1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The User Datagram Protocol (UDP). 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 11 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 12 * Hirokazu Takahashi, <taka@valinux.co.jp> 13 * 14 * Fixes: 15 * Alan Cox : verify_area() calls 16 * Alan Cox : stopped close while in use off icmp 17 * messages. Not a fix but a botch that 18 * for udp at least is 'valid'. 19 * Alan Cox : Fixed icmp handling properly 20 * Alan Cox : Correct error for oversized datagrams 21 * Alan Cox : Tidied select() semantics. 22 * Alan Cox : udp_err() fixed properly, also now 23 * select and read wake correctly on errors 24 * Alan Cox : udp_send verify_area moved to avoid mem leak 25 * Alan Cox : UDP can count its memory 26 * Alan Cox : send to an unknown connection causes 27 * an ECONNREFUSED off the icmp, but 28 * does NOT close. 29 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 31 * bug no longer crashes it. 32 * Fred Van Kempen : Net2e support for sk->broadcast. 33 * Alan Cox : Uses skb_free_datagram 34 * Alan Cox : Added get/set sockopt support. 35 * Alan Cox : Broadcasting without option set returns EACCES. 36 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 37 * Alan Cox : Use ip_tos and ip_ttl 38 * Alan Cox : SNMP Mibs 39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 40 * Matt Dillon : UDP length checks. 41 * Alan Cox : Smarter af_inet used properly. 42 * Alan Cox : Use new kernel side addressing. 43 * Alan Cox : Incorrect return on truncated datagram receive. 44 * Arnt Gulbrandsen : New udp_send and stuff 45 * Alan Cox : Cache last socket 46 * Alan Cox : Route cache 47 * Jon Peatfield : Minor efficiency fix to sendto(). 48 * Mike Shaver : RFC1122 checks. 49 * Alan Cox : Nonblocking error fix. 50 * Willy Konynenberg : Transparent proxying support. 51 * Mike McLagan : Routing by source 52 * David S. Miller : New socket lookup architecture. 53 * Last socket cache retained as it 54 * does have a high hit rate. 55 * Olaf Kirch : Don't linearise iovec on sendmsg. 56 * Andi Kleen : Some cleanups, cache destination entry 57 * for connect. 58 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 59 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 60 * return ENOTCONN for unconnected sockets (POSIX) 61 * Janos Farkas : don't deliver multi/broadcasts to a different 62 * bound-to-device socket 63 * Hirokazu Takahashi : HW checksumming for outgoing UDP 64 * datagrams. 65 * Hirokazu Takahashi : sendfile() on UDP works now. 66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 69 * a single port at the same time. 70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 71 * James Chapman : Add L2TP encapsulation type. 72 * 73 * 74 * This program is free software; you can redistribute it and/or 75 * modify it under the terms of the GNU General Public License 76 * as published by the Free Software Foundation; either version 77 * 2 of the License, or (at your option) any later version. 78 */ 79 80 #include <asm/system.h> 81 #include <asm/uaccess.h> 82 #include <asm/ioctls.h> 83 #include <linux/bootmem.h> 84 #include <linux/highmem.h> 85 #include <linux/swap.h> 86 #include <linux/types.h> 87 #include <linux/fcntl.h> 88 #include <linux/module.h> 89 #include <linux/socket.h> 90 #include <linux/sockios.h> 91 #include <linux/igmp.h> 92 #include <linux/in.h> 93 #include <linux/errno.h> 94 #include <linux/timer.h> 95 #include <linux/mm.h> 96 #include <linux/inet.h> 97 #include <linux/netdevice.h> 98 #include <net/tcp_states.h> 99 #include <linux/skbuff.h> 100 #include <linux/proc_fs.h> 101 #include <linux/seq_file.h> 102 #include <net/net_namespace.h> 103 #include <net/icmp.h> 104 #include <net/route.h> 105 #include <net/checksum.h> 106 #include <net/xfrm.h> 107 #include "udp_impl.h" 108 109 struct udp_table udp_table; 110 EXPORT_SYMBOL(udp_table); 111 112 int sysctl_udp_mem[3] __read_mostly; 113 EXPORT_SYMBOL(sysctl_udp_mem); 114 115 int sysctl_udp_rmem_min __read_mostly; 116 EXPORT_SYMBOL(sysctl_udp_rmem_min); 117 118 int sysctl_udp_wmem_min __read_mostly; 119 EXPORT_SYMBOL(sysctl_udp_wmem_min); 120 121 atomic_t udp_memory_allocated; 122 EXPORT_SYMBOL(udp_memory_allocated); 123 124 #define PORTS_PER_CHAIN (65536 / UDP_HTABLE_SIZE) 125 126 static int udp_lib_lport_inuse(struct net *net, __u16 num, 127 const struct udp_hslot *hslot, 128 unsigned long *bitmap, 129 struct sock *sk, 130 int (*saddr_comp)(const struct sock *sk1, 131 const struct sock *sk2)) 132 { 133 struct sock *sk2; 134 struct hlist_nulls_node *node; 135 136 sk_nulls_for_each(sk2, node, &hslot->head) 137 if (net_eq(sock_net(sk2), net) && 138 sk2 != sk && 139 (bitmap || sk2->sk_hash == num) && 140 (!sk2->sk_reuse || !sk->sk_reuse) && 141 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if 142 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 143 (*saddr_comp)(sk, sk2)) { 144 if (bitmap) 145 __set_bit(sk2->sk_hash / UDP_HTABLE_SIZE, 146 bitmap); 147 else 148 return 1; 149 } 150 return 0; 151 } 152 153 /** 154 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 155 * 156 * @sk: socket struct in question 157 * @snum: port number to look up 158 * @saddr_comp: AF-dependent comparison of bound local IP addresses 159 */ 160 int udp_lib_get_port(struct sock *sk, unsigned short snum, 161 int (*saddr_comp)(const struct sock *sk1, 162 const struct sock *sk2)) 163 { 164 struct udp_hslot *hslot; 165 struct udp_table *udptable = sk->sk_prot->h.udp_table; 166 int error = 1; 167 struct net *net = sock_net(sk); 168 169 if (!snum) { 170 int low, high, remaining; 171 unsigned rand; 172 unsigned short first, last; 173 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 174 175 inet_get_local_port_range(&low, &high); 176 remaining = (high - low) + 1; 177 178 rand = net_random(); 179 first = (((u64)rand * remaining) >> 32) + low; 180 /* 181 * force rand to be an odd multiple of UDP_HTABLE_SIZE 182 */ 183 rand = (rand | 1) * UDP_HTABLE_SIZE; 184 for (last = first + UDP_HTABLE_SIZE; first != last; first++) { 185 hslot = &udptable->hash[udp_hashfn(net, first)]; 186 bitmap_zero(bitmap, PORTS_PER_CHAIN); 187 spin_lock_bh(&hslot->lock); 188 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 189 saddr_comp); 190 191 snum = first; 192 /* 193 * Iterate on all possible values of snum for this hash. 194 * Using steps of an odd multiple of UDP_HTABLE_SIZE 195 * give us randomization and full range coverage. 196 */ 197 do { 198 if (low <= snum && snum <= high && 199 !test_bit(snum / UDP_HTABLE_SIZE, bitmap)) 200 goto found; 201 snum += rand; 202 } while (snum != first); 203 spin_unlock_bh(&hslot->lock); 204 } 205 goto fail; 206 } else { 207 hslot = &udptable->hash[udp_hashfn(net, snum)]; 208 spin_lock_bh(&hslot->lock); 209 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, saddr_comp)) 210 goto fail_unlock; 211 } 212 found: 213 inet_sk(sk)->num = snum; 214 sk->sk_hash = snum; 215 if (sk_unhashed(sk)) { 216 sk_nulls_add_node_rcu(sk, &hslot->head); 217 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 218 } 219 error = 0; 220 fail_unlock: 221 spin_unlock_bh(&hslot->lock); 222 fail: 223 return error; 224 } 225 EXPORT_SYMBOL(udp_lib_get_port); 226 227 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 228 { 229 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 230 231 return (!ipv6_only_sock(sk2) && 232 (!inet1->rcv_saddr || !inet2->rcv_saddr || 233 inet1->rcv_saddr == inet2->rcv_saddr)); 234 } 235 236 int udp_v4_get_port(struct sock *sk, unsigned short snum) 237 { 238 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal); 239 } 240 241 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, 242 unsigned short hnum, 243 __be16 sport, __be32 daddr, __be16 dport, int dif) 244 { 245 int score = -1; 246 247 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum && 248 !ipv6_only_sock(sk)) { 249 struct inet_sock *inet = inet_sk(sk); 250 251 score = (sk->sk_family == PF_INET ? 1 : 0); 252 if (inet->rcv_saddr) { 253 if (inet->rcv_saddr != daddr) 254 return -1; 255 score += 2; 256 } 257 if (inet->daddr) { 258 if (inet->daddr != saddr) 259 return -1; 260 score += 2; 261 } 262 if (inet->dport) { 263 if (inet->dport != sport) 264 return -1; 265 score += 2; 266 } 267 if (sk->sk_bound_dev_if) { 268 if (sk->sk_bound_dev_if != dif) 269 return -1; 270 score += 2; 271 } 272 } 273 return score; 274 } 275 276 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 277 * harder than this. -DaveM 278 */ 279 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 280 __be16 sport, __be32 daddr, __be16 dport, 281 int dif, struct udp_table *udptable) 282 { 283 struct sock *sk, *result; 284 struct hlist_nulls_node *node; 285 unsigned short hnum = ntohs(dport); 286 unsigned int hash = udp_hashfn(net, hnum); 287 struct udp_hslot *hslot = &udptable->hash[hash]; 288 int score, badness; 289 290 rcu_read_lock(); 291 begin: 292 result = NULL; 293 badness = -1; 294 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 295 score = compute_score(sk, net, saddr, hnum, sport, 296 daddr, dport, dif); 297 if (score > badness) { 298 result = sk; 299 badness = score; 300 } 301 } 302 /* 303 * if the nulls value we got at the end of this lookup is 304 * not the expected one, we must restart lookup. 305 * We probably met an item that was moved to another chain. 306 */ 307 if (get_nulls_value(node) != hash) 308 goto begin; 309 310 if (result) { 311 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) 312 result = NULL; 313 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 314 daddr, dport, dif) < badness)) { 315 sock_put(result); 316 goto begin; 317 } 318 } 319 rcu_read_unlock(); 320 return result; 321 } 322 323 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 324 __be16 sport, __be16 dport, 325 struct udp_table *udptable) 326 { 327 struct sock *sk; 328 const struct iphdr *iph = ip_hdr(skb); 329 330 if (unlikely(sk = skb_steal_sock(skb))) 331 return sk; 332 else 333 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 334 iph->daddr, dport, inet_iif(skb), 335 udptable); 336 } 337 338 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 339 __be32 daddr, __be16 dport, int dif) 340 { 341 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 342 } 343 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 344 345 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 346 __be16 loc_port, __be32 loc_addr, 347 __be16 rmt_port, __be32 rmt_addr, 348 int dif) 349 { 350 struct hlist_nulls_node *node; 351 struct sock *s = sk; 352 unsigned short hnum = ntohs(loc_port); 353 354 sk_nulls_for_each_from(s, node) { 355 struct inet_sock *inet = inet_sk(s); 356 357 if (!net_eq(sock_net(s), net) || 358 s->sk_hash != hnum || 359 (inet->daddr && inet->daddr != rmt_addr) || 360 (inet->dport != rmt_port && inet->dport) || 361 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) || 362 ipv6_only_sock(s) || 363 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 364 continue; 365 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 366 continue; 367 goto found; 368 } 369 s = NULL; 370 found: 371 return s; 372 } 373 374 /* 375 * This routine is called by the ICMP module when it gets some 376 * sort of error condition. If err < 0 then the socket should 377 * be closed and the error returned to the user. If err > 0 378 * it's just the icmp type << 8 | icmp code. 379 * Header points to the ip header of the error packet. We move 380 * on past this. Then (as it used to claim before adjustment) 381 * header points to the first 8 bytes of the udp header. We need 382 * to find the appropriate port. 383 */ 384 385 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 386 { 387 struct inet_sock *inet; 388 struct iphdr *iph = (struct iphdr *)skb->data; 389 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 390 const int type = icmp_hdr(skb)->type; 391 const int code = icmp_hdr(skb)->code; 392 struct sock *sk; 393 int harderr; 394 int err; 395 struct net *net = dev_net(skb->dev); 396 397 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 398 iph->saddr, uh->source, skb->dev->ifindex, udptable); 399 if (sk == NULL) { 400 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 401 return; /* No socket for error */ 402 } 403 404 err = 0; 405 harderr = 0; 406 inet = inet_sk(sk); 407 408 switch (type) { 409 default: 410 case ICMP_TIME_EXCEEDED: 411 err = EHOSTUNREACH; 412 break; 413 case ICMP_SOURCE_QUENCH: 414 goto out; 415 case ICMP_PARAMETERPROB: 416 err = EPROTO; 417 harderr = 1; 418 break; 419 case ICMP_DEST_UNREACH: 420 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 421 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 422 err = EMSGSIZE; 423 harderr = 1; 424 break; 425 } 426 goto out; 427 } 428 err = EHOSTUNREACH; 429 if (code <= NR_ICMP_UNREACH) { 430 harderr = icmp_err_convert[code].fatal; 431 err = icmp_err_convert[code].errno; 432 } 433 break; 434 } 435 436 /* 437 * RFC1122: OK. Passes ICMP errors back to application, as per 438 * 4.1.3.3. 439 */ 440 if (!inet->recverr) { 441 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 442 goto out; 443 } else { 444 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 445 } 446 sk->sk_err = err; 447 sk->sk_error_report(sk); 448 out: 449 sock_put(sk); 450 } 451 452 void udp_err(struct sk_buff *skb, u32 info) 453 { 454 __udp4_lib_err(skb, info, &udp_table); 455 } 456 457 /* 458 * Throw away all pending data and cancel the corking. Socket is locked. 459 */ 460 void udp_flush_pending_frames(struct sock *sk) 461 { 462 struct udp_sock *up = udp_sk(sk); 463 464 if (up->pending) { 465 up->len = 0; 466 up->pending = 0; 467 ip_flush_pending_frames(sk); 468 } 469 } 470 EXPORT_SYMBOL(udp_flush_pending_frames); 471 472 /** 473 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 474 * @sk: socket we are sending on 475 * @skb: sk_buff containing the filled-in UDP header 476 * (checksum field must be zeroed out) 477 */ 478 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 479 __be32 src, __be32 dst, int len) 480 { 481 unsigned int offset; 482 struct udphdr *uh = udp_hdr(skb); 483 __wsum csum = 0; 484 485 if (skb_queue_len(&sk->sk_write_queue) == 1) { 486 /* 487 * Only one fragment on the socket. 488 */ 489 skb->csum_start = skb_transport_header(skb) - skb->head; 490 skb->csum_offset = offsetof(struct udphdr, check); 491 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 492 } else { 493 /* 494 * HW-checksum won't work as there are two or more 495 * fragments on the socket so that all csums of sk_buffs 496 * should be together 497 */ 498 offset = skb_transport_offset(skb); 499 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 500 501 skb->ip_summed = CHECKSUM_NONE; 502 503 skb_queue_walk(&sk->sk_write_queue, skb) { 504 csum = csum_add(csum, skb->csum); 505 } 506 507 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 508 if (uh->check == 0) 509 uh->check = CSUM_MANGLED_0; 510 } 511 } 512 513 /* 514 * Push out all pending data as one UDP datagram. Socket is locked. 515 */ 516 static int udp_push_pending_frames(struct sock *sk) 517 { 518 struct udp_sock *up = udp_sk(sk); 519 struct inet_sock *inet = inet_sk(sk); 520 struct flowi *fl = &inet->cork.fl; 521 struct sk_buff *skb; 522 struct udphdr *uh; 523 int err = 0; 524 int is_udplite = IS_UDPLITE(sk); 525 __wsum csum = 0; 526 527 /* Grab the skbuff where UDP header space exists. */ 528 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 529 goto out; 530 531 /* 532 * Create a UDP header 533 */ 534 uh = udp_hdr(skb); 535 uh->source = fl->fl_ip_sport; 536 uh->dest = fl->fl_ip_dport; 537 uh->len = htons(up->len); 538 uh->check = 0; 539 540 if (is_udplite) /* UDP-Lite */ 541 csum = udplite_csum_outgoing(sk, skb); 542 543 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 544 545 skb->ip_summed = CHECKSUM_NONE; 546 goto send; 547 548 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 549 550 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len); 551 goto send; 552 553 } else /* `normal' UDP */ 554 csum = udp_csum_outgoing(sk, skb); 555 556 /* add protocol-dependent pseudo-header */ 557 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 558 sk->sk_protocol, csum); 559 if (uh->check == 0) 560 uh->check = CSUM_MANGLED_0; 561 562 send: 563 err = ip_push_pending_frames(sk); 564 if (err) { 565 if (err == -ENOBUFS && !inet->recverr) { 566 UDP_INC_STATS_USER(sock_net(sk), 567 UDP_MIB_SNDBUFERRORS, is_udplite); 568 err = 0; 569 } 570 } else 571 UDP_INC_STATS_USER(sock_net(sk), 572 UDP_MIB_OUTDATAGRAMS, is_udplite); 573 out: 574 up->len = 0; 575 up->pending = 0; 576 return err; 577 } 578 579 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 580 size_t len) 581 { 582 struct inet_sock *inet = inet_sk(sk); 583 struct udp_sock *up = udp_sk(sk); 584 int ulen = len; 585 struct ipcm_cookie ipc; 586 struct rtable *rt = NULL; 587 int free = 0; 588 int connected = 0; 589 __be32 daddr, faddr, saddr; 590 __be16 dport; 591 u8 tos; 592 int err, is_udplite = IS_UDPLITE(sk); 593 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 594 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 595 596 if (len > 0xFFFF) 597 return -EMSGSIZE; 598 599 /* 600 * Check the flags. 601 */ 602 603 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 604 return -EOPNOTSUPP; 605 606 ipc.opt = NULL; 607 ipc.shtx.flags = 0; 608 609 if (up->pending) { 610 /* 611 * There are pending frames. 612 * The socket lock must be held while it's corked. 613 */ 614 lock_sock(sk); 615 if (likely(up->pending)) { 616 if (unlikely(up->pending != AF_INET)) { 617 release_sock(sk); 618 return -EINVAL; 619 } 620 goto do_append_data; 621 } 622 release_sock(sk); 623 } 624 ulen += sizeof(struct udphdr); 625 626 /* 627 * Get and verify the address. 628 */ 629 if (msg->msg_name) { 630 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name; 631 if (msg->msg_namelen < sizeof(*usin)) 632 return -EINVAL; 633 if (usin->sin_family != AF_INET) { 634 if (usin->sin_family != AF_UNSPEC) 635 return -EAFNOSUPPORT; 636 } 637 638 daddr = usin->sin_addr.s_addr; 639 dport = usin->sin_port; 640 if (dport == 0) 641 return -EINVAL; 642 } else { 643 if (sk->sk_state != TCP_ESTABLISHED) 644 return -EDESTADDRREQ; 645 daddr = inet->daddr; 646 dport = inet->dport; 647 /* Open fast path for connected socket. 648 Route will not be used, if at least one option is set. 649 */ 650 connected = 1; 651 } 652 ipc.addr = inet->saddr; 653 654 ipc.oif = sk->sk_bound_dev_if; 655 err = sock_tx_timestamp(msg, sk, &ipc.shtx); 656 if (err) 657 return err; 658 if (msg->msg_controllen) { 659 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 660 if (err) 661 return err; 662 if (ipc.opt) 663 free = 1; 664 connected = 0; 665 } 666 if (!ipc.opt) 667 ipc.opt = inet->opt; 668 669 saddr = ipc.addr; 670 ipc.addr = faddr = daddr; 671 672 if (ipc.opt && ipc.opt->srr) { 673 if (!daddr) 674 return -EINVAL; 675 faddr = ipc.opt->faddr; 676 connected = 0; 677 } 678 tos = RT_TOS(inet->tos); 679 if (sock_flag(sk, SOCK_LOCALROUTE) || 680 (msg->msg_flags & MSG_DONTROUTE) || 681 (ipc.opt && ipc.opt->is_strictroute)) { 682 tos |= RTO_ONLINK; 683 connected = 0; 684 } 685 686 if (ipv4_is_multicast(daddr)) { 687 if (!ipc.oif) 688 ipc.oif = inet->mc_index; 689 if (!saddr) 690 saddr = inet->mc_addr; 691 connected = 0; 692 } 693 694 if (connected) 695 rt = (struct rtable *)sk_dst_check(sk, 0); 696 697 if (rt == NULL) { 698 struct flowi fl = { .oif = ipc.oif, 699 .mark = sk->sk_mark, 700 .nl_u = { .ip4_u = 701 { .daddr = faddr, 702 .saddr = saddr, 703 .tos = tos } }, 704 .proto = sk->sk_protocol, 705 .flags = inet_sk_flowi_flags(sk), 706 .uli_u = { .ports = 707 { .sport = inet->sport, 708 .dport = dport } } }; 709 struct net *net = sock_net(sk); 710 711 security_sk_classify_flow(sk, &fl); 712 err = ip_route_output_flow(net, &rt, &fl, sk, 1); 713 if (err) { 714 if (err == -ENETUNREACH) 715 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 716 goto out; 717 } 718 719 err = -EACCES; 720 if ((rt->rt_flags & RTCF_BROADCAST) && 721 !sock_flag(sk, SOCK_BROADCAST)) 722 goto out; 723 if (connected) 724 sk_dst_set(sk, dst_clone(&rt->u.dst)); 725 } 726 727 if (msg->msg_flags&MSG_CONFIRM) 728 goto do_confirm; 729 back_from_confirm: 730 731 saddr = rt->rt_src; 732 if (!ipc.addr) 733 daddr = ipc.addr = rt->rt_dst; 734 735 lock_sock(sk); 736 if (unlikely(up->pending)) { 737 /* The socket is already corked while preparing it. */ 738 /* ... which is an evident application bug. --ANK */ 739 release_sock(sk); 740 741 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 742 err = -EINVAL; 743 goto out; 744 } 745 /* 746 * Now cork the socket to pend data. 747 */ 748 inet->cork.fl.fl4_dst = daddr; 749 inet->cork.fl.fl_ip_dport = dport; 750 inet->cork.fl.fl4_src = saddr; 751 inet->cork.fl.fl_ip_sport = inet->sport; 752 up->pending = AF_INET; 753 754 do_append_data: 755 up->len += ulen; 756 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 757 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 758 sizeof(struct udphdr), &ipc, &rt, 759 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 760 if (err) 761 udp_flush_pending_frames(sk); 762 else if (!corkreq) 763 err = udp_push_pending_frames(sk); 764 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 765 up->pending = 0; 766 release_sock(sk); 767 768 out: 769 ip_rt_put(rt); 770 if (free) 771 kfree(ipc.opt); 772 if (!err) 773 return len; 774 /* 775 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 776 * ENOBUFS might not be good (it's not tunable per se), but otherwise 777 * we don't have a good statistic (IpOutDiscards but it can be too many 778 * things). We could add another new stat but at least for now that 779 * seems like overkill. 780 */ 781 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 782 UDP_INC_STATS_USER(sock_net(sk), 783 UDP_MIB_SNDBUFERRORS, is_udplite); 784 } 785 return err; 786 787 do_confirm: 788 dst_confirm(&rt->u.dst); 789 if (!(msg->msg_flags&MSG_PROBE) || len) 790 goto back_from_confirm; 791 err = 0; 792 goto out; 793 } 794 EXPORT_SYMBOL(udp_sendmsg); 795 796 int udp_sendpage(struct sock *sk, struct page *page, int offset, 797 size_t size, int flags) 798 { 799 struct udp_sock *up = udp_sk(sk); 800 int ret; 801 802 if (!up->pending) { 803 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 804 805 /* Call udp_sendmsg to specify destination address which 806 * sendpage interface can't pass. 807 * This will succeed only when the socket is connected. 808 */ 809 ret = udp_sendmsg(NULL, sk, &msg, 0); 810 if (ret < 0) 811 return ret; 812 } 813 814 lock_sock(sk); 815 816 if (unlikely(!up->pending)) { 817 release_sock(sk); 818 819 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 820 return -EINVAL; 821 } 822 823 ret = ip_append_page(sk, page, offset, size, flags); 824 if (ret == -EOPNOTSUPP) { 825 release_sock(sk); 826 return sock_no_sendpage(sk->sk_socket, page, offset, 827 size, flags); 828 } 829 if (ret < 0) { 830 udp_flush_pending_frames(sk); 831 goto out; 832 } 833 834 up->len += size; 835 if (!(up->corkflag || (flags&MSG_MORE))) 836 ret = udp_push_pending_frames(sk); 837 if (!ret) 838 ret = size; 839 out: 840 release_sock(sk); 841 return ret; 842 } 843 844 845 /** 846 * first_packet_length - return length of first packet in receive queue 847 * @sk: socket 848 * 849 * Drops all bad checksum frames, until a valid one is found. 850 * Returns the length of found skb, or 0 if none is found. 851 */ 852 static unsigned int first_packet_length(struct sock *sk) 853 { 854 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 855 struct sk_buff *skb; 856 unsigned int res; 857 858 __skb_queue_head_init(&list_kill); 859 860 spin_lock_bh(&rcvq->lock); 861 while ((skb = skb_peek(rcvq)) != NULL && 862 udp_lib_checksum_complete(skb)) { 863 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 864 IS_UDPLITE(sk)); 865 __skb_unlink(skb, rcvq); 866 __skb_queue_tail(&list_kill, skb); 867 } 868 res = skb ? skb->len : 0; 869 spin_unlock_bh(&rcvq->lock); 870 871 if (!skb_queue_empty(&list_kill)) { 872 lock_sock(sk); 873 __skb_queue_purge(&list_kill); 874 sk_mem_reclaim_partial(sk); 875 release_sock(sk); 876 } 877 return res; 878 } 879 880 /* 881 * IOCTL requests applicable to the UDP protocol 882 */ 883 884 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 885 { 886 switch (cmd) { 887 case SIOCOUTQ: 888 { 889 int amount = sk_wmem_alloc_get(sk); 890 891 return put_user(amount, (int __user *)arg); 892 } 893 894 case SIOCINQ: 895 { 896 unsigned int amount = first_packet_length(sk); 897 898 if (amount) 899 /* 900 * We will only return the amount 901 * of this packet since that is all 902 * that will be read. 903 */ 904 amount -= sizeof(struct udphdr); 905 906 return put_user(amount, (int __user *)arg); 907 } 908 909 default: 910 return -ENOIOCTLCMD; 911 } 912 913 return 0; 914 } 915 EXPORT_SYMBOL(udp_ioctl); 916 917 /* 918 * This should be easy, if there is something there we 919 * return it, otherwise we block. 920 */ 921 922 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 923 size_t len, int noblock, int flags, int *addr_len) 924 { 925 struct inet_sock *inet = inet_sk(sk); 926 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 927 struct sk_buff *skb; 928 unsigned int ulen, copied; 929 int peeked; 930 int err; 931 int is_udplite = IS_UDPLITE(sk); 932 933 /* 934 * Check any passed addresses 935 */ 936 if (addr_len) 937 *addr_len = sizeof(*sin); 938 939 if (flags & MSG_ERRQUEUE) 940 return ip_recv_error(sk, msg, len); 941 942 try_again: 943 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 944 &peeked, &err); 945 if (!skb) 946 goto out; 947 948 ulen = skb->len - sizeof(struct udphdr); 949 copied = len; 950 if (copied > ulen) 951 copied = ulen; 952 else if (copied < ulen) 953 msg->msg_flags |= MSG_TRUNC; 954 955 /* 956 * If checksum is needed at all, try to do it while copying the 957 * data. If the data is truncated, or if we only want a partial 958 * coverage checksum (UDP-Lite), do it before the copy. 959 */ 960 961 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 962 if (udp_lib_checksum_complete(skb)) 963 goto csum_copy_err; 964 } 965 966 if (skb_csum_unnecessary(skb)) 967 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 968 msg->msg_iov, copied); 969 else { 970 err = skb_copy_and_csum_datagram_iovec(skb, 971 sizeof(struct udphdr), 972 msg->msg_iov); 973 974 if (err == -EINVAL) 975 goto csum_copy_err; 976 } 977 978 if (err) 979 goto out_free; 980 981 if (!peeked) 982 UDP_INC_STATS_USER(sock_net(sk), 983 UDP_MIB_INDATAGRAMS, is_udplite); 984 985 sock_recv_timestamp(msg, sk, skb); 986 987 /* Copy the address. */ 988 if (sin) { 989 sin->sin_family = AF_INET; 990 sin->sin_port = udp_hdr(skb)->source; 991 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 992 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 993 } 994 if (inet->cmsg_flags) 995 ip_cmsg_recv(msg, skb); 996 997 err = copied; 998 if (flags & MSG_TRUNC) 999 err = ulen; 1000 1001 out_free: 1002 skb_free_datagram_locked(sk, skb); 1003 out: 1004 return err; 1005 1006 csum_copy_err: 1007 lock_sock(sk); 1008 if (!skb_kill_datagram(sk, skb, flags)) 1009 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1010 release_sock(sk); 1011 1012 if (noblock) 1013 return -EAGAIN; 1014 goto try_again; 1015 } 1016 1017 1018 int udp_disconnect(struct sock *sk, int flags) 1019 { 1020 struct inet_sock *inet = inet_sk(sk); 1021 /* 1022 * 1003.1g - break association. 1023 */ 1024 1025 sk->sk_state = TCP_CLOSE; 1026 inet->daddr = 0; 1027 inet->dport = 0; 1028 sk->sk_bound_dev_if = 0; 1029 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1030 inet_reset_saddr(sk); 1031 1032 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1033 sk->sk_prot->unhash(sk); 1034 inet->sport = 0; 1035 } 1036 sk_dst_reset(sk); 1037 return 0; 1038 } 1039 EXPORT_SYMBOL(udp_disconnect); 1040 1041 void udp_lib_unhash(struct sock *sk) 1042 { 1043 if (sk_hashed(sk)) { 1044 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1045 unsigned int hash = udp_hashfn(sock_net(sk), sk->sk_hash); 1046 struct udp_hslot *hslot = &udptable->hash[hash]; 1047 1048 spin_lock_bh(&hslot->lock); 1049 if (sk_nulls_del_node_init_rcu(sk)) { 1050 inet_sk(sk)->num = 0; 1051 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1052 } 1053 spin_unlock_bh(&hslot->lock); 1054 } 1055 } 1056 EXPORT_SYMBOL(udp_lib_unhash); 1057 1058 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1059 { 1060 int is_udplite = IS_UDPLITE(sk); 1061 int rc; 1062 1063 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) { 1064 /* Note that an ENOMEM error is charged twice */ 1065 if (rc == -ENOMEM) { 1066 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1067 is_udplite); 1068 atomic_inc(&sk->sk_drops); 1069 } 1070 goto drop; 1071 } 1072 1073 return 0; 1074 1075 drop: 1076 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1077 kfree_skb(skb); 1078 return -1; 1079 } 1080 1081 /* returns: 1082 * -1: error 1083 * 0: success 1084 * >0: "udp encap" protocol resubmission 1085 * 1086 * Note that in the success and error cases, the skb is assumed to 1087 * have either been requeued or freed. 1088 */ 1089 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1090 { 1091 struct udp_sock *up = udp_sk(sk); 1092 int rc; 1093 int is_udplite = IS_UDPLITE(sk); 1094 1095 /* 1096 * Charge it to the socket, dropping if the queue is full. 1097 */ 1098 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1099 goto drop; 1100 nf_reset(skb); 1101 1102 if (up->encap_type) { 1103 /* 1104 * This is an encapsulation socket so pass the skb to 1105 * the socket's udp_encap_rcv() hook. Otherwise, just 1106 * fall through and pass this up the UDP socket. 1107 * up->encap_rcv() returns the following value: 1108 * =0 if skb was successfully passed to the encap 1109 * handler or was discarded by it. 1110 * >0 if skb should be passed on to UDP. 1111 * <0 if skb should be resubmitted as proto -N 1112 */ 1113 1114 /* if we're overly short, let UDP handle it */ 1115 if (skb->len > sizeof(struct udphdr) && 1116 up->encap_rcv != NULL) { 1117 int ret; 1118 1119 ret = (*up->encap_rcv)(sk, skb); 1120 if (ret <= 0) { 1121 UDP_INC_STATS_BH(sock_net(sk), 1122 UDP_MIB_INDATAGRAMS, 1123 is_udplite); 1124 return -ret; 1125 } 1126 } 1127 1128 /* FALLTHROUGH -- it's a UDP Packet */ 1129 } 1130 1131 /* 1132 * UDP-Lite specific tests, ignored on UDP sockets 1133 */ 1134 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1135 1136 /* 1137 * MIB statistics other than incrementing the error count are 1138 * disabled for the following two types of errors: these depend 1139 * on the application settings, not on the functioning of the 1140 * protocol stack as such. 1141 * 1142 * RFC 3828 here recommends (sec 3.3): "There should also be a 1143 * way ... to ... at least let the receiving application block 1144 * delivery of packets with coverage values less than a value 1145 * provided by the application." 1146 */ 1147 if (up->pcrlen == 0) { /* full coverage was set */ 1148 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1149 "%d while full coverage %d requested\n", 1150 UDP_SKB_CB(skb)->cscov, skb->len); 1151 goto drop; 1152 } 1153 /* The next case involves violating the min. coverage requested 1154 * by the receiver. This is subtle: if receiver wants x and x is 1155 * greater than the buffersize/MTU then receiver will complain 1156 * that it wants x while sender emits packets of smaller size y. 1157 * Therefore the above ...()->partial_cov statement is essential. 1158 */ 1159 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1160 LIMIT_NETDEBUG(KERN_WARNING 1161 "UDPLITE: coverage %d too small, need min %d\n", 1162 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1163 goto drop; 1164 } 1165 } 1166 1167 if (sk->sk_filter) { 1168 if (udp_lib_checksum_complete(skb)) 1169 goto drop; 1170 } 1171 1172 rc = 0; 1173 1174 bh_lock_sock(sk); 1175 if (!sock_owned_by_user(sk)) 1176 rc = __udp_queue_rcv_skb(sk, skb); 1177 else 1178 sk_add_backlog(sk, skb); 1179 bh_unlock_sock(sk); 1180 1181 return rc; 1182 1183 drop: 1184 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1185 kfree_skb(skb); 1186 return -1; 1187 } 1188 1189 /* 1190 * Multicasts and broadcasts go to each listener. 1191 * 1192 * Note: called only from the BH handler context, 1193 * so we don't need to lock the hashes. 1194 */ 1195 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1196 struct udphdr *uh, 1197 __be32 saddr, __be32 daddr, 1198 struct udp_table *udptable) 1199 { 1200 struct sock *sk; 1201 struct udp_hslot *hslot = &udptable->hash[udp_hashfn(net, ntohs(uh->dest))]; 1202 int dif; 1203 1204 spin_lock(&hslot->lock); 1205 sk = sk_nulls_head(&hslot->head); 1206 dif = skb->dev->ifindex; 1207 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1208 if (sk) { 1209 struct sock *sknext = NULL; 1210 1211 do { 1212 struct sk_buff *skb1 = skb; 1213 1214 sknext = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1215 daddr, uh->source, saddr, 1216 dif); 1217 if (sknext) 1218 skb1 = skb_clone(skb, GFP_ATOMIC); 1219 1220 if (skb1) { 1221 int ret = udp_queue_rcv_skb(sk, skb1); 1222 if (ret > 0) 1223 /* we should probably re-process instead 1224 * of dropping packets here. */ 1225 kfree_skb(skb1); 1226 } 1227 sk = sknext; 1228 } while (sknext); 1229 } else 1230 consume_skb(skb); 1231 spin_unlock(&hslot->lock); 1232 return 0; 1233 } 1234 1235 /* Initialize UDP checksum. If exited with zero value (success), 1236 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1237 * Otherwise, csum completion requires chacksumming packet body, 1238 * including udp header and folding it to skb->csum. 1239 */ 1240 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1241 int proto) 1242 { 1243 const struct iphdr *iph; 1244 int err; 1245 1246 UDP_SKB_CB(skb)->partial_cov = 0; 1247 UDP_SKB_CB(skb)->cscov = skb->len; 1248 1249 if (proto == IPPROTO_UDPLITE) { 1250 err = udplite_checksum_init(skb, uh); 1251 if (err) 1252 return err; 1253 } 1254 1255 iph = ip_hdr(skb); 1256 if (uh->check == 0) { 1257 skb->ip_summed = CHECKSUM_UNNECESSARY; 1258 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1259 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1260 proto, skb->csum)) 1261 skb->ip_summed = CHECKSUM_UNNECESSARY; 1262 } 1263 if (!skb_csum_unnecessary(skb)) 1264 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1265 skb->len, proto, 0); 1266 /* Probably, we should checksum udp header (it should be in cache 1267 * in any case) and data in tiny packets (< rx copybreak). 1268 */ 1269 1270 return 0; 1271 } 1272 1273 /* 1274 * All we need to do is get the socket, and then do a checksum. 1275 */ 1276 1277 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1278 int proto) 1279 { 1280 struct sock *sk; 1281 struct udphdr *uh; 1282 unsigned short ulen; 1283 struct rtable *rt = skb_rtable(skb); 1284 __be32 saddr, daddr; 1285 struct net *net = dev_net(skb->dev); 1286 1287 /* 1288 * Validate the packet. 1289 */ 1290 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1291 goto drop; /* No space for header. */ 1292 1293 uh = udp_hdr(skb); 1294 ulen = ntohs(uh->len); 1295 if (ulen > skb->len) 1296 goto short_packet; 1297 1298 if (proto == IPPROTO_UDP) { 1299 /* UDP validates ulen. */ 1300 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1301 goto short_packet; 1302 uh = udp_hdr(skb); 1303 } 1304 1305 if (udp4_csum_init(skb, uh, proto)) 1306 goto csum_error; 1307 1308 saddr = ip_hdr(skb)->saddr; 1309 daddr = ip_hdr(skb)->daddr; 1310 1311 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1312 return __udp4_lib_mcast_deliver(net, skb, uh, 1313 saddr, daddr, udptable); 1314 1315 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1316 1317 if (sk != NULL) { 1318 int ret = udp_queue_rcv_skb(sk, skb); 1319 sock_put(sk); 1320 1321 /* a return value > 0 means to resubmit the input, but 1322 * it wants the return to be -protocol, or 0 1323 */ 1324 if (ret > 0) 1325 return -ret; 1326 return 0; 1327 } 1328 1329 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1330 goto drop; 1331 nf_reset(skb); 1332 1333 /* No socket. Drop packet silently, if checksum is wrong */ 1334 if (udp_lib_checksum_complete(skb)) 1335 goto csum_error; 1336 1337 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1338 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1339 1340 /* 1341 * Hmm. We got an UDP packet to a port to which we 1342 * don't wanna listen. Ignore it. 1343 */ 1344 kfree_skb(skb); 1345 return 0; 1346 1347 short_packet: 1348 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1349 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1350 &saddr, 1351 ntohs(uh->source), 1352 ulen, 1353 skb->len, 1354 &daddr, 1355 ntohs(uh->dest)); 1356 goto drop; 1357 1358 csum_error: 1359 /* 1360 * RFC1122: OK. Discards the bad packet silently (as far as 1361 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1362 */ 1363 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1364 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1365 &saddr, 1366 ntohs(uh->source), 1367 &daddr, 1368 ntohs(uh->dest), 1369 ulen); 1370 drop: 1371 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1372 kfree_skb(skb); 1373 return 0; 1374 } 1375 1376 int udp_rcv(struct sk_buff *skb) 1377 { 1378 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1379 } 1380 1381 void udp_destroy_sock(struct sock *sk) 1382 { 1383 lock_sock(sk); 1384 udp_flush_pending_frames(sk); 1385 release_sock(sk); 1386 } 1387 1388 /* 1389 * Socket option code for UDP 1390 */ 1391 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1392 char __user *optval, unsigned int optlen, 1393 int (*push_pending_frames)(struct sock *)) 1394 { 1395 struct udp_sock *up = udp_sk(sk); 1396 int val; 1397 int err = 0; 1398 int is_udplite = IS_UDPLITE(sk); 1399 1400 if (optlen < sizeof(int)) 1401 return -EINVAL; 1402 1403 if (get_user(val, (int __user *)optval)) 1404 return -EFAULT; 1405 1406 switch (optname) { 1407 case UDP_CORK: 1408 if (val != 0) { 1409 up->corkflag = 1; 1410 } else { 1411 up->corkflag = 0; 1412 lock_sock(sk); 1413 (*push_pending_frames)(sk); 1414 release_sock(sk); 1415 } 1416 break; 1417 1418 case UDP_ENCAP: 1419 switch (val) { 1420 case 0: 1421 case UDP_ENCAP_ESPINUDP: 1422 case UDP_ENCAP_ESPINUDP_NON_IKE: 1423 up->encap_rcv = xfrm4_udp_encap_rcv; 1424 /* FALLTHROUGH */ 1425 case UDP_ENCAP_L2TPINUDP: 1426 up->encap_type = val; 1427 break; 1428 default: 1429 err = -ENOPROTOOPT; 1430 break; 1431 } 1432 break; 1433 1434 /* 1435 * UDP-Lite's partial checksum coverage (RFC 3828). 1436 */ 1437 /* The sender sets actual checksum coverage length via this option. 1438 * The case coverage > packet length is handled by send module. */ 1439 case UDPLITE_SEND_CSCOV: 1440 if (!is_udplite) /* Disable the option on UDP sockets */ 1441 return -ENOPROTOOPT; 1442 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1443 val = 8; 1444 else if (val > USHORT_MAX) 1445 val = USHORT_MAX; 1446 up->pcslen = val; 1447 up->pcflag |= UDPLITE_SEND_CC; 1448 break; 1449 1450 /* The receiver specifies a minimum checksum coverage value. To make 1451 * sense, this should be set to at least 8 (as done below). If zero is 1452 * used, this again means full checksum coverage. */ 1453 case UDPLITE_RECV_CSCOV: 1454 if (!is_udplite) /* Disable the option on UDP sockets */ 1455 return -ENOPROTOOPT; 1456 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1457 val = 8; 1458 else if (val > USHORT_MAX) 1459 val = USHORT_MAX; 1460 up->pcrlen = val; 1461 up->pcflag |= UDPLITE_RECV_CC; 1462 break; 1463 1464 default: 1465 err = -ENOPROTOOPT; 1466 break; 1467 } 1468 1469 return err; 1470 } 1471 EXPORT_SYMBOL(udp_lib_setsockopt); 1472 1473 int udp_setsockopt(struct sock *sk, int level, int optname, 1474 char __user *optval, unsigned int optlen) 1475 { 1476 if (level == SOL_UDP || level == SOL_UDPLITE) 1477 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1478 udp_push_pending_frames); 1479 return ip_setsockopt(sk, level, optname, optval, optlen); 1480 } 1481 1482 #ifdef CONFIG_COMPAT 1483 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1484 char __user *optval, unsigned int optlen) 1485 { 1486 if (level == SOL_UDP || level == SOL_UDPLITE) 1487 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1488 udp_push_pending_frames); 1489 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1490 } 1491 #endif 1492 1493 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1494 char __user *optval, int __user *optlen) 1495 { 1496 struct udp_sock *up = udp_sk(sk); 1497 int val, len; 1498 1499 if (get_user(len, optlen)) 1500 return -EFAULT; 1501 1502 len = min_t(unsigned int, len, sizeof(int)); 1503 1504 if (len < 0) 1505 return -EINVAL; 1506 1507 switch (optname) { 1508 case UDP_CORK: 1509 val = up->corkflag; 1510 break; 1511 1512 case UDP_ENCAP: 1513 val = up->encap_type; 1514 break; 1515 1516 /* The following two cannot be changed on UDP sockets, the return is 1517 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1518 case UDPLITE_SEND_CSCOV: 1519 val = up->pcslen; 1520 break; 1521 1522 case UDPLITE_RECV_CSCOV: 1523 val = up->pcrlen; 1524 break; 1525 1526 default: 1527 return -ENOPROTOOPT; 1528 } 1529 1530 if (put_user(len, optlen)) 1531 return -EFAULT; 1532 if (copy_to_user(optval, &val, len)) 1533 return -EFAULT; 1534 return 0; 1535 } 1536 EXPORT_SYMBOL(udp_lib_getsockopt); 1537 1538 int udp_getsockopt(struct sock *sk, int level, int optname, 1539 char __user *optval, int __user *optlen) 1540 { 1541 if (level == SOL_UDP || level == SOL_UDPLITE) 1542 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1543 return ip_getsockopt(sk, level, optname, optval, optlen); 1544 } 1545 1546 #ifdef CONFIG_COMPAT 1547 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1548 char __user *optval, int __user *optlen) 1549 { 1550 if (level == SOL_UDP || level == SOL_UDPLITE) 1551 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1552 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1553 } 1554 #endif 1555 /** 1556 * udp_poll - wait for a UDP event. 1557 * @file - file struct 1558 * @sock - socket 1559 * @wait - poll table 1560 * 1561 * This is same as datagram poll, except for the special case of 1562 * blocking sockets. If application is using a blocking fd 1563 * and a packet with checksum error is in the queue; 1564 * then it could get return from select indicating data available 1565 * but then block when reading it. Add special case code 1566 * to work around these arguably broken applications. 1567 */ 1568 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1569 { 1570 unsigned int mask = datagram_poll(file, sock, wait); 1571 struct sock *sk = sock->sk; 1572 1573 /* Check for false positives due to checksum errors */ 1574 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 1575 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 1576 mask &= ~(POLLIN | POLLRDNORM); 1577 1578 return mask; 1579 1580 } 1581 EXPORT_SYMBOL(udp_poll); 1582 1583 struct proto udp_prot = { 1584 .name = "UDP", 1585 .owner = THIS_MODULE, 1586 .close = udp_lib_close, 1587 .connect = ip4_datagram_connect, 1588 .disconnect = udp_disconnect, 1589 .ioctl = udp_ioctl, 1590 .destroy = udp_destroy_sock, 1591 .setsockopt = udp_setsockopt, 1592 .getsockopt = udp_getsockopt, 1593 .sendmsg = udp_sendmsg, 1594 .recvmsg = udp_recvmsg, 1595 .sendpage = udp_sendpage, 1596 .backlog_rcv = __udp_queue_rcv_skb, 1597 .hash = udp_lib_hash, 1598 .unhash = udp_lib_unhash, 1599 .get_port = udp_v4_get_port, 1600 .memory_allocated = &udp_memory_allocated, 1601 .sysctl_mem = sysctl_udp_mem, 1602 .sysctl_wmem = &sysctl_udp_wmem_min, 1603 .sysctl_rmem = &sysctl_udp_rmem_min, 1604 .obj_size = sizeof(struct udp_sock), 1605 .slab_flags = SLAB_DESTROY_BY_RCU, 1606 .h.udp_table = &udp_table, 1607 #ifdef CONFIG_COMPAT 1608 .compat_setsockopt = compat_udp_setsockopt, 1609 .compat_getsockopt = compat_udp_getsockopt, 1610 #endif 1611 }; 1612 EXPORT_SYMBOL(udp_prot); 1613 1614 /* ------------------------------------------------------------------------ */ 1615 #ifdef CONFIG_PROC_FS 1616 1617 static struct sock *udp_get_first(struct seq_file *seq, int start) 1618 { 1619 struct sock *sk; 1620 struct udp_iter_state *state = seq->private; 1621 struct net *net = seq_file_net(seq); 1622 1623 for (state->bucket = start; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) { 1624 struct hlist_nulls_node *node; 1625 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 1626 spin_lock_bh(&hslot->lock); 1627 sk_nulls_for_each(sk, node, &hslot->head) { 1628 if (!net_eq(sock_net(sk), net)) 1629 continue; 1630 if (sk->sk_family == state->family) 1631 goto found; 1632 } 1633 spin_unlock_bh(&hslot->lock); 1634 } 1635 sk = NULL; 1636 found: 1637 return sk; 1638 } 1639 1640 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1641 { 1642 struct udp_iter_state *state = seq->private; 1643 struct net *net = seq_file_net(seq); 1644 1645 do { 1646 sk = sk_nulls_next(sk); 1647 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1648 1649 if (!sk) { 1650 if (state->bucket < UDP_HTABLE_SIZE) 1651 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1652 return udp_get_first(seq, state->bucket + 1); 1653 } 1654 return sk; 1655 } 1656 1657 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1658 { 1659 struct sock *sk = udp_get_first(seq, 0); 1660 1661 if (sk) 1662 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1663 --pos; 1664 return pos ? NULL : sk; 1665 } 1666 1667 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1668 { 1669 struct udp_iter_state *state = seq->private; 1670 state->bucket = UDP_HTABLE_SIZE; 1671 1672 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 1673 } 1674 1675 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1676 { 1677 struct sock *sk; 1678 1679 if (v == SEQ_START_TOKEN) 1680 sk = udp_get_idx(seq, 0); 1681 else 1682 sk = udp_get_next(seq, v); 1683 1684 ++*pos; 1685 return sk; 1686 } 1687 1688 static void udp_seq_stop(struct seq_file *seq, void *v) 1689 { 1690 struct udp_iter_state *state = seq->private; 1691 1692 if (state->bucket < UDP_HTABLE_SIZE) 1693 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1694 } 1695 1696 static int udp_seq_open(struct inode *inode, struct file *file) 1697 { 1698 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1699 struct udp_iter_state *s; 1700 int err; 1701 1702 err = seq_open_net(inode, file, &afinfo->seq_ops, 1703 sizeof(struct udp_iter_state)); 1704 if (err < 0) 1705 return err; 1706 1707 s = ((struct seq_file *)file->private_data)->private; 1708 s->family = afinfo->family; 1709 s->udp_table = afinfo->udp_table; 1710 return err; 1711 } 1712 1713 /* ------------------------------------------------------------------------ */ 1714 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 1715 { 1716 struct proc_dir_entry *p; 1717 int rc = 0; 1718 1719 afinfo->seq_fops.open = udp_seq_open; 1720 afinfo->seq_fops.read = seq_read; 1721 afinfo->seq_fops.llseek = seq_lseek; 1722 afinfo->seq_fops.release = seq_release_net; 1723 1724 afinfo->seq_ops.start = udp_seq_start; 1725 afinfo->seq_ops.next = udp_seq_next; 1726 afinfo->seq_ops.stop = udp_seq_stop; 1727 1728 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 1729 &afinfo->seq_fops, afinfo); 1730 if (!p) 1731 rc = -ENOMEM; 1732 return rc; 1733 } 1734 EXPORT_SYMBOL(udp_proc_register); 1735 1736 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 1737 { 1738 proc_net_remove(net, afinfo->name); 1739 } 1740 EXPORT_SYMBOL(udp_proc_unregister); 1741 1742 /* ------------------------------------------------------------------------ */ 1743 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 1744 int bucket, int *len) 1745 { 1746 struct inet_sock *inet = inet_sk(sp); 1747 __be32 dest = inet->daddr; 1748 __be32 src = inet->rcv_saddr; 1749 __u16 destp = ntohs(inet->dport); 1750 __u16 srcp = ntohs(inet->sport); 1751 1752 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 1753 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 1754 bucket, src, srcp, dest, destp, sp->sk_state, 1755 sk_wmem_alloc_get(sp), 1756 sk_rmem_alloc_get(sp), 1757 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1758 atomic_read(&sp->sk_refcnt), sp, 1759 atomic_read(&sp->sk_drops), len); 1760 } 1761 1762 int udp4_seq_show(struct seq_file *seq, void *v) 1763 { 1764 if (v == SEQ_START_TOKEN) 1765 seq_printf(seq, "%-127s\n", 1766 " sl local_address rem_address st tx_queue " 1767 "rx_queue tr tm->when retrnsmt uid timeout " 1768 "inode ref pointer drops"); 1769 else { 1770 struct udp_iter_state *state = seq->private; 1771 int len; 1772 1773 udp4_format_sock(v, seq, state->bucket, &len); 1774 seq_printf(seq, "%*s\n", 127 - len, ""); 1775 } 1776 return 0; 1777 } 1778 1779 /* ------------------------------------------------------------------------ */ 1780 static struct udp_seq_afinfo udp4_seq_afinfo = { 1781 .name = "udp", 1782 .family = AF_INET, 1783 .udp_table = &udp_table, 1784 .seq_fops = { 1785 .owner = THIS_MODULE, 1786 }, 1787 .seq_ops = { 1788 .show = udp4_seq_show, 1789 }, 1790 }; 1791 1792 static int udp4_proc_init_net(struct net *net) 1793 { 1794 return udp_proc_register(net, &udp4_seq_afinfo); 1795 } 1796 1797 static void udp4_proc_exit_net(struct net *net) 1798 { 1799 udp_proc_unregister(net, &udp4_seq_afinfo); 1800 } 1801 1802 static struct pernet_operations udp4_net_ops = { 1803 .init = udp4_proc_init_net, 1804 .exit = udp4_proc_exit_net, 1805 }; 1806 1807 int __init udp4_proc_init(void) 1808 { 1809 return register_pernet_subsys(&udp4_net_ops); 1810 } 1811 1812 void udp4_proc_exit(void) 1813 { 1814 unregister_pernet_subsys(&udp4_net_ops); 1815 } 1816 #endif /* CONFIG_PROC_FS */ 1817 1818 void __init udp_table_init(struct udp_table *table) 1819 { 1820 int i; 1821 1822 for (i = 0; i < UDP_HTABLE_SIZE; i++) { 1823 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 1824 spin_lock_init(&table->hash[i].lock); 1825 } 1826 } 1827 1828 void __init udp_init(void) 1829 { 1830 unsigned long nr_pages, limit; 1831 1832 udp_table_init(&udp_table); 1833 /* Set the pressure threshold up by the same strategy of TCP. It is a 1834 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 1835 * toward zero with the amount of memory, with a floor of 128 pages. 1836 */ 1837 nr_pages = totalram_pages - totalhigh_pages; 1838 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 1839 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 1840 limit = max(limit, 128UL); 1841 sysctl_udp_mem[0] = limit / 4 * 3; 1842 sysctl_udp_mem[1] = limit; 1843 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 1844 1845 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 1846 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 1847 } 1848 1849 int udp4_ufo_send_check(struct sk_buff *skb) 1850 { 1851 const struct iphdr *iph; 1852 struct udphdr *uh; 1853 1854 if (!pskb_may_pull(skb, sizeof(*uh))) 1855 return -EINVAL; 1856 1857 iph = ip_hdr(skb); 1858 uh = udp_hdr(skb); 1859 1860 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1861 IPPROTO_UDP, 0); 1862 skb->csum_start = skb_transport_header(skb) - skb->head; 1863 skb->csum_offset = offsetof(struct udphdr, check); 1864 skb->ip_summed = CHECKSUM_PARTIAL; 1865 return 0; 1866 } 1867 1868 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features) 1869 { 1870 struct sk_buff *segs = ERR_PTR(-EINVAL); 1871 unsigned int mss; 1872 int offset; 1873 __wsum csum; 1874 1875 mss = skb_shinfo(skb)->gso_size; 1876 if (unlikely(skb->len <= mss)) 1877 goto out; 1878 1879 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 1880 /* Packet is from an untrusted source, reset gso_segs. */ 1881 int type = skb_shinfo(skb)->gso_type; 1882 1883 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || 1884 !(type & (SKB_GSO_UDP)))) 1885 goto out; 1886 1887 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 1888 1889 segs = NULL; 1890 goto out; 1891 } 1892 1893 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot 1894 * do checksum of UDP packets sent as multiple IP fragments. 1895 */ 1896 offset = skb->csum_start - skb_headroom(skb); 1897 csum = skb_checksum(skb, offset, skb->len - offset, 0); 1898 offset += skb->csum_offset; 1899 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 1900 skb->ip_summed = CHECKSUM_NONE; 1901 1902 /* Fragment the skb. IP headers of the fragments are updated in 1903 * inet_gso_segment() 1904 */ 1905 segs = skb_segment(skb, features); 1906 out: 1907 return segs; 1908 } 1909 1910