]>
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 | struct state { | |
34 | char *devname; | |
35 | char devnm[32]; /* to sync with mdstat info */ | |
36 | unsigned int 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; | |
129 | struct alert_info info; | |
130 | struct mddev_ident *mdlist; | |
131 | ||
132 | if (!mailaddr) { | |
133 | mailaddr = conf_get_mailaddr(); | |
134 | if (mailaddr && ! c->scan) | |
135 | pr_err("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 && !c->scan) | |
143 | pr_err("Monitor using program \"%s\" from config file\n", | |
144 | alert_cmd); | |
145 | } | |
146 | if (c->scan && !mailaddr && !alert_cmd && !dosyslog) { | |
147 | pr_err("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(c->scan)) | |
163 | return 1; | |
164 | ||
165 | if (devlist == NULL) { | |
166 | mdlist = conf_get_ident(NULL); | |
167 | for (; mdlist; mdlist = mdlist->next) { | |
168 | struct state *st; | |
169 | ||
170 | if (mdlist->devname == NULL) | |
171 | continue; | |
172 | if (strcasecmp(mdlist->devname, "<ignore>") == 0) | |
173 | continue; | |
174 | st = xcalloc(1, sizeof *st); | |
175 | if (mdlist->devname[0] == '/') | |
176 | st->devname = xstrdup(mdlist->devname); | |
177 | else { | |
178 | st->devname = xmalloc(8+strlen(mdlist->devname)+1); | |
179 | strcpy(strcpy(st->devname, "/dev/md/"), | |
180 | mdlist->devname); | |
181 | } | |
182 | st->next = statelist; | |
183 | st->devnm[0] = 0; | |
184 | st->percent = RESYNC_UNKNOWN; | |
185 | st->from_config = 1; | |
186 | st->expected_spares = mdlist->spare_disks; | |
187 | if (mdlist->spare_group) | |
188 | st->spare_group = xstrdup(mdlist->spare_group); | |
189 | statelist = st; | |
190 | } | |
191 | } else { | |
192 | struct mddev_dev *dv; | |
193 | ||
194 | for (dv = devlist; dv; dv = dv->next) { | |
195 | struct state *st = xcalloc(1, sizeof *st); | |
196 | mdlist = conf_get_ident(dv->devname); | |
197 | st->devname = xstrdup(dv->devname); | |
198 | st->next = statelist; | |
199 | st->devnm[0] = 0; | |
200 | st->percent = RESYNC_UNKNOWN; | |
201 | st->expected_spares = -1; | |
202 | if (mdlist) { | |
203 | st->expected_spares = mdlist->spare_disks; | |
204 | if (mdlist->spare_group) | |
205 | st->spare_group = xstrdup(mdlist->spare_group); | |
206 | } | |
207 | statelist = st; | |
208 | } | |
209 | } | |
210 | ||
211 | while (!finished) { | |
212 | int new_found = 0; | |
213 | struct state *st, **stp; | |
214 | int anydegraded = 0; | |
215 | ||
216 | if (mdstat) | |
217 | free_mdstat(mdstat); | |
218 | mdstat = mdstat_read(oneshot ? 0 : 1, 0); | |
219 | if (!mdstat) | |
220 | mdstat_close(); | |
221 | ||
222 | for (st = statelist; st; st = st->next) | |
223 | if (check_array(st, mdstat, c->test, &info, | |
224 | increments, c->prefer)) | |
225 | anydegraded = 1; | |
226 | ||
227 | /* now check if there are any new devices found in mdstat */ | |
228 | if (c->scan) | |
229 | new_found = add_new_arrays(mdstat, &statelist, c->test, | |
230 | &info); | |
231 | ||
232 | /* If an array has active < raid && spare == 0 && spare_group != NULL | |
233 | * Look for another array with spare > 0 and active == raid and same spare_group | |
234 | * if found, choose a device and hotremove/hotadd | |
235 | */ | |
236 | if (share && anydegraded) | |
237 | try_spare_migration(statelist, &info); | |
238 | if (!new_found) { | |
239 | if (oneshot) | |
240 | break; | |
241 | else | |
242 | mdstat_wait(c->delay); | |
243 | } | |
244 | c->test = 0; | |
245 | ||
246 | for (stp = &statelist; (st = *stp) != NULL; ) { | |
247 | if (st->from_auto && st->err > 5) { | |
248 | *stp = st->next; | |
249 | free(st->devname); | |
250 | free(st->spare_group); | |
251 | free(st); | |
252 | } else | |
253 | stp = &st->next; | |
254 | } | |
255 | } | |
256 | for (st2 = statelist; st2; st2 = statelist) { | |
257 | statelist = st2->next; | |
258 | free(st2); | |
259 | } | |
260 | ||
261 | if (pidfile) | |
262 | unlink(pidfile); | |
263 | return 0; | |
264 | } | |
265 | ||
266 | static int make_daemon(char *pidfile) | |
267 | { | |
268 | /* Return: | |
269 | * -1 in the forked daemon | |
270 | * 0 in the parent | |
271 | * 1 on error | |
272 | * so a none-negative becomes the exit code. | |
273 | */ | |
274 | int pid = fork(); | |
275 | if (pid > 0) { | |
276 | if (!pidfile) | |
277 | printf("%d\n", pid); | |
278 | else { | |
279 | FILE *pid_file; | |
280 | pid_file=fopen(pidfile, "w"); | |
281 | if (!pid_file) | |
282 | perror("cannot create pid file"); | |
283 | else { | |
284 | fprintf(pid_file,"%d\n", pid); | |
285 | fclose(pid_file); | |
286 | } | |
287 | } | |
288 | return 0; | |
289 | } | |
290 | if (pid < 0) { | |
291 | perror("daemonise"); | |
292 | return 1; | |
293 | } | |
294 | close(0); | |
295 | open("/dev/null", O_RDWR); | |
296 | dup2(0, 1); | |
297 | dup2(0, 2); | |
298 | setsid(); | |
299 | return -1; | |
300 | } | |
301 | ||
302 | static int check_one_sharer(int scan) | |
303 | { | |
304 | int pid; | |
305 | FILE *comm_fp; | |
306 | FILE *fp; | |
307 | char comm_path[100]; | |
308 | char path[100]; | |
309 | char comm[20]; | |
310 | ||
311 | sprintf(path, "%s/autorebuild.pid", MDMON_DIR); | |
312 | fp = fopen(path, "r"); | |
313 | if (fp) { | |
314 | if (fscanf(fp, "%d", &pid) != 1) | |
315 | pid = -1; | |
316 | snprintf(comm_path, sizeof(comm_path), | |
317 | "/proc/%d/comm", pid); | |
318 | comm_fp = fopen(comm_path, "r"); | |
319 | if (comm_fp) { | |
320 | if (fscanf(comm_fp, "%s", comm) && | |
321 | strncmp(basename(comm), Name, strlen(Name)) == 0) { | |
322 | if (scan) { | |
323 | pr_err("Only one autorebuild process allowed in scan mode, aborting\n"); | |
324 | fclose(comm_fp); | |
325 | fclose(fp); | |
326 | return 1; | |
327 | } else { | |
328 | pr_err("Warning: One autorebuild process already running.\n"); | |
329 | } | |
330 | } | |
331 | fclose(comm_fp); | |
332 | } | |
333 | fclose(fp); | |
334 | } | |
335 | if (scan) { | |
336 | if (mkdir(MDMON_DIR, S_IRWXU) < 0 && errno != EEXIST) { | |
337 | pr_err("Can't create autorebuild.pid file\n"); | |
338 | } else { | |
339 | fp = fopen(path, "w"); | |
340 | if (!fp) | |
341 | pr_err("Cannot create autorebuild.pidfile\n"); | |
342 | else { | |
343 | pid = getpid(); | |
344 | fprintf(fp, "%d\n", pid); | |
345 | fclose(fp); | |
346 | } | |
347 | } | |
348 | } | |
349 | return 0; | |
350 | } | |
351 | ||
352 | static void alert(char *event, char *dev, char *disc, struct alert_info *info) | |
353 | { | |
354 | int priority; | |
355 | ||
356 | if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) { | |
357 | time_t now = time(0); | |
358 | ||
359 | printf("%1.15s: %s on %s %s\n", ctime(&now) + 4, | |
360 | event, dev, disc?disc:"unknown device"); | |
361 | } | |
362 | if (info->alert_cmd) { | |
363 | int pid = fork(); | |
364 | switch(pid) { | |
365 | default: | |
366 | waitpid(pid, NULL, 0); | |
367 | break; | |
368 | case -1: | |
369 | break; | |
370 | case 0: | |
371 | execl(info->alert_cmd, info->alert_cmd, | |
372 | event, dev, disc, NULL); | |
373 | exit(2); | |
374 | } | |
375 | } | |
376 | if (info->mailaddr && (strncmp(event, "Fail", 4) == 0 || | |
377 | strncmp(event, "Test", 4) == 0 || | |
378 | strncmp(event, "Spares", 6) == 0 || | |
379 | strncmp(event, "Degrade", 7) == 0)) { | |
380 | FILE *mp = popen(Sendmail, "w"); | |
381 | if (mp) { | |
382 | FILE *mdstat; | |
383 | char hname[256]; | |
384 | gethostname(hname, sizeof(hname)); | |
385 | signal(SIGPIPE, SIG_IGN); | |
386 | if (info->mailfrom) | |
387 | fprintf(mp, "From: %s\n", info->mailfrom); | |
388 | else | |
389 | fprintf(mp, "From: %s monitoring <root>\n", | |
390 | Name); | |
391 | fprintf(mp, "To: %s\n", info->mailaddr); | |
392 | fprintf(mp, "Subject: %s event on %s:%s\n\n", | |
393 | event, dev, hname); | |
394 | ||
395 | fprintf(mp, | |
396 | "This is an automatically generated mail message from %s\n", Name); | |
397 | fprintf(mp, "running on %s\n\n", hname); | |
398 | ||
399 | fprintf(mp, | |
400 | "A %s event had been detected on md device %s.\n\n", event, dev); | |
401 | ||
402 | if (disc && disc[0] != ' ') | |
403 | fprintf(mp, | |
404 | "It could be related to component device %s.\n\n", disc); | |
405 | if (disc && disc[0] == ' ') | |
406 | fprintf(mp, "Extra information:%s.\n\n", disc); | |
407 | ||
408 | fprintf(mp, "Faithfully yours, etc.\n"); | |
409 | ||
410 | mdstat = fopen("/proc/mdstat", "r"); | |
411 | if (mdstat) { | |
412 | char buf[8192]; | |
413 | int n; | |
414 | fprintf(mp, | |
415 | "\nP.S. The /proc/mdstat file currently contains the following:\n\n"); | |
416 | while ((n = fread(buf, 1, sizeof(buf), | |
417 | mdstat)) > 0) | |
418 | n = fwrite(buf, 1, n, mp); | |
419 | fclose(mdstat); | |
420 | } | |
421 | pclose(mp); | |
422 | } | |
423 | } | |
424 | ||
425 | /* log the event to syslog maybe */ | |
426 | if (info->dosyslog) { | |
427 | /* Log at a different severity depending on the event. | |
428 | * | |
429 | * These are the critical events: */ | |
430 | if (strncmp(event, "Fail", 4) == 0 || | |
431 | strncmp(event, "Degrade", 7) == 0 || | |
432 | strncmp(event, "DeviceDisappeared", 17) == 0) | |
433 | priority = LOG_CRIT; | |
434 | /* Good to know about, but are not failures: */ | |
435 | else if (strncmp(event, "Rebuild", 7) == 0 || | |
436 | strncmp(event, "MoveSpare", 9) == 0 || | |
437 | strncmp(event, "Spares", 6) != 0) | |
438 | priority = LOG_WARNING; | |
439 | /* Everything else: */ | |
440 | else | |
441 | priority = LOG_INFO; | |
442 | ||
443 | if (disc && disc[0] != ' ') | |
444 | syslog(priority, | |
445 | "%s event detected on md device %s, component device %s", event, dev, disc); | |
446 | else if (disc) | |
447 | syslog(priority, | |
448 | "%s event detected on md device %s: %s", | |
449 | event, dev, disc); | |
450 | else | |
451 | syslog(priority, | |
452 | "%s event detected on md device %s", | |
453 | event, dev); | |
454 | } | |
455 | } | |
456 | ||
457 | static int check_array(struct state *st, struct mdstat_ent *mdstat, | |
458 | int test, struct alert_info *ainfo, | |
459 | int increments, char *prefer) | |
460 | { | |
461 | /* Update the state 'st' to reflect any changes shown in mdstat, | |
462 | * or found by directly examining the array, and return | |
463 | * '1' if the array is degraded, or '0' if it is optimal (or dead). | |
464 | */ | |
465 | struct { int state, major, minor; } info[MAX_DISKS]; | |
466 | struct mdinfo *sra = NULL; | |
467 | mdu_array_info_t array; | |
468 | struct mdstat_ent *mse = NULL, *mse2; | |
469 | char *dev = st->devname; | |
470 | int fd; | |
471 | int i; | |
472 | int remaining_disks; | |
473 | int last_disk; | |
474 | int new_array = 0; | |
475 | int retval; | |
476 | int is_container = 0; | |
477 | unsigned long redundancy_only_flags = 0; | |
478 | ||
479 | if (test) | |
480 | alert("TestMessage", dev, NULL, ainfo); | |
481 | ||
482 | retval = 0; | |
483 | ||
484 | fd = open(dev, O_RDONLY); | |
485 | if (fd < 0) | |
486 | goto disappeared; | |
487 | ||
488 | if (st->devnm[0] == 0) | |
489 | strcpy(st->devnm, fd2devnm(fd)); | |
490 | ||
491 | for (mse2 = mdstat; mse2; mse2 = mse2->next) | |
492 | if (strcmp(mse2->devnm, st->devnm) == 0) { | |
493 | mse2->devnm[0] = 0; /* 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 | */ | |
501 | st->err++; | |
502 | goto out; | |
503 | } | |
504 | ||
505 | if (mse->level == NULL) | |
506 | is_container = 1; | |
507 | ||
508 | if (!is_container && !md_array_active(fd)) | |
509 | goto disappeared; | |
510 | ||
511 | fcntl(fd, F_SETFD, FD_CLOEXEC); | |
512 | if (md_get_array_info(fd, &array) < 0) | |
513 | goto disappeared; | |
514 | ||
515 | if (!is_container && map_name(pers, mse->level) > 0) | |
516 | redundancy_only_flags |= GET_MISMATCH; | |
517 | ||
518 | sra = sysfs_read(-1, st->devnm, GET_LEVEL | GET_DISKS | GET_DEVS | | |
519 | GET_STATE | redundancy_only_flags); | |
520 | ||
521 | if (!sra) | |
522 | goto disappeared; | |
523 | ||
524 | /* It's much easier to list what array levels can't | |
525 | * have a device disappear than all of them that can | |
526 | */ | |
527 | if (sra->array.level == 0 || sra->array.level == -1) { | |
528 | if (!st->err && !st->from_config) | |
529 | alert("DeviceDisappeared", dev, " Wrong-Level", ainfo); | |
530 | st->err++; | |
531 | goto out; | |
532 | } | |
533 | ||
534 | /* this array is in /proc/mdstat */ | |
535 | if (array.utime == 0) | |
536 | /* external arrays don't update utime, so | |
537 | * just make sure it is always different. */ | |
538 | array.utime = st->utime + 1;; | |
539 | ||
540 | if (st->err) { | |
541 | /* New array appeared where previously had an error */ | |
542 | st->err = 0; | |
543 | st->percent = RESYNC_NONE; | |
544 | new_array = 1; | |
545 | alert("NewArray", st->devname, NULL, ainfo); | |
546 | } | |
547 | ||
548 | if (st->utime == array.utime && st->failed == sra->array.failed_disks && | |
549 | st->working == sra->array.working_disks && | |
550 | st->spare == sra->array.spare_disks && | |
551 | (mse == NULL || (mse->percent == st->percent))) { | |
552 | if ((st->active < st->raid) && st->spare == 0) | |
553 | retval = 1; | |
554 | goto out; | |
555 | } | |
556 | if (st->utime == 0 && /* new array */ | |
557 | mse->pattern && strchr(mse->pattern, '_') /* degraded */) | |
558 | alert("DegradedArray", dev, NULL, ainfo); | |
559 | ||
560 | if (st->utime == 0 && /* new array */ st->expected_spares > 0 && | |
561 | sra->array.spare_disks < st->expected_spares) | |
562 | alert("SparesMissing", dev, NULL, ainfo); | |
563 | if (st->percent < 0 && st->percent != RESYNC_UNKNOWN && | |
564 | mse->percent >= 0) | |
565 | alert("RebuildStarted", dev, NULL, ainfo); | |
566 | if (st->percent >= 0 && mse->percent >= 0 && | |
567 | (mse->percent / increments) > (st->percent / increments)) { | |
568 | char percentalert[18]; | |
569 | /* | |
570 | * "RebuildNN" (10 chars) or "RebuildStarted" (15 chars) | |
571 | */ | |
572 | ||
573 | if((mse->percent / increments) == 0) | |
574 | snprintf(percentalert, sizeof(percentalert), | |
575 | "RebuildStarted"); | |
576 | else | |
577 | snprintf(percentalert, sizeof(percentalert), | |
578 | "Rebuild%02d", mse->percent); | |
579 | ||
580 | alert(percentalert, dev, NULL, ainfo); | |
581 | } | |
582 | ||
583 | if (mse->percent == RESYNC_NONE && st->percent >= 0) { | |
584 | /* Rebuild/sync/whatever just finished. | |
585 | * If there is a number in /mismatch_cnt, | |
586 | * we should report that. | |
587 | */ | |
588 | if (sra && sra->mismatch_cnt > 0) { | |
589 | char cnt[80]; | |
590 | snprintf(cnt, sizeof(cnt), | |
591 | " mismatches found: %d (on raid level %d)", | |
592 | sra->mismatch_cnt, sra->array.level); | |
593 | alert("RebuildFinished", dev, cnt, ainfo); | |
594 | } else | |
595 | alert("RebuildFinished", dev, NULL, ainfo); | |
596 | } | |
597 | st->percent = mse->percent; | |
598 | ||
599 | remaining_disks = sra->array.nr_disks; | |
600 | for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) { | |
601 | mdu_disk_info_t disc; | |
602 | disc.number = i; | |
603 | if (md_get_disk_info(fd, &disc) >= 0) { | |
604 | info[i].state = disc.state; | |
605 | info[i].major = disc.major; | |
606 | info[i].minor = disc.minor; | |
607 | if (disc.major || disc.minor) | |
608 | remaining_disks --; | |
609 | } else | |
610 | info[i].major = info[i].minor = 0; | |
611 | } | |
612 | last_disk = i; | |
613 | ||
614 | if (mse->metadata_version && | |
615 | strncmp(mse->metadata_version, "external:", 9) == 0 && | |
616 | is_subarray(mse->metadata_version+9)) { | |
617 | char *sl; | |
618 | strcpy(st->parent_devnm, mse->metadata_version + 10); | |
619 | sl = strchr(st->parent_devnm, '/'); | |
620 | if (sl) | |
621 | *sl = 0; | |
622 | } else | |
623 | st->parent_devnm[0] = 0; | |
624 | if (st->metadata == NULL && st->parent_devnm[0] == 0) | |
625 | st->metadata = super_by_fd(fd, NULL); | |
626 | ||
627 | for (i = 0; i < MAX_DISKS; i++) { | |
628 | mdu_disk_info_t disc = {0, 0, 0, 0, 0}; | |
629 | int newstate = 0; | |
630 | int change; | |
631 | char *dv = NULL; | |
632 | disc.number = i; | |
633 | if (i < last_disk && (info[i].major || info[i].minor)) { | |
634 | newstate = info[i].state; | |
635 | dv = map_dev_preferred(info[i].major, info[i].minor, 1, | |
636 | prefer); | |
637 | disc.state = newstate; | |
638 | disc.major = info[i].major; | |
639 | disc.minor = info[i].minor; | |
640 | } else | |
641 | newstate = (1 << MD_DISK_REMOVED); | |
642 | ||
643 | if (dv == NULL && st->devid[i]) | |
644 | dv = map_dev_preferred(major(st->devid[i]), | |
645 | minor(st->devid[i]), 1, prefer); | |
646 | change = newstate ^ st->devstate[i]; | |
647 | if (st->utime && change && !st->err && !new_array) { | |
648 | if ((st->devstate[i]&change) & (1 << MD_DISK_SYNC)) | |
649 | alert("Fail", dev, dv, ainfo); | |
650 | else if ((newstate & (1 << MD_DISK_FAULTY)) && | |
651 | (disc.major || disc.minor) && | |
652 | st->devid[i] == makedev(disc.major, | |
653 | disc.minor)) | |
654 | alert("FailSpare", dev, dv, ainfo); | |
655 | else if ((newstate&change) & (1 << MD_DISK_SYNC)) | |
656 | alert("SpareActive", dev, dv, ainfo); | |
657 | } | |
658 | st->devstate[i] = newstate; | |
659 | st->devid[i] = makedev(disc.major, disc.minor); | |
660 | } | |
661 | st->active = sra->array.active_disks; | |
662 | st->working = sra->array.working_disks; | |
663 | st->spare = sra->array.spare_disks; | |
664 | st->failed = sra->array.failed_disks; | |
665 | st->utime = array.utime; | |
666 | st->raid = sra->array.raid_disks; | |
667 | st->err = 0; | |
668 | if ((st->active < st->raid) && st->spare == 0) | |
669 | retval = 1; | |
670 | ||
671 | out: | |
672 | if (sra) | |
673 | sysfs_free(sra); | |
674 | if (fd >= 0) | |
675 | close(fd); | |
676 | return retval; | |
677 | ||
678 | disappeared: | |
679 | if (!st->err) | |
680 | alert("DeviceDisappeared", dev, NULL, ainfo); | |
681 | st->err++; | |
682 | goto out; | |
683 | } | |
684 | ||
685 | static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist, | |
686 | int test, struct alert_info *info) | |
687 | { | |
688 | struct mdstat_ent *mse; | |
689 | int new_found = 0; | |
690 | char *name; | |
691 | ||
692 | for (mse = mdstat; mse; mse = mse->next) | |
693 | if (mse->devnm[0] && (!mse->level || /* retrieve containers */ | |
694 | (strcmp(mse->level, "raid0") != 0 && | |
695 | strcmp(mse->level, "linear") != 0))) { | |
696 | struct state *st = xcalloc(1, sizeof *st); | |
697 | mdu_array_info_t array; | |
698 | int fd; | |
699 | ||
700 | name = get_md_name(mse->devnm); | |
701 | if (!name) { | |
702 | free(st); | |
703 | continue; | |
704 | } | |
705 | ||
706 | st->devname = xstrdup(name); | |
707 | if ((fd = open(st->devname, O_RDONLY)) < 0 || | |
708 | md_get_array_info(fd, &array) < 0) { | |
709 | /* no such array */ | |
710 | if (fd >= 0) | |
711 | close(fd); | |
712 | put_md_name(st->devname); | |
713 | free(st->devname); | |
714 | if (st->metadata) { | |
715 | st->metadata->ss->free_super(st->metadata); | |
716 | free(st->metadata); | |
717 | } | |
718 | free(st); | |
719 | continue; | |
720 | } | |
721 | close(fd); | |
722 | st->next = *statelist; | |
723 | st->err = 1; | |
724 | st->from_auto = 1; | |
725 | strcpy(st->devnm, mse->devnm); | |
726 | st->percent = RESYNC_UNKNOWN; | |
727 | st->expected_spares = -1; | |
728 | if (mse->metadata_version && | |
729 | strncmp(mse->metadata_version, | |
730 | "external:", 9) == 0 && | |
731 | is_subarray(mse->metadata_version+9)) { | |
732 | char *sl; | |
733 | strcpy(st->parent_devnm, | |
734 | mse->metadata_version+10); | |
735 | sl = strchr(st->parent_devnm, '/'); | |
736 | *sl = 0; | |
737 | } else | |
738 | st->parent_devnm[0] = 0; | |
739 | *statelist = st; | |
740 | if (test) | |
741 | alert("TestMessage", st->devname, NULL, info); | |
742 | new_found = 1; | |
743 | } | |
744 | return new_found; | |
745 | } | |
746 | ||
747 | static int get_required_spare_criteria(struct state *st, | |
748 | struct spare_criteria *sc) | |
749 | { | |
750 | int fd; | |
751 | ||
752 | if (!st->metadata || !st->metadata->ss->get_spare_criteria) { | |
753 | sc->min_size = 0; | |
754 | sc->sector_size = 0; | |
755 | return 0; | |
756 | } | |
757 | ||
758 | fd = open(st->devname, O_RDONLY); | |
759 | if (fd < 0) | |
760 | return 1; | |
761 | if (st->metadata->ss->external) | |
762 | st->metadata->ss->load_container(st->metadata, fd, st->devname); | |
763 | else | |
764 | st->metadata->ss->load_super(st->metadata, fd, st->devname); | |
765 | close(fd); | |
766 | if (!st->metadata->sb) | |
767 | return 1; | |
768 | ||
769 | st->metadata->ss->get_spare_criteria(st->metadata, sc); | |
770 | st->metadata->ss->free_super(st->metadata); | |
771 | ||
772 | return 0; | |
773 | } | |
774 | ||
775 | static int check_donor(struct state *from, struct state *to) | |
776 | { | |
777 | struct state *sub; | |
778 | ||
779 | if (from == to) | |
780 | return 0; | |
781 | if (from->parent) | |
782 | /* Cannot move from a member */ | |
783 | return 0; | |
784 | if (from->err) | |
785 | return 0; | |
786 | for (sub = from->subarray; sub; sub = sub->subarray) | |
787 | /* If source array has degraded subarrays, don't | |
788 | * remove anything | |
789 | */ | |
790 | if (sub->active < sub->raid) | |
791 | return 0; | |
792 | if (from->metadata->ss->external == 0) | |
793 | if (from->active < from->raid) | |
794 | return 0; | |
795 | if (from->spare <= 0) | |
796 | return 0; | |
797 | return 1; | |
798 | } | |
799 | ||
800 | static dev_t choose_spare(struct state *from, struct state *to, | |
801 | struct domainlist *domlist, struct spare_criteria *sc) | |
802 | { | |
803 | int d; | |
804 | dev_t dev = 0; | |
805 | ||
806 | for (d = from->raid; !dev && d < MAX_DISKS; d++) { | |
807 | if (from->devid[d] > 0 && from->devstate[d] == 0) { | |
808 | struct dev_policy *pol; | |
809 | unsigned long long dev_size; | |
810 | unsigned int dev_sector_size; | |
811 | ||
812 | if (to->metadata->ss->external && | |
813 | test_partition_from_id(from->devid[d])) | |
814 | continue; | |
815 | ||
816 | if (sc->min_size && | |
817 | dev_size_from_id(from->devid[d], &dev_size) && | |
818 | dev_size < sc->min_size) | |
819 | continue; | |
820 | ||
821 | if (sc->sector_size && | |
822 | dev_sector_size_from_id(from->devid[d], | |
823 | &dev_sector_size) && | |
824 | sc->sector_size != dev_sector_size) | |
825 | continue; | |
826 | ||
827 | pol = devid_policy(from->devid[d]); | |
828 | if (from->spare_group) | |
829 | pol_add(&pol, pol_domain, | |
830 | from->spare_group, NULL); | |
831 | if (domain_test(domlist, pol, | |
832 | to->metadata->ss->name) == 1) | |
833 | dev = from->devid[d]; | |
834 | dev_policy_free(pol); | |
835 | } | |
836 | } | |
837 | return dev; | |
838 | } | |
839 | ||
840 | static dev_t container_choose_spare(struct state *from, struct state *to, | |
841 | struct domainlist *domlist, | |
842 | struct spare_criteria *sc, int active) | |
843 | { | |
844 | /* This is similar to choose_spare, but we cannot trust devstate, | |
845 | * so we need to read the metadata instead | |
846 | */ | |
847 | struct mdinfo *list; | |
848 | struct supertype *st = from->metadata; | |
849 | int fd = open(from->devname, O_RDONLY); | |
850 | int err; | |
851 | dev_t dev = 0; | |
852 | ||
853 | if (fd < 0) | |
854 | return 0; | |
855 | if (!st->ss->getinfo_super_disks) { | |
856 | close(fd); | |
857 | return 0; | |
858 | } | |
859 | ||
860 | err = st->ss->load_container(st, fd, NULL); | |
861 | close(fd); | |
862 | if (err) | |
863 | return 0; | |
864 | ||
865 | if (from == to) { | |
866 | /* We must check if number of active disks has not increased | |
867 | * since ioctl in main loop. mdmon may have added spare | |
868 | * to subarray. If so we do not need to look for more spares | |
869 | * so return non zero value */ | |
870 | int active_cnt = 0; | |
871 | struct mdinfo *dp; | |
872 | list = st->ss->getinfo_super_disks(st); | |
873 | if (!list) { | |
874 | st->ss->free_super(st); | |
875 | return 1; | |
876 | } | |
877 | dp = list->devs; | |
878 | while (dp) { | |
879 | if (dp->disk.state & (1 << MD_DISK_SYNC) && | |
880 | !(dp->disk.state & (1 << MD_DISK_FAULTY))) | |
881 | active_cnt++; | |
882 | dp = dp->next; | |
883 | } | |
884 | sysfs_free(list); | |
885 | if (active < active_cnt) { | |
886 | /* Spare just activated.*/ | |
887 | st->ss->free_super(st); | |
888 | return 1; | |
889 | } | |
890 | } | |
891 | ||
892 | /* We only need one spare so full list not needed */ | |
893 | list = container_choose_spares(st, sc, domlist, from->spare_group, | |
894 | to->metadata->ss->name, 1); | |
895 | if (list) { | |
896 | struct mdinfo *disks = list->devs; | |
897 | if (disks) | |
898 | dev = makedev(disks->disk.major, disks->disk.minor); | |
899 | sysfs_free(list); | |
900 | } | |
901 | st->ss->free_super(st); | |
902 | return dev; | |
903 | } | |
904 | ||
905 | static void try_spare_migration(struct state *statelist, struct alert_info *info) | |
906 | { | |
907 | struct state *from; | |
908 | struct state *st; | |
909 | struct spare_criteria sc; | |
910 | ||
911 | link_containers_with_subarrays(statelist); | |
912 | for (st = statelist; st; st = st->next) | |
913 | if (st->active < st->raid && st->spare == 0 && !st->err) { | |
914 | struct domainlist *domlist = NULL; | |
915 | int d; | |
916 | struct state *to = st; | |
917 | ||
918 | if (to->parent_devnm[0] && !to->parent) | |
919 | /* subarray monitored without parent container | |
920 | * we can't move spares here */ | |
921 | continue; | |
922 | ||
923 | if (to->parent) | |
924 | /* member of a container */ | |
925 | to = to->parent; | |
926 | ||
927 | if (get_required_spare_criteria(to, &sc)) | |
928 | continue; | |
929 | if (to->metadata->ss->external) { | |
930 | /* We must make sure there is | |
931 | * no suitable spare in container already. | |
932 | * If there is we don't add more */ | |
933 | dev_t devid = container_choose_spare( | |
934 | to, to, NULL, &sc, st->active); | |
935 | if (devid > 0) | |
936 | continue; | |
937 | } | |
938 | for (d = 0; d < MAX_DISKS; d++) | |
939 | if (to->devid[d]) | |
940 | domainlist_add_dev(&domlist, | |
941 | to->devid[d], | |
942 | to->metadata->ss->name); | |
943 | if (to->spare_group) | |
944 | domain_add(&domlist, to->spare_group); | |
945 | /* | |
946 | * No spare migration if the destination | |
947 | * has no domain. Skip this array. | |
948 | */ | |
949 | if (!domlist) | |
950 | continue; | |
951 | for (from=statelist ; from ; from=from->next) { | |
952 | dev_t devid; | |
953 | if (!check_donor(from, to)) | |
954 | continue; | |
955 | if (from->metadata->ss->external) | |
956 | devid = container_choose_spare( | |
957 | from, to, domlist, &sc, 0); | |
958 | else | |
959 | devid = choose_spare(from, to, domlist, | |
960 | &sc); | |
961 | if (devid > 0 && | |
962 | move_spare(from->devname, to->devname, | |
963 | devid)) { | |
964 | alert("MoveSpare", to->devname, | |
965 | from->devname, info); | |
966 | break; | |
967 | } | |
968 | } | |
969 | domain_free(domlist); | |
970 | } | |
971 | } | |
972 | ||
973 | /* search the statelist to connect external | |
974 | * metadata subarrays with their containers | |
975 | * We always completely rebuild the tree from scratch as | |
976 | * that is safest considering the possibility of entries | |
977 | * disappearing or changing. | |
978 | */ | |
979 | static void link_containers_with_subarrays(struct state *list) | |
980 | { | |
981 | struct state *st; | |
982 | struct state *cont; | |
983 | for (st = list; st; st = st->next) { | |
984 | st->parent = NULL; | |
985 | st->subarray = NULL; | |
986 | } | |
987 | for (st = list; st; st = st->next) | |
988 | if (st->parent_devnm[0]) | |
989 | for (cont = list; cont; cont = cont->next) | |
990 | if (!cont->err && cont->parent_devnm[0] == 0 && | |
991 | strcmp(cont->devnm, st->parent_devnm) == 0) { | |
992 | st->parent = cont; | |
993 | st->subarray = cont->subarray; | |
994 | cont->subarray = st; | |
995 | break; | |
996 | } | |
997 | } | |
998 | ||
999 | /* Not really Monitor but ... */ | |
1000 | int Wait(char *dev) | |
1001 | { | |
1002 | char devnm[32]; | |
1003 | dev_t rdev; | |
1004 | char *tmp; | |
1005 | int rv = 1; | |
1006 | int frozen_remaining = 3; | |
1007 | ||
1008 | if (!stat_is_blkdev(dev, &rdev)) | |
1009 | return 2; | |
1010 | ||
1011 | tmp = devid2devnm(rdev); | |
1012 | if (!tmp) { | |
1013 | pr_err("Cannot get md device name.\n"); | |
1014 | return 2; | |
1015 | } | |
1016 | ||
1017 | strcpy(devnm, tmp); | |
1018 | ||
1019 | while(1) { | |
1020 | struct mdstat_ent *ms = mdstat_read(1, 0); | |
1021 | struct mdstat_ent *e; | |
1022 | ||
1023 | for (e = ms; e; e = e->next) | |
1024 | if (strcmp(e->devnm, devnm) == 0) | |
1025 | break; | |
1026 | ||
1027 | if (e && e->percent == RESYNC_NONE) { | |
1028 | /* We could be in the brief pause before something | |
1029 | * starts. /proc/mdstat doesn't show that, but | |
1030 | * sync_action does. | |
1031 | */ | |
1032 | struct mdinfo mdi; | |
1033 | char buf[21]; | |
1034 | ||
1035 | if (sysfs_init(&mdi, -1, devnm)) | |
1036 | return 2; | |
1037 | if (sysfs_get_str(&mdi, NULL, "sync_action", | |
1038 | buf, 20) > 0 && | |
1039 | strcmp(buf,"idle\n") != 0) { | |
1040 | e->percent = RESYNC_UNKNOWN; | |
1041 | if (strcmp(buf, "frozen\n") == 0) { | |
1042 | if (frozen_remaining == 0) | |
1043 | e->percent = RESYNC_NONE; | |
1044 | else | |
1045 | frozen_remaining -= 1; | |
1046 | } | |
1047 | } | |
1048 | } | |
1049 | if (!e || e->percent == RESYNC_NONE) { | |
1050 | if (e && e->metadata_version && | |
1051 | strncmp(e->metadata_version, "external:", 9) == 0) { | |
1052 | if (is_subarray(&e->metadata_version[9])) | |
1053 | ping_monitor(&e->metadata_version[9]); | |
1054 | else | |
1055 | ping_monitor(devnm); | |
1056 | } | |
1057 | free_mdstat(ms); | |
1058 | return rv; | |
1059 | } | |
1060 | free_mdstat(ms); | |
1061 | rv = 0; | |
1062 | mdstat_wait(5); | |
1063 | } | |
1064 | } | |
1065 | ||
1066 | /* The state "broken" is used only for RAID0/LINEAR - it's the same as | |
1067 | * "clean", but used in case the array has one or more members missing. | |
1068 | */ | |
1069 | static char *clean_states[] = { | |
1070 | "clear", "inactive", "readonly", "read-auto", "clean", "broken", NULL }; | |
1071 | ||
1072 | int WaitClean(char *dev, int verbose) | |
1073 | { | |
1074 | int fd; | |
1075 | struct mdinfo *mdi; | |
1076 | int rv = 1; | |
1077 | char devnm[32]; | |
1078 | ||
1079 | if (!stat_is_blkdev(dev, NULL)) | |
1080 | return 2; | |
1081 | fd = open(dev, O_RDONLY); | |
1082 | if (fd < 0) { | |
1083 | if (verbose) | |
1084 | pr_err("Couldn't open %s: %s\n", dev, strerror(errno)); | |
1085 | return 1; | |
1086 | } | |
1087 | ||
1088 | strcpy(devnm, fd2devnm(fd)); | |
1089 | mdi = sysfs_read(fd, devnm, GET_VERSION|GET_LEVEL|GET_SAFEMODE); | |
1090 | if (!mdi) { | |
1091 | if (verbose) | |
1092 | pr_err("Failed to read sysfs attributes for %s\n", dev); | |
1093 | close(fd); | |
1094 | return 0; | |
1095 | } | |
1096 | ||
1097 | switch(mdi->array.level) { | |
1098 | case LEVEL_LINEAR: | |
1099 | case LEVEL_MULTIPATH: | |
1100 | case 0: | |
1101 | /* safemode delay is irrelevant for these levels */ | |
1102 | rv = 0; | |
1103 | } | |
1104 | ||
1105 | /* for internal metadata the kernel handles the final clean | |
1106 | * transition, containers can never be dirty | |
1107 | */ | |
1108 | if (!is_subarray(mdi->text_version)) | |
1109 | rv = 0; | |
1110 | ||
1111 | /* safemode disabled ? */ | |
1112 | if (mdi->safe_mode_delay == 0) | |
1113 | rv = 0; | |
1114 | ||
1115 | if (rv) { | |
1116 | int state_fd = sysfs_open(fd2devnm(fd), NULL, "array_state"); | |
1117 | char buf[20]; | |
1118 | int delay = 5000; | |
1119 | ||
1120 | /* minimize the safe_mode_delay and prepare to wait up to 5s | |
1121 | * for writes to quiesce | |
1122 | */ | |
1123 | sysfs_set_safemode(mdi, 1); | |
1124 | ||
1125 | /* wait for array_state to be clean */ | |
1126 | while (1) { | |
1127 | rv = read(state_fd, buf, sizeof(buf)); | |
1128 | if (rv < 0) | |
1129 | break; | |
1130 | if (sysfs_match_word(buf, clean_states) < | |
1131 | (int)ARRAY_SIZE(clean_states) - 1) | |
1132 | break; | |
1133 | rv = sysfs_wait(state_fd, &delay); | |
1134 | if (rv < 0 && errno != EINTR) | |
1135 | break; | |
1136 | lseek(state_fd, 0, SEEK_SET); | |
1137 | } | |
1138 | if (rv < 0) | |
1139 | rv = 1; | |
1140 | else if (ping_monitor(mdi->text_version) == 0) { | |
1141 | /* we need to ping to close the window between array | |
1142 | * state transitioning to clean and the metadata being | |
1143 | * marked clean | |
1144 | */ | |
1145 | rv = 0; | |
1146 | } else { | |
1147 | rv = 1; | |
1148 | pr_err("Error connecting monitor with %s\n", dev); | |
1149 | } | |
1150 | if (rv && verbose) | |
1151 | pr_err("Error waiting for %s to be clean\n", dev); | |
1152 | ||
1153 | /* restore the original safe_mode_delay */ | |
1154 | sysfs_set_safemode(mdi, mdi->safe_mode_delay); | |
1155 | close(state_fd); | |
1156 | } | |
1157 | ||
1158 | sysfs_free(mdi); | |
1159 | close(fd); | |
1160 | ||
1161 | return rv; | |
1162 | } |