[dpdk-users] rte_flow() usage of htonl() for ipv4 addr masks?

[Stephen Hemminger]

On Sun, 9 Aug 2020 22:14:28 -0500
Arvind Narayanan <webguru2688 at gmail.com> wrote:

> On Sun, Aug 9, 2020 at 9:56 PM Cliff Burdick <shaklee3 at gmail.com> wrote:
> >
> > It should convert to network order, although many applications it won't matter since they use all F's. If you follow the code in flow_filtering, indeed it's using:
> >
> > #define FULL_MASK 0xffffffff /* full mask */
> >
> > So it won't make any difference. The example should probably be updated, though..  
> 
> Thanks Cliff! Yes, when it's all Fs, it doesn't matter.
> But I am trying to install rte_flow rules for subnets by parsing a
> file which has IPv4 ranges mentioned using CIDR format.
> 
> I have it working for say /24 ranges, but as I go to /30 or /29, the
> same implementation is not working. I followed the flow classify
> example. https://doc.dpdk.org/guides/sample_app_ug/flow_classify.html
> as it does the same thing.
> 
> ```
> static uint32_t
> convert_depth_to_bitmask(uint32_t depth_val) {
>   uint32_t bitmask = 0;
>   int i, j;
> 
>   for (i = depth_val, j = 0; i > 0; i--, j++)
>     bitmask |= (1 << (31 - j));
>   return bitmask;
> }
> 
> ip_mask.hdr.dst_addr = htonl(convert_depth_to_bitmask(29))
> ```
> and https://github.com/DPDK/dpdk/blob/master/examples/flow_classify/flow_classify.c#L377-L396
> for ipv4 parsing.
> 
> I'll keep digging. As always, it seems too trivial to fix as a bug,
> but it's been driving me crazy.. haha
> 
> - Arvind
> 
> 
> >
> > On Sun, Aug 9, 2020 at 7:03 PM Arvind Narayanan <webguru2688 at gmail.com> wrote:  
> >>
> >> Hi,
> >>
> >> In the flow_filtering sample application, the IP's mask was set without
> >> using htonl().
> >> https://github.com/DPDK/dpdk/blob/master/examples/flow_filtering/flow_blocks.c#L85
> >>
> >> Another DPDK page <https://doc.dpdk.org/guides/howto/rte_flow.html> shows
> >> how a testpmd command is translated to C code.
> >> On this page though, Example 4.2 (Range IPv4 drop) has used htonl() to set
> >> the mask.
> >>
> >> Any clarification on how to load the mask would be helpful.

In DPDK please use rte_cpu_to_be_32() instead of htonl().