netfilter-devel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Pablo Neira Ayuso <pablo@netfilter.org>
To: Duncan Roe <duncan_roe@optusnet.com.au>
Cc: netfilter-devel@vger.kernel.org
Subject: Re: [PATCH libnetfilter_queue v2] src: Add faster alternatives to pktb_alloc()
Date: Wed, 19 Feb 2020 19:04:10 +0100	[thread overview]
Message-ID: <20200219180410.e56psjovne3y43rc@salvia> (raw)
In-Reply-To: <20200201062127.4729-1-duncan_roe@optusnet.com.au>

On Sat, Feb 01, 2020 at 05:21:27PM +1100, Duncan Roe wrote:
> Functions pktb_alloc_data, pktb_make and pktb_make_data are defined.
> The pktb_make pair are syggested as replacements for the pktb_alloc (now) pair
> because they are always faster.
> 
> - Add prototypes to include/libnetfilter_queue/pktbuff.h
> - Add pktb_alloc_data much as per Pablo's email of Wed, 8 Jan 2020
>   speedup: point to packet data in netlink receive buffer rather than copy to
>            area immediately following pktb struct
> - Add pktb_make much like pktb_usebuf proposed on 10 Dec 2019
>   2 sppedups: 1. Use an existing buffer rather than calloc and (later) free one.
>               2. Only zero struct and extra parts of pktb - the rest is
>                  overwritten by copy (calloc has to zero the lot).
> - Add pktb_make_data
>   3 speedups: All of the above
> - Document the new functions
> - Move pktb_alloc and pktb_alloc_data into the "other functions" group since
>   they are slower than the "make" equivalent functions
> 
> Signed-off-by: Duncan Roe <duncan_roe@optusnet.com.au>
> ---
>  include/libnetfilter_queue/pktbuff.h |   3 +
>  src/extra/pktbuff.c                  | 296 ++++++++++++++++++++++++++++++-----
>  2 files changed, 261 insertions(+), 38 deletions(-)
> 
> diff --git a/include/libnetfilter_queue/pktbuff.h b/include/libnetfilter_queue/pktbuff.h
> index 42bc153..fc6bf01 100644
> --- a/include/libnetfilter_queue/pktbuff.h
> +++ b/include/libnetfilter_queue/pktbuff.h
> @@ -4,6 +4,9 @@
>  struct pkt_buff;
>  
>  struct pkt_buff *pktb_alloc(int family, void *data, size_t len, size_t extra);
> +struct pkt_buff *pktb_alloc_data(int family, void *data, size_t len);
> +struct pkt_buff *pktb_make(int family, void *data, size_t len, size_t extra, void *buf, size_t bufsize);
> +struct pkt_buff *pktb_make_data(int family, void *data, size_t len, void *buf, size_t bufsize);

Hm, when I delivered the patch to you, I forgot that you main point
was that you wanted to skip the memory allocation.

I wonder if all these new functions can be consolidated into one
single function, something like:

        struct pkt_buff *pktb_alloc2(int family, void *head, size_t head_size, void *data, size_t len, size_t extra);

The idea is that:

* head is the memory area that is large enough for the struct pkt_buff
  (metadata). You can add a new pktb_head_size() function that returns
  the size of opaque struct pkt_buff structure (whose layout is not
  exposed to the user). I think you can this void *buf in your pktb_make
  function.

* data is the memory area to store the network packet itself.

Then, you can allocate head and data in the stack and skip
malloc/calloc.

Would this work for you? I would prefer if there is just one single
advanced function to do this.

Thanks for your patience.

  parent reply	other threads:[~2020-02-19 18:04 UTC|newest]

Thread overview: 19+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-12-10 11:26 [PATCH libnetfilter_queue 0/1] New pktb_usebuf() function Duncan Roe
2019-12-10 11:26 ` [PATCH libnetfilter_queue 1/1] src: Add alternative function to pktb_alloc to avoid malloc / free overhead Duncan Roe
2019-12-22  2:09 ` [PATCH libnetfilter_queue 0/1] New pktb_usebuf() function Duncan Roe
2020-01-03  2:47 ` Duncan Roe
2020-01-06  3:17 ` [PATCH libnetfilter_queue v2 0/1] New pktb_make() function Duncan Roe
2020-01-08 22:53   ` Pablo Neira Ayuso
2020-01-10  2:27     ` Duncan Roe
2020-01-13 18:51       ` Pablo Neira Ayuso
2020-01-27  2:11         ` Duncan Roe
2020-01-27  1:44     ` Duncan Roe
2020-02-01  6:21     ` [PATCH libnetfilter_queue v2] src: Add faster alternatives to pktb_alloc() Duncan Roe
2020-02-07 22:39       ` Duncan Roe
2020-02-19 18:04       ` Pablo Neira Ayuso [this message]
2020-02-20 23:22         ` Duncan Roe
2020-02-20 23:50           ` Duncan Roe
2020-04-06  6:17         ` Duncan Roe
2020-04-11  7:24         ` [PATCH libnetfilter_queue 0/1] src & doc: pktb_alloc2 Duncan Roe
2020-04-11  7:24         ` [PATCH libnetfilter_queue 1/1] New faster pktb_alloc2 replaces pktb_alloc & pktb_free Duncan Roe
2020-01-06  3:17 ` [PATCH libnetfilter_queue v2 1/1] src: Add alternative function to pktb_alloc to avoid malloc / free overhead Duncan Roe

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200219180410.e56psjovne3y43rc@salvia \
    --to=pablo@netfilter.org \
    --cc=duncan_roe@optusnet.com.au \
    --cc=netfilter-devel@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).