]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - mdmon.c
2 * mdmon - monitor external metadata arrays
4 * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de>
5 * Copyright (C) 2007-2009 Intel Corporation
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.
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
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.
23 * When md arrays have user-space managed metadata, this is the program
24 * that does the managing.
26 * Given one argument: the name of the array (e.g. /dev/md0) that is
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.
34 * The main thread communicates with the priority thread by writing over
36 * Separate programs can communicate with the main thread via Unix-domain
38 * The two threads share address space and open file table.
48 #include <sys/types.h>
50 #include <sys/socket.h>
53 #include <sys/syscall.h>
69 char const Name
[] = "mdmon";
71 struct active_array
*discard_this
;
72 struct active_array
*pending_discard
;
79 static void *run_child(void *v
)
81 struct supertype
*c
= v
;
83 mon_tid
= syscall(SYS_gettid
);
88 static int clone_monitor(struct supertype
*container
)
95 pthread_attr_init(&attr
);
96 pthread_attr_setstacksize(&attr
, 4096);
97 pthread_attr_setdetachstate(&attr
, PTHREAD_CREATE_DETACHED
);
98 rc
= pthread_create(&thread
, &attr
, run_child
, container
);
101 while (mon_tid
== -1)
103 pthread_attr_destroy(&attr
);
105 mgr_tid
= syscall(SYS_gettid
);
109 #else /* USE_PTHREADS */
110 static int run_child(void *v
)
112 struct supertype
*c
= v
;
119 int __clone2(int (*fn
)(void *),
120 void *child_stack_base
, size_t stack_size
,
121 int flags
, void *arg
, ...
122 /* pid_t *pid, struct user_desc *tls, pid_t *ctid */ );
124 static int clone_monitor(struct supertype
*container
)
126 static char stack
[4096];
129 mon_tid
= __clone2(run_child
, stack
, sizeof(stack
),
130 CLONE_FS
|CLONE_FILES
|CLONE_VM
|CLONE_SIGHAND
|CLONE_THREAD
,
133 mon_tid
= clone(run_child
, stack
+4096-64,
134 CLONE_FS
|CLONE_FILES
|CLONE_VM
|CLONE_SIGHAND
|CLONE_THREAD
,
138 mgr_tid
= syscall(SYS_gettid
);
142 #endif /* USE_PTHREADS */
144 static int make_pidfile(char *devname
)
151 if (mkdir(MDMON_DIR
, 0755) < 0 &&
154 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devname
);
156 fd
= open(path
, O_RDWR
|O_CREAT
|O_EXCL
, 0600);
159 sprintf(pid
, "%d\n", getpid());
160 n
= write(fd
, pid
, strlen(pid
));
167 static void try_kill_monitor(pid_t pid
, char *devname
, int sock
)
175 /* first rule of survival... don't off yourself */
179 /* kill this process if it is mdmon */
180 sprintf(buf
, "/proc/%lu/cmdline", (unsigned long) pid
);
181 fd
= open(buf
, O_RDONLY
);
185 n
= read(fd
, buf
, sizeof(buf
)-1);
186 buf
[sizeof(buf
)-1] = 0;
189 if (n
< 0 || !(strstr(buf
, "mdmon") ||
190 strstr(buf
, "@dmon")))
198 /* Wait for monitor to exit by reading from the socket, after
199 * clearing the non-blocking flag */
200 fl
= fcntl(sock
, F_GETFL
, 0);
202 fcntl(sock
, F_SETFL
, fl
);
203 n
= read(sock
, buf
, 100);
205 /* If there is I/O going on it might took some time to get to
206 * clean state. Wait for monitor to exit fully to avoid races.
207 * Ping it with SIGUSR1 in case that it is sleeping */
208 for (n
= 0; n
< 25; n
++) {
209 rv
= kill(pid
, SIGUSR1
);
216 void remove_pidfile(char *devname
)
220 sprintf(buf
, "%s/%s.pid", MDMON_DIR
, devname
);
222 sprintf(buf
, "%s/%s.sock", MDMON_DIR
, devname
);
226 static int make_control_sock(char *devname
)
231 struct sockaddr_un addr
;
236 sprintf(path
, "%s/%s.sock", MDMON_DIR
, devname
);
238 sfd
= socket(PF_LOCAL
, SOCK_STREAM
, 0);
242 addr
.sun_family
= PF_LOCAL
;
243 strcpy(addr
.sun_path
, path
);
244 umask(077); /* ensure no world write access */
245 if (bind(sfd
, (struct sockaddr
*)&addr
, sizeof(addr
)) < 0) {
250 fl
= fcntl(sfd
, F_GETFL
, 0);
252 fcntl(sfd
, F_SETFL
, fl
);
256 static void term(int sig
)
261 static void wake_me(int sig
)
266 /* if we are debugging and starting mdmon by hand then don't fork */
267 static int do_fork(void)
270 if (check_env("MDADM_NO_MDMON"))
280 "Usage: mdmon [options] CONTAINER\n"
283 " --help -h : This message\n"
284 " --all -a : All devices\n"
285 " --foreground -F : Run in foreground (do not fork)\n"
286 " --takeover -t : Takeover container\n"
291 static bool is_duplicate_opt(const int opt
, const int set_val
, const char *long_name
)
293 if (opt
== set_val
) {
294 pr_err("--%s option duplicated!\n", long_name
);
300 static int mdmon(char *devnm
, int must_fork
, int takeover
);
302 int main(int argc
, char *argv
[])
304 char *container_name
= NULL
;
312 static struct option options
[] = {
313 {"all", 0, NULL
, 'a'},
314 {"takeover", 0, NULL
, 't'},
315 {"help", 0, NULL
, 'h'},
316 {"offroot", 0, NULL
, OffRootOpt
},
317 {"foreground", 0, NULL
, 'F'},
321 while ((opt
= getopt_long(argc
, argv
, "thaF", options
, NULL
)) != -1) {
324 if (is_duplicate_opt(all
, 1, "all"))
326 container_name
= argv
[optind
-1];
330 if (is_duplicate_opt(takeover
, 1, "takeover"))
335 if (is_duplicate_opt(dofork
, 0, "foreground"))
340 if (is_duplicate_opt(argv
[0][0], '@', "offroot"))
345 if (is_duplicate_opt(help
, true, "help"))
358 * set first char of argv[0] to @. This is used by
359 * systemd to signal that the task was launched from
360 * initrd/initramfs and should be preserved during shutdown
365 if (all
== 0 && container_name
== NULL
) {
367 container_name
= argv
[optind
];
370 if (container_name
== NULL
)
373 if (argc
- optind
> 1)
376 if (strcmp(container_name
, "/proc/mdstat") == 0)
383 struct mdstat_ent
*mdstat
, *e
;
384 int container_len
= strlen(container_name
);
386 /* launch an mdmon instance for each container found */
387 mdstat
= mdstat_read(0, 0);
388 for (e
= mdstat
; e
; e
= e
->next
) {
389 if (e
->metadata_version
&&
390 strncmp(e
->metadata_version
, "external:", 9) == 0 &&
391 !is_subarray(&e
->metadata_version
[9])) {
392 /* update cmdline so this mdmon instance can be
393 * distinguished from others in a call to ps(1)
395 if (strlen(e
->devnm
) <= (unsigned)container_len
) {
396 memset(container_name
, 0, container_len
);
397 sprintf(container_name
, "%s", e
->devnm
);
399 status
|= mdmon(e
->devnm
, 1, takeover
);
405 } else if (strncmp(container_name
, "md", 2) == 0) {
406 int id
= devnm2devid(container_name
);
408 devnm
= container_name
;
412 if (stat(container_name
, &st
) == 0)
413 devnm
= xstrdup(stat2devnm(&st
));
417 pr_err("%s is not a valid md device name\n",
421 return mdmon(devnm
, dofork
&& do_fork(), takeover
);
424 static int mdmon(char *devnm
, int must_fork
, int takeover
)
427 struct mdinfo
*mdi
, *di
;
428 struct supertype
*container
;
430 struct sigaction act
;
435 int victim_sock
= -1;
437 dprintf("starting mdmon for %s\n", devnm
);
439 mdfd
= open_dev(devnm
);
441 pr_err("%s: %s\n", devnm
, strerror(errno
));
445 /* Fork, and have the child tell us when they are ready */
447 if (pipe(pfd
) != 0) {
448 pr_err("failed to create pipe\n");
453 pr_err("failed to fork: %s\n", strerror(errno
));
458 default: /* parent */
460 if (read(pfd
[0], &status
, sizeof(status
)) != sizeof(status
)) {
462 status
= WEXITSTATUS(status
);
468 pfd
[0] = pfd
[1] = -1;
470 container
= xcalloc(1, sizeof(*container
));
471 strcpy(container
->devnm
, devnm
);
472 container
->arrays
= NULL
;
473 container
->sock
= -1;
475 mdi
= sysfs_read(mdfd
, container
->devnm
, GET_VERSION
|GET_LEVEL
|GET_DEVS
);
478 pr_err("failed to load sysfs info for %s\n", container
->devnm
);
481 if (mdi
->array
.level
!= UnSet
) {
482 pr_err("%s is not a container - cannot monitor\n", devnm
);
485 if (mdi
->array
.major_version
!= -1 ||
486 mdi
->array
.minor_version
!= -2) {
487 pr_err("%s does not use external metadata - cannot monitor\n",
492 container
->ss
= version_to_superswitch(mdi
->text_version
);
493 if (container
->ss
== NULL
) {
494 pr_err("%s uses unsupported metadata: %s\n",
495 devnm
, mdi
->text_version
);
499 container
->devs
= NULL
;
500 for (di
= mdi
->devs
; di
; di
= di
->next
) {
501 struct mdinfo
*cd
= xmalloc(sizeof(*cd
));
503 cd
->next
= container
->devs
;
504 container
->devs
= cd
;
508 /* SIGUSR is sent between parent and child. So both block it
509 * and enable it only with pselect.
512 sigaddset(&set
, SIGUSR1
);
513 sigaddset(&set
, SIGTERM
);
514 sigprocmask(SIG_BLOCK
, &set
, NULL
);
515 act
.sa_handler
= wake_me
;
517 sigaction(SIGUSR1
, &act
, NULL
);
518 act
.sa_handler
= term
;
519 sigaction(SIGTERM
, &act
, NULL
);
520 act
.sa_handler
= SIG_IGN
;
521 sigaction(SIGPIPE
, &act
, NULL
);
523 victim
= mdmon_pid(container
->devnm
);
525 victim_sock
= connect_monitor(container
->devnm
);
528 if (!takeover
&& victim
> 0 && victim_sock
>= 0) {
529 if (fping_monitor(victim_sock
) == 0) {
530 pr_err("%s already managed\n", container
->devnm
);
536 if (container
->ss
->load_container(container
, mdfd
, devnm
)) {
537 pr_err("Cannot load metadata for %s\n", devnm
);
542 /* Ok, this is close enough. We can say goodbye to our parent now.
545 remove_pidfile(devnm
);
546 if (make_pidfile(devnm
) < 0) {
549 container
->sock
= make_control_sock(devnm
);
553 if (write(pfd
[1], &status
, sizeof(status
)) < 0)
554 pr_err("failed to notify our parent: %d\n",
559 mlockall(MCL_CURRENT
| MCL_FUTURE
);
561 if (clone_monitor(container
) < 0) {
562 pr_err("failed to start monitor process: %s\n",
568 try_kill_monitor(victim
, container
->devnm
, victim_sock
);
569 if (victim_sock
>= 0)
576 /* This silliness is to stop the compiler complaining
577 * that we ignore 'ignore'
582 do_manager(container
);
587 /* Some stub functions so super-* can link with us */
588 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
589 struct supertype
*st
, unsigned long blocks
,
590 int *fds
, unsigned long long *offsets
,
591 int dests
, int *destfd
, unsigned long long *destoffsets
)
596 int restore_stripes(int *dest
, unsigned long long *offsets
,
597 int raid_disks
, int chunk_size
, int level
, int layout
,
598 int source
, unsigned long long read_offset
,
599 unsigned long long start
, unsigned long long length
,
605 int save_stripes(int *source
, unsigned long long *offsets
,
606 int raid_disks
, int chunk_size
, int level
, int layout
,
607 int nwrites
, int *dest
,
608 unsigned long long start
, unsigned long long length
,
614 struct superswitch super0
= {
617 struct superswitch super1
= {