]>
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 sprintf(path
, "%s/%s.pid", pid_dir
, devname
);
125 fd
= open(path
, O_RDWR
|O_CREAT
|O_EXCL
, 0600);
128 sprintf(pid
, "%d\n", getpid());
129 n
= write(fd
, pid
, strlen(pid
));
136 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
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
)] != '/')
148 static void try_kill_monitor(pid_t pid
, char *devname
, int sock
)
155 /* first rule of survival... don't off yourself */
159 /* kill this process if it is mdmon */
160 sprintf(buf
, "/proc/%lu/cmdline", (unsigned long) pid
);
161 fd
= open(buf
, O_RDONLY
);
165 n
= read(fd
, buf
, sizeof(buf
)-1);
166 buf
[sizeof(buf
)-1] = 0;
169 if (n
< 0 || !strstr(buf
, "mdmon"))
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);
178 fcntl(sock
, F_SETFL
, fl
);
179 read(sock
, buf
, 100);
182 void remove_pidfile(char *devname
)
186 sprintf(buf
, "%s/%s.pid", pid_dir
, devname
);
188 sprintf(buf
, "%s/%s.sock", pid_dir
, devname
);
190 if (strcmp(pid_dir
, ALT_RUN
) == 0)
191 /* try to clean up when we are finished with this dir */
195 static int make_control_sock(char *devname
)
200 struct sockaddr_un addr
;
205 sprintf(path
, "%s/%s.sock", pid_dir
, devname
);
207 sfd
= socket(PF_LOCAL
, SOCK_STREAM
, 0);
211 addr
.sun_family
= PF_LOCAL
;
212 strcpy(addr
.sun_path
, path
);
213 if (bind(sfd
, &addr
, sizeof(addr
)) < 0) {
218 fl
= fcntl(sfd
, F_GETFL
, 0);
220 fcntl(sfd
, F_SETFL
, fl
);
224 static void term(int sig
)
229 static void wake_me(int sig
)
234 /* if we are debugging and starting mdmon by hand then don't fork */
235 static int do_fork(void)
238 if (check_env("MDADM_NO_MDMON"))
247 fprintf(stderr
, "Usage: mdmon /device/name/for/container [target_dir]\n");
251 static int mdmon(char *devname
, int devnum
, int must_fork
, int takeover
);
253 int main(int argc
, char *argv
[])
255 char *container_name
= NULL
;
263 for (arg
= 1; arg
< argc
; arg
++) {
264 if (strcmp(argv
[arg
], "--all") == 0 ||
265 strcmp(argv
[arg
], "/proc/mdstat") == 0)
267 else if (strcmp(argv
[arg
], "--takeover") == 0)
269 else if (container_name
== NULL
)
270 container_name
= argv
[arg
];
276 struct mdstat_ent
*mdstat
, *e
;
278 /* launch an mdmon instance for each container found */
279 mdstat
= mdstat_read(0, 0);
280 for (e
= mdstat
; e
; e
= e
->next
) {
281 if (strncmp(e
->metadata_version
, "external:", 9) == 0 &&
282 !is_subarray(&e
->metadata_version
[9])) {
283 devname
= devnum2devname(e
->devnum
);
284 /* update cmdline so this mdmon instance can be
285 * distinguished from others in a call to ps(1)
287 if (strlen(devname
) <= strlen(container_name
)) {
288 memset(container_name
, 0, strlen(container_name
));
289 sprintf(container_name
, "%s", devname
);
291 status
|= mdmon(devname
, e
->devnum
, 1,
298 } else if (strncmp(container_name
, "md", 2) == 0) {
299 devnum
= devname2devnum(container_name
);
300 devname
= devnum2devname(devnum
);
301 if (strcmp(container_name
, devname
) != 0)
307 if (stat(container_name
, &st
) == 0)
308 devnum
= stat2devnum(&st
);
309 if (devnum
== NoMdDev
)
312 devname
= devnum2devname(devnum
);
316 fprintf(stderr
, "mdmon: %s is not a valid md device name\n",
320 return mdmon(devname
, devnum
, do_fork(), takeover
);
323 static int mdmon(char *devname
, int devnum
, int must_fork
, int takeover
)
326 struct mdinfo
*mdi
, *di
;
327 struct supertype
*container
;
329 struct sigaction act
;
334 int victim_sock
= -1;
336 dprintf("starting mdmon for %s\n", devname
);
338 mdfd
= open_dev(devnum
);
340 fprintf(stderr
, "mdmon: %s: %s\n", devname
,
344 if (md_get_version(mdfd
) < 0) {
345 fprintf(stderr
, "mdmon: %s: Not an md device\n",
350 /* Fork, and have the child tell us when they are ready */
352 if (pipe(pfd
) != 0) {
353 fprintf(stderr
, "mdmon: failed to create pipe\n");
358 fprintf(stderr
, "mdmon: failed to fork: %s\n",
364 default: /* parent */
366 if (read(pfd
[0], &status
, sizeof(status
)) != sizeof(status
)) {
368 status
= WEXITSTATUS(status
);
373 pfd
[0] = pfd
[1] = -1;
375 container
= calloc(1, sizeof(*container
));
376 container
->devnum
= devnum
;
377 container
->devname
= devname
;
378 container
->arrays
= NULL
;
379 container
->subarray
[0] = 0;
380 container
->sock
= -1;
382 if (!container
->devname
) {
383 fprintf(stderr
, "mdmon: failed to allocate container name string\n");
387 mdi
= sysfs_read(mdfd
, container
->devnum
,
388 GET_VERSION
|GET_LEVEL
|GET_DEVS
|SKIP_GONE_DEVS
);
391 fprintf(stderr
, "mdmon: failed to load sysfs info for %s\n",
395 if (mdi
->array
.level
!= UnSet
) {
396 fprintf(stderr
, "mdmon: %s is not a container - cannot monitor\n",
400 if (mdi
->array
.major_version
!= -1 ||
401 mdi
->array
.minor_version
!= -2) {
402 fprintf(stderr
, "mdmon: %s does not use external metadata - cannot monitor\n",
407 container
->ss
= find_metadata_methods(mdi
->text_version
);
408 if (container
->ss
== NULL
) {
409 fprintf(stderr
, "mdmon: %s uses unknown metadata: %s\n",
410 devname
, mdi
->text_version
);
414 container
->devs
= NULL
;
415 for (di
= mdi
->devs
; di
; di
= di
->next
) {
416 struct mdinfo
*cd
= malloc(sizeof(*cd
));
418 cd
->next
= container
->devs
;
419 container
->devs
= cd
;
423 /* SIGUSR is sent between parent and child. So both block it
424 * and enable it only with pselect.
427 sigaddset(&set
, SIGUSR1
);
428 sigaddset(&set
, SIGTERM
);
429 sigprocmask(SIG_BLOCK
, &set
, NULL
);
430 act
.sa_handler
= wake_me
;
432 sigaction(SIGUSR1
, &act
, NULL
);
433 act
.sa_handler
= term
;
434 sigaction(SIGTERM
, &act
, NULL
);
435 act
.sa_handler
= SIG_IGN
;
436 sigaction(SIGPIPE
, &act
, NULL
);
440 victim
= mdmon_pid(container
->devnum
);
443 victim
= mdmon_pid(container
->devnum
);
446 victim_sock
= connect_monitor(container
->devname
);
451 if (ping_monitor(container
->devname
) == 0) {
452 fprintf(stderr
, "mdmon: %s already managed\n",
456 /* if there is a pid file, kill whoever is there just in case */
457 victim
= mdmon_pid(container
->devnum
);
459 if (container
->ss
->load_super(container
, mdfd
, devname
)) {
460 fprintf(stderr
, "mdmon: Cannot load metadata for %s\n",
466 /* Ok, this is close enough. We can say goodbye to our parent now.
469 remove_pidfile(devname
);
470 if (mkdir(VAR_RUN
, 0600) >= 0 || errno
== EEXIST
)
472 else if (mkdir(ALT_RUN
, 0600) >= 0 || errno
== EEXIST
)
475 fprintf(stderr
, "mdmon: Neither %s nor %s are writable\n"
476 " cannot create .pid or .sock files. Aborting\n",
480 if (make_pidfile(devname
) < 0) {
481 fprintf(stderr
, "mdmon: Cannot create pid file in %s - aborting.\n",
485 container
->sock
= make_control_sock(devname
);
488 if (write(pfd
[1], &status
, sizeof(status
)) < 0)
489 fprintf(stderr
, "mdmon: failed to notify our parent: %d\n",
493 mlockall(MCL_CURRENT
| MCL_FUTURE
);
495 if (clone_monitor(container
) < 0) {
496 fprintf(stderr
, "mdmon: failed to start monitor process: %s\n",
502 try_kill_monitor(victim
, container
->devname
, victim_sock
);
508 open("/dev/null", O_RDWR
);
516 do_manager(container
);