]> 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
imsm: FIX: Use definitions for migration record disk operations
[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);
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)
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, increments))
225 anydegraded = 1;
226
227 /* now check if there are any new devices found in mdstat */
228 if (scan)
229 new_found = add_new_arrays(mdstat, &statelist, test,
230 &info);
231
232 /* If an array has active < raid && spare == 0 && spare_group != NULL
233 * Look for another array with spare > 0 and active == raid and same spare_group
234 * if found, choose a device and hotremove/hotadd
235 */
236 if (share && anydegraded)
237 try_spare_migration(statelist, &info);
238 if (!new_found) {
239 if (oneshot)
240 break;
241 else
242 mdstat_wait(period);
243 }
244 test = 0;
245 }
246 for (st2 = statelist; st2; st2 = statelist) {
247 statelist = st2->next;
248 free(st2);
249 }
250
251 if (pidfile)
252 unlink(pidfile);
253 return 0;
254 }
255
256 static int make_daemon(char *pidfile)
257 {
258 /* Return:
259 * -1 in the forked daemon
260 * 0 in the parent
261 * 1 on error
262 * so a none-negative becomes the exit code.
263 */
264 int pid = fork();
265 if (pid > 0) {
266 if (!pidfile)
267 printf("%d\n", pid);
268 else {
269 FILE *pid_file;
270 pid_file=fopen(pidfile, "w");
271 if (!pid_file)
272 perror("cannot create pid file");
273 else {
274 fprintf(pid_file,"%d\n", pid);
275 fclose(pid_file);
276 }
277 }
278 return 0;
279 }
280 if (pid < 0) {
281 perror("daemonise");
282 return 1;
283 }
284 close(0);
285 open("/dev/null", O_RDWR);
286 dup2(0,1);
287 dup2(0,2);
288 setsid();
289 return -1;
290 }
291
292 static int check_one_sharer(int scan)
293 {
294 int pid, rv;
295 FILE *fp;
296 char dir[20];
297 struct stat buf;
298 fp = fopen("/var/run/mdadm/autorebuild.pid", "r");
299 if (fp) {
300 if (fscanf(fp, "%d", &pid) != 1)
301 pid = -1;
302 sprintf(dir, "/proc/%d", pid);
303 rv = stat(dir, &buf);
304 if (rv != -1) {
305 if (scan) {
306 fprintf(stderr, Name ": Only one "
307 "autorebuild process allowed"
308 " in scan mode, aborting\n");
309 fclose(fp);
310 return 1;
311 } else {
312 fprintf(stderr, Name ": Warning: One"
313 " autorebuild process already"
314 " running.\n");
315 }
316 }
317 fclose(fp);
318 }
319 if (scan) {
320 if (mkdir("/var/run/mdadm", S_IRWXU) < 0 &&
321 errno != EEXIST) {
322 fprintf(stderr, Name ": Can't create "
323 "autorebuild.pid file\n");
324 } else {
325 fp = fopen("/var/run/mdadm/autorebuild.pid", "w");
326 if (!fp)
327 fprintf(stderr, Name ": Cannot create"
328 " autorebuild.pid"
329 "file\n");
330 else {
331 pid = getpid();
332 fprintf(fp, "%d\n", pid);
333 fclose(fp);
334 }
335 }
336 }
337 return 0;
338 }
339
340 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
341 {
342 int priority;
343
344 if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
345 time_t now = time(0);
346
347 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
348 }
349 if (info->alert_cmd) {
350 int pid = fork();
351 switch(pid) {
352 default:
353 waitpid(pid, NULL, 0);
354 break;
355 case -1:
356 break;
357 case 0:
358 execl(info->alert_cmd, info->alert_cmd,
359 event, dev, disc, NULL);
360 exit(2);
361 }
362 }
363 if (info->mailaddr &&
364 (strncmp(event, "Fail", 4)==0 ||
365 strncmp(event, "Test", 4)==0 ||
366 strncmp(event, "Spares", 6)==0 ||
367 strncmp(event, "Degrade", 7)==0)) {
368 FILE *mp = popen(Sendmail, "w");
369 if (mp) {
370 FILE *mdstat;
371 char hname[256];
372 gethostname(hname, sizeof(hname));
373 signal(SIGPIPE, SIG_IGN);
374 if (info->mailfrom)
375 fprintf(mp, "From: %s\n", info->mailfrom);
376 else
377 fprintf(mp, "From: " Name " monitoring <root>\n");
378 fprintf(mp, "To: %s\n", info->mailaddr);
379 fprintf(mp, "Subject: %s event on %s:%s\n\n",
380 event, dev, hname);
381
382 fprintf(mp,
383 "This is an automatically generated"
384 " mail message from " Name "\n");
385 fprintf(mp, "running on %s\n\n", hname);
386
387 fprintf(mp,
388 "A %s event had been detected on"
389 " md device %s.\n\n", event, dev);
390
391 if (disc && disc[0] != ' ')
392 fprintf(mp,
393 "It could be related to"
394 " component device %s.\n\n", disc);
395 if (disc && disc[0] == ' ')
396 fprintf(mp, "Extra information:%s.\n\n", disc);
397
398 fprintf(mp, "Faithfully yours, etc.\n");
399
400 mdstat = fopen("/proc/mdstat", "r");
401 if (mdstat) {
402 char buf[8192];
403 int n;
404 fprintf(mp,
405 "\nP.S. The /proc/mdstat file"
406 " currently contains the following:\n\n");
407 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
408 n=fwrite(buf, 1, n, mp);
409 fclose(mdstat);
410 }
411 pclose(mp);
412 }
413 }
414
415 /* log the event to syslog maybe */
416 if (info->dosyslog) {
417 /* Log at a different severity depending on the event.
418 *
419 * These are the critical events: */
420 if (strncmp(event, "Fail", 4)==0 ||
421 strncmp(event, "Degrade", 7)==0 ||
422 strncmp(event, "DeviceDisappeared", 17)==0)
423 priority = LOG_CRIT;
424 /* Good to know about, but are not failures: */
425 else if (strncmp(event, "Rebuild", 7)==0 ||
426 strncmp(event, "MoveSpare", 9)==0 ||
427 strncmp(event, "Spares", 6) != 0)
428 priority = LOG_WARNING;
429 /* Everything else: */
430 else
431 priority = LOG_INFO;
432
433 if (disc)
434 syslog(priority,
435 "%s event detected on md device %s,"
436 " component device %s", event, dev, disc);
437 else
438 syslog(priority,
439 "%s event detected on md device %s",
440 event, dev);
441 }
442 }
443
444 static int check_array(struct state *st, struct mdstat_ent *mdstat,
445 int test, struct alert_info *ainfo,
446 int increments)
447 {
448 /* Update the state 'st' to reflect any changes shown in mdstat,
449 * or found by directly examining the array, and return
450 * '1' if the array is degraded, or '0' if it is optimal (or dead).
451 */
452 struct { int state, major, minor; } info[MaxDisks];
453 mdu_array_info_t array;
454 struct mdstat_ent *mse = NULL, *mse2;
455 char *dev = st->devname;
456 int fd;
457 int i;
458 int remaining_disks;
459 int last_disk;
460
461 if (test)
462 alert("TestMessage", dev, NULL, ainfo);
463 fd = open(dev, O_RDONLY);
464 if (fd < 0) {
465 if (!st->err)
466 alert("DeviceDisappeared", dev, NULL, ainfo);
467 st->err=1;
468 return 0;
469 }
470 fcntl(fd, F_SETFD, FD_CLOEXEC);
471 if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
472 if (!st->err)
473 alert("DeviceDisappeared", dev, NULL, ainfo);
474 st->err=1;
475 close(fd);
476 return 0;
477 }
478 /* It's much easier to list what array levels can't
479 * have a device disappear than all of them that can
480 */
481 if (array.level == 0 || array.level == -1) {
482 if (!st->err)
483 alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
484 st->err = 1;
485 close(fd);
486 return 0;
487 }
488 if (st->devnum == INT_MAX) {
489 struct stat stb;
490 if (fstat(fd, &stb) == 0 &&
491 (S_IFMT&stb.st_mode)==S_IFBLK) {
492 if (major(stb.st_rdev) == MD_MAJOR)
493 st->devnum = minor(stb.st_rdev);
494 else
495 st->devnum = -1- (minor(stb.st_rdev)>>6);
496 }
497 }
498
499 for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
500 if (mse2->devnum == st->devnum) {
501 mse2->devnum = INT_MAX; /* flag it as "used" */
502 mse = mse2;
503 }
504
505 if (!mse) {
506 /* duplicated array in statelist
507 * or re-created after reading mdstat*/
508 st->err = 1;
509 close(fd);
510 return 0;
511 }
512 /* this array is in /proc/mdstat */
513 if (array.utime == 0)
514 /* external arrays don't update utime, so
515 * just make sure it is always different. */
516 array.utime = st->utime + 1;;
517
518 if (st->utime == array.utime &&
519 st->failed == array.failed_disks &&
520 st->working == array.working_disks &&
521 st->spare == array.spare_disks &&
522 (mse == NULL || (
523 mse->percent == st->percent
524 ))) {
525 close(fd);
526 st->err = 0;
527 if ((st->active < st->raid) && st->spare == 0)
528 return 1;
529 else
530 return 0;
531 }
532 if (st->utime == 0 && /* new array */
533 mse->pattern && strchr(mse->pattern, '_') /* degraded */
534 )
535 alert("DegradedArray", dev, NULL, ainfo);
536
537 if (st->utime == 0 && /* new array */
538 st->expected_spares > 0 &&
539 array.spare_disks < st->expected_spares)
540 alert("SparesMissing", dev, NULL, ainfo);
541 if (st->percent == -1 &&
542 mse->percent >= 0)
543 alert("RebuildStarted", dev, NULL, ainfo);
544 if (st->percent >= 0 &&
545 mse->percent >= 0 &&
546 (mse->percent / increments) > (st->percent / increments)) {
547 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
548
549 if((mse->percent / increments) == 0)
550 snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
551 else
552 snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
553
554 alert(percentalert, dev, NULL, ainfo);
555 }
556
557 if (mse->percent == -1 &&
558 st->percent >= 0) {
559 /* Rebuild/sync/whatever just finished.
560 * If there is a number in /mismatch_cnt,
561 * we should report that.
562 */
563 struct mdinfo *sra =
564 sysfs_read(-1, st->devnum, GET_MISMATCH);
565 if (sra && sra->mismatch_cnt > 0) {
566 char cnt[40];
567 sprintf(cnt, " mismatches found: %d (on raid level %d)",
568 sra->mismatch_cnt, array.level);
569 alert("RebuildFinished", dev, cnt, ainfo);
570 } else
571 alert("RebuildFinished", dev, NULL, ainfo);
572 if (sra)
573 free(sra);
574 }
575 st->percent = mse->percent;
576
577 remaining_disks = array.nr_disks;
578 for (i=0; i<MaxDisks && remaining_disks > 0;
579 i++) {
580 mdu_disk_info_t disc;
581 disc.number = i;
582 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
583 info[i].state = disc.state;
584 info[i].major = disc.major;
585 info[i].minor = disc.minor;
586 if (disc.major || disc.minor)
587 remaining_disks --;
588 } else
589 info[i].major = info[i].minor = 0;
590 }
591 last_disk = i;
592
593 if (mse->metadata_version &&
594 strncmp(mse->metadata_version, "external:", 9) == 0 &&
595 is_subarray(mse->metadata_version+9))
596 st->parent_dev =
597 devname2devnum(mse->metadata_version+10);
598 else
599 st->parent_dev = NoMdDev;
600 if (st->metadata == NULL &&
601 st->parent_dev == NoMdDev)
602 st->metadata = super_by_fd(fd, NULL);
603
604 close(fd);
605
606 for (i=0; i<MaxDisks; i++) {
607 mdu_disk_info_t disc = {0,0,0,0,0};
608 int newstate=0;
609 int change;
610 char *dv = NULL;
611 disc.number = i;
612 if (i >= last_disk) {
613 newstate = 0;
614 disc.major = disc.minor = 0;
615 } else if (info[i].major || info[i].minor) {
616 newstate = info[i].state;
617 dv = map_dev(info[i].major, info[i].minor, 1);
618 disc.state = newstate;
619 disc.major = info[i].major;
620 disc.minor = info[i].minor;
621 } else if (mse && mse->pattern && i < (int)strlen(mse->pattern)) {
622 switch(mse->pattern[i]) {
623 case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
624 case '_': newstate = 8 /* REMOVED */; break;
625 }
626 disc.major = disc.minor = 0;
627 }
628 if (dv == NULL && st->devid[i])
629 dv = map_dev(major(st->devid[i]),
630 minor(st->devid[i]), 1);
631 change = newstate ^ st->devstate[i];
632 if (st->utime && change && !st->err) {
633 if (i < array.raid_disks &&
634 (((newstate&change)&(1<<MD_DISK_FAULTY)) ||
635 ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) ||
636 ((st->devstate[i]&change)&(1<<MD_DISK_SYNC)))
637 )
638 alert("Fail", dev, dv, ainfo);
639 else if (i >= array.raid_disks &&
640 (disc.major || disc.minor) &&
641 st->devid[i] == makedev(disc.major, disc.minor) &&
642 ((newstate&change)&(1<<MD_DISK_FAULTY))
643 )
644 alert("FailSpare", dev, dv, ainfo);
645 else if (i < array.raid_disks &&
646 ! (newstate & (1<<MD_DISK_REMOVED)) &&
647 (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) ||
648 ((newstate&change)&(1<<MD_DISK_ACTIVE)) ||
649 ((newstate&change)&(1<<MD_DISK_SYNC)))
650 )
651 alert("SpareActive", dev, dv, ainfo);
652 }
653 st->devstate[i] = newstate;
654 st->devid[i] = makedev(disc.major, disc.minor);
655 }
656 st->active = array.active_disks;
657 st->working = array.working_disks;
658 st->spare = array.spare_disks;
659 st->failed = array.failed_disks;
660 st->utime = array.utime;
661 st->raid = array.raid_disks;
662 st->err = 0;
663 if ((st->active < st->raid) && st->spare == 0)
664 return 1;
665 return 0;
666 }
667
668 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
669 int test, struct alert_info *info)
670 {
671 struct mdstat_ent *mse;
672 int new_found = 0;
673
674 for (mse=mdstat; mse; mse=mse->next)
675 if (mse->devnum != INT_MAX &&
676 (!mse->level || /* retrieve containers */
677 (strcmp(mse->level, "raid0") != 0 &&
678 strcmp(mse->level, "linear") != 0))
679 ) {
680 struct state *st = calloc(1, sizeof *st);
681 mdu_array_info_t array;
682 int fd;
683 if (st == NULL)
684 continue;
685 st->devname = strdup(get_md_name(mse->devnum));
686 if ((fd = open(st->devname, O_RDONLY)) < 0 ||
687 ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
688 /* no such array */
689 if (fd >=0) close(fd);
690 put_md_name(st->devname);
691 free(st->devname);
692 if (st->metadata) {
693 st->metadata->ss->free_super(st->metadata);
694 free(st->metadata);
695 }
696 free(st);
697 continue;
698 }
699 close(fd);
700 st->next = *statelist;
701 st->err = 1;
702 st->devnum = mse->devnum;
703 st->percent = -2;
704 st->expected_spares = -1;
705 if (mse->metadata_version &&
706 strncmp(mse->metadata_version, "external:", 9) == 0 &&
707 is_subarray(mse->metadata_version+9))
708 st->parent_dev =
709 devname2devnum(mse->metadata_version+10);
710 else
711 st->parent_dev = NoMdDev;
712 *statelist = st;
713 if (test)
714 alert("TestMessage", st->devname, NULL, info);
715 alert("NewArray", st->devname, NULL, info);
716 new_found = 1;
717 }
718 return new_found;
719 }
720
721 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
722 {
723 int fd;
724
725 if (!st->metadata ||
726 !st->metadata->ss->min_acceptable_spare_size) {
727 *sizep = 0;
728 return 0;
729 }
730
731 fd = open(st->devname, O_RDONLY);
732 if (fd < 0)
733 return 1;
734 if (st->metadata->ss->external)
735 st->metadata->ss->load_container(st->metadata, fd, st->devname);
736 else
737 st->metadata->ss->load_super(st->metadata, fd, st->devname);
738 close(fd);
739 if (!st->metadata->sb)
740 return 1;
741 *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
742 st->metadata->ss->free_super(st->metadata);
743
744 return 0;
745 }
746
747 static int check_donor(struct state *from, struct state *to)
748 {
749 struct state *sub;
750
751 if (from == to)
752 return 0;
753 if (from->parent)
754 /* Cannot move from a member */
755 return 0;
756 if (from->err)
757 return 0;
758 for (sub = from->subarray; sub; sub = sub->subarray)
759 /* If source array has degraded subarrays, don't
760 * remove anything
761 */
762 if (sub->active < sub->raid)
763 return 0;
764 if (from->metadata->ss->external == 0)
765 if (from->active < from->raid)
766 return 0;
767 if (from->spare <= 0)
768 return 0;
769 return 1;
770 }
771
772 static dev_t choose_spare(struct state *from, struct state *to,
773 struct domainlist *domlist, unsigned long long min_size)
774 {
775 int d;
776 dev_t dev = 0;
777
778 for (d = from->raid; !dev && d < MaxDisks; d++) {
779 if (from->devid[d] > 0 &&
780 from->devstate[d] == 0) {
781 struct dev_policy *pol;
782 unsigned long long dev_size;
783
784 if (to->metadata->ss->external &&
785 test_partition_from_id(from->devid[d]))
786 continue;
787
788 if (min_size &&
789 dev_size_from_id(from->devid[d], &dev_size) &&
790 dev_size < min_size)
791 continue;
792
793 pol = devnum_policy(from->devid[d]);
794 if (from->spare_group)
795 pol_add(&pol, pol_domain,
796 from->spare_group, NULL);
797 if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
798 dev = from->devid[d];
799 dev_policy_free(pol);
800 }
801 }
802 return dev;
803 }
804
805 static dev_t container_choose_spare(struct state *from, struct state *to,
806 struct domainlist *domlist,
807 unsigned long long min_size, int active)
808 {
809 /* This is similar to choose_spare, but we cannot trust devstate,
810 * so we need to read the metadata instead
811 */
812 struct mdinfo *list;
813 struct supertype *st = from->metadata;
814 int fd = open(from->devname, O_RDONLY);
815 int err;
816 dev_t dev = 0;
817
818 if (fd < 0)
819 return 0;
820 if (!st->ss->getinfo_super_disks) {
821 close(fd);
822 return 0;
823 }
824
825 err = st->ss->load_container(st, fd, NULL);
826 close(fd);
827 if (err)
828 return 0;
829
830 if (from == to) {
831 /* We must check if number of active disks has not increased
832 * since ioctl in main loop. mdmon may have added spare
833 * to subarray. If so we do not need to look for more spares
834 * so return non zero value */
835 int active_cnt = 0;
836 struct mdinfo *dp;
837 list = st->ss->getinfo_super_disks(st);
838 if (!list) {
839 st->ss->free_super(st);
840 return 1;
841 }
842 dp = list->devs;
843 while (dp) {
844 if (dp->disk.state & (1<<MD_DISK_SYNC) &&
845 !(dp->disk.state & (1<<MD_DISK_FAULTY)))
846 active_cnt++;
847 dp = dp->next;
848 }
849 sysfs_free(list);
850 if (active < active_cnt) {
851 /* Spare just activated.*/
852 st->ss->free_super(st);
853 return 1;
854 }
855 }
856
857 /* We only need one spare so full list not needed */
858 list = container_choose_spares(st, min_size, domlist, from->spare_group,
859 to->metadata->ss->name, 1);
860 if (list) {
861 struct mdinfo *disks = list->devs;
862 if (disks)
863 dev = makedev(disks->disk.major, disks->disk.minor);
864 sysfs_free(list);
865 }
866 st->ss->free_super(st);
867 return dev;
868 }
869
870
871 static void try_spare_migration(struct state *statelist, struct alert_info *info)
872 {
873 struct state *from;
874 struct state *st;
875
876 link_containers_with_subarrays(statelist);
877 for (st = statelist; st; st = st->next)
878 if (st->active < st->raid &&
879 st->spare == 0 && !st->err) {
880 struct domainlist *domlist = NULL;
881 int d;
882 struct state *to = st;
883 unsigned long long min_size;
884
885 if (to->parent_dev != NoMdDev && !to->parent)
886 /* subarray monitored without parent container
887 * we can't move spares here */
888 continue;
889
890 if (to->parent)
891 /* member of a container */
892 to = to->parent;
893
894 if (get_min_spare_size_required(to, &min_size))
895 continue;
896 if (to->metadata->ss->external) {
897 /* We must make sure there is
898 * no suitable spare in container already.
899 * If there is we don't add more */
900 dev_t devid = container_choose_spare(
901 to, to, NULL, min_size, st->active);
902 if (devid > 0)
903 continue;
904 }
905 for (d = 0; d < MaxDisks; d++)
906 if (to->devid[d])
907 domainlist_add_dev(&domlist,
908 to->devid[d],
909 to->metadata->ss->name);
910 if (to->spare_group)
911 domain_add(&domlist, to->spare_group);
912 /*
913 * No spare migration if the destination
914 * has no domain. Skip this array.
915 */
916 if (!domlist)
917 continue;
918 for (from=statelist ; from ; from=from->next) {
919 dev_t devid;
920 if (!check_donor(from, to))
921 continue;
922 if (from->metadata->ss->external)
923 devid = container_choose_spare(
924 from, to, domlist, min_size, 0);
925 else
926 devid = choose_spare(from, to, domlist,
927 min_size);
928 if (devid > 0
929 && move_spare(from->devname, to->devname, devid)) {
930 alert("MoveSpare", to->devname, from->devname, info);
931 break;
932 }
933 }
934 domain_free(domlist);
935 }
936 }
937
938 /* search the statelist to connect external
939 * metadata subarrays with their containers
940 * We always completely rebuild the tree from scratch as
941 * that is safest considering the possibility of entries
942 * disappearing or changing.
943 */
944 static void link_containers_with_subarrays(struct state *list)
945 {
946 struct state *st;
947 struct state *cont;
948 for (st = list; st; st = st->next) {
949 st->parent = NULL;
950 st->subarray = NULL;
951 }
952 for (st = list; st; st = st->next)
953 if (st->parent_dev != NoMdDev)
954 for (cont = list; cont; cont = cont->next)
955 if (!cont->err &&
956 cont->parent_dev == NoMdDev &&
957 cont->devnum == st->parent_dev) {
958 st->parent = cont;
959 st->subarray = cont->subarray;
960 cont->subarray = st;
961 break;
962 }
963 }
964
965 /* Not really Monitor but ... */
966 int Wait(char *dev)
967 {
968 struct stat stb;
969 int devnum;
970 int rv = 1;
971
972 if (stat(dev, &stb) != 0) {
973 fprintf(stderr, Name ": Cannot find %s: %s\n", dev,
974 strerror(errno));
975 return 2;
976 }
977 devnum = stat2devnum(&stb);
978
979 while(1) {
980 struct mdstat_ent *ms = mdstat_read(1, 0);
981 struct mdstat_ent *e;
982
983 for (e=ms ; e; e=e->next)
984 if (e->devnum == devnum)
985 break;
986
987 if (!e || e->percent < 0) {
988 if (e && e->metadata_version &&
989 strncmp(e->metadata_version, "external:", 9) == 0) {
990 if (is_subarray(&e->metadata_version[9]))
991 ping_monitor(&e->metadata_version[9]);
992 else
993 ping_monitor_by_id(devnum);
994 }
995 free_mdstat(ms);
996 return rv;
997 }
998 free_mdstat(ms);
999 rv = 0;
1000 mdstat_wait(5);
1001 }
1002 }
1003
1004 #ifndef MDASSEMBLE
1005
1006 static char *clean_states[] = {
1007 "clear", "inactive", "readonly", "read-auto", "clean", NULL };
1008
1009 int WaitClean(char *dev, int sock, int verbose)
1010 {
1011 int fd;
1012 struct mdinfo *mdi;
1013 int rv = 1;
1014 int devnum;
1015
1016 fd = open(dev, O_RDONLY);
1017 if (fd < 0) {
1018 if (verbose)
1019 fprintf(stderr, Name ": Couldn't open %s: %s\n", dev, strerror(errno));
1020 return 1;
1021 }
1022
1023 devnum = fd2devnum(fd);
1024 mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1025 if (!mdi) {
1026 if (verbose)
1027 fprintf(stderr, Name ": Failed to read sysfs attributes for "
1028 "%s\n", dev);
1029 close(fd);
1030 return 0;
1031 }
1032
1033 switch(mdi->array.level) {
1034 case LEVEL_LINEAR:
1035 case LEVEL_MULTIPATH:
1036 case 0:
1037 /* safemode delay is irrelevant for these levels */
1038 rv = 0;
1039 }
1040
1041 /* for internal metadata the kernel handles the final clean
1042 * transition, containers can never be dirty
1043 */
1044 if (!is_subarray(mdi->text_version))
1045 rv = 0;
1046
1047 /* safemode disabled ? */
1048 if (mdi->safe_mode_delay == 0)
1049 rv = 0;
1050
1051 if (rv) {
1052 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
1053 char buf[20];
1054 fd_set fds;
1055 struct timeval tm;
1056
1057 /* minimize the safe_mode_delay and prepare to wait up to 5s
1058 * for writes to quiesce
1059 */
1060 sysfs_set_safemode(mdi, 1);
1061 tm.tv_sec = 5;
1062 tm.tv_usec = 0;
1063
1064 FD_ZERO(&fds);
1065
1066 /* wait for array_state to be clean */
1067 while (1) {
1068 rv = read(state_fd, buf, sizeof(buf));
1069 if (rv < 0)
1070 break;
1071 if (sysfs_match_word(buf, clean_states) <= 4)
1072 break;
1073 FD_SET(state_fd, &fds);
1074 rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
1075 if (rv < 0 && errno != EINTR)
1076 break;
1077 lseek(state_fd, 0, SEEK_SET);
1078 }
1079 if (rv < 0)
1080 rv = 1;
1081 else if (fping_monitor(sock) == 0 ||
1082 ping_monitor(mdi->text_version) == 0) {
1083 /* we need to ping to close the window between array
1084 * state transitioning to clean and the metadata being
1085 * marked clean
1086 */
1087 rv = 0;
1088 } else
1089 rv = 1;
1090 if (rv && verbose)
1091 fprintf(stderr, Name ": Error waiting for %s to be clean\n",
1092 dev);
1093
1094 /* restore the original safe_mode_delay */
1095 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1096 close(state_fd);
1097 }
1098
1099 sysfs_free(mdi);
1100 close(fd);
1101
1102 return rv;
1103 }
1104 #endif /* MDASSEMBLE */
Unnamed repository; edit this file 'description' to name the repository.