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, err;
  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         err = ipv6_hopopt_jumbo_remove(skb);
  91         if (err)
  92                 return ERR_PTR(err);
  93         nhoff = skb_network_header(skb) - skb_mac_header(skb);
  94         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  95                 goto out;
  96
  97         encap = SKB_GSO_CB(skb)->encap_level > 0;
  98         if (encap)
  99                 features &= skb->dev->hw_enc_features;
 100         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
 101
 102         ipv6h = ipv6_hdr(skb);
 103         __skb_pull(skb, sizeof(*ipv6h));
 104         segs = ERR_PTR(-EPROTONOSUPPORT);
 105
 106         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
 107
 108         if (skb->encapsulation &&
 109             skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
 110                 udpfrag = proto == IPPROTO_UDP && encap &&
 111                           (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
 112         else
 113                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
 114                           (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
 115
 116         ops = rcu_dereference(inet6_offloads[proto]);
 117         if (likely(ops && ops->callbacks.gso_segment)) {
 118                 skb_reset_transport_header(skb);
 119                 segs = ops->callbacks.gso_segment(skb, features);
 120                 if (!segs)
 121                         skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
 122         }
 123
 124         if (IS_ERR_OR_NULL(segs))
 125                 goto out;
 126
 127         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
 128
 129         for (skb = segs; skb; skb = skb->next) {
 130                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 131                 if (gso_partial && skb_is_gso(skb))
 132                         payload_len = skb_shinfo(skb)->gso_size +
 133                                       SKB_GSO_CB(skb)->data_offset +
 134                                       skb->head - (unsigned char *)(ipv6h + 1);
 135                 else
 136                         payload_len = skb->len - nhoff - sizeof(*ipv6h);
 137                 ipv6h->payload_len = htons(payload_len);
 138                 skb->network_header = (u8 *)ipv6h - skb->head;
 139                 skb_reset_mac_len(skb);
 140
 141                 if (udpfrag) {
 142                         int err = ip6_find_1stfragopt(skb, &prevhdr);
 143                         if (err < 0) {
 144                                 kfree_skb_list(segs);
 145                                 return ERR_PTR(err);
 146                         }
 147                         fptr = (struct frag_hdr *)((u8 *)ipv6h + err);
 148                         fptr->frag_off = htons(offset);
 149                         if (skb->next)
 150                                 fptr->frag_off |= htons(IP6_MF);
 151                         offset += (ntohs(ipv6h->payload_len) -
 152                                    sizeof(struct frag_hdr));
 153                 }
 154                 if (encap)
 155                         skb_reset_inner_headers(skb);
 156         }
 157
 158 out:
 159         return segs;
 160 }
 161
 162 /* Return the total length of all the extension hdrs, following the same
 163  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 164  */
 165 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 166                             const struct net_offload **opps)
 167 {
 168         struct ipv6_opt_hdr *opth = (void *)iph;
 169         int len = 0, proto, optlen = sizeof(*iph);
 170
 171         proto = iph->nexthdr;
 172         for (;;) {
 173                 if (proto != NEXTHDR_HOP) {
 174                         *opps = rcu_dereference(inet6_offloads[proto]);
 175                         if (unlikely(!(*opps)))
 176                                 break;
 177                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 178                                 break;
 179                 }
 180                 opth = (void *)opth + optlen;
 181                 optlen = ipv6_optlen(opth);
 182                 len += optlen;
 183                 proto = opth->nexthdr;
 184         }
 185         return len;
 186 }
 187
 188 INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
 189                                                          struct sk_buff *skb)
 190 {
 191         const struct net_offload *ops;
 192         struct sk_buff *pp = NULL;
 193         struct sk_buff *p;
 194         struct ipv6hdr *iph;
 195         unsigned int nlen;
 196         unsigned int hlen;
 197         unsigned int off;
 198         u16 flush = 1;
 199         int proto;
 200
 201         off = skb_gro_offset(skb);
 202         hlen = off + sizeof(*iph);
 203         iph = skb_gro_header(skb, hlen, off);
 204         if (unlikely(!iph))
 205                 goto out;
 206
 207         skb_set_network_header(skb, off);
 208         skb_gro_pull(skb, sizeof(*iph));
 209         skb_set_transport_header(skb, skb_gro_offset(skb));
 210
 211         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 212
 213         proto = iph->nexthdr;
 214         ops = rcu_dereference(inet6_offloads[proto]);
 215         if (!ops || !ops->callbacks.gro_receive) {
 216                 pskb_pull(skb, skb_gro_offset(skb));
 217                 skb_gro_frag0_invalidate(skb);
 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]);
 224                 if (!ops || !ops->callbacks.gro_receive)
 225                         goto out;
 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
 235         list_for_each_entry(p, head, list) {
 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
 242                 iph2 = (struct ipv6hdr *)(p->data + off);
 243                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 244
 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)))
 249                  * memcmp() alone below is sufficient, right?
 250                  */
 251                  if ((first_word & htonl(0xF00FFFFF)) ||
 252                      !ipv6_addr_equal(&iph->saddr, &iph2->saddr) ||
 253                      !ipv6_addr_equal(&iph->daddr, &iph2->daddr) ||
 254                      iph->nexthdr != iph2->nexthdr) {
 255 not_same_flow:
 256                         NAPI_GRO_CB(p)->same_flow = 0;
 257                         continue;
 258                 }
 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                 }
 264                 /* flush if Traffic Class fields are different */
 265                 NAPI_GRO_CB(p)->flush |= !!((first_word & htonl(0x0FF00000)) |
 266                         (__force __be32)(iph->hop_limit ^ iph2->hop_limit));
 267                 NAPI_GRO_CB(p)->flush |= flush;
 268
 269                 /* If the previous IP ID value was based on an atomic
 270                  * datagram we can overwrite the value and ignore it.
 271                  */
 272                 if (NAPI_GRO_CB(skb)->is_atomic)
 273                         NAPI_GRO_CB(p)->flush_id = 0;
 274         }
 275
 276         NAPI_GRO_CB(skb)->is_atomic = true;
 277         NAPI_GRO_CB(skb)->flush |= flush;
 278
 279         skb_gro_postpull_rcsum(skb, iph, nlen);
 280
 281         pp = indirect_call_gro_receive_l4(tcp6_gro_receive, udp6_gro_receive,
 282                                          ops->callbacks.gro_receive, head, skb);
 283
 284 out:
 285         skb_gro_flush_final(skb, pp, flush);
 286
 287         return pp;
 288 }
 289
 290 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
 291                                               struct sk_buff *skb)
 292 {
 293         /* Common GRO receive for SIT and IP6IP6 */
 294
 295         if (NAPI_GRO_CB(skb)->encap_mark) {
 296                 NAPI_GRO_CB(skb)->flush = 1;
 297                 return NULL;
 298         }
 299
 300         NAPI_GRO_CB(skb)->encap_mark = 1;
 301
 302         return ipv6_gro_receive(head, skb);
 303 }
 304
 305 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
 306                                           struct sk_buff *skb)
 307 {
 308         /* Common GRO receive for SIT and IP6IP6 */
 309
 310         if (NAPI_GRO_CB(skb)->encap_mark) {
 311                 NAPI_GRO_CB(skb)->flush = 1;
 312                 return NULL;
 313         }
 314
 315         NAPI_GRO_CB(skb)->encap_mark = 1;
 316
 317         return inet_gro_receive(head, skb);
 318 }
 319
 320 INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 321 {
 322         const struct net_offload *ops;
 323         struct ipv6hdr *iph;
 324         int err = -ENOSYS;
 325         u32 payload_len;
 326
 327         if (skb->encapsulation) {
 328                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
 329                 skb_set_inner_network_header(skb, nhoff);
 330         }
 331
 332         payload_len = skb->len - nhoff - sizeof(*iph);
 333         if (unlikely(payload_len > IPV6_MAXPLEN)) {
 334                 struct hop_jumbo_hdr *hop_jumbo;
 335                 int hoplen = sizeof(*hop_jumbo);
 336
 337                 /* Move network header left */
 338                 memmove(skb_mac_header(skb) - hoplen, skb_mac_header(skb),
 339                         skb->transport_header - skb->mac_header);
 340                 skb->data -= hoplen;
 341                 skb->len += hoplen;
 342                 skb->mac_header -= hoplen;
 343                 skb->network_header -= hoplen;
 344                 iph = (struct ipv6hdr *)(skb->data + nhoff);
 345                 hop_jumbo = (struct hop_jumbo_hdr *)(iph + 1);
 346
 347                 /* Build hop-by-hop options */
 348                 hop_jumbo->nexthdr = iph->nexthdr;
 349                 hop_jumbo->hdrlen = 0;
 350                 hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
 351                 hop_jumbo->tlv_len = 4;
 352                 hop_jumbo->jumbo_payload_len = htonl(payload_len + hoplen);
 353
 354                 iph->nexthdr = NEXTHDR_HOP;
 355                 iph->payload_len = 0;
 356         } else {
 357                 iph = (struct ipv6hdr *)(skb->data + nhoff);
 358                 iph->payload_len = htons(payload_len);
 359         }
 360
 361         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 362         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 363                 goto out;
 364
 365         err = INDIRECT_CALL_L4(ops->callbacks.gro_complete, tcp6_gro_complete,
 366                                udp6_gro_complete, skb, nhoff);
 367
 368 out:
 369         return err;
 370 }
 371
 372 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 373 {
 374         skb->encapsulation = 1;
 375         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
 376         return ipv6_gro_complete(skb, nhoff);
 377 }
 378
 379 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
 380 {
 381         skb->encapsulation = 1;
 382         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 383         return ipv6_gro_complete(skb, nhoff);
 384 }
 385
 386 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
 387 {
 388         skb->encapsulation = 1;
 389         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 390         return inet_gro_complete(skb, nhoff);
 391 }
 392
 393 static struct packet_offload ipv6_packet_offload __read_mostly = {
 394         .type = cpu_to_be16(ETH_P_IPV6),
 395         .callbacks = {
 396                 .gso_segment = ipv6_gso_segment,
 397                 .gro_receive = ipv6_gro_receive,
 398                 .gro_complete = ipv6_gro_complete,
 399         },
 400 };
 401
 402 static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
 403                                        netdev_features_t features)
 404 {
 405         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
 406                 return ERR_PTR(-EINVAL);
 407
 408         return ipv6_gso_segment(skb, features);
 409 }
 410
 411 static struct sk_buff *ip4ip6_gso_segment(struct sk_buff *skb,
 412                                           netdev_features_t features)
 413 {
 414         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
 415                 return ERR_PTR(-EINVAL);
 416
 417         return inet_gso_segment(skb, features);
 418 }
 419
 420 static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb,
 421                                           netdev_features_t features)
 422 {
 423         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
 424                 return ERR_PTR(-EINVAL);
 425
 426         return ipv6_gso_segment(skb, features);
 427 }
 428
 429 static const struct net_offload sit_offload = {
 430         .callbacks = {
 431                 .gso_segment    = sit_gso_segment,
 432                 .gro_receive    = sit_ip6ip6_gro_receive,
 433                 .gro_complete   = sit_gro_complete,
 434         },
 435 };
 436
 437 static const struct net_offload ip4ip6_offload = {
 438         .callbacks = {
 439                 .gso_segment    = ip4ip6_gso_segment,
 440                 .gro_receive    = ip4ip6_gro_receive,
 441                 .gro_complete   = ip4ip6_gro_complete,
 442         },
 443 };
 444
 445 static const struct net_offload ip6ip6_offload = {
 446         .callbacks = {
 447                 .gso_segment    = ip6ip6_gso_segment,
 448                 .gro_receive    = sit_ip6ip6_gro_receive,
 449                 .gro_complete   = ip6ip6_gro_complete,
 450         },
 451 };
 452 static int __init ipv6_offload_init(void)
 453 {
 454
 455         if (tcpv6_offload_init() < 0)
 456                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 457         if (ipv6_exthdrs_offload_init() < 0)
 458                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 459
 460         dev_add_offload(&ipv6_packet_offload);
 461
 462         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 463         inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
 464         inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
 465
 466         return 0;
 467 }
 468
 469 fs_initcall(ipv6_offload_init);