Merge branch 'stable-3.17' of git://git.infradead.org/users/pcmoore/selinux
[deliverable/linux.git] / net / ipv4 / tcp_offload.c
CommitLineData
28850dc7
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 * TCPv4 GSO/GRO support
11 */
12
13#include <linux/skbuff.h>
14#include <net/tcp.h>
15#include <net/protocol.h>
16
f066e2b0
WB
17static void tcp_gso_tstamp(struct sk_buff *skb, unsigned int ts_seq,
18 unsigned int seq, unsigned int mss)
4ed2d765
WB
19{
20 while (skb) {
f066e2b0
WB
21 if (before(ts_seq, seq + mss)) {
22 skb_shinfo(skb)->tx_flags |= SKBTX_SW_TSTAMP;
4ed2d765
WB
23 skb_shinfo(skb)->tskey = ts_seq;
24 return;
25 }
26
27 skb = skb->next;
28 seq += mss;
29 }
30}
31
28be6e07 32struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
28850dc7
DB
33 netdev_features_t features)
34{
35 struct sk_buff *segs = ERR_PTR(-EINVAL);
0d08c42c 36 unsigned int sum_truesize = 0;
28850dc7
DB
37 struct tcphdr *th;
38 unsigned int thlen;
39 unsigned int seq;
40 __be32 delta;
41 unsigned int oldlen;
42 unsigned int mss;
43 struct sk_buff *gso_skb = skb;
44 __sum16 newcheck;
45 bool ooo_okay, copy_destructor;
46
47 if (!pskb_may_pull(skb, sizeof(*th)))
48 goto out;
49
50 th = tcp_hdr(skb);
51 thlen = th->doff * 4;
52 if (thlen < sizeof(*th))
53 goto out;
54
55 if (!pskb_may_pull(skb, thlen))
56 goto out;
57
58 oldlen = (u16)~skb->len;
59 __skb_pull(skb, thlen);
60
61 mss = tcp_skb_mss(skb);
62 if (unlikely(skb->len <= mss))
63 goto out;
64
65 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
66 /* Packet is from an untrusted source, reset gso_segs. */
67 int type = skb_shinfo(skb)->gso_type;
68
69 if (unlikely(type &
70 ~(SKB_GSO_TCPV4 |
71 SKB_GSO_DODGY |
72 SKB_GSO_TCP_ECN |
73 SKB_GSO_TCPV6 |
74 SKB_GSO_GRE |
4749c09c 75 SKB_GSO_GRE_CSUM |
cb32f511 76 SKB_GSO_IPIP |
61c1db7f 77 SKB_GSO_SIT |
28850dc7
DB
78 SKB_GSO_MPLS |
79 SKB_GSO_UDP_TUNNEL |
0f4f4ffa 80 SKB_GSO_UDP_TUNNEL_CSUM |
28850dc7
DB
81 0) ||
82 !(type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))))
83 goto out;
84
85 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
86
87 segs = NULL;
88 goto out;
89 }
90
91 copy_destructor = gso_skb->destructor == tcp_wfree;
92 ooo_okay = gso_skb->ooo_okay;
93 /* All segments but the first should have ooo_okay cleared */
94 skb->ooo_okay = 0;
95
96 segs = skb_segment(skb, features);
97 if (IS_ERR(segs))
98 goto out;
99
100 /* Only first segment might have ooo_okay set */
101 segs->ooo_okay = ooo_okay;
102
103 delta = htonl(oldlen + (thlen + mss));
104
105 skb = segs;
106 th = tcp_hdr(skb);
107 seq = ntohl(th->seq);
108
4ed2d765
WB
109 if (unlikely(skb_shinfo(gso_skb)->tx_flags & SKBTX_SW_TSTAMP))
110 tcp_gso_tstamp(segs, skb_shinfo(gso_skb)->tskey, seq, mss);
111
28850dc7
DB
112 newcheck = ~csum_fold((__force __wsum)((__force u32)th->check +
113 (__force u32)delta));
114
115 do {
116 th->fin = th->psh = 0;
117 th->check = newcheck;
118
119 if (skb->ip_summed != CHECKSUM_PARTIAL)
e9c3a24b 120 th->check = gso_make_checksum(skb, ~th->check);
28850dc7
DB
121
122 seq += mss;
123 if (copy_destructor) {
124 skb->destructor = gso_skb->destructor;
125 skb->sk = gso_skb->sk;
0d08c42c 126 sum_truesize += skb->truesize;
28850dc7
DB
127 }
128 skb = skb->next;
129 th = tcp_hdr(skb);
130
131 th->seq = htonl(seq);
132 th->cwr = 0;
133 } while (skb->next);
134
135 /* Following permits TCP Small Queues to work well with GSO :
136 * The callback to TCP stack will be called at the time last frag
137 * is freed at TX completion, and not right now when gso_skb
138 * is freed by GSO engine
139 */
140 if (copy_destructor) {
141 swap(gso_skb->sk, skb->sk);
142 swap(gso_skb->destructor, skb->destructor);
0d08c42c
ED
143 sum_truesize += skb->truesize;
144 atomic_add(sum_truesize - gso_skb->truesize,
145 &skb->sk->sk_wmem_alloc);
28850dc7
DB
146 }
147
148 delta = htonl(oldlen + (skb_tail_pointer(skb) -
149 skb_transport_header(skb)) +
150 skb->data_len);
151 th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
152 (__force u32)delta));
153 if (skb->ip_summed != CHECKSUM_PARTIAL)
e9c3a24b 154 th->check = gso_make_checksum(skb, ~th->check);
28850dc7
DB
155out:
156 return segs;
157}
28850dc7
DB
158
159struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb)
160{
161 struct sk_buff **pp = NULL;
162 struct sk_buff *p;
163 struct tcphdr *th;
164 struct tcphdr *th2;
165 unsigned int len;
166 unsigned int thlen;
167 __be32 flags;
168 unsigned int mss = 1;
169 unsigned int hlen;
170 unsigned int off;
171 int flush = 1;
172 int i;
173
174 off = skb_gro_offset(skb);
175 hlen = off + sizeof(*th);
176 th = skb_gro_header_fast(skb, off);
177 if (skb_gro_header_hard(skb, hlen)) {
178 th = skb_gro_header_slow(skb, hlen, off);
179 if (unlikely(!th))
180 goto out;
181 }
182
183 thlen = th->doff * 4;
184 if (thlen < sizeof(*th))
185 goto out;
186
187 hlen = off + thlen;
188 if (skb_gro_header_hard(skb, hlen)) {
189 th = skb_gro_header_slow(skb, hlen, off);
190 if (unlikely(!th))
191 goto out;
192 }
193
194 skb_gro_pull(skb, thlen);
195
196 len = skb_gro_len(skb);
197 flags = tcp_flag_word(th);
198
199 for (; (p = *head); head = &p->next) {
200 if (!NAPI_GRO_CB(p)->same_flow)
201 continue;
202
203 th2 = tcp_hdr(p);
204
205 if (*(u32 *)&th->source ^ *(u32 *)&th2->source) {
206 NAPI_GRO_CB(p)->same_flow = 0;
207 continue;
208 }
209
210 goto found;
211 }
212
213 goto out_check_final;
214
215found:
bf5a755f
JC
216 /* Include the IP ID check below from the inner most IP hdr */
217 flush = NAPI_GRO_CB(p)->flush | NAPI_GRO_CB(p)->flush_id;
28850dc7
DB
218 flush |= (__force int)(flags & TCP_FLAG_CWR);
219 flush |= (__force int)((flags ^ tcp_flag_word(th2)) &
220 ~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH));
221 flush |= (__force int)(th->ack_seq ^ th2->ack_seq);
222 for (i = sizeof(*th); i < thlen; i += 4)
223 flush |= *(u32 *)((u8 *)th + i) ^
224 *(u32 *)((u8 *)th2 + i);
225
226 mss = tcp_skb_mss(p);
227
228 flush |= (len - 1) >= mss;
229 flush |= (ntohl(th2->seq) + skb_gro_len(p)) ^ ntohl(th->seq);
230
231 if (flush || skb_gro_receive(head, skb)) {
232 mss = 1;
233 goto out_check_final;
234 }
235
236 p = *head;
237 th2 = tcp_hdr(p);
238 tcp_flag_word(th2) |= flags & (TCP_FLAG_FIN | TCP_FLAG_PSH);
239
240out_check_final:
241 flush = len < mss;
242 flush |= (__force int)(flags & (TCP_FLAG_URG | TCP_FLAG_PSH |
243 TCP_FLAG_RST | TCP_FLAG_SYN |
244 TCP_FLAG_FIN));
245
246 if (p && (!NAPI_GRO_CB(skb)->same_flow || flush))
247 pp = head;
248
249out:
bf5a755f 250 NAPI_GRO_CB(skb)->flush |= (flush != 0);
28850dc7
DB
251
252 return pp;
253}
28850dc7
DB
254
255int tcp_gro_complete(struct sk_buff *skb)
256{
257 struct tcphdr *th = tcp_hdr(skb);
258
299603e8 259 skb->csum_start = (unsigned char *)th - skb->head;
28850dc7
DB
260 skb->csum_offset = offsetof(struct tcphdr, check);
261 skb->ip_summed = CHECKSUM_PARTIAL;
262
263 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
264
265 if (th->cwr)
266 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
267
268 return 0;
269}
270EXPORT_SYMBOL(tcp_gro_complete);
271
272static int tcp_v4_gso_send_check(struct sk_buff *skb)
273{
274 const struct iphdr *iph;
275 struct tcphdr *th;
276
277 if (!pskb_may_pull(skb, sizeof(*th)))
278 return -EINVAL;
279
280 iph = ip_hdr(skb);
281 th = tcp_hdr(skb);
282
283 th->check = 0;
284 skb->ip_summed = CHECKSUM_PARTIAL;
285 __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
286 return 0;
287}
288
289static struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
290{
299603e8 291 /* Use the IP hdr immediately proceeding for this transport */
28850dc7
DB
292 const struct iphdr *iph = skb_gro_network_header(skb);
293 __wsum wsum;
28850dc7 294
cc5c00bb
HX
295 /* Don't bother verifying checksum if we're going to flush anyway. */
296 if (NAPI_GRO_CB(skb)->flush)
297 goto skip_csum;
298
bf5a755f 299 wsum = NAPI_GRO_CB(skb)->csum;
b8ee93ba 300
28850dc7 301 switch (skb->ip_summed) {
b8ee93ba
HX
302 case CHECKSUM_NONE:
303 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb),
304 0);
305
306 /* fall through */
307
28850dc7
DB
308 case CHECKSUM_COMPLETE:
309 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
b8ee93ba 310 wsum)) {
28850dc7
DB
311 skb->ip_summed = CHECKSUM_UNNECESSARY;
312 break;
313 }
b8ee93ba 314
28850dc7
DB
315 NAPI_GRO_CB(skb)->flush = 1;
316 return NULL;
28850dc7
DB
317 }
318
cc5c00bb 319skip_csum:
28850dc7
DB
320 return tcp_gro_receive(head, skb);
321}
322
299603e8 323static int tcp4_gro_complete(struct sk_buff *skb, int thoff)
28850dc7
DB
324{
325 const struct iphdr *iph = ip_hdr(skb);
326 struct tcphdr *th = tcp_hdr(skb);
327
299603e8
JC
328 th->check = ~tcp_v4_check(skb->len - thoff, iph->saddr,
329 iph->daddr, 0);
c3caf119 330 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4;
28850dc7
DB
331
332 return tcp_gro_complete(skb);
333}
334
335static const struct net_offload tcpv4_offload = {
336 .callbacks = {
337 .gso_send_check = tcp_v4_gso_send_check,
28be6e07 338 .gso_segment = tcp_gso_segment,
28850dc7
DB
339 .gro_receive = tcp4_gro_receive,
340 .gro_complete = tcp4_gro_complete,
341 },
342};
343
344int __init tcpv4_offload_init(void)
345{
346 return inet_add_offload(&tcpv4_offload, IPPROTO_TCP);
347}
This page took 0.254737 seconds and 5 git commands to generate.