Thanks for the review answer inlined. Thanks, Alin.
> -----Mesaj original----- > De la: Nithin Raju [mailto:nit...@vmware.com] > Trimis: Wednesday, June 8, 2016 10:10 PM > Către: Alin Serdean <aserd...@cloudbasesolutions.com>; > dev@openvswitch.org > Subiect: Re: [ovs-dev] [PATCH v3] datapath-windows: Add GRE checksum > > > >+ /* Get a contignuous buffer for the maxmimum length of a GRE > >+ header > >*/ > >+ bufferStart = NdisGetDataBuffer(curNb, OVS_MAX_GRE_LGTH, NULL, 1, > >+ 0); > > Sorry we have to go back and forth on this. Like I mentioned in the previous > email, the idea is to get a contiguous chunk of memory so we can walk all the > header until the GRE header. It is a good idea to use > NdisGetDataBuffer() instead of copying the NBL. But, we won¹t avoid the > copied NBL anyway since decap has to happen on the copied NBL. [Alin Gabriel Serdean: ] Agreed. I only wanted a check to make sure the whole ETH, IP, GRE was contiguous. > > In any case, NdisGetDataBuffer() has some pitfalls: > "If the requested data in the buffer is contiguous, this return value is a > pointer to a location that NDIS provides. If the data is not contiguous, NDIS > uses the Storage parameter as follows: > > * If the Storage parameter is non-NULL, NDIS copies the data to the buffer at > Storage. This return value is the pointer passed to the Storage parameter. > * If the Storage parameter is NULL, this return value is NULL.² > > So, if the first MDL does not fit the headers until the GRE headers, we need > to pass explicit memory to NdisGetDataBuffer() in argument #3 in order for > NDIS to copy it over to a contiguous chunk of memory. [Alin Gabriel Serdean: ] We have to take into consideration that NDIS 6.3 supports NVGRE and in NDIS 6.4 GRE (https://support.microsoft.com/en-us/kb/3022776 ). Packets can be split in multiple Mdl's, if the miniport supports it, as early as at the beginning of the upper-layer-protocol (https://msdn.microsoft.com/en-us/library/windows/hardware/ff571065(v=vs.85).aspx) and if the split header is not greater than the maximum header size which is configurable. In the case of VXLAN / STT we can be in the case above, because of the UDP/TCP like packet, but for GRE things differ since there are cards which support offloading: https://msdn.microsoft.com/en-us/library/windows/hardware/dn605832(v=vs.85).aspx "The protocol driver will guarantee that all the prepended encapsulation headers (ETH, IP, GRE) will be physically contiguous and will be in the first MDL of the packet." I do not think we will be in a situation in which a GRE header will not be continuous. If you consider we should still have a fragmented header problem, I will allocate a buffer and execute another NdisGetDataBuffer call if the first one failed. > > Thanks, > -- Nithin _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
