1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * IPV6 GSO/GRO offload support 4 * Linux INET6 implementation 5 */ 6 7 #include <linux/kernel.h> 8 #include <linux/socket.h> 9 #include <linux/netdevice.h> 10 #include <linux/skbuff.h> 11 #include <linux/printk.h> 12 13 #include <net/protocol.h> 14 #include <net/ipv6.h> 15 #include <net/inet_common.h> 16 #include <net/tcp.h> 17 #include <net/udp.h> 18 #include <net/gro.h> 19 #include <net/gso.h> 20 21 #include "ip6_offload.h" 22 #include "tcpv6_offload.c" 23 24 static int ipv6_gro_pull_exthdrs(struct sk_buff *skb, int off, int proto) 25 { 26 const struct net_offload *ops = NULL; 27 struct ipv6_opt_hdr *opth; 28 29 for (;;) { 30 int len; 31 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 opth = skb_gro_header(skb, off + sizeof(*opth), off); 41 if (unlikely(!opth)) 42 break; 43 44 len = ipv6_optlen(opth); 45 46 opth = skb_gro_header(skb, off + len, off); 47 if (unlikely(!opth)) 48 break; 49 proto = opth->nexthdr; 50 51 off += len; 52 } 53 54 skb_gro_pull(skb, off - skb_gro_receive_network_offset(skb)); 55 return proto; 56 } 57 58 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto) 59 { 60 const struct net_offload *ops = NULL; 61 62 for (;;) { 63 struct ipv6_opt_hdr *opth; 64 int len; 65 66 ops = rcu_dereference(inet6_offloads[proto]); 67 68 if (unlikely(!ops)) 69 break; 70 71 if (!(ops->flags & INET6_PROTO_GSO_EXTHDR)) 72 break; 73 74 if (unlikely(!pskb_may_pull(skb, 8))) 75 break; 76 77 opth = (void *)skb->data; 78 len = ipv6_optlen(opth); 79 80 if (unlikely(!pskb_may_pull(skb, len))) 81 break; 82 83 opth = (void *)skb->data; 84 proto = opth->nexthdr; 85 __skb_pull(skb, len); 86 } 87 88 return proto; 89 } 90 91 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb, 92 netdev_features_t features) 93 { 94 struct sk_buff *segs = ERR_PTR(-EINVAL); 95 struct ipv6hdr *ipv6h; 96 const struct net_offload *ops; 97 int proto, err; 98 struct frag_hdr *fptr; 99 unsigned int payload_len; 100 u8 *prevhdr; 101 int offset = 0; 102 bool encap, udpfrag; 103 int nhoff; 104 bool gso_partial; 105 106 skb_reset_network_header(skb); 107 err = ipv6_hopopt_jumbo_remove(skb); 108 if (err) 109 return ERR_PTR(err); 110 nhoff = skb_network_header(skb) - skb_mac_header(skb); 111 if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h)))) 112 goto out; 113 114 encap = SKB_GSO_CB(skb)->encap_level > 0; 115 if (encap) 116 features &= skb->dev->hw_enc_features; 117 SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h); 118 119 ipv6h = ipv6_hdr(skb); 120 __skb_pull(skb, sizeof(*ipv6h)); 121 segs = ERR_PTR(-EPROTONOSUPPORT); 122 123 proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr); 124 125 if (skb->encapsulation && 126 skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6)) 127 udpfrag = proto == IPPROTO_UDP && encap && 128 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 129 else 130 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation && 131 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 132 133 ops = rcu_dereference(inet6_offloads[proto]); 134 if (likely(ops && ops->callbacks.gso_segment)) { 135 if (!skb_reset_transport_header_careful(skb)) 136 goto out; 137 138 segs = ops->callbacks.gso_segment(skb, features); 139 if (!segs) 140 skb->network_header = skb_mac_header(skb) + nhoff - skb->head; 141 } 142 143 if (IS_ERR_OR_NULL(segs)) 144 goto out; 145 146 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 147 148 for (skb = segs; skb; skb = skb->next) { 149 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff); 150 if (gso_partial && skb_is_gso(skb)) 151 payload_len = skb_shinfo(skb)->gso_size + 152 SKB_GSO_CB(skb)->data_offset + 153 skb->head - (unsigned char *)(ipv6h + 1); 154 else 155 payload_len = skb->len - nhoff - sizeof(*ipv6h); 156 ipv6h->payload_len = htons(payload_len); 157 skb->network_header = (u8 *)ipv6h - skb->head; 158 skb_reset_mac_len(skb); 159 160 if (udpfrag) { 161 int err = ip6_find_1stfragopt(skb, &prevhdr); 162 if (err < 0) { 163 kfree_skb_list(segs); 164 return ERR_PTR(err); 165 } 166 fptr = (struct frag_hdr *)((u8 *)ipv6h + err); 167 fptr->frag_off = htons(offset); 168 if (skb->next) 169 fptr->frag_off |= htons(IP6_MF); 170 offset += (ntohs(ipv6h->payload_len) - 171 sizeof(struct frag_hdr)); 172 } 173 if (encap) 174 skb_reset_inner_headers(skb); 175 } 176 177 out: 178 return segs; 179 } 180 181 /* Return the total length of all the extension hdrs, following the same 182 * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs. 183 */ 184 static int ipv6_exthdrs_len(struct ipv6hdr *iph, 185 const struct net_offload **opps) 186 { 187 struct ipv6_opt_hdr *opth = (void *)iph; 188 int len = 0, proto, optlen = sizeof(*iph); 189 190 proto = iph->nexthdr; 191 for (;;) { 192 *opps = rcu_dereference(inet6_offloads[proto]); 193 if (unlikely(!(*opps))) 194 break; 195 if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR)) 196 break; 197 198 opth = (void *)opth + optlen; 199 optlen = ipv6_optlen(opth); 200 len += optlen; 201 proto = opth->nexthdr; 202 } 203 return len; 204 } 205 206 INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head, 207 struct sk_buff *skb) 208 { 209 const struct net_offload *ops; 210 struct sk_buff *pp = NULL; 211 struct sk_buff *p; 212 struct ipv6hdr *iph; 213 unsigned int nlen; 214 unsigned int hlen; 215 unsigned int off; 216 u16 flush = 1; 217 int proto; 218 219 off = skb_gro_offset(skb); 220 hlen = off + sizeof(*iph); 221 iph = skb_gro_header(skb, hlen, off); 222 if (unlikely(!iph)) 223 goto out; 224 225 NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off; 226 227 flush += ntohs(iph->payload_len) != skb->len - hlen; 228 229 proto = iph->nexthdr; 230 ops = rcu_dereference(inet6_offloads[proto]); 231 if (!ops || !ops->callbacks.gro_receive) { 232 proto = ipv6_gro_pull_exthdrs(skb, hlen, proto); 233 234 ops = rcu_dereference(inet6_offloads[proto]); 235 if (!ops || !ops->callbacks.gro_receive) 236 goto out; 237 238 iph = skb_gro_network_header(skb); 239 } else { 240 skb_gro_pull(skb, sizeof(*iph)); 241 } 242 243 skb_set_transport_header(skb, skb_gro_offset(skb)); 244 245 NAPI_GRO_CB(skb)->proto = proto; 246 247 flush--; 248 nlen = skb_gro_offset(skb) - off; 249 250 list_for_each_entry(p, head, list) { 251 const struct ipv6hdr *iph2; 252 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */ 253 254 if (!NAPI_GRO_CB(p)->same_flow) 255 continue; 256 257 iph2 = (struct ipv6hdr *)(p->data + off); 258 first_word = *(__be32 *)iph ^ *(__be32 *)iph2; 259 260 /* All fields must match except length and Traffic Class. 261 * XXX skbs on the gro_list have all been parsed and pulled 262 * already so we don't need to compare nlen 263 * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops))) 264 * memcmp() alone below is sufficient, right? 265 */ 266 if ((first_word & htonl(0xF00FFFFF)) || 267 !ipv6_addr_equal(&iph->saddr, &iph2->saddr) || 268 !ipv6_addr_equal(&iph->daddr, &iph2->daddr) || 269 iph->nexthdr != iph2->nexthdr) { 270 not_same_flow: 271 NAPI_GRO_CB(p)->same_flow = 0; 272 continue; 273 } 274 if (unlikely(nlen > sizeof(struct ipv6hdr))) { 275 if (memcmp(iph + 1, iph2 + 1, 276 nlen - sizeof(struct ipv6hdr))) 277 goto not_same_flow; 278 } 279 } 280 281 NAPI_GRO_CB(skb)->flush |= flush; 282 283 skb_gro_postpull_rcsum(skb, iph, nlen); 284 285 if (unlikely(gro_recursion_inc_test(skb))) { 286 flush = 1; 287 goto out; 288 } 289 290 if (likely(proto == IPPROTO_TCP)) 291 pp = tcp6_gro_receive(head, skb); 292 #if IS_BUILTIN(CONFIG_IPV6) 293 else if (likely(proto == IPPROTO_UDP)) 294 pp = udp6_gro_receive(head, skb); 295 #endif 296 else 297 pp = ops->callbacks.gro_receive(head, skb); 298 out: 299 skb_gro_flush_final(skb, pp, flush); 300 301 return pp; 302 } 303 304 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head, 305 struct sk_buff *skb) 306 { 307 /* Common GRO receive for SIT and IP6IP6 */ 308 309 if (NAPI_GRO_CB(skb)->encap_mark) { 310 NAPI_GRO_CB(skb)->flush = 1; 311 return NULL; 312 } 313 314 NAPI_GRO_CB(skb)->encap_mark = 1; 315 316 return ipv6_gro_receive(head, skb); 317 } 318 319 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head, 320 struct sk_buff *skb) 321 { 322 /* Common GRO receive for SIT and IP6IP6 */ 323 324 if (NAPI_GRO_CB(skb)->encap_mark) { 325 NAPI_GRO_CB(skb)->flush = 1; 326 return NULL; 327 } 328 329 NAPI_GRO_CB(skb)->encap_mark = 1; 330 331 return inet_gro_receive(head, skb); 332 } 333 334 INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff) 335 { 336 const struct net_offload *ops; 337 struct ipv6hdr *iph; 338 int err = -ENOSYS; 339 u32 payload_len; 340 341 if (skb->encapsulation) { 342 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6)); 343 skb_set_inner_network_header(skb, nhoff); 344 } 345 346 payload_len = skb->len - nhoff - sizeof(*iph); 347 if (unlikely(payload_len > IPV6_MAXPLEN)) { 348 struct hop_jumbo_hdr *hop_jumbo; 349 int hoplen = sizeof(*hop_jumbo); 350 351 /* Move network header left */ 352 memmove(skb_mac_header(skb) - hoplen, skb_mac_header(skb), 353 skb->transport_header - skb->mac_header); 354 skb->data -= hoplen; 355 skb->len += hoplen; 356 skb->mac_header -= hoplen; 357 skb->network_header -= hoplen; 358 iph = (struct ipv6hdr *)(skb->data + nhoff); 359 hop_jumbo = (struct hop_jumbo_hdr *)(iph + 1); 360 361 /* Build hop-by-hop options */ 362 hop_jumbo->nexthdr = iph->nexthdr; 363 hop_jumbo->hdrlen = 0; 364 hop_jumbo->tlv_type = IPV6_TLV_JUMBO; 365 hop_jumbo->tlv_len = 4; 366 hop_jumbo->jumbo_payload_len = htonl(payload_len + hoplen); 367 368 iph->nexthdr = NEXTHDR_HOP; 369 iph->payload_len = 0; 370 } else { 371 iph = (struct ipv6hdr *)(skb->data + nhoff); 372 iph->payload_len = htons(payload_len); 373 } 374 375 nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops); 376 377 if (likely(ops == &net_hotdata.tcpv6_offload)) 378 return tcp6_gro_complete(skb, nhoff); 379 #if IS_BUILTIN(CONFIG_IPV6) 380 if (ops == &net_hotdata.udpv6_offload) 381 return udp6_gro_complete(skb, nhoff); 382 #endif 383 384 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 385 goto out; 386 387 err = ops->callbacks.gro_complete(skb, nhoff); 388 389 out: 390 return err; 391 } 392 393 static int sit_gro_complete(struct sk_buff *skb, int nhoff) 394 { 395 skb->encapsulation = 1; 396 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 397 return ipv6_gro_complete(skb, nhoff); 398 } 399 400 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff) 401 { 402 skb->encapsulation = 1; 403 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6; 404 return ipv6_gro_complete(skb, nhoff); 405 } 406 407 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff) 408 { 409 skb->encapsulation = 1; 410 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6; 411 return inet_gro_complete(skb, nhoff); 412 } 413 414 415 static struct sk_buff *sit_gso_segment(struct sk_buff *skb, 416 netdev_features_t features) 417 { 418 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) 419 return ERR_PTR(-EINVAL); 420 421 return ipv6_gso_segment(skb, features); 422 } 423 424 static struct sk_buff *ip4ip6_gso_segment(struct sk_buff *skb, 425 netdev_features_t features) 426 { 427 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6)) 428 return ERR_PTR(-EINVAL); 429 430 return inet_gso_segment(skb, features); 431 } 432 433 static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb, 434 netdev_features_t features) 435 { 436 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6)) 437 return ERR_PTR(-EINVAL); 438 439 return ipv6_gso_segment(skb, features); 440 } 441 442 static const struct net_offload sit_offload = { 443 .callbacks = { 444 .gso_segment = sit_gso_segment, 445 .gro_receive = sit_ip6ip6_gro_receive, 446 .gro_complete = sit_gro_complete, 447 }, 448 }; 449 450 static const struct net_offload ip4ip6_offload = { 451 .callbacks = { 452 .gso_segment = ip4ip6_gso_segment, 453 .gro_receive = ip4ip6_gro_receive, 454 .gro_complete = ip4ip6_gro_complete, 455 }, 456 }; 457 458 static const struct net_offload ip6ip6_offload = { 459 .callbacks = { 460 .gso_segment = ip6ip6_gso_segment, 461 .gro_receive = sit_ip6ip6_gro_receive, 462 .gro_complete = ip6ip6_gro_complete, 463 }, 464 }; 465 static int __init ipv6_offload_init(void) 466 { 467 468 if (tcpv6_offload_init() < 0) 469 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 470 if (ipv6_exthdrs_offload_init() < 0) 471 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__); 472 473 net_hotdata.ipv6_packet_offload = (struct packet_offload) { 474 .type = cpu_to_be16(ETH_P_IPV6), 475 .callbacks = { 476 .gso_segment = ipv6_gso_segment, 477 .gro_receive = ipv6_gro_receive, 478 .gro_complete = ipv6_gro_complete, 479 }, 480 }; 481 dev_add_offload(&net_hotdata.ipv6_packet_offload); 482 483 inet_add_offload(&sit_offload, IPPROTO_IPV6); 484 inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6); 485 inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP); 486 487 return 0; 488 } 489 490 fs_initcall(ipv6_offload_init); 491