]> git.ipfire.org Git - thirdparty/mdadm.git/blob - mdmon.c
mdmon: fix missing ->subarray initialization
[thirdparty/mdadm.git] / mdmon.c
1 /*
2 * mdmon - monitor external metadata arrays
3 *
4 * Copyright (C) 2007-2008 Neil Brown <neilb@suse.de>
5 * Copyright (C) 2007-2008 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
117 int make_pidfile(char *devname, int o_excl)
118 {
119 char path[100];
120 char pid[10];
121 int fd;
122 int n;
123
124 if (sigterm)
125 return -1;
126
127 sprintf(path, "/var/run/mdadm/%s.pid", devname);
128
129 fd = open(path, O_RDWR|O_CREAT|o_excl, 0600);
130 if (fd < 0)
131 return -errno;
132 sprintf(pid, "%d\n", getpid());
133 n = write(fd, pid, strlen(pid));
134 close(fd);
135 if (n < 0)
136 return -errno;
137 return 0;
138 }
139
140 int is_container_member(struct mdstat_ent *mdstat, char *container)
141 {
142 if (mdstat->metadata_version == NULL ||
143 strncmp(mdstat->metadata_version, "external:", 9) != 0 ||
144 !is_subarray(mdstat->metadata_version+9) ||
145 strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 ||
146 mdstat->metadata_version[10+strlen(container)] != '/')
147 return 0;
148
149 return 1;
150 }
151
152 void remove_pidfile(char *devname);
153 static void try_kill_monitor(char *devname)
154 {
155 char buf[100];
156 int fd;
157 pid_t pid;
158 struct mdstat_ent *mdstat;
159
160 sprintf(buf, "/var/run/mdadm/%s.pid", devname);
161 fd = open(buf, O_RDONLY);
162 if (fd < 0)
163 return;
164
165 if (read(fd, buf, sizeof(buf)) < 0) {
166 close(fd);
167 return;
168 }
169
170 close(fd);
171 pid = strtoul(buf, NULL, 10);
172
173 /* first rule of survival... don't off yourself */
174 if (pid == getpid())
175 return;
176
177 /* kill this process if it is mdmon */
178 sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid);
179 fd = open(buf, O_RDONLY);
180 if (fd < 0)
181 return;
182
183 if (read(fd, buf, sizeof(buf)) < 0) {
184 close(fd);
185 return;
186 }
187
188 if (!strstr(buf, "mdmon"))
189 return;
190
191 kill(pid, SIGTERM);
192
193 mdstat = mdstat_read(0, 0);
194 for ( ; mdstat; mdstat = mdstat->next)
195 if (is_container_member(mdstat, devname)) {
196 sprintf(buf, "/dev/%s", mdstat->dev);
197 WaitClean(buf, 0);
198 }
199 free_mdstat(mdstat);
200 remove_pidfile(devname);
201 }
202
203 void remove_pidfile(char *devname)
204 {
205 char buf[100];
206
207 if (sigterm)
208 return;
209
210 sprintf(buf, "/var/run/mdadm/%s.pid", devname);
211 unlink(buf);
212 sprintf(buf, "/var/run/mdadm/%s.sock", devname);
213 unlink(buf);
214 }
215
216 int make_control_sock(char *devname)
217 {
218 char path[100];
219 int sfd;
220 long fl;
221 struct sockaddr_un addr;
222
223 if (sigterm)
224 return -1;
225
226 sprintf(path, "/var/run/mdadm/%s.sock", devname);
227 unlink(path);
228 sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
229 if (sfd < 0)
230 return -1;
231
232 addr.sun_family = PF_LOCAL;
233 strcpy(addr.sun_path, path);
234 if (bind(sfd, &addr, sizeof(addr)) < 0) {
235 close(sfd);
236 return -1;
237 }
238 listen(sfd, 10);
239 fl = fcntl(sfd, F_GETFL, 0);
240 fl |= O_NONBLOCK;
241 fcntl(sfd, F_SETFL, fl);
242 return sfd;
243 }
244
245 int socket_hup_requested;
246 static void hup(int sig)
247 {
248 socket_hup_requested = 1;
249 }
250
251 static void term(int sig)
252 {
253 sigterm = 1;
254 }
255
256 static void wake_me(int sig)
257 {
258
259 }
260
261 /* if we are debugging and starting mdmon by hand then don't fork */
262 static int do_fork(void)
263 {
264 #ifdef DEBUG
265 if (check_env("MDADM_NO_MDMON"))
266 return 0;
267 #endif
268
269 return 1;
270 }
271
272 void usage(void)
273 {
274 fprintf(stderr, "Usage: mdmon [--switch-root dir] /device/name/for/container\n");
275 exit(2);
276 }
277
278 int main(int argc, char *argv[])
279 {
280 int mdfd;
281 struct mdinfo *mdi, *di;
282 struct supertype *container;
283 sigset_t set;
284 struct sigaction act;
285 int pfd[2];
286 int status;
287 int ignore;
288 char *container_name = NULL;
289 char *switchroot = NULL;
290 int devnum;
291 char *devname;
292
293 switch (argc) {
294 case 2:
295 container_name = argv[1];
296 break;
297 case 4:
298 if (strcmp(argv[1], "--switch-root") != 0) {
299 fprintf(stderr, "mdmon: unknown argument %s\n", argv[1]);
300 usage();
301 }
302 switchroot = argv[2];
303 container_name = argv[3];
304 break;
305 default:
306 usage();
307 }
308
309 devnum = devname2devnum(container_name);
310 devname = devnum2devname(devnum);
311 if (strcmp(container_name, devname) != 0) {
312 fprintf(stderr, "mdmon: %s is not a valid md device name\n",
313 container_name);
314 exit(1);
315 }
316 mdfd = open_dev(devnum);
317 if (mdfd < 0) {
318 fprintf(stderr, "mdmon: %s: %s\n", container_name,
319 strerror(errno));
320 exit(1);
321 }
322 if (md_get_version(mdfd) < 0) {
323 fprintf(stderr, "mdmon: %s: Not an md device\n",
324 container_name);
325 exit(1);
326 }
327
328 /* Fork, and have the child tell us when they are ready */
329 if (do_fork()) {
330 if (pipe(pfd) != 0) {
331 fprintf(stderr, "mdmon: failed to create pipe\n");
332 exit(1);
333 }
334 switch(fork()) {
335 case -1:
336 fprintf(stderr, "mdmon: failed to fork: %s\n",
337 strerror(errno));
338 exit(1);
339 case 0: /* child */
340 close(pfd[0]);
341 break;
342 default: /* parent */
343 close(pfd[1]);
344 if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) {
345 wait(&status);
346 status = WEXITSTATUS(status);
347 }
348 exit(status);
349 }
350 } else
351 pfd[0] = pfd[1] = -1;
352
353 container = malloc(sizeof(*container));
354 container->devnum = devnum;
355 container->devname = devname;
356 container->arrays = NULL;
357 container->subarray[0] = 0;
358
359 if (!container->devname) {
360 fprintf(stderr, "mdmon: failed to allocate container name string\n");
361 exit(3);
362 }
363
364 mdi = sysfs_read(mdfd, container->devnum,
365 GET_VERSION|GET_LEVEL|GET_DEVS);
366
367 if (!mdi) {
368 fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
369 container->devname);
370 exit(3);
371 }
372 if (mdi->array.level != UnSet) {
373 fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
374 container_name);
375 exit(3);
376 }
377 if (mdi->array.major_version != -1 ||
378 mdi->array.minor_version != -2) {
379 fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
380 container_name);
381 exit(3);
382 }
383
384 container->ss = find_metadata_methods(mdi->text_version);
385 if (container->ss == NULL) {
386 fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
387 container_name, mdi->text_version);
388 exit(3);
389 }
390
391 container->devs = NULL;
392 for (di = mdi->devs; di; di = di->next) {
393 struct mdinfo *cd = malloc(sizeof(*cd));
394 *cd = *di;
395 cd->next = container->devs;
396 container->devs = cd;
397 }
398 sysfs_free(mdi);
399
400 /* SIGUSR is sent between parent and child. So both block it
401 * and enable it only with pselect.
402 */
403 sigemptyset(&set);
404 sigaddset(&set, SIGUSR1);
405 sigaddset(&set, SIGHUP);
406 sigaddset(&set, SIGALRM);
407 sigaddset(&set, SIGTERM);
408 sigprocmask(SIG_BLOCK, &set, NULL);
409 act.sa_handler = wake_me;
410 act.sa_flags = 0;
411 sigaction(SIGUSR1, &act, NULL);
412 sigaction(SIGALRM, &act, NULL);
413 act.sa_handler = hup;
414 sigaction(SIGHUP, &act, NULL);
415 act.sa_handler = term;
416 sigaction(SIGTERM, &act, NULL);
417 act.sa_handler = SIG_IGN;
418 sigaction(SIGPIPE, &act, NULL);
419
420 if (switchroot) {
421 /* we assume we assume that /sys /proc /dev are available in
422 * the new root (see nash:setuproot)
423 *
424 * kill any monitors in the current namespace and change
425 * to the new one
426 */
427 try_kill_monitor(container->devname);
428 if (chroot(switchroot) != 0) {
429 fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
430 switchroot, strerror(errno));
431 exit(4);
432 }
433 }
434
435 /* If this fails, we hope it already exists
436 * pid file lives in /var/run/mdadm/mdXX.pid
437 */
438 mkdir("/var", 0600);
439 mkdir("/var/run", 0600);
440 mkdir("/var/run/mdadm", 0600);
441 ignore = chdir("/");
442 if (make_pidfile(container->devname, O_EXCL) < 0) {
443 if (ping_monitor(container->devname) == 0) {
444 fprintf(stderr, "mdmon: %s already managed\n",
445 container->devname);
446 exit(3);
447 } else {
448 int err;
449
450 /* cleanup the old monitor, this one is taking over */
451 try_kill_monitor(container->devname);
452 err = make_pidfile(container->devname, 0);
453 if (err < 0) {
454 fprintf(stderr, "mdmon: %s Cannot create pidfile\n",
455 container->devname);
456 if (err == -EROFS) {
457 /* FIXME implement a mechanism to
458 * prevent duplicate monitor instances
459 */
460 fprintf(stderr,
461 "mdmon: continuing on read-only file system\n");
462 } else
463 exit(3);
464 }
465 }
466 }
467 container->sock = make_control_sock(container->devname);
468
469 if (container->ss->load_super(container, mdfd, container_name)) {
470 fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
471 container_name);
472 exit(3);
473 }
474 close(mdfd);
475
476 /* Ok, this is close enough. We can say goodbye to our parent now.
477 */
478 status = 0;
479 if (write(pfd[1], &status, sizeof(status)) < 0)
480 fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
481 getppid());
482 close(pfd[1]);
483
484 setsid();
485 close(0);
486 open("/dev/null", O_RDWR);
487 close(1);
488 ignore = dup(0);
489 #ifndef DEBUG
490 close(2);
491 ignore = dup(0);
492 #endif
493
494 mlockall(MCL_FUTURE);
495
496 if (clone_monitor(container) < 0) {
497 fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
498 strerror(errno));
499 exit(2);
500 }
501
502 do_manager(container);
503
504 exit(0);
505 }