[thirdparty/mdadm.git] / mdmon.c

/*
 * mdmon - monitor external metadata arrays
 *
 * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de>
 * Copyright (C) 2007-2009 Intel Corporation
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms and conditions of the GNU General Public License,
 * version 2, as published by the Free Software Foundation.
 *
 * This program is distributed in the hope it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 *
 * You should have received a copy of the GNU General Public License along with
 * this program; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
 */

/*
 * md array manager.
 * When md arrays have user-space managed metadata, this is the program
 * that does the managing.
 *
 * Given one argument: the name of the array (e.g. /dev/md0) that is
 * the container.
 * We fork off a helper that runs high priority and mlocked.  It responds to
 * device failures and other events that might stop writeout, or that are
 * trivial to deal with.
 * The main thread then watches for new arrays being created in the container
 * and starts monitoring them too ... along with a few other tasks.
 *
 * The main thread communicates with the priority thread by writing over
 * a pipe.
 * Separate programs can communicate with the main thread via Unix-domain
 * socket.
 * The two threads share address space and open file table.
 *
 */

#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif

#include	<unistd.h>
#include	<stdlib.h>
#include	<sys/types.h>
#include	<sys/stat.h>
#include	<sys/socket.h>
#include	<sys/un.h>
#include	<sys/mman.h>
#include	<sys/syscall.h>
#include	<sys/wait.h>
#include	<stdio.h>
#include	<errno.h>
#include	<string.h>
#include	<fcntl.h>
#include	<signal.h>
#include	<dirent.h>

#include	<sched.h>

#include	"mdadm.h"
#include	"mdmon.h"

struct active_array *discard_this;
struct active_array *pending_discard;

int mon_tid, mgr_tid;

int sigterm;

int run_child(void *v)
{
	struct supertype *c = v;

	do_monitor(c);
	return 0;
}

#ifdef __ia64__
int __clone2(int (*fn)(void *),
	    void *child_stack_base, size_t stack_size,
	    int flags, void *arg, ...
	 /* pid_t *pid, struct user_desc *tls, pid_t *ctid */ );
#endif
 int clone_monitor(struct supertype *container)
{
	static char stack[4096];

#ifdef __ia64__
	mon_tid = __clone2(run_child, stack, sizeof(stack),
		   CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD,
		   container);
#else
	mon_tid = clone(run_child, stack+4096-64,
		   CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD,
		   container);
#endif

	mgr_tid = syscall(SYS_gettid);

	return mon_tid;
}

static struct superswitch *find_metadata_methods(char *vers)
{
	if (strcmp(vers, "ddf") == 0)
		return &super_ddf;
	if (strcmp(vers, "imsm") == 0)
		return &super_imsm;
	return NULL;
}

static int test_pidfile(char *devname)
{
	char path[100];
	struct stat st;

	sprintf(path, "/var/run/mdadm/%s.pid", devname);
	return stat(path, &st);
}

int make_pidfile(char *devname, int o_excl)
{
	char path[100];
	char pid[10];
	int fd;
	int n;

	if (sigterm)
		return -1;

	sprintf(path, "/var/run/mdadm/%s.pid", devname);

	fd = open(path, O_RDWR|O_CREAT|o_excl, 0600);
	if (fd < 0)
		return -errno;
	sprintf(pid, "%d\n", getpid());
	n = write(fd, pid, strlen(pid));
	close(fd);
	if (n < 0)
		return -errno;
	return 0;
}

int is_container_member(struct mdstat_ent *mdstat, char *container)
{
	if (mdstat->metadata_version == NULL ||
	    strncmp(mdstat->metadata_version, "external:", 9) != 0 ||
	    !is_subarray(mdstat->metadata_version+9) ||
	    strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 ||
	    mdstat->metadata_version[10+strlen(container)] != '/')
		return 0;

	return 1;
}

pid_t devname2mdmon(char *devname)
{
	char buf[100];
	pid_t pid = -1;
	int fd;

	sprintf(buf, "/var/run/mdadm/%s.pid", devname);
	fd = open(buf, O_RDONLY|O_NOATIME);
	if (fd < 0)
		return -1;

	if (read(fd, buf, sizeof(buf)) > 0)
		sscanf(buf, "%d\n", &pid);
	close(fd);

	return pid;
}

static void try_kill_monitor(pid_t pid, char *devname)
{
	char buf[100];
	int fd;
	struct mdstat_ent *mdstat;

	/* first rule of survival... don't off yourself */
	if (pid == getpid())
		return;

	/* kill this process if it is mdmon */
	sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid);
	fd = open(buf, O_RDONLY);
	if (fd < 0)
		return;

	if (read(fd, buf, sizeof(buf)) < 0) {
		close(fd);
		return;
	}

	if (!strstr(buf, "mdmon"))
		return;

	kill(pid, SIGTERM);

	mdstat = mdstat_read(0, 0);
	for ( ; mdstat; mdstat = mdstat->next)
		if (is_container_member(mdstat, devname)) {
			sprintf(buf, "/dev/%s", mdstat->dev);
			WaitClean(buf, 0);
		}
	free_mdstat(mdstat);
}

void remove_pidfile(char *devname)
{
	char buf[100];

	if (sigterm)
		return;

	sprintf(buf, "/var/run/mdadm/%s.pid", devname);
	unlink(buf);
	sprintf(buf, "/var/run/mdadm/%s.sock", devname);
	unlink(buf);
}

int make_control_sock(char *devname)
{
	char path[100];
	int sfd;
	long fl;
	struct sockaddr_un addr;

	if (sigterm)
		return -1;

	sprintf(path, "/var/run/mdadm/%s.sock", devname);
	unlink(path);
	sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
	if (sfd < 0)
		return -1;

	addr.sun_family = PF_LOCAL;
	strcpy(addr.sun_path, path);
	if (bind(sfd, &addr, sizeof(addr)) < 0) {
		close(sfd);
		return -1;
	}
	listen(sfd, 10);
	fl = fcntl(sfd, F_GETFL, 0);
	fl |= O_NONBLOCK;
	fcntl(sfd, F_SETFL, fl);
	return sfd;
}

int socket_hup_requested;
static void hup(int sig)
{
	socket_hup_requested = 1;
}

static void term(int sig)
{
	sigterm = 1;
}

static void wake_me(int sig)
{

}

/* if we are debugging and starting mdmon by hand then don't fork */
static int do_fork(void)
{
	#ifdef DEBUG
	if (check_env("MDADM_NO_MDMON"))
		return 0;
	#endif

	return 1;
}

void usage(void)
{
	fprintf(stderr, "Usage: mdmon /device/name/for/container [target_dir]\n");
	exit(2);
}

int mdmon(char *devname, int devnum, int scan, char *switchroot);

int main(int argc, char *argv[])
{
	char *container_name = NULL;
	char *switchroot = NULL;
	int devnum;
	char *devname;
	int scan = 0;
	int status = 0;

	switch (argc) {
	case 3:
		switchroot = argv[2];
	case 2:
		container_name = argv[1];
		break;
	default:
		usage();
	}

	if (strcmp(container_name, "/proc/mdstat") == 0) {
		struct mdstat_ent *mdstat, *e;

		/* launch an mdmon instance for each container found */
		scan = 1;
		mdstat = mdstat_read(0, 0);
		for (e = mdstat; e; e = e->next) {
			if (strncmp(e->metadata_version, "external:", 9) == 0 &&
			    !is_subarray(&e->metadata_version[9])) {
				devname = devnum2devname(e->devnum);
				/* update cmdline so this mdmon instance can be
				 * distinguished from others in a call to ps(1)
				 */
				if (strlen(devname) <= strlen(container_name)) {
					memset(container_name, 0, strlen(container_name));
					sprintf(container_name, "%s", devname);
				}
				status |= mdmon(devname, e->devnum, scan,
						switchroot);
			}
		}
		free_mdstat(mdstat);

		return status;
	} else if (strncmp(container_name, "md", 2) == 0) {
		devnum = devname2devnum(container_name);
		devname = devnum2devname(devnum);
		if (strcmp(container_name, devname) != 0)
			devname = NULL;
	} else {
		struct stat st;

		devnum = NoMdDev;
		if (stat(container_name, &st) == 0)
			devnum = stat2devnum(&st);
		if (devnum == NoMdDev)
			devname = NULL;
		else
			devname = devnum2devname(devnum);
	}

	if (!devname) {
		fprintf(stderr, "mdmon: %s is not a valid md device name\n",
			container_name);
		exit(1);
	}
	return mdmon(devname, devnum, scan, switchroot);
}

int mdmon(char *devname, int devnum, int scan, char *switchroot)
{
	int mdfd;
	struct mdinfo *mdi, *di;
	struct supertype *container;
	sigset_t set;
	struct sigaction act;
	int pfd[2];
	int status;
	int ignore;
	pid_t victim = -1;

	dprintf("starting mdmon for %s in %s\n",
		devname, switchroot ? : "/");

	/* try to spawn mdmon instances from the target file system */
	if (switchroot && strcmp(switchroot, "/") != 0) {
		char path[1024];
		pid_t pid;

		sprintf(path, "%s/sbin/mdmon", switchroot);
		switch (fork()) {
		case 0:
			execl(path, "mdmon", devname, NULL);
			exit(1);
		case -1:
			return 1;
		default:
			pid = wait(&status);
			if (pid > -1 && WIFEXITED(status) &&
			    WEXITSTATUS(status) == 0)
				return 0;
			else
				return 1;
		}
	}

	mdfd = open_dev(devnum);
	if (mdfd < 0) {
		fprintf(stderr, "mdmon: %s: %s\n", devname,
			strerror(errno));
		return 1;
	}
	if (md_get_version(mdfd) < 0) {
		fprintf(stderr, "mdmon: %s: Not an md device\n",
			devname);
		return 1;
	}

	/* Fork, and have the child tell us when they are ready */
	if (do_fork() || scan) {
		if (pipe(pfd) != 0) {
			fprintf(stderr, "mdmon: failed to create pipe\n");
			return 1;
		}
		switch(fork()) {
		case -1:
			fprintf(stderr, "mdmon: failed to fork: %s\n",
				strerror(errno));
			return 1;
		case 0: /* child */
			close(pfd[0]);
			break;
		default: /* parent */
			close(pfd[1]);
			if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) {
				wait(&status);
				status = WEXITSTATUS(status);
			}
			return status;
		}
	} else
		pfd[0] = pfd[1] = -1;

	container = calloc(1, sizeof(*container));
	container->devnum = devnum;
	container->devname = devname;
	container->arrays = NULL;
	container->subarray[0] = 0;
	container->sock = -1;

	if (!container->devname) {
		fprintf(stderr, "mdmon: failed to allocate container name string\n");
		exit(3);
	}

	mdi = sysfs_read(mdfd, container->devnum,
			 GET_VERSION|GET_LEVEL|GET_DEVS|SKIP_GONE_DEVS);

	if (!mdi) {
		fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
			container->devname);
		exit(3);
	}
	if (mdi->array.level != UnSet) {
		fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
			devname);
		exit(3);
	}
	if (mdi->array.major_version != -1 ||
	    mdi->array.minor_version != -2) {
		fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
			devname);
		exit(3);
	}

	container->ss = find_metadata_methods(mdi->text_version);
	if (container->ss == NULL) {
		fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
			devname, mdi->text_version);
		exit(3);
	}

	container->devs = NULL;
	for (di = mdi->devs; di; di = di->next) {
		struct mdinfo *cd = malloc(sizeof(*cd));
		*cd = *di;
		cd->next = container->devs;
		container->devs = cd;
	}
	sysfs_free(mdi);

	/* SIGUSR is sent between parent and child.  So both block it
	 * and enable it only with pselect.
	 */
	sigemptyset(&set);
	sigaddset(&set, SIGUSR1);
	sigaddset(&set, SIGHUP);
	sigaddset(&set, SIGALRM);
	sigaddset(&set, SIGTERM);
	sigprocmask(SIG_BLOCK, &set, NULL);
	act.sa_handler = wake_me;
	act.sa_flags = 0;
	sigaction(SIGUSR1, &act, NULL);
	sigaction(SIGALRM, &act, NULL);
	act.sa_handler = hup;
	sigaction(SIGHUP, &act, NULL);
	act.sa_handler = term;
	sigaction(SIGTERM, &act, NULL);
	act.sa_handler = SIG_IGN;
	sigaction(SIGPIPE, &act, NULL);

	if (switchroot) {
		/* we assume we assume that /sys /proc /dev are available in
		 * the new root
		 */
		victim = devname2mdmon(container->devname);
		if (chroot(switchroot) != 0) {
			fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
				switchroot, strerror(errno));
			exit(4);
		}
	}

	ignore = chdir("/");
	if (victim < 0 && test_pidfile(container->devname) == 0) {
		if (ping_monitor(container->devname) == 0) {
			fprintf(stderr, "mdmon: %s already managed\n",
				container->devname);
			exit(3);
		} else if (victim < 0)
			victim = devname2mdmon(container->devname);
	}
	if (container->ss->load_super(container, mdfd, devname)) {
		fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
			devname);
		exit(3);
	}
	close(mdfd);

	/* Ok, this is close enough.  We can say goodbye to our parent now.
	 */
	status = 0;
	if (write(pfd[1], &status, sizeof(status)) < 0)
		fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
			getppid());
	close(pfd[1]);

	setsid();
	close(0);
	open("/dev/null", O_RDWR);
	close(1);
	ignore = dup(0);
#ifndef DEBUG
	close(2);
	ignore = dup(0);
#endif

	mlockall(MCL_FUTURE);

	if (clone_monitor(container) < 0) {
		fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
			strerror(errno));
		exit(2);
	}

	if (victim > -1)
		try_kill_monitor(victim, container->devname);
	do_manager(container);

	exit(0);
}
Commit	Line	Data
a54d5262 DW	1	/*
	2	* mdmon - monitor external metadata arrays
	3	*
e736b623 N	4	* Copyright (C) 2007-2009 Neil Brown <neilb@suse.de>
e736b623 N	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	67	struct active_array *discard_this;
549e9569 NB	68	struct active_array *pending_discard;
4d43913c NB	69
4d43913c NB	70	int mon_tid, mgr_tid;
549e9569	71
6144ed44 DW	72	int sigterm;
6144ed44 DW	73
549e9569 NB	74	int 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__
	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 / );
	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,
549e9569 NB	99	container);
97f734fd	100	#endif
3e70c845	101
4d43913c	102	mgr_tid = syscall(SYS_gettid);
2cc98f9e DW	103
2cc98f9e DW	104	return mon_tid;
549e9569 NB	105	}
549e9569 NB	106
ce744c97 DW	107	static 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
96a8270d DW	116	static int test_pidfile(char *devname)
	117	{
	118	char path[100];
	119	struct stat st;
	120
	121	sprintf(path, "/var/run/mdadm/%s.pid", devname);
	122	return stat(path, &st);
	123	}
549e9569	124
295646b3	125	int make_pidfile(char *devname, int o_excl)
549e9569 NB	126	{
	127	char path[100];
	128	char pid[10];
	129	int fd;
3d2c4fc7 DW	130	int n;
3d2c4fc7 DW	131
6144ed44 DW	132	if (sigterm)
	133	return -1;
	134
549e9569 NB	135	sprintf(path, "/var/run/mdadm/%s.pid", devname);
549e9569 NB	136
b109d928	137	fd = open(path, O_RDWR\|O_CREAT\|o_excl, 0600);
549e9569	138	if (fd < 0)
295646b3	139	return -errno;
549e9569	140	sprintf(pid, "%d\n", getpid());
3d2c4fc7	141	n = write(fd, pid, strlen(pid));
549e9569	142	close(fd);
3d2c4fc7 DW	143	if (n < 0)
3d2c4fc7 DW	144	return -errno;
549e9569 NB	145	return 0;
	146	}
	147
883a6142 DW	148	int is_container_member(struct mdstat_ent mdstat, char container)
	149	{
	150	if (mdstat->metadata_version == NULL \|\|
	151	strncmp(mdstat->metadata_version, "external:", 9) != 0 \|\|
	152	!is_subarray(mdstat->metadata_version+9) \|\|
	153	strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 \|\|
	154	mdstat->metadata_version[10+strlen(container)] != '/')
	155	return 0;
	156
	157	return 1;
	158	}
	159
96a8270d	160	pid_t devname2mdmon(char *devname)
b109d928 DW	161	{
b109d928 DW	162	char buf[100];
96a8270d	163	pid_t pid = -1;
b109d928	164	int fd;
b109d928 DW	165
b109d928 DW	166	sprintf(buf, "/var/run/mdadm/%s.pid", devname);
96a8270d	167	fd = open(buf, O_RDONLY\|O_NOATIME);
b109d928	168	if (fd < 0)
96a8270d	169	return -1;
b109d928	170
96a8270d DW	171	if (read(fd, buf, sizeof(buf)) > 0)
96a8270d DW	172	sscanf(buf, "%d\n", &pid);
b109d928	173	close(fd);
96a8270d DW	174
	175	return pid;
	176	}
	177
	178	static void try_kill_monitor(pid_t pid, char *devname)
	179	{
	180	char buf[100];
	181	int fd;
	182	struct mdstat_ent *mdstat;
b109d928	183
8aae4219 DW	184	/* first rule of survival... don't off yourself */
	185	if (pid == getpid())
	186	return;
	187
b109d928 DW	188	/* kill this process if it is mdmon */
	189	sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid);
	190	fd = open(buf, O_RDONLY);
	191	if (fd < 0)
	192	return;
	193
	194	if (read(fd, buf, sizeof(buf)) < 0) {
	195	close(fd);
	196	return;
	197	}
	198
883a6142 DW	199	if (!strstr(buf, "mdmon"))
	200	return;
	201
	202	kill(pid, SIGTERM);
	203
	204	mdstat = mdstat_read(0, 0);
	205	for ( ; mdstat; mdstat = mdstat->next)
	206	if (is_container_member(mdstat, devname)) {
	207	sprintf(buf, "/dev/%s", mdstat->dev);
27dec8fa	208	WaitClean(buf, 0);
883a6142 DW	209	}
883a6142 DW	210	free_mdstat(mdstat);
b109d928 DW	211	}
b109d928 DW	212
e0d6609f NB	213	void remove_pidfile(char *devname)
	214	{
	215	char buf[100];
	216
6144ed44 DW	217	if (sigterm)
	218	return;
	219
e0d6609f NB	220	sprintf(buf, "/var/run/mdadm/%s.pid", devname);
e0d6609f NB	221	unlink(buf);
57752795 N	222	sprintf(buf, "/var/run/mdadm/%s.sock", devname);
57752795 N	223	unlink(buf);
e0d6609f NB	224	}
e0d6609f NB	225
295646b3	226	int make_control_sock(char *devname)
549e9569 NB	227	{
	228	char path[100];
	229	int sfd;
	230	long fl;
	231	struct sockaddr_un addr;
	232
6144ed44 DW	233	if (sigterm)
	234	return -1;
	235
549e9569 NB	236	sprintf(path, "/var/run/mdadm/%s.sock", devname);
	237	unlink(path);
	238	sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
	239	if (sfd < 0)
	240	return -1;
	241
	242	addr.sun_family = PF_LOCAL;
	243	strcpy(addr.sun_path, path);
	244	if (bind(sfd, &addr, sizeof(addr)) < 0) {
	245	close(sfd);
	246	return -1;
	247	}
	248	listen(sfd, 10);
	249	fl = fcntl(sfd, F_GETFL, 0);
	250	fl \|= O_NONBLOCK;
	251	fcntl(sfd, F_SETFL, fl);
	252	return sfd;
	253	}
	254
295646b3 DW	255	int socket_hup_requested;
	256	static void hup(int sig)
	257	{
	258	socket_hup_requested = 1;
	259	}
	260
6144ed44 DW	261	static void term(int sig)
	262	{
	263	sigterm = 1;
	264	}
	265
4d43913c NB	266	static void wake_me(int sig)
	267	{
	268
	269	}
	270
16ddab0d DW	271	/* if we are debugging and starting mdmon by hand then don't fork */
	272	static int do_fork(void)
	273	{
	274	#ifdef DEBUG
40ebbb9c	275	if (check_env("MDADM_NO_MDMON"))
16ddab0d DW	276	return 0;
	277	#endif
	278
	279	return 1;
	280	}
	281
13047e4c DW	282	void usage(void)
13047e4c DW	283	{
5746141e	284	fprintf(stderr, "Usage: mdmon /device/name/for/container [target_dir]\n");
13047e4c DW	285	exit(2);
13047e4c DW	286	}
16ddab0d	287
1ffd2840 DW	288	int mdmon(char devname, int devnum, int scan, char switchroot);
1ffd2840 DW	289
549e9569 NB	290	int main(int argc, char *argv[])
549e9569 NB	291	{
13047e4c DW	292	char *container_name = NULL;
13047e4c DW	293	char *switchroot = NULL;
e8a70c89 N	294	int devnum;
e8a70c89 N	295	char *devname;
1ffd2840 DW	296	int scan = 0;
1ffd2840 DW	297	int status = 0;
13047e4c DW	298
13047e4c DW	299	switch (argc) {
5746141e DW	300	case 3:
5746141e DW	301	switchroot = argv[2];
13047e4c DW	302	case 2:
	303	container_name = argv[1];
	304	break;
13047e4c DW	305	default:
13047e4c DW	306	usage();
549e9569	307	}
13047e4c	308
1ffd2840 DW	309	if (strcmp(container_name, "/proc/mdstat") == 0) {
	310	struct mdstat_ent mdstat, e;
	311
	312	/* launch an mdmon instance for each container found */
	313	scan = 1;
	314	mdstat = mdstat_read(0, 0);
	315	for (e = mdstat; e; e = e->next) {
	316	if (strncmp(e->metadata_version, "external:", 9) == 0 &&
	317	!is_subarray(&e->metadata_version[9])) {
	318	devname = devnum2devname(e->devnum);
1b34f519 DW	319	/* update cmdline so this mdmon instance can be
	320	* distinguished from others in a call to ps(1)
	321	*/
	322	if (strlen(devname) <= strlen(container_name)) {
	323	memset(container_name, 0, strlen(container_name));
	324	sprintf(container_name, "%s", devname);
	325	}
1ffd2840 DW	326	status \|= mdmon(devname, e->devnum, scan,
	327	switchroot);
	328	}
	329	}
	330	free_mdstat(mdstat);
	331
	332	return status;
	333	} else if (strncmp(container_name, "md", 2) == 0) {
6f4098a6 DW	334	devnum = devname2devnum(container_name);
	335	devname = devnum2devname(devnum);
	336	if (strcmp(container_name, devname) != 0)
	337	devname = NULL;
	338	} else {
	339	struct stat st;
	340
	341	devnum = NoMdDev;
	342	if (stat(container_name, &st) == 0)
	343	devnum = stat2devnum(&st);
	344	if (devnum == NoMdDev)
	345	devname = NULL;
	346	else
	347	devname = devnum2devname(devnum);
	348	}
	349
	350	if (!devname) {
e8a70c89 N	351	fprintf(stderr, "mdmon: %s is not a valid md device name\n",
	352	container_name);
	353	exit(1);
	354	}
1ffd2840 DW	355	return mdmon(devname, devnum, scan, switchroot);
	356	}
	357
	358	int mdmon(char devname, int devnum, int scan, char switchroot)
	359	{
	360	int mdfd;
	361	struct mdinfo mdi, di;
	362	struct supertype *container;
	363	sigset_t set;
	364	struct sigaction act;
	365	int pfd[2];
	366	int status;
	367	int ignore;
96a8270d	368	pid_t victim = -1;
1ffd2840	369
5746141e DW	370	dprintf("starting mdmon for %s in %s\n",
5746141e DW	371	devname, switchroot ? : "/");
b928b5a0 DW	372
	373	/* try to spawn mdmon instances from the target file system */
	374	if (switchroot && strcmp(switchroot, "/") != 0) {
	375	char path[1024];
	376	pid_t pid;
	377
	378	sprintf(path, "%s/sbin/mdmon", switchroot);
	379	switch (fork()) {
	380	case 0:
	381	execl(path, "mdmon", devname, NULL);
	382	exit(1);
	383	case -1:
	384	return 1;
	385	default:
	386	pid = wait(&status);
	387	if (pid > -1 && WIFEXITED(status) &&
	388	WEXITSTATUS(status) == 0)
	389	return 0;
	390	else
	391	return 1;
	392	}
	393	}
	394
e8a70c89	395	mdfd = open_dev(devnum);
549e9569	396	if (mdfd < 0) {
1ffd2840	397	fprintf(stderr, "mdmon: %s: %s\n", devname,
549e9569	398	strerror(errno));
1ffd2840	399	return 1;
549e9569 NB	400	}
549e9569 NB	401	if (md_get_version(mdfd) < 0) {
13047e4c	402	fprintf(stderr, "mdmon: %s: Not an md device\n",
1ffd2840 DW	403	devname);
1ffd2840 DW	404	return 1;
549e9569 NB	405	}
549e9569 NB	406
9fe32043	407	/* Fork, and have the child tell us when they are ready */
1ffd2840	408	if (do_fork() \|\| scan) {
3d2c4fc7 DW	409	if (pipe(pfd) != 0) {
3d2c4fc7 DW	410	fprintf(stderr, "mdmon: failed to create pipe\n");
1ffd2840	411	return 1;
3d2c4fc7	412	}
16ddab0d DW	413	switch(fork()) {
	414	case -1:
	415	fprintf(stderr, "mdmon: failed to fork: %s\n",
	416	strerror(errno));
1ffd2840	417	return 1;
16ddab0d DW	418	case 0: /* child */
	419	close(pfd[0]);
	420	break;
	421	default: /* parent */
	422	close(pfd[1]);
	423	if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) {
	424	wait(&status);
	425	status = WEXITSTATUS(status);
	426	}
1ffd2840	427	return status;
9fe32043	428	}
16ddab0d DW	429	} else
16ddab0d DW	430	pfd[0] = pfd[1] = -1;
549e9569	431
f5df5d69	432	container = calloc(1, sizeof(*container));
e8a70c89 N	433	container->devnum = devnum;
e8a70c89 N	434	container->devname = devname;
13047e4c	435	container->arrays = NULL;
c1363b40	436	container->subarray[0] = 0;
96a8270d	437	container->sock = -1;
13047e4c DW	438
	439	if (!container->devname) {
	440	fprintf(stderr, "mdmon: failed to allocate container name string\n");
	441	exit(3);
	442	}
	443
	444	mdi = sysfs_read(mdfd, container->devnum,
7da80e6f	445	GET_VERSION\|GET_LEVEL\|GET_DEVS\|SKIP_GONE_DEVS);
13047e4c DW	446
	447	if (!mdi) {
	448	fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
	449	container->devname);
	450	exit(3);
	451	}
	452	if (mdi->array.level != UnSet) {
	453	fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
1ffd2840	454	devname);
13047e4c DW	455	exit(3);
	456	}
	457	if (mdi->array.major_version != -1 \|\|
	458	mdi->array.minor_version != -2) {
	459	fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
1ffd2840	460	devname);
13047e4c DW	461	exit(3);
	462	}
	463
	464	container->ss = find_metadata_methods(mdi->text_version);
	465	if (container->ss == NULL) {
	466	fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
1ffd2840	467	devname, mdi->text_version);
13047e4c DW	468	exit(3);
	469	}
	470
	471	container->devs = NULL;
	472	for (di = mdi->devs; di; di = di->next) {
	473	struct mdinfo cd = malloc(sizeof(cd));
	474	cd = di;
	475	cd->next = container->devs;
	476	container->devs = cd;
	477	}
	478	sysfs_free(mdi);
549e9569	479
883a6142 DW	480	/* SIGUSR is sent between parent and child. So both block it
	481	* and enable it only with pselect.
	482	*/
	483	sigemptyset(&set);
	484	sigaddset(&set, SIGUSR1);
	485	sigaddset(&set, SIGHUP);
	486	sigaddset(&set, SIGALRM);
	487	sigaddset(&set, SIGTERM);
	488	sigprocmask(SIG_BLOCK, &set, NULL);
	489	act.sa_handler = wake_me;
	490	act.sa_flags = 0;
	491	sigaction(SIGUSR1, &act, NULL);
	492	sigaction(SIGALRM, &act, NULL);
	493	act.sa_handler = hup;
	494	sigaction(SIGHUP, &act, NULL);
	495	act.sa_handler = term;
	496	sigaction(SIGTERM, &act, NULL);
	497	act.sa_handler = SIG_IGN;
	498	sigaction(SIGPIPE, &act, NULL);
	499
13047e4c DW	500	if (switchroot) {
13047e4c DW	501	/* we assume we assume that /sys /proc /dev are available in
96a8270d	502	* the new root
13047e4c	503	*/
96a8270d	504	victim = devname2mdmon(container->devname);
13047e4c DW	505	if (chroot(switchroot) != 0) {
	506	fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
	507	switchroot, strerror(errno));
	508	exit(4);
	509	}
	510	}
	511
13047e4c	512	ignore = chdir("/");
96a8270d	513	if (victim < 0 && test_pidfile(container->devname) == 0) {
b109d928 DW	514	if (ping_monitor(container->devname) == 0) {
	515	fprintf(stderr, "mdmon: %s already managed\n",
	516	container->devname);
	517	exit(3);
96a8270d DW	518	} else if (victim < 0)
96a8270d DW	519	victim = devname2mdmon(container->devname);
549e9569	520	}
1ffd2840	521	if (container->ss->load_super(container, mdfd, devname)) {
549e9569	522	fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
1ffd2840	523	devname);
549e9569 NB	524	exit(3);
549e9569 NB	525	}
e8a70c89	526	close(mdfd);
549e9569	527
9fe32043 N	528	/* Ok, this is close enough. We can say goodbye to our parent now.
	529	*/
	530	status = 0;
3d2c4fc7 DW	531	if (write(pfd[1], &status, sizeof(status)) < 0)
	532	fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
	533	getppid());
9fe32043 N	534	close(pfd[1]);
9fe32043 N	535
9fe32043 N	536	setsid();
	537	close(0);
	538	open("/dev/null", O_RDWR);
	539	close(1);
3d2c4fc7	540	ignore = dup(0);
9fe32043 N	541	#ifndef DEBUG
9fe32043 N	542	close(2);
3d2c4fc7	543	ignore = dup(0);
9fe32043 N	544	#endif
9fe32043 N	545
549e9569 NB	546	mlockall(MCL_FUTURE);
549e9569 NB	547
3e70c845	548	if (clone_monitor(container) < 0) {
295646b3	549	fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
549e9569 NB	550	strerror(errno));
	551	exit(2);
	552	}
	553
96a8270d DW	554	if (victim > -1)
96a8270d DW	555	try_kill_monitor(victim, container->devname);
549e9569 NB	556	do_manager(container);
	557
	558	exit(0);
	559	}