1 Segmentation Offloads in the Linux Networking Stack
6 This document describes a set of techniques in the Linux networking stack
7 to take advantage of segmentation offload capabilities of various NICs.
9 The following technologies are described:
10 * TCP Segmentation Offload - TSO
11 * UDP Fragmentation Offload - UFO
12 * IPIP, SIT, GRE, and UDP Tunnel Offloads
13 * Generic Segmentation Offload - GSO
14 * Generic Receive Offload - GRO
15 * Partial Generic Segmentation Offload - GSO_PARTIAL
17 TCP Segmentation Offload
18 ========================
20 TCP segmentation allows a device to segment a single frame into multiple
21 frames with a data payload size specified in skb_shinfo()->gso_size.
22 When TCP segmentation requested the bit for either SKB_GSO_TCP or
23 SKB_GSO_TCP6 should be set in skb_shinfo()->gso_type and
24 skb_shinfo()->gso_size should be set to a non-zero value.
26 TCP segmentation is dependent on support for the use of partial checksum
27 offload. For this reason TSO is normally disabled if the Tx checksum
28 offload for a given device is disabled.
30 In order to support TCP segmentation offload it is necessary to populate
31 the network and transport header offsets of the skbuff so that the device
32 drivers will be able determine the offsets of the IP or IPv6 header and the
33 TCP header. In addition as CHECKSUM_PARTIAL is required csum_start should
34 also point to the TCP header of the packet.
36 For IPv4 segmentation we support one of two types in terms of the IP ID.
37 The default behavior is to increment the IP ID with every segment. If the
38 GSO type SKB_GSO_TCP_FIXEDID is specified then we will not increment the IP
39 ID and all segments will use the same IP ID. If a device has
40 NETIF_F_TSO_MANGLEID set then the IP ID can be ignored when performing TSO
41 and we will either increment the IP ID for all frames, or leave it at a
42 static value based on driver preference.
44 UDP Fragmentation Offload
45 =========================
47 UDP fragmentation offload allows a device to fragment an oversized UDP
48 datagram into multiple IPv4 fragments. Many of the requirements for UDP
49 fragmentation offload are the same as TSO. However the IPv4 ID for
50 fragments should not increment as a single IPv4 datagram is fragmented.
52 IPIP, SIT, GRE, UDP Tunnel, and Remote Checksum Offloads
53 ========================================================
55 In addition to the offloads described above it is possible for a frame to
56 contain additional headers such as an outer tunnel. In order to account
57 for such instances an additional set of segmentation offload types were
58 introduced including SKB_GSO_IPXIP4, SKB_GSO_IPXIP6, SKB_GSO_GRE, and
59 SKB_GSO_UDP_TUNNEL. These extra segmentation types are used to identify
60 cases where there are more than just 1 set of headers. For example in the
61 case of IPIP and SIT we should have the network and transport headers moved
62 from the standard list of headers to "inner" header offsets.
64 Currently only two levels of headers are supported. The convention is to
65 refer to the tunnel headers as the outer headers, while the encapsulated
66 data is normally referred to as the inner headers. Below is the list of
67 calls to access the given headers:
72 Network skb_network_header skb_inner_network_header
73 Transport skb_transport_header
77 MAC skb_mac_header skb_inner_mac_header
78 Network skb_network_header skb_inner_network_header
79 Transport skb_transport_header skb_inner_transport_header
81 In addition to the above tunnel types there are also SKB_GSO_GRE_CSUM and
82 SKB_GSO_UDP_TUNNEL_CSUM. These two additional tunnel types reflect the
83 fact that the outer header also requests to have a non-zero checksum
84 included in the outer header.
86 Finally there is SKB_GSO_REMCSUM which indicates that a given tunnel header
87 has requested a remote checksum offload. In this case the inner headers
88 will be left with a partial checksum and only the outer header checksum
91 Generic Segmentation Offload
92 ============================
94 Generic segmentation offload is a pure software offload that is meant to
95 deal with cases where device drivers cannot perform the offloads described
96 above. What occurs in GSO is that a given skbuff will have its data broken
97 out over multiple skbuffs that have been resized to match the MSS provided
98 via skb_shinfo()->gso_size.
100 Before enabling any hardware segmentation offload a corresponding software
101 offload is required in GSO. Otherwise it becomes possible for a frame to
102 be re-routed between devices and end up being unable to be transmitted.
104 Generic Receive Offload
105 =======================
107 Generic receive offload is the complement to GSO. Ideally any frame
108 assembled by GRO should be segmented to create an identical sequence of
109 frames using GSO, and any sequence of frames segmented by GSO should be
110 able to be reassembled back to the original by GRO. The only exception to
111 this is IPv4 ID in the case that the DF bit is set for a given IP header.
112 If the value of the IPv4 ID is not sequentially incrementing it will be
113 altered so that it is when a frame assembled via GRO is segmented via GSO.
115 Partial Generic Segmentation Offload
116 ====================================
118 Partial generic segmentation offload is a hybrid between TSO and GSO. What
119 it effectively does is take advantage of certain traits of TCP and tunnels
120 so that instead of having to rewrite the packet headers for each segment
121 only the inner-most transport header and possibly the outer-most network
122 header need to be updated. This allows devices that do not support tunnel
123 offloads or tunnel offloads with checksum to still make use of segmentation.
125 With the partial offload what occurs is that all headers excluding the
126 inner transport header are updated such that they will contain the correct
127 values for if the header was simply duplicated. The one exception to this
128 is the outer IPv4 ID field. It is up to the device drivers to guarantee
129 that the IPv4 ID field is incremented in the case that a given header does
130 not have the DF bit set.