]> git.ipfire.org Git - thirdparty/mdadm.git/blob - mdmon.c
mdmon: support scanning for containers
[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 mdmon(char *devname, int devnum, int scan, char *switchroot);
279
280 int main(int argc, char *argv[])
281 {
282 char *container_name = NULL;
283 char *switchroot = NULL;
284 int devnum;
285 char *devname;
286 int scan = 0;
287 int status = 0;
288
289 switch (argc) {
290 case 2:
291 container_name = argv[1];
292 break;
293 case 4:
294 if (strcmp(argv[1], "--switch-root") != 0) {
295 fprintf(stderr, "mdmon: unknown argument %s\n", argv[1]);
296 usage();
297 }
298 switchroot = argv[2];
299 container_name = argv[3];
300 break;
301 default:
302 usage();
303 }
304
305 if (strcmp(container_name, "/proc/mdstat") == 0) {
306 struct mdstat_ent *mdstat, *e;
307
308 /* launch an mdmon instance for each container found */
309 scan = 1;
310 mdstat = mdstat_read(0, 0);
311 for (e = mdstat; e; e = e->next) {
312 if (strncmp(e->metadata_version, "external:", 9) == 0 &&
313 !is_subarray(&e->metadata_version[9])) {
314 devname = devnum2devname(e->devnum);
315 status |= mdmon(devname, e->devnum, scan,
316 switchroot);
317 }
318 }
319 free_mdstat(mdstat);
320
321 return status;
322 } else if (strncmp(container_name, "md", 2) == 0) {
323 devnum = devname2devnum(container_name);
324 devname = devnum2devname(devnum);
325 if (strcmp(container_name, devname) != 0)
326 devname = NULL;
327 } else {
328 struct stat st;
329
330 devnum = NoMdDev;
331 if (stat(container_name, &st) == 0)
332 devnum = stat2devnum(&st);
333 if (devnum == NoMdDev)
334 devname = NULL;
335 else
336 devname = devnum2devname(devnum);
337 }
338
339 if (!devname) {
340 fprintf(stderr, "mdmon: %s is not a valid md device name\n",
341 container_name);
342 exit(1);
343 }
344 return mdmon(devname, devnum, scan, switchroot);
345 }
346
347 int mdmon(char *devname, int devnum, int scan, char *switchroot)
348 {
349 int mdfd;
350 struct mdinfo *mdi, *di;
351 struct supertype *container;
352 sigset_t set;
353 struct sigaction act;
354 int pfd[2];
355 int status;
356 int ignore;
357
358 mdfd = open_dev(devnum);
359 if (mdfd < 0) {
360 fprintf(stderr, "mdmon: %s: %s\n", devname,
361 strerror(errno));
362 return 1;
363 }
364 if (md_get_version(mdfd) < 0) {
365 fprintf(stderr, "mdmon: %s: Not an md device\n",
366 devname);
367 return 1;
368 }
369
370 /* Fork, and have the child tell us when they are ready */
371 if (do_fork() || scan) {
372 if (pipe(pfd) != 0) {
373 fprintf(stderr, "mdmon: failed to create pipe\n");
374 return 1;
375 }
376 switch(fork()) {
377 case -1:
378 fprintf(stderr, "mdmon: failed to fork: %s\n",
379 strerror(errno));
380 return 1;
381 case 0: /* child */
382 close(pfd[0]);
383 break;
384 default: /* parent */
385 close(pfd[1]);
386 if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) {
387 wait(&status);
388 status = WEXITSTATUS(status);
389 }
390 return status;
391 }
392 } else
393 pfd[0] = pfd[1] = -1;
394
395 container = malloc(sizeof(*container));
396 container->devnum = devnum;
397 container->devname = devname;
398 container->arrays = NULL;
399 container->subarray[0] = 0;
400
401 if (!container->devname) {
402 fprintf(stderr, "mdmon: failed to allocate container name string\n");
403 exit(3);
404 }
405
406 mdi = sysfs_read(mdfd, container->devnum,
407 GET_VERSION|GET_LEVEL|GET_DEVS);
408
409 if (!mdi) {
410 fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
411 container->devname);
412 exit(3);
413 }
414 if (mdi->array.level != UnSet) {
415 fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
416 devname);
417 exit(3);
418 }
419 if (mdi->array.major_version != -1 ||
420 mdi->array.minor_version != -2) {
421 fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
422 devname);
423 exit(3);
424 }
425
426 container->ss = find_metadata_methods(mdi->text_version);
427 if (container->ss == NULL) {
428 fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
429 devname, mdi->text_version);
430 exit(3);
431 }
432
433 container->devs = NULL;
434 for (di = mdi->devs; di; di = di->next) {
435 struct mdinfo *cd = malloc(sizeof(*cd));
436 *cd = *di;
437 cd->next = container->devs;
438 container->devs = cd;
439 }
440 sysfs_free(mdi);
441
442 /* SIGUSR is sent between parent and child. So both block it
443 * and enable it only with pselect.
444 */
445 sigemptyset(&set);
446 sigaddset(&set, SIGUSR1);
447 sigaddset(&set, SIGHUP);
448 sigaddset(&set, SIGALRM);
449 sigaddset(&set, SIGTERM);
450 sigprocmask(SIG_BLOCK, &set, NULL);
451 act.sa_handler = wake_me;
452 act.sa_flags = 0;
453 sigaction(SIGUSR1, &act, NULL);
454 sigaction(SIGALRM, &act, NULL);
455 act.sa_handler = hup;
456 sigaction(SIGHUP, &act, NULL);
457 act.sa_handler = term;
458 sigaction(SIGTERM, &act, NULL);
459 act.sa_handler = SIG_IGN;
460 sigaction(SIGPIPE, &act, NULL);
461
462 if (switchroot) {
463 /* we assume we assume that /sys /proc /dev are available in
464 * the new root (see nash:setuproot)
465 *
466 * kill any monitors in the current namespace and change
467 * to the new one
468 */
469 try_kill_monitor(container->devname);
470 if (chroot(switchroot) != 0) {
471 fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
472 switchroot, strerror(errno));
473 exit(4);
474 }
475 }
476
477 /* If this fails, we hope it already exists
478 * pid file lives in /var/run/mdadm/mdXX.pid
479 */
480 mkdir("/var", 0600);
481 mkdir("/var/run", 0600);
482 mkdir("/var/run/mdadm", 0600);
483 ignore = chdir("/");
484 if (make_pidfile(container->devname, O_EXCL) < 0) {
485 if (ping_monitor(container->devname) == 0) {
486 fprintf(stderr, "mdmon: %s already managed\n",
487 container->devname);
488 exit(3);
489 } else {
490 int err;
491
492 /* cleanup the old monitor, this one is taking over */
493 try_kill_monitor(container->devname);
494 err = make_pidfile(container->devname, 0);
495 if (err < 0) {
496 fprintf(stderr, "mdmon: %s Cannot create pidfile\n",
497 container->devname);
498 if (err == -EROFS) {
499 /* FIXME implement a mechanism to
500 * prevent duplicate monitor instances
501 */
502 fprintf(stderr,
503 "mdmon: continuing on read-only file system\n");
504 } else
505 exit(3);
506 }
507 }
508 }
509 container->sock = make_control_sock(container->devname);
510
511 if (container->ss->load_super(container, mdfd, devname)) {
512 fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
513 devname);
514 exit(3);
515 }
516 close(mdfd);
517
518 /* Ok, this is close enough. We can say goodbye to our parent now.
519 */
520 status = 0;
521 if (write(pfd[1], &status, sizeof(status)) < 0)
522 fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
523 getppid());
524 close(pfd[1]);
525
526 setsid();
527 close(0);
528 open("/dev/null", O_RDWR);
529 close(1);
530 ignore = dup(0);
531 #ifndef DEBUG
532 close(2);
533 ignore = dup(0);
534 #endif
535
536 mlockall(MCL_FUTURE);
537
538 if (clone_monitor(container) < 0) {
539 fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
540 strerror(errno));
541 exit(2);
542 }
543
544 do_manager(container);
545
546 exit(0);
547 }