int ret = -1; | |
169 | ||
170 | if (!cluster_name) | |
171 | return -1; | |
172 | ||
173 | ret = dlm_hooks->ls_unlock(dlm_lock_res->ls, lockid, 0, | |
174 | &dlm_lock_res->lksb, dlm_lock_res); | |
175 | if (ret) { | |
176 | pr_err("error %d happened when unlock\n", errno); | |
177 | /* XXX make sure the lock is unlocked eventually */ | |
178 | goto out; | |
179 | } | |
180 | ||
181 | /* Wait for it to complete */ | |
182 | poll_for_ast(dlm_lock_res->ls); | |
183 | ||
184 | errno = dlm_lock_res->lksb.sb_status; | |
185 | if (errno != EUNLOCK) { | |
186 | pr_err("error %d happened in ast when unlock lockspace\n", errno); | |
187 | /* XXX make sure the lockspace is unlocked eventually */ | |
188 | goto out; | |
189 | } | |
190 | ||
191 | ret = dlm_hooks->release_lockspace(cluster_name, dlm_lock_res->ls, 1); | |
192 | if (ret) { | |
193 | pr_err("error %d happened when release lockspace\n", errno); | |
194 | /* XXX make sure the lockspace is released eventually */ | |
195 | goto out; | |
196 | } | |
197 | free(dlm_lock_res); | |
198 | ||
199 | out: | |
200 | return ret; | |
201 | } | |
202 | ||
203 | /* | |
204 | * Get array info from the kernel. Longer term we want to deprecate the | |
205 | * ioctl and get it from sysfs. | |
206 | */ | |
207 | int md_get_array_info(int fd, struct mdu_array_info_s *array) | |
208 | { | |
209 | return ioctl(fd, GET_ARRAY_INFO, array); | |
210 | } | |
211 | ||
212 | /* | |
213 | * Set array info | |
214 | */ | |
215 | int md_set_array_info(int fd, struct mdu_array_info_s *array) | |
216 | { | |
217 | return ioctl(fd, SET_ARRAY_INFO, array); | |
218 | } | |
219 | ||
220 | /* | |
221 | * Get disk info from the kernel. | |
222 | */ | |
223 | int md_get_disk_info(int fd, struct mdu_disk_info_s *disk) | |
224 | { | |
225 | return ioctl(fd, GET_DISK_INFO, disk); | |
226 | } | |
227 | ||
228 | /* | |
229 | * Parse a 128 bit uuid in 4 integers | |
230 | * format is 32 hexx nibbles with options :.<space> separator | |
231 | * If not exactly 32 hex digits are found, return 0 | |
232 | * else return 1 | |
233 | */ | |
234 | int parse_uuid(char *str, int uuid[4]) | |
235 | { | |
236 | int hit = 0; /* number of Hex digIT */ | |
237 | int i; | |
238 | char c; | |
239 | for (i = 0; i < 4; i++) | |
240 | uuid[i] = 0; | |
241 | ||
242 | while ((c = *str++) != 0) { | |
243 | int n; | |
244 | if (c >= '0' && c <= '9') | |
245 | n = c-'0'; | |
246 | else if (c >= 'a' && c <= 'f') | |
247 | n = 10 + c - 'a'; | |
248 | else if (c >= 'A' && c <= 'F') | |
249 | n = 10 + c - 'A'; | |
250 | else if (strchr(":. -", c)) | |
251 | continue; | |
252 | else return 0; | |
253 | ||
254 | if (hit<32) { | |
255 | uuid[hit/8] <<= 4; | |
256 | uuid[hit/8] += n; | |
257 | } | |
258 | hit++; | |
259 | } | |
260 | if (hit == 32) | |
261 | return 1; | |
262 | return 0; | |
263 | } | |
264 | ||
265 | int get_linux_version() | |
266 | { | |
267 | struct utsname name; | |
268 | char *cp; | |
269 | int a = 0, b = 0,c = 0; | |
270 | if (uname(&name) <0) | |
271 | return -1; | |
272 | ||
273 | cp = name.release; | |
274 | a = strtoul(cp, &cp, 10); | |
275 | if (*cp == '.') | |
276 | b = strtoul(cp+1, &cp, 10); | |
277 | if (*cp == '.') | |
278 | c = strtoul(cp+1, &cp, 10); | |
279 | ||
280 | return (a*1000000)+(b*1000)+c; | |
281 | } | |
282 | ||
283 | int mdadm_version(char *version) | |
284 | { | |
285 | int a, b, c; | |
286 | char *cp; | |
287 | ||
288 | if (!version) | |
289 | version = Version; | |
290 | ||
291 | cp = strchr(version, '-'); | |
292 | if (!cp || *(cp+1) != ' ' || *(cp+2) != 'v') | |
293 | return -1; | |
294 | cp += 3; | |
295 | a = strtoul(cp, &cp, 10); | |
296 | if (*cp != '.') | |
297 | return -1; | |
298 | b = strtoul(cp+1, &cp, 10); | |
299 | if (*cp == '.') | |
300 | c = strtoul(cp+1, &cp, 10); | |
301 | else | |
302 | c = 0; | |
303 | if (*cp != ' ' && *cp != '-') | |
304 | return -1; | |
305 | return (a*1000000)+(b*1000)+c; | |
306 | } | |
307 | ||
308 | unsigned long long parse_size(char *size) | |
309 | { | |
310 | /* parse 'size' which should be a number optionally | |
311 | * followed by 'K', 'M', or 'G'. | |
312 | * Without a suffix, K is assumed. | |
313 | * Number returned is in sectors (half-K) | |
314 | * INVALID_SECTORS returned on error. | |
315 | */ | |
316 | char *c; | |
317 | long long s = strtoll(size, &c, 10); | |
318 | if (s > 0) { | |
319 | switch (*c) { | |
320 | case 'K': | |
321 | c++; | |
322 | default: | |
323 | s *= 2; | |
324 | break; | |
325 | case 'M': | |
326 | c++; | |
327 | s *= 1024 * 2; | |
328 | break; | |
329 | case 'G': | |
330 | c++; | |
331 | s *= 1024 * 1024 * 2; | |
332 | break; | |
333 | case 's': /* sectors */ | |
334 | c++; | |
335 | break; | |
336 | } | |
337 | } else | |
338 | s = INVALID_SECTORS; | |
339 | if (*c) | |
340 | s = INVALID_SECTORS; | |
341 | return s; | |
342 | } | |
343 | ||
344 | int parse_layout_10(char *layout) | |
345 | { | |
346 | int copies, rv; | |
347 | char *cp; | |
348 | /* Parse the layout string for raid10 */ | |
349 | /* 'f', 'o' or 'n' followed by a number <= raid_disks */ | |
350 | if ((layout[0] != 'n' && layout[0] != 'f' && layout[0] != 'o') || | |
351 | (copies = strtoul(layout+1, &cp, 10)) < 1 || | |
352 | copies > 200 || | |
353 | *cp) | |
354 | return -1; | |
355 | if (layout[0] == 'n') | |
356 | rv = 256 + copies; | |
357 | else if (layout[0] == 'o') | |
358 | rv = 0x10000 + (copies<<8) + 1; | |
359 | else | |
360 | rv = 1 + (copies<<8); | |
361 | return rv; | |
362 | } | |
363 | ||
364 | int parse_layout_faulty(char *layout) | |
365 | { | |
366 | /* Parse the layout string for 'faulty' */ | |
367 | int ln = strcspn(layout, "0123456789"); | |
368 | char *m = xstrdup(layout); | |
369 | int mode; | |
370 | m[ln] = 0; | |
371 | mode = map_name(faultylayout, m); | |
372 | if (mode == UnSet) | |
373 | return -1; | |
374 | ||
375 | return mode | (atoi(layout+ln)<< ModeShift); | |
376 | } | |
377 | ||
378 | long parse_num(char *num) | |
379 | { | |
380 | /* Either return a valid number, or -1 */ | |
381 | char *c; | |
382 | long rv = strtol(num, &c, 10); | |
383 | if (rv < 0 || *c || !num[0]) | |
384 | return -1; | |
385 | else | |
386 | return rv; | |
387 | } | |
388 | ||
389 | int parse_cluster_confirm_arg(char *input, char **devname, int *slot) | |
390 | { | |
391 | char *dev; | |
392 | *slot = strtoul(input, &dev, 10); | |
393 | if (dev == input || dev[0] != ':') | |
394 | return -1; | |
395 | *devname = dev+1; | |
396 | return 0; | |
397 | } | |
398 | ||
399 | void remove_partitions(int fd) | |
400 | { | |
401 | /* remove partitions from this block devices. | |
402 | * This is used for components added to an array | |
403 | */ | |
404 | #ifdef BLKPG_DEL_PARTITION | |
405 | struct blkpg_ioctl_arg a; | |
406 | struct blkpg_partition p; | |
407 | ||
408 | a.op = BLKPG_DEL_PARTITION; | |
409 | a.data = (void*)&p; | |
410 | a.datalen = sizeof(p); | |
411 | a.flags = 0; | |
412 | memset(a.data, 0, a.datalen); | |
413 | for (p.pno = 0; p.pno < 16; p.pno++) | |
414 | ioctl(fd, BLKPG, &a); | |
415 | #endif | |
416 | } | |
417 | ||
418 | int test_partition(int fd) | |
419 | { | |
420 | /* Check if fd is a whole-disk or a partition. | |
421 | * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION | |
422 | * will return ENXIO on an invalid partition number. | |
423 | */ | |
424 | struct blkpg_ioctl_arg a; | |
425 | struct blkpg_partition p; | |
426 | a.op = BLKPG_DEL_PARTITION; | |
427 | a.data = (void*)&p; | |
428 | a.datalen = sizeof(p); | |
429 | a.flags = 0; | |
430 | memset(a.data, 0, a.datalen); | |
431 | p.pno = 1<<30; | |
432 | if (ioctl(fd, BLKPG, &a) == 0) | |
433 | /* Very unlikely, but not a partition */ | |
434 | return 0; | |
435 | if (errno == ENXIO || errno == ENOTTY) | |
436 | /* not a partition */ | |
437 | return 0; | |
438 | ||
439 | return 1; | |
440 | } | |
441 | ||
442 | int test_partition_from_id(dev_t id) | |
443 | { | |
444 | char buf[20]; | |
445 | int fd, rv; | |
446 | ||
447 | sprintf(buf, "%d:%d", major(id), minor(id)); | |
448 | fd = dev_open(buf, O_RDONLY); | |
449 | if (fd < 0) | |
450 | return -1; | |
451 | rv = test_partition(fd); | |
452 | close(fd); | |
453 | return rv; | |
454 | } | |
455 | ||
456 | int enough(int level, int raid_disks, int layout, int clean, char *avail) | |
457 | { | |
458 | int copies, first; | |
459 | int i; | |
460 | int avail_disks = 0; | |
461 | ||
462 | for (i = 0; i < raid_disks; i++) | |
463 | avail_disks += !!avail[i]; | |
464 | ||
465 | switch (level) { | |
466 | case 10: | |
467 | /* This is the tricky one - we need to check | |
468 | * which actual disks are present. | |
469 | */ | |
470 | copies = (layout&255)* ((layout>>8) & 255); | |
471 | first = 0; | |
472 | do { | |
473 | /* there must be one of the 'copies' form 'first' */ | |
474 | int n = copies; | |
475 | int cnt = 0; | |
476 | int this = first; | |
477 | while (n--) { | |
478 | if (avail[this]) | |
479 | cnt++; | |
480 | this = (this+1) % raid_disks; | |
481 | } | |
482 | if (cnt == 0) | |
483 | return 0; | |
484 | first = (first+(layout&255)) % raid_disks; | |
485 | } while (first != 0); | |
486 | return 1; | |
487 | ||
488 | case LEVEL_MULTIPATH: | |
489 | return avail_disks>= 1; | |
490 | case LEVEL_LINEAR: | |
491 | case 0: | |
492 | return avail_disks == raid_disks; | |
493 | case 1: | |
494 | return avail_disks >= 1; | |
495 | case 4: | |
496 | if (avail_disks == raid_disks - 1 && | |
497 | !avail[raid_disks - 1]) | |
498 | /* If just the parity device is missing, then we | |
499 | * have enough, even if not clean | |
500 | */ | |
501 | return 1; | |
502 | /* FALL THROUGH */ | |
503 | case 5: | |
504 | if (clean) | |
505 | return avail_disks >= raid_disks-1; | |
506 | else | |
507 | return avail_disks >= raid_disks; | |
508 | case 6: | |
509 | if (clean) | |
510 | return avail_disks >= raid_disks-2; | |
511 | else | |
512 | return avail_disks >= raid_disks; | |
513 | default: | |
514 | return 0; | |
515 | } | |
516 | } | |
517 | ||
518 | int enough_fd(int fd) | |
519 | { | |
520 | struct mdu_array_info_s array; | |
521 | struct mdu_disk_info_s disk; | |
522 | int i, rv; | |
523 | char *avail; | |
524 | ||
525 | if (md_get_array_info(fd, &array) != 0 || array.raid_disks <= 0) | |
526 | return 0; | |
527 | avail = xcalloc(array.raid_disks, 1); | |
528 | for (i = 0; i < MAX_DISKS && array.nr_disks > 0; i++) { | |
529 | disk.number = i; | |
530 | if (md_get_disk_info(fd, &disk) != 0) | |
531 | continue; | |
532 | if (disk.major == 0 && disk.minor == 0) | |
533 | continue; | |
534 | array.nr_disks--; | |
535 | ||
536 | if (! (disk.state & (1<<MD_DISK_SYNC))) | |
537 | continue; | |
538 | if (disk.raid_disk < 0 || disk.raid_disk >= array.raid_disks) | |
539 | continue; | |
540 | avail[disk.raid_disk] = 1; | |
541 | } | |
542 | /* This is used on an active array, so assume it is clean */ | |
543 | rv = enough(array.level, array.raid_disks, array.layout, | |
544 | 1, avail); | |
545 | free(avail); | |
546 | return rv; | |
547 | } | |
548 | ||
549 | const int uuid_zero[4] = { 0, 0, 0, 0 }; | |
550 | ||
551 | int same_uuid(int a[4], int b[4], int swapuuid) | |
552 | { | |
553 | if (swapuuid) { | |
554 | /* parse uuids are hostendian. | |
555 | * uuid's from some superblocks are big-ending | |
556 | * if there is a difference, we need to swap.. | |
557 | */ | |
558 | unsigned char *ac = (unsigned char *)a; | |
559 | unsigned char *bc = (unsigned char *)b; | |
560 | int i; | |
561 | for (i = 0; i < 16; i += 4) { | |
562 | if (ac[i+0] != bc[i+3] || | |
563 | ac[i+1] != bc[i+2] || | |
564 | ac[i+2] != bc[i+1] || | |
565 | ac[i+3] != bc[i+0]) | |
566 | return 0; | |
567 | } | |
568 | return 1; | |
569 | } else { | |
570 | if (a[0]==b[0] && | |
571 | a[1]==b[1] && | |
572 | a[2]==b[2] && | |
573 | a[3]==b[3]) | |
574 | return 1; | |
575 | return 0; | |
576 | } | |
577 | } | |
578 | ||
579 | void copy_uuid(void *a, int b[4], int swapuuid) | |
580 | { | |
581 | if (swapuuid) { | |
582 | /* parse uuids are hostendian. | |
583 | * uuid's from some superblocks are big-ending | |
584 | * if there is a difference, we need to swap.. | |
585 | */ | |
586 | unsigned char *ac = (unsigned char *)a; | |
587 | unsigned char *bc = (unsigned char *)b; | |
588 | int i; | |
589 | for (i = 0; i < 16; i += 4) { | |
590 | ac[i+0] = bc[i+3]; | |
591 | ac[i+1] = bc[i+2]; | |
592 | ac[i+2] = bc[i+1]; | |
593 | ac[i+3] = bc[i+0]; | |
594 | } | |
595 | } else | |
596 | memcpy(a, b, 16); | |
597 | } | |
598 | ||
599 | char *__fname_from_uuid(int id[4], int swap, char *buf, char sep) | |
600 | { | |
601 | int i, j; | |
602 | char uuid[16]; | |
603 | char *c = buf; | |
604 | strcpy(c, "UUID-"); | |
605 | c += strlen(c); | |
606 | copy_uuid(uuid, id, swap); | |
607 | for (i = 0; i < 4; i++) { | |
608 | if (i) | |
609 | *c++ = sep; | |
610 | for (j = 3; j >= 0; j--) { | |
611 | sprintf(c,"%02x", (unsigned char) uuid[j+4*i]); | |
612 | c+= 2; | |
613 | } | |
614 | } | |
615 | return buf; | |
616 | ||
617 | } | |
618 | ||
619 | char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep) | |
620 | { | |
621 | // dirty hack to work around an issue with super1 superblocks... | |
622 | // super1 superblocks need swapuuid set in order for assembly to | |
623 | // work, but can't have it set if we want this printout to match | |
624 | // all the other uuid printouts in super1.c, so we force swapuuid | |
625 | // to 1 to make our printout match the rest of super1 | |
626 | return __fname_from_uuid(info->uuid, (st->ss == &super1) ? 1 : st->ss->swapuuid, buf, sep); | |
627 | } | |
628 | ||
629 | int check_ext2(int fd, char *name) | |
630 | { | |
631 | /* | |
632 | * Check for an ext2fs file system. | |
633 | * Superblock is always 1K at 1K offset | |
634 | * | |
635 | * s_magic is le16 at 56 == 0xEF53 | |
636 | * report mtime - le32 at 44 | |
637 | * blocks - le32 at 4 | |
638 | * logblksize - le32 at 24 | |
639 | */ | |
640 | unsigned char sb[1024]; | |
641 | time_t mtime; | |
642 | unsigned long long size; | |
643 | int bsize; | |
644 | if (lseek(fd, 1024,0)!= 1024) | |
645 | return 0; | |
646 | if (read(fd, sb, 1024)!= 1024) | |
647 | return 0; | |
648 | if (sb[56] != 0x53 || sb[57] != 0xef) | |
649 | return 0; | |
650 | ||
651 | mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8; | |
652 | bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8; | |
653 | size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8; | |
654 | size <<= bsize; | |
655 | pr_err("%s appears to contain an ext2fs file system\n", | |
656 | name); | |
657 | cont_err("size=%lluK mtime=%s", size, ctime(&mtime)); | |
658 | return 1; | |
659 | } | |
660 | ||
661 | int check_reiser(int fd, char *name) | |
662 | { | |
663 | /* | |
664 | * superblock is at 64K | |
665 | * size is 1024; | |
666 | * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52 | |
667 | * | |
668 | */ | |
669 | unsigned char sb[1024]; | |
670 | unsigned long long size; | |
671 | if (lseek(fd, 64*1024, 0) != 64*1024) | |
672 | return 0; | |
673 | if (read(fd, sb, 1024) != 1024) | |
674 | return 0; | |
675 | if (strncmp((char*)sb+52, "ReIsErFs",8) != 0 && | |
676 | strncmp((char*)sb+52, "ReIsEr2Fs",9) != 0) | |
677 | return 0; | |
678 | pr_err("%s appears to contain a reiserfs file system\n",name); | |
679 | size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8; | |
680 | cont_err("size = %lluK\n", size*4); | |
681 | ||
682 | return 1; | |
683 | } | |
684 | ||
685 | int check_raid(int fd, char *name) | |
686 | { | |
687 | struct mdinfo info; | |
688 | time_t crtime; | |
689 | char *level; | |
690 | struct supertype *st = guess_super(fd); | |
691 | ||
692 | if (!st) | |
693 | return 0; | |
694 | if (st->ss->add_to_super != NULL) { | |
695 | st->ss->load_super(st, fd, name); | |
696 | /* Looks like a raid array .. */ | |
697 | pr_err("%s appears to be part of a raid array:\n", name); | |
698 | st->ss->getinfo_super(st, &info, NULL); | |
699 | st->ss->free_super(st); | |
700 | crtime = info.array.ctime; | |
701 | level = map_num(pers, info.array.level); | |
702 | if (!level) | |
703 | level = "-unknown-"; | |
704 | cont_err("level=%s devices=%d ctime=%s", | |
705 | level, info.array.raid_disks, ctime(&crtime)); | |
706 | } else { | |
707 | /* Looks like GPT or MBR */ | |
708 | pr_err("partition table exists on %s\n", name); | |
709 | } | |
710 | return 1; | |
711 | } | |
712 | ||
713 | int ask(char *mesg) | |
714 | { | |
715 | char *add = ""; | |
716 | int i; | |
717 | for (i = 0; i < 5; i++) { | |
718 | char buf[100]; | |
719 | fprintf(stderr, "%s%s", mesg, add); | |
720 | fflush(stderr); | |
721 | if (fgets(buf, 100, stdin)==NULL) | |
722 | return 0; | |
723 | if (buf[0]=='y' || buf[0]=='Y') | |
724 | return 1; | |
725 | if (buf[0]=='n' || buf[0]=='N') | |
726 | return 0; | |
727 | add = "(y/n) "; | |
728 | } | |
729 | pr_err("assuming 'no'\n"); | |
730 | return 0; | |
731 | } | |
732 | ||
733 | int is_standard(char *dev, int *nump) | |
734 | { | |
735 | /* tests if dev is a "standard" md dev name. | |
736 | * i.e if the last component is "/dNN" or "/mdNN", | |
737 | * where NN is a string of digits | |
738 | * Returns 1 if a partitionable standard, | |
739 | * -1 if non-partitonable, | |
740 | * 0 if not a standard name. | |
741 | */ | |
742 | char *d = strrchr(dev, '/'); | |
743 | int type = 0; | |
744 | int num; | |
745 | if (!d) | |
746 | return 0; | |
747 | if (strncmp(d, "/d",2) == 0) | |
748 | d += 2, type = 1; /* /dev/md/dN{pM} */ | |
749 | else if (strncmp(d, "/md_d", 5) == 0) | |
750 | d += 5, type = 1; /* /dev/md_dN{pM} */ | |
751 | else if (strncmp(d, "/md", 3) == 0) | |
752 | d += 3, type = -1; /* /dev/mdN */ | |
753 | else if (d-dev > 3 && strncmp(d-2, "md/", 3) == 0) | |
754 | d += 1, type = -1; /* /dev/md/N */ | |
755 | else | |
756 | return 0; | |
757 | if (!*d) | |
758 | return 0; | |
759 | num = atoi(d); | |
760 | while (isdigit(*d)) | |
761 | d++; | |
762 | if (*d) | |
763 | return 0; | |
764 | if (nump) *nump = num; | |
765 | ||
766 | return type; | |
767 | } | |
768 | ||
769 | unsigned long calc_csum(void *super, int bytes) | |
770 | { | |
771 | unsigned long long newcsum = 0; | |
772 | int i; | |
773 | unsigned int csum; | |
774 | unsigned int *superc = (unsigned int*) super; | |
775 | ||
776 | for(i = 0; i < bytes/4; i++) | |
777 | newcsum += superc[i]; | |
778 | csum = (newcsum& 0xffffffff) + (newcsum>>32); | |
779 | #ifdef __alpha__ | |
780 | /* The in-kernel checksum calculation is always 16bit on | |
781 | * the alpha, though it is 32 bit on i386... | |
782 | * I wonder what it is elsewhere... (it uses an API in | |
783 | * a way that it shouldn't). | |
784 | */ | |
785 | csum = (csum & 0xffff) + (csum >> 16); | |
786 | csum = (csum & 0xffff) + (csum >> 16); | |
787 | #endif | |
788 | return csum; | |
789 | } | |
790 | ||
791 | char *human_size(long long bytes) | |
792 | { | |
793 | static char buf[47]; | |
794 | ||
795 | /* We convert bytes to either centi-M{ega,ibi}bytes or | |
796 | * centi-G{igi,ibi}bytes, with appropriate rounding, | |
797 | * and then print 1/100th of those as a decimal. | |
798 | * We allow upto 2048Megabytes before converting to | |
799 | * gigabytes, as that shows more precision and isn't | |
800 | * too large a number. | |
801 | * Terabytes are not yet handled. | |
802 | */ | |
803 | ||
804 | if (bytes < 5000*1024) | |
805 | buf[0] = 0; | |
806 | else if (bytes < 2*1024LL*1024LL*1024LL) { | |
807 | long cMiB = (bytes * 200LL / (1LL<<20) + 1) / 2; | |
808 | long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2; | |
809 | snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)", | |
810 | cMiB/100, cMiB % 100, cMB/100, cMB % 100); | |
811 | } else { | |
812 | long cGiB = (bytes * 200LL / (1LL<<30) +1) / 2; | |
813 | long cGB = (bytes / (1000000000LL/200LL ) +1) /2; | |
814 | snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)", | |
815 | cGiB/100, cGiB % 100, cGB/100, cGB % 100); | |
816 | } | |
817 | return buf; | |
818 | } | |
819 | ||
820 | char *human_size_brief(long long bytes, int prefix) | |
821 | { | |
822 | static char buf[30]; | |
823 | ||
824 | /* We convert bytes to either centi-M{ega,ibi}bytes or | |
825 | * centi-G{igi,ibi}bytes, with appropriate rounding, | |
826 | * and then print 1/100th of those as a decimal. | |
827 | * We allow upto 2048Megabytes before converting to | |
828 | * gigabytes, as that shows more precision and isn't | |
829 | * too large a number. | |
830 | * Terabytes are not yet handled. | |
831 | * | |
832 | * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc. | |
833 | * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc. | |
834 | */ | |
835 | ||
836 | if (bytes < 5000*1024) | |
837 | buf[0] = 0; | |
838 | else if (prefix == IEC) { | |
839 | if (bytes < 2*1024LL*1024LL*1024LL) { | |
840 | long cMiB = (bytes * 200LL / (1LL<<20) +1) /2; | |
841 | snprintf(buf, sizeof(buf), "%ld.%02ldMiB", | |
842 | cMiB/100, cMiB % 100); | |
843 | } else { | |
844 | long cGiB = (bytes * 200LL / (1LL<<30) +1) /2; | |
845 | snprintf(buf, sizeof(buf), "%ld.%02ldGiB", | |
846 | cGiB/100, cGiB % 100); | |
847 | } | |
848 | } | |
849 | else if (prefix == JEDEC) { | |
850 | if (bytes < 2*1024LL*1024LL*1024LL) { | |
851 | long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2; | |
852 | snprintf(buf, sizeof(buf), "%ld.%02ldMB", | |
853 | cMB/100, cMB % 100); | |
854 | } else { | |
855 | long cGB = (bytes / (1000000000LL/200LL ) +1) /2; | |
856 | snprintf(buf, sizeof(buf), "%ld.%02ldGB", | |
857 | cGB/100, cGB % 100); | |
858 | } | |
859 | } | |
860 | else | |
861 | buf[0] = 0; | |
862 | ||
863 | return buf; | |
864 | } | |
865 | ||
866 | void print_r10_layout(int layout) | |
867 | { | |
868 | int near = layout & 255; | |
869 | int far = (layout >> 8) & 255; | |
870 | int offset = (layout&0x10000); | |
871 | char *sep = ""; | |
872 | ||
873 | if (near != 1) { | |
874 | printf("%s near=%d", sep, near); | |
875 | sep = ","; | |
876 | } | |
877 | if (far != 1) | |
878 | printf("%s %s=%d", sep, offset?"offset":"far", far); | |
879 | if (near*far == 1) | |
880 | printf("NO REDUNDANCY"); | |
881 | } | |
882 | ||
883 | unsigned long long calc_array_size(int level, int raid_disks, int layout, | |
884 | int chunksize, unsigned long long devsize) | |
885 | { | |
886 | if (level == 1) | |
887 | return devsize; | |
888 | devsize &= ~(unsigned long long)((chunksize>>9)-1); | |
889 | return get_data_disks(level, layout, raid_disks) * devsize; | |
890 | } | |
891 | ||
892 | int get_data_disks(int level, int layout, int raid_disks) | |
893 | { | |
894 | int data_disks = 0; | |
895 | switch (level) { | |
896 | case 0: data_disks = raid_disks; | |
897 | break; | |
898 | case 1: data_disks = 1; | |
899 | break; | |
900 | case 4: | |
901 | case 5: data_disks = raid_disks - 1; | |
902 | break; | |
903 | case 6: data_disks = raid_disks - 2; | |
904 | break; | |
905 | case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255); | |
906 | break; | |
907 | } | |
908 | ||
909 | return data_disks; | |
910 | } | |
911 | ||
912 | dev_t devnm2devid(char *devnm) | |
913 | { | |
914 | /* First look in /sys/block/$DEVNM/dev for %d:%d | |
915 | * If that fails, try parsing out a number | |
916 | */ | |
917 | char path[100]; | |
918 | char *ep; | |
919 | int fd; | |
920 | int mjr,mnr; | |
921 | ||
922 | sprintf(path, "/sys/block/%s/dev", devnm); | |
923 | fd = open(path, O_RDONLY); | |
924 | if (fd >= 0) { | |
925 | char buf[20]; | |
926 | int n = read(fd, buf, sizeof(buf)); | |
927 | close(fd); | |
928 | if (n > 0) | |
929 | buf[n] = 0; | |
930 | if (n > 0 && sscanf(buf, "%d:%d\n", &mjr, &mnr) == 2) | |
931 | return makedev(mjr, mnr); | |
932 | } | |
933 | if (strncmp(devnm, "md_d", 4) == 0 && | |
934 | isdigit(devnm[4]) && | |
935 | (mnr = strtoul(devnm+4, &ep, 10)) >= 0 && | |
936 | ep > devnm && *ep == 0) | |
937 | return makedev(get_mdp_major(), mnr << MdpMinorShift); | |
938 | ||
939 | if (strncmp(devnm, "md", 2) == 0 && | |
940 | isdigit(devnm[2]) && | |
941 | (mnr = strtoul(devnm+2, &ep, 10)) >= 0 && | |
942 | ep > devnm && *ep == 0) | |
943 | return makedev(MD_MAJOR, mnr); | |
944 | ||
945 | return 0; | |
946 | } | |
947 | ||
948 | char *get_md_name(char *devnm) | |
949 | { | |
950 | /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */ | |
951 | /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */ | |
952 | ||
953 | static char devname[50]; | |
954 | struct stat stb; | |
955 | dev_t rdev = devnm2devid(devnm); | |
956 | char *dn; | |
957 | ||
958 | if (rdev == 0) | |
959 | return 0; | |
960 | if (strncmp(devnm, "md_", 3) == 0) { | |
961 | snprintf(devname, sizeof(devname), "/dev/md/%s", | |
962 | devnm + 3); | |
963 | if (stat(devname, &stb) == 0 | |
964 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
965 | && (stb.st_rdev == rdev)) | |
966 | return devname; | |
967 | } | |
968 | snprintf(devname, sizeof(devname), "/dev/%s", devnm); | |
969 | if (stat(devname, &stb) == 0 | |
970 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
971 | && (stb.st_rdev == rdev)) | |
972 | return devname; | |
973 | ||
974 | snprintf(devname, sizeof(devname), "/dev/md/%s", devnm+2); | |
975 | if (stat(devname, &stb) == 0 | |
976 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
977 | && (stb.st_rdev == rdev)) | |
978 | return devname; | |
979 | ||
980 | dn = map_dev(major(rdev), minor(rdev), 0); | |
981 | if (dn) | |
982 | return dn; | |
983 | snprintf(devname, sizeof(devname), "/dev/.tmp.%s", devnm); | |
984 | if (mknod(devname, S_IFBLK | 0600, rdev) == -1) | |
985 | if (errno != EEXIST) | |
986 | return NULL; | |
987 | ||
988 | if (stat(devname, &stb) == 0 | |
989 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
990 | && (stb.st_rdev == rdev)) | |
991 | return devname; | |
992 | unlink(devname); | |
993 | return NULL; | |
994 | } | |
995 | ||
996 | void put_md_name(char *name) | |
997 | { | |
998 | if (strncmp(name, "/dev/.tmp.md", 12) == 0) | |
999 | unlink(name); | |
1000 | } | |
1001 | ||
1002 | int get_maj_min(char *dev, int *major, int *minor) | |
1003 | { | |
1004 | char *e; | |
1005 | *major = strtoul(dev, &e, 0); | |
1006 | return (e > dev && *e == ':' && e[1] && | |
1007 | (*minor = strtoul(e+1, &e, 0)) >= 0 && | |
1008 | *e == 0); | |
1009 | } | |
1010 | ||
1011 | int dev_open(char *dev, int flags) | |
1012 | { | |
1013 | /* like 'open', but if 'dev' matches %d:%d, create a temp | |
1014 | * block device and open that | |
1015 | */ | |
1016 | int fd = -1; | |
1017 | char devname[32]; | |
1018 | int major; | |
1019 | int minor; | |
1020 | ||
1021 | if (!dev) | |
1022 | return -1; | |
1023 | flags |= O_DIRECT; | |
1024 | ||
1025 | if (get_maj_min(dev, &major, &minor)) { | |
1026 | snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d", | |
1027 | (int)getpid(), major, minor); | |
1028 | if (mknod(devname, S_IFBLK|0600, makedev(major, minor)) == 0) { | |
1029 | fd = open(devname, flags); | |
1030 | unlink(devname); | |
1031 | } | |
1032 | if (fd < 0) { | |
1033 | /* Try /tmp as /dev appear to be read-only */ | |
1034 | snprintf(devname, sizeof(devname), "/tmp/.tmp.md.%d:%d:%d", | |
1035 | (int)getpid(), major, minor); | |
1036 | if (mknod(devname, S_IFBLK|0600, makedev(major, minor)) == 0) { | |
1037 | fd = open(devname, flags); | |
1038 | unlink(devname); | |
1039 | } | |
1040 | } | |
1041 | } else | |
1042 | fd = open(dev, flags); | |
1043 | return fd; | |
1044 | } | |
1045 | ||
1046 | int open_dev_flags(char *devnm, int flags) | |
1047 | { | |
1048 | dev_t devid; | |
1049 | char buf[20]; | |
1050 | ||
1051 | devid = devnm2devid(devnm); | |
1052 | sprintf(buf, "%d:%d", major(devid), minor(devid)); | |
1053 | return dev_open(buf, flags); | |
1054 | } | |
1055 | ||
1056 | int open_dev(char *devnm) | |
1057 | { | |
1058 | return open_dev_flags(devnm, O_RDONLY); | |
1059 | } | |
1060 | ||
1061 | int open_dev_excl(char *devnm) | |
1062 | { | |
1063 | char buf[20]; | |
1064 | int i; | |
1065 | int flags = O_RDWR; | |
1066 | dev_t devid = devnm2devid(devnm); | |
1067 | long delay = 1000; | |
1068 | ||
1069 | sprintf(buf, "%d:%d", major(devid), minor(devid)); | |
1070 | for (i = 0; i < 25; i++) { | |
1071 | int fd = dev_open(buf, flags|O_EXCL); | |
1072 | if (fd >= 0) | |
1073 | return fd; | |
1074 | if (errno == EACCES && flags == O_RDWR) { | |
1075 | flags = O_RDONLY; | |
1076 | continue; | |
1077 | } | |
1078 | if (errno != EBUSY) | |
1079 | return fd; | |
1080 | usleep(delay); | |
1081 | if (delay < 200000) | |
1082 | delay *= 2; | |
1083 | } | |
1084 | return -1; | |
1085 | } | |
1086 | ||
1087 | int same_dev(char *one, char *two) | |
1088 | { | |
1089 | struct stat st1, st2; | |
1090 | if (stat(one, &st1) != 0) | |
1091 | return 0; | |
1092 | if (stat(two, &st2) != 0) | |
1093 | return 0; | |
1094 | if ((st1.st_mode & S_IFMT) != S_IFBLK) | |
1095 | return 0; | |
1096 | if ((st2.st_mode & S_IFMT) != S_IFBLK) | |
1097 | return 0; | |
1098 | return st1.st_rdev == st2.st_rdev; | |
1099 | } | |
1100 | ||
1101 | void wait_for(char *dev, int fd) | |
1102 | { | |
1103 | int i; | |
1104 | struct stat stb_want; | |
1105 | long delay = 1000; | |
1106 | ||
1107 | if (fstat(fd, &stb_want) != 0 || | |
1108 | (stb_want.st_mode & S_IFMT) != S_IFBLK) | |
1109 | return; | |
1110 | ||
1111 | for (i = 0; i < 25; i++) { | |
1112 | struct stat stb; | |
1113 | if (stat(dev, &stb) == 0 && | |
1114 | (stb.st_mode & S_IFMT) == S_IFBLK && | |
1115 | (stb.st_rdev == stb_want.st_rdev)) | |
1116 | return; | |
1117 | usleep(delay); | |
1118 | if (delay < 200000) | |
1119 | delay *= 2; | |
1120 | } | |
1121 | if (i == 25) | |
1122 | dprintf("timeout waiting for %s\n", dev); | |
1123 | } | |
1124 | ||
1125 | struct superswitch *superlist[] = | |
1126 | { | |
1127 | &super0, &super1, | |
1128 | &super_ddf, &super_imsm, | |
1129 | &mbr, &gpt, | |
1130 | NULL | |
1131 | }; | |
1132 | ||
1133 | struct supertype *super_by_fd(int fd, char **subarrayp) | |
1134 | { | |
1135 | mdu_array_info_t array; | |
1136 | int vers; | |
1137 | int minor; | |
1138 | struct supertype *st = NULL; | |
1139 | struct mdinfo *sra; | |
1140 | char *verstr; | |
1141 | char version[20]; | |
1142 | int i; | |
1143 | char *subarray = NULL; | |
1144 | char container[32] = ""; | |
1145 | ||
1146 | sra = sysfs_read(fd, NULL, GET_VERSION); | |
1147 | ||
1148 | if (sra) { | |
1149 | vers = sra->array.major_version; | |
1150 | minor = sra->array.minor_version; | |
1151 | verstr = sra->text_version; | |
1152 | } else { | |
1153 | if (md_get_array_info(fd, &array)) | |
1154 | array.major_version = array.minor_version = 0; | |
1155 | vers = array.major_version; | |
1156 | minor = array.minor_version; | |
1157 | verstr = ""; | |
1158 | } | |
1159 | ||
1160 | if (vers != -1) { | |
1161 | sprintf(version, "%d.%d", vers, minor); | |
1162 | verstr = version; | |
1163 | } | |
1164 | if (minor == -2 && is_subarray(verstr)) { | |
1165 | char *dev = verstr+1; | |
1166 | ||
1167 | subarray = strchr(dev, '/'); | |
1168 | if (subarray) { | |
1169 | *subarray++ = '\0'; | |
1170 | subarray = xstrdup(subarray); | |
1171 | } | |
1172 | strcpy(container, dev); | |
1173 | sysfs_free(sra); | |
1174 | sra = sysfs_read(-1, container, GET_VERSION); | |
1175 | if (sra && sra->text_version[0]) | |
1176 | verstr = sra->text_version; | |
1177 | else | |
1178 | verstr = "-no-metadata-"; | |
1179 | } | |
1180 | ||
1181 | for (i = 0; st == NULL && superlist[i]; i++) | |
1182 | st = superlist[i]->match_metadata_desc(verstr); | |
1183 | ||
1184 | sysfs_free(sra); | |
1185 | if (st) { | |
1186 | st->sb = NULL; | |
1187 | if (subarrayp) | |
1188 | *subarrayp = subarray; | |
1189 | strcpy(st->container_devnm, container); | |
1190 | strcpy(st->devnm, fd2devnm(fd)); | |
1191 | } else | |
1192 | free(subarray); | |
1193 | ||
1194 | return st; | |
1195 | } | |
1196 | ||
1197 | int dev_size_from_id(dev_t id, unsigned long long *size) | |
1198 | { | |
1199 | char buf[20]; | |
1200 | int fd; | |
1201 | ||
1202 | sprintf(buf, "%d:%d", major(id), minor(id)); | |
1203 | fd = dev_open(buf, O_RDONLY); | |
1204 | if (fd < 0) | |
1205 | return 0; | |
1206 | if (get_dev_size(fd, NULL, size)) { | |
1207 | close(fd); | |
1208 | return 1; | |
1209 | } | |
1210 | close(fd); | |
1211 | return 0; | |
1212 | } | |
1213 | ||
1214 | struct supertype *dup_super(struct supertype *orig) | |
1215 | { | |
1216 | struct supertype *st; | |
1217 | ||
1218 | if (!orig) | |
1219 | return orig; | |
1220 | st = xcalloc(1, sizeof(*st)); | |
1221 | st->ss = orig->ss; | |
1222 | st->max_devs = orig->max_devs; | |
1223 | st->minor_version = orig->minor_version; | |
1224 | st->ignore_hw_compat = orig->ignore_hw_compat; | |
1225 | st->data_offset = orig->data_offset; | |
1226 | st->sb = NULL; | |
1227 | st->info = NULL; | |
1228 | return st; | |
1229 | } | |
1230 | ||
1231 | struct supertype *guess_super_type(int fd, enum guess_types guess_type) | |
1232 | { | |
1233 | /* try each load_super to find the best match, | |
1234 | * and return the best superswitch | |
1235 | */ | |
1236 | struct superswitch *ss; | |
1237 | struct supertype *st; | |
1238 | unsigned int besttime = 0; | |
1239 | int bestsuper = -1; | |
1240 | int i; | |
1241 | ||
1242 | st = xcalloc(1, sizeof(*st)); | |
1243 | st->container_devnm[0] = 0; | |
1244 | ||
1245 | for (i = 0; superlist[i]; i++) { | |
1246 | int rv; | |
1247 | ss = superlist[i]; | |
1248 | if (guess_type == guess_array && ss->add_to_super == NULL) | |
1249 | continue; | |
1250 | if (guess_type == guess_partitions && ss->add_to_super != NULL) | |
1251 | continue; | |
1252 | memset(st, 0, sizeof(*st)); | |
1253 | st->ignore_hw_compat = 1; | |
1254 | rv = ss->load_super(st, fd, NULL); | |
1255 | if (rv == 0) { | |
1256 | struct mdinfo info; | |
1257 | st->ss->getinfo_super(st, &info, NULL); | |
1258 | if (bestsuper == -1 || | |
1259 | besttime < info.array.ctime) { | |
1260 | bestsuper = i; | |
1261 | besttime = info.array.ctime; | |
1262 | } | |
1263 | ss->free_super(st); | |
1264 | } | |
1265 | } | |
1266 | if (bestsuper != -1) { | |
1267 | int rv; | |
1268 | memset(st, 0, sizeof(*st)); | |
1269 | st->ignore_hw_compat = 1; | |
1270 | rv = superlist[bestsuper]->load_super(st, fd, NULL); | |
1271 | if (rv == 0) { | |
1272 | superlist[bestsuper]->free_super(st); | |
1273 | return st; | |
1274 | } | |
1275 | } | |
1276 | free(st); | |
1277 | return NULL; | |
1278 | } | |
1279 | ||
1280 | /* Return size of device in bytes */ | |
1281 | int get_dev_size(int fd, char *dname, unsigned long long *sizep) | |
1282 | { | |
1283 | unsigned long long ldsize; | |
1284 | struct stat st; | |
1285 | ||
1286 | if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode)) | |
1287 | ldsize = (unsigned long long)st.st_size; | |
1288 | else | |
1289 | #ifdef BLKGETSIZE64 | |
1290 | if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0) | |
1291 | #endif | |
1292 | { | |
1293 | unsigned long dsize; | |
1294 | if (ioctl(fd, BLKGETSIZE, &dsize) == 0) { | |
1295 | ldsize = dsize; | |
1296 | ldsize <<= 9; | |
1297 | } else { | |
1298 | if (dname) | |
1299 | pr_err("Cannot get size of %s: %s\n", | |
1300 | dname, strerror(errno)); | |
1301 | return 0; | |
1302 | } | |
1303 | } | |
1304 | *sizep = ldsize; | |
1305 | return 1; | |
1306 | } | |
1307 | ||
1308 | /* Return sector size of device in bytes */ | |
1309 | int get_dev_sector_size(int fd, char *dname, unsigned int *sectsizep) | |
1310 | { | |
1311 | unsigned int sectsize; | |
1312 | ||
1313 | if (ioctl(fd, BLKSSZGET, §size) != 0) { | |
1314 | if (dname) | |
1315 | pr_err("Cannot get sector size of %s: %s\n", | |
1316 | dname, strerror(errno)); | |
1317 | return 0; | |
1318 | } | |
1319 | ||
1320 | *sectsizep = sectsize; | |
1321 | return 1; | |
1322 | } | |
1323 | ||
1324 | /* Return true if this can only be a container, not a member device. | |
1325 | * i.e. is and md device and size is zero | |
1326 | */ | |
1327 | int must_be_container(int fd) | |
1328 | { | |
1329 | struct mdinfo *mdi; | |
1330 | unsigned long long size; | |
1331 | ||
1332 | mdi = sysfs_read(fd, NULL, GET_VERSION); | |
1333 | if (!mdi) | |
1334 | return 0; | |
1335 | sysfs_free(mdi); | |
1336 | ||
1337 | if (get_dev_size(fd, NULL, &size) == 0) | |
1338 | return 1; | |
1339 | if (size == 0) | |
1340 | return 1; | |
1341 | return 0; | |
1342 | } | |
1343 | ||
1344 | /* Sets endofpart parameter to the last block used by the last GPT partition on the device. | |
1345 | * Returns: 1 if successful | |
1346 | * -1 for unknown partition type | |
1347 | * 0 for other errors | |
1348 | */ | |
1349 | static int get_gpt_last_partition_end(int fd, unsigned long long *endofpart) | |
1350 | { | |
1351 | struct GPT gpt; | |
1352 | unsigned char empty_gpt_entry[16]= {0}; | |
1353 | struct GPT_part_entry *part; | |
1354 | char buf[512]; | |
1355 | unsigned long long curr_part_end; | |
1356 | unsigned all_partitions, entry_size; | |
1357 | unsigned part_nr; | |
1358 | unsigned int sector_size = 0; | |
1359 | ||
1360 | *endofpart = 0; | |
1361 | ||
1362 | BUILD_BUG_ON(sizeof(gpt) != 512); | |
1363 | /* skip protective MBR */ | |
1364 | if (!get_dev_sector_size(fd, NULL, §or_size)) | |
1365 | return 0; | |
1366 | lseek(fd, sector_size, SEEK_SET); | |
1367 | /* read GPT header */ | |
1368 | if (read(fd, &gpt, 512) != 512) | |
1369 | return 0; | |
1370 | ||
1371 | /* get the number of partition entries and the entry size */ | |
1372 | all_partitions = __le32_to_cpu(gpt.part_cnt); | |
1373 | entry_size = __le32_to_cpu(gpt.part_size); | |
1374 | ||
1375 | /* Check GPT signature*/ | |
1376 | if (gpt.magic != GPT_SIGNATURE_MAGIC) | |
1377 | return -1; | |
1378 | ||
1379 | /* sanity checks */ | |
1380 | if (all_partitions > 1024 || | |
1381 | entry_size > sizeof(buf)) | |
1382 | return -1; | |
1383 | ||
1384 | part = (struct GPT_part_entry *)buf; | |
1385 | ||
1386 | /* set offset to third block (GPT entries) */ | |
1387 | lseek(fd, sector_size*2, SEEK_SET); | |
1388 | for (part_nr = 0; part_nr < all_partitions; part_nr++) { | |
1389 | /* read partition entry */ | |
1390 | if (read(fd, buf, entry_size) != (ssize_t)entry_size) | |
1391 | return 0; | |
1392 | ||
1393 | /* is this valid partition? */ | |
1394 | if (memcmp(part->type_guid, empty_gpt_entry, 16) != 0) { | |
1395 | /* check the last lba for the current partition */ | |
1396 | curr_part_end = __le64_to_cpu(part->ending_lba); | |
1397 | if (curr_part_end > *endofpart) | |
1398 | *endofpart = curr_part_end; | |
1399 | } | |
1400 | ||
1401 | } | |
1402 | return 1; | |
1403 | } | |
1404 | ||
1405 | /* Sets endofpart parameter to the last block used by the last partition on the device. | |
1406 | * Returns: 1 if successful | |
1407 | * -1 for unknown partition type | |
1408 | * 0 for other errors | |
1409 | */ | |
1410 | static int get_last_partition_end(int fd, unsigned long long *endofpart) | |
1411 | { | |
1412 | struct MBR boot_sect; | |
1413 | unsigned long long curr_part_end; | |
1414 | unsigned part_nr; | |
1415 | unsigned int sector_size; | |
1416 | int retval = 0; | |
1417 | ||
1418 | *endofpart = 0; | |
1419 | ||
1420 | BUILD_BUG_ON(sizeof(boot_sect) != 512); | |
1421 | /* read MBR */ | |
1422 | lseek(fd, 0, 0); | |
1423 | if (read(fd, &boot_sect, 512) != 512) | |
1424 | goto abort; | |
1425 | ||
1426 | /* check MBP signature */ | |
1427 | if (boot_sect.magic == MBR_SIGNATURE_MAGIC) { | |
1428 | retval = 1; | |
1429 | /* found the correct signature */ | |
1430 | ||
1431 | for (part_nr = 0; part_nr < MBR_PARTITIONS; part_nr++) { | |
1432 | /* | |
1433 | * Have to make every access through boot_sect rather | |
1434 | * than using a pointer to the partition table (or an | |
1435 | * entry), since the entries are not properly aligned. | |
1436 | */ | |
1437 | ||
1438 | /* check for GPT type */ | |
1439 | if (boot_sect.parts[part_nr].part_type == | |
1440 | 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 = | |
1446 | __le32_to_cpu(boot_sect.parts[part_nr].first_sect_lba) + | |
1447 | __le32_to_cpu(boot_sect.parts[part_nr].blocks_num); | |
1448 | if (curr_part_end > *endofpart) | |
1449 | *endofpart = curr_part_end; | |
1450 | } | |
1451 | } else { | |
1452 | /* Unknown partition table */ | |
1453 | retval = -1; | |
1454 | } | |
1455 | /* calculate number of 512-byte blocks */ | |
1456 | if (get_dev_sector_size(fd, NULL, §or_size)) | |
1457 | *endofpart *= (sector_size / 512); | |
1458 | abort: | |
1459 | return retval; | |
1460 | } | |
1461 | ||
1462 | int check_partitions(int fd, char *dname, unsigned long long freesize, | |
1463 | unsigned long long size) | |
1464 | { | |
1465 | /* | |
1466 | * Check where the last partition ends | |
1467 | */ | |
1468 | unsigned long long endofpart; | |
1469 | ||
1470 | if (get_last_partition_end(fd, &endofpart) > 0) { | |
1471 | /* There appears to be a partition table here */ | |
1472 | if (freesize == 0) { | |
1473 | /* partitions will not be visible in new device */ | |
1474 | pr_err("partition table exists on %s but will be lost or\n" | |
1475 | " meaningless after creating array\n", | |
1476 | dname); | |
1477 | return 1; | |
1478 | } else if (endofpart > freesize) { | |
1479 | /* last partition overlaps metadata */ | |
1480 | pr_err("metadata will over-write last partition on %s.\n", | |
1481 | dname); | |
1482 | return 1; | |
1483 | } else if (size && endofpart > size) { | |
1484 | /* partitions will be truncated in new device */ | |
1485 | pr_err("array size is too small to cover all partitions on %s.\n", | |
1486 | dname); | |
1487 | return 1; | |
1488 | } | |
1489 | } | |
1490 | return 0; | |
1491 | } | |
1492 | ||
1493 | int open_container(int fd) | |
1494 | { | |
1495 | /* 'fd' is a block device. Find out if it is in use | |
1496 | * by a container, and return an open fd on that container. | |
1497 | */ | |
1498 | char path[256]; | |
1499 | char *e; | |
1500 | DIR *dir; | |
1501 | struct dirent *de; | |
1502 | int dfd, n; | |
1503 | char buf[200]; | |
1504 | int major, minor; | |
1505 | struct stat st; | |
1506 | ||
1507 | if (fstat(fd, &st) != 0) | |
1508 | return -1; | |
1509 | sprintf(path, "/sys/dev/block/%d:%d/holders", | |
1510 | (int)major(st.st_rdev), (int)minor(st.st_rdev)); | |
1511 | e = path + strlen(path); | |
1512 | ||
1513 | dir = opendir(path); | |
1514 | if (!dir) | |
1515 | return -1; | |
1516 | while ((de = readdir(dir))) { | |
1517 | if (de->d_ino == 0) | |
1518 | continue; | |
1519 | if (de->d_name[0] == '.') | |
1520 | continue; | |
1521 | /* Need to make sure it is a container and not a volume */ | |
1522 | sprintf(e, "/%s/md/metadata_version", de->d_name); | |
1523 | dfd = open(path, O_RDONLY); | |
1524 | if (dfd < 0) | |
1525 | continue; | |
1526 | n = read(dfd, buf, sizeof(buf)); | |
1527 | close(dfd); | |
1528 | if (n <= 0 || (unsigned)n >= sizeof(buf)) | |
1529 | continue; | |
1530 | buf[n] = 0; | |
1531 | if (strncmp(buf, "external", 8) != 0 || | |
1532 | n < 10 || | |
1533 | buf[9] == '/') | |
1534 | continue; | |
1535 | sprintf(e, "/%s/dev", de->d_name); | |
1536 | dfd = open(path, O_RDONLY); | |
1537 | if (dfd < 0) | |
1538 | continue; | |
1539 | n = read(dfd, buf, sizeof(buf)); | |
1540 | close(dfd); | |
1541 | if (n <= 0 || (unsigned)n >= sizeof(buf)) | |
1542 | continue; | |
1543 | buf[n] = 0; | |
1544 | if (sscanf(buf, "%d:%d", &major, &minor) != 2) | |
1545 | continue; | |
1546 | sprintf(buf, "%d:%d", major, minor); | |
1547 | dfd = dev_open(buf, O_RDONLY); | |
1548 | if (dfd >= 0) { | |
1549 | closedir(dir); | |
1550 | return dfd; | |
1551 | } | |
1552 | } | |
1553 | closedir(dir); | |
1554 | return -1; | |
1555 | } | |
1556 | ||
1557 | struct superswitch *version_to_superswitch(char *vers) | |
1558 | { | |
1559 | int i; | |
1560 | ||
1561 | for (i = 0; superlist[i]; i++) { | |
1562 | struct superswitch *ss = superlist[i]; | |
1563 | ||
1564 | if (strcmp(vers, ss->name) == 0) | |
1565 | return ss; | |
1566 | } | |
1567 | ||
1568 | return NULL; | |
1569 | } | |
1570 | ||
1571 | int metadata_container_matches(char *metadata, char *devnm) | |
1572 | { | |
1573 | /* Check if 'devnm' is the container named in 'metadata' | |
1574 | * which is | |
1575 | * /containername/componentname or | |
1576 | * -containername/componentname | |
1577 | */ | |
1578 | int l; | |
1579 | if (*metadata != '/' && *metadata != '-') | |
1580 | return 0; | |
1581 | l = strlen(devnm); | |
1582 | if (strncmp(metadata+1, devnm, l) != 0) | |
1583 | return 0; | |
1584 | if (metadata[l+1] != '/') | |
1585 | return 0; | |
1586 | return 1; | |
1587 | } | |
1588 | ||
1589 | int metadata_subdev_matches(char *metadata, char *devnm) | |
1590 | { | |
1591 | /* Check if 'devnm' is the subdev named in 'metadata' | |
1592 | * which is | |
1593 | * /containername/subdev or | |
1594 | * -containername/subdev | |
1595 | */ | |
1596 | char *sl; | |
1597 | if (*metadata != '/' && *metadata != '-') | |
1598 | return 0; | |
1599 | sl = strchr(metadata+1, '/'); | |
1600 | if (!sl) | |
1601 | return 0; | |
1602 | if (strcmp(sl+1, devnm) == 0) | |
1603 | return 1; | |
1604 | return 0; | |
1605 | } | |
1606 | ||
1607 | int is_container_member(struct mdstat_ent *mdstat, char *container) | |
1608 | { | |
1609 | if (mdstat->metadata_version == NULL || | |
1610 | strncmp(mdstat->metadata_version, "external:", 9) != 0 || | |
1611 | !metadata_container_matches(mdstat->metadata_version+9, container)) | |
1612 | return 0; | |
1613 | ||
1614 | return 1; | |
1615 | } | |
1616 | ||
1617 | int is_subarray_active(char *subarray, char *container) | |
1618 | { | |
1619 | struct mdstat_ent *mdstat = mdstat_read(0, 0); | |
1620 | struct mdstat_ent *ent; | |
1621 | ||
1622 | for (ent = mdstat; ent; ent = ent->next) | |
1623 | if (is_container_member(ent, container)) | |
1624 | if (strcmp(to_subarray(ent, container), subarray) == 0) | |
1625 | break; | |
1626 | ||
1627 | free_mdstat(mdstat); | |
1628 | ||
1629 | return ent != NULL; | |
1630 | } | |
1631 | ||
1632 | /* open_subarray - opens a subarray in a container | |
1633 | * @dev: container device name | |
1634 | * @st: empty supertype | |
1635 | * @quiet: block reporting errors flag | |
1636 | * | |
1637 | * On success returns an fd to a container and fills in *st | |
1638 | */ | |
1639 | int open_subarray(char *dev, char *subarray, struct supertype *st, int quiet) | |
1640 | { | |
1641 | struct mdinfo *mdi; | |
1642 | struct mdinfo *info; | |
1643 | int fd, err = 1; | |
1644 | char *_devnm; | |
1645 | ||
1646 | fd = open(dev, O_RDWR|O_EXCL); | |
1647 | if (fd < 0) { | |
1648 | if (!quiet) | |
1649 | pr_err("Couldn't open %s, aborting\n", | |
1650 | dev); | |
1651 | return -1; | |
1652 | } | |
1653 | ||
1654 | _devnm = fd2devnm(fd); | |
1655 | if (_devnm == NULL) { | |
1656 | if (!quiet) | |
1657 | pr_err("Failed to determine device number for %s\n", | |
1658 | dev); | |
1659 | goto close_fd; | |
1660 | } | |
1661 | strcpy(st->devnm, _devnm); | |
1662 | ||
1663 | mdi = sysfs_read(fd, st->devnm, GET_VERSION|GET_LEVEL); | |
1664 | if (!mdi) { | |
1665 | if (!quiet) | |
1666 | pr_err("Failed to read sysfs for %s\n", | |
1667 | dev); | |
1668 | goto close_fd; | |
1669 | } | |
1670 | ||
1671 | if (mdi->array.level != UnSet) { | |
1672 | if (!quiet) | |
1673 | pr_err("%s is not a container\n", dev); | |
1674 | goto free_sysfs; | |
1675 | } | |
1676 | ||
1677 | st->ss = version_to_superswitch(mdi->text_version); | |
1678 | if (!st->ss) { | |
1679 | if (!quiet) | |
1680 | pr_err("Operation not supported for %s metadata\n", | |
1681 | mdi->text_version); | |
1682 | goto free_sysfs; | |
1683 | } | |
1684 | ||
1685 | if (st->devnm[0] == 0) { | |
1686 | if (!quiet) | |
1687 | pr_err("Failed to allocate device name\n"); | |
1688 | goto free_sysfs; | |
1689 | } | |
1690 | ||
1691 | if (!st->ss->load_container) { | |
1692 | if (!quiet) | |
1693 | pr_err("%s is not a container\n", dev); | |
1694 | goto free_sysfs; | |
1695 | } | |
1696 | ||
1697 | if (st->ss->load_container(st, fd, NULL)) { | |
1698 | if (!quiet) | |
1699 | pr_err("Failed to load metadata for %s\n", | |
1700 | dev); | |
1701 | goto free_sysfs; | |
1702 | } | |
1703 | ||
1704 | info = st->ss->container_content(st, subarray); | |
1705 | if (!info) { | |
1706 | if (!quiet) | |
1707 | pr_err("Failed to find subarray-%s in %s\n", | |
1708 | subarray, dev); | |
1709 | goto free_super; | |
1710 | } | |
1711 | free(info); | |
1712 | ||
1713 | err = 0; | |
1714 | ||
1715 | free_super: | |
1716 | if (err) | |
1717 | st->ss->free_super(st); | |
1718 | free_sysfs: | |
1719 | sysfs_free(mdi); | |
1720 | close_fd: | |
1721 | if (err) | |
1722 | close(fd); | |
1723 | ||
1724 | if (err) | |
1725 | return -1; | |
1726 | else | |
1727 | return fd; | |
1728 | } | |
1729 | ||
1730 | int add_disk(int mdfd, struct supertype *st, | |
1731 | struct mdinfo *sra, struct mdinfo *info) | |
1732 | { | |
1733 | /* Add a device to an array, in one of 2 ways. */ | |
1734 | int rv; | |
1735 | ||
1736 | if (st->ss->external) { | |
1737 | if (info->disk.state & (1<<MD_DISK_SYNC)) | |
1738 | info->recovery_start = MaxSector; | |
1739 | else | |
1740 | info->recovery_start = 0; | |
1741 | rv = sysfs_add_disk(sra, info, 0); | |
1742 | if (! rv) { | |
1743 | struct mdinfo *sd2; | |
1744 | for (sd2 = sra->devs; sd2; sd2=sd2->next) | |
1745 | if (sd2 == info) | |
1746 | break; | |
1747 | if (sd2 == NULL) { | |
1748 | sd2 = xmalloc(sizeof(*sd2)); | |
1749 | *sd2 = *info; | |
1750 | sd2->next = sra->devs; | |
1751 | sra->devs = sd2; | |
1752 | } | |
1753 | } | |
1754 | } else | |
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 | ||
1764 | /* Remove the disk given by 'info' from the array */ | |
1765 | if (st->ss->external) | |
1766 | rv = sysfs_set_str(sra, info, "slot", "none"); | |
1767 | else | |
1768 | rv = ioctl(mdfd, HOT_REMOVE_DISK, makedev(info->disk.major, | |
1769 | info->disk.minor)); | |
1770 | return rv; | |
1771 | } | |
1772 | ||
1773 | int hot_remove_disk(int mdfd, unsigned long dev, int force) | |
1774 | { | |
1775 | int cnt = force ? 500 : 5; | |
1776 | int ret; | |
1777 | ||
1778 | /* HOT_REMOVE_DISK can fail with EBUSY if there are | |
1779 | * outstanding IO requests to the device. | |
1780 | * In this case, it can be helpful to wait a little while, | |
1781 | * up to 5 seconds if 'force' is set, or 50 msec if not. | |
1782 | */ | |
1783 | while ((ret = ioctl(mdfd, HOT_REMOVE_DISK, dev)) == -1 && | |
1784 | errno == EBUSY && | |
1785 | cnt-- > 0) | |
1786 | usleep(10000); | |
1787 | ||
1788 | return ret; | |
1789 | } | |
1790 | ||
1791 | int sys_hot_remove_disk(int statefd, int force) | |
1792 | { | |
1793 | int cnt = force ? 500 : 5; | |
1794 | int ret; | |
1795 | ||
1796 | while ((ret = write(statefd, "remove", 6)) == -1 && | |
1797 | errno == EBUSY && | |
1798 | cnt-- > 0) | |
1799 | usleep(10000); | |
1800 | return ret == 6 ? 0 : -1; | |
1801 | } | |
1802 | ||
1803 | int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info) | |
1804 | { | |
1805 | /* Initialise kernel's knowledge of array. | |
1806 | * This varies between externally managed arrays | |
1807 | * and older kernels | |
1808 | */ | |
1809 | mdu_array_info_t inf; | |
1810 | int rv; | |
1811 | ||
1812 | if (st->ss->external) | |
1813 | return sysfs_set_array(info, 9003); | |
1814 | ||
1815 | memset(&inf, 0, sizeof(inf)); | |
1816 | inf.major_version = info->array.major_version; | |
1817 | inf.minor_version = info->array.minor_version; | |
1818 | rv = md_set_array_info(mdfd, &inf); | |
1819 | ||
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 | int flush_metadata_updates(struct supertype *st) | |
1994 | { | |
1995 | int sfd; | |
1996 | if (!st->updates) { | |
1997 | st->update_tail = NULL; | |
1998 | return -1; | |
1999 | } | |
2000 | ||
2001 | sfd = connect_monitor(st->container_devnm); | |
2002 | if (sfd < 0) | |
2003 | return -1; | |
2004 | ||
2005 | while (st->updates) { | |
2006 | struct metadata_update *mu = st->updates; | |
2007 | st->updates = mu->next; | |
2008 | ||
2009 | send_message(sfd, mu, 0); | |
2010 | wait_reply(sfd, 0); | |
2011 | free(mu->buf); | |
2012 | free(mu); | |
2013 | } | |
2014 | ack(sfd, 0); | |
2015 | wait_reply(sfd, 0); | |
2016 | close(sfd); | |
2017 | st->update_tail = NULL; | |
2018 | return 0; | |
2019 | } | |
2020 | ||
2021 | void append_metadata_update(struct supertype *st, void *buf, int len) | |
2022 | { | |
2023 | ||
2024 | struct metadata_update *mu = xmalloc(sizeof(*mu)); | |
2025 | ||
2026 | mu->buf = buf; | |
2027 | mu->len = len; | |
2028 | mu->space = NULL; | |
2029 | mu->space_list = NULL; | |
2030 | mu->next = NULL; | |
2031 | *st->update_tail = mu; | |
2032 | st->update_tail = &mu->next; | |
2033 | } | |
2034 | ||
2035 | #ifdef __TINYC__ | |
2036 | /* tinyc doesn't optimize this check in ioctl.h out ... */ | |
2037 | unsigned int __invalid_size_argument_for_IOC = 0; | |
2038 | #endif | |
2039 | ||
2040 | int experimental(void) | |
2041 | { | |
2042 | if (check_env("MDADM_EXPERIMENTAL")) | |
2043 | return 1; | |
2044 | else { | |
2045 | pr_err("To use this feature MDADM_EXPERIMENTAL environment variable has to be defined.\n"); | |
2046 | return 0; | |
2047 | } | |
2048 | } | |
2049 | ||
2050 | /* Pick all spares matching given criteria from a container | |
2051 | * if min_size == 0 do not check size | |
2052 | * if domlist == NULL do not check domains | |
2053 | * if spare_group given add it to domains of each spare | |
2054 | * metadata allows to test domains using metadata of destination array */ | |
2055 | struct mdinfo *container_choose_spares(struct supertype *st, | |
2056 | unsigned long long min_size, | |
2057 | struct domainlist *domlist, | |
2058 | char *spare_group, | |
2059 | const char *metadata, int get_one) | |
2060 | { | |
2061 | struct mdinfo *d, **dp, *disks = NULL; | |
2062 | ||
2063 | /* get list of all disks in container */ | |
2064 | if (st->ss->getinfo_super_disks) | |
2065 | disks = st->ss->getinfo_super_disks(st); | |
2066 | ||
2067 | if (!disks) | |
2068 | return disks; | |
2069 | /* find spare devices on the list */ | |
2070 | dp = &disks->devs; | |
2071 | disks->array.spare_disks = 0; | |
2072 | while (*dp) { | |
2073 | int found = 0; | |
2074 | d = *dp; | |
2075 | if (d->disk.state == 0) { | |
2076 | /* check if size is acceptable */ | |
2077 | unsigned long long dev_size; | |
2078 | dev_t dev = makedev(d->disk.major,d->disk.minor); | |
2079 | ||
2080 | if (!min_size || | |
2081 | (dev_size_from_id(dev, &dev_size) && | |
2082 | dev_size >= min_size)) | |
2083 | found = 1; | |
2084 | /* check if domain matches */ | |
2085 | if (found && domlist) { | |
2086 | struct dev_policy *pol = devid_policy(dev); | |
2087 | if (spare_group) | |
2088 | pol_add(&pol, pol_domain, | |
2089 | spare_group, NULL); | |
2090 | if (domain_test(domlist, pol, metadata) != 1) | |
2091 | found = 0; | |
2092 | dev_policy_free(pol); | |
2093 | } | |
2094 | } | |
2095 | if (found) { | |
2096 | dp = &d->next; | |
2097 | disks->array.spare_disks++; | |
2098 | if (get_one) { | |
2099 | sysfs_free(*dp); | |
2100 | d->next = NULL; | |
2101 | } | |
2102 | } else { | |
2103 | *dp = d->next; | |
2104 | d->next = NULL; | |
2105 | sysfs_free(d); | |
2106 | } | |
2107 | } | |
2108 | return disks; | |
2109 | } | |
2110 | ||
2111 | /* Checks if paths point to the same device | |
2112 | * Returns 0 if they do. | |
2113 | * Returns 1 if they don't. | |
2114 | * Returns -1 if something went wrong, | |
2115 | * e.g. paths are empty or the files | |
2116 | * they point to don't exist */ | |
2117 | int compare_paths (char* path1, char* path2) | |
2118 | { | |
2119 | struct stat st1,st2; | |
2120 | ||
2121 | if (path1 == NULL || path2 == NULL) | |
2122 | return -1; | |
2123 | if (stat(path1,&st1) != 0) | |
2124 | return -1; | |
2125 | if (stat(path2,&st2) != 0) | |
2126 | return -1; | |
2127 | if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev)) | |
2128 | return 0; | |
2129 | return 1; | |
2130 | } | |
2131 | ||
2132 | /* Make sure we can open as many devices as needed */ | |
2133 | void enable_fds(int devices) | |
2134 | { | |
2135 | unsigned int fds = 20 + devices; | |
2136 | struct rlimit lim; | |
2137 | if (getrlimit(RLIMIT_NOFILE, &lim) != 0 | |
2138 | || lim.rlim_cur >= fds) | |
2139 | return; | |
2140 | if (lim.rlim_max < fds) | |
2141 | lim.rlim_max = fds; | |
2142 | lim.rlim_cur = fds; | |
2143 | setrlimit(RLIMIT_NOFILE, &lim); | |
2144 | } | |
2145 | ||
2146 | int in_initrd(void) | |
2147 | { | |
2148 | /* This is based on similar function in systemd. */ | |
2149 | struct statfs s; | |
2150 | /* statfs.f_type is signed long on s390x and MIPS, causing all | |
2151 | sorts of sign extension problems with RAMFS_MAGIC being | |
2152 | defined as 0x858458f6 */ | |
2153 | return statfs("/", &s) >= 0 && | |
2154 | ((unsigned long)s.f_type == TMPFS_MAGIC || | |
2155 | ((unsigned long)s.f_type & 0xFFFFFFFFUL) == | |
2156 | ((unsigned long)RAMFS_MAGIC & 0xFFFFFFFFUL)); | |
2157 | } | |
2158 | ||
2159 | void reopen_mddev(int mdfd) | |
2160 | { | |
2161 | /* Re-open without any O_EXCL, but keep | |
2162 | * the same fd | |
2163 | */ | |
2164 | char *devnm; | |
2165 | int fd; | |
2166 | devnm = fd2devnm(mdfd); | |
2167 | close(mdfd); | |
2168 | fd = open_dev(devnm); | |
2169 | if (fd >= 0 && fd != mdfd) | |
2170 | dup2(fd, mdfd); | |
2171 | } | |
2172 | ||
2173 | static struct cmap_hooks *cmap_hooks = NULL; | |
2174 | static int is_cmap_hooks_ready = 0; | |
2175 | ||
2176 | void set_cmap_hooks(void) | |
2177 | { | |
2178 | cmap_hooks = xmalloc(sizeof(struct cmap_hooks)); | |
2179 | cmap_hooks->cmap_handle = dlopen("libcmap.so.4", RTLD_NOW | RTLD_LOCAL); | |
2180 | if (!cmap_hooks->cmap_handle) | |
2181 | return; | |
2182 | ||
2183 | cmap_hooks->initialize = dlsym(cmap_hooks->cmap_handle, "cmap_initialize"); | |
2184 | cmap_hooks->get_string = dlsym(cmap_hooks->cmap_handle, "cmap_get_string"); | |
2185 | cmap_hooks->finalize = dlsym(cmap_hooks->cmap_handle, "cmap_finalize"); | |
2186 | ||
2187 | if (!cmap_hooks->initialize || !cmap_hooks->get_string || | |
2188 | !cmap_hooks->finalize) | |
2189 | dlclose(cmap_hooks->cmap_handle); | |
2190 | else | |
2191 | is_cmap_hooks_ready = 1; | |
2192 | } | |
2193 | ||
2194 | int get_cluster_name(char **cluster_name) | |
2195 | { | |
2196 | int rv = -1; | |
2197 | cmap_handle_t handle; | |
2198 | ||
2199 | if (!is_cmap_hooks_ready) | |
2200 | return rv; | |
2201 | ||
2202 | rv = cmap_hooks->initialize(&handle); | |
2203 | if (rv != CS_OK) | |
2204 | goto out; | |
2205 | ||
2206 | rv = cmap_hooks->get_string(handle, "totem.cluster_name", cluster_name); | |
2207 | if (rv != CS_OK) { | |
2208 | free(*cluster_name); | |
2209 | rv = -1; | |
2210 | goto name_err; | |
2211 | } | |
2212 | ||
2213 | rv = 0; | |
2214 | name_err: | |
2215 | cmap_hooks->finalize(handle); | |
2216 | out: | |
2217 | return rv; | |
2218 | } | |
2219 | ||
2220 | void set_dlm_hooks(void) | |
2221 | { | |
2222 | dlm_hooks = xmalloc(sizeof(struct dlm_hooks)); | |
2223 | dlm_hooks->dlm_handle = dlopen("libdlm_lt.so.3", RTLD_NOW | RTLD_LOCAL); | |
2224 | if (!dlm_hooks->dlm_handle) | |
2225 | return; | |
2226 | ||
2227 | dlm_hooks->create_lockspace = dlsym(dlm_hooks->dlm_handle, "dlm_create_lockspace"); | |
2228 | dlm_hooks->release_lockspace = dlsym(dlm_hooks->dlm_handle, "dlm_release_lockspace"); | |
2229 | dlm_hooks->ls_lock = dlsym(dlm_hooks->dlm_handle, "dlm_ls_lock"); | |
2230 | dlm_hooks->ls_unlock = dlsym(dlm_hooks->dlm_handle, "dlm_ls_unlock"); | |
2231 | dlm_hooks->ls_get_fd = dlsym(dlm_hooks->dlm_handle, "dlm_ls_get_fd"); | |
2232 | dlm_hooks->dispatch = dlsym(dlm_hooks->dlm_handle, "dlm_dispatch"); | |
2233 | ||
2234 | if (!dlm_hooks->create_lockspace || !dlm_hooks->ls_lock || | |
2235 | !dlm_hooks->ls_unlock || !dlm_hooks->release_lockspace || | |
2236 | !dlm_hooks->ls_get_fd || !dlm_hooks->dispatch) | |
2237 | dlclose(dlm_hooks->dlm_handle); | |
2238 | else | |
2239 | is_dlm_hooks_ready = 1; | |
2240 | } | |
2241 | ||
2242 | void set_hooks(void) | |
2243 | { | |
2244 | set_dlm_hooks(); | |
2245 | set_cmap_hooks(); | |
2246 | } |