X-Git-Url: http://git.ipfire.org/?p=thirdparty%2Fmdadm.git;a=blobdiff_plain;f=mdmon.c;h=b84d4d9520485c04e55e7ebaccc844e0e81f2f67;hp=bfad18a7e6c2a88440304fd736c945e3ec63fc66;hb=fdcd157a8034fe5c896e49c4764d8ceef72e3741;hpb=6c766cf10178b9bae4ddcfd6b8de2d70f9d84568

diff --git a/mdmon.c b/mdmon.c
index bfad18a7..b84d4d95 100644
--- a/mdmon.c
+++ b/mdmon.c
@@ -1,8 +1,8 @@
 /*
  * mdmon - monitor external metadata arrays
  *
- * Copyright (C) 2007-2008 Neil Brown <neilb@suse.de>
- * Copyright (C) 2007-2008 Intel Corporation
+ * 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,
@@ -58,8 +58,11 @@
 #include	<fcntl.h>
 #include	<signal.h>
 #include	<dirent.h>
-
+#ifdef USE_PTHREADS
+#include	<pthread.h>
+#else
 #include	<sched.h>
+#endif
 
 #include	"mdadm.h"
 #include	"mdmon.h"
@@ -71,7 +74,39 @@ int mon_tid, mgr_tid;
 
 int sigterm;
 
-int run_child(void *v)
+#ifdef USE_PTHREADS
+static void *run_child(void *v)
+{
+	struct supertype *c = v;
+
+	mon_tid = syscall(SYS_gettid);
+	do_monitor(c);
+	return 0;
+}
+
+static int clone_monitor(struct supertype *container)
+{
+	pthread_attr_t attr;
+	pthread_t thread;
+	int rc;
+
+	mon_tid = -1;
+	pthread_attr_init(&attr);
+	pthread_attr_setstacksize(&attr, 4096);
+	pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
+	rc = pthread_create(&thread, &attr, run_child, container);
+	if (rc)
+		return rc;
+	while (mon_tid == -1)
+		usleep(10);
+	pthread_attr_destroy(&attr);
+
+	mgr_tid = syscall(SYS_gettid);
+
+	return mon_tid;
+}
+#else /* USE_PTHREADS */
+static int run_child(void *v)
 {
 	struct supertype *c = v;
 
@@ -79,33 +114,45 @@ int run_child(void *v)
 	return 0;
 }
 
-int clone_monitor(struct supertype *container)
+#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
+static 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;
 }
+#endif /* USE_PTHREADS */
 
-
-int make_pidfile(char *devname, int o_excl)
+static int make_pidfile(char *devname)
 {
 	char path[100];
 	char pid[10];
 	int fd;
 	int n;
 
-	if (sigterm)
-		return -1;
-
-	sprintf(path, "/var/run/mdadm/%s.pid", devname);
+	if (mkdir(MDMON_DIR, 0755) < 0 &&
+	    errno != EEXIST)
+		return -errno;
+	sprintf(path, "%s/%s.pid", MDMON_DIR, devname);
 
-	fd = open(path, O_RDWR|O_CREAT|o_excl, 0600);
+	fd = open(path, O_RDWR|O_CREAT|O_EXCL, 0600);
 	if (fd < 0)
 		return -errno;
 	sprintf(pid, "%d\n", getpid());
@@ -116,38 +163,12 @@ int make_pidfile(char *devname, int o_excl)
 	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;
-}
-
-void remove_pidfile(char *devname);
-static void try_kill_monitor(char *devname)
+static void try_kill_monitor(pid_t pid, char *devname, int sock)
 {
 	char buf[100];
 	int fd;
-	pid_t pid;
-	struct mdstat_ent *mdstat;
-
-	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);
+	int n;
+	long fl;
 
 	/* first rule of survival... don't off yourself */
 	if (pid == getpid())
@@ -159,40 +180,41 @@ static void try_kill_monitor(char *devname)
 	if (fd < 0)
 		return;
 
-	if (read(fd, buf, sizeof(buf)) < 0) {
-		close(fd);
-		return;
-	}
+	n = read(fd, buf, sizeof(buf)-1);
+	buf[sizeof(buf)-1] = 0;
+	close(fd);
 
-	if (!strstr(buf, "mdmon"))
+	if (n < 0 || !(strstr(buf, "mdmon") ||
+		       strstr(buf, "@dmon")))
 		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);
-	remove_pidfile(devname);
+	if (sock < 0)
+		return;
+
+	/* Wait for monitor to exit by reading from the socket, after
+	 * clearing the non-blocking flag */
+	fl = fcntl(sock, F_GETFL, 0);
+	fl &= ~O_NONBLOCK;
+	fcntl(sock, F_SETFL, fl);
+	n = read(sock, buf, 100);
+	/* Ignore result, it is just the wait that
+	 * matters
+	 */
 }
 
 void remove_pidfile(char *devname)
 {
 	char buf[100];
 
-	if (sigterm)
-		return;
-
-	sprintf(buf, "/var/run/mdadm/%s.pid", devname);
+	sprintf(buf, "%s/%s.pid", MDMON_DIR, devname);
 	unlink(buf);
-	sprintf(buf, "/var/run/mdadm/%s.sock", devname);
+	sprintf(buf, "%s/%s.sock", MDMON_DIR, devname);
 	unlink(buf);
 }
 
-int make_control_sock(char *devname)
+static int make_control_sock(char *devname)
 {
 	char path[100];
 	int sfd;
@@ -202,7 +224,7 @@ int make_control_sock(char *devname)
 	if (sigterm)
 		return -1;
 
-	sprintf(path, "/var/run/mdadm/%s.sock", devname);
+	sprintf(path, "%s/%s.sock", MDMON_DIR, devname);
 	unlink(path);
 	sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
 	if (sfd < 0)
@@ -221,12 +243,6 @@ int make_control_sock(char *devname)
 	return sfd;
 }
 
-int socket_hup_requested;
-static void hup(int sig)
-{
-	socket_hup_requested = 1;
-}
-
 static void term(int sig)
 {
 	sigterm = 1;
@@ -241,7 +257,7 @@ static void wake_me(int sig)
 static int do_fork(void)
 {
 	#ifdef DEBUG
-	if (env_no_mdmon())
+	if (check_env("MDADM_NO_MDMON"))
 		return 0;
 	#endif
 
@@ -250,11 +266,126 @@ static int do_fork(void)
 
 void usage(void)
 {
-	fprintf(stderr, "Usage: mdmon [--switch-root dir] /device/name/for/container\n");
+	fprintf(stderr,
+"Usage: mdmon [options] CONTAINER\n"
+"\n"
+"Options are:\n"
+"  --help        -h   : This message\n"
+"  --all         -a   : All devices\n"
+"  --foreground  -F   : Run in foreground (do not fork)\n"
+"  --takeover    -t   : Takeover container\n"
+);
 	exit(2);
 }
 
+static int mdmon(char *devnm, int must_fork, int takeover);
+
 int main(int argc, char *argv[])
+{
+	char *container_name = NULL;
+	char *devnm = NULL;
+	int status = 0;
+	int opt;
+	int all = 0;
+	int takeover = 0;
+	int dofork = 1;
+	static struct option options[] = {
+		{"all", 0, NULL, 'a'},
+		{"takeover", 0, NULL, 't'},
+		{"help", 0, NULL, 'h'},
+		{"offroot", 0, NULL, OffRootOpt},
+		{"foreground", 0, NULL, 'F'},
+		{NULL, 0, NULL, 0}
+	};
+
+	if (in_initrd()) {
+		/*
+		 * set first char of argv[0] to @. This is used by
+		 * systemd to signal that the task was launched from
+		 * initrd/initramfs and should be preserved during shutdown
+		 */
+		argv[0][0] = '@';
+	}
+
+	while ((opt = getopt_long(argc, argv, "thaF", options, NULL)) != -1) {
+		switch (opt) {
+		case 'a':
+			container_name = argv[optind-1];
+			all = 1;
+			break;
+		case 't':
+			takeover = 1;
+			break;
+		case 'F':
+			dofork = 0;
+			break;
+		case OffRootOpt:
+			argv[0][0] = '@';
+			break;
+		case 'h':
+		default:
+			usage();
+			break;
+		}
+	}
+
+	if (all == 0 && container_name == NULL) {
+		if (argv[optind])
+			container_name = argv[optind];
+	}
+
+	if (container_name == NULL)
+		usage();
+
+	if (argc - optind > 1)
+		usage();
+
+	if (strcmp(container_name, "/proc/mdstat") == 0)
+		all = 1;
+
+	if (all) {
+		struct mdstat_ent *mdstat, *e;
+		int container_len = strlen(container_name);
+
+		/* launch an mdmon instance for each container found */
+		mdstat = mdstat_read(0, 0);
+		for (e = mdstat; e; e = e->next) {
+			if (e->metadata_version &&
+			    strncmp(e->metadata_version, "external:", 9) == 0 &&
+			    !is_subarray(&e->metadata_version[9])) {
+				/* update cmdline so this mdmon instance can be
+				 * distinguished from others in a call to ps(1)
+				 */
+				if (strlen(e->devnm) <= (unsigned)container_len) {
+					memset(container_name, 0, container_len);
+					sprintf(container_name, "%s", e->devnm);
+				}
+				status |= mdmon(e->devnm, 1, takeover);
+			}
+		}
+		free_mdstat(mdstat);
+
+		return status;
+	} else if (strncmp(container_name, "md", 2) == 0) {
+		int id = devnm2devid(container_name);
+		if (id)
+			devnm = container_name;
+	} else {
+		struct stat st;
+
+		if (stat(container_name, &st) == 0)
+			devnm = xstrdup(stat2devnm(&st));
+	}
+
+	if (!devnm) {
+		pr_err("%s is not a valid md device name\n",
+			container_name);
+		exit(1);
+	}
+	return mdmon(devnm, dofork && do_fork(), takeover);
+}
+
+static int mdmon(char *devnm, int must_fork, int takeover)
 {
 	int mdfd;
 	struct mdinfo *mdi, *di;
@@ -264,48 +395,31 @@ int main(int argc, char *argv[])
 	int pfd[2];
 	int status;
 	int ignore;
-	char *container_name = NULL;
-	char *switchroot = NULL;
-
-	switch (argc) {
-	case 2:
-		container_name = argv[1];
-		break;
-	case 4:
-		if (strcmp(argv[1], "--switch-root") != 0) {
-			fprintf(stderr, "mdmon: unknown argument %s\n", argv[1]);
-			usage();
-		}
-		switchroot = argv[2];
-		container_name = argv[3];
-		break;
-	default:
-		usage();
-	}
+	pid_t victim = -1;
+	int victim_sock = -1;
+
+	dprintf("starting mdmon for %s\n", devnm);
 
-	mdfd = open(container_name, O_RDWR);
+	mdfd = open_dev(devnm);
 	if (mdfd < 0) {
-		fprintf(stderr, "mdmon: %s: %s\n", container_name,
-			strerror(errno));
-		exit(1);
+		pr_err("%s: %s\n", devnm, strerror(errno));
+		return 1;
 	}
 	if (md_get_version(mdfd) < 0) {
-		fprintf(stderr, "mdmon: %s: Not an md device\n",
-			container_name);
-		exit(1);
+		pr_err("%s: Not an md device\n", devnm);
+		return 1;
 	}
 
 	/* Fork, and have the child tell us when they are ready */
-	if (do_fork()) {
+	if (must_fork) {
 		if (pipe(pfd) != 0) {
-			fprintf(stderr, "mdmon: failed to create pipe\n");
-			exit(1);
+			pr_err("failed to create pipe\n");
+			return 1;
 		}
 		switch(fork()) {
 		case -1:
-			fprintf(stderr, "mdmon: failed to fork: %s\n",
-				strerror(errno));
-			exit(1);
+			pr_err("failed to fork: %s\n", strerror(errno));
+			return 1;
 		case 0: /* child */
 			close(pfd[0]);
 			break;
@@ -315,52 +429,44 @@ int main(int argc, char *argv[])
 				wait(&status);
 				status = WEXITSTATUS(status);
 			}
-			exit(status);
+			close(pfd[0]);
+			return status;
 		}
 	} else
 		pfd[0] = pfd[1] = -1;
 
-	container = malloc(sizeof(*container));
-	container->devnum = fd2devnum(mdfd);
-	container->devname = devnum2devname(container->devnum);
-	container->device_name = container_name;
+	container = xcalloc(1, sizeof(*container));
+	strcpy(container->devnm, devnm);
 	container->arrays = NULL;
+	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);
+	mdi = sysfs_read(mdfd, container->devnm, GET_VERSION|GET_LEVEL|GET_DEVS);
 
 	if (!mdi) {
-		fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
-			container->devname);
+		pr_err("failed to load sysfs info for %s\n", container->devnm);
 		exit(3);
 	}
 	if (mdi->array.level != UnSet) {
-		fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
-			container_name);
+		pr_err("%s is not a container - cannot monitor\n", devnm);
 		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",
-			container_name);
+		pr_err("%s does not use external metadata - cannot monitor\n",
+			devnm);
 		exit(3);
 	}
 
-	container->ss = find_metadata_methods(mdi->text_version);
+	container->ss = version_to_superswitch(mdi->text_version);
 	if (container->ss == NULL) {
-		fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
-			container_name, mdi->text_version);
+		pr_err("%s uses unsupported metadata: %s\n",
+			devnm, mdi->text_version);
 		exit(3);
 	}
 
 	container->devs = NULL;
 	for (di = mdi->devs; di; di = di->next) {
-		struct mdinfo *cd = malloc(sizeof(*cd));
+		struct mdinfo *cd = xmalloc(sizeof(*cd));
 		*cd = *di;
 		cd->next = container->devs;
 		container->devs = cd;
@@ -372,83 +478,65 @@ int main(int argc, char *argv[])
 	 */
 	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 (see nash:setuproot)
-		 *
-		 * kill any monitors in the current namespace and change
-		 * to the new one
-		 */
-		try_kill_monitor(container->devname);
-		if (chroot(switchroot) != 0) {
-			fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
-				switchroot, strerror(errno));
-			exit(4);
-		}
-	}
+	victim = mdmon_pid(container->devnm);
+	if (victim >= 0)
+		victim_sock = connect_monitor(container->devnm);
 
-	/* If this fails, we hope it already exists 
-	 * pid file lives in /var/run/mdadm/mdXX.pid
-	 */
-	mkdir("/var", 0600);
-	mkdir("/var/run", 0600);
-	mkdir("/var/run/mdadm", 0600);
 	ignore = chdir("/");
-	if (make_pidfile(container->devname, O_EXCL) < 0) {
-		if (ping_monitor(container->devname) == 0) {
-			fprintf(stderr, "mdmon: %s already managed\n",
-				container->devname);
+	if (!takeover && victim > 0 && victim_sock >= 0) {
+		if (fping_monitor(victim_sock) == 0) {
+			pr_err("%s already managed\n", container->devnm);
 			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);
-			}
 		}
+		close(victim_sock);
+		victim_sock = -1;
 	}
-	container->sock = make_control_sock(container->devname);
-
-	if (container->ss->load_super(container, mdfd, container_name)) {
-		fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
-			container_name);
+	if (container->ss->load_container(container, mdfd, devnm)) {
+		pr_err("Cannot load metadata for %s\n", devnm);
 		exit(3);
 	}
+	close(mdfd);
 
 	/* Ok, this is close enough.  We can say goodbye to our parent now.
 	 */
+	if (victim > 0)
+		remove_pidfile(devnm);
+	if (make_pidfile(devnm) < 0) {
+		exit(3);
+	}
+	container->sock = make_control_sock(devnm);
+
 	status = 0;
-	if (write(pfd[1], &status, sizeof(status)) < 0)
-		fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
-			getppid());
-	close(pfd[1]);
+	if (pfd[1] >= 0) {
+		if (write(pfd[1], &status, sizeof(status)) < 0)
+			pr_err("failed to notify our parent: %d\n",
+			       getppid());
+		close(pfd[1]);
+	}
+
+	mlockall(MCL_CURRENT | MCL_FUTURE);
+
+	if (clone_monitor(container) < 0) {
+		pr_err("failed to start monitor process: %s\n",
+			strerror(errno));
+		exit(2);
+	}
+
+	if (victim > 0) {
+		try_kill_monitor(victim, container->devnm, victim_sock);
+		if (victim_sock >= 0)
+			close(victim_sock);
+	}
 
 	setsid();
 	close(0);
@@ -460,15 +548,45 @@ int main(int argc, char *argv[])
 	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);
-	}
+	/* This silliness is to stop the compiler complaining
+	 * that we ignore 'ignore'
+	 */
+	if (ignore)
+		ignore++;
 
 	do_manager(container);
 
 	exit(0);
 }
+
+/* Some stub functions so super-* can link with us */
+int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
+		  struct supertype *st, unsigned long blocks,
+		  int *fds, unsigned long long *offsets,
+		  int dests, int *destfd, unsigned long long *destoffsets)
+{
+	return 0;
+}
+
+int restore_stripes(int *dest, unsigned long long *offsets,
+		    int raid_disks, int chunk_size, int level, int layout,
+		    int source, unsigned long long read_offset,
+		    unsigned long long start, unsigned long long length,
+		    char *src_buf)
+{
+	return 1;
+}
+
+void abort_reshape(struct mdinfo *sra)
+{
+	return;
+}
+
+int save_stripes(int *source, unsigned long long *offsets,
+		 int raid_disks, int chunk_size, int level, int layout,
+		 int nwrites, int *dest,
+		 unsigned long long start, unsigned long long length,
+		 char *buf)
+{
+	return 0;
+}