pasta: By default, quit if filesystem-bound net namespace goes away

This should be convenient for users managing filesystem-bound network
namespaces: monitor the base directory of the namespace and exit if
the namespace given as PATH or NAME target is deleted. We can't add
an inotify watch directly on the namespace directory, that won't work
with nsfs.

Add an option to disable this behaviour, --no-netns-quit.

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>

7 files changed, 107 insertions, 12 deletions

diff --git a/Makefile b/Makefile
index 8477cf0..28ef316 100644
--- a/Makefile
+++ b/Makefile
@@ -153,6 +153,7 @@ pkgs:
 # - android-cloexec-pipe
 # - android-cloexec-pipe2
 # - android-cloexec-epoll-create1
+# - android-cloexec-inotify-init1
 #	TODO: check, fix except for the few cases where we need to share fds
 #
 # - bugprone-narrowing-conversions
@@ -197,7 +198,7 @@ clang-tidy: $(wildcard *.c) $(wildcard *.h)
 	-cppcoreguidelines-avoid-magic-numbers,\
 	-readability-isolate-declaration,\
 	-android-cloexec-open,-android-cloexec-pipe,-android-cloexec-pipe2,\
-	-android-cloexec-epoll-create1,\
+	-android-cloexec-epoll-create1,-android-cloexec-inotify-init1,\
 	-bugprone-narrowing-conversions,\
 	-cppcoreguidelines-narrowing-conversions,\
 	-cppcoreguidelines-avoid-non-const-global-variables,\
diff --git a/conf.c b/conf.c
index 21e9bc0..9851575 100644
--- a/conf.c
+++ b/conf.c
@@ -20,6 +20,7 @@
 #include <sched.h>
 #include <sys/types.h>
 #include <sys/stat.h>
+#include <libgen.h>
 #include <limits.h>
 #include <stdlib.h>
 #include <stdint.h>
@@ -414,20 +415,34 @@ static int conf_ns_opt(struct ctx *c,
 
 		nfd = open(netns, O_RDONLY);
 
-		if (nfd >= 0 && (ufd >= 0 || c->netns_only)) {
-			c->pasta_netns_fd = nfd;
-			c->pasta_userns_fd = ufd;
+		if (nfd == -1 || (ufd == -1 && !c->netns_only)) {
+			if (nfd >= 0)
+				close(nfd);
 
-			NS_CALL(conf_ns_check, c);
-			if (c->pasta_netns_fd >= 0)
-				return 0;
+			if (ufd >= 0)
+				close(ufd);
+
+			continue;
 		}
 
-		if (nfd >= 0)
-			close(nfd);
+		c->pasta_netns_fd = nfd;
+		c->pasta_userns_fd = ufd;
+
+		NS_CALL(conf_ns_check, c);
+
+		if (c->pasta_netns_fd >= 0) {
+			char buf[PATH_MAX];
+
+			if (try == 0 || c->no_netns_quit)
+				return 0;
+
+			strncpy(buf, netns, PATH_MAX);
+			strncpy(c->netns_base, basename(buf), PATH_MAX - 1);
+			strncpy(buf, netns, PATH_MAX);
+			strncpy(c->netns_dir, dirname(buf), PATH_MAX - 1);
 
-		if (ufd >= 0)
-			close(ufd);
+			return 0;
+		}
 	}
 
 	c->netns_only = netns_only_reset;
@@ -813,6 +828,7 @@ void conf(struct ctx *c, int argc, char **argv)
 		{"dhcp-search", no_argument,		NULL,		7 },
 		{"no-dhcp-search", no_argument,		NULL,		8 },
 		{"dns-forward",	required_argument,	NULL,		9 },
+		{"no-netns-quit", no_argument,		NULL,		10 },
 		{ 0 },
 	};
 	struct get_bound_ports_ns_arg ns_ports_arg = { .c = c };
@@ -937,6 +953,13 @@ void conf(struct ctx *c, int argc, char **argv)
 			err("Invalid DNS forwarding address: %s", optarg);
 			usage(argv[0]);
 			break;
+		case 10:
+			if (c->mode != MODE_PASTA) {
+				err("--no-netns-quit is for pasta mode only");
+				usage(argv[0]);
+			}
+			c->no_netns_quit = 1;
+			break;
 		case 'd':
 			if (c->debug) {
 				err("Multiple --debug options given");
diff --git a/passt.1 b/passt.1
index 7070a31..485e1db 100644
--- a/passt.1
+++ b/passt.1
@@ -427,6 +427,11 @@ for sandboxing purposes either. This is implied if PATH or NAME are given
 without \-\-userns.
 
 .TP
+.BR \-\-no-netns-quit
+If the target network namespace is bound to the filesystem (that is, if PATH or
+NAME are given as target), do not exit once the network namespace is deleted.
+
+.TP
 .BR \-\-nsrun-dir " " \fIpath
 Directory for nsfs mountpoints, used as path prefix for names of namespaces.
 
diff --git a/passt.c b/passt.c
index 67ad1c7..36f0161 100644
--- a/passt.c
+++ b/passt.c
@@ -301,7 +301,7 @@ void exit_handler(int signal)
  */
 int main(int argc, char **argv)
 {
-	int nfds, i, devnull_fd = -1, pidfile_fd = -1;
+	int nfds, i, devnull_fd = -1, pidfile_fd = -1, quit_fd;
 	struct epoll_event events[EPOLL_EVENTS];
 	struct ctx c = { 0 };
 	struct rlimit limit;
@@ -357,6 +357,8 @@ int main(int argc, char **argv)
 		exit(EXIT_FAILURE);
 	}
 
+	quit_fd = pasta_netns_quit_init(&c);
+
 	if (getrlimit(RLIMIT_NOFILE, &limit)) {
 		perror("getrlimit");
 		exit(EXIT_FAILURE);
@@ -416,6 +418,7 @@ int main(int argc, char **argv)
 	seccomp(&c);
 
 	timer_init(&c, &now);
+
 loop:
 	nfds = epoll_wait(c.epollfd, events, EPOLL_EVENTS, TIMER_INTERVAL);
 	if (nfds == -1 && errno != EINTR) {
@@ -431,6 +434,8 @@ loop:
 
 		if (fd == c.fd_tap || fd == c.fd_tap_listen)
 			tap_handler(&c, fd, events[i].events, &now);
+		else if (fd == quit_fd)
+			pasta_netns_quit_handler(&c, fd);
 		else
 			sock_handler(&c, ref, events[i].events, &now);
 	}
diff --git a/passt.h b/passt.h
index 2589ee7..042f760 100644
--- a/passt.h
+++ b/passt.h
@@ -101,6 +101,9 @@ enum passt_modes {
  * @pasta_netns_fd:	File descriptor for network namespace in pasta mode
  * @pasta_userns_fd:	Descriptor for user namespace to join, -1 once joined
  * @netns_only:		In pasta mode, don't join or create a user namespace
+ * @no_netns_quit:	In pasta mode, don't exit if fs-bound namespace is gone
+ * @netns_base:		Base name for fs-bound namespace, if any, in pasta mode
+ * @netns_dir:		Directory of fs-bound namespace, if any, in pasta mode
  * @proc_net_tcp:	Stored handles for /proc/net/tcp{,6} in init and ns
  * @proc_net_udp:	Stored handles for /proc/net/udp{,6} in init and ns
  * @epollfd:		File descriptor for epoll instance
@@ -161,6 +164,10 @@ struct ctx {
 	int pasta_userns_fd;
 	int netns_only;
 
+	int no_netns_quit;
+	char netns_base[PATH_MAX];
+	char netns_dir[PATH_MAX];
+
 	int proc_net_tcp[IP_VERSIONS][2];
 	int proc_net_udp[IP_VERSIONS][2];
 
diff --git a/pasta.c b/pasta.c
index 972cbcf..e45cc92 100644
--- a/pasta.c
+++ b/pasta.c
@@ -24,6 +24,8 @@
 #include <stdint.h>
 #include <unistd.h>
 #include <syslog.h>
+#include <sys/epoll.h>
+#include <sys/inotify.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <fcntl.h>
@@ -219,3 +221,53 @@ void pasta_ns_conf(struct ctx *c)
 
 	proto_update_l2_buf(c->mac_guest, NULL, NULL);
 }
+
+/**
+ * pasta_netns_quit_init() - Watch network namespace to quit once it's gone
+ * @c:		Execution context
+ *
+ * Return: inotify file descriptor, -1 on failure or if not needed/applicable
+ */
+int pasta_netns_quit_init(struct ctx *c)
+{
+	struct epoll_event ev = { .events = EPOLLIN };
+	int inotify_fd;
+
+	if (c->mode != MODE_PASTA || c->no_netns_quit || !*c->netns_base)
+		return -1;
+
+	if ((inotify_fd = inotify_init1(O_NONBLOCK)) < 0) {
+		perror("inotify_init(): won't quit once netns is gone");
+		return -1;
+	}
+
+	if (inotify_add_watch(inotify_fd, c->netns_dir, IN_DELETE) < 0) {
+		perror("inotify_add_watch(): won't quit once netns is gone");
+		return -1;
+	}
+
+	ev.data.fd = inotify_fd;
+	epoll_ctl(c->epollfd, EPOLL_CTL_ADD, inotify_fd, &ev);
+
+	return inotify_fd;
+}
+
+/**
+ * pasta_netns_quit_handler() - Handle ns directory events, exit if ns is gone
+ * @c:		Execution context
+ * @inotify_fd:	inotify file descriptor with watch on namespace directory
+ */
+void pasta_netns_quit_handler(struct ctx *c, int inotify_fd)
+{
+	char buf[sizeof(struct inotify_event) + NAME_MAX + 1];
+	struct inotify_event *in_ev = (struct inotify_event *)buf;
+
+	if (read(inotify_fd, buf, sizeof(buf)) < (ssize_t)sizeof(*in_ev))
+		return;
+
+	if (strncmp(in_ev->name, c->netns_base, sizeof(c->netns_base)))
+		return;
+
+	info("Namespace %s is gone, exiting", c->netns_base);
+	exit(EXIT_SUCCESS);
+}
diff --git a/pasta.h b/pasta.h
index 1fcd6a9..235bfb9 100644
--- a/pasta.h
+++ b/pasta.h
@@ -6,3 +6,5 @@
 void pasta_start_ns(struct ctx *c);
 void pasta_ns_conf(struct ctx *c);
 void pasta_child_handler(int signal);
+int pasta_netns_quit_init(struct ctx *c);
+void pasta_netns_quit_handler(struct ctx *c, int inotify_fd);