]>
Commit | Line | Data |
---|---|---|
52826846 | 1 | /* |
9a9dab36 | 2 | * mdadm - manage Linux "md" devices aka RAID arrays. |
52826846 | 3 | * |
cd29a5c8 | 4 | * Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au> |
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 | |
22 | * Email: <neilb@cse.unsw.edu.au> | |
23 | * Paper: Neil Brown | |
24 | * School of Computer Science and Engineering | |
25 | * The University of New South Wales | |
26 | * Sydney, 2052 | |
27 | * Australia | |
28 | */ | |
29 | ||
9a9dab36 | 30 | #include "mdadm.h" |
52826846 NB |
31 | #include "md_p.h" |
32 | #include "md_u.h" | |
e0d19036 | 33 | #include <sys/wait.h> |
52826846 NB |
34 | #include <sys/signal.h> |
35 | ||
36 | static void alert(char *event, char *dev, char *disc, char *mailaddr, char *cmd); | |
37 | ||
e0d19036 NB |
38 | static char *percentalerts[] = { |
39 | "RebuildStarted", | |
40 | "Rebuild20", | |
41 | "Rebuild40", | |
42 | "Rebuild60", | |
43 | "Rebuild80", | |
44 | }; | |
45 | ||
cd29a5c8 | 46 | int Monitor(mddev_dev_t devlist, |
52826846 | 47 | char *mailaddr, char *alert_cmd, |
aa88f531 | 48 | int period, int daemonise, int scan, int oneshot, |
52826846 NB |
49 | char *config) |
50 | { | |
51 | /* | |
52 | * Every few seconds, scan every md device looking for changes | |
53 | * When a change is found, log it, possibly run the alert command, | |
54 | * and possibly send Email | |
55 | * | |
56 | * For each array, we record: | |
57 | * Update time | |
58 | * active/working/failed/spare drives | |
59 | * State of each device. | |
e0d19036 | 60 | * %rebuilt if rebuilding |
52826846 NB |
61 | * |
62 | * If the update time changes, check out all the data again | |
63 | * It is possible that we cannot get the state of each device | |
64 | * due to bugs in the md kernel module. | |
e0d19036 NB |
65 | * We also read /proc/mdstat to get rebuild percent, |
66 | * and to get state on all active devices incase of kernel bug. | |
52826846 | 67 | * |
e0d19036 NB |
68 | * Events are: |
69 | * Fail | |
70 | * An active device had Faulty set or Active/Sync removed | |
71 | * FailSpare | |
72 | * A spare device had Faulty set | |
73 | * SpareActive | |
74 | * An active device had a reverse transition | |
75 | * RebuildStarted | |
76 | * percent went from -1 to +ve | |
77 | * Rebuild20 Rebuild40 Rebuild60 Rebuild80 | |
78 | * percent went from below to not-below that number | |
79 | * DeviceDisappeared | |
80 | * Couldn't access a device which was previously visible | |
52826846 NB |
81 | * |
82 | * if we detect an array with active<raid and spare==0 | |
83 | * we look at other arrays that have same spare-group | |
84 | * If we find one with active==raid and spare>0, | |
85 | * and if we can get_disk_info and find a name | |
86 | * Then we hot-remove and hot-add to the other array | |
87 | * | |
e0d19036 NB |
88 | * If devlist is NULL, then we can monitor everything because --scan |
89 | * was given. We get an initial list from config file and add anything | |
90 | * that appears in /proc/mdstat | |
52826846 NB |
91 | */ |
92 | ||
93 | struct state { | |
94 | char *devname; | |
e0d19036 | 95 | int devnum; /* to sync with mdstat info */ |
52826846 NB |
96 | long utime; |
97 | int err; | |
e0d19036 NB |
98 | char *spare_group; |
99 | int active, working, failed, spare, raid; | |
feb716e9 | 100 | int expected_spares; |
52826846 | 101 | int devstate[MD_SB_DISKS]; |
e0d19036 NB |
102 | int devid[MD_SB_DISKS]; |
103 | int percent; | |
52826846 NB |
104 | struct state *next; |
105 | } *statelist = NULL; | |
106 | int finished = 0; | |
e0d19036 NB |
107 | struct mdstat_ent *mdstat = NULL; |
108 | ||
d013a55e | 109 | if (!mailaddr) { |
e0d19036 | 110 | mailaddr = conf_get_mailaddr(config); |
d013a55e | 111 | if (mailaddr && ! scan) |
56eedc1a | 112 | fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n", |
d013a55e NB |
113 | mailaddr); |
114 | } | |
115 | if (!alert_cmd) { | |
e0d19036 | 116 | alert_cmd = conf_get_program(config); |
d013a55e | 117 | if (alert_cmd && ! scan) |
56eedc1a | 118 | fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n", |
d013a55e NB |
119 | alert_cmd); |
120 | } | |
56eedc1a NB |
121 | if (scan && !mailaddr && !alert_cmd) { |
122 | fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n"); | |
d013a55e | 123 | return 1; |
56eedc1a | 124 | } |
d013a55e NB |
125 | |
126 | if (daemonise) { | |
127 | int pid = fork(); | |
128 | if (pid > 0) { | |
129 | printf("%d\n", pid); | |
130 | return 0; | |
131 | } | |
132 | if (pid < 0) { | |
133 | perror("daemonise"); | |
134 | return 1; | |
135 | } | |
136 | close(0); | |
137 | open("/dev/null", 3); | |
138 | dup2(0,1); | |
139 | dup2(0,2); | |
140 | setsid(); | |
141 | } | |
e0d19036 NB |
142 | |
143 | if (devlist == NULL) { | |
144 | mddev_ident_t mdlist = conf_get_ident(config, NULL); | |
145 | for (; mdlist; mdlist=mdlist->next) { | |
146 | struct state *st = malloc(sizeof *st); | |
147 | if (st == NULL) | |
148 | continue; | |
149 | st->devname = strdup(mdlist->devname); | |
150 | st->utime = 0; | |
151 | st->next = statelist; | |
d013a55e | 152 | st->err = 0; |
e0d19036 NB |
153 | st->devnum = -1; |
154 | st->percent = -2; | |
feb716e9 | 155 | st->expected_spares = mdlist->spare_disks; |
e0d19036 NB |
156 | if (mdlist->spare_group) |
157 | st->spare_group = strdup(mdlist->spare_group); | |
158 | else | |
159 | st->spare_group = NULL; | |
160 | statelist = st; | |
161 | } | |
162 | } else { | |
cd29a5c8 | 163 | mddev_dev_t dv; |
e0d19036 NB |
164 | for (dv=devlist ; dv; dv=dv->next) { |
165 | struct state *st = malloc(sizeof *st); | |
166 | if (st == NULL) | |
167 | continue; | |
168 | st->devname = strdup(dv->devname); | |
169 | st->utime = 0; | |
170 | st->next = statelist; | |
d013a55e | 171 | st->err = 0; |
e0d19036 NB |
172 | st->devnum = -1; |
173 | st->percent = -2; | |
feb716e9 | 174 | st->expected_spares = -1; |
e0d19036 NB |
175 | st->spare_group = NULL; |
176 | statelist = st; | |
177 | } | |
178 | } | |
179 | ||
180 | ||
181 | while (! finished) { | |
aa88f531 | 182 | int new_found = 0; |
e0d19036 NB |
183 | struct state *st; |
184 | ||
185 | if (mdstat) | |
186 | free_mdstat(mdstat); | |
187 | mdstat = mdstat_read(); | |
188 | ||
189 | for (st=statelist; st; st=st->next) { | |
52826846 | 190 | mdu_array_info_t array; |
e0d19036 NB |
191 | struct mdstat_ent *mse; |
192 | char *dev = st->devname; | |
52826846 | 193 | int fd; |
52826846 | 194 | int i; |
e0d19036 | 195 | |
52826846 NB |
196 | fd = open(dev, O_RDONLY); |
197 | if (fd < 0) { | |
198 | if (!st->err) | |
e0d19036 NB |
199 | alert("DeviceDisappeared", dev, NULL, |
200 | mailaddr, alert_cmd); | |
201 | /* fprintf(stderr, Name ": cannot open %s: %s\n", | |
52826846 | 202 | dev, strerror(errno)); |
e0d19036 | 203 | */ st->err=1; |
52826846 NB |
204 | continue; |
205 | } | |
206 | if (ioctl(fd, GET_ARRAY_INFO, &array)<0) { | |
207 | if (!st->err) | |
e0d19036 NB |
208 | alert("DeviceDisappeared", dev, NULL, |
209 | mailaddr, alert_cmd); | |
210 | /* fprintf(stderr, Name ": cannot get array info for %s: %s\n", | |
52826846 | 211 | dev, strerror(errno)); |
e0d19036 | 212 | */ st->err=1; |
52826846 NB |
213 | close(fd); |
214 | continue; | |
215 | } | |
5787fa49 NB |
216 | if (array.level != 1 && array.level != 5 && array.level != -4) { |
217 | if (!st->err) | |
218 | alert("DeviceDisappeared", dev, "Wrong-Level", | |
219 | mailaddr, alert_cmd); | |
220 | st->err = 1; | |
221 | close(fd); | |
222 | continue; | |
223 | } | |
e0d19036 NB |
224 | if (st->devnum < 0) { |
225 | struct stat stb; | |
226 | if (fstat(fd, &stb) == 0 && | |
227 | (S_IFMT&stb.st_mode)==S_IFBLK) | |
228 | st->devnum = MINOR(stb.st_rdev); | |
229 | } | |
230 | ||
231 | for (mse = mdstat ; mse ; mse=mse->next) | |
232 | if (mse->devnum == st->devnum) { | |
233 | mse->devnum = -1; /* flag it as "used" */ | |
234 | break; | |
235 | } | |
236 | ||
52826846 | 237 | if (st->utime == array.utime && |
e0d19036 NB |
238 | st->failed == array.failed_disks && |
239 | st->working == array.working_disks && | |
240 | st->spare == array.spare_disks && | |
241 | (mse == NULL || ( | |
242 | mse->percent == st->percent | |
243 | ))) { | |
52826846 | 244 | close(fd); |
e0d19036 | 245 | st->err = 0; |
52826846 NB |
246 | continue; |
247 | } | |
aa88f531 NB |
248 | if (st->utime == 0 && /* new array */ |
249 | mse && /* is in /proc/mdstat */ | |
250 | mse->pattern && strchr(mse->pattern, '_') /* degraded */ | |
251 | ) | |
252 | alert("DegradedArray", dev, NULL, mailaddr, alert_cmd); | |
253 | ||
feb716e9 NB |
254 | if (st->utime == 0 && /* new array */ |
255 | st->expected_spares > 0 && | |
256 | array.spare_disks < st->expected_spares) | |
257 | alert("SparesMissing", dev, NULL, mailaddr, alert_cmd); | |
e0d19036 NB |
258 | if (mse && |
259 | st->percent == -1 && | |
260 | mse->percent >= 0) | |
261 | alert("RebuildStarted", dev, NULL, mailaddr, alert_cmd); | |
262 | if (mse && | |
263 | st->percent >= 0 && | |
264 | mse->percent >= 0 && | |
265 | (mse->percent / 20) > (st->percent / 20)) | |
266 | alert(percentalerts[mse->percent/20], | |
267 | dev, NULL, mailaddr, alert_cmd); | |
268 | ||
269 | if (mse) | |
270 | st->percent = mse->percent; | |
271 | ||
272 | for (i=0; i<MD_SB_DISKS; i++) { | |
52826846 | 273 | mdu_disk_info_t disc; |
e0d19036 NB |
274 | int newstate=0; |
275 | int change; | |
276 | char *dv = NULL; | |
52826846 NB |
277 | disc.number = i; |
278 | if (ioctl(fd, GET_DISK_INFO, &disc)>= 0) { | |
e0d19036 NB |
279 | newstate = disc.state; |
280 | dv = map_dev(disc.major, disc.minor); | |
5787fa49 | 281 | } else if (mse && mse->pattern && i < strlen(mse->pattern)) |
e0d19036 NB |
282 | switch(mse->pattern[i]) { |
283 | case 'U': newstate = 6 /* ACTIVE/SYNC */; break; | |
284 | case '_': newstate = 0; break; | |
52826846 | 285 | } |
e0d19036 NB |
286 | change = newstate ^ st->devstate[i]; |
287 | if (st->utime && change && !st->err) { | |
288 | if (i < array.raid_disks && | |
289 | (((newstate&change)&(1<<MD_DISK_FAULTY)) || | |
290 | ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) || | |
291 | ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))) | |
292 | ) | |
293 | alert("Fail", dev, dv, mailaddr, alert_cmd); | |
294 | else if (i>=array.raid_disks && | |
295 | (disc.major || disc.minor) && | |
296 | st->devid[i] == MKDEV(disc.major, disc.minor) && | |
297 | ((newstate&change)&(1<<MD_DISK_FAULTY)) | |
298 | ) | |
299 | alert("FailSpare", dev, dv, mailaddr, alert_cmd); | |
300 | else if (i < array.raid_disks && | |
301 | (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) || | |
302 | ((newstate&change)&(1<<MD_DISK_ACTIVE)) || | |
303 | ((newstate&change)&(1<<MD_DISK_SYNC))) | |
304 | ) | |
305 | alert("SpareActive", dev, dv, mailaddr, alert_cmd); | |
52826846 | 306 | } |
e0d19036 NB |
307 | st->devstate[i] = disc.state; |
308 | st->devid[i] = MKDEV(disc.major, disc.minor); | |
52826846 NB |
309 | } |
310 | close(fd); | |
311 | st->active = array.active_disks; | |
312 | st->working = array.working_disks; | |
313 | st->spare = array.spare_disks; | |
314 | st->failed = array.failed_disks; | |
315 | st->utime = array.utime; | |
e0d19036 NB |
316 | st->raid = array.raid_disks; |
317 | st->err = 0; | |
52826846 | 318 | } |
e0d19036 NB |
319 | /* now check if there are any new devices found in mdstat */ |
320 | if (scan) { | |
321 | struct mdstat_ent *mse; | |
322 | for (mse=mdstat; mse; mse=mse->next) | |
5787fa49 NB |
323 | if (mse->devnum >= 0 && |
324 | (strcmp(mse->level, "raid1")==0 || | |
325 | strcmp(mse->level, "raid5")==0 || | |
326 | strcmp(mse->level, "multipath")==0) | |
327 | ) { | |
e0d19036 NB |
328 | struct state *st = malloc(sizeof *st); |
329 | if (st == NULL) | |
330 | continue; | |
331 | st->devname = strdup(get_md_name(mse->devnum)); | |
332 | st->utime = 0; | |
333 | st->next = statelist; | |
334 | st->err = 1; | |
335 | st->devnum = mse->devnum; | |
336 | st->percent = -2; | |
337 | st->spare_group = NULL; | |
338 | statelist = st; | |
339 | alert("NewArray", st->devname, NULL, mailaddr, alert_cmd); | |
aa88f531 | 340 | new_found = 1; |
e0d19036 NB |
341 | } |
342 | } | |
343 | /* If an array has active < raid && spare == 0 && spare_group != NULL | |
344 | * Look for another array with spare > 0 and active == raid and same spare_group | |
345 | * if found, choose a device and hotremove/hotadd | |
346 | */ | |
347 | for (st = statelist; st; st=st->next) | |
348 | if (st->active < st->raid && | |
349 | st->spare == 0 && | |
350 | st->spare_group != NULL) { | |
351 | struct state *st2; | |
352 | for (st2=statelist ; st2 ; st2=st2->next) | |
353 | if (st2 != st && | |
354 | st2->spare > 0 && | |
355 | st2->active == st2->raid && | |
356 | st2->spare_group != NULL && | |
357 | strcmp(st->spare_group, st2->spare_group) == 0) { | |
358 | /* try to remove and add */ | |
359 | int fd1 = open(st->devname, O_RDONLY); | |
360 | int fd2 = open(st2->devname, O_RDONLY); | |
361 | int dev = -1; | |
362 | int d; | |
363 | if (fd1 < 0 || fd2 < 0) { | |
364 | if (fd1>=0) close(fd1); | |
365 | if (fd2>=0) close(fd2); | |
366 | continue; | |
367 | } | |
368 | for (d=st2->raid; d<MD_SB_DISKS; d++) { | |
369 | if (st2->devid[d] > 0 && | |
370 | st2->devstate[d] == 0) { | |
371 | dev = st2->devid[d]; | |
372 | break; | |
373 | } | |
374 | } | |
375 | if (dev > 0) { | |
376 | if (ioctl(fd2, HOT_REMOVE_DISK, | |
377 | (unsigned long)dev) == 0) { | |
378 | if (ioctl(fd1, HOT_ADD_DISK, | |
379 | (unsigned long)dev) == 0) { | |
380 | alert("MoveSpare", st->devname, st2->devname, mailaddr, alert_cmd); | |
381 | close(fd1); | |
382 | close(fd2); | |
383 | break; | |
384 | } | |
385 | else ioctl(fd2, HOT_ADD_DISK, (unsigned long) dev); | |
386 | } | |
387 | } | |
388 | close(fd1); | |
389 | close(fd2); | |
390 | } | |
391 | } | |
aa88f531 NB |
392 | if (!new_found) { |
393 | if (oneshot) | |
394 | break; | |
395 | else | |
396 | sleep(period); | |
397 | } | |
52826846 NB |
398 | } |
399 | return 0; | |
400 | } | |
401 | ||
402 | ||
403 | static void alert(char *event, char *dev, char *disc, char *mailaddr, char *cmd) | |
404 | { | |
cd29a5c8 NB |
405 | if (!cmd && !mailaddr) { |
406 | time_t now = time(0); | |
407 | ||
e0d19036 | 408 | printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device"); |
cd29a5c8 | 409 | } |
52826846 NB |
410 | if (cmd) { |
411 | int pid = fork(); | |
412 | switch(pid) { | |
413 | default: | |
414 | waitpid(pid, NULL, 0); | |
415 | break; | |
416 | case -1: | |
417 | break; | |
418 | case 0: | |
419 | execl(cmd, cmd, event, dev, disc, NULL); | |
420 | exit(2); | |
421 | } | |
422 | } | |
aa88f531 NB |
423 | if (mailaddr && |
424 | (strncmp(event, "Fail", 4)==0 || | |
425 | strncmp(event, "Degrade", 7)==0)) { | |
52826846 NB |
426 | FILE *mp = popen(Sendmail, "w"); |
427 | if (mp) { | |
428 | char hname[256]; | |
429 | gethostname(hname, sizeof(hname)); | |
430 | signal(SIGPIPE, SIG_IGN); | |
431 | fprintf(mp, "From: " Name " monitoring <root>\n"); | |
432 | fprintf(mp, "To: %s\n", mailaddr); | |
433 | fprintf(mp, "Subject: %s event on %s:%s\n\n", event, dev, hname); | |
434 | ||
435 | fprintf(mp, "This is an automatically generated mail message from " Name "\n"); | |
436 | fprintf(mp, "running on %s\n\n", hname); | |
437 | ||
438 | fprintf(mp, "A %s event had been detected on md device %s.\n\n", event, dev); | |
439 | ||
440 | if (disc) | |
56eedc1a | 441 | fprintf(mp, "It could be related to component device %s.\n\n", disc); |
52826846 NB |
442 | |
443 | fprintf(mp, "Faithfully yours, etc.\n"); | |
444 | fclose(mp); | |
445 | } | |
446 | ||
447 | } | |
448 | /* FIXME log the event to syslog maybe */ | |
449 | } |