]>
Commit | Line | Data |
---|---|---|
52826846 | 1 | /* |
9a9dab36 | 2 | * mdadm - manage Linux "md" devices aka RAID arrays. |
52826846 | 3 | * |
4f589ad0 | 4 | * Copyright (C) 2001-2006 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 | |
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> |
98127a6d | 34 | #include <signal.h> |
4450e59f | 35 | #include <limits.h> |
773135f5 | 36 | #include <syslog.h> |
52826846 | 37 | |
4948b8f7 NB |
38 | static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mailfrom, |
39 | char *cmd, int dosyslog); | |
52826846 | 40 | |
aba69144 | 41 | static char *percentalerts[] = { |
e0d19036 NB |
42 | "RebuildStarted", |
43 | "Rebuild20", | |
44 | "Rebuild40", | |
45 | "Rebuild60", | |
46 | "Rebuild80", | |
47 | }; | |
48 | ||
e27d562b NB |
49 | /* The largest number of disks current arrays can manage is 384 |
50 | * This really should be dynamically, but that will have to wait | |
51 | * At least it isn't MD_SB_DISKS. | |
52 | */ | |
53 | #define MaxDisks 384 | |
cd29a5c8 | 54 | int Monitor(mddev_dev_t devlist, |
52826846 | 55 | char *mailaddr, char *alert_cmd, |
aa88f531 | 56 | int period, int daemonise, int scan, int oneshot, |
8aec876d | 57 | int dosyslog, int test, char* pidfile) |
52826846 NB |
58 | { |
59 | /* | |
60 | * Every few seconds, scan every md device looking for changes | |
61 | * When a change is found, log it, possibly run the alert command, | |
62 | * and possibly send Email | |
63 | * | |
64 | * For each array, we record: | |
65 | * Update time | |
66 | * active/working/failed/spare drives | |
67 | * State of each device. | |
e0d19036 | 68 | * %rebuilt if rebuilding |
52826846 NB |
69 | * |
70 | * If the update time changes, check out all the data again | |
71 | * It is possible that we cannot get the state of each device | |
72 | * due to bugs in the md kernel module. | |
e0d19036 NB |
73 | * We also read /proc/mdstat to get rebuild percent, |
74 | * and to get state on all active devices incase of kernel bug. | |
52826846 | 75 | * |
e0d19036 NB |
76 | * Events are: |
77 | * Fail | |
78 | * An active device had Faulty set or Active/Sync removed | |
79 | * FailSpare | |
80 | * A spare device had Faulty set | |
81 | * SpareActive | |
82 | * An active device had a reverse transition | |
83 | * RebuildStarted | |
84 | * percent went from -1 to +ve | |
85 | * Rebuild20 Rebuild40 Rebuild60 Rebuild80 | |
86 | * percent went from below to not-below that number | |
87 | * DeviceDisappeared | |
88 | * Couldn't access a device which was previously visible | |
52826846 NB |
89 | * |
90 | * if we detect an array with active<raid and spare==0 | |
91 | * we look at other arrays that have same spare-group | |
92 | * If we find one with active==raid and spare>0, | |
93 | * and if we can get_disk_info and find a name | |
94 | * Then we hot-remove and hot-add to the other array | |
95 | * | |
e0d19036 NB |
96 | * If devlist is NULL, then we can monitor everything because --scan |
97 | * was given. We get an initial list from config file and add anything | |
98 | * that appears in /proc/mdstat | |
52826846 NB |
99 | */ |
100 | ||
101 | struct state { | |
102 | char *devname; | |
e0d19036 | 103 | int devnum; /* to sync with mdstat info */ |
52826846 NB |
104 | long utime; |
105 | int err; | |
e0d19036 NB |
106 | char *spare_group; |
107 | int active, working, failed, spare, raid; | |
feb716e9 | 108 | int expected_spares; |
e27d562b NB |
109 | int devstate[MaxDisks]; |
110 | int devid[MaxDisks]; | |
e0d19036 | 111 | int percent; |
52826846 NB |
112 | struct state *next; |
113 | } *statelist = NULL; | |
114 | int finished = 0; | |
e0d19036 | 115 | struct mdstat_ent *mdstat = NULL; |
4948b8f7 | 116 | char *mailfrom = NULL; |
e0d19036 | 117 | |
d013a55e | 118 | if (!mailaddr) { |
8aec876d | 119 | mailaddr = conf_get_mailaddr(); |
d013a55e | 120 | if (mailaddr && ! scan) |
56eedc1a | 121 | fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n", |
d013a55e NB |
122 | mailaddr); |
123 | } | |
8aec876d | 124 | mailfrom = conf_get_mailfrom(); |
4948b8f7 | 125 | |
d013a55e | 126 | if (!alert_cmd) { |
8aec876d | 127 | alert_cmd = conf_get_program(); |
d013a55e | 128 | if (alert_cmd && ! scan) |
56eedc1a | 129 | fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n", |
d013a55e NB |
130 | alert_cmd); |
131 | } | |
56eedc1a NB |
132 | if (scan && !mailaddr && !alert_cmd) { |
133 | fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n"); | |
d013a55e | 134 | return 1; |
56eedc1a | 135 | } |
d013a55e NB |
136 | |
137 | if (daemonise) { | |
138 | int pid = fork(); | |
139 | if (pid > 0) { | |
b5e64645 NB |
140 | if (!pidfile) |
141 | printf("%d\n", pid); | |
142 | else { | |
143 | FILE *pid_file; | |
144 | pid_file=fopen(pidfile, "w"); | |
145 | if (!pid_file) | |
146 | perror("cannot create pid file"); | |
147 | else { | |
148 | fprintf(pid_file,"%d\n", pid); | |
149 | fclose(pid_file); | |
150 | } | |
151 | } | |
d013a55e NB |
152 | return 0; |
153 | } | |
154 | if (pid < 0) { | |
155 | perror("daemonise"); | |
156 | return 1; | |
157 | } | |
158 | close(0); | |
fb97b4d6 | 159 | open("/dev/null", O_RDWR); |
d013a55e NB |
160 | dup2(0,1); |
161 | dup2(0,2); | |
162 | setsid(); | |
163 | } | |
e0d19036 NB |
164 | |
165 | if (devlist == NULL) { | |
8aec876d | 166 | mddev_ident_t mdlist = conf_get_ident(NULL); |
e0d19036 | 167 | for (; mdlist; mdlist=mdlist->next) { |
fe056d1f N |
168 | struct state *st; |
169 | if (mdlist->devname == NULL) | |
170 | continue; | |
112cace6 N |
171 | if (strcasecmp(mdlist->devname, "<ignore>") == 0) |
172 | continue; | |
fe056d1f | 173 | st = malloc(sizeof *st); |
e0d19036 NB |
174 | if (st == NULL) |
175 | continue; | |
176 | st->devname = strdup(mdlist->devname); | |
177 | st->utime = 0; | |
178 | st->next = statelist; | |
d013a55e | 179 | st->err = 0; |
4450e59f | 180 | st->devnum = INT_MAX; |
e0d19036 | 181 | st->percent = -2; |
feb716e9 | 182 | st->expected_spares = mdlist->spare_disks; |
e0d19036 NB |
183 | if (mdlist->spare_group) |
184 | st->spare_group = strdup(mdlist->spare_group); | |
185 | else | |
186 | st->spare_group = NULL; | |
187 | statelist = st; | |
188 | } | |
189 | } else { | |
cd29a5c8 | 190 | mddev_dev_t dv; |
e0d19036 | 191 | for (dv=devlist ; dv; dv=dv->next) { |
8aec876d | 192 | mddev_ident_t mdlist = conf_get_ident(dv->devname); |
e0d19036 NB |
193 | struct state *st = malloc(sizeof *st); |
194 | if (st == NULL) | |
195 | continue; | |
196 | st->devname = strdup(dv->devname); | |
197 | st->utime = 0; | |
198 | st->next = statelist; | |
d013a55e | 199 | st->err = 0; |
4450e59f | 200 | st->devnum = INT_MAX; |
e0d19036 | 201 | st->percent = -2; |
feb716e9 | 202 | st->expected_spares = -1; |
e0d19036 | 203 | st->spare_group = NULL; |
e5329c37 NB |
204 | if (mdlist) { |
205 | st->expected_spares = mdlist->spare_disks; | |
206 | if (mdlist->spare_group) | |
207 | st->spare_group = strdup(mdlist->spare_group); | |
208 | } | |
e0d19036 NB |
209 | statelist = st; |
210 | } | |
211 | } | |
212 | ||
213 | ||
214 | while (! finished) { | |
aa88f531 | 215 | int new_found = 0; |
e0d19036 NB |
216 | struct state *st; |
217 | ||
218 | if (mdstat) | |
219 | free_mdstat(mdstat); | |
22a88995 | 220 | mdstat = mdstat_read(oneshot?0:1, 0); |
e0d19036 NB |
221 | |
222 | for (st=statelist; st; st=st->next) { | |
ab5303d6 | 223 | struct { int state, major, minor; } info[MaxDisks]; |
52826846 | 224 | mdu_array_info_t array; |
dd0781e5 | 225 | struct mdstat_ent *mse = NULL, *mse2; |
e0d19036 | 226 | char *dev = st->devname; |
52826846 | 227 | int fd; |
98c6faba | 228 | unsigned int i; |
e0d19036 | 229 | |
98c6faba | 230 | if (test) |
4948b8f7 | 231 | alert("TestMessage", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
52826846 NB |
232 | fd = open(dev, O_RDONLY); |
233 | if (fd < 0) { | |
234 | if (!st->err) | |
e0d19036 | 235 | alert("DeviceDisappeared", dev, NULL, |
4948b8f7 | 236 | mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 | 237 | /* fprintf(stderr, Name ": cannot open %s: %s\n", |
52826846 | 238 | dev, strerror(errno)); |
e0d19036 | 239 | */ st->err=1; |
52826846 NB |
240 | continue; |
241 | } | |
e4dc5106 | 242 | fcntl(fd, F_SETFD, FD_CLOEXEC); |
52826846 NB |
243 | if (ioctl(fd, GET_ARRAY_INFO, &array)<0) { |
244 | if (!st->err) | |
e0d19036 | 245 | alert("DeviceDisappeared", dev, NULL, |
4948b8f7 | 246 | mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 | 247 | /* fprintf(stderr, Name ": cannot get array info for %s: %s\n", |
52826846 | 248 | dev, strerror(errno)); |
e0d19036 | 249 | */ st->err=1; |
52826846 NB |
250 | close(fd); |
251 | continue; | |
252 | } | |
66f8bbbe DL |
253 | /* It's much easier to list what array levels can't |
254 | * have a device disappear than all of them that can | |
255 | */ | |
256 | if (array.level == 0 || array.level == -1) { | |
5787fa49 NB |
257 | if (!st->err) |
258 | alert("DeviceDisappeared", dev, "Wrong-Level", | |
4948b8f7 | 259 | mailaddr, mailfrom, alert_cmd, dosyslog); |
5787fa49 NB |
260 | st->err = 1; |
261 | close(fd); | |
262 | continue; | |
263 | } | |
4450e59f | 264 | if (st->devnum == INT_MAX) { |
e0d19036 NB |
265 | struct stat stb; |
266 | if (fstat(fd, &stb) == 0 && | |
98c6faba | 267 | (S_IFMT&stb.st_mode)==S_IFBLK) { |
0df46c2a NB |
268 | if (major(stb.st_rdev) == MD_MAJOR) |
269 | st->devnum = minor(stb.st_rdev); | |
98c6faba | 270 | else |
0df46c2a | 271 | st->devnum = -1- (minor(stb.st_rdev)>>6); |
98c6faba | 272 | } |
e0d19036 NB |
273 | } |
274 | ||
dd0781e5 NB |
275 | for (mse2 = mdstat ; mse2 ; mse2=mse2->next) |
276 | if (mse2->devnum == st->devnum) { | |
4450e59f | 277 | mse2->devnum = INT_MAX; /* flag it as "used" */ |
dd0781e5 NB |
278 | mse = mse2; |
279 | } | |
e0d19036 | 280 | |
52826846 | 281 | if (st->utime == array.utime && |
e0d19036 NB |
282 | st->failed == array.failed_disks && |
283 | st->working == array.working_disks && | |
284 | st->spare == array.spare_disks && | |
285 | (mse == NULL || ( | |
286 | mse->percent == st->percent | |
287 | ))) { | |
52826846 | 288 | close(fd); |
e0d19036 | 289 | st->err = 0; |
52826846 NB |
290 | continue; |
291 | } | |
aa88f531 NB |
292 | if (st->utime == 0 && /* new array */ |
293 | mse && /* is in /proc/mdstat */ | |
294 | mse->pattern && strchr(mse->pattern, '_') /* degraded */ | |
295 | ) | |
4948b8f7 | 296 | alert("DegradedArray", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
aa88f531 | 297 | |
feb716e9 | 298 | if (st->utime == 0 && /* new array */ |
aba69144 NB |
299 | st->expected_spares > 0 && |
300 | array.spare_disks < st->expected_spares) | |
4948b8f7 | 301 | alert("SparesMissing", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 | 302 | if (mse && |
aba69144 | 303 | st->percent == -1 && |
e0d19036 | 304 | mse->percent >= 0) |
4948b8f7 | 305 | alert("RebuildStarted", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 NB |
306 | if (mse && |
307 | st->percent >= 0 && | |
308 | mse->percent >= 0 && | |
309 | (mse->percent / 20) > (st->percent / 20)) | |
310 | alert(percentalerts[mse->percent/20], | |
4948b8f7 | 311 | dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 | 312 | |
98c6faba NB |
313 | if (mse && |
314 | mse->percent == -1 && | |
37dfc3d6 NB |
315 | st->percent >= 0) { |
316 | /* Rebuild/sync/whatever just finished. | |
317 | * If there is a number in /mismatch_cnt, | |
318 | * we should report that. | |
319 | */ | |
7e0f6979 | 320 | struct mdinfo *sra = |
37dfc3d6 NB |
321 | sysfs_read(-1, st->devnum, GET_MISMATCH); |
322 | if (sra && sra->mismatch_cnt > 0) { | |
323 | char cnt[40]; | |
324 | sprintf(cnt, " mismatches found: %d", sra->mismatch_cnt); | |
325 | alert("RebuildFinished", dev, cnt, mailaddr, mailfrom, alert_cmd, dosyslog); | |
326 | } else | |
327 | alert("RebuildFinished", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
328 | if (sra) | |
329 | free(sra); | |
330 | } | |
98c6faba | 331 | |
e0d19036 NB |
332 | if (mse) |
333 | st->percent = mse->percent; | |
e27d562b | 334 | |
ab5303d6 NB |
335 | |
336 | for (i=0; i<MaxDisks && i <= array.raid_disks + array.nr_disks; | |
337 | i++) { | |
338 | mdu_disk_info_t disc; | |
01d9299c | 339 | disc.number = i; |
ab5303d6 NB |
340 | if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) { |
341 | info[i].state = disc.state; | |
342 | info[i].major = disc.major; | |
343 | info[i].minor = disc.minor; | |
344 | } else | |
345 | info[i].major = info[i].minor = 0; | |
346 | } | |
347 | close(fd); | |
348 | ||
e27d562b | 349 | for (i=0; i<MaxDisks; i++) { |
f8409e54 | 350 | mdu_disk_info_t disc = {0}; |
e0d19036 NB |
351 | int newstate=0; |
352 | int change; | |
353 | char *dv = NULL; | |
52826846 | 354 | disc.number = i; |
e27d562b NB |
355 | if (i > array.raid_disks + array.nr_disks) { |
356 | newstate = 0; | |
357 | disc.major = disc.minor = 0; | |
ab5303d6 NB |
358 | } else if (info[i].major || info[i].minor) { |
359 | newstate = info[i].state; | |
360 | dv = map_dev(info[i].major, info[i].minor, 1); | |
361 | disc.state = newstate; | |
362 | disc.major = info[i].major; | |
363 | disc.minor = info[i].minor; | |
e27d562b | 364 | } else if (mse && mse->pattern && i < strlen(mse->pattern)) { |
e0d19036 NB |
365 | switch(mse->pattern[i]) { |
366 | case 'U': newstate = 6 /* ACTIVE/SYNC */; break; | |
367 | case '_': newstate = 0; break; | |
52826846 | 368 | } |
e27d562b NB |
369 | disc.major = disc.minor = 0; |
370 | } | |
fe394e5e NB |
371 | if (dv == NULL && st->devid[i]) |
372 | dv = map_dev(major(st->devid[i]), | |
16c6fa80 | 373 | minor(st->devid[i]), 1); |
e0d19036 NB |
374 | change = newstate ^ st->devstate[i]; |
375 | if (st->utime && change && !st->err) { | |
98c6faba | 376 | if (i < (unsigned)array.raid_disks && |
e0d19036 NB |
377 | (((newstate&change)&(1<<MD_DISK_FAULTY)) || |
378 | ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) || | |
379 | ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))) | |
380 | ) | |
4948b8f7 | 381 | alert("Fail", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog); |
98c6faba | 382 | else if (i >= (unsigned)array.raid_disks && |
e0d19036 | 383 | (disc.major || disc.minor) && |
0df46c2a | 384 | st->devid[i] == makedev(disc.major, disc.minor) && |
e0d19036 NB |
385 | ((newstate&change)&(1<<MD_DISK_FAULTY)) |
386 | ) | |
4948b8f7 | 387 | alert("FailSpare", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog); |
98c6faba | 388 | else if (i < (unsigned)array.raid_disks && |
e0d19036 NB |
389 | (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) || |
390 | ((newstate&change)&(1<<MD_DISK_ACTIVE)) || | |
391 | ((newstate&change)&(1<<MD_DISK_SYNC))) | |
392 | ) | |
4948b8f7 | 393 | alert("SpareActive", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog); |
52826846 | 394 | } |
3e6944b2 | 395 | st->devstate[i] = newstate; |
0df46c2a | 396 | st->devid[i] = makedev(disc.major, disc.minor); |
52826846 | 397 | } |
52826846 NB |
398 | st->active = array.active_disks; |
399 | st->working = array.working_disks; | |
400 | st->spare = array.spare_disks; | |
401 | st->failed = array.failed_disks; | |
402 | st->utime = array.utime; | |
e0d19036 NB |
403 | st->raid = array.raid_disks; |
404 | st->err = 0; | |
52826846 | 405 | } |
e0d19036 NB |
406 | /* now check if there are any new devices found in mdstat */ |
407 | if (scan) { | |
408 | struct mdstat_ent *mse; | |
aba69144 | 409 | for (mse=mdstat; mse; mse=mse->next) |
4450e59f | 410 | if (mse->devnum != INT_MAX && |
2cdb6489 | 411 | mse->level && |
66f8bbbe DL |
412 | (strcmp(mse->level, "raid0")!=0 && |
413 | strcmp(mse->level, "linear")!=0) | |
5787fa49 | 414 | ) { |
e0d19036 | 415 | struct state *st = malloc(sizeof *st); |
98c6faba NB |
416 | mdu_array_info_t array; |
417 | int fd; | |
e0d19036 NB |
418 | if (st == NULL) |
419 | continue; | |
420 | st->devname = strdup(get_md_name(mse->devnum)); | |
98c6faba NB |
421 | if ((fd = open(st->devname, O_RDONLY)) < 0 || |
422 | ioctl(fd, GET_ARRAY_INFO, &array)< 0) { | |
423 | /* no such array */ | |
424 | if (fd >=0) close(fd); | |
8bd2e0c3 | 425 | put_md_name(st->devname); |
98c6faba NB |
426 | free(st->devname); |
427 | free(st); | |
428 | continue; | |
429 | } | |
dd0781e5 | 430 | close(fd); |
e0d19036 NB |
431 | st->utime = 0; |
432 | st->next = statelist; | |
433 | st->err = 1; | |
434 | st->devnum = mse->devnum; | |
435 | st->percent = -2; | |
436 | st->spare_group = NULL; | |
98c6faba | 437 | st->expected_spares = -1; |
e0d19036 | 438 | statelist = st; |
bc854448 N |
439 | if (test) |
440 | alert("TestMessage", st->devname, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); | |
4948b8f7 | 441 | alert("NewArray", st->devname, NULL, mailaddr, mailfrom, alert_cmd, dosyslog); |
aa88f531 | 442 | new_found = 1; |
e0d19036 NB |
443 | } |
444 | } | |
445 | /* If an array has active < raid && spare == 0 && spare_group != NULL | |
446 | * Look for another array with spare > 0 and active == raid and same spare_group | |
447 | * if found, choose a device and hotremove/hotadd | |
448 | */ | |
449 | for (st = statelist; st; st=st->next) | |
450 | if (st->active < st->raid && | |
451 | st->spare == 0 && | |
452 | st->spare_group != NULL) { | |
453 | struct state *st2; | |
454 | for (st2=statelist ; st2 ; st2=st2->next) | |
455 | if (st2 != st && | |
456 | st2->spare > 0 && | |
457 | st2->active == st2->raid && | |
458 | st2->spare_group != NULL && | |
459 | strcmp(st->spare_group, st2->spare_group) == 0) { | |
460 | /* try to remove and add */ | |
461 | int fd1 = open(st->devname, O_RDONLY); | |
462 | int fd2 = open(st2->devname, O_RDONLY); | |
463 | int dev = -1; | |
464 | int d; | |
465 | if (fd1 < 0 || fd2 < 0) { | |
466 | if (fd1>=0) close(fd1); | |
467 | if (fd2>=0) close(fd2); | |
468 | continue; | |
469 | } | |
e27d562b | 470 | for (d=st2->raid; d < MaxDisks; d++) { |
e0d19036 NB |
471 | if (st2->devid[d] > 0 && |
472 | st2->devstate[d] == 0) { | |
473 | dev = st2->devid[d]; | |
474 | break; | |
475 | } | |
476 | } | |
477 | if (dev > 0) { | |
aba69144 | 478 | if (ioctl(fd2, HOT_REMOVE_DISK, |
e0d19036 NB |
479 | (unsigned long)dev) == 0) { |
480 | if (ioctl(fd1, HOT_ADD_DISK, | |
481 | (unsigned long)dev) == 0) { | |
4948b8f7 | 482 | alert("MoveSpare", st->devname, st2->devname, mailaddr, mailfrom, alert_cmd, dosyslog); |
e0d19036 NB |
483 | close(fd1); |
484 | close(fd2); | |
485 | break; | |
486 | } | |
487 | else ioctl(fd2, HOT_ADD_DISK, (unsigned long) dev); | |
488 | } | |
489 | } | |
490 | close(fd1); | |
491 | close(fd2); | |
492 | } | |
493 | } | |
aa88f531 NB |
494 | if (!new_found) { |
495 | if (oneshot) | |
496 | break; | |
497 | else | |
dd0781e5 | 498 | mdstat_wait(period); |
aa88f531 | 499 | } |
98c6faba | 500 | test = 0; |
52826846 | 501 | } |
b5e64645 NB |
502 | if (pidfile) |
503 | unlink(pidfile); | |
52826846 NB |
504 | return 0; |
505 | } | |
506 | ||
507 | ||
4948b8f7 | 508 | static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mailfrom, char *cmd, |
773135f5 | 509 | int dosyslog) |
52826846 | 510 | { |
773135f5 NB |
511 | int priority; |
512 | ||
cd29a5c8 NB |
513 | if (!cmd && !mailaddr) { |
514 | time_t now = time(0); | |
aba69144 | 515 | |
e0d19036 | 516 | printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device"); |
cd29a5c8 | 517 | } |
52826846 NB |
518 | if (cmd) { |
519 | int pid = fork(); | |
520 | switch(pid) { | |
521 | default: | |
522 | waitpid(pid, NULL, 0); | |
523 | break; | |
524 | case -1: | |
525 | break; | |
526 | case 0: | |
527 | execl(cmd, cmd, event, dev, disc, NULL); | |
528 | exit(2); | |
529 | } | |
530 | } | |
aba69144 NB |
531 | if (mailaddr && |
532 | (strncmp(event, "Fail", 4)==0 || | |
98c6faba | 533 | strncmp(event, "Test", 4)==0 || |
d1732eeb | 534 | strncmp(event, "Spares", 6)==0 || |
aa88f531 | 535 | strncmp(event, "Degrade", 7)==0)) { |
52826846 NB |
536 | FILE *mp = popen(Sendmail, "w"); |
537 | if (mp) { | |
008e1100 | 538 | FILE *mdstat; |
52826846 NB |
539 | char hname[256]; |
540 | gethostname(hname, sizeof(hname)); | |
541 | signal(SIGPIPE, SIG_IGN); | |
4948b8f7 NB |
542 | if (mailfrom) |
543 | fprintf(mp, "From: %s\n", mailfrom); | |
544 | else | |
545 | fprintf(mp, "From: " Name " monitoring <root>\n"); | |
52826846 NB |
546 | fprintf(mp, "To: %s\n", mailaddr); |
547 | fprintf(mp, "Subject: %s event on %s:%s\n\n", event, dev, hname); | |
548 | ||
549 | fprintf(mp, "This is an automatically generated mail message from " Name "\n"); | |
550 | fprintf(mp, "running on %s\n\n", hname); | |
551 | ||
552 | fprintf(mp, "A %s event had been detected on md device %s.\n\n", event, dev); | |
553 | ||
37dfc3d6 | 554 | if (disc && disc[0] != ' ') |
56eedc1a | 555 | fprintf(mp, "It could be related to component device %s.\n\n", disc); |
37dfc3d6 NB |
556 | if (disc && disc[0] == ' ') |
557 | fprintf(mp, "Extra information:%s.\n\n", disc); | |
52826846 NB |
558 | |
559 | fprintf(mp, "Faithfully yours, etc.\n"); | |
008e1100 NB |
560 | |
561 | mdstat = fopen("/proc/mdstat", "r"); | |
562 | if (mdstat) { | |
563 | char buf[8192]; | |
564 | int n; | |
a524a7ee | 565 | fprintf(mp, "\nP.S. The /proc/mdstat file currently contains the following:\n\n"); |
008e1100 | 566 | while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0) |
9fca7d62 | 567 | n=fwrite(buf, 1, n, mp); /* yes, i don't care about the result */ |
008e1100 NB |
568 | fclose(mdstat); |
569 | } | |
52826846 NB |
570 | fclose(mp); |
571 | } | |
572 | ||
573 | } | |
773135f5 NB |
574 | |
575 | /* log the event to syslog maybe */ | |
576 | if (dosyslog) { | |
577 | /* Log at a different severity depending on the event. | |
578 | * | |
579 | * These are the critical events: */ | |
580 | if (strncmp(event, "Fail", 4)==0 || | |
581 | strncmp(event, "Degrade", 7)==0 || | |
582 | strncmp(event, "DeviceDisappeared", 17)==0) | |
583 | priority = LOG_CRIT; | |
584 | /* Good to know about, but are not failures: */ | |
585 | else if (strncmp(event, "Rebuild", 7)==0 || | |
d1732eeb NB |
586 | strncmp(event, "MoveSpare", 9)==0 || |
587 | strncmp(event, "Spares", 6) != 0) | |
773135f5 NB |
588 | priority = LOG_WARNING; |
589 | /* Everything else: */ | |
590 | else | |
591 | priority = LOG_INFO; | |
592 | ||
593 | if (disc) | |
594 | syslog(priority, "%s event detected on md device %s, component device %s", event, dev, disc); | |
595 | else | |
596 | syslog(priority, "%s event detected on md device %s", event, dev); | |
597 | } | |
52826846 | 598 | } |
b90c0e9a NB |
599 | |
600 | /* Not really Monitor but ... */ | |
601 | int Wait(char *dev) | |
602 | { | |
603 | struct stat stb; | |
604 | int devnum; | |
605 | int rv = 1; | |
606 | ||
607 | if (stat(dev, &stb) != 0) { | |
608 | fprintf(stderr, Name ": Cannot find %s: %s\n", dev, | |
609 | strerror(errno)); | |
610 | return 2; | |
611 | } | |
c94709e8 | 612 | devnum = stat2devnum(&stb); |
b90c0e9a NB |
613 | |
614 | while(1) { | |
615 | struct mdstat_ent *ms = mdstat_read(1, 0); | |
616 | struct mdstat_ent *e; | |
617 | ||
618 | for (e=ms ; e; e=e->next) | |
619 | if (e->devnum == devnum) | |
620 | break; | |
621 | ||
622 | if (!e || e->percent < 0) { | |
e7783ee6 | 623 | if (e && e->metadata_version && |
c94709e8 DW |
624 | strncmp(e->metadata_version, "external:", 9) == 0) { |
625 | if (is_subarray(&e->metadata_version[9])) | |
626 | ping_monitor(&e->metadata_version[9]); | |
627 | else | |
628 | ping_monitor(devnum2devname(devnum)); | |
629 | } | |
b90c0e9a NB |
630 | free_mdstat(ms); |
631 | return rv; | |
632 | } | |
89a10d84 | 633 | free_mdstat(ms); |
b90c0e9a NB |
634 | rv = 0; |
635 | mdstat_wait(5); | |
636 | } | |
637 | } | |
1770662b DW |
638 | |
639 | static char *clean_states[] = { | |
640 | "clear", "inactive", "readonly", "read-auto", "clean", NULL }; | |
641 | ||
27dec8fa | 642 | int WaitClean(char *dev, int verbose) |
1770662b DW |
643 | { |
644 | int fd; | |
645 | struct mdinfo *mdi; | |
646 | int rv = 1; | |
647 | int devnum; | |
648 | ||
649 | fd = open(dev, O_RDONLY); | |
650 | if (fd < 0) { | |
27dec8fa DW |
651 | if (verbose) |
652 | fprintf(stderr, Name ": Couldn't open %s: %s\n", dev, strerror(errno)); | |
1770662b DW |
653 | return 1; |
654 | } | |
655 | ||
656 | devnum = fd2devnum(fd); | |
657 | mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE); | |
658 | if (!mdi) { | |
27dec8fa DW |
659 | if (verbose) |
660 | fprintf(stderr, Name ": Failed to read sysfs attributes for " | |
661 | "%s\n", dev); | |
1770662b DW |
662 | close(fd); |
663 | return 0; | |
664 | } | |
665 | ||
666 | switch(mdi->array.level) { | |
667 | case LEVEL_LINEAR: | |
668 | case LEVEL_MULTIPATH: | |
669 | case 0: | |
670 | /* safemode delay is irrelevant for these levels */ | |
671 | rv = 0; | |
672 | ||
673 | } | |
674 | ||
675 | /* for internal metadata the kernel handles the final clean | |
676 | * transition, containers can never be dirty | |
677 | */ | |
678 | if (!is_subarray(mdi->text_version)) | |
679 | rv = 0; | |
680 | ||
681 | /* safemode disabled ? */ | |
682 | if (mdi->safe_mode_delay == 0) | |
683 | rv = 0; | |
684 | ||
685 | if (rv) { | |
686 | int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state"); | |
1770662b | 687 | char buf[20]; |
0dd3ba30 DW |
688 | fd_set fds; |
689 | struct timeval tm; | |
1770662b | 690 | |
0dd3ba30 DW |
691 | /* minimize the safe_mode_delay and prepare to wait up to 5s |
692 | * for writes to quiesce | |
693 | */ | |
694 | sysfs_set_safemode(mdi, 1); | |
695 | tm.tv_sec = 5; | |
696 | tm.tv_usec = 0; | |
697 | ||
7146ec6a DW |
698 | /* give mdmon a chance to checkpoint resync */ |
699 | sysfs_set_str(mdi, NULL, "sync_action", "idle"); | |
700 | ||
0dd3ba30 | 701 | FD_ZERO(&fds); |
1770662b | 702 | |
0dd3ba30 DW |
703 | /* wait for array_state to be clean */ |
704 | while (1) { | |
1770662b DW |
705 | rv = read(state_fd, buf, sizeof(buf)); |
706 | if (rv < 0) | |
707 | break; | |
708 | if (sysfs_match_word(buf, clean_states) <= 4) | |
709 | break; | |
0dd3ba30 | 710 | FD_SET(state_fd, &fds); |
28005287 | 711 | rv = select(state_fd + 1, NULL, NULL, &fds, &tm); |
0dd3ba30 DW |
712 | if (rv < 0 && errno != EINTR) |
713 | break; | |
1770662b DW |
714 | lseek(state_fd, 0, SEEK_SET); |
715 | } | |
716 | if (rv < 0) | |
717 | rv = 1; | |
0dd3ba30 | 718 | else if (ping_monitor(mdi->text_version) == 0) { |
1770662b DW |
719 | /* we need to ping to close the window between array |
720 | * state transitioning to clean and the metadata being | |
721 | * marked clean | |
722 | */ | |
0dd3ba30 DW |
723 | rv = 0; |
724 | } else | |
725 | rv = 1; | |
27dec8fa | 726 | if (rv && verbose) |
1770662b DW |
727 | fprintf(stderr, Name ": Error waiting for %s to be clean\n", |
728 | dev); | |
729 | ||
0dd3ba30 DW |
730 | /* restore the original safe_mode_delay */ |
731 | sysfs_set_safemode(mdi, mdi->safe_mode_delay); | |
1770662b DW |
732 | close(state_fd); |
733 | } | |
734 | ||
0dd3ba30 | 735 | sysfs_free(mdi); |
1770662b DW |
736 | close(fd); |
737 | ||
738 | return rv; | |
739 | } | |
740 | ||
741 |