Skip to content

TCP: short writes handling

Jan Včelák requested to merge tcp-sndbuf-size into master

This changeset enables handling of short writes on TCP connections. Also, the default send/receive buffer sizes for server TCP sockets is increased to make fast-path processing more likely.

Formerly, Knot DNS running on FreeBSD 10 failed to answer large AXFR queries. The default send buffer is too small (32768 bytes) to fit the whole message. Since the sockets are non-blocking, a short write happened and the TCP handler code terminated the connection.

Merge request reports