tcp: annotate sk->sk_wmem_queued lockless reads
[linux-2.6-block.git] / net / ipv6 / ip6_offload.c
CommitLineData
2874c5fd 1// SPDX-License-Identifier: GPL-2.0-or-later
d1da932e
VY
2/*
3 * IPV6 GSO/GRO offload support
4 * Linux INET6 implementation
d1da932e
VY
5 */
6
7#include <linux/kernel.h>
8#include <linux/socket.h>
9#include <linux/netdevice.h>
10#include <linux/skbuff.h>
c6b641a4 11#include <linux/printk.h>
d1da932e
VY
12
13#include <net/protocol.h>
14#include <net/ipv6.h>
b8921ca8 15#include <net/inet_common.h>
d1da932e
VY
16
17#include "ip6_offload.h"
18
028e0a47
PA
19/* All GRO functions are always builtin, except UDP over ipv6, which lays in
20 * ipv6 module, as it depends on UDPv6 lookup function, so we need special care
21 * when ipv6 is built as a module
22 */
23#if IS_BUILTIN(CONFIG_IPV6)
24#define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_2(f, f2, f1, __VA_ARGS__)
25#else
26#define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_1(f, f2, __VA_ARGS__)
27#endif
28
29#define indirect_call_gro_receive_l4(f2, f1, cb, head, skb) \
30({ \
31 unlikely(gro_recursion_inc_test(skb)) ? \
32 NAPI_GRO_CB(skb)->flush |= 1, NULL : \
33 INDIRECT_CALL_L4(cb, f2, f1, head, skb); \
34})
35
d1da932e
VY
36static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
37{
38 const struct net_offload *ops = NULL;
39
40 for (;;) {
41 struct ipv6_opt_hdr *opth;
42 int len;
43
44 if (proto != NEXTHDR_HOP) {
45 ops = rcu_dereference(inet6_offloads[proto]);
46
47 if (unlikely(!ops))
48 break;
49
50 if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
51 break;
52 }
53
54 if (unlikely(!pskb_may_pull(skb, 8)))
55 break;
56
57 opth = (void *)skb->data;
58 len = ipv6_optlen(opth);
59
60 if (unlikely(!pskb_may_pull(skb, len)))
61 break;
62
fc6fb41c 63 opth = (void *)skb->data;
d1da932e
VY
64 proto = opth->nexthdr;
65 __skb_pull(skb, len);
66 }
67
68 return proto;
69}
70
d1da932e
VY
71static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
72 netdev_features_t features)
73{
74 struct sk_buff *segs = ERR_PTR(-EINVAL);
75 struct ipv6hdr *ipv6h;
76 const struct net_offload *ops;
77 int proto;
78 struct frag_hdr *fptr;
802ab55a 79 unsigned int payload_len;
d1da932e
VY
80 u8 *prevhdr;
81 int offset = 0;
91a48a2e 82 bool encap, udpfrag;
d3e5e006 83 int nhoff;
07b26c94 84 bool gso_partial;
d1da932e 85
d3e5e006
ED
86 skb_reset_network_header(skb);
87 nhoff = skb_network_header(skb) - skb_mac_header(skb);
d1da932e
VY
88 if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
89 goto out;
90
91a48a2e
HFS
91 encap = SKB_GSO_CB(skb)->encap_level > 0;
92 if (encap)
1e16aa3d 93 features &= skb->dev->hw_enc_features;
d3e5e006
ED
94 SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
95
d1da932e
VY
96 ipv6h = ipv6_hdr(skb);
97 __skb_pull(skb, sizeof(*ipv6h));
98 segs = ERR_PTR(-EPROTONOSUPPORT);
99
100 proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
b917eb15 101
91a48a2e 102 if (skb->encapsulation &&
7e13318d 103 skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
ee80d1eb
WB
104 udpfrag = proto == IPPROTO_UDP && encap &&
105 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
91a48a2e 106 else
ee80d1eb
WB
107 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
108 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
91a48a2e 109
d1da932e 110 ops = rcu_dereference(inet6_offloads[proto]);
f191a1d1 111 if (likely(ops && ops->callbacks.gso_segment)) {
d1da932e 112 skb_reset_transport_header(skb);
f191a1d1 113 segs = ops->callbacks.gso_segment(skb, features);
d1da932e 114 }
d1da932e 115
6b6ebb6b 116 if (IS_ERR_OR_NULL(segs))
d1da932e
VY
117 goto out;
118
07b26c94
SK
119 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
120
d1da932e 121 for (skb = segs; skb; skb = skb->next) {
d3e5e006 122 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
3d0241d5 123 if (gso_partial && skb_is_gso(skb))
802ab55a
AD
124 payload_len = skb_shinfo(skb)->gso_size +
125 SKB_GSO_CB(skb)->data_offset +
126 skb->head - (unsigned char *)(ipv6h + 1);
127 else
128 payload_len = skb->len - nhoff - sizeof(*ipv6h);
129 ipv6h->payload_len = htons(payload_len);
d3e5e006 130 skb->network_header = (u8 *)ipv6h - skb->head;
c56cae23 131 skb_reset_mac_len(skb);
d3e5e006 132
91a48a2e 133 if (udpfrag) {
7dd7eb95 134 int err = ip6_find_1stfragopt(skb, &prevhdr);
e3e86b51
DM
135 if (err < 0) {
136 kfree_skb_list(segs);
7dd7eb95 137 return ERR_PTR(err);
e3e86b51 138 }
7dd7eb95 139 fptr = (struct frag_hdr *)((u8 *)ipv6h + err);
d1da932e 140 fptr->frag_off = htons(offset);
53b24b8f 141 if (skb->next)
d1da932e
VY
142 fptr->frag_off |= htons(IP6_MF);
143 offset += (ntohs(ipv6h->payload_len) -
144 sizeof(struct frag_hdr));
145 }
91a48a2e
HFS
146 if (encap)
147 skb_reset_inner_headers(skb);
d1da932e
VY
148 }
149
150out:
151 return segs;
152}
153
299603e8
JC
154/* Return the total length of all the extension hdrs, following the same
155 * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
156 */
157static int ipv6_exthdrs_len(struct ipv6hdr *iph,
158 const struct net_offload **opps)
159{
810c23a3
JC
160 struct ipv6_opt_hdr *opth = (void *)iph;
161 int len = 0, proto, optlen = sizeof(*iph);
299603e8
JC
162
163 proto = iph->nexthdr;
164 for (;;) {
165 if (proto != NEXTHDR_HOP) {
166 *opps = rcu_dereference(inet6_offloads[proto]);
167 if (unlikely(!(*opps)))
168 break;
169 if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
170 break;
171 }
810c23a3
JC
172 opth = (void *)opth + optlen;
173 optlen = ipv6_optlen(opth);
299603e8
JC
174 len += optlen;
175 proto = opth->nexthdr;
176 }
177 return len;
178}
179
028e0a47
PA
180INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *,
181 struct sk_buff *));
182INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp6_gro_receive(struct list_head *,
183 struct sk_buff *));
aaa5d90b
PA
184INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
185 struct sk_buff *skb)
d1da932e
VY
186{
187 const struct net_offload *ops;
d4546c25 188 struct sk_buff *pp = NULL;
d1da932e
VY
189 struct sk_buff *p;
190 struct ipv6hdr *iph;
191 unsigned int nlen;
192 unsigned int hlen;
193 unsigned int off;
bf5a755f 194 u16 flush = 1;
d1da932e 195 int proto;
d1da932e
VY
196
197 off = skb_gro_offset(skb);
198 hlen = off + sizeof(*iph);
199 iph = skb_gro_header_fast(skb, off);
200 if (skb_gro_header_hard(skb, hlen)) {
201 iph = skb_gro_header_slow(skb, hlen, off);
202 if (unlikely(!iph))
203 goto out;
204 }
205
299603e8 206 skb_set_network_header(skb, off);
d1da932e
VY
207 skb_gro_pull(skb, sizeof(*iph));
208 skb_set_transport_header(skb, skb_gro_offset(skb));
209
210 flush += ntohs(iph->payload_len) != skb_gro_len(skb);
211
212 rcu_read_lock();
213 proto = iph->nexthdr;
214 ops = rcu_dereference(inet6_offloads[proto]);
f191a1d1 215 if (!ops || !ops->callbacks.gro_receive) {
d1da932e 216 __pskb_pull(skb, skb_gro_offset(skb));
57ea52a8 217 skb_gro_frag0_invalidate(skb);
d1da932e
VY
218 proto = ipv6_gso_pull_exthdrs(skb, proto);
219 skb_gro_pull(skb, -skb_transport_offset(skb));
220 skb_reset_transport_header(skb);
221 __skb_push(skb, skb_gro_offset(skb));
222
223 ops = rcu_dereference(inet6_offloads[proto]);
f191a1d1 224 if (!ops || !ops->callbacks.gro_receive)
d1da932e
VY
225 goto out_unlock;
226
227 iph = ipv6_hdr(skb);
228 }
229
230 NAPI_GRO_CB(skb)->proto = proto;
231
232 flush--;
233 nlen = skb_network_header_len(skb);
234
d4546c25 235 list_for_each_entry(p, head, list) {
d1da932e
VY
236 const struct ipv6hdr *iph2;
237 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
238
239 if (!NAPI_GRO_CB(p)->same_flow)
240 continue;
241
299603e8 242 iph2 = (struct ipv6hdr *)(p->data + off);
67ba4152 243 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
d1da932e 244
299603e8
JC
245 /* All fields must match except length and Traffic Class.
246 * XXX skbs on the gro_list have all been parsed and pulled
247 * already so we don't need to compare nlen
248 * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
0b215b97 249 * memcmp() alone below is sufficient, right?
299603e8
JC
250 */
251 if ((first_word & htonl(0xF00FFFFF)) ||
0b215b97
ED
252 !ipv6_addr_equal(&iph->saddr, &iph2->saddr) ||
253 !ipv6_addr_equal(&iph->daddr, &iph2->daddr) ||
254 *(u16 *)&iph->nexthdr != *(u16 *)&iph2->nexthdr) {
255not_same_flow:
d1da932e
VY
256 NAPI_GRO_CB(p)->same_flow = 0;
257 continue;
258 }
0b215b97
ED
259 if (unlikely(nlen > sizeof(struct ipv6hdr))) {
260 if (memcmp(iph + 1, iph2 + 1,
261 nlen - sizeof(struct ipv6hdr)))
262 goto not_same_flow;
263 }
d1da932e
VY
264 /* flush if Traffic Class fields are different */
265 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
266 NAPI_GRO_CB(p)->flush |= flush;
03d56daa 267
1530545e
AD
268 /* If the previous IP ID value was based on an atomic
269 * datagram we can overwrite the value and ignore it.
270 */
271 if (NAPI_GRO_CB(skb)->is_atomic)
272 NAPI_GRO_CB(p)->flush_id = 0;
d1da932e
VY
273 }
274
1530545e 275 NAPI_GRO_CB(skb)->is_atomic = true;
d1da932e
VY
276 NAPI_GRO_CB(skb)->flush |= flush;
277
4de462ab 278 skb_gro_postpull_rcsum(skb, iph, nlen);
d1da932e 279
028e0a47
PA
280 pp = indirect_call_gro_receive_l4(tcp6_gro_receive, udp6_gro_receive,
281 ops->callbacks.gro_receive, head, skb);
d1da932e 282
d1da932e
VY
283out_unlock:
284 rcu_read_unlock();
285
286out:
5f114163 287 skb_gro_flush_final(skb, pp, flush);
d1da932e
VY
288
289 return pp;
290}
291
d4546c25
DM
292static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
293 struct sk_buff *skb)
fac8e0f5 294{
815d22e5
TH
295 /* Common GRO receive for SIT and IP6IP6 */
296
fac8e0f5
JG
297 if (NAPI_GRO_CB(skb)->encap_mark) {
298 NAPI_GRO_CB(skb)->flush = 1;
299 return NULL;
300 }
301
302 NAPI_GRO_CB(skb)->encap_mark = 1;
303
304 return ipv6_gro_receive(head, skb);
305}
306
d4546c25
DM
307static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
308 struct sk_buff *skb)
b8921ca8
TH
309{
310 /* Common GRO receive for SIT and IP6IP6 */
311
312 if (NAPI_GRO_CB(skb)->encap_mark) {
313 NAPI_GRO_CB(skb)->flush = 1;
314 return NULL;
315 }
316
317 NAPI_GRO_CB(skb)->encap_mark = 1;
318
319 return inet_gro_receive(head, skb);
320}
321
028e0a47
PA
322INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *, int));
323INDIRECT_CALLABLE_DECLARE(int udp6_gro_complete(struct sk_buff *, int));
aaa5d90b 324INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
d1da932e
VY
325{
326 const struct net_offload *ops;
299603e8 327 struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
d1da932e
VY
328 int err = -ENOSYS;
329
294acf1c
PA
330 if (skb->encapsulation) {
331 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
feec0cb3 332 skb_set_inner_network_header(skb, nhoff);
294acf1c 333 }
feec0cb3 334
299603e8 335 iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
d1da932e
VY
336
337 rcu_read_lock();
299603e8
JC
338
339 nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
f191a1d1 340 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
d1da932e
VY
341 goto out_unlock;
342
028e0a47
PA
343 err = INDIRECT_CALL_L4(ops->callbacks.gro_complete, tcp6_gro_complete,
344 udp6_gro_complete, skb, nhoff);
d1da932e
VY
345
346out_unlock:
347 rcu_read_unlock();
348
349 return err;
350}
351
feec0cb3
ED
352static int sit_gro_complete(struct sk_buff *skb, int nhoff)
353{
354 skb->encapsulation = 1;
7e13318d 355 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
feec0cb3
ED
356 return ipv6_gro_complete(skb, nhoff);
357}
358
815d22e5
TH
359static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
360{
361 skb->encapsulation = 1;
362 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
363 return ipv6_gro_complete(skb, nhoff);
364}
365
b8921ca8
TH
366static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
367{
368 skb->encapsulation = 1;
369 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
370 return inet_gro_complete(skb, nhoff);
371}
372
d1da932e
VY
373static struct packet_offload ipv6_packet_offload __read_mostly = {
374 .type = cpu_to_be16(ETH_P_IPV6),
f191a1d1 375 .callbacks = {
f191a1d1
VY
376 .gso_segment = ipv6_gso_segment,
377 .gro_receive = ipv6_gro_receive,
378 .gro_complete = ipv6_gro_complete,
379 },
d1da932e
VY
380};
381
418e897e
WB
382static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
383 netdev_features_t features)
384{
385 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
386 return ERR_PTR(-EINVAL);
387
388 return ipv6_gso_segment(skb, features);
389}
390
391static struct sk_buff *ip4ip6_gso_segment(struct sk_buff *skb,
392 netdev_features_t features)
393{
394 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
395 return ERR_PTR(-EINVAL);
396
397 return inet_gso_segment(skb, features);
398}
399
400static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb,
401 netdev_features_t features)
402{
403 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
404 return ERR_PTR(-EINVAL);
405
406 return ipv6_gso_segment(skb, features);
407}
408
61c1db7f
ED
409static const struct net_offload sit_offload = {
410 .callbacks = {
418e897e 411 .gso_segment = sit_gso_segment,
815d22e5 412 .gro_receive = sit_ip6ip6_gro_receive,
feec0cb3 413 .gro_complete = sit_gro_complete,
61c1db7f
ED
414 },
415};
416
b8921ca8
TH
417static const struct net_offload ip4ip6_offload = {
418 .callbacks = {
418e897e 419 .gso_segment = ip4ip6_gso_segment,
b8921ca8
TH
420 .gro_receive = ip4ip6_gro_receive,
421 .gro_complete = ip4ip6_gro_complete,
422 },
423};
424
815d22e5
TH
425static const struct net_offload ip6ip6_offload = {
426 .callbacks = {
418e897e 427 .gso_segment = ip6ip6_gso_segment,
815d22e5
TH
428 .gro_receive = sit_ip6ip6_gro_receive,
429 .gro_complete = ip6ip6_gro_complete,
430 },
431};
c6b641a4 432static int __init ipv6_offload_init(void)
d1da932e 433{
c6b641a4
VY
434
435 if (tcpv6_offload_init() < 0)
436 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
c6b641a4
VY
437 if (ipv6_exthdrs_offload_init() < 0)
438 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
439
d1da932e 440 dev_add_offload(&ipv6_packet_offload);
61c1db7f
ED
441
442 inet_add_offload(&sit_offload, IPPROTO_IPV6);
815d22e5 443 inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
b8921ca8 444 inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
61c1db7f 445
c6b641a4 446 return 0;
d1da932e
VY
447}
448
c6b641a4 449fs_initcall(ipv6_offload_init);