On 9/16/24 17:59, Adrián Moreno wrote: > On Mon, Sep 02, 2024 at 10:14:34AM GMT, Eelco Chaudron via discuss wrote: >> Hi, >> >> I’m not an expert in this area, but I can see the metadata is explicitly >> cleared when traffic goes over a patch port. >> >> https://github.com/openvswitch/ovs/blob/0051785f00c52cbbc2897989a454a5512e82dce0/ofproto/ofproto-dpif-xlate.c#L4131 >> >> If I go over the Git history, I can see this has always been the case since >> the introduction in commit 0a740f48293e. >> >> It also explicitly mentions this: >> >> /* If 'struct flow' gets additional metadata, we'll need to zero it out >> * before traversing a patch port. */ >> >> So I assume the documentation might be off, or the commit was wrong (and >> it’s a bug), or there is another reason. Maybe Ilya or others know, as this >> was done 12 years ago :) > > My 2 cents: > > The documentation says regarding REGISTER fields: > " > These fields give an OpenFlow switch space for temporary storage > while the pipeline is running. Whereas metadata fields can have a > meaningful initial value and can persist across some hops across > OpenFlow switches, registers are always initially 0 and their > values never persist across inter-switch hops (not even across > patch ports). > " > > I think the documentation refers to METADATA Fields, not the REGISTER > field that is (very confusingly) called "metadata". IIUC, it refers to > the following fields: > in_port > in_port_oxm > skb_priority > pkt_mark > actset_output > packet_type
Yeah, only above fields may be preserved and the OXM_OF_METADATA is a register and not a metadata field. Unfortunate naming. > > From these, the only field that is treated specially when a packet > traverses a patch port to another bridge is "in_port" that is set to > such patch port. > > Thanks. > Adrián > >> >> Cheers, >> >> Eelco >> >> >> On 30 Aug 2024, at 4:27, 征途 via discuss wrote: >> >>> hello，In the ovs fields documentation, it says that metadata can retain >>> values across the ovs bridge >>> >>> >>> https://man7.org/linux/man-pages/man7/ovs-fields.7.html#METADATA_FIELDS >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> But in my tests, it doesn't seem to retain the value, can you help us to >>> answer the design in the first place? >>> Does metadata support retention of values across the ovs bridge or >>> not？thanks. >> >>> _______________________________________________ >>> discuss mailing list >>> disc...@openvswitch.org >>> https://mail.openvswitch.org/mailman/listinfo/ovs-discuss > >> _______________________________________________ >> discuss mailing list >> disc...@openvswitch.org >> https://mail.openvswitch.org/mailman/listinfo/ovs-discuss > _______________________________________________ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
