1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * IPV4 GSO/GRO offload support
4 * Linux INET implementation
5 *
6 * GRE GSO support
7 */
8
9 #include <linux/skbuff.h>
10 #include <linux/init.h>
11 #include <net/protocol.h>
12 #include <net/gre.h>
13
gre_gso_segment(struct sk_buff * skb,netdev_features_t features)14 static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
15 netdev_features_t features)
16 {
17 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
18 bool need_csum, offload_csum, gso_partial, need_ipsec;
19 struct sk_buff *segs = ERR_PTR(-EINVAL);
20 u16 mac_offset = skb->mac_header;
21 __be16 protocol = skb->protocol;
22 u16 mac_len = skb->mac_len;
23 int gre_offset, outer_hlen;
24
25 if (!skb->encapsulation)
26 goto out;
27
28 if (unlikely(tnl_hlen < sizeof(struct gre_base_hdr)))
29 goto out;
30
31 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
32 goto out;
33
34 /* setup inner skb. */
35 skb->encapsulation = 0;
36 SKB_GSO_CB(skb)->encap_level = 0;
37 __skb_pull(skb, tnl_hlen);
38 skb_reset_mac_header(skb);
39 skb_set_network_header(skb, skb_inner_network_offset(skb));
40 skb->mac_len = skb_inner_network_offset(skb);
41 skb->protocol = skb->inner_protocol;
42
43 need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_GRE_CSUM);
44 skb->encap_hdr_csum = need_csum;
45
46 features &= skb->dev->hw_enc_features;
47 if (need_csum)
48 features &= ~NETIF_F_SCTP_CRC;
49
50 need_ipsec = skb_dst(skb) && dst_xfrm(skb_dst(skb));
51 /* Try to offload checksum if possible */
52 offload_csum = !!(need_csum && !need_ipsec &&
53 (skb->dev->features & NETIF_F_HW_CSUM));
54
55 /* segment inner packet. */
56 segs = skb_mac_gso_segment(skb, features);
57 if (IS_ERR_OR_NULL(segs)) {
58 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
59 mac_len);
60 goto out;
61 }
62
63 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
64
65 outer_hlen = skb_tnl_header_len(skb);
66 gre_offset = outer_hlen - tnl_hlen;
67 skb = segs;
68 do {
69 struct gre_base_hdr *greh;
70 __sum16 *pcsum;
71
72 /* Set up inner headers if we are offloading inner checksum */
73 if (skb->ip_summed == CHECKSUM_PARTIAL) {
74 skb_reset_inner_headers(skb);
75 skb->encapsulation = 1;
76 }
77
78 skb->mac_len = mac_len;
79 skb->protocol = protocol;
80
81 __skb_push(skb, outer_hlen);
82 skb_reset_mac_header(skb);
83 skb_set_network_header(skb, mac_len);
84 skb_set_transport_header(skb, gre_offset);
85
86 if (!need_csum)
87 continue;
88
89 greh = (struct gre_base_hdr *)skb_transport_header(skb);
90 pcsum = (__sum16 *)(greh + 1);
91
92 if (gso_partial && skb_is_gso(skb)) {
93 unsigned int partial_adj;
94
95 /* Adjust checksum to account for the fact that
96 * the partial checksum is based on actual size
97 * whereas headers should be based on MSS size.
98 */
99 partial_adj = skb->len + skb_headroom(skb) -
100 SKB_GSO_CB(skb)->data_offset -
101 skb_shinfo(skb)->gso_size;
102 *pcsum = ~csum_fold((__force __wsum)htonl(partial_adj));
103 } else {
104 *pcsum = 0;
105 }
106
107 *(pcsum + 1) = 0;
108 if (skb->encapsulation || !offload_csum) {
109 *pcsum = gso_make_checksum(skb, 0);
110 } else {
111 skb->ip_summed = CHECKSUM_PARTIAL;
112 skb->csum_start = skb_transport_header(skb) - skb->head;
113 skb->csum_offset = sizeof(*greh);
114 }
115 } while ((skb = skb->next));
116 out:
117 return segs;
118 }
119
gre_gro_receive(struct list_head * head,struct sk_buff * skb)120 static struct sk_buff *gre_gro_receive(struct list_head *head,
121 struct sk_buff *skb)
122 {
123 struct sk_buff *pp = NULL;
124 struct sk_buff *p;
125 const struct gre_base_hdr *greh;
126 unsigned int hlen, grehlen;
127 unsigned int off;
128 int flush = 1;
129 struct packet_offload *ptype;
130 __be16 type;
131
132 if (NAPI_GRO_CB(skb)->encap_mark)
133 goto out;
134
135 NAPI_GRO_CB(skb)->encap_mark = 1;
136
137 off = skb_gro_offset(skb);
138 hlen = off + sizeof(*greh);
139 greh = skb_gro_header_fast(skb, off);
140 if (skb_gro_header_hard(skb, hlen)) {
141 greh = skb_gro_header_slow(skb, hlen, off);
142 if (unlikely(!greh))
143 goto out;
144 }
145
146 /* Only support version 0 and K (key), C (csum) flags. Note that
147 * although the support for the S (seq#) flag can be added easily
148 * for GRO, this is problematic for GSO hence can not be enabled
149 * here because a GRO pkt may end up in the forwarding path, thus
150 * requiring GSO support to break it up correctly.
151 */
152 if ((greh->flags & ~(GRE_KEY|GRE_CSUM)) != 0)
153 goto out;
154
155 /* We can only support GRE_CSUM if we can track the location of
156 * the GRE header. In the case of FOU/GUE we cannot because the
157 * outer UDP header displaces the GRE header leaving us in a state
158 * of limbo.
159 */
160 if ((greh->flags & GRE_CSUM) && NAPI_GRO_CB(skb)->is_fou)
161 goto out;
162
163 type = greh->protocol;
164
165 rcu_read_lock();
166 ptype = gro_find_receive_by_type(type);
167 if (!ptype)
168 goto out_unlock;
169
170 grehlen = GRE_HEADER_SECTION;
171
172 if (greh->flags & GRE_KEY)
173 grehlen += GRE_HEADER_SECTION;
174
175 if (greh->flags & GRE_CSUM)
176 grehlen += GRE_HEADER_SECTION;
177
178 hlen = off + grehlen;
179 if (skb_gro_header_hard(skb, hlen)) {
180 greh = skb_gro_header_slow(skb, hlen, off);
181 if (unlikely(!greh))
182 goto out_unlock;
183 }
184
185 /* Don't bother verifying checksum if we're going to flush anyway. */
186 if ((greh->flags & GRE_CSUM) && !NAPI_GRO_CB(skb)->flush) {
187 if (skb_gro_checksum_simple_validate(skb))
188 goto out_unlock;
189
190 skb_gro_checksum_try_convert(skb, IPPROTO_GRE,
191 null_compute_pseudo);
192 }
193
194 list_for_each_entry(p, head, list) {
195 const struct gre_base_hdr *greh2;
196
197 if (!NAPI_GRO_CB(p)->same_flow)
198 continue;
199
200 /* The following checks are needed to ensure only pkts
201 * from the same tunnel are considered for aggregation.
202 * The criteria for "the same tunnel" includes:
203 * 1) same version (we only support version 0 here)
204 * 2) same protocol (we only support ETH_P_IP for now)
205 * 3) same set of flags
206 * 4) same key if the key field is present.
207 */
208 greh2 = (struct gre_base_hdr *)(p->data + off);
209
210 if (greh2->flags != greh->flags ||
211 greh2->protocol != greh->protocol) {
212 NAPI_GRO_CB(p)->same_flow = 0;
213 continue;
214 }
215 if (greh->flags & GRE_KEY) {
216 /* compare keys */
217 if (*(__be32 *)(greh2+1) != *(__be32 *)(greh+1)) {
218 NAPI_GRO_CB(p)->same_flow = 0;
219 continue;
220 }
221 }
222 }
223
224 skb_gro_pull(skb, grehlen);
225
226 /* Adjusted NAPI_GRO_CB(skb)->csum after skb_gro_pull()*/
227 skb_gro_postpull_rcsum(skb, greh, grehlen);
228
229 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
230 flush = 0;
231
232 out_unlock:
233 rcu_read_unlock();
234 out:
235 skb_gro_flush_final(skb, pp, flush);
236
237 return pp;
238 }
239
gre_gro_complete(struct sk_buff * skb,int nhoff)240 static int gre_gro_complete(struct sk_buff *skb, int nhoff)
241 {
242 struct gre_base_hdr *greh = (struct gre_base_hdr *)(skb->data + nhoff);
243 struct packet_offload *ptype;
244 unsigned int grehlen = sizeof(*greh);
245 int err = -ENOENT;
246 __be16 type;
247
248 skb->encapsulation = 1;
249 skb_shinfo(skb)->gso_type = SKB_GSO_GRE;
250
251 type = greh->protocol;
252 if (greh->flags & GRE_KEY)
253 grehlen += GRE_HEADER_SECTION;
254
255 if (greh->flags & GRE_CSUM)
256 grehlen += GRE_HEADER_SECTION;
257
258 rcu_read_lock();
259 ptype = gro_find_complete_by_type(type);
260 if (ptype)
261 err = ptype->callbacks.gro_complete(skb, nhoff + grehlen);
262
263 rcu_read_unlock();
264
265 skb_set_inner_mac_header(skb, nhoff + grehlen);
266
267 return err;
268 }
269
270 static const struct net_offload gre_offload = {
271 .callbacks = {
272 .gso_segment = gre_gso_segment,
273 .gro_receive = gre_gro_receive,
274 .gro_complete = gre_gro_complete,
275 },
276 };
277
gre_offload_init(void)278 static int __init gre_offload_init(void)
279 {
280 int err;
281
282 err = inet_add_offload(&gre_offload, IPPROTO_GRE);
283 #if IS_ENABLED(CONFIG_IPV6)
284 if (err)
285 return err;
286
287 err = inet6_add_offload(&gre_offload, IPPROTO_GRE);
288 if (err)
289 inet_del_offload(&gre_offload, IPPROTO_GRE);
290 #endif
291
292 return err;
293 }
294 device_initcall(gre_offload_init);
295