]> git.ipfire.org Git - thirdparty/mdadm.git/blame - mdmon.c
Merge branch 'master' into devel-3.0
[thirdparty/mdadm.git] / mdmon.c
CommitLineData
a54d5262
DW
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 */
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
549e9569 116
295646b3 117int make_pidfile(char *devname, int o_excl)
549e9569
NB
118{
119 char path[100];
120 char pid[10];
121 int fd;
3d2c4fc7
DW
122 int n;
123
6144ed44
DW
124 if (sigterm)
125 return -1;
126
549e9569
NB
127 sprintf(path, "/var/run/mdadm/%s.pid", devname);
128
b109d928 129 fd = open(path, O_RDWR|O_CREAT|o_excl, 0600);
549e9569 130 if (fd < 0)
295646b3 131 return -errno;
549e9569 132 sprintf(pid, "%d\n", getpid());
3d2c4fc7 133 n = write(fd, pid, strlen(pid));
549e9569 134 close(fd);
3d2c4fc7
DW
135 if (n < 0)
136 return -errno;
549e9569
NB
137 return 0;
138}
139
883a6142
DW
140int 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
152void remove_pidfile(char *devname);
b109d928
DW
153static void try_kill_monitor(char *devname)
154{
155 char buf[100];
156 int fd;
157 pid_t pid;
883a6142 158 struct mdstat_ent *mdstat;
b109d928
DW
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
8aae4219
DW
173 /* first rule of survival... don't off yourself */
174 if (pid == getpid())
175 return;
176
b109d928
DW
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
883a6142
DW
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);
27dec8fa 197 WaitClean(buf, 0);
883a6142
DW
198 }
199 free_mdstat(mdstat);
200 remove_pidfile(devname);
b109d928
DW
201}
202
e0d6609f
NB
203void remove_pidfile(char *devname)
204{
205 char buf[100];
206
6144ed44
DW
207 if (sigterm)
208 return;
209
e0d6609f
NB
210 sprintf(buf, "/var/run/mdadm/%s.pid", devname);
211 unlink(buf);
57752795
N
212 sprintf(buf, "/var/run/mdadm/%s.sock", devname);
213 unlink(buf);
e0d6609f
NB
214}
215
295646b3 216int make_control_sock(char *devname)
549e9569
NB
217{
218 char path[100];
219 int sfd;
220 long fl;
221 struct sockaddr_un addr;
222
6144ed44
DW
223 if (sigterm)
224 return -1;
225
549e9569
NB
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
295646b3
DW
245int socket_hup_requested;
246static void hup(int sig)
247{
248 socket_hup_requested = 1;
249}
250
6144ed44
DW
251static void term(int sig)
252{
253 sigterm = 1;
254}
255
4d43913c
NB
256static void wake_me(int sig)
257{
258
259}
260
16ddab0d
DW
261/* if we are debugging and starting mdmon by hand then don't fork */
262static int do_fork(void)
263{
264 #ifdef DEBUG
40ebbb9c 265 if (check_env("MDADM_NO_MDMON"))
16ddab0d
DW
266 return 0;
267 #endif
268
269 return 1;
270}
271
13047e4c
DW
272void usage(void)
273{
274 fprintf(stderr, "Usage: mdmon [--switch-root dir] /device/name/for/container\n");
275 exit(2);
276}
16ddab0d 277
549e9569
NB
278int main(int argc, char *argv[])
279{
280 int mdfd;
549e9569
NB
281 struct mdinfo *mdi, *di;
282 struct supertype *container;
4d43913c 283 sigset_t set;
bfa44e2e 284 struct sigaction act;
9fe32043
N
285 int pfd[2];
286 int status;
3d2c4fc7 287 int ignore;
13047e4c
DW
288 char *container_name = NULL;
289 char *switchroot = NULL;
e8a70c89
N
290 int devnum;
291 char *devname;
13047e4c
DW
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();
549e9569 307 }
13047e4c 308
e8a70c89
N
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);
549e9569 317 if (mdfd < 0) {
13047e4c 318 fprintf(stderr, "mdmon: %s: %s\n", container_name,
549e9569
NB
319 strerror(errno));
320 exit(1);
321 }
322 if (md_get_version(mdfd) < 0) {
13047e4c
DW
323 fprintf(stderr, "mdmon: %s: Not an md device\n",
324 container_name);
549e9569
NB
325 exit(1);
326 }
327
9fe32043 328 /* Fork, and have the child tell us when they are ready */
16ddab0d 329 if (do_fork()) {
3d2c4fc7
DW
330 if (pipe(pfd) != 0) {
331 fprintf(stderr, "mdmon: failed to create pipe\n");
332 exit(1);
333 }
16ddab0d
DW
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);
9fe32043 349 }
16ddab0d
DW
350 } else
351 pfd[0] = pfd[1] = -1;
549e9569
NB
352
353 container = malloc(sizeof(*container));
e8a70c89
N
354 container->devnum = devnum;
355 container->devname = devname;
13047e4c
DW
356 container->arrays = NULL;
357
358 if (!container->devname) {
359 fprintf(stderr, "mdmon: failed to allocate container name string\n");
360 exit(3);
361 }
362
363 mdi = sysfs_read(mdfd, container->devnum,
364 GET_VERSION|GET_LEVEL|GET_DEVS);
365
366 if (!mdi) {
367 fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
368 container->devname);
369 exit(3);
370 }
371 if (mdi->array.level != UnSet) {
372 fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
373 container_name);
374 exit(3);
375 }
376 if (mdi->array.major_version != -1 ||
377 mdi->array.minor_version != -2) {
378 fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
379 container_name);
380 exit(3);
381 }
382
383 container->ss = find_metadata_methods(mdi->text_version);
384 if (container->ss == NULL) {
385 fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
386 container_name, mdi->text_version);
387 exit(3);
388 }
389
390 container->devs = NULL;
391 for (di = mdi->devs; di; di = di->next) {
392 struct mdinfo *cd = malloc(sizeof(*cd));
393 *cd = *di;
394 cd->next = container->devs;
395 container->devs = cd;
396 }
397 sysfs_free(mdi);
549e9569 398
883a6142
DW
399 /* SIGUSR is sent between parent and child. So both block it
400 * and enable it only with pselect.
401 */
402 sigemptyset(&set);
403 sigaddset(&set, SIGUSR1);
404 sigaddset(&set, SIGHUP);
405 sigaddset(&set, SIGALRM);
406 sigaddset(&set, SIGTERM);
407 sigprocmask(SIG_BLOCK, &set, NULL);
408 act.sa_handler = wake_me;
409 act.sa_flags = 0;
410 sigaction(SIGUSR1, &act, NULL);
411 sigaction(SIGALRM, &act, NULL);
412 act.sa_handler = hup;
413 sigaction(SIGHUP, &act, NULL);
414 act.sa_handler = term;
415 sigaction(SIGTERM, &act, NULL);
416 act.sa_handler = SIG_IGN;
417 sigaction(SIGPIPE, &act, NULL);
418
13047e4c
DW
419 if (switchroot) {
420 /* we assume we assume that /sys /proc /dev are available in
421 * the new root (see nash:setuproot)
422 *
423 * kill any monitors in the current namespace and change
424 * to the new one
425 */
426 try_kill_monitor(container->devname);
427 if (chroot(switchroot) != 0) {
428 fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
429 switchroot, strerror(errno));
430 exit(4);
431 }
432 }
433
434 /* If this fails, we hope it already exists
435 * pid file lives in /var/run/mdadm/mdXX.pid
436 */
437 mkdir("/var", 0600);
438 mkdir("/var/run", 0600);
549e9569 439 mkdir("/var/run/mdadm", 0600);
13047e4c 440 ignore = chdir("/");
b109d928
DW
441 if (make_pidfile(container->devname, O_EXCL) < 0) {
442 if (ping_monitor(container->devname) == 0) {
443 fprintf(stderr, "mdmon: %s already managed\n",
444 container->devname);
445 exit(3);
446 } else {
295646b3
DW
447 int err;
448
b109d928
DW
449 /* cleanup the old monitor, this one is taking over */
450 try_kill_monitor(container->devname);
295646b3
DW
451 err = make_pidfile(container->devname, 0);
452 if (err < 0) {
b109d928
DW
453 fprintf(stderr, "mdmon: %s Cannot create pidfile\n",
454 container->devname);
295646b3
DW
455 if (err == -EROFS) {
456 /* FIXME implement a mechanism to
457 * prevent duplicate monitor instances
458 */
459 fprintf(stderr,
460 "mdmon: continuing on read-only file system\n");
461 } else
462 exit(3);
b109d928
DW
463 }
464 }
549e9569 465 }
549e9569 466 container->sock = make_control_sock(container->devname);
549e9569 467
13047e4c 468 if (container->ss->load_super(container, mdfd, container_name)) {
549e9569 469 fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
13047e4c 470 container_name);
549e9569
NB
471 exit(3);
472 }
e8a70c89 473 close(mdfd);
549e9569 474
9fe32043
N
475 /* Ok, this is close enough. We can say goodbye to our parent now.
476 */
477 status = 0;
3d2c4fc7
DW
478 if (write(pfd[1], &status, sizeof(status)) < 0)
479 fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
480 getppid());
9fe32043
N
481 close(pfd[1]);
482
9fe32043
N
483 setsid();
484 close(0);
485 open("/dev/null", O_RDWR);
486 close(1);
3d2c4fc7 487 ignore = dup(0);
9fe32043
N
488#ifndef DEBUG
489 close(2);
3d2c4fc7 490 ignore = dup(0);
9fe32043
N
491#endif
492
549e9569
NB
493 mlockall(MCL_FUTURE);
494
3e70c845 495 if (clone_monitor(container) < 0) {
295646b3 496 fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
549e9569
NB
497 strerror(errno));
498 exit(2);
499 }
500
501 do_manager(container);
502
503 exit(0);
504}