]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Monitor.c
projects / thirdparty / mdadm.git / blob
? search:


b5231d215da131c674d3b48dd09ac4e9b0e05065
[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 <signal.h>
30 #include <limits.h>
31 #include <syslog.h>
32
33 struct state {
34 char *devname;
35 char devnm[32]; /* to sync with mdstat info */
36 unsigned int utime;
37 int err;
38 char *spare_group;
39 int active, working, failed, spare, raid;
40 int from_config;
41 int from_auto;
42 int expected_spares;
43 int devstate[MAX_DISKS];
44 dev_t devid[MAX_DISKS];
45 int percent;
46 char parent_devnm[32]; /* For subarray, devnm of parent.
47 * For others, ""
48 */
49 struct supertype *metadata;
50 struct state *subarray;/* for a container it is a link to first subarray
51 * for a subarray it is a link to next subarray
52 * in the same container */
53 struct state *parent; /* for a subarray it is a link to its container
54 */
55 struct state *next;
56 };
57
58 struct alert_info {
59 char *mailaddr;
60 char *mailfrom;
61 char *alert_cmd;
62 int dosyslog;
63 };
64 static int make_daemon(char *pidfile);
65 static int check_one_sharer(int scan);
66 static void alert(char *event, char *dev, char *disc, struct alert_info *info);
67 static int check_array(struct state *st, struct mdstat_ent *mdstat,
68 int test, struct alert_info *info,
69 int increments, char *prefer);
70 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
71 int test, struct alert_info *info);
72 static void try_spare_migration(struct state *statelist, struct alert_info *info);
73 static void link_containers_with_subarrays(struct state *list);
74
75 int Monitor(struct mddev_dev *devlist,
76 char *mailaddr, char *alert_cmd,
77 struct context *c,
78 int daemonise, int oneshot,
79 int dosyslog, char *pidfile, int increments,
80 int share)
81 {
82 /*
83 * Every few seconds, scan every md device looking for changes
84 * When a change is found, log it, possibly run the alert command,
85 * and possibly send Email
86 *
87 * For each array, we record:
88 * Update time
89 * active/working/failed/spare drives
90 * State of each device.
91 * %rebuilt if rebuilding
92 *
93 * If the update time changes, check out all the data again
94 * It is possible that we cannot get the state of each device
95 * due to bugs in the md kernel module.
96 * We also read /proc/mdstat to get rebuild percent,
97 * and to get state on all active devices incase of kernel bug.
98 *
99 * Events are:
100 * Fail
101 * An active device had Faulty set or Active/Sync removed
102 * FailSpare
103 * A spare device had Faulty set
104 * SpareActive
105 * An active device had a reverse transition
106 * RebuildStarted
107 * percent went from -1 to +ve
108 * RebuildNN
109 * percent went from below to not-below NN%
110 * DeviceDisappeared
111 * Couldn't access a device which was previously visible
112 *
113 * if we detect an array with active<raid and spare==0
114 * we look at other arrays that have same spare-group
115 * If we find one with active==raid and spare>0,
116 * and if we can get_disk_info and find a name
117 * Then we hot-remove and hot-add to the other array
118 *
119 * If devlist is NULL, then we can monitor everything because --scan
120 * was given. We get an initial list from config file and add anything
121 * that appears in /proc/mdstat
122 */
123
124 struct state *statelist = NULL;
125 struct state *st2;
126 int finished = 0;
127 struct mdstat_ent *mdstat = NULL;
128 char *mailfrom = NULL;
129 struct alert_info info;
130
131 if (!mailaddr) {
132 mailaddr = conf_get_mailaddr();
133 if (mailaddr && ! c->scan)
134 pr_err("Monitor using email address \"%s\" from config file\n",
135 mailaddr);
136 }
137 mailfrom = conf_get_mailfrom();
138
139 if (!alert_cmd) {
140 alert_cmd = conf_get_program();
141 if (alert_cmd && ! c->scan)
142 pr_err("Monitor using program \"%s\" from config file\n",
143 alert_cmd);
144 }
145 if (c->scan && !mailaddr && !alert_cmd && !dosyslog) {
146 pr_err("No mail address or alert command - not monitoring.\n");
147 return 1;
148 }
149 info.alert_cmd = alert_cmd;
150 info.mailaddr = mailaddr;
151 info.mailfrom = mailfrom;
152 info.dosyslog = dosyslog;
153
154 if (daemonise) {
155 int rv = make_daemon(pidfile);
156 if (rv >= 0)
157 return rv;
158 }
159
160 if (share)
161 if (check_one_sharer(c->scan))
162 return 1;
163
164 if (devlist == NULL) {
165 struct mddev_ident *mdlist = conf_get_ident(NULL);
166 for (; mdlist; mdlist=mdlist->next) {
167 struct state *st;
168 if (mdlist->devname == NULL)
169 continue;
170 if (strcasecmp(mdlist->devname, "<ignore>") == 0)
171 continue;
172 st = xcalloc(1, sizeof *st);
173 if (mdlist->devname[0] == '/')
174 st->devname = xstrdup(mdlist->devname);
175 else {
176 st->devname = xmalloc(8+strlen(mdlist->devname)+1);
177 strcpy(strcpy(st->devname, "/dev/md/"),
178 mdlist->devname);
179 }
180 st->next = statelist;
181 st->devnm[0] = 0;
182 st->percent = RESYNC_UNKNOWN;
183 st->from_config = 1;
184 st->expected_spares = mdlist->spare_disks;
185 if (mdlist->spare_group)
186 st->spare_group = xstrdup(mdlist->spare_group);
187 statelist = st;
188 }
189 } else {
190 struct mddev_dev *dv;
191 for (dv=devlist ; dv; dv=dv->next) {
192 struct mddev_ident *mdlist = conf_get_ident(dv->devname);
193 struct state *st = xcalloc(1, sizeof *st);
194 st->devname = xstrdup(dv->devname);
195 st->next = statelist;
196 st->devnm[0] = 0;
197 st->percent = RESYNC_UNKNOWN;
198 st->expected_spares = -1;
199 if (mdlist) {
200 st->expected_spares = mdlist->spare_disks;
201 if (mdlist->spare_group)
202 st->spare_group = xstrdup(mdlist->spare_group);
203 }
204 statelist = st;
205 }
206 }
207
208 while (! finished) {
209 int new_found = 0;
210 struct state *st, **stp;
211 int anydegraded = 0;
212
213 if (mdstat)
214 free_mdstat(mdstat);
215 mdstat = mdstat_read(oneshot?0:1, 0);
216 if (!mdstat)
217 mdstat_close();
218
219 for (st=statelist; st; st=st->next)
220 if (check_array(st, mdstat, c->test, &info,
221 increments, c->prefer))
222 anydegraded = 1;
223
224 /* now check if there are any new devices found in mdstat */
225 if (c->scan)
226 new_found = add_new_arrays(mdstat, &statelist, c->test,
227 &info);
228
229 /* If an array has active < raid && spare == 0 && spare_group != NULL
230 * Look for another array with spare > 0 and active == raid and same spare_group
231 * if found, choose a device and hotremove/hotadd
232 */
233 if (share && anydegraded)
234 try_spare_migration(statelist, &info);
235 if (!new_found) {
236 if (oneshot)
237 break;
238 else
239 mdstat_wait(c->delay);
240 }
241 c->test = 0;
242
243 for (stp = &statelist; (st = *stp) != NULL; ) {
244 if (st->from_auto && st->err > 5) {
245 *stp = st->next;
246 free(st->devname);
247 free(st->spare_group);
248 free(st);
249 } else
250 stp = &st->next;
251 }
252 }
253 for (st2 = statelist; st2; st2 = statelist) {
254 statelist = st2->next;
255 free(st2);
256 }
257
258 if (pidfile)
259 unlink(pidfile);
260 return 0;
261 }
262
263 static int make_daemon(char *pidfile)
264 {
265 /* Return:
266 * -1 in the forked daemon
267 * 0 in the parent
268 * 1 on error
269 * so a none-negative becomes the exit code.
270 */
271 int pid = fork();
272 if (pid > 0) {
273 if (!pidfile)
274 printf("%d\n", pid);
275 else {
276 FILE *pid_file;
277 pid_file=fopen(pidfile, "w");
278 if (!pid_file)
279 perror("cannot create pid file");
280 else {
281 fprintf(pid_file,"%d\n", pid);
282 fclose(pid_file);
283 }
284 }
285 return 0;
286 }
287 if (pid < 0) {
288 perror("daemonise");
289 return 1;
290 }
291 close(0);
292 open("/dev/null", O_RDWR);
293 dup2(0,1);
294 dup2(0,2);
295 setsid();
296 return -1;
297 }
298
299 static int check_one_sharer(int scan)
300 {
301 int pid, rv;
302 FILE *fp;
303 char dir[20];
304 char path[100];
305 struct stat buf;
306 sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
307 fp = fopen(path, "r");
308 if (fp) {
309 if (fscanf(fp, "%d", &pid) != 1)
310 pid = -1;
311 sprintf(dir, "/proc/%d", pid);
312 rv = stat(dir, &buf);
313 if (rv != -1) {
314 if (scan) {
315 pr_err("Only one autorebuild process allowed in scan mode, aborting\n");
316 fclose(fp);
317 return 1;
318 } else {
319 pr_err("Warning: One autorebuild process already running.\n");
320 }
321 }
322 fclose(fp);
323 }
324 if (scan) {
325 if (mkdir(MDMON_DIR, S_IRWXU) < 0 &&
326 errno != EEXIST) {
327 pr_err("Can't create autorebuild.pid file\n");
328 } else {
329 fp = fopen(path, "w");
330 if (!fp)
331 pr_err("Cannot create autorebuild.pidfile\n");
332 else {
333 pid = getpid();
334 fprintf(fp, "%d\n", pid);
335 fclose(fp);
336 }
337 }
338 }
339 return 0;
340 }
341
342 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
343 {
344 int priority;
345
346 if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
347 time_t now = time(0);
348
349 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
350 }
351 if (info->alert_cmd) {
352 int pid = fork();
353 switch(pid) {
354 default:
355 waitpid(pid, NULL, 0);
356 break;
357 case -1:
358 break;
359 case 0:
360 execl(info->alert_cmd, info->alert_cmd,
361 event, dev, disc, NULL);
362 exit(2);
363 }
364 }
365 if (info->mailaddr &&
366 (strncmp(event, "Fail", 4)==0 ||
367 strncmp(event, "Test", 4)==0 ||
368 strncmp(event, "Spares", 6)==0 ||
369 strncmp(event, "Degrade", 7)==0)) {
370 FILE *mp = popen(Sendmail, "w");
371 if (mp) {
372 FILE *mdstat;
373 char hname[256];
374 gethostname(hname, sizeof(hname));
375 signal(SIGPIPE, SIG_IGN);
376 if (info->mailfrom)
377 fprintf(mp, "From: %s\n", info->mailfrom);
378 else
379 fprintf(mp, "From: %s monitoring <root>\n", Name);
380 fprintf(mp, "To: %s\n", info->mailaddr);
381 fprintf(mp, "Subject: %s event on %s:%s\n\n",
382 event, dev, hname);
383
384 fprintf(mp,
385 "This is an automatically generated mail message from %s\n", Name);
386 fprintf(mp, "running on %s\n\n", hname);
387
388 fprintf(mp,
389 "A %s event had been detected on md device %s.\n\n", event, dev);
390
391 if (disc && disc[0] != ' ')
392 fprintf(mp,
393 "It could be related to component device %s.\n\n", disc);
394 if (disc && disc[0] == ' ')
395 fprintf(mp, "Extra information:%s.\n\n", disc);
396
397 fprintf(mp, "Faithfully yours, etc.\n");
398
399 mdstat = fopen("/proc/mdstat", "r");
400 if (mdstat) {
401 char buf[8192];
402 int n;
403 fprintf(mp,
404 "\nP.S. The /proc/mdstat file currently contains the following:\n\n");
405 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
406 n=fwrite(buf, 1, n, mp);
407 fclose(mdstat);
408 }
409 pclose(mp);
410 }
411 }
412
413 /* log the event to syslog maybe */
414 if (info->dosyslog) {
415 /* Log at a different severity depending on the event.
416 *
417 * These are the critical events: */
418 if (strncmp(event, "Fail", 4)==0 ||
419 strncmp(event, "Degrade", 7)==0 ||
420 strncmp(event, "DeviceDisappeared", 17)==0)
421 priority = LOG_CRIT;
422 /* Good to know about, but are not failures: */
423 else if (strncmp(event, "Rebuild", 7)==0 ||
424 strncmp(event, "MoveSpare", 9)==0 ||
425 strncmp(event, "Spares", 6) != 0)
426 priority = LOG_WARNING;
427 /* Everything else: */
428 else
429 priority = LOG_INFO;
430
431 if (disc && disc[0] != ' ')
432 syslog(priority,
433 "%s event detected on md device %s, component device %s", event, dev, disc);
434 else if (disc)
435 syslog(priority,
436 "%s event detected on md device %s: %s",
437 event, dev, disc);
438 else
439 syslog(priority,
440 "%s event detected on md device %s",
441 event, dev);
442 }
443 }
444
445 static int check_array(struct state *st, struct mdstat_ent *mdstat,
446 int test, struct alert_info *ainfo,
447 int increments, char *prefer)
448 {
449 /* Update the state 'st' to reflect any changes shown in mdstat,
450 * or found by directly examining the array, and return
451 * '1' if the array is degraded, or '0' if it is optimal (or dead).
452 */
453 struct { int state, major, minor; } info[MAX_DISKS];
454 mdu_array_info_t array;
455 struct mdstat_ent *mse = NULL, *mse2;
456 char *dev = st->devname;
457 int fd;
458 int i;
459 int remaining_disks;
460 int last_disk;
461 int new_array = 0;
462
463 if (test)
464 alert("TestMessage", dev, NULL, ainfo);
465
466 fd = open(dev, O_RDONLY);
467 if (fd < 0) {
468 if (!st->err)
469 alert("DeviceDisappeared", dev, NULL, ainfo);
470 st->err++;
471 return 0;
472 }
473
474 if (!md_array_active(fd)) {
475 close(fd);
476 if (!st->err)
477 alert("DeviceDisappeared", dev, NULL, ainfo);
478 st->err++;
479 return 0;
480 }
481
482 fcntl(fd, F_SETFD, FD_CLOEXEC);
483 if (md_get_array_info(fd, &array) < 0) {
484 if (!st->err)
485 alert("DeviceDisappeared", dev, NULL, ainfo);
486 st->err++;
487 close(fd);
488 return 0;
489 }
490 /* It's much easier to list what array levels can't
491 * have a device disappear than all of them that can
492 */
493 if (array.level == 0 || array.level == -1) {
494 if (!st->err && !st->from_config)
495 alert("DeviceDisappeared", dev, " Wrong-Level", ainfo);
496 st->err++;
497 close(fd);
498 return 0;
499 }
500 if (st->devnm[0] == 0)
501 strcpy(st->devnm, fd2devnm(fd));
502
503 for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
504 if (strcmp(mse2->devnm, st->devnm) == 0) {
505 mse2->devnm[0] = 0; /* flag it as "used" */
506 mse = mse2;
507 }
508
509 if (!mse) {
510 /* duplicated array in statelist
511 * or re-created after reading mdstat*/
512 st->err++;
513 close(fd);
514 return 0;
515 }
516 /* this array is in /proc/mdstat */
517 if (array.utime == 0)
518 /* external arrays don't update utime, so
519 * just make sure it is always different. */
520 array.utime = st->utime + 1;;
521
522 if (st->err) {
523 /* New array appeared where previously had an error */
524 st->err = 0;
525 st->percent = RESYNC_NONE;
526 new_array = 1;
527 alert("NewArray", st->devname, NULL, ainfo);
528 }
529
530 if (st->utime == array.utime &&
531 st->failed == array.failed_disks &&
532 st->working == array.working_disks &&
533 st->spare == array.spare_disks &&
534 (mse == NULL || (
535 mse->percent == st->percent
536 ))) {
537 close(fd);
538 if ((st->active < st->raid) && st->spare == 0)
539 return 1;
540 else
541 return 0;
542 }
543 if (st->utime == 0 && /* new array */
544 mse->pattern && strchr(mse->pattern, '_') /* degraded */
545 )
546 alert("DegradedArray", dev, NULL, ainfo);
547
548 if (st->utime == 0 && /* new array */
549 st->expected_spares > 0 &&
550 array.spare_disks < st->expected_spares)
551 alert("SparesMissing", dev, NULL, ainfo);
552 if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
553 mse->percent >= 0)
554 alert("RebuildStarted", dev, NULL, ainfo);
555 if (st->percent >= 0 &&
556 mse->percent >= 0 &&
557 (mse->percent / increments) > (st->percent / increments)) {
558 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
559
560 if((mse->percent / increments) == 0)
561 snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
562 else
563 snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
564
565 alert(percentalert, dev, NULL, ainfo);
566 }
567
568 if (mse->percent == RESYNC_NONE &&
569 st->percent >= 0) {
570 /* Rebuild/sync/whatever just finished.
571 * If there is a number in /mismatch_cnt,
572 * we should report that.
573 */
574 struct mdinfo *sra =
575 sysfs_read(-1, st->devnm, GET_MISMATCH);
576 if (sra && sra->mismatch_cnt > 0) {
577 char cnt[80];
578 snprintf(cnt, sizeof(cnt),
579 " mismatches found: %d (on raid level %d)",
580 sra->mismatch_cnt, array.level);
581 alert("RebuildFinished", dev, cnt, ainfo);
582 } else
583 alert("RebuildFinished", dev, NULL, ainfo);
584 if (sra)
585 sysfs_free(sra);
586 }
587 st->percent = mse->percent;
588
589 remaining_disks = array.nr_disks;
590 for (i=0; i<MAX_DISKS && remaining_disks > 0;
591 i++) {
592 mdu_disk_info_t disc;
593 disc.number = i;
594 if (md_get_disk_info(fd, &disc) >= 0) {
595 info[i].state = disc.state;
596 info[i].major = disc.major;
597 info[i].minor = disc.minor;
598 if (disc.major || disc.minor)
599 remaining_disks --;
600 } else
601 info[i].major = info[i].minor = 0;
602 }
603 last_disk = i;
604
605 if (mse->metadata_version &&
606 strncmp(mse->metadata_version, "external:", 9) == 0 &&
607 is_subarray(mse->metadata_version+9)) {
608 char *sl;
609 strcpy(st->parent_devnm,
610 mse->metadata_version+10);
611 sl = strchr(st->parent_devnm, '/');
612 if (sl)
613 *sl = 0;
614 } else
615 st->parent_devnm[0] = 0;
616 if (st->metadata == NULL &&
617 st->parent_devnm[0] == 0)
618 st->metadata = super_by_fd(fd, NULL);
619
620 close(fd);
621
622 for (i=0; i<MAX_DISKS; i++) {
623 mdu_disk_info_t disc = {0,0,0,0,0};
624 int newstate=0;
625 int change;
626 char *dv = NULL;
627 disc.number = i;
628 if (i < last_disk &&
629 (info[i].major || info[i].minor)) {
630 newstate = info[i].state;
631 dv = map_dev_preferred(
632 info[i].major, info[i].minor, 1,
633 prefer);
634 disc.state = newstate;
635 disc.major = info[i].major;
636 disc.minor = info[i].minor;
637 } else
638 newstate = (1 << MD_DISK_REMOVED);
639
640 if (dv == NULL && st->devid[i])
641 dv = map_dev_preferred(
642 major(st->devid[i]),
643 minor(st->devid[i]), 1, prefer);
644 change = newstate ^ st->devstate[i];
645 if (st->utime && change && !st->err && !new_array) {
646 if ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))
647 alert("Fail", dev, dv, ainfo);
648 else if ((newstate & (1<<MD_DISK_FAULTY)) &&
649 (disc.major || disc.minor) &&
650 st->devid[i] == makedev(disc.major, disc.minor))
651 alert("FailSpare", dev, dv, ainfo);
652 else if ((newstate&change)&(1<<MD_DISK_SYNC))
653 alert("SpareActive", dev, dv, ainfo);
654 }
655 st->devstate[i] = newstate;
656 st->devid[i] = makedev(disc.major, disc.minor);
657 }
658 st->active = array.active_disks;
659 st->working = array.working_disks;
660 st->spare = array.spare_disks;
661 st->failed = array.failed_disks;
662 st->utime = array.utime;
663 st->raid = array.raid_disks;
664 st->err = 0;
665 if ((st->active < st->raid) && st->spare == 0)
666 return 1;
667 return 0;
668 }
669
670 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
671 int test, struct alert_info *info)
672 {
673 struct mdstat_ent *mse;
674 int new_found = 0;
675 char *name;
676
677 for (mse=mdstat; mse; mse=mse->next)
678 if (mse->devnm[0] &&
679 (!mse->level || /* retrieve containers */
680 (strcmp(mse->level, "raid0") != 0 &&
681 strcmp(mse->level, "linear") != 0))
682 ) {
683 struct state *st = xcalloc(1, sizeof *st);
684 mdu_array_info_t array;
685 int fd;
686
687 name = get_md_name(mse->devnm);
688 if (!name) {
689 free(st);
690 continue;
691 }
692
693 st->devname = xstrdup(name);
694 if ((fd = open(st->devname, O_RDONLY)) < 0 ||
695 md_get_array_info(fd, &array) < 0) {
696 /* no such array */
697 if (fd >= 0)
698 close(fd);
699 put_md_name(st->devname);
700 free(st->devname);
701 if (st->metadata) {
702 st->metadata->ss->free_super(st->metadata);
703 free(st->metadata);
704 }
705 free(st);
706 continue;
707 }
708 close(fd);
709 st->next = *statelist;
710 st->err = 1;
711 st->from_auto = 1;
712 strcpy(st->devnm, mse->devnm);
713 st->percent = RESYNC_UNKNOWN;
714 st->expected_spares = -1;
715 if (mse->metadata_version &&
716 strncmp(mse->metadata_version, "external:", 9) == 0 &&
717 is_subarray(mse->metadata_version+9)) {
718 char *sl;
719 strcpy(st->parent_devnm,
720 mse->metadata_version+10);
721 sl = strchr(st->parent_devnm, '/');
722 *sl = 0;
723 } else
724 st->parent_devnm[0] = 0;
725 *statelist = st;
726 if (test)
727 alert("TestMessage", st->devname, NULL, info);
728 new_found = 1;
729 }
730 return new_found;
731 }
732
733 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
734 {
735 int fd;
736
737 if (!st->metadata ||
738 !st->metadata->ss->min_acceptable_spare_size) {
739 *sizep = 0;
740 return 0;
741 }
742
743 fd = open(st->devname, O_RDONLY);
744 if (fd < 0)
745 return 1;
746 if (st->metadata->ss->external)
747 st->metadata->ss->load_container(st->metadata, fd, st->devname);
748 else
749 st->metadata->ss->load_super(st->metadata, fd, st->devname);
750 close(fd);
751 if (!st->metadata->sb)
752 return 1;
753 *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
754 st->metadata->ss->free_super(st->metadata);
755
756 return 0;
757 }
758
759 static int check_donor(struct state *from, struct state *to)
760 {
761 struct state *sub;
762
763 if (from == to)
764 return 0;
765 if (from->parent)
766 /* Cannot move from a member */
767 return 0;
768 if (from->err)
769 return 0;
770 for (sub = from->subarray; sub; sub = sub->subarray)
771 /* If source array has degraded subarrays, don't
772 * remove anything
773 */
774 if (sub->active < sub->raid)
775 return 0;
776 if (from->metadata->ss->external == 0)
777 if (from->active < from->raid)
778 return 0;
779 if (from->spare <= 0)
780 return 0;
781 return 1;
782 }
783
784 static dev_t choose_spare(struct state *from, struct state *to,
785 struct domainlist *domlist, unsigned long long min_size)
786 {
787 int d;
788 dev_t dev = 0;
789
790 for (d = from->raid; !dev && d < MAX_DISKS; d++) {
791 if (from->devid[d] > 0 &&
792 from->devstate[d] == 0) {
793 struct dev_policy *pol;
794 unsigned long long dev_size;
795
796 if (to->metadata->ss->external &&
797 test_partition_from_id(from->devid[d]))
798 continue;
799
800 if (min_size &&
801 dev_size_from_id(from->devid[d], &dev_size) &&
802 dev_size < min_size)
803 continue;
804
805 pol = devid_policy(from->devid[d]);
806 if (from->spare_group)
807 pol_add(&pol, pol_domain,
808 from->spare_group, NULL);
809 if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
810 dev = from->devid[d];
811 dev_policy_free(pol);
812 }
813 }
814 return dev;
815 }
816
817 static dev_t container_choose_spare(struct state *from, struct state *to,
818 struct domainlist *domlist,
819 unsigned long long min_size, int active)
820 {
821 /* This is similar to choose_spare, but we cannot trust devstate,
822 * so we need to read the metadata instead
823 */
824 struct mdinfo *list;
825 struct supertype *st = from->metadata;
826 int fd = open(from->devname, O_RDONLY);
827 int err;
828 dev_t dev = 0;
829
830 if (fd < 0)
831 return 0;
832 if (!st->ss->getinfo_super_disks) {
833 close(fd);
834 return 0;
835 }
836
837 err = st->ss->load_container(st, fd, NULL);
838 close(fd);
839 if (err)
840 return 0;
841
842 if (from == to) {
843 /* We must check if number of active disks has not increased
844 * since ioctl in main loop. mdmon may have added spare
845 * to subarray. If so we do not need to look for more spares
846 * so return non zero value */
847 int active_cnt = 0;
848 struct mdinfo *dp;
849 list = st->ss->getinfo_super_disks(st);
850 if (!list) {
851 st->ss->free_super(st);
852 return 1;
853 }
854 dp = list->devs;
855 while (dp) {
856 if (dp->disk.state & (1<<MD_DISK_SYNC) &&
857 !(dp->disk.state & (1<<MD_DISK_FAULTY)))
858 active_cnt++;
859 dp = dp->next;
860 }
861 sysfs_free(list);
862 if (active < active_cnt) {
863 /* Spare just activated.*/
864 st->ss->free_super(st);
865 return 1;
866 }
867 }
868
869 /* We only need one spare so full list not needed */
870 list = container_choose_spares(st, min_size, domlist, from->spare_group,
871 to->metadata->ss->name, 1);
872 if (list) {
873 struct mdinfo *disks = list->devs;
874 if (disks)
875 dev = makedev(disks->disk.major, disks->disk.minor);
876 sysfs_free(list);
877 }
878 st->ss->free_super(st);
879 return dev;
880 }
881
882 static void try_spare_migration(struct state *statelist, struct alert_info *info)
883 {
884 struct state *from;
885 struct state *st;
886
887 link_containers_with_subarrays(statelist);
888 for (st = statelist; st; st = st->next)
889 if (st->active < st->raid &&
890 st->spare == 0 && !st->err) {
891 struct domainlist *domlist = NULL;
892 int d;
893 struct state *to = st;
894 unsigned long long min_size;
895
896 if (to->parent_devnm[0] && !to->parent)
897 /* subarray monitored without parent container
898 * we can't move spares here */
899 continue;
900
901 if (to->parent)
902 /* member of a container */
903 to = to->parent;
904
905 if (get_min_spare_size_required(to, &min_size))
906 continue;
907 if (to->metadata->ss->external) {
908 /* We must make sure there is
909 * no suitable spare in container already.
910 * If there is we don't add more */
911 dev_t devid = container_choose_spare(
912 to, to, NULL, min_size, st->active);
913 if (devid > 0)
914 continue;
915 }
916 for (d = 0; d < MAX_DISKS; d++)
917 if (to->devid[d])
918 domainlist_add_dev(&domlist,
919 to->devid[d],
920 to->metadata->ss->name);
921 if (to->spare_group)
922 domain_add(&domlist, to->spare_group);
923 /*
924 * No spare migration if the destination
925 * has no domain. Skip this array.
926 */
927 if (!domlist)
928 continue;
929 for (from=statelist ; from ; from=from->next) {
930 dev_t devid;
931 if (!check_donor(from, to))
932 continue;
933 if (from->metadata->ss->external)
934 devid = container_choose_spare(
935 from, to, domlist, min_size, 0);
936 else
937 devid = choose_spare(from, to, domlist,
938 min_size);
939 if (devid > 0
940 && move_spare(from->devname, to->devname, devid)) {
941 alert("MoveSpare", to->devname, from->devname, info);
942 break;
943 }
944 }
945 domain_free(domlist);
946 }
947 }
948
949 /* search the statelist to connect external
950 * metadata subarrays with their containers
951 * We always completely rebuild the tree from scratch as
952 * that is safest considering the possibility of entries
953 * disappearing or changing.
954 */
955 static void link_containers_with_subarrays(struct state *list)
956 {
957 struct state *st;
958 struct state *cont;
959 for (st = list; st; st = st->next) {
960 st->parent = NULL;
961 st->subarray = NULL;
962 }
963 for (st = list; st; st = st->next)
964 if (st->parent_devnm[0])
965 for (cont = list; cont; cont = cont->next)
966 if (!cont->err &&
967 cont->parent_devnm[0] == 0 &&
968 strcmp(cont->devnm, st->parent_devnm) == 0) {
969 st->parent = cont;
970 st->subarray = cont->subarray;
971 cont->subarray = st;
972 break;
973 }
974 }
975
976 /* Not really Monitor but ... */
977 int Wait(char *dev)
978 {
979 char devnm[32];
980 int rv = 1;
981 int frozen_remaining = 3;
982
983 if (!stat_is_blkdev(dev, NULL))
984 return 2;
985 strcpy(devnm, dev);
986
987 while(1) {
988 struct mdstat_ent *ms = mdstat_read(1, 0);
989 struct mdstat_ent *e;
990
991 for (e=ms ; e; e=e->next)
992 if (strcmp(e->devnm, devnm) == 0)
993 break;
994
995 if (e && e->percent == RESYNC_NONE) {
996 /* We could be in the brief pause before something
997 * starts. /proc/mdstat doesn't show that, but
998 * sync_action does.
999 */
1000 struct mdinfo mdi;
1001 char buf[21];
1002
1003 if (sysfs_init(&mdi, -1, devnm))
1004 return 2;
1005 if (sysfs_get_str(&mdi, NULL, "sync_action",
1006 buf, 20) > 0 &&
1007 strcmp(buf,"idle\n") != 0) {
1008 e->percent = RESYNC_UNKNOWN;
1009 if (strcmp(buf, "frozen\n") == 0) {
1010 if (frozen_remaining == 0)
1011 e->percent = RESYNC_NONE;
1012 else
1013 frozen_remaining -= 1;
1014 }
1015 }
1016 }
1017 if (!e || e->percent == RESYNC_NONE) {
1018 if (e && e->metadata_version &&
1019 strncmp(e->metadata_version, "external:", 9) == 0) {
1020 if (is_subarray(&e->metadata_version[9]))
1021 ping_monitor(&e->metadata_version[9]);
1022 else
1023 ping_monitor(devnm);
1024 }
1025 free_mdstat(ms);
1026 return rv;
1027 }
1028 free_mdstat(ms);
1029 rv = 0;
1030 mdstat_wait(5);
1031 }
1032 }
1033
1034 static char *clean_states[] = {
1035 "clear", "inactive", "readonly", "read-auto", "clean", NULL };
1036
1037 int WaitClean(char *dev, int sock, int verbose)
1038 {
1039 int fd;
1040 struct mdinfo *mdi;
1041 int rv = 1;
1042 char devnm[32];
1043
1044 if (!stat_is_blkdev(dev, NULL))
1045 return 2;
1046 fd = open(dev, O_RDONLY);
1047 if (fd < 0) {
1048 if (verbose)
1049 pr_err("Couldn't open %s: %s\n", dev, strerror(errno));
1050 return 1;
1051 }
1052
1053 strcpy(devnm, fd2devnm(fd));
1054 mdi = sysfs_read(fd, devnm, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1055 if (!mdi) {
1056 if (verbose)
1057 pr_err("Failed to read sysfs attributes for %s\n", dev);
1058 close(fd);
1059 return 0;
1060 }
1061
1062 switch(mdi->array.level) {
1063 case LEVEL_LINEAR:
1064 case LEVEL_MULTIPATH:
1065 case 0:
1066 /* safemode delay is irrelevant for these levels */
1067 rv = 0;
1068 }
1069
1070 /* for internal metadata the kernel handles the final clean
1071 * transition, containers can never be dirty
1072 */
1073 if (!is_subarray(mdi->text_version))
1074 rv = 0;
1075
1076 /* safemode disabled ? */
1077 if (mdi->safe_mode_delay == 0)
1078 rv = 0;
1079
1080 if (rv) {
1081 int state_fd = sysfs_open(fd2devnm(fd), NULL, "array_state");
1082 char buf[20];
1083 int delay = 5000;
1084
1085 /* minimize the safe_mode_delay and prepare to wait up to 5s
1086 * for writes to quiesce
1087 */
1088 sysfs_set_safemode(mdi, 1);
1089
1090 /* wait for array_state to be clean */
1091 while (1) {
1092 rv = read(state_fd, buf, sizeof(buf));
1093 if (rv < 0)
1094 break;
1095 if (sysfs_match_word(buf, clean_states) <= 4)
1096 break;
1097 rv = sysfs_wait(state_fd, &delay);
1098 if (rv < 0 && errno != EINTR)
1099 break;
1100 lseek(state_fd, 0, SEEK_SET);
1101 }
1102 if (rv < 0)
1103 rv = 1;
1104 else if (fping_monitor(sock) == 0 ||
1105 ping_monitor(mdi->text_version) == 0) {
1106 /* we need to ping to close the window between array
1107 * state transitioning to clean and the metadata being
1108 * marked clean
1109 */
1110 rv = 0;
1111 } else
1112 rv = 1;
1113 if (rv && verbose)
1114 pr_err("Error waiting for %s to be clean\n",
1115 dev);
1116
1117 /* restore the original safe_mode_delay */
1118 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1119 close(state_fd);
1120 }
1121
1122 sysfs_free(mdi);
1123 close(fd);
1124
1125 return rv;
1126 }
Unnamed repository; edit this file 'description' to name the repository.