]> 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
Change Monitor to take a struct context
[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 struct context *c,
76 int daemonise, int oneshot,
77 int dosyslog, char *pidfile, int increments,
78 int share)
79 {
80 /*
81 * Every few seconds, scan every md device looking for changes
82 * When a change is found, log it, possibly run the alert command,
83 * and possibly send Email
84 *
85 * For each array, we record:
86 * Update time
87 * active/working/failed/spare drives
88 * State of each device.
89 * %rebuilt if rebuilding
90 *
91 * If the update time changes, check out all the data again
92 * It is possible that we cannot get the state of each device
93 * due to bugs in the md kernel module.
94 * We also read /proc/mdstat to get rebuild percent,
95 * and to get state on all active devices incase of kernel bug.
96 *
97 * Events are:
98 * Fail
99 * An active device had Faulty set or Active/Sync removed
100 * FailSpare
101 * A spare device had Faulty set
102 * SpareActive
103 * An active device had a reverse transition
104 * RebuildStarted
105 * percent went from -1 to +ve
106 * RebuildNN
107 * percent went from below to not-below NN%
108 * DeviceDisappeared
109 * Couldn't access a device which was previously visible
110 *
111 * if we detect an array with active<raid and spare==0
112 * we look at other arrays that have same spare-group
113 * If we find one with active==raid and spare>0,
114 * and if we can get_disk_info and find a name
115 * Then we hot-remove and hot-add to the other array
116 *
117 * If devlist is NULL, then we can monitor everything because --scan
118 * was given. We get an initial list from config file and add anything
119 * that appears in /proc/mdstat
120 */
121
122 struct state *statelist = NULL;
123 struct state *st2;
124 int finished = 0;
125 struct mdstat_ent *mdstat = NULL;
126 char *mailfrom = NULL;
127 struct alert_info info;
128
129 if (!mailaddr) {
130 mailaddr = conf_get_mailaddr();
131 if (mailaddr && ! c->scan)
132 pr_err("Monitor using email address \"%s\" from config file\n",
133 mailaddr);
134 }
135 mailfrom = conf_get_mailfrom();
136
137 if (!alert_cmd) {
138 alert_cmd = conf_get_program();
139 if (alert_cmd && ! c->scan)
140 pr_err("Monitor using program \"%s\" from config file\n",
141 alert_cmd);
142 }
143 if (c->scan && !mailaddr && !alert_cmd && !dosyslog) {
144 pr_err("No mail address or alert command - not monitoring.\n");
145 return 1;
146 }
147 info.alert_cmd = alert_cmd;
148 info.mailaddr = mailaddr;
149 info.mailfrom = mailfrom;
150 info.dosyslog = dosyslog;
151
152 if (daemonise) {
153 int rv = make_daemon(pidfile);
154 if (rv >= 0)
155 return rv;
156 }
157
158 if (share)
159 if (check_one_sharer(c->scan))
160 return 1;
161
162 if (devlist == NULL) {
163 struct mddev_ident *mdlist = conf_get_ident(NULL);
164 for (; mdlist; mdlist=mdlist->next) {
165 struct state *st;
166 if (mdlist->devname == NULL)
167 continue;
168 if (strcasecmp(mdlist->devname, "<ignore>") == 0)
169 continue;
170 st = xcalloc(1, sizeof *st);
171 if (mdlist->devname[0] == '/')
172 st->devname = xstrdup(mdlist->devname);
173 else {
174 st->devname = xmalloc(8+strlen(mdlist->devname)+1);
175 strcpy(strcpy(st->devname, "/dev/md/"),
176 mdlist->devname);
177 }
178 st->next = statelist;
179 st->devnum = INT_MAX;
180 st->percent = RESYNC_UNKNOWN;
181 st->expected_spares = mdlist->spare_disks;
182 if (mdlist->spare_group)
183 st->spare_group = xstrdup(mdlist->spare_group);
184 statelist = st;
185 }
186 } else {
187 struct mddev_dev *dv;
188 for (dv=devlist ; dv; dv=dv->next) {
189 struct mddev_ident *mdlist = conf_get_ident(dv->devname);
190 struct state *st = xcalloc(1, sizeof *st);
191 st->devname = xstrdup(dv->devname);
192 st->next = statelist;
193 st->devnum = INT_MAX;
194 st->percent = RESYNC_UNKNOWN;
195 st->expected_spares = -1;
196 if (mdlist) {
197 st->expected_spares = mdlist->spare_disks;
198 if (mdlist->spare_group)
199 st->spare_group = xstrdup(mdlist->spare_group);
200 }
201 statelist = st;
202 }
203 }
204
205
206 while (! finished) {
207 int new_found = 0;
208 struct state *st;
209 int anydegraded = 0;
210
211 if (mdstat)
212 free_mdstat(mdstat);
213 mdstat = mdstat_read(oneshot?0:1, 0);
214
215 for (st=statelist; st; st=st->next)
216 if (check_array(st, mdstat, c->test, &info,
217 increments, c->prefer))
218 anydegraded = 1;
219
220 /* now check if there are any new devices found in mdstat */
221 if (c->scan)
222 new_found = add_new_arrays(mdstat, &statelist, c->test,
223 &info);
224
225 /* If an array has active < raid && spare == 0 && spare_group != NULL
226 * Look for another array with spare > 0 and active == raid and same spare_group
227 * if found, choose a device and hotremove/hotadd
228 */
229 if (share && anydegraded)
230 try_spare_migration(statelist, &info);
231 if (!new_found) {
232 if (oneshot)
233 break;
234 else
235 mdstat_wait(c->delay);
236 }
237 c->test = 0;
238 }
239 for (st2 = statelist; st2; st2 = statelist) {
240 statelist = st2->next;
241 free(st2);
242 }
243
244 if (pidfile)
245 unlink(pidfile);
246 return 0;
247 }
248
249 static int make_daemon(char *pidfile)
250 {
251 /* Return:
252 * -1 in the forked daemon
253 * 0 in the parent
254 * 1 on error
255 * so a none-negative becomes the exit code.
256 */
257 int pid = fork();
258 if (pid > 0) {
259 if (!pidfile)
260 printf("%d\n", pid);
261 else {
262 FILE *pid_file;
263 pid_file=fopen(pidfile, "w");
264 if (!pid_file)
265 perror("cannot create pid file");
266 else {
267 fprintf(pid_file,"%d\n", pid);
268 fclose(pid_file);
269 }
270 }
271 return 0;
272 }
273 if (pid < 0) {
274 perror("daemonise");
275 return 1;
276 }
277 close(0);
278 open("/dev/null", O_RDWR);
279 dup2(0,1);
280 dup2(0,2);
281 setsid();
282 return -1;
283 }
284
285 static int check_one_sharer(int scan)
286 {
287 int pid, rv;
288 FILE *fp;
289 char dir[20];
290 char path[100];
291 struct stat buf;
292 sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
293 fp = fopen(path, "r");
294 if (fp) {
295 if (fscanf(fp, "%d", &pid) != 1)
296 pid = -1;
297 sprintf(dir, "/proc/%d", pid);
298 rv = stat(dir, &buf);
299 if (rv != -1) {
300 if (scan) {
301 pr_err("Only one "
302 "autorebuild process allowed"
303 " in scan mode, aborting\n");
304 fclose(fp);
305 return 1;
306 } else {
307 pr_err("Warning: One"
308 " autorebuild process already"
309 " running.\n");
310 }
311 }
312 fclose(fp);
313 }
314 if (scan) {
315 if (mkdir(MDMON_DIR, S_IRWXU) < 0 &&
316 errno != EEXIST) {
317 pr_err("Can't create "
318 "autorebuild.pid file\n");
319 } else {
320 fp = fopen(path, "w");
321 if (!fp)
322 pr_err("Cannot create"
323 " autorebuild.pid"
324 "file\n");
325 else {
326 pid = getpid();
327 fprintf(fp, "%d\n", pid);
328 fclose(fp);
329 }
330 }
331 }
332 return 0;
333 }
334
335 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
336 {
337 int priority;
338
339 if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
340 time_t now = time(0);
341
342 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
343 }
344 if (info->alert_cmd) {
345 int pid = fork();
346 switch(pid) {
347 default:
348 waitpid(pid, NULL, 0);
349 break;
350 case -1:
351 break;
352 case 0:
353 execl(info->alert_cmd, info->alert_cmd,
354 event, dev, disc, NULL);
355 exit(2);
356 }
357 }
358 if (info->mailaddr &&
359 (strncmp(event, "Fail", 4)==0 ||
360 strncmp(event, "Test", 4)==0 ||
361 strncmp(event, "Spares", 6)==0 ||
362 strncmp(event, "Degrade", 7)==0)) {
363 FILE *mp = popen(Sendmail, "w");
364 if (mp) {
365 FILE *mdstat;
366 char hname[256];
367 gethostname(hname, sizeof(hname));
368 signal(SIGPIPE, SIG_IGN);
369 if (info->mailfrom)
370 fprintf(mp, "From: %s\n", info->mailfrom);
371 else
372 fprintf(mp, "From: " Name " monitoring <root>\n");
373 fprintf(mp, "To: %s\n", info->mailaddr);
374 fprintf(mp, "Subject: %s event on %s:%s\n\n",
375 event, dev, hname);
376
377 fprintf(mp,
378 "This is an automatically generated"
379 " mail message from " Name "\n");
380 fprintf(mp, "running on %s\n\n", hname);
381
382 fprintf(mp,
383 "A %s event had been detected on"
384 " md device %s.\n\n", event, dev);
385
386 if (disc && disc[0] != ' ')
387 fprintf(mp,
388 "It could be related to"
389 " component device %s.\n\n", disc);
390 if (disc && disc[0] == ' ')
391 fprintf(mp, "Extra information:%s.\n\n", disc);
392
393 fprintf(mp, "Faithfully yours, etc.\n");
394
395 mdstat = fopen("/proc/mdstat", "r");
396 if (mdstat) {
397 char buf[8192];
398 int n;
399 fprintf(mp,
400 "\nP.S. The /proc/mdstat file"
401 " currently contains the following:\n\n");
402 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
403 n=fwrite(buf, 1, n, mp);
404 fclose(mdstat);
405 }
406 pclose(mp);
407 }
408 }
409
410 /* log the event to syslog maybe */
411 if (info->dosyslog) {
412 /* Log at a different severity depending on the event.
413 *
414 * These are the critical events: */
415 if (strncmp(event, "Fail", 4)==0 ||
416 strncmp(event, "Degrade", 7)==0 ||
417 strncmp(event, "DeviceDisappeared", 17)==0)
418 priority = LOG_CRIT;
419 /* Good to know about, but are not failures: */
420 else if (strncmp(event, "Rebuild", 7)==0 ||
421 strncmp(event, "MoveSpare", 9)==0 ||
422 strncmp(event, "Spares", 6) != 0)
423 priority = LOG_WARNING;
424 /* Everything else: */
425 else
426 priority = LOG_INFO;
427
428 if (disc)
429 syslog(priority,
430 "%s event detected on md device %s,"
431 " component device %s", event, dev, disc);
432 else
433 syslog(priority,
434 "%s event detected on md device %s",
435 event, dev);
436 }
437 }
438
439 static int check_array(struct state *st, struct mdstat_ent *mdstat,
440 int test, struct alert_info *ainfo,
441 int increments, char *prefer)
442 {
443 /* Update the state 'st' to reflect any changes shown in mdstat,
444 * or found by directly examining the array, and return
445 * '1' if the array is degraded, or '0' if it is optimal (or dead).
446 */
447 struct { int state, major, minor; } info[MAX_DISKS];
448 mdu_array_info_t array;
449 struct mdstat_ent *mse = NULL, *mse2;
450 char *dev = st->devname;
451 int fd;
452 int i;
453 int remaining_disks;
454 int last_disk;
455 int new_array = 0;
456
457 if (test)
458 alert("TestMessage", dev, NULL, ainfo);
459 fd = open(dev, O_RDONLY);
460 if (fd < 0) {
461 if (!st->err)
462 alert("DeviceDisappeared", dev, NULL, ainfo);
463 st->err=1;
464 return 0;
465 }
466 fcntl(fd, F_SETFD, FD_CLOEXEC);
467 if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
468 if (!st->err)
469 alert("DeviceDisappeared", dev, NULL, ainfo);
470 st->err=1;
471 close(fd);
472 return 0;
473 }
474 /* It's much easier to list what array levels can't
475 * have a device disappear than all of them that can
476 */
477 if (array.level == 0 || array.level == -1) {
478 if (!st->err)
479 alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
480 st->err = 1;
481 close(fd);
482 return 0;
483 }
484 if (st->devnum == INT_MAX) {
485 struct stat stb;
486 if (fstat(fd, &stb) == 0 &&
487 (S_IFMT&stb.st_mode)==S_IFBLK) {
488 if (major(stb.st_rdev) == MD_MAJOR)
489 st->devnum = minor(stb.st_rdev);
490 else
491 st->devnum = -1- (minor(stb.st_rdev)>>6);
492 }
493 }
494
495 for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
496 if (mse2->devnum == st->devnum) {
497 mse2->devnum = INT_MAX; /* flag it as "used" */
498 mse = mse2;
499 }
500
501 if (!mse) {
502 /* duplicated array in statelist
503 * or re-created after reading mdstat*/
504 st->err = 1;
505 close(fd);
506 return 0;
507 }
508 /* this array is in /proc/mdstat */
509 if (array.utime == 0)
510 /* external arrays don't update utime, so
511 * just make sure it is always different. */
512 array.utime = st->utime + 1;;
513
514 if (st->err) {
515 /* New array appeared where previously had and error */
516 st->err = 0;
517 st->percent = RESYNC_NONE;
518 new_array = 1;
519 alert("NewArray", st->devname, NULL, ainfo);
520 }
521
522 if (st->utime == array.utime &&
523 st->failed == array.failed_disks &&
524 st->working == array.working_disks &&
525 st->spare == array.spare_disks &&
526 (mse == NULL || (
527 mse->percent == st->percent
528 ))) {
529 close(fd);
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 < 0 && st->percent != RESYNC_UNKNOWN &&
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 == RESYNC_NONE &&
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<MAX_DISKS && 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<MAX_DISKS; 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 (info[i].major || info[i].minor)) {
618 newstate = info[i].state;
619 dv = map_dev_preferred(
620 info[i].major, info[i].minor, 1,
621 prefer);
622 disc.state = newstate;
623 disc.major = info[i].major;
624 disc.minor = info[i].minor;
625 } else
626 newstate = (1 << MD_DISK_REMOVED);
627
628 if (dv == NULL && st->devid[i])
629 dv = map_dev_preferred(
630 major(st->devid[i]),
631 minor(st->devid[i]), 1, prefer);
632 change = newstate ^ st->devstate[i];
633 if (st->utime && change && !st->err && !new_array) {
634 if ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))
635 alert("Fail", dev, dv, ainfo);
636 else if ((newstate & (1<<MD_DISK_FAULTY)) &&
637 (disc.major || disc.minor) &&
638 st->devid[i] == makedev(disc.major, disc.minor))
639 alert("FailSpare", dev, dv, ainfo);
640 else if ((newstate&change)&(1<<MD_DISK_SYNC))
641 alert("SpareActive", dev, dv, ainfo);
642 }
643 st->devstate[i] = newstate;
644 st->devid[i] = makedev(disc.major, disc.minor);
645 }
646 st->active = array.active_disks;
647 st->working = array.working_disks;
648 st->spare = array.spare_disks;
649 st->failed = array.failed_disks;
650 st->utime = array.utime;
651 st->raid = array.raid_disks;
652 st->err = 0;
653 if ((st->active < st->raid) && st->spare == 0)
654 return 1;
655 return 0;
656 }
657
658 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
659 int test, struct alert_info *info)
660 {
661 struct mdstat_ent *mse;
662 int new_found = 0;
663
664 for (mse=mdstat; mse; mse=mse->next)
665 if (mse->devnum != INT_MAX &&
666 (!mse->level || /* retrieve containers */
667 (strcmp(mse->level, "raid0") != 0 &&
668 strcmp(mse->level, "linear") != 0))
669 ) {
670 struct state *st = xcalloc(1, sizeof *st);
671 mdu_array_info_t array;
672 int fd;
673 st->devname = xstrdup(get_md_name(mse->devnum));
674 if ((fd = open(st->devname, O_RDONLY)) < 0 ||
675 ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
676 /* no such array */
677 if (fd >=0) close(fd);
678 put_md_name(st->devname);
679 free(st->devname);
680 if (st->metadata) {
681 st->metadata->ss->free_super(st->metadata);
682 free(st->metadata);
683 }
684 free(st);
685 continue;
686 }
687 close(fd);
688 st->next = *statelist;
689 st->err = 1;
690 st->devnum = mse->devnum;
691 st->percent = RESYNC_UNKNOWN;
692 st->expected_spares = -1;
693 if (mse->metadata_version &&
694 strncmp(mse->metadata_version, "external:", 9) == 0 &&
695 is_subarray(mse->metadata_version+9))
696 st->parent_dev =
697 devname2devnum(mse->metadata_version+10);
698 else
699 st->parent_dev = NoMdDev;
700 *statelist = st;
701 if (test)
702 alert("TestMessage", st->devname, NULL, info);
703 new_found = 1;
704 }
705 return new_found;
706 }
707
708 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
709 {
710 int fd;
711
712 if (!st->metadata ||
713 !st->metadata->ss->min_acceptable_spare_size) {
714 *sizep = 0;
715 return 0;
716 }
717
718 fd = open(st->devname, O_RDONLY);
719 if (fd < 0)
720 return 1;
721 if (st->metadata->ss->external)
722 st->metadata->ss->load_container(st->metadata, fd, st->devname);
723 else
724 st->metadata->ss->load_super(st->metadata, fd, st->devname);
725 close(fd);
726 if (!st->metadata->sb)
727 return 1;
728 *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
729 st->metadata->ss->free_super(st->metadata);
730
731 return 0;
732 }
733
734 static int check_donor(struct state *from, struct state *to)
735 {
736 struct state *sub;
737
738 if (from == to)
739 return 0;
740 if (from->parent)
741 /* Cannot move from a member */
742 return 0;
743 if (from->err)
744 return 0;
745 for (sub = from->subarray; sub; sub = sub->subarray)
746 /* If source array has degraded subarrays, don't
747 * remove anything
748 */
749 if (sub->active < sub->raid)
750 return 0;
751 if (from->metadata->ss->external == 0)
752 if (from->active < from->raid)
753 return 0;
754 if (from->spare <= 0)
755 return 0;
756 return 1;
757 }
758
759 static dev_t choose_spare(struct state *from, struct state *to,
760 struct domainlist *domlist, unsigned long long min_size)
761 {
762 int d;
763 dev_t dev = 0;
764
765 for (d = from->raid; !dev && d < MAX_DISKS; d++) {
766 if (from->devid[d] > 0 &&
767 from->devstate[d] == 0) {
768 struct dev_policy *pol;
769 unsigned long long dev_size;
770
771 if (to->metadata->ss->external &&
772 test_partition_from_id(from->devid[d]))
773 continue;
774
775 if (min_size &&
776 dev_size_from_id(from->devid[d], &dev_size) &&
777 dev_size < min_size)
778 continue;
779
780 pol = devnum_policy(from->devid[d]);
781 if (from->spare_group)
782 pol_add(&pol, pol_domain,
783 from->spare_group, NULL);
784 if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
785 dev = from->devid[d];
786 dev_policy_free(pol);
787 }
788 }
789 return dev;
790 }
791
792 static dev_t container_choose_spare(struct state *from, struct state *to,
793 struct domainlist *domlist,
794 unsigned long long min_size, int active)
795 {
796 /* This is similar to choose_spare, but we cannot trust devstate,
797 * so we need to read the metadata instead
798 */
799 struct mdinfo *list;
800 struct supertype *st = from->metadata;
801 int fd = open(from->devname, O_RDONLY);
802 int err;
803 dev_t dev = 0;
804
805 if (fd < 0)
806 return 0;
807 if (!st->ss->getinfo_super_disks) {
808 close(fd);
809 return 0;
810 }
811
812 err = st->ss->load_container(st, fd, NULL);
813 close(fd);
814 if (err)
815 return 0;
816
817 if (from == to) {
818 /* We must check if number of active disks has not increased
819 * since ioctl in main loop. mdmon may have added spare
820 * to subarray. If so we do not need to look for more spares
821 * so return non zero value */
822 int active_cnt = 0;
823 struct mdinfo *dp;
824 list = st->ss->getinfo_super_disks(st);
825 if (!list) {
826 st->ss->free_super(st);
827 return 1;
828 }
829 dp = list->devs;
830 while (dp) {
831 if (dp->disk.state & (1<<MD_DISK_SYNC) &&
832 !(dp->disk.state & (1<<MD_DISK_FAULTY)))
833 active_cnt++;
834 dp = dp->next;
835 }
836 sysfs_free(list);
837 if (active < active_cnt) {
838 /* Spare just activated.*/
839 st->ss->free_super(st);
840 return 1;
841 }
842 }
843
844 /* We only need one spare so full list not needed */
845 list = container_choose_spares(st, min_size, domlist, from->spare_group,
846 to->metadata->ss->name, 1);
847 if (list) {
848 struct mdinfo *disks = list->devs;
849 if (disks)
850 dev = makedev(disks->disk.major, disks->disk.minor);
851 sysfs_free(list);
852 }
853 st->ss->free_super(st);
854 return dev;
855 }
856
857
858 static void try_spare_migration(struct state *statelist, struct alert_info *info)
859 {
860 struct state *from;
861 struct state *st;
862
863 link_containers_with_subarrays(statelist);
864 for (st = statelist; st; st = st->next)
865 if (st->active < st->raid &&
866 st->spare == 0 && !st->err) {
867 struct domainlist *domlist = NULL;
868 int d;
869 struct state *to = st;
870 unsigned long long min_size;
871
872 if (to->parent_dev != NoMdDev && !to->parent)
873 /* subarray monitored without parent container
874 * we can't move spares here */
875 continue;
876
877 if (to->parent)
878 /* member of a container */
879 to = to->parent;
880
881 if (get_min_spare_size_required(to, &min_size))
882 continue;
883 if (to->metadata->ss->external) {
884 /* We must make sure there is
885 * no suitable spare in container already.
886 * If there is we don't add more */
887 dev_t devid = container_choose_spare(
888 to, to, NULL, min_size, st->active);
889 if (devid > 0)
890 continue;
891 }
892 for (d = 0; d < MAX_DISKS; d++)
893 if (to->devid[d])
894 domainlist_add_dev(&domlist,
895 to->devid[d],
896 to->metadata->ss->name);
897 if (to->spare_group)
898 domain_add(&domlist, to->spare_group);
899 /*
900 * No spare migration if the destination
901 * has no domain. Skip this array.
902 */
903 if (!domlist)
904 continue;
905 for (from=statelist ; from ; from=from->next) {
906 dev_t devid;
907 if (!check_donor(from, to))
908 continue;
909 if (from->metadata->ss->external)
910 devid = container_choose_spare(
911 from, to, domlist, min_size, 0);
912 else
913 devid = choose_spare(from, to, domlist,
914 min_size);
915 if (devid > 0
916 && move_spare(from->devname, to->devname, devid)) {
917 alert("MoveSpare", to->devname, from->devname, info);
918 break;
919 }
920 }
921 domain_free(domlist);
922 }
923 }
924
925 /* search the statelist to connect external
926 * metadata subarrays with their containers
927 * We always completely rebuild the tree from scratch as
928 * that is safest considering the possibility of entries
929 * disappearing or changing.
930 */
931 static void link_containers_with_subarrays(struct state *list)
932 {
933 struct state *st;
934 struct state *cont;
935 for (st = list; st; st = st->next) {
936 st->parent = NULL;
937 st->subarray = NULL;
938 }
939 for (st = list; st; st = st->next)
940 if (st->parent_dev != NoMdDev)
941 for (cont = list; cont; cont = cont->next)
942 if (!cont->err &&
943 cont->parent_dev == NoMdDev &&
944 cont->devnum == st->parent_dev) {
945 st->parent = cont;
946 st->subarray = cont->subarray;
947 cont->subarray = st;
948 break;
949 }
950 }
951
952 /* Not really Monitor but ... */
953 int Wait(char *dev)
954 {
955 struct stat stb;
956 int devnum;
957 int rv = 1;
958
959 if (stat(dev, &stb) != 0) {
960 pr_err("Cannot find %s: %s\n", dev,
961 strerror(errno));
962 return 2;
963 }
964 devnum = stat2devnum(&stb);
965
966 while(1) {
967 struct mdstat_ent *ms = mdstat_read(1, 0);
968 struct mdstat_ent *e;
969
970 for (e=ms ; e; e=e->next)
971 if (e->devnum == devnum)
972 break;
973
974 if (!e || e->percent < 0) {
975 if (e && e->metadata_version &&
976 strncmp(e->metadata_version, "external:", 9) == 0) {
977 if (is_subarray(&e->metadata_version[9]))
978 ping_monitor(&e->metadata_version[9]);
979 else
980 ping_monitor_by_id(devnum);
981 }
982 free_mdstat(ms);
983 return rv;
984 }
985 free_mdstat(ms);
986 rv = 0;
987 mdstat_wait(5);
988 }
989 }
990
991 #ifndef MDASSEMBLE
992
993 static char *clean_states[] = {
994 "clear", "inactive", "readonly", "read-auto", "clean", NULL };
995
996 int WaitClean(char *dev, int sock, int verbose)
997 {
998 int fd;
999 struct mdinfo *mdi;
1000 int rv = 1;
1001 int devnum;
1002
1003 fd = open(dev, O_RDONLY);
1004 if (fd < 0) {
1005 if (verbose)
1006 pr_err("Couldn't open %s: %s\n", dev, strerror(errno));
1007 return 1;
1008 }
1009
1010 devnum = fd2devnum(fd);
1011 mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1012 if (!mdi) {
1013 if (verbose)
1014 pr_err("Failed to read sysfs attributes for "
1015 "%s\n", dev);
1016 close(fd);
1017 return 0;
1018 }
1019
1020 switch(mdi->array.level) {
1021 case LEVEL_LINEAR:
1022 case LEVEL_MULTIPATH:
1023 case 0:
1024 /* safemode delay is irrelevant for these levels */
1025 rv = 0;
1026 }
1027
1028 /* for internal metadata the kernel handles the final clean
1029 * transition, containers can never be dirty
1030 */
1031 if (!is_subarray(mdi->text_version))
1032 rv = 0;
1033
1034 /* safemode disabled ? */
1035 if (mdi->safe_mode_delay == 0)
1036 rv = 0;
1037
1038 if (rv) {
1039 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
1040 char buf[20];
1041 fd_set fds;
1042 struct timeval tm;
1043
1044 /* minimize the safe_mode_delay and prepare to wait up to 5s
1045 * for writes to quiesce
1046 */
1047 sysfs_set_safemode(mdi, 1);
1048 tm.tv_sec = 5;
1049 tm.tv_usec = 0;
1050
1051 FD_ZERO(&fds);
1052
1053 /* wait for array_state to be clean */
1054 while (1) {
1055 rv = read(state_fd, buf, sizeof(buf));
1056 if (rv < 0)
1057 break;
1058 if (sysfs_match_word(buf, clean_states) <= 4)
1059 break;
1060 FD_SET(state_fd, &fds);
1061 rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
1062 if (rv < 0 && errno != EINTR)
1063 break;
1064 lseek(state_fd, 0, SEEK_SET);
1065 }
1066 if (rv < 0)
1067 rv = 1;
1068 else if (fping_monitor(sock) == 0 ||
1069 ping_monitor(mdi->text_version) == 0) {
1070 /* we need to ping to close the window between array
1071 * state transitioning to clean and the metadata being
1072 * marked clean
1073 */
1074 rv = 0;
1075 } else
1076 rv = 1;
1077 if (rv && verbose)
1078 pr_err("Error waiting for %s to be clean\n",
1079 dev);
1080
1081 /* restore the original safe_mode_delay */
1082 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1083 close(state_fd);
1084 }
1085
1086 sysfs_free(mdi);
1087 close(fd);
1088
1089 return rv;
1090 }
1091 #endif /* MDASSEMBLE */
Unnamed repository; edit this file 'description' to name the repository.