]> git.ipfire.org Git - thirdparty/mdadm.git/blame - mdmon.c
mdmon: recognise --all in place of /proc/mdstat
[thirdparty/mdadm.git] / mdmon.c
CommitLineData
a54d5262
DW
1/*
2 * mdmon - monitor external metadata arrays
3 *
e736b623
N
4 * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de>
5 * Copyright (C) 2007-2009 Intel Corporation
a54d5262
DW
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 */
549e9569
NB
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>
4d43913c 48#include <sys/types.h>
549e9569
NB
49#include <sys/stat.h>
50#include <sys/socket.h>
51#include <sys/un.h>
52#include <sys/mman.h>
4d43913c 53#include <sys/syscall.h>
9fe32043 54#include <sys/wait.h>
549e9569
NB
55#include <stdio.h>
56#include <errno.h>
57#include <string.h>
58#include <fcntl.h>
b109d928 59#include <signal.h>
13047e4c 60#include <dirent.h>
549e9569
NB
61
62#include <sched.h>
63
64#include "mdadm.h"
65#include "mdmon.h"
66
549e9569
NB
67struct active_array *discard_this;
68struct active_array *pending_discard;
4d43913c
NB
69
70int mon_tid, mgr_tid;
549e9569 71
6144ed44
DW
72int sigterm;
73
549e9569
NB
74int run_child(void *v)
75{
76 struct supertype *c = v;
1ed3f387 77
549e9569
NB
78 do_monitor(c);
79 return 0;
80}
81
97f734fd
N
82#ifdef __ia64__
83int __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)
549e9569 89{
549e9569 90 static char stack[4096];
549e9569 91
97f734fd
N
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
2cc98f9e 97 mon_tid = clone(run_child, stack+4096-64,
549e9569
NB
98 CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD,
99 container);
97f734fd 100#endif
3e70c845 101
4d43913c 102 mgr_tid = syscall(SYS_gettid);
2cc98f9e
DW
103
104 return mon_tid;
549e9569
NB
105}
106
ce744c97
DW
107static 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
5d4d1b26 116int make_pidfile(char *devname)
549e9569
NB
117{
118 char path[100];
119 char pid[10];
120 int fd;
3d2c4fc7
DW
121 int n;
122
5d4d1b26 123 sprintf(path, "%s/%s.pid", pid_dir, devname);
549e9569 124
5d4d1b26 125 fd = open(path, O_RDWR|O_CREAT|O_EXCL, 0600);
549e9569 126 if (fd < 0)
295646b3 127 return -errno;
549e9569 128 sprintf(pid, "%d\n", getpid());
3d2c4fc7 129 n = write(fd, pid, strlen(pid));
549e9569 130 close(fd);
3d2c4fc7
DW
131 if (n < 0)
132 return -errno;
549e9569
NB
133 return 0;
134}
135
883a6142
DW
136int is_container_member(struct mdstat_ent *mdstat, char *container)
137{
138 if (mdstat->metadata_version == NULL ||
139 strncmp(mdstat->metadata_version, "external:", 9) != 0 ||
140 !is_subarray(mdstat->metadata_version+9) ||
141 strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 ||
142 mdstat->metadata_version[10+strlen(container)] != '/')
143 return 0;
144
145 return 1;
146}
147
9f1da824 148static void try_kill_monitor(pid_t pid, char *devname, int sock)
96a8270d
DW
149{
150 char buf[100];
151 int fd;
417a4b04 152 int n;
af7ca334 153 long fl;
b109d928 154
8aae4219
DW
155 /* first rule of survival... don't off yourself */
156 if (pid == getpid())
157 return;
158
b109d928
DW
159 /* kill this process if it is mdmon */
160 sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid);
161 fd = open(buf, O_RDONLY);
162 if (fd < 0)
163 return;
164
417a4b04
N
165 n = read(fd, buf, sizeof(buf)-1);
166 buf[sizeof(buf)-1] = 0;
167 close(fd);
b109d928 168
417a4b04 169 if (n < 0 || !strstr(buf, "mdmon"))
883a6142
DW
170 return;
171
172 kill(pid, SIGTERM);
173
af7ca334
N
174 /* Wait for monitor to exit by reading from the socket, after
175 * clearing the non-blocking flag */
176 fl = fcntl(sock, F_GETFL, 0);
177 fl &= ~O_NONBLOCK;
178 fcntl(sock, F_SETFL, fl);
179 read(sock, buf, 100);
b109d928
DW
180}
181
e0d6609f
NB
182void remove_pidfile(char *devname)
183{
184 char buf[100];
185
5d4d1b26 186 sprintf(buf, "%s/%s.pid", pid_dir, devname);
e0d6609f 187 unlink(buf);
5d4d1b26 188 sprintf(buf, "%s/%s.sock", pid_dir, devname);
57752795 189 unlink(buf);
5d4d1b26
N
190 if (strcmp(pid_dir, ALT_RUN) == 0)
191 /* try to clean up when we are finished with this dir */
192 rmdir(pid_dir);
e0d6609f
NB
193}
194
295646b3 195int make_control_sock(char *devname)
549e9569
NB
196{
197 char path[100];
198 int sfd;
199 long fl;
200 struct sockaddr_un addr;
201
6144ed44
DW
202 if (sigterm)
203 return -1;
204
5d4d1b26 205 sprintf(path, "%s/%s.sock", pid_dir, devname);
549e9569
NB
206 unlink(path);
207 sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
208 if (sfd < 0)
209 return -1;
210
211 addr.sun_family = PF_LOCAL;
212 strcpy(addr.sun_path, path);
213 if (bind(sfd, &addr, sizeof(addr)) < 0) {
214 close(sfd);
215 return -1;
216 }
217 listen(sfd, 10);
218 fl = fcntl(sfd, F_GETFL, 0);
219 fl |= O_NONBLOCK;
220 fcntl(sfd, F_SETFL, fl);
221 return sfd;
222}
223
295646b3
DW
224int socket_hup_requested;
225static void hup(int sig)
226{
227 socket_hup_requested = 1;
228}
229
6144ed44
DW
230static void term(int sig)
231{
232 sigterm = 1;
233}
234
4d43913c
NB
235static void wake_me(int sig)
236{
237
238}
239
16ddab0d
DW
240/* if we are debugging and starting mdmon by hand then don't fork */
241static int do_fork(void)
242{
243 #ifdef DEBUG
40ebbb9c 244 if (check_env("MDADM_NO_MDMON"))
16ddab0d
DW
245 return 0;
246 #endif
247
248 return 1;
249}
250
13047e4c
DW
251void usage(void)
252{
5746141e 253 fprintf(stderr, "Usage: mdmon /device/name/for/container [target_dir]\n");
13047e4c
DW
254 exit(2);
255}
16ddab0d 256
3e7312a9 257static int mdmon(char *devname, int devnum, int must_fork, char *switchroot);
1ffd2840 258
549e9569
NB
259int main(int argc, char *argv[])
260{
13047e4c
DW
261 char *container_name = NULL;
262 char *switchroot = NULL;
e8a70c89
N
263 int devnum;
264 char *devname;
1ffd2840 265 int status = 0;
13047e4c
DW
266
267 switch (argc) {
5746141e
DW
268 case 3:
269 switchroot = argv[2];
13047e4c
DW
270 case 2:
271 container_name = argv[1];
272 break;
13047e4c
DW
273 default:
274 usage();
549e9569 275 }
13047e4c 276
a9596b79
N
277 if (strcmp(container_name, "/proc/mdstat") == 0 ||
278 strcmp(container_name, "--all") == 0) {
1ffd2840
DW
279 struct mdstat_ent *mdstat, *e;
280
281 /* launch an mdmon instance for each container found */
1ffd2840
DW
282 mdstat = mdstat_read(0, 0);
283 for (e = mdstat; e; e = e->next) {
284 if (strncmp(e->metadata_version, "external:", 9) == 0 &&
285 !is_subarray(&e->metadata_version[9])) {
286 devname = devnum2devname(e->devnum);
1b34f519
DW
287 /* update cmdline so this mdmon instance can be
288 * distinguished from others in a call to ps(1)
289 */
290 if (strlen(devname) <= strlen(container_name)) {
291 memset(container_name, 0, strlen(container_name));
292 sprintf(container_name, "%s", devname);
293 }
3e7312a9 294 status |= mdmon(devname, e->devnum, 1,
1ffd2840
DW
295 switchroot);
296 }
297 }
298 free_mdstat(mdstat);
299
300 return status;
301 } else if (strncmp(container_name, "md", 2) == 0) {
6f4098a6
DW
302 devnum = devname2devnum(container_name);
303 devname = devnum2devname(devnum);
304 if (strcmp(container_name, devname) != 0)
305 devname = NULL;
306 } else {
307 struct stat st;
308
309 devnum = NoMdDev;
310 if (stat(container_name, &st) == 0)
311 devnum = stat2devnum(&st);
312 if (devnum == NoMdDev)
313 devname = NULL;
314 else
315 devname = devnum2devname(devnum);
316 }
317
318 if (!devname) {
e8a70c89
N
319 fprintf(stderr, "mdmon: %s is not a valid md device name\n",
320 container_name);
321 exit(1);
322 }
3e7312a9 323 return mdmon(devname, devnum, do_fork(), switchroot);
1ffd2840
DW
324}
325
3e7312a9 326static int mdmon(char *devname, int devnum, int must_fork, char *switchroot)
1ffd2840
DW
327{
328 int mdfd;
329 struct mdinfo *mdi, *di;
330 struct supertype *container;
331 sigset_t set;
332 struct sigaction act;
333 int pfd[2];
334 int status;
335 int ignore;
96a8270d 336 pid_t victim = -1;
9f1da824 337 int victim_sock = -1;
1ffd2840 338
5746141e
DW
339 dprintf("starting mdmon for %s in %s\n",
340 devname, switchroot ? : "/");
b928b5a0 341
e98ef225
N
342 /* switchroot is either a path name starting with '/', or a
343 * pid of the original mdmon (we have already done the chroot).
344 * In the latter case, stdin is a socket connected to the original
345 * mdmon.
346 */
347
b928b5a0 348 /* try to spawn mdmon instances from the target file system */
e98ef225
N
349 if (switchroot && switchroot[0] == '/' &&
350 strcmp(switchroot, "/") != 0) {
b928b5a0 351 pid_t pid;
e98ef225 352 char buf[20];
b928b5a0 353
b928b5a0
DW
354 switch (fork()) {
355 case 0:
24f6f99b 356 victim = mdmon_pid(devnum);
e98ef225
N
357 victim_sock = connect_monitor(devname);
358 if (chroot(switchroot) != 0) {
359 fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
360 switchroot, strerror(errno));
361 exit(4);
362 }
363 ignore = chdir("/");
364 sprintf(buf, "%d", victim);
365 if (victim_sock) {
366 close(0);
367 dup(victim_sock);
368 close(victim_sock);
369 }
370 execl("/sbin/mdmon", "mdmon", devname, buf, NULL);
b928b5a0
DW
371 exit(1);
372 case -1:
373 return 1;
374 default:
375 pid = wait(&status);
376 if (pid > -1 && WIFEXITED(status) &&
377 WEXITSTATUS(status) == 0)
378 return 0;
379 else
380 return 1;
381 }
382 }
383
e8a70c89 384 mdfd = open_dev(devnum);
549e9569 385 if (mdfd < 0) {
1ffd2840 386 fprintf(stderr, "mdmon: %s: %s\n", devname,
549e9569 387 strerror(errno));
1ffd2840 388 return 1;
549e9569
NB
389 }
390 if (md_get_version(mdfd) < 0) {
13047e4c 391 fprintf(stderr, "mdmon: %s: Not an md device\n",
1ffd2840
DW
392 devname);
393 return 1;
549e9569
NB
394 }
395
9fe32043 396 /* Fork, and have the child tell us when they are ready */
3e7312a9 397 if (must_fork) {
3d2c4fc7
DW
398 if (pipe(pfd) != 0) {
399 fprintf(stderr, "mdmon: failed to create pipe\n");
1ffd2840 400 return 1;
3d2c4fc7 401 }
16ddab0d
DW
402 switch(fork()) {
403 case -1:
404 fprintf(stderr, "mdmon: failed to fork: %s\n",
405 strerror(errno));
1ffd2840 406 return 1;
16ddab0d
DW
407 case 0: /* child */
408 close(pfd[0]);
409 break;
410 default: /* parent */
411 close(pfd[1]);
412 if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) {
413 wait(&status);
414 status = WEXITSTATUS(status);
415 }
1ffd2840 416 return status;
9fe32043 417 }
16ddab0d
DW
418 } else
419 pfd[0] = pfd[1] = -1;
549e9569 420
f5df5d69 421 container = calloc(1, sizeof(*container));
e8a70c89
N
422 container->devnum = devnum;
423 container->devname = devname;
13047e4c 424 container->arrays = NULL;
c1363b40 425 container->subarray[0] = 0;
96a8270d 426 container->sock = -1;
13047e4c
DW
427
428 if (!container->devname) {
429 fprintf(stderr, "mdmon: failed to allocate container name string\n");
430 exit(3);
431 }
432
433 mdi = sysfs_read(mdfd, container->devnum,
7da80e6f 434 GET_VERSION|GET_LEVEL|GET_DEVS|SKIP_GONE_DEVS);
13047e4c
DW
435
436 if (!mdi) {
437 fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
438 container->devname);
439 exit(3);
440 }
441 if (mdi->array.level != UnSet) {
442 fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
1ffd2840 443 devname);
13047e4c
DW
444 exit(3);
445 }
446 if (mdi->array.major_version != -1 ||
447 mdi->array.minor_version != -2) {
448 fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
1ffd2840 449 devname);
13047e4c
DW
450 exit(3);
451 }
452
453 container->ss = find_metadata_methods(mdi->text_version);
454 if (container->ss == NULL) {
455 fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
1ffd2840 456 devname, mdi->text_version);
13047e4c
DW
457 exit(3);
458 }
459
460 container->devs = NULL;
461 for (di = mdi->devs; di; di = di->next) {
462 struct mdinfo *cd = malloc(sizeof(*cd));
463 *cd = *di;
464 cd->next = container->devs;
465 container->devs = cd;
466 }
467 sysfs_free(mdi);
549e9569 468
883a6142
DW
469 /* SIGUSR is sent between parent and child. So both block it
470 * and enable it only with pselect.
471 */
472 sigemptyset(&set);
473 sigaddset(&set, SIGUSR1);
474 sigaddset(&set, SIGHUP);
475 sigaddset(&set, SIGALRM);
476 sigaddset(&set, SIGTERM);
477 sigprocmask(SIG_BLOCK, &set, NULL);
478 act.sa_handler = wake_me;
479 act.sa_flags = 0;
480 sigaction(SIGUSR1, &act, NULL);
481 sigaction(SIGALRM, &act, NULL);
482 act.sa_handler = hup;
483 sigaction(SIGHUP, &act, NULL);
484 act.sa_handler = term;
485 sigaction(SIGTERM, &act, NULL);
486 act.sa_handler = SIG_IGN;
487 sigaction(SIGPIPE, &act, NULL);
488
13047e4c
DW
489 if (switchroot) {
490 /* we assume we assume that /sys /proc /dev are available in
96a8270d 491 * the new root
13047e4c 492 */
e98ef225 493 if (switchroot[0] == '/') {
24f6f99b 494 victim = mdmon_pid(container->devnum);
e98ef225
N
495 victim_sock = connect_monitor(container->devname);
496 } else {
497 victim = atoi(switchroot);
498 victim_sock = 0;
13047e4c
DW
499 }
500 }
501
13047e4c 502 ignore = chdir("/");
24cfdbc5 503 if (victim < 0) {
b109d928
DW
504 if (ping_monitor(container->devname) == 0) {
505 fprintf(stderr, "mdmon: %s already managed\n",
506 container->devname);
507 exit(3);
24cfdbc5
N
508 }
509 /* if there is a pid file, kill whoever is there just in case */
24f6f99b 510 victim = mdmon_pid(container->devnum);
549e9569 511 }
1ffd2840 512 if (container->ss->load_super(container, mdfd, devname)) {
549e9569 513 fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
1ffd2840 514 devname);
549e9569
NB
515 exit(3);
516 }
e8a70c89 517 close(mdfd);
549e9569 518
9fe32043
N
519 /* Ok, this is close enough. We can say goodbye to our parent now.
520 */
521 status = 0;
3d2c4fc7
DW
522 if (write(pfd[1], &status, sizeof(status)) < 0)
523 fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
524 getppid());
9fe32043
N
525 close(pfd[1]);
526
1373b07d 527 mlockall(MCL_CURRENT | MCL_FUTURE);
549e9569 528
3e70c845 529 if (clone_monitor(container) < 0) {
295646b3 530 fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
549e9569
NB
531 strerror(errno));
532 exit(2);
533 }
534
9f1da824
DW
535 if (victim > -1) {
536 try_kill_monitor(victim, container->devname, victim_sock);
537 close(victim_sock);
538 }
e98ef225
N
539
540 setsid();
541 close(0);
542 open("/dev/null", O_RDWR);
543 close(1);
544 ignore = dup(0);
545#ifndef DEBUG
546 close(2);
547 ignore = dup(0);
548#endif
549
549e9569
NB
550 do_manager(container);
551
552 exit(0);
553}