X-Git-Url: http://git.ipfire.org/?p=thirdparty%2Fmdadm.git;a=blobdiff_plain;f=mdmon.c;h=d20bb3e490c01bbde2067bffb70c31ae50dcab04;hp=acb36a2d6f2621ad9ec26daf95522bbe04b3c4aa;hb=24f6f99b3630b1a89aaa57930c5c9de8a3df9ded;hpb=5746141e3f48246cd51f74562e3f947e8d0ddf6a

diff --git a/mdmon.c b/mdmon.c
index acb36a2d..d20bb3e4 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,
@@ -113,7 +113,6 @@ static struct superswitch *find_metadata_methods(char *vers)
 	return NULL;
 }
 
-
 int make_pidfile(char *devname, int o_excl)
 {
 	char path[100];
@@ -149,26 +148,12 @@ int is_container_member(struct mdstat_ent *mdstat, char *container)
 	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())
@@ -180,24 +165,21 @@ 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"))
 		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);
+	/* 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);
+	read(sock, buf, 100);
 }
 
 void remove_pidfile(char *devname)
@@ -275,7 +257,7 @@ void usage(void)
 	exit(2);
 }
 
-int mdmon(char *devname, int devnum, int scan, char *switchroot);
+static int mdmon(char *devname, int devnum, int must_fork, char *switchroot);
 
 int main(int argc, char *argv[])
 {
@@ -283,7 +265,6 @@ int main(int argc, char *argv[])
 	char *switchroot = NULL;
 	int devnum;
 	char *devname;
-	int scan = 0;
 	int status = 0;
 
 	switch (argc) {
@@ -300,13 +281,19 @@ int main(int argc, char *argv[])
 		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);
-				status |= mdmon(devname, e->devnum, scan,
+				/* 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, 1,
 						switchroot);
 			}
 		}
@@ -335,10 +322,10 @@ int main(int argc, char *argv[])
 			container_name);
 		exit(1);
 	}
-	return mdmon(devname, devnum, scan, switchroot);
+	return mdmon(devname, devnum, do_fork(), switchroot);
 }
 
-int mdmon(char *devname, int devnum, int scan, char *switchroot)
+static int mdmon(char *devname, int devnum, int must_fork, char *switchroot)
 {
 	int mdfd;
 	struct mdinfo *mdi, *di;
@@ -348,9 +335,54 @@ int mdmon(char *devname, int devnum, int scan, char *switchroot)
 	int pfd[2];
 	int status;
 	int ignore;
+	pid_t victim = -1;
+	int victim_sock = -1;
 
 	dprintf("starting mdmon for %s in %s\n",
 		devname, switchroot ? : "/");
+
+	/* switchroot is either a path name starting with '/', or a
+	 * pid of the original mdmon (we have already done the chroot).
+	 * In the latter case, stdin is a socket connected to the original
+	 * mdmon.
+	 */
+
+	/* try to spawn mdmon instances from the target file system */
+	if (switchroot && switchroot[0] == '/' &&
+	    strcmp(switchroot, "/") != 0) {
+		pid_t pid;
+		char buf[20];
+
+		switch (fork()) {
+		case 0:
+			victim = mdmon_pid(devnum);
+			victim_sock = connect_monitor(devname);
+			if (chroot(switchroot) != 0) {
+				fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
+					switchroot, strerror(errno));
+				exit(4);
+			}
+			ignore = chdir("/");
+			sprintf(buf, "%d", victim);
+			if (victim_sock) {
+				close(0);
+				dup(victim_sock);
+				close(victim_sock);
+			}
+			execl("/sbin/mdmon", "mdmon", devname, buf, 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,
@@ -364,7 +396,7 @@ int mdmon(char *devname, int devnum, int scan, char *switchroot)
 	}
 
 	/* Fork, and have the child tell us when they are ready */
-	if (do_fork() || scan) {
+	if (must_fork) {
 		if (pipe(pfd) != 0) {
 			fprintf(stderr, "mdmon: failed to create pipe\n");
 			return 1;
@@ -388,11 +420,12 @@ int mdmon(char *devname, int devnum, int scan, char *switchroot)
 	} else
 		pfd[0] = pfd[1] = -1;
 
-	container = malloc(sizeof(*container));
+	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");
@@ -400,7 +433,7 @@ int mdmon(char *devname, int devnum, int scan, char *switchroot)
 	}
 
 	mdi = sysfs_read(mdfd, container->devnum,
-			 GET_VERSION|GET_LEVEL|GET_DEVS);
+			 GET_VERSION|GET_LEVEL|GET_DEVS|SKIP_GONE_DEVS);
 
 	if (!mdi) {
 		fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
@@ -457,53 +490,27 @@ int mdmon(char *devname, int devnum, int scan, char *switchroot)
 
 	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
+		 * the new root
 		 */
-		try_kill_monitor(container->devname);
-		if (chroot(switchroot) != 0) {
-			fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
-				switchroot, strerror(errno));
-			exit(4);
+		if (switchroot[0] == '/') {
+			victim = mdmon_pid(container->devnum);
+			victim_sock = connect_monitor(container->devname);
+		} else {
+			victim = atoi(switchroot);
+			victim_sock = 0;
 		}
 	}
 
-	/* 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 (victim < 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);
-			}
 		}
+		/* if there is a pid file, kill whoever is there just in case */
+		victim = mdmon_pid(container->devnum);
 	}
-	container->sock = make_control_sock(container->devname);
-
 	if (container->ss->load_super(container, mdfd, devname)) {
 		fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
 			devname);
@@ -519,6 +526,19 @@ int mdmon(char *devname, int devnum, int scan, char *switchroot)
 			getppid());
 	close(pfd[1]);
 
+	mlockall(MCL_CURRENT | 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, victim_sock);
+		close(victim_sock);
+	}
+
 	setsid();
 	close(0);
 	open("/dev/null", O_RDWR);
@@ -529,14 +549,6 @@ int mdmon(char *devname, int devnum, int scan, char *switchroot)
 	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);
-	}
-
 	do_manager(container);
 
 	exit(0);