]> git.baikalelectronics.ru Git - kernel.git/commit
net: Do not clear the sock TX queue in sk_set_socket()
authorTariq Toukan <tariqt@mellanox.com>
Mon, 22 Jun 2020 20:26:04 +0000 (23:26 +0300)
committerDavid S. Miller <davem@davemloft.net>
Wed, 24 Jun 2020 03:49:12 +0000 (20:49 -0700)
commitfeaae454e18242bfa6e5530811501a198a4776a9
tree9822e7a28e953a0dd5ab520782a44eb5781585ca
parentf203b2d7921c700f5c65d92387589d07656b7e30
net: Do not clear the sock TX queue in sk_set_socket()

Clearing the sock TX queue in sk_set_socket() might cause unexpected
out-of-order transmit when called from sock_orphan(), as outstanding
packets can pick a different TX queue and bypass the ones already queued.

This is undesired in general. More specifically, it breaks the in-order
scheduling property guarantee for device-offloaded TLS sockets.

Remove the call to sk_tx_queue_clear() in sk_set_socket(), and add it
explicitly only where needed.

Fixes: 20e5169d94b8 ("net: Introduce sk_tx_queue_mapping")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/sock.h
net/core/sock.c