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
  17 #include "ip6_offload.h"
  18
  19 /* All GRO functions are always builtin, except UDP over ipv6, which lays in
  20  * ipv6 module, as it depends on UDPv6 lookup function, so we need special care
  21  * when ipv6 is built as a module
  22  */
  23 #if IS_BUILTIN(CONFIG_IPV6)
  24 #define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_2(f, f2, f1, __VA_ARGS__)
  25 #else
  26 #define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_1(f, f2, __VA_ARGS__)
  27 #endif
  28
  29 #define indirect_call_gro_receive_l4(f2, f1, cb, head, skb)     \
  30 ({                                                              \
  31         unlikely(gro_recursion_inc_test(skb)) ?                 \
  32                 NAPI_GRO_CB(skb)->flush |= 1, NULL :            \
  33                 INDIRECT_CALL_L4(cb, f2, f1, head, skb);        \
  34 })
  35
  36 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
  37 {
  38         const struct net_offload *ops = NULL;
  39
  40         for (;;) {
  41                 struct ipv6_opt_hdr *opth;
  42                 int len;
  43
  44                 if (proto != NEXTHDR_HOP) {
  45                         ops = rcu_dereference(inet6_offloads[proto]);
  46
  47                         if (unlikely(!ops))
  48                                 break;
  49
  50                         if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
  51                                 break;
  52                 }
  53
  54                 if (unlikely(!pskb_may_pull(skb, 8)))
  55                         break;
  56
  57                 opth = (void *)skb->data;
  58                 len = ipv6_optlen(opth);
  59
  60                 if (unlikely(!pskb_may_pull(skb, len)))
  61                         break;
  62
  63                 opth = (void *)skb->data;
  64                 proto = opth->nexthdr;
  65                 __skb_pull(skb, len);
  66         }
  67
  68         return proto;
  69 }
  70
  71 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
  72         netdev_features_t features)
  73 {
  74         struct sk_buff *segs = ERR_PTR(-EINVAL);
  75         struct ipv6hdr *ipv6h;
  76         const struct net_offload *ops;
  77         int proto;
  78         struct frag_hdr *fptr;
  79         unsigned int payload_len;
  80         u8 *prevhdr;
  81         int offset = 0;
  82         bool encap, udpfrag;
  83         int nhoff;
  84         bool gso_partial;
  85
  86         skb_reset_network_header(skb);
  87         nhoff = skb_network_header(skb) - skb_mac_header(skb);
  88         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  89                 goto out;
  90
  91         encap = SKB_GSO_CB(skb)->encap_level > 0;
  92         if (encap)
  93                 features &= skb->dev->hw_enc_features;
  94         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
  95
  96         ipv6h = ipv6_hdr(skb);
  97         __skb_pull(skb, sizeof(*ipv6h));
  98         segs = ERR_PTR(-EPROTONOSUPPORT);
  99
 100         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
 101
 102         if (skb->encapsulation &&
 103             skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
 104                 udpfrag = proto == IPPROTO_UDP && encap &&
 105                           (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
 106         else
 107                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
 108                           (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
 109
 110         ops = rcu_dereference(inet6_offloads[proto]);
 111         if (likely(ops && ops->callbacks.gso_segment)) {
 112                 skb_reset_transport_header(skb);
 113                 segs = ops->callbacks.gso_segment(skb, features);
 114                 if (!segs)
 115                         skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
 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_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *,
 183                                                            struct sk_buff *));
 184 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp6_gro_receive(struct list_head *,
 185                                                            struct sk_buff *));
 186 INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
 187                                                          struct sk_buff *skb)
 188 {
 189         const struct net_offload *ops;
 190         struct sk_buff *pp = NULL;
 191         struct sk_buff *p;
 192         struct ipv6hdr *iph;
 193         unsigned int nlen;
 194         unsigned int hlen;
 195         unsigned int off;
 196         u16 flush = 1;
 197         int proto;
 198
 199         off = skb_gro_offset(skb);
 200         hlen = off + sizeof(*iph);
 201         iph = skb_gro_header_fast(skb, off);
 202         if (skb_gro_header_hard(skb, hlen)) {
 203                 iph = skb_gro_header_slow(skb, hlen, off);
 204                 if (unlikely(!iph))
 205                         goto out;
 206         }
 207
 208         skb_set_network_header(skb, off);
 209         skb_gro_pull(skb, sizeof(*iph));
 210         skb_set_transport_header(skb, skb_gro_offset(skb));
 211
 212         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 213
 214         rcu_read_lock();
 215         proto = iph->nexthdr;
 216         ops = rcu_dereference(inet6_offloads[proto]);
 217         if (!ops || !ops->callbacks.gro_receive) {
 218                 __pskb_pull(skb, skb_gro_offset(skb));
 219                 skb_gro_frag0_invalidate(skb);
 220                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 221                 skb_gro_pull(skb, -skb_transport_offset(skb));
 222                 skb_reset_transport_header(skb);
 223                 __skb_push(skb, skb_gro_offset(skb));
 224
 225                 ops = rcu_dereference(inet6_offloads[proto]);
 226                 if (!ops || !ops->callbacks.gro_receive)
 227                         goto out_unlock;
 228
 229                 iph = ipv6_hdr(skb);
 230         }
 231
 232         NAPI_GRO_CB(skb)->proto = proto;
 233
 234         flush--;
 235         nlen = skb_network_header_len(skb);
 236
 237         list_for_each_entry(p, head, list) {
 238                 const struct ipv6hdr *iph2;
 239                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 240
 241                 if (!NAPI_GRO_CB(p)->same_flow)
 242                         continue;
 243
 244                 iph2 = (struct ipv6hdr *)(p->data + off);
 245                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 246
 247                 /* All fields must match except length and Traffic Class.
 248                  * XXX skbs on the gro_list have all been parsed and pulled
 249                  * already so we don't need to compare nlen
 250                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 251                  * memcmp() alone below is sufficient, right?
 252                  */
 253                  if ((first_word & htonl(0xF00FFFFF)) ||
 254                     !ipv6_addr_equal(&iph->saddr, &iph2->saddr) ||
 255                     !ipv6_addr_equal(&iph->daddr, &iph2->daddr) ||
 256                     *(u16 *)&iph->nexthdr != *(u16 *)&iph2->nexthdr) {
 257 not_same_flow:
 258                         NAPI_GRO_CB(p)->same_flow = 0;
 259                         continue;
 260                 }
 261                 if (unlikely(nlen > sizeof(struct ipv6hdr))) {
 262                         if (memcmp(iph + 1, iph2 + 1,
 263                                    nlen - sizeof(struct ipv6hdr)))
 264                                 goto not_same_flow;
 265                 }
 266                 /* flush if Traffic Class fields are different */
 267                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 268                 NAPI_GRO_CB(p)->flush |= flush;
 269
 270                 /* If the previous IP ID value was based on an atomic
 271                  * datagram we can overwrite the value and ignore it.
 272                  */
 273                 if (NAPI_GRO_CB(skb)->is_atomic)
 274                         NAPI_GRO_CB(p)->flush_id = 0;
 275         }
 276
 277         NAPI_GRO_CB(skb)->is_atomic = true;
 278         NAPI_GRO_CB(skb)->flush |= flush;
 279
 280         skb_gro_postpull_rcsum(skb, iph, nlen);
 281
 282         pp = indirect_call_gro_receive_l4(tcp6_gro_receive, udp6_gro_receive,
 283                                          ops->callbacks.gro_receive, head, skb);
 284
 285 out_unlock:
 286         rcu_read_unlock();
 287
 288 out:
 289         skb_gro_flush_final(skb, pp, flush);
 290
 291         return pp;
 292 }
 293
 294 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
 295                                               struct sk_buff *skb)
 296 {
 297         /* Common GRO receive for SIT and IP6IP6 */
 298
 299         if (NAPI_GRO_CB(skb)->encap_mark) {
 300                 NAPI_GRO_CB(skb)->flush = 1;
 301                 return NULL;
 302         }
 303
 304         NAPI_GRO_CB(skb)->encap_mark = 1;
 305
 306         return ipv6_gro_receive(head, skb);
 307 }
 308
 309 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
 310                                           struct sk_buff *skb)
 311 {
 312         /* Common GRO receive for SIT and IP6IP6 */
 313
 314         if (NAPI_GRO_CB(skb)->encap_mark) {
 315                 NAPI_GRO_CB(skb)->flush = 1;
 316                 return NULL;
 317         }
 318
 319         NAPI_GRO_CB(skb)->encap_mark = 1;
 320
 321         return inet_gro_receive(head, skb);
 322 }
 323
 324 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *, int));
 325 INDIRECT_CALLABLE_DECLARE(int udp6_gro_complete(struct sk_buff *, int));
 326 INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 327 {
 328         const struct net_offload *ops;
 329         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 330         int err = -ENOSYS;
 331
 332         if (skb->encapsulation) {
 333                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
 334                 skb_set_inner_network_header(skb, nhoff);
 335         }
 336
 337         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 338
 339         rcu_read_lock();
 340
 341         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 342         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 343                 goto out_unlock;
 344
 345         err = INDIRECT_CALL_L4(ops->callbacks.gro_complete, tcp6_gro_complete,
 346                                udp6_gro_complete, skb, nhoff);
 347
 348 out_unlock:
 349         rcu_read_unlock();
 350
 351         return err;
 352 }
 353
 354 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 355 {
 356         skb->encapsulation = 1;
 357         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
 358         return ipv6_gro_complete(skb, nhoff);
 359 }
 360
 361 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
 362 {
 363         skb->encapsulation = 1;
 364         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 365         return ipv6_gro_complete(skb, nhoff);
 366 }
 367
 368 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
 369 {
 370         skb->encapsulation = 1;
 371         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 372         return inet_gro_complete(skb, nhoff);
 373 }
 374
 375 static struct packet_offload ipv6_packet_offload __read_mostly = {
 376         .type = cpu_to_be16(ETH_P_IPV6),
 377         .callbacks = {
 378                 .gso_segment = ipv6_gso_segment,
 379                 .gro_receive = ipv6_gro_receive,
 380                 .gro_complete = ipv6_gro_complete,
 381         },
 382 };
 383
 384 static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
 385                                        netdev_features_t features)
 386 {
 387         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
 388                 return ERR_PTR(-EINVAL);
 389
 390         return ipv6_gso_segment(skb, features);
 391 }
 392
 393 static struct sk_buff *ip4ip6_gso_segment(struct sk_buff *skb,
 394                                           netdev_features_t features)
 395 {
 396         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
 397                 return ERR_PTR(-EINVAL);
 398
 399         return inet_gso_segment(skb, features);
 400 }
 401
 402 static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb,
 403                                           netdev_features_t features)
 404 {
 405         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
 406                 return ERR_PTR(-EINVAL);
 407
 408         return ipv6_gso_segment(skb, features);
 409 }
 410
 411 static const struct net_offload sit_offload = {
 412         .callbacks = {
 413                 .gso_segment    = sit_gso_segment,
 414                 .gro_receive    = sit_ip6ip6_gro_receive,
 415                 .gro_complete   = sit_gro_complete,
 416         },
 417 };
 418
 419 static const struct net_offload ip4ip6_offload = {
 420         .callbacks = {
 421                 .gso_segment    = ip4ip6_gso_segment,
 422                 .gro_receive    = ip4ip6_gro_receive,
 423                 .gro_complete   = ip4ip6_gro_complete,
 424         },
 425 };
 426
 427 static const struct net_offload ip6ip6_offload = {
 428         .callbacks = {
 429                 .gso_segment    = ip6ip6_gso_segment,
 430                 .gro_receive    = sit_ip6ip6_gro_receive,
 431                 .gro_complete   = ip6ip6_gro_complete,
 432         },
 433 };
 434 static int __init ipv6_offload_init(void)
 435 {
 436
 437         if (tcpv6_offload_init() < 0)
 438                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 439         if (ipv6_exthdrs_offload_init() < 0)
 440                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 441
 442         dev_add_offload(&ipv6_packet_offload);
 443
 444         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 445         inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
 446         inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
 447
 448         return 0;
 449 }
 450
 451 fs_initcall(ipv6_offload_init);