Hi Daniel,

On 6/1/07, Daniel Hazelton <dhazelton@enter.net> wrote:
quoted text
> On Wednesday 30 May 2007 19:02:28 Mark Adler wrote:
> > On May 30, 2007, at 6:30 AM, Satyam Sharma wrote:
> > > [1] For your reference, here is the user code in question:
> >
> > ...
> >
> > >        if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
> > >         ((data_in[0] & 0x0f) == Z_DEFLATED) &&
> > >         !(((data_in[0]<<8) + data_in[1]) % 31)) {
> >
> > The funny thing here is that the author felt compelled to use a
> > #defined constant for the dictionary bit (PRESET_DICT), but had no
> > problem with a numeric constant to isolate the compression method
> > (0x0f), or for that matter extracting the window bits from the
> > header.  The easy way to avoid the use of an internal zlib header
> > file here is to simply replace PRESET_DICT with 0x20.  That constant
> > will never change -- it is part of the definition of the zlib header
> > in RFC 1950.
>
> If there is no objection, I'll put together a patch that changes the use in
> JFFS2 into a "magic number", complete with documentation on it, and also
> moves all of the zlib stuff into a single directory.

Right, s/PRESET_DICT/0x20/ would have the least fall-out / unknown
side-effects.

quoted text
> > The slightly more involved patch to avoid the problem is to let
> > inflate() do all that work for you, including the integrity check on
> > the zlib header (% 31).  Also this corrects an error in the original
> > code, which is that it continues to try to decompress after finding
> > that a dictionary is needed or that the zlib header is invalid.  In
> > this version, a bad header simply returns an error:
> >
>
> Does anyone know if doing as Mark suggests would negatively impact the
> performance of JFFS2 to such a degree that it could be considered a
> regression? I, unfortunately, don't have the hardware to properly test the
> code. And, at this point in time, I also don't have enough familiarity with
> the JFFS2 code to make such a change myself.

David would have to comment on that, but you could simultaneously
make and submit the patch as suggested by Mark with both your
signed-offs-by. That'll naturally need to go through David's tree, so we'll
know if he likes/accepts the suggested modifications ...
-
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/