]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Monitor.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
mdadm: Add Documentation entries to systemd services
[thirdparty/mdadm.git] / Monitor.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2009 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 "md_u.h"
28 #include <sys/wait.h>
29 #include <limits.h>
30 #include <syslog.h>
31 #ifndef NO_LIBUDEV
32 #include <libudev.h>
33 #endif
34
35 struct state {
36 char devname[MD_NAME_MAX + sizeof("/dev/md/")]; /* length of "/dev/md/" + device name + terminating byte*/
37 char devnm[MD_NAME_MAX]; /* to sync with mdstat info */
38 unsigned int utime;
39 int err;
40 char *spare_group;
41 int active, working, failed, spare, raid;
42 int from_config;
43 int from_auto;
44 int expected_spares;
45 int devstate[MAX_DISKS];
46 dev_t devid[MAX_DISKS];
47 int percent;
48 char parent_devnm[MD_NAME_MAX]; /* For subarray, devnm of parent.
49 * For others, ""
50 */
51 struct supertype *metadata;
52 struct state *subarray;/* for a container it is a link to first subarray
53 * for a subarray it is a link to next subarray
54 * in the same container */
55 struct state *parent; /* for a subarray it is a link to its container
56 */
57 struct state *next;
58 };
59
60 struct alert_info {
61 char *mailaddr;
62 char *mailfrom;
63 char *alert_cmd;
64 int dosyslog;
65 };
66 static int make_daemon(char *pidfile);
67 static int check_one_sharer(int scan);
68 static void write_autorebuild_pid(void);
69 static void alert(char *event, char *dev, char *disc, struct alert_info *info);
70 static int check_array(struct state *st, struct mdstat_ent *mdstat,
71 int test, struct alert_info *info,
72 int increments, char *prefer);
73 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
74 int test, struct alert_info *info);
75 static void try_spare_migration(struct state *statelist, struct alert_info *info);
76 static void link_containers_with_subarrays(struct state *list);
77 static void free_statelist(struct state *statelist);
78 #ifndef NO_LIBUDEV
79 static int check_udev_activity(void);
80 #endif
81
82 int Monitor(struct mddev_dev *devlist,
83 char *mailaddr, char *alert_cmd,
84 struct context *c,
85 int daemonise, int oneshot,
86 int dosyslog, char *pidfile, int increments,
87 int share)
88 {
89 /*
90 * Every few seconds, scan every md device looking for changes
91 * When a change is found, log it, possibly run the alert command,
92 * and possibly send Email
93 *
94 * For each array, we record:
95 * Update time
96 * active/working/failed/spare drives
97 * State of each device.
98 * %rebuilt if rebuilding
99 *
100 * If the update time changes, check out all the data again
101 * It is possible that we cannot get the state of each device
102 * due to bugs in the md kernel module.
103 * We also read /proc/mdstat to get rebuild percent,
104 * and to get state on all active devices incase of kernel bug.
105 *
106 * Events are:
107 * Fail
108 * An active device had Faulty set or Active/Sync removed
109 * FailSpare
110 * A spare device had Faulty set
111 * SpareActive
112 * An active device had a reverse transition
113 * RebuildStarted
114 * percent went from -1 to +ve
115 * RebuildNN
116 * percent went from below to not-below NN%
117 * DeviceDisappeared
118 * Couldn't access a device which was previously visible
119 *
120 * if we detect an array with active<raid and spare==0
121 * we look at other arrays that have same spare-group
122 * If we find one with active==raid and spare>0,
123 * and if we can get_disk_info and find a name
124 * Then we hot-remove and hot-add to the other array
125 *
126 * If devlist is NULL, then we can monitor everything because --scan
127 * was given. We get an initial list from config file and add anything
128 * that appears in /proc/mdstat
129 */
130
131 struct state *statelist = NULL;
132 int finished = 0;
133 struct mdstat_ent *mdstat = NULL;
134 char *mailfrom;
135 struct alert_info info;
136 struct mddev_ident *mdlist;
137 int delay_for_event = c->delay;
138
139 if (!mailaddr)
140 mailaddr = conf_get_mailaddr();
141
142 if (!alert_cmd)
143 alert_cmd = conf_get_program();
144
145 mailfrom = conf_get_mailfrom();
146
147 if (c->scan && !mailaddr && !alert_cmd && !dosyslog) {
148 pr_err("No mail address or alert command - not monitoring.\n");
149 return 1;
150 }
151
152 if (c->verbose) {
153 pr_err("Monitor is started with delay %ds\n", c->delay);
154 if (mailaddr)
155 pr_err("Monitor using email address %s\n", mailaddr);
156 if (alert_cmd)
157 pr_err("Monitor using program %s\n", alert_cmd);
158 }
159
160 info.alert_cmd = alert_cmd;
161 info.mailaddr = mailaddr;
162 info.mailfrom = mailfrom;
163 info.dosyslog = dosyslog;
164
165 if (share){
166 if (check_one_sharer(c->scan))
167 return 1;
168 }
169
170 if (daemonise) {
171 int rv = make_daemon(pidfile);
172 if (rv >= 0)
173 return rv;
174 }
175
176 if (share)
177 write_autorebuild_pid();
178
179 if (devlist == NULL) {
180 mdlist = conf_get_ident(NULL);
181 for (; mdlist; mdlist = mdlist->next) {
182 struct state *st;
183
184 if (mdlist->devname == NULL)
185 continue;
186 if (strcasecmp(mdlist->devname, "<ignore>") == 0)
187 continue;
188 if (!is_mddev(mdlist->devname)) {
189 free_statelist(statelist);
190 return 1;
191 }
192
193 st = xcalloc(1, sizeof *st);
194 snprintf(st->devname, MD_NAME_MAX + sizeof("/dev/md/"),
195 "/dev/md/%s", basename(mdlist->devname));
196 st->next = statelist;
197 st->devnm[0] = 0;
198 st->percent = RESYNC_UNKNOWN;
199 st->from_config = 1;
200 st->expected_spares = mdlist->spare_disks;
201 if (mdlist->spare_group)
202 st->spare_group = xstrdup(mdlist->spare_group);
203 statelist = st;
204 }
205 } else {
206 struct mddev_dev *dv;
207
208 for (dv = devlist; dv; dv = dv->next) {
209 struct state *st;
210
211 if (!is_mddev(dv->devname)) {
212 free_statelist(statelist);
213 return 1;
214 }
215
216 st = xcalloc(1, sizeof *st);
217 mdlist = conf_get_ident(dv->devname);
218 snprintf(st->devname, MD_NAME_MAX + sizeof("/dev/md/"), "%s", dv->devname);
219 st->next = statelist;
220 st->devnm[0] = 0;
221 st->percent = RESYNC_UNKNOWN;
222 st->expected_spares = -1;
223 if (mdlist) {
224 st->expected_spares = mdlist->spare_disks;
225 if (mdlist->spare_group)
226 st->spare_group = xstrdup(mdlist->spare_group);
227 }
228 statelist = st;
229 }
230 }
231
232 while (!finished) {
233 int new_found = 0;
234 struct state *st, **stp;
235 int anydegraded = 0;
236 int anyredundant = 0;
237
238 if (mdstat)
239 free_mdstat(mdstat);
240 mdstat = mdstat_read(oneshot ? 0 : 1, 0);
241
242 for (st = statelist; st; st = st->next) {
243 if (check_array(st, mdstat, c->test, &info,
244 increments, c->prefer))
245 anydegraded = 1;
246 /* for external arrays, metadata is filled for
247 * containers only
248 */
249 if (st->metadata && st->metadata->ss->external)
250 continue;
251 if (st->err == 0 && !anyredundant)
252 anyredundant = 1;
253 }
254
255 /* now check if there are any new devices found in mdstat */
256 if (c->scan)
257 new_found = add_new_arrays(mdstat, &statelist, c->test,
258 &info);
259
260 /* If an array has active < raid && spare == 0 && spare_group != NULL
261 * Look for another array with spare > 0 and active == raid and same spare_group
262 * if found, choose a device and hotremove/hotadd
263 */
264 if (share && anydegraded)
265 try_spare_migration(statelist, &info);
266 if (!new_found) {
267 if (oneshot)
268 break;
269 else if (!anyredundant) {
270 pr_err("No array with redundancy detected, stopping\n");
271 break;
272 }
273 else {
274 #ifndef NO_LIBUDEV
275 /*
276 * Wait for udevd to finish new devices
277 * processing.
278 */
279 if (mdstat_wait(delay_for_event) &&
280 check_udev_activity())
281 pr_err("Error while waiting for UDEV to complete new devices processing\n");
282 #else
283 int wait_result = mdstat_wait(delay_for_event);
284 /*
285 * Give chance to process new device
286 */
287 if (wait_result != 0) {
288 if (c->delay > 5)
289 delay_for_event = 5;
290 } else
291 delay_for_event = c->delay;
292 #endif
293 mdstat_close();
294 }
295 }
296 c->test = 0;
297
298 for (stp = &statelist; (st = *stp) != NULL; ) {
299 if (st->from_auto && st->err > 5) {
300 *stp = st->next;
301 if (st->spare_group)
302 free(st->spare_group);
303
304 free(st);
305 } else
306 stp = &st->next;
307 }
308 }
309
310 free_statelist(statelist);
311
312 if (pidfile)
313 unlink(pidfile);
314 return 0;
315 }
316
317 static int make_daemon(char *pidfile)
318 {
319 /* Return:
320 * -1 in the forked daemon
321 * 0 in the parent
322 * 1 on error
323 * so a none-negative becomes the exit code.
324 */
325 int pid = fork();
326 if (pid > 0) {
327 if (!pidfile)
328 printf("%d\n", pid);
329 else {
330 FILE *pid_file = NULL;
331 int fd = open(pidfile, O_WRONLY | O_CREAT | O_TRUNC,
332 0644);
333 if (fd >= 0)
334 pid_file = fdopen(fd, "w");
335 if (!pid_file)
336 perror("cannot create pid file");
337 else {
338 fprintf(pid_file,"%d\n", pid);
339 fclose(pid_file);
340 }
341 }
342 return 0;
343 }
344 if (pid < 0) {
345 perror("daemonise");
346 return 1;
347 }
348 manage_fork_fds(0);
349 setsid();
350 return -1;
351 }
352
353 static int check_one_sharer(int scan)
354 {
355 int pid;
356 FILE *comm_fp;
357 FILE *fp;
358 char comm_path[PATH_MAX];
359 char path[PATH_MAX];
360 char comm[20];
361
362 sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
363 fp = fopen(path, "r");
364 if (fp) {
365 if (fscanf(fp, "%d", &pid) != 1)
366 pid = -1;
367 snprintf(comm_path, sizeof(comm_path),
368 "/proc/%d/comm", pid);
369 comm_fp = fopen(comm_path, "r");
370 if (comm_fp) {
371 if (fscanf(comm_fp, "%19s", comm) &&
372 strncmp(basename(comm), Name, strlen(Name)) == 0) {
373 if (scan) {
374 pr_err("Only one autorebuild process allowed in scan mode, aborting\n");
375 fclose(comm_fp);
376 fclose(fp);
377 return 1;
378 } else {
379 pr_err("Warning: One autorebuild process already running.\n");
380 }
381 }
382 fclose(comm_fp);
383 }
384 fclose(fp);
385 }
386 return 0;
387 }
388
389 static void write_autorebuild_pid()
390 {
391 char path[PATH_MAX];
392 int pid;
393 FILE *fp = NULL;
394 sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
395
396 if (mkdir(MDMON_DIR, 0700) < 0 && errno != EEXIST) {
397 pr_err("Can't create autorebuild.pid file\n");
398 } else {
399 int fd = open(path, O_WRONLY | O_CREAT | O_TRUNC, 0700);
400
401 if (fd >= 0)
402 fp = fdopen(fd, "w");
403
404 if (!fp)
405 pr_err("Can't create autorebuild.pid file\n");
406 else {
407 pid = getpid();
408 fprintf(fp, "%d\n", pid);
409 fclose(fp);
410 }
411 }
412 }
413
414 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
415 {
416 int priority;
417
418 if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
419 time_t now = time(0);
420
421 printf("%1.15s: %s on %s %s\n", ctime(&now) + 4,
422 event, dev, disc?disc:"unknown device");
423 }
424 if (info->alert_cmd) {
425 int pid = fork();
426 switch(pid) {
427 default:
428 waitpid(pid, NULL, 0);
429 break;
430 case -1:
431 break;
432 case 0:
433 execl(info->alert_cmd, info->alert_cmd,
434 event, dev, disc, NULL);
435 exit(2);
436 }
437 }
438 if (info->mailaddr && (strncmp(event, "Fail", 4) == 0 ||
439 strncmp(event, "Test", 4) == 0 ||
440 strncmp(event, "Spares", 6) == 0 ||
441 strncmp(event, "Degrade", 7) == 0)) {
442 FILE *mp = popen(Sendmail, "w");
443 if (mp) {
444 FILE *mdstat;
445 char hname[256];
446
447 gethostname(hname, sizeof(hname));
448 signal_s(SIGPIPE, SIG_IGN);
449
450 if (info->mailfrom)
451 fprintf(mp, "From: %s\n", info->mailfrom);
452 else
453 fprintf(mp, "From: %s monitoring <root>\n",
454 Name);
455 fprintf(mp, "To: %s\n", info->mailaddr);
456 fprintf(mp, "Subject: %s event on %s:%s\n\n",
457 event, dev, hname);
458
459 fprintf(mp,
460 "This is an automatically generated mail message from %s\n", Name);
461 fprintf(mp, "running on %s\n\n", hname);
462
463 fprintf(mp,
464 "A %s event had been detected on md device %s.\n\n", event, dev);
465
466 if (disc && disc[0] != ' ')
467 fprintf(mp,
468 "It could be related to component device %s.\n\n", disc);
469 if (disc && disc[0] == ' ')
470 fprintf(mp, "Extra information:%s.\n\n", disc);
471
472 fprintf(mp, "Faithfully yours, etc.\n");
473
474 mdstat = fopen("/proc/mdstat", "r");
475 if (mdstat) {
476 char buf[8192];
477 int n;
478 fprintf(mp,
479 "\nP.S. The /proc/mdstat file currently contains the following:\n\n");
480 while ((n = fread(buf, 1, sizeof(buf),
481 mdstat)) > 0)
482 n = fwrite(buf, 1, n, mp);
483 fclose(mdstat);
484 }
485 pclose(mp);
486 }
487 }
488
489 /* log the event to syslog maybe */
490 if (info->dosyslog) {
491 /* Log at a different severity depending on the event.
492 *
493 * These are the critical events: */
494 if (strncmp(event, "Fail", 4) == 0 ||
495 strncmp(event, "Degrade", 7) == 0 ||
496 strncmp(event, "DeviceDisappeared", 17) == 0)
497 priority = LOG_CRIT;
498 /* Good to know about, but are not failures: */
499 else if (strncmp(event, "Rebuild", 7) == 0 ||
500 strncmp(event, "MoveSpare", 9) == 0 ||
501 strncmp(event, "Spares", 6) != 0)
502 priority = LOG_WARNING;
503 /* Everything else: */
504 else
505 priority = LOG_INFO;
506
507 if (disc && disc[0] != ' ')
508 syslog(priority,
509 "%s event detected on md device %s, component device %s", event, dev, disc);
510 else if (disc)
511 syslog(priority,
512 "%s event detected on md device %s: %s",
513 event, dev, disc);
514 else
515 syslog(priority,
516 "%s event detected on md device %s",
517 event, dev);
518 }
519 }
520
521 static int check_array(struct state *st, struct mdstat_ent *mdstat,
522 int test, struct alert_info *ainfo,
523 int increments, char *prefer)
524 {
525 /* Update the state 'st' to reflect any changes shown in mdstat,
526 * or found by directly examining the array, and return
527 * '1' if the array is degraded, or '0' if it is optimal (or dead).
528 */
529 struct { int state, major, minor; } info[MAX_DISKS];
530 struct mdinfo *sra = NULL;
531 mdu_array_info_t array;
532 struct mdstat_ent *mse = NULL, *mse2;
533 char *dev = st->devname;
534 int fd;
535 int i;
536 int remaining_disks;
537 int last_disk;
538 int new_array = 0;
539 int retval;
540 int is_container = 0;
541 unsigned long redundancy_only_flags = 0;
542
543 if (test)
544 alert("TestMessage", dev, NULL, ainfo);
545
546 retval = 0;
547
548 fd = open(dev, O_RDONLY);
549 if (fd < 0)
550 goto disappeared;
551
552 if (st->devnm[0] == 0)
553 snprintf(st->devnm, MD_NAME_MAX, "%s", fd2devnm(fd));
554
555 for (mse2 = mdstat; mse2; mse2 = mse2->next)
556 if (strcmp(mse2->devnm, st->devnm) == 0) {
557 mse2->devnm[0] = 0; /* flag it as "used" */
558 mse = mse2;
559 }
560
561 if (!mse) {
562 /* duplicated array in statelist
563 * or re-created after reading mdstat
564 */
565 st->err++;
566 goto out;
567 }
568
569 if (mse->level == NULL)
570 is_container = 1;
571
572 if (!is_container && !md_array_active(fd))
573 goto disappeared;
574
575 fcntl(fd, F_SETFD, FD_CLOEXEC);
576 if (md_get_array_info(fd, &array) < 0)
577 goto disappeared;
578
579 if (!is_container && map_name(pers, mse->level) > 0)
580 redundancy_only_flags |= GET_MISMATCH;
581
582 sra = sysfs_read(-1, st->devnm, GET_LEVEL | GET_DISKS | GET_DEVS |
583 GET_STATE | redundancy_only_flags);
584
585 if (!sra)
586 goto disappeared;
587
588 /* It's much easier to list what array levels can't
589 * have a device disappear than all of them that can
590 */
591 if (sra->array.level == 0 || sra->array.level == -1) {
592 if (!st->err && !st->from_config)
593 alert("DeviceDisappeared", dev, " Wrong-Level", ainfo);
594 st->err++;
595 goto out;
596 }
597
598 /* this array is in /proc/mdstat */
599 if (array.utime == 0)
600 /* external arrays don't update utime, so
601 * just make sure it is always different. */
602 array.utime = st->utime + 1;;
603
604 if (st->err) {
605 /* New array appeared where previously had an error */
606 st->err = 0;
607 st->percent = RESYNC_NONE;
608 new_array = 1;
609 if (!is_container)
610 alert("NewArray", st->devname, NULL, ainfo);
611 }
612
613 if (st->utime == array.utime && st->failed == sra->array.failed_disks &&
614 st->working == sra->array.working_disks &&
615 st->spare == sra->array.spare_disks &&
616 (mse == NULL || (mse->percent == st->percent))) {
617 if ((st->active < st->raid) && st->spare == 0)
618 retval = 1;
619 goto out;
620 }
621 if (st->utime == 0 && /* new array */
622 mse->pattern && strchr(mse->pattern, '_') /* degraded */)
623 alert("DegradedArray", dev, NULL, ainfo);
624
625 if (st->utime == 0 && /* new array */ st->expected_spares > 0 &&
626 sra->array.spare_disks < st->expected_spares)
627 alert("SparesMissing", dev, NULL, ainfo);
628 if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
629 mse->percent >= 0)
630 alert("RebuildStarted", dev, NULL, ainfo);
631 if (st->percent >= 0 && mse->percent >= 0 &&
632 (mse->percent / increments) > (st->percent / increments)) {
633 char percentalert[18];
634 /*
635 * "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
636 */
637
638 if((mse->percent / increments) == 0)
639 snprintf(percentalert, sizeof(percentalert),
640 "RebuildStarted");
641 else
642 snprintf(percentalert, sizeof(percentalert),
643 "Rebuild%02d", mse->percent);
644
645 alert(percentalert, dev, NULL, ainfo);
646 }
647
648 if (mse->percent == RESYNC_NONE && st->percent >= 0) {
649 /* Rebuild/sync/whatever just finished.
650 * If there is a number in /mismatch_cnt,
651 * we should report that.
652 */
653 if (sra && sra->mismatch_cnt > 0) {
654 char cnt[80];
655 snprintf(cnt, sizeof(cnt),
656 " mismatches found: %d (on raid level %d)",
657 sra->mismatch_cnt, sra->array.level);
658 alert("RebuildFinished", dev, cnt, ainfo);
659 } else
660 alert("RebuildFinished", dev, NULL, ainfo);
661 }
662 st->percent = mse->percent;
663
664 remaining_disks = sra->array.nr_disks;
665 for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) {
666 mdu_disk_info_t disc;
667 disc.number = i;
668 if (md_get_disk_info(fd, &disc) >= 0) {
669 info[i].state = disc.state;
670 info[i].major = disc.major;
671 info[i].minor = disc.minor;
672 if (disc.major || disc.minor)
673 remaining_disks --;
674 } else
675 info[i].major = info[i].minor = 0;
676 }
677 last_disk = i;
678
679 if (mse->metadata_version &&
680 strncmp(mse->metadata_version, "external:", 9) == 0 &&
681 is_subarray(mse->metadata_version+9)) {
682 char *sl;
683 snprintf(st->parent_devnm, MD_NAME_MAX, "%s", mse->metadata_version + 10);
684 sl = strchr(st->parent_devnm, '/');
685 if (sl)
686 *sl = 0;
687 } else
688 st->parent_devnm[0] = 0;
689 if (st->metadata == NULL && st->parent_devnm[0] == 0)
690 st->metadata = super_by_fd(fd, NULL);
691
692 for (i = 0; i < MAX_DISKS; i++) {
693 mdu_disk_info_t disc = {0, 0, 0, 0, 0};
694 int newstate = 0;
695 int change;
696 char *dv = NULL;
697 disc.number = i;
698 if (i < last_disk && (info[i].major || info[i].minor)) {
699 newstate = info[i].state;
700 dv = map_dev_preferred(info[i].major, info[i].minor, 1,
701 prefer);
702 disc.state = newstate;
703 disc.major = info[i].major;
704 disc.minor = info[i].minor;
705 } else
706 newstate = (1 << MD_DISK_REMOVED);
707
708 if (dv == NULL && st->devid[i])
709 dv = map_dev_preferred(major(st->devid[i]),
710 minor(st->devid[i]), 1, prefer);
711 change = newstate ^ st->devstate[i];
712 if (st->utime && change && !st->err && !new_array) {
713 if ((st->devstate[i]&change) & (1 << MD_DISK_SYNC))
714 alert("Fail", dev, dv, ainfo);
715 else if ((newstate & (1 << MD_DISK_FAULTY)) &&
716 (disc.major || disc.minor) &&
717 st->devid[i] == makedev(disc.major,
718 disc.minor))
719 alert("FailSpare", dev, dv, ainfo);
720 else if ((newstate&change) & (1 << MD_DISK_SYNC))
721 alert("SpareActive", dev, dv, ainfo);
722 }
723 st->devstate[i] = newstate;
724 st->devid[i] = makedev(disc.major, disc.minor);
725 }
726 st->active = sra->array.active_disks;
727 st->working = sra->array.working_disks;
728 st->spare = sra->array.spare_disks;
729 st->failed = sra->array.failed_disks;
730 st->utime = array.utime;
731 st->raid = sra->array.raid_disks;
732 st->err = 0;
733 if ((st->active < st->raid) && st->spare == 0)
734 retval = 1;
735
736 out:
737 if (sra)
738 sysfs_free(sra);
739 if (fd >= 0)
740 close(fd);
741 return retval;
742
743 disappeared:
744 if (!st->err && !is_container)
745 alert("DeviceDisappeared", dev, NULL, ainfo);
746 st->err++;
747 goto out;
748 }
749
750 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
751 int test, struct alert_info *info)
752 {
753 struct mdstat_ent *mse;
754 int new_found = 0;
755 char *name;
756
757 for (mse = mdstat; mse; mse = mse->next)
758 if (mse->devnm[0] && (!mse->level || /* retrieve containers */
759 (strcmp(mse->level, "raid0") != 0 &&
760 strcmp(mse->level, "linear") != 0))) {
761 struct state *st = xcalloc(1, sizeof *st);
762 mdu_array_info_t array;
763 int fd;
764
765 name = get_md_name(mse->devnm);
766 if (!name) {
767 free(st);
768 continue;
769 }
770
771 snprintf(st->devname, MD_NAME_MAX + sizeof("/dev/md/"), "%s", name);
772 if ((fd = open(st->devname, O_RDONLY)) < 0 ||
773 md_get_array_info(fd, &array) < 0) {
774 /* no such array */
775 if (fd >= 0)
776 close(fd);
777 put_md_name(st->devname);
778 if (st->metadata) {
779 st->metadata->ss->free_super(st->metadata);
780 free(st->metadata);
781 }
782 free(st);
783 continue;
784 }
785 close(fd);
786 st->next = *statelist;
787 st->err = 1;
788 st->from_auto = 1;
789 snprintf(st->devnm, MD_NAME_MAX, "%s", mse->devnm);
790 st->percent = RESYNC_UNKNOWN;
791 st->expected_spares = -1;
792 if (mse->metadata_version &&
793 strncmp(mse->metadata_version,
794 "external:", 9) == 0 &&
795 is_subarray(mse->metadata_version+9)) {
796 char *sl;
797 snprintf(st->parent_devnm, MD_NAME_MAX,
798 "%s", mse->metadata_version + 10);
799 sl = strchr(st->parent_devnm, '/');
800 *sl = 0;
801 } else
802 st->parent_devnm[0] = 0;
803 *statelist = st;
804 if (test)
805 alert("TestMessage", st->devname, NULL, info);
806 new_found = 1;
807 }
808 return new_found;
809 }
810
811 static int get_required_spare_criteria(struct state *st,
812 struct spare_criteria *sc)
813 {
814 int fd;
815
816 if (!st->metadata || !st->metadata->ss->get_spare_criteria) {
817 sc->min_size = 0;
818 sc->sector_size = 0;
819 return 0;
820 }
821
822 fd = open(st->devname, O_RDONLY);
823 if (fd < 0)
824 return 1;
825 if (st->metadata->ss->external)
826 st->metadata->ss->load_container(st->metadata, fd, st->devname);
827 else
828 st->metadata->ss->load_super(st->metadata, fd, st->devname);
829 close(fd);
830 if (!st->metadata->sb)
831 return 1;
832
833 st->metadata->ss->get_spare_criteria(st->metadata, sc);
834 st->metadata->ss->free_super(st->metadata);
835
836 return 0;
837 }
838
839 static int check_donor(struct state *from, struct state *to)
840 {
841 struct state *sub;
842
843 if (from == to)
844 return 0;
845 if (from->parent)
846 /* Cannot move from a member */
847 return 0;
848 if (from->err)
849 return 0;
850 for (sub = from->subarray; sub; sub = sub->subarray)
851 /* If source array has degraded subarrays, don't
852 * remove anything
853 */
854 if (sub->active < sub->raid)
855 return 0;
856 if (from->metadata->ss->external == 0)
857 if (from->active < from->raid)
858 return 0;
859 if (from->spare <= 0)
860 return 0;
861 return 1;
862 }
863
864 static dev_t choose_spare(struct state *from, struct state *to,
865 struct domainlist *domlist, struct spare_criteria *sc)
866 {
867 int d;
868 dev_t dev = 0;
869
870 for (d = from->raid; !dev && d < MAX_DISKS; d++) {
871 if (from->devid[d] > 0 && from->devstate[d] == 0) {
872 struct dev_policy *pol;
873 unsigned long long dev_size;
874 unsigned int dev_sector_size;
875
876 if (to->metadata->ss->external &&
877 test_partition_from_id(from->devid[d]))
878 continue;
879
880 if (sc->min_size &&
881 dev_size_from_id(from->devid[d], &dev_size) &&
882 dev_size < sc->min_size)
883 continue;
884
885 if (sc->sector_size &&
886 dev_sector_size_from_id(from->devid[d],
887 &dev_sector_size) &&
888 sc->sector_size != dev_sector_size)
889 continue;
890
891 pol = devid_policy(from->devid[d]);
892 if (from->spare_group)
893 pol_add(&pol, pol_domain,
894 from->spare_group, NULL);
895 if (domain_test(domlist, pol,
896 to->metadata->ss->name) == 1)
897 dev = from->devid[d];
898 dev_policy_free(pol);
899 }
900 }
901 return dev;
902 }
903
904 static dev_t container_choose_spare(struct state *from, struct state *to,
905 struct domainlist *domlist,
906 struct spare_criteria *sc, int active)
907 {
908 /* This is similar to choose_spare, but we cannot trust devstate,
909 * so we need to read the metadata instead
910 */
911 struct mdinfo *list;
912 struct supertype *st = from->metadata;
913 int fd = open(from->devname, O_RDONLY);
914 int err;
915 dev_t dev = 0;
916
917 if (fd < 0)
918 return 0;
919 if (!st->ss->getinfo_super_disks) {
920 close(fd);
921 return 0;
922 }
923
924 err = st->ss->load_container(st, fd, NULL);
925 close(fd);
926 if (err)
927 return 0;
928
929 if (from == to) {
930 /* We must check if number of active disks has not increased
931 * since ioctl in main loop. mdmon may have added spare
932 * to subarray. If so we do not need to look for more spares
933 * so return non zero value */
934 int active_cnt = 0;
935 struct mdinfo *dp;
936 list = st->ss->getinfo_super_disks(st);
937 if (!list) {
938 st->ss->free_super(st);
939 return 1;
940 }
941 dp = list->devs;
942 while (dp) {
943 if (dp->disk.state & (1 << MD_DISK_SYNC) &&
944 !(dp->disk.state & (1 << MD_DISK_FAULTY)))
945 active_cnt++;
946 dp = dp->next;
947 }
948 sysfs_free(list);
949 if (active < active_cnt) {
950 /* Spare just activated.*/
951 st->ss->free_super(st);
952 return 1;
953 }
954 }
955
956 /* We only need one spare so full list not needed */
957 list = container_choose_spares(st, sc, domlist, from->spare_group,
958 to->metadata->ss->name, 1);
959 if (list) {
960 struct mdinfo *disks = list->devs;
961 if (disks)
962 dev = makedev(disks->disk.major, disks->disk.minor);
963 sysfs_free(list);
964 }
965 st->ss->free_super(st);
966 return dev;
967 }
968
969 static void try_spare_migration(struct state *statelist, struct alert_info *info)
970 {
971 struct state *from;
972 struct state *st;
973 struct spare_criteria sc;
974
975 link_containers_with_subarrays(statelist);
976 for (st = statelist; st; st = st->next)
977 if (st->active < st->raid && st->spare == 0 && !st->err) {
978 struct domainlist *domlist = NULL;
979 int d;
980 struct state *to = st;
981
982 if (to->parent_devnm[0] && !to->parent)
983 /* subarray monitored without parent container
984 * we can't move spares here */
985 continue;
986
987 if (to->parent)
988 /* member of a container */
989 to = to->parent;
990
991 if (get_required_spare_criteria(to, &sc))
992 continue;
993 if (to->metadata->ss->external) {
994 /* We must make sure there is
995 * no suitable spare in container already.
996 * If there is we don't add more */
997 dev_t devid = container_choose_spare(
998 to, to, NULL, &sc, st->active);
999 if (devid > 0)
1000 continue;
1001 }
1002 for (d = 0; d < MAX_DISKS; d++)
1003 if (to->devid[d])
1004 domainlist_add_dev(&domlist,
1005 to->devid[d],
1006 to->metadata->ss->name);
1007 if (to->spare_group)
1008 domain_add(&domlist, to->spare_group);
1009 /*
1010 * No spare migration if the destination
1011 * has no domain. Skip this array.
1012 */
1013 if (!domlist)
1014 continue;
1015 for (from=statelist ; from ; from=from->next) {
1016 dev_t devid;
1017 if (!check_donor(from, to))
1018 continue;
1019 if (from->metadata->ss->external)
1020 devid = container_choose_spare(
1021 from, to, domlist, &sc, 0);
1022 else
1023 devid = choose_spare(from, to, domlist,
1024 &sc);
1025 if (devid > 0 &&
1026 move_spare(from->devname, to->devname,
1027 devid)) {
1028 alert("MoveSpare", to->devname,
1029 from->devname, info);
1030 break;
1031 }
1032 }
1033 domain_free(domlist);
1034 }
1035 }
1036
1037 /* search the statelist to connect external
1038 * metadata subarrays with their containers
1039 * We always completely rebuild the tree from scratch as
1040 * that is safest considering the possibility of entries
1041 * disappearing or changing.
1042 */
1043 static void link_containers_with_subarrays(struct state *list)
1044 {
1045 struct state *st;
1046 struct state *cont;
1047 for (st = list; st; st = st->next) {
1048 st->parent = NULL;
1049 st->subarray = NULL;
1050 }
1051 for (st = list; st; st = st->next)
1052 if (st->parent_devnm[0])
1053 for (cont = list; cont; cont = cont->next)
1054 if (!cont->err && cont->parent_devnm[0] == 0 &&
1055 strcmp(cont->devnm, st->parent_devnm) == 0) {
1056 st->parent = cont;
1057 st->subarray = cont->subarray;
1058 cont->subarray = st;
1059 break;
1060 }
1061 }
1062
1063 /**
1064 * free_statelist() - Frees statelist.
1065 * @statelist: statelist to free
1066 */
1067 static void free_statelist(struct state *statelist)
1068 {
1069 struct state *tmp = NULL;
1070
1071 while (statelist) {
1072 if (statelist->spare_group)
1073 free(statelist->spare_group);
1074
1075 tmp = statelist;
1076 statelist = statelist->next;
1077 free(tmp);
1078 }
1079 }
1080
1081 #ifndef NO_LIBUDEV
1082 /* function: check_udev_activity
1083 * Description: Function waits for udev to finish
1084 * events processing.
1085 * Returns:
1086 * 1 - detected error while opening udev
1087 * 2 - timeout
1088 * 0 - successfull completion
1089 */
1090 static int check_udev_activity(void)
1091 {
1092 struct udev *udev = NULL;
1093 struct udev_queue *udev_queue = NULL;
1094 int timeout_cnt = 30;
1095 int rc = 0;
1096
1097 /*
1098 * In rare cases systemd may not have udevm,
1099 * in such cases just exit with rc 0
1100 */
1101 if (!use_udev())
1102 goto out;
1103
1104 udev = udev_new();
1105 if (!udev) {
1106 rc = 1;
1107 goto out;
1108 }
1109
1110 udev_queue = udev_queue_new(udev);
1111 if (!udev_queue) {
1112 rc = 1;
1113 goto out;
1114 }
1115
1116 if (udev_queue_get_queue_is_empty(udev_queue))
1117 goto out;
1118
1119 while (!udev_queue_get_queue_is_empty(udev_queue)) {
1120 sleep(1);
1121
1122 if (timeout_cnt)
1123 timeout_cnt--;
1124 else {
1125 rc = 2;
1126 goto out;
1127 }
1128 }
1129
1130 out:
1131 if (udev_queue)
1132 udev_queue_unref(udev_queue);
1133 if (udev)
1134 udev_unref(udev);
1135 return rc;
1136 }
1137 #endif
1138
1139 /* Not really Monitor but ... */
1140 int Wait(char *dev)
1141 {
1142 char devnm[32];
1143 dev_t rdev;
1144 char *tmp;
1145 int rv = 1;
1146 int frozen_remaining = 3;
1147
1148 if (!stat_is_blkdev(dev, &rdev))
1149 return 2;
1150
1151 tmp = devid2devnm(rdev);
1152 if (!tmp) {
1153 pr_err("Cannot get md device name.\n");
1154 return 2;
1155 }
1156
1157 strcpy(devnm, tmp);
1158
1159 while(1) {
1160 struct mdstat_ent *ms = mdstat_read(1, 0);
1161 struct mdstat_ent *e;
1162
1163 for (e = ms; e; e = e->next)
1164 if (strcmp(e->devnm, devnm) == 0)
1165 break;
1166
1167 if (e && e->percent == RESYNC_NONE) {
1168 /* We could be in the brief pause before something
1169 * starts. /proc/mdstat doesn't show that, but
1170 * sync_action does.
1171 */
1172 struct mdinfo mdi;
1173 char buf[21];
1174
1175 if (sysfs_init(&mdi, -1, devnm))
1176 return 2;
1177 if (sysfs_get_str(&mdi, NULL, "sync_action",
1178 buf, 20) > 0 &&
1179 strcmp(buf,"idle\n") != 0) {
1180 e->percent = RESYNC_UNKNOWN;
1181 if (strcmp(buf, "frozen\n") == 0) {
1182 if (frozen_remaining == 0)
1183 e->percent = RESYNC_NONE;
1184 else
1185 frozen_remaining -= 1;
1186 }
1187 }
1188 }
1189 if (!e || e->percent == RESYNC_NONE) {
1190 if (e && e->metadata_version &&
1191 strncmp(e->metadata_version, "external:", 9) == 0) {
1192 if (is_subarray(&e->metadata_version[9]))
1193 ping_monitor(&e->metadata_version[9]);
1194 else
1195 ping_monitor(devnm);
1196 }
1197 free_mdstat(ms);
1198 return rv;
1199 }
1200 free_mdstat(ms);
1201 rv = 0;
1202 mdstat_wait(5);
1203 }
1204 }
1205
1206 /* The state "broken" is used only for RAID0/LINEAR - it's the same as
1207 * "clean", but used in case the array has one or more members missing.
1208 */
1209 static char *clean_states[] = {
1210 "clear", "inactive", "readonly", "read-auto", "clean", "broken", NULL };
1211
1212 int WaitClean(char *dev, int verbose)
1213 {
1214 int fd;
1215 struct mdinfo *mdi;
1216 int rv = 1;
1217 char devnm[32];
1218
1219 if (!stat_is_blkdev(dev, NULL))
1220 return 2;
1221 fd = open(dev, O_RDONLY);
1222 if (fd < 0) {
1223 if (verbose)
1224 pr_err("Couldn't open %s: %s\n", dev, strerror(errno));
1225 return 1;
1226 }
1227
1228 strcpy(devnm, fd2devnm(fd));
1229 mdi = sysfs_read(fd, devnm, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1230 if (!mdi) {
1231 if (verbose)
1232 pr_err("Failed to read sysfs attributes for %s\n", dev);
1233 close(fd);
1234 return 0;
1235 }
1236
1237 switch(mdi->array.level) {
1238 case LEVEL_LINEAR:
1239 case LEVEL_MULTIPATH:
1240 case 0:
1241 /* safemode delay is irrelevant for these levels */
1242 rv = 0;
1243 }
1244
1245 /* for internal metadata the kernel handles the final clean
1246 * transition, containers can never be dirty
1247 */
1248 if (!is_subarray(mdi->text_version))
1249 rv = 0;
1250
1251 /* safemode disabled ? */
1252 if (mdi->safe_mode_delay == 0)
1253 rv = 0;
1254
1255 if (rv) {
1256 int state_fd = sysfs_open(fd2devnm(fd), NULL, "array_state");
1257 char buf[20];
1258 int delay = 5000;
1259
1260 /* minimize the safe_mode_delay and prepare to wait up to 5s
1261 * for writes to quiesce
1262 */
1263 sysfs_set_safemode(mdi, 1);
1264
1265 /* wait for array_state to be clean */
1266 while (1) {
1267 rv = read(state_fd, buf, sizeof(buf));
1268 if (rv < 0)
1269 break;
1270 if (sysfs_match_word(buf, clean_states) <
1271 (int)ARRAY_SIZE(clean_states) - 1)
1272 break;
1273 rv = sysfs_wait(state_fd, &delay);
1274 if (rv < 0 && errno != EINTR)
1275 break;
1276 lseek(state_fd, 0, SEEK_SET);
1277 }
1278 if (rv < 0)
1279 rv = 1;
1280 else if (ping_monitor(mdi->text_version) == 0) {
1281 /* we need to ping to close the window between array
1282 * state transitioning to clean and the metadata being
1283 * marked clean
1284 */
1285 rv = 0;
1286 } else {
1287 rv = 1;
1288 pr_err("Error connecting monitor with %s\n", dev);
1289 }
1290 if (rv && verbose)
1291 pr_err("Error waiting for %s to be clean\n", dev);
1292
1293 /* restore the original safe_mode_delay */
1294 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1295 close(state_fd);
1296 }
1297
1298 sysfs_free(mdi);
1299 close(fd);
1300
1301 return rv;
1302 }
Unnamed repository; edit this file 'description' to name the repository.