[PATCH gwhttpd 06/14] gwhttpd: Avoid endless busy spinning on `send()`

[Ammar Faizi <ammarfaizi2@gnuweeb.org>]

When we fail to send() due to EAGAIN, we retry the send(). However,
if we are spinning on this retry for so long, this will eat CPU cycle
and slow down the entire application. We should stop retrying at
some point. Add a loop counter and return -ENETDOWN if we are failing
too many times in the send() retry loop cycle.

Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
---
 gwhttpd.cpp | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/gwhttpd.cpp b/gwhttpd.cpp
index fbe219d..73a1d84 100644
--- a/gwhttpd.cpp
+++ b/gwhttpd.cpp
@@ -312,10 +312,14 @@ static int handle_new_client(struct server_state *state)
 static ssize_t send_to_client(struct client_sess *sess, const char *buf,
 			      size_t len)
 {
+	constexpr uint32_t max_try = 10;
+	uint32_t try_count = 0;
 	ssize_t ret;
 	int tmp;
 
 repeat:
+	if (unlikely(try_count++ >= max_try))
+		return -ENETDOWN;
 	ret = send(sess->fd, buf, len, MSG_DONTWAIT);
 	if (unlikely(ret < 0)) {
 		tmp = errno;
-- 
Ammar Faizi