]> 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: Report NewArray when an array the disappeared, reappears.
[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
459 if (test)
460 alert("TestMessage", dev, NULL, ainfo);
461 fd = open(dev, O_RDONLY);
462 if (fd < 0) {
463 if (!st->err)
464 alert("DeviceDisappeared", dev, NULL, ainfo);
465 st->err=1;
466 return 0;
467 }
468 fcntl(fd, F_SETFD, FD_CLOEXEC);
469 if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
470 if (!st->err)
471 alert("DeviceDisappeared", dev, NULL, ainfo);
472 st->err=1;
473 close(fd);
474 return 0;
475 }
476 /* It's much easier to list what array levels can't
477 * have a device disappear than all of them that can
478 */
479 if (array.level == 0 || array.level == -1) {
480 if (!st->err)
481 alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
482 st->err = 1;
483 close(fd);
484 return 0;
485 }
486 if (st->devnum == INT_MAX) {
487 struct stat stb;
488 if (fstat(fd, &stb) == 0 &&
489 (S_IFMT&stb.st_mode)==S_IFBLK) {
490 if (major(stb.st_rdev) == MD_MAJOR)
491 st->devnum = minor(stb.st_rdev);
492 else
493 st->devnum = -1- (minor(stb.st_rdev)>>6);
494 }
495 }
496
497 for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
498 if (mse2->devnum == st->devnum) {
499 mse2->devnum = INT_MAX; /* flag it as "used" */
500 mse = mse2;
501 }
502
503 if (!mse) {
504 /* duplicated array in statelist
505 * or re-created after reading mdstat*/
506 st->err = 1;
507 close(fd);
508 return 0;
509 }
510 /* this array is in /proc/mdstat */
511 if (array.utime == 0)
512 /* external arrays don't update utime, so
513 * just make sure it is always different. */
514 array.utime = st->utime + 1;;
515
516 if (st->err) {
517 /* New array appeared where previously had and error */
518 st->err = 0;
519 st->percent = RESYNC_NONE;
520 alert("NewArray", st->devname, NULL, ainfo);
521 }
522
523 if (st->utime == array.utime &&
524 st->failed == array.failed_disks &&
525 st->working == array.working_disks &&
526 st->spare == array.spare_disks &&
527 (mse == NULL || (
528 mse->percent == st->percent
529 ))) {
530 close(fd);
531 if ((st->active < st->raid) && st->spare == 0)
532 return 1;
533 else
534 return 0;
535 }
536 if (st->utime == 0 && /* new array */
537 mse->pattern && strchr(mse->pattern, '_') /* degraded */
538 )
539 alert("DegradedArray", dev, NULL, ainfo);
540
541 if (st->utime == 0 && /* new array */
542 st->expected_spares > 0 &&
543 array.spare_disks < st->expected_spares)
544 alert("SparesMissing", dev, NULL, ainfo);
545 if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
546 mse->percent >= 0)
547 alert("RebuildStarted", dev, NULL, ainfo);
548 if (st->percent >= 0 &&
549 mse->percent >= 0 &&
550 (mse->percent / increments) > (st->percent / increments)) {
551 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
552
553 if((mse->percent / increments) == 0)
554 snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
555 else
556 snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
557
558 alert(percentalert, dev, NULL, ainfo);
559 }
560
561 if (mse->percent == RESYNC_NONE &&
562 st->percent >= 0) {
563 /* Rebuild/sync/whatever just finished.
564 * If there is a number in /mismatch_cnt,
565 * we should report that.
566 */
567 struct mdinfo *sra =
568 sysfs_read(-1, st->devnum, GET_MISMATCH);
569 if (sra && sra->mismatch_cnt > 0) {
570 char cnt[80];
571 snprintf(cnt, sizeof(cnt),
572 " mismatches found: %d (on raid level %d)",
573 sra->mismatch_cnt, array.level);
574 alert("RebuildFinished", dev, cnt, ainfo);
575 } else
576 alert("RebuildFinished", dev, NULL, ainfo);
577 if (sra)
578 free(sra);
579 }
580 st->percent = mse->percent;
581
582 remaining_disks = array.nr_disks;
583 for (i=0; i<MAX_DISKS && remaining_disks > 0;
584 i++) {
585 mdu_disk_info_t disc;
586 disc.number = i;
587 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
588 info[i].state = disc.state;
589 info[i].major = disc.major;
590 info[i].minor = disc.minor;
591 if (disc.major || disc.minor)
592 remaining_disks --;
593 } else
594 info[i].major = info[i].minor = 0;
595 }
596 last_disk = i;
597
598 if (mse->metadata_version &&
599 strncmp(mse->metadata_version, "external:", 9) == 0 &&
600 is_subarray(mse->metadata_version+9))
601 st->parent_dev =
602 devname2devnum(mse->metadata_version+10);
603 else
604 st->parent_dev = NoMdDev;
605 if (st->metadata == NULL &&
606 st->parent_dev == NoMdDev)
607 st->metadata = super_by_fd(fd, NULL);
608
609 close(fd);
610
611 for (i=0; i<MAX_DISKS; i++) {
612 mdu_disk_info_t disc = {0,0,0,0,0};
613 int newstate=0;
614 int change;
615 char *dv = NULL;
616 disc.number = i;
617 if (i >= last_disk) {
618 newstate = 0;
619 disc.major = disc.minor = 0;
620 } else if (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 if (mse && mse->pattern && i < (int)strlen(mse->pattern)) {
629 switch(mse->pattern[i]) {
630 case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
631 case '_': newstate = 8 /* REMOVED */; break;
632 }
633 disc.major = disc.minor = 0;
634 }
635 if (dv == NULL && st->devid[i])
636 dv = map_dev_preferred(
637 major(st->devid[i]),
638 minor(st->devid[i]), 1, prefer);
639 change = newstate ^ st->devstate[i];
640 if (st->utime && change && !st->err) {
641 if (i < array.raid_disks &&
642 (((newstate&change)&(1<<MD_DISK_FAULTY)) ||
643 ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) ||
644 ((st->devstate[i]&change)&(1<<MD_DISK_SYNC)))
645 )
646 alert("Fail", dev, dv, ainfo);
647 else if (i >= array.raid_disks &&
648 (disc.major || disc.minor) &&
649 st->devid[i] == makedev(disc.major, disc.minor) &&
650 ((newstate&change)&(1<<MD_DISK_FAULTY))
651 )
652 alert("FailSpare", dev, dv, ainfo);
653 else if (i < array.raid_disks &&
654 ! (newstate & (1<<MD_DISK_REMOVED)) &&
655 (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) ||
656 ((newstate&change)&(1<<MD_DISK_ACTIVE)) ||
657 ((newstate&change)&(1<<MD_DISK_SYNC)))
658 )
659 alert("SpareActive", dev, dv, ainfo);
660 }
661 st->devstate[i] = newstate;
662 st->devid[i] = makedev(disc.major, disc.minor);
663 }
664 st->active = array.active_disks;
665 st->working = array.working_disks;
666 st->spare = array.spare_disks;
667 st->failed = array.failed_disks;
668 st->utime = array.utime;
669 st->raid = array.raid_disks;
670 st->err = 0;
671 if ((st->active < st->raid) && st->spare == 0)
672 return 1;
673 return 0;
674 }
675
676 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
677 int test, struct alert_info *info)
678 {
679 struct mdstat_ent *mse;
680 int new_found = 0;
681
682 for (mse=mdstat; mse; mse=mse->next)
683 if (mse->devnum != INT_MAX &&
684 (!mse->level || /* retrieve containers */
685 (strcmp(mse->level, "raid0") != 0 &&
686 strcmp(mse->level, "linear") != 0))
687 ) {
688 struct state *st = calloc(1, sizeof *st);
689 mdu_array_info_t array;
690 int fd;
691 if (st == NULL)
692 continue;
693 st->devname = strdup(get_md_name(mse->devnum));
694 if ((fd = open(st->devname, O_RDONLY)) < 0 ||
695 ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
696 /* no such array */
697 if (fd >=0) close(fd);
698 put_md_name(st->devname);
699 free(st->devname);
700 if (st->metadata) {
701 st->metadata->ss->free_super(st->metadata);
702 free(st->metadata);
703 }
704 free(st);
705 continue;
706 }
707 close(fd);
708 st->next = *statelist;
709 st->err = 1;
710 st->devnum = mse->devnum;
711 st->percent = RESYNC_UNKNOWN;
712 st->expected_spares = -1;
713 if (mse->metadata_version &&
714 strncmp(mse->metadata_version, "external:", 9) == 0 &&
715 is_subarray(mse->metadata_version+9))
716 st->parent_dev =
717 devname2devnum(mse->metadata_version+10);
718 else
719 st->parent_dev = NoMdDev;
720 *statelist = st;
721 if (test)
722 alert("TestMessage", st->devname, NULL, info);
723 new_found = 1;
724 }
725 return new_found;
726 }
727
728 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
729 {
730 int fd;
731
732 if (!st->metadata ||
733 !st->metadata->ss->min_acceptable_spare_size) {
734 *sizep = 0;
735 return 0;
736 }
737
738 fd = open(st->devname, O_RDONLY);
739 if (fd < 0)
740 return 1;
741 if (st->metadata->ss->external)
742 st->metadata->ss->load_container(st->metadata, fd, st->devname);
743 else
744 st->metadata->ss->load_super(st->metadata, fd, st->devname);
745 close(fd);
746 if (!st->metadata->sb)
747 return 1;
748 *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
749 st->metadata->ss->free_super(st->metadata);
750
751 return 0;
752 }
753
754 static int check_donor(struct state *from, struct state *to)
755 {
756 struct state *sub;
757
758 if (from == to)
759 return 0;
760 if (from->parent)
761 /* Cannot move from a member */
762 return 0;
763 if (from->err)
764 return 0;
765 for (sub = from->subarray; sub; sub = sub->subarray)
766 /* If source array has degraded subarrays, don't
767 * remove anything
768 */
769 if (sub->active < sub->raid)
770 return 0;
771 if (from->metadata->ss->external == 0)
772 if (from->active < from->raid)
773 return 0;
774 if (from->spare <= 0)
775 return 0;
776 return 1;
777 }
778
779 static dev_t choose_spare(struct state *from, struct state *to,
780 struct domainlist *domlist, unsigned long long min_size)
781 {
782 int d;
783 dev_t dev = 0;
784
785 for (d = from->raid; !dev && d < MAX_DISKS; d++) {
786 if (from->devid[d] > 0 &&
787 from->devstate[d] == 0) {
788 struct dev_policy *pol;
789 unsigned long long dev_size;
790
791 if (to->metadata->ss->external &&
792 test_partition_from_id(from->devid[d]))
793 continue;
794
795 if (min_size &&
796 dev_size_from_id(from->devid[d], &dev_size) &&
797 dev_size < min_size)
798 continue;
799
800 pol = devnum_policy(from->devid[d]);
801 if (from->spare_group)
802 pol_add(&pol, pol_domain,
803 from->spare_group, NULL);
804 if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
805 dev = from->devid[d];
806 dev_policy_free(pol);
807 }
808 }
809 return dev;
810 }
811
812 static dev_t container_choose_spare(struct state *from, struct state *to,
813 struct domainlist *domlist,
814 unsigned long long min_size, int active)
815 {
816 /* This is similar to choose_spare, but we cannot trust devstate,
817 * so we need to read the metadata instead
818 */
819 struct mdinfo *list;
820 struct supertype *st = from->metadata;
821 int fd = open(from->devname, O_RDONLY);
822 int err;
823 dev_t dev = 0;
824
825 if (fd < 0)
826 return 0;
827 if (!st->ss->getinfo_super_disks) {
828 close(fd);
829 return 0;
830 }
831
832 err = st->ss->load_container(st, fd, NULL);
833 close(fd);
834 if (err)
835 return 0;
836
837 if (from == to) {
838 /* We must check if number of active disks has not increased
839 * since ioctl in main loop. mdmon may have added spare
840 * to subarray. If so we do not need to look for more spares
841 * so return non zero value */
842 int active_cnt = 0;
843 struct mdinfo *dp;
844 list = st->ss->getinfo_super_disks(st);
845 if (!list) {
846 st->ss->free_super(st);
847 return 1;
848 }
849 dp = list->devs;
850 while (dp) {
851 if (dp->disk.state & (1<<MD_DISK_SYNC) &&
852 !(dp->disk.state & (1<<MD_DISK_FAULTY)))
853 active_cnt++;
854 dp = dp->next;
855 }
856 sysfs_free(list);
857 if (active < active_cnt) {
858 /* Spare just activated.*/
859 st->ss->free_super(st);
860 return 1;
861 }
862 }
863
864 /* We only need one spare so full list not needed */
865 list = container_choose_spares(st, min_size, domlist, from->spare_group,
866 to->metadata->ss->name, 1);
867 if (list) {
868 struct mdinfo *disks = list->devs;
869 if (disks)
870 dev = makedev(disks->disk.major, disks->disk.minor);
871 sysfs_free(list);
872 }
873 st->ss->free_super(st);
874 return dev;
875 }
876
877
878 static void try_spare_migration(struct state *statelist, struct alert_info *info)
879 {
880 struct state *from;
881 struct state *st;
882
883 link_containers_with_subarrays(statelist);
884 for (st = statelist; st; st = st->next)
885 if (st->active < st->raid &&
886 st->spare == 0 && !st->err) {
887 struct domainlist *domlist = NULL;
888 int d;
889 struct state *to = st;
890 unsigned long long min_size;
891
892 if (to->parent_dev != NoMdDev && !to->parent)
893 /* subarray monitored without parent container
894 * we can't move spares here */
895 continue;
896
897 if (to->parent)
898 /* member of a container */
899 to = to->parent;
900
901 if (get_min_spare_size_required(to, &min_size))
902 continue;
903 if (to->metadata->ss->external) {
904 /* We must make sure there is
905 * no suitable spare in container already.
906 * If there is we don't add more */
907 dev_t devid = container_choose_spare(
908 to, to, NULL, min_size, st->active);
909 if (devid > 0)
910 continue;
911 }
912 for (d = 0; d < MAX_DISKS; d++)
913 if (to->devid[d])
914 domainlist_add_dev(&domlist,
915 to->devid[d],
916 to->metadata->ss->name);
917 if (to->spare_group)
918 domain_add(&domlist, to->spare_group);
919 /*
920 * No spare migration if the destination
921 * has no domain. Skip this array.
922 */
923 if (!domlist)
924 continue;
925 for (from=statelist ; from ; from=from->next) {
926 dev_t devid;
927 if (!check_donor(from, to))
928 continue;
929 if (from->metadata->ss->external)
930 devid = container_choose_spare(
931 from, to, domlist, min_size, 0);
932 else
933 devid = choose_spare(from, to, domlist,
934 min_size);
935 if (devid > 0
936 && move_spare(from->devname, to->devname, devid)) {
937 alert("MoveSpare", to->devname, from->devname, info);
938 break;
939 }
940 }
941 domain_free(domlist);
942 }
943 }
944
945 /* search the statelist to connect external
946 * metadata subarrays with their containers
947 * We always completely rebuild the tree from scratch as
948 * that is safest considering the possibility of entries
949 * disappearing or changing.
950 */
951 static void link_containers_with_subarrays(struct state *list)
952 {
953 struct state *st;
954 struct state *cont;
955 for (st = list; st; st = st->next) {
956 st->parent = NULL;
957 st->subarray = NULL;
958 }
959 for (st = list; st; st = st->next)
960 if (st->parent_dev != NoMdDev)
961 for (cont = list; cont; cont = cont->next)
962 if (!cont->err &&
963 cont->parent_dev == NoMdDev &&
964 cont->devnum == st->parent_dev) {
965 st->parent = cont;
966 st->subarray = cont->subarray;
967 cont->subarray = st;
968 break;
969 }
970 }
971
972 /* Not really Monitor but ... */
973 int Wait(char *dev)
974 {
975 struct stat stb;
976 int devnum;
977 int rv = 1;
978
979 if (stat(dev, &stb) != 0) {
980 fprintf(stderr, Name ": Cannot find %s: %s\n", dev,
981 strerror(errno));
982 return 2;
983 }
984 devnum = stat2devnum(&stb);
985
986 while(1) {
987 struct mdstat_ent *ms = mdstat_read(1, 0);
988 struct mdstat_ent *e;
989
990 for (e=ms ; e; e=e->next)
991 if (e->devnum == devnum)
992 break;
993
994 if (!e || e->percent < 0) {
995 if (e && e->metadata_version &&
996 strncmp(e->metadata_version, "external:", 9) == 0) {
997 if (is_subarray(&e->metadata_version[9]))
998 ping_monitor(&e->metadata_version[9]);
999 else
1000 ping_monitor_by_id(devnum);
1001 }
1002 free_mdstat(ms);
1003 return rv;
1004 }
1005 free_mdstat(ms);
1006 rv = 0;
1007 mdstat_wait(5);
1008 }
1009 }
1010
1011 #ifndef MDASSEMBLE
1012
1013 static char *clean_states[] = {
1014 "clear", "inactive", "readonly", "read-auto", "clean", NULL };
1015
1016 int WaitClean(char *dev, int sock, int verbose)
1017 {
1018 int fd;
1019 struct mdinfo *mdi;
1020 int rv = 1;
1021 int devnum;
1022
1023 fd = open(dev, O_RDONLY);
1024 if (fd < 0) {
1025 if (verbose)
1026 fprintf(stderr, Name ": Couldn't open %s: %s\n", dev, strerror(errno));
1027 return 1;
1028 }
1029
1030 devnum = fd2devnum(fd);
1031 mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1032 if (!mdi) {
1033 if (verbose)
1034 fprintf(stderr, Name ": Failed to read sysfs attributes for "
1035 "%s\n", dev);
1036 close(fd);
1037 return 0;
1038 }
1039
1040 switch(mdi->array.level) {
1041 case LEVEL_LINEAR:
1042 case LEVEL_MULTIPATH:
1043 case 0:
1044 /* safemode delay is irrelevant for these levels */
1045 rv = 0;
1046 }
1047
1048 /* for internal metadata the kernel handles the final clean
1049 * transition, containers can never be dirty
1050 */
1051 if (!is_subarray(mdi->text_version))
1052 rv = 0;
1053
1054 /* safemode disabled ? */
1055 if (mdi->safe_mode_delay == 0)
1056 rv = 0;
1057
1058 if (rv) {
1059 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
1060 char buf[20];
1061 fd_set fds;
1062 struct timeval tm;
1063
1064 /* minimize the safe_mode_delay and prepare to wait up to 5s
1065 * for writes to quiesce
1066 */
1067 sysfs_set_safemode(mdi, 1);
1068 tm.tv_sec = 5;
1069 tm.tv_usec = 0;
1070
1071 FD_ZERO(&fds);
1072
1073 /* wait for array_state to be clean */
1074 while (1) {
1075 rv = read(state_fd, buf, sizeof(buf));
1076 if (rv < 0)
1077 break;
1078 if (sysfs_match_word(buf, clean_states) <= 4)
1079 break;
1080 FD_SET(state_fd, &fds);
1081 rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
1082 if (rv < 0 && errno != EINTR)
1083 break;
1084 lseek(state_fd, 0, SEEK_SET);
1085 }
1086 if (rv < 0)
1087 rv = 1;
1088 else if (fping_monitor(sock) == 0 ||
1089 ping_monitor(mdi->text_version) == 0) {
1090 /* we need to ping to close the window between array
1091 * state transitioning to clean and the metadata being
1092 * marked clean
1093 */
1094 rv = 0;
1095 } else
1096 rv = 1;
1097 if (rv && verbose)
1098 fprintf(stderr, Name ": Error waiting for %s to be clean\n",
1099 dev);
1100
1101 /* restore the original safe_mode_delay */
1102 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1103 close(state_fd);
1104 }
1105
1106 sysfs_free(mdi);
1107 close(fd);
1108
1109 return rv;
1110 }
1111 #endif /* MDASSEMBLE */
Unnamed repository; edit this file 'description' to name the repository.