From owner-freebsd-net  Tue Oct 23 11:32: 2 2001
Delivered-To: freebsd-net@freebsd.org
Received: from elvis.mu.org (elvis.mu.org [216.33.66.196])
	by hub.freebsd.org (Postfix) with ESMTP id 81BA637B403
	for <net@freebsd.org>; Tue, 23 Oct 2001 11:31:59 -0700 (PDT)
Received: by elvis.mu.org (Postfix, from userid 1192)
	id 7167A81D01; Tue, 23 Oct 2001 13:31:59 -0500 (CDT)
Date: Tue, 23 Oct 2001 13:31:59 -0500
From: Alfred Perlstein <bright@mu.org>
To: Luigi Rizzo <rizzo@aciri.org>
Cc: net@freebsd.org
Subject: Re: performance issues with M_PREPEND on clusters
Message-ID: <20011023133159.J15052@elvis.mu.org>
References: <20011023110307.A34494@iguana.aciri.org> <20011023132813.I15052@elvis.mu.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.2.5i
In-Reply-To: <20011023132813.I15052@elvis.mu.org>; from bright@mu.org on Tue, Oct 23, 2001 at 01:28:13PM -0500
Sender: owner-freebsd-net@FreeBSD.ORG
Precedence: bulk
List-ID: <freebsd-net.FreeBSD.ORG>
List-Archive: <http://docs.freebsd.org/mail/> (Web Archive)
List-Help: <mailto:majordomo@FreeBSD.ORG?subject=help> (List Instructions)
List-Subscribe: <mailto:majordomo@FreeBSD.ORG?subject=subscribe%20freebsd-net>
List-Unsubscribe: <mailto:majordomo@FreeBSD.ORG?subject=unsubscribe%20freebsd-net>
X-Loop: FreeBSD.org

* Alfred Perlstein <bright@mu.org> [011023 13:28] wrote:
> * Luigi Rizzo <rizzo@aciri.org> [011023 13:06] wrote:
> > 
> > Hi,
> > don't know what is the right forum to discuss this, but this is
> > certainly one. I have also Bcc-ed some people who might need to be
> > involved.
> > 
> > In the mbuf code, M_LEADINGSPACE always returns 0 when M_EXT is
> > set, instead of calling the second part of M_WRITABLE to check
> > whether there is a chance of writing into the cluster. This means
> > that both M_PREPEND() and m_pullup() will have to allocate a second
> > mbuf when they are invoked on a cluster, which these days is
> > basically true for most of the incoming traffic (several network
> > drivers pass up the full packet in a cluster).
> 
> I've seen this brought up before, this is not what you want to
> do otherwise you risk corrupting EXT data.  The right thing to
> do is have the people allocating writable EXT bufs mark them
> as such.  Other EXT type buffers such as sendfile bufs can not
> be written to no matter what the sharecount is.
> 
> I hope this helps.

Oh yeah, using this method you could "fix" MCLGET in -stable to
from this:

#define MCLGET(m, how) do {                                             \
        struct mbuf *_mm = (m);                                         \
                                                                        \
        MCLALLOC(_mm->m_ext.ext_buf, (how));                            \
        if (_mm->m_ext.ext_buf != NULL) {                               \
                _mm->m_data = _mm->m_ext.ext_buf;                       \
                _mm->m_flags |= M_EXT;                                  \
                _mm->m_ext.ext_free = NULL;                             \
                _mm->m_ext.ext_ref = NULL;                              \
                _mm->m_ext.ext_size = MCLBYTES;                         \
        }                                                               \
} while (0)

to:

#define MCLGET(m, how) do {                                             \
        struct mbuf *_mm = (m);                                         \
                                                                        \
        MCLALLOC(_mm->m_ext.ext_buf, (how));                            \
        if (_mm->m_ext.ext_buf != NULL) {                               \
                _mm->m_data = _mm->m_ext.ext_buf;                       \
                _mm->m_flags |= M_EXT | M_EXTWRITEABLE;                 \
                _mm->m_ext.ext_free = NULL;                             \
                _mm->m_ext.ext_ref = NULL;                              \
                _mm->m_ext.ext_size = MCLBYTES;                         \
        }                                                               \
} while (0)

-- 
-Alfred Perlstein [alfred@freebsd.org]
'Instead of asking why a piece of software is using "1970s technology,"
 start asking why software is ignoring 30 years of accumulated wisdom.'
                           http://www.morons.org/rants/gpl-harmful.php3

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-net" in the body of the message