mdmon: fix wrong array state when disk fails during mdmon startup
[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) 2006-2009 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 #define MAX_SYSFS_PATH_LEN 120
31
32 int load_sys(char *path, char *buf, int len)
33 {
34 int fd = open(path, O_RDONLY);
35 int n;
36 if (fd < 0)
37 return -1;
38 n = read(fd, buf, len);
39 close(fd);
40 if (n <0 || n >= len)
41 return -1;
42 buf[n] = 0;
43 if (n && buf[n-1] == '\n')
44 buf[n-1] = 0;
45 return 0;
46 }
47
48 void sysfs_free(struct mdinfo *sra)
49 {
50 while (sra) {
51 struct mdinfo *sra2 = sra->next;
52 while (sra->devs) {
53 struct mdinfo *d = sra->devs;
54 sra->devs = d->next;
55 free(d->bb.entries);
56 free(d);
57 }
58 free(sra->bb.entries);
59 free(sra);
60 sra = sra2;
61 }
62 }
63
64 int sysfs_open(char *devnm, char *devname, char *attr)
65 {
66 char fname[MAX_SYSFS_PATH_LEN];
67 int fd;
68
69 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/", devnm);
70 if (devname) {
71 strncat(fname, devname, MAX_SYSFS_PATH_LEN - strlen(fname));
72 strncat(fname, "/", MAX_SYSFS_PATH_LEN - strlen(fname));
73 }
74 strncat(fname, attr, MAX_SYSFS_PATH_LEN - strlen(fname));
75 fd = open(fname, O_RDWR);
76 if (fd < 0 && errno == EACCES)
77 fd = open(fname, O_RDONLY);
78 return fd;
79 }
80
81 void sysfs_init_dev(struct mdinfo *mdi, dev_t devid)
82 {
83 snprintf(mdi->sys_name,
84 sizeof(mdi->sys_name), "dev-%s", devid2kname(devid));
85 }
86
87 int sysfs_init(struct mdinfo *mdi, int fd, char *devnm)
88 {
89 struct stat stb;
90 char fname[MAX_SYSFS_PATH_LEN];
91 int retval = -ENODEV;
92
93 mdi->sys_name[0] = 0;
94 if (fd >= 0)
95 devnm = fd2devnm(fd);
96
97 if (devnm == NULL)
98 goto out;
99
100 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md", devnm);
101
102 if (stat(fname, &stb))
103 goto out;
104 if (!S_ISDIR(stb.st_mode))
105 goto out;
106 strcpy(mdi->sys_name, devnm);
107
108 retval = 0;
109 out:
110 return retval;
111 }
112
113 struct mdinfo *sysfs_read(int fd, char *devnm, unsigned long options)
114 {
115 char fname[PATH_MAX];
116 char buf[PATH_MAX];
117 char *base;
118 char *dbase;
119 struct mdinfo *sra;
120 struct mdinfo *dev, **devp;
121 DIR *dir = NULL;
122 struct dirent *de;
123
124 sra = xcalloc(1, sizeof(*sra));
125 if (sysfs_init(sra, fd, devnm)) {
126 free(sra);
127 return NULL;
128 }
129
130 sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
131 base = fname + strlen(fname);
132
133 sra->devs = NULL;
134 if (options & GET_VERSION) {
135 strcpy(base, "metadata_version");
136 if (load_sys(fname, buf, sizeof(buf)))
137 goto abort;
138 if (strncmp(buf, "none", 4) == 0) {
139 sra->array.major_version =
140 sra->array.minor_version = -1;
141 strcpy(sra->text_version, "");
142 } else if (strncmp(buf, "external:", 9) == 0) {
143 sra->array.major_version = -1;
144 sra->array.minor_version = -2;
145 strcpy(sra->text_version, buf+9);
146 } else {
147 sscanf(buf, "%d.%d",
148 &sra->array.major_version,
149 &sra->array.minor_version);
150 strcpy(sra->text_version, buf);
151 }
152 }
153 if (options & GET_LEVEL) {
154 strcpy(base, "level");
155 if (load_sys(fname, buf, sizeof(buf)))
156 goto abort;
157 sra->array.level = map_name(pers, buf);
158 }
159 if (options & GET_LAYOUT) {
160 strcpy(base, "layout");
161 if (load_sys(fname, buf, sizeof(buf)))
162 goto abort;
163 sra->array.layout = strtoul(buf, NULL, 0);
164 }
165 if (options & (GET_DISKS|GET_STATE)) {
166 strcpy(base, "raid_disks");
167 if (load_sys(fname, buf, sizeof(buf)))
168 goto abort;
169 sra->array.raid_disks = strtoul(buf, NULL, 0);
170 }
171 if (options & GET_COMPONENT) {
172 strcpy(base, "component_size");
173 if (load_sys(fname, buf, sizeof(buf)))
174 goto abort;
175 sra->component_size = strtoull(buf, NULL, 0);
176 /* sysfs reports "K", but we want sectors */
177 sra->component_size *= 2;
178 }
179 if (options & GET_CHUNK) {
180 strcpy(base, "chunk_size");
181 if (load_sys(fname, buf, sizeof(buf)))
182 goto abort;
183 sra->array.chunk_size = strtoul(buf, NULL, 0);
184 }
185 if (options & GET_CACHE) {
186 strcpy(base, "stripe_cache_size");
187 if (load_sys(fname, buf, sizeof(buf)))
188 /* Probably level doesn't support it */
189 sra->cache_size = 0;
190 else
191 sra->cache_size = strtoul(buf, NULL, 0);
192 }
193 if (options & GET_MISMATCH) {
194 strcpy(base, "mismatch_cnt");
195 if (load_sys(fname, buf, sizeof(buf)))
196 goto abort;
197 sra->mismatch_cnt = strtoul(buf, NULL, 0);
198 }
199 if (options & GET_SAFEMODE) {
200 int scale = 1;
201 int dot = 0;
202 unsigned i;
203 unsigned long msec;
204 size_t len;
205
206 strcpy(base, "safe_mode_delay");
207 if (load_sys(fname, buf, sizeof(buf)))
208 goto abort;
209
210 /* remove a period, and count digits after it */
211 len = strlen(buf);
212 for (i = 0; i < len; i++) {
213 if (dot) {
214 if (isdigit(buf[i])) {
215 buf[i-1] = buf[i];
216 scale *= 10;
217 }
218 buf[i] = 0;
219 } else if (buf[i] == '.') {
220 dot=1;
221 buf[i] = 0;
222 }
223 }
224 msec = strtoul(buf, NULL, 10);
225 msec = (msec * 1000) / scale;
226 sra->safe_mode_delay = msec;
227 }
228 if (options & GET_BITMAP_LOCATION) {
229 strcpy(base, "bitmap/location");
230 if (load_sys(fname, buf, sizeof(buf)))
231 goto abort;
232 if (strncmp(buf, "file", 4) == 0)
233 sra->bitmap_offset = 1;
234 else if (strncmp(buf, "none", 4) == 0)
235 sra->bitmap_offset = 0;
236 else if (buf[0] == '+')
237 sra->bitmap_offset = strtol(buf+1, NULL, 10);
238 else
239 goto abort;
240 }
241
242 if (options & GET_ARRAY_STATE) {
243 strcpy(base, "array_state");
244 if (load_sys(fname, buf, sizeof(buf)))
245 goto abort;
246 sra->array_state = map_name(sysfs_array_states, buf);
247 }
248
249 if (options & GET_CONSISTENCY_POLICY) {
250 strcpy(base, "consistency_policy");
251 if (load_sys(fname, buf, sizeof(buf)))
252 sra->consistency_policy = CONSISTENCY_POLICY_UNKNOWN;
253 else
254 sra->consistency_policy = map_name(consistency_policies,
255 buf);
256 }
257
258 if (! (options & GET_DEVS))
259 return sra;
260
261 /* Get all the devices as well */
262 *base = 0;
263 dir = opendir(fname);
264 if (!dir)
265 goto abort;
266 sra->array.spare_disks = 0;
267 sra->array.active_disks = 0;
268 sra->array.failed_disks = 0;
269 sra->array.working_disks = 0;
270
271 devp = &sra->devs;
272 sra->devs = NULL;
273 while ((de = readdir(dir)) != NULL) {
274 char *ep;
275 if (de->d_ino == 0 ||
276 strncmp(de->d_name, "dev-", 4) != 0)
277 continue;
278 strcpy(base, de->d_name);
279 dbase = base + strlen(base);
280 *dbase++ = '/';
281
282 dev = xcalloc(1, sizeof(*dev));
283
284 /* Always get slot, major, minor */
285 strcpy(dbase, "slot");
286 if (load_sys(fname, buf, sizeof(buf))) {
287 /* hmm... unable to read 'slot' maybe the device
288 * is going away?
289 */
290 strcpy(dbase, "block");
291 if (readlink(fname, buf, sizeof(buf)) < 0 &&
292 errno != ENAMETOOLONG) {
293 /* ...yup device is gone */
294 free(dev);
295 continue;
296 } else {
297 /* slot is unreadable but 'block' link
298 * still intact... something bad is happening
299 * so abort
300 */
301 free(dev);
302 goto abort;
303 }
304
305 }
306 strcpy(dev->sys_name, de->d_name);
307 dev->disk.raid_disk = strtoul(buf, &ep, 10);
308 if (*ep) dev->disk.raid_disk = -1;
309
310 sra->array.nr_disks++;
311 strcpy(dbase, "block/dev");
312 if (load_sys(fname, buf, sizeof(buf))) {
313 /* assume this is a stale reference to a hot
314 * removed device
315 */
316 if (!(options & GET_DEVS_ALL)) {
317 free(dev);
318 continue;
319 }
320 } else {
321 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
322 }
323
324 if (!(options & GET_DEVS_ALL)) {
325 /* special case check for block devices that can go 'offline' */
326 strcpy(dbase, "block/device/state");
327 if (load_sys(fname, buf, sizeof(buf)) == 0 &&
328 strncmp(buf, "offline", 7) == 0) {
329 free(dev);
330 continue;
331 }
332 }
333
334 /* finally add this disk to the array */
335 *devp = dev;
336 devp = & dev->next;
337 dev->next = NULL;
338
339 if (options & GET_OFFSET) {
340 strcpy(dbase, "offset");
341 if (load_sys(fname, buf, sizeof(buf)))
342 goto abort;
343 dev->data_offset = strtoull(buf, NULL, 0);
344 strcpy(dbase, "new_offset");
345 if (load_sys(fname, buf, sizeof(buf)) == 0)
346 dev->new_data_offset = strtoull(buf, NULL, 0);
347 else
348 dev->new_data_offset = dev->data_offset;
349 }
350 if (options & GET_SIZE) {
351 strcpy(dbase, "size");
352 if (load_sys(fname, buf, sizeof(buf)))
353 goto abort;
354 dev->component_size = strtoull(buf, NULL, 0) * 2;
355 }
356 if (options & GET_STATE) {
357 dev->disk.state = 0;
358 strcpy(dbase, "state");
359 if (load_sys(fname, buf, sizeof(buf)))
360 goto abort;
361 if (strstr(buf, "faulty"))
362 dev->disk.state |= (1<<MD_DISK_FAULTY);
363 else {
364 sra->array.working_disks++;
365 if (strstr(buf, "in_sync")) {
366 dev->disk.state |= (1<<MD_DISK_SYNC);
367 sra->array.active_disks++;
368 }
369 if (dev->disk.state == 0)
370 sra->array.spare_disks++;
371 }
372 }
373 if (options & GET_ERROR) {
374 strcpy(buf, "errors");
375 if (load_sys(fname, buf, sizeof(buf)))
376 goto abort;
377 dev->errors = strtoul(buf, NULL, 0);
378 }
379 }
380
381 if ((options & GET_STATE) && sra->array.raid_disks)
382 sra->array.failed_disks = sra->array.raid_disks -
383 sra->array.active_disks - sra->array.spare_disks;
384
385 closedir(dir);
386 return sra;
387
388 abort:
389 if (dir)
390 closedir(dir);
391 sysfs_free(sra);
392 return NULL;
393 }
394
395 int sysfs_attr_match(const char *attr, const char *str)
396 {
397 /* See if attr, read from a sysfs file, matches
398 * str. They must either be the same, or attr can
399 * have a trailing newline or comma
400 */
401 while (*attr && *str && *attr == *str) {
402 attr++;
403 str++;
404 }
405
406 if (*str || (*attr && *attr != ',' && *attr != '\n'))
407 return 0;
408 return 1;
409 }
410
411 int sysfs_match_word(const char *word, char **list)
412 {
413 int n;
414 for (n=0; list[n]; n++)
415 if (sysfs_attr_match(word, list[n]))
416 break;
417 return n;
418 }
419
420 unsigned long long get_component_size(int fd)
421 {
422 /* Find out the component size of the array.
423 * We cannot trust GET_ARRAY_INFO ioctl as it's
424 * size field is only 32bits.
425 * So look in /sys/block/mdXXX/md/component_size
426 *
427 * This returns in units of sectors.
428 */
429 struct stat stb;
430 char fname[MAX_SYSFS_PATH_LEN];
431 int n;
432 if (fstat(fd, &stb))
433 return 0;
434 snprintf(fname, MAX_SYSFS_PATH_LEN,
435 "/sys/block/%s/md/component_size", stat2devnm(&stb));
436 fd = open(fname, O_RDONLY);
437 if (fd < 0)
438 return 0;
439 n = read(fd, fname, sizeof(fname));
440 close(fd);
441 if (n < 0 || n == sizeof(fname))
442 return 0;
443 fname[n] = 0;
444 return strtoull(fname, NULL, 10) * 2;
445 }
446
447 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
448 char *name, char *val)
449 {
450 char fname[MAX_SYSFS_PATH_LEN];
451 unsigned int n;
452 int fd;
453
454 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
455 sra->sys_name, dev?dev->sys_name:"", name);
456 fd = open(fname, O_WRONLY);
457 if (fd < 0)
458 return -1;
459 n = write(fd, val, strlen(val));
460 close(fd);
461 if (n != strlen(val)) {
462 dprintf("failed to write '%s' to '%s' (%s)\n",
463 val, fname, strerror(errno));
464 return -1;
465 }
466 return 0;
467 }
468
469 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
470 char *name, unsigned long long val)
471 {
472 char valstr[50];
473 sprintf(valstr, "%llu", val);
474 return sysfs_set_str(sra, dev, name, valstr);
475 }
476
477 int sysfs_set_num_signed(struct mdinfo *sra, struct mdinfo *dev,
478 char *name, long long val)
479 {
480 char valstr[50];
481 sprintf(valstr, "%lli", val);
482 return sysfs_set_str(sra, dev, name, valstr);
483 }
484
485 int sysfs_uevent(struct mdinfo *sra, char *event)
486 {
487 char fname[MAX_SYSFS_PATH_LEN];
488 int n;
489 int fd;
490
491 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/uevent",
492 sra->sys_name);
493 fd = open(fname, O_WRONLY);
494 if (fd < 0)
495 return -1;
496 n = write(fd, event, strlen(event));
497 close(fd);
498 if (n != (int)strlen(event)) {
499 dprintf("failed to write '%s' to '%s' (%s)\n",
500 event, fname, strerror(errno));
501 return -1;
502 }
503 return 0;
504 }
505
506 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
507 {
508 char fname[MAX_SYSFS_PATH_LEN];
509 struct stat st;
510
511 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
512 sra->sys_name, dev?dev->sys_name:"", name);
513
514 return stat(fname, &st) == 0;
515 }
516
517 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
518 char *name)
519 {
520 char fname[MAX_SYSFS_PATH_LEN];
521 int fd;
522
523 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
524 sra->sys_name, dev?dev->sys_name:"", name);
525 fd = open(fname, O_RDWR);
526 if (fd < 0)
527 fd = open(fname, O_RDONLY);
528 return fd;
529 }
530
531 int sysfs_fd_get_ll(int fd, unsigned long long *val)
532 {
533 char buf[50];
534 int n;
535 char *ep;
536
537 lseek(fd, 0, 0);
538 n = read(fd, buf, sizeof(buf));
539 if (n <= 0 || n == sizeof(buf))
540 return -2;
541 buf[n] = 0;
542 *val = strtoull(buf, &ep, 0);
543 if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
544 return -1;
545 return 0;
546 }
547
548 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
549 char *name, unsigned long long *val)
550 {
551 int n;
552 int fd;
553
554 fd = sysfs_get_fd(sra, dev, name);
555 if (fd < 0)
556 return -1;
557 n = sysfs_fd_get_ll(fd, val);
558 close(fd);
559 return n;
560 }
561
562 int sysfs_fd_get_two(int fd, unsigned long long *v1, unsigned long long *v2)
563 {
564 /* two numbers in this sysfs file, either
565 * NNN (NNN)
566 * or
567 * NNN / NNN
568 */
569 char buf[80];
570 int n;
571 char *ep, *ep2;
572
573 lseek(fd, 0, 0);
574 n = read(fd, buf, sizeof(buf));
575 if (n <= 0 || n == sizeof(buf))
576 return -2;
577 buf[n] = 0;
578 *v1 = strtoull(buf, &ep, 0);
579 if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
580 return -1;
581 while (*ep == ' ' || *ep == '/' || *ep == '(')
582 ep++;
583 *v2 = strtoull(ep, &ep2, 0);
584 if (ep2 == ep || (*ep2 != 0 && *ep2 != '\n' && *ep2 != ' ' && *ep2 != ')')) {
585 *v2 = *v1;
586 return 1;
587 }
588 return 2;
589 }
590
591 int sysfs_get_two(struct mdinfo *sra, struct mdinfo *dev,
592 char *name, unsigned long long *v1, unsigned long long *v2)
593 {
594 int n;
595 int fd;
596
597 fd = sysfs_get_fd(sra, dev, name);
598 if (fd < 0)
599 return -1;
600 n = sysfs_fd_get_two(fd, v1, v2);
601 close(fd);
602 return n;
603 }
604
605 int sysfs_fd_get_str(int fd, char *val, int size)
606 {
607 int n;
608
609 lseek(fd, 0, 0);
610 n = read(fd, val, size);
611 if (n <= 0 || n == size)
612 return -1;
613 val[n] = 0;
614 return n;
615 }
616
617 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
618 char *name, char *val, int size)
619 {
620 int n;
621 int fd;
622
623 fd = sysfs_get_fd(sra, dev, name);
624 if (fd < 0)
625 return -1;
626 n = sysfs_fd_get_str(fd, val, size);
627 close(fd);
628 return n;
629 }
630
631 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
632 {
633 unsigned long sec;
634 unsigned long msec;
635 char delay[30];
636
637 sec = ms / 1000;
638 msec = ms % 1000;
639
640 sprintf(delay, "%ld.%03ld\n", sec, msec);
641 /* this '\n' ^ needed for kernels older than 2.6.28 */
642 return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
643 }
644
645 int sysfs_set_array(struct mdinfo *info, int vers)
646 {
647 int rv = 0;
648 char ver[100];
649 int raid_disks = info->array.raid_disks;
650
651 ver[0] = 0;
652 if (info->array.major_version == -1 &&
653 info->array.minor_version == -2) {
654 char buf[1024];
655
656 strcat(strcpy(ver, "external:"), info->text_version);
657
658 /* meta version might already be set if we are setting
659 * new geometry for a reshape. In that case we don't
660 * want to over-write the 'readonly' flag that is
661 * stored in the metadata version. So read the current
662 * version first, and preserve the flag
663 */
664 if (sysfs_get_str(info, NULL, "metadata_version",
665 buf, 1024) > 0)
666 if (strlen(buf) >= 9 && buf[9] == '-')
667 ver[9] = '-';
668
669 if ((vers % 100) < 2 ||
670 sysfs_set_str(info, NULL, "metadata_version",
671 ver) < 0) {
672 pr_err("This kernel does not support external metadata.\n");
673 return 1;
674 }
675 }
676 if (info->array.level < 0)
677 return 0; /* FIXME */
678 rv |= sysfs_set_str(info, NULL, "level",
679 map_num(pers, info->array.level));
680 if (info->reshape_active && info->delta_disks != UnSet)
681 raid_disks -= info->delta_disks;
682 rv |= sysfs_set_num(info, NULL, "raid_disks", raid_disks);
683 rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
684 rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
685 rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
686 if (info->custom_array_size) {
687 int rc;
688
689 rc = sysfs_set_num(info, NULL, "array_size",
690 info->custom_array_size/2);
691 if (rc && errno == ENOENT) {
692 pr_err("This kernel does not have the md/array_size attribute, the array may be larger than expected\n");
693 rc = 0;
694 }
695 rv |= rc;
696 }
697
698 if (info->array.level > 0)
699 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
700
701 if (info->reshape_active) {
702 rv |= sysfs_set_num(info, NULL, "reshape_position",
703 info->reshape_progress);
704 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
705 rv |= sysfs_set_num(info, NULL, "layout", info->new_layout);
706 rv |= sysfs_set_num(info, NULL, "raid_disks",
707 info->array.raid_disks);
708 /* We don't set 'new_level' here. That can only happen
709 * once the reshape completes.
710 */
711 }
712
713 if (info->consistency_policy == CONSISTENCY_POLICY_PPL) {
714 if (sysfs_set_str(info, NULL, "consistency_policy",
715 map_num(consistency_policies,
716 info->consistency_policy))) {
717 pr_err("This kernel does not support PPL. Falling back to consistency-policy=resync.\n");
718 info->consistency_policy = CONSISTENCY_POLICY_RESYNC;
719 }
720 }
721
722 return rv;
723 }
724
725 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
726 {
727 char dv[PATH_MAX];
728 char nm[PATH_MAX];
729 char *dname;
730 int rv;
731 int i;
732
733 sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
734 rv = sysfs_set_str(sra, NULL, "new_dev", dv);
735 if (rv)
736 return rv;
737
738 memset(nm, 0, sizeof(nm));
739 dname = devid2kname(makedev(sd->disk.major, sd->disk.minor));
740 strcpy(sd->sys_name, "dev-");
741 strcpy(sd->sys_name+4, dname);
742
743 /* test write to see if 'recovery_start' is available */
744 if (resume && sd->recovery_start < MaxSector &&
745 sysfs_set_num(sra, sd, "recovery_start", 0)) {
746 sysfs_set_str(sra, sd, "state", "remove");
747 return -1;
748 }
749
750 rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
751 rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
752 if (sra->array.level != LEVEL_CONTAINER) {
753 if (sra->consistency_policy == CONSISTENCY_POLICY_PPL) {
754 rv |= sysfs_set_num(sra, sd, "ppl_sector", sd->ppl_sector);
755 rv |= sysfs_set_num(sra, sd, "ppl_size", sd->ppl_size);
756 }
757 if (sd->recovery_start == MaxSector)
758 /* This can correctly fail if array isn't started,
759 * yet, so just ignore status for now.
760 */
761 sysfs_set_str(sra, sd, "state", "insync");
762 if (sd->disk.raid_disk >= 0)
763 rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
764 if (resume)
765 sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
766 }
767 if (sd->bb.supported) {
768 if (sysfs_set_str(sra, sd, "state", "external_bbl")) {
769 /*
770 * backward compatibility - if kernel doesn't support
771 * bad blocks for external metadata, let it continue
772 * as long as there are none known so far
773 */
774 if (sd->bb.count) {
775 pr_err("The kernel has no support for bad blocks in external metadata\n");
776 return -1;
777 }
778 }
779
780 for (i = 0; i < sd->bb.count; i++) {
781 char s[30];
782 const struct md_bb_entry *entry = &sd->bb.entries[i];
783
784 snprintf(s, sizeof(s) - 1, "%llu %d\n", entry->sector,
785 entry->length);
786 rv |= sysfs_set_str(sra, sd, "bad_blocks", s);
787 }
788 }
789 return rv;
790 }
791
792 #if 0
793 int sysfs_disk_to_sg(int fd)
794 {
795 /* from an open block device, try find and open its corresponding
796 * scsi_generic interface
797 */
798 struct stat st;
799 char path[256];
800 char sg_path[256];
801 char sg_major_minor[10];
802 char *c;
803 DIR *dir;
804 struct dirent *de;
805 int major, minor, rv;
806
807 if (fstat(fd, &st))
808 return -1;
809
810 snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
811 major(st.st_rdev), minor(st.st_rdev));
812
813 dir = opendir(path);
814 if (!dir)
815 return -1;
816
817 de = readdir(dir);
818 while (de) {
819 if (strncmp("scsi_generic:", de->d_name,
820 strlen("scsi_generic:")) == 0)
821 break;
822 de = readdir(dir);
823 }
824 closedir(dir);
825
826 if (!de)
827 return -1;
828
829 snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
830 fd = open(sg_path, O_RDONLY);
831 if (fd < 0)
832 return fd;
833
834 rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
835 close(fd);
836 if (rv < 0 || rv == sizeof(sg_major_minor))
837 return -1;
838 else
839 sg_major_minor[rv - 1] = '\0';
840
841 c = strchr(sg_major_minor, ':');
842 *c = '\0';
843 c++;
844 major = strtol(sg_major_minor, NULL, 10);
845 minor = strtol(c, NULL, 10);
846 snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
847 (int) getpid(), major, minor);
848 if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
849 fd = open(path, O_RDONLY);
850 unlink(path);
851 return fd;
852 }
853
854 return -1;
855 }
856 #endif
857
858 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
859 {
860 /* from an open block device, try to retrieve it scsi_id */
861 struct stat st;
862 char path[256];
863 DIR *dir;
864 struct dirent *de;
865 int host, bus, target, lun;
866
867 if (fstat(fd, &st))
868 return 1;
869
870 snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
871 major(st.st_rdev), minor(st.st_rdev));
872
873 dir = opendir(path);
874 if (!dir)
875 return 1;
876
877 for (de = readdir(dir); de; de = readdir(dir)) {
878 int count;
879
880 if (de->d_type != DT_DIR)
881 continue;
882
883 count = sscanf(de->d_name, "%d:%d:%d:%d", &host, &bus, &target, &lun);
884 if (count == 4)
885 break;
886 }
887 closedir(dir);
888
889 if (!de)
890 return 1;
891
892 *id = (host << 24) | (bus << 16) | (target << 8) | (lun << 0);
893 return 0;
894 }
895
896 int sysfs_unique_holder(char *devnm, long rdev)
897 {
898 /* Check that devnm is a holder of rdev,
899 * and is the only holder.
900 * we should be locked against races by
901 * an O_EXCL on devnm
902 * Return values:
903 * 0 - not unique, not even a holder
904 * 1 - unique, this is the only holder.
905 * 2/3 - not unique, there is another holder
906 * -1 - error, cannot find the holders
907 */
908 DIR *dir;
909 struct dirent *de;
910 char dirname[100];
911 char l;
912 int ret = 0;
913 sprintf(dirname, "/sys/dev/block/%d:%d/holders",
914 major(rdev), minor(rdev));
915 dir = opendir(dirname);
916 if (!dir)
917 return -1;
918 l = strlen(dirname);
919 while ((de = readdir(dir)) != NULL) {
920 char buf[100];
921 char *sl;
922 int n;
923
924 if (de->d_ino == 0)
925 continue;
926 if (de->d_name[0] == '.')
927 continue;
928 strcpy(dirname+l, "/");
929 strcat(dirname+l, de->d_name);
930 n = readlink(dirname, buf, sizeof(buf)-1);
931 if (n <= 0)
932 continue;
933 buf[n] = 0;
934 sl = strrchr(buf, '/');
935 if (!sl)
936 continue;
937 sl++;
938
939 if (strcmp(devnm, sl) == 0)
940 ret |= 1;
941 else
942 ret |= 2;
943 }
944 closedir(dir);
945 return ret;
946 }
947
948 int sysfs_freeze_array(struct mdinfo *sra)
949 {
950 /* Try to freeze resync/rebuild on this array/container.
951 * Return -1 if the array is busy,
952 * return 0 if this kernel doesn't support 'frozen'
953 * return 1 if it worked.
954 */
955 char buf[20];
956
957 if (!sysfs_attribute_available(sra, NULL, "sync_action"))
958 return 1; /* no sync_action == frozen */
959 if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
960 return 0;
961 if (strcmp(buf, "frozen\n") == 0)
962 /* Already frozen */
963 return 0;
964 if (strcmp(buf, "idle\n") != 0 && strcmp(buf, "recover\n") != 0)
965 return -1;
966 if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
967 return 0;
968 return 1;
969 }
970
971 int sysfs_wait(int fd, int *msec)
972 {
973 /* Wait up to '*msec' for fd to have an exception condition.
974 * if msec == NULL, wait indefinitely.
975 */
976 fd_set fds;
977 int n;
978 FD_ZERO(&fds);
979 FD_SET(fd, &fds);
980 if (msec == NULL)
981 n = select(fd+1, NULL, NULL, &fds, NULL);
982 else if (*msec < 0)
983 n = 0;
984 else {
985 struct timeval start, end, tv;
986 gettimeofday(&start, NULL);
987 if (*msec < 1000) {
988 tv.tv_sec = 0;
989 tv.tv_usec = (*msec)*1000;
990 } else {
991 tv.tv_sec = (*msec)/1000;
992 tv.tv_usec = 0;
993 }
994 n = select(fd+1, NULL, NULL, &fds, &tv);
995 gettimeofday(&end, NULL);
996 end.tv_sec -= start.tv_sec;
997 *msec -= (end.tv_sec * 1000 + end.tv_usec/1000
998 - start.tv_usec/1000) + 1;
999 }
1000 return n;
1001 }