passt, pasta: Namespace-based sandboxing, defer seccomp policy application

To reach (at least) a conceptually equivalent security level as
implemented by --enable-sandbox in slirp4netns, we need to create a
new mount namespace and pivot_root() into a new (empty) mountpoint, so
that passt and pasta can't access any filesystem resource after
initialisation.

While at it, also detach IPC, PID (only for passt, to prevent
vulnerabilities based on the knowledge of a target PID), and UTS
namespaces.

With this approach, if we apply the seccomp filters right after the
configuration step, the number of allowed syscalls grows further. To
prevent this, defer the application of seccomp policies after the
initialisation phase, before the main loop, that's where we expect bad
things to happen, potentially. This way, we get back to 22 allowed
syscalls for passt and 34 for pasta, on x86_64.

While at it, move #syscalls notes to specific code paths wherever it
conceptually makes sense.

We have to open all the file handles we'll ever need before
sandboxing:

- the packet capture file can only be opened once, drop instance
  numbers from the default path and use the (pre-sandbox) PID instead

- /proc/net/tcp{,v6} and /proc/net/udp{,v6}, for automatic detection
  of bound ports in pasta mode, are now opened only once, before
  sandboxing, and their handles are stored in the execution context

- the UNIX domain socket for passt is also bound only once, before
  sandboxing: to reject clients after the first one, instead of
  closing the listening socket, keep it open, accept and immediately
  discard new connection if we already have a valid one

Clarify the (unchanged) behaviour for --netns-only in the man page.

To actually make passt and pasta processes run in a separate PID
namespace, we need to unshare(CLONE_NEWPID) before forking to
background (if configured to do so). Introduce a small daemon()
implementation, __daemon(), that additionally saves the PID file
before forking. While running in foreground, the process itself can't
move to a new PID namespace (a process can't change the notion of its
own PID): mention that in the man page.

For some reason, fork() in a detached PID namespace causes SIGTERM
and SIGQUIT to be ignored, even if the handler is still reported as
SIG_DFL: add a signal handler that just exits.

We can now drop most of the pasta_child_handler() implementation,
that took care of terminating all processes running in the same
namespace, if pasta started a shell: the shell itself is now the
init process in that namespace, and all children will terminate
once the init process exits.

Issuing 'echo $$' in a detached PID namespace won't return the
actual namespace PID as seen from the init namespace: adapt
demo and test setup scripts to reflect that.

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

README.md

@@ -232,9 +232,10 @@ speeding up local connections, and usually requiring NAT. _pasta_:
   `seccomp`](/passt/tree/seccomp.sh))
 * ✅ root operation not allowed outside user namespaces
 * ✅ all capabilities dropped, other than `CAP_NET_BIND_SERVICE` (if granted)
+* ✅ with default options, user, mount, IPC, UTS, PID namespaces are detached
 * ✅ no external dependencies (other than a standard C library)
-* ✅ restrictive seccomp profiles (50 syscalls allowed for _passt_, 62 for
+* ✅ restrictive seccomp profiles (22 syscalls allowed for _passt_, 34 for
-  _pasta_)
+  _pasta_ on x86_64)
 * ✅ static checkers in continuous integration (clang-tidy, cppcheck)
 * 🛠️ rework of TCP state machine (flags instead of states), TCP timers, and code
   de-duplication

conf.c

@@ -10,8 +10,6 @@
  *
  * Copyright (c) 2020-2021 Red Hat GmbH
  * Author: Stefano Brivio <sbrivio@redhat.com>
- *
- * #syscalls stat|statx
  */
 
 #include <arpa/inet.h>
@@ -46,31 +44,31 @@
  */
 void get_bound_ports(struct ctx *c, int ns, uint8_t proto)
 {
-	uint8_t *udp_map, *udp_exclude, *tcp_map, *tcp_exclude;
+	uint8_t *udp_map, *udp_excl, *tcp_map, *tcp_excl;
 
 	if (ns) {
 		udp_map = c->udp.port_to_tap;
-		udp_exclude = c->udp.port_to_init;
+		udp_excl = c->udp.port_to_init;
 		tcp_map = c->tcp.port_to_tap;
-		tcp_exclude = c->tcp.port_to_init;
+		tcp_excl = c->tcp.port_to_init;
 	} else {
 		udp_map = c->udp.port_to_init;
-		udp_exclude = c->udp.port_to_tap;
+		udp_excl = c->udp.port_to_tap;
 		tcp_map = c->tcp.port_to_init;
-		tcp_exclude = c->tcp.port_to_tap;
+		tcp_excl = c->tcp.port_to_tap;
 	}
 
 	if (proto == IPPROTO_UDP) {
 		memset(udp_map, 0, USHRT_MAX / 8);
-		procfs_scan_listen("udp",  udp_map, udp_exclude);
+		procfs_scan_listen(c, IPPROTO_UDP, V4, ns, udp_map, udp_excl);
-		procfs_scan_listen("udp6", udp_map, udp_exclude);
+		procfs_scan_listen(c, IPPROTO_UDP, V6, ns, udp_map, udp_excl);
 
-		procfs_scan_listen("tcp",  udp_map, udp_exclude);
+		procfs_scan_listen(c, IPPROTO_TCP, V4, ns, udp_map, udp_excl);
-		procfs_scan_listen("tcp6", udp_map, udp_exclude);
+		procfs_scan_listen(c, IPPROTO_TCP, V6, ns, udp_map, udp_excl);
 	} else if (proto == IPPROTO_TCP) {
 		memset(tcp_map, 0, USHRT_MAX / 8);
-		procfs_scan_listen("tcp",  tcp_map, tcp_exclude);
+		procfs_scan_listen(c, IPPROTO_TCP, V4, ns, tcp_map, tcp_excl);
-		procfs_scan_listen("tcp6", tcp_map, tcp_exclude);
+		procfs_scan_listen(c, IPPROTO_TCP, V6, ns, tcp_map, tcp_excl);
 	}
 }
 
@@ -367,7 +365,7 @@ static int conf_ns_check(void *arg)
 static int conf_ns_opt(struct ctx *c,
 		       char *nsdir, char *conf_userns, const char *optarg)
 {
-	int ufd = 0, nfd = 0, try, ret, netns_only_reset = c->netns_only;
+	int ufd = -1, nfd = -1, try, ret, netns_only_reset = c->netns_only;
 	char userns[PATH_MAX] = { 0 }, netns[PATH_MAX];
 	char *endptr;
 	pid_t pid;
@@ -416,7 +414,7 @@ static int conf_ns_opt(struct ctx *c,
 
 		nfd = open(netns, O_RDONLY);
 
-		if (nfd >= 0 && ufd >= 0) {
+		if (nfd >= 0 && (ufd >= 0 || c->netns_only)) {
 			c->pasta_netns_fd = nfd;
 			c->pasta_userns_fd = ufd;
 
@@ -425,10 +423,10 @@ static int conf_ns_opt(struct ctx *c,
 				return 0;
 		}
 
-		if (nfd > 0)
+		if (nfd >= 0)
 			close(nfd);
 
-		if (ufd > 0)
+		if (ufd >= 0)
 			close(ufd);
 	}
 
@@ -565,9 +563,9 @@ static void usage(const char *name)
 	info(   "    if FILE is not given, log to:");
 
 	if (strstr(name, "pasta") || strstr(name, "passt4netns"))
-		info("      /tmp/pasta_ISO8601-TIMESTAMP_INSTANCE-NUMBER.pcap");
+		info("      /tmp/pasta_ISO8601-TIMESTAMP_PID.pcap");
 	else
-		info("      /tmp/passt_ISO8601-TIMESTAMP_INSTANCE-NUMBER.pcap");
+		info("      /tmp/passt_ISO8601-TIMESTAMP_PID.pcap");
 
 	info(   "  -P, --pid FILE	Write own PID to the given file");
 	info(   "  -m, --mtu MTU	Assign MTU via DHCP/NDP");
@@ -664,7 +662,7 @@ pasta_opts:
 	info(   "    SPEC is as described above");
 	info(   "    default: auto");
 	info(   "  --userns NSPATH 	Target user namespace to join");
-	info(   "  --netns-only		Don't join or create user namespace");
+	info(   "  --netns-only		Don't join existing user namespace");
 	info(   "    implied if PATH or NAME are given without --userns");
 	info(   "  --nsrun-dir		Directory for nsfs mountpoints");
 	info(   "    default: " NETNS_RUN_DIR);
@@ -1170,7 +1168,7 @@ void conf(struct ctx *c, int argc, char **argv)
 		usage(argv[0]);
 	}
 
-	if (c->mode == MODE_PASTA && c->pasta_netns_fd <= 0)
+	if (c->mode == MODE_PASTA && c->pasta_netns_fd == -1)
 		pasta_start_ns(c);
 
 	if (nl_sock_init(c)) {
@@ -1216,6 +1214,11 @@ void conf(struct ctx *c, int argc, char **argv)
 	c->tcp.init_detect_ports = c->udp.init_detect_ports = 0;
 
 	if (c->mode == MODE_PASTA) {
+		c->proc_net_tcp[V4][0] = c->proc_net_tcp[V4][1] = -1;
+		c->proc_net_tcp[V6][0] = c->proc_net_tcp[V6][1] = -1;
+		c->proc_net_udp[V4][0] = c->proc_net_udp[V4][1] = -1;
+		c->proc_net_udp[V6][0] = c->proc_net_udp[V6][1] = -1;
+
 		if (!tcp_tap || tcp_tap == PORT_AUTO) {
 			c->tcp.ns_detect_ports = 1;
 			ns_ports_arg.proto = IPPROTO_TCP;

passt.1

@@ -80,7 +80,8 @@ Don't print informational messages.
 
 .TP
 .BR \-f ", " \-\-foreground
-Don't run in background.
+Don't run in background. This implies that the process is not moved to a
+detached PID namespace after starting, because the PID itself cannot change.
 Default is to fork into background.
 
 .TP
@@ -100,14 +101,13 @@ Capture tap-facing (that is, guest-side or namespace-side) network packets to
 
 If \fIfile\fR is not given, capture packets to
 
-	\fB/tmp/passt_\fIISO8601-timestamp\fR_\fIinstance-number\fB.pcap\fR
+	\fB/tmp/passt_\fIISO8601-timestamp\fR_\fIPID\fB.pcap\fR
 
 in \fBpasst\fR mode and to
 
-	\fB/tmp/pasta_\fIISO8601-timestamp\fR_\fIinstance-number\fB.pcap\fR
+	\fB/tmp/pasta_\fIISO8601-timestamp\fR_\fIPID\fB.pcap\fR
 
-in \fBpasta\fR mode, where \fIinstance-number\fR is a progressive count of
+in \fBpasta\fR mode, where \fIPID\fR is the ID of the running process.
-other detected instances running on the same host.
 
 .TP
 .BR \-P ", " \-\-pid " " \fIfile
@@ -379,8 +379,9 @@ This option requires PID, PATH or NAME to be specified.
 
 .TP
 .BR \-\-netns-only
-Join or create only the network namespace, not a user namespace. This is implied
+Join only a target network namespace, not a user namespace, and don't create one
-if PATH or NAME are given without \-\-userns.
+for sandboxing purposes either. This is implied if PATH or NAME are given
+without \-\-userns.
 
 .TP
 .BR \-\-nsrun-dir " " \fIpath

passt.c

@@ -30,7 +30,9 @@
 #include <sys/mman.h>
 #include <sys/resource.h>
 #include <sys/uio.h>
+#include <sys/syscall.h>
 #include <sys/wait.h>
+#include <sys/mount.h>
 #include <netinet/ip.h>
 #include <net/ethernet.h>
 #include <stdlib.h>
@@ -53,7 +55,6 @@
 #include <linux/seccomp.h>
 #include <linux/audit.h>
 #include <linux/filter.h>
-#include <linux/capability.h>
 #include <linux/icmpv6.h>
 
 #include "util.h"
@@ -228,42 +229,61 @@ static void check_root(void)
 }
 
 /**
- * drop_caps() - Drop capabilities we might have except for CAP_NET_BIND_SERVICE
+ * sandbox() - Unshare IPC, mount, PID, UTS, and user namespaces, "unmount" root
+ *
+ * Return: negative error code on failure, zero on success
  */
-static void drop_caps(void)
+static int sandbox(struct ctx *c)
 {
-	int i;
+	int flags = CLONE_NEWIPC | CLONE_NEWNS | CLONE_NEWUTS;
 
-	for (i = 0; i < 64; i++) {
+	errno = 0;
-		if (i == CAP_NET_BIND_SERVICE)
-			continue;
 
-		prctl(PR_CAPBSET_DROP, i, 0, 0, 0);
+	if (!c->netns_only) {
+		if (c->pasta_userns_fd == -1)
+			flags |= CLONE_NEWUSER;
+		else
+			setns(c->pasta_userns_fd, CLONE_NEWUSER);
 	}
+
+	c->pasta_userns_fd = -1;
+
+	/* If we run in foreground, we have no chance to actually move to a new
+	 * PID namespace. For passt, use CLONE_NEWPID anyway, in case somebody
+	 * ever gets around seccomp profiles -- there's no harm in passing it.
+	 */
+	if (!c->foreground || c->mode == MODE_PASST)
+		flags |= CLONE_NEWPID;
+
+	unshare(flags);
+
+	mount("", "/", "", MS_UNBINDABLE | MS_REC, NULL);
+	mount("", TMPDIR, "tmpfs", MS_NODEV | MS_NOEXEC | MS_NOSUID | MS_RDONLY,
+	      "nr_inodes=2,nr_blocks=0");
+	chdir(TMPDIR);
+	syscall(SYS_pivot_root, ".", ".");
+	umount2(".", MNT_DETACH | UMOUNT_NOFOLLOW);
+
+	if (errno)
+		return -errno;
+
+	drop_caps();	/* Relative to the new user namespace this time. */
+
+	return 0;
 }
 
 /**
- * pid_file() - Write own PID to file, if configured
+ * exit_handler() - Signal handler for SIGQUIT and SIGTERM
- * @c:		Execution context
+ * @unused:	Unused, handler deals with SIGQUIT and SIGTERM only
+ *
+ * TODO: After unsharing the PID namespace and forking, SIG_DFL for SIGTERM and
+ * SIGQUIT unexpectedly doesn't cause the process to terminate, figure out why.
  */
-static void pid_file(struct ctx *c) {
+void exit_handler(int signal)
-	char pid_buf[12];
+{
-	int pid_fd, n;
+	(void)signal;
 
-	if (!*c->pid_file)
+	exit(EXIT_SUCCESS);
-		return;
-
-	pid_fd = open(c->pid_file, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
-	if (pid_fd < 0)
-		return;
-
-	n = snprintf(pid_buf, sizeof(pid_buf), "%i\n", getpid());
-
-	if (write(pid_fd, pid_buf, n) < 0) {
-		perror("PID file write");
-		exit(EXIT_FAILURE);
-	}
-	close(pid_fd);
 }
 
 /**
@@ -273,36 +293,36 @@ static void pid_file(struct ctx *c) {
  *
  * Return: non-zero on failure
  *
- * #syscalls read write open|openat close fork|clone dup2|dup3 ioctl writev
+ * #syscalls read write writev
- * #syscalls socket bind connect getsockopt setsockopt recvfrom sendto shutdown
+ * #syscalls socket bind connect getsockopt setsockopt s390x:socketcall close
- * #syscalls accept4 accept listen set_robust_list getrlimit setrlimit
+ * #syscalls recvfrom sendto shutdown ppc64le:recv ppc64le:send
- * #syscalls openat fcntl lseek clone setsid exit exit_group getpid chdir
+ * #syscalls accept4|accept listen
- * #syscalls epoll_ctl epoll_create1 epoll_wait|epoll_pwait epoll_pwait
+ * #syscalls epoll_ctl epoll_wait|epoll_pwait epoll_pwait clock_gettime
- * #syscalls prlimit64 clock_gettime fstat|newfstat newfstatat syslog
- * #syscalls ppc64le:_llseek ppc64le:recv ppc64le:send ppc64le:getuid
- * #syscalls ppc64:_llseek ppc64:recv ppc64:send ppc64:getuid ppc64:ugetrlimit
- * #syscalls s390x:socketcall s390x:sigreturn
- * #syscalls:pasta rt_sigreturn|sigreturn ppc64:sigreturn ppc64:fcntl64
  */
 int main(int argc, char **argv)
 {
+	int nfds, i, devnull_fd = -1, pidfile_fd = -1;
 	struct epoll_event events[EPOLL_EVENTS];
 	struct ctx c = { 0 };
 	struct rlimit limit;
 	struct timespec now;
+	struct sigaction sa;
 	char *log_name;
-	int nfds, i;
 
 #ifndef PASST_LEGACY_NO_OPTIONS
 	check_root();
 #endif
 	drop_caps();
 
-	if (strstr(argv[0], "pasta") || strstr(argv[0], "passt4netns")) {
+	c.pasta_userns_fd = c.pasta_netns_fd = c.fd_tap = c.fd_tap_listen = -1;
-		struct sigaction sa;
 
-		sigemptyset(&sa.sa_mask);
+	sigemptyset(&sa.sa_mask);
-		sa.sa_flags = 0;
+	sa.sa_flags = 0;
+	sa.sa_handler = exit_handler;
+	sigaction(SIGTERM, &sa, NULL);
+	sigaction(SIGQUIT, &sa, NULL);
+
+	if (strstr(argv[0], "pasta") || strstr(argv[0], "passt4netns")) {
 		sa.sa_handler = pasta_child_handler;
 		sigaction(SIGCHLD, &sa, NULL);
 		signal(SIGPIPE, SIG_IGN);
@@ -323,8 +343,6 @@ int main(int argc, char **argv)
 
 	conf(&c, argc, argv);
 
-	seccomp(&c);
-
 	if (!c.debug && (c.stderr || isatty(fileno(stdout))))
 		__openlog(log_name, LOG_PERROR, LOG_DAEMON);
 
@@ -369,12 +387,26 @@ int main(int argc, char **argv)
 	else
 		__setlogmask(LOG_UPTO(LOG_INFO));
 
-	if (!c.foreground && daemon(0, 0)) {
+	pcap_init(&c);
-		perror("daemon");
+
+	if (!c.foreground)
+		devnull_fd = open("/dev/null", O_RDWR);
+
+	if (*c.pid_file)
+		pidfile_fd = open(c.pid_file,
+				  O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
+
+	if (sandbox(&c)) {
+		err("Failed to sandbox process, exiting\n");
 		exit(EXIT_FAILURE);
 	}
 
-	pid_file(&c);
+	if (!c.foreground)
+		__daemon(pidfile_fd, devnull_fd);
+	else
+		write_pidfile(pidfile_fd, getpid());
+
+	seccomp(&c);
 
 	timer_init(&c, &now);
 loop:

passt.h

@@ -99,8 +99,10 @@ enum passt_modes {
  * @pcap:		Path for packet capture file
  * @pid_file:		Path to PID file, empty string if not configured
  * @pasta_netns_fd:	File descriptor for network namespace in pasta mode
- * @pasta_userns_fd:	File descriptor for user 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
+ * @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
  * @fd_tap_listen:	File descriptor for listening AF_UNIX socket, if any
  * @fd_tap:		File descriptor for AF_UNIX socket or tuntap device
@@ -155,6 +157,9 @@ struct ctx {
 	int pasta_userns_fd;
 	int netns_only;
 
+	int proc_net_tcp[IP_VERSIONS][2];
+	int proc_net_udp[IP_VERSIONS][2];
+
 	int epollfd;
 	int fd_tap_listen;
 	int fd_tap;

pasta.c

@@ -11,9 +11,8 @@
  * Copyright (c) 2020-2021 Red Hat GmbH
  * Author: Stefano Brivio <sbrivio@redhat.com>
  *
- * #syscalls:pasta clone unshare waitid kill execve exit_group rt_sigprocmask
+ * #syscalls:pasta clone waitid exit exit_group rt_sigprocmask
- * #syscalls:pasta geteuid getdents64|getdents readlink|readlinkat setsid
+ * #syscalls:pasta rt_sigreturn|sigreturn ppc64:sigreturn s390x:sigreturn
- * #syscalls:pasta nanosleep clock_nanosleep
  */
 
 #include <sched.h>
@@ -40,75 +39,8 @@
 #include "passt.h"
 #include "netlink.h"
 
-/* PID of child, in case we created a namespace, and its procfs link */
+/* PID of child, in case we created a namespace */
 static int pasta_child_pid;
-static char pasta_child_ns[PATH_MAX];
-
-/**
- * pasta_ns_cleanup() - Look for processes in namespace, terminate them
- */
-static void pasta_ns_cleanup(void)
-{
-	char proc_path[PATH_MAX], ns_link[PATH_MAX], buf[BUFSIZ];
-	int recheck = 0, found = 0, waited = 0;
-	int dir_fd, n;
-
-	if (!*pasta_child_ns)
-		return;
-
-loop:
-	if ((dir_fd = open("/proc", O_RDONLY | O_DIRECTORY)) < 0)
-		return;
-
-	while ((n = syscall(SYS_getdents64, dir_fd, buf, BUFSIZ)) > 0) {
-		struct dirent *dp = (struct dirent *)buf;
-		int pos = 0;
-
-		while (dp->d_reclen && pos < n) {
-			pid_t pid;
-
-			errno = 0;
-			pid = strtol(dp->d_name, NULL, 0);
-			if (!pid || errno)
-				goto next;
-
-			snprintf(proc_path, PATH_MAX, "/proc/%i/ns/net", pid);
-			if (readlink(proc_path, ns_link, PATH_MAX) < 0)
-				goto next;
-
-			if (!strncmp(ns_link, pasta_child_ns, PATH_MAX)) {
-				found = 1;
-				if (waited)
-					kill(pid, SIGKILL);
-				else
-					kill(pid, SIGQUIT);
-			}
-next:
-			dp = (struct dirent *)(buf + (pos += dp->d_reclen));
-		}
-	}
-
-	close(dir_fd);
-
-	if (!found)
-		return;
-
-	if (waited) {
-		if (recheck) {
-			info("Some processes in namespace didn't quit");
-		} else {
-			found = 0;
-			recheck = 1;
-			goto loop;
-		}
-		return;
-	}
-
-	info("Waiting for all processes in namespace to terminate");
-	sleep(1);
-	waited = 1;
-	goto loop;
-}
 
 /**
  * pasta_child_handler() - Exit once shell exits (if we started it), reap clones
@@ -120,12 +52,14 @@ void pasta_child_handler(int signal)
 
 	(void)signal;
 
+	if (signal != SIGCHLD)
+		return;
+
 	if (pasta_child_pid &&
 	    !waitid(P_PID, pasta_child_pid, &infop, WEXITED | WNOHANG)) {
-		if (infop.si_pid == pasta_child_pid) {
+		if (infop.si_pid == pasta_child_pid)
-			pasta_ns_cleanup();
 			exit(EXIT_SUCCESS);
-		}
+			/* Nothing to do, detached PID namespace going away */
 	}
 
 	waitid(P_ALL, 0, NULL, WEXITED | WNOHANG);
@@ -163,45 +97,31 @@ netns:
 }
 
 /**
- * pasta_start_ns() - Fork shell in new namespace if target ns is not given
+ * struct pasta_setup_ns_arg - Argument for pasta_setup_ns()
  * @c:		Execution context
+ * @euid:	Effective UID of caller
  */
-void pasta_start_ns(struct ctx *c)
+struct pasta_setup_ns_arg {
+	struct ctx *c;
+	int euid;
+};
+
+/**
+ * pasta_setup_ns() - Map credentials, enable access to ping sockets, run shell
+ * @arg:	See @pasta_setup_ns_arg
+ *
+ * Return: this function never returns
+ */
+static int pasta_setup_ns(void *arg)
 {
-	int euid = geteuid(), fd;
+	struct pasta_setup_ns_arg *a = (struct pasta_setup_ns_arg *)arg;
 	char *shell;
+	int fd;
 
-	c->foreground = 1;
+	if (!a->c->netns_only) {
-	if (!c->debug)
-		c->quiet = 1;
-
-	if ((pasta_child_pid = fork()) == -1) {
-		perror("fork");
-		exit(EXIT_FAILURE);
-	}
-
-	if (pasta_child_pid) {
-		char proc_path[PATH_MAX];
-
-		NS_CALL(pasta_wait_for_ns, c);
-
-		snprintf(proc_path, PATH_MAX, "/proc/%i/ns/net",
-			 pasta_child_pid);
-		if (readlink(proc_path, pasta_child_ns, PATH_MAX) < 0)
-			warn("Cannot read link to ns, won't clean up on exit");
-
-		return;
-	}
-
-	if (unshare(CLONE_NEWNET | (c->netns_only ? 0 : CLONE_NEWUSER))) {
-		perror("unshare");
-		exit(EXIT_FAILURE);
-	}
-
-	if (!c->netns_only) {
 		char buf[BUFSIZ];
 
-		snprintf(buf, BUFSIZ, "%i %i %i", 0, euid, 1);
+		snprintf(buf, BUFSIZ, "%i %i %i", 0, a->euid, 1);
 
 		fd = open("/proc/self/uid_map", O_WRONLY);
 		if (write(fd, buf, strlen(buf)) < 0)
@@ -234,6 +154,39 @@ void pasta_start_ns(struct ctx *c)
 	exit(EXIT_FAILURE);
 }
 
+/**
+ * pasta_start_ns() - Fork shell in new namespace if target ns is not given
+ * @c:		Execution context
+ */
+void pasta_start_ns(struct ctx *c)
+{
+	struct pasta_setup_ns_arg arg = { .c = c, .euid = geteuid() };
+	char ns_fn_stack[NS_FN_STACK_SIZE];
+
+	c->foreground = 1;
+	if (!c->debug)
+		c->quiet = 1;
+
+	pasta_child_pid = clone(pasta_setup_ns,
+				ns_fn_stack + sizeof(ns_fn_stack) / 2,
+				(c->netns_only ? 0 : CLONE_NEWNET) |
+				CLONE_NEWIPC | CLONE_NEWPID | CLONE_NEWUSER |
+				CLONE_NEWUTS,
+				(void *)&arg);
+
+	if (pasta_child_pid == -1) {
+		perror("clone");
+		exit(EXIT_FAILURE);
+	}
+
+	drop_caps();
+
+	if (pasta_child_pid) {
+		NS_CALL(pasta_wait_for_ns, c);
+		return;
+	}
+}
+
 /**
  * pasta_ns_conf() - Set up loopback and tap interfaces in namespace as needed
  * @c:		Execution context

pcap.c

@@ -167,9 +167,8 @@ fail:
 /**
  * pcap_init() - Initialise pcap file
  * @c:		Execution context
- * @index:	pcap name index: passt instance number or pasta netns socket
  */
-void pcap_init(struct ctx *c, int index)
+void pcap_init(struct ctx *c)
 {
 	struct timeval tv;
 
@@ -196,7 +195,7 @@ void pcap_init(struct ctx *c, int index)
 		snprintf(name + strlen(PCAP_PREFIX) + strlen(PCAP_ISO8601_STR),
 			 sizeof(name) - strlen(PCAP_PREFIX) -
 					strlen(PCAP_ISO8601_STR),
-			 "_%i.pcap", index);
+			 "_%i.pcap", getpid());
 
 		strncpy(c->pcap, name, PATH_MAX);
 	}

pcap.h

@@ -6,4 +6,4 @@
 void pcap(char *pkt, size_t len);
 void pcapm(struct msghdr *mh);
 void pcapmm(struct mmsghdr *mmh, unsigned int vlen);
-void pcap_init(struct ctx *c, int sock_index);
+void pcap_init(struct ctx *c);

slirp4netns.sh

@@ -10,7 +10,7 @@
 #
 # slirp4netns.sh - Compatibility wrapper for pasta, behaving like slirp4netns(1)
 #
-# WARNING: Draft quality, not really tested, --enable-sandbox not supported yet
+# WARNING: Draft quality, not really tested
 #
 # Copyright (c) 2021 Red Hat GmbH
 # Author: Stefano Brivio <sbrivio@redhat.com>

tap.c

@@ -11,7 +11,6 @@
  * Copyright (c) 2020-2021 Red Hat GmbH
  * Author: Stefano Brivio <sbrivio@redhat.com>
  *
- * #syscalls recvfrom sendto
  */
 
 #include <sched.h>
@@ -769,12 +768,10 @@ restart:
 }
 
 /**
- * tap_sock_init_unix() - Create and bind AF_UNIX socket, listen for connection
+ * tap_sock_unix_init() - Create and bind AF_UNIX socket, listen for connection
  * @c:		Execution context
- *
- * #syscalls:passt unlink|unlinkat
  */
-static void tap_sock_init_unix(struct ctx *c)
+static void tap_sock_unix_init(struct ctx *c)
 {
 	int fd = socket(AF_UNIX, SOCK_STREAM, 0), ex;
 	struct epoll_event ev = { 0 };
@@ -783,11 +780,6 @@ static void tap_sock_init_unix(struct ctx *c)
 	};
 	int i, ret;
 
-	if (c->fd_tap_listen != -1) {
-		epoll_ctl(c->epollfd, EPOLL_CTL_DEL, c->fd_tap_listen, &ev);
-		close(c->fd_tap_listen);
-	}
-
 	if (fd < 0) {
 		perror("UNIX socket");
 		exit(EXIT_FAILURE);
@@ -834,8 +826,6 @@ static void tap_sock_init_unix(struct ctx *c)
 	      S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
 #endif
 
-	pcap_init(c, i);
-
 	listen(fd, 0);
 
 	ev.data.fd = c->fd_tap_listen = fd;
@@ -852,19 +842,26 @@ static void tap_sock_init_unix(struct ctx *c)
 }
 
 /**
- * tap_sock_accept_unix() - Accept connection on listening socket
+ * tap_sock_unix_new() - Handle new connection on listening socket
  * @c:		Execution context
  */
-static void tap_sock_accept_unix(struct ctx *c)
+static void tap_sock_unix_new(struct ctx *c)
 {
 	struct epoll_event ev = { 0 };
 	int v = INT_MAX / 2;
 
-	c->fd_tap = accept(c->fd_tap_listen, NULL, NULL);
+	/* Another client is already connected: accept and close right away. */
+	if (c->fd_tap != -1) {
+		int discard = accept4(c->fd_tap_listen, NULL, NULL,
+				      SOCK_NONBLOCK);
 
-	epoll_ctl(c->epollfd, EPOLL_CTL_DEL, c->fd_tap_listen, &ev);
+		if (discard != -1)
-	close(c->fd_tap_listen);
+			close(discard);
-	c->fd_tap_listen = -1;
+
+		return;
+	}
+
+	c->fd_tap = accept4(c->fd_tap_listen, NULL, NULL, 0);
 
 	if (!c->low_rmem)
 		setsockopt(c->fd_tap, SOL_SOCKET, SO_RCVBUF, &v, sizeof(v));
@@ -884,8 +881,6 @@ static int tun_ns_fd = -1;
  * @c:		Execution context
  *
  * Return: 0
- *
- * #syscalls:pasta ioctl
  */
 static int tap_ns_tun(void *arg)
 {
@@ -907,7 +902,7 @@ static int tap_ns_tun(void *arg)
  * tap_sock_init_tun() - Set up tuntap file descriptor
  * @c:		Execution context
  */
-static void tap_sock_init_tun(struct ctx *c)
+static void tap_sock_tun_init(struct ctx *c)
 {
 	struct epoll_event ev = { 0 };
 
@@ -919,8 +914,6 @@ static void tap_sock_init_tun(struct ctx *c)
 
 	pasta_ns_conf(c);
 
-	pcap_init(c, c->pasta_netns_fd);
-
 	c->fd_tap = tun_ns_fd;
 
 	ev.data.fd = c->fd_tap;
@@ -937,12 +930,15 @@ void tap_sock_init(struct ctx *c)
 	if (c->fd_tap != -1) {
 		epoll_ctl(c->epollfd, EPOLL_CTL_DEL, c->fd_tap, NULL);
 		close(c->fd_tap);
+		c->fd_tap = -1;
 	}
 
-	if (c->mode == MODE_PASST)
+	if (c->mode == MODE_PASST) {
-		tap_sock_init_unix(c);
+		if (c->fd_tap_listen == -1)
-	else
+			tap_sock_unix_init(c);
-		tap_sock_init_tun(c);
+	} else {
+		tap_sock_tun_init(c);
+	}
 }
 
 /**
@@ -955,18 +951,18 @@ void tap_sock_init(struct ctx *c)
 void tap_handler(struct ctx *c, int fd, uint32_t events, struct timespec *now)
 {
 	if (fd == c->fd_tap_listen && events == EPOLLIN) {
-		tap_sock_accept_unix(c);
+		tap_sock_unix_new(c);
 		return;
 	}
 
 	if (events & (EPOLLRDHUP | EPOLLHUP | EPOLLERR))
-		goto fail;
+		goto reinit;
 
 	if ((c->mode == MODE_PASST && tap_handler_passt(c, now)) ||
 	    (c->mode == MODE_PASTA && tap_handler_pasta(c, now)))
-		goto fail;
+		goto reinit;
 
 	return;
-fail:
+reinit:
 	tap_sock_init(c);
 }

tcp.c

@@ -304,7 +304,7 @@
  * - SPLICE_FIN_TO:		FIN (EPOLLRDHUP) seen from connected socket
  * - SPLICE_FIN_BOTH:		FIN (EPOLLRDHUP) seen from both sides
  *
- * #syscalls pipe|pipe2 pipe2
+ * #syscalls:pasta pipe2|pipe fcntl ppc64:fcntl64
  */
 
 #include <sched.h>
@@ -3028,7 +3028,7 @@ static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref,
  * @ref:	epoll reference
  * @events:	epoll events bitmap
  *
- * #syscalls splice
+ * #syscalls:pasta splice
  */
 void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref,
 			     uint32_t events)
@@ -3374,7 +3374,7 @@ static void tcp_set_pipe_size(struct ctx *c)
 
 smaller:
 	for (i = 0; i < TCP_SPLICE_PIPE_POOL_SIZE * 2; i++) {
-		if (pipe(probe_pipe[i])) {
+		if (pipe2(probe_pipe[i], 0)) {
 			i++;
 			break;
 		}
@@ -3493,7 +3493,7 @@ static void tcp_sock_init_one(struct ctx *c, int ns, in_port_t port)
  * tcp_sock_init_ns() - Bind sockets in namespace for inbound connections
  * @arg:	Execution context
  *
- * Return: 0 on success, -1 on failure
+ * Return: 0
  */
 static int tcp_sock_init_ns(void *arg)
 {
@@ -3560,8 +3560,7 @@ static int tcp_sock_refill(void *arg)
 	int i, *p4, *p6;
 
 	if (a->ns) {
-		if (ns_enter(a->c))
+		ns_enter(a->c);
-			return 0;
 		p4 = ns_sock_pool4;
 		p6 = ns_sock_pool6;
 	} else {
@@ -3594,8 +3593,6 @@ static int tcp_sock_refill(void *arg)
  * @c:		Execution context
  *
  * Return: 0 on success, -1 on failure
- *
- * #syscalls getrandom
  */
 int tcp_sock_init(struct ctx *c, struct timespec *now)
 {

test/demo/passt

@@ -84,7 +84,8 @@ say	Now let's run 'passt' in the new namespace, and
 nl
 say	  enter this namespace from the guest terminal too.
 sleep	3
-pout	TARGET_PID echo $$
+guest	pstree -p | grep pasta
+gout	TARGET_PID pstree -p | grep pasta | sed -n 's/.*(\([0-9].*\))$/\1/p'
 sleep	1
 
 passtb	./passt -f -t 5201,5203

test/demo/pasta

@@ -58,7 +58,8 @@ say	For convenience, let's enter this namespace
 nl
 say	  from another terminal.
 sleep	3
-pout	TARGET_PID echo $$
+ns	pstree -p | grep pasta
+nsout	TARGET_PID pstree -p | grep pasta | sed -n 's/.*(\([0-9].*\))$/\1/p'
 sleep	1
 
 ns	nsenter -t __TARGET_PID__ -U -n --preserve-credentials
@@ -172,7 +173,7 @@ sleep	2
 passtb	perf record -g ./pasta
 sleep	2
 
-pout	TARGET_PID echo $$
+nsout	TARGET_PID pstree -p | grep pasta | sed -n 's/.*(\([0-9].*\))$/\1/p'
 sleep	1
 ns	nsenter -t __TARGET_PID__ -U -n --preserve-credentials
 sleep	5

test/lib/setup

@@ -115,13 +115,14 @@ setup_passt_in_ns() {
 	[ ${PCAP} -eq 1 ] && __opts="${__opts} -p /tmp/pasta_with_passt.pcap"
 	[ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
 
-	pane_run PASST "./pasta ${__opts} -t 10001,10002,10011,10012 -T 10003,10013 -u 10001,10002,10011,10012 -U 10003,10013"
+	__pid_file="$(mktemp)"
+	pane_run PASST "./pasta ${__opts} -t 10001,10002,10011,10012 -T 10003,10013 -u 10001,10002,10011,10012 -U 10003,10013 -P ${__pid_file}"
 	sleep 1
 	pane_run PASST ''
 	pane_wait PASST
-	pane_run PASST 'echo $$'
+	__pasta_pid="$(cat "${__pid_file}")"
-	pane_wait PASST
+	__ns_pid="$(cat /proc/${__pasta_pid}/task/${__pasta_pid}/children | cut -f1 -d' ')"
-	__ns_pid="$(pane_parse PASST)"
+	rm "${__pid_file}"
 
 	pane_run GUEST "nsenter -t ${__ns_pid} -U -n --preserve-credentials"
 	pane_run NS "nsenter -t ${__ns_pid} -U -n --preserve-credentials"
@@ -172,15 +173,18 @@ setup_two_guests() {
 	#  10004            | as server |  to init  |  to guest  |  to ns #2
 	#  10005            |           |           |  as server |  to ns #2
 
+	__pid1_file="$(mktemp)"
+	__pid2_file="$(mktemp)"
+
 	__opts=
 	[ ${PCAP} -eq 1 ] && __opts="${__opts} -p /tmp/pasta_1.pcap"
 	[ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
-	pane_run PASST_1 "./pasta ${__opts} -t 10001,10002 -T 10003,10004 -u 10001,10002 -U 10003,10004"
+	pane_run PASST_1 "./pasta ${__opts} -P ${__pid1_file} -t 10001,10002 -T 10003,10004 -u 10001,10002 -U 10003,10004"
 
 	__opts=
 	[ ${PCAP} -eq 1 ] && __opts="${__opts} -p /tmp/pasta_2.pcap"
 	[ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
-	pane_run PASST_2 "./pasta ${__opts} -t 10004,10005 -T 10003,10001 -u 10004,10005 -U 10003,10001"
+	pane_run PASST_2 "./pasta ${__opts} -P ${__pid2_file} -t 10004,10005 -T 10003,10001 -u 10004,10005 -U 10003,10001"
 
 	sleep 1
 	pane_run PASST_1 ''
@@ -188,12 +192,12 @@ setup_two_guests() {
 
 	pane_wait PASST_1
 	pane_wait PASST_2
-	pane_run PASST_1 'echo $$'
+	__pasta1_pid="$(cat "${__pid1_file}")"
-	pane_run PASST_2 'echo $$'
+	__ns1_pid="$(cat /proc/${__pasta1_pid}/task/${__pasta1_pid}/children | cut -f1 -d' ')"
-	pane_wait PASST_1
+	rm "${__pid1_file}"
-	pane_wait PASST_2
+	__pasta2_pid="$(cat "${__pid2_file}")"
-	__ns1_pid="$(pane_parse PASST_1)"
+	__ns2_pid="$(cat /proc/${__pasta2_pid}/task/${__pasta2_pid}/children | cut -f1 -d' ')"
-	__ns2_pid="$(pane_parse PASST_2)"
+	rm "${__pid2_file}"
 
 	pane_run GUEST_1 "nsenter -t ${__ns1_pid} -U -n --preserve-credentials"
 	pane_run GUEST_2 "nsenter -t ${__ns2_pid} -U -n --preserve-credentials"

udp.c

@@ -529,7 +529,9 @@ static int udp_splice_connect_ns(void *arg)
 
 	a = (struct udp_splice_connect_ns_arg *)arg;
 
-	ns_enter(a->c);
+	if (ns_enter(a->c))
+		return 0;
+
 	a->s = udp_splice_connect(a->c, a->v6, a->bound_sock, a->src, a->dst,
 				  UDP_BACK_TO_INIT);
 
@@ -1029,7 +1031,8 @@ int udp_sock_init_ns(void *arg)
 	struct ctx *c = (struct ctx *)arg;
 	int dst;
 
-	ns_enter(c);
+	if (ns_enter(c))
+		return 0;
 
 	for (dst = 0; dst < USHRT_MAX; dst++) {
 		if (!bitmap_isset(c->udp.port_to_init, dst))

util.c

@@ -16,6 +16,7 @@
 #include <stdio.h>
 #include <stdint.h>
 #include <stddef.h>
+#include <stdlib.h>
 #include <unistd.h>
 #include <arpa/inet.h>
 #include <net/ethernet.h>
@@ -23,6 +24,7 @@
 #include <netinet/tcp.h>
 #include <netinet/udp.h>
 #include <sys/epoll.h>
+#include <sys/prctl.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <fcntl.h>
@@ -32,6 +34,8 @@
 #include <time.h>
 #include <errno.h>
 
+#include <linux/capability.h>
+
 #include "util.h"
 #include "passt.h"
 
@@ -431,31 +435,51 @@ char *line_read(char *buf, size_t len, int fd)
 
 /**
  * procfs_scan_listen() - Set bits for listening TCP or UDP sockets from procfs
- * @name:	Corresponding name of file under /proc/net/
+ * @proto:	IPPROTO_TCP or IPPROTO_UDP
+ * @ip_version:	IP version, V4 or V6
+ * @ns:		Use saved file descriptors for namespace if set
  * @map:	Bitmap where numbers of ports in listening state will be set
  * @exclude:	Bitmap of ports to exclude from setting (and clear)
+ *
+ * #syscalls:pasta lseek ppc64le:_llseek ppc64:_llseek
  */
-void procfs_scan_listen(char *name, uint8_t *map, uint8_t *exclude)
+void procfs_scan_listen(struct ctx *c, uint8_t proto, int ip_version, int ns,
+			uint8_t *map, uint8_t *exclude)
 {
-	char line[BUFSIZ], path[PATH_MAX];
+	char line[BUFSIZ], *path;
 	unsigned long port;
 	unsigned int state;
-	int fd;
+	int *fd;
 
-	snprintf(path, PATH_MAX, "/proc/net/%s", name);
+	if (proto == IPPROTO_TCP) {
-	if ((fd = open(path, O_RDONLY)) < 0)
+		fd = &c->proc_net_tcp[ip_version][ns];
+		if (ip_version == V4)
+			path = "/proc/net/tcp";
+		else
+			path = "/proc/net/tcp6";
+	} else {
+		fd = &c->proc_net_udp[ip_version][ns];
+		if (ip_version == V4)
+			path = "/proc/net/udp";
+		else
+			path = "/proc/net/udp6";
+	}
+
+	if (*fd != -1)
+		lseek(*fd, 0, SEEK_SET);
+	else if ((*fd = open(path, O_RDONLY)) < 0)
 		return;
 
 	*line = 0;
-	line_read(line, sizeof(line), fd);
+	line_read(line, sizeof(line), *fd);
-	while (line_read(line, sizeof(line), fd)) {
+	while (line_read(line, sizeof(line), *fd)) {
 		/* NOLINTNEXTLINE(cert-err34-c): != 2 if conversion fails */
 		if (sscanf(line, "%*u: %*x:%lx %*x:%*x %x", &port, &state) != 2)
 			continue;
 
 		/* See enum in kernel's include/net/tcp_states.h */
-		if ((strstr(name, "tcp") && state != 0x0a) ||
+		if ((proto == IPPROTO_TCP && state != 0x0a) ||
-		    (strstr(name, "udp") && state != 0x07))
+		    (proto == IPPROTO_UDP && state != 0x07))
 			continue;
 
 		if (bitmap_isset(exclude, port))
@@ -463,25 +487,98 @@ void procfs_scan_listen(char *name, uint8_t *map, uint8_t *exclude)
 		else
 			bitmap_set(map, port);
 	}
-
-	close(fd);
 }
 
 /**
- * ns_enter() - Enter configured network and user namespaces
+ * drop_caps() - Drop capabilities we might have except for CAP_NET_BIND_SERVICE
+ */
+void drop_caps(void)
+{
+	int i;
+
+	for (i = 0; i < 64; i++) {
+		if (i == CAP_NET_BIND_SERVICE)
+			continue;
+
+		prctl(PR_CAPBSET_DROP, i, 0, 0, 0);
+	}
+}
+
+/**
+ * ns_enter() - Enter configured user (unless already joined) and network ns
  * @c:		Execution context
  *
- * Return: 0 on success, -1 on failure
+ * Return: 0, won't return on failure
  *
  * #syscalls:pasta setns
  */
 int ns_enter(struct ctx *c)
 {
-	if (!c->netns_only && setns(c->pasta_userns_fd, CLONE_NEWUSER))
+	if (!c->netns_only &&
-		return -errno;
+	    c->pasta_userns_fd != -1 &&
+	    setns(c->pasta_userns_fd, CLONE_NEWUSER))
+		exit(EXIT_FAILURE);
 
 	if (setns(c->pasta_netns_fd, CLONE_NEWNET))
-		return -errno;
+		exit(EXIT_FAILURE);
+
+	return 0;
+}
+
+/**
+ * pid_file() - Write PID to file, if requested to do so, and close it
+ * @fd:		Open PID file descriptor, closed on exit, -1 to skip writing it
+ * @pid:	PID value to write
+ */
+void write_pidfile(int fd, pid_t pid) {
+	char pid_buf[12];
+	int n;
+
+	if (fd == -1)
+		return;
+
+	n = snprintf(pid_buf, sizeof(pid_buf), "%i\n", pid);
+
+	if (write(fd, pid_buf, n) < 0) {
+		perror("PID file write");
+		exit(EXIT_FAILURE);
+	}
+
+	close(fd);
+}
+
+/**
+ * __daemon() - daemon()-like function writing PID file before parent exits
+ * @pidfile_fd:	Open PID file descriptor
+ * @devnull_fd:	Open file descriptor for /dev/null
+ *
+ * Return: child PID on success, won't return on failure
+ */
+int __daemon(int pidfile_fd, int devnull_fd)
+{
+	pid_t pid = fork();
+
+	if (pid == -1) {
+		perror("fork");
+		exit(EXIT_FAILURE);
+	}
+
+	if (pid) {
+		write_pidfile(pidfile_fd, pid);
+		exit(EXIT_SUCCESS);
+	}
+
+	errno = 0;
+
+	setsid();
+
+	dup2(devnull_fd, STDIN_FILENO);
+	dup2(devnull_fd, STDOUT_FILENO);
+	dup2(devnull_fd, STDERR_FILENO);
+	close(devnull_fd);
+
+	if (errno)
+		exit(EXIT_FAILURE);
 
 	return 0;
 }

util.h

@@ -54,6 +54,12 @@ void debug(const char *format, ...);
 #define STRINGIFY(x)	#x
 #define STR(x)		STRINGIFY(x)
 
+#ifdef P_tmpdir
+#define TMPDIR		P_tmpdir
+#else
+#define TMPDIR		"/tmp"
+#endif
+
 #define V4		0
 #define V6		1
 #define IP_VERSIONS	2
@@ -202,5 +208,9 @@ void bitmap_set(uint8_t *map, int bit);
 void bitmap_clear(uint8_t *map, int bit);
 int bitmap_isset(const uint8_t *map, int bit);
 char *line_read(char *buf, size_t len, int fd);
-void procfs_scan_listen(char *name, uint8_t *map, uint8_t *exclude);
+void procfs_scan_listen(struct ctx *c, uint8_t proto, int ip_version, int ns,
+			uint8_t *map, uint8_t *exclude);
+void drop_caps(void);
 int ns_enter(struct ctx *c);
+void write_pidfile(int fd, pid_t pid);
+int __daemon(int pidfile_fd, int devnull_fd);