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