]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Monitor.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
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.
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.
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
22 * Email: <neilb@cse.unsw.edu.au>
24 * School of Computer Science and Engineering
25 * The University of New South Wales
34 #include <sys/signal.h>
37 static void alert(char *event
, char *dev
, char *disc
, char *mailaddr
, char *cmd
);
39 static char *percentalerts
[] = {
47 int Monitor(mddev_dev_t devlist
,
48 char *mailaddr
, char *alert_cmd
,
49 int period
, int daemonise
, int scan
, int oneshot
,
50 char *config
, int test
, char* pidfile
)
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
57 * For each array, we record:
59 * active/working/failed/spare drives
60 * State of each device.
61 * %rebuilt if rebuilding
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.
66 * We also read /proc/mdstat to get rebuild percent,
67 * and to get state on all active devices incase of kernel bug.
71 * An active device had Faulty set or Active/Sync removed
73 * A spare device had Faulty set
75 * An active device had a reverse transition
77 * percent went from -1 to +ve
78 * Rebuild20 Rebuild40 Rebuild60 Rebuild80
79 * percent went from below to not-below that number
81 * Couldn't access a device which was previously visible
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
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
96 int devnum
; /* to sync with mdstat info */
100 int active
, working
, failed
, spare
, raid
;
102 int devstate
[MD_SB_DISKS
];
103 int devid
[MD_SB_DISKS
];
108 struct mdstat_ent
*mdstat
= NULL
;
111 mailaddr
= conf_get_mailaddr(config
);
112 if (mailaddr
&& ! scan
)
113 fprintf(stderr
, Name
": Monitor using email address \"%s\" from config file\n",
117 alert_cmd
= conf_get_program(config
);
118 if (alert_cmd
&& ! scan
)
119 fprintf(stderr
, Name
": Monitor using program \"%s\" from config file\n",
122 if (scan
&& !mailaddr
&& !alert_cmd
) {
123 fprintf(stderr
, Name
": No mail address or alert command - not monitoring.\n");
134 pid_file
=fopen(pidfile
, "w");
136 perror("cannot create pid file");
138 fprintf(pid_file
,"%d\n", pid
);
149 open("/dev/null", 3);
155 if (devlist
== NULL
) {
156 mddev_ident_t mdlist
= conf_get_ident(config
, NULL
);
157 for (; mdlist
; mdlist
=mdlist
->next
) {
158 struct state
*st
= malloc(sizeof *st
);
161 st
->devname
= strdup(mdlist
->devname
);
163 st
->next
= statelist
;
167 st
->expected_spares
= mdlist
->spare_disks
;
168 if (mdlist
->spare_group
)
169 st
->spare_group
= strdup(mdlist
->spare_group
);
171 st
->spare_group
= NULL
;
176 for (dv
=devlist
; dv
; dv
=dv
->next
) {
177 mddev_ident_t mdlist
= conf_get_ident(config
, dv
->devname
);
178 struct state
*st
= malloc(sizeof *st
);
181 st
->devname
= strdup(dv
->devname
);
183 st
->next
= statelist
;
187 st
->expected_spares
= -1;
188 st
->spare_group
= NULL
;
190 st
->expected_spares
= mdlist
->spare_disks
;
191 if (mdlist
->spare_group
)
192 st
->spare_group
= strdup(mdlist
->spare_group
);
205 mdstat
= mdstat_read(oneshot
?0:1);
207 for (st
=statelist
; st
; st
=st
->next
) {
208 mdu_array_info_t array
;
209 struct mdstat_ent
*mse
= NULL
, *mse2
;
210 char *dev
= st
->devname
;
215 alert("TestMessage", dev
, NULL
, mailaddr
, alert_cmd
);
216 fd
= open(dev
, O_RDONLY
);
219 alert("DeviceDisappeared", dev
, NULL
,
220 mailaddr
, alert_cmd
);
221 /* fprintf(stderr, Name ": cannot open %s: %s\n",
222 dev, strerror(errno));
226 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)<0) {
228 alert("DeviceDisappeared", dev
, NULL
,
229 mailaddr
, alert_cmd
);
230 /* fprintf(stderr, Name ": cannot get array info for %s: %s\n",
231 dev, strerror(errno));
236 if (array
.level
!= 1 && array
.level
!= 5 && array
.level
!= -4 &&
237 array
.level
!= 6 && array
.level
!= 10) {
239 alert("DeviceDisappeared", dev
, "Wrong-Level",
240 mailaddr
, alert_cmd
);
245 if (st
->devnum
== MAXINT
) {
247 if (fstat(fd
, &stb
) == 0 &&
248 (S_IFMT
&stb
.st_mode
)==S_IFBLK
) {
249 if (major(stb
.st_rdev
) == MD_MAJOR
)
250 st
->devnum
= minor(stb
.st_rdev
);
252 st
->devnum
= -1- (minor(stb
.st_rdev
)>>6);
256 for (mse2
= mdstat
; mse2
; mse2
=mse2
->next
)
257 if (mse2
->devnum
== st
->devnum
) {
258 mse2
->devnum
= MAXINT
; /* flag it as "used" */
262 if (st
->utime
== array
.utime
&&
263 st
->failed
== array
.failed_disks
&&
264 st
->working
== array
.working_disks
&&
265 st
->spare
== array
.spare_disks
&&
267 mse
->percent
== st
->percent
273 if (st
->utime
== 0 && /* new array */
274 mse
&& /* is in /proc/mdstat */
275 mse
->pattern
&& strchr(mse
->pattern
, '_') /* degraded */
277 alert("DegradedArray", dev
, NULL
, mailaddr
, alert_cmd
);
279 if (st
->utime
== 0 && /* new array */
280 st
->expected_spares
> 0 &&
281 array
.spare_disks
< st
->expected_spares
)
282 alert("SparesMissing", dev
, NULL
, mailaddr
, alert_cmd
);
286 alert("RebuildStarted", dev
, NULL
, mailaddr
, alert_cmd
);
290 (mse
->percent
/ 20) > (st
->percent
/ 20))
291 alert(percentalerts
[mse
->percent
/20],
292 dev
, NULL
, mailaddr
, alert_cmd
);
295 mse
->percent
== -1 &&
297 alert("RebuildFinished", dev
, NULL
, mailaddr
, alert_cmd
);
300 st
->percent
= mse
->percent
;
302 for (i
=0; i
<MD_SB_DISKS
; i
++) {
303 mdu_disk_info_t disc
;
308 if (ioctl(fd
, GET_DISK_INFO
, &disc
)>= 0) {
309 newstate
= disc
.state
;
310 dv
= map_dev(disc
.major
, disc
.minor
);
311 } else if (mse
&& mse
->pattern
&& i
< strlen(mse
->pattern
))
312 switch(mse
->pattern
[i
]) {
313 case 'U': newstate
= 6 /* ACTIVE/SYNC */; break;
314 case '_': newstate
= 0; break;
316 change
= newstate
^ st
->devstate
[i
];
317 if (st
->utime
&& change
&& !st
->err
) {
318 if (i
< (unsigned)array
.raid_disks
&&
319 (((newstate
&change
)&(1<<MD_DISK_FAULTY
)) ||
320 ((st
->devstate
[i
]&change
)&(1<<MD_DISK_ACTIVE
)) ||
321 ((st
->devstate
[i
]&change
)&(1<<MD_DISK_SYNC
)))
323 alert("Fail", dev
, dv
, mailaddr
, alert_cmd
);
324 else if (i
>= (unsigned)array
.raid_disks
&&
325 (disc
.major
|| disc
.minor
) &&
326 st
->devid
[i
] == makedev(disc
.major
, disc
.minor
) &&
327 ((newstate
&change
)&(1<<MD_DISK_FAULTY
))
329 alert("FailSpare", dev
, dv
, mailaddr
, alert_cmd
);
330 else if (i
< (unsigned)array
.raid_disks
&&
331 (((st
->devstate
[i
]&change
)&(1<<MD_DISK_FAULTY
)) ||
332 ((newstate
&change
)&(1<<MD_DISK_ACTIVE
)) ||
333 ((newstate
&change
)&(1<<MD_DISK_SYNC
)))
335 alert("SpareActive", dev
, dv
, mailaddr
, alert_cmd
);
337 st
->devstate
[i
] = disc
.state
;
338 st
->devid
[i
] = makedev(disc
.major
, disc
.minor
);
341 st
->active
= array
.active_disks
;
342 st
->working
= array
.working_disks
;
343 st
->spare
= array
.spare_disks
;
344 st
->failed
= array
.failed_disks
;
345 st
->utime
= array
.utime
;
346 st
->raid
= array
.raid_disks
;
349 /* now check if there are any new devices found in mdstat */
351 struct mdstat_ent
*mse
;
352 for (mse
=mdstat
; mse
; mse
=mse
->next
)
353 if (mse
->devnum
!= MAXINT
&&
354 (strcmp(mse
->level
, "raid1")==0 ||
355 strcmp(mse
->level
, "raid5")==0 ||
356 strcmp(mse
->level
, "multipath")==0)
358 struct state
*st
= malloc(sizeof *st
);
359 mdu_array_info_t array
;
363 st
->devname
= strdup(get_md_name(mse
->devnum
));
364 if ((fd
= open(st
->devname
, O_RDONLY
)) < 0 ||
365 ioctl(fd
, GET_ARRAY_INFO
, &array
)< 0) {
367 if (fd
>=0) close(fd
);
374 st
->next
= statelist
;
376 st
->devnum
= mse
->devnum
;
378 st
->spare_group
= NULL
;
379 st
->expected_spares
= -1;
381 alert("NewArray", st
->devname
, NULL
, mailaddr
, alert_cmd
);
385 /* If an array has active < raid && spare == 0 && spare_group != NULL
386 * Look for another array with spare > 0 and active == raid and same spare_group
387 * if found, choose a device and hotremove/hotadd
389 for (st
= statelist
; st
; st
=st
->next
)
390 if (st
->active
< st
->raid
&&
392 st
->spare_group
!= NULL
) {
394 for (st2
=statelist
; st2
; st2
=st2
->next
)
397 st2
->active
== st2
->raid
&&
398 st2
->spare_group
!= NULL
&&
399 strcmp(st
->spare_group
, st2
->spare_group
) == 0) {
400 /* try to remove and add */
401 int fd1
= open(st
->devname
, O_RDONLY
);
402 int fd2
= open(st2
->devname
, O_RDONLY
);
405 if (fd1
< 0 || fd2
< 0) {
406 if (fd1
>=0) close(fd1
);
407 if (fd2
>=0) close(fd2
);
410 for (d
=st2
->raid
; d
<MD_SB_DISKS
; d
++) {
411 if (st2
->devid
[d
] > 0 &&
412 st2
->devstate
[d
] == 0) {
418 if (ioctl(fd2
, HOT_REMOVE_DISK
,
419 (unsigned long)dev
) == 0) {
420 if (ioctl(fd1
, HOT_ADD_DISK
,
421 (unsigned long)dev
) == 0) {
422 alert("MoveSpare", st
->devname
, st2
->devname
, mailaddr
, alert_cmd
);
427 else ioctl(fd2
, HOT_ADD_DISK
, (unsigned long) dev
);
448 static void alert(char *event
, char *dev
, char *disc
, char *mailaddr
, char *cmd
)
450 if (!cmd
&& !mailaddr
) {
451 time_t now
= time(0);
453 printf("%1.15s: %s on %s %s\n", ctime(&now
)+4, event
, dev
, disc
?disc
:"unknown device");
459 waitpid(pid
, NULL
, 0);
464 execl(cmd
, cmd
, event
, dev
, disc
, NULL
);
469 (strncmp(event
, "Fail", 4)==0 ||
470 strncmp(event
, "Test", 4)==0 ||
471 strncmp(event
, "Degrade", 7)==0)) {
472 FILE *mp
= popen(Sendmail
, "w");
475 gethostname(hname
, sizeof(hname
));
476 signal(SIGPIPE
, SIG_IGN
);
477 fprintf(mp
, "From: " Name
" monitoring <root>\n");
478 fprintf(mp
, "To: %s\n", mailaddr
);
479 fprintf(mp
, "Subject: %s event on %s:%s\n\n", event
, dev
, hname
);
481 fprintf(mp
, "This is an automatically generated mail message from " Name
"\n");
482 fprintf(mp
, "running on %s\n\n", hname
);
484 fprintf(mp
, "A %s event had been detected on md device %s.\n\n", event
, dev
);
487 fprintf(mp
, "It could be related to component device %s.\n\n", disc
);
489 fprintf(mp
, "Faithfully yours, etc.\n");
494 /* FIXME log the event to syslog maybe */