On Thu, Aug 11, 2016 at 03:17:51PM +0200, Florian Westphal wrote:
> Since commit d188ba86dd07a72eb ("xfrm: add rcu protection to sk->sk_policy[]")
> sk_policy can rely on rcu protection.
>
> This change allows to also use rcu in xfrm_policy_lookup_bytype and to
> avoid grabbing the read-sie policy lock during lookups.
>
> read-side policy rwlock is converted to pure rcu, then the policy_lock is
> changed to a plain spinlock as only write-sides remain.
>
> First few patches do some preparation work, the later ones remove
> the read-side locks, last patch converts rwlock to spinlock.
>
> I tested this with rcu debug enabled and simple esp tunnel
> forwarding udp packets.
>
> If you have better tests for this please let me know and I can re-run with
> that.
>
> Florian Westphal (8):
> xfrm: policy: use rcu versions for iteration and list add/del
> xfrm: policy: prepare policy_bydst hash for rcu lookups
> xfrm: policy: add sequence count to synchronize reads with hash resizes
> xfrm: policy: use atomic_inc_not_zero in rcu section
> xfrm: policy: make xfrm_policy_lookup_bytype lockless
> xfrm: policy: only use rcu in xfrm_sk_policy_lookup
> xfrm: policy: don't acquire policy lock in xfrm_spd_getinfo
> xfrm: policy: convert policy_lock to spinlock
>
> include/net/netns/xfrm.h | 4 -
> net/xfrm/xfrm_policy.c | 145
> +++++++++++++++++++++++++++--------------------
> 2 files changed, 88 insertions(+), 61 deletions(-)
Looks good, I've applied to already to the ipsec-next testing branch.
However, I'll be off for a week starting this afternoon, so I don't
see if the patchset passed all tests before I leave. For that I defer
the applying to master until I'm back.
Thanks!
