]> 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
Add --prefer option for --detail and --monitor
[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 /* The largest number of disks current arrays can manage is 384
34 * This really should be dynamically, but that will have to wait
35 * At least it isn't MD_SB_DISKS.
36 */
37 #define MaxDisks 384
38 struct state {
39 char *devname;
40 int devnum; /* to sync with mdstat info */
41 long utime;
42 int err;
43 char *spare_group;
44 int active, working, failed, spare, raid;
45 int expected_spares;
46 int devstate[MaxDisks];
47 dev_t devid[MaxDisks];
48 int percent;
49 int parent_dev; /* For subarray, devnum of parent.
50 * For others, NoMdDev
51 */
52 struct supertype *metadata;
53 struct state *subarray;/* for a container it is a link to first subarray
54 * for a subarray it is a link to next subarray
55 * in the same container */
56 struct state *parent; /* for a subarray it is a link to its container
57 */
58 struct state *next;
59 };
60
61 struct alert_info {
62 char *mailaddr;
63 char *mailfrom;
64 char *alert_cmd;
65 int dosyslog;
66 };
67 static int make_daemon(char *pidfile);
68 static int check_one_sharer(int scan);
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
78 int Monitor(struct mddev_dev *devlist,
79 char *mailaddr, char *alert_cmd,
80 int period, int daemonise, int scan, int oneshot,
81 int dosyslog, int test, char *pidfile, int increments,
82 int share, char *prefer)
83 {
84 /*
85 * Every few seconds, scan every md device looking for changes
86 * When a change is found, log it, possibly run the alert command,
87 * and possibly send Email
88 *
89 * For each array, we record:
90 * Update time
91 * active/working/failed/spare drives
92 * State of each device.
93 * %rebuilt if rebuilding
94 *
95 * If the update time changes, check out all the data again
96 * It is possible that we cannot get the state of each device
97 * due to bugs in the md kernel module.
98 * We also read /proc/mdstat to get rebuild percent,
99 * and to get state on all active devices incase of kernel bug.
100 *
101 * Events are:
102 * Fail
103 * An active device had Faulty set or Active/Sync removed
104 * FailSpare
105 * A spare device had Faulty set
106 * SpareActive
107 * An active device had a reverse transition
108 * RebuildStarted
109 * percent went from -1 to +ve
110 * RebuildNN
111 * percent went from below to not-below NN%
112 * DeviceDisappeared
113 * Couldn't access a device which was previously visible
114 *
115 * if we detect an array with active<raid and spare==0
116 * we look at other arrays that have same spare-group
117 * If we find one with active==raid and spare>0,
118 * and if we can get_disk_info and find a name
119 * Then we hot-remove and hot-add to the other array
120 *
121 * If devlist is NULL, then we can monitor everything because --scan
122 * was given. We get an initial list from config file and add anything
123 * that appears in /proc/mdstat
124 */
125
126 struct state *statelist = NULL;
127 struct state *st2;
128 int finished = 0;
129 struct mdstat_ent *mdstat = NULL;
130 char *mailfrom = NULL;
131 struct alert_info info;
132
133 if (!mailaddr) {
134 mailaddr = conf_get_mailaddr();
135 if (mailaddr && ! scan)
136 fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n",
137 mailaddr);
138 }
139 mailfrom = conf_get_mailfrom();
140
141 if (!alert_cmd) {
142 alert_cmd = conf_get_program();
143 if (alert_cmd && ! scan)
144 fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n",
145 alert_cmd);
146 }
147 if (scan && !mailaddr && !alert_cmd && !dosyslog) {
148 fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n");
149 return 1;
150 }
151 info.alert_cmd = alert_cmd;
152 info.mailaddr = mailaddr;
153 info.mailfrom = mailfrom;
154 info.dosyslog = dosyslog;
155
156 if (daemonise) {
157 int rv = make_daemon(pidfile);
158 if (rv >= 0)
159 return rv;
160 }
161
162 if (share)
163 if (check_one_sharer(scan))
164 return 1;
165
166 if (devlist == NULL) {
167 struct mddev_ident *mdlist = conf_get_ident(NULL);
168 for (; mdlist; mdlist=mdlist->next) {
169 struct state *st;
170 if (mdlist->devname == NULL)
171 continue;
172 if (strcasecmp(mdlist->devname, "<ignore>") == 0)
173 continue;
174 st = calloc(1, sizeof *st);
175 if (st == NULL)
176 continue;
177 if (mdlist->devname[0] == '/')
178 st->devname = strdup(mdlist->devname);
179 else {
180 st->devname = malloc(8+strlen(mdlist->devname)+1);
181 strcpy(strcpy(st->devname, "/dev/md/"),
182 mdlist->devname);
183 }
184 st->next = statelist;
185 st->devnum = INT_MAX;
186 st->percent = -2;
187 st->expected_spares = mdlist->spare_disks;
188 if (mdlist->spare_group)
189 st->spare_group = strdup(mdlist->spare_group);
190 statelist = st;
191 }
192 } else {
193 struct mddev_dev *dv;
194 for (dv=devlist ; dv; dv=dv->next) {
195 struct mddev_ident *mdlist = conf_get_ident(dv->devname);
196 struct state *st = calloc(1, sizeof *st);
197 if (st == NULL)
198 continue;
199 st->devname = strdup(dv->devname);
200 st->next = statelist;
201 st->devnum = INT_MAX;
202 st->percent = -2;
203 st->expected_spares = -1;
204 if (mdlist) {
205 st->expected_spares = mdlist->spare_disks;
206 if (mdlist->spare_group)
207 st->spare_group = strdup(mdlist->spare_group);
208 }
209 statelist = st;
210 }
211 }
212
213
214 while (! finished) {
215 int new_found = 0;
216 struct state *st;
217 int anydegraded = 0;
218
219 if (mdstat)
220 free_mdstat(mdstat);
221 mdstat = mdstat_read(oneshot?0:1, 0);
222
223 for (st=statelist; st; st=st->next)
224 if (check_array(st, mdstat, test, &info,
225 increments, prefer))
226 anydegraded = 1;
227
228 /* now check if there are any new devices found in mdstat */
229 if (scan)
230 new_found = add_new_arrays(mdstat, &statelist, test,
231 &info);
232
233 /* If an array has active < raid && spare == 0 && spare_group != NULL
234 * Look for another array with spare > 0 and active == raid and same spare_group
235 * if found, choose a device and hotremove/hotadd
236 */
237 if (share && anydegraded)
238 try_spare_migration(statelist, &info);
239 if (!new_found) {
240 if (oneshot)
241 break;
242 else
243 mdstat_wait(period);
244 }
245 test = 0;
246 }
247 for (st2 = statelist; st2; st2 = statelist) {
248 statelist = st2->next;
249 free(st2);
250 }
251
252 if (pidfile)
253 unlink(pidfile);
254 return 0;
255 }
256
257 static int make_daemon(char *pidfile)
258 {
259 /* Return:
260 * -1 in the forked daemon
261 * 0 in the parent
262 * 1 on error
263 * so a none-negative becomes the exit code.
264 */
265 int pid = fork();
266 if (pid > 0) {
267 if (!pidfile)
268 printf("%d\n", pid);
269 else {
270 FILE *pid_file;
271 pid_file=fopen(pidfile, "w");
272 if (!pid_file)
273 perror("cannot create pid file");
274 else {
275 fprintf(pid_file,"%d\n", pid);
276 fclose(pid_file);
277 }
278 }
279 return 0;
280 }
281 if (pid < 0) {
282 perror("daemonise");
283 return 1;
284 }
285 close(0);
286 open("/dev/null", O_RDWR);
287 dup2(0,1);
288 dup2(0,2);
289 setsid();
290 return -1;
291 }
292
293 static int check_one_sharer(int scan)
294 {
295 int pid, rv;
296 FILE *fp;
297 char dir[20];
298 char path[100];
299 struct stat buf;
300 sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
301 fp = fopen(path, "r");
302 if (fp) {
303 if (fscanf(fp, "%d", &pid) != 1)
304 pid = -1;
305 sprintf(dir, "/proc/%d", pid);
306 rv = stat(dir, &buf);
307 if (rv != -1) {
308 if (scan) {
309 fprintf(stderr, Name ": Only one "
310 "autorebuild process allowed"
311 " in scan mode, aborting\n");
312 fclose(fp);
313 return 1;
314 } else {
315 fprintf(stderr, Name ": Warning: One"
316 " autorebuild process already"
317 " running.\n");
318 }
319 }
320 fclose(fp);
321 }
322 if (scan) {
323 if (mkdir(MDMON_DIR, S_IRWXU) < 0 &&
324 errno != EEXIST) {
325 fprintf(stderr, Name ": Can't create "
326 "autorebuild.pid file\n");
327 } else {
328 fp = fopen(path, "w");
329 if (!fp)
330 fprintf(stderr, Name ": Cannot create"
331 " autorebuild.pid"
332 "file\n");
333 else {
334 pid = getpid();
335 fprintf(fp, "%d\n", pid);
336 fclose(fp);
337 }
338 }
339 }
340 return 0;
341 }
342
343 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
344 {
345 int priority;
346
347 if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
348 time_t now = time(0);
349
350 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
351 }
352 if (info->alert_cmd) {
353 int pid = fork();
354 switch(pid) {
355 default:
356 waitpid(pid, NULL, 0);
357 break;
358 case -1:
359 break;
360 case 0:
361 execl(info->alert_cmd, info->alert_cmd,
362 event, dev, disc, NULL);
363 exit(2);
364 }
365 }
366 if (info->mailaddr &&
367 (strncmp(event, "Fail", 4)==0 ||
368 strncmp(event, "Test", 4)==0 ||
369 strncmp(event, "Spares", 6)==0 ||
370 strncmp(event, "Degrade", 7)==0)) {
371 FILE *mp = popen(Sendmail, "w");
372 if (mp) {
373 FILE *mdstat;
374 char hname[256];
375 gethostname(hname, sizeof(hname));
376 signal(SIGPIPE, SIG_IGN);
377 if (info->mailfrom)
378 fprintf(mp, "From: %s\n", info->mailfrom);
379 else
380 fprintf(mp, "From: " Name " monitoring <root>\n");
381 fprintf(mp, "To: %s\n", info->mailaddr);
382 fprintf(mp, "Subject: %s event on %s:%s\n\n",
383 event, dev, hname);
384
385 fprintf(mp,
386 "This is an automatically generated"
387 " mail message from " Name "\n");
388 fprintf(mp, "running on %s\n\n", hname);
389
390 fprintf(mp,
391 "A %s event had been detected on"
392 " md device %s.\n\n", event, dev);
393
394 if (disc && disc[0] != ' ')
395 fprintf(mp,
396 "It could be related to"
397 " component device %s.\n\n", disc);
398 if (disc && disc[0] == ' ')
399 fprintf(mp, "Extra information:%s.\n\n", disc);
400
401 fprintf(mp, "Faithfully yours, etc.\n");
402
403 mdstat = fopen("/proc/mdstat", "r");
404 if (mdstat) {
405 char buf[8192];
406 int n;
407 fprintf(mp,
408 "\nP.S. The /proc/mdstat file"
409 " currently contains the following:\n\n");
410 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
411 n=fwrite(buf, 1, n, mp);
412 fclose(mdstat);
413 }
414 pclose(mp);
415 }
416 }
417
418 /* log the event to syslog maybe */
419 if (info->dosyslog) {
420 /* Log at a different severity depending on the event.
421 *
422 * These are the critical events: */
423 if (strncmp(event, "Fail", 4)==0 ||
424 strncmp(event, "Degrade", 7)==0 ||
425 strncmp(event, "DeviceDisappeared", 17)==0)
426 priority = LOG_CRIT;
427 /* Good to know about, but are not failures: */
428 else if (strncmp(event, "Rebuild", 7)==0 ||
429 strncmp(event, "MoveSpare", 9)==0 ||
430 strncmp(event, "Spares", 6) != 0)
431 priority = LOG_WARNING;
432 /* Everything else: */
433 else
434 priority = LOG_INFO;
435
436 if (disc)
437 syslog(priority,
438 "%s event detected on md device %s,"
439 " component device %s", event, dev, disc);
440 else
441 syslog(priority,
442 "%s event detected on md device %s",
443 event, dev);
444 }
445 }
446
447 static int check_array(struct state *st, struct mdstat_ent *mdstat,
448 int test, struct alert_info *ainfo,
449 int increments, char *prefer)
450 {
451 /* Update the state 'st' to reflect any changes shown in mdstat,
452 * or found by directly examining the array, and return
453 * '1' if the array is degraded, or '0' if it is optimal (or dead).
454 */
455 struct { int state, major, minor; } info[MaxDisks];
456 mdu_array_info_t array;
457 struct mdstat_ent *mse = NULL, *mse2;
458 char *dev = st->devname;
459 int fd;
460 int i;
461 int remaining_disks;
462 int last_disk;
463
464 if (test)
465 alert("TestMessage", dev, NULL, ainfo);
466 fd = open(dev, O_RDONLY);
467 if (fd < 0) {
468 if (!st->err)
469 alert("DeviceDisappeared", dev, NULL, ainfo);
470 st->err=1;
471 return 0;
472 }
473 fcntl(fd, F_SETFD, FD_CLOEXEC);
474 if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
475 if (!st->err)
476 alert("DeviceDisappeared", dev, NULL, ainfo);
477 st->err=1;
478 close(fd);
479 return 0;
480 }
481 /* It's much easier to list what array levels can't
482 * have a device disappear than all of them that can
483 */
484 if (array.level == 0 || array.level == -1) {
485 if (!st->err)
486 alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
487 st->err = 1;
488 close(fd);
489 return 0;
490 }
491 if (st->devnum == INT_MAX) {
492 struct stat stb;
493 if (fstat(fd, &stb) == 0 &&
494 (S_IFMT&stb.st_mode)==S_IFBLK) {
495 if (major(stb.st_rdev) == MD_MAJOR)
496 st->devnum = minor(stb.st_rdev);
497 else
498 st->devnum = -1- (minor(stb.st_rdev)>>6);
499 }
500 }
501
502 for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
503 if (mse2->devnum == st->devnum) {
504 mse2->devnum = INT_MAX; /* flag it as "used" */
505 mse = mse2;
506 }
507
508 if (!mse) {
509 /* duplicated array in statelist
510 * or re-created after reading mdstat*/
511 st->err = 1;
512 close(fd);
513 return 0;
514 }
515 /* this array is in /proc/mdstat */
516 if (array.utime == 0)
517 /* external arrays don't update utime, so
518 * just make sure it is always different. */
519 array.utime = st->utime + 1;;
520
521 if (st->utime == array.utime &&
522 st->failed == array.failed_disks &&
523 st->working == array.working_disks &&
524 st->spare == array.spare_disks &&
525 (mse == NULL || (
526 mse->percent == st->percent
527 ))) {
528 close(fd);
529 st->err = 0;
530 if ((st->active < st->raid) && st->spare == 0)
531 return 1;
532 else
533 return 0;
534 }
535 if (st->utime == 0 && /* new array */
536 mse->pattern && strchr(mse->pattern, '_') /* degraded */
537 )
538 alert("DegradedArray", dev, NULL, ainfo);
539
540 if (st->utime == 0 && /* new array */
541 st->expected_spares > 0 &&
542 array.spare_disks < st->expected_spares)
543 alert("SparesMissing", dev, NULL, ainfo);
544 if (st->percent == -1 &&
545 mse->percent >= 0)
546 alert("RebuildStarted", dev, NULL, ainfo);
547 if (st->percent >= 0 &&
548 mse->percent >= 0 &&
549 (mse->percent / increments) > (st->percent / increments)) {
550 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
551
552 if((mse->percent / increments) == 0)
553 snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
554 else
555 snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
556
557 alert(percentalert, dev, NULL, ainfo);
558 }
559
560 if (mse->percent == -1 &&
561 st->percent >= 0) {
562 /* Rebuild/sync/whatever just finished.
563 * If there is a number in /mismatch_cnt,
564 * we should report that.
565 */
566 struct mdinfo *sra =
567 sysfs_read(-1, st->devnum, GET_MISMATCH);
568 if (sra && sra->mismatch_cnt > 0) {
569 char cnt[80];
570 snprintf(cnt, sizeof(cnt),
571 " mismatches found: %d (on raid level %d)",
572 sra->mismatch_cnt, array.level);
573 alert("RebuildFinished", dev, cnt, ainfo);
574 } else
575 alert("RebuildFinished", dev, NULL, ainfo);
576 if (sra)
577 free(sra);
578 }
579 st->percent = mse->percent;
580
581 remaining_disks = array.nr_disks;
582 for (i=0; i<MaxDisks && remaining_disks > 0;
583 i++) {
584 mdu_disk_info_t disc;
585 disc.number = i;
586 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
587 info[i].state = disc.state;
588 info[i].major = disc.major;
589 info[i].minor = disc.minor;
590 if (disc.major || disc.minor)
591 remaining_disks --;
592 } else
593 info[i].major = info[i].minor = 0;
594 }
595 last_disk = i;
596
597 if (mse->metadata_version &&
598 strncmp(mse->metadata_version, "external:", 9) == 0 &&
599 is_subarray(mse->metadata_version+9))
600 st->parent_dev =
601 devname2devnum(mse->metadata_version+10);
602 else
603 st->parent_dev = NoMdDev;
604 if (st->metadata == NULL &&
605 st->parent_dev == NoMdDev)
606 st->metadata = super_by_fd(fd, NULL);
607
608 close(fd);
609
610 for (i=0; i<MaxDisks; i++) {
611 mdu_disk_info_t disc = {0,0,0,0,0};
612 int newstate=0;
613 int change;
614 char *dv = NULL;
615 disc.number = i;
616 if (i >= last_disk) {
617 newstate = 0;
618 disc.major = disc.minor = 0;
619 } else if (info[i].major || info[i].minor) {
620 newstate = info[i].state;
621 dv = map_dev_preferred(
622 info[i].major, info[i].minor, 1,
623 prefer);
624 disc.state = newstate;
625 disc.major = info[i].major;
626 disc.minor = info[i].minor;
627 } else if (mse && mse->pattern && i < (int)strlen(mse->pattern)) {
628 switch(mse->pattern[i]) {
629 case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
630 case '_': newstate = 8 /* REMOVED */; break;
631 }
632 disc.major = disc.minor = 0;
633 }
634 if (dv == NULL && st->devid[i])
635 dv = map_dev_preferred(
636 major(st->devid[i]),
637 minor(st->devid[i]), 1, prefer);
638 change = newstate ^ st->devstate[i];
639 if (st->utime && change && !st->err) {
640 if (i < array.raid_disks &&
641 (((newstate&change)&(1<<MD_DISK_FAULTY)) ||
642 ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) ||
643 ((st->devstate[i]&change)&(1<<MD_DISK_SYNC)))
644 )
645 alert("Fail", dev, dv, ainfo);
646 else if (i >= array.raid_disks &&
647 (disc.major || disc.minor) &&
648 st->devid[i] == makedev(disc.major, disc.minor) &&
649 ((newstate&change)&(1<<MD_DISK_FAULTY))
650 )
651 alert("FailSpare", dev, dv, ainfo);
652 else if (i < array.raid_disks &&
653 ! (newstate & (1<<MD_DISK_REMOVED)) &&
654 (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) ||
655 ((newstate&change)&(1<<MD_DISK_ACTIVE)) ||
656 ((newstate&change)&(1<<MD_DISK_SYNC)))
657 )
658 alert("SpareActive", dev, dv, ainfo);
659 }
660 st->devstate[i] = newstate;
661 st->devid[i] = makedev(disc.major, disc.minor);
662 }
663 st->active = array.active_disks;
664 st->working = array.working_disks;
665 st->spare = array.spare_disks;
666 st->failed = array.failed_disks;
667 st->utime = array.utime;
668 st->raid = array.raid_disks;
669 st->err = 0;
670 if ((st->active < st->raid) && st->spare == 0)
671 return 1;
672 return 0;
673 }
674
675 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
676 int test, struct alert_info *info)
677 {
678 struct mdstat_ent *mse;
679 int new_found = 0;
680
681 for (mse=mdstat; mse; mse=mse->next)
682 if (mse->devnum != INT_MAX &&
683 (!mse->level || /* retrieve containers */
684 (strcmp(mse->level, "raid0") != 0 &&
685 strcmp(mse->level, "linear") != 0))
686 ) {
687 struct state *st = calloc(1, sizeof *st);
688 mdu_array_info_t array;
689 int fd;
690 if (st == NULL)
691 continue;
692 st->devname = strdup(get_md_name(mse->devnum));
693 if ((fd = open(st->devname, O_RDONLY)) < 0 ||
694 ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
695 /* no such array */
696 if (fd >=0) close(fd);
697 put_md_name(st->devname);
698 free(st->devname);
699 if (st->metadata) {
700 st->metadata->ss->free_super(st->metadata);
701 free(st->metadata);
702 }
703 free(st);
704 continue;
705 }
706 close(fd);
707 st->next = *statelist;
708 st->err = 1;
709 st->devnum = mse->devnum;
710 st->percent = -2;
711 st->expected_spares = -1;
712 if (mse->metadata_version &&
713 strncmp(mse->metadata_version, "external:", 9) == 0 &&
714 is_subarray(mse->metadata_version+9))
715 st->parent_dev =
716 devname2devnum(mse->metadata_version+10);
717 else
718 st->parent_dev = NoMdDev;
719 *statelist = st;
720 if (test)
721 alert("TestMessage", st->devname, NULL, info);
722 alert("NewArray", 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 < MaxDisks; 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 < MaxDisks; 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.