]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdmon - monitor external metadata arrays | |
3 | * | |
4 | * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de> | |
5 | * Copyright (C) 2007-2009 Intel Corporation | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify it | |
8 | * under the terms and conditions of the GNU General Public License, | |
9 | * version 2, as published by the Free Software Foundation. | |
10 | * | |
11 | * This program is distributed in the hope it will be useful, but WITHOUT | |
12 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
13 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
14 | * more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License along with | |
17 | * this program; if not, write to the Free Software Foundation, Inc., | |
18 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
19 | */ | |
20 | ||
21 | /* | |
22 | * md array manager. | |
23 | * When md arrays have user-space managed metadata, this is the program | |
24 | * that does the managing. | |
25 | * | |
26 | * Given one argument: the name of the array (e.g. /dev/md0) that is | |
27 | * the container. | |
28 | * We fork off a helper that runs high priority and mlocked. It responds to | |
29 | * device failures and other events that might stop writeout, or that are | |
30 | * trivial to deal with. | |
31 | * The main thread then watches for new arrays being created in the container | |
32 | * and starts monitoring them too ... along with a few other tasks. | |
33 | * | |
34 | * The main thread communicates with the priority thread by writing over | |
35 | * a pipe. | |
36 | * Separate programs can communicate with the main thread via Unix-domain | |
37 | * socket. | |
38 | * The two threads share address space and open file table. | |
39 | * | |
40 | */ | |
41 | ||
42 | #ifndef _GNU_SOURCE | |
43 | #define _GNU_SOURCE | |
44 | #endif | |
45 | ||
46 | #include <unistd.h> | |
47 | #include <stdlib.h> | |
48 | #include <sys/types.h> | |
49 | #include <sys/stat.h> | |
50 | #include <sys/socket.h> | |
51 | #include <sys/un.h> | |
52 | #include <sys/mman.h> | |
53 | #include <sys/syscall.h> | |
54 | #include <sys/wait.h> | |
55 | #include <stdio.h> | |
56 | #include <errno.h> | |
57 | #include <string.h> | |
58 | #include <fcntl.h> | |
59 | #include <signal.h> | |
60 | #include <dirent.h> | |
61 | ||
62 | #include <sched.h> | |
63 | ||
64 | #include "mdadm.h" | |
65 | #include "mdmon.h" | |
66 | ||
67 | struct active_array *discard_this; | |
68 | struct active_array *pending_discard; | |
69 | ||
70 | int mon_tid, mgr_tid; | |
71 | ||
72 | int sigterm; | |
73 | ||
74 | int run_child(void *v) | |
75 | { | |
76 | struct supertype *c = v; | |
77 | ||
78 | do_monitor(c); | |
79 | return 0; | |
80 | } | |
81 | ||
82 | #ifdef __ia64__ | |
83 | int __clone2(int (*fn)(void *), | |
84 | void *child_stack_base, size_t stack_size, | |
85 | int flags, void *arg, ... | |
86 | /* pid_t *pid, struct user_desc *tls, pid_t *ctid */ ); | |
87 | #endif | |
88 | int clone_monitor(struct supertype *container) | |
89 | { | |
90 | static char stack[4096]; | |
91 | ||
92 | #ifdef __ia64__ | |
93 | mon_tid = __clone2(run_child, stack, sizeof(stack), | |
94 | CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD, | |
95 | container); | |
96 | #else | |
97 | mon_tid = clone(run_child, stack+4096-64, | |
98 | CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD, | |
99 | container); | |
100 | #endif | |
101 | ||
102 | mgr_tid = syscall(SYS_gettid); | |
103 | ||
104 | return mon_tid; | |
105 | } | |
106 | ||
107 | static struct superswitch *find_metadata_methods(char *vers) | |
108 | { | |
109 | if (strcmp(vers, "ddf") == 0) | |
110 | return &super_ddf; | |
111 | if (strcmp(vers, "imsm") == 0) | |
112 | return &super_imsm; | |
113 | return NULL; | |
114 | } | |
115 | ||
116 | int make_pidfile(char *devname, int o_excl) | |
117 | { | |
118 | char path[100]; | |
119 | char pid[10]; | |
120 | int fd; | |
121 | int n; | |
122 | ||
123 | if (sigterm) | |
124 | return -1; | |
125 | ||
126 | sprintf(path, "/var/run/mdadm/%s.pid", devname); | |
127 | ||
128 | fd = open(path, O_RDWR|O_CREAT|o_excl, 0600); | |
129 | if (fd < 0) | |
130 | return -errno; | |
131 | sprintf(pid, "%d\n", getpid()); | |
132 | n = write(fd, pid, strlen(pid)); | |
133 | close(fd); | |
134 | if (n < 0) | |
135 | return -errno; | |
136 | return 0; | |
137 | } | |
138 | ||
139 | int is_container_member(struct mdstat_ent *mdstat, char *container) | |
140 | { | |
141 | if (mdstat->metadata_version == NULL || | |
142 | strncmp(mdstat->metadata_version, "external:", 9) != 0 || | |
143 | !is_subarray(mdstat->metadata_version+9) || | |
144 | strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 || | |
145 | mdstat->metadata_version[10+strlen(container)] != '/') | |
146 | return 0; | |
147 | ||
148 | return 1; | |
149 | } | |
150 | ||
151 | static void try_kill_monitor(pid_t pid, char *devname, int sock) | |
152 | { | |
153 | char buf[100]; | |
154 | int fd; | |
155 | int n; | |
156 | long fl; | |
157 | ||
158 | /* first rule of survival... don't off yourself */ | |
159 | if (pid == getpid()) | |
160 | return; | |
161 | ||
162 | /* kill this process if it is mdmon */ | |
163 | sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid); | |
164 | fd = open(buf, O_RDONLY); | |
165 | if (fd < 0) | |
166 | return; | |
167 | ||
168 | n = read(fd, buf, sizeof(buf)-1); | |
169 | buf[sizeof(buf)-1] = 0; | |
170 | close(fd); | |
171 | ||
172 | if (n < 0 || !strstr(buf, "mdmon")) | |
173 | return; | |
174 | ||
175 | kill(pid, SIGTERM); | |
176 | ||
177 | /* Wait for monitor to exit by reading from the socket, after | |
178 | * clearing the non-blocking flag */ | |
179 | fl = fcntl(sock, F_GETFL, 0); | |
180 | fl &= ~O_NONBLOCK; | |
181 | fcntl(sock, F_SETFL, fl); | |
182 | read(sock, buf, 100); | |
183 | } | |
184 | ||
185 | void remove_pidfile(char *devname) | |
186 | { | |
187 | char buf[100]; | |
188 | ||
189 | if (sigterm) | |
190 | return; | |
191 | ||
192 | sprintf(buf, "/var/run/mdadm/%s.pid", devname); | |
193 | unlink(buf); | |
194 | sprintf(buf, "/var/run/mdadm/%s.sock", devname); | |
195 | unlink(buf); | |
196 | } | |
197 | ||
198 | int make_control_sock(char *devname) | |
199 | { | |
200 | char path[100]; | |
201 | int sfd; | |
202 | long fl; | |
203 | struct sockaddr_un addr; | |
204 | ||
205 | if (sigterm) | |
206 | return -1; | |
207 | ||
208 | sprintf(path, "/var/run/mdadm/%s.sock", devname); | |
209 | unlink(path); | |
210 | sfd = socket(PF_LOCAL, SOCK_STREAM, 0); | |
211 | if (sfd < 0) | |
212 | return -1; | |
213 | ||
214 | addr.sun_family = PF_LOCAL; | |
215 | strcpy(addr.sun_path, path); | |
216 | if (bind(sfd, &addr, sizeof(addr)) < 0) { | |
217 | close(sfd); | |
218 | return -1; | |
219 | } | |
220 | listen(sfd, 10); | |
221 | fl = fcntl(sfd, F_GETFL, 0); | |
222 | fl |= O_NONBLOCK; | |
223 | fcntl(sfd, F_SETFL, fl); | |
224 | return sfd; | |
225 | } | |
226 | ||
227 | int socket_hup_requested; | |
228 | static void hup(int sig) | |
229 | { | |
230 | socket_hup_requested = 1; | |
231 | } | |
232 | ||
233 | static void term(int sig) | |
234 | { | |
235 | sigterm = 1; | |
236 | } | |
237 | ||
238 | static void wake_me(int sig) | |
239 | { | |
240 | ||
241 | } | |
242 | ||
243 | /* if we are debugging and starting mdmon by hand then don't fork */ | |
244 | static int do_fork(void) | |
245 | { | |
246 | #ifdef DEBUG | |
247 | if (check_env("MDADM_NO_MDMON")) | |
248 | return 0; | |
249 | #endif | |
250 | ||
251 | return 1; | |
252 | } | |
253 | ||
254 | void usage(void) | |
255 | { | |
256 | fprintf(stderr, "Usage: mdmon /device/name/for/container [target_dir]\n"); | |
257 | exit(2); | |
258 | } | |
259 | ||
260 | static int mdmon(char *devname, int devnum, int must_fork, char *switchroot); | |
261 | ||
262 | int main(int argc, char *argv[]) | |
263 | { | |
264 | char *container_name = NULL; | |
265 | char *switchroot = NULL; | |
266 | int devnum; | |
267 | char *devname; | |
268 | int status = 0; | |
269 | ||
270 | switch (argc) { | |
271 | case 3: | |
272 | switchroot = argv[2]; | |
273 | case 2: | |
274 | container_name = argv[1]; | |
275 | break; | |
276 | default: | |
277 | usage(); | |
278 | } | |
279 | ||
280 | if (strcmp(container_name, "/proc/mdstat") == 0) { | |
281 | struct mdstat_ent *mdstat, *e; | |
282 | ||
283 | /* launch an mdmon instance for each container found */ | |
284 | mdstat = mdstat_read(0, 0); | |
285 | for (e = mdstat; e; e = e->next) { | |
286 | if (strncmp(e->metadata_version, "external:", 9) == 0 && | |
287 | !is_subarray(&e->metadata_version[9])) { | |
288 | devname = devnum2devname(e->devnum); | |
289 | /* update cmdline so this mdmon instance can be | |
290 | * distinguished from others in a call to ps(1) | |
291 | */ | |
292 | if (strlen(devname) <= strlen(container_name)) { | |
293 | memset(container_name, 0, strlen(container_name)); | |
294 | sprintf(container_name, "%s", devname); | |
295 | } | |
296 | status |= mdmon(devname, e->devnum, 1, | |
297 | switchroot); | |
298 | } | |
299 | } | |
300 | free_mdstat(mdstat); | |
301 | ||
302 | return status; | |
303 | } else if (strncmp(container_name, "md", 2) == 0) { | |
304 | devnum = devname2devnum(container_name); | |
305 | devname = devnum2devname(devnum); | |
306 | if (strcmp(container_name, devname) != 0) | |
307 | devname = NULL; | |
308 | } else { | |
309 | struct stat st; | |
310 | ||
311 | devnum = NoMdDev; | |
312 | if (stat(container_name, &st) == 0) | |
313 | devnum = stat2devnum(&st); | |
314 | if (devnum == NoMdDev) | |
315 | devname = NULL; | |
316 | else | |
317 | devname = devnum2devname(devnum); | |
318 | } | |
319 | ||
320 | if (!devname) { | |
321 | fprintf(stderr, "mdmon: %s is not a valid md device name\n", | |
322 | container_name); | |
323 | exit(1); | |
324 | } | |
325 | return mdmon(devname, devnum, do_fork(), switchroot); | |
326 | } | |
327 | ||
328 | static int mdmon(char *devname, int devnum, int must_fork, char *switchroot) | |
329 | { | |
330 | int mdfd; | |
331 | struct mdinfo *mdi, *di; | |
332 | struct supertype *container; | |
333 | sigset_t set; | |
334 | struct sigaction act; | |
335 | int pfd[2]; | |
336 | int status; | |
337 | int ignore; | |
338 | pid_t victim = -1; | |
339 | int victim_sock = -1; | |
340 | ||
341 | dprintf("starting mdmon for %s in %s\n", | |
342 | devname, switchroot ? : "/"); | |
343 | ||
344 | /* switchroot is either a path name starting with '/', or a | |
345 | * pid of the original mdmon (we have already done the chroot). | |
346 | * In the latter case, stdin is a socket connected to the original | |
347 | * mdmon. | |
348 | */ | |
349 | ||
350 | /* try to spawn mdmon instances from the target file system */ | |
351 | if (switchroot && switchroot[0] == '/' && | |
352 | strcmp(switchroot, "/") != 0) { | |
353 | pid_t pid; | |
354 | char buf[20]; | |
355 | ||
356 | switch (fork()) { | |
357 | case 0: | |
358 | victim = mdmon_pid(devnum); | |
359 | victim_sock = connect_monitor(devname); | |
360 | if (chroot(switchroot) != 0) { | |
361 | fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n", | |
362 | switchroot, strerror(errno)); | |
363 | exit(4); | |
364 | } | |
365 | ignore = chdir("/"); | |
366 | sprintf(buf, "%d", victim); | |
367 | if (victim_sock) { | |
368 | close(0); | |
369 | dup(victim_sock); | |
370 | close(victim_sock); | |
371 | } | |
372 | execl("/sbin/mdmon", "mdmon", devname, buf, NULL); | |
373 | exit(1); | |
374 | case -1: | |
375 | return 1; | |
376 | default: | |
377 | pid = wait(&status); | |
378 | if (pid > -1 && WIFEXITED(status) && | |
379 | WEXITSTATUS(status) == 0) | |
380 | return 0; | |
381 | else | |
382 | return 1; | |
383 | } | |
384 | } | |
385 | ||
386 | mdfd = open_dev(devnum); | |
387 | if (mdfd < 0) { | |
388 | fprintf(stderr, "mdmon: %s: %s\n", devname, | |
389 | strerror(errno)); | |
390 | return 1; | |
391 | } | |
392 | if (md_get_version(mdfd) < 0) { | |
393 | fprintf(stderr, "mdmon: %s: Not an md device\n", | |
394 | devname); | |
395 | return 1; | |
396 | } | |
397 | ||
398 | /* Fork, and have the child tell us when they are ready */ | |
399 | if (must_fork) { | |
400 | if (pipe(pfd) != 0) { | |
401 | fprintf(stderr, "mdmon: failed to create pipe\n"); | |
402 | return 1; | |
403 | } | |
404 | switch(fork()) { | |
405 | case -1: | |
406 | fprintf(stderr, "mdmon: failed to fork: %s\n", | |
407 | strerror(errno)); | |
408 | return 1; | |
409 | case 0: /* child */ | |
410 | close(pfd[0]); | |
411 | break; | |
412 | default: /* parent */ | |
413 | close(pfd[1]); | |
414 | if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) { | |
415 | wait(&status); | |
416 | status = WEXITSTATUS(status); | |
417 | } | |
418 | return status; | |
419 | } | |
420 | } else | |
421 | pfd[0] = pfd[1] = -1; | |
422 | ||
423 | container = calloc(1, sizeof(*container)); | |
424 | container->devnum = devnum; | |
425 | container->devname = devname; | |
426 | container->arrays = NULL; | |
427 | container->subarray[0] = 0; | |
428 | container->sock = -1; | |
429 | ||
430 | if (!container->devname) { | |
431 | fprintf(stderr, "mdmon: failed to allocate container name string\n"); | |
432 | exit(3); | |
433 | } | |
434 | ||
435 | mdi = sysfs_read(mdfd, container->devnum, | |
436 | GET_VERSION|GET_LEVEL|GET_DEVS|SKIP_GONE_DEVS); | |
437 | ||
438 | if (!mdi) { | |
439 | fprintf(stderr, "mdmon: failed to load sysfs info for %s\n", | |
440 | container->devname); | |
441 | exit(3); | |
442 | } | |
443 | if (mdi->array.level != UnSet) { | |
444 | fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n", | |
445 | devname); | |
446 | exit(3); | |
447 | } | |
448 | if (mdi->array.major_version != -1 || | |
449 | mdi->array.minor_version != -2) { | |
450 | fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n", | |
451 | devname); | |
452 | exit(3); | |
453 | } | |
454 | ||
455 | container->ss = find_metadata_methods(mdi->text_version); | |
456 | if (container->ss == NULL) { | |
457 | fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n", | |
458 | devname, mdi->text_version); | |
459 | exit(3); | |
460 | } | |
461 | ||
462 | container->devs = NULL; | |
463 | for (di = mdi->devs; di; di = di->next) { | |
464 | struct mdinfo *cd = malloc(sizeof(*cd)); | |
465 | *cd = *di; | |
466 | cd->next = container->devs; | |
467 | container->devs = cd; | |
468 | } | |
469 | sysfs_free(mdi); | |
470 | ||
471 | /* SIGUSR is sent between parent and child. So both block it | |
472 | * and enable it only with pselect. | |
473 | */ | |
474 | sigemptyset(&set); | |
475 | sigaddset(&set, SIGUSR1); | |
476 | sigaddset(&set, SIGHUP); | |
477 | sigaddset(&set, SIGALRM); | |
478 | sigaddset(&set, SIGTERM); | |
479 | sigprocmask(SIG_BLOCK, &set, NULL); | |
480 | act.sa_handler = wake_me; | |
481 | act.sa_flags = 0; | |
482 | sigaction(SIGUSR1, &act, NULL); | |
483 | sigaction(SIGALRM, &act, NULL); | |
484 | act.sa_handler = hup; | |
485 | sigaction(SIGHUP, &act, NULL); | |
486 | act.sa_handler = term; | |
487 | sigaction(SIGTERM, &act, NULL); | |
488 | act.sa_handler = SIG_IGN; | |
489 | sigaction(SIGPIPE, &act, NULL); | |
490 | ||
491 | if (switchroot) { | |
492 | /* we assume we assume that /sys /proc /dev are available in | |
493 | * the new root | |
494 | */ | |
495 | if (switchroot[0] == '/') { | |
496 | victim = mdmon_pid(container->devnum); | |
497 | victim_sock = connect_monitor(container->devname); | |
498 | } else { | |
499 | victim = atoi(switchroot); | |
500 | victim_sock = 0; | |
501 | } | |
502 | } | |
503 | ||
504 | ignore = chdir("/"); | |
505 | if (victim < 0) { | |
506 | if (ping_monitor(container->devname) == 0) { | |
507 | fprintf(stderr, "mdmon: %s already managed\n", | |
508 | container->devname); | |
509 | exit(3); | |
510 | } | |
511 | /* if there is a pid file, kill whoever is there just in case */ | |
512 | victim = mdmon_pid(container->devnum); | |
513 | } | |
514 | if (container->ss->load_super(container, mdfd, devname)) { | |
515 | fprintf(stderr, "mdmon: Cannot load metadata for %s\n", | |
516 | devname); | |
517 | exit(3); | |
518 | } | |
519 | close(mdfd); | |
520 | ||
521 | /* Ok, this is close enough. We can say goodbye to our parent now. | |
522 | */ | |
523 | status = 0; | |
524 | if (write(pfd[1], &status, sizeof(status)) < 0) | |
525 | fprintf(stderr, "mdmon: failed to notify our parent: %d\n", | |
526 | getppid()); | |
527 | close(pfd[1]); | |
528 | ||
529 | mlockall(MCL_CURRENT | MCL_FUTURE); | |
530 | ||
531 | if (clone_monitor(container) < 0) { | |
532 | fprintf(stderr, "mdmon: failed to start monitor process: %s\n", | |
533 | strerror(errno)); | |
534 | exit(2); | |
535 | } | |
536 | ||
537 | if (victim > -1) { | |
538 | try_kill_monitor(victim, container->devname, victim_sock); | |
539 | close(victim_sock); | |
540 | } | |
541 | ||
542 | setsid(); | |
543 | close(0); | |
544 | open("/dev/null", O_RDWR); | |
545 | close(1); | |
546 | ignore = dup(0); | |
547 | #ifndef DEBUG | |
548 | close(2); | |
549 | ignore = dup(0); | |
550 | #endif | |
551 | ||
552 | do_manager(container); | |
553 | ||
554 | exit(0); | |
555 | } |