Commit | Line | Data |
---|---|---|
c50cd357 DB |
1 | /* |
2 | * IPV4 GSO/GRO offload support | |
3 | * Linux INET implementation | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or | |
6 | * modify it under the terms of the GNU General Public License | |
7 | * as published by the Free Software Foundation; either version | |
8 | * 2 of the License, or (at your option) any later version. | |
9 | * | |
10 | * GRE GSO support | |
11 | */ | |
12 | ||
13 | #include <linux/skbuff.h> | |
cf172283 | 14 | #include <linux/init.h> |
c50cd357 DB |
15 | #include <net/protocol.h> |
16 | #include <net/gre.h> | |
17 | ||
c50cd357 DB |
18 | static struct sk_buff *gre_gso_segment(struct sk_buff *skb, |
19 | netdev_features_t features) | |
20 | { | |
2e598af7 | 21 | int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb); |
c50cd357 | 22 | struct sk_buff *segs = ERR_PTR(-EINVAL); |
7a7ffbab | 23 | u16 mac_offset = skb->mac_header; |
c50cd357 | 24 | __be16 protocol = skb->protocol; |
2e598af7 AD |
25 | u16 mac_len = skb->mac_len; |
26 | int gre_offset, outer_hlen; | |
22463876 | 27 | bool need_csum, ufo; |
c50cd357 | 28 | |
53e50398 TH |
29 | if (!skb->encapsulation) |
30 | goto out; | |
31 | ||
2e598af7 | 32 | if (unlikely(tnl_hlen < sizeof(struct gre_base_hdr))) |
c50cd357 DB |
33 | goto out; |
34 | ||
2e598af7 | 35 | if (unlikely(!pskb_may_pull(skb, tnl_hlen))) |
b884b1a4 NC |
36 | goto out; |
37 | ||
c50cd357 | 38 | /* setup inner skb. */ |
c50cd357 | 39 | skb->encapsulation = 0; |
5197f349 | 40 | SKB_GSO_CB(skb)->encap_level = 0; |
2e598af7 | 41 | __skb_pull(skb, tnl_hlen); |
c50cd357 DB |
42 | skb_reset_mac_header(skb); |
43 | skb_set_network_header(skb, skb_inner_network_offset(skb)); | |
44 | skb->mac_len = skb_inner_network_offset(skb); | |
38720352 | 45 | skb->protocol = skb->inner_protocol; |
c50cd357 | 46 | |
2e598af7 AD |
47 | need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_GRE_CSUM); |
48 | skb->encap_hdr_csum = need_csum; | |
49 | ||
22463876 AD |
50 | ufo = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); |
51 | ||
bef3c6c9 AD |
52 | features &= skb->dev->hw_enc_features; |
53 | ||
22463876 AD |
54 | /* The only checksum offload we care about from here on out is the |
55 | * outer one so strip the existing checksum feature flags based | |
56 | * on the fact that we will be computing our checksum in software. | |
57 | */ | |
58 | if (ufo) { | |
59 | features &= ~NETIF_F_CSUM_MASK; | |
60 | if (!need_csum) | |
61 | features |= NETIF_F_HW_CSUM; | |
62 | } | |
63 | ||
c50cd357 | 64 | /* segment inner packet. */ |
bef3c6c9 | 65 | segs = skb_mac_gso_segment(skb, features); |
5a8dbf03 | 66 | if (IS_ERR_OR_NULL(segs)) { |
2e598af7 AD |
67 | skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset, |
68 | mac_len); | |
c50cd357 | 69 | goto out; |
7a7ffbab | 70 | } |
c50cd357 | 71 | |
2e598af7 AD |
72 | outer_hlen = skb_tnl_header_len(skb); |
73 | gre_offset = outer_hlen - tnl_hlen; | |
c50cd357 | 74 | skb = segs; |
c50cd357 | 75 | do { |
38720352 | 76 | struct gre_base_hdr *greh; |
802ab55a | 77 | __sum16 *pcsum; |
c50cd357 | 78 | |
22463876 AD |
79 | /* Set up inner headers if we are offloading inner checksum */ |
80 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
81 | skb_reset_inner_headers(skb); | |
82 | skb->encapsulation = 1; | |
83 | } | |
cdbaa0bb | 84 | |
c50cd357 DB |
85 | skb->mac_len = mac_len; |
86 | skb->protocol = protocol; | |
2e598af7 AD |
87 | |
88 | __skb_push(skb, outer_hlen); | |
89 | skb_reset_mac_header(skb); | |
90 | skb_set_network_header(skb, mac_len); | |
91 | skb_set_transport_header(skb, gre_offset); | |
92 | ||
93 | if (!need_csum) | |
94 | continue; | |
95 | ||
96 | greh = (struct gre_base_hdr *)skb_transport_header(skb); | |
802ab55a AD |
97 | pcsum = (__sum16 *)(greh + 1); |
98 | ||
99 | if (skb_is_gso(skb)) { | |
100 | unsigned int partial_adj; | |
101 | ||
102 | /* Adjust checksum to account for the fact that | |
103 | * the partial checksum is based on actual size | |
104 | * whereas headers should be based on MSS size. | |
105 | */ | |
106 | partial_adj = skb->len + skb_headroom(skb) - | |
107 | SKB_GSO_CB(skb)->data_offset - | |
108 | skb_shinfo(skb)->gso_size; | |
109 | *pcsum = ~csum_fold((__force __wsum)htonl(partial_adj)); | |
110 | } else { | |
111 | *pcsum = 0; | |
112 | } | |
2e598af7 | 113 | |
802ab55a AD |
114 | *(pcsum + 1) = 0; |
115 | *pcsum = gso_make_checksum(skb, 0); | |
c50cd357 DB |
116 | } while ((skb = skb->next)); |
117 | out: | |
118 | return segs; | |
119 | } | |
120 | ||
bf5a755f JC |
121 | static struct sk_buff **gre_gro_receive(struct sk_buff **head, |
122 | struct sk_buff *skb) | |
123 | { | |
124 | struct sk_buff **pp = NULL; | |
125 | struct sk_buff *p; | |
126 | const struct gre_base_hdr *greh; | |
127 | unsigned int hlen, grehlen; | |
128 | unsigned int off; | |
129 | int flush = 1; | |
130 | struct packet_offload *ptype; | |
131 | __be16 type; | |
132 | ||
fac8e0f5 JG |
133 | if (NAPI_GRO_CB(skb)->encap_mark) |
134 | goto out; | |
135 | ||
136 | NAPI_GRO_CB(skb)->encap_mark = 1; | |
137 | ||
bf5a755f JC |
138 | off = skb_gro_offset(skb); |
139 | hlen = off + sizeof(*greh); | |
140 | greh = skb_gro_header_fast(skb, off); | |
141 | if (skb_gro_header_hard(skb, hlen)) { | |
142 | greh = skb_gro_header_slow(skb, hlen, off); | |
143 | if (unlikely(!greh)) | |
144 | goto out; | |
145 | } | |
146 | ||
147 | /* Only support version 0 and K (key), C (csum) flags. Note that | |
148 | * although the support for the S (seq#) flag can be added easily | |
149 | * for GRO, this is problematic for GSO hence can not be enabled | |
150 | * here because a GRO pkt may end up in the forwarding path, thus | |
151 | * requiring GSO support to break it up correctly. | |
152 | */ | |
153 | if ((greh->flags & ~(GRE_KEY|GRE_CSUM)) != 0) | |
154 | goto out; | |
155 | ||
a0ca153f AD |
156 | /* We can only support GRE_CSUM if we can track the location of |
157 | * the GRE header. In the case of FOU/GUE we cannot because the | |
158 | * outer UDP header displaces the GRE header leaving us in a state | |
159 | * of limbo. | |
160 | */ | |
161 | if ((greh->flags & GRE_CSUM) && NAPI_GRO_CB(skb)->is_fou) | |
162 | goto out; | |
163 | ||
bf5a755f JC |
164 | type = greh->protocol; |
165 | ||
166 | rcu_read_lock(); | |
167 | ptype = gro_find_receive_by_type(type); | |
51456b29 | 168 | if (!ptype) |
bf5a755f JC |
169 | goto out_unlock; |
170 | ||
171 | grehlen = GRE_HEADER_SECTION; | |
172 | ||
173 | if (greh->flags & GRE_KEY) | |
174 | grehlen += GRE_HEADER_SECTION; | |
175 | ||
176 | if (greh->flags & GRE_CSUM) | |
177 | grehlen += GRE_HEADER_SECTION; | |
178 | ||
179 | hlen = off + grehlen; | |
180 | if (skb_gro_header_hard(skb, hlen)) { | |
181 | greh = skb_gro_header_slow(skb, hlen, off); | |
182 | if (unlikely(!greh)) | |
183 | goto out_unlock; | |
184 | } | |
758f75d1 TH |
185 | |
186 | /* Don't bother verifying checksum if we're going to flush anyway. */ | |
884d338c TH |
187 | if ((greh->flags & GRE_CSUM) && !NAPI_GRO_CB(skb)->flush) { |
188 | if (skb_gro_checksum_simple_validate(skb)) | |
bf5a755f | 189 | goto out_unlock; |
758f75d1 | 190 | |
884d338c TH |
191 | skb_gro_checksum_try_convert(skb, IPPROTO_GRE, 0, |
192 | null_compute_pseudo); | |
193 | } | |
194 | ||
bf5a755f JC |
195 | for (p = *head; p; p = p->next) { |
196 | const struct gre_base_hdr *greh2; | |
197 | ||
198 | if (!NAPI_GRO_CB(p)->same_flow) | |
199 | continue; | |
200 | ||
201 | /* The following checks are needed to ensure only pkts | |
202 | * from the same tunnel are considered for aggregation. | |
203 | * The criteria for "the same tunnel" includes: | |
204 | * 1) same version (we only support version 0 here) | |
205 | * 2) same protocol (we only support ETH_P_IP for now) | |
206 | * 3) same set of flags | |
207 | * 4) same key if the key field is present. | |
208 | */ | |
209 | greh2 = (struct gre_base_hdr *)(p->data + off); | |
210 | ||
211 | if (greh2->flags != greh->flags || | |
212 | greh2->protocol != greh->protocol) { | |
213 | NAPI_GRO_CB(p)->same_flow = 0; | |
214 | continue; | |
215 | } | |
216 | if (greh->flags & GRE_KEY) { | |
217 | /* compare keys */ | |
218 | if (*(__be32 *)(greh2+1) != *(__be32 *)(greh+1)) { | |
219 | NAPI_GRO_CB(p)->same_flow = 0; | |
220 | continue; | |
221 | } | |
222 | } | |
223 | } | |
224 | ||
225 | skb_gro_pull(skb, grehlen); | |
226 | ||
227 | /* Adjusted NAPI_GRO_CB(skb)->csum after skb_gro_pull()*/ | |
228 | skb_gro_postpull_rcsum(skb, greh, grehlen); | |
229 | ||
230 | pp = ptype->callbacks.gro_receive(head, skb); | |
c194cf93 | 231 | flush = 0; |
bf5a755f JC |
232 | |
233 | out_unlock: | |
234 | rcu_read_unlock(); | |
235 | out: | |
236 | NAPI_GRO_CB(skb)->flush |= flush; | |
237 | ||
238 | return pp; | |
239 | } | |
240 | ||
d10dbad2 | 241 | static int gre_gro_complete(struct sk_buff *skb, int nhoff) |
bf5a755f JC |
242 | { |
243 | struct gre_base_hdr *greh = (struct gre_base_hdr *)(skb->data + nhoff); | |
244 | struct packet_offload *ptype; | |
245 | unsigned int grehlen = sizeof(*greh); | |
246 | int err = -ENOENT; | |
247 | __be16 type; | |
248 | ||
c3caf119 JC |
249 | skb->encapsulation = 1; |
250 | skb_shinfo(skb)->gso_type = SKB_GSO_GRE; | |
251 | ||
bf5a755f JC |
252 | type = greh->protocol; |
253 | if (greh->flags & GRE_KEY) | |
254 | grehlen += GRE_HEADER_SECTION; | |
255 | ||
256 | if (greh->flags & GRE_CSUM) | |
257 | grehlen += GRE_HEADER_SECTION; | |
258 | ||
259 | rcu_read_lock(); | |
260 | ptype = gro_find_complete_by_type(type); | |
00db4124 | 261 | if (ptype) |
bf5a755f JC |
262 | err = ptype->callbacks.gro_complete(skb, nhoff + grehlen); |
263 | ||
264 | rcu_read_unlock(); | |
6fb2a756 TH |
265 | |
266 | skb_set_inner_mac_header(skb, nhoff + grehlen); | |
267 | ||
bf5a755f JC |
268 | return err; |
269 | } | |
270 | ||
c50cd357 DB |
271 | static const struct net_offload gre_offload = { |
272 | .callbacks = { | |
c50cd357 | 273 | .gso_segment = gre_gso_segment, |
bf5a755f JC |
274 | .gro_receive = gre_gro_receive, |
275 | .gro_complete = gre_gro_complete, | |
c50cd357 DB |
276 | }, |
277 | }; | |
278 | ||
438e38fa | 279 | static int __init gre_offload_init(void) |
c50cd357 | 280 | { |
e0c20967 AD |
281 | int err; |
282 | ||
283 | err = inet_add_offload(&gre_offload, IPPROTO_GRE); | |
284 | #if IS_ENABLED(CONFIG_IPV6) | |
285 | if (err) | |
286 | return err; | |
287 | ||
288 | err = inet6_add_offload(&gre_offload, IPPROTO_GRE); | |
289 | if (err) | |
290 | inet_del_offload(&gre_offload, IPPROTO_GRE); | |
291 | #endif | |
292 | ||
293 | return err; | |
c50cd357 | 294 | } |
cf172283 | 295 | device_initcall(gre_offload_init); |