mbox series

[RFC,0/5] net: remove sk skb caches

Message ID cover.1631888517.git.pabeni@redhat.com (mailing list archive)
Headers show
Series net: remove sk skb caches | expand

Message

Paolo Abeni Sept. 17, 2021, 3:38 p.m. UTC
Eric noted we would be better off reverting the sk
skb caches.

MPTCP relies on such a feature, so we need a
little refactor of the MPTCP tx path before the mentioned
revert.

The first patch avoids that the next one will cause a name
clash. The second exposes additional TCP helpers. The 3rd patch
changes the MPTCP code to do locally the whole skb allocation
and updating, so it does not rely anymore on core TCP helpers
for that nor the sk skb cache.

As a side effect, we can drop the tcp_build_frag helper.

Finally, we can pull Eric's revert.

Note that patch 3/5 will conflict with the pending -net fix
for a recently reported syzkaller splat.

Eric Dumazet (1):
  tcp: remove sk_{tr}x_skb_cache

Paolo Abeni (4):
  chtls: rename skb_entail() to chtls_skb_entail()
  tcp: expose the tcp_mark_push() and skb_entail() helpers
  mptcp: stop relying on tcp_tx_skb_cache
  Partially revert "tcp: factor out tcp_build_frag()"

 Documentation/networking/ip-sysctl.rst        |   8 -
 .../chelsio/inline_crypto/chtls/chtls.h       |   2 +-
 .../chelsio/inline_crypto/chtls/chtls_cm.c    |   2 +-
 .../chelsio/inline_crypto/chtls/chtls_io.c    |  10 +-
 include/net/sock.h                            |  19 ---
 include/net/tcp.h                             |   4 +-
 net/ipv4/af_inet.c                            |   4 -
 net/ipv4/sysctl_net_ipv4.c                    |  12 --
 net/ipv4/tcp.c                                | 147 +++++++-----------
 net/ipv4/tcp_ipv4.c                           |   6 -
 net/ipv6/tcp_ipv6.c                           |   6 -
 net/mptcp/protocol.c                          | 137 +++++++++-------
 12 files changed, 139 insertions(+), 218 deletions(-)

Comments

Eric Dumazet Sept. 17, 2021, 4:47 p.m. UTC | #1
On Fri, Sep 17, 2021 at 8:39 AM Paolo Abeni <pabeni@redhat.com> wrote:
>
> Eric noted we would be better off reverting the sk
> skb caches.
>
> MPTCP relies on such a feature, so we need a
> little refactor of the MPTCP tx path before the mentioned
> revert.
>
> The first patch avoids that the next one will cause a name
> clash. The second exposes additional TCP helpers. The 3rd patch
> changes the MPTCP code to do locally the whole skb allocation
> and updating, so it does not rely anymore on core TCP helpers
> for that nor the sk skb cache.
>
> As a side effect, we can drop the tcp_build_frag helper.
>
> Finally, we can pull Eric's revert.
>
> Note that patch 3/5 will conflict with the pending -net fix
> for a recently reported syzkaller splat.
>

Thanks for working on this Paolo !