]>
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 | static int make_pidfile(char *devname) | |
117 | { | |
118 | char path[100]; | |
119 | char pid[10]; | |
120 | int fd; | |
121 | int n; | |
122 | ||
123 | if (mkdir(pid_dir, 0600) < 0 && | |
124 | errno != EEXIST) | |
125 | return -errno; | |
126 | sprintf(path, "%s/%s.pid", pid_dir, 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 | n = read(sock, buf, 100); | |
183 | /* Ignore result, it is just the wait that | |
184 | * matters | |
185 | */ | |
186 | } | |
187 | ||
188 | void remove_pidfile(char *devname) | |
189 | { | |
190 | char buf[100]; | |
191 | ||
192 | sprintf(buf, "%s/%s.pid", pid_dir, devname); | |
193 | unlink(buf); | |
194 | sprintf(buf, "%s/%s.sock", pid_dir, devname); | |
195 | unlink(buf); | |
196 | if (strcmp(pid_dir, ALT_RUN) == 0) | |
197 | /* try to clean up when we are finished with this dir */ | |
198 | rmdir(pid_dir); | |
199 | } | |
200 | ||
201 | static int make_control_sock(char *devname) | |
202 | { | |
203 | char path[100]; | |
204 | int sfd; | |
205 | long fl; | |
206 | struct sockaddr_un addr; | |
207 | ||
208 | if (sigterm) | |
209 | return -1; | |
210 | ||
211 | sprintf(path, "%s/%s.sock", pid_dir, devname); | |
212 | unlink(path); | |
213 | sfd = socket(PF_LOCAL, SOCK_STREAM, 0); | |
214 | if (sfd < 0) | |
215 | return -1; | |
216 | ||
217 | addr.sun_family = PF_LOCAL; | |
218 | strcpy(addr.sun_path, path); | |
219 | if (bind(sfd, &addr, sizeof(addr)) < 0) { | |
220 | close(sfd); | |
221 | return -1; | |
222 | } | |
223 | listen(sfd, 10); | |
224 | fl = fcntl(sfd, F_GETFL, 0); | |
225 | fl |= O_NONBLOCK; | |
226 | fcntl(sfd, F_SETFL, fl); | |
227 | return sfd; | |
228 | } | |
229 | ||
230 | static void term(int sig) | |
231 | { | |
232 | sigterm = 1; | |
233 | } | |
234 | ||
235 | static void wake_me(int sig) | |
236 | { | |
237 | ||
238 | } | |
239 | ||
240 | /* if we are debugging and starting mdmon by hand then don't fork */ | |
241 | static int do_fork(void) | |
242 | { | |
243 | #ifdef DEBUG | |
244 | if (check_env("MDADM_NO_MDMON")) | |
245 | return 0; | |
246 | #endif | |
247 | ||
248 | return 1; | |
249 | } | |
250 | ||
251 | void usage(void) | |
252 | { | |
253 | fprintf(stderr, "Usage: mdmon [--all] [--takeover] CONTAINER\n"); | |
254 | exit(2); | |
255 | } | |
256 | ||
257 | static int mdmon(char *devname, int devnum, int must_fork, int takeover); | |
258 | ||
259 | int main(int argc, char *argv[]) | |
260 | { | |
261 | char *container_name = NULL; | |
262 | int devnum; | |
263 | char *devname; | |
264 | int status = 0; | |
265 | int arg; | |
266 | int all = 0; | |
267 | int takeover = 0; | |
268 | ||
269 | for (arg = 1; arg < argc; arg++) { | |
270 | if (strncmp(argv[arg], "--all",5) == 0 || | |
271 | strcmp(argv[arg], "/proc/mdstat") == 0) { | |
272 | container_name = argv[arg]; | |
273 | all = 1; | |
274 | } else if (strcmp(argv[arg], "--takeover") == 0) | |
275 | takeover = 1; | |
276 | else if (container_name == NULL) | |
277 | container_name = argv[arg]; | |
278 | else | |
279 | usage(); | |
280 | } | |
281 | if (container_name == NULL) | |
282 | usage(); | |
283 | ||
284 | if (all) { | |
285 | struct mdstat_ent *mdstat, *e; | |
286 | int container_len = strlen(container_name); | |
287 | ||
288 | /* launch an mdmon instance for each container found */ | |
289 | mdstat = mdstat_read(0, 0); | |
290 | for (e = mdstat; e; e = e->next) { | |
291 | if (strncmp(e->metadata_version, "external:", 9) == 0 && | |
292 | !is_subarray(&e->metadata_version[9])) { | |
293 | devname = devnum2devname(e->devnum); | |
294 | /* update cmdline so this mdmon instance can be | |
295 | * distinguished from others in a call to ps(1) | |
296 | */ | |
297 | if (strlen(devname) <= container_len) { | |
298 | memset(container_name, 0, container_len); | |
299 | sprintf(container_name, "%s", devname); | |
300 | } | |
301 | status |= mdmon(devname, e->devnum, 1, | |
302 | takeover); | |
303 | } | |
304 | } | |
305 | free_mdstat(mdstat); | |
306 | ||
307 | return status; | |
308 | } else if (strncmp(container_name, "md", 2) == 0) { | |
309 | devnum = devname2devnum(container_name); | |
310 | devname = devnum2devname(devnum); | |
311 | if (strcmp(container_name, devname) != 0) | |
312 | devname = NULL; | |
313 | } else { | |
314 | struct stat st; | |
315 | ||
316 | devnum = NoMdDev; | |
317 | if (stat(container_name, &st) == 0) | |
318 | devnum = stat2devnum(&st); | |
319 | if (devnum == NoMdDev) | |
320 | devname = NULL; | |
321 | else | |
322 | devname = devnum2devname(devnum); | |
323 | } | |
324 | ||
325 | if (!devname) { | |
326 | fprintf(stderr, "mdmon: %s is not a valid md device name\n", | |
327 | container_name); | |
328 | exit(1); | |
329 | } | |
330 | return mdmon(devname, devnum, do_fork(), takeover); | |
331 | } | |
332 | ||
333 | static int mdmon(char *devname, int devnum, int must_fork, int takeover) | |
334 | { | |
335 | int mdfd; | |
336 | struct mdinfo *mdi, *di; | |
337 | struct supertype *container; | |
338 | sigset_t set; | |
339 | struct sigaction act; | |
340 | int pfd[2]; | |
341 | int status; | |
342 | int ignore; | |
343 | pid_t victim = -1; | |
344 | int victim_sock = -1; | |
345 | ||
346 | dprintf("starting mdmon for %s\n", devname); | |
347 | ||
348 | mdfd = open_dev(devnum); | |
349 | if (mdfd < 0) { | |
350 | fprintf(stderr, "mdmon: %s: %s\n", devname, | |
351 | strerror(errno)); | |
352 | return 1; | |
353 | } | |
354 | if (md_get_version(mdfd) < 0) { | |
355 | fprintf(stderr, "mdmon: %s: Not an md device\n", | |
356 | devname); | |
357 | return 1; | |
358 | } | |
359 | ||
360 | /* Fork, and have the child tell us when they are ready */ | |
361 | if (must_fork) { | |
362 | if (pipe(pfd) != 0) { | |
363 | fprintf(stderr, "mdmon: failed to create pipe\n"); | |
364 | return 1; | |
365 | } | |
366 | switch(fork()) { | |
367 | case -1: | |
368 | fprintf(stderr, "mdmon: failed to fork: %s\n", | |
369 | strerror(errno)); | |
370 | return 1; | |
371 | case 0: /* child */ | |
372 | close(pfd[0]); | |
373 | break; | |
374 | default: /* parent */ | |
375 | close(pfd[1]); | |
376 | if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) { | |
377 | wait(&status); | |
378 | status = WEXITSTATUS(status); | |
379 | } | |
380 | return status; | |
381 | } | |
382 | } else | |
383 | pfd[0] = pfd[1] = -1; | |
384 | ||
385 | container = calloc(1, sizeof(*container)); | |
386 | container->devnum = devnum; | |
387 | container->devname = devname; | |
388 | container->arrays = NULL; | |
389 | container->subarray[0] = 0; | |
390 | container->sock = -1; | |
391 | ||
392 | if (!container->devname) { | |
393 | fprintf(stderr, "mdmon: failed to allocate container name string\n"); | |
394 | exit(3); | |
395 | } | |
396 | ||
397 | mdi = sysfs_read(mdfd, container->devnum, | |
398 | GET_VERSION|GET_LEVEL|GET_DEVS|SKIP_GONE_DEVS); | |
399 | ||
400 | if (!mdi) { | |
401 | fprintf(stderr, "mdmon: failed to load sysfs info for %s\n", | |
402 | container->devname); | |
403 | exit(3); | |
404 | } | |
405 | if (mdi->array.level != UnSet) { | |
406 | fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n", | |
407 | devname); | |
408 | exit(3); | |
409 | } | |
410 | if (mdi->array.major_version != -1 || | |
411 | mdi->array.minor_version != -2) { | |
412 | fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n", | |
413 | devname); | |
414 | exit(3); | |
415 | } | |
416 | ||
417 | container->ss = find_metadata_methods(mdi->text_version); | |
418 | if (container->ss == NULL) { | |
419 | fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n", | |
420 | devname, mdi->text_version); | |
421 | exit(3); | |
422 | } | |
423 | ||
424 | container->devs = NULL; | |
425 | for (di = mdi->devs; di; di = di->next) { | |
426 | struct mdinfo *cd = malloc(sizeof(*cd)); | |
427 | *cd = *di; | |
428 | cd->next = container->devs; | |
429 | container->devs = cd; | |
430 | } | |
431 | sysfs_free(mdi); | |
432 | ||
433 | /* SIGUSR is sent between parent and child. So both block it | |
434 | * and enable it only with pselect. | |
435 | */ | |
436 | sigemptyset(&set); | |
437 | sigaddset(&set, SIGUSR1); | |
438 | sigaddset(&set, SIGTERM); | |
439 | sigprocmask(SIG_BLOCK, &set, NULL); | |
440 | act.sa_handler = wake_me; | |
441 | act.sa_flags = 0; | |
442 | sigaction(SIGUSR1, &act, NULL); | |
443 | act.sa_handler = term; | |
444 | sigaction(SIGTERM, &act, NULL); | |
445 | act.sa_handler = SIG_IGN; | |
446 | sigaction(SIGPIPE, &act, NULL); | |
447 | ||
448 | if (takeover) { | |
449 | pid_dir = VAR_RUN; | |
450 | victim = mdmon_pid(container->devnum); | |
451 | if (victim < 0) { | |
452 | pid_dir = ALT_RUN; | |
453 | victim = mdmon_pid(container->devnum); | |
454 | } | |
455 | if (victim >= 0) | |
456 | victim_sock = connect_monitor(container->devname); | |
457 | } | |
458 | ||
459 | ignore = chdir("/"); | |
460 | if (victim < 0) { | |
461 | if (ping_monitor(container->devname) == 0) { | |
462 | fprintf(stderr, "mdmon: %s already managed\n", | |
463 | container->devname); | |
464 | exit(3); | |
465 | } | |
466 | /* if there is a pid file, kill whoever is there just in case */ | |
467 | victim = mdmon_pid(container->devnum); | |
468 | } | |
469 | if (container->ss->load_super(container, mdfd, devname)) { | |
470 | fprintf(stderr, "mdmon: Cannot load metadata for %s\n", | |
471 | devname); | |
472 | exit(3); | |
473 | } | |
474 | close(mdfd); | |
475 | ||
476 | /* Ok, this is close enough. We can say goodbye to our parent now. | |
477 | */ | |
478 | if (victim > 0) | |
479 | remove_pidfile(devname); | |
480 | pid_dir = VAR_RUN; | |
481 | if (make_pidfile(devname) < 0) { | |
482 | /* Try the alternate */ | |
483 | pid_dir = ALT_RUN; | |
484 | if (make_pidfile(devname) < 0) { | |
485 | fprintf(stderr, "mdmon: Neither %s nor %s are writable\n" | |
486 | " cannot create .pid or .sock files. Aborting\n", | |
487 | VAR_RUN, ALT_RUN); | |
488 | exit(3); | |
489 | } | |
490 | } | |
491 | container->sock = make_control_sock(devname); | |
492 | ||
493 | status = 0; | |
494 | if (write(pfd[1], &status, sizeof(status)) < 0) | |
495 | fprintf(stderr, "mdmon: failed to notify our parent: %d\n", | |
496 | getppid()); | |
497 | close(pfd[1]); | |
498 | ||
499 | mlockall(MCL_CURRENT | MCL_FUTURE); | |
500 | ||
501 | if (clone_monitor(container) < 0) { | |
502 | fprintf(stderr, "mdmon: failed to start monitor process: %s\n", | |
503 | strerror(errno)); | |
504 | exit(2); | |
505 | } | |
506 | ||
507 | if (victim > 0) { | |
508 | try_kill_monitor(victim, container->devname, victim_sock); | |
509 | close(victim_sock); | |
510 | } | |
511 | ||
512 | setsid(); | |
513 | close(0); | |
514 | open("/dev/null", O_RDWR); | |
515 | close(1); | |
516 | ignore = dup(0); | |
517 | #ifndef DEBUG | |
518 | close(2); | |
519 | ignore = dup(0); | |
520 | #endif | |
521 | ||
522 | do_manager(container); | |
523 | ||
524 | exit(0); | |
525 | } |