xref: /linux/Documentation/networking/xsk-tx-metadata.rst (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
1.. SPDX-License-Identifier: GPL-2.0
2
3==================
4AF_XDP TX Metadata
5==================
6
7This document describes how to enable offloads when transmitting packets
8via :doc:`af_xdp`. Refer to :doc:`xdp-rx-metadata` on how to access similar
9metadata on the receive side.
10
11General Design
12==============
13
14The headroom for the metadata is reserved via ``tx_metadata_len`` in
15``struct xdp_umem_reg``. The metadata length is therefore the same for
16every socket that shares the same umem. The metadata layout is a fixed UAPI,
17refer to ``union xsk_tx_metadata`` in ``include/uapi/linux/if_xdp.h``.
18Thus, generally, the ``tx_metadata_len`` field above should contain
19``sizeof(union xsk_tx_metadata)``.
20
21The headroom and the metadata itself should be located right before
22``xdp_desc->addr`` in the umem frame. Within a frame, the metadata
23layout is as follows::
24
25           tx_metadata_len
26     /                         \
27    +-----------------+---------+----------------------------+
28    | xsk_tx_metadata | padding |          payload           |
29    +-----------------+---------+----------------------------+
30                                ^
31                                |
32                          xdp_desc->addr
33
34An AF_XDP application can request headrooms larger than ``sizeof(struct
35xsk_tx_metadata)``. The kernel will ignore the padding (and will still
36use ``xdp_desc->addr - tx_metadata_len`` to locate
37the ``xsk_tx_metadata``). For the frames that shouldn't carry
38any metadata (i.e., the ones that don't have ``XDP_TX_METADATA`` option),
39the metadata area is ignored by the kernel as well.
40
41The flags field enables the particular offload:
42
43- ``XDP_TXMD_FLAGS_TIMESTAMP``: requests the device to put transmission
44  timestamp into ``tx_timestamp`` field of ``union xsk_tx_metadata``.
45- ``XDP_TXMD_FLAGS_CHECKSUM``: requests the device to calculate L4
46  checksum. ``csum_start`` specifies byte offset of where the checksumming
47  should start and ``csum_offset`` specifies byte offset where the
48  device should store the computed checksum.
49
50Besides the flags above, in order to trigger the offloads, the first
51packet's ``struct xdp_desc`` descriptor should set ``XDP_TX_METADATA``
52bit in the ``options`` field. Also note that in a multi-buffer packet
53only the first chunk should carry the metadata.
54
55Software TX Checksum
56====================
57
58For development and testing purposes its possible to pass
59``XDP_UMEM_TX_SW_CSUM`` flag to ``XDP_UMEM_REG`` UMEM registration call.
60In this case, when running in ``XDK_COPY`` mode, the TX checksum
61is calculated on the CPU. Do not enable this option in production because
62it will negatively affect performance.
63
64Querying Device Capabilities
65============================
66
67Every devices exports its offloads capabilities via netlink netdev family.
68Refer to ``xsk-flags`` features bitmask in
69``Documentation/netlink/specs/netdev.yaml``.
70
71- ``tx-timestamp``: device supports ``XDP_TXMD_FLAGS_TIMESTAMP``
72- ``tx-checksum``: device supports ``XDP_TXMD_FLAGS_CHECKSUM``
73
74See ``tools/net/ynl/samples/netdev.c`` on how to query this information.
75
76Example
77=======
78
79See ``tools/testing/selftests/bpf/xdp_hw_metadata.c`` for an example
80program that handles TX metadata. Also see https://github.com/fomichev/xskgen
81for a more bare-bones example.
82