[PATCH net-next 00/11] tls: rx: avoid skb_cow_data()

[Jakub Kicinski <kuba@kernel.org>]

TLS calls skb_cow_data() on the skb it received from strparser
whenever it needs to hold onto the skb with the decrypted data.
(The alternative being decrypting directly to a user space buffer
in whic case the input skb doesn't get modified or used after.)
TLS needs the decrypted skb:
 - almost always with TLS 1.3 (unless the new NoPad is enabled);
 - when user space buffer is too small to fit the record;
 - when BPF sockmap is enabled.

Most of the time the skb we get out of strparser is a clone of
a 64kB data unit coalsced by GRO. To make things worse skb_cow_data()
tries to output a linear skb and allocates it with GFP_ATOMIC.
This occasionally fails even under moderate memory pressure.

This patch set rejigs the TLS Rx so that we don't expect decryption
in place. The decryption handlers return an skb which may or may not
be the skb from strparser. For TLS 1.3 this results in a 20-30%
performance improvement without NoPad enabled.

Jakub Kicinski (11):
  tls: rx: allow only one reader at a time
  tls: rx: don't try to keep the skbs always on the list
  tls: rx: don't keep decrypted skbs on ctx->recv_pkt
  tls: rx: remove the message decrypted tracking
  tls: rx: factor out device darg update
  tls: rx: read the input skb from ctx->recv_pkt
  tls: rx: return the decrypted skb via darg
  tls: rx: async: adjust record geometry immediately
  tls: rx: async: hold onto the input skb
  tls: rx: async: don't put async zc on the list
  tls: rx: decrypt into a fresh skb

 include/net/strparser.h |   1 -
 include/net/tls.h       |   4 +
 net/tls/Makefile        |   2 +-
 net/tls/tls.h           |  20 +-
 net/tls/tls_device.c    |  25 ++-
 net/tls/tls_strp.c      |  17 ++
 net/tls/tls_sw.c        | 458 ++++++++++++++++++++++++----------------
 7 files changed, 333 insertions(+), 194 deletions(-)
 create mode 100644 net/tls/tls_strp.c

-- 
2.36.1