On Thu, Oct 12, 2023 at 10:14:31PM +0200, Pablo Neira Ayuso wrote: > On Thu, Oct 12, 2023 at 09:47:38PM +0200, Pablo Neira Ayuso wrote: > > On Thu, Oct 12, 2023 at 06:08:42PM +0200, Phil Sutter wrote: > > > String match indeed returns a match as long as the given pattern starts > > > in the range of --from and --to, update the text accordingly. > > > Also add a note regarding fragment boundaries. > > > > > > Closes: https://bugzilla.netfilter.org/show_bug.cgi?id=1707 > > > Signed-off-by: Phil Sutter <phil@xxxxxx> > > > --- > > > extensions/libxt_string.man | 8 ++++++-- > > > 1 file changed, 6 insertions(+), 2 deletions(-) > > > > > > diff --git a/extensions/libxt_string.man b/extensions/libxt_string.man > > > index 2a470ece19c9d..efdda492ae78d 100644 > > > --- a/extensions/libxt_string.man > > > +++ b/extensions/libxt_string.man > > > @@ -7,9 +7,13 @@ Select the pattern matching strategy. (bm = Boyer-Moore, kmp = Knuth-Pratt-Morri > > > Set the offset from which it starts looking for any matching. If not passed, default is 0. > > > .TP > > > \fB\-\-to\fP \fIoffset\fP > > > -Set the offset up to which should be scanned. That is, byte \fIoffset\fP-1 > > > -(counting from 0) is the last one that is scanned. > > > +Set the offset up to which should be scanned. If the pattern does not start > > > +within this offset, it is not considered a match. > > > If not passed, default is the packet size. > > > +A second function of this parameter is instructing the kernel how much data > > > +from the packet should be provided. With non-linear skbuffs (e.g. due to > > > +fragmentation), a pattern extending past this offset may not be found. Also see > > > +the related note below about Boyer-Moore algorithm in these cases. > > > > Then, matching on: > > > > - linear skbuff: if the pattern falls within from-to, all good. > > - non-linear skbuff: if pattern falls within from-to, but remaining > > patter > > > > This is clearly broken, the fix is just to document this? > > > > No attempt to fix it this from the kernel? > > I would align linear skbuff behaviour to how non-linear skbuff works, > that is, pattern matching stops at to. This requires a small patch for > xt_string. Sounds reasonable. My efforts at documenting the current behaviour probably show how unintuitive this currently is. Thanks for the suggestion! Cheers, Phil