1.. SPDX-License-Identifier: GPL-2.0
2
3=====================================================
4Netdev features mess and how to get out from it alive
5=====================================================
6
7Author:
8	Michał Mirosław <mirq-linux@rere.qmqm.pl>
9
10
11
12Part I: Feature sets
13====================
14
15Long gone are the days when a network card would just take and give packets
16verbatim.  Today's devices add multiple features and bugs (read: offloads)
17that relieve an OS of various tasks like generating and checking checksums,
18splitting packets, classifying them.  Those capabilities and their state
19are commonly referred to as netdev features in Linux kernel world.
20
21There are currently three sets of features relevant to the driver, and
22one used internally by network core:
23
24 1. netdev->hw_features set contains features whose state may possibly
25    be changed (enabled or disabled) for a particular device by user's
26    request.  This set should be initialized in ndo_init callback and not
27    changed later.
28
29 2. netdev->features set contains features which are currently enabled
30    for a device.  This should be changed only by network core or in
31    error paths of ndo_set_features callback.
32
33 3. netdev->vlan_features set contains features whose state is inherited
34    by child VLAN devices (limits netdev->features set).  This is currently
35    used for all VLAN devices whether tags are stripped or inserted in
36    hardware or software.
37
38 4. netdev->wanted_features set contains feature set requested by user.
39    This set is filtered by ndo_fix_features callback whenever it or
40    some device-specific conditions change. This set is internal to
41    networking core and should not be referenced in drivers.
42
43
44
45Part II: Controlling enabled features
46=====================================
47
48When current feature set (netdev->features) is to be changed, new set
49is calculated and filtered by calling ndo_fix_features callback
50and netdev_fix_features(). If the resulting set differs from current
51set, it is passed to ndo_set_features callback and (if the callback
52returns success) replaces value stored in netdev->features.
53NETDEV_FEAT_CHANGE notification is issued after that whenever current
54set might have changed.
55
56The following events trigger recalculation:
57 1. device's registration, after ndo_init returned success
58 2. user requested changes in features state
59 3. netdev_update_features() is called
60
61ndo_*_features callbacks are called with rtnl_lock held. Missing callbacks
62are treated as always returning success.
63
64A driver that wants to trigger recalculation must do so by calling
65netdev_update_features() while holding rtnl_lock. This should not be done
66from ndo_*_features callbacks. netdev->features should not be modified by
67driver except by means of ndo_fix_features callback.
68
69
70
71Part III: Implementation hints
72==============================
73
74 * ndo_fix_features:
75
76All dependencies between features should be resolved here. The resulting
77set can be reduced further by networking core imposed limitations (as coded
78in netdev_fix_features()). For this reason it is safer to disable a feature
79when its dependencies are not met instead of forcing the dependency on.
80
81This callback should not modify hardware nor driver state (should be
82stateless).  It can be called multiple times between successive
83ndo_set_features calls.
84
85Callback must not alter features contained in NETIF_F_SOFT_FEATURES or
86NETIF_F_NEVER_CHANGE sets. The exception is NETIF_F_VLAN_CHALLENGED but
87care must be taken as the change won't affect already configured VLANs.
88
89 * ndo_set_features:
90
91Hardware should be reconfigured to match passed feature set. The set
92should not be altered unless some error condition happens that can't
93be reliably detected in ndo_fix_features. In this case, the callback
94should update netdev->features to match resulting hardware state.
95Errors returned are not (and cannot be) propagated anywhere except dmesg.
96(Note: successful return is zero, >0 means silent error.)
97
98
99
100Part IV: Features
101=================
102
103For current list of features, see include/linux/netdev_features.h.
104This section describes semantics of some of them.
105
106 * Transmit checksumming
107
108For complete description, see comments near the top of include/linux/skbuff.h.
109
110Note: NETIF_F_HW_CSUM is a superset of NETIF_F_IP_CSUM + NETIF_F_IPV6_CSUM.
111It means that device can fill TCP/UDP-like checksum anywhere in the packets
112whatever headers there might be.
113
114 * Transmit TCP segmentation offload
115
116NETIF_F_TSO_ECN means that hardware can properly split packets with CWR bit
117set, be it TCPv4 (when NETIF_F_TSO is enabled) or TCPv6 (NETIF_F_TSO6).
118
119 * Transmit UDP segmentation offload
120
121NETIF_F_GSO_UDP_L4 accepts a single UDP header with a payload that exceeds
122gso_size. On segmentation, it segments the payload on gso_size boundaries and
123replicates the network and UDP headers (fixing up the last one if less than
124gso_size).
125
126 * Transmit DMA from high memory
127
128On platforms where this is relevant, NETIF_F_HIGHDMA signals that
129ndo_start_xmit can handle skbs with frags in high memory.
130
131 * Transmit scatter-gather
132
133Those features say that ndo_start_xmit can handle fragmented skbs:
134NETIF_F_SG --- paged skbs (skb_shinfo()->frags), NETIF_F_FRAGLIST ---
135chained skbs (skb->next/prev list).
136
137 * Software features
138
139Features contained in NETIF_F_SOFT_FEATURES are features of networking
140stack. Driver should not change behaviour based on them.
141
142 * LLTX driver (deprecated for hardware drivers)
143
144NETIF_F_LLTX is meant to be used by drivers that don't need locking at all,
145e.g. software tunnels.
146
147This is also used in a few legacy drivers that implement their
148own locking, don't use it for new (hardware) drivers.
149
150 * netns-local device
151
152NETIF_F_NETNS_LOCAL is set for devices that are not allowed to move between
153network namespaces (e.g. loopback).
154
155Don't use it in drivers.
156
157 * VLAN challenged
158
159NETIF_F_VLAN_CHALLENGED should be set for devices which can't cope with VLAN
160headers. Some drivers set this because the cards can't handle the bigger MTU.
161[FIXME: Those cases could be fixed in VLAN code by allowing only reduced-MTU
162VLANs. This may be not useful, though.]
163
164*  rx-fcs
165
166This requests that the NIC append the Ethernet Frame Checksum (FCS)
167to the end of the skb data.  This allows sniffers and other tools to
168read the CRC recorded by the NIC on receipt of the packet.
169
170*  rx-all
171
172This requests that the NIC receive all possible frames, including errored
173frames (such as bad FCS, etc).  This can be helpful when sniffing a link with
174bad packets on it.  Some NICs may receive more packets if also put into normal
175PROMISC mode.
176
177*  rx-gro-hw
178
179This requests that the NIC enables Hardware GRO (generic receive offload).
180Hardware GRO is basically the exact reverse of TSO, and is generally
181stricter than Hardware LRO.  A packet stream merged by Hardware GRO must
182be re-segmentable by GSO or TSO back to the exact original packet stream.
183Hardware GRO is dependent on RXCSUM since every packet successfully merged
184by hardware must also have the checksum verified by hardware.
185