[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/stat.h>
#include	<sys/socket.h>
#include	<sys/un.h>
#include	<sys/mman.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;
struct md_generic_cmd *active_cmd;

int run_child(void *v)
{
	struct supertype *c = v;
	sigset_t set;
	/* SIGUSR is sent from child to parent,  So child must block it */
	sigemptyset(&set);
	sigaddset(&set, SIGUSR1);
	sigprocmask(SIG_BLOCK, &set, NULL);

	do_monitor(c);
	return 0;
}

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

	rv = pipe(container->mgr_pipe);
	if (rv < 0)
		return rv;
	rv = pipe(container->mon_pipe);
	if (rv < 0)
		goto err_mon_pipe;

	rv = clone(run_child, stack+4096-64,
		   CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD,
		   container);
	if (rv < 0)
		goto err_clone;
	else
		return rv;

 err_clone:
	close(container->mon_pipe[0]);
	close(container->mon_pipe[1]);
 err_mon_pipe:
	close(container->mgr_pipe[0]);
	close(container->mgr_pipe[1]);

	return rv;
}

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


static int make_pidfile(char *devname, int o_excl)
{
	char path[100];
	char pid[10];
	int fd;
	sprintf(path, "/var/run/mdadm/%s.pid", devname);

	fd = open(path, O_RDWR|O_CREAT|o_excl, 0600);
	if (fd < 0)
		return -1;
	sprintf(pid, "%d\n", getpid());
	write(fd, pid, strlen(pid));
	close(fd);
	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);
}

static 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 main(int argc, char *argv[])
{
	int mdfd;
	struct mdinfo *mdi, *di;
	struct supertype *container;
	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);
	}

	/* 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 {
			/* cleanup the old monitor, this one is taking over */
			try_kill_monitor(container->devname);
			if (make_pidfile(container->devname, 0) < 0) {
				fprintf(stderr, "mdmon: %s Cannot create pidfile\n",
					container->devname);
				exit(3);
			}
		}
	}

	container->sock = make_control_sock(container->devname);
	if (container->sock < 0) {
		fprintf(stderr, "mdmon: Cannot create socket in /var/run/mdadm\n");
		exit(3);
	}
	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);
	}
	close(mdfd);
	close(mdfd);

	mlockall(MCL_FUTURE);

	if (clone_monitor(container) < 0) {
		fprintf(stderr, "md-manage: 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>
	29	#include <sys/stat.h>
	30	#include <sys/socket.h>
	31	#include <sys/un.h>
	32	#include <sys/mman.h>
	33	#include <stdio.h>
	34	#include <errno.h>
	35	#include <string.h>
	36	#include <fcntl.h>
b109d928	37	#include <signal.h>
549e9569 NB	38
	39	#include <sched.h>
	40
	41	#include "mdadm.h"
	42	#include "mdmon.h"
	43
549e9569 NB	44	struct active_array *discard_this;
549e9569 NB	45	struct active_array *pending_discard;
3e70c845	46	struct md_generic_cmd *active_cmd;
549e9569 NB	47
	48	int run_child(void *v)
	49	{
	50	struct supertype *c = v;
1ed3f387 NB	51	sigset_t set;
	52	/* SIGUSR is sent from child to parent, So child must block it */
	53	sigemptyset(&set);
	54	sigaddset(&set, SIGUSR1);
	55	sigprocmask(SIG_BLOCK, &set, NULL);
	56
549e9569 NB	57	do_monitor(c);
	58	return 0;
	59	}
	60
	61	int clone_monitor(struct supertype *container)
	62	{
549e9569 NB	63	static char stack[4096];
	64	int rv;
	65
3e70c845 DW	66	rv = pipe(container->mgr_pipe);
	67	if (rv < 0)
	68	return rv;
	69	rv = pipe(container->mon_pipe);
	70	if (rv < 0)
	71	goto err_mon_pipe;
549e9569 NB	72
	73	rv = clone(run_child, stack+4096-64,
	74	CLONE_FS\|CLONE_FILES\|CLONE_VM\|CLONE_SIGHAND\|CLONE_THREAD,
	75	container);
549e9569	76	if (rv < 0)
3e70c845 DW	77	goto err_clone;
3e70c845 DW	78	else
549e9569	79	return rv;
3e70c845 DW	80
	81	err_clone:
	82	close(container->mon_pipe[0]);
	83	close(container->mon_pipe[1]);
	84	err_mon_pipe:
	85	close(container->mgr_pipe[0]);
	86	close(container->mgr_pipe[1]);
	87
	88	return rv;
549e9569 NB	89	}
	90
	91	static struct superswitch find_metadata_methods(char vers)
	92	{
	93	if (strcmp(vers, "ddf") == 0)
	94	return &super_ddf;
	95	return NULL;
	96	}
	97
	98
b109d928	99	static int make_pidfile(char *devname, int o_excl)
549e9569 NB	100	{
	101	char path[100];
	102	char pid[10];
	103	int fd;
	104	sprintf(path, "/var/run/mdadm/%s.pid", devname);
	105
b109d928	106	fd = open(path, O_RDWR\|O_CREAT\|o_excl, 0600);
549e9569 NB	107	if (fd < 0)
	108	return -1;
	109	sprintf(pid, "%d\n", getpid());
	110	write(fd, pid, strlen(pid));
	111	close(fd);
	112	return 0;
	113	}
	114
b109d928 DW	115	static void try_kill_monitor(char *devname)
	116	{
	117	char buf[100];
	118	int fd;
	119	pid_t pid;
	120
	121	sprintf(buf, "/var/run/mdadm/%s.pid", devname);
	122	fd = open(buf, O_RDONLY);
	123	if (fd < 0)
	124	return;
	125
	126	if (read(fd, buf, sizeof(buf)) < 0) {
	127	close(fd);
	128	return;
	129	}
	130
	131	close(fd);
	132	pid = strtoul(buf, NULL, 10);
	133
	134	/* kill this process if it is mdmon */
	135	sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid);
	136	fd = open(buf, O_RDONLY);
	137	if (fd < 0)
	138	return;
	139
	140	if (read(fd, buf, sizeof(buf)) < 0) {
	141	close(fd);
	142	return;
	143	}
	144
	145	if (strstr(buf, "mdmon") != NULL)
	146	kill(pid, SIGTERM);
	147	}
	148
e0d6609f NB	149	void remove_pidfile(char *devname)
	150	{
	151	char buf[100];
	152
	153	sprintf(buf, "/var/run/mdadm/%s.pid", devname);
	154	unlink(buf);
	155	}
	156
549e9569 NB	157	static int make_control_sock(char *devname)
	158	{
	159	char path[100];
	160	int sfd;
	161	long fl;
	162	struct sockaddr_un addr;
	163
	164	sprintf(path, "/var/run/mdadm/%s.sock", devname);
	165	unlink(path);
	166	sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
	167	if (sfd < 0)
	168	return -1;
	169
	170	addr.sun_family = PF_LOCAL;
	171	strcpy(addr.sun_path, path);
	172	if (bind(sfd, &addr, sizeof(addr)) < 0) {
	173	close(sfd);
	174	return -1;
	175	}
	176	listen(sfd, 10);
	177	fl = fcntl(sfd, F_GETFL, 0);
	178	fl \|= O_NONBLOCK;
	179	fcntl(sfd, F_SETFL, fl);
	180	return sfd;
	181	}
	182
	183	int main(int argc, char *argv[])
	184	{
	185	int mdfd;
549e9569 NB	186	struct mdinfo mdi, di;
	187	struct supertype *container;
	188	if (argc != 2) {
	189	fprintf(stderr, "Usage: md-manage /device/name/for/container\n");
	190	exit(2);
	191	}
	192	mdfd = open(argv[1], O_RDWR);
	193	if (mdfd < 0) {
	194	fprintf(stderr, "md-manage: %s: %s\n", argv[1],
	195	strerror(errno));
	196	exit(1);
	197	}
	198	if (md_get_version(mdfd) < 0) {
	199	fprintf(stderr, "md-manage: %s: Not an md device\n",
	200	argv[1]);
	201	exit(1);
	202	}
	203
	204	/* hopefully it is a container - we'll check later */
	205
	206	container = malloc(sizeof(*container));
549e9569 NB	207	container->devnum = fd2devnum(mdfd);
549e9569 NB	208	container->devname = devnum2devname(container->devnum);
e0d6609f	209	container->device_name = argv[1];
549e9569 NB	210
	211	/* If this fails, we hope it already exists */
	212	mkdir("/var/run/mdadm", 0600);
	213	/* pid file lives in /var/run/mdadm/mdXX.pid */
b109d928 DW	214	if (make_pidfile(container->devname, O_EXCL) < 0) {
	215	if (ping_monitor(container->devname) == 0) {
	216	fprintf(stderr, "mdmon: %s already managed\n",
	217	container->devname);
	218	exit(3);
	219	} else {
	220	/* cleanup the old monitor, this one is taking over */
	221	try_kill_monitor(container->devname);
	222	if (make_pidfile(container->devname, 0) < 0) {
	223	fprintf(stderr, "mdmon: %s Cannot create pidfile\n",
	224	container->devname);
	225	exit(3);
	226	}
	227	}
549e9569 NB	228	}
	229
	230	container->sock = make_control_sock(container->devname);
	231	if (container->sock < 0) {
	232	fprintf(stderr, "mdmon: Cannot create socket in /var/run/mdadm\n");
	233	exit(3);
	234	}
	235	container->arrays = NULL;
	236
	237	mdi = sysfs_read(mdfd, container->devnum,
	238	GET_VERSION\|GET_LEVEL\|GET_DEVS);
	239
	240	if (!mdi) {
	241	fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
	242	container->devname);
	243	exit(3);
	244	}
	245	if (mdi->array.level != UnSet) {
	246	fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
	247	argv[1]);
	248	exit(3);
	249	}
	250	if (mdi->array.major_version != -1 \|\|
	251	mdi->array.minor_version != -2) {
	252	fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
	253	argv[1]);
	254	exit(3);
	255	}
	256
	257	container->ss = find_metadata_methods(mdi->text_version);
	258	if (container->ss == NULL) {
	259	fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
	260	argv[1], mdi->text_version);
	261	exit(3);
	262	}
	263
	264	container->devs = NULL;
	265	for (di = mdi->devs; di; di = di->next) {
	266	struct mdinfo cd = malloc(sizeof(cd));
	267	cd = di;
	268	cd->next = container->devs;
	269	container->devs = cd;
	270	}
	271	sysfs_free(mdi);
	272
	273
	274	if (container->ss->load_super(container, mdfd, argv[1])) {
	275	fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
	276	argv[1]);
	277	exit(3);
	278	}
e0d6609f	279	close(mdfd);
5869a76c	280	close(mdfd);
549e9569 NB	281
	282	mlockall(MCL_FUTURE);
	283
3e70c845	284	if (clone_monitor(container) < 0) {
549e9569 NB	285	fprintf(stderr, "md-manage: failed to start monitor process: %s\n",
	286	strerror(errno));
	287	exit(2);
	288	}
	289
	290	do_manager(container);
	291
	292	exit(0);
	293	}