]>
Commit | Line | Data |
---|---|---|
d1da932e VY |
1 | /* |
2 | * IPV6 GSO/GRO offload support | |
3 | * Linux INET6 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 | ||
11 | #include <linux/kernel.h> | |
12 | #include <linux/socket.h> | |
13 | #include <linux/netdevice.h> | |
14 | #include <linux/skbuff.h> | |
c6b641a4 | 15 | #include <linux/printk.h> |
d1da932e VY |
16 | |
17 | #include <net/protocol.h> | |
18 | #include <net/ipv6.h> | |
19 | ||
20 | #include "ip6_offload.h" | |
21 | ||
22 | static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto) | |
23 | { | |
24 | const struct net_offload *ops = NULL; | |
25 | ||
26 | for (;;) { | |
27 | struct ipv6_opt_hdr *opth; | |
28 | int len; | |
29 | ||
30 | if (proto != NEXTHDR_HOP) { | |
31 | ops = rcu_dereference(inet6_offloads[proto]); | |
32 | ||
33 | if (unlikely(!ops)) | |
34 | break; | |
35 | ||
36 | if (!(ops->flags & INET6_PROTO_GSO_EXTHDR)) | |
37 | break; | |
38 | } | |
39 | ||
40 | if (unlikely(!pskb_may_pull(skb, 8))) | |
41 | break; | |
42 | ||
43 | opth = (void *)skb->data; | |
44 | len = ipv6_optlen(opth); | |
45 | ||
46 | if (unlikely(!pskb_may_pull(skb, len))) | |
47 | break; | |
48 | ||
fc6fb41c | 49 | opth = (void *)skb->data; |
d1da932e VY |
50 | proto = opth->nexthdr; |
51 | __skb_pull(skb, len); | |
52 | } | |
53 | ||
54 | return proto; | |
55 | } | |
56 | ||
d1da932e VY |
57 | static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb, |
58 | netdev_features_t features) | |
59 | { | |
60 | struct sk_buff *segs = ERR_PTR(-EINVAL); | |
61 | struct ipv6hdr *ipv6h; | |
62 | const struct net_offload *ops; | |
63 | int proto; | |
64 | struct frag_hdr *fptr; | |
65 | unsigned int unfrag_ip6hlen; | |
66 | u8 *prevhdr; | |
67 | int offset = 0; | |
91a48a2e | 68 | bool encap, udpfrag; |
d3e5e006 | 69 | int nhoff; |
d1da932e | 70 | |
d1da932e | 71 | if (unlikely(skb_shinfo(skb)->gso_type & |
b6fef4c6 AD |
72 | ~(SKB_GSO_TCPV4 | |
73 | SKB_GSO_UDP | | |
d1da932e VY |
74 | SKB_GSO_DODGY | |
75 | SKB_GSO_TCP_ECN | | |
cbc53e08 AD |
76 | SKB_GSO_TCP_FIXEDID | |
77 | SKB_GSO_TCPV6 | | |
68c33163 | 78 | SKB_GSO_GRE | |
4749c09c | 79 | SKB_GSO_GRE_CSUM | |
cb32f511 | 80 | SKB_GSO_IPIP | |
61c1db7f | 81 | SKB_GSO_SIT | |
73136267 | 82 | SKB_GSO_UDP_TUNNEL | |
0f4f4ffa | 83 | SKB_GSO_UDP_TUNNEL_CSUM | |
e585f236 | 84 | SKB_GSO_TUNNEL_REMCSUM | |
d1da932e VY |
85 | 0))) |
86 | goto out; | |
87 | ||
d3e5e006 ED |
88 | skb_reset_network_header(skb); |
89 | nhoff = skb_network_header(skb) - skb_mac_header(skb); | |
d1da932e VY |
90 | if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h)))) |
91 | goto out; | |
92 | ||
91a48a2e HFS |
93 | encap = SKB_GSO_CB(skb)->encap_level > 0; |
94 | if (encap) | |
1e16aa3d | 95 | features &= skb->dev->hw_enc_features; |
d3e5e006 ED |
96 | SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h); |
97 | ||
d1da932e VY |
98 | ipv6h = ipv6_hdr(skb); |
99 | __skb_pull(skb, sizeof(*ipv6h)); | |
100 | segs = ERR_PTR(-EPROTONOSUPPORT); | |
101 | ||
102 | proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr); | |
b917eb15 | 103 | |
91a48a2e HFS |
104 | if (skb->encapsulation && |
105 | skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP)) | |
106 | udpfrag = proto == IPPROTO_UDP && encap; | |
107 | else | |
108 | udpfrag = proto == IPPROTO_UDP && !skb->encapsulation; | |
109 | ||
d1da932e | 110 | ops = rcu_dereference(inet6_offloads[proto]); |
f191a1d1 | 111 | if (likely(ops && ops->callbacks.gso_segment)) { |
d1da932e | 112 | skb_reset_transport_header(skb); |
f191a1d1 | 113 | segs = ops->callbacks.gso_segment(skb, features); |
d1da932e | 114 | } |
d1da932e VY |
115 | |
116 | if (IS_ERR(segs)) | |
117 | goto out; | |
118 | ||
119 | for (skb = segs; skb; skb = skb->next) { | |
d3e5e006 ED |
120 | ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff); |
121 | ipv6h->payload_len = htons(skb->len - nhoff - sizeof(*ipv6h)); | |
d3e5e006 ED |
122 | skb->network_header = (u8 *)ipv6h - skb->head; |
123 | ||
91a48a2e | 124 | if (udpfrag) { |
d1da932e | 125 | unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr); |
d3e5e006 | 126 | fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen); |
d1da932e | 127 | fptr->frag_off = htons(offset); |
53b24b8f | 128 | if (skb->next) |
d1da932e VY |
129 | fptr->frag_off |= htons(IP6_MF); |
130 | offset += (ntohs(ipv6h->payload_len) - | |
131 | sizeof(struct frag_hdr)); | |
132 | } | |
91a48a2e HFS |
133 | if (encap) |
134 | skb_reset_inner_headers(skb); | |
d1da932e VY |
135 | } |
136 | ||
137 | out: | |
138 | return segs; | |
139 | } | |
140 | ||
299603e8 JC |
141 | /* Return the total length of all the extension hdrs, following the same |
142 | * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs. | |
143 | */ | |
144 | static int ipv6_exthdrs_len(struct ipv6hdr *iph, | |
145 | const struct net_offload **opps) | |
146 | { | |
810c23a3 JC |
147 | struct ipv6_opt_hdr *opth = (void *)iph; |
148 | int len = 0, proto, optlen = sizeof(*iph); | |
299603e8 JC |
149 | |
150 | proto = iph->nexthdr; | |
151 | for (;;) { | |
152 | if (proto != NEXTHDR_HOP) { | |
153 | *opps = rcu_dereference(inet6_offloads[proto]); | |
154 | if (unlikely(!(*opps))) | |
155 | break; | |
156 | if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR)) | |
157 | break; | |
158 | } | |
810c23a3 JC |
159 | opth = (void *)opth + optlen; |
160 | optlen = ipv6_optlen(opth); | |
299603e8 JC |
161 | len += optlen; |
162 | proto = opth->nexthdr; | |
163 | } | |
164 | return len; | |
165 | } | |
166 | ||
d1da932e VY |
167 | static struct sk_buff **ipv6_gro_receive(struct sk_buff **head, |
168 | struct sk_buff *skb) | |
169 | { | |
170 | const struct net_offload *ops; | |
171 | struct sk_buff **pp = NULL; | |
172 | struct sk_buff *p; | |
173 | struct ipv6hdr *iph; | |
174 | unsigned int nlen; | |
175 | unsigned int hlen; | |
176 | unsigned int off; | |
bf5a755f | 177 | u16 flush = 1; |
d1da932e | 178 | int proto; |
d1da932e VY |
179 | |
180 | off = skb_gro_offset(skb); | |
181 | hlen = off + sizeof(*iph); | |
182 | iph = skb_gro_header_fast(skb, off); | |
183 | if (skb_gro_header_hard(skb, hlen)) { | |
184 | iph = skb_gro_header_slow(skb, hlen, off); | |
185 | if (unlikely(!iph)) | |
186 | goto out; | |
187 | } | |
188 | ||
299603e8 | 189 | skb_set_network_header(skb, off); |
d1da932e VY |
190 | skb_gro_pull(skb, sizeof(*iph)); |
191 | skb_set_transport_header(skb, skb_gro_offset(skb)); | |
192 | ||
193 | flush += ntohs(iph->payload_len) != skb_gro_len(skb); | |
194 | ||
195 | rcu_read_lock(); | |
196 | proto = iph->nexthdr; | |
197 | ops = rcu_dereference(inet6_offloads[proto]); | |
f191a1d1 | 198 | if (!ops || !ops->callbacks.gro_receive) { |
d1da932e VY |
199 | __pskb_pull(skb, skb_gro_offset(skb)); |
200 | proto = ipv6_gso_pull_exthdrs(skb, proto); | |
201 | skb_gro_pull(skb, -skb_transport_offset(skb)); | |
202 | skb_reset_transport_header(skb); | |
203 | __skb_push(skb, skb_gro_offset(skb)); | |
204 | ||
205 | ops = rcu_dereference(inet6_offloads[proto]); | |
f191a1d1 | 206 | if (!ops || !ops->callbacks.gro_receive) |
d1da932e VY |
207 | goto out_unlock; |
208 | ||
209 | iph = ipv6_hdr(skb); | |
210 | } | |
211 | ||
212 | NAPI_GRO_CB(skb)->proto = proto; | |
213 | ||
214 | flush--; | |
215 | nlen = skb_network_header_len(skb); | |
216 | ||
217 | for (p = *head; p; p = p->next) { | |
218 | const struct ipv6hdr *iph2; | |
219 | __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */ | |
220 | ||
221 | if (!NAPI_GRO_CB(p)->same_flow) | |
222 | continue; | |
223 | ||
299603e8 | 224 | iph2 = (struct ipv6hdr *)(p->data + off); |
67ba4152 | 225 | first_word = *(__be32 *)iph ^ *(__be32 *)iph2; |
d1da932e | 226 | |
299603e8 JC |
227 | /* All fields must match except length and Traffic Class. |
228 | * XXX skbs on the gro_list have all been parsed and pulled | |
229 | * already so we don't need to compare nlen | |
230 | * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops))) | |
231 | * memcmp() alone below is suffcient, right? | |
232 | */ | |
233 | if ((first_word & htonl(0xF00FFFFF)) || | |
d1da932e VY |
234 | memcmp(&iph->nexthdr, &iph2->nexthdr, |
235 | nlen - offsetof(struct ipv6hdr, nexthdr))) { | |
236 | NAPI_GRO_CB(p)->same_flow = 0; | |
237 | continue; | |
238 | } | |
239 | /* flush if Traffic Class fields are different */ | |
240 | NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000)); | |
241 | NAPI_GRO_CB(p)->flush |= flush; | |
03d56daa TH |
242 | |
243 | /* Clear flush_id, there's really no concept of ID in IPv6. */ | |
244 | NAPI_GRO_CB(p)->flush_id = 0; | |
d1da932e VY |
245 | } |
246 | ||
247 | NAPI_GRO_CB(skb)->flush |= flush; | |
248 | ||
4de462ab | 249 | skb_gro_postpull_rcsum(skb, iph, nlen); |
d1da932e | 250 | |
f191a1d1 | 251 | pp = ops->callbacks.gro_receive(head, skb); |
d1da932e | 252 | |
d1da932e VY |
253 | out_unlock: |
254 | rcu_read_unlock(); | |
255 | ||
256 | out: | |
257 | NAPI_GRO_CB(skb)->flush |= flush; | |
258 | ||
259 | return pp; | |
260 | } | |
261 | ||
fac8e0f5 JG |
262 | static struct sk_buff **sit_gro_receive(struct sk_buff **head, |
263 | struct sk_buff *skb) | |
264 | { | |
265 | if (NAPI_GRO_CB(skb)->encap_mark) { | |
266 | NAPI_GRO_CB(skb)->flush = 1; | |
267 | return NULL; | |
268 | } | |
269 | ||
270 | NAPI_GRO_CB(skb)->encap_mark = 1; | |
271 | ||
272 | return ipv6_gro_receive(head, skb); | |
273 | } | |
274 | ||
299603e8 | 275 | static int ipv6_gro_complete(struct sk_buff *skb, int nhoff) |
d1da932e VY |
276 | { |
277 | const struct net_offload *ops; | |
299603e8 | 278 | struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff); |
d1da932e VY |
279 | int err = -ENOSYS; |
280 | ||
feec0cb3 ED |
281 | if (skb->encapsulation) |
282 | skb_set_inner_network_header(skb, nhoff); | |
283 | ||
299603e8 | 284 | iph->payload_len = htons(skb->len - nhoff - sizeof(*iph)); |
d1da932e VY |
285 | |
286 | rcu_read_lock(); | |
299603e8 JC |
287 | |
288 | nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops); | |
f191a1d1 | 289 | if (WARN_ON(!ops || !ops->callbacks.gro_complete)) |
d1da932e VY |
290 | goto out_unlock; |
291 | ||
299603e8 | 292 | err = ops->callbacks.gro_complete(skb, nhoff); |
d1da932e VY |
293 | |
294 | out_unlock: | |
295 | rcu_read_unlock(); | |
296 | ||
297 | return err; | |
298 | } | |
299 | ||
feec0cb3 ED |
300 | static int sit_gro_complete(struct sk_buff *skb, int nhoff) |
301 | { | |
302 | skb->encapsulation = 1; | |
303 | skb_shinfo(skb)->gso_type |= SKB_GSO_SIT; | |
304 | return ipv6_gro_complete(skb, nhoff); | |
305 | } | |
306 | ||
d1da932e VY |
307 | static struct packet_offload ipv6_packet_offload __read_mostly = { |
308 | .type = cpu_to_be16(ETH_P_IPV6), | |
f191a1d1 | 309 | .callbacks = { |
f191a1d1 VY |
310 | .gso_segment = ipv6_gso_segment, |
311 | .gro_receive = ipv6_gro_receive, | |
312 | .gro_complete = ipv6_gro_complete, | |
313 | }, | |
d1da932e VY |
314 | }; |
315 | ||
61c1db7f ED |
316 | static const struct net_offload sit_offload = { |
317 | .callbacks = { | |
61c1db7f | 318 | .gso_segment = ipv6_gso_segment, |
fac8e0f5 | 319 | .gro_receive = sit_gro_receive, |
feec0cb3 | 320 | .gro_complete = sit_gro_complete, |
61c1db7f ED |
321 | }, |
322 | }; | |
323 | ||
c6b641a4 | 324 | static int __init ipv6_offload_init(void) |
d1da932e | 325 | { |
c6b641a4 VY |
326 | |
327 | if (tcpv6_offload_init() < 0) | |
328 | pr_crit("%s: Cannot add TCP protocol offload\n", __func__); | |
c6b641a4 VY |
329 | if (ipv6_exthdrs_offload_init() < 0) |
330 | pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__); | |
331 | ||
d1da932e | 332 | dev_add_offload(&ipv6_packet_offload); |
61c1db7f ED |
333 | |
334 | inet_add_offload(&sit_offload, IPPROTO_IPV6); | |
335 | ||
c6b641a4 | 336 | return 0; |
d1da932e VY |
337 | } |
338 | ||
c6b641a4 | 339 | fs_initcall(ipv6_offload_init); |