]>
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>
67 struct active_array
*discard_this
;
68 struct active_array
*pending_discard
;
74 int run_child(void *v
)
76 struct supertype
*c
= v
;
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 */ );
88 int clone_monitor(struct supertype
*container
)
90 static char stack
[4096];
93 mon_tid
= __clone2(run_child
, stack
, sizeof(stack
),
94 CLONE_FS
|CLONE_FILES
|CLONE_VM
|CLONE_SIGHAND
|CLONE_THREAD
,
97 mon_tid
= clone(run_child
, stack
+4096-64,
98 CLONE_FS
|CLONE_FILES
|CLONE_VM
|CLONE_SIGHAND
|CLONE_THREAD
,
102 mgr_tid
= syscall(SYS_gettid
);
107 static struct superswitch
*find_metadata_methods(char *vers
)
109 if (strcmp(vers
, "ddf") == 0)
111 if (strcmp(vers
, "imsm") == 0)
116 static int make_pidfile(char *devname
)
123 if (mkdir(pid_dir
, 0700) < 0 &&
126 sprintf(path
, "%s/%s.pid", pid_dir
, devname
);
128 fd
= open(path
, O_RDWR
|O_CREAT
|O_EXCL
, 0600);
131 sprintf(pid
, "%d\n", getpid());
132 n
= write(fd
, pid
, strlen(pid
));
139 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
141 if (mdstat
->metadata_version
== NULL
||
142 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
143 !is_subarray(mdstat
->metadata_version
+9) ||
144 strncmp(mdstat
->metadata_version
+10, container
, strlen(container
)) != 0 ||
145 mdstat
->metadata_version
[10+strlen(container
)] != '/')
151 static void try_kill_monitor(pid_t pid
, char *devname
, int sock
)
158 /* first rule of survival... don't off yourself */
162 /* kill this process if it is mdmon */
163 sprintf(buf
, "/proc/%lu/cmdline", (unsigned long) pid
);
164 fd
= open(buf
, O_RDONLY
);
168 n
= read(fd
, buf
, sizeof(buf
)-1);
169 buf
[sizeof(buf
)-1] = 0;
172 if (n
< 0 || !strstr(buf
, "mdmon"))
177 /* Wait for monitor to exit by reading from the socket, after
178 * clearing the non-blocking flag */
179 fl
= fcntl(sock
, F_GETFL
, 0);
181 fcntl(sock
, F_SETFL
, fl
);
182 n
= read(sock
, buf
, 100);
183 /* Ignore result, it is just the wait that
188 void remove_pidfile(char *devname
)
192 sprintf(buf
, "%s/%s.pid", pid_dir
, devname
);
194 sprintf(buf
, "%s/%s.sock", pid_dir
, devname
);
196 if (strcmp(pid_dir
, ALT_RUN
) == 0)
197 /* try to clean up when we are finished with this dir */
201 static int make_control_sock(char *devname
)
206 struct sockaddr_un addr
;
211 sprintf(path
, "%s/%s.sock", pid_dir
, devname
);
213 sfd
= socket(PF_LOCAL
, SOCK_STREAM
, 0);
217 addr
.sun_family
= PF_LOCAL
;
218 strcpy(addr
.sun_path
, path
);
219 if (bind(sfd
, &addr
, sizeof(addr
)) < 0) {
224 fl
= fcntl(sfd
, F_GETFL
, 0);
226 fcntl(sfd
, F_SETFL
, fl
);
230 static void term(int sig
)
235 static void wake_me(int sig
)
240 /* if we are debugging and starting mdmon by hand then don't fork */
241 static int do_fork(void)
244 if (check_env("MDADM_NO_MDMON"))
253 fprintf(stderr
, "Usage: mdmon [--all] [--takeover] CONTAINER\n");
257 static int mdmon(char *devname
, int devnum
, int must_fork
, int takeover
);
259 int main(int argc
, char *argv
[])
261 char *container_name
= NULL
;
269 for (arg
= 1; arg
< argc
; arg
++) {
270 if (strncmp(argv
[arg
], "--all",5) == 0 ||
271 strcmp(argv
[arg
], "/proc/mdstat") == 0) {
272 container_name
= argv
[arg
];
274 } else if (strcmp(argv
[arg
], "--takeover") == 0)
276 else if (container_name
== NULL
)
277 container_name
= argv
[arg
];
281 if (container_name
== NULL
)
285 struct mdstat_ent
*mdstat
, *e
;
286 int container_len
= strlen(container_name
);
288 /* launch an mdmon instance for each container found */
289 mdstat
= mdstat_read(0, 0);
290 for (e
= mdstat
; e
; e
= e
->next
) {
291 if (strncmp(e
->metadata_version
, "external:", 9) == 0 &&
292 !is_subarray(&e
->metadata_version
[9])) {
293 devname
= devnum2devname(e
->devnum
);
294 /* update cmdline so this mdmon instance can be
295 * distinguished from others in a call to ps(1)
297 if (strlen(devname
) <= container_len
) {
298 memset(container_name
, 0, container_len
);
299 sprintf(container_name
, "%s", devname
);
301 status
|= mdmon(devname
, e
->devnum
, 1,
308 } else if (strncmp(container_name
, "md", 2) == 0) {
309 devnum
= devname2devnum(container_name
);
310 devname
= devnum2devname(devnum
);
311 if (strcmp(container_name
, devname
) != 0)
317 if (stat(container_name
, &st
) == 0)
318 devnum
= stat2devnum(&st
);
319 if (devnum
== NoMdDev
)
322 devname
= devnum2devname(devnum
);
326 fprintf(stderr
, "mdmon: %s is not a valid md device name\n",
330 return mdmon(devname
, devnum
, do_fork(), takeover
);
333 static int mdmon(char *devname
, int devnum
, int must_fork
, int takeover
)
336 struct mdinfo
*mdi
, *di
;
337 struct supertype
*container
;
339 struct sigaction act
;
344 int victim_sock
= -1;
346 dprintf("starting mdmon for %s\n", devname
);
348 mdfd
= open_dev(devnum
);
350 fprintf(stderr
, "mdmon: %s: %s\n", devname
,
354 if (md_get_version(mdfd
) < 0) {
355 fprintf(stderr
, "mdmon: %s: Not an md device\n",
360 /* Fork, and have the child tell us when they are ready */
362 if (pipe(pfd
) != 0) {
363 fprintf(stderr
, "mdmon: failed to create pipe\n");
368 fprintf(stderr
, "mdmon: failed to fork: %s\n",
374 default: /* parent */
376 if (read(pfd
[0], &status
, sizeof(status
)) != sizeof(status
)) {
378 status
= WEXITSTATUS(status
);
383 pfd
[0] = pfd
[1] = -1;
385 container
= calloc(1, sizeof(*container
));
386 container
->devnum
= devnum
;
387 container
->devname
= devname
;
388 container
->arrays
= NULL
;
389 container
->subarray
[0] = 0;
390 container
->sock
= -1;
392 if (!container
->devname
) {
393 fprintf(stderr
, "mdmon: failed to allocate container name string\n");
397 mdi
= sysfs_read(mdfd
, container
->devnum
,
398 GET_VERSION
|GET_LEVEL
|GET_DEVS
|SKIP_GONE_DEVS
);
401 fprintf(stderr
, "mdmon: failed to load sysfs info for %s\n",
405 if (mdi
->array
.level
!= UnSet
) {
406 fprintf(stderr
, "mdmon: %s is not a container - cannot monitor\n",
410 if (mdi
->array
.major_version
!= -1 ||
411 mdi
->array
.minor_version
!= -2) {
412 fprintf(stderr
, "mdmon: %s does not use external metadata - cannot monitor\n",
417 container
->ss
= find_metadata_methods(mdi
->text_version
);
418 if (container
->ss
== NULL
) {
419 fprintf(stderr
, "mdmon: %s uses unknown metadata: %s\n",
420 devname
, mdi
->text_version
);
424 container
->devs
= NULL
;
425 for (di
= mdi
->devs
; di
; di
= di
->next
) {
426 struct mdinfo
*cd
= malloc(sizeof(*cd
));
428 cd
->next
= container
->devs
;
429 container
->devs
= cd
;
433 /* SIGUSR is sent between parent and child. So both block it
434 * and enable it only with pselect.
437 sigaddset(&set
, SIGUSR1
);
438 sigaddset(&set
, SIGTERM
);
439 sigprocmask(SIG_BLOCK
, &set
, NULL
);
440 act
.sa_handler
= wake_me
;
442 sigaction(SIGUSR1
, &act
, NULL
);
443 act
.sa_handler
= term
;
444 sigaction(SIGTERM
, &act
, NULL
);
445 act
.sa_handler
= SIG_IGN
;
446 sigaction(SIGPIPE
, &act
, NULL
);
449 victim
= mdmon_pid(container
->devnum
);
452 victim
= mdmon_pid(container
->devnum
);
455 victim_sock
= connect_monitor(container
->devname
);
458 if (!takeover
&& victim
> 0 && victim_sock
>= 0) {
459 if (fping_monitor(victim_sock
) == 0) {
460 fprintf(stderr
, "mdmon: %s already managed\n",
466 if (container
->ss
->load_super(container
, mdfd
, devname
)) {
467 fprintf(stderr
, "mdmon: Cannot load metadata for %s\n",
473 /* Ok, this is close enough. We can say goodbye to our parent now.
476 remove_pidfile(devname
);
478 if (make_pidfile(devname
) < 0) {
479 /* Try the alternate */
481 if (make_pidfile(devname
) < 0) {
482 fprintf(stderr
, "mdmon: Neither %s nor %s are writable\n"
483 " cannot create .pid or .sock files. Aborting\n",
488 container
->sock
= make_control_sock(devname
);
491 if (write(pfd
[1], &status
, sizeof(status
)) < 0)
492 fprintf(stderr
, "mdmon: failed to notify our parent: %d\n",
496 mlockall(MCL_CURRENT
| MCL_FUTURE
);
498 if (clone_monitor(container
) < 0) {
499 fprintf(stderr
, "mdmon: failed to start monitor process: %s\n",
505 try_kill_monitor(victim
, container
->devname
, victim_sock
);
511 open("/dev/null", O_RDWR
);
519 do_manager(container
);