]> 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
Grow: fix resize of array component size to > 32bits
[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 char devnm[32]; /* to sync with mdstat info */
36 long utime;
37 int err;
38 char *spare_group;
39 int active, working, failed, spare, raid;
40 int from_config;
41 int from_auto;
42 int expected_spares;
43 int devstate[MAX_DISKS];
44 dev_t devid[MAX_DISKS];
45 int percent;
46 char parent_devnm[32]; /* For subarray, devnm of parent.
47 * For others, ""
48 */
49 struct supertype *metadata;
50 struct state *subarray;/* for a container it is a link to first subarray
51 * for a subarray it is a link to next subarray
52 * in the same container */
53 struct state *parent; /* for a subarray it is a link to its container
54 */
55 struct state *next;
56 };
57
58 struct alert_info {
59 char *mailaddr;
60 char *mailfrom;
61 char *alert_cmd;
62 int dosyslog;
63 };
64 static int make_daemon(char *pidfile);
65 static int check_one_sharer(int scan);
66 static void alert(char *event, char *dev, char *disc, struct alert_info *info);
67 static int check_array(struct state *st, struct mdstat_ent *mdstat,
68 int test, struct alert_info *info,
69 int increments, char *prefer);
70 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
71 int test, struct alert_info *info);
72 static void try_spare_migration(struct state *statelist, struct alert_info *info);
73 static void link_containers_with_subarrays(struct state *list);
74
75 int Monitor(struct mddev_dev *devlist,
76 char *mailaddr, char *alert_cmd,
77 struct context *c,
78 int daemonise, int oneshot,
79 int dosyslog, char *pidfile, int increments,
80 int share)
81 {
82 /*
83 * Every few seconds, scan every md device looking for changes
84 * When a change is found, log it, possibly run the alert command,
85 * and possibly send Email
86 *
87 * For each array, we record:
88 * Update time
89 * active/working/failed/spare drives
90 * State of each device.
91 * %rebuilt if rebuilding
92 *
93 * If the update time changes, check out all the data again
94 * It is possible that we cannot get the state of each device
95 * due to bugs in the md kernel module.
96 * We also read /proc/mdstat to get rebuild percent,
97 * and to get state on all active devices incase of kernel bug.
98 *
99 * Events are:
100 * Fail
101 * An active device had Faulty set or Active/Sync removed
102 * FailSpare
103 * A spare device had Faulty set
104 * SpareActive
105 * An active device had a reverse transition
106 * RebuildStarted
107 * percent went from -1 to +ve
108 * RebuildNN
109 * percent went from below to not-below NN%
110 * DeviceDisappeared
111 * Couldn't access a device which was previously visible
112 *
113 * if we detect an array with active<raid and spare==0
114 * we look at other arrays that have same spare-group
115 * If we find one with active==raid and spare>0,
116 * and if we can get_disk_info and find a name
117 * Then we hot-remove and hot-add to the other array
118 *
119 * If devlist is NULL, then we can monitor everything because --scan
120 * was given. We get an initial list from config file and add anything
121 * that appears in /proc/mdstat
122 */
123
124 struct state *statelist = NULL;
125 struct state *st2;
126 int finished = 0;
127 struct mdstat_ent *mdstat = NULL;
128 char *mailfrom = NULL;
129 struct alert_info info;
130
131 if (!mailaddr) {
132 mailaddr = conf_get_mailaddr();
133 if (mailaddr && ! c->scan)
134 pr_err("Monitor using email address \"%s\" from config file\n",
135 mailaddr);
136 }
137 mailfrom = conf_get_mailfrom();
138
139 if (!alert_cmd) {
140 alert_cmd = conf_get_program();
141 if (alert_cmd && ! c->scan)
142 pr_err("Monitor using program \"%s\" from config file\n",
143 alert_cmd);
144 }
145 if (c->scan && !mailaddr && !alert_cmd && !dosyslog) {
146 pr_err("No mail address or alert command - not monitoring.\n");
147 return 1;
148 }
149 info.alert_cmd = alert_cmd;
150 info.mailaddr = mailaddr;
151 info.mailfrom = mailfrom;
152 info.dosyslog = dosyslog;
153
154 if (daemonise) {
155 int rv = make_daemon(pidfile);
156 if (rv >= 0)
157 return rv;
158 }
159
160 if (share)
161 if (check_one_sharer(c->scan))
162 return 1;
163
164 if (devlist == NULL) {
165 struct mddev_ident *mdlist = conf_get_ident(NULL);
166 for (; mdlist; mdlist=mdlist->next) {
167 struct state *st;
168 if (mdlist->devname == NULL)
169 continue;
170 if (strcasecmp(mdlist->devname, "<ignore>") == 0)
171 continue;
172 st = xcalloc(1, sizeof *st);
173 if (mdlist->devname[0] == '/')
174 st->devname = xstrdup(mdlist->devname);
175 else {
176 st->devname = xmalloc(8+strlen(mdlist->devname)+1);
177 strcpy(strcpy(st->devname, "/dev/md/"),
178 mdlist->devname);
179 }
180 st->next = statelist;
181 st->devnm[0] = 0;
182 st->percent = RESYNC_UNKNOWN;
183 st->from_config = 1;
184 st->expected_spares = mdlist->spare_disks;
185 if (mdlist->spare_group)
186 st->spare_group = xstrdup(mdlist->spare_group);
187 statelist = st;
188 }
189 } else {
190 struct mddev_dev *dv;
191 for (dv=devlist ; dv; dv=dv->next) {
192 struct mddev_ident *mdlist = conf_get_ident(dv->devname);
193 struct state *st = xcalloc(1, sizeof *st);
194 st->devname = xstrdup(dv->devname);
195 st->next = statelist;
196 st->devnm[0] = 0;
197 st->percent = RESYNC_UNKNOWN;
198 st->expected_spares = -1;
199 if (mdlist) {
200 st->expected_spares = mdlist->spare_disks;
201 if (mdlist->spare_group)
202 st->spare_group = xstrdup(mdlist->spare_group);
203 }
204 statelist = st;
205 }
206 }
207
208 while (! finished) {
209 int new_found = 0;
210 struct state *st, **stp;
211 int anydegraded = 0;
212
213 if (mdstat)
214 free_mdstat(mdstat);
215 mdstat = mdstat_read(oneshot?0:1, 0);
216
217 for (st=statelist; st; st=st->next)
218 if (check_array(st, mdstat, c->test, &info,
219 increments, c->prefer))
220 anydegraded = 1;
221
222 /* now check if there are any new devices found in mdstat */
223 if (c->scan)
224 new_found = add_new_arrays(mdstat, &statelist, c->test,
225 &info);
226
227 /* If an array has active < raid && spare == 0 && spare_group != NULL
228 * Look for another array with spare > 0 and active == raid and same spare_group
229 * if found, choose a device and hotremove/hotadd
230 */
231 if (share && anydegraded)
232 try_spare_migration(statelist, &info);
233 if (!new_found) {
234 if (oneshot)
235 break;
236 else
237 mdstat_wait(c->delay);
238 }
239 c->test = 0;
240
241 for (stp = &statelist; (st = *stp) != NULL; ) {
242 if (st->from_auto && st->err > 5) {
243 *stp = st->next;
244 free(st->devname);
245 free(st->spare_group);
246 free(st);
247 } else
248 stp = &st->next;
249 }
250 }
251 for (st2 = statelist; st2; st2 = statelist) {
252 statelist = st2->next;
253 free(st2);
254 }
255
256 if (pidfile)
257 unlink(pidfile);
258 return 0;
259 }
260
261 static int make_daemon(char *pidfile)
262 {
263 /* Return:
264 * -1 in the forked daemon
265 * 0 in the parent
266 * 1 on error
267 * so a none-negative becomes the exit code.
268 */
269 int pid = fork();
270 if (pid > 0) {
271 if (!pidfile)
272 printf("%d\n", pid);
273 else {
274 FILE *pid_file;
275 pid_file=fopen(pidfile, "w");
276 if (!pid_file)
277 perror("cannot create pid file");
278 else {
279 fprintf(pid_file,"%d\n", pid);
280 fclose(pid_file);
281 }
282 }
283 return 0;
284 }
285 if (pid < 0) {
286 perror("daemonise");
287 return 1;
288 }
289 close(0);
290 open("/dev/null", O_RDWR);
291 dup2(0,1);
292 dup2(0,2);
293 setsid();
294 return -1;
295 }
296
297 static int check_one_sharer(int scan)
298 {
299 int pid, rv;
300 FILE *fp;
301 char dir[20];
302 char path[100];
303 struct stat buf;
304 sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
305 fp = fopen(path, "r");
306 if (fp) {
307 if (fscanf(fp, "%d", &pid) != 1)
308 pid = -1;
309 sprintf(dir, "/proc/%d", pid);
310 rv = stat(dir, &buf);
311 if (rv != -1) {
312 if (scan) {
313 pr_err("Only one "
314 "autorebuild process allowed"
315 " in scan mode, aborting\n");
316 fclose(fp);
317 return 1;
318 } else {
319 pr_err("Warning: One"
320 " autorebuild process already"
321 " running.\n");
322 }
323 }
324 fclose(fp);
325 }
326 if (scan) {
327 if (mkdir(MDMON_DIR, S_IRWXU) < 0 &&
328 errno != EEXIST) {
329 pr_err("Can't create "
330 "autorebuild.pid file\n");
331 } else {
332 fp = fopen(path, "w");
333 if (!fp)
334 pr_err("Cannot create"
335 " autorebuild.pid"
336 "file\n");
337 else {
338 pid = getpid();
339 fprintf(fp, "%d\n", pid);
340 fclose(fp);
341 }
342 }
343 }
344 return 0;
345 }
346
347 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
348 {
349 int priority;
350
351 if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
352 time_t now = time(0);
353
354 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
355 }
356 if (info->alert_cmd) {
357 int pid = fork();
358 switch(pid) {
359 default:
360 waitpid(pid, NULL, 0);
361 break;
362 case -1:
363 break;
364 case 0:
365 execl(info->alert_cmd, info->alert_cmd,
366 event, dev, disc, NULL);
367 exit(2);
368 }
369 }
370 if (info->mailaddr &&
371 (strncmp(event, "Fail", 4)==0 ||
372 strncmp(event, "Test", 4)==0 ||
373 strncmp(event, "Spares", 6)==0 ||
374 strncmp(event, "Degrade", 7)==0)) {
375 FILE *mp = popen(Sendmail, "w");
376 if (mp) {
377 FILE *mdstat;
378 char hname[256];
379 gethostname(hname, sizeof(hname));
380 signal(SIGPIPE, SIG_IGN);
381 if (info->mailfrom)
382 fprintf(mp, "From: %s\n", info->mailfrom);
383 else
384 fprintf(mp, "From: " Name " monitoring <root>\n");
385 fprintf(mp, "To: %s\n", info->mailaddr);
386 fprintf(mp, "Subject: %s event on %s:%s\n\n",
387 event, dev, hname);
388
389 fprintf(mp,
390 "This is an automatically generated"
391 " mail message from " Name "\n");
392 fprintf(mp, "running on %s\n\n", hname);
393
394 fprintf(mp,
395 "A %s event had been detected on"
396 " md device %s.\n\n", event, dev);
397
398 if (disc && disc[0] != ' ')
399 fprintf(mp,
400 "It could be related to"
401 " component device %s.\n\n", disc);
402 if (disc && disc[0] == ' ')
403 fprintf(mp, "Extra information:%s.\n\n", disc);
404
405 fprintf(mp, "Faithfully yours, etc.\n");
406
407 mdstat = fopen("/proc/mdstat", "r");
408 if (mdstat) {
409 char buf[8192];
410 int n;
411 fprintf(mp,
412 "\nP.S. The /proc/mdstat file"
413 " currently contains the following:\n\n");
414 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
415 n=fwrite(buf, 1, n, mp);
416 fclose(mdstat);
417 }
418 pclose(mp);
419 }
420 }
421
422 /* log the event to syslog maybe */
423 if (info->dosyslog) {
424 /* Log at a different severity depending on the event.
425 *
426 * These are the critical events: */
427 if (strncmp(event, "Fail", 4)==0 ||
428 strncmp(event, "Degrade", 7)==0 ||
429 strncmp(event, "DeviceDisappeared", 17)==0)
430 priority = LOG_CRIT;
431 /* Good to know about, but are not failures: */
432 else if (strncmp(event, "Rebuild", 7)==0 ||
433 strncmp(event, "MoveSpare", 9)==0 ||
434 strncmp(event, "Spares", 6) != 0)
435 priority = LOG_WARNING;
436 /* Everything else: */
437 else
438 priority = LOG_INFO;
439
440 if (disc)
441 syslog(priority,
442 "%s event detected on md device %s,"
443 " component device %s", event, dev, disc);
444 else
445 syslog(priority,
446 "%s event detected on md device %s",
447 event, dev);
448 }
449 }
450
451 static int check_array(struct state *st, struct mdstat_ent *mdstat,
452 int test, struct alert_info *ainfo,
453 int increments, char *prefer)
454 {
455 /* Update the state 'st' to reflect any changes shown in mdstat,
456 * or found by directly examining the array, and return
457 * '1' if the array is degraded, or '0' if it is optimal (or dead).
458 */
459 struct { int state, major, minor; } info[MAX_DISKS];
460 mdu_array_info_t array;
461 struct mdstat_ent *mse = NULL, *mse2;
462 char *dev = st->devname;
463 int fd;
464 int i;
465 int remaining_disks;
466 int last_disk;
467 int new_array = 0;
468
469 if (test)
470 alert("TestMessage", dev, NULL, ainfo);
471 fd = open(dev, O_RDONLY);
472 if (fd < 0) {
473 if (!st->err)
474 alert("DeviceDisappeared", dev, NULL, ainfo);
475 st->err++;
476 return 0;
477 }
478 fcntl(fd, F_SETFD, FD_CLOEXEC);
479 if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
480 if (!st->err)
481 alert("DeviceDisappeared", dev, NULL, ainfo);
482 st->err++;
483 close(fd);
484 return 0;
485 }
486 /* It's much easier to list what array levels can't
487 * have a device disappear than all of them that can
488 */
489 if (array.level == 0 || array.level == -1) {
490 if (!st->err && !st->from_config)
491 alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
492 st->err++;
493 close(fd);
494 return 0;
495 }
496 if (st->devnm[0] == 0)
497 strcpy(st->devnm, fd2devnm(fd));
498
499 for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
500 if (strcmp(mse2->devnm, st->devnm) == 0) {
501 mse2->devnm[0] = 0; /* 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++;
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->err) {
519 /* New array appeared where previously had an error */
520 st->err = 0;
521 st->percent = RESYNC_NONE;
522 new_array = 1;
523 alert("NewArray", st->devname, NULL, ainfo);
524 }
525
526 if (st->utime == array.utime &&
527 st->failed == array.failed_disks &&
528 st->working == array.working_disks &&
529 st->spare == array.spare_disks &&
530 (mse == NULL || (
531 mse->percent == st->percent
532 ))) {
533 close(fd);
534 if ((st->active < st->raid) && st->spare == 0)
535 return 1;
536 else
537 return 0;
538 }
539 if (st->utime == 0 && /* new array */
540 mse->pattern && strchr(mse->pattern, '_') /* degraded */
541 )
542 alert("DegradedArray", dev, NULL, ainfo);
543
544 if (st->utime == 0 && /* new array */
545 st->expected_spares > 0 &&
546 array.spare_disks < st->expected_spares)
547 alert("SparesMissing", dev, NULL, ainfo);
548 if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
549 mse->percent >= 0)
550 alert("RebuildStarted", dev, NULL, ainfo);
551 if (st->percent >= 0 &&
552 mse->percent >= 0 &&
553 (mse->percent / increments) > (st->percent / increments)) {
554 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
555
556 if((mse->percent / increments) == 0)
557 snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
558 else
559 snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
560
561 alert(percentalert, dev, NULL, ainfo);
562 }
563
564 if (mse->percent == RESYNC_NONE &&
565 st->percent >= 0) {
566 /* Rebuild/sync/whatever just finished.
567 * If there is a number in /mismatch_cnt,
568 * we should report that.
569 */
570 struct mdinfo *sra =
571 sysfs_read(-1, st->devnm, GET_MISMATCH);
572 if (sra && sra->mismatch_cnt > 0) {
573 char cnt[80];
574 snprintf(cnt, sizeof(cnt),
575 " mismatches found: %d (on raid level %d)",
576 sra->mismatch_cnt, array.level);
577 alert("RebuildFinished", dev, cnt, ainfo);
578 } else
579 alert("RebuildFinished", dev, NULL, ainfo);
580 if (sra)
581 free(sra);
582 }
583 st->percent = mse->percent;
584
585 remaining_disks = array.nr_disks;
586 for (i=0; i<MAX_DISKS && remaining_disks > 0;
587 i++) {
588 mdu_disk_info_t disc;
589 disc.number = i;
590 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
591 info[i].state = disc.state;
592 info[i].major = disc.major;
593 info[i].minor = disc.minor;
594 if (disc.major || disc.minor)
595 remaining_disks --;
596 } else
597 info[i].major = info[i].minor = 0;
598 }
599 last_disk = i;
600
601 if (mse->metadata_version &&
602 strncmp(mse->metadata_version, "external:", 9) == 0 &&
603 is_subarray(mse->metadata_version+9)) {
604 char *sl;
605 strcpy(st->parent_devnm,
606 mse->metadata_version+10);
607 sl = strchr(st->parent_devnm, '/');
608 if (sl)
609 *sl = 0;
610 } else
611 st->parent_devnm[0] = 0;
612 if (st->metadata == NULL &&
613 st->parent_devnm[0] == 0)
614 st->metadata = super_by_fd(fd, NULL);
615
616 close(fd);
617
618 for (i=0; i<MAX_DISKS; i++) {
619 mdu_disk_info_t disc = {0,0,0,0,0};
620 int newstate=0;
621 int change;
622 char *dv = NULL;
623 disc.number = i;
624 if (i < last_disk &&
625 (info[i].major || info[i].minor)) {
626 newstate = info[i].state;
627 dv = map_dev_preferred(
628 info[i].major, info[i].minor, 1,
629 prefer);
630 disc.state = newstate;
631 disc.major = info[i].major;
632 disc.minor = info[i].minor;
633 } else
634 newstate = (1 << MD_DISK_REMOVED);
635
636 if (dv == NULL && st->devid[i])
637 dv = map_dev_preferred(
638 major(st->devid[i]),
639 minor(st->devid[i]), 1, prefer);
640 change = newstate ^ st->devstate[i];
641 if (st->utime && change && !st->err && !new_array) {
642 if ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))
643 alert("Fail", dev, dv, ainfo);
644 else if ((newstate & (1<<MD_DISK_FAULTY)) &&
645 (disc.major || disc.minor) &&
646 st->devid[i] == makedev(disc.major, disc.minor))
647 alert("FailSpare", dev, dv, ainfo);
648 else if ((newstate&change)&(1<<MD_DISK_SYNC))
649 alert("SpareActive", dev, dv, ainfo);
650 }
651 st->devstate[i] = newstate;
652 st->devid[i] = makedev(disc.major, disc.minor);
653 }
654 st->active = array.active_disks;
655 st->working = array.working_disks;
656 st->spare = array.spare_disks;
657 st->failed = array.failed_disks;
658 st->utime = array.utime;
659 st->raid = array.raid_disks;
660 st->err = 0;
661 if ((st->active < st->raid) && st->spare == 0)
662 return 1;
663 return 0;
664 }
665
666 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
667 int test, struct alert_info *info)
668 {
669 struct mdstat_ent *mse;
670 int new_found = 0;
671
672 for (mse=mdstat; mse; mse=mse->next)
673 if (mse->devnm[0] &&
674 (!mse->level || /* retrieve containers */
675 (strcmp(mse->level, "raid0") != 0 &&
676 strcmp(mse->level, "linear") != 0))
677 ) {
678 struct state *st = xcalloc(1, sizeof *st);
679 mdu_array_info_t array;
680 int fd;
681 st->devname = xstrdup(get_md_name(mse->devnm));
682 if ((fd = open(st->devname, O_RDONLY)) < 0 ||
683 ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
684 /* no such array */
685 if (fd >=0) close(fd);
686 put_md_name(st->devname);
687 free(st->devname);
688 if (st->metadata) {
689 st->metadata->ss->free_super(st->metadata);
690 free(st->metadata);
691 }
692 free(st);
693 continue;
694 }
695 close(fd);
696 st->next = *statelist;
697 st->err = 1;
698 st->from_auto = 1;
699 strcpy(st->devnm, mse->devnm);
700 st->percent = RESYNC_UNKNOWN;
701 st->expected_spares = -1;
702 if (mse->metadata_version &&
703 strncmp(mse->metadata_version, "external:", 9) == 0 &&
704 is_subarray(mse->metadata_version+9)) {
705 char *sl;
706 strcpy(st->parent_devnm,
707 mse->metadata_version+10);
708 sl = strchr(st->parent_devnm, '/');
709 *sl = 0;
710 } else
711 st->parent_devnm[0] = 0;
712 *statelist = st;
713 if (test)
714 alert("TestMessage", st->devname, NULL, info);
715 new_found = 1;
716 }
717 return new_found;
718 }
719
720 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
721 {
722 int fd;
723
724 if (!st->metadata ||
725 !st->metadata->ss->min_acceptable_spare_size) {
726 *sizep = 0;
727 return 0;
728 }
729
730 fd = open(st->devname, O_RDONLY);
731 if (fd < 0)
732 return 1;
733 if (st->metadata->ss->external)
734 st->metadata->ss->load_container(st->metadata, fd, st->devname);
735 else
736 st->metadata->ss->load_super(st->metadata, fd, st->devname);
737 close(fd);
738 if (!st->metadata->sb)
739 return 1;
740 *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
741 st->metadata->ss->free_super(st->metadata);
742
743 return 0;
744 }
745
746 static int check_donor(struct state *from, struct state *to)
747 {
748 struct state *sub;
749
750 if (from == to)
751 return 0;
752 if (from->parent)
753 /* Cannot move from a member */
754 return 0;
755 if (from->err)
756 return 0;
757 for (sub = from->subarray; sub; sub = sub->subarray)
758 /* If source array has degraded subarrays, don't
759 * remove anything
760 */
761 if (sub->active < sub->raid)
762 return 0;
763 if (from->metadata->ss->external == 0)
764 if (from->active < from->raid)
765 return 0;
766 if (from->spare <= 0)
767 return 0;
768 return 1;
769 }
770
771 static dev_t choose_spare(struct state *from, struct state *to,
772 struct domainlist *domlist, unsigned long long min_size)
773 {
774 int d;
775 dev_t dev = 0;
776
777 for (d = from->raid; !dev && d < MAX_DISKS; d++) {
778 if (from->devid[d] > 0 &&
779 from->devstate[d] == 0) {
780 struct dev_policy *pol;
781 unsigned long long dev_size;
782
783 if (to->metadata->ss->external &&
784 test_partition_from_id(from->devid[d]))
785 continue;
786
787 if (min_size &&
788 dev_size_from_id(from->devid[d], &dev_size) &&
789 dev_size < min_size)
790 continue;
791
792 pol = devid_policy(from->devid[d]);
793 if (from->spare_group)
794 pol_add(&pol, pol_domain,
795 from->spare_group, NULL);
796 if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
797 dev = from->devid[d];
798 dev_policy_free(pol);
799 }
800 }
801 return dev;
802 }
803
804 static dev_t container_choose_spare(struct state *from, struct state *to,
805 struct domainlist *domlist,
806 unsigned long long min_size, int active)
807 {
808 /* This is similar to choose_spare, but we cannot trust devstate,
809 * so we need to read the metadata instead
810 */
811 struct mdinfo *list;
812 struct supertype *st = from->metadata;
813 int fd = open(from->devname, O_RDONLY);
814 int err;
815 dev_t dev = 0;
816
817 if (fd < 0)
818 return 0;
819 if (!st->ss->getinfo_super_disks) {
820 close(fd);
821 return 0;
822 }
823
824 err = st->ss->load_container(st, fd, NULL);
825 close(fd);
826 if (err)
827 return 0;
828
829 if (from == to) {
830 /* We must check if number of active disks has not increased
831 * since ioctl in main loop. mdmon may have added spare
832 * to subarray. If so we do not need to look for more spares
833 * so return non zero value */
834 int active_cnt = 0;
835 struct mdinfo *dp;
836 list = st->ss->getinfo_super_disks(st);
837 if (!list) {
838 st->ss->free_super(st);
839 return 1;
840 }
841 dp = list->devs;
842 while (dp) {
843 if (dp->disk.state & (1<<MD_DISK_SYNC) &&
844 !(dp->disk.state & (1<<MD_DISK_FAULTY)))
845 active_cnt++;
846 dp = dp->next;
847 }
848 sysfs_free(list);
849 if (active < active_cnt) {
850 /* Spare just activated.*/
851 st->ss->free_super(st);
852 return 1;
853 }
854 }
855
856 /* We only need one spare so full list not needed */
857 list = container_choose_spares(st, min_size, domlist, from->spare_group,
858 to->metadata->ss->name, 1);
859 if (list) {
860 struct mdinfo *disks = list->devs;
861 if (disks)
862 dev = makedev(disks->disk.major, disks->disk.minor);
863 sysfs_free(list);
864 }
865 st->ss->free_super(st);
866 return dev;
867 }
868
869 static void try_spare_migration(struct state *statelist, struct alert_info *info)
870 {
871 struct state *from;
872 struct state *st;
873
874 link_containers_with_subarrays(statelist);
875 for (st = statelist; st; st = st->next)
876 if (st->active < st->raid &&
877 st->spare == 0 && !st->err) {
878 struct domainlist *domlist = NULL;
879 int d;
880 struct state *to = st;
881 unsigned long long min_size;
882
883 if (to->parent_devnm[0] && !to->parent)
884 /* subarray monitored without parent container
885 * we can't move spares here */
886 continue;
887
888 if (to->parent)
889 /* member of a container */
890 to = to->parent;
891
892 if (get_min_spare_size_required(to, &min_size))
893 continue;
894 if (to->metadata->ss->external) {
895 /* We must make sure there is
896 * no suitable spare in container already.
897 * If there is we don't add more */
898 dev_t devid = container_choose_spare(
899 to, to, NULL, min_size, st->active);
900 if (devid > 0)
901 continue;
902 }
903 for (d = 0; d < MAX_DISKS; d++)
904 if (to->devid[d])
905 domainlist_add_dev(&domlist,
906 to->devid[d],
907 to->metadata->ss->name);
908 if (to->spare_group)
909 domain_add(&domlist, to->spare_group);
910 /*
911 * No spare migration if the destination
912 * has no domain. Skip this array.
913 */
914 if (!domlist)
915 continue;
916 for (from=statelist ; from ; from=from->next) {
917 dev_t devid;
918 if (!check_donor(from, to))
919 continue;
920 if (from->metadata->ss->external)
921 devid = container_choose_spare(
922 from, to, domlist, min_size, 0);
923 else
924 devid = choose_spare(from, to, domlist,
925 min_size);
926 if (devid > 0
927 && move_spare(from->devname, to->devname, devid)) {
928 alert("MoveSpare", to->devname, from->devname, info);
929 break;
930 }
931 }
932 domain_free(domlist);
933 }
934 }
935
936 /* search the statelist to connect external
937 * metadata subarrays with their containers
938 * We always completely rebuild the tree from scratch as
939 * that is safest considering the possibility of entries
940 * disappearing or changing.
941 */
942 static void link_containers_with_subarrays(struct state *list)
943 {
944 struct state *st;
945 struct state *cont;
946 for (st = list; st; st = st->next) {
947 st->parent = NULL;
948 st->subarray = NULL;
949 }
950 for (st = list; st; st = st->next)
951 if (st->parent_devnm[0])
952 for (cont = list; cont; cont = cont->next)
953 if (!cont->err &&
954 cont->parent_devnm[0] == 0 &&
955 strcmp(cont->devnm, st->parent_devnm) == 0) {
956 st->parent = cont;
957 st->subarray = cont->subarray;
958 cont->subarray = st;
959 break;
960 }
961 }
962
963 /* Not really Monitor but ... */
964 int Wait(char *dev)
965 {
966 struct stat stb;
967 char devnm[32];
968 int rv = 1;
969
970 if (stat(dev, &stb) != 0) {
971 pr_err("Cannot find %s: %s\n", dev,
972 strerror(errno));
973 return 2;
974 }
975 strcpy(devnm, stat2devnm(&stb));
976
977 while(1) {
978 struct mdstat_ent *ms = mdstat_read(1, 0);
979 struct mdstat_ent *e;
980
981 for (e=ms ; e; e=e->next)
982 if (strcmp(e->devnm, devnm) == 0)
983 break;
984
985 if (e->percent == RESYNC_NONE) {
986 /* We could be in the brief pause before something
987 * starts. /proc/mdstat doesn't show that, but
988 * sync_action does.
989 */
990 struct mdinfo mdi;
991 char buf[21];
992 sysfs_init(&mdi, -1, devnm);
993 if (sysfs_get_str(&mdi, NULL, "sync_action",
994 buf, 20) > 0 &&
995 strcmp(buf,"idle\n") != 0)
996 e->percent = RESYNC_UNKNOWN;
997 }
998 if (!e || e->percent == RESYNC_NONE) {
999 if (e && e->metadata_version &&
1000 strncmp(e->metadata_version, "external:", 9) == 0) {
1001 if (is_subarray(&e->metadata_version[9]))
1002 ping_monitor(&e->metadata_version[9]);
1003 else
1004 ping_monitor(devnm);
1005 }
1006 free_mdstat(ms);
1007 return rv;
1008 }
1009 free_mdstat(ms);
1010 rv = 0;
1011 mdstat_wait(5);
1012 }
1013 }
1014
1015 #ifndef MDASSEMBLE
1016
1017 static char *clean_states[] = {
1018 "clear", "inactive", "readonly", "read-auto", "clean", NULL };
1019
1020 int WaitClean(char *dev, int sock, int verbose)
1021 {
1022 int fd;
1023 struct mdinfo *mdi;
1024 int rv = 1;
1025 char devnm[32];
1026
1027 fd = open(dev, O_RDONLY);
1028 if (fd < 0) {
1029 if (verbose)
1030 pr_err("Couldn't open %s: %s\n", dev, strerror(errno));
1031 return 1;
1032 }
1033
1034 strcpy(devnm, fd2devnm(fd));
1035 mdi = sysfs_read(fd, devnm, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1036 if (!mdi) {
1037 if (verbose)
1038 pr_err("Failed to read sysfs attributes for "
1039 "%s\n", dev);
1040 close(fd);
1041 return 0;
1042 }
1043
1044 switch(mdi->array.level) {
1045 case LEVEL_LINEAR:
1046 case LEVEL_MULTIPATH:
1047 case 0:
1048 /* safemode delay is irrelevant for these levels */
1049 rv = 0;
1050 }
1051
1052 /* for internal metadata the kernel handles the final clean
1053 * transition, containers can never be dirty
1054 */
1055 if (!is_subarray(mdi->text_version))
1056 rv = 0;
1057
1058 /* safemode disabled ? */
1059 if (mdi->safe_mode_delay == 0)
1060 rv = 0;
1061
1062 if (rv) {
1063 int state_fd = sysfs_open(fd2devnm(fd), NULL, "array_state");
1064 char buf[20];
1065 int delay = 5000;
1066
1067 /* minimize the safe_mode_delay and prepare to wait up to 5s
1068 * for writes to quiesce
1069 */
1070 sysfs_set_safemode(mdi, 1);
1071
1072 /* wait for array_state to be clean */
1073 while (1) {
1074 rv = read(state_fd, buf, sizeof(buf));
1075 if (rv < 0)
1076 break;
1077 if (sysfs_match_word(buf, clean_states) <= 4)
1078 break;
1079 rv = sysfs_wait(state_fd, &delay);
1080 if (rv < 0 && errno != EINTR)
1081 break;
1082 lseek(state_fd, 0, SEEK_SET);
1083 }
1084 if (rv < 0)
1085 rv = 1;
1086 else if (fping_monitor(sock) == 0 ||
1087 ping_monitor(mdi->text_version) == 0) {
1088 /* we need to ping to close the window between array
1089 * state transitioning to clean and the metadata being
1090 * marked clean
1091 */
1092 rv = 0;
1093 } else
1094 rv = 1;
1095 if (rv && verbose)
1096 pr_err("Error waiting for %s to be clean\n",
1097 dev);
1098
1099 /* restore the original safe_mode_delay */
1100 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1101 close(state_fd);
1102 }
1103
1104 sysfs_free(mdi);
1105 close(fd);
1106
1107 return rv;
1108 }
1109 #endif /* MDASSEMBLE */
Unnamed repository; edit this file 'description' to name the repository.