> In linux.dev.kernel, article <XFMail.961101042606.rriggs@tesser.com>,
> Rob Riggs <rriggs@tesser.com> writes:
> >
> > - The point Alan brought up does not affect Joe Average User,
> > but rather very busy network boxes. And it will only be a
> > problem on very busy network boxes that are tight on memory
> > _and_ needs to access the /proc/net files.
> >
> Umm, I'd hesitate to call six MBytes of free RAM just to read
> /proc/net/routes (or whatever) "tight on memory". I'd call it "completely
> thrashes memory on the system".
and the original idea was to use big kernel-space buffers to cache stuff
... kernel buffers are unswappable and we open up a new set of denial of
service attacks here?
unless someone has bright new ideas about /proc in general, the 1 page
limitation looks like a reasonable protection against overallocation?
-- mingo