On 16/12/18 - 22:31:41, Eric Dumazet wrote:
>
>
> On 12/14/2018 02:40 PM, Christoph Paasch wrote:
> > Instead of having a single TFO-context, we now have a list of
> > tcp_fastopen_context, bounded by TCP_FASTOPEN_CTXT_LEN (set to 2).
> >
> > This enables us to do a rolling TFO-key update that allows the server to
> > accept old cookies and at the same time announce new ones to the client
> > (see follow-up patch).
> >
> > Signed-off-by: Christoph Paasch <cpaa...@apple.com>
> > ---
> > include/net/tcp.h | 2 ++
> > net/ipv4/tcp_fastopen.c | 52
> > +++++++++++++++++++++++++++++++++++++++++++++----
> > 2 files changed, 50 insertions(+), 4 deletions(-)
> >
> > diff --git a/include/net/tcp.h b/include/net/tcp.h
> > index e0a65c067662..e629ea2e6c9d 100644
> > --- a/include/net/tcp.h
> > +++ b/include/net/tcp.h
> > @@ -1622,9 +1622,11 @@ bool tcp_fastopen_cookie_check(struct sock *sk, u16
> > *mss,
> > struct tcp_fastopen_cookie *cookie);
> > bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
> > #define TCP_FASTOPEN_KEY_LENGTH 16
> > +#define TCP_FASTOPEN_CTXT_LEN 2
> >
> > /* Fastopen key context */
> > struct tcp_fastopen_context {
> > + struct tcp_fastopen_context __rcu *next;
> > struct crypto_cipher *tfm;
> > __u8 key[TCP_FASTOPEN_KEY_LENGTH];
> > struct rcu_head rcu;
> > diff --git a/net/ipv4/tcp_fastopen.c b/net/ipv4/tcp_fastopen.c
> > index 018a48477355..c52d5b8eabf0 100644
> > --- a/net/ipv4/tcp_fastopen.c
> > +++ b/net/ipv4/tcp_fastopen.c
> > @@ -37,8 +37,14 @@ static void tcp_fastopen_ctx_free(struct rcu_head *head)
> > {
> > struct tcp_fastopen_context *ctx =
> > container_of(head, struct tcp_fastopen_context, rcu);
> > - crypto_free_cipher(ctx->tfm);
> > - kfree(ctx);
> > +
> > + while (ctx) {
> > + struct tcp_fastopen_context *prev = ctx;
> > + /* We own ctx, thus no need to hold the Fastopen-lock */
> > + ctx = rcu_dereference_protected(ctx->next, 1);
> > + crypto_free_cipher(prev->tfm);
> > + kfree(prev);
> > + }
> >
>
> It seems this function does not need to be changed, since at most one context
> should be freed per run ?
It gets called from tcp_fastopen_destroy_cipher() though (to destroy the
socket's TFO-keys when the socket gets closed). There it has to destroy the
whole list.
Same when going through exit_batch for the namespace.
We could of course split it in tcp_fastopen_ctx_free_one() and
tcp_fastopen_ctx_free_all(). But maybe that's overkill as it's a rare thing
to do?
Christoph
