poll(2): Add POLLRDHUP.

Teach poll(2) to support Linux-style POLLRDHUP events for sockets, if
requested.  Triggered when the remote peer shuts down writing or closes
its end.

Reviewed by:	kib
MFC after:	1 month
Differential Revision:	https://reviews.freebsd.org/D29757

5 files changed, 159 insertions, 3 deletions

diff --git a/lib/libc/sys/poll.2 b/lib/libc/sys/poll.2
index bea4aac82bd3..fec82db08944 100644
--- a/lib/libc/sys/poll.2
+++ b/lib/libc/sys/poll.2
@@ -28,7 +28,7 @@
 .\" (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 .\" THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .\"
-.Dd February 27, 2019
+.Dd April 27, 2021
 .Dt POLL 2
 .Os
 .Sh NAME
@@ -126,6 +126,15 @@ POLLOUT
 should never be present in the
 .Fa revents
 bitmask at the same time.
+.It POLLRDHUP
+Remote peer closed connection, or shut down writing.
+Unlike
+POLLHUP,
+POLLRDHUP
+must be present in the
+.Fa events
+bitmask to be reported.
+Applies only to stream sockets.
 .It POLLNVAL
 The file descriptor is not open,
 or in capability mode the file descriptor has insufficient rights.
@@ -261,6 +270,9 @@ function conforms to
 The
 .Fn ppoll
 is not specified by POSIX.
+The
+POLLRDHUP
+flag is not specified by POSIX, but is compatible with Linux and illumos.
 .Sh HISTORY
 The
 .Fn poll
diff --git a/sys/kern/uipc_socket.c b/sys/kern/uipc_socket.c
index 92a204aafef0..ae678136bade 100644
--- a/sys/kern/uipc_socket.c
+++ b/sys/kern/uipc_socket.c
@@ -3571,9 +3571,11 @@ sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
 					revents |= POLLHUP;
 			}
 		}
+		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
+			revents |= events & POLLRDHUP;
 		if (revents == 0) {
 			if (events &
-			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
+			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
 				selrecord(td, &so->so_rdsel);
 				so->so_rcv.sb_flags |= SB_SEL;
 			}
diff --git a/sys/sys/poll.h b/sys/sys/poll.h
index 6805704ca39f..02347ed09a13 100644
--- a/sys/sys/poll.h
+++ b/sys/sys/poll.h
@@ -71,6 +71,7 @@ struct pollfd {
 #if __BSD_VISIBLE
 /* General FreeBSD extension (currently only supported for sockets): */
 #define	POLLINIGNEOF	0x2000		/* like POLLIN, except ignore EOF */
+#define	POLLRDHUP	0x4000		/* half shut down */
 #endif
 
 /*
diff --git a/tests/sys/netinet/socket_afinet.c b/tests/sys/netinet/socket_afinet.c
index 54585086da23..985d67d83c99 100644
--- a/tests/sys/netinet/socket_afinet.c
+++ b/tests/sys/netinet/socket_afinet.c
@@ -31,6 +31,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/errno.h>
 #include <sys/socket.h>
 #include <netinet/in.h>
+#include <poll.h>
 
 #include <atf-c.h>
 
@@ -89,12 +90,152 @@ ATF_TC_BODY(socket_afinet_bind_ok, tc)
 	close(sd);
 }
 
+ATF_TC_WITHOUT_HEAD(socket_afinet_poll_no_rdhup);
+ATF_TC_BODY(socket_afinet_poll_no_rdhup, tc)
+{
+	int ss, ss2, cs, rc;
+	struct sockaddr_in sin;
+	struct pollfd pfd;
+	int one = 1;
+
+	/* Verify that we don't expose POLLRDHUP if not requested. */
+
+	/* Server setup. */
+	ss = socket(PF_INET, SOCK_STREAM, 0);
+	ATF_CHECK(ss >= 0);
+	rc = setsockopt(ss, SOL_SOCKET, SO_REUSEPORT, &one, sizeof(one));
+	ATF_CHECK_EQ(0, rc);
+	bzero(&sin, sizeof(sin));
+	sin.sin_family = AF_INET;
+	sin.sin_len = sizeof(sin);
+	sin.sin_port = htons(6666);
+	sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
+	rc = bind(ss, (struct sockaddr *)&sin, sizeof(sin));
+	ATF_CHECK_EQ(0, rc);
+	rc = listen(ss, 1);
+	ATF_CHECK_EQ(0, rc);
+
+	/* Client connects, server accepts. */
+	cs = socket(PF_INET, SOCK_STREAM, 0);
+	ATF_CHECK(cs >= 0);
+	rc = connect(cs, (struct sockaddr *)&sin, sizeof(sin));
+	ATF_CHECK_EQ(0, rc);
+	ss2 = accept(ss, NULL, NULL);
+	ATF_CHECK(ss2 >= 0);
+
+	/* Server can write, sees only POLLOUT. */
+	pfd.fd = ss2;
+	pfd.events = POLLIN | POLLOUT;
+	rc = poll(&pfd, 1, 0);
+	ATF_CHECK_EQ(1, rc);
+	ATF_CHECK_EQ(POLLOUT, pfd.revents);
+
+	/* Client closes socket! */
+	rc = close(cs);
+	ATF_CHECK_EQ(0, rc);
+
+	/*
+	 * Server now sees POLLIN, but not POLLRDHUP because we didn't ask.
+	 * Need non-zero timeout to wait for the FIN to arrive and trigger the
+	 * socket to become readable.
+	 */
+	pfd.fd = ss2;
+	pfd.events = POLLIN;
+	rc = poll(&pfd, 1, 60000);
+	ATF_CHECK_EQ(1, rc);
+	ATF_CHECK_EQ(POLLIN, pfd.revents);
+
+	close(ss2);
+	close(ss);
+}
+
+ATF_TC_WITHOUT_HEAD(socket_afinet_poll_rdhup);
+ATF_TC_BODY(socket_afinet_poll_rdhup, tc)
+{
+	int ss, ss2, cs, rc;
+	struct sockaddr_in sin;
+	struct pollfd pfd;
+	char buffer;
+	int one = 1;
+
+	/* Verify that server sees POLLRDHUP if it asks for it. */
+
+	/* Server setup. */
+	ss = socket(PF_INET, SOCK_STREAM, 0);
+	ATF_CHECK(ss >= 0);
+	rc = setsockopt(ss, SOL_SOCKET, SO_REUSEPORT, &one, sizeof(one));
+	ATF_CHECK_EQ(0, rc);
+	bzero(&sin, sizeof(sin));
+	sin.sin_family = AF_INET;
+	sin.sin_len = sizeof(sin);
+	sin.sin_port = htons(6666);
+	sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
+	rc = bind(ss, (struct sockaddr *)&sin, sizeof(sin));
+	ATF_CHECK_EQ(0, rc);
+	rc = listen(ss, 1);
+	ATF_CHECK_EQ(0, rc);
+
+	/* Client connects, server accepts. */
+	cs = socket(PF_INET, SOCK_STREAM, 0);
+	ATF_CHECK(cs >= 0);
+	rc = connect(cs, (struct sockaddr *)&sin, sizeof(sin));
+	ATF_CHECK_EQ(0, rc);
+	ss2 = accept(ss, NULL, NULL);
+	ATF_CHECK(ss2 >= 0);
+
+	/* Server can write, so sees POLLOUT. */
+	pfd.fd = ss2;
+	pfd.events = POLLIN | POLLOUT | POLLRDHUP;
+	rc = poll(&pfd, 1, 0);
+	ATF_CHECK_EQ(1, rc);
+	ATF_CHECK_EQ(POLLOUT, pfd.revents);
+
+	/* Client writes two bytes, server reads only one of them. */
+	rc = write(cs, "xx", 2);
+	ATF_CHECK_EQ(2, rc);
+	rc = read(ss2, &buffer, 1);
+	ATF_CHECK_EQ(1, rc);
+
+	/* Server can read, so sees POLLIN. */
+	pfd.fd = ss2;
+	pfd.events = POLLIN | POLLOUT | POLLRDHUP;
+	rc = poll(&pfd, 1, 0);
+	ATF_CHECK_EQ(1, rc);
+	ATF_CHECK_EQ(POLLIN | POLLOUT, pfd.revents);
+
+	/* Client closes socket! */
+	rc = close(cs);
+	ATF_CHECK_EQ(0, rc);
+
+	/*
+	 * Server sees Linux-style POLLRDHUP.  Note that this is the case even
+	 * though one byte of data remains unread.
+	 *
+	 * This races against the delivery of FIN caused by the close() above.
+	 * Sometimes (more likely when run under truss or if another system
+	 * call is added in between) it hits the path where sopoll_generic()
+	 * immediately sees SBS_CANTRCVMORE, and sometimes it sleeps with flag
+	 * SB_SEL so that it's woken up almost immediately and runs again,
+	 * which is why we need a non-zero timeout here.
+	 */
+	pfd.fd = ss2;
+	pfd.events = POLLRDHUP;
+	rc = poll(&pfd, 1, 60000);
+	ATF_CHECK_EQ(1, rc);
+	ATF_CHECK_EQ(POLLRDHUP, pfd.revents);
+
+	close(ss2);
+	close(ss);
+}
+
 ATF_TP_ADD_TCS(tp)
 {
 
 	ATF_TP_ADD_TC(tp, socket_afinet);
 	ATF_TP_ADD_TC(tp, socket_afinet_bind_zero);
 	ATF_TP_ADD_TC(tp, socket_afinet_bind_ok);
+	ATF_TP_ADD_TC(tp, socket_afinet_poll_no_rdhup);
+	ATF_TP_ADD_TC(tp, socket_afinet_poll_rdhup);
 
 	return atf_no_error();
 }
diff --git a/usr.bin/truss/syscalls.c b/usr.bin/truss/syscalls.c
index eaea3ad96765..798cd299582c 100644
--- a/usr.bin/truss/syscalls.c
+++ b/usr.bin/truss/syscalls.c
@@ -726,7 +726,7 @@ struct xlat {
 static struct xlat poll_flags[] = {
 	X(POLLSTANDARD) X(POLLIN) X(POLLPRI) X(POLLOUT) X(POLLERR)
 	X(POLLHUP) X(POLLNVAL) X(POLLRDNORM) X(POLLRDBAND)
-	X(POLLWRBAND) X(POLLINIGNEOF) XEND
+	X(POLLWRBAND) X(POLLINIGNEOF) X(POLLRDHUP) XEND
 };
 
 static struct xlat sigaction_flags[] = {