diff options
| author |   Willem de Bruijn <willemb@google.com> | 2017-12-22 19:00:19 -0500 |
|---|---|---|
| committer |   David S. Miller <davem@davemloft.net> | 2017-12-27 16:44:13 -0500 |
| commit | 02583adeff12fa0a4d72558853a926867afb226c (patch) | |
| tree | 61bb4d25bbfbbaa8c0912f6781f1be2214b1252a /net | |
| parent | tcp: push full zerocopy packets (diff) | |
| download | linux-dev-02583adeff12fa0a4d72558853a926867afb226c.tar.xz linux-dev-02583adeff12fa0a4d72558853a926867afb226c.zip | |
tcp: place all zerocopy payload in frags
This avoids an unnecessary copy of 1-2KB and improves tso_fragment,
which has to fall back to tcp_fragment if skb->len != skb_data_len.
It also avoids a surprising inconsistency in notifications:
Zerocopy packets sent over loopback have their frags copied, so set
SO_EE_CODE_ZEROCOPY_COPIED in the notification. But this currently
does not happen for small packets, because when all data fits in the
linear fragment, data is not copied in skb_orphan_frags_rx.
Reported-by: Tom Deseyn <tom.deseyn@gmail.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net')
| -rw-r--r-- | net/ipv4/tcp.c | 9 |
1 files changed, 5 insertions, 4 deletions
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index 44102484a76f..947348872c3e 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -1186,7 +1186,7 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size) int flags, err, copied = 0; int mss_now = 0, size_goal, copied_syn = 0; bool process_backlog = false; - bool sg; + bool sg, zc = false; long timeo; flags = msg->msg_flags; @@ -1204,7 +1204,8 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size) goto out_err; } - if (!(sk_check_csum_caps(sk) && sk->sk_route_caps & NETIF_F_SG)) + zc = sk_check_csum_caps(sk) && sk->sk_route_caps & NETIF_F_SG; + if (!zc) uarg->zerocopy = 0; } @@ -1325,13 +1326,13 @@ new_segment: copy = msg_data_left(msg); /* Where to copy to? */ - if (skb_availroom(skb) > 0) { + if (skb_availroom(skb) > 0 && !zc) { /* We have some space in skb head. Superb! */ copy = min_t(int, copy, skb_availroom(skb)); err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy); if (err) goto do_fault; - } else if (!uarg || !uarg->zerocopy) { + } else if (!zc) { bool merge = true; int i = skb_shinfo(skb)->nr_frags; struct page_frag *pfrag = sk_page_frag(sk); |