]>
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 | |
4948b8f7 NB |
33 | static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mailfrom, |
34 | char *cmd, int dosyslog); | |
52826846 | 35 | |
aba69144 | 36 | static char *percentalerts[] = { |
e0d19036 NB |
37 | "RebuildStarted", |
38 | "Rebuild20", | |
39 | "Rebuild40", | |
40 | "Rebuild60", | |
41 | "Rebuild80", | |
42 | }; | |
43 | ||
e27d562b NB |
44 | /* The largest number of disks current arrays can manage is 384 |
45 | * This really should be dynamically, but that will have to wait | |
46 | * At least it isn't MD_SB_DISKS. | |
47 | */ | |
48 | #define MaxDisks 384 | |
cd29a5c8 | 49 | int Monitor(mddev_dev_t devlist, |
52826846 | 50 | char *mailaddr, char *alert_cmd, |
aa88f531 | 51 | int period, int daemonise, int scan, int oneshot, |
8aec876d | 52 | int dosyslog, int test, char* pidfile) |
52826846 NB |
53 | { |
54 | /* | |
55 | * Every few seconds, scan every md device looking for changes | |
56 | * When a change is found, log it, possibly run the alert command, | |
57 | * and possibly send Email | |
58 | * | |
59 | * For each array, we record: | |
60 | * Update time | |
61 | * active/working/failed/spare drives | |
62 | * State of each device. | |
e0d19036 | 63 | * %rebuilt if rebuilding |
52826846 NB |
64 | * |
65 | * If the update time changes, check out all the data again | |
66 | * It is possible that we cannot get the state of each device | |
67 | * due to bugs in the md kernel module. | |
e0d19036 NB |
68 | * We also read /proc/mdstat to get rebuild percent, |
69 | * and to get state on all active devices incase of kernel bug. | |
52826846 | 70 | * |
e0d19036 NB |
71 | * Events are: |
72 | * Fail | |
73 | * An active device had Faulty set or Active/Sync removed | |
74 | * FailSpare | |
75 | * A spare device had Faulty set | |
76 | * SpareActive | |
77 | * An active device had a reverse transition | |
78 | * RebuildStarted | |
79 | * percent went from -1 to +ve | |
80 | * Rebuild20 Rebuild40 Rebuild60 Rebuild80 | |
81 | * percent went from below to not-below that number | |
82 | * DeviceDisappeared | |
83 | * Couldn't access a device which was previously visible | |
52826846 NB |
84 | * |
85 | * if we detect an array with active<raid and spare==0 | |
86 | * we look at other arrays that have same spare-group | |
87 | * If we find one with active==raid and spare>0, | |
88 | * and if we can get_disk_info and find a name | |
89 | * Then we hot-remove and hot-add to the other array | |
90 | * | |
e0d19036 NB |
91 | * If devlist is NULL, then we can monitor everything because --scan |
92 | * was given. We get an initial list from config file and add anything | |
93 | * that appears in /proc/mdstat | |
52826846 NB |
94 | */ |
95 | ||
96 | struct state { | |
97 | char *devname; | |
e0d19036 | 98 | int devnum; /* to sync with mdstat info */ |
52826846 NB |
99 | long utime; |
100 | int err; | |
e0d19036 NB |
101 | char *spare_group; |
102 | int active, working, failed, spare, raid; | |
feb716e9 | 103 | int expected_spares; |
e27d562b NB |
104 | int devstate[MaxDisks]; |
105 | int devid[MaxDisks]; | |
e0d19036 | 106 | int percent; |
52826846 NB |
107 | struct state *next; |
108 | } *statelist = NULL; | |
109 | int finished = 0; | |
e0d19036 | 110 | struct mdstat_ent *mdstat = NULL; |
4948b8f7 | 111 | char *mailfrom = NULL; |
e0d19036 | 112 | |
d013a55e | 113 | if (!mailaddr) { |
8aec876d | 114 | mailaddr = conf_get_mailaddr(); |
d013a55e | 115 | if (mailaddr && ! scan) |
56eedc1a | 116 | fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n", |
d013a55e NB |
117 | mailaddr); |
118 | } | |
8aec876d | 119 | mailfrom = conf_get_mailfrom(); |
4948b8f7 | 120 | |
d013a55e | 121 | if (!alert_cmd) { |
8aec876d | 122 | alert_cmd = conf_get_program(); |
d013a55e | 123 | if (alert_cmd && ! scan) |
56eedc1a | 124 | fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n", |
d013a55e NB |
125 | alert_cmd); |
126 | } | |
56eedc1a NB |
127 | if (scan && !mailaddr && !alert_cmd) { |
128 | fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n"); | |
d013a55e | 129 | return 1; |
56eedc1a | 130 | } |
d013a55e NB |
131 | |
132 | if (daemonise) { | |
133 | int pid = fork(); | |
134 | if (pid > 0) { | |
b5e64645 NB |
135 | if (!pidfile) |
136 | printf("%d\n", pid); | |
137 | else { | |
138 | FILE *pid_file; | |
139 | pid_file=fopen(pidfile, "w"); | |
140 | if (!pid_file) | |
141 | perror("cannot create pid file"); | |
142 | else { | |
143 | fprintf(pid_file,"%d\n", pid); | |
144 | fclose(pid_file); | |
145 | } | |
146 | } | |
d013a55e NB |
147 | return 0; |
148 | } | |
149 | if (pid < 0) { | |
150 | perror("daemonise"); | |
151 | return 1; | |
152 | } | |
153 | close(0); | |
fb97b4d6 | 154 | open("/dev/null", O_RDWR); |
d013a55e NB |
155 | dup2(0,1); |
156 | dup2(0,2); | |
157 | setsid(); | |
158 | } | |
e0d19036 NB |
159 | |
160 | if (devlist == NULL) { | |
8aec876d | 161 | mddev_ident_t mdlist = conf_get_ident(NULL); |
e0d19036 | 162 | for (; mdlist; mdlist=mdlist->next) { |
fe056d1f N |
163 | struct state *st; |
164 | if (mdlist->devname == NULL) | |
165 | continue; | |
112cace6 N |
166 | if (strcasecmp(mdlist->devname, "<ignore>") == 0) |
167 | continue; | |
fe056d1f | 168 | st = malloc(sizeof *st); |
e0d19036 NB |
169 | if (st == NULL) |
170 | continue; | |
db2d001c N |
171 | if (mdlist->devname[0] == '/') |
172 | st->devname = strdup(mdlist->devname); | |
173 | else { | |
174 | st->devname = malloc(8+strlen(mdlist->devname)+1); | |
175 | strcpy(strcpy(st->devname, "/dev/md/"), | |
176 | mdlist->devname); | |
177 | } | |
e0d19036 NB |
178 | st->utime = 0; |
179 | st->next = statelist; | |
d013a55e | 180 | st->err = 0; |
4450e59f | 181 | st->devnum = INT_MAX; |
e0d19036 | 182 | st->percent = -2; |
feb716e9 | 183 | st->expected_spares = mdlist->spare_disks; |
e0d19036 NB |
184 | if (mdlist->spare_group) |
185 | st->spare_group = strdup(mdlist->spare_group); | |
186 | else | |
187 | st->spare_group = NULL; | |
188 | statelist = st; | |
189 | } | |
190 | } else { | |
cd29a5c8 | 191 | mddev_dev_t dv; |
e0d19036 | 192 | for (dv=devlist ; dv; dv=dv->next) { |
8aec876d | 193 | mddev_ident_t mdlist = conf_get_ident(dv->devname); |
e0d19036 NB |
194 | struct state *st = malloc(sizeof *st); |
195 | if (st == NULL) | |
196 | continue; | |
197 | st->devname = strdup(dv->devname); | |
198 | st->utime = 0; | |
199 | st->next = statelist; | |
d013a55e | 200 | st->err = 0; |
4450e59f | 201 | st->devnum = INT_MAX; |
e0d19036 | 202 | st->percent = -2; |
feb716e9 | 203 | st->expected_spares = -1; |
e0d19036 | 204 | st->spare_group = NULL; |
e5329c37 NB |
205 | if (mdlist) { |
206 | st->expected_spares = mdlist->spare_disks; | |
207 | if (mdlist->spare_group) | |
208 | st->spare_group = strdup(mdlist->spare_group); | |
209 | } | |
e0d19036 NB |
210 | statelist = st; |
211 | } | |
212 | } | |
213 | ||
214 | ||
215 | while (! finished) { | |
aa88f531 | 216 | int new_found = 0; |
e0d19036 NB |
217 | struct state *st; |
218 | ||
219 | if (mdstat) | |
220 | free_mdstat(mdstat); | |
22a88995 | 221 | mdstat = mdstat_read(oneshot?0:1, 0); |
e0d19036 NB |
222 | |
223 | for (st=statelist; st; st=st->next) { | |
ab5303d6 | 224 | struct { int state, major, minor; } info[MaxDisks]; |
52826846 | 225 | mdu_array_info_t array; |
dd0781e5 | 226 | struct mdstat_ent *mse = NULL, *mse2; |
e0d19036 | 227 | char *dev = st->devname; |
52826846 | 228 | int fd; |
98c6faba | 229 | unsigned int i; |
e0d19036 | 230 | |
98c6faba | 231 | if (test) |
4948b8f7 | 232 | alert("TestMessage", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
52826846 NB |
233 | fd = open(dev, O_RDONLY); |
234 | if (fd < 0) { | |
235 | if (!st->err) | |
e0d19036 | 236 | alert("DeviceDisappeared", dev, NULL, |
4948b8f7 | 237 | mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 | 238 | /* fprintf(stderr, Name ": cannot open %s: %s\n", |
52826846 | 239 | dev, strerror(errno)); |
e0d19036 | 240 | */ st->err=1; |
52826846 NB |
241 | continue; |
242 | } | |
e4dc5106 | 243 | fcntl(fd, F_SETFD, FD_CLOEXEC); |
52826846 NB |
244 | if (ioctl(fd, GET_ARRAY_INFO, &array)<0) { |
245 | if (!st->err) | |
e0d19036 | 246 | alert("DeviceDisappeared", dev, NULL, |
4948b8f7 | 247 | mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 | 248 | /* fprintf(stderr, Name ": cannot get array info for %s: %s\n", |
52826846 | 249 | dev, strerror(errno)); |
e0d19036 | 250 | */ st->err=1; |
52826846 NB |
251 | close(fd); |
252 | continue; | |
253 | } | |
66f8bbbe DL |
254 | /* It's much easier to list what array levels can't |
255 | * have a device disappear than all of them that can | |
256 | */ | |
257 | if (array.level == 0 || array.level == -1) { | |
5787fa49 NB |
258 | if (!st->err) |
259 | alert("DeviceDisappeared", dev, "Wrong-Level", | |
4948b8f7 | 260 | mailaddr, mailfrom, alert_cmd, dosyslog); |
5787fa49 NB |
261 | st->err = 1; |
262 | close(fd); | |
263 | continue; | |
264 | } | |
4450e59f | 265 | if (st->devnum == INT_MAX) { |
e0d19036 NB |
266 | struct stat stb; |
267 | if (fstat(fd, &stb) == 0 && | |
98c6faba | 268 | (S_IFMT&stb.st_mode)==S_IFBLK) { |
0df46c2a NB |
269 | if (major(stb.st_rdev) == MD_MAJOR) |
270 | st->devnum = minor(stb.st_rdev); | |
98c6faba | 271 | else |
0df46c2a | 272 | st->devnum = -1- (minor(stb.st_rdev)>>6); |
98c6faba | 273 | } |
e0d19036 NB |
274 | } |
275 | ||
dd0781e5 NB |
276 | for (mse2 = mdstat ; mse2 ; mse2=mse2->next) |
277 | if (mse2->devnum == st->devnum) { | |
4450e59f | 278 | mse2->devnum = INT_MAX; /* flag it as "used" */ |
dd0781e5 NB |
279 | mse = mse2; |
280 | } | |
e0d19036 | 281 | |
4736b5dc N |
282 | if (array.utime == 0) |
283 | /* external arrays don't update utime */ | |
284 | array.utime = time(0); | |
285 | ||
52826846 | 286 | if (st->utime == array.utime && |
e0d19036 NB |
287 | st->failed == array.failed_disks && |
288 | st->working == array.working_disks && | |
289 | st->spare == array.spare_disks && | |
290 | (mse == NULL || ( | |
291 | mse->percent == st->percent | |
292 | ))) { | |
52826846 | 293 | close(fd); |
e0d19036 | 294 | st->err = 0; |
52826846 NB |
295 | continue; |
296 | } | |
aa88f531 NB |
297 | if (st->utime == 0 && /* new array */ |
298 | mse && /* is in /proc/mdstat */ | |
299 | mse->pattern && strchr(mse->pattern, '_') /* degraded */ | |
300 | ) | |
4948b8f7 | 301 | alert("DegradedArray", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
aa88f531 | 302 | |
feb716e9 | 303 | if (st->utime == 0 && /* new array */ |
aba69144 NB |
304 | st->expected_spares > 0 && |
305 | array.spare_disks < st->expected_spares) | |
4948b8f7 | 306 | alert("SparesMissing", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 | 307 | if (mse && |
aba69144 | 308 | st->percent == -1 && |
e0d19036 | 309 | mse->percent >= 0) |
4948b8f7 | 310 | alert("RebuildStarted", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 NB |
311 | if (mse && |
312 | st->percent >= 0 && | |
313 | mse->percent >= 0 && | |
314 | (mse->percent / 20) > (st->percent / 20)) | |
315 | alert(percentalerts[mse->percent/20], | |
4948b8f7 | 316 | dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 | 317 | |
98c6faba NB |
318 | if (mse && |
319 | mse->percent == -1 && | |
37dfc3d6 NB |
320 | st->percent >= 0) { |
321 | /* Rebuild/sync/whatever just finished. | |
322 | * If there is a number in /mismatch_cnt, | |
323 | * we should report that. | |
324 | */ | |
7e0f6979 | 325 | struct mdinfo *sra = |
37dfc3d6 NB |
326 | sysfs_read(-1, st->devnum, GET_MISMATCH); |
327 | if (sra && sra->mismatch_cnt > 0) { | |
328 | char cnt[40]; | |
329 | sprintf(cnt, " mismatches found: %d", sra->mismatch_cnt); | |
330 | alert("RebuildFinished", dev, cnt, mailaddr, mailfrom, alert_cmd, dosyslog); | |
331 | } else | |
332 | alert("RebuildFinished", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
333 | if (sra) | |
334 | free(sra); | |
335 | } | |
98c6faba | 336 | |
e0d19036 NB |
337 | if (mse) |
338 | st->percent = mse->percent; | |
e27d562b | 339 | |
ab5303d6 NB |
340 | |
341 | for (i=0; i<MaxDisks && i <= array.raid_disks + array.nr_disks; | |
342 | i++) { | |
343 | mdu_disk_info_t disc; | |
01d9299c | 344 | disc.number = i; |
ab5303d6 NB |
345 | if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) { |
346 | info[i].state = disc.state; | |
347 | info[i].major = disc.major; | |
348 | info[i].minor = disc.minor; | |
349 | } else | |
350 | info[i].major = info[i].minor = 0; | |
351 | } | |
352 | close(fd); | |
353 | ||
e27d562b | 354 | for (i=0; i<MaxDisks; i++) { |
f8409e54 | 355 | mdu_disk_info_t disc = {0}; |
e0d19036 NB |
356 | int newstate=0; |
357 | int change; | |
358 | char *dv = NULL; | |
52826846 | 359 | disc.number = i; |
e27d562b NB |
360 | if (i > array.raid_disks + array.nr_disks) { |
361 | newstate = 0; | |
362 | disc.major = disc.minor = 0; | |
ab5303d6 NB |
363 | } else if (info[i].major || info[i].minor) { |
364 | newstate = info[i].state; | |
365 | dv = map_dev(info[i].major, info[i].minor, 1); | |
366 | disc.state = newstate; | |
367 | disc.major = info[i].major; | |
368 | disc.minor = info[i].minor; | |
e27d562b | 369 | } else if (mse && mse->pattern && i < strlen(mse->pattern)) { |
e0d19036 NB |
370 | switch(mse->pattern[i]) { |
371 | case 'U': newstate = 6 /* ACTIVE/SYNC */; break; | |
372 | case '_': newstate = 0; break; | |
52826846 | 373 | } |
e27d562b NB |
374 | disc.major = disc.minor = 0; |
375 | } | |
fe394e5e NB |
376 | if (dv == NULL && st->devid[i]) |
377 | dv = map_dev(major(st->devid[i]), | |
16c6fa80 | 378 | minor(st->devid[i]), 1); |
e0d19036 NB |
379 | change = newstate ^ st->devstate[i]; |
380 | if (st->utime && change && !st->err) { | |
98c6faba | 381 | if (i < (unsigned)array.raid_disks && |
e0d19036 NB |
382 | (((newstate&change)&(1<<MD_DISK_FAULTY)) || |
383 | ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) || | |
384 | ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))) | |
385 | ) | |
4948b8f7 | 386 | alert("Fail", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog); |
98c6faba | 387 | else if (i >= (unsigned)array.raid_disks && |
e0d19036 | 388 | (disc.major || disc.minor) && |
0df46c2a | 389 | st->devid[i] == makedev(disc.major, disc.minor) && |
e0d19036 NB |
390 | ((newstate&change)&(1<<MD_DISK_FAULTY)) |
391 | ) | |
4948b8f7 | 392 | alert("FailSpare", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog); |
98c6faba | 393 | else if (i < (unsigned)array.raid_disks && |
e0d19036 NB |
394 | (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) || |
395 | ((newstate&change)&(1<<MD_DISK_ACTIVE)) || | |
396 | ((newstate&change)&(1<<MD_DISK_SYNC))) | |
397 | ) | |
4948b8f7 | 398 | alert("SpareActive", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog); |
52826846 | 399 | } |
3e6944b2 | 400 | st->devstate[i] = newstate; |
0df46c2a | 401 | st->devid[i] = makedev(disc.major, disc.minor); |
52826846 | 402 | } |
52826846 NB |
403 | st->active = array.active_disks; |
404 | st->working = array.working_disks; | |
405 | st->spare = array.spare_disks; | |
406 | st->failed = array.failed_disks; | |
407 | st->utime = array.utime; | |
e0d19036 NB |
408 | st->raid = array.raid_disks; |
409 | st->err = 0; | |
52826846 | 410 | } |
e0d19036 NB |
411 | /* now check if there are any new devices found in mdstat */ |
412 | if (scan) { | |
413 | struct mdstat_ent *mse; | |
aba69144 | 414 | for (mse=mdstat; mse; mse=mse->next) |
4450e59f | 415 | if (mse->devnum != INT_MAX && |
2cdb6489 | 416 | mse->level && |
66f8bbbe DL |
417 | (strcmp(mse->level, "raid0")!=0 && |
418 | strcmp(mse->level, "linear")!=0) | |
5787fa49 | 419 | ) { |
e0d19036 | 420 | struct state *st = malloc(sizeof *st); |
98c6faba NB |
421 | mdu_array_info_t array; |
422 | int fd; | |
e0d19036 NB |
423 | if (st == NULL) |
424 | continue; | |
425 | st->devname = strdup(get_md_name(mse->devnum)); | |
98c6faba NB |
426 | if ((fd = open(st->devname, O_RDONLY)) < 0 || |
427 | ioctl(fd, GET_ARRAY_INFO, &array)< 0) { | |
428 | /* no such array */ | |
429 | if (fd >=0) close(fd); | |
8bd2e0c3 | 430 | put_md_name(st->devname); |
98c6faba NB |
431 | free(st->devname); |
432 | free(st); | |
433 | continue; | |
434 | } | |
dd0781e5 | 435 | close(fd); |
e0d19036 NB |
436 | st->utime = 0; |
437 | st->next = statelist; | |
438 | st->err = 1; | |
439 | st->devnum = mse->devnum; | |
440 | st->percent = -2; | |
441 | st->spare_group = NULL; | |
98c6faba | 442 | st->expected_spares = -1; |
e0d19036 | 443 | statelist = st; |
bc854448 N |
444 | if (test) |
445 | alert("TestMessage", st->devname, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
4948b8f7 | 446 | alert("NewArray", st->devname, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
aa88f531 | 447 | new_found = 1; |
e0d19036 NB |
448 | } |
449 | } | |
450 | /* If an array has active < raid && spare == 0 && spare_group != NULL | |
451 | * Look for another array with spare > 0 and active == raid and same spare_group | |
452 | * if found, choose a device and hotremove/hotadd | |
453 | */ | |
454 | for (st = statelist; st; st=st->next) | |
455 | if (st->active < st->raid && | |
456 | st->spare == 0 && | |
457 | st->spare_group != NULL) { | |
458 | struct state *st2; | |
459 | for (st2=statelist ; st2 ; st2=st2->next) | |
460 | if (st2 != st && | |
461 | st2->spare > 0 && | |
462 | st2->active == st2->raid && | |
463 | st2->spare_group != NULL && | |
464 | strcmp(st->spare_group, st2->spare_group) == 0) { | |
465 | /* try to remove and add */ | |
466 | int fd1 = open(st->devname, O_RDONLY); | |
467 | int fd2 = open(st2->devname, O_RDONLY); | |
468 | int dev = -1; | |
469 | int d; | |
470 | if (fd1 < 0 || fd2 < 0) { | |
471 | if (fd1>=0) close(fd1); | |
472 | if (fd2>=0) close(fd2); | |
473 | continue; | |
474 | } | |
e27d562b | 475 | for (d=st2->raid; d < MaxDisks; d++) { |
e0d19036 NB |
476 | if (st2->devid[d] > 0 && |
477 | st2->devstate[d] == 0) { | |
478 | dev = st2->devid[d]; | |
479 | break; | |
480 | } | |
481 | } | |
482 | if (dev > 0) { | |
c7b47447 N |
483 | struct mddev_dev_s devlist; |
484 | char devname[20]; | |
485 | devlist.next = NULL; | |
486 | devlist.used = 0; | |
487 | devlist.re_add = 0; | |
488 | devlist.writemostly = 0; | |
489 | devlist.devname = devname; | |
490 | sprintf(devname, "%d:%d", major(dev), minor(dev)); | |
491 | ||
492 | devlist.disposition = 'r'; | |
493 | if (Manage_subdevs(st2->devname, fd2, &devlist, -1) == 0) { | |
494 | devlist.disposition = 'a'; | |
495 | if (Manage_subdevs(st->devname, fd1, &devlist, -1) == 0) { | |
4948b8f7 | 496 | alert("MoveSpare", st->devname, st2->devname, mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 NB |
497 | close(fd1); |
498 | close(fd2); | |
499 | break; | |
500 | } | |
c7b47447 | 501 | else Manage_subdevs(st2->devname, fd2, &devlist, -1); |
e0d19036 NB |
502 | } |
503 | } | |
504 | close(fd1); | |
505 | close(fd2); | |
506 | } | |
507 | } | |
aa88f531 NB |
508 | if (!new_found) { |
509 | if (oneshot) | |
510 | break; | |
511 | else | |
dd0781e5 | 512 | mdstat_wait(period); |
aa88f531 | 513 | } |
98c6faba | 514 | test = 0; |
52826846 | 515 | } |
b5e64645 NB |
516 | if (pidfile) |
517 | unlink(pidfile); | |
52826846 NB |
518 | return 0; |
519 | } | |
520 | ||
521 | ||
4948b8f7 | 522 | static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mailfrom, char *cmd, |
773135f5 | 523 | int dosyslog) |
52826846 | 524 | { |
773135f5 NB |
525 | int priority; |
526 | ||
cd29a5c8 NB |
527 | if (!cmd && !mailaddr) { |
528 | time_t now = time(0); | |
aba69144 | 529 | |
e0d19036 | 530 | printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device"); |
cd29a5c8 | 531 | } |
52826846 NB |
532 | if (cmd) { |
533 | int pid = fork(); | |
534 | switch(pid) { | |
535 | default: | |
536 | waitpid(pid, NULL, 0); | |
537 | break; | |
538 | case -1: | |
539 | break; | |
540 | case 0: | |
541 | execl(cmd, cmd, event, dev, disc, NULL); | |
542 | exit(2); | |
543 | } | |
544 | } | |
aba69144 NB |
545 | if (mailaddr && |
546 | (strncmp(event, "Fail", 4)==0 || | |
98c6faba | 547 | strncmp(event, "Test", 4)==0 || |
d1732eeb | 548 | strncmp(event, "Spares", 6)==0 || |
aa88f531 | 549 | strncmp(event, "Degrade", 7)==0)) { |
52826846 NB |
550 | FILE *mp = popen(Sendmail, "w"); |
551 | if (mp) { | |
008e1100 | 552 | FILE *mdstat; |
52826846 NB |
553 | char hname[256]; |
554 | gethostname(hname, sizeof(hname)); | |
555 | signal(SIGPIPE, SIG_IGN); | |
4948b8f7 NB |
556 | if (mailfrom) |
557 | fprintf(mp, "From: %s\n", mailfrom); | |
558 | else | |
559 | fprintf(mp, "From: " Name " monitoring <root>\n"); | |
52826846 NB |
560 | fprintf(mp, "To: %s\n", mailaddr); |
561 | fprintf(mp, "Subject: %s event on %s:%s\n\n", event, dev, hname); | |
562 | ||
563 | fprintf(mp, "This is an automatically generated mail message from " Name "\n"); | |
564 | fprintf(mp, "running on %s\n\n", hname); | |
565 | ||
566 | fprintf(mp, "A %s event had been detected on md device %s.\n\n", event, dev); | |
567 | ||
37dfc3d6 | 568 | if (disc && disc[0] != ' ') |
56eedc1a | 569 | fprintf(mp, "It could be related to component device %s.\n\n", disc); |
37dfc3d6 NB |
570 | if (disc && disc[0] == ' ') |
571 | fprintf(mp, "Extra information:%s.\n\n", disc); | |
52826846 NB |
572 | |
573 | fprintf(mp, "Faithfully yours, etc.\n"); | |
008e1100 NB |
574 | |
575 | mdstat = fopen("/proc/mdstat", "r"); | |
576 | if (mdstat) { | |
577 | char buf[8192]; | |
578 | int n; | |
a524a7ee | 579 | fprintf(mp, "\nP.S. The /proc/mdstat file currently contains the following:\n\n"); |
008e1100 | 580 | while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0) |
9fca7d62 | 581 | n=fwrite(buf, 1, n, mp); /* yes, i don't care about the result */ |
008e1100 NB |
582 | fclose(mdstat); |
583 | } | |
6278fb3a | 584 | pclose(mp); |
52826846 NB |
585 | } |
586 | ||
587 | } | |
773135f5 NB |
588 | |
589 | /* log the event to syslog maybe */ | |
590 | if (dosyslog) { | |
591 | /* Log at a different severity depending on the event. | |
592 | * | |
593 | * These are the critical events: */ | |
594 | if (strncmp(event, "Fail", 4)==0 || | |
595 | strncmp(event, "Degrade", 7)==0 || | |
596 | strncmp(event, "DeviceDisappeared", 17)==0) | |
597 | priority = LOG_CRIT; | |
598 | /* Good to know about, but are not failures: */ | |
599 | else if (strncmp(event, "Rebuild", 7)==0 || | |
d1732eeb NB |
600 | strncmp(event, "MoveSpare", 9)==0 || |
601 | strncmp(event, "Spares", 6) != 0) | |
773135f5 NB |
602 | priority = LOG_WARNING; |
603 | /* Everything else: */ | |
604 | else | |
605 | priority = LOG_INFO; | |
606 | ||
607 | if (disc) | |
608 | syslog(priority, "%s event detected on md device %s, component device %s", event, dev, disc); | |
609 | else | |
610 | syslog(priority, "%s event detected on md device %s", event, dev); | |
611 | } | |
52826846 | 612 | } |
b90c0e9a NB |
613 | |
614 | /* Not really Monitor but ... */ | |
615 | int Wait(char *dev) | |
616 | { | |
617 | struct stat stb; | |
618 | int devnum; | |
619 | int rv = 1; | |
620 | ||
621 | if (stat(dev, &stb) != 0) { | |
622 | fprintf(stderr, Name ": Cannot find %s: %s\n", dev, | |
623 | strerror(errno)); | |
624 | return 2; | |
625 | } | |
c94709e8 | 626 | devnum = stat2devnum(&stb); |
b90c0e9a NB |
627 | |
628 | while(1) { | |
629 | struct mdstat_ent *ms = mdstat_read(1, 0); | |
630 | struct mdstat_ent *e; | |
631 | ||
632 | for (e=ms ; e; e=e->next) | |
633 | if (e->devnum == devnum) | |
634 | break; | |
635 | ||
636 | if (!e || e->percent < 0) { | |
e7783ee6 | 637 | if (e && e->metadata_version && |
c94709e8 DW |
638 | strncmp(e->metadata_version, "external:", 9) == 0) { |
639 | if (is_subarray(&e->metadata_version[9])) | |
640 | ping_monitor(&e->metadata_version[9]); | |
641 | else | |
642 | ping_monitor(devnum2devname(devnum)); | |
643 | } | |
b90c0e9a NB |
644 | free_mdstat(ms); |
645 | return rv; | |
646 | } | |
89a10d84 | 647 | free_mdstat(ms); |
b90c0e9a NB |
648 | rv = 0; |
649 | mdstat_wait(5); | |
650 | } | |
651 | } |