]>
Commit | Line | Data |
---|---|---|
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 | int finished = 0; | |
128 | struct mdstat_ent *mdstat = NULL; | |
129 | char *mailfrom = NULL; | |
130 | struct alert_info info; | |
131 | ||
132 | if (!mailaddr) { | |
133 | mailaddr = conf_get_mailaddr(); | |
134 | if (mailaddr && ! scan) | |
135 | fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n", | |
136 | mailaddr); | |
137 | } | |
138 | mailfrom = conf_get_mailfrom(); | |
139 | ||
140 | if (!alert_cmd) { | |
141 | alert_cmd = conf_get_program(); | |
142 | if (alert_cmd && ! scan) | |
143 | fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n", | |
144 | alert_cmd); | |
145 | } | |
146 | if (scan && !mailaddr && !alert_cmd && !dosyslog) { | |
147 | fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n"); | |
148 | return 1; | |
149 | } | |
150 | info.alert_cmd = alert_cmd; | |
151 | info.mailaddr = mailaddr; | |
152 | info.mailfrom = mailfrom; | |
153 | info.dosyslog = dosyslog; | |
154 | ||
155 | if (daemonise) { | |
156 | int rv = make_daemon(pidfile); | |
157 | if (rv >= 0) | |
158 | return rv; | |
159 | } | |
160 | ||
161 | if (share) | |
162 | if (check_one_sharer(scan)) | |
163 | return 1; | |
164 | ||
165 | if (devlist == NULL) { | |
166 | struct mddev_ident *mdlist = conf_get_ident(NULL); | |
167 | for (; mdlist; mdlist=mdlist->next) { | |
168 | struct state *st; | |
169 | if (mdlist->devname == NULL) | |
170 | continue; | |
171 | if (strcasecmp(mdlist->devname, "<ignore>") == 0) | |
172 | continue; | |
173 | st = calloc(1, sizeof *st); | |
174 | if (st == NULL) | |
175 | continue; | |
176 | if (mdlist->devname[0] == '/') | |
177 | st->devname = strdup(mdlist->devname); | |
178 | else { | |
179 | st->devname = malloc(8+strlen(mdlist->devname)+1); | |
180 | strcpy(strcpy(st->devname, "/dev/md/"), | |
181 | mdlist->devname); | |
182 | } | |
183 | st->next = statelist; | |
184 | st->devnum = INT_MAX; | |
185 | st->percent = -2; | |
186 | st->expected_spares = mdlist->spare_disks; | |
187 | if (mdlist->spare_group) | |
188 | st->spare_group = strdup(mdlist->spare_group); | |
189 | statelist = st; | |
190 | } | |
191 | } else { | |
192 | struct mddev_dev *dv; | |
193 | for (dv=devlist ; dv; dv=dv->next) { | |
194 | struct mddev_ident *mdlist = conf_get_ident(dv->devname); | |
195 | struct state *st = calloc(1, sizeof *st); | |
196 | if (st == NULL) | |
197 | continue; | |
198 | st->devname = strdup(dv->devname); | |
199 | st->next = statelist; | |
200 | st->devnum = INT_MAX; | |
201 | st->percent = -2; | |
202 | st->expected_spares = -1; | |
203 | if (mdlist) { | |
204 | st->expected_spares = mdlist->spare_disks; | |
205 | if (mdlist->spare_group) | |
206 | st->spare_group = strdup(mdlist->spare_group); | |
207 | } | |
208 | statelist = st; | |
209 | } | |
210 | } | |
211 | ||
212 | ||
213 | while (! finished) { | |
214 | int new_found = 0; | |
215 | struct state *st; | |
216 | int anydegraded = 0; | |
217 | ||
218 | if (mdstat) | |
219 | free_mdstat(mdstat); | |
220 | mdstat = mdstat_read(oneshot?0:1, 0); | |
221 | ||
222 | for (st=statelist; st; st=st->next) | |
223 | if (check_array(st, mdstat, test, &info, increments)) | |
224 | anydegraded = 1; | |
225 | ||
226 | /* now check if there are any new devices found in mdstat */ | |
227 | if (scan) | |
228 | new_found = add_new_arrays(mdstat, &statelist, test, | |
229 | &info); | |
230 | ||
231 | /* If an array has active < raid && spare == 0 && spare_group != NULL | |
232 | * Look for another array with spare > 0 and active == raid and same spare_group | |
233 | * if found, choose a device and hotremove/hotadd | |
234 | */ | |
235 | if (share && anydegraded) | |
236 | try_spare_migration(statelist, &info); | |
237 | if (!new_found) { | |
238 | if (oneshot) | |
239 | break; | |
240 | else | |
241 | mdstat_wait(period); | |
242 | } | |
243 | test = 0; | |
244 | } | |
245 | if (pidfile) | |
246 | unlink(pidfile); | |
247 | return 0; | |
248 | } | |
249 | ||
250 | static int make_daemon(char *pidfile) | |
251 | { | |
252 | /* Return: | |
253 | * -1 in the forked daemon | |
254 | * 0 in the parent | |
255 | * 1 on error | |
256 | * so a none-negative becomes the exit code. | |
257 | */ | |
258 | int pid = fork(); | |
259 | if (pid > 0) { | |
260 | if (!pidfile) | |
261 | printf("%d\n", pid); | |
262 | else { | |
263 | FILE *pid_file; | |
264 | pid_file=fopen(pidfile, "w"); | |
265 | if (!pid_file) | |
266 | perror("cannot create pid file"); | |
267 | else { | |
268 | fprintf(pid_file,"%d\n", pid); | |
269 | fclose(pid_file); | |
270 | } | |
271 | } | |
272 | return 0; | |
273 | } | |
274 | if (pid < 0) { | |
275 | perror("daemonise"); | |
276 | return 1; | |
277 | } | |
278 | close(0); | |
279 | open("/dev/null", O_RDWR); | |
280 | dup2(0,1); | |
281 | dup2(0,2); | |
282 | setsid(); | |
283 | return -1; | |
284 | } | |
285 | ||
286 | static int check_one_sharer(int scan) | |
287 | { | |
288 | int pid, rv; | |
289 | FILE *fp; | |
290 | char dir[20]; | |
291 | struct stat buf; | |
292 | fp = fopen("/var/run/mdadm/autorebuild.pid", "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 | fprintf(stderr, Name ": Only one " | |
301 | "autorebuild process allowed" | |
302 | " in scan mode, aborting\n"); | |
303 | fclose(fp); | |
304 | return 1; | |
305 | } else { | |
306 | fprintf(stderr, Name ": Warning: One" | |
307 | " autorebuild process already" | |
308 | " running.\n"); | |
309 | } | |
310 | } | |
311 | fclose(fp); | |
312 | } | |
313 | if (scan) { | |
314 | if (mkdir("/var/run/mdadm", S_IRWXU) < 0 && | |
315 | errno != EEXIST) { | |
316 | fprintf(stderr, Name ": Can't create " | |
317 | "autorebuild.pid file\n"); | |
318 | } else { | |
319 | fp = fopen("/var/run/mdadm/autorebuild.pid", "w"); | |
320 | if (!fp) | |
321 | fprintf(stderr, Name ": 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) | |
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[MaxDisks]; | |
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 | ||
453 | if (test) | |
454 | alert("TestMessage", dev, NULL, ainfo); | |
455 | fd = open(dev, O_RDONLY); | |
456 | if (fd < 0) { | |
457 | if (!st->err) | |
458 | alert("DeviceDisappeared", dev, NULL, ainfo); | |
459 | st->err=1; | |
460 | return 0; | |
461 | } | |
462 | fcntl(fd, F_SETFD, FD_CLOEXEC); | |
463 | if (ioctl(fd, GET_ARRAY_INFO, &array)<0) { | |
464 | if (!st->err) | |
465 | alert("DeviceDisappeared", dev, NULL, ainfo); | |
466 | st->err=1; | |
467 | close(fd); | |
468 | return 0; | |
469 | } | |
470 | /* It's much easier to list what array levels can't | |
471 | * have a device disappear than all of them that can | |
472 | */ | |
473 | if (array.level == 0 || array.level == -1) { | |
474 | if (!st->err) | |
475 | alert("DeviceDisappeared", dev, "Wrong-Level", ainfo); | |
476 | st->err = 1; | |
477 | close(fd); | |
478 | return 0; | |
479 | } | |
480 | if (st->devnum == INT_MAX) { | |
481 | struct stat stb; | |
482 | if (fstat(fd, &stb) == 0 && | |
483 | (S_IFMT&stb.st_mode)==S_IFBLK) { | |
484 | if (major(stb.st_rdev) == MD_MAJOR) | |
485 | st->devnum = minor(stb.st_rdev); | |
486 | else | |
487 | st->devnum = -1- (minor(stb.st_rdev)>>6); | |
488 | } | |
489 | } | |
490 | ||
491 | for (mse2 = mdstat ; mse2 ; mse2=mse2->next) | |
492 | if (mse2->devnum == st->devnum) { | |
493 | mse2->devnum = INT_MAX; /* flag it as "used" */ | |
494 | mse = mse2; | |
495 | } | |
496 | ||
497 | if (!mse) { | |
498 | /* duplicated array in statelist | |
499 | * or re-created after reading mdstat*/ | |
500 | st->err = 1; | |
501 | close(fd); | |
502 | return 0; | |
503 | } | |
504 | /* this array is in /proc/mdstat */ | |
505 | if (array.utime == 0) | |
506 | /* external arrays don't update utime, so | |
507 | * just make sure it is always different. */ | |
508 | array.utime = st->utime + 1;; | |
509 | ||
510 | if (st->utime == array.utime && | |
511 | st->failed == array.failed_disks && | |
512 | st->working == array.working_disks && | |
513 | st->spare == array.spare_disks && | |
514 | (mse == NULL || ( | |
515 | mse->percent == st->percent | |
516 | ))) { | |
517 | close(fd); | |
518 | st->err = 0; | |
519 | if ((st->active < st->raid) && st->spare == 0) | |
520 | return 1; | |
521 | else | |
522 | return 0; | |
523 | } | |
524 | if (st->utime == 0 && /* new array */ | |
525 | mse->pattern && strchr(mse->pattern, '_') /* degraded */ | |
526 | ) | |
527 | alert("DegradedArray", dev, NULL, ainfo); | |
528 | ||
529 | if (st->utime == 0 && /* new array */ | |
530 | st->expected_spares > 0 && | |
531 | array.spare_disks < st->expected_spares) | |
532 | alert("SparesMissing", dev, NULL, ainfo); | |
533 | if (st->percent == -1 && | |
534 | mse->percent >= 0) | |
535 | alert("RebuildStarted", dev, NULL, ainfo); | |
536 | if (st->percent >= 0 && | |
537 | mse->percent >= 0 && | |
538 | (mse->percent / increments) > (st->percent / increments)) { | |
539 | char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars) | |
540 | ||
541 | if((mse->percent / increments) == 0) | |
542 | snprintf(percentalert, sizeof(percentalert), "RebuildStarted"); | |
543 | else | |
544 | snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent); | |
545 | ||
546 | alert(percentalert, dev, NULL, ainfo); | |
547 | } | |
548 | ||
549 | if (mse->percent == -1 && | |
550 | st->percent >= 0) { | |
551 | /* Rebuild/sync/whatever just finished. | |
552 | * If there is a number in /mismatch_cnt, | |
553 | * we should report that. | |
554 | */ | |
555 | struct mdinfo *sra = | |
556 | sysfs_read(-1, st->devnum, GET_MISMATCH); | |
557 | if (sra && sra->mismatch_cnt > 0) { | |
558 | char cnt[40]; | |
559 | sprintf(cnt, " mismatches found: %d", sra->mismatch_cnt); | |
560 | alert("RebuildFinished", dev, cnt, ainfo); | |
561 | } else | |
562 | alert("RebuildFinished", dev, NULL, ainfo); | |
563 | if (sra) | |
564 | free(sra); | |
565 | } | |
566 | st->percent = mse->percent; | |
567 | ||
568 | for (i=0; i<MaxDisks && i <= array.raid_disks + array.nr_disks; | |
569 | i++) { | |
570 | mdu_disk_info_t disc; | |
571 | disc.number = i; | |
572 | if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) { | |
573 | info[i].state = disc.state; | |
574 | info[i].major = disc.major; | |
575 | info[i].minor = disc.minor; | |
576 | } else | |
577 | info[i].major = info[i].minor = 0; | |
578 | } | |
579 | ||
580 | if (strncmp(mse->metadata_version, "external:", 9) == 0 && | |
581 | is_subarray(mse->metadata_version+9)) | |
582 | st->parent_dev = | |
583 | devname2devnum(mse->metadata_version+10); | |
584 | else | |
585 | st->parent_dev = NoMdDev; | |
586 | if (st->metadata == NULL && | |
587 | st->parent_dev == NoMdDev) | |
588 | st->metadata = super_by_fd(fd, NULL); | |
589 | ||
590 | close(fd); | |
591 | ||
592 | for (i=0; i<MaxDisks; i++) { | |
593 | mdu_disk_info_t disc = {0,0,0,0,0}; | |
594 | int newstate=0; | |
595 | int change; | |
596 | char *dv = NULL; | |
597 | disc.number = i; | |
598 | if (i > array.raid_disks + array.nr_disks) { | |
599 | newstate = 0; | |
600 | disc.major = disc.minor = 0; | |
601 | } else if (info[i].major || info[i].minor) { | |
602 | newstate = info[i].state; | |
603 | dv = map_dev(info[i].major, info[i].minor, 1); | |
604 | disc.state = newstate; | |
605 | disc.major = info[i].major; | |
606 | disc.minor = info[i].minor; | |
607 | } else if (mse && mse->pattern && i < (int)strlen(mse->pattern)) { | |
608 | switch(mse->pattern[i]) { | |
609 | case 'U': newstate = 6 /* ACTIVE/SYNC */; break; | |
610 | case '_': newstate = 8 /* REMOVED */; break; | |
611 | } | |
612 | disc.major = disc.minor = 0; | |
613 | } | |
614 | if (dv == NULL && st->devid[i]) | |
615 | dv = map_dev(major(st->devid[i]), | |
616 | minor(st->devid[i]), 1); | |
617 | change = newstate ^ st->devstate[i]; | |
618 | if (st->utime && change && !st->err) { | |
619 | if (i < array.raid_disks && | |
620 | (((newstate&change)&(1<<MD_DISK_FAULTY)) || | |
621 | ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) || | |
622 | ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))) | |
623 | ) | |
624 | alert("Fail", dev, dv, ainfo); | |
625 | else if (i >= array.raid_disks && | |
626 | (disc.major || disc.minor) && | |
627 | st->devid[i] == makedev(disc.major, disc.minor) && | |
628 | ((newstate&change)&(1<<MD_DISK_FAULTY)) | |
629 | ) | |
630 | alert("FailSpare", dev, dv, ainfo); | |
631 | else if (i < array.raid_disks && | |
632 | ! (newstate & (1<<MD_DISK_REMOVED)) && | |
633 | (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) || | |
634 | ((newstate&change)&(1<<MD_DISK_ACTIVE)) || | |
635 | ((newstate&change)&(1<<MD_DISK_SYNC))) | |
636 | ) | |
637 | alert("SpareActive", dev, dv, ainfo); | |
638 | } | |
639 | st->devstate[i] = newstate; | |
640 | st->devid[i] = makedev(disc.major, disc.minor); | |
641 | } | |
642 | st->active = array.active_disks; | |
643 | st->working = array.working_disks; | |
644 | st->spare = array.spare_disks; | |
645 | st->failed = array.failed_disks; | |
646 | st->utime = array.utime; | |
647 | st->raid = array.raid_disks; | |
648 | st->err = 0; | |
649 | if ((st->active < st->raid) && st->spare == 0) | |
650 | return 1; | |
651 | return 0; | |
652 | } | |
653 | ||
654 | static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist, | |
655 | int test, struct alert_info *info) | |
656 | { | |
657 | struct mdstat_ent *mse; | |
658 | int new_found = 0; | |
659 | ||
660 | for (mse=mdstat; mse; mse=mse->next) | |
661 | if (mse->devnum != INT_MAX && | |
662 | (!mse->level || /* retrieve containers */ | |
663 | (strcmp(mse->level, "raid0") != 0 && | |
664 | strcmp(mse->level, "linear") != 0)) | |
665 | ) { | |
666 | struct state *st = calloc(1, sizeof *st); | |
667 | mdu_array_info_t array; | |
668 | int fd; | |
669 | if (st == NULL) | |
670 | continue; | |
671 | st->devname = strdup(get_md_name(mse->devnum)); | |
672 | if ((fd = open(st->devname, O_RDONLY)) < 0 || | |
673 | ioctl(fd, GET_ARRAY_INFO, &array)< 0) { | |
674 | /* no such array */ | |
675 | if (fd >=0) close(fd); | |
676 | put_md_name(st->devname); | |
677 | free(st->devname); | |
678 | if (st->metadata) { | |
679 | st->metadata->ss->free_super(st->metadata); | |
680 | free(st->metadata); | |
681 | } | |
682 | free(st); | |
683 | continue; | |
684 | } | |
685 | close(fd); | |
686 | st->next = *statelist; | |
687 | st->err = 1; | |
688 | st->devnum = mse->devnum; | |
689 | st->percent = -2; | |
690 | st->expected_spares = -1; | |
691 | if (mse->metadata_version && | |
692 | strncmp(mse->metadata_version, "external:", 9) == 0 && | |
693 | is_subarray(mse->metadata_version+9)) | |
694 | st->parent_dev = | |
695 | devname2devnum(mse->metadata_version+10); | |
696 | else | |
697 | st->parent_dev = NoMdDev; | |
698 | *statelist = st; | |
699 | if (test) | |
700 | alert("TestMessage", st->devname, NULL, info); | |
701 | alert("NewArray", 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 < MaxDisks; 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 < MaxDisks; 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 | fprintf(stderr, Name ": 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 | fprintf(stderr, Name ": 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 | fprintf(stderr, Name ": 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 | fprintf(stderr, Name ": 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 */ |