]> git.ipfire.org Git - thirdparty/mdadm.git/blob - sysfs.c
imsm: set array size at Create/Assemble
[thirdparty/mdadm.git] / sysfs.c
1 /*
2 * sysfs - extract md related information from sysfs. Part of:
3 * mdadm - manage Linux "md" devices aka RAID arrays.
4 *
5 * Copyright (C) 2006 Neil Brown <neilb@suse.de>
6 *
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 *
22 * Author: Neil Brown
23 * Email: <neilb@suse.de>
24 */
25
26 #include "mdadm.h"
27 #include <dirent.h>
28 #include <ctype.h>
29
30 int load_sys(char *path, char *buf)
31 {
32 int fd = open(path, O_RDONLY);
33 int n;
34 if (fd < 0)
35 return -1;
36 n = read(fd, buf, 1024);
37 close(fd);
38 if (n <0 || n >= 1024)
39 return -1;
40 buf[n] = 0;
41 if (n && buf[n-1] == '\n')
42 buf[n-1] = 0;
43 return 0;
44 }
45
46 void sysfs_free(struct mdinfo *sra)
47 {
48 while (sra) {
49 struct mdinfo *sra2 = sra->next;
50 while (sra->devs) {
51 struct mdinfo *d = sra->devs;
52 sra->devs = d->next;
53 free(d);
54 }
55 free(sra);
56 sra = sra2;
57 }
58 }
59
60 int sysfs_open(int devnum, char *devname, char *attr)
61 {
62 char fname[50];
63 int fd;
64 char *mdname = devnum2devname(devnum);
65
66 if (!mdname)
67 return -1;
68
69 sprintf(fname, "/sys/block/%s/md/", mdname);
70 if (devname) {
71 strcat(fname, devname);
72 strcat(fname, "/");
73 }
74 strcat(fname, attr);
75 fd = open(fname, O_RDWR);
76 if (fd < 0 && errno == EACCES)
77 fd = open(fname, O_RDONLY);
78 free(mdname);
79 return fd;
80 }
81
82 void sysfs_init(struct mdinfo *mdi, int fd, int devnum)
83 {
84 mdi->sys_name[0] = 0;
85 if (fd >= 0) {
86 mdu_version_t vers;
87 if (ioctl(fd, RAID_VERSION, &vers) != 0)
88 return;
89 devnum = fd2devnum(fd);
90 }
91 if (devnum == NoMdDev)
92 return;
93 if (devnum >= 0)
94 sprintf(mdi->sys_name, "md%d", devnum);
95 else
96 sprintf(mdi->sys_name, "md_d%d",
97 -1-devnum);
98 }
99
100
101 struct mdinfo *sysfs_read(int fd, int devnum, unsigned long options)
102 {
103 /* Longest possible name in sysfs, mounted at /sys, is
104 * /sys/block/md_dXXX/md/dev-XXXXX/block/dev
105 * /sys/block/md_dXXX/md/metadata_version
106 * which is about 41 characters. 50 should do for now
107 */
108 char fname[50];
109 char buf[1024];
110 char *base;
111 char *dbase;
112 struct mdinfo *sra;
113 struct mdinfo *dev;
114 DIR *dir = NULL;
115 struct dirent *de;
116
117 sra = malloc(sizeof(*sra));
118 if (sra == NULL)
119 return sra;
120 memset(sra, 0, sizeof(*sra));
121 sysfs_init(sra, fd, devnum);
122 if (sra->sys_name[0] == 0) {
123 free(sra);
124 return NULL;
125 }
126
127 sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
128 base = fname + strlen(fname);
129
130 sra->devs = NULL;
131 if (options & GET_VERSION) {
132 strcpy(base, "metadata_version");
133 if (load_sys(fname, buf))
134 goto abort;
135 if (strncmp(buf, "none", 4) == 0) {
136 sra->array.major_version =
137 sra->array.minor_version = -1;
138 strcpy(sra->text_version, "");
139 } else if (strncmp(buf, "external:", 9) == 0) {
140 sra->array.major_version = -1;
141 sra->array.minor_version = -2;
142 strcpy(sra->text_version, buf+9);
143 } else {
144 sscanf(buf, "%d.%d",
145 &sra->array.major_version,
146 &sra->array.minor_version);
147 strcpy(sra->text_version, buf);
148 }
149 }
150 if (options & GET_LEVEL) {
151 strcpy(base, "level");
152 if (load_sys(fname, buf))
153 goto abort;
154 sra->array.level = map_name(pers, buf);
155 }
156 if (options & GET_LAYOUT) {
157 strcpy(base, "layout");
158 if (load_sys(fname, buf))
159 goto abort;
160 sra->array.layout = strtoul(buf, NULL, 0);
161 }
162 if (options & GET_DISKS) {
163 strcpy(base, "raid_disks");
164 if (load_sys(fname, buf))
165 goto abort;
166 sra->array.raid_disks = strtoul(buf, NULL, 0);
167 }
168 if (options & GET_DEGRADED) {
169 strcpy(base, "degraded");
170 if (load_sys(fname, buf))
171 goto abort;
172 sra->array.failed_disks = strtoul(buf, NULL, 0);
173 }
174 if (options & GET_COMPONENT) {
175 strcpy(base, "component_size");
176 if (load_sys(fname, buf))
177 goto abort;
178 sra->component_size = strtoull(buf, NULL, 0);
179 /* sysfs reports "K", but we want sectors */
180 sra->component_size *= 2;
181 }
182 if (options & GET_CHUNK) {
183 strcpy(base, "chunk_size");
184 if (load_sys(fname, buf))
185 goto abort;
186 sra->array.chunk_size = strtoul(buf, NULL, 0);
187 }
188 if (options & GET_CACHE) {
189 strcpy(base, "stripe_cache_size");
190 if (load_sys(fname, buf))
191 goto abort;
192 sra->cache_size = strtoul(buf, NULL, 0);
193 }
194 if (options & GET_MISMATCH) {
195 strcpy(base, "mismatch_cnt");
196 if (load_sys(fname, buf))
197 goto abort;
198 sra->mismatch_cnt = strtoul(buf, NULL, 0);
199 }
200 if (options & GET_SAFEMODE) {
201 int scale = 1;
202 int dot = 0;
203 int i;
204 unsigned long msec;
205 size_t len;
206
207 strcpy(base, "safe_mode_delay");
208 if (load_sys(fname, buf))
209 goto abort;
210
211 /* remove a period, and count digits after it */
212 len = strlen(buf);
213 for (i = 0; i < len; i++) {
214 if (dot) {
215 if (isdigit(buf[i])) {
216 buf[i-1] = buf[i];
217 scale *= 10;
218 }
219 buf[i] = 0;
220 } else if (buf[i] == '.') {
221 dot=1;
222 buf[i] = 0;
223 }
224 }
225 msec = strtoul(buf, NULL, 10);
226 msec = (msec * 1000) / scale;
227 sra->safe_mode_delay = msec;
228 }
229
230 if (! (options & GET_DEVS))
231 return sra;
232
233 /* Get all the devices as well */
234 *base = 0;
235 dir = opendir(fname);
236 if (!dir)
237 goto abort;
238 sra->array.spare_disks = 0;
239
240 while ((de = readdir(dir)) != NULL) {
241 char *ep;
242 if (de->d_ino == 0 ||
243 strncmp(de->d_name, "dev-", 4) != 0)
244 continue;
245 strcpy(base, de->d_name);
246 dbase = base + strlen(base);
247 *dbase++ = '/';
248
249 dev = malloc(sizeof(*dev));
250 if (!dev)
251 goto abort;
252
253 /* Always get slot, major, minor */
254 strcpy(dbase, "slot");
255 if (load_sys(fname, buf)) {
256 /* hmm... unable to read 'slot' maybe the device
257 * is going away?
258 */
259 strcpy(dbase, "block");
260 if (readlink(fname, buf, sizeof(buf)) < 0 &&
261 errno != ENAMETOOLONG) {
262 /* ...yup device is gone */
263 free(dev);
264 continue;
265 } else {
266 /* slot is unreadable but 'block' link
267 * still intact... something bad is happening
268 * so abort
269 */
270 free(dev);
271 goto abort;
272 }
273
274 }
275 strcpy(dev->sys_name, de->d_name);
276 dev->disk.raid_disk = strtoul(buf, &ep, 10);
277 if (*ep) dev->disk.raid_disk = -1;
278
279 strcpy(dbase, "block/dev");
280 if (load_sys(fname, buf)) {
281 free(dev);
282 if (options & SKIP_GONE_DEVS)
283 continue;
284 else
285 goto abort;
286 }
287 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
288
289 /* special case check for block devices that can go 'offline' */
290 if (options & SKIP_GONE_DEVS) {
291 strcpy(dbase, "block/device/state");
292 if (load_sys(fname, buf) == 0 &&
293 strncmp(buf, "offline", 7) == 0) {
294 free(dev);
295 continue;
296 }
297 }
298
299 /* finally add this disk to the array */
300 dev->next = sra->devs;
301 sra->devs = dev;
302
303 if (options & GET_OFFSET) {
304 strcpy(dbase, "offset");
305 if (load_sys(fname, buf))
306 goto abort;
307 dev->data_offset = strtoull(buf, NULL, 0);
308 }
309 if (options & GET_SIZE) {
310 strcpy(dbase, "size");
311 if (load_sys(fname, buf))
312 goto abort;
313 dev->component_size = strtoull(buf, NULL, 0) * 2;
314 }
315 if (options & GET_STATE) {
316 dev->disk.state = 0;
317 strcpy(dbase, "state");
318 if (load_sys(fname, buf))
319 goto abort;
320 if (strstr(buf, "in_sync"))
321 dev->disk.state |= (1<<MD_DISK_SYNC);
322 if (strstr(buf, "faulty"))
323 dev->disk.state |= (1<<MD_DISK_FAULTY);
324 if (dev->disk.state == 0)
325 sra->array.spare_disks++;
326 }
327 if (options & GET_ERROR) {
328 strcpy(buf, "errors");
329 if (load_sys(fname, buf))
330 goto abort;
331 dev->errors = strtoul(buf, NULL, 0);
332 }
333 }
334 closedir(dir);
335 return sra;
336
337 abort:
338 if (dir)
339 closedir(dir);
340 sysfs_free(sra);
341 return NULL;
342 }
343
344 int sysfs_attr_match(const char *attr, const char *str)
345 {
346 /* See if attr, read from a sysfs file, matches
347 * str. They must either be the same, or attr can
348 * have a trailing newline or comma
349 */
350 while (*attr && *str && *attr == *str) {
351 attr++;
352 str++;
353 }
354
355 if (*str || (*attr && *attr != ',' && *attr != '\n'))
356 return 0;
357 return 1;
358 }
359
360 int sysfs_match_word(const char *word, char **list)
361 {
362 int n;
363 for (n=0; list[n]; n++)
364 if (sysfs_attr_match(word, list[n]))
365 break;
366 return n;
367 }
368
369 unsigned long long get_component_size(int fd)
370 {
371 /* Find out the component size of the array.
372 * We cannot trust GET_ARRAY_INFO ioctl as it's
373 * size field is only 32bits.
374 * So look in /sys/block/mdXXX/md/component_size
375 *
376 * This returns in units of sectors.
377 */
378 struct stat stb;
379 char fname[50];
380 int n;
381 if (fstat(fd, &stb)) return 0;
382 if (major(stb.st_rdev) != get_mdp_major())
383 sprintf(fname, "/sys/block/md%d/md/component_size",
384 (int)minor(stb.st_rdev));
385 else
386 sprintf(fname, "/sys/block/md_d%d/md/component_size",
387 (int)minor(stb.st_rdev)>>MdpMinorShift);
388 fd = open(fname, O_RDONLY);
389 if (fd < 0)
390 return 0;
391 n = read(fd, fname, sizeof(fname));
392 close(fd);
393 if (n == sizeof(fname))
394 return 0;
395 fname[n] = 0;
396 return strtoull(fname, NULL, 10) * 2;
397 }
398
399 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
400 char *name, char *val)
401 {
402 char fname[50];
403 int n;
404 int fd;
405
406 sprintf(fname, "/sys/block/%s/md/%s/%s",
407 sra->sys_name, dev?dev->sys_name:"", name);
408 fd = open(fname, O_WRONLY);
409 if (fd < 0)
410 return -1;
411 n = write(fd, val, strlen(val));
412 close(fd);
413 if (n != strlen(val)) {
414 dprintf(Name ": failed to write '%s' to '%s' (%s)\n",
415 val, fname, strerror(errno));
416 return -1;
417 }
418 return 0;
419 }
420
421 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
422 char *name, unsigned long long val)
423 {
424 char valstr[50];
425 sprintf(valstr, "%llu", val);
426 return sysfs_set_str(sra, dev, name, valstr);
427 }
428
429 int sysfs_uevent(struct mdinfo *sra, char *event)
430 {
431 char fname[50];
432 int n;
433 int fd;
434
435 sprintf(fname, "/sys/block/%s/uevent",
436 sra->sys_name);
437 fd = open(fname, O_WRONLY);
438 if (fd < 0)
439 return -1;
440 n = write(fd, event, strlen(event));
441 close(fd);
442 return 0;
443 }
444
445 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
446 char *name, unsigned long long *val)
447 {
448 char fname[50];
449 char buf[50];
450 int n;
451 int fd;
452 char *ep;
453 sprintf(fname, "/sys/block/%s/md/%s/%s",
454 sra->sys_name, dev?dev->sys_name:"", name);
455 fd = open(fname, O_RDONLY);
456 if (fd < 0)
457 return -1;
458 n = read(fd, buf, sizeof(buf));
459 close(fd);
460 if (n <= 0)
461 return -1;
462 buf[n] = 0;
463 *val = strtoull(buf, &ep, 0);
464 if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
465 return -1;
466 return 0;
467 }
468
469 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
470 {
471 unsigned long sec;
472 unsigned long msec;
473 char delay[30];
474
475 sec = ms / 1000;
476 msec = ms % 1000;
477
478 sprintf(delay, "%ld.%03ld\n", sec, msec);
479 /* this '\n' ^ needed for kernels older than 2.6.28 */
480 return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
481 }
482
483 int sysfs_set_array(struct mdinfo *info, int vers)
484 {
485 int rv = 0;
486 char ver[100];
487
488 ver[0] = 0;
489 if (info->array.major_version == -1 &&
490 info->array.minor_version == -2) {
491 strcat(strcpy(ver, "external:"), info->text_version);
492
493 if ((vers % 100) < 2 ||
494 sysfs_set_str(info, NULL, "metadata_version",
495 ver) < 0) {
496 fprintf(stderr, Name ": This kernel does not "
497 "support external metadata.\n");
498 return 1;
499 }
500 }
501 if (info->array.level < 0)
502 return 0; /* FIXME */
503 rv |= sysfs_set_str(info, NULL, "level",
504 map_num(pers, info->array.level));
505 rv |= sysfs_set_num(info, NULL, "raid_disks", info->array.raid_disks);
506 rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
507 rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
508 rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
509 if (info->custom_array_size) {
510 int rc;
511
512 rc = sysfs_set_num(info, NULL, "array_size",
513 info->custom_array_size/2);
514 if (rc && errno == ENOENT) {
515 fprintf(stderr, Name ": This kernel does not "
516 "have the md/array_size attribute, "
517 "the array may be larger than expected\n");
518 rc = 0;
519 }
520 rv |= rc;
521 }
522
523 if (info->array.level > 0)
524 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
525 return rv;
526 }
527
528 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd)
529 {
530 char dv[100];
531 char nm[100];
532 char *dname;
533 int rv;
534
535 sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
536 rv = sysfs_set_str(sra, NULL, "new_dev", dv);
537 if (rv)
538 return rv;
539
540 memset(nm, 0, sizeof(nm));
541 sprintf(dv, "/sys/dev/block/%d:%d", sd->disk.major, sd->disk.minor);
542 rv = readlink(dv, nm, sizeof(nm));
543 if (rv <= 0)
544 return -1;
545 nm[rv] = '\0';
546 dname = strrchr(nm, '/');
547 if (dname) dname++;
548 strcpy(sd->sys_name, "dev-");
549 strcpy(sd->sys_name+4, dname);
550
551 rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
552 rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
553 if (sra->array.level != LEVEL_CONTAINER) {
554 rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
555 // rv |= sysfs_set_str(sra, sd, "state", "in_sync");
556 }
557 return rv;
558 }
559
560 #if 0
561 int sysfs_disk_to_sg(int fd)
562 {
563 /* from an open block device, try find and open its corresponding
564 * scsi_generic interface
565 */
566 struct stat st;
567 char path[256];
568 char sg_path[256];
569 char sg_major_minor[8];
570 char *c;
571 DIR *dir;
572 struct dirent *de;
573 int major, minor, rv;
574
575 if (fstat(fd, &st))
576 return -1;
577
578 snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
579 major(st.st_rdev), minor(st.st_rdev));
580
581 dir = opendir(path);
582 if (!dir)
583 return -1;
584
585 de = readdir(dir);
586 while (de) {
587 if (strncmp("scsi_generic:", de->d_name,
588 strlen("scsi_generic:")) == 0)
589 break;
590 de = readdir(dir);
591 }
592 closedir(dir);
593
594 if (!de)
595 return -1;
596
597 snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
598 fd = open(sg_path, O_RDONLY);
599 if (fd < 0)
600 return fd;
601
602 rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
603 close(fd);
604 if (rv < 0)
605 return -1;
606 else
607 sg_major_minor[rv - 1] = '\0';
608
609 c = strchr(sg_major_minor, ':');
610 *c = '\0';
611 c++;
612 major = strtol(sg_major_minor, NULL, 10);
613 minor = strtol(c, NULL, 10);
614 snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
615 (int) getpid(), major, minor);
616 if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
617 fd = open(path, O_RDONLY);
618 unlink(path);
619 return fd;
620 }
621
622 return -1;
623 }
624 #endif
625
626 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
627 {
628 /* from an open block device, try to retrieve it scsi_id */
629 struct stat st;
630 char path[256];
631 char *c1, *c2;
632 DIR *dir;
633 struct dirent *de;
634
635 if (fstat(fd, &st))
636 return 1;
637
638 snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
639 major(st.st_rdev), minor(st.st_rdev));
640
641 dir = opendir(path);
642 if (!dir)
643 return 1;
644
645 de = readdir(dir);
646 while (de) {
647 if (strncmp("scsi_disk:", de->d_name,
648 strlen("scsi_disk:")) == 0)
649 break;
650 de = readdir(dir);
651 }
652 closedir(dir);
653
654 if (!de)
655 return 1;
656
657 c1 = strchr(de->d_name, ':');
658 c1++;
659 c2 = strchr(c1, ':');
660 *c2 = '\0';
661 *id = strtol(c1, NULL, 10) << 24; /* host */
662 c1 = c2 + 1;
663 c2 = strchr(c1, ':');
664 *c2 = '\0';
665 *id |= strtol(c1, NULL, 10) << 16; /* channel */
666 c1 = c2 + 1;
667 c2 = strchr(c1, ':');
668 *c2 = '\0';
669 *id |= strtol(c1, NULL, 10) << 8; /* lun */
670 c1 = c2 + 1;
671 *id |= strtol(c1, NULL, 10); /* id */
672
673 return 0;
674 }
675
676
677 int sysfs_unique_holder(int devnum, long rdev)
678 {
679 /* Check that devnum is a holder of rdev,
680 * and is the only holder.
681 * we should be locked against races by
682 * an O_EXCL on devnum
683 */
684 DIR *dir;
685 struct dirent *de;
686 char dirname[100];
687 char l;
688 int found = 0;
689 sprintf(dirname, "/sys/dev/block/%d:%d/holders",
690 major(rdev), minor(rdev));
691 dir = opendir(dirname);
692 errno = ENOENT;
693 if (!dir)
694 return 0;
695 l = strlen(dirname);
696 while ((de = readdir(dir)) != NULL) {
697 char buf[10];
698 int n;
699 int mj, mn;
700 char c;
701 int fd;
702
703 if (de->d_ino == 0)
704 continue;
705 if (de->d_name[0] == '.')
706 continue;
707 strcpy(dirname+l, "/");
708 strcat(dirname+l, de->d_name);
709 strcat(dirname+l, "/dev");
710 fd = open(dirname, O_RDONLY);
711 if (fd < 0) {
712 errno = ENOENT;
713 break;
714 }
715 n = read(fd, buf, sizeof(buf)-1);
716 close(fd);
717 buf[n] = 0;
718 if (sscanf(buf, "%d:%d%c", &mj, &mn, &c) != 3 ||
719 c != '\n') {
720 errno = ENOENT;
721 break;
722 }
723 if (mj != MD_MAJOR)
724 mn = -1-(mn>>6);
725
726 if (devnum != mn) {
727 errno = EEXIST;
728 break;
729 }
730 found = 1;
731 }
732 closedir(dir);
733 if (de)
734 return 0;
735 else
736 return found;
737 }