[PATCH 5/6] tap: Re-introduce EPOLLET for tap connections

Since 4684f603446b ("tap: Don't use EPOLLET on Qemu sockets") we've
only
used level-triggered events for the tap device.  Prior to that we used it
inconsistently which caused some problems.

We want to add support for EPOLLOUT events on the tap connection, and
without EPOLLET that would require toggling EPOLLOUT on and off, which is
awkward.  So, re-introduce EPOLLET, but now use it uniformly for all tap
modes.  The main change this requires is making sure on EPOLLIN we loop
until all there's no more data to process.

Signed-off-by: David Gibson <david(a)gibson.dropbear.id.au>
---
 tap.c | 28 +++++++++++++++++++---------
 1 file changed, 19 insertions(+), 9 deletions(-)

diff --git a/tap.c b/tap.c
index 2fbcef04..d7d3fc19 100644
--- a/tap.c
+++ b/tap.c
@@ -985,8 +985,10 @@ static void tap_sock_reset(struct ctx *c)
  * tap_passt_input() - Handler for new data on the socket to qemu
  * @c:		Execution context
  * @now:	Current timestamp
+ *
+ * Return: true if there may be additional data to read, false otherwise
  */
-static void tap_passt_input(struct ctx *c, const struct timespec *now)
+static bool tap_passt_input(struct ctx *c, const struct timespec *now)
 {
 	static const char *partial_frame;
 	static ssize_t partial_len = 0;
@@ -1013,7 +1015,7 @@ static void tap_passt_input(struct ctx *c, const struct timespec
*now)
 			err_perror("Receive error on guest connection, reset");
 			tap_sock_reset(c);
 		}
-		return;
+		return false;
 	}
 
 	p = pkt_buf;
@@ -1025,7 +1027,7 @@ static void tap_passt_input(struct ctx *c, const struct timespec
*now)
 		if (l2len < sizeof(struct ethhdr) || l2len > ETH_MAX_MTU) {
 			err("Bad frame size from guest, resetting connection");
 			tap_sock_reset(c);
-			return;
+			return false;
 		}
 
 		if (l2len + sizeof(uint32_t) > (size_t)n)
@@ -1045,6 +1047,8 @@ static void tap_passt_input(struct ctx *c, const struct timespec
*now)
 	partial_frame = p;
 
 	tap_handler(c, now);
+
+	return true;
 }
 
 /**
@@ -1062,15 +1066,18 @@ void tap_handler_passt(struct ctx *c, uint32_t events,
 	}
 
 	if (events & EPOLLIN)
-		tap_passt_input(c, now);
+		while (tap_passt_input(c, now))
+			;
 }
 
 /**
  * tap_passt_input() - Handler for new data on the socket to qemu
  * @c:		Execution context
  * @now:	Current timestamp
+ *
+ * Return: true if there may be additional data to read, false otherwise
  */
-static void tap_pasta_input(struct ctx *c, const struct timespec *now)
+static bool tap_pasta_input(struct ctx *c, const struct timespec *now)
 {
 	ssize_t n, len;
 
@@ -1101,6 +1108,8 @@ static void tap_pasta_input(struct ctx *c, const struct timespec
*now)
 		die("EOF on tap device, exiting");
 
 	tap_handler(c, now);
+
+	return len > 0;
 }
 
 /**
@@ -1116,7 +1125,8 @@ void tap_handler_pasta(struct ctx *c, uint32_t events,
 		die("Disconnect event on /dev/net/tun device, exiting");
 
 	if (events & EPOLLIN)
-		tap_pasta_input(c, now);
+		while (tap_pasta_input(c, now))
+			;
 }
 
 /**
@@ -1250,7 +1260,7 @@ void tap_listen_handler(struct ctx *c, uint32_t events)
 		trace("tap: failed to set SO_SNDBUF to %i", v);
 
 	ref.fd = c->fd_tap;
-	ev.events = EPOLLIN | EPOLLRDHUP;
+	ev.events = EPOLLIN | EPOLLRDHUP | EPOLLET;
 	ev.data.u64 = ref.u64;
 	epoll_ctl(c->epollfd, EPOLL_CTL_ADD, c->fd_tap, &ev);
 }
@@ -1306,7 +1316,7 @@ static void tap_sock_tun_init(struct ctx *c)
 	pasta_ns_conf(c);
 
 	ref.fd = c->fd_tap;
-	ev.events = EPOLLIN | EPOLLRDHUP;
+	ev.events = EPOLLIN | EPOLLRDHUP | EPOLLET;
 	ev.data.u64 = ref.u64;
 	epoll_ctl(c->epollfd, EPOLL_CTL_ADD, c->fd_tap, &ev);
 }
@@ -1339,7 +1349,7 @@ void tap_sock_init(struct ctx *c)
 		else
 			ref.type = EPOLL_TYPE_TAP_PASTA;
 
-		ev.events = EPOLLIN | EPOLLRDHUP;
+		ev.events = EPOLLIN | EPOLLRDHUP | EPOLLET;
 		ev.data.u64 = ref.u64;
 		epoll_ctl(c->epollfd, EPOLL_CTL_ADD, c->fd_tap, &ev);
 		return;
-- 
2.46.0