Subject: Re: Some observations about DMA segment count
To: None <firstname.lastname@example.org>
From: Jason R Thorpe <email@example.com>
Date: 07/26/2001 12:29:07
On Fri, Jul 27, 2001 at 04:25:19AM +0900, firstname.lastname@example.org wrote:
> traditionally it was a requirement for the caller of the M_*SPACE to
> check if the mbuf is under M_READONLY situation or not. the check
> is not normally made on the input path as we are sure it is not
> shared on input path (WARNING: some packet filter directives, like
> "duplicate this traffic", break the assumption).
> the above change can choke certain code (which uses M_LEADINGSPACE when
> doing some mbuf length computation, not just for playing with m_data
> for example), IMHO.
...except M_LEADINGSPACE() was already making a (broken) read-only check...
it was checking M_EXT, and if that was set, was assuming the buffer was
read-only. M_TRAILINGSPACE() was not making this check, assuming that the
buffer was always writable.
Seems like making M_LEADINGSPACE() and M_TRAILINGSPACE() perform these checks
is better, esp. considering the "duplicate this traffic" filter example you
-- Jason R. Thorpe <email@example.com>