]> git.ipfire.org Git - thirdparty/mdadm.git/blob - sysfs.c
imsm: fix spare promotion
[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 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_fd(struct mdinfo *sra, struct mdinfo *dev,
446 char *name)
447 {
448 char fname[50];
449 int fd;
450
451 sprintf(fname, "/sys/block/%s/md/%s/%s",
452 sra->sys_name, dev?dev->sys_name:"", name);
453 fd = open(fname, O_RDWR);
454 if (fd < 0)
455 fd = open(fname, O_RDONLY);
456 return fd;
457 }
458
459 int sysfs_fd_get_ll(int fd, unsigned long long *val)
460 {
461 char buf[50];
462 int n;
463 char *ep;
464
465 lseek(fd, 0, 0);
466 n = read(fd, buf, sizeof(buf));
467 if (n <= 0)
468 return -1;
469 buf[n] = 0;
470 *val = strtoull(buf, &ep, 0);
471 if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
472 return -1;
473 return 0;
474 }
475
476 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
477 char *name, unsigned long long *val)
478 {
479 int n;
480 int fd;
481
482 fd = sysfs_get_fd(sra, dev, name);
483 if (fd < 0)
484 return -1;
485 n = sysfs_fd_get_ll(fd, val);
486 close(fd);
487 return n;
488 }
489
490 int sysfs_fd_get_str(int fd, char *val, int size)
491 {
492 int n;
493
494 lseek(fd, 0, 0);
495 n = read(fd, val, size);
496 if (n <= 0)
497 return -1;
498 val[n] = 0;
499 return n;
500 }
501
502 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
503 char *name, char *val, int size)
504 {
505 int n;
506 int fd;
507
508 fd = sysfs_get_fd(sra, dev, name);
509 if (fd < 0)
510 return -1;
511 n = sysfs_fd_get_str(fd, val, size);
512 close(fd);
513 return n;
514 }
515
516 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
517 {
518 unsigned long sec;
519 unsigned long msec;
520 char delay[30];
521
522 sec = ms / 1000;
523 msec = ms % 1000;
524
525 sprintf(delay, "%ld.%03ld\n", sec, msec);
526 /* this '\n' ^ needed for kernels older than 2.6.28 */
527 return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
528 }
529
530 int sysfs_set_array(struct mdinfo *info, int vers)
531 {
532 int rv = 0;
533 char ver[100];
534
535 ver[0] = 0;
536 if (info->array.major_version == -1 &&
537 info->array.minor_version == -2) {
538 strcat(strcpy(ver, "external:"), info->text_version);
539
540 if ((vers % 100) < 2 ||
541 sysfs_set_str(info, NULL, "metadata_version",
542 ver) < 0) {
543 fprintf(stderr, Name ": This kernel does not "
544 "support external metadata.\n");
545 return 1;
546 }
547 }
548 if (info->array.level < 0)
549 return 0; /* FIXME */
550 rv |= sysfs_set_str(info, NULL, "level",
551 map_num(pers, info->array.level));
552 rv |= sysfs_set_num(info, NULL, "raid_disks", info->array.raid_disks);
553 rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
554 rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
555 rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
556 if (info->custom_array_size) {
557 int rc;
558
559 rc = sysfs_set_num(info, NULL, "array_size",
560 info->custom_array_size/2);
561 if (rc && errno == ENOENT) {
562 fprintf(stderr, Name ": This kernel does not "
563 "have the md/array_size attribute, "
564 "the array may be larger than expected\n");
565 rc = 0;
566 }
567 rv |= rc;
568 }
569
570 if (info->array.level > 0)
571 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
572 return rv;
573 }
574
575 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int in_sync)
576 {
577 char dv[100];
578 char nm[100];
579 char *dname;
580 int rv;
581
582 sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
583 rv = sysfs_set_str(sra, NULL, "new_dev", dv);
584 if (rv)
585 return rv;
586
587 memset(nm, 0, sizeof(nm));
588 sprintf(dv, "/sys/dev/block/%d:%d", sd->disk.major, sd->disk.minor);
589 rv = readlink(dv, nm, sizeof(nm));
590 if (rv <= 0)
591 return -1;
592 nm[rv] = '\0';
593 dname = strrchr(nm, '/');
594 if (dname) dname++;
595 strcpy(sd->sys_name, "dev-");
596 strcpy(sd->sys_name+4, dname);
597
598 rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
599 rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
600 if (sra->array.level != LEVEL_CONTAINER) {
601 if (in_sync)
602 /* This can correctly fail if array isn't started,
603 * yet, so just ignore status for now.
604 */
605 sysfs_set_str(sra, sd, "state", "in_sync");
606 rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
607 }
608 return rv;
609 }
610
611 #if 0
612 int sysfs_disk_to_sg(int fd)
613 {
614 /* from an open block device, try find and open its corresponding
615 * scsi_generic interface
616 */
617 struct stat st;
618 char path[256];
619 char sg_path[256];
620 char sg_major_minor[8];
621 char *c;
622 DIR *dir;
623 struct dirent *de;
624 int major, minor, rv;
625
626 if (fstat(fd, &st))
627 return -1;
628
629 snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
630 major(st.st_rdev), minor(st.st_rdev));
631
632 dir = opendir(path);
633 if (!dir)
634 return -1;
635
636 de = readdir(dir);
637 while (de) {
638 if (strncmp("scsi_generic:", de->d_name,
639 strlen("scsi_generic:")) == 0)
640 break;
641 de = readdir(dir);
642 }
643 closedir(dir);
644
645 if (!de)
646 return -1;
647
648 snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
649 fd = open(sg_path, O_RDONLY);
650 if (fd < 0)
651 return fd;
652
653 rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
654 close(fd);
655 if (rv < 0)
656 return -1;
657 else
658 sg_major_minor[rv - 1] = '\0';
659
660 c = strchr(sg_major_minor, ':');
661 *c = '\0';
662 c++;
663 major = strtol(sg_major_minor, NULL, 10);
664 minor = strtol(c, NULL, 10);
665 snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
666 (int) getpid(), major, minor);
667 if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
668 fd = open(path, O_RDONLY);
669 unlink(path);
670 return fd;
671 }
672
673 return -1;
674 }
675 #endif
676
677 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
678 {
679 /* from an open block device, try to retrieve it scsi_id */
680 struct stat st;
681 char path[256];
682 char *c1, *c2;
683 DIR *dir;
684 struct dirent *de;
685
686 if (fstat(fd, &st))
687 return 1;
688
689 snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
690 major(st.st_rdev), minor(st.st_rdev));
691
692 dir = opendir(path);
693 if (!dir)
694 return 1;
695
696 de = readdir(dir);
697 while (de) {
698 if (strncmp("scsi_disk:", de->d_name,
699 strlen("scsi_disk:")) == 0)
700 break;
701 de = readdir(dir);
702 }
703 closedir(dir);
704
705 if (!de)
706 return 1;
707
708 c1 = strchr(de->d_name, ':');
709 c1++;
710 c2 = strchr(c1, ':');
711 *c2 = '\0';
712 *id = strtol(c1, NULL, 10) << 24; /* host */
713 c1 = c2 + 1;
714 c2 = strchr(c1, ':');
715 *c2 = '\0';
716 *id |= strtol(c1, NULL, 10) << 16; /* channel */
717 c1 = c2 + 1;
718 c2 = strchr(c1, ':');
719 *c2 = '\0';
720 *id |= strtol(c1, NULL, 10) << 8; /* lun */
721 c1 = c2 + 1;
722 *id |= strtol(c1, NULL, 10); /* id */
723
724 return 0;
725 }
726
727
728 int sysfs_unique_holder(int devnum, long rdev)
729 {
730 /* Check that devnum is a holder of rdev,
731 * and is the only holder.
732 * we should be locked against races by
733 * an O_EXCL on devnum
734 */
735 DIR *dir;
736 struct dirent *de;
737 char dirname[100];
738 char l;
739 int found = 0;
740 sprintf(dirname, "/sys/dev/block/%d:%d/holders",
741 major(rdev), minor(rdev));
742 dir = opendir(dirname);
743 errno = ENOENT;
744 if (!dir)
745 return 0;
746 l = strlen(dirname);
747 while ((de = readdir(dir)) != NULL) {
748 char buf[10];
749 int n;
750 int mj, mn;
751 char c;
752 int fd;
753
754 if (de->d_ino == 0)
755 continue;
756 if (de->d_name[0] == '.')
757 continue;
758 strcpy(dirname+l, "/");
759 strcat(dirname+l, de->d_name);
760 strcat(dirname+l, "/dev");
761 fd = open(dirname, O_RDONLY);
762 if (fd < 0) {
763 errno = ENOENT;
764 break;
765 }
766 n = read(fd, buf, sizeof(buf)-1);
767 close(fd);
768 buf[n] = 0;
769 if (sscanf(buf, "%d:%d%c", &mj, &mn, &c) != 3 ||
770 c != '\n') {
771 errno = ENOENT;
772 break;
773 }
774 if (mj != MD_MAJOR)
775 mn = -1-(mn>>6);
776
777 if (devnum != mn) {
778 errno = EEXIST;
779 break;
780 }
781 found = 1;
782 }
783 closedir(dir);
784 if (de)
785 return 0;
786 else
787 return found;
788 }
789
790 #ifndef MDASSEMBLE
791
792 static char *clean_states[] = {
793 "clear", "inactive", "readonly", "read-auto", "clean", NULL };
794
795 int WaitClean(char *dev, int sock, int verbose)
796 {
797 int fd;
798 struct mdinfo *mdi;
799 int rv = 1;
800 int devnum;
801
802 fd = open(dev, O_RDONLY);
803 if (fd < 0) {
804 if (verbose)
805 fprintf(stderr, Name ": Couldn't open %s: %s\n", dev, strerror(errno));
806 return 1;
807 }
808
809 devnum = fd2devnum(fd);
810 mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
811 if (!mdi) {
812 if (verbose)
813 fprintf(stderr, Name ": Failed to read sysfs attributes for "
814 "%s\n", dev);
815 close(fd);
816 return 0;
817 }
818
819 switch(mdi->array.level) {
820 case LEVEL_LINEAR:
821 case LEVEL_MULTIPATH:
822 case 0:
823 /* safemode delay is irrelevant for these levels */
824 rv = 0;
825
826 }
827
828 /* for internal metadata the kernel handles the final clean
829 * transition, containers can never be dirty
830 */
831 if (!is_subarray(mdi->text_version))
832 rv = 0;
833
834 /* safemode disabled ? */
835 if (mdi->safe_mode_delay == 0)
836 rv = 0;
837
838 if (rv) {
839 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
840 char buf[20];
841 fd_set fds;
842 struct timeval tm;
843
844 /* minimize the safe_mode_delay and prepare to wait up to 5s
845 * for writes to quiesce
846 */
847 sysfs_set_safemode(mdi, 1);
848 tm.tv_sec = 5;
849 tm.tv_usec = 0;
850
851 /* give mdmon a chance to checkpoint resync */
852 sysfs_set_str(mdi, NULL, "sync_action", "idle");
853
854 FD_ZERO(&fds);
855
856 /* wait for array_state to be clean */
857 while (1) {
858 rv = read(state_fd, buf, sizeof(buf));
859 if (rv < 0)
860 break;
861 if (sysfs_match_word(buf, clean_states) <= 4)
862 break;
863 FD_SET(state_fd, &fds);
864 rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
865 if (rv < 0 && errno != EINTR)
866 break;
867 lseek(state_fd, 0, SEEK_SET);
868 }
869 if (rv < 0)
870 rv = 1;
871 else if (fping_monitor(sock) == 0 ||
872 ping_monitor(mdi->text_version) == 0) {
873 /* we need to ping to close the window between array
874 * state transitioning to clean and the metadata being
875 * marked clean
876 */
877 rv = 0;
878 } else
879 rv = 1;
880 if (rv && verbose)
881 fprintf(stderr, Name ": Error waiting for %s to be clean\n",
882 dev);
883
884 /* restore the original safe_mode_delay */
885 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
886 close(state_fd);
887 }
888
889 sysfs_free(mdi);
890 close(fd);
891
892 return rv;
893 }
894 #endif /* MDASSEMBLE */