I have been testing ktls and sockmap lately and noticed that neither
was handling sk_write_space events correctly. We need to ensure
these events are pushed down to the lower layer in all cases to
handle the case where the lower layer sendpage call has called
sk_wait_event and needs to be woken up. Without this I see
occosional stalls of sndtimeo length while we wait for the
timeout value even though space is available.
Two fixes below. Thanks.
---
John Fastabend (2):
tls: possible hang when do_tcp_sendpages hits sndbuf is full case
bpf: sockmap: write_space events need to be passed to TCP handler
kernel/bpf/sockmap.c | 3 +++
net/tls/tls_main.c | 9 +++++++--
2 files changed, 10 insertions(+), 2 deletions(-)
--
Signature
