net/ipv6/ip6_offload.c

   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
  20 #include "ip6_offload.h"
  21
  22 /* All GRO functions are always builtin, except UDP over ipv6, which lays in
  23  * ipv6 module, as it depends on UDPv6 lookup function, so we need special care
  24  * when ipv6 is built as a module
  25  */
  26 #if IS_BUILTIN(CONFIG_IPV6)
  27 #define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_2(f, f2, f1, __VA_ARGS__)
  28 #else
  29 #define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_1(f, f2, __VA_ARGS__)
  30 #endif
  31
  32 #define indirect_call_gro_receive_l4(f2, f1, cb, head, skb)     \
  33 ({                                                              \
  34         unlikely(gro_recursion_inc_test(skb)) ?                 \
  35                 NAPI_GRO_CB(skb)->flush |= 1, NULL :            \
  36                 INDIRECT_CALL_L4(cb, f2, f1, head, skb);        \
  37 })
  38
  39 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
  40 {
  41         const struct net_offload *ops = NULL;
  42
  43         for (;;) {
  44                 struct ipv6_opt_hdr *opth;
  45                 int len;
  46
  47                 if (proto != NEXTHDR_HOP) {
  48                         ops = rcu_dereference(inet6_offloads[proto]);
  49
  50                         if (unlikely(!ops))
  51                                 break;
  52
  53                         if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
  54                                 break;
  55                 }
  56
  57                 if (unlikely(!pskb_may_pull(skb, 8)))
  58                         break;
  59
  60                 opth = (void *)skb->data;
  61                 len = ipv6_optlen(opth);
  62
  63                 if (unlikely(!pskb_may_pull(skb, len)))
  64                         break;
  65
  66                 opth = (void *)skb->data;
  67                 proto = opth->nexthdr;
  68                 __skb_pull(skb, len);
  69         }
  70
  71         return proto;
  72 }
  73
  74 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
  75         netdev_features_t features)
  76 {
  77         struct sk_buff *segs = ERR_PTR(-EINVAL);
  78         struct ipv6hdr *ipv6h;
  79         const struct net_offload *ops;
  80         int proto, nexthdr;
  81         struct frag_hdr *fptr;
  82         unsigned int payload_len;
  83         u8 *prevhdr;
  84         int offset = 0;
  85         bool encap, udpfrag;
  86         int nhoff;
  87         bool gso_partial;
  88
  89         skb_reset_network_header(skb);
  90         nexthdr = ipv6_has_hopopt_jumbo(skb);
  91         if (nexthdr) {
  92                 const int hophdr_len = sizeof(struct hop_jumbo_hdr);
  93                 int err;
  94
  95                 err = skb_cow_head(skb, 0);
  96                 if (err < 0)
  97                         return ERR_PTR(err);
  98
  99                 /* remove the HBH header.
 100                  * Layout: [Ethernet header][IPv6 header][HBH][TCP header]
 101                  */
 102                 memmove(skb_mac_header(skb) + hophdr_len,
 103                         skb_mac_header(skb),
 104                         ETH_HLEN + sizeof(struct ipv6hdr));
 105                 skb->data += hophdr_len;
 106                 skb->len -= hophdr_len;
 107                 skb->network_header += hophdr_len;
 108                 skb->mac_header += hophdr_len;
 109                 ipv6h = (struct ipv6hdr *)skb->data;
 110                 ipv6h->nexthdr = nexthdr;
 111         }
 112         nhoff = skb_network_header(skb) - skb_mac_header(skb);
 113         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
 114                 goto out;
 115
 116         encap = SKB_GSO_CB(skb)->encap_level > 0;
 117         if (encap)
 118                 features &= skb->dev->hw_enc_features;
 119         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
 120
 121         ipv6h = ipv6_hdr(skb);
 122         __skb_pull(skb, sizeof(*ipv6h));
 123         segs = ERR_PTR(-EPROTONOSUPPORT);
 124
 125         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
 126
 127         if (skb->encapsulation &&
 128             skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
 129                 udpfrag = proto == IPPROTO_UDP && encap &&
 130                           (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
 131         else
 132                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
 133                           (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
 134
 135         ops = rcu_dereference(inet6_offloads[proto]);
 136         if (likely(ops && ops->callbacks.gso_segment)) {
 137                 skb_reset_transport_header(skb);
 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                 if (proto != NEXTHDR_HOP) {
 193                         *opps = rcu_dereference(inet6_offloads[proto]);
 194                         if (unlikely(!(*opps)))
 195                                 break;
 196                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 197                                 break;
 198                 }
 199                 opth = (void *)opth + optlen;
 200                 optlen = ipv6_optlen(opth);
 201                 len += optlen;
 202                 proto = opth->nexthdr;
 203         }
 204         return len;
 205 }
 206
 207 INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
 208                                                          struct sk_buff *skb)
 209 {
 210         const struct net_offload *ops;
 211         struct sk_buff *pp = NULL;
 212         struct sk_buff *p;
 213         struct ipv6hdr *iph;
 214         unsigned int nlen;
 215         unsigned int hlen;
 216         unsigned int off;
 217         u16 flush = 1;
 218         int proto;
 219
 220         off = skb_gro_offset(skb);
 221         hlen = off + sizeof(*iph);
 222         iph = skb_gro_header_fast(skb, off);
 223         if (skb_gro_header_hard(skb, hlen)) {
 224                 iph = skb_gro_header_slow(skb, hlen, off);
 225                 if (unlikely(!iph))
 226                         goto out;
 227         }
 228
 229         skb_set_network_header(skb, off);
 230         skb_gro_pull(skb, sizeof(*iph));
 231         skb_set_transport_header(skb, skb_gro_offset(skb));
 232
 233         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 234
 235         proto = iph->nexthdr;
 236         ops = rcu_dereference(inet6_offloads[proto]);
 237         if (!ops || !ops->callbacks.gro_receive) {
 238                 __pskb_pull(skb, skb_gro_offset(skb));
 239                 skb_gro_frag0_invalidate(skb);
 240                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 241                 skb_gro_pull(skb, -skb_transport_offset(skb));
 242                 skb_reset_transport_header(skb);
 243                 __skb_push(skb, skb_gro_offset(skb));
 244
 245                 ops = rcu_dereference(inet6_offloads[proto]);
 246                 if (!ops || !ops->callbacks.gro_receive)
 247                         goto out;
 248
 249                 iph = ipv6_hdr(skb);
 250         }
 251
 252         NAPI_GRO_CB(skb)->proto = proto;
 253
 254         flush--;
 255         nlen = skb_network_header_len(skb);
 256
 257         list_for_each_entry(p, head, list) {
 258                 const struct ipv6hdr *iph2;
 259                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 260
 261                 if (!NAPI_GRO_CB(p)->same_flow)
 262                         continue;
 263
 264                 iph2 = (struct ipv6hdr *)(p->data + off);
 265                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 266
 267                 /* All fields must match except length and Traffic Class.
 268                  * XXX skbs on the gro_list have all been parsed and pulled
 269                  * already so we don't need to compare nlen
 270                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 271                  * memcmp() alone below is sufficient, right?
 272                  */
 273                  if ((first_word & htonl(0xF00FFFFF)) ||
 274                      !ipv6_addr_equal(&iph->saddr, &iph2->saddr) ||
 275                      !ipv6_addr_equal(&iph->daddr, &iph2->daddr) ||
 276                      iph->nexthdr != iph2->nexthdr) {
 277 not_same_flow:
 278                         NAPI_GRO_CB(p)->same_flow = 0;
 279                         continue;
 280                 }
 281                 if (unlikely(nlen > sizeof(struct ipv6hdr))) {
 282                         if (memcmp(iph + 1, iph2 + 1,
 283                                    nlen - sizeof(struct ipv6hdr)))
 284                                 goto not_same_flow;
 285                 }
 286                 /* flush if Traffic Class fields are different */
 287                 NAPI_GRO_CB(p)->flush |= !!((first_word & htonl(0x0FF00000)) |
 288                         (__force __be32)(iph->hop_limit ^ iph2->hop_limit));
 289                 NAPI_GRO_CB(p)->flush |= flush;
 290
 291                 /* If the previous IP ID value was based on an atomic
 292                  * datagram we can overwrite the value and ignore it.
 293                  */
 294                 if (NAPI_GRO_CB(skb)->is_atomic)
 295                         NAPI_GRO_CB(p)->flush_id = 0;
 296         }
 297
 298         NAPI_GRO_CB(skb)->is_atomic = true;
 299         NAPI_GRO_CB(skb)->flush |= flush;
 300
 301         skb_gro_postpull_rcsum(skb, iph, nlen);
 302
 303         pp = indirect_call_gro_receive_l4(tcp6_gro_receive, udp6_gro_receive,
 304                                          ops->callbacks.gro_receive, head, skb);
 305
 306 out:
 307         skb_gro_flush_final(skb, pp, flush);
 308
 309         return pp;
 310 }
 311
 312 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
 313                                               struct sk_buff *skb)
 314 {
 315         /* Common GRO receive for SIT and IP6IP6 */
 316
 317         if (NAPI_GRO_CB(skb)->encap_mark) {
 318                 NAPI_GRO_CB(skb)->flush = 1;
 319                 return NULL;
 320         }
 321
 322         NAPI_GRO_CB(skb)->encap_mark = 1;
 323
 324         return ipv6_gro_receive(head, skb);
 325 }
 326
 327 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
 328                                           struct sk_buff *skb)
 329 {
 330         /* Common GRO receive for SIT and IP6IP6 */
 331
 332         if (NAPI_GRO_CB(skb)->encap_mark) {
 333                 NAPI_GRO_CB(skb)->flush = 1;
 334                 return NULL;
 335         }
 336
 337         NAPI_GRO_CB(skb)->encap_mark = 1;
 338
 339         return inet_gro_receive(head, skb);
 340 }
 341
 342 INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 343 {
 344         const struct net_offload *ops;
 345         struct ipv6hdr *iph;
 346         int err = -ENOSYS;
 347         u32 payload_len;
 348
 349         if (skb->encapsulation) {
 350                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
 351                 skb_set_inner_network_header(skb, nhoff);
 352         }
 353
 354         payload_len = skb->len - nhoff - sizeof(*iph);
 355         if (unlikely(payload_len > IPV6_MAXPLEN)) {
 356                 struct hop_jumbo_hdr *hop_jumbo;
 357                 int hoplen = sizeof(*hop_jumbo);
 358
 359                 /* Move network header left */
 360                 memmove(skb_mac_header(skb) - hoplen, skb_mac_header(skb),
 361                         skb->transport_header - skb->mac_header);
 362                 skb->data -= hoplen;
 363                 skb->len += hoplen;
 364                 skb->mac_header -= hoplen;
 365                 skb->network_header -= hoplen;
 366                 iph = (struct ipv6hdr *)(skb->data + nhoff);
 367                 hop_jumbo = (struct hop_jumbo_hdr *)(iph + 1);
 368
 369                 /* Build hop-by-hop options */
 370                 hop_jumbo->nexthdr = iph->nexthdr;
 371                 hop_jumbo->hdrlen = 0;
 372                 hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
 373                 hop_jumbo->tlv_len = 4;
 374                 hop_jumbo->jumbo_payload_len = htonl(payload_len + hoplen);
 375
 376                 iph->nexthdr = NEXTHDR_HOP;
 377                 iph->payload_len = 0;
 378         } else {
 379                 iph = (struct ipv6hdr *)(skb->data + nhoff);
 380                 iph->payload_len = htons(payload_len);
 381         }
 382
 383         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 384         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 385                 goto out;
 386
 387         err = INDIRECT_CALL_L4(ops->callbacks.gro_complete, tcp6_gro_complete,
 388                                udp6_gro_complete, skb, nhoff);
 389
 390 out:
 391         return err;
 392 }
 393
 394 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 395 {
 396         skb->encapsulation = 1;
 397         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
 398         return ipv6_gro_complete(skb, nhoff);
 399 }
 400
 401 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
 402 {
 403         skb->encapsulation = 1;
 404         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 405         return ipv6_gro_complete(skb, nhoff);
 406 }
 407
 408 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
 409 {
 410         skb->encapsulation = 1;
 411         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 412         return inet_gro_complete(skb, nhoff);
 413 }
 414
 415 static struct packet_offload ipv6_packet_offload __read_mostly = {
 416         .type = cpu_to_be16(ETH_P_IPV6),
 417         .callbacks = {
 418                 .gso_segment = ipv6_gso_segment,
 419                 .gro_receive = ipv6_gro_receive,
 420                 .gro_complete = ipv6_gro_complete,
 421         },
 422 };
 423
 424 static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
 425                                        netdev_features_t features)
 426 {
 427         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
 428                 return ERR_PTR(-EINVAL);
 429
 430         return ipv6_gso_segment(skb, features);
 431 }
 432
 433 static struct sk_buff *ip4ip6_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 inet_gso_segment(skb, features);
 440 }
 441
 442 static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb,
 443                                           netdev_features_t features)
 444 {
 445         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
 446                 return ERR_PTR(-EINVAL);
 447
 448         return ipv6_gso_segment(skb, features);
 449 }
 450
 451 static const struct net_offload sit_offload = {
 452         .callbacks = {
 453                 .gso_segment    = sit_gso_segment,
 454                 .gro_receive    = sit_ip6ip6_gro_receive,
 455                 .gro_complete   = sit_gro_complete,
 456         },
 457 };
 458
 459 static const struct net_offload ip4ip6_offload = {
 460         .callbacks = {
 461                 .gso_segment    = ip4ip6_gso_segment,
 462                 .gro_receive    = ip4ip6_gro_receive,
 463                 .gro_complete   = ip4ip6_gro_complete,
 464         },
 465 };
 466
 467 static const struct net_offload ip6ip6_offload = {
 468         .callbacks = {
 469                 .gso_segment    = ip6ip6_gso_segment,
 470                 .gro_receive    = sit_ip6ip6_gro_receive,
 471                 .gro_complete   = ip6ip6_gro_complete,
 472         },
 473 };
 474 static int __init ipv6_offload_init(void)
 475 {
 476
 477         if (tcpv6_offload_init() < 0)
 478                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 479         if (ipv6_exthdrs_offload_init() < 0)
 480                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 481
 482         dev_add_offload(&ipv6_packet_offload);
 483
 484         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 485         inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
 486         inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
 487
 488         return 0;
 489 }
 490
 491 fs_initcall(ipv6_offload_init);