| 1 | /* |
| 2 | * UDP over IPv6 |
| 3 | * Linux INET6 implementation |
| 4 | * |
| 5 | * Authors: |
| 6 | * Pedro Roque <roque@di.fc.ul.pt> |
| 7 | * |
| 8 | * Based on linux/ipv4/udp.c |
| 9 | * |
| 10 | * Fixes: |
| 11 | * Hideaki YOSHIFUJI : sin6_scope_id support |
| 12 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which |
| 13 | * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind |
| 14 | * a single port at the same time. |
| 15 | * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data |
| 16 | * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file. |
| 17 | * |
| 18 | * This program is free software; you can redistribute it and/or |
| 19 | * modify it under the terms of the GNU General Public License |
| 20 | * as published by the Free Software Foundation; either version |
| 21 | * 2 of the License, or (at your option) any later version. |
| 22 | */ |
| 23 | |
| 24 | #include <linux/errno.h> |
| 25 | #include <linux/types.h> |
| 26 | #include <linux/socket.h> |
| 27 | #include <linux/sockios.h> |
| 28 | #include <linux/net.h> |
| 29 | #include <linux/in6.h> |
| 30 | #include <linux/netdevice.h> |
| 31 | #include <linux/if_arp.h> |
| 32 | #include <linux/ipv6.h> |
| 33 | #include <linux/icmpv6.h> |
| 34 | #include <linux/init.h> |
| 35 | #include <linux/module.h> |
| 36 | #include <linux/skbuff.h> |
| 37 | #include <linux/slab.h> |
| 38 | #include <asm/uaccess.h> |
| 39 | |
| 40 | #include <net/ndisc.h> |
| 41 | #include <net/protocol.h> |
| 42 | #include <net/transp_v6.h> |
| 43 | #include <net/ip6_route.h> |
| 44 | #include <net/raw.h> |
| 45 | #include <net/tcp_states.h> |
| 46 | #include <net/ip6_checksum.h> |
| 47 | #include <net/xfrm.h> |
| 48 | #include <net/inet6_hashtables.h> |
| 49 | #include <net/busy_poll.h> |
| 50 | |
| 51 | #include <linux/proc_fs.h> |
| 52 | #include <linux/seq_file.h> |
| 53 | #include <trace/events/skb.h> |
| 54 | #include "udp_impl.h" |
| 55 | |
| 56 | static unsigned int udp6_ehashfn(struct net *net, |
| 57 | const struct in6_addr *laddr, |
| 58 | const u16 lport, |
| 59 | const struct in6_addr *faddr, |
| 60 | const __be16 fport) |
| 61 | { |
| 62 | static u32 udp6_ehash_secret __read_mostly; |
| 63 | static u32 udp_ipv6_hash_secret __read_mostly; |
| 64 | |
| 65 | u32 lhash, fhash; |
| 66 | |
| 67 | net_get_random_once(&udp6_ehash_secret, |
| 68 | sizeof(udp6_ehash_secret)); |
| 69 | net_get_random_once(&udp_ipv6_hash_secret, |
| 70 | sizeof(udp_ipv6_hash_secret)); |
| 71 | |
| 72 | lhash = (__force u32)laddr->s6_addr32[3]; |
| 73 | fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); |
| 74 | |
| 75 | return __inet6_ehashfn(lhash, lport, fhash, fport, |
| 76 | udp_ipv6_hash_secret + net_hash_mix(net)); |
| 77 | } |
| 78 | |
| 79 | int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2) |
| 80 | { |
| 81 | const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2); |
| 82 | int sk_ipv6only = ipv6_only_sock(sk); |
| 83 | int sk2_ipv6only = inet_v6_ipv6only(sk2); |
| 84 | int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr); |
| 85 | int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED; |
| 86 | |
| 87 | /* if both are mapped, treat as IPv4 */ |
| 88 | if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) |
| 89 | return (!sk2_ipv6only && |
| 90 | (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr || |
| 91 | sk->sk_rcv_saddr == sk2->sk_rcv_saddr)); |
| 92 | |
| 93 | if (addr_type2 == IPV6_ADDR_ANY && |
| 94 | !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED)) |
| 95 | return 1; |
| 96 | |
| 97 | if (addr_type == IPV6_ADDR_ANY && |
| 98 | !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED)) |
| 99 | return 1; |
| 100 | |
| 101 | if (sk2_rcv_saddr6 && |
| 102 | ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6)) |
| 103 | return 1; |
| 104 | |
| 105 | return 0; |
| 106 | } |
| 107 | |
| 108 | static unsigned int udp6_portaddr_hash(struct net *net, |
| 109 | const struct in6_addr *addr6, |
| 110 | unsigned int port) |
| 111 | { |
| 112 | unsigned int hash, mix = net_hash_mix(net); |
| 113 | |
| 114 | if (ipv6_addr_any(addr6)) |
| 115 | hash = jhash_1word(0, mix); |
| 116 | else if (ipv6_addr_v4mapped(addr6)) |
| 117 | hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix); |
| 118 | else |
| 119 | hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix); |
| 120 | |
| 121 | return hash ^ port; |
| 122 | } |
| 123 | |
| 124 | |
| 125 | int udp_v6_get_port(struct sock *sk, unsigned short snum) |
| 126 | { |
| 127 | unsigned int hash2_nulladdr = |
| 128 | udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum); |
| 129 | unsigned int hash2_partial = |
| 130 | udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); |
| 131 | |
| 132 | /* precompute partial secondary hash */ |
| 133 | udp_sk(sk)->udp_portaddr_hash = hash2_partial; |
| 134 | return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr); |
| 135 | } |
| 136 | |
| 137 | static void udp_v6_rehash(struct sock *sk) |
| 138 | { |
| 139 | u16 new_hash = udp6_portaddr_hash(sock_net(sk), |
| 140 | &sk->sk_v6_rcv_saddr, |
| 141 | inet_sk(sk)->inet_num); |
| 142 | |
| 143 | udp_lib_rehash(sk, new_hash); |
| 144 | } |
| 145 | |
| 146 | static inline int compute_score(struct sock *sk, struct net *net, |
| 147 | unsigned short hnum, |
| 148 | const struct in6_addr *saddr, __be16 sport, |
| 149 | const struct in6_addr *daddr, __be16 dport, |
| 150 | int dif) |
| 151 | { |
| 152 | int score = -1; |
| 153 | |
| 154 | if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && |
| 155 | sk->sk_family == PF_INET6) { |
| 156 | struct inet_sock *inet = inet_sk(sk); |
| 157 | |
| 158 | score = 0; |
| 159 | if (inet->inet_dport) { |
| 160 | if (inet->inet_dport != sport) |
| 161 | return -1; |
| 162 | score++; |
| 163 | } |
| 164 | if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { |
| 165 | if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) |
| 166 | return -1; |
| 167 | score++; |
| 168 | } |
| 169 | if (!ipv6_addr_any(&sk->sk_v6_daddr)) { |
| 170 | if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) |
| 171 | return -1; |
| 172 | score++; |
| 173 | } |
| 174 | if (sk->sk_bound_dev_if) { |
| 175 | if (sk->sk_bound_dev_if != dif) |
| 176 | return -1; |
| 177 | score++; |
| 178 | } |
| 179 | } |
| 180 | return score; |
| 181 | } |
| 182 | |
| 183 | #define SCORE2_MAX (1 + 1 + 1) |
| 184 | static inline int compute_score2(struct sock *sk, struct net *net, |
| 185 | const struct in6_addr *saddr, __be16 sport, |
| 186 | const struct in6_addr *daddr, unsigned short hnum, |
| 187 | int dif) |
| 188 | { |
| 189 | int score = -1; |
| 190 | |
| 191 | if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && |
| 192 | sk->sk_family == PF_INET6) { |
| 193 | struct inet_sock *inet = inet_sk(sk); |
| 194 | |
| 195 | if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) |
| 196 | return -1; |
| 197 | score = 0; |
| 198 | if (inet->inet_dport) { |
| 199 | if (inet->inet_dport != sport) |
| 200 | return -1; |
| 201 | score++; |
| 202 | } |
| 203 | if (!ipv6_addr_any(&sk->sk_v6_daddr)) { |
| 204 | if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) |
| 205 | return -1; |
| 206 | score++; |
| 207 | } |
| 208 | if (sk->sk_bound_dev_if) { |
| 209 | if (sk->sk_bound_dev_if != dif) |
| 210 | return -1; |
| 211 | score++; |
| 212 | } |
| 213 | } |
| 214 | return score; |
| 215 | } |
| 216 | |
| 217 | |
| 218 | /* called with read_rcu_lock() */ |
| 219 | static struct sock *udp6_lib_lookup2(struct net *net, |
| 220 | const struct in6_addr *saddr, __be16 sport, |
| 221 | const struct in6_addr *daddr, unsigned int hnum, int dif, |
| 222 | struct udp_hslot *hslot2, unsigned int slot2) |
| 223 | { |
| 224 | struct sock *sk, *result; |
| 225 | struct hlist_nulls_node *node; |
| 226 | int score, badness, matches = 0, reuseport = 0; |
| 227 | u32 hash = 0; |
| 228 | |
| 229 | begin: |
| 230 | result = NULL; |
| 231 | badness = -1; |
| 232 | udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { |
| 233 | score = compute_score2(sk, net, saddr, sport, |
| 234 | daddr, hnum, dif); |
| 235 | if (score > badness) { |
| 236 | result = sk; |
| 237 | badness = score; |
| 238 | reuseport = sk->sk_reuseport; |
| 239 | if (reuseport) { |
| 240 | hash = udp6_ehashfn(net, daddr, hnum, |
| 241 | saddr, sport); |
| 242 | matches = 1; |
| 243 | } else if (score == SCORE2_MAX) |
| 244 | goto exact_match; |
| 245 | } else if (score == badness && reuseport) { |
| 246 | matches++; |
| 247 | if (((u64)hash * matches) >> 32 == 0) |
| 248 | result = sk; |
| 249 | hash = next_pseudo_random32(hash); |
| 250 | } |
| 251 | } |
| 252 | /* |
| 253 | * if the nulls value we got at the end of this lookup is |
| 254 | * not the expected one, we must restart lookup. |
| 255 | * We probably met an item that was moved to another chain. |
| 256 | */ |
| 257 | if (get_nulls_value(node) != slot2) |
| 258 | goto begin; |
| 259 | |
| 260 | if (result) { |
| 261 | exact_match: |
| 262 | if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) |
| 263 | result = NULL; |
| 264 | else if (unlikely(compute_score2(result, net, saddr, sport, |
| 265 | daddr, hnum, dif) < badness)) { |
| 266 | sock_put(result); |
| 267 | goto begin; |
| 268 | } |
| 269 | } |
| 270 | return result; |
| 271 | } |
| 272 | |
| 273 | struct sock *__udp6_lib_lookup(struct net *net, |
| 274 | const struct in6_addr *saddr, __be16 sport, |
| 275 | const struct in6_addr *daddr, __be16 dport, |
| 276 | int dif, struct udp_table *udptable) |
| 277 | { |
| 278 | struct sock *sk, *result; |
| 279 | struct hlist_nulls_node *node; |
| 280 | unsigned short hnum = ntohs(dport); |
| 281 | unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); |
| 282 | struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; |
| 283 | int score, badness, matches = 0, reuseport = 0; |
| 284 | u32 hash = 0; |
| 285 | |
| 286 | rcu_read_lock(); |
| 287 | if (hslot->count > 10) { |
| 288 | hash2 = udp6_portaddr_hash(net, daddr, hnum); |
| 289 | slot2 = hash2 & udptable->mask; |
| 290 | hslot2 = &udptable->hash2[slot2]; |
| 291 | if (hslot->count < hslot2->count) |
| 292 | goto begin; |
| 293 | |
| 294 | result = udp6_lib_lookup2(net, saddr, sport, |
| 295 | daddr, hnum, dif, |
| 296 | hslot2, slot2); |
| 297 | if (!result) { |
| 298 | hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum); |
| 299 | slot2 = hash2 & udptable->mask; |
| 300 | hslot2 = &udptable->hash2[slot2]; |
| 301 | if (hslot->count < hslot2->count) |
| 302 | goto begin; |
| 303 | |
| 304 | result = udp6_lib_lookup2(net, saddr, sport, |
| 305 | &in6addr_any, hnum, dif, |
| 306 | hslot2, slot2); |
| 307 | } |
| 308 | rcu_read_unlock(); |
| 309 | return result; |
| 310 | } |
| 311 | begin: |
| 312 | result = NULL; |
| 313 | badness = -1; |
| 314 | sk_nulls_for_each_rcu(sk, node, &hslot->head) { |
| 315 | score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif); |
| 316 | if (score > badness) { |
| 317 | result = sk; |
| 318 | badness = score; |
| 319 | reuseport = sk->sk_reuseport; |
| 320 | if (reuseport) { |
| 321 | hash = udp6_ehashfn(net, daddr, hnum, |
| 322 | saddr, sport); |
| 323 | matches = 1; |
| 324 | } |
| 325 | } else if (score == badness && reuseport) { |
| 326 | matches++; |
| 327 | if (((u64)hash * matches) >> 32 == 0) |
| 328 | result = sk; |
| 329 | hash = next_pseudo_random32(hash); |
| 330 | } |
| 331 | } |
| 332 | /* |
| 333 | * if the nulls value we got at the end of this lookup is |
| 334 | * not the expected one, we must restart lookup. |
| 335 | * We probably met an item that was moved to another chain. |
| 336 | */ |
| 337 | if (get_nulls_value(node) != slot) |
| 338 | goto begin; |
| 339 | |
| 340 | if (result) { |
| 341 | if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) |
| 342 | result = NULL; |
| 343 | else if (unlikely(compute_score(result, net, hnum, saddr, sport, |
| 344 | daddr, dport, dif) < badness)) { |
| 345 | sock_put(result); |
| 346 | goto begin; |
| 347 | } |
| 348 | } |
| 349 | rcu_read_unlock(); |
| 350 | return result; |
| 351 | } |
| 352 | EXPORT_SYMBOL_GPL(__udp6_lib_lookup); |
| 353 | |
| 354 | static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, |
| 355 | __be16 sport, __be16 dport, |
| 356 | struct udp_table *udptable) |
| 357 | { |
| 358 | struct sock *sk; |
| 359 | const struct ipv6hdr *iph = ipv6_hdr(skb); |
| 360 | |
| 361 | if (unlikely(sk = skb_steal_sock(skb))) |
| 362 | return sk; |
| 363 | return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport, |
| 364 | &iph->daddr, dport, inet6_iif(skb), |
| 365 | udptable); |
| 366 | } |
| 367 | |
| 368 | struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, |
| 369 | const struct in6_addr *daddr, __be16 dport, int dif) |
| 370 | { |
| 371 | return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); |
| 372 | } |
| 373 | EXPORT_SYMBOL_GPL(udp6_lib_lookup); |
| 374 | |
| 375 | |
| 376 | /* |
| 377 | * This should be easy, if there is something there we |
| 378 | * return it, otherwise we block. |
| 379 | */ |
| 380 | |
| 381 | int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk, |
| 382 | struct msghdr *msg, size_t len, |
| 383 | int noblock, int flags, int *addr_len) |
| 384 | { |
| 385 | struct ipv6_pinfo *np = inet6_sk(sk); |
| 386 | struct inet_sock *inet = inet_sk(sk); |
| 387 | struct sk_buff *skb; |
| 388 | unsigned int ulen, copied; |
| 389 | int peeked, off = 0; |
| 390 | int err; |
| 391 | int is_udplite = IS_UDPLITE(sk); |
| 392 | int is_udp4; |
| 393 | bool slow; |
| 394 | |
| 395 | if (flags & MSG_ERRQUEUE) |
| 396 | return ipv6_recv_error(sk, msg, len, addr_len); |
| 397 | |
| 398 | if (np->rxpmtu && np->rxopt.bits.rxpmtu) |
| 399 | return ipv6_recv_rxpmtu(sk, msg, len, addr_len); |
| 400 | |
| 401 | try_again: |
| 402 | skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), |
| 403 | &peeked, &off, &err); |
| 404 | if (!skb) |
| 405 | goto out; |
| 406 | |
| 407 | ulen = skb->len - sizeof(struct udphdr); |
| 408 | copied = len; |
| 409 | if (copied > ulen) |
| 410 | copied = ulen; |
| 411 | else if (copied < ulen) |
| 412 | msg->msg_flags |= MSG_TRUNC; |
| 413 | |
| 414 | is_udp4 = (skb->protocol == htons(ETH_P_IP)); |
| 415 | |
| 416 | /* |
| 417 | * If checksum is needed at all, try to do it while copying the |
| 418 | * data. If the data is truncated, or if we only want a partial |
| 419 | * coverage checksum (UDP-Lite), do it before the copy. |
| 420 | */ |
| 421 | |
| 422 | if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { |
| 423 | if (udp_lib_checksum_complete(skb)) |
| 424 | goto csum_copy_err; |
| 425 | } |
| 426 | |
| 427 | if (skb_csum_unnecessary(skb)) |
| 428 | err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), |
| 429 | msg->msg_iov, copied); |
| 430 | else { |
| 431 | err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); |
| 432 | if (err == -EINVAL) |
| 433 | goto csum_copy_err; |
| 434 | } |
| 435 | if (unlikely(err)) { |
| 436 | trace_kfree_skb(skb, udpv6_recvmsg); |
| 437 | if (!peeked) { |
| 438 | atomic_inc(&sk->sk_drops); |
| 439 | if (is_udp4) |
| 440 | UDP_INC_STATS_USER(sock_net(sk), |
| 441 | UDP_MIB_INERRORS, |
| 442 | is_udplite); |
| 443 | else |
| 444 | UDP6_INC_STATS_USER(sock_net(sk), |
| 445 | UDP_MIB_INERRORS, |
| 446 | is_udplite); |
| 447 | } |
| 448 | goto out_free; |
| 449 | } |
| 450 | if (!peeked) { |
| 451 | if (is_udp4) |
| 452 | UDP_INC_STATS_USER(sock_net(sk), |
| 453 | UDP_MIB_INDATAGRAMS, is_udplite); |
| 454 | else |
| 455 | UDP6_INC_STATS_USER(sock_net(sk), |
| 456 | UDP_MIB_INDATAGRAMS, is_udplite); |
| 457 | } |
| 458 | |
| 459 | sock_recv_ts_and_drops(msg, sk, skb); |
| 460 | |
| 461 | /* Copy the address. */ |
| 462 | if (msg->msg_name) { |
| 463 | DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); |
| 464 | sin6->sin6_family = AF_INET6; |
| 465 | sin6->sin6_port = udp_hdr(skb)->source; |
| 466 | sin6->sin6_flowinfo = 0; |
| 467 | |
| 468 | if (is_udp4) { |
| 469 | ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, |
| 470 | &sin6->sin6_addr); |
| 471 | sin6->sin6_scope_id = 0; |
| 472 | } else { |
| 473 | sin6->sin6_addr = ipv6_hdr(skb)->saddr; |
| 474 | sin6->sin6_scope_id = |
| 475 | ipv6_iface_scope_id(&sin6->sin6_addr, |
| 476 | IP6CB(skb)->iif); |
| 477 | } |
| 478 | *addr_len = sizeof(*sin6); |
| 479 | } |
| 480 | |
| 481 | if (np->rxopt.all) |
| 482 | ip6_datagram_recv_common_ctl(sk, msg, skb); |
| 483 | |
| 484 | if (is_udp4) { |
| 485 | if (inet->cmsg_flags) |
| 486 | ip_cmsg_recv(msg, skb); |
| 487 | } else { |
| 488 | if (np->rxopt.all) |
| 489 | ip6_datagram_recv_specific_ctl(sk, msg, skb); |
| 490 | } |
| 491 | |
| 492 | err = copied; |
| 493 | if (flags & MSG_TRUNC) |
| 494 | err = ulen; |
| 495 | |
| 496 | out_free: |
| 497 | skb_free_datagram_locked(sk, skb); |
| 498 | out: |
| 499 | return err; |
| 500 | |
| 501 | csum_copy_err: |
| 502 | slow = lock_sock_fast(sk); |
| 503 | if (!skb_kill_datagram(sk, skb, flags)) { |
| 504 | if (is_udp4) { |
| 505 | UDP_INC_STATS_USER(sock_net(sk), |
| 506 | UDP_MIB_CSUMERRORS, is_udplite); |
| 507 | UDP_INC_STATS_USER(sock_net(sk), |
| 508 | UDP_MIB_INERRORS, is_udplite); |
| 509 | } else { |
| 510 | UDP6_INC_STATS_USER(sock_net(sk), |
| 511 | UDP_MIB_CSUMERRORS, is_udplite); |
| 512 | UDP6_INC_STATS_USER(sock_net(sk), |
| 513 | UDP_MIB_INERRORS, is_udplite); |
| 514 | } |
| 515 | } |
| 516 | unlock_sock_fast(sk, slow); |
| 517 | |
| 518 | if (noblock) |
| 519 | return -EAGAIN; |
| 520 | |
| 521 | /* starting over for a new packet */ |
| 522 | msg->msg_flags &= ~MSG_TRUNC; |
| 523 | goto try_again; |
| 524 | } |
| 525 | |
| 526 | void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, |
| 527 | u8 type, u8 code, int offset, __be32 info, |
| 528 | struct udp_table *udptable) |
| 529 | { |
| 530 | struct ipv6_pinfo *np; |
| 531 | const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; |
| 532 | const struct in6_addr *saddr = &hdr->saddr; |
| 533 | const struct in6_addr *daddr = &hdr->daddr; |
| 534 | struct udphdr *uh = (struct udphdr*)(skb->data+offset); |
| 535 | struct sock *sk; |
| 536 | int err; |
| 537 | |
| 538 | sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest, |
| 539 | saddr, uh->source, inet6_iif(skb), udptable); |
| 540 | if (sk == NULL) |
| 541 | return; |
| 542 | |
| 543 | if (type == ICMPV6_PKT_TOOBIG) { |
| 544 | if (!ip6_sk_accept_pmtu(sk)) |
| 545 | goto out; |
| 546 | ip6_sk_update_pmtu(skb, sk, info); |
| 547 | } |
| 548 | if (type == NDISC_REDIRECT) { |
| 549 | ip6_sk_redirect(skb, sk); |
| 550 | goto out; |
| 551 | } |
| 552 | |
| 553 | np = inet6_sk(sk); |
| 554 | |
| 555 | if (!icmpv6_err_convert(type, code, &err) && !np->recverr) |
| 556 | goto out; |
| 557 | |
| 558 | if (sk->sk_state != TCP_ESTABLISHED && !np->recverr) |
| 559 | goto out; |
| 560 | |
| 561 | if (np->recverr) |
| 562 | ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); |
| 563 | |
| 564 | sk->sk_err = err; |
| 565 | sk->sk_error_report(sk); |
| 566 | out: |
| 567 | sock_put(sk); |
| 568 | } |
| 569 | |
| 570 | static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
| 571 | { |
| 572 | int rc; |
| 573 | |
| 574 | if (!ipv6_addr_any(&sk->sk_v6_daddr)) { |
| 575 | sock_rps_save_rxhash(sk, skb); |
| 576 | sk_mark_napi_id(sk, skb); |
| 577 | } |
| 578 | |
| 579 | rc = sock_queue_rcv_skb(sk, skb); |
| 580 | if (rc < 0) { |
| 581 | int is_udplite = IS_UDPLITE(sk); |
| 582 | |
| 583 | /* Note that an ENOMEM error is charged twice */ |
| 584 | if (rc == -ENOMEM) |
| 585 | UDP6_INC_STATS_BH(sock_net(sk), |
| 586 | UDP_MIB_RCVBUFERRORS, is_udplite); |
| 587 | UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
| 588 | kfree_skb(skb); |
| 589 | return -1; |
| 590 | } |
| 591 | return 0; |
| 592 | } |
| 593 | |
| 594 | static __inline__ void udpv6_err(struct sk_buff *skb, |
| 595 | struct inet6_skb_parm *opt, u8 type, |
| 596 | u8 code, int offset, __be32 info ) |
| 597 | { |
| 598 | __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); |
| 599 | } |
| 600 | |
| 601 | static struct static_key udpv6_encap_needed __read_mostly; |
| 602 | void udpv6_encap_enable(void) |
| 603 | { |
| 604 | if (!static_key_enabled(&udpv6_encap_needed)) |
| 605 | static_key_slow_inc(&udpv6_encap_needed); |
| 606 | } |
| 607 | EXPORT_SYMBOL(udpv6_encap_enable); |
| 608 | |
| 609 | int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
| 610 | { |
| 611 | struct udp_sock *up = udp_sk(sk); |
| 612 | int rc; |
| 613 | int is_udplite = IS_UDPLITE(sk); |
| 614 | |
| 615 | if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) |
| 616 | goto drop; |
| 617 | |
| 618 | if (static_key_false(&udpv6_encap_needed) && up->encap_type) { |
| 619 | int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); |
| 620 | |
| 621 | /* |
| 622 | * This is an encapsulation socket so pass the skb to |
| 623 | * the socket's udp_encap_rcv() hook. Otherwise, just |
| 624 | * fall through and pass this up the UDP socket. |
| 625 | * up->encap_rcv() returns the following value: |
| 626 | * =0 if skb was successfully passed to the encap |
| 627 | * handler or was discarded by it. |
| 628 | * >0 if skb should be passed on to UDP. |
| 629 | * <0 if skb should be resubmitted as proto -N |
| 630 | */ |
| 631 | |
| 632 | /* if we're overly short, let UDP handle it */ |
| 633 | encap_rcv = ACCESS_ONCE(up->encap_rcv); |
| 634 | if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) { |
| 635 | int ret; |
| 636 | |
| 637 | /* Verify checksum before giving to encap */ |
| 638 | if (udp_lib_checksum_complete(skb)) |
| 639 | goto csum_error; |
| 640 | |
| 641 | ret = encap_rcv(sk, skb); |
| 642 | if (ret <= 0) { |
| 643 | UDP_INC_STATS_BH(sock_net(sk), |
| 644 | UDP_MIB_INDATAGRAMS, |
| 645 | is_udplite); |
| 646 | return -ret; |
| 647 | } |
| 648 | } |
| 649 | |
| 650 | /* FALLTHROUGH -- it's a UDP Packet */ |
| 651 | } |
| 652 | |
| 653 | /* |
| 654 | * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). |
| 655 | */ |
| 656 | if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { |
| 657 | |
| 658 | if (up->pcrlen == 0) { /* full coverage was set */ |
| 659 | LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage" |
| 660 | " %d while full coverage %d requested\n", |
| 661 | UDP_SKB_CB(skb)->cscov, skb->len); |
| 662 | goto drop; |
| 663 | } |
| 664 | if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { |
| 665 | LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d " |
| 666 | "too small, need min %d\n", |
| 667 | UDP_SKB_CB(skb)->cscov, up->pcrlen); |
| 668 | goto drop; |
| 669 | } |
| 670 | } |
| 671 | |
| 672 | if (rcu_access_pointer(sk->sk_filter)) { |
| 673 | if (udp_lib_checksum_complete(skb)) |
| 674 | goto csum_error; |
| 675 | } |
| 676 | |
| 677 | if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) { |
| 678 | UDP6_INC_STATS_BH(sock_net(sk), |
| 679 | UDP_MIB_RCVBUFERRORS, is_udplite); |
| 680 | goto drop; |
| 681 | } |
| 682 | |
| 683 | skb_dst_drop(skb); |
| 684 | |
| 685 | bh_lock_sock(sk); |
| 686 | rc = 0; |
| 687 | if (!sock_owned_by_user(sk)) |
| 688 | rc = __udpv6_queue_rcv_skb(sk, skb); |
| 689 | else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { |
| 690 | bh_unlock_sock(sk); |
| 691 | goto drop; |
| 692 | } |
| 693 | bh_unlock_sock(sk); |
| 694 | |
| 695 | return rc; |
| 696 | |
| 697 | csum_error: |
| 698 | UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
| 699 | drop: |
| 700 | UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
| 701 | atomic_inc(&sk->sk_drops); |
| 702 | kfree_skb(skb); |
| 703 | return -1; |
| 704 | } |
| 705 | |
| 706 | static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk, |
| 707 | __be16 loc_port, const struct in6_addr *loc_addr, |
| 708 | __be16 rmt_port, const struct in6_addr *rmt_addr, |
| 709 | int dif) |
| 710 | { |
| 711 | struct hlist_nulls_node *node; |
| 712 | unsigned short num = ntohs(loc_port); |
| 713 | |
| 714 | sk_nulls_for_each_from(sk, node) { |
| 715 | struct inet_sock *inet = inet_sk(sk); |
| 716 | |
| 717 | if (!net_eq(sock_net(sk), net)) |
| 718 | continue; |
| 719 | |
| 720 | if (udp_sk(sk)->udp_port_hash == num && |
| 721 | sk->sk_family == PF_INET6) { |
| 722 | if (inet->inet_dport) { |
| 723 | if (inet->inet_dport != rmt_port) |
| 724 | continue; |
| 725 | } |
| 726 | if (!ipv6_addr_any(&sk->sk_v6_daddr) && |
| 727 | !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) |
| 728 | continue; |
| 729 | |
| 730 | if (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) |
| 731 | continue; |
| 732 | |
| 733 | if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { |
| 734 | if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)) |
| 735 | continue; |
| 736 | } |
| 737 | if (!inet6_mc_check(sk, loc_addr, rmt_addr)) |
| 738 | continue; |
| 739 | return sk; |
| 740 | } |
| 741 | } |
| 742 | return NULL; |
| 743 | } |
| 744 | |
| 745 | static void flush_stack(struct sock **stack, unsigned int count, |
| 746 | struct sk_buff *skb, unsigned int final) |
| 747 | { |
| 748 | struct sk_buff *skb1 = NULL; |
| 749 | struct sock *sk; |
| 750 | unsigned int i; |
| 751 | |
| 752 | for (i = 0; i < count; i++) { |
| 753 | sk = stack[i]; |
| 754 | if (likely(skb1 == NULL)) |
| 755 | skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); |
| 756 | if (!skb1) { |
| 757 | atomic_inc(&sk->sk_drops); |
| 758 | UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, |
| 759 | IS_UDPLITE(sk)); |
| 760 | UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, |
| 761 | IS_UDPLITE(sk)); |
| 762 | } |
| 763 | |
| 764 | if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0) |
| 765 | skb1 = NULL; |
| 766 | } |
| 767 | if (unlikely(skb1)) |
| 768 | kfree_skb(skb1); |
| 769 | } |
| 770 | |
| 771 | static void udp6_csum_zero_error(struct sk_buff *skb) |
| 772 | { |
| 773 | /* RFC 2460 section 8.1 says that we SHOULD log |
| 774 | * this error. Well, it is reasonable. |
| 775 | */ |
| 776 | LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", |
| 777 | &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), |
| 778 | &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); |
| 779 | } |
| 780 | |
| 781 | /* |
| 782 | * Note: called only from the BH handler context, |
| 783 | * so we don't need to lock the hashes. |
| 784 | */ |
| 785 | static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, |
| 786 | const struct in6_addr *saddr, const struct in6_addr *daddr, |
| 787 | struct udp_table *udptable) |
| 788 | { |
| 789 | struct sock *sk, *stack[256 / sizeof(struct sock *)]; |
| 790 | const struct udphdr *uh = udp_hdr(skb); |
| 791 | struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); |
| 792 | int dif; |
| 793 | unsigned int i, count = 0; |
| 794 | |
| 795 | spin_lock(&hslot->lock); |
| 796 | sk = sk_nulls_head(&hslot->head); |
| 797 | dif = inet6_iif(skb); |
| 798 | sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); |
| 799 | while (sk) { |
| 800 | /* If zero checksum and no_check is not on for |
| 801 | * the socket then skip it. |
| 802 | */ |
| 803 | if (uh->check || udp_sk(sk)->no_check6_rx) |
| 804 | stack[count++] = sk; |
| 805 | |
| 806 | sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr, |
| 807 | uh->source, saddr, dif); |
| 808 | if (unlikely(count == ARRAY_SIZE(stack))) { |
| 809 | if (!sk) |
| 810 | break; |
| 811 | flush_stack(stack, count, skb, ~0); |
| 812 | count = 0; |
| 813 | } |
| 814 | } |
| 815 | /* |
| 816 | * before releasing the lock, we must take reference on sockets |
| 817 | */ |
| 818 | for (i = 0; i < count; i++) |
| 819 | sock_hold(stack[i]); |
| 820 | |
| 821 | spin_unlock(&hslot->lock); |
| 822 | |
| 823 | if (count) { |
| 824 | flush_stack(stack, count, skb, count - 1); |
| 825 | |
| 826 | for (i = 0; i < count; i++) |
| 827 | sock_put(stack[i]); |
| 828 | } else { |
| 829 | kfree_skb(skb); |
| 830 | } |
| 831 | return 0; |
| 832 | } |
| 833 | |
| 834 | int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, |
| 835 | int proto) |
| 836 | { |
| 837 | struct net *net = dev_net(skb->dev); |
| 838 | struct sock *sk; |
| 839 | struct udphdr *uh; |
| 840 | const struct in6_addr *saddr, *daddr; |
| 841 | u32 ulen = 0; |
| 842 | |
| 843 | if (!pskb_may_pull(skb, sizeof(struct udphdr))) |
| 844 | goto discard; |
| 845 | |
| 846 | saddr = &ipv6_hdr(skb)->saddr; |
| 847 | daddr = &ipv6_hdr(skb)->daddr; |
| 848 | uh = udp_hdr(skb); |
| 849 | |
| 850 | ulen = ntohs(uh->len); |
| 851 | if (ulen > skb->len) |
| 852 | goto short_packet; |
| 853 | |
| 854 | if (proto == IPPROTO_UDP) { |
| 855 | /* UDP validates ulen. */ |
| 856 | |
| 857 | /* Check for jumbo payload */ |
| 858 | if (ulen == 0) |
| 859 | ulen = skb->len; |
| 860 | |
| 861 | if (ulen < sizeof(*uh)) |
| 862 | goto short_packet; |
| 863 | |
| 864 | if (ulen < skb->len) { |
| 865 | if (pskb_trim_rcsum(skb, ulen)) |
| 866 | goto short_packet; |
| 867 | saddr = &ipv6_hdr(skb)->saddr; |
| 868 | daddr = &ipv6_hdr(skb)->daddr; |
| 869 | uh = udp_hdr(skb); |
| 870 | } |
| 871 | } |
| 872 | |
| 873 | if (udp6_csum_init(skb, uh, proto)) |
| 874 | goto csum_error; |
| 875 | |
| 876 | /* |
| 877 | * Multicast receive code |
| 878 | */ |
| 879 | if (ipv6_addr_is_multicast(daddr)) |
| 880 | return __udp6_lib_mcast_deliver(net, skb, |
| 881 | saddr, daddr, udptable); |
| 882 | |
| 883 | /* Unicast */ |
| 884 | |
| 885 | /* |
| 886 | * check socket cache ... must talk to Alan about his plans |
| 887 | * for sock caches... i'll skip this for now. |
| 888 | */ |
| 889 | sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); |
| 890 | if (sk != NULL) { |
| 891 | int ret; |
| 892 | |
| 893 | if (!uh->check && !udp_sk(sk)->no_check6_rx) { |
| 894 | sock_put(sk); |
| 895 | udp6_csum_zero_error(skb); |
| 896 | goto csum_error; |
| 897 | } |
| 898 | |
| 899 | ret = udpv6_queue_rcv_skb(sk, skb); |
| 900 | sock_put(sk); |
| 901 | |
| 902 | /* a return value > 0 means to resubmit the input, but |
| 903 | * it wants the return to be -protocol, or 0 |
| 904 | */ |
| 905 | if (ret > 0) |
| 906 | return -ret; |
| 907 | |
| 908 | return 0; |
| 909 | } |
| 910 | |
| 911 | if (!uh->check) { |
| 912 | udp6_csum_zero_error(skb); |
| 913 | goto csum_error; |
| 914 | } |
| 915 | |
| 916 | if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) |
| 917 | goto discard; |
| 918 | |
| 919 | if (udp_lib_checksum_complete(skb)) |
| 920 | goto csum_error; |
| 921 | |
| 922 | UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); |
| 923 | icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); |
| 924 | |
| 925 | kfree_skb(skb); |
| 926 | return 0; |
| 927 | |
| 928 | short_packet: |
| 929 | LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", |
| 930 | proto == IPPROTO_UDPLITE ? "-Lite" : "", |
| 931 | saddr, |
| 932 | ntohs(uh->source), |
| 933 | ulen, |
| 934 | skb->len, |
| 935 | daddr, |
| 936 | ntohs(uh->dest)); |
| 937 | goto discard; |
| 938 | csum_error: |
| 939 | UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); |
| 940 | discard: |
| 941 | UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); |
| 942 | kfree_skb(skb); |
| 943 | return 0; |
| 944 | } |
| 945 | |
| 946 | static __inline__ int udpv6_rcv(struct sk_buff *skb) |
| 947 | { |
| 948 | return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); |
| 949 | } |
| 950 | |
| 951 | /* |
| 952 | * Throw away all pending data and cancel the corking. Socket is locked. |
| 953 | */ |
| 954 | static void udp_v6_flush_pending_frames(struct sock *sk) |
| 955 | { |
| 956 | struct udp_sock *up = udp_sk(sk); |
| 957 | |
| 958 | if (up->pending == AF_INET) |
| 959 | udp_flush_pending_frames(sk); |
| 960 | else if (up->pending) { |
| 961 | up->len = 0; |
| 962 | up->pending = 0; |
| 963 | ip6_flush_pending_frames(sk); |
| 964 | } |
| 965 | } |
| 966 | |
| 967 | /** |
| 968 | * udp6_hwcsum_outgoing - handle outgoing HW checksumming |
| 969 | * @sk: socket we are sending on |
| 970 | * @skb: sk_buff containing the filled-in UDP header |
| 971 | * (checksum field must be zeroed out) |
| 972 | */ |
| 973 | static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, |
| 974 | const struct in6_addr *saddr, |
| 975 | const struct in6_addr *daddr, int len) |
| 976 | { |
| 977 | unsigned int offset; |
| 978 | struct udphdr *uh = udp_hdr(skb); |
| 979 | __wsum csum = 0; |
| 980 | |
| 981 | if (skb_queue_len(&sk->sk_write_queue) == 1) { |
| 982 | /* Only one fragment on the socket. */ |
| 983 | skb->csum_start = skb_transport_header(skb) - skb->head; |
| 984 | skb->csum_offset = offsetof(struct udphdr, check); |
| 985 | uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); |
| 986 | } else { |
| 987 | /* |
| 988 | * HW-checksum won't work as there are two or more |
| 989 | * fragments on the socket so that all csums of sk_buffs |
| 990 | * should be together |
| 991 | */ |
| 992 | offset = skb_transport_offset(skb); |
| 993 | skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); |
| 994 | |
| 995 | skb->ip_summed = CHECKSUM_NONE; |
| 996 | |
| 997 | skb_queue_walk(&sk->sk_write_queue, skb) { |
| 998 | csum = csum_add(csum, skb->csum); |
| 999 | } |
| 1000 | |
| 1001 | uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, |
| 1002 | csum); |
| 1003 | if (uh->check == 0) |
| 1004 | uh->check = CSUM_MANGLED_0; |
| 1005 | } |
| 1006 | } |
| 1007 | |
| 1008 | /* |
| 1009 | * Sending |
| 1010 | */ |
| 1011 | |
| 1012 | static int udp_v6_push_pending_frames(struct sock *sk) |
| 1013 | { |
| 1014 | struct sk_buff *skb; |
| 1015 | struct udphdr *uh; |
| 1016 | struct udp_sock *up = udp_sk(sk); |
| 1017 | struct inet_sock *inet = inet_sk(sk); |
| 1018 | struct flowi6 *fl6; |
| 1019 | int err = 0; |
| 1020 | int is_udplite = IS_UDPLITE(sk); |
| 1021 | __wsum csum = 0; |
| 1022 | |
| 1023 | if (up->pending == AF_INET) |
| 1024 | return udp_push_pending_frames(sk); |
| 1025 | |
| 1026 | fl6 = &inet->cork.fl.u.ip6; |
| 1027 | |
| 1028 | /* Grab the skbuff where UDP header space exists. */ |
| 1029 | if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) |
| 1030 | goto out; |
| 1031 | |
| 1032 | /* |
| 1033 | * Create a UDP header |
| 1034 | */ |
| 1035 | uh = udp_hdr(skb); |
| 1036 | uh->source = fl6->fl6_sport; |
| 1037 | uh->dest = fl6->fl6_dport; |
| 1038 | uh->len = htons(up->len); |
| 1039 | uh->check = 0; |
| 1040 | |
| 1041 | if (is_udplite) |
| 1042 | csum = udplite_csum_outgoing(sk, skb); |
| 1043 | else if (up->no_check6_tx) { /* UDP csum disabled */ |
| 1044 | skb->ip_summed = CHECKSUM_NONE; |
| 1045 | goto send; |
| 1046 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ |
| 1047 | udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, |
| 1048 | up->len); |
| 1049 | goto send; |
| 1050 | } else |
| 1051 | csum = udp_csum_outgoing(sk, skb); |
| 1052 | |
| 1053 | /* add protocol-dependent pseudo-header */ |
| 1054 | uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, |
| 1055 | up->len, fl6->flowi6_proto, csum); |
| 1056 | if (uh->check == 0) |
| 1057 | uh->check = CSUM_MANGLED_0; |
| 1058 | |
| 1059 | send: |
| 1060 | err = ip6_push_pending_frames(sk); |
| 1061 | if (err) { |
| 1062 | if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { |
| 1063 | UDP6_INC_STATS_USER(sock_net(sk), |
| 1064 | UDP_MIB_SNDBUFERRORS, is_udplite); |
| 1065 | err = 0; |
| 1066 | } |
| 1067 | } else |
| 1068 | UDP6_INC_STATS_USER(sock_net(sk), |
| 1069 | UDP_MIB_OUTDATAGRAMS, is_udplite); |
| 1070 | out: |
| 1071 | up->len = 0; |
| 1072 | up->pending = 0; |
| 1073 | return err; |
| 1074 | } |
| 1075 | |
| 1076 | int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk, |
| 1077 | struct msghdr *msg, size_t len) |
| 1078 | { |
| 1079 | struct ipv6_txoptions opt_space; |
| 1080 | struct udp_sock *up = udp_sk(sk); |
| 1081 | struct inet_sock *inet = inet_sk(sk); |
| 1082 | struct ipv6_pinfo *np = inet6_sk(sk); |
| 1083 | DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); |
| 1084 | struct in6_addr *daddr, *final_p, final; |
| 1085 | struct ipv6_txoptions *opt = NULL; |
| 1086 | struct ip6_flowlabel *flowlabel = NULL; |
| 1087 | struct flowi6 fl6; |
| 1088 | struct dst_entry *dst; |
| 1089 | int addr_len = msg->msg_namelen; |
| 1090 | int ulen = len; |
| 1091 | int hlimit = -1; |
| 1092 | int tclass = -1; |
| 1093 | int dontfrag = -1; |
| 1094 | int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; |
| 1095 | int err; |
| 1096 | int connected = 0; |
| 1097 | int is_udplite = IS_UDPLITE(sk); |
| 1098 | int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); |
| 1099 | |
| 1100 | /* destination address check */ |
| 1101 | if (sin6) { |
| 1102 | if (addr_len < offsetof(struct sockaddr, sa_data)) |
| 1103 | return -EINVAL; |
| 1104 | |
| 1105 | switch (sin6->sin6_family) { |
| 1106 | case AF_INET6: |
| 1107 | if (addr_len < SIN6_LEN_RFC2133) |
| 1108 | return -EINVAL; |
| 1109 | daddr = &sin6->sin6_addr; |
| 1110 | break; |
| 1111 | case AF_INET: |
| 1112 | goto do_udp_sendmsg; |
| 1113 | case AF_UNSPEC: |
| 1114 | msg->msg_name = sin6 = NULL; |
| 1115 | msg->msg_namelen = addr_len = 0; |
| 1116 | daddr = NULL; |
| 1117 | break; |
| 1118 | default: |
| 1119 | return -EINVAL; |
| 1120 | } |
| 1121 | } else if (!up->pending) { |
| 1122 | if (sk->sk_state != TCP_ESTABLISHED) |
| 1123 | return -EDESTADDRREQ; |
| 1124 | daddr = &sk->sk_v6_daddr; |
| 1125 | } else |
| 1126 | daddr = NULL; |
| 1127 | |
| 1128 | if (daddr) { |
| 1129 | if (ipv6_addr_v4mapped(daddr)) { |
| 1130 | struct sockaddr_in sin; |
| 1131 | sin.sin_family = AF_INET; |
| 1132 | sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; |
| 1133 | sin.sin_addr.s_addr = daddr->s6_addr32[3]; |
| 1134 | msg->msg_name = &sin; |
| 1135 | msg->msg_namelen = sizeof(sin); |
| 1136 | do_udp_sendmsg: |
| 1137 | if (__ipv6_only_sock(sk)) |
| 1138 | return -ENETUNREACH; |
| 1139 | return udp_sendmsg(iocb, sk, msg, len); |
| 1140 | } |
| 1141 | } |
| 1142 | |
| 1143 | if (up->pending == AF_INET) |
| 1144 | return udp_sendmsg(iocb, sk, msg, len); |
| 1145 | |
| 1146 | /* Rough check on arithmetic overflow, |
| 1147 | better check is made in ip6_append_data(). |
| 1148 | */ |
| 1149 | if (len > INT_MAX - sizeof(struct udphdr)) |
| 1150 | return -EMSGSIZE; |
| 1151 | |
| 1152 | if (up->pending) { |
| 1153 | /* |
| 1154 | * There are pending frames. |
| 1155 | * The socket lock must be held while it's corked. |
| 1156 | */ |
| 1157 | lock_sock(sk); |
| 1158 | if (likely(up->pending)) { |
| 1159 | if (unlikely(up->pending != AF_INET6)) { |
| 1160 | release_sock(sk); |
| 1161 | return -EAFNOSUPPORT; |
| 1162 | } |
| 1163 | dst = NULL; |
| 1164 | goto do_append_data; |
| 1165 | } |
| 1166 | release_sock(sk); |
| 1167 | } |
| 1168 | ulen += sizeof(struct udphdr); |
| 1169 | |
| 1170 | memset(&fl6, 0, sizeof(fl6)); |
| 1171 | |
| 1172 | if (sin6) { |
| 1173 | if (sin6->sin6_port == 0) |
| 1174 | return -EINVAL; |
| 1175 | |
| 1176 | fl6.fl6_dport = sin6->sin6_port; |
| 1177 | daddr = &sin6->sin6_addr; |
| 1178 | |
| 1179 | if (np->sndflow) { |
| 1180 | fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; |
| 1181 | if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { |
| 1182 | flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); |
| 1183 | if (flowlabel == NULL) |
| 1184 | return -EINVAL; |
| 1185 | } |
| 1186 | } |
| 1187 | |
| 1188 | /* |
| 1189 | * Otherwise it will be difficult to maintain |
| 1190 | * sk->sk_dst_cache. |
| 1191 | */ |
| 1192 | if (sk->sk_state == TCP_ESTABLISHED && |
| 1193 | ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) |
| 1194 | daddr = &sk->sk_v6_daddr; |
| 1195 | |
| 1196 | if (addr_len >= sizeof(struct sockaddr_in6) && |
| 1197 | sin6->sin6_scope_id && |
| 1198 | __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) |
| 1199 | fl6.flowi6_oif = sin6->sin6_scope_id; |
| 1200 | } else { |
| 1201 | if (sk->sk_state != TCP_ESTABLISHED) |
| 1202 | return -EDESTADDRREQ; |
| 1203 | |
| 1204 | fl6.fl6_dport = inet->inet_dport; |
| 1205 | daddr = &sk->sk_v6_daddr; |
| 1206 | fl6.flowlabel = np->flow_label; |
| 1207 | connected = 1; |
| 1208 | } |
| 1209 | |
| 1210 | if (!fl6.flowi6_oif) |
| 1211 | fl6.flowi6_oif = sk->sk_bound_dev_if; |
| 1212 | |
| 1213 | if (!fl6.flowi6_oif) |
| 1214 | fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; |
| 1215 | |
| 1216 | fl6.flowi6_mark = sk->sk_mark; |
| 1217 | |
| 1218 | if (msg->msg_controllen) { |
| 1219 | opt = &opt_space; |
| 1220 | memset(opt, 0, sizeof(struct ipv6_txoptions)); |
| 1221 | opt->tot_len = sizeof(*opt); |
| 1222 | |
| 1223 | err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt, |
| 1224 | &hlimit, &tclass, &dontfrag); |
| 1225 | if (err < 0) { |
| 1226 | fl6_sock_release(flowlabel); |
| 1227 | return err; |
| 1228 | } |
| 1229 | if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { |
| 1230 | flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); |
| 1231 | if (flowlabel == NULL) |
| 1232 | return -EINVAL; |
| 1233 | } |
| 1234 | if (!(opt->opt_nflen|opt->opt_flen)) |
| 1235 | opt = NULL; |
| 1236 | connected = 0; |
| 1237 | } |
| 1238 | if (opt == NULL) |
| 1239 | opt = np->opt; |
| 1240 | if (flowlabel) |
| 1241 | opt = fl6_merge_options(&opt_space, flowlabel, opt); |
| 1242 | opt = ipv6_fixup_options(&opt_space, opt); |
| 1243 | |
| 1244 | fl6.flowi6_proto = sk->sk_protocol; |
| 1245 | if (!ipv6_addr_any(daddr)) |
| 1246 | fl6.daddr = *daddr; |
| 1247 | else |
| 1248 | fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ |
| 1249 | if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) |
| 1250 | fl6.saddr = np->saddr; |
| 1251 | fl6.fl6_sport = inet->inet_sport; |
| 1252 | |
| 1253 | final_p = fl6_update_dst(&fl6, opt, &final); |
| 1254 | if (final_p) |
| 1255 | connected = 0; |
| 1256 | |
| 1257 | if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { |
| 1258 | fl6.flowi6_oif = np->mcast_oif; |
| 1259 | connected = 0; |
| 1260 | } else if (!fl6.flowi6_oif) |
| 1261 | fl6.flowi6_oif = np->ucast_oif; |
| 1262 | |
| 1263 | security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); |
| 1264 | |
| 1265 | dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p); |
| 1266 | if (IS_ERR(dst)) { |
| 1267 | err = PTR_ERR(dst); |
| 1268 | dst = NULL; |
| 1269 | goto out; |
| 1270 | } |
| 1271 | |
| 1272 | if (hlimit < 0) |
| 1273 | hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); |
| 1274 | |
| 1275 | if (tclass < 0) |
| 1276 | tclass = np->tclass; |
| 1277 | |
| 1278 | if (msg->msg_flags&MSG_CONFIRM) |
| 1279 | goto do_confirm; |
| 1280 | back_from_confirm: |
| 1281 | |
| 1282 | lock_sock(sk); |
| 1283 | if (unlikely(up->pending)) { |
| 1284 | /* The socket is already corked while preparing it. */ |
| 1285 | /* ... which is an evident application bug. --ANK */ |
| 1286 | release_sock(sk); |
| 1287 | |
| 1288 | LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); |
| 1289 | err = -EINVAL; |
| 1290 | goto out; |
| 1291 | } |
| 1292 | |
| 1293 | up->pending = AF_INET6; |
| 1294 | |
| 1295 | do_append_data: |
| 1296 | if (dontfrag < 0) |
| 1297 | dontfrag = np->dontfrag; |
| 1298 | up->len += ulen; |
| 1299 | getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; |
| 1300 | err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen, |
| 1301 | sizeof(struct udphdr), hlimit, tclass, opt, &fl6, |
| 1302 | (struct rt6_info*)dst, |
| 1303 | corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag); |
| 1304 | if (err) |
| 1305 | udp_v6_flush_pending_frames(sk); |
| 1306 | else if (!corkreq) |
| 1307 | err = udp_v6_push_pending_frames(sk); |
| 1308 | else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) |
| 1309 | up->pending = 0; |
| 1310 | |
| 1311 | if (dst) { |
| 1312 | if (connected) { |
| 1313 | ip6_dst_store(sk, dst, |
| 1314 | ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ? |
| 1315 | &sk->sk_v6_daddr : NULL, |
| 1316 | #ifdef CONFIG_IPV6_SUBTREES |
| 1317 | ipv6_addr_equal(&fl6.saddr, &np->saddr) ? |
| 1318 | &np->saddr : |
| 1319 | #endif |
| 1320 | NULL); |
| 1321 | } else { |
| 1322 | dst_release(dst); |
| 1323 | } |
| 1324 | dst = NULL; |
| 1325 | } |
| 1326 | |
| 1327 | if (err > 0) |
| 1328 | err = np->recverr ? net_xmit_errno(err) : 0; |
| 1329 | release_sock(sk); |
| 1330 | out: |
| 1331 | dst_release(dst); |
| 1332 | fl6_sock_release(flowlabel); |
| 1333 | if (!err) |
| 1334 | return len; |
| 1335 | /* |
| 1336 | * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting |
| 1337 | * ENOBUFS might not be good (it's not tunable per se), but otherwise |
| 1338 | * we don't have a good statistic (IpOutDiscards but it can be too many |
| 1339 | * things). We could add another new stat but at least for now that |
| 1340 | * seems like overkill. |
| 1341 | */ |
| 1342 | if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { |
| 1343 | UDP6_INC_STATS_USER(sock_net(sk), |
| 1344 | UDP_MIB_SNDBUFERRORS, is_udplite); |
| 1345 | } |
| 1346 | return err; |
| 1347 | |
| 1348 | do_confirm: |
| 1349 | dst_confirm(dst); |
| 1350 | if (!(msg->msg_flags&MSG_PROBE) || len) |
| 1351 | goto back_from_confirm; |
| 1352 | err = 0; |
| 1353 | goto out; |
| 1354 | } |
| 1355 | |
| 1356 | void udpv6_destroy_sock(struct sock *sk) |
| 1357 | { |
| 1358 | struct udp_sock *up = udp_sk(sk); |
| 1359 | lock_sock(sk); |
| 1360 | udp_v6_flush_pending_frames(sk); |
| 1361 | release_sock(sk); |
| 1362 | |
| 1363 | if (static_key_false(&udpv6_encap_needed) && up->encap_type) { |
| 1364 | void (*encap_destroy)(struct sock *sk); |
| 1365 | encap_destroy = ACCESS_ONCE(up->encap_destroy); |
| 1366 | if (encap_destroy) |
| 1367 | encap_destroy(sk); |
| 1368 | } |
| 1369 | |
| 1370 | inet6_destroy_sock(sk); |
| 1371 | } |
| 1372 | |
| 1373 | /* |
| 1374 | * Socket option code for UDP |
| 1375 | */ |
| 1376 | int udpv6_setsockopt(struct sock *sk, int level, int optname, |
| 1377 | char __user *optval, unsigned int optlen) |
| 1378 | { |
| 1379 | if (level == SOL_UDP || level == SOL_UDPLITE) |
| 1380 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, |
| 1381 | udp_v6_push_pending_frames); |
| 1382 | return ipv6_setsockopt(sk, level, optname, optval, optlen); |
| 1383 | } |
| 1384 | |
| 1385 | #ifdef CONFIG_COMPAT |
| 1386 | int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, |
| 1387 | char __user *optval, unsigned int optlen) |
| 1388 | { |
| 1389 | if (level == SOL_UDP || level == SOL_UDPLITE) |
| 1390 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, |
| 1391 | udp_v6_push_pending_frames); |
| 1392 | return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); |
| 1393 | } |
| 1394 | #endif |
| 1395 | |
| 1396 | int udpv6_getsockopt(struct sock *sk, int level, int optname, |
| 1397 | char __user *optval, int __user *optlen) |
| 1398 | { |
| 1399 | if (level == SOL_UDP || level == SOL_UDPLITE) |
| 1400 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); |
| 1401 | return ipv6_getsockopt(sk, level, optname, optval, optlen); |
| 1402 | } |
| 1403 | |
| 1404 | #ifdef CONFIG_COMPAT |
| 1405 | int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, |
| 1406 | char __user *optval, int __user *optlen) |
| 1407 | { |
| 1408 | if (level == SOL_UDP || level == SOL_UDPLITE) |
| 1409 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); |
| 1410 | return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); |
| 1411 | } |
| 1412 | #endif |
| 1413 | |
| 1414 | static const struct inet6_protocol udpv6_protocol = { |
| 1415 | .handler = udpv6_rcv, |
| 1416 | .err_handler = udpv6_err, |
| 1417 | .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, |
| 1418 | }; |
| 1419 | |
| 1420 | /* ------------------------------------------------------------------------ */ |
| 1421 | #ifdef CONFIG_PROC_FS |
| 1422 | int udp6_seq_show(struct seq_file *seq, void *v) |
| 1423 | { |
| 1424 | if (v == SEQ_START_TOKEN) { |
| 1425 | seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); |
| 1426 | } else { |
| 1427 | int bucket = ((struct udp_iter_state *)seq->private)->bucket; |
| 1428 | struct inet_sock *inet = inet_sk(v); |
| 1429 | __u16 srcp = ntohs(inet->inet_sport); |
| 1430 | __u16 destp = ntohs(inet->inet_dport); |
| 1431 | ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket); |
| 1432 | } |
| 1433 | return 0; |
| 1434 | } |
| 1435 | |
| 1436 | static const struct file_operations udp6_afinfo_seq_fops = { |
| 1437 | .owner = THIS_MODULE, |
| 1438 | .open = udp_seq_open, |
| 1439 | .read = seq_read, |
| 1440 | .llseek = seq_lseek, |
| 1441 | .release = seq_release_net |
| 1442 | }; |
| 1443 | |
| 1444 | static struct udp_seq_afinfo udp6_seq_afinfo = { |
| 1445 | .name = "udp6", |
| 1446 | .family = AF_INET6, |
| 1447 | .udp_table = &udp_table, |
| 1448 | .seq_fops = &udp6_afinfo_seq_fops, |
| 1449 | .seq_ops = { |
| 1450 | .show = udp6_seq_show, |
| 1451 | }, |
| 1452 | }; |
| 1453 | |
| 1454 | int __net_init udp6_proc_init(struct net *net) |
| 1455 | { |
| 1456 | return udp_proc_register(net, &udp6_seq_afinfo); |
| 1457 | } |
| 1458 | |
| 1459 | void udp6_proc_exit(struct net *net) { |
| 1460 | udp_proc_unregister(net, &udp6_seq_afinfo); |
| 1461 | } |
| 1462 | #endif /* CONFIG_PROC_FS */ |
| 1463 | |
| 1464 | void udp_v6_clear_sk(struct sock *sk, int size) |
| 1465 | { |
| 1466 | struct inet_sock *inet = inet_sk(sk); |
| 1467 | |
| 1468 | /* we do not want to clear pinet6 field, because of RCU lookups */ |
| 1469 | sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6)); |
| 1470 | |
| 1471 | size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6); |
| 1472 | memset(&inet->pinet6 + 1, 0, size); |
| 1473 | } |
| 1474 | |
| 1475 | /* ------------------------------------------------------------------------ */ |
| 1476 | |
| 1477 | struct proto udpv6_prot = { |
| 1478 | .name = "UDPv6", |
| 1479 | .owner = THIS_MODULE, |
| 1480 | .close = udp_lib_close, |
| 1481 | .connect = ip6_datagram_connect, |
| 1482 | .disconnect = udp_disconnect, |
| 1483 | .ioctl = udp_ioctl, |
| 1484 | .destroy = udpv6_destroy_sock, |
| 1485 | .setsockopt = udpv6_setsockopt, |
| 1486 | .getsockopt = udpv6_getsockopt, |
| 1487 | .sendmsg = udpv6_sendmsg, |
| 1488 | .recvmsg = udpv6_recvmsg, |
| 1489 | .backlog_rcv = __udpv6_queue_rcv_skb, |
| 1490 | .hash = udp_lib_hash, |
| 1491 | .unhash = udp_lib_unhash, |
| 1492 | .rehash = udp_v6_rehash, |
| 1493 | .get_port = udp_v6_get_port, |
| 1494 | .memory_allocated = &udp_memory_allocated, |
| 1495 | .sysctl_mem = sysctl_udp_mem, |
| 1496 | .sysctl_wmem = &sysctl_udp_wmem_min, |
| 1497 | .sysctl_rmem = &sysctl_udp_rmem_min, |
| 1498 | .obj_size = sizeof(struct udp6_sock), |
| 1499 | .slab_flags = SLAB_DESTROY_BY_RCU, |
| 1500 | .h.udp_table = &udp_table, |
| 1501 | #ifdef CONFIG_COMPAT |
| 1502 | .compat_setsockopt = compat_udpv6_setsockopt, |
| 1503 | .compat_getsockopt = compat_udpv6_getsockopt, |
| 1504 | #endif |
| 1505 | .clear_sk = udp_v6_clear_sk, |
| 1506 | }; |
| 1507 | |
| 1508 | static struct inet_protosw udpv6_protosw = { |
| 1509 | .type = SOCK_DGRAM, |
| 1510 | .protocol = IPPROTO_UDP, |
| 1511 | .prot = &udpv6_prot, |
| 1512 | .ops = &inet6_dgram_ops, |
| 1513 | .flags = INET_PROTOSW_PERMANENT, |
| 1514 | }; |
| 1515 | |
| 1516 | |
| 1517 | int __init udpv6_init(void) |
| 1518 | { |
| 1519 | int ret; |
| 1520 | |
| 1521 | ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); |
| 1522 | if (ret) |
| 1523 | goto out; |
| 1524 | |
| 1525 | ret = inet6_register_protosw(&udpv6_protosw); |
| 1526 | if (ret) |
| 1527 | goto out_udpv6_protocol; |
| 1528 | out: |
| 1529 | return ret; |
| 1530 | |
| 1531 | out_udpv6_protocol: |
| 1532 | inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); |
| 1533 | goto out; |
| 1534 | } |
| 1535 | |
| 1536 | void udpv6_exit(void) |
| 1537 | { |
| 1538 | inet6_unregister_protosw(&udpv6_protosw); |
| 1539 | inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); |
| 1540 | } |