]>
Commit | Line | Data |
---|---|---|
52826846 | 1 | /* |
9a9dab36 | 2 | * mdadm - manage Linux "md" devices aka RAID arrays. |
52826846 | 3 | * |
e736b623 | 4 | * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de> |
52826846 NB |
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 | |
e736b623 | 22 | * Email: <neilb@suse.de> |
52826846 NB |
23 | */ |
24 | ||
9a9dab36 | 25 | #include "mdadm.h" |
52826846 NB |
26 | #include "md_p.h" |
27 | #include "md_u.h" | |
e0d19036 | 28 | #include <sys/wait.h> |
98127a6d | 29 | #include <signal.h> |
4450e59f | 30 | #include <limits.h> |
773135f5 | 31 | #include <syslog.h> |
52826846 | 32 | |
e27d562b NB |
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 | |
2e0172b1 N |
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 | unsigned 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 *next; | |
54 | }; | |
55 | ||
56 | static int make_daemon(char *pidfile); | |
57 | static int check_one_sharer(int scan); | |
58 | static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mailfrom, | |
59 | char *cmd, int dosyslog); | |
60 | static void check_array(struct state *st, struct mdstat_ent *mdstat, | |
61 | int test, char *mailaddr, | |
62 | char *mailfrom, char *alert_cmd, int dosyslog, | |
63 | int increments); | |
64 | static int add_new_arrays(struct mdstat_ent *mdstat, struct state *statelist, | |
65 | int test, char *mailaddr, char *mailfrom, | |
66 | char *alert_cmd, int dosyslog); | |
67 | static void try_spare_migration(struct state *statelist, | |
68 | char *mailaddr, char *mailfrom, | |
69 | char *alert_cmd, int dosyslog); | |
70 | ||
a655e550 | 71 | int Monitor(struct mddev_dev *devlist, |
52826846 | 72 | char *mailaddr, char *alert_cmd, |
aa88f531 | 73 | int period, int daemonise, int scan, int oneshot, |
edde9560 AC |
74 | int dosyslog, int test, char *pidfile, int increments, |
75 | int share) | |
52826846 NB |
76 | { |
77 | /* | |
78 | * Every few seconds, scan every md device looking for changes | |
79 | * When a change is found, log it, possibly run the alert command, | |
80 | * and possibly send Email | |
81 | * | |
82 | * For each array, we record: | |
83 | * Update time | |
84 | * active/working/failed/spare drives | |
85 | * State of each device. | |
e0d19036 | 86 | * %rebuilt if rebuilding |
52826846 NB |
87 | * |
88 | * If the update time changes, check out all the data again | |
89 | * It is possible that we cannot get the state of each device | |
90 | * due to bugs in the md kernel module. | |
e0d19036 NB |
91 | * We also read /proc/mdstat to get rebuild percent, |
92 | * and to get state on all active devices incase of kernel bug. | |
52826846 | 93 | * |
e0d19036 NB |
94 | * Events are: |
95 | * Fail | |
96 | * An active device had Faulty set or Active/Sync removed | |
97 | * FailSpare | |
98 | * A spare device had Faulty set | |
99 | * SpareActive | |
100 | * An active device had a reverse transition | |
101 | * RebuildStarted | |
102 | * percent went from -1 to +ve | |
9a36a9b7 ZB |
103 | * RebuildNN |
104 | * percent went from below to not-below NN% | |
e0d19036 NB |
105 | * DeviceDisappeared |
106 | * Couldn't access a device which was previously visible | |
52826846 NB |
107 | * |
108 | * if we detect an array with active<raid and spare==0 | |
109 | * we look at other arrays that have same spare-group | |
110 | * If we find one with active==raid and spare>0, | |
111 | * and if we can get_disk_info and find a name | |
112 | * Then we hot-remove and hot-add to the other array | |
113 | * | |
e0d19036 NB |
114 | * If devlist is NULL, then we can monitor everything because --scan |
115 | * was given. We get an initial list from config file and add anything | |
116 | * that appears in /proc/mdstat | |
52826846 NB |
117 | */ |
118 | ||
2e0172b1 | 119 | struct state *statelist = NULL; |
52826846 | 120 | int finished = 0; |
e0d19036 | 121 | struct mdstat_ent *mdstat = NULL; |
4948b8f7 | 122 | char *mailfrom = NULL; |
e0d19036 | 123 | |
d013a55e | 124 | if (!mailaddr) { |
8aec876d | 125 | mailaddr = conf_get_mailaddr(); |
d013a55e | 126 | if (mailaddr && ! scan) |
56eedc1a | 127 | fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n", |
d013a55e NB |
128 | mailaddr); |
129 | } | |
8aec876d | 130 | mailfrom = conf_get_mailfrom(); |
4948b8f7 | 131 | |
d013a55e | 132 | if (!alert_cmd) { |
8aec876d | 133 | alert_cmd = conf_get_program(); |
d013a55e | 134 | if (alert_cmd && ! scan) |
56eedc1a | 135 | fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n", |
d013a55e NB |
136 | alert_cmd); |
137 | } | |
56eedc1a NB |
138 | if (scan && !mailaddr && !alert_cmd) { |
139 | fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n"); | |
d013a55e | 140 | return 1; |
56eedc1a | 141 | } |
d013a55e | 142 | |
2e0172b1 N |
143 | if (daemonise) |
144 | if (make_daemon(pidfile)) | |
d013a55e | 145 | return 1; |
e0d19036 | 146 | |
2e0172b1 N |
147 | if (share) |
148 | if (check_one_sharer(scan)) | |
149 | return 1; | |
edde9560 | 150 | |
e0d19036 | 151 | if (devlist == NULL) { |
fa56eddb | 152 | struct mddev_ident *mdlist = conf_get_ident(NULL); |
e0d19036 | 153 | for (; mdlist; mdlist=mdlist->next) { |
fe056d1f N |
154 | struct state *st; |
155 | if (mdlist->devname == NULL) | |
156 | continue; | |
112cace6 N |
157 | if (strcasecmp(mdlist->devname, "<ignore>") == 0) |
158 | continue; | |
fe056d1f | 159 | st = malloc(sizeof *st); |
e0d19036 NB |
160 | if (st == NULL) |
161 | continue; | |
db2d001c N |
162 | if (mdlist->devname[0] == '/') |
163 | st->devname = strdup(mdlist->devname); | |
164 | else { | |
165 | st->devname = malloc(8+strlen(mdlist->devname)+1); | |
166 | strcpy(strcpy(st->devname, "/dev/md/"), | |
167 | mdlist->devname); | |
168 | } | |
e0d19036 NB |
169 | st->utime = 0; |
170 | st->next = statelist; | |
d013a55e | 171 | st->err = 0; |
4450e59f | 172 | st->devnum = INT_MAX; |
e0d19036 | 173 | st->percent = -2; |
feb716e9 | 174 | st->expected_spares = mdlist->spare_disks; |
e0d19036 NB |
175 | if (mdlist->spare_group) |
176 | st->spare_group = strdup(mdlist->spare_group); | |
177 | else | |
178 | st->spare_group = NULL; | |
179 | statelist = st; | |
180 | } | |
181 | } else { | |
a655e550 | 182 | struct mddev_dev *dv; |
e0d19036 | 183 | for (dv=devlist ; dv; dv=dv->next) { |
fa56eddb | 184 | struct mddev_ident *mdlist = conf_get_ident(dv->devname); |
e0d19036 NB |
185 | struct state *st = malloc(sizeof *st); |
186 | if (st == NULL) | |
187 | continue; | |
188 | st->devname = strdup(dv->devname); | |
189 | st->utime = 0; | |
190 | st->next = statelist; | |
d013a55e | 191 | st->err = 0; |
4450e59f | 192 | st->devnum = INT_MAX; |
e0d19036 | 193 | st->percent = -2; |
feb716e9 | 194 | st->expected_spares = -1; |
e0d19036 | 195 | st->spare_group = NULL; |
e5329c37 NB |
196 | if (mdlist) { |
197 | st->expected_spares = mdlist->spare_disks; | |
198 | if (mdlist->spare_group) | |
199 | st->spare_group = strdup(mdlist->spare_group); | |
200 | } | |
e0d19036 NB |
201 | statelist = st; |
202 | } | |
203 | } | |
204 | ||
205 | ||
206 | while (! finished) { | |
aa88f531 | 207 | int new_found = 0; |
e0d19036 NB |
208 | struct state *st; |
209 | ||
210 | if (mdstat) | |
211 | free_mdstat(mdstat); | |
22a88995 | 212 | mdstat = mdstat_read(oneshot?0:1, 0); |
e0d19036 | 213 | |
2e0172b1 N |
214 | for (st=statelist; st; st=st->next) |
215 | check_array(st, mdstat, test, mailaddr, mailfrom, | |
216 | alert_cmd, dosyslog, increments); | |
217 | ||
e0d19036 | 218 | /* now check if there are any new devices found in mdstat */ |
2e0172b1 N |
219 | if (scan) |
220 | new_found = add_new_arrays(mdstat, statelist, test, | |
221 | mailaddr, mailfrom, alert_cmd, | |
222 | dosyslog); | |
223 | ||
e0d19036 NB |
224 | /* If an array has active < raid && spare == 0 && spare_group != NULL |
225 | * Look for another array with spare > 0 and active == raid and same spare_group | |
226 | * if found, choose a device and hotremove/hotadd | |
227 | */ | |
2e0172b1 N |
228 | if (share) |
229 | try_spare_migration(statelist, mailaddr, mailfrom, | |
230 | alert_cmd, dosyslog); | |
aa88f531 NB |
231 | if (!new_found) { |
232 | if (oneshot) | |
233 | break; | |
234 | else | |
dd0781e5 | 235 | mdstat_wait(period); |
aa88f531 | 236 | } |
98c6faba | 237 | test = 0; |
52826846 | 238 | } |
b5e64645 NB |
239 | if (pidfile) |
240 | unlink(pidfile); | |
52826846 NB |
241 | return 0; |
242 | } | |
243 | ||
2e0172b1 N |
244 | static int make_daemon(char *pidfile) |
245 | { | |
246 | int pid = fork(); | |
247 | if (pid > 0) { | |
248 | if (!pidfile) | |
249 | printf("%d\n", pid); | |
250 | else { | |
251 | FILE *pid_file; | |
252 | pid_file=fopen(pidfile, "w"); | |
253 | if (!pid_file) | |
254 | perror("cannot create pid file"); | |
255 | else { | |
256 | fprintf(pid_file,"%d\n", pid); | |
257 | fclose(pid_file); | |
258 | } | |
259 | } | |
260 | return 0; | |
261 | } | |
262 | if (pid < 0) { | |
263 | perror("daemonise"); | |
264 | return 1; | |
265 | } | |
266 | close(0); | |
267 | open("/dev/null", O_RDWR); | |
268 | dup2(0,1); | |
269 | dup2(0,2); | |
270 | setsid(); | |
271 | return 0; | |
272 | } | |
273 | ||
274 | static int check_one_sharer(int scan) | |
275 | { | |
276 | int pid, rv; | |
277 | FILE *fp; | |
278 | char dir[20]; | |
279 | struct stat buf; | |
280 | fp = fopen("/var/run/mdadm/autorebuild.pid", "r"); | |
281 | if (fp) { | |
282 | fscanf(fp, "%d", &pid); | |
283 | sprintf(dir, "/proc/%d", pid); | |
284 | rv = stat(dir, &buf); | |
285 | if (rv != -1) { | |
286 | if (scan) { | |
287 | fprintf(stderr, Name ": Only one " | |
288 | "autorebuild process allowed" | |
289 | " in scan mode, aborting\n"); | |
290 | fclose(fp); | |
291 | return 1; | |
292 | } else { | |
293 | fprintf(stderr, Name ": Warning: One" | |
294 | " autorebuild process already" | |
295 | " running."); | |
296 | } | |
297 | } | |
298 | fclose(fp); | |
299 | } | |
300 | if (scan) { | |
301 | fp = fopen("/var/run/mdadm/autorebuild.pid", "w"); | |
302 | if (!fp) | |
303 | fprintf(stderr, Name ": Cannot create" | |
304 | " autorebuild.pid " | |
305 | "file\n"); | |
306 | else { | |
307 | pid = getpid(); | |
308 | fprintf(fp, "%d\n", pid); | |
309 | fclose(fp); | |
310 | } | |
311 | } | |
312 | return 0; | |
313 | } | |
52826846 | 314 | |
4948b8f7 | 315 | static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mailfrom, char *cmd, |
773135f5 | 316 | int dosyslog) |
52826846 | 317 | { |
773135f5 NB |
318 | int priority; |
319 | ||
cd29a5c8 NB |
320 | if (!cmd && !mailaddr) { |
321 | time_t now = time(0); | |
aba69144 | 322 | |
e0d19036 | 323 | printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device"); |
cd29a5c8 | 324 | } |
52826846 NB |
325 | if (cmd) { |
326 | int pid = fork(); | |
327 | switch(pid) { | |
328 | default: | |
329 | waitpid(pid, NULL, 0); | |
330 | break; | |
331 | case -1: | |
332 | break; | |
333 | case 0: | |
334 | execl(cmd, cmd, event, dev, disc, NULL); | |
335 | exit(2); | |
336 | } | |
337 | } | |
aba69144 NB |
338 | if (mailaddr && |
339 | (strncmp(event, "Fail", 4)==0 || | |
98c6faba | 340 | strncmp(event, "Test", 4)==0 || |
d1732eeb | 341 | strncmp(event, "Spares", 6)==0 || |
aa88f531 | 342 | strncmp(event, "Degrade", 7)==0)) { |
52826846 NB |
343 | FILE *mp = popen(Sendmail, "w"); |
344 | if (mp) { | |
008e1100 | 345 | FILE *mdstat; |
52826846 NB |
346 | char hname[256]; |
347 | gethostname(hname, sizeof(hname)); | |
348 | signal(SIGPIPE, SIG_IGN); | |
4948b8f7 NB |
349 | if (mailfrom) |
350 | fprintf(mp, "From: %s\n", mailfrom); | |
351 | else | |
352 | fprintf(mp, "From: " Name " monitoring <root>\n"); | |
52826846 NB |
353 | fprintf(mp, "To: %s\n", mailaddr); |
354 | fprintf(mp, "Subject: %s event on %s:%s\n\n", event, dev, hname); | |
355 | ||
356 | fprintf(mp, "This is an automatically generated mail message from " Name "\n"); | |
357 | fprintf(mp, "running on %s\n\n", hname); | |
358 | ||
359 | fprintf(mp, "A %s event had been detected on md device %s.\n\n", event, dev); | |
360 | ||
37dfc3d6 | 361 | if (disc && disc[0] != ' ') |
56eedc1a | 362 | fprintf(mp, "It could be related to component device %s.\n\n", disc); |
37dfc3d6 NB |
363 | if (disc && disc[0] == ' ') |
364 | fprintf(mp, "Extra information:%s.\n\n", disc); | |
52826846 NB |
365 | |
366 | fprintf(mp, "Faithfully yours, etc.\n"); | |
008e1100 NB |
367 | |
368 | mdstat = fopen("/proc/mdstat", "r"); | |
369 | if (mdstat) { | |
370 | char buf[8192]; | |
371 | int n; | |
a524a7ee | 372 | fprintf(mp, "\nP.S. The /proc/mdstat file currently contains the following:\n\n"); |
008e1100 | 373 | while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0) |
9fca7d62 | 374 | n=fwrite(buf, 1, n, mp); /* yes, i don't care about the result */ |
008e1100 NB |
375 | fclose(mdstat); |
376 | } | |
6278fb3a | 377 | pclose(mp); |
52826846 NB |
378 | } |
379 | ||
380 | } | |
773135f5 NB |
381 | |
382 | /* log the event to syslog maybe */ | |
383 | if (dosyslog) { | |
384 | /* Log at a different severity depending on the event. | |
385 | * | |
386 | * These are the critical events: */ | |
387 | if (strncmp(event, "Fail", 4)==0 || | |
388 | strncmp(event, "Degrade", 7)==0 || | |
389 | strncmp(event, "DeviceDisappeared", 17)==0) | |
390 | priority = LOG_CRIT; | |
391 | /* Good to know about, but are not failures: */ | |
392 | else if (strncmp(event, "Rebuild", 7)==0 || | |
d1732eeb NB |
393 | strncmp(event, "MoveSpare", 9)==0 || |
394 | strncmp(event, "Spares", 6) != 0) | |
773135f5 NB |
395 | priority = LOG_WARNING; |
396 | /* Everything else: */ | |
397 | else | |
398 | priority = LOG_INFO; | |
399 | ||
400 | if (disc) | |
401 | syslog(priority, "%s event detected on md device %s, component device %s", event, dev, disc); | |
402 | else | |
403 | syslog(priority, "%s event detected on md device %s", event, dev); | |
404 | } | |
52826846 | 405 | } |
b90c0e9a | 406 | |
2e0172b1 N |
407 | static void check_array(struct state *st, struct mdstat_ent *mdstat, |
408 | int test, char *mailaddr, | |
409 | char *mailfrom, char *alert_cmd, int dosyslog, | |
410 | int increments) | |
411 | { | |
412 | struct { int state, major, minor; } info[MaxDisks]; | |
413 | mdu_array_info_t array; | |
414 | struct mdstat_ent *mse = NULL, *mse2; | |
415 | char *dev = st->devname; | |
416 | int fd; | |
417 | int i; | |
418 | ||
419 | if (test) | |
420 | alert("TestMessage", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
421 | fd = open(dev, O_RDONLY); | |
422 | if (fd < 0) { | |
423 | if (!st->err) | |
424 | alert("DeviceDisappeared", dev, NULL, | |
425 | mailaddr, mailfrom, alert_cmd, dosyslog); | |
426 | /* fprintf(stderr, Name ": cannot open %s: %s\n", | |
427 | dev, strerror(errno)); | |
428 | */ st->err=1; | |
429 | return; | |
430 | } | |
431 | fcntl(fd, F_SETFD, FD_CLOEXEC); | |
432 | if (ioctl(fd, GET_ARRAY_INFO, &array)<0) { | |
433 | if (!st->err) | |
434 | alert("DeviceDisappeared", dev, NULL, | |
435 | mailaddr, mailfrom, alert_cmd, dosyslog); | |
436 | /* fprintf(stderr, Name ": cannot get array info for %s: %s\n", | |
437 | dev, strerror(errno)); | |
438 | */ st->err=1; | |
439 | close(fd); | |
440 | return; | |
441 | } | |
442 | /* It's much easier to list what array levels can't | |
443 | * have a device disappear than all of them that can | |
444 | */ | |
445 | if (array.level == 0 || array.level == -1) { | |
446 | if (!st->err) | |
447 | alert("DeviceDisappeared", dev, "Wrong-Level", | |
448 | mailaddr, mailfrom, alert_cmd, dosyslog); | |
449 | st->err = 1; | |
450 | close(fd); | |
451 | return; | |
452 | } | |
453 | if (st->devnum == INT_MAX) { | |
454 | struct stat stb; | |
455 | if (fstat(fd, &stb) == 0 && | |
456 | (S_IFMT&stb.st_mode)==S_IFBLK) { | |
457 | if (major(stb.st_rdev) == MD_MAJOR) | |
458 | st->devnum = minor(stb.st_rdev); | |
459 | else | |
460 | st->devnum = -1- (minor(stb.st_rdev)>>6); | |
461 | } | |
462 | } | |
463 | ||
464 | for (mse2 = mdstat ; mse2 ; mse2=mse2->next) | |
465 | if (mse2->devnum == st->devnum) { | |
466 | mse2->devnum = INT_MAX; /* flag it as "used" */ | |
467 | mse = mse2; | |
468 | } | |
469 | ||
470 | if (!mse) { | |
471 | /* duplicated array in statelist | |
472 | * or re-created after reading mdstat*/ | |
473 | st->err = 1; | |
474 | close(fd); | |
475 | return; | |
476 | } | |
477 | /* this array is in /proc/mdstat */ | |
478 | if (array.utime == 0) | |
479 | /* external arrays don't update utime, so | |
480 | * just make sure it is always different. */ | |
481 | array.utime = st->utime + 1;; | |
482 | ||
483 | if (st->utime == array.utime && | |
484 | st->failed == array.failed_disks && | |
485 | st->working == array.working_disks && | |
486 | st->spare == array.spare_disks && | |
487 | (mse == NULL || ( | |
488 | mse->percent == st->percent | |
489 | ))) { | |
490 | close(fd); | |
491 | st->err = 0; | |
492 | return; | |
493 | } | |
494 | if (st->utime == 0 && /* new array */ | |
495 | mse->pattern && strchr(mse->pattern, '_') /* degraded */ | |
496 | ) | |
497 | alert("DegradedArray", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
498 | ||
499 | if (st->utime == 0 && /* new array */ | |
500 | st->expected_spares > 0 && | |
501 | array.spare_disks < st->expected_spares) | |
502 | alert("SparesMissing", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
503 | if (st->percent == -1 && | |
504 | mse->percent >= 0) | |
505 | alert("RebuildStarted", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
506 | if (st->percent >= 0 && | |
507 | mse->percent >= 0 && | |
508 | (mse->percent / increments) > (st->percent / increments)) { | |
509 | char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars) | |
510 | ||
511 | if((mse->percent / increments) == 0) | |
512 | snprintf(percentalert, sizeof(percentalert), "RebuildStarted"); | |
513 | else | |
514 | snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent); | |
515 | ||
516 | alert(percentalert, | |
517 | dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
518 | } | |
519 | ||
520 | if (mse->percent == -1 && | |
521 | st->percent >= 0) { | |
522 | /* Rebuild/sync/whatever just finished. | |
523 | * If there is a number in /mismatch_cnt, | |
524 | * we should report that. | |
525 | */ | |
526 | struct mdinfo *sra = | |
527 | sysfs_read(-1, st->devnum, GET_MISMATCH); | |
528 | if (sra && sra->mismatch_cnt > 0) { | |
529 | char cnt[40]; | |
530 | sprintf(cnt, " mismatches found: %d", sra->mismatch_cnt); | |
531 | alert("RebuildFinished", dev, cnt, mailaddr, mailfrom, alert_cmd, dosyslog); | |
532 | } else | |
533 | alert("RebuildFinished", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
534 | if (sra) | |
535 | free(sra); | |
536 | } | |
537 | st->percent = mse->percent; | |
538 | ||
539 | for (i=0; i<MaxDisks && i <= array.raid_disks + array.nr_disks; | |
540 | i++) { | |
541 | mdu_disk_info_t disc; | |
542 | disc.number = i; | |
543 | if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) { | |
544 | info[i].state = disc.state; | |
545 | info[i].major = disc.major; | |
546 | info[i].minor = disc.minor; | |
547 | } else | |
548 | info[i].major = info[i].minor = 0; | |
549 | } | |
550 | ||
551 | if (strncmp(mse->metadata_version, "external:", 9) == 0 && | |
552 | is_subarray(mse->metadata_version+9)) | |
553 | st->parent_dev = | |
554 | devname2devnum(mse->metadata_version+10); | |
555 | else | |
556 | st->parent_dev = NoMdDev; | |
557 | if (st->metadata == NULL && | |
558 | st->parent_dev == NoMdDev) | |
559 | st->metadata = super_by_fd(fd, NULL); | |
560 | ||
561 | close(fd); | |
562 | ||
563 | for (i=0; i<MaxDisks; i++) { | |
564 | mdu_disk_info_t disc = {0,0,0,0,0}; | |
565 | int newstate=0; | |
566 | int change; | |
567 | char *dv = NULL; | |
568 | disc.number = i; | |
569 | if (i > array.raid_disks + array.nr_disks) { | |
570 | newstate = 0; | |
571 | disc.major = disc.minor = 0; | |
572 | } else if (info[i].major || info[i].minor) { | |
573 | newstate = info[i].state; | |
574 | dv = map_dev(info[i].major, info[i].minor, 1); | |
575 | disc.state = newstate; | |
576 | disc.major = info[i].major; | |
577 | disc.minor = info[i].minor; | |
578 | } else if (mse && mse->pattern && i < (int)strlen(mse->pattern)) { | |
579 | switch(mse->pattern[i]) { | |
580 | case 'U': newstate = 6 /* ACTIVE/SYNC */; break; | |
581 | case '_': newstate = 0; break; | |
582 | } | |
583 | disc.major = disc.minor = 0; | |
584 | } | |
585 | if (dv == NULL && st->devid[i]) | |
586 | dv = map_dev(major(st->devid[i]), | |
587 | minor(st->devid[i]), 1); | |
588 | change = newstate ^ st->devstate[i]; | |
589 | if (st->utime && change && !st->err) { | |
590 | if (i < array.raid_disks && | |
591 | (((newstate&change)&(1<<MD_DISK_FAULTY)) || | |
592 | ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) || | |
593 | ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))) | |
594 | ) | |
595 | alert("Fail", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog); | |
596 | else if (i >= array.raid_disks && | |
597 | (disc.major || disc.minor) && | |
598 | st->devid[i] == makedev(disc.major, disc.minor) && | |
599 | ((newstate&change)&(1<<MD_DISK_FAULTY)) | |
600 | ) | |
601 | alert("FailSpare", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog); | |
602 | else if (i < array.raid_disks && | |
603 | ! (newstate & (1<<MD_DISK_REMOVED)) && | |
604 | (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) || | |
605 | ((newstate&change)&(1<<MD_DISK_ACTIVE)) || | |
606 | ((newstate&change)&(1<<MD_DISK_SYNC))) | |
607 | ) | |
608 | alert("SpareActive", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog); | |
609 | } | |
610 | st->devstate[i] = newstate; | |
611 | st->devid[i] = makedev(disc.major, disc.minor); | |
612 | } | |
613 | st->active = array.active_disks; | |
614 | st->working = array.working_disks; | |
615 | st->spare = array.spare_disks; | |
616 | st->failed = array.failed_disks; | |
617 | st->utime = array.utime; | |
618 | st->raid = array.raid_disks; | |
619 | st->err = 0; | |
620 | } | |
621 | ||
622 | static int add_new_arrays(struct mdstat_ent *mdstat, struct state *statelist, | |
623 | int test, char *mailaddr, char *mailfrom, | |
624 | char *alert_cmd, int dosyslog) | |
625 | { | |
626 | struct mdstat_ent *mse; | |
627 | int new_found = 0; | |
628 | ||
629 | for (mse=mdstat; mse; mse=mse->next) | |
630 | if (mse->devnum != INT_MAX && | |
631 | (!mse->level || /* retrieve containers */ | |
632 | (strcmp(mse->level, "raid0") != 0 && | |
633 | strcmp(mse->level, "linear") != 0)) | |
634 | ) { | |
635 | struct state *st = malloc(sizeof *st); | |
636 | mdu_array_info_t array; | |
637 | int fd; | |
638 | if (st == NULL) | |
639 | continue; | |
640 | st->devname = strdup(get_md_name(mse->devnum)); | |
641 | if ((fd = open(st->devname, O_RDONLY)) < 0 || | |
642 | ioctl(fd, GET_ARRAY_INFO, &array)< 0) { | |
643 | /* no such array */ | |
644 | if (fd >=0) close(fd); | |
645 | put_md_name(st->devname); | |
646 | free(st->devname); | |
647 | if (st->metadata) { | |
648 | st->metadata->ss->free_super(st->metadata); | |
649 | free(st->metadata); | |
650 | } | |
651 | free(st); | |
652 | continue; | |
653 | } | |
654 | close(fd); | |
655 | st->utime = 0; | |
656 | st->next = statelist; | |
657 | st->err = 1; | |
658 | st->devnum = mse->devnum; | |
659 | st->percent = -2; | |
660 | st->spare_group = NULL; | |
661 | st->expected_spares = -1; | |
662 | if (strncmp(mse->metadata_version, "external:", 9) == 0 && | |
663 | is_subarray(mse->metadata_version+9)) | |
664 | st->parent_dev = | |
665 | devname2devnum(mse->metadata_version+10); | |
666 | else | |
667 | st->parent_dev = NoMdDev; | |
668 | st->metadata = NULL; | |
669 | statelist = st; | |
670 | if (test) | |
671 | alert("TestMessage", st->devname, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
672 | alert("NewArray", st->devname, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
673 | new_found = 1; | |
674 | } | |
675 | return new_found; | |
676 | } | |
677 | ||
678 | static void try_spare_migration(struct state *statelist, | |
679 | char *mailaddr, char *mailfrom, | |
680 | char *alert_cmd, int dosyslog) | |
681 | { | |
682 | struct state *st; | |
683 | for (st = statelist; st; st=st->next) | |
684 | if (st->active < st->raid && | |
685 | st->spare == 0 && | |
686 | st->spare_group != NULL) { | |
687 | struct state *st2; | |
688 | for (st2=statelist ; st2 ; st2=st2->next) | |
689 | if (st2 != st && | |
690 | st2->spare > 0 && | |
691 | st2->active == st2->raid && | |
692 | st2->spare_group != NULL && | |
693 | strcmp(st->spare_group, st2->spare_group) == 0) { | |
694 | /* try to remove and add */ | |
695 | int fd1 = open(st->devname, O_RDONLY); | |
696 | int fd2 = open(st2->devname, O_RDONLY); | |
697 | int dev = -1; | |
698 | int d; | |
699 | if (fd1 < 0 || fd2 < 0) { | |
700 | if (fd1>=0) close(fd1); | |
701 | if (fd2>=0) close(fd2); | |
702 | continue; | |
703 | } | |
704 | for (d=st2->raid; d < MaxDisks; d++) { | |
705 | if (st2->devid[d] > 0 && | |
706 | st2->devstate[d] == 0) { | |
707 | dev = st2->devid[d]; | |
708 | break; | |
709 | } | |
710 | } | |
711 | if (dev > 0) { | |
712 | struct mddev_dev devlist; | |
713 | char devname[20]; | |
714 | devlist.next = NULL; | |
715 | devlist.used = 0; | |
716 | devlist.re_add = 0; | |
717 | devlist.writemostly = 0; | |
718 | devlist.devname = devname; | |
719 | sprintf(devname, "%d:%d", major(dev), minor(dev)); | |
720 | ||
721 | devlist.disposition = 'r'; | |
722 | if (Manage_subdevs(st2->devname, fd2, &devlist, -1, 0) == 0) { | |
723 | devlist.disposition = 'a'; | |
724 | if (Manage_subdevs(st->devname, fd1, &devlist, -1, 0) == 0) { | |
725 | alert("MoveSpare", st->devname, st2->devname, mailaddr, mailfrom, alert_cmd, dosyslog); | |
726 | close(fd1); | |
727 | close(fd2); | |
728 | break; | |
729 | } | |
730 | else Manage_subdevs(st2->devname, fd2, &devlist, -1, 0); | |
731 | } | |
732 | } | |
733 | close(fd1); | |
734 | close(fd2); | |
735 | } | |
736 | } | |
737 | } | |
b90c0e9a NB |
738 | /* Not really Monitor but ... */ |
739 | int Wait(char *dev) | |
740 | { | |
741 | struct stat stb; | |
742 | int devnum; | |
743 | int rv = 1; | |
744 | ||
745 | if (stat(dev, &stb) != 0) { | |
746 | fprintf(stderr, Name ": Cannot find %s: %s\n", dev, | |
747 | strerror(errno)); | |
748 | return 2; | |
749 | } | |
c94709e8 | 750 | devnum = stat2devnum(&stb); |
b90c0e9a NB |
751 | |
752 | while(1) { | |
753 | struct mdstat_ent *ms = mdstat_read(1, 0); | |
754 | struct mdstat_ent *e; | |
755 | ||
756 | for (e=ms ; e; e=e->next) | |
757 | if (e->devnum == devnum) | |
758 | break; | |
759 | ||
760 | if (!e || e->percent < 0) { | |
e7783ee6 | 761 | if (e && e->metadata_version && |
c94709e8 DW |
762 | strncmp(e->metadata_version, "external:", 9) == 0) { |
763 | if (is_subarray(&e->metadata_version[9])) | |
764 | ping_monitor(&e->metadata_version[9]); | |
765 | else | |
766 | ping_monitor(devnum2devname(devnum)); | |
767 | } | |
b90c0e9a NB |
768 | free_mdstat(ms); |
769 | return rv; | |
770 | } | |
89a10d84 | 771 | free_mdstat(ms); |
b90c0e9a NB |
772 | rv = 0; |
773 | mdstat_wait(5); | |
774 | } | |
775 | } |