2 * IPV6 GSO/GRO offload support
3 * Linux INET6 implementation
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.
11 #include <linux/kernel.h>
12 #include <linux/socket.h>
13 #include <linux/netdevice.h>
14 #include <linux/skbuff.h>
15 #include <linux/printk.h>
17 #include <net/protocol.h>
20 #include "ip6_offload.h"
22 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
24 const struct net_offload *ops = NULL;
27 struct ipv6_opt_hdr *opth;
30 if (proto != NEXTHDR_HOP) {
31 ops = rcu_dereference(inet6_offloads[proto]);
36 if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
40 if (unlikely(!pskb_may_pull(skb, 8)))
43 opth = (void *)skb->data;
44 len = ipv6_optlen(opth);
46 if (unlikely(!pskb_may_pull(skb, len)))
49 opth = (void *)skb->data;
50 proto = opth->nexthdr;
57 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
58 netdev_features_t features)
60 struct sk_buff *segs = ERR_PTR(-EINVAL);
61 struct ipv6hdr *ipv6h;
62 const struct net_offload *ops;
64 struct frag_hdr *fptr;
65 unsigned int unfrag_ip6hlen;
71 if (unlikely(skb_shinfo(skb)->gso_type &
83 SKB_GSO_UDP_TUNNEL_CSUM |
84 SKB_GSO_TUNNEL_REMCSUM |
88 skb_reset_network_header(skb);
89 nhoff = skb_network_header(skb) - skb_mac_header(skb);
90 if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
93 encap = SKB_GSO_CB(skb)->encap_level > 0;
95 features &= skb->dev->hw_enc_features;
96 SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
98 ipv6h = ipv6_hdr(skb);
99 __skb_pull(skb, sizeof(*ipv6h));
100 segs = ERR_PTR(-EPROTONOSUPPORT);
102 proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
104 if (skb->encapsulation &&
105 skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
106 udpfrag = proto == IPPROTO_UDP && encap;
108 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
110 ops = rcu_dereference(inet6_offloads[proto]);
111 if (likely(ops && ops->callbacks.gso_segment)) {
112 skb_reset_transport_header(skb);
113 segs = ops->callbacks.gso_segment(skb, features);
119 for (skb = segs; skb; skb = skb->next) {
120 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
121 ipv6h->payload_len = htons(skb->len - nhoff - sizeof(*ipv6h));
122 skb->network_header = (u8 *)ipv6h - skb->head;
125 unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
126 fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen);
127 fptr->frag_off = htons(offset);
129 fptr->frag_off |= htons(IP6_MF);
130 offset += (ntohs(ipv6h->payload_len) -
131 sizeof(struct frag_hdr));
134 skb_reset_inner_headers(skb);
141 /* Return the total length of all the extension hdrs, following the same
142 * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
144 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
145 const struct net_offload **opps)
147 struct ipv6_opt_hdr *opth = (void *)iph;
148 int len = 0, proto, optlen = sizeof(*iph);
150 proto = iph->nexthdr;
152 if (proto != NEXTHDR_HOP) {
153 *opps = rcu_dereference(inet6_offloads[proto]);
154 if (unlikely(!(*opps)))
156 if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
159 opth = (void *)opth + optlen;
160 optlen = ipv6_optlen(opth);
162 proto = opth->nexthdr;
167 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
170 const struct net_offload *ops;
171 struct sk_buff **pp = NULL;
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);
189 skb_set_network_header(skb, off);
190 skb_gro_pull(skb, sizeof(*iph));
191 skb_set_transport_header(skb, skb_gro_offset(skb));
193 flush += ntohs(iph->payload_len) != skb_gro_len(skb);
196 proto = iph->nexthdr;
197 ops = rcu_dereference(inet6_offloads[proto]);
198 if (!ops || !ops->callbacks.gro_receive) {
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));
205 ops = rcu_dereference(inet6_offloads[proto]);
206 if (!ops || !ops->callbacks.gro_receive)
212 NAPI_GRO_CB(skb)->proto = proto;
215 nlen = skb_network_header_len(skb);
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> */
221 if (!NAPI_GRO_CB(p)->same_flow)
224 iph2 = (struct ipv6hdr *)(p->data + off);
225 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
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?
233 if ((first_word & htonl(0xF00FFFFF)) ||
234 memcmp(&iph->nexthdr, &iph2->nexthdr,
235 nlen - offsetof(struct ipv6hdr, nexthdr))) {
236 NAPI_GRO_CB(p)->same_flow = 0;
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;
243 /* If the previous IP ID value was based on an atomic
244 * datagram we can overwrite the value and ignore it.
246 if (NAPI_GRO_CB(skb)->is_atomic)
247 NAPI_GRO_CB(p)->flush_id = 0;
250 NAPI_GRO_CB(skb)->is_atomic = true;
251 NAPI_GRO_CB(skb)->flush |= flush;
253 skb_gro_postpull_rcsum(skb, iph, nlen);
255 pp = ops->callbacks.gro_receive(head, skb);
261 NAPI_GRO_CB(skb)->flush |= flush;
266 static struct sk_buff **sit_gro_receive(struct sk_buff **head,
269 if (NAPI_GRO_CB(skb)->encap_mark) {
270 NAPI_GRO_CB(skb)->flush = 1;
274 NAPI_GRO_CB(skb)->encap_mark = 1;
276 return ipv6_gro_receive(head, skb);
279 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
281 const struct net_offload *ops;
282 struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
285 if (skb->encapsulation)
286 skb_set_inner_network_header(skb, nhoff);
288 iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
292 nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
293 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
296 err = ops->callbacks.gro_complete(skb, nhoff);
304 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
306 skb->encapsulation = 1;
307 skb_shinfo(skb)->gso_type |= SKB_GSO_SIT;
308 return ipv6_gro_complete(skb, nhoff);
311 static struct packet_offload ipv6_packet_offload __read_mostly = {
312 .type = cpu_to_be16(ETH_P_IPV6),
314 .gso_segment = ipv6_gso_segment,
315 .gro_receive = ipv6_gro_receive,
316 .gro_complete = ipv6_gro_complete,
320 static const struct net_offload sit_offload = {
322 .gso_segment = ipv6_gso_segment,
323 .gro_receive = sit_gro_receive,
324 .gro_complete = sit_gro_complete,
328 static int __init ipv6_offload_init(void)
331 if (tcpv6_offload_init() < 0)
332 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
333 if (ipv6_exthdrs_offload_init() < 0)
334 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
336 dev_add_offload(&ipv6_packet_offload);
338 inet_add_offload(&sit_offload, IPPROTO_IPV6);
343 fs_initcall(ipv6_offload_init);