blob: eb92deb12666fb56b6a289c55b8205fc27117933 [file] [log] [blame]
Daniel Borkmannc50cd352013-07-01 19:24:00 +02001/*
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>
Paul Gortmakercf172282014-01-15 11:19:55 -050014#include <linux/init.h>
Daniel Borkmannc50cd352013-07-01 19:24:00 +020015#include <net/protocol.h>
16#include <net/gre.h>
17
18static int gre_gso_send_check(struct sk_buff *skb)
19{
20 if (!skb->encapsulation)
21 return -EINVAL;
22 return 0;
23}
24
25static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
26 netdev_features_t features)
27{
28 struct sk_buff *segs = ERR_PTR(-EINVAL);
29 netdev_features_t enc_features;
Neal Cardwellb884b1a2014-01-09 20:47:17 -050030 int ghl;
Daniel Borkmannc50cd352013-07-01 19:24:00 +020031 struct gre_base_hdr *greh;
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -080032 u16 mac_offset = skb->mac_header;
Daniel Borkmannc50cd352013-07-01 19:24:00 +020033 int mac_len = skb->mac_len;
34 __be16 protocol = skb->protocol;
35 int tnl_hlen;
36 bool csum;
37
38 if (unlikely(skb_shinfo(skb)->gso_type &
39 ~(SKB_GSO_TCPV4 |
40 SKB_GSO_TCPV6 |
41 SKB_GSO_UDP |
42 SKB_GSO_DODGY |
43 SKB_GSO_TCP_ECN |
Eric Dumazetcb32f512013-10-19 11:42:57 -070044 SKB_GSO_GRE |
Tom Herbert4749c092014-06-04 17:20:23 -070045 SKB_GSO_GRE_CSUM |
Eric Dumazetcb32f512013-10-19 11:42:57 -070046 SKB_GSO_IPIP)))
Daniel Borkmannc50cd352013-07-01 19:24:00 +020047 goto out;
48
49 if (unlikely(!pskb_may_pull(skb, sizeof(*greh))))
50 goto out;
51
52 greh = (struct gre_base_hdr *)skb_transport_header(skb);
53
Neal Cardwellb884b1a2014-01-09 20:47:17 -050054 ghl = skb_inner_network_header(skb) - skb_transport_header(skb);
55 if (unlikely(ghl < sizeof(*greh)))
56 goto out;
57
58 csum = !!(greh->flags & GRE_CSUM);
Tom Herbert4749c092014-06-04 17:20:23 -070059 if (csum)
60 skb->encap_hdr_csum = 1;
Daniel Borkmannc50cd352013-07-01 19:24:00 +020061
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -080062 if (unlikely(!pskb_may_pull(skb, ghl)))
63 goto out;
64
Daniel Borkmannc50cd352013-07-01 19:24:00 +020065 /* setup inner skb. */
66 skb->protocol = greh->protocol;
67 skb->encapsulation = 0;
68
Daniel Borkmannc50cd352013-07-01 19:24:00 +020069 __skb_pull(skb, ghl);
70 skb_reset_mac_header(skb);
71 skb_set_network_header(skb, skb_inner_network_offset(skb));
72 skb->mac_len = skb_inner_network_offset(skb);
73
74 /* segment inner packet. */
75 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb);
76 segs = skb_mac_gso_segment(skb, enc_features);
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -080077 if (!segs || IS_ERR(segs)) {
78 skb_gso_error_unwind(skb, protocol, ghl, mac_offset, mac_len);
Daniel Borkmannc50cd352013-07-01 19:24:00 +020079 goto out;
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -080080 }
Daniel Borkmannc50cd352013-07-01 19:24:00 +020081
82 skb = segs;
83 tnl_hlen = skb_tnl_header_len(skb);
84 do {
85 __skb_push(skb, ghl);
86 if (csum) {
87 __be32 *pcsum;
88
89 if (skb_has_shared_frag(skb)) {
90 int err;
91
92 err = __skb_linearize(skb);
93 if (err) {
David S. Miller0c1072a2013-07-03 14:50:41 -070094 kfree_skb_list(segs);
Daniel Borkmannc50cd352013-07-01 19:24:00 +020095 segs = ERR_PTR(err);
96 goto out;
97 }
98 }
99
Tom Herbert4749c092014-06-04 17:20:23 -0700100 skb_reset_transport_header(skb);
101
102 greh = (struct gre_base_hdr *)
103 skb_transport_header(skb);
Daniel Borkmannc50cd352013-07-01 19:24:00 +0200104 pcsum = (__be32 *)(greh + 1);
105 *pcsum = 0;
Tom Herbert4749c092014-06-04 17:20:23 -0700106 *(__sum16 *)pcsum = gso_make_checksum(skb, 0);
Daniel Borkmannc50cd352013-07-01 19:24:00 +0200107 }
108 __skb_push(skb, tnl_hlen - ghl);
109
Alexander Duyckcdbaa0b2013-07-10 17:05:06 -0700110 skb_reset_inner_headers(skb);
111 skb->encapsulation = 1;
112
Daniel Borkmannc50cd352013-07-01 19:24:00 +0200113 skb_reset_mac_header(skb);
114 skb_set_network_header(skb, mac_len);
115 skb->mac_len = mac_len;
116 skb->protocol = protocol;
117 } while ((skb = skb->next));
118out:
119 return segs;
120}
121
Jerry Chubf5a7552014-01-07 10:23:19 -0800122/* Compute the whole skb csum in s/w and store it, then verify GRO csum
123 * starting from gro_offset.
124 */
125static __sum16 gro_skb_checksum(struct sk_buff *skb)
126{
127 __sum16 sum;
128
129 skb->csum = skb_checksum(skb, 0, skb->len, 0);
130 NAPI_GRO_CB(skb)->csum = csum_sub(skb->csum,
131 csum_partial(skb->data, skb_gro_offset(skb), 0));
132 sum = csum_fold(NAPI_GRO_CB(skb)->csum);
133 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) {
Tom Herbert7e3cead2014-06-10 18:54:19 -0700134 if (unlikely(!sum) && !skb->csum_complete_sw)
Jerry Chubf5a7552014-01-07 10:23:19 -0800135 netdev_rx_csum_fault(skb->dev);
Tom Herbert7e3cead2014-06-10 18:54:19 -0700136 } else {
Jerry Chubf5a7552014-01-07 10:23:19 -0800137 skb->ip_summed = CHECKSUM_COMPLETE;
Tom Herbert7e3cead2014-06-10 18:54:19 -0700138 skb->csum_complete_sw = 1;
139 }
Jerry Chubf5a7552014-01-07 10:23:19 -0800140
141 return sum;
142}
143
144static struct sk_buff **gre_gro_receive(struct sk_buff **head,
145 struct sk_buff *skb)
146{
147 struct sk_buff **pp = NULL;
148 struct sk_buff *p;
149 const struct gre_base_hdr *greh;
150 unsigned int hlen, grehlen;
151 unsigned int off;
152 int flush = 1;
153 struct packet_offload *ptype;
154 __be16 type;
155
156 off = skb_gro_offset(skb);
157 hlen = off + sizeof(*greh);
158 greh = skb_gro_header_fast(skb, off);
159 if (skb_gro_header_hard(skb, hlen)) {
160 greh = skb_gro_header_slow(skb, hlen, off);
161 if (unlikely(!greh))
162 goto out;
163 }
164
165 /* Only support version 0 and K (key), C (csum) flags. Note that
166 * although the support for the S (seq#) flag can be added easily
167 * for GRO, this is problematic for GSO hence can not be enabled
168 * here because a GRO pkt may end up in the forwarding path, thus
169 * requiring GSO support to break it up correctly.
170 */
171 if ((greh->flags & ~(GRE_KEY|GRE_CSUM)) != 0)
172 goto out;
173
174 type = greh->protocol;
175
176 rcu_read_lock();
177 ptype = gro_find_receive_by_type(type);
178 if (ptype == NULL)
179 goto out_unlock;
180
181 grehlen = GRE_HEADER_SECTION;
182
183 if (greh->flags & GRE_KEY)
184 grehlen += GRE_HEADER_SECTION;
185
186 if (greh->flags & GRE_CSUM)
187 grehlen += GRE_HEADER_SECTION;
188
189 hlen = off + grehlen;
190 if (skb_gro_header_hard(skb, hlen)) {
191 greh = skb_gro_header_slow(skb, hlen, off);
192 if (unlikely(!greh))
193 goto out_unlock;
194 }
195 if (greh->flags & GRE_CSUM) { /* Need to verify GRE csum first */
196 __sum16 csum = 0;
197
198 if (skb->ip_summed == CHECKSUM_COMPLETE)
199 csum = csum_fold(NAPI_GRO_CB(skb)->csum);
200 /* Don't trust csum error calculated/reported by h/w */
201 if (skb->ip_summed == CHECKSUM_NONE || csum != 0)
202 csum = gro_skb_checksum(skb);
203
204 /* GRE CSUM is the 1's complement of the 1's complement sum
205 * of the GRE hdr plus payload so it should add up to 0xffff
206 * (and 0 after csum_fold()) just like the IPv4 hdr csum.
207 */
208 if (csum)
209 goto out_unlock;
210 }
211 flush = 0;
212
213 for (p = *head; p; p = p->next) {
214 const struct gre_base_hdr *greh2;
215
216 if (!NAPI_GRO_CB(p)->same_flow)
217 continue;
218
219 /* The following checks are needed to ensure only pkts
220 * from the same tunnel are considered for aggregation.
221 * The criteria for "the same tunnel" includes:
222 * 1) same version (we only support version 0 here)
223 * 2) same protocol (we only support ETH_P_IP for now)
224 * 3) same set of flags
225 * 4) same key if the key field is present.
226 */
227 greh2 = (struct gre_base_hdr *)(p->data + off);
228
229 if (greh2->flags != greh->flags ||
230 greh2->protocol != greh->protocol) {
231 NAPI_GRO_CB(p)->same_flow = 0;
232 continue;
233 }
234 if (greh->flags & GRE_KEY) {
235 /* compare keys */
236 if (*(__be32 *)(greh2+1) != *(__be32 *)(greh+1)) {
237 NAPI_GRO_CB(p)->same_flow = 0;
238 continue;
239 }
240 }
241 }
242
243 skb_gro_pull(skb, grehlen);
244
245 /* Adjusted NAPI_GRO_CB(skb)->csum after skb_gro_pull()*/
246 skb_gro_postpull_rcsum(skb, greh, grehlen);
247
248 pp = ptype->callbacks.gro_receive(head, skb);
249
250out_unlock:
251 rcu_read_unlock();
252out:
253 NAPI_GRO_CB(skb)->flush |= flush;
254
255 return pp;
256}
257
Wei Yongjund10dbad2014-01-09 22:22:05 +0800258static int gre_gro_complete(struct sk_buff *skb, int nhoff)
Jerry Chubf5a7552014-01-07 10:23:19 -0800259{
260 struct gre_base_hdr *greh = (struct gre_base_hdr *)(skb->data + nhoff);
261 struct packet_offload *ptype;
262 unsigned int grehlen = sizeof(*greh);
263 int err = -ENOENT;
264 __be16 type;
265
266 type = greh->protocol;
267 if (greh->flags & GRE_KEY)
268 grehlen += GRE_HEADER_SECTION;
269
270 if (greh->flags & GRE_CSUM)
271 grehlen += GRE_HEADER_SECTION;
272
273 rcu_read_lock();
274 ptype = gro_find_complete_by_type(type);
275 if (ptype != NULL)
276 err = ptype->callbacks.gro_complete(skb, nhoff + grehlen);
277
278 rcu_read_unlock();
279 return err;
280}
281
Daniel Borkmannc50cd352013-07-01 19:24:00 +0200282static const struct net_offload gre_offload = {
283 .callbacks = {
284 .gso_send_check = gre_gso_send_check,
285 .gso_segment = gre_gso_segment,
Jerry Chubf5a7552014-01-07 10:23:19 -0800286 .gro_receive = gre_gro_receive,
287 .gro_complete = gre_gro_complete,
Daniel Borkmannc50cd352013-07-01 19:24:00 +0200288 },
289};
290
Eric Dumazet438e38f2014-01-06 14:03:07 -0800291static int __init gre_offload_init(void)
Daniel Borkmannc50cd352013-07-01 19:24:00 +0200292{
293 return inet_add_offload(&gre_offload, IPPROTO_GRE);
294}
Paul Gortmakercf172282014-01-15 11:19:55 -0500295device_initcall(gre_offload_init);