]> 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
Remove scattered checks for malloc success.
[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 pr_err("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 pr_err("Monitor using program \"%s\" from config file\n",
140 alert_cmd);
141 }
142 if (scan && !mailaddr && !alert_cmd && !dosyslog) {
143 pr_err("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 = xcalloc(1, sizeof *st);
170 if (mdlist->devname[0] == '/')
171 st->devname = xstrdup(mdlist->devname);
172 else {
173 st->devname = xmalloc(8+strlen(mdlist->devname)+1);
174 strcpy(strcpy(st->devname, "/dev/md/"),
175 mdlist->devname);
176 }
177 st->next = statelist;
178 st->devnum = INT_MAX;
179 st->percent = RESYNC_UNKNOWN;
180 st->expected_spares = mdlist->spare_disks;
181 if (mdlist->spare_group)
182 st->spare_group = xstrdup(mdlist->spare_group);
183 statelist = st;
184 }
185 } else {
186 struct mddev_dev *dv;
187 for (dv=devlist ; dv; dv=dv->next) {
188 struct mddev_ident *mdlist = conf_get_ident(dv->devname);
189 struct state *st = xcalloc(1, sizeof *st);
190 st->devname = xstrdup(dv->devname);
191 st->next = statelist;
192 st->devnum = INT_MAX;
193 st->percent = RESYNC_UNKNOWN;
194 st->expected_spares = -1;
195 if (mdlist) {
196 st->expected_spares = mdlist->spare_disks;
197 if (mdlist->spare_group)
198 st->spare_group = xstrdup(mdlist->spare_group);
199 }
200 statelist = st;
201 }
202 }
203
204
205 while (! finished) {
206 int new_found = 0;
207 struct state *st;
208 int anydegraded = 0;
209
210 if (mdstat)
211 free_mdstat(mdstat);
212 mdstat = mdstat_read(oneshot?0:1, 0);
213
214 for (st=statelist; st; st=st->next)
215 if (check_array(st, mdstat, test, &info,
216 increments, prefer))
217 anydegraded = 1;
218
219 /* now check if there are any new devices found in mdstat */
220 if (scan)
221 new_found = add_new_arrays(mdstat, &statelist, test,
222 &info);
223
224 /* If an array has active < raid && spare == 0 && spare_group != NULL
225 * Look for another array with spare > 0 and active == raid and same spare_group
226 * if found, choose a device and hotremove/hotadd
227 */
228 if (share && anydegraded)
229 try_spare_migration(statelist, &info);
230 if (!new_found) {
231 if (oneshot)
232 break;
233 else
234 mdstat_wait(period);
235 }
236 test = 0;
237 }
238 for (st2 = statelist; st2; st2 = statelist) {
239 statelist = st2->next;
240 free(st2);
241 }
242
243 if (pidfile)
244 unlink(pidfile);
245 return 0;
246 }
247
248 static int make_daemon(char *pidfile)
249 {
250 /* Return:
251 * -1 in the forked daemon
252 * 0 in the parent
253 * 1 on error
254 * so a none-negative becomes the exit code.
255 */
256 int pid = fork();
257 if (pid > 0) {
258 if (!pidfile)
259 printf("%d\n", pid);
260 else {
261 FILE *pid_file;
262 pid_file=fopen(pidfile, "w");
263 if (!pid_file)
264 perror("cannot create pid file");
265 else {
266 fprintf(pid_file,"%d\n", pid);
267 fclose(pid_file);
268 }
269 }
270 return 0;
271 }
272 if (pid < 0) {
273 perror("daemonise");
274 return 1;
275 }
276 close(0);
277 open("/dev/null", O_RDWR);
278 dup2(0,1);
279 dup2(0,2);
280 setsid();
281 return -1;
282 }
283
284 static int check_one_sharer(int scan)
285 {
286 int pid, rv;
287 FILE *fp;
288 char dir[20];
289 char path[100];
290 struct stat buf;
291 sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
292 fp = fopen(path, "r");
293 if (fp) {
294 if (fscanf(fp, "%d", &pid) != 1)
295 pid = -1;
296 sprintf(dir, "/proc/%d", pid);
297 rv = stat(dir, &buf);
298 if (rv != -1) {
299 if (scan) {
300 pr_err("Only one "
301 "autorebuild process allowed"
302 " in scan mode, aborting\n");
303 fclose(fp);
304 return 1;
305 } else {
306 pr_err("Warning: One"
307 " autorebuild process already"
308 " running.\n");
309 }
310 }
311 fclose(fp);
312 }
313 if (scan) {
314 if (mkdir(MDMON_DIR, S_IRWXU) < 0 &&
315 errno != EEXIST) {
316 pr_err("Can't create "
317 "autorebuild.pid file\n");
318 } else {
319 fp = fopen(path, "w");
320 if (!fp)
321 pr_err("Cannot create"
322 " autorebuild.pid"
323 "file\n");
324 else {
325 pid = getpid();
326 fprintf(fp, "%d\n", pid);
327 fclose(fp);
328 }
329 }
330 }
331 return 0;
332 }
333
334 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
335 {
336 int priority;
337
338 if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
339 time_t now = time(0);
340
341 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
342 }
343 if (info->alert_cmd) {
344 int pid = fork();
345 switch(pid) {
346 default:
347 waitpid(pid, NULL, 0);
348 break;
349 case -1:
350 break;
351 case 0:
352 execl(info->alert_cmd, info->alert_cmd,
353 event, dev, disc, NULL);
354 exit(2);
355 }
356 }
357 if (info->mailaddr &&
358 (strncmp(event, "Fail", 4)==0 ||
359 strncmp(event, "Test", 4)==0 ||
360 strncmp(event, "Spares", 6)==0 ||
361 strncmp(event, "Degrade", 7)==0)) {
362 FILE *mp = popen(Sendmail, "w");
363 if (mp) {
364 FILE *mdstat;
365 char hname[256];
366 gethostname(hname, sizeof(hname));
367 signal(SIGPIPE, SIG_IGN);
368 if (info->mailfrom)
369 fprintf(mp, "From: %s\n", info->mailfrom);
370 else
371 fprintf(mp, "From: " Name " monitoring <root>\n");
372 fprintf(mp, "To: %s\n", info->mailaddr);
373 fprintf(mp, "Subject: %s event on %s:%s\n\n",
374 event, dev, hname);
375
376 fprintf(mp,
377 "This is an automatically generated"
378 " mail message from " Name "\n");
379 fprintf(mp, "running on %s\n\n", hname);
380
381 fprintf(mp,
382 "A %s event had been detected on"
383 " md device %s.\n\n", event, dev);
384
385 if (disc && disc[0] != ' ')
386 fprintf(mp,
387 "It could be related to"
388 " component device %s.\n\n", disc);
389 if (disc && disc[0] == ' ')
390 fprintf(mp, "Extra information:%s.\n\n", disc);
391
392 fprintf(mp, "Faithfully yours, etc.\n");
393
394 mdstat = fopen("/proc/mdstat", "r");
395 if (mdstat) {
396 char buf[8192];
397 int n;
398 fprintf(mp,
399 "\nP.S. The /proc/mdstat file"
400 " currently contains the following:\n\n");
401 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
402 n=fwrite(buf, 1, n, mp);
403 fclose(mdstat);
404 }
405 pclose(mp);
406 }
407 }
408
409 /* log the event to syslog maybe */
410 if (info->dosyslog) {
411 /* Log at a different severity depending on the event.
412 *
413 * These are the critical events: */
414 if (strncmp(event, "Fail", 4)==0 ||
415 strncmp(event, "Degrade", 7)==0 ||
416 strncmp(event, "DeviceDisappeared", 17)==0)
417 priority = LOG_CRIT;
418 /* Good to know about, but are not failures: */
419 else if (strncmp(event, "Rebuild", 7)==0 ||
420 strncmp(event, "MoveSpare", 9)==0 ||
421 strncmp(event, "Spares", 6) != 0)
422 priority = LOG_WARNING;
423 /* Everything else: */
424 else
425 priority = LOG_INFO;
426
427 if (disc)
428 syslog(priority,
429 "%s event detected on md device %s,"
430 " component device %s", event, dev, disc);
431 else
432 syslog(priority,
433 "%s event detected on md device %s",
434 event, dev);
435 }
436 }
437
438 static int check_array(struct state *st, struct mdstat_ent *mdstat,
439 int test, struct alert_info *ainfo,
440 int increments, char *prefer)
441 {
442 /* Update the state 'st' to reflect any changes shown in mdstat,
443 * or found by directly examining the array, and return
444 * '1' if the array is degraded, or '0' if it is optimal (or dead).
445 */
446 struct { int state, major, minor; } info[MAX_DISKS];
447 mdu_array_info_t array;
448 struct mdstat_ent *mse = NULL, *mse2;
449 char *dev = st->devname;
450 int fd;
451 int i;
452 int remaining_disks;
453 int last_disk;
454 int new_array = 0;
455
456 if (test)
457 alert("TestMessage", dev, NULL, ainfo);
458 fd = open(dev, O_RDONLY);
459 if (fd < 0) {
460 if (!st->err)
461 alert("DeviceDisappeared", dev, NULL, ainfo);
462 st->err=1;
463 return 0;
464 }
465 fcntl(fd, F_SETFD, FD_CLOEXEC);
466 if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
467 if (!st->err)
468 alert("DeviceDisappeared", dev, NULL, ainfo);
469 st->err=1;
470 close(fd);
471 return 0;
472 }
473 /* It's much easier to list what array levels can't
474 * have a device disappear than all of them that can
475 */
476 if (array.level == 0 || array.level == -1) {
477 if (!st->err)
478 alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
479 st->err = 1;
480 close(fd);
481 return 0;
482 }
483 if (st->devnum == INT_MAX) {
484 struct stat stb;
485 if (fstat(fd, &stb) == 0 &&
486 (S_IFMT&stb.st_mode)==S_IFBLK) {
487 if (major(stb.st_rdev) == MD_MAJOR)
488 st->devnum = minor(stb.st_rdev);
489 else
490 st->devnum = -1- (minor(stb.st_rdev)>>6);
491 }
492 }
493
494 for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
495 if (mse2->devnum == st->devnum) {
496 mse2->devnum = INT_MAX; /* flag it as "used" */
497 mse = mse2;
498 }
499
500 if (!mse) {
501 /* duplicated array in statelist
502 * or re-created after reading mdstat*/
503 st->err = 1;
504 close(fd);
505 return 0;
506 }
507 /* this array is in /proc/mdstat */
508 if (array.utime == 0)
509 /* external arrays don't update utime, so
510 * just make sure it is always different. */
511 array.utime = st->utime + 1;;
512
513 if (st->err) {
514 /* New array appeared where previously had and error */
515 st->err = 0;
516 st->percent = RESYNC_NONE;
517 new_array = 1;
518 alert("NewArray", st->devname, NULL, ainfo);
519 }
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 if ((st->active < st->raid) && st->spare == 0)
530 return 1;
531 else
532 return 0;
533 }
534 if (st->utime == 0 && /* new array */
535 mse->pattern && strchr(mse->pattern, '_') /* degraded */
536 )
537 alert("DegradedArray", dev, NULL, ainfo);
538
539 if (st->utime == 0 && /* new array */
540 st->expected_spares > 0 &&
541 array.spare_disks < st->expected_spares)
542 alert("SparesMissing", dev, NULL, ainfo);
543 if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
544 mse->percent >= 0)
545 alert("RebuildStarted", dev, NULL, ainfo);
546 if (st->percent >= 0 &&
547 mse->percent >= 0 &&
548 (mse->percent / increments) > (st->percent / increments)) {
549 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
550
551 if((mse->percent / increments) == 0)
552 snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
553 else
554 snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
555
556 alert(percentalert, dev, NULL, ainfo);
557 }
558
559 if (mse->percent == RESYNC_NONE &&
560 st->percent >= 0) {
561 /* Rebuild/sync/whatever just finished.
562 * If there is a number in /mismatch_cnt,
563 * we should report that.
564 */
565 struct mdinfo *sra =
566 sysfs_read(-1, st->devnum, GET_MISMATCH);
567 if (sra && sra->mismatch_cnt > 0) {
568 char cnt[80];
569 snprintf(cnt, sizeof(cnt),
570 " mismatches found: %d (on raid level %d)",
571 sra->mismatch_cnt, array.level);
572 alert("RebuildFinished", dev, cnt, ainfo);
573 } else
574 alert("RebuildFinished", dev, NULL, ainfo);
575 if (sra)
576 free(sra);
577 }
578 st->percent = mse->percent;
579
580 remaining_disks = array.nr_disks;
581 for (i=0; i<MAX_DISKS && remaining_disks > 0;
582 i++) {
583 mdu_disk_info_t disc;
584 disc.number = i;
585 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
586 info[i].state = disc.state;
587 info[i].major = disc.major;
588 info[i].minor = disc.minor;
589 if (disc.major || disc.minor)
590 remaining_disks --;
591 } else
592 info[i].major = info[i].minor = 0;
593 }
594 last_disk = i;
595
596 if (mse->metadata_version &&
597 strncmp(mse->metadata_version, "external:", 9) == 0 &&
598 is_subarray(mse->metadata_version+9))
599 st->parent_dev =
600 devname2devnum(mse->metadata_version+10);
601 else
602 st->parent_dev = NoMdDev;
603 if (st->metadata == NULL &&
604 st->parent_dev == NoMdDev)
605 st->metadata = super_by_fd(fd, NULL);
606
607 close(fd);
608
609 for (i=0; i<MAX_DISKS; i++) {
610 mdu_disk_info_t disc = {0,0,0,0,0};
611 int newstate=0;
612 int change;
613 char *dv = NULL;
614 disc.number = i;
615 if (i < last_disk &&
616 (info[i].major || info[i].minor)) {
617 newstate = info[i].state;
618 dv = map_dev_preferred(
619 info[i].major, info[i].minor, 1,
620 prefer);
621 disc.state = newstate;
622 disc.major = info[i].major;
623 disc.minor = info[i].minor;
624 } else
625 newstate = (1 << MD_DISK_REMOVED);
626
627 if (dv == NULL && st->devid[i])
628 dv = map_dev_preferred(
629 major(st->devid[i]),
630 minor(st->devid[i]), 1, prefer);
631 change = newstate ^ st->devstate[i];
632 if (st->utime && change && !st->err && !new_array) {
633 if ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))
634 alert("Fail", dev, dv, ainfo);
635 else if ((newstate & (1<<MD_DISK_FAULTY)) &&
636 (disc.major || disc.minor) &&
637 st->devid[i] == makedev(disc.major, disc.minor))
638 alert("FailSpare", dev, dv, ainfo);
639 else if ((newstate&change)&(1<<MD_DISK_SYNC))
640 alert("SpareActive", dev, dv, ainfo);
641 }
642 st->devstate[i] = newstate;
643 st->devid[i] = makedev(disc.major, disc.minor);
644 }
645 st->active = array.active_disks;
646 st->working = array.working_disks;
647 st->spare = array.spare_disks;
648 st->failed = array.failed_disks;
649 st->utime = array.utime;
650 st->raid = array.raid_disks;
651 st->err = 0;
652 if ((st->active < st->raid) && st->spare == 0)
653 return 1;
654 return 0;
655 }
656
657 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
658 int test, struct alert_info *info)
659 {
660 struct mdstat_ent *mse;
661 int new_found = 0;
662
663 for (mse=mdstat; mse; mse=mse->next)
664 if (mse->devnum != INT_MAX &&
665 (!mse->level || /* retrieve containers */
666 (strcmp(mse->level, "raid0") != 0 &&
667 strcmp(mse->level, "linear") != 0))
668 ) {
669 struct state *st = xcalloc(1, sizeof *st);
670 mdu_array_info_t array;
671 int fd;
672 st->devname = xstrdup(get_md_name(mse->devnum));
673 if ((fd = open(st->devname, O_RDONLY)) < 0 ||
674 ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
675 /* no such array */
676 if (fd >=0) close(fd);
677 put_md_name(st->devname);
678 free(st->devname);
679 if (st->metadata) {
680 st->metadata->ss->free_super(st->metadata);
681 free(st->metadata);
682 }
683 free(st);
684 continue;
685 }
686 close(fd);
687 st->next = *statelist;
688 st->err = 1;
689 st->devnum = mse->devnum;
690 st->percent = RESYNC_UNKNOWN;
691 st->expected_spares = -1;
692 if (mse->metadata_version &&
693 strncmp(mse->metadata_version, "external:", 9) == 0 &&
694 is_subarray(mse->metadata_version+9))
695 st->parent_dev =
696 devname2devnum(mse->metadata_version+10);
697 else
698 st->parent_dev = NoMdDev;
699 *statelist = st;
700 if (test)
701 alert("TestMessage", st->devname, NULL, info);
702 new_found = 1;
703 }
704 return new_found;
705 }
706
707 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
708 {
709 int fd;
710
711 if (!st->metadata ||
712 !st->metadata->ss->min_acceptable_spare_size) {
713 *sizep = 0;
714 return 0;
715 }
716
717 fd = open(st->devname, O_RDONLY);
718 if (fd < 0)
719 return 1;
720 if (st->metadata->ss->external)
721 st->metadata->ss->load_container(st->metadata, fd, st->devname);
722 else
723 st->metadata->ss->load_super(st->metadata, fd, st->devname);
724 close(fd);
725 if (!st->metadata->sb)
726 return 1;
727 *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
728 st->metadata->ss->free_super(st->metadata);
729
730 return 0;
731 }
732
733 static int check_donor(struct state *from, struct state *to)
734 {
735 struct state *sub;
736
737 if (from == to)
738 return 0;
739 if (from->parent)
740 /* Cannot move from a member */
741 return 0;
742 if (from->err)
743 return 0;
744 for (sub = from->subarray; sub; sub = sub->subarray)
745 /* If source array has degraded subarrays, don't
746 * remove anything
747 */
748 if (sub->active < sub->raid)
749 return 0;
750 if (from->metadata->ss->external == 0)
751 if (from->active < from->raid)
752 return 0;
753 if (from->spare <= 0)
754 return 0;
755 return 1;
756 }
757
758 static dev_t choose_spare(struct state *from, struct state *to,
759 struct domainlist *domlist, unsigned long long min_size)
760 {
761 int d;
762 dev_t dev = 0;
763
764 for (d = from->raid; !dev && d < MAX_DISKS; d++) {
765 if (from->devid[d] > 0 &&
766 from->devstate[d] == 0) {
767 struct dev_policy *pol;
768 unsigned long long dev_size;
769
770 if (to->metadata->ss->external &&
771 test_partition_from_id(from->devid[d]))
772 continue;
773
774 if (min_size &&
775 dev_size_from_id(from->devid[d], &dev_size) &&
776 dev_size < min_size)
777 continue;
778
779 pol = devnum_policy(from->devid[d]);
780 if (from->spare_group)
781 pol_add(&pol, pol_domain,
782 from->spare_group, NULL);
783 if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
784 dev = from->devid[d];
785 dev_policy_free(pol);
786 }
787 }
788 return dev;
789 }
790
791 static dev_t container_choose_spare(struct state *from, struct state *to,
792 struct domainlist *domlist,
793 unsigned long long min_size, int active)
794 {
795 /* This is similar to choose_spare, but we cannot trust devstate,
796 * so we need to read the metadata instead
797 */
798 struct mdinfo *list;
799 struct supertype *st = from->metadata;
800 int fd = open(from->devname, O_RDONLY);
801 int err;
802 dev_t dev = 0;
803
804 if (fd < 0)
805 return 0;
806 if (!st->ss->getinfo_super_disks) {
807 close(fd);
808 return 0;
809 }
810
811 err = st->ss->load_container(st, fd, NULL);
812 close(fd);
813 if (err)
814 return 0;
815
816 if (from == to) {
817 /* We must check if number of active disks has not increased
818 * since ioctl in main loop. mdmon may have added spare
819 * to subarray. If so we do not need to look for more spares
820 * so return non zero value */
821 int active_cnt = 0;
822 struct mdinfo *dp;
823 list = st->ss->getinfo_super_disks(st);
824 if (!list) {
825 st->ss->free_super(st);
826 return 1;
827 }
828 dp = list->devs;
829 while (dp) {
830 if (dp->disk.state & (1<<MD_DISK_SYNC) &&
831 !(dp->disk.state & (1<<MD_DISK_FAULTY)))
832 active_cnt++;
833 dp = dp->next;
834 }
835 sysfs_free(list);
836 if (active < active_cnt) {
837 /* Spare just activated.*/
838 st->ss->free_super(st);
839 return 1;
840 }
841 }
842
843 /* We only need one spare so full list not needed */
844 list = container_choose_spares(st, min_size, domlist, from->spare_group,
845 to->metadata->ss->name, 1);
846 if (list) {
847 struct mdinfo *disks = list->devs;
848 if (disks)
849 dev = makedev(disks->disk.major, disks->disk.minor);
850 sysfs_free(list);
851 }
852 st->ss->free_super(st);
853 return dev;
854 }
855
856
857 static void try_spare_migration(struct state *statelist, struct alert_info *info)
858 {
859 struct state *from;
860 struct state *st;
861
862 link_containers_with_subarrays(statelist);
863 for (st = statelist; st; st = st->next)
864 if (st->active < st->raid &&
865 st->spare == 0 && !st->err) {
866 struct domainlist *domlist = NULL;
867 int d;
868 struct state *to = st;
869 unsigned long long min_size;
870
871 if (to->parent_dev != NoMdDev && !to->parent)
872 /* subarray monitored without parent container
873 * we can't move spares here */
874 continue;
875
876 if (to->parent)
877 /* member of a container */
878 to = to->parent;
879
880 if (get_min_spare_size_required(to, &min_size))
881 continue;
882 if (to->metadata->ss->external) {
883 /* We must make sure there is
884 * no suitable spare in container already.
885 * If there is we don't add more */
886 dev_t devid = container_choose_spare(
887 to, to, NULL, min_size, st->active);
888 if (devid > 0)
889 continue;
890 }
891 for (d = 0; d < MAX_DISKS; d++)
892 if (to->devid[d])
893 domainlist_add_dev(&domlist,
894 to->devid[d],
895 to->metadata->ss->name);
896 if (to->spare_group)
897 domain_add(&domlist, to->spare_group);
898 /*
899 * No spare migration if the destination
900 * has no domain. Skip this array.
901 */
902 if (!domlist)
903 continue;
904 for (from=statelist ; from ; from=from->next) {
905 dev_t devid;
906 if (!check_donor(from, to))
907 continue;
908 if (from->metadata->ss->external)
909 devid = container_choose_spare(
910 from, to, domlist, min_size, 0);
911 else
912 devid = choose_spare(from, to, domlist,
913 min_size);
914 if (devid > 0
915 && move_spare(from->devname, to->devname, devid)) {
916 alert("MoveSpare", to->devname, from->devname, info);
917 break;
918 }
919 }
920 domain_free(domlist);
921 }
922 }
923
924 /* search the statelist to connect external
925 * metadata subarrays with their containers
926 * We always completely rebuild the tree from scratch as
927 * that is safest considering the possibility of entries
928 * disappearing or changing.
929 */
930 static void link_containers_with_subarrays(struct state *list)
931 {
932 struct state *st;
933 struct state *cont;
934 for (st = list; st; st = st->next) {
935 st->parent = NULL;
936 st->subarray = NULL;
937 }
938 for (st = list; st; st = st->next)
939 if (st->parent_dev != NoMdDev)
940 for (cont = list; cont; cont = cont->next)
941 if (!cont->err &&
942 cont->parent_dev == NoMdDev &&
943 cont->devnum == st->parent_dev) {
944 st->parent = cont;
945 st->subarray = cont->subarray;
946 cont->subarray = st;
947 break;
948 }
949 }
950
951 /* Not really Monitor but ... */
952 int Wait(char *dev)
953 {
954 struct stat stb;
955 int devnum;
956 int rv = 1;
957
958 if (stat(dev, &stb) != 0) {
959 pr_err("Cannot find %s: %s\n", dev,
960 strerror(errno));
961 return 2;
962 }
963 devnum = stat2devnum(&stb);
964
965 while(1) {
966 struct mdstat_ent *ms = mdstat_read(1, 0);
967 struct mdstat_ent *e;
968
969 for (e=ms ; e; e=e->next)
970 if (e->devnum == devnum)
971 break;
972
973 if (!e || e->percent < 0) {
974 if (e && e->metadata_version &&
975 strncmp(e->metadata_version, "external:", 9) == 0) {
976 if (is_subarray(&e->metadata_version[9]))
977 ping_monitor(&e->metadata_version[9]);
978 else
979 ping_monitor_by_id(devnum);
980 }
981 free_mdstat(ms);
982 return rv;
983 }
984 free_mdstat(ms);
985 rv = 0;
986 mdstat_wait(5);
987 }
988 }
989
990 #ifndef MDASSEMBLE
991
992 static char *clean_states[] = {
993 "clear", "inactive", "readonly", "read-auto", "clean", NULL };
994
995 int WaitClean(char *dev, int sock, int verbose)
996 {
997 int fd;
998 struct mdinfo *mdi;
999 int rv = 1;
1000 int devnum;
1001
1002 fd = open(dev, O_RDONLY);
1003 if (fd < 0) {
1004 if (verbose)
1005 pr_err("Couldn't open %s: %s\n", dev, strerror(errno));
1006 return 1;
1007 }
1008
1009 devnum = fd2devnum(fd);
1010 mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1011 if (!mdi) {
1012 if (verbose)
1013 pr_err("Failed to read sysfs attributes for "
1014 "%s\n", dev);
1015 close(fd);
1016 return 0;
1017 }
1018
1019 switch(mdi->array.level) {
1020 case LEVEL_LINEAR:
1021 case LEVEL_MULTIPATH:
1022 case 0:
1023 /* safemode delay is irrelevant for these levels */
1024 rv = 0;
1025 }
1026
1027 /* for internal metadata the kernel handles the final clean
1028 * transition, containers can never be dirty
1029 */
1030 if (!is_subarray(mdi->text_version))
1031 rv = 0;
1032
1033 /* safemode disabled ? */
1034 if (mdi->safe_mode_delay == 0)
1035 rv = 0;
1036
1037 if (rv) {
1038 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
1039 char buf[20];
1040 fd_set fds;
1041 struct timeval tm;
1042
1043 /* minimize the safe_mode_delay and prepare to wait up to 5s
1044 * for writes to quiesce
1045 */
1046 sysfs_set_safemode(mdi, 1);
1047 tm.tv_sec = 5;
1048 tm.tv_usec = 0;
1049
1050 FD_ZERO(&fds);
1051
1052 /* wait for array_state to be clean */
1053 while (1) {
1054 rv = read(state_fd, buf, sizeof(buf));
1055 if (rv < 0)
1056 break;
1057 if (sysfs_match_word(buf, clean_states) <= 4)
1058 break;
1059 FD_SET(state_fd, &fds);
1060 rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
1061 if (rv < 0 && errno != EINTR)
1062 break;
1063 lseek(state_fd, 0, SEEK_SET);
1064 }
1065 if (rv < 0)
1066 rv = 1;
1067 else if (fping_monitor(sock) == 0 ||
1068 ping_monitor(mdi->text_version) == 0) {
1069 /* we need to ping to close the window between array
1070 * state transitioning to clean and the metadata being
1071 * marked clean
1072 */
1073 rv = 0;
1074 } else
1075 rv = 1;
1076 if (rv && verbose)
1077 pr_err("Error waiting for %s to be clean\n",
1078 dev);
1079
1080 /* restore the original safe_mode_delay */
1081 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1082 close(state_fd);
1083 }
1084
1085 sysfs_free(mdi);
1086 close(fd);
1087
1088 return rv;
1089 }
1090 #endif /* MDASSEMBLE */
Unnamed repository; edit this file 'description' to name the repository.