1 /* 2 * IPV6 GSO/GRO offload support 3 * Linux INET6 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 11 #include <linux/kernel.h> 12 #include <linux/socket.h> 13 #include <linux/netdevice.h> 14 #include <linux/skbuff.h> 15 #include <linux/printk.h> 16 17 #include <net/protocol.h> 18 #include <net/ipv6.h> 19 #include <net/inet_common.h> 20 21 #include "ip6_offload.h" 22 23 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto) 24 { 25 const struct net_offload *ops = NULL; 26 27 for (;;) { 28 struct ipv6_opt_hdr *opth; 29 int len; 30 31 if (proto != NEXTHDR_HOP) { 32 ops = rcu_dereference(inet6_offloads[proto]); 33 34 if (unlikely(!ops)) 35 break; 36 37 if (!(ops->flags & INET6_PROTO_GSO_EXTHDR)) 38 break; 39 } 40 41 if (unlikely(!pskb_may_pull(skb, 8))) 42 break; 43 44 opth = (void *)skb->data; 45 len = ipv6_optlen(opth); 46 47 if (unlikely(!pskb_may_pull(skb, len))) 48 break; 49 50 opth = (void *)skb->data; 51 proto = opth->nexthdr; 52 __skb_pull(skb, len); 53 } 54 55 return proto; 56 } 57 58 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb, 59 netdev_features_t features) 60 { 61 struct sk_buff *segs = ERR_PTR(-EINVAL); 62 struct ipv6hdr *ipv6h; 63 const struct net_offload *ops; 64 int proto; 65 struct frag_hdr *fptr; 66 unsigned int payload_len; 67 u8 *prevhdr; 68 int offset = 0; 69 bool encap, udpfrag; 70 int nhoff; 71 bool gso_partial; 72 73 skb_reset_network_header(skb); 74 nhoff = skb_network_header(skb) - skb_mac_header(skb); 75 if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h)))) 76 goto out; 77 78 encap = SKB_GSO_CB(skb)->encap_level > 0; 79 if (encap) 80 features &= skb->dev->hw_enc_features; 81 SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h); 82 83 ipv6h = ipv6_hdr(skb); 84 __skb_pull(skb, sizeof(*ipv6h)); 85 segs = ERR_PTR(-EPROTONOSUPPORT); 86 87 proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr); 88 89 if (skb->encapsulation && 90 skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6)) 91 udpfrag = proto == IPPROTO_UDP && encap && 92 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 93 else 94 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation && 95 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 96 97 ops = rcu_dereference(inet6_offloads[proto]); 98 if (likely(ops && ops->callbacks.gso_segment)) { 99 skb_reset_transport_header(skb); 100 segs = ops->callbacks.gso_segment(skb, features); 101 } 102 103 if (IS_ERR_OR_NULL(segs)) 104 goto out; 105 106 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 107 108 for (skb = segs; skb; skb = skb->next) { 109 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff); 110 if (gso_partial && skb_is_gso(skb)) 111 payload_len = skb_shinfo(skb)->gso_size + 112 SKB_GSO_CB(skb)->data_offset + 113 skb->head - (unsigned char *)(ipv6h + 1); 114 else 115 payload_len = skb->len - nhoff - sizeof(*ipv6h); 116 ipv6h->payload_len = htons(payload_len); 117 skb->network_header = (u8 *)ipv6h - skb->head; 118 skb_reset_mac_len(skb); 119 120 if (udpfrag) { 121 int err = ip6_find_1stfragopt(skb, &prevhdr); 122 if (err < 0) { 123 kfree_skb_list(segs); 124 return ERR_PTR(err); 125 } 126 fptr = (struct frag_hdr *)((u8 *)ipv6h + err); 127 fptr->frag_off = htons(offset); 128 if (skb->next) 129 fptr->frag_off |= htons(IP6_MF); 130 offset += (ntohs(ipv6h->payload_len) - 131 sizeof(struct frag_hdr)); 132 } 133 if (encap) 134 skb_reset_inner_headers(skb); 135 } 136 137 out: 138 return segs; 139 } 140 141 /* Return the total length of all the extension hdrs, following the same 142 * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs. 143 */ 144 static int ipv6_exthdrs_len(struct ipv6hdr *iph, 145 const struct net_offload **opps) 146 { 147 struct ipv6_opt_hdr *opth = (void *)iph; 148 int len = 0, proto, optlen = sizeof(*iph); 149 150 proto = iph->nexthdr; 151 for (;;) { 152 if (proto != NEXTHDR_HOP) { 153 *opps = rcu_dereference(inet6_offloads[proto]); 154 if (unlikely(!(*opps))) 155 break; 156 if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR)) 157 break; 158 } 159 opth = (void *)opth + optlen; 160 optlen = ipv6_optlen(opth); 161 len += optlen; 162 proto = opth->nexthdr; 163 } 164 return len; 165 } 166 167 static struct sk_buff *ipv6_gro_receive(struct list_head *head, 168 struct sk_buff *skb) 169 { 170 const struct net_offload *ops; 171 struct sk_buff *pp = NULL; 172 struct sk_buff *p; 173 struct ipv6hdr *iph; 174 unsigned int nlen; 175 unsigned int hlen; 176 unsigned int off; 177 u16 flush = 1; 178 int proto; 179 180 off = skb_gro_offset(skb); 181 hlen = off + sizeof(*iph); 182 iph = skb_gro_header_fast(skb, off); 183 if (skb_gro_header_hard(skb, hlen)) { 184 iph = skb_gro_header_slow(skb, hlen, off); 185 if (unlikely(!iph)) 186 goto out; 187 } 188 189 skb_set_network_header(skb, off); 190 skb_gro_pull(skb, sizeof(*iph)); 191 skb_set_transport_header(skb, skb_gro_offset(skb)); 192 193 flush += ntohs(iph->payload_len) != skb_gro_len(skb); 194 195 rcu_read_lock(); 196 proto = iph->nexthdr; 197 ops = rcu_dereference(inet6_offloads[proto]); 198 if (!ops || !ops->callbacks.gro_receive) { 199 __pskb_pull(skb, skb_gro_offset(skb)); 200 skb_gro_frag0_invalidate(skb); 201 proto = ipv6_gso_pull_exthdrs(skb, proto); 202 skb_gro_pull(skb, -skb_transport_offset(skb)); 203 skb_reset_transport_header(skb); 204 __skb_push(skb, skb_gro_offset(skb)); 205 206 ops = rcu_dereference(inet6_offloads[proto]); 207 if (!ops || !ops->callbacks.gro_receive) 208 goto out_unlock; 209 210 iph = ipv6_hdr(skb); 211 } 212 213 NAPI_GRO_CB(skb)->proto = proto; 214 215 flush--; 216 nlen = skb_network_header_len(skb); 217 218 list_for_each_entry(p, head, list) { 219 const struct ipv6hdr *iph2; 220 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */ 221 222 if (!NAPI_GRO_CB(p)->same_flow) 223 continue; 224 225 iph2 = (struct ipv6hdr *)(p->data + off); 226 first_word = *(__be32 *)iph ^ *(__be32 *)iph2; 227 228 /* All fields must match except length and Traffic Class. 229 * XXX skbs on the gro_list have all been parsed and pulled 230 * already so we don't need to compare nlen 231 * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops))) 232 * memcmp() alone below is suffcient, right? 233 */ 234 if ((first_word & htonl(0xF00FFFFF)) || 235 memcmp(&iph->nexthdr, &iph2->nexthdr, 236 nlen - offsetof(struct ipv6hdr, nexthdr))) { 237 NAPI_GRO_CB(p)->same_flow = 0; 238 continue; 239 } 240 /* flush if Traffic Class fields are different */ 241 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000)); 242 NAPI_GRO_CB(p)->flush |= flush; 243 244 /* If the previous IP ID value was based on an atomic 245 * datagram we can overwrite the value and ignore it. 246 */ 247 if (NAPI_GRO_CB(skb)->is_atomic) 248 NAPI_GRO_CB(p)->flush_id = 0; 249 } 250 251 NAPI_GRO_CB(skb)->is_atomic = true; 252 NAPI_GRO_CB(skb)->flush |= flush; 253 254 skb_gro_postpull_rcsum(skb, iph, nlen); 255 256 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb); 257 258 out_unlock: 259 rcu_read_unlock(); 260 261 out: 262 skb_gro_flush_final(skb, pp, flush); 263 264 return pp; 265 } 266 267 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head, 268 struct sk_buff *skb) 269 { 270 /* Common GRO receive for SIT and IP6IP6 */ 271 272 if (NAPI_GRO_CB(skb)->encap_mark) { 273 NAPI_GRO_CB(skb)->flush = 1; 274 return NULL; 275 } 276 277 NAPI_GRO_CB(skb)->encap_mark = 1; 278 279 return ipv6_gro_receive(head, skb); 280 } 281 282 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head, 283 struct sk_buff *skb) 284 { 285 /* Common GRO receive for SIT and IP6IP6 */ 286 287 if (NAPI_GRO_CB(skb)->encap_mark) { 288 NAPI_GRO_CB(skb)->flush = 1; 289 return NULL; 290 } 291 292 NAPI_GRO_CB(skb)->encap_mark = 1; 293 294 return inet_gro_receive(head, skb); 295 } 296 297 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff) 298 { 299 const struct net_offload *ops; 300 struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff); 301 int err = -ENOSYS; 302 303 if (skb->encapsulation) { 304 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6)); 305 skb_set_inner_network_header(skb, nhoff); 306 } 307 308 iph->payload_len = htons(skb->len - nhoff - sizeof(*iph)); 309 310 rcu_read_lock(); 311 312 nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops); 313 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 314 goto out_unlock; 315 316 err = ops->callbacks.gro_complete(skb, nhoff); 317 318 out_unlock: 319 rcu_read_unlock(); 320 321 return err; 322 } 323 324 static int sit_gro_complete(struct sk_buff *skb, int nhoff) 325 { 326 skb->encapsulation = 1; 327 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 328 return ipv6_gro_complete(skb, nhoff); 329 } 330 331 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff) 332 { 333 skb->encapsulation = 1; 334 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6; 335 return ipv6_gro_complete(skb, nhoff); 336 } 337 338 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff) 339 { 340 skb->encapsulation = 1; 341 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6; 342 return inet_gro_complete(skb, nhoff); 343 } 344 345 static struct packet_offload ipv6_packet_offload __read_mostly = { 346 .type = cpu_to_be16(ETH_P_IPV6), 347 .callbacks = { 348 .gso_segment = ipv6_gso_segment, 349 .gro_receive = ipv6_gro_receive, 350 .gro_complete = ipv6_gro_complete, 351 }, 352 }; 353 354 static const struct net_offload sit_offload = { 355 .callbacks = { 356 .gso_segment = ipv6_gso_segment, 357 .gro_receive = sit_ip6ip6_gro_receive, 358 .gro_complete = sit_gro_complete, 359 }, 360 }; 361 362 static const struct net_offload ip4ip6_offload = { 363 .callbacks = { 364 .gso_segment = inet_gso_segment, 365 .gro_receive = ip4ip6_gro_receive, 366 .gro_complete = ip4ip6_gro_complete, 367 }, 368 }; 369 370 static const struct net_offload ip6ip6_offload = { 371 .callbacks = { 372 .gso_segment = ipv6_gso_segment, 373 .gro_receive = sit_ip6ip6_gro_receive, 374 .gro_complete = ip6ip6_gro_complete, 375 }, 376 }; 377 static int __init ipv6_offload_init(void) 378 { 379 380 if (tcpv6_offload_init() < 0) 381 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 382 if (ipv6_exthdrs_offload_init() < 0) 383 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__); 384 385 dev_add_offload(&ipv6_packet_offload); 386 387 inet_add_offload(&sit_offload, IPPROTO_IPV6); 388 inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6); 389 inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP); 390 391 return 0; 392 } 393 394 fs_initcall(ipv6_offload_init); 395