[thirdparty/mdadm.git] / mdmon.c


/*
 * 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	<sched.h>

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

struct active_array *discard_this;
struct active_array *pending_discard;

int mon_tid, mgr_tid;

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

	do_monitor(c);
	return 0;
}

int clone_monitor(struct supertype *container)
{
	static char stack[4096];

	mon_tid = clone(run_child, stack+4096-64,
		   CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD,
		   container);

	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;
}


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

	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;
}

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

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

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

	close(fd);
	pid = strtoul(buf, NULL, 10);

	/* 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") != NULL)
		kill(pid, SIGTERM);
}

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

	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;

	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 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 (env_no_mdmon())
		return 0;
	#endif

	return 1;
}


int main(int argc, char *argv[])
{
	int mdfd;
	struct mdinfo *mdi, *di;
	struct supertype *container;
	sigset_t set;
	struct sigaction act;
	int pfd[2];
	int status;
	int ignore;

	if (argc != 2) {
		fprintf(stderr, "Usage: md-manage /device/name/for/container\n");
		exit(2);
	}
	mdfd = open(argv[1], O_RDWR);
	if (mdfd < 0) {
		fprintf(stderr, "md-manage: %s: %s\n", argv[1],
			strerror(errno));
		exit(1);
	}
	if (md_get_version(mdfd) < 0) {
		fprintf(stderr, "md-manage: %s: Not an md device\n",
			argv[1]);
		exit(1);
	}

	/* Fork, and have the child tell us when they are ready */
	if (do_fork()) {
		if (pipe(pfd) != 0) {
			fprintf(stderr, "mdmon: failed to create pipe\n");
			exit(1);
		}
		switch(fork()) {
		case -1:
			fprintf(stderr, "mdmon: failed to fork: %s\n",
				strerror(errno));
			exit(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);
			}
			exit(status);
		}
	} else
		pfd[0] = pfd[1] = -1;
	/* hopefully it is a container - we'll check later */

	container = malloc(sizeof(*container));
	container->devnum = fd2devnum(mdfd);
	container->devname = devnum2devname(container->devnum);
	container->device_name = argv[1];

	/* If this fails, we hope it already exists */
	mkdir("/var/run/mdadm", 0600);
	/* pid file lives in /var/run/mdadm/mdXX.pid */
	if (make_pidfile(container->devname, O_EXCL) < 0) {
		if (ping_monitor(container->devname) == 0) {
			fprintf(stderr, "mdmon: %s already managed\n",
				container->devname);
			exit(3);
		} else {
			int err;

			/* cleanup the old monitor, this one is taking over */
			try_kill_monitor(container->devname);
			err = make_pidfile(container->devname, 0);
			if (err < 0) {
				fprintf(stderr, "mdmon: %s Cannot create pidfile\n",
					container->devname);
				if (err == -EROFS) {
					/* FIXME implement a mechanism to
					 * prevent duplicate monitor instances
					 */
					fprintf(stderr,
						"mdmon: continuing on read-only file system\n");
				} else
					exit(3);
			}
		}
	}

	container->sock = make_control_sock(container->devname);
	container->arrays = NULL;

	mdi = sysfs_read(mdfd, container->devnum,
			 GET_VERSION|GET_LEVEL|GET_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",
			argv[1]);
		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",
			argv[1]);
		exit(3);
	}

	container->ss = find_metadata_methods(mdi->text_version);
	if (container->ss == NULL) {
		fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
			argv[1], 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);


	if (container->ss->load_super(container, mdfd, argv[1])) {
		fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
			argv[1]);
		exit(3);
	}

	/* 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]);

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

	mlockall(MCL_FUTURE);

	/* 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);
	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 = SIG_IGN;
	sigaction(SIGPIPE, &act, NULL);

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

	do_manager(container);

	exit(0);
}
Commit	Line	Data
549e9569 NB	1
	2	/*
	3	* md array manager.
	4	* When md arrays have user-space managed metadata, this is the program
	5	* that does the managing.
	6	*
	7	* Given one argument: the name of the array (e.g. /dev/md0) that is
	8	* the container.
	9	* We fork off a helper that runs high priority and mlocked. It responds to
	10	* device failures and other events that might stop writeout, or that are
	11	* trivial to deal with.
	12	* The main thread then watches for new arrays being created in the container
	13	* and starts monitoring them too ... along with a few other tasks.
	14	*
	15	* The main thread communicates with the priority thread by writing over
	16	* a pipe.
	17	* Separate programs can communicate with the main thread via Unix-domain
	18	* socket.
	19	* The two threads share address space and open file table.
	20	*
	21	*/
	22
	23	#ifndef _GNU_SOURCE
	24	#define _GNU_SOURCE
	25	#endif
	26
	27	#include <unistd.h>
	28	#include <stdlib.h>
4d43913c	29	#include <sys/types.h>
549e9569 NB	30	#include <sys/stat.h>
	31	#include <sys/socket.h>
	32	#include <sys/un.h>
	33	#include <sys/mman.h>
4d43913c	34	#include <sys/syscall.h>
9fe32043	35	#include <sys/wait.h>
549e9569 NB	36	#include <stdio.h>
	37	#include <errno.h>
	38	#include <string.h>
	39	#include <fcntl.h>
b109d928	40	#include <signal.h>
549e9569 NB	41
	42	#include <sched.h>
	43
	44	#include "mdadm.h"
	45	#include "mdmon.h"
	46
549e9569 NB	47	struct active_array *discard_this;
549e9569 NB	48	struct active_array *pending_discard;
4d43913c NB	49
4d43913c NB	50	int mon_tid, mgr_tid;
549e9569 NB	51
	52	int run_child(void *v)
	53	{
	54	struct supertype *c = v;
1ed3f387	55
549e9569 NB	56	do_monitor(c);
	57	return 0;
	58	}
	59
	60	int clone_monitor(struct supertype *container)
	61	{
549e9569	62	static char stack[4096];
549e9569	63
2cc98f9e	64	mon_tid = clone(run_child, stack+4096-64,
549e9569 NB	65	CLONE_FS\|CLONE_FILES\|CLONE_VM\|CLONE_SIGHAND\|CLONE_THREAD,
549e9569 NB	66	container);
3e70c845	67
4d43913c	68	mgr_tid = syscall(SYS_gettid);
2cc98f9e DW	69
2cc98f9e DW	70	return mon_tid;
549e9569 NB	71	}
	72
	73	static struct superswitch find_metadata_methods(char vers)
	74	{
	75	if (strcmp(vers, "ddf") == 0)
	76	return &super_ddf;
5b65005f DW	77	if (strcmp(vers, "imsm") == 0)
5b65005f DW	78	return &super_imsm;
549e9569 NB	79	return NULL;
	80	}
	81
	82
295646b3	83	int make_pidfile(char *devname, int o_excl)
549e9569 NB	84	{
	85	char path[100];
	86	char pid[10];
	87	int fd;
3d2c4fc7 DW	88	int n;
3d2c4fc7 DW	89
549e9569 NB	90	sprintf(path, "/var/run/mdadm/%s.pid", devname);
549e9569 NB	91
b109d928	92	fd = open(path, O_RDWR\|O_CREAT\|o_excl, 0600);
549e9569	93	if (fd < 0)
295646b3	94	return -errno;
549e9569	95	sprintf(pid, "%d\n", getpid());
3d2c4fc7	96	n = write(fd, pid, strlen(pid));
549e9569	97	close(fd);
3d2c4fc7 DW	98	if (n < 0)
3d2c4fc7 DW	99	return -errno;
549e9569 NB	100	return 0;
	101	}
	102
b109d928 DW	103	static void try_kill_monitor(char *devname)
	104	{
	105	char buf[100];
	106	int fd;
	107	pid_t pid;
	108
	109	sprintf(buf, "/var/run/mdadm/%s.pid", devname);
	110	fd = open(buf, O_RDONLY);
	111	if (fd < 0)
	112	return;
	113
	114	if (read(fd, buf, sizeof(buf)) < 0) {
	115	close(fd);
	116	return;
	117	}
	118
	119	close(fd);
	120	pid = strtoul(buf, NULL, 10);
	121
	122	/* kill this process if it is mdmon */
	123	sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid);
	124	fd = open(buf, O_RDONLY);
	125	if (fd < 0)
	126	return;
	127
	128	if (read(fd, buf, sizeof(buf)) < 0) {
	129	close(fd);
	130	return;
	131	}
	132
	133	if (strstr(buf, "mdmon") != NULL)
	134	kill(pid, SIGTERM);
	135	}
	136
e0d6609f NB	137	void remove_pidfile(char *devname)
	138	{
	139	char buf[100];
	140
	141	sprintf(buf, "/var/run/mdadm/%s.pid", devname);
	142	unlink(buf);
57752795 N	143	sprintf(buf, "/var/run/mdadm/%s.sock", devname);
57752795 N	144	unlink(buf);
e0d6609f NB	145	}
e0d6609f NB	146
295646b3	147	int make_control_sock(char *devname)
549e9569 NB	148	{
	149	char path[100];
	150	int sfd;
	151	long fl;
	152	struct sockaddr_un addr;
	153
	154	sprintf(path, "/var/run/mdadm/%s.sock", devname);
	155	unlink(path);
	156	sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
	157	if (sfd < 0)
	158	return -1;
	159
	160	addr.sun_family = PF_LOCAL;
	161	strcpy(addr.sun_path, path);
	162	if (bind(sfd, &addr, sizeof(addr)) < 0) {
	163	close(sfd);
	164	return -1;
	165	}
	166	listen(sfd, 10);
	167	fl = fcntl(sfd, F_GETFL, 0);
	168	fl \|= O_NONBLOCK;
	169	fcntl(sfd, F_SETFL, fl);
	170	return sfd;
	171	}
	172
295646b3 DW	173	int socket_hup_requested;
	174	static void hup(int sig)
	175	{
	176	socket_hup_requested = 1;
	177	}
	178
4d43913c NB	179	static void wake_me(int sig)
	180	{
	181
	182	}
	183
16ddab0d DW	184	/* if we are debugging and starting mdmon by hand then don't fork */
	185	static int do_fork(void)
	186	{
	187	#ifdef DEBUG
	188	if (env_no_mdmon())
	189	return 0;
	190	#endif
	191
	192	return 1;
	193	}
	194
	195
	196
549e9569 NB	197	int main(int argc, char *argv[])
	198	{
	199	int mdfd;
549e9569 NB	200	struct mdinfo mdi, di;
549e9569 NB	201	struct supertype *container;
4d43913c	202	sigset_t set;
bfa44e2e	203	struct sigaction act;
9fe32043 N	204	int pfd[2];
9fe32043 N	205	int status;
3d2c4fc7	206	int ignore;
4d43913c	207
549e9569 NB	208	if (argc != 2) {
	209	fprintf(stderr, "Usage: md-manage /device/name/for/container\n");
	210	exit(2);
	211	}
	212	mdfd = open(argv[1], O_RDWR);
	213	if (mdfd < 0) {
	214	fprintf(stderr, "md-manage: %s: %s\n", argv[1],
	215	strerror(errno));
	216	exit(1);
	217	}
	218	if (md_get_version(mdfd) < 0) {
	219	fprintf(stderr, "md-manage: %s: Not an md device\n",
	220	argv[1]);
	221	exit(1);
	222	}
	223
9fe32043	224	/* Fork, and have the child tell us when they are ready */
16ddab0d	225	if (do_fork()) {
3d2c4fc7 DW	226	if (pipe(pfd) != 0) {
	227	fprintf(stderr, "mdmon: failed to create pipe\n");
	228	exit(1);
	229	}
16ddab0d DW	230	switch(fork()) {
	231	case -1:
	232	fprintf(stderr, "mdmon: failed to fork: %s\n",
	233	strerror(errno));
	234	exit(1);
	235	case 0: /* child */
	236	close(pfd[0]);
	237	break;
	238	default: /* parent */
	239	close(pfd[1]);
	240	if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) {
	241	wait(&status);
	242	status = WEXITSTATUS(status);
	243	}
	244	exit(status);
9fe32043	245	}
16ddab0d DW	246	} else
16ddab0d DW	247	pfd[0] = pfd[1] = -1;
549e9569 NB	248	/* hopefully it is a container - we'll check later */
	249
	250	container = malloc(sizeof(*container));
549e9569 NB	251	container->devnum = fd2devnum(mdfd);
549e9569 NB	252	container->devname = devnum2devname(container->devnum);
e0d6609f	253	container->device_name = argv[1];
549e9569 NB	254
	255	/* If this fails, we hope it already exists */
	256	mkdir("/var/run/mdadm", 0600);
	257	/* pid file lives in /var/run/mdadm/mdXX.pid */
b109d928 DW	258	if (make_pidfile(container->devname, O_EXCL) < 0) {
	259	if (ping_monitor(container->devname) == 0) {
	260	fprintf(stderr, "mdmon: %s already managed\n",
	261	container->devname);
	262	exit(3);
	263	} else {
295646b3 DW	264	int err;
295646b3 DW	265
b109d928 DW	266	/* cleanup the old monitor, this one is taking over */
b109d928 DW	267	try_kill_monitor(container->devname);
295646b3 DW	268	err = make_pidfile(container->devname, 0);
295646b3 DW	269	if (err < 0) {
b109d928 DW	270	fprintf(stderr, "mdmon: %s Cannot create pidfile\n",
b109d928 DW	271	container->devname);
295646b3 DW	272	if (err == -EROFS) {
	273	/* FIXME implement a mechanism to
	274	* prevent duplicate monitor instances
	275	*/
	276	fprintf(stderr,
	277	"mdmon: continuing on read-only file system\n");
	278	} else
	279	exit(3);
b109d928 DW	280	}
b109d928 DW	281	}
549e9569 NB	282	}
	283
	284	container->sock = make_control_sock(container->devname);
549e9569 NB	285	container->arrays = NULL;
	286
	287	mdi = sysfs_read(mdfd, container->devnum,
	288	GET_VERSION\|GET_LEVEL\|GET_DEVS);
	289
	290	if (!mdi) {
	291	fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
	292	container->devname);
	293	exit(3);
	294	}
	295	if (mdi->array.level != UnSet) {
	296	fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
	297	argv[1]);
	298	exit(3);
	299	}
	300	if (mdi->array.major_version != -1 \|\|
	301	mdi->array.minor_version != -2) {
	302	fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
	303	argv[1]);
	304	exit(3);
	305	}
	306
	307	container->ss = find_metadata_methods(mdi->text_version);
	308	if (container->ss == NULL) {
	309	fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
	310	argv[1], mdi->text_version);
	311	exit(3);
	312	}
	313
	314	container->devs = NULL;
	315	for (di = mdi->devs; di; di = di->next) {
	316	struct mdinfo cd = malloc(sizeof(cd));
7bc1962f	317	cd = di;
549e9569 NB	318	cd->next = container->devs;
	319	container->devs = cd;
	320	}
	321	sysfs_free(mdi);
	322
	323
	324	if (container->ss->load_super(container, mdfd, argv[1])) {
	325	fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
	326	argv[1]);
	327	exit(3);
	328	}
549e9569	329
9fe32043 N	330	/* Ok, this is close enough. We can say goodbye to our parent now.
	331	*/
	332	status = 0;
3d2c4fc7 DW	333	if (write(pfd[1], &status, sizeof(status)) < 0)
	334	fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
	335	getppid());
9fe32043 N	336	close(pfd[1]);
9fe32043 N	337
3d2c4fc7	338	ignore = chdir("/");
9fe32043 N	339	setsid();
	340	close(0);
	341	open("/dev/null", O_RDWR);
	342	close(1);
3d2c4fc7	343	ignore = dup(0);
9fe32043 N	344	#ifndef DEBUG
9fe32043 N	345	close(2);
3d2c4fc7	346	ignore = dup(0);
9fe32043 N	347	#endif
9fe32043 N	348
549e9569 NB	349	mlockall(MCL_FUTURE);
549e9569 NB	350
4d43913c NB	351	/* SIGUSR is sent between parent and child. So both block it
	352	* and enable it only with pselect.
	353	*/
	354	sigemptyset(&set);
	355	sigaddset(&set, SIGUSR1);
295646b3	356	sigaddset(&set, SIGHUP);
695154b2	357	sigaddset(&set, SIGALRM);
4d43913c	358	sigprocmask(SIG_BLOCK, &set, NULL);
bfa44e2e NB	359	act.sa_handler = wake_me;
	360	act.sa_flags = 0;
	361	sigaction(SIGUSR1, &act, NULL);
695154b2	362	sigaction(SIGALRM, &act, NULL);
295646b3 DW	363	act.sa_handler = hup;
295646b3 DW	364	sigaction(SIGHUP, &act, NULL);
bfa44e2e NB	365	act.sa_handler = SIG_IGN;
bfa44e2e NB	366	sigaction(SIGPIPE, &act, NULL);
4d43913c	367
3e70c845	368	if (clone_monitor(container) < 0) {
295646b3	369	fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
549e9569 NB	370	strerror(errno));
	371	exit(2);
	372	}
	373
	374	do_manager(container);
	375
	376	exit(0);
	377	}