On Wed, Oct 30, 2024 at 10:38:11PM +0100, Pablo Neira Ayuso wrote:
> On Wed, Oct 30, 2024 at 07:19:47PM +0100, Phil Sutter wrote:
> > On Tue, Oct 29, 2024 at 09:12:19PM +0100, Florian Westphal wrote:
> > [...]
> > > diff --git a/tests/monitor/testcases/set-simple.t b/tests/monitor/testcases/set-simple.t
> > > index 8ca4f32463fd..6853a0ebbb0c 100644
> > > --- a/tests/monitor/testcases/set-simple.t
> > > +++ b/tests/monitor/testcases/set-simple.t
> > > @@ -37,9 +37,10 @@ J {"add": {"element": {"family": "ip", "table": "t", "name": "portrange", "elem"
> > > # make sure half open before other element works
> > > I add element ip t portrange { 1024-65535 }
> > > I add element ip t portrange { 100-200 }
> > > -O -
> > > -J {"add": {"element": {"family": "ip", "table": "t", "name": "portrange", "elem": {"set": [{"range": [1024, 65535]}]}}}}
> > > +O add element ip t portrange { 100-200 }
> > > +O add element ip t portrange { 1024-65535 }
> > > J {"add": {"element": {"family": "ip", "table": "t", "name": "portrange", "elem": {"set": [{"range": [100, 200]}]}}}}
> > > +J {"add": {"element": {"family": "ip", "table": "t", "name": "portrange", "elem": {"set": [{"range": [1024, 65535]}]}}}}
> >
> > This is odd: Why does monitor output reverse input? If nft reorders
> > input, the test ("make sure half open before other element works") is
> > probably moot anyway.
>
> Because elements are collapsed in one single command.
>
> I add element ip t portrange { 1024-65535 }
> I add element ip t portrange { 100-200 }
>
> and there is qsort() to search for interval overlaps.
>
> this becomes one single command with { 100-200, 1024-65535 }
>
> I have a patch to remove the overlap check using qsort() from
> userspace which should remove this reordering, but it seems there is a
> overlap case that kernel does not handle yet.
I can also revert the patch if you don't like, but it is saving _a
lot_ of memory from userspace for the silly one element per line case.
