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> | |
14 | #include <net/protocol.h> | |
15 | #include <net/gre.h> | |
16 | ||
17 | static int gre_gso_send_check(struct sk_buff *skb) | |
18 | { | |
19 | if (!skb->encapsulation) | |
20 | return -EINVAL; | |
21 | return 0; | |
22 | } | |
23 | ||
24 | static struct sk_buff *gre_gso_segment(struct sk_buff *skb, | |
25 | netdev_features_t features) | |
26 | { | |
27 | struct sk_buff *segs = ERR_PTR(-EINVAL); | |
28 | netdev_features_t enc_features; | |
b884b1a4 | 29 | int ghl; |
c50cd357 | 30 | struct gre_base_hdr *greh; |
7a7ffbab | 31 | u16 mac_offset = skb->mac_header; |
c50cd357 DB |
32 | int mac_len = skb->mac_len; |
33 | __be16 protocol = skb->protocol; | |
34 | int tnl_hlen; | |
35 | bool csum; | |
36 | ||
37 | if (unlikely(skb_shinfo(skb)->gso_type & | |
38 | ~(SKB_GSO_TCPV4 | | |
39 | SKB_GSO_TCPV6 | | |
40 | SKB_GSO_UDP | | |
41 | SKB_GSO_DODGY | | |
42 | SKB_GSO_TCP_ECN | | |
cb32f511 ED |
43 | SKB_GSO_GRE | |
44 | SKB_GSO_IPIP))) | |
c50cd357 DB |
45 | goto out; |
46 | ||
47 | if (unlikely(!pskb_may_pull(skb, sizeof(*greh)))) | |
48 | goto out; | |
49 | ||
50 | greh = (struct gre_base_hdr *)skb_transport_header(skb); | |
51 | ||
b884b1a4 NC |
52 | ghl = skb_inner_network_header(skb) - skb_transport_header(skb); |
53 | if (unlikely(ghl < sizeof(*greh))) | |
54 | goto out; | |
55 | ||
56 | csum = !!(greh->flags & GRE_CSUM); | |
c50cd357 | 57 | |
7a7ffbab WCC |
58 | if (unlikely(!pskb_may_pull(skb, ghl))) |
59 | goto out; | |
60 | ||
c50cd357 DB |
61 | /* setup inner skb. */ |
62 | skb->protocol = greh->protocol; | |
63 | skb->encapsulation = 0; | |
64 | ||
c50cd357 DB |
65 | __skb_pull(skb, ghl); |
66 | skb_reset_mac_header(skb); | |
67 | skb_set_network_header(skb, skb_inner_network_offset(skb)); | |
68 | skb->mac_len = skb_inner_network_offset(skb); | |
69 | ||
70 | /* segment inner packet. */ | |
71 | enc_features = skb->dev->hw_enc_features & netif_skb_features(skb); | |
72 | segs = skb_mac_gso_segment(skb, enc_features); | |
7a7ffbab WCC |
73 | if (!segs || IS_ERR(segs)) { |
74 | skb_gso_error_unwind(skb, protocol, ghl, mac_offset, mac_len); | |
c50cd357 | 75 | goto out; |
7a7ffbab | 76 | } |
c50cd357 DB |
77 | |
78 | skb = segs; | |
79 | tnl_hlen = skb_tnl_header_len(skb); | |
80 | do { | |
81 | __skb_push(skb, ghl); | |
82 | if (csum) { | |
83 | __be32 *pcsum; | |
84 | ||
85 | if (skb_has_shared_frag(skb)) { | |
86 | int err; | |
87 | ||
88 | err = __skb_linearize(skb); | |
89 | if (err) { | |
0c1072ae | 90 | kfree_skb_list(segs); |
c50cd357 DB |
91 | segs = ERR_PTR(err); |
92 | goto out; | |
93 | } | |
94 | } | |
95 | ||
96 | greh = (struct gre_base_hdr *)(skb->data); | |
97 | pcsum = (__be32 *)(greh + 1); | |
98 | *pcsum = 0; | |
99 | *(__sum16 *)pcsum = csum_fold(skb_checksum(skb, 0, skb->len, 0)); | |
100 | } | |
101 | __skb_push(skb, tnl_hlen - ghl); | |
102 | ||
cdbaa0bb AD |
103 | skb_reset_inner_headers(skb); |
104 | skb->encapsulation = 1; | |
105 | ||
c50cd357 DB |
106 | skb_reset_mac_header(skb); |
107 | skb_set_network_header(skb, mac_len); | |
108 | skb->mac_len = mac_len; | |
109 | skb->protocol = protocol; | |
110 | } while ((skb = skb->next)); | |
111 | out: | |
112 | return segs; | |
113 | } | |
114 | ||
bf5a755f JC |
115 | /* Compute the whole skb csum in s/w and store it, then verify GRO csum |
116 | * starting from gro_offset. | |
117 | */ | |
118 | static __sum16 gro_skb_checksum(struct sk_buff *skb) | |
119 | { | |
120 | __sum16 sum; | |
121 | ||
122 | skb->csum = skb_checksum(skb, 0, skb->len, 0); | |
123 | NAPI_GRO_CB(skb)->csum = csum_sub(skb->csum, | |
124 | csum_partial(skb->data, skb_gro_offset(skb), 0)); | |
125 | sum = csum_fold(NAPI_GRO_CB(skb)->csum); | |
126 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) { | |
127 | if (unlikely(!sum)) | |
128 | netdev_rx_csum_fault(skb->dev); | |
129 | } else | |
130 | skb->ip_summed = CHECKSUM_COMPLETE; | |
131 | ||
132 | return sum; | |
133 | } | |
134 | ||
135 | static struct sk_buff **gre_gro_receive(struct sk_buff **head, | |
136 | struct sk_buff *skb) | |
137 | { | |
138 | struct sk_buff **pp = NULL; | |
139 | struct sk_buff *p; | |
140 | const struct gre_base_hdr *greh; | |
141 | unsigned int hlen, grehlen; | |
142 | unsigned int off; | |
143 | int flush = 1; | |
144 | struct packet_offload *ptype; | |
145 | __be16 type; | |
146 | ||
147 | off = skb_gro_offset(skb); | |
148 | hlen = off + sizeof(*greh); | |
149 | greh = skb_gro_header_fast(skb, off); | |
150 | if (skb_gro_header_hard(skb, hlen)) { | |
151 | greh = skb_gro_header_slow(skb, hlen, off); | |
152 | if (unlikely(!greh)) | |
153 | goto out; | |
154 | } | |
155 | ||
156 | /* Only support version 0 and K (key), C (csum) flags. Note that | |
157 | * although the support for the S (seq#) flag can be added easily | |
158 | * for GRO, this is problematic for GSO hence can not be enabled | |
159 | * here because a GRO pkt may end up in the forwarding path, thus | |
160 | * requiring GSO support to break it up correctly. | |
161 | */ | |
162 | if ((greh->flags & ~(GRE_KEY|GRE_CSUM)) != 0) | |
163 | goto out; | |
164 | ||
165 | type = greh->protocol; | |
166 | ||
167 | rcu_read_lock(); | |
168 | ptype = gro_find_receive_by_type(type); | |
169 | if (ptype == NULL) | |
170 | goto out_unlock; | |
171 | ||
172 | grehlen = GRE_HEADER_SECTION; | |
173 | ||
174 | if (greh->flags & GRE_KEY) | |
175 | grehlen += GRE_HEADER_SECTION; | |
176 | ||
177 | if (greh->flags & GRE_CSUM) | |
178 | grehlen += GRE_HEADER_SECTION; | |
179 | ||
180 | hlen = off + grehlen; | |
181 | if (skb_gro_header_hard(skb, hlen)) { | |
182 | greh = skb_gro_header_slow(skb, hlen, off); | |
183 | if (unlikely(!greh)) | |
184 | goto out_unlock; | |
185 | } | |
186 | if (greh->flags & GRE_CSUM) { /* Need to verify GRE csum first */ | |
187 | __sum16 csum = 0; | |
188 | ||
189 | if (skb->ip_summed == CHECKSUM_COMPLETE) | |
190 | csum = csum_fold(NAPI_GRO_CB(skb)->csum); | |
191 | /* Don't trust csum error calculated/reported by h/w */ | |
192 | if (skb->ip_summed == CHECKSUM_NONE || csum != 0) | |
193 | csum = gro_skb_checksum(skb); | |
194 | ||
195 | /* GRE CSUM is the 1's complement of the 1's complement sum | |
196 | * of the GRE hdr plus payload so it should add up to 0xffff | |
197 | * (and 0 after csum_fold()) just like the IPv4 hdr csum. | |
198 | */ | |
199 | if (csum) | |
200 | goto out_unlock; | |
201 | } | |
202 | flush = 0; | |
203 | ||
204 | for (p = *head; p; p = p->next) { | |
205 | const struct gre_base_hdr *greh2; | |
206 | ||
207 | if (!NAPI_GRO_CB(p)->same_flow) | |
208 | continue; | |
209 | ||
210 | /* The following checks are needed to ensure only pkts | |
211 | * from the same tunnel are considered for aggregation. | |
212 | * The criteria for "the same tunnel" includes: | |
213 | * 1) same version (we only support version 0 here) | |
214 | * 2) same protocol (we only support ETH_P_IP for now) | |
215 | * 3) same set of flags | |
216 | * 4) same key if the key field is present. | |
217 | */ | |
218 | greh2 = (struct gre_base_hdr *)(p->data + off); | |
219 | ||
220 | if (greh2->flags != greh->flags || | |
221 | greh2->protocol != greh->protocol) { | |
222 | NAPI_GRO_CB(p)->same_flow = 0; | |
223 | continue; | |
224 | } | |
225 | if (greh->flags & GRE_KEY) { | |
226 | /* compare keys */ | |
227 | if (*(__be32 *)(greh2+1) != *(__be32 *)(greh+1)) { | |
228 | NAPI_GRO_CB(p)->same_flow = 0; | |
229 | continue; | |
230 | } | |
231 | } | |
232 | } | |
233 | ||
234 | skb_gro_pull(skb, grehlen); | |
235 | ||
236 | /* Adjusted NAPI_GRO_CB(skb)->csum after skb_gro_pull()*/ | |
237 | skb_gro_postpull_rcsum(skb, greh, grehlen); | |
238 | ||
239 | pp = ptype->callbacks.gro_receive(head, skb); | |
240 | ||
241 | out_unlock: | |
242 | rcu_read_unlock(); | |
243 | out: | |
244 | NAPI_GRO_CB(skb)->flush |= flush; | |
245 | ||
246 | return pp; | |
247 | } | |
248 | ||
d10dbad2 | 249 | static int gre_gro_complete(struct sk_buff *skb, int nhoff) |
bf5a755f JC |
250 | { |
251 | struct gre_base_hdr *greh = (struct gre_base_hdr *)(skb->data + nhoff); | |
252 | struct packet_offload *ptype; | |
253 | unsigned int grehlen = sizeof(*greh); | |
254 | int err = -ENOENT; | |
255 | __be16 type; | |
256 | ||
257 | type = greh->protocol; | |
258 | if (greh->flags & GRE_KEY) | |
259 | grehlen += GRE_HEADER_SECTION; | |
260 | ||
261 | if (greh->flags & GRE_CSUM) | |
262 | grehlen += GRE_HEADER_SECTION; | |
263 | ||
264 | rcu_read_lock(); | |
265 | ptype = gro_find_complete_by_type(type); | |
266 | if (ptype != NULL) | |
267 | err = ptype->callbacks.gro_complete(skb, nhoff + grehlen); | |
268 | ||
269 | rcu_read_unlock(); | |
270 | return err; | |
271 | } | |
272 | ||
c50cd357 DB |
273 | static const struct net_offload gre_offload = { |
274 | .callbacks = { | |
275 | .gso_send_check = gre_gso_send_check, | |
276 | .gso_segment = gre_gso_segment, | |
bf5a755f JC |
277 | .gro_receive = gre_gro_receive, |
278 | .gro_complete = gre_gro_complete, | |
c50cd357 DB |
279 | }, |
280 | }; | |
281 | ||
438e38fa | 282 | static int __init gre_offload_init(void) |
c50cd357 DB |
283 | { |
284 | return inet_add_offload(&gre_offload, IPPROTO_GRE); | |
285 | } | |
286 | ||
438e38fa | 287 | static void __exit gre_offload_exit(void) |
c50cd357 DB |
288 | { |
289 | inet_del_offload(&gre_offload, IPPROTO_GRE); | |
290 | } | |
438e38fa ED |
291 | |
292 | module_init(gre_offload_init); | |
293 | module_exit(gre_offload_exit); |