[ipsec-next,v2,0/5] xfrm: remove xfrm replay indirections

Message

Florian Westphal June 18, 2021, 1:51 p.m. UTC

This is v2 of an older patchset that got stuck in backlog hell. Changes:

 - drop bogus "get rid of duplicated notification code" patch. As noted
   by Sabrina it does change behavior.
 - fix a compiler warning in patch 2.

ipsec.c selftest passes.

The xfrm replay logic is implemented via indirect calls.

xfrm_state struct holds a pointer to a
'struct xfrm_replay', which is one of several replay protection
backends.

XFRM then invokes the backend via state->repl->callback().
Due to retpoline all indirect calls have become a lot more
expensive.  Fortunately, there are no 'replay modules', all are available
for direct calls.

This series removes the 'struct xfrm_replay' and adds replay
functions that can be called instead of the redirection.

Example:
  -  err = x->repl->overflow(x, skb);
  +  err = xfrm_replay_overflow(x, skb);

Instead of a pointer to a struct with function pointers, xfrm_state
now holds an enum that tells the replay core what kind of replay
test is to be done.

Florian Westphal (5):
  xfrm: replay: avoid xfrm replay notify indirection
  xfrm: replay: remove advance indirection
  xfrm: replay: remove recheck indirection
  xfrm: replay: avoid replay indirection
  xfrm: replay: remove last replay indirection

 include/net/xfrm.h     |  29 ++++---
 net/xfrm/xfrm_input.c  |   6 +-
 net/xfrm/xfrm_output.c |   2 +-
 net/xfrm/xfrm_replay.c | 171 +++++++++++++++++++++++++----------------
 net/xfrm/xfrm_state.c  |   2 +-
 5 files changed, 123 insertions(+), 87 deletions(-)

Comments

Steffen Klassert June 22, 2021, 7:07 a.m. UTC | #1

On Fri, Jun 18, 2021 at 03:51:55PM +0200, Florian Westphal wrote:
> This is v2 of an older patchset that got stuck in backlog hell. Changes:
> 
>  - drop bogus "get rid of duplicated notification code" patch. As noted
>    by Sabrina it does change behavior.
>  - fix a compiler warning in patch 2.
> 
> ipsec.c selftest passes.
> 
> The xfrm replay logic is implemented via indirect calls.
> 
> xfrm_state struct holds a pointer to a
> 'struct xfrm_replay', which is one of several replay protection
> backends.
> 
> XFRM then invokes the backend via state->repl->callback().
> Due to retpoline all indirect calls have become a lot more
> expensive.  Fortunately, there are no 'replay modules', all are available
> for direct calls.
> 
> This series removes the 'struct xfrm_replay' and adds replay
> functions that can be called instead of the redirection.
> 
> Example:
>   -  err = x->repl->overflow(x, skb);
>   +  err = xfrm_replay_overflow(x, skb);
> 
> Instead of a pointer to a struct with function pointers, xfrm_state
> now holds an enum that tells the replay core what kind of replay
> test is to be done.
> 
> Florian Westphal (5):
>   xfrm: replay: avoid xfrm replay notify indirection
>   xfrm: replay: remove advance indirection
>   xfrm: replay: remove recheck indirection
>   xfrm: replay: avoid replay indirection
>   xfrm: replay: remove last replay indirection

All applied, thanks a lot Florian!