Re: [PATCH v4 01/21] net/tcp: Prepare tcp_md5sig_pool for TCP-AO

[Herbert Xu]

On Wed, Feb 15, 2023 at 06:33:15PM +0000, Dmitry Safonov wrote:
> TCP-AO similarly to TCP-MD5 needs to allocate tfms on a slow-path, which
> is setsockopt() and use crypto ahash requests on fast paths, which are
> RX/TX softirqs. It as well needs a temporary/scratch buffer for
> preparing the hashing request.
> 
> Extend tcp_md5sig_pool to support other hashing algorithms than MD5.
> Move it in a separate file.
> 
> This patch was previously submitted as more generic crypto_pool [1],
> but Herbert nacked making it generic crypto API. His view is that crypto
> requests should be atomically allocated on fast-paths. So, in this
> version I don't move this pool anywhere outside TCP, only extending it
> for TCP-AO use-case. It can be converted once there will be per-request
> hashing crypto keys.
> 
> [1]: https://lore.kernel.org/all/20230118214111.394416-1-dima@xxxxxxxxxx/T/#u
> Signed-off-by: Dmitry Safonov <dima@xxxxxxxxxx>
> ---
>  include/net/tcp.h        |  48 ++++--
>  net/ipv4/Kconfig         |   4 +
>  net/ipv4/Makefile        |   1 +
>  net/ipv4/tcp.c           | 103 +++---------
>  net/ipv4/tcp_ipv4.c      |  97 +++++++-----
>  net/ipv4/tcp_minisocks.c |  21 ++-
>  net/ipv4/tcp_sigpool.c   | 333 +++++++++++++++++++++++++++++++++++++++
>  net/ipv6/tcp_ipv6.c      |  58 +++----
>  8 files changed, 493 insertions(+), 172 deletions(-)
>  create mode 100644 net/ipv4/tcp_sigpool.c

Please wait for my per-request hash work before you resubmit this.
Once that's in place all you need is a single tfm for the whole
system.

As to request pools what exactly is the point of that? Just kmalloc
them on demand.

Cheers,
-- 
Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt