]> 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
Monitor: fix reporting for Fail vs FailSpare etc.
[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 int devnum; /* to sync with mdstat info */
36 long utime;
37 int err;
38 char *spare_group;
39 int active, working, failed, spare, raid;
40 int expected_spares;
41 int devstate[MAX_DISKS];
42 dev_t devid[MAX_DISKS];
43 int percent;
44 int parent_dev; /* For subarray, devnum of parent.
45 * For others, NoMdDev
46 */
47 struct supertype *metadata;
48 struct state *subarray;/* for a container it is a link to first subarray
49 * for a subarray it is a link to next subarray
50 * in the same container */
51 struct state *parent; /* for a subarray it is a link to its container
52 */
53 struct state *next;
54 };
55
56 struct alert_info {
57 char *mailaddr;
58 char *mailfrom;
59 char *alert_cmd;
60 int dosyslog;
61 };
62 static int make_daemon(char *pidfile);
63 static int check_one_sharer(int scan);
64 static void alert(char *event, char *dev, char *disc, struct alert_info *info);
65 static int check_array(struct state *st, struct mdstat_ent *mdstat,
66 int test, struct alert_info *info,
67 int increments, char *prefer);
68 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
69 int test, struct alert_info *info);
70 static void try_spare_migration(struct state *statelist, struct alert_info *info);
71 static void link_containers_with_subarrays(struct state *list);
72
73 int Monitor(struct mddev_dev *devlist,
74 char *mailaddr, char *alert_cmd,
75 int period, int daemonise, int scan, int oneshot,
76 int dosyslog, int test, char *pidfile, int increments,
77 int share, char *prefer)
78 {
79 /*
80 * Every few seconds, scan every md device looking for changes
81 * When a change is found, log it, possibly run the alert command,
82 * and possibly send Email
83 *
84 * For each array, we record:
85 * Update time
86 * active/working/failed/spare drives
87 * State of each device.
88 * %rebuilt if rebuilding
89 *
90 * If the update time changes, check out all the data again
91 * It is possible that we cannot get the state of each device
92 * due to bugs in the md kernel module.
93 * We also read /proc/mdstat to get rebuild percent,
94 * and to get state on all active devices incase of kernel bug.
95 *
96 * Events are:
97 * Fail
98 * An active device had Faulty set or Active/Sync removed
99 * FailSpare
100 * A spare device had Faulty set
101 * SpareActive
102 * An active device had a reverse transition
103 * RebuildStarted
104 * percent went from -1 to +ve
105 * RebuildNN
106 * percent went from below to not-below NN%
107 * DeviceDisappeared
108 * Couldn't access a device which was previously visible
109 *
110 * if we detect an array with active<raid and spare==0
111 * we look at other arrays that have same spare-group
112 * If we find one with active==raid and spare>0,
113 * and if we can get_disk_info and find a name
114 * Then we hot-remove and hot-add to the other array
115 *
116 * If devlist is NULL, then we can monitor everything because --scan
117 * was given. We get an initial list from config file and add anything
118 * that appears in /proc/mdstat
119 */
120
121 struct state *statelist = NULL;
122 struct state *st2;
123 int finished = 0;
124 struct mdstat_ent *mdstat = NULL;
125 char *mailfrom = NULL;
126 struct alert_info info;
127
128 if (!mailaddr) {
129 mailaddr = conf_get_mailaddr();
130 if (mailaddr && ! scan)
131 fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n",
132 mailaddr);
133 }
134 mailfrom = conf_get_mailfrom();
135
136 if (!alert_cmd) {
137 alert_cmd = conf_get_program();
138 if (alert_cmd && ! scan)
139 fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n",
140 alert_cmd);
141 }
142 if (scan && !mailaddr && !alert_cmd && !dosyslog) {
143 fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n");
144 return 1;
145 }
146 info.alert_cmd = alert_cmd;
147 info.mailaddr = mailaddr;
148 info.mailfrom = mailfrom;
149 info.dosyslog = dosyslog;
150
151 if (daemonise) {
152 int rv = make_daemon(pidfile);
153 if (rv >= 0)
154 return rv;
155 }
156
157 if (share)
158 if (check_one_sharer(scan))
159 return 1;
160
161 if (devlist == NULL) {
162 struct mddev_ident *mdlist = conf_get_ident(NULL);
163 for (; mdlist; mdlist=mdlist->next) {
164 struct state *st;
165 if (mdlist->devname == NULL)
166 continue;
167 if (strcasecmp(mdlist->devname, "<ignore>") == 0)
168 continue;
169 st = calloc(1, sizeof *st);
170 if (st == NULL)
171 continue;
172 if (mdlist->devname[0] == '/')
173 st->devname = strdup(mdlist->devname);
174 else {
175 st->devname = malloc(8+strlen(mdlist->devname)+1);
176 strcpy(strcpy(st->devname, "/dev/md/"),
177 mdlist->devname);
178 }
179 st->next = statelist;
180 st->devnum = INT_MAX;
181 st->percent = RESYNC_UNKNOWN;
182 st->expected_spares = mdlist->spare_disks;
183 if (mdlist->spare_group)
184 st->spare_group = strdup(mdlist->spare_group);
185 statelist = st;
186 }
187 } else {
188 struct mddev_dev *dv;
189 for (dv=devlist ; dv; dv=dv->next) {
190 struct mddev_ident *mdlist = conf_get_ident(dv->devname);
191 struct state *st = calloc(1, sizeof *st);
192 if (st == NULL)
193 continue;
194 st->devname = strdup(dv->devname);
195 st->next = statelist;
196 st->devnum = INT_MAX;
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 = strdup(mdlist->spare_group);
203 }
204 statelist = st;
205 }
206 }
207
208
209 while (! finished) {
210 int new_found = 0;
211 struct state *st;
212 int anydegraded = 0;
213
214 if (mdstat)
215 free_mdstat(mdstat);
216 mdstat = mdstat_read(oneshot?0:1, 0);
217
218 for (st=statelist; st; st=st->next)
219 if (check_array(st, mdstat, test, &info,
220 increments, prefer))
221 anydegraded = 1;
222
223 /* now check if there are any new devices found in mdstat */
224 if (scan)
225 new_found = add_new_arrays(mdstat, &statelist, test,
226 &info);
227
228 /* If an array has active < raid && spare == 0 && spare_group != NULL
229 * Look for another array with spare > 0 and active == raid and same spare_group
230 * if found, choose a device and hotremove/hotadd
231 */
232 if (share && anydegraded)
233 try_spare_migration(statelist, &info);
234 if (!new_found) {
235 if (oneshot)
236 break;
237 else
238 mdstat_wait(period);
239 }
240 test = 0;
241 }
242 for (st2 = statelist; st2; st2 = statelist) {
243 statelist = st2->next;
244 free(st2);
245 }
246
247 if (pidfile)
248 unlink(pidfile);
249 return 0;
250 }
251
252 static int make_daemon(char *pidfile)
253 {
254 /* Return:
255 * -1 in the forked daemon
256 * 0 in the parent
257 * 1 on error
258 * so a none-negative becomes the exit code.
259 */
260 int pid = fork();
261 if (pid > 0) {
262 if (!pidfile)
263 printf("%d\n", pid);
264 else {
265 FILE *pid_file;
266 pid_file=fopen(pidfile, "w");
267 if (!pid_file)
268 perror("cannot create pid file");
269 else {
270 fprintf(pid_file,"%d\n", pid);
271 fclose(pid_file);
272 }
273 }
274 return 0;
275 }
276 if (pid < 0) {
277 perror("daemonise");
278 return 1;
279 }
280 close(0);
281 open("/dev/null", O_RDWR);
282 dup2(0,1);
283 dup2(0,2);
284 setsid();
285 return -1;
286 }
287
288 static int check_one_sharer(int scan)
289 {
290 int pid, rv;
291 FILE *fp;
292 char dir[20];
293 char path[100];
294 struct stat buf;
295 sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
296 fp = fopen(path, "r");
297 if (fp) {
298 if (fscanf(fp, "%d", &pid) != 1)
299 pid = -1;
300 sprintf(dir, "/proc/%d", pid);
301 rv = stat(dir, &buf);
302 if (rv != -1) {
303 if (scan) {
304 fprintf(stderr, Name ": Only one "
305 "autorebuild process allowed"
306 " in scan mode, aborting\n");
307 fclose(fp);
308 return 1;
309 } else {
310 fprintf(stderr, Name ": Warning: One"
311 " autorebuild process already"
312 " running.\n");
313 }
314 }
315 fclose(fp);
316 }
317 if (scan) {
318 if (mkdir(MDMON_DIR, S_IRWXU) < 0 &&
319 errno != EEXIST) {
320 fprintf(stderr, Name ": Can't create "
321 "autorebuild.pid file\n");
322 } else {
323 fp = fopen(path, "w");
324 if (!fp)
325 fprintf(stderr, Name ": Cannot create"
326 " autorebuild.pid"
327 "file\n");
328 else {
329 pid = getpid();
330 fprintf(fp, "%d\n", pid);
331 fclose(fp);
332 }
333 }
334 }
335 return 0;
336 }
337
338 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
339 {
340 int priority;
341
342 if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
343 time_t now = time(0);
344
345 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
346 }
347 if (info->alert_cmd) {
348 int pid = fork();
349 switch(pid) {
350 default:
351 waitpid(pid, NULL, 0);
352 break;
353 case -1:
354 break;
355 case 0:
356 execl(info->alert_cmd, info->alert_cmd,
357 event, dev, disc, NULL);
358 exit(2);
359 }
360 }
361 if (info->mailaddr &&
362 (strncmp(event, "Fail", 4)==0 ||
363 strncmp(event, "Test", 4)==0 ||
364 strncmp(event, "Spares", 6)==0 ||
365 strncmp(event, "Degrade", 7)==0)) {
366 FILE *mp = popen(Sendmail, "w");
367 if (mp) {
368 FILE *mdstat;
369 char hname[256];
370 gethostname(hname, sizeof(hname));
371 signal(SIGPIPE, SIG_IGN);
372 if (info->mailfrom)
373 fprintf(mp, "From: %s\n", info->mailfrom);
374 else
375 fprintf(mp, "From: " Name " monitoring <root>\n");
376 fprintf(mp, "To: %s\n", info->mailaddr);
377 fprintf(mp, "Subject: %s event on %s:%s\n\n",
378 event, dev, hname);
379
380 fprintf(mp,
381 "This is an automatically generated"
382 " mail message from " Name "\n");
383 fprintf(mp, "running on %s\n\n", hname);
384
385 fprintf(mp,
386 "A %s event had been detected on"
387 " md device %s.\n\n", event, dev);
388
389 if (disc && disc[0] != ' ')
390 fprintf(mp,
391 "It could be related to"
392 " component device %s.\n\n", disc);
393 if (disc && disc[0] == ' ')
394 fprintf(mp, "Extra information:%s.\n\n", disc);
395
396 fprintf(mp, "Faithfully yours, etc.\n");
397
398 mdstat = fopen("/proc/mdstat", "r");
399 if (mdstat) {
400 char buf[8192];
401 int n;
402 fprintf(mp,
403 "\nP.S. The /proc/mdstat file"
404 " 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)
432 syslog(priority,
433 "%s event detected on md device %s,"
434 " component device %s", event, dev, disc);
435 else
436 syslog(priority,
437 "%s event detected on md device %s",
438 event, dev);
439 }
440 }
441
442 static int check_array(struct state *st, struct mdstat_ent *mdstat,
443 int test, struct alert_info *ainfo,
444 int increments, char *prefer)
445 {
446 /* Update the state 'st' to reflect any changes shown in mdstat,
447 * or found by directly examining the array, and return
448 * '1' if the array is degraded, or '0' if it is optimal (or dead).
449 */
450 struct { int state, major, minor; } info[MAX_DISKS];
451 mdu_array_info_t array;
452 struct mdstat_ent *mse = NULL, *mse2;
453 char *dev = st->devname;
454 int fd;
455 int i;
456 int remaining_disks;
457 int last_disk;
458 int new_array = 0;
459
460 if (test)
461 alert("TestMessage", dev, NULL, ainfo);
462 fd = open(dev, O_RDONLY);
463 if (fd < 0) {
464 if (!st->err)
465 alert("DeviceDisappeared", dev, NULL, ainfo);
466 st->err=1;
467 return 0;
468 }
469 fcntl(fd, F_SETFD, FD_CLOEXEC);
470 if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
471 if (!st->err)
472 alert("DeviceDisappeared", dev, NULL, ainfo);
473 st->err=1;
474 close(fd);
475 return 0;
476 }
477 /* It's much easier to list what array levels can't
478 * have a device disappear than all of them that can
479 */
480 if (array.level == 0 || array.level == -1) {
481 if (!st->err)
482 alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
483 st->err = 1;
484 close(fd);
485 return 0;
486 }
487 if (st->devnum == INT_MAX) {
488 struct stat stb;
489 if (fstat(fd, &stb) == 0 &&
490 (S_IFMT&stb.st_mode)==S_IFBLK) {
491 if (major(stb.st_rdev) == MD_MAJOR)
492 st->devnum = minor(stb.st_rdev);
493 else
494 st->devnum = -1- (minor(stb.st_rdev)>>6);
495 }
496 }
497
498 for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
499 if (mse2->devnum == st->devnum) {
500 mse2->devnum = INT_MAX; /* flag it as "used" */
501 mse = mse2;
502 }
503
504 if (!mse) {
505 /* duplicated array in statelist
506 * or re-created after reading mdstat*/
507 st->err = 1;
508 close(fd);
509 return 0;
510 }
511 /* this array is in /proc/mdstat */
512 if (array.utime == 0)
513 /* external arrays don't update utime, so
514 * just make sure it is always different. */
515 array.utime = st->utime + 1;;
516
517 if (st->err) {
518 /* New array appeared where previously had and error */
519 st->err = 0;
520 st->percent = RESYNC_NONE;
521 new_array = 1;
522 alert("NewArray", st->devname, NULL, ainfo);
523 }
524
525 if (st->utime == array.utime &&
526 st->failed == array.failed_disks &&
527 st->working == array.working_disks &&
528 st->spare == array.spare_disks &&
529 (mse == NULL || (
530 mse->percent == st->percent
531 ))) {
532 close(fd);
533 if ((st->active < st->raid) && st->spare == 0)
534 return 1;
535 else
536 return 0;
537 }
538 if (st->utime == 0 && /* new array */
539 mse->pattern && strchr(mse->pattern, '_') /* degraded */
540 )
541 alert("DegradedArray", dev, NULL, ainfo);
542
543 if (st->utime == 0 && /* new array */
544 st->expected_spares > 0 &&
545 array.spare_disks < st->expected_spares)
546 alert("SparesMissing", dev, NULL, ainfo);
547 if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
548 mse->percent >= 0)
549 alert("RebuildStarted", dev, NULL, ainfo);
550 if (st->percent >= 0 &&
551 mse->percent >= 0 &&
552 (mse->percent / increments) > (st->percent / increments)) {
553 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
554
555 if((mse->percent / increments) == 0)
556 snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
557 else
558 snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
559
560 alert(percentalert, dev, NULL, ainfo);
561 }
562
563 if (mse->percent == RESYNC_NONE &&
564 st->percent >= 0) {
565 /* Rebuild/sync/whatever just finished.
566 * If there is a number in /mismatch_cnt,
567 * we should report that.
568 */
569 struct mdinfo *sra =
570 sysfs_read(-1, st->devnum, GET_MISMATCH);
571 if (sra && sra->mismatch_cnt > 0) {
572 char cnt[80];
573 snprintf(cnt, sizeof(cnt),
574 " mismatches found: %d (on raid level %d)",
575 sra->mismatch_cnt, array.level);
576 alert("RebuildFinished", dev, cnt, ainfo);
577 } else
578 alert("RebuildFinished", dev, NULL, ainfo);
579 if (sra)
580 free(sra);
581 }
582 st->percent = mse->percent;
583
584 remaining_disks = array.nr_disks;
585 for (i=0; i<MAX_DISKS && remaining_disks > 0;
586 i++) {
587 mdu_disk_info_t disc;
588 disc.number = i;
589 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
590 info[i].state = disc.state;
591 info[i].major = disc.major;
592 info[i].minor = disc.minor;
593 if (disc.major || disc.minor)
594 remaining_disks --;
595 } else
596 info[i].major = info[i].minor = 0;
597 }
598 last_disk = i;
599
600 if (mse->metadata_version &&
601 strncmp(mse->metadata_version, "external:", 9) == 0 &&
602 is_subarray(mse->metadata_version+9))
603 st->parent_dev =
604 devname2devnum(mse->metadata_version+10);
605 else
606 st->parent_dev = NoMdDev;
607 if (st->metadata == NULL &&
608 st->parent_dev == NoMdDev)
609 st->metadata = super_by_fd(fd, NULL);
610
611 close(fd);
612
613 for (i=0; i<MAX_DISKS; i++) {
614 mdu_disk_info_t disc = {0,0,0,0,0};
615 int newstate=0;
616 int change;
617 char *dv = NULL;
618 disc.number = i;
619 if (i < last_disk &&
620 (info[i].major || info[i].minor)) {
621 newstate = info[i].state;
622 dv = map_dev_preferred(
623 info[i].major, info[i].minor, 1,
624 prefer);
625 disc.state = newstate;
626 disc.major = info[i].major;
627 disc.minor = info[i].minor;
628 } else
629 newstate = (1 << MD_DISK_REMOVED);
630
631 if (dv == NULL && st->devid[i])
632 dv = map_dev_preferred(
633 major(st->devid[i]),
634 minor(st->devid[i]), 1, prefer);
635 change = newstate ^ st->devstate[i];
636 if (st->utime && change && !st->err && !new_array) {
637 if ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))
638 alert("Fail", dev, dv, ainfo);
639 else if ((newstate & (1<<MD_DISK_FAULTY)) &&
640 (disc.major || disc.minor) &&
641 st->devid[i] == makedev(disc.major, disc.minor))
642 alert("FailSpare", dev, dv, ainfo);
643 else if ((newstate&change)&(1<<MD_DISK_SYNC))
644 alert("SpareActive", dev, dv, ainfo);
645 }
646 st->devstate[i] = newstate;
647 st->devid[i] = makedev(disc.major, disc.minor);
648 }
649 st->active = array.active_disks;
650 st->working = array.working_disks;
651 st->spare = array.spare_disks;
652 st->failed = array.failed_disks;
653 st->utime = array.utime;
654 st->raid = array.raid_disks;
655 st->err = 0;
656 if ((st->active < st->raid) && st->spare == 0)
657 return 1;
658 return 0;
659 }
660
661 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
662 int test, struct alert_info *info)
663 {
664 struct mdstat_ent *mse;
665 int new_found = 0;
666
667 for (mse=mdstat; mse; mse=mse->next)
668 if (mse->devnum != INT_MAX &&
669 (!mse->level || /* retrieve containers */
670 (strcmp(mse->level, "raid0") != 0 &&
671 strcmp(mse->level, "linear") != 0))
672 ) {
673 struct state *st = calloc(1, sizeof *st);
674 mdu_array_info_t array;
675 int fd;
676 if (st == NULL)
677 continue;
678 st->devname = strdup(get_md_name(mse->devnum));
679 if ((fd = open(st->devname, O_RDONLY)) < 0 ||
680 ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
681 /* no such array */
682 if (fd >=0) close(fd);
683 put_md_name(st->devname);
684 free(st->devname);
685 if (st->metadata) {
686 st->metadata->ss->free_super(st->metadata);
687 free(st->metadata);
688 }
689 free(st);
690 continue;
691 }
692 close(fd);
693 st->next = *statelist;
694 st->err = 1;
695 st->devnum = mse->devnum;
696 st->percent = RESYNC_UNKNOWN;
697 st->expected_spares = -1;
698 if (mse->metadata_version &&
699 strncmp(mse->metadata_version, "external:", 9) == 0 &&
700 is_subarray(mse->metadata_version+9))
701 st->parent_dev =
702 devname2devnum(mse->metadata_version+10);
703 else
704 st->parent_dev = NoMdDev;
705 *statelist = st;
706 if (test)
707 alert("TestMessage", st->devname, NULL, info);
708 new_found = 1;
709 }
710 return new_found;
711 }
712
713 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
714 {
715 int fd;
716
717 if (!st->metadata ||
718 !st->metadata->ss->min_acceptable_spare_size) {
719 *sizep = 0;
720 return 0;
721 }
722
723 fd = open(st->devname, O_RDONLY);
724 if (fd < 0)
725 return 1;
726 if (st->metadata->ss->external)
727 st->metadata->ss->load_container(st->metadata, fd, st->devname);
728 else
729 st->metadata->ss->load_super(st->metadata, fd, st->devname);
730 close(fd);
731 if (!st->metadata->sb)
732 return 1;
733 *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
734 st->metadata->ss->free_super(st->metadata);
735
736 return 0;
737 }
738
739 static int check_donor(struct state *from, struct state *to)
740 {
741 struct state *sub;
742
743 if (from == to)
744 return 0;
745 if (from->parent)
746 /* Cannot move from a member */
747 return 0;
748 if (from->err)
749 return 0;
750 for (sub = from->subarray; sub; sub = sub->subarray)
751 /* If source array has degraded subarrays, don't
752 * remove anything
753 */
754 if (sub->active < sub->raid)
755 return 0;
756 if (from->metadata->ss->external == 0)
757 if (from->active < from->raid)
758 return 0;
759 if (from->spare <= 0)
760 return 0;
761 return 1;
762 }
763
764 static dev_t choose_spare(struct state *from, struct state *to,
765 struct domainlist *domlist, unsigned long long min_size)
766 {
767 int d;
768 dev_t dev = 0;
769
770 for (d = from->raid; !dev && d < MAX_DISKS; d++) {
771 if (from->devid[d] > 0 &&
772 from->devstate[d] == 0) {
773 struct dev_policy *pol;
774 unsigned long long dev_size;
775
776 if (to->metadata->ss->external &&
777 test_partition_from_id(from->devid[d]))
778 continue;
779
780 if (min_size &&
781 dev_size_from_id(from->devid[d], &dev_size) &&
782 dev_size < min_size)
783 continue;
784
785 pol = devnum_policy(from->devid[d]);
786 if (from->spare_group)
787 pol_add(&pol, pol_domain,
788 from->spare_group, NULL);
789 if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
790 dev = from->devid[d];
791 dev_policy_free(pol);
792 }
793 }
794 return dev;
795 }
796
797 static dev_t container_choose_spare(struct state *from, struct state *to,
798 struct domainlist *domlist,
799 unsigned long long min_size, int active)
800 {
801 /* This is similar to choose_spare, but we cannot trust devstate,
802 * so we need to read the metadata instead
803 */
804 struct mdinfo *list;
805 struct supertype *st = from->metadata;
806 int fd = open(from->devname, O_RDONLY);
807 int err;
808 dev_t dev = 0;
809
810 if (fd < 0)
811 return 0;
812 if (!st->ss->getinfo_super_disks) {
813 close(fd);
814 return 0;
815 }
816
817 err = st->ss->load_container(st, fd, NULL);
818 close(fd);
819 if (err)
820 return 0;
821
822 if (from == to) {
823 /* We must check if number of active disks has not increased
824 * since ioctl in main loop. mdmon may have added spare
825 * to subarray. If so we do not need to look for more spares
826 * so return non zero value */
827 int active_cnt = 0;
828 struct mdinfo *dp;
829 list = st->ss->getinfo_super_disks(st);
830 if (!list) {
831 st->ss->free_super(st);
832 return 1;
833 }
834 dp = list->devs;
835 while (dp) {
836 if (dp->disk.state & (1<<MD_DISK_SYNC) &&
837 !(dp->disk.state & (1<<MD_DISK_FAULTY)))
838 active_cnt++;
839 dp = dp->next;
840 }
841 sysfs_free(list);
842 if (active < active_cnt) {
843 /* Spare just activated.*/
844 st->ss->free_super(st);
845 return 1;
846 }
847 }
848
849 /* We only need one spare so full list not needed */
850 list = container_choose_spares(st, min_size, domlist, from->spare_group,
851 to->metadata->ss->name, 1);
852 if (list) {
853 struct mdinfo *disks = list->devs;
854 if (disks)
855 dev = makedev(disks->disk.major, disks->disk.minor);
856 sysfs_free(list);
857 }
858 st->ss->free_super(st);
859 return dev;
860 }
861
862
863 static void try_spare_migration(struct state *statelist, struct alert_info *info)
864 {
865 struct state *from;
866 struct state *st;
867
868 link_containers_with_subarrays(statelist);
869 for (st = statelist; st; st = st->next)
870 if (st->active < st->raid &&
871 st->spare == 0 && !st->err) {
872 struct domainlist *domlist = NULL;
873 int d;
874 struct state *to = st;
875 unsigned long long min_size;
876
877 if (to->parent_dev != NoMdDev && !to->parent)
878 /* subarray monitored without parent container
879 * we can't move spares here */
880 continue;
881
882 if (to->parent)
883 /* member of a container */
884 to = to->parent;
885
886 if (get_min_spare_size_required(to, &min_size))
887 continue;
888 if (to->metadata->ss->external) {
889 /* We must make sure there is
890 * no suitable spare in container already.
891 * If there is we don't add more */
892 dev_t devid = container_choose_spare(
893 to, to, NULL, min_size, st->active);
894 if (devid > 0)
895 continue;
896 }
897 for (d = 0; d < MAX_DISKS; d++)
898 if (to->devid[d])
899 domainlist_add_dev(&domlist,
900 to->devid[d],
901 to->metadata->ss->name);
902 if (to->spare_group)
903 domain_add(&domlist, to->spare_group);
904 /*
905 * No spare migration if the destination
906 * has no domain. Skip this array.
907 */
908 if (!domlist)
909 continue;
910 for (from=statelist ; from ; from=from->next) {
911 dev_t devid;
912 if (!check_donor(from, to))
913 continue;
914 if (from->metadata->ss->external)
915 devid = container_choose_spare(
916 from, to, domlist, min_size, 0);
917 else
918 devid = choose_spare(from, to, domlist,
919 min_size);
920 if (devid > 0
921 && move_spare(from->devname, to->devname, devid)) {
922 alert("MoveSpare", to->devname, from->devname, info);
923 break;
924 }
925 }
926 domain_free(domlist);
927 }
928 }
929
930 /* search the statelist to connect external
931 * metadata subarrays with their containers
932 * We always completely rebuild the tree from scratch as
933 * that is safest considering the possibility of entries
934 * disappearing or changing.
935 */
936 static void link_containers_with_subarrays(struct state *list)
937 {
938 struct state *st;
939 struct state *cont;
940 for (st = list; st; st = st->next) {
941 st->parent = NULL;
942 st->subarray = NULL;
943 }
944 for (st = list; st; st = st->next)
945 if (st->parent_dev != NoMdDev)
946 for (cont = list; cont; cont = cont->next)
947 if (!cont->err &&
948 cont->parent_dev == NoMdDev &&
949 cont->devnum == st->parent_dev) {
950 st->parent = cont;
951 st->subarray = cont->subarray;
952 cont->subarray = st;
953 break;
954 }
955 }
956
957 /* Not really Monitor but ... */
958 int Wait(char *dev)
959 {
960 struct stat stb;
961 int devnum;
962 int rv = 1;
963
964 if (stat(dev, &stb) != 0) {
965 fprintf(stderr, Name ": Cannot find %s: %s\n", dev,
966 strerror(errno));
967 return 2;
968 }
969 devnum = stat2devnum(&stb);
970
971 while(1) {
972 struct mdstat_ent *ms = mdstat_read(1, 0);
973 struct mdstat_ent *e;
974
975 for (e=ms ; e; e=e->next)
976 if (e->devnum == devnum)
977 break;
978
979 if (!e || e->percent < 0) {
980 if (e && e->metadata_version &&
981 strncmp(e->metadata_version, "external:", 9) == 0) {
982 if (is_subarray(&e->metadata_version[9]))
983 ping_monitor(&e->metadata_version[9]);
984 else
985 ping_monitor_by_id(devnum);
986 }
987 free_mdstat(ms);
988 return rv;
989 }
990 free_mdstat(ms);
991 rv = 0;
992 mdstat_wait(5);
993 }
994 }
995
996 #ifndef MDASSEMBLE
997
998 static char *clean_states[] = {
999 "clear", "inactive", "readonly", "read-auto", "clean", NULL };
1000
1001 int WaitClean(char *dev, int sock, int verbose)
1002 {
1003 int fd;
1004 struct mdinfo *mdi;
1005 int rv = 1;
1006 int devnum;
1007
1008 fd = open(dev, O_RDONLY);
1009 if (fd < 0) {
1010 if (verbose)
1011 fprintf(stderr, Name ": Couldn't open %s: %s\n", dev, strerror(errno));
1012 return 1;
1013 }
1014
1015 devnum = fd2devnum(fd);
1016 mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1017 if (!mdi) {
1018 if (verbose)
1019 fprintf(stderr, Name ": Failed to read sysfs attributes for "
1020 "%s\n", dev);
1021 close(fd);
1022 return 0;
1023 }
1024
1025 switch(mdi->array.level) {
1026 case LEVEL_LINEAR:
1027 case LEVEL_MULTIPATH:
1028 case 0:
1029 /* safemode delay is irrelevant for these levels */
1030 rv = 0;
1031 }
1032
1033 /* for internal metadata the kernel handles the final clean
1034 * transition, containers can never be dirty
1035 */
1036 if (!is_subarray(mdi->text_version))
1037 rv = 0;
1038
1039 /* safemode disabled ? */
1040 if (mdi->safe_mode_delay == 0)
1041 rv = 0;
1042
1043 if (rv) {
1044 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
1045 char buf[20];
1046 fd_set fds;
1047 struct timeval tm;
1048
1049 /* minimize the safe_mode_delay and prepare to wait up to 5s
1050 * for writes to quiesce
1051 */
1052 sysfs_set_safemode(mdi, 1);
1053 tm.tv_sec = 5;
1054 tm.tv_usec = 0;
1055
1056 FD_ZERO(&fds);
1057
1058 /* wait for array_state to be clean */
1059 while (1) {
1060 rv = read(state_fd, buf, sizeof(buf));
1061 if (rv < 0)
1062 break;
1063 if (sysfs_match_word(buf, clean_states) <= 4)
1064 break;
1065 FD_SET(state_fd, &fds);
1066 rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
1067 if (rv < 0 && errno != EINTR)
1068 break;
1069 lseek(state_fd, 0, SEEK_SET);
1070 }
1071 if (rv < 0)
1072 rv = 1;
1073 else if (fping_monitor(sock) == 0 ||
1074 ping_monitor(mdi->text_version) == 0) {
1075 /* we need to ping to close the window between array
1076 * state transitioning to clean and the metadata being
1077 * marked clean
1078 */
1079 rv = 0;
1080 } else
1081 rv = 1;
1082 if (rv && verbose)
1083 fprintf(stderr, Name ": Error waiting for %s to be clean\n",
1084 dev);
1085
1086 /* restore the original safe_mode_delay */
1087 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1088 close(state_fd);
1089 }
1090
1091 sysfs_free(mdi);
1092 close(fd);
1093
1094 return rv;
1095 }
1096 #endif /* MDASSEMBLE */
Unnamed repository; edit this file 'description' to name the repository.