_PACKET_IN_DATA_OFFSET calculation conflicts with ASSERT for OF1.3 only

[samwhitlock]

I'm working with a recent version of Indigo and the loxigen-artifacts submodule (merged the changes from today).

how to trigger

Here are the important steps in my code to trigger this bug:

1. Create an OpenFlow 1.3 packet in object

of_packet_in_t *packet_in = (of_packet_in_t*) of_packet_in_new(OF_VERSION_1_3);

\2. Set the data in it to send to the controller (i.e., don't buffer the packet locally)

of_packet_in_data_set(packet_in, &octets);

\3. the _PACKET_IN_DATA_OFFSET macro sets the offset to 26, and the call to _END_LEN returns 8. Packet length is 34 when I observe this behavior.

\4. Because cur_len != 0 and cur_len != value->bytes in of_wire_buffer_octets_data_set, the assertion fails.

---

I don't observe this behavior when I use OpenFlow 1.2 packets.

[rlane]

Here's a minimal failing testcase:

    uint8_t data[] = { 1, 2, 3, 4 };
    of_octets_t octets = { .bytes = 4, .data = data };
    of_packet_in_t *packet_in = (of_packet_in_t*) of_packet_in_new(OF_VERSION_1_3);
    of_packet_in_data_set(packet_in, &octets);

This works:

    of_match_t match;
    uint8_t data[] = { 1, 2, 3, 4 };
    memset(&match, 0, sizeof(match));
    match.version = OF_VERSION_1_3;
    of_octets_t octets = { .bytes = 4, .data = data };
    of_packet_in_t *packet_in = (of_packet_in_t*) of_packet_in_new(OF_VERSION_1_3);
    of_packet_in_match_set(packet_in, &match);
    of_packet_in_data_set(packet_in, &octets);

I don't know what's causing this bug yet, but setting an empty match is an easy workaround. We've never seen this because we always set the match in packet-ins.

[bjlee72]

Is this issue resolved?

[capveg]

This looks like a resolved issue so unless I hear otherwise, I'm going to close it.

[andi-bigswitch]

@rlane is this related to cd6c1d6?

[rlane]

I don't think so, given that the padding also exists in OF 1.2.

[rlane]

I found the issue. There's a hack in gen_new_fn_body that manually initializes the length field (but not the type) of embedded matches. It uses the offset from OF 1.2, but OF 1.3 added a cookie field, changing the offset of the match. So for OF 1.3, the initial length is written into the cookie field instead of the match length field.

I'll fix this after the OF 1.4 changes are merged.

[andi-bigswitch]

@rlane - did we end up fixing this?