]>
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 | 34 | #include <sys/signal.h> |
98c6faba | 35 | #include <values.h> |
52826846 NB |
36 | |
37 | static void alert(char *event, char *dev, char *disc, char *mailaddr, char *cmd); | |
38 | ||
e0d19036 NB |
39 | static char *percentalerts[] = { |
40 | "RebuildStarted", | |
41 | "Rebuild20", | |
42 | "Rebuild40", | |
43 | "Rebuild60", | |
44 | "Rebuild80", | |
45 | }; | |
46 | ||
cd29a5c8 | 47 | int Monitor(mddev_dev_t devlist, |
52826846 | 48 | char *mailaddr, char *alert_cmd, |
aa88f531 | 49 | int period, int daemonise, int scan, int oneshot, |
98c6faba | 50 | char *config, int test) |
52826846 NB |
51 | { |
52 | /* | |
53 | * Every few seconds, scan every md device looking for changes | |
54 | * When a change is found, log it, possibly run the alert command, | |
55 | * and possibly send Email | |
56 | * | |
57 | * For each array, we record: | |
58 | * Update time | |
59 | * active/working/failed/spare drives | |
60 | * State of each device. | |
e0d19036 | 61 | * %rebuilt if rebuilding |
52826846 NB |
62 | * |
63 | * If the update time changes, check out all the data again | |
64 | * It is possible that we cannot get the state of each device | |
65 | * due to bugs in the md kernel module. | |
e0d19036 NB |
66 | * We also read /proc/mdstat to get rebuild percent, |
67 | * and to get state on all active devices incase of kernel bug. | |
52826846 | 68 | * |
e0d19036 NB |
69 | * Events are: |
70 | * Fail | |
71 | * An active device had Faulty set or Active/Sync removed | |
72 | * FailSpare | |
73 | * A spare device had Faulty set | |
74 | * SpareActive | |
75 | * An active device had a reverse transition | |
76 | * RebuildStarted | |
77 | * percent went from -1 to +ve | |
78 | * Rebuild20 Rebuild40 Rebuild60 Rebuild80 | |
79 | * percent went from below to not-below that number | |
80 | * DeviceDisappeared | |
81 | * Couldn't access a device which was previously visible | |
52826846 NB |
82 | * |
83 | * if we detect an array with active<raid and spare==0 | |
84 | * we look at other arrays that have same spare-group | |
85 | * If we find one with active==raid and spare>0, | |
86 | * and if we can get_disk_info and find a name | |
87 | * Then we hot-remove and hot-add to the other array | |
88 | * | |
e0d19036 NB |
89 | * If devlist is NULL, then we can monitor everything because --scan |
90 | * was given. We get an initial list from config file and add anything | |
91 | * that appears in /proc/mdstat | |
52826846 NB |
92 | */ |
93 | ||
94 | struct state { | |
95 | char *devname; | |
e0d19036 | 96 | int devnum; /* to sync with mdstat info */ |
52826846 NB |
97 | long utime; |
98 | int err; | |
e0d19036 NB |
99 | char *spare_group; |
100 | int active, working, failed, spare, raid; | |
feb716e9 | 101 | int expected_spares; |
52826846 | 102 | int devstate[MD_SB_DISKS]; |
e0d19036 NB |
103 | int devid[MD_SB_DISKS]; |
104 | int percent; | |
52826846 NB |
105 | struct state *next; |
106 | } *statelist = NULL; | |
107 | int finished = 0; | |
e0d19036 NB |
108 | struct mdstat_ent *mdstat = NULL; |
109 | ||
d013a55e | 110 | if (!mailaddr) { |
e0d19036 | 111 | mailaddr = conf_get_mailaddr(config); |
d013a55e | 112 | if (mailaddr && ! scan) |
56eedc1a | 113 | fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n", |
d013a55e NB |
114 | mailaddr); |
115 | } | |
116 | if (!alert_cmd) { | |
e0d19036 | 117 | alert_cmd = conf_get_program(config); |
d013a55e | 118 | if (alert_cmd && ! scan) |
56eedc1a | 119 | fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n", |
d013a55e NB |
120 | alert_cmd); |
121 | } | |
56eedc1a NB |
122 | if (scan && !mailaddr && !alert_cmd) { |
123 | fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n"); | |
d013a55e | 124 | return 1; |
56eedc1a | 125 | } |
d013a55e NB |
126 | |
127 | if (daemonise) { | |
128 | int pid = fork(); | |
129 | if (pid > 0) { | |
130 | printf("%d\n", pid); | |
131 | return 0; | |
132 | } | |
133 | if (pid < 0) { | |
134 | perror("daemonise"); | |
135 | return 1; | |
136 | } | |
137 | close(0); | |
138 | open("/dev/null", 3); | |
139 | dup2(0,1); | |
140 | dup2(0,2); | |
141 | setsid(); | |
142 | } | |
e0d19036 NB |
143 | |
144 | if (devlist == NULL) { | |
145 | mddev_ident_t mdlist = conf_get_ident(config, NULL); | |
146 | for (; mdlist; mdlist=mdlist->next) { | |
147 | struct state *st = malloc(sizeof *st); | |
148 | if (st == NULL) | |
149 | continue; | |
150 | st->devname = strdup(mdlist->devname); | |
151 | st->utime = 0; | |
152 | st->next = statelist; | |
d013a55e | 153 | st->err = 0; |
98c6faba | 154 | st->devnum = MAXINT; |
e0d19036 | 155 | st->percent = -2; |
feb716e9 | 156 | st->expected_spares = mdlist->spare_disks; |
e0d19036 NB |
157 | if (mdlist->spare_group) |
158 | st->spare_group = strdup(mdlist->spare_group); | |
159 | else | |
160 | st->spare_group = NULL; | |
161 | statelist = st; | |
162 | } | |
163 | } else { | |
cd29a5c8 | 164 | mddev_dev_t dv; |
e0d19036 NB |
165 | for (dv=devlist ; dv; dv=dv->next) { |
166 | struct state *st = malloc(sizeof *st); | |
167 | if (st == NULL) | |
168 | continue; | |
169 | st->devname = strdup(dv->devname); | |
170 | st->utime = 0; | |
171 | st->next = statelist; | |
d013a55e | 172 | st->err = 0; |
98c6faba | 173 | st->devnum = MAXINT; |
e0d19036 | 174 | st->percent = -2; |
feb716e9 | 175 | st->expected_spares = -1; |
e0d19036 NB |
176 | st->spare_group = NULL; |
177 | statelist = st; | |
178 | } | |
179 | } | |
180 | ||
181 | ||
182 | while (! finished) { | |
aa88f531 | 183 | int new_found = 0; |
e0d19036 NB |
184 | struct state *st; |
185 | ||
186 | if (mdstat) | |
187 | free_mdstat(mdstat); | |
dd0781e5 | 188 | mdstat = mdstat_read(oneshot?0:1); |
e0d19036 NB |
189 | |
190 | for (st=statelist; st; st=st->next) { | |
52826846 | 191 | mdu_array_info_t array; |
dd0781e5 | 192 | struct mdstat_ent *mse = NULL, *mse2; |
e0d19036 | 193 | char *dev = st->devname; |
52826846 | 194 | int fd; |
98c6faba | 195 | unsigned int i; |
e0d19036 | 196 | |
98c6faba NB |
197 | if (test) |
198 | alert("TestMessage", dev, NULL, mailaddr, alert_cmd); | |
52826846 NB |
199 | fd = open(dev, O_RDONLY); |
200 | if (fd < 0) { | |
201 | if (!st->err) | |
e0d19036 NB |
202 | alert("DeviceDisappeared", dev, NULL, |
203 | mailaddr, alert_cmd); | |
204 | /* fprintf(stderr, Name ": cannot open %s: %s\n", | |
52826846 | 205 | dev, strerror(errno)); |
e0d19036 | 206 | */ st->err=1; |
52826846 NB |
207 | continue; |
208 | } | |
209 | if (ioctl(fd, GET_ARRAY_INFO, &array)<0) { | |
210 | if (!st->err) | |
e0d19036 NB |
211 | alert("DeviceDisappeared", dev, NULL, |
212 | mailaddr, alert_cmd); | |
213 | /* fprintf(stderr, Name ": cannot get array info for %s: %s\n", | |
52826846 | 214 | dev, strerror(errno)); |
e0d19036 | 215 | */ st->err=1; |
52826846 NB |
216 | close(fd); |
217 | continue; | |
218 | } | |
5787fa49 NB |
219 | if (array.level != 1 && array.level != 5 && array.level != -4) { |
220 | if (!st->err) | |
221 | alert("DeviceDisappeared", dev, "Wrong-Level", | |
222 | mailaddr, alert_cmd); | |
223 | st->err = 1; | |
224 | close(fd); | |
225 | continue; | |
226 | } | |
98c6faba | 227 | if (st->devnum == MAXINT) { |
e0d19036 NB |
228 | struct stat stb; |
229 | if (fstat(fd, &stb) == 0 && | |
98c6faba | 230 | (S_IFMT&stb.st_mode)==S_IFBLK) { |
dd0781e5 | 231 | if (MAJOR(stb.st_rdev) == MD_MAJOR) |
98c6faba NB |
232 | st->devnum = MINOR(stb.st_rdev); |
233 | else | |
234 | st->devnum = -1- (MINOR(stb.st_rdev)>>6); | |
235 | } | |
e0d19036 NB |
236 | } |
237 | ||
dd0781e5 NB |
238 | for (mse2 = mdstat ; mse2 ; mse2=mse2->next) |
239 | if (mse2->devnum == st->devnum) { | |
240 | mse2->devnum = MAXINT; /* flag it as "used" */ | |
241 | mse = mse2; | |
242 | } | |
e0d19036 | 243 | |
52826846 | 244 | if (st->utime == array.utime && |
e0d19036 NB |
245 | st->failed == array.failed_disks && |
246 | st->working == array.working_disks && | |
247 | st->spare == array.spare_disks && | |
248 | (mse == NULL || ( | |
249 | mse->percent == st->percent | |
250 | ))) { | |
52826846 | 251 | close(fd); |
e0d19036 | 252 | st->err = 0; |
52826846 NB |
253 | continue; |
254 | } | |
aa88f531 NB |
255 | if (st->utime == 0 && /* new array */ |
256 | mse && /* is in /proc/mdstat */ | |
257 | mse->pattern && strchr(mse->pattern, '_') /* degraded */ | |
258 | ) | |
259 | alert("DegradedArray", dev, NULL, mailaddr, alert_cmd); | |
260 | ||
feb716e9 NB |
261 | if (st->utime == 0 && /* new array */ |
262 | st->expected_spares > 0 && | |
263 | array.spare_disks < st->expected_spares) | |
264 | alert("SparesMissing", dev, NULL, mailaddr, alert_cmd); | |
e0d19036 NB |
265 | if (mse && |
266 | st->percent == -1 && | |
267 | mse->percent >= 0) | |
268 | alert("RebuildStarted", dev, NULL, mailaddr, alert_cmd); | |
269 | if (mse && | |
270 | st->percent >= 0 && | |
271 | mse->percent >= 0 && | |
272 | (mse->percent / 20) > (st->percent / 20)) | |
273 | alert(percentalerts[mse->percent/20], | |
274 | dev, NULL, mailaddr, alert_cmd); | |
275 | ||
98c6faba NB |
276 | if (mse && |
277 | mse->percent == -1 && | |
278 | st->percent >= 0) | |
279 | alert("RebuildFinished", dev, NULL, mailaddr, alert_cmd); | |
280 | ||
e0d19036 NB |
281 | if (mse) |
282 | st->percent = mse->percent; | |
283 | ||
284 | for (i=0; i<MD_SB_DISKS; i++) { | |
52826846 | 285 | mdu_disk_info_t disc; |
e0d19036 NB |
286 | int newstate=0; |
287 | int change; | |
288 | char *dv = NULL; | |
52826846 NB |
289 | disc.number = i; |
290 | if (ioctl(fd, GET_DISK_INFO, &disc)>= 0) { | |
e0d19036 NB |
291 | newstate = disc.state; |
292 | dv = map_dev(disc.major, disc.minor); | |
5787fa49 | 293 | } else if (mse && mse->pattern && i < strlen(mse->pattern)) |
e0d19036 NB |
294 | switch(mse->pattern[i]) { |
295 | case 'U': newstate = 6 /* ACTIVE/SYNC */; break; | |
296 | case '_': newstate = 0; break; | |
52826846 | 297 | } |
e0d19036 NB |
298 | change = newstate ^ st->devstate[i]; |
299 | if (st->utime && change && !st->err) { | |
98c6faba | 300 | if (i < (unsigned)array.raid_disks && |
e0d19036 NB |
301 | (((newstate&change)&(1<<MD_DISK_FAULTY)) || |
302 | ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) || | |
303 | ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))) | |
304 | ) | |
305 | alert("Fail", dev, dv, mailaddr, alert_cmd); | |
98c6faba | 306 | else if (i >= (unsigned)array.raid_disks && |
e0d19036 NB |
307 | (disc.major || disc.minor) && |
308 | st->devid[i] == MKDEV(disc.major, disc.minor) && | |
309 | ((newstate&change)&(1<<MD_DISK_FAULTY)) | |
310 | ) | |
311 | alert("FailSpare", dev, dv, mailaddr, alert_cmd); | |
98c6faba | 312 | else if (i < (unsigned)array.raid_disks && |
e0d19036 NB |
313 | (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) || |
314 | ((newstate&change)&(1<<MD_DISK_ACTIVE)) || | |
315 | ((newstate&change)&(1<<MD_DISK_SYNC))) | |
316 | ) | |
317 | alert("SpareActive", dev, dv, mailaddr, alert_cmd); | |
52826846 | 318 | } |
e0d19036 NB |
319 | st->devstate[i] = disc.state; |
320 | st->devid[i] = MKDEV(disc.major, disc.minor); | |
52826846 NB |
321 | } |
322 | close(fd); | |
323 | st->active = array.active_disks; | |
324 | st->working = array.working_disks; | |
325 | st->spare = array.spare_disks; | |
326 | st->failed = array.failed_disks; | |
327 | st->utime = array.utime; | |
e0d19036 NB |
328 | st->raid = array.raid_disks; |
329 | st->err = 0; | |
52826846 | 330 | } |
e0d19036 NB |
331 | /* now check if there are any new devices found in mdstat */ |
332 | if (scan) { | |
333 | struct mdstat_ent *mse; | |
334 | for (mse=mdstat; mse; mse=mse->next) | |
98c6faba | 335 | if (mse->devnum != MAXINT && |
5787fa49 NB |
336 | (strcmp(mse->level, "raid1")==0 || |
337 | strcmp(mse->level, "raid5")==0 || | |
338 | strcmp(mse->level, "multipath")==0) | |
339 | ) { | |
e0d19036 | 340 | struct state *st = malloc(sizeof *st); |
98c6faba NB |
341 | mdu_array_info_t array; |
342 | int fd; | |
e0d19036 NB |
343 | if (st == NULL) |
344 | continue; | |
345 | st->devname = strdup(get_md_name(mse->devnum)); | |
98c6faba NB |
346 | if ((fd = open(st->devname, O_RDONLY)) < 0 || |
347 | ioctl(fd, GET_ARRAY_INFO, &array)< 0) { | |
348 | /* no such array */ | |
349 | if (fd >=0) close(fd); | |
350 | free(st->devname); | |
351 | free(st); | |
352 | continue; | |
353 | } | |
dd0781e5 | 354 | close(fd); |
e0d19036 NB |
355 | st->utime = 0; |
356 | st->next = statelist; | |
357 | st->err = 1; | |
358 | st->devnum = mse->devnum; | |
359 | st->percent = -2; | |
360 | st->spare_group = NULL; | |
98c6faba | 361 | st->expected_spares = -1; |
e0d19036 NB |
362 | statelist = st; |
363 | alert("NewArray", st->devname, NULL, mailaddr, alert_cmd); | |
aa88f531 | 364 | new_found = 1; |
e0d19036 NB |
365 | } |
366 | } | |
367 | /* If an array has active < raid && spare == 0 && spare_group != NULL | |
368 | * Look for another array with spare > 0 and active == raid and same spare_group | |
369 | * if found, choose a device and hotremove/hotadd | |
370 | */ | |
371 | for (st = statelist; st; st=st->next) | |
372 | if (st->active < st->raid && | |
373 | st->spare == 0 && | |
374 | st->spare_group != NULL) { | |
375 | struct state *st2; | |
376 | for (st2=statelist ; st2 ; st2=st2->next) | |
377 | if (st2 != st && | |
378 | st2->spare > 0 && | |
379 | st2->active == st2->raid && | |
380 | st2->spare_group != NULL && | |
381 | strcmp(st->spare_group, st2->spare_group) == 0) { | |
382 | /* try to remove and add */ | |
383 | int fd1 = open(st->devname, O_RDONLY); | |
384 | int fd2 = open(st2->devname, O_RDONLY); | |
385 | int dev = -1; | |
386 | int d; | |
387 | if (fd1 < 0 || fd2 < 0) { | |
388 | if (fd1>=0) close(fd1); | |
389 | if (fd2>=0) close(fd2); | |
390 | continue; | |
391 | } | |
392 | for (d=st2->raid; d<MD_SB_DISKS; d++) { | |
393 | if (st2->devid[d] > 0 && | |
394 | st2->devstate[d] == 0) { | |
395 | dev = st2->devid[d]; | |
396 | break; | |
397 | } | |
398 | } | |
399 | if (dev > 0) { | |
400 | if (ioctl(fd2, HOT_REMOVE_DISK, | |
401 | (unsigned long)dev) == 0) { | |
402 | if (ioctl(fd1, HOT_ADD_DISK, | |
403 | (unsigned long)dev) == 0) { | |
404 | alert("MoveSpare", st->devname, st2->devname, mailaddr, alert_cmd); | |
405 | close(fd1); | |
406 | close(fd2); | |
407 | break; | |
408 | } | |
409 | else ioctl(fd2, HOT_ADD_DISK, (unsigned long) dev); | |
410 | } | |
411 | } | |
412 | close(fd1); | |
413 | close(fd2); | |
414 | } | |
415 | } | |
aa88f531 NB |
416 | if (!new_found) { |
417 | if (oneshot) | |
418 | break; | |
419 | else | |
dd0781e5 | 420 | mdstat_wait(period); |
aa88f531 | 421 | } |
98c6faba | 422 | test = 0; |
52826846 NB |
423 | } |
424 | return 0; | |
425 | } | |
426 | ||
427 | ||
428 | static void alert(char *event, char *dev, char *disc, char *mailaddr, char *cmd) | |
429 | { | |
cd29a5c8 NB |
430 | if (!cmd && !mailaddr) { |
431 | time_t now = time(0); | |
432 | ||
e0d19036 | 433 | printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device"); |
cd29a5c8 | 434 | } |
52826846 NB |
435 | if (cmd) { |
436 | int pid = fork(); | |
437 | switch(pid) { | |
438 | default: | |
439 | waitpid(pid, NULL, 0); | |
440 | break; | |
441 | case -1: | |
442 | break; | |
443 | case 0: | |
444 | execl(cmd, cmd, event, dev, disc, NULL); | |
445 | exit(2); | |
446 | } | |
447 | } | |
aa88f531 NB |
448 | if (mailaddr && |
449 | (strncmp(event, "Fail", 4)==0 || | |
98c6faba | 450 | strncmp(event, "Test", 4)==0 || |
aa88f531 | 451 | strncmp(event, "Degrade", 7)==0)) { |
52826846 NB |
452 | FILE *mp = popen(Sendmail, "w"); |
453 | if (mp) { | |
454 | char hname[256]; | |
455 | gethostname(hname, sizeof(hname)); | |
456 | signal(SIGPIPE, SIG_IGN); | |
457 | fprintf(mp, "From: " Name " monitoring <root>\n"); | |
458 | fprintf(mp, "To: %s\n", mailaddr); | |
459 | fprintf(mp, "Subject: %s event on %s:%s\n\n", event, dev, hname); | |
460 | ||
461 | fprintf(mp, "This is an automatically generated mail message from " Name "\n"); | |
462 | fprintf(mp, "running on %s\n\n", hname); | |
463 | ||
464 | fprintf(mp, "A %s event had been detected on md device %s.\n\n", event, dev); | |
465 | ||
466 | if (disc) | |
56eedc1a | 467 | fprintf(mp, "It could be related to component device %s.\n\n", disc); |
52826846 NB |
468 | |
469 | fprintf(mp, "Faithfully yours, etc.\n"); | |
470 | fclose(mp); | |
471 | } | |
472 | ||
473 | } | |
474 | /* FIXME log the event to syslog maybe */ | |
475 | } |