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 __read_mostly; 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 MAX_UDP_PORTS 65536 125 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 126 127 static int udp_lib_lport_inuse(struct net *net, __u16 num, 128 const struct udp_hslot *hslot, 129 unsigned long *bitmap, 130 struct sock *sk, 131 int (*saddr_comp)(const struct sock *sk1, 132 const struct sock *sk2), 133 unsigned int log) 134 { 135 struct sock *sk2; 136 struct hlist_nulls_node *node; 137 138 sk_nulls_for_each(sk2, node, &hslot->head) 139 if (net_eq(sock_net(sk2), net) && 140 sk2 != sk && 141 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 142 (!sk2->sk_reuse || !sk->sk_reuse) && 143 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 144 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 145 (*saddr_comp)(sk, sk2)) { 146 if (bitmap) 147 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 148 bitmap); 149 else 150 return 1; 151 } 152 return 0; 153 } 154 155 /* 156 * Note: we still hold spinlock of primary hash chain, so no other writer 157 * can insert/delete a socket with local_port == num 158 */ 159 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 160 struct udp_hslot *hslot2, 161 struct sock *sk, 162 int (*saddr_comp)(const struct sock *sk1, 163 const struct sock *sk2)) 164 { 165 struct sock *sk2; 166 struct hlist_nulls_node *node; 167 int res = 0; 168 169 spin_lock(&hslot2->lock); 170 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) 171 if (net_eq(sock_net(sk2), net) && 172 sk2 != sk && 173 (udp_sk(sk2)->udp_port_hash == num) && 174 (!sk2->sk_reuse || !sk->sk_reuse) && 175 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 176 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 177 (*saddr_comp)(sk, sk2)) { 178 res = 1; 179 break; 180 } 181 spin_unlock(&hslot2->lock); 182 return res; 183 } 184 185 /** 186 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 187 * 188 * @sk: socket struct in question 189 * @snum: port number to look up 190 * @saddr_comp: AF-dependent comparison of bound local IP addresses 191 * @hash2_nulladdr: AF-dependant hash value in secondary hash chains, 192 * with NULL address 193 */ 194 int udp_lib_get_port(struct sock *sk, unsigned short snum, 195 int (*saddr_comp)(const struct sock *sk1, 196 const struct sock *sk2), 197 unsigned int hash2_nulladdr) 198 { 199 struct udp_hslot *hslot, *hslot2; 200 struct udp_table *udptable = sk->sk_prot->h.udp_table; 201 int error = 1; 202 struct net *net = sock_net(sk); 203 204 if (!snum) { 205 int low, high, remaining; 206 unsigned rand; 207 unsigned short first, last; 208 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 209 210 inet_get_local_port_range(&low, &high); 211 remaining = (high - low) + 1; 212 213 rand = net_random(); 214 first = (((u64)rand * remaining) >> 32) + low; 215 /* 216 * force rand to be an odd multiple of UDP_HTABLE_SIZE 217 */ 218 rand = (rand | 1) * (udptable->mask + 1); 219 for (last = first + udptable->mask + 1; 220 first != last; 221 first++) { 222 hslot = udp_hashslot(udptable, net, first); 223 bitmap_zero(bitmap, PORTS_PER_CHAIN); 224 spin_lock_bh(&hslot->lock); 225 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 226 saddr_comp, udptable->log); 227 228 snum = first; 229 /* 230 * Iterate on all possible values of snum for this hash. 231 * Using steps of an odd multiple of UDP_HTABLE_SIZE 232 * give us randomization and full range coverage. 233 */ 234 do { 235 if (low <= snum && snum <= high && 236 !test_bit(snum >> udptable->log, bitmap)) 237 goto found; 238 snum += rand; 239 } while (snum != first); 240 spin_unlock_bh(&hslot->lock); 241 } 242 goto fail; 243 } else { 244 hslot = udp_hashslot(udptable, net, snum); 245 spin_lock_bh(&hslot->lock); 246 if (hslot->count > 10) { 247 int exist; 248 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 249 250 slot2 &= udptable->mask; 251 hash2_nulladdr &= udptable->mask; 252 253 hslot2 = udp_hashslot2(udptable, slot2); 254 if (hslot->count < hslot2->count) 255 goto scan_primary_hash; 256 257 exist = udp_lib_lport_inuse2(net, snum, hslot2, 258 sk, saddr_comp); 259 if (!exist && (hash2_nulladdr != slot2)) { 260 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 261 exist = udp_lib_lport_inuse2(net, snum, hslot2, 262 sk, saddr_comp); 263 } 264 if (exist) 265 goto fail_unlock; 266 else 267 goto found; 268 } 269 scan_primary_hash: 270 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 271 saddr_comp, 0)) 272 goto fail_unlock; 273 } 274 found: 275 inet_sk(sk)->inet_num = snum; 276 udp_sk(sk)->udp_port_hash = snum; 277 udp_sk(sk)->udp_portaddr_hash ^= snum; 278 if (sk_unhashed(sk)) { 279 sk_nulls_add_node_rcu(sk, &hslot->head); 280 hslot->count++; 281 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 282 283 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 284 spin_lock(&hslot2->lock); 285 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 286 &hslot2->head); 287 hslot2->count++; 288 spin_unlock(&hslot2->lock); 289 } 290 error = 0; 291 fail_unlock: 292 spin_unlock_bh(&hslot->lock); 293 fail: 294 return error; 295 } 296 EXPORT_SYMBOL(udp_lib_get_port); 297 298 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 299 { 300 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 301 302 return (!ipv6_only_sock(sk2) && 303 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 304 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 305 } 306 307 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr, 308 unsigned int port) 309 { 310 return jhash_1word(saddr, net_hash_mix(net)) ^ port; 311 } 312 313 int udp_v4_get_port(struct sock *sk, unsigned short snum) 314 { 315 unsigned int hash2_nulladdr = 316 udp4_portaddr_hash(sock_net(sk), INADDR_ANY, snum); 317 unsigned int hash2_partial = 318 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 319 320 /* precompute partial secondary hash */ 321 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 322 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 323 } 324 325 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, 326 unsigned short hnum, 327 __be16 sport, __be32 daddr, __be16 dport, int dif) 328 { 329 int score = -1; 330 331 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 332 !ipv6_only_sock(sk)) { 333 struct inet_sock *inet = inet_sk(sk); 334 335 score = (sk->sk_family == PF_INET ? 1 : 0); 336 if (inet->inet_rcv_saddr) { 337 if (inet->inet_rcv_saddr != daddr) 338 return -1; 339 score += 2; 340 } 341 if (inet->inet_daddr) { 342 if (inet->inet_daddr != saddr) 343 return -1; 344 score += 2; 345 } 346 if (inet->inet_dport) { 347 if (inet->inet_dport != sport) 348 return -1; 349 score += 2; 350 } 351 if (sk->sk_bound_dev_if) { 352 if (sk->sk_bound_dev_if != dif) 353 return -1; 354 score += 2; 355 } 356 } 357 return score; 358 } 359 360 /* 361 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 362 */ 363 #define SCORE2_MAX (1 + 2 + 2 + 2) 364 static inline int compute_score2(struct sock *sk, struct net *net, 365 __be32 saddr, __be16 sport, 366 __be32 daddr, unsigned int hnum, int dif) 367 { 368 int score = -1; 369 370 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) { 371 struct inet_sock *inet = inet_sk(sk); 372 373 if (inet->inet_rcv_saddr != daddr) 374 return -1; 375 if (inet->inet_num != hnum) 376 return -1; 377 378 score = (sk->sk_family == PF_INET ? 1 : 0); 379 if (inet->inet_daddr) { 380 if (inet->inet_daddr != saddr) 381 return -1; 382 score += 2; 383 } 384 if (inet->inet_dport) { 385 if (inet->inet_dport != sport) 386 return -1; 387 score += 2; 388 } 389 if (sk->sk_bound_dev_if) { 390 if (sk->sk_bound_dev_if != dif) 391 return -1; 392 score += 2; 393 } 394 } 395 return score; 396 } 397 398 399 /* called with read_rcu_lock() */ 400 static struct sock *udp4_lib_lookup2(struct net *net, 401 __be32 saddr, __be16 sport, 402 __be32 daddr, unsigned int hnum, int dif, 403 struct udp_hslot *hslot2, unsigned int slot2) 404 { 405 struct sock *sk, *result; 406 struct hlist_nulls_node *node; 407 int score, badness; 408 409 begin: 410 result = NULL; 411 badness = -1; 412 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 413 score = compute_score2(sk, net, saddr, sport, 414 daddr, hnum, dif); 415 if (score > badness) { 416 result = sk; 417 badness = score; 418 if (score == SCORE2_MAX) 419 goto exact_match; 420 } 421 } 422 /* 423 * if the nulls value we got at the end of this lookup is 424 * not the expected one, we must restart lookup. 425 * We probably met an item that was moved to another chain. 426 */ 427 if (get_nulls_value(node) != slot2) 428 goto begin; 429 430 if (result) { 431 exact_match: 432 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) 433 result = NULL; 434 else if (unlikely(compute_score2(result, net, saddr, sport, 435 daddr, hnum, dif) < badness)) { 436 sock_put(result); 437 goto begin; 438 } 439 } 440 return result; 441 } 442 443 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 444 * harder than this. -DaveM 445 */ 446 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 447 __be16 sport, __be32 daddr, __be16 dport, 448 int dif, struct udp_table *udptable) 449 { 450 struct sock *sk, *result; 451 struct hlist_nulls_node *node; 452 unsigned short hnum = ntohs(dport); 453 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 454 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 455 int score, badness; 456 457 rcu_read_lock(); 458 if (hslot->count > 10) { 459 hash2 = udp4_portaddr_hash(net, daddr, hnum); 460 slot2 = hash2 & udptable->mask; 461 hslot2 = &udptable->hash2[slot2]; 462 if (hslot->count < hslot2->count) 463 goto begin; 464 465 result = udp4_lib_lookup2(net, saddr, sport, 466 daddr, hnum, dif, 467 hslot2, slot2); 468 if (!result) { 469 hash2 = udp4_portaddr_hash(net, INADDR_ANY, hnum); 470 slot2 = hash2 & udptable->mask; 471 hslot2 = &udptable->hash2[slot2]; 472 if (hslot->count < hslot2->count) 473 goto begin; 474 475 result = udp4_lib_lookup2(net, INADDR_ANY, sport, 476 daddr, hnum, dif, 477 hslot2, slot2); 478 } 479 rcu_read_unlock(); 480 return result; 481 } 482 begin: 483 result = NULL; 484 badness = -1; 485 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 486 score = compute_score(sk, net, saddr, hnum, sport, 487 daddr, dport, dif); 488 if (score > badness) { 489 result = sk; 490 badness = score; 491 } 492 } 493 /* 494 * if the nulls value we got at the end of this lookup is 495 * not the expected one, we must restart lookup. 496 * We probably met an item that was moved to another chain. 497 */ 498 if (get_nulls_value(node) != slot) 499 goto begin; 500 501 if (result) { 502 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) 503 result = NULL; 504 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 505 daddr, dport, dif) < badness)) { 506 sock_put(result); 507 goto begin; 508 } 509 } 510 rcu_read_unlock(); 511 return result; 512 } 513 514 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 515 __be16 sport, __be16 dport, 516 struct udp_table *udptable) 517 { 518 struct sock *sk; 519 const struct iphdr *iph = ip_hdr(skb); 520 521 if (unlikely(sk = skb_steal_sock(skb))) 522 return sk; 523 else 524 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 525 iph->daddr, dport, inet_iif(skb), 526 udptable); 527 } 528 529 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 530 __be32 daddr, __be16 dport, int dif) 531 { 532 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 533 } 534 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 535 536 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 537 __be16 loc_port, __be32 loc_addr, 538 __be16 rmt_port, __be32 rmt_addr, 539 int dif) 540 { 541 struct hlist_nulls_node *node; 542 struct sock *s = sk; 543 unsigned short hnum = ntohs(loc_port); 544 545 sk_nulls_for_each_from(s, node) { 546 struct inet_sock *inet = inet_sk(s); 547 548 if (!net_eq(sock_net(s), net) || 549 udp_sk(s)->udp_port_hash != hnum || 550 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 551 (inet->inet_dport != rmt_port && inet->inet_dport) || 552 (inet->inet_rcv_saddr && 553 inet->inet_rcv_saddr != loc_addr) || 554 ipv6_only_sock(s) || 555 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 556 continue; 557 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 558 continue; 559 goto found; 560 } 561 s = NULL; 562 found: 563 return s; 564 } 565 566 /* 567 * This routine is called by the ICMP module when it gets some 568 * sort of error condition. If err < 0 then the socket should 569 * be closed and the error returned to the user. If err > 0 570 * it's just the icmp type << 8 | icmp code. 571 * Header points to the ip header of the error packet. We move 572 * on past this. Then (as it used to claim before adjustment) 573 * header points to the first 8 bytes of the udp header. We need 574 * to find the appropriate port. 575 */ 576 577 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 578 { 579 struct inet_sock *inet; 580 struct iphdr *iph = (struct iphdr *)skb->data; 581 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 582 const int type = icmp_hdr(skb)->type; 583 const int code = icmp_hdr(skb)->code; 584 struct sock *sk; 585 int harderr; 586 int err; 587 struct net *net = dev_net(skb->dev); 588 589 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 590 iph->saddr, uh->source, skb->dev->ifindex, udptable); 591 if (sk == NULL) { 592 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 593 return; /* No socket for error */ 594 } 595 596 err = 0; 597 harderr = 0; 598 inet = inet_sk(sk); 599 600 switch (type) { 601 default: 602 case ICMP_TIME_EXCEEDED: 603 err = EHOSTUNREACH; 604 break; 605 case ICMP_SOURCE_QUENCH: 606 goto out; 607 case ICMP_PARAMETERPROB: 608 err = EPROTO; 609 harderr = 1; 610 break; 611 case ICMP_DEST_UNREACH: 612 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 613 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 614 err = EMSGSIZE; 615 harderr = 1; 616 break; 617 } 618 goto out; 619 } 620 err = EHOSTUNREACH; 621 if (code <= NR_ICMP_UNREACH) { 622 harderr = icmp_err_convert[code].fatal; 623 err = icmp_err_convert[code].errno; 624 } 625 break; 626 } 627 628 /* 629 * RFC1122: OK. Passes ICMP errors back to application, as per 630 * 4.1.3.3. 631 */ 632 if (!inet->recverr) { 633 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 634 goto out; 635 } else { 636 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 637 } 638 sk->sk_err = err; 639 sk->sk_error_report(sk); 640 out: 641 sock_put(sk); 642 } 643 644 void udp_err(struct sk_buff *skb, u32 info) 645 { 646 __udp4_lib_err(skb, info, &udp_table); 647 } 648 649 /* 650 * Throw away all pending data and cancel the corking. Socket is locked. 651 */ 652 void udp_flush_pending_frames(struct sock *sk) 653 { 654 struct udp_sock *up = udp_sk(sk); 655 656 if (up->pending) { 657 up->len = 0; 658 up->pending = 0; 659 ip_flush_pending_frames(sk); 660 } 661 } 662 EXPORT_SYMBOL(udp_flush_pending_frames); 663 664 /** 665 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 666 * @sk: socket we are sending on 667 * @skb: sk_buff containing the filled-in UDP header 668 * (checksum field must be zeroed out) 669 */ 670 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 671 __be32 src, __be32 dst, int len) 672 { 673 unsigned int offset; 674 struct udphdr *uh = udp_hdr(skb); 675 __wsum csum = 0; 676 677 if (skb_queue_len(&sk->sk_write_queue) == 1) { 678 /* 679 * Only one fragment on the socket. 680 */ 681 skb->csum_start = skb_transport_header(skb) - skb->head; 682 skb->csum_offset = offsetof(struct udphdr, check); 683 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 684 } else { 685 /* 686 * HW-checksum won't work as there are two or more 687 * fragments on the socket so that all csums of sk_buffs 688 * should be together 689 */ 690 offset = skb_transport_offset(skb); 691 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 692 693 skb->ip_summed = CHECKSUM_NONE; 694 695 skb_queue_walk(&sk->sk_write_queue, skb) { 696 csum = csum_add(csum, skb->csum); 697 } 698 699 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 700 if (uh->check == 0) 701 uh->check = CSUM_MANGLED_0; 702 } 703 } 704 705 /* 706 * Push out all pending data as one UDP datagram. Socket is locked. 707 */ 708 static int udp_push_pending_frames(struct sock *sk) 709 { 710 struct udp_sock *up = udp_sk(sk); 711 struct inet_sock *inet = inet_sk(sk); 712 struct flowi *fl = &inet->cork.fl; 713 struct sk_buff *skb; 714 struct udphdr *uh; 715 int err = 0; 716 int is_udplite = IS_UDPLITE(sk); 717 __wsum csum = 0; 718 719 /* Grab the skbuff where UDP header space exists. */ 720 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 721 goto out; 722 723 /* 724 * Create a UDP header 725 */ 726 uh = udp_hdr(skb); 727 uh->source = fl->fl_ip_sport; 728 uh->dest = fl->fl_ip_dport; 729 uh->len = htons(up->len); 730 uh->check = 0; 731 732 if (is_udplite) /* UDP-Lite */ 733 csum = udplite_csum_outgoing(sk, skb); 734 735 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 736 737 skb->ip_summed = CHECKSUM_NONE; 738 goto send; 739 740 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 741 742 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len); 743 goto send; 744 745 } else /* `normal' UDP */ 746 csum = udp_csum_outgoing(sk, skb); 747 748 /* add protocol-dependent pseudo-header */ 749 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 750 sk->sk_protocol, csum); 751 if (uh->check == 0) 752 uh->check = CSUM_MANGLED_0; 753 754 send: 755 err = ip_push_pending_frames(sk); 756 if (err) { 757 if (err == -ENOBUFS && !inet->recverr) { 758 UDP_INC_STATS_USER(sock_net(sk), 759 UDP_MIB_SNDBUFERRORS, is_udplite); 760 err = 0; 761 } 762 } else 763 UDP_INC_STATS_USER(sock_net(sk), 764 UDP_MIB_OUTDATAGRAMS, is_udplite); 765 out: 766 up->len = 0; 767 up->pending = 0; 768 return err; 769 } 770 771 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 772 size_t len) 773 { 774 struct inet_sock *inet = inet_sk(sk); 775 struct udp_sock *up = udp_sk(sk); 776 int ulen = len; 777 struct ipcm_cookie ipc; 778 struct rtable *rt = NULL; 779 int free = 0; 780 int connected = 0; 781 __be32 daddr, faddr, saddr; 782 __be16 dport; 783 u8 tos; 784 int err, is_udplite = IS_UDPLITE(sk); 785 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 786 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 787 788 if (len > 0xFFFF) 789 return -EMSGSIZE; 790 791 /* 792 * Check the flags. 793 */ 794 795 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 796 return -EOPNOTSUPP; 797 798 ipc.opt = NULL; 799 ipc.shtx.flags = 0; 800 801 if (up->pending) { 802 /* 803 * There are pending frames. 804 * The socket lock must be held while it's corked. 805 */ 806 lock_sock(sk); 807 if (likely(up->pending)) { 808 if (unlikely(up->pending != AF_INET)) { 809 release_sock(sk); 810 return -EINVAL; 811 } 812 goto do_append_data; 813 } 814 release_sock(sk); 815 } 816 ulen += sizeof(struct udphdr); 817 818 /* 819 * Get and verify the address. 820 */ 821 if (msg->msg_name) { 822 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name; 823 if (msg->msg_namelen < sizeof(*usin)) 824 return -EINVAL; 825 if (usin->sin_family != AF_INET) { 826 if (usin->sin_family != AF_UNSPEC) 827 return -EAFNOSUPPORT; 828 } 829 830 daddr = usin->sin_addr.s_addr; 831 dport = usin->sin_port; 832 if (dport == 0) 833 return -EINVAL; 834 } else { 835 if (sk->sk_state != TCP_ESTABLISHED) 836 return -EDESTADDRREQ; 837 daddr = inet->inet_daddr; 838 dport = inet->inet_dport; 839 /* Open fast path for connected socket. 840 Route will not be used, if at least one option is set. 841 */ 842 connected = 1; 843 } 844 ipc.addr = inet->inet_saddr; 845 846 ipc.oif = sk->sk_bound_dev_if; 847 err = sock_tx_timestamp(msg, sk, &ipc.shtx); 848 if (err) 849 return err; 850 if (msg->msg_controllen) { 851 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 852 if (err) 853 return err; 854 if (ipc.opt) 855 free = 1; 856 connected = 0; 857 } 858 if (!ipc.opt) 859 ipc.opt = inet->opt; 860 861 saddr = ipc.addr; 862 ipc.addr = faddr = daddr; 863 864 if (ipc.opt && ipc.opt->srr) { 865 if (!daddr) 866 return -EINVAL; 867 faddr = ipc.opt->faddr; 868 connected = 0; 869 } 870 tos = RT_TOS(inet->tos); 871 if (sock_flag(sk, SOCK_LOCALROUTE) || 872 (msg->msg_flags & MSG_DONTROUTE) || 873 (ipc.opt && ipc.opt->is_strictroute)) { 874 tos |= RTO_ONLINK; 875 connected = 0; 876 } 877 878 if (ipv4_is_multicast(daddr)) { 879 if (!ipc.oif) 880 ipc.oif = inet->mc_index; 881 if (!saddr) 882 saddr = inet->mc_addr; 883 connected = 0; 884 } 885 886 if (connected) 887 rt = (struct rtable *)sk_dst_check(sk, 0); 888 889 if (rt == NULL) { 890 struct flowi fl = { .oif = ipc.oif, 891 .mark = sk->sk_mark, 892 .nl_u = { .ip4_u = 893 { .daddr = faddr, 894 .saddr = saddr, 895 .tos = tos } }, 896 .proto = sk->sk_protocol, 897 .flags = inet_sk_flowi_flags(sk), 898 .uli_u = { .ports = 899 { .sport = inet->inet_sport, 900 .dport = dport } } }; 901 struct net *net = sock_net(sk); 902 903 security_sk_classify_flow(sk, &fl); 904 err = ip_route_output_flow(net, &rt, &fl, sk, 1); 905 if (err) { 906 if (err == -ENETUNREACH) 907 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 908 goto out; 909 } 910 911 err = -EACCES; 912 if ((rt->rt_flags & RTCF_BROADCAST) && 913 !sock_flag(sk, SOCK_BROADCAST)) 914 goto out; 915 if (connected) 916 sk_dst_set(sk, dst_clone(&rt->u.dst)); 917 } 918 919 if (msg->msg_flags&MSG_CONFIRM) 920 goto do_confirm; 921 back_from_confirm: 922 923 saddr = rt->rt_src; 924 if (!ipc.addr) 925 daddr = ipc.addr = rt->rt_dst; 926 927 lock_sock(sk); 928 if (unlikely(up->pending)) { 929 /* The socket is already corked while preparing it. */ 930 /* ... which is an evident application bug. --ANK */ 931 release_sock(sk); 932 933 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 934 err = -EINVAL; 935 goto out; 936 } 937 /* 938 * Now cork the socket to pend data. 939 */ 940 inet->cork.fl.fl4_dst = daddr; 941 inet->cork.fl.fl_ip_dport = dport; 942 inet->cork.fl.fl4_src = saddr; 943 inet->cork.fl.fl_ip_sport = inet->inet_sport; 944 up->pending = AF_INET; 945 946 do_append_data: 947 up->len += ulen; 948 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 949 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 950 sizeof(struct udphdr), &ipc, &rt, 951 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 952 if (err) 953 udp_flush_pending_frames(sk); 954 else if (!corkreq) 955 err = udp_push_pending_frames(sk); 956 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 957 up->pending = 0; 958 release_sock(sk); 959 960 out: 961 ip_rt_put(rt); 962 if (free) 963 kfree(ipc.opt); 964 if (!err) 965 return len; 966 /* 967 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 968 * ENOBUFS might not be good (it's not tunable per se), but otherwise 969 * we don't have a good statistic (IpOutDiscards but it can be too many 970 * things). We could add another new stat but at least for now that 971 * seems like overkill. 972 */ 973 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 974 UDP_INC_STATS_USER(sock_net(sk), 975 UDP_MIB_SNDBUFERRORS, is_udplite); 976 } 977 return err; 978 979 do_confirm: 980 dst_confirm(&rt->u.dst); 981 if (!(msg->msg_flags&MSG_PROBE) || len) 982 goto back_from_confirm; 983 err = 0; 984 goto out; 985 } 986 EXPORT_SYMBOL(udp_sendmsg); 987 988 int udp_sendpage(struct sock *sk, struct page *page, int offset, 989 size_t size, int flags) 990 { 991 struct udp_sock *up = udp_sk(sk); 992 int ret; 993 994 if (!up->pending) { 995 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 996 997 /* Call udp_sendmsg to specify destination address which 998 * sendpage interface can't pass. 999 * This will succeed only when the socket is connected. 1000 */ 1001 ret = udp_sendmsg(NULL, sk, &msg, 0); 1002 if (ret < 0) 1003 return ret; 1004 } 1005 1006 lock_sock(sk); 1007 1008 if (unlikely(!up->pending)) { 1009 release_sock(sk); 1010 1011 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 1012 return -EINVAL; 1013 } 1014 1015 ret = ip_append_page(sk, page, offset, size, flags); 1016 if (ret == -EOPNOTSUPP) { 1017 release_sock(sk); 1018 return sock_no_sendpage(sk->sk_socket, page, offset, 1019 size, flags); 1020 } 1021 if (ret < 0) { 1022 udp_flush_pending_frames(sk); 1023 goto out; 1024 } 1025 1026 up->len += size; 1027 if (!(up->corkflag || (flags&MSG_MORE))) 1028 ret = udp_push_pending_frames(sk); 1029 if (!ret) 1030 ret = size; 1031 out: 1032 release_sock(sk); 1033 return ret; 1034 } 1035 1036 1037 /** 1038 * first_packet_length - return length of first packet in receive queue 1039 * @sk: socket 1040 * 1041 * Drops all bad checksum frames, until a valid one is found. 1042 * Returns the length of found skb, or 0 if none is found. 1043 */ 1044 static unsigned int first_packet_length(struct sock *sk) 1045 { 1046 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1047 struct sk_buff *skb; 1048 unsigned int res; 1049 1050 __skb_queue_head_init(&list_kill); 1051 1052 spin_lock_bh(&rcvq->lock); 1053 while ((skb = skb_peek(rcvq)) != NULL && 1054 udp_lib_checksum_complete(skb)) { 1055 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1056 IS_UDPLITE(sk)); 1057 atomic_inc(&sk->sk_drops); 1058 __skb_unlink(skb, rcvq); 1059 __skb_queue_tail(&list_kill, skb); 1060 } 1061 res = skb ? skb->len : 0; 1062 spin_unlock_bh(&rcvq->lock); 1063 1064 if (!skb_queue_empty(&list_kill)) { 1065 lock_sock(sk); 1066 __skb_queue_purge(&list_kill); 1067 sk_mem_reclaim_partial(sk); 1068 release_sock(sk); 1069 } 1070 return res; 1071 } 1072 1073 /* 1074 * IOCTL requests applicable to the UDP protocol 1075 */ 1076 1077 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1078 { 1079 switch (cmd) { 1080 case SIOCOUTQ: 1081 { 1082 int amount = sk_wmem_alloc_get(sk); 1083 1084 return put_user(amount, (int __user *)arg); 1085 } 1086 1087 case SIOCINQ: 1088 { 1089 unsigned int amount = first_packet_length(sk); 1090 1091 if (amount) 1092 /* 1093 * We will only return the amount 1094 * of this packet since that is all 1095 * that will be read. 1096 */ 1097 amount -= sizeof(struct udphdr); 1098 1099 return put_user(amount, (int __user *)arg); 1100 } 1101 1102 default: 1103 return -ENOIOCTLCMD; 1104 } 1105 1106 return 0; 1107 } 1108 EXPORT_SYMBOL(udp_ioctl); 1109 1110 /* 1111 * This should be easy, if there is something there we 1112 * return it, otherwise we block. 1113 */ 1114 1115 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1116 size_t len, int noblock, int flags, int *addr_len) 1117 { 1118 struct inet_sock *inet = inet_sk(sk); 1119 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 1120 struct sk_buff *skb; 1121 unsigned int ulen, copied; 1122 int peeked; 1123 int err; 1124 int is_udplite = IS_UDPLITE(sk); 1125 1126 /* 1127 * Check any passed addresses 1128 */ 1129 if (addr_len) 1130 *addr_len = sizeof(*sin); 1131 1132 if (flags & MSG_ERRQUEUE) 1133 return ip_recv_error(sk, msg, len); 1134 1135 try_again: 1136 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1137 &peeked, &err); 1138 if (!skb) 1139 goto out; 1140 1141 ulen = skb->len - sizeof(struct udphdr); 1142 copied = len; 1143 if (copied > ulen) 1144 copied = ulen; 1145 else if (copied < ulen) 1146 msg->msg_flags |= MSG_TRUNC; 1147 1148 /* 1149 * If checksum is needed at all, try to do it while copying the 1150 * data. If the data is truncated, or if we only want a partial 1151 * coverage checksum (UDP-Lite), do it before the copy. 1152 */ 1153 1154 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1155 if (udp_lib_checksum_complete(skb)) 1156 goto csum_copy_err; 1157 } 1158 1159 if (skb_csum_unnecessary(skb)) 1160 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 1161 msg->msg_iov, copied); 1162 else { 1163 err = skb_copy_and_csum_datagram_iovec(skb, 1164 sizeof(struct udphdr), 1165 msg->msg_iov); 1166 1167 if (err == -EINVAL) 1168 goto csum_copy_err; 1169 } 1170 1171 if (err) 1172 goto out_free; 1173 1174 if (!peeked) 1175 UDP_INC_STATS_USER(sock_net(sk), 1176 UDP_MIB_INDATAGRAMS, is_udplite); 1177 1178 sock_recv_ts_and_drops(msg, sk, skb); 1179 1180 /* Copy the address. */ 1181 if (sin) { 1182 sin->sin_family = AF_INET; 1183 sin->sin_port = udp_hdr(skb)->source; 1184 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1185 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1186 } 1187 if (inet->cmsg_flags) 1188 ip_cmsg_recv(msg, skb); 1189 1190 err = copied; 1191 if (flags & MSG_TRUNC) 1192 err = ulen; 1193 1194 out_free: 1195 skb_free_datagram_locked(sk, skb); 1196 out: 1197 return err; 1198 1199 csum_copy_err: 1200 lock_sock(sk); 1201 if (!skb_kill_datagram(sk, skb, flags)) 1202 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1203 release_sock(sk); 1204 1205 if (noblock) 1206 return -EAGAIN; 1207 goto try_again; 1208 } 1209 1210 1211 int udp_disconnect(struct sock *sk, int flags) 1212 { 1213 struct inet_sock *inet = inet_sk(sk); 1214 /* 1215 * 1003.1g - break association. 1216 */ 1217 1218 sk->sk_state = TCP_CLOSE; 1219 inet->inet_daddr = 0; 1220 inet->inet_dport = 0; 1221 sk->sk_bound_dev_if = 0; 1222 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1223 inet_reset_saddr(sk); 1224 1225 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1226 sk->sk_prot->unhash(sk); 1227 inet->inet_sport = 0; 1228 } 1229 sk_dst_reset(sk); 1230 return 0; 1231 } 1232 EXPORT_SYMBOL(udp_disconnect); 1233 1234 void udp_lib_unhash(struct sock *sk) 1235 { 1236 if (sk_hashed(sk)) { 1237 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1238 struct udp_hslot *hslot, *hslot2; 1239 1240 hslot = udp_hashslot(udptable, sock_net(sk), 1241 udp_sk(sk)->udp_port_hash); 1242 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1243 1244 spin_lock_bh(&hslot->lock); 1245 if (sk_nulls_del_node_init_rcu(sk)) { 1246 hslot->count--; 1247 inet_sk(sk)->inet_num = 0; 1248 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1249 1250 spin_lock(&hslot2->lock); 1251 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1252 hslot2->count--; 1253 spin_unlock(&hslot2->lock); 1254 } 1255 spin_unlock_bh(&hslot->lock); 1256 } 1257 } 1258 EXPORT_SYMBOL(udp_lib_unhash); 1259 1260 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1261 { 1262 int rc = sock_queue_rcv_skb(sk, skb); 1263 1264 if (rc < 0) { 1265 int is_udplite = IS_UDPLITE(sk); 1266 1267 /* Note that an ENOMEM error is charged twice */ 1268 if (rc == -ENOMEM) 1269 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1270 is_udplite); 1271 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1272 kfree_skb(skb); 1273 return -1; 1274 } 1275 1276 return 0; 1277 1278 } 1279 1280 /* returns: 1281 * -1: error 1282 * 0: success 1283 * >0: "udp encap" protocol resubmission 1284 * 1285 * Note that in the success and error cases, the skb is assumed to 1286 * have either been requeued or freed. 1287 */ 1288 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1289 { 1290 struct udp_sock *up = udp_sk(sk); 1291 int rc; 1292 int is_udplite = IS_UDPLITE(sk); 1293 1294 /* 1295 * Charge it to the socket, dropping if the queue is full. 1296 */ 1297 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1298 goto drop; 1299 nf_reset(skb); 1300 1301 if (up->encap_type) { 1302 /* 1303 * This is an encapsulation socket so pass the skb to 1304 * the socket's udp_encap_rcv() hook. Otherwise, just 1305 * fall through and pass this up the UDP socket. 1306 * up->encap_rcv() returns the following value: 1307 * =0 if skb was successfully passed to the encap 1308 * handler or was discarded by it. 1309 * >0 if skb should be passed on to UDP. 1310 * <0 if skb should be resubmitted as proto -N 1311 */ 1312 1313 /* if we're overly short, let UDP handle it */ 1314 if (skb->len > sizeof(struct udphdr) && 1315 up->encap_rcv != NULL) { 1316 int ret; 1317 1318 ret = (*up->encap_rcv)(sk, skb); 1319 if (ret <= 0) { 1320 UDP_INC_STATS_BH(sock_net(sk), 1321 UDP_MIB_INDATAGRAMS, 1322 is_udplite); 1323 return -ret; 1324 } 1325 } 1326 1327 /* FALLTHROUGH -- it's a UDP Packet */ 1328 } 1329 1330 /* 1331 * UDP-Lite specific tests, ignored on UDP sockets 1332 */ 1333 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1334 1335 /* 1336 * MIB statistics other than incrementing the error count are 1337 * disabled for the following two types of errors: these depend 1338 * on the application settings, not on the functioning of the 1339 * protocol stack as such. 1340 * 1341 * RFC 3828 here recommends (sec 3.3): "There should also be a 1342 * way ... to ... at least let the receiving application block 1343 * delivery of packets with coverage values less than a value 1344 * provided by the application." 1345 */ 1346 if (up->pcrlen == 0) { /* full coverage was set */ 1347 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1348 "%d while full coverage %d requested\n", 1349 UDP_SKB_CB(skb)->cscov, skb->len); 1350 goto drop; 1351 } 1352 /* The next case involves violating the min. coverage requested 1353 * by the receiver. This is subtle: if receiver wants x and x is 1354 * greater than the buffersize/MTU then receiver will complain 1355 * that it wants x while sender emits packets of smaller size y. 1356 * Therefore the above ...()->partial_cov statement is essential. 1357 */ 1358 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1359 LIMIT_NETDEBUG(KERN_WARNING 1360 "UDPLITE: coverage %d too small, need min %d\n", 1361 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1362 goto drop; 1363 } 1364 } 1365 1366 if (sk->sk_filter) { 1367 if (udp_lib_checksum_complete(skb)) 1368 goto drop; 1369 } 1370 1371 rc = 0; 1372 1373 bh_lock_sock(sk); 1374 if (!sock_owned_by_user(sk)) 1375 rc = __udp_queue_rcv_skb(sk, skb); 1376 else 1377 sk_add_backlog(sk, skb); 1378 bh_unlock_sock(sk); 1379 1380 return rc; 1381 1382 drop: 1383 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1384 atomic_inc(&sk->sk_drops); 1385 kfree_skb(skb); 1386 return -1; 1387 } 1388 1389 1390 static void flush_stack(struct sock **stack, unsigned int count, 1391 struct sk_buff *skb, unsigned int final) 1392 { 1393 unsigned int i; 1394 struct sk_buff *skb1 = NULL; 1395 struct sock *sk; 1396 1397 for (i = 0; i < count; i++) { 1398 sk = stack[i]; 1399 if (likely(skb1 == NULL)) 1400 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1401 1402 if (!skb1) { 1403 atomic_inc(&sk->sk_drops); 1404 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1405 IS_UDPLITE(sk)); 1406 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1407 IS_UDPLITE(sk)); 1408 } 1409 1410 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1411 skb1 = NULL; 1412 } 1413 if (unlikely(skb1)) 1414 kfree_skb(skb1); 1415 } 1416 1417 /* 1418 * Multicasts and broadcasts go to each listener. 1419 * 1420 * Note: called only from the BH handler context. 1421 */ 1422 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1423 struct udphdr *uh, 1424 __be32 saddr, __be32 daddr, 1425 struct udp_table *udptable) 1426 { 1427 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1428 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1429 int dif; 1430 unsigned int i, count = 0; 1431 1432 spin_lock(&hslot->lock); 1433 sk = sk_nulls_head(&hslot->head); 1434 dif = skb->dev->ifindex; 1435 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1436 while (sk) { 1437 stack[count++] = sk; 1438 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1439 daddr, uh->source, saddr, dif); 1440 if (unlikely(count == ARRAY_SIZE(stack))) { 1441 if (!sk) 1442 break; 1443 flush_stack(stack, count, skb, ~0); 1444 count = 0; 1445 } 1446 } 1447 /* 1448 * before releasing chain lock, we must take a reference on sockets 1449 */ 1450 for (i = 0; i < count; i++) 1451 sock_hold(stack[i]); 1452 1453 spin_unlock(&hslot->lock); 1454 1455 /* 1456 * do the slow work with no lock held 1457 */ 1458 if (count) { 1459 flush_stack(stack, count, skb, count - 1); 1460 1461 for (i = 0; i < count; i++) 1462 sock_put(stack[i]); 1463 } else { 1464 kfree_skb(skb); 1465 } 1466 return 0; 1467 } 1468 1469 /* Initialize UDP checksum. If exited with zero value (success), 1470 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1471 * Otherwise, csum completion requires chacksumming packet body, 1472 * including udp header and folding it to skb->csum. 1473 */ 1474 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1475 int proto) 1476 { 1477 const struct iphdr *iph; 1478 int err; 1479 1480 UDP_SKB_CB(skb)->partial_cov = 0; 1481 UDP_SKB_CB(skb)->cscov = skb->len; 1482 1483 if (proto == IPPROTO_UDPLITE) { 1484 err = udplite_checksum_init(skb, uh); 1485 if (err) 1486 return err; 1487 } 1488 1489 iph = ip_hdr(skb); 1490 if (uh->check == 0) { 1491 skb->ip_summed = CHECKSUM_UNNECESSARY; 1492 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1493 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1494 proto, skb->csum)) 1495 skb->ip_summed = CHECKSUM_UNNECESSARY; 1496 } 1497 if (!skb_csum_unnecessary(skb)) 1498 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1499 skb->len, proto, 0); 1500 /* Probably, we should checksum udp header (it should be in cache 1501 * in any case) and data in tiny packets (< rx copybreak). 1502 */ 1503 1504 return 0; 1505 } 1506 1507 /* 1508 * All we need to do is get the socket, and then do a checksum. 1509 */ 1510 1511 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1512 int proto) 1513 { 1514 struct sock *sk; 1515 struct udphdr *uh; 1516 unsigned short ulen; 1517 struct rtable *rt = skb_rtable(skb); 1518 __be32 saddr, daddr; 1519 struct net *net = dev_net(skb->dev); 1520 1521 /* 1522 * Validate the packet. 1523 */ 1524 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1525 goto drop; /* No space for header. */ 1526 1527 uh = udp_hdr(skb); 1528 ulen = ntohs(uh->len); 1529 if (ulen > skb->len) 1530 goto short_packet; 1531 1532 if (proto == IPPROTO_UDP) { 1533 /* UDP validates ulen. */ 1534 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1535 goto short_packet; 1536 uh = udp_hdr(skb); 1537 } 1538 1539 if (udp4_csum_init(skb, uh, proto)) 1540 goto csum_error; 1541 1542 saddr = ip_hdr(skb)->saddr; 1543 daddr = ip_hdr(skb)->daddr; 1544 1545 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1546 return __udp4_lib_mcast_deliver(net, skb, uh, 1547 saddr, daddr, udptable); 1548 1549 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1550 1551 if (sk != NULL) { 1552 int ret = udp_queue_rcv_skb(sk, skb); 1553 sock_put(sk); 1554 1555 /* a return value > 0 means to resubmit the input, but 1556 * it wants the return to be -protocol, or 0 1557 */ 1558 if (ret > 0) 1559 return -ret; 1560 return 0; 1561 } 1562 1563 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1564 goto drop; 1565 nf_reset(skb); 1566 1567 /* No socket. Drop packet silently, if checksum is wrong */ 1568 if (udp_lib_checksum_complete(skb)) 1569 goto csum_error; 1570 1571 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1572 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1573 1574 /* 1575 * Hmm. We got an UDP packet to a port to which we 1576 * don't wanna listen. Ignore it. 1577 */ 1578 kfree_skb(skb); 1579 return 0; 1580 1581 short_packet: 1582 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1583 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1584 &saddr, 1585 ntohs(uh->source), 1586 ulen, 1587 skb->len, 1588 &daddr, 1589 ntohs(uh->dest)); 1590 goto drop; 1591 1592 csum_error: 1593 /* 1594 * RFC1122: OK. Discards the bad packet silently (as far as 1595 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1596 */ 1597 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1598 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1599 &saddr, 1600 ntohs(uh->source), 1601 &daddr, 1602 ntohs(uh->dest), 1603 ulen); 1604 drop: 1605 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1606 kfree_skb(skb); 1607 return 0; 1608 } 1609 1610 int udp_rcv(struct sk_buff *skb) 1611 { 1612 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1613 } 1614 1615 void udp_destroy_sock(struct sock *sk) 1616 { 1617 lock_sock(sk); 1618 udp_flush_pending_frames(sk); 1619 release_sock(sk); 1620 } 1621 1622 /* 1623 * Socket option code for UDP 1624 */ 1625 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1626 char __user *optval, unsigned int optlen, 1627 int (*push_pending_frames)(struct sock *)) 1628 { 1629 struct udp_sock *up = udp_sk(sk); 1630 int val; 1631 int err = 0; 1632 int is_udplite = IS_UDPLITE(sk); 1633 1634 if (optlen < sizeof(int)) 1635 return -EINVAL; 1636 1637 if (get_user(val, (int __user *)optval)) 1638 return -EFAULT; 1639 1640 switch (optname) { 1641 case UDP_CORK: 1642 if (val != 0) { 1643 up->corkflag = 1; 1644 } else { 1645 up->corkflag = 0; 1646 lock_sock(sk); 1647 (*push_pending_frames)(sk); 1648 release_sock(sk); 1649 } 1650 break; 1651 1652 case UDP_ENCAP: 1653 switch (val) { 1654 case 0: 1655 case UDP_ENCAP_ESPINUDP: 1656 case UDP_ENCAP_ESPINUDP_NON_IKE: 1657 up->encap_rcv = xfrm4_udp_encap_rcv; 1658 /* FALLTHROUGH */ 1659 case UDP_ENCAP_L2TPINUDP: 1660 up->encap_type = val; 1661 break; 1662 default: 1663 err = -ENOPROTOOPT; 1664 break; 1665 } 1666 break; 1667 1668 /* 1669 * UDP-Lite's partial checksum coverage (RFC 3828). 1670 */ 1671 /* The sender sets actual checksum coverage length via this option. 1672 * The case coverage > packet length is handled by send module. */ 1673 case UDPLITE_SEND_CSCOV: 1674 if (!is_udplite) /* Disable the option on UDP sockets */ 1675 return -ENOPROTOOPT; 1676 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1677 val = 8; 1678 else if (val > USHORT_MAX) 1679 val = USHORT_MAX; 1680 up->pcslen = val; 1681 up->pcflag |= UDPLITE_SEND_CC; 1682 break; 1683 1684 /* The receiver specifies a minimum checksum coverage value. To make 1685 * sense, this should be set to at least 8 (as done below). If zero is 1686 * used, this again means full checksum coverage. */ 1687 case UDPLITE_RECV_CSCOV: 1688 if (!is_udplite) /* Disable the option on UDP sockets */ 1689 return -ENOPROTOOPT; 1690 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1691 val = 8; 1692 else if (val > USHORT_MAX) 1693 val = USHORT_MAX; 1694 up->pcrlen = val; 1695 up->pcflag |= UDPLITE_RECV_CC; 1696 break; 1697 1698 default: 1699 err = -ENOPROTOOPT; 1700 break; 1701 } 1702 1703 return err; 1704 } 1705 EXPORT_SYMBOL(udp_lib_setsockopt); 1706 1707 int udp_setsockopt(struct sock *sk, int level, int optname, 1708 char __user *optval, unsigned int optlen) 1709 { 1710 if (level == SOL_UDP || level == SOL_UDPLITE) 1711 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1712 udp_push_pending_frames); 1713 return ip_setsockopt(sk, level, optname, optval, optlen); 1714 } 1715 1716 #ifdef CONFIG_COMPAT 1717 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1718 char __user *optval, unsigned int optlen) 1719 { 1720 if (level == SOL_UDP || level == SOL_UDPLITE) 1721 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1722 udp_push_pending_frames); 1723 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1724 } 1725 #endif 1726 1727 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1728 char __user *optval, int __user *optlen) 1729 { 1730 struct udp_sock *up = udp_sk(sk); 1731 int val, len; 1732 1733 if (get_user(len, optlen)) 1734 return -EFAULT; 1735 1736 len = min_t(unsigned int, len, sizeof(int)); 1737 1738 if (len < 0) 1739 return -EINVAL; 1740 1741 switch (optname) { 1742 case UDP_CORK: 1743 val = up->corkflag; 1744 break; 1745 1746 case UDP_ENCAP: 1747 val = up->encap_type; 1748 break; 1749 1750 /* The following two cannot be changed on UDP sockets, the return is 1751 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1752 case UDPLITE_SEND_CSCOV: 1753 val = up->pcslen; 1754 break; 1755 1756 case UDPLITE_RECV_CSCOV: 1757 val = up->pcrlen; 1758 break; 1759 1760 default: 1761 return -ENOPROTOOPT; 1762 } 1763 1764 if (put_user(len, optlen)) 1765 return -EFAULT; 1766 if (copy_to_user(optval, &val, len)) 1767 return -EFAULT; 1768 return 0; 1769 } 1770 EXPORT_SYMBOL(udp_lib_getsockopt); 1771 1772 int udp_getsockopt(struct sock *sk, int level, int optname, 1773 char __user *optval, int __user *optlen) 1774 { 1775 if (level == SOL_UDP || level == SOL_UDPLITE) 1776 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1777 return ip_getsockopt(sk, level, optname, optval, optlen); 1778 } 1779 1780 #ifdef CONFIG_COMPAT 1781 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1782 char __user *optval, int __user *optlen) 1783 { 1784 if (level == SOL_UDP || level == SOL_UDPLITE) 1785 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1786 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1787 } 1788 #endif 1789 /** 1790 * udp_poll - wait for a UDP event. 1791 * @file - file struct 1792 * @sock - socket 1793 * @wait - poll table 1794 * 1795 * This is same as datagram poll, except for the special case of 1796 * blocking sockets. If application is using a blocking fd 1797 * and a packet with checksum error is in the queue; 1798 * then it could get return from select indicating data available 1799 * but then block when reading it. Add special case code 1800 * to work around these arguably broken applications. 1801 */ 1802 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1803 { 1804 unsigned int mask = datagram_poll(file, sock, wait); 1805 struct sock *sk = sock->sk; 1806 1807 /* Check for false positives due to checksum errors */ 1808 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 1809 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 1810 mask &= ~(POLLIN | POLLRDNORM); 1811 1812 return mask; 1813 1814 } 1815 EXPORT_SYMBOL(udp_poll); 1816 1817 struct proto udp_prot = { 1818 .name = "UDP", 1819 .owner = THIS_MODULE, 1820 .close = udp_lib_close, 1821 .connect = ip4_datagram_connect, 1822 .disconnect = udp_disconnect, 1823 .ioctl = udp_ioctl, 1824 .destroy = udp_destroy_sock, 1825 .setsockopt = udp_setsockopt, 1826 .getsockopt = udp_getsockopt, 1827 .sendmsg = udp_sendmsg, 1828 .recvmsg = udp_recvmsg, 1829 .sendpage = udp_sendpage, 1830 .backlog_rcv = __udp_queue_rcv_skb, 1831 .hash = udp_lib_hash, 1832 .unhash = udp_lib_unhash, 1833 .get_port = udp_v4_get_port, 1834 .memory_allocated = &udp_memory_allocated, 1835 .sysctl_mem = sysctl_udp_mem, 1836 .sysctl_wmem = &sysctl_udp_wmem_min, 1837 .sysctl_rmem = &sysctl_udp_rmem_min, 1838 .obj_size = sizeof(struct udp_sock), 1839 .slab_flags = SLAB_DESTROY_BY_RCU, 1840 .h.udp_table = &udp_table, 1841 #ifdef CONFIG_COMPAT 1842 .compat_setsockopt = compat_udp_setsockopt, 1843 .compat_getsockopt = compat_udp_getsockopt, 1844 #endif 1845 }; 1846 EXPORT_SYMBOL(udp_prot); 1847 1848 /* ------------------------------------------------------------------------ */ 1849 #ifdef CONFIG_PROC_FS 1850 1851 static struct sock *udp_get_first(struct seq_file *seq, int start) 1852 { 1853 struct sock *sk; 1854 struct udp_iter_state *state = seq->private; 1855 struct net *net = seq_file_net(seq); 1856 1857 for (state->bucket = start; state->bucket <= state->udp_table->mask; 1858 ++state->bucket) { 1859 struct hlist_nulls_node *node; 1860 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 1861 1862 if (hlist_nulls_empty(&hslot->head)) 1863 continue; 1864 1865 spin_lock_bh(&hslot->lock); 1866 sk_nulls_for_each(sk, node, &hslot->head) { 1867 if (!net_eq(sock_net(sk), net)) 1868 continue; 1869 if (sk->sk_family == state->family) 1870 goto found; 1871 } 1872 spin_unlock_bh(&hslot->lock); 1873 } 1874 sk = NULL; 1875 found: 1876 return sk; 1877 } 1878 1879 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1880 { 1881 struct udp_iter_state *state = seq->private; 1882 struct net *net = seq_file_net(seq); 1883 1884 do { 1885 sk = sk_nulls_next(sk); 1886 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1887 1888 if (!sk) { 1889 if (state->bucket <= state->udp_table->mask) 1890 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1891 return udp_get_first(seq, state->bucket + 1); 1892 } 1893 return sk; 1894 } 1895 1896 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1897 { 1898 struct sock *sk = udp_get_first(seq, 0); 1899 1900 if (sk) 1901 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1902 --pos; 1903 return pos ? NULL : sk; 1904 } 1905 1906 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1907 { 1908 struct udp_iter_state *state = seq->private; 1909 state->bucket = MAX_UDP_PORTS; 1910 1911 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 1912 } 1913 1914 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1915 { 1916 struct sock *sk; 1917 1918 if (v == SEQ_START_TOKEN) 1919 sk = udp_get_idx(seq, 0); 1920 else 1921 sk = udp_get_next(seq, v); 1922 1923 ++*pos; 1924 return sk; 1925 } 1926 1927 static void udp_seq_stop(struct seq_file *seq, void *v) 1928 { 1929 struct udp_iter_state *state = seq->private; 1930 1931 if (state->bucket <= state->udp_table->mask) 1932 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1933 } 1934 1935 static int udp_seq_open(struct inode *inode, struct file *file) 1936 { 1937 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1938 struct udp_iter_state *s; 1939 int err; 1940 1941 err = seq_open_net(inode, file, &afinfo->seq_ops, 1942 sizeof(struct udp_iter_state)); 1943 if (err < 0) 1944 return err; 1945 1946 s = ((struct seq_file *)file->private_data)->private; 1947 s->family = afinfo->family; 1948 s->udp_table = afinfo->udp_table; 1949 return err; 1950 } 1951 1952 /* ------------------------------------------------------------------------ */ 1953 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 1954 { 1955 struct proc_dir_entry *p; 1956 int rc = 0; 1957 1958 afinfo->seq_fops.open = udp_seq_open; 1959 afinfo->seq_fops.read = seq_read; 1960 afinfo->seq_fops.llseek = seq_lseek; 1961 afinfo->seq_fops.release = seq_release_net; 1962 1963 afinfo->seq_ops.start = udp_seq_start; 1964 afinfo->seq_ops.next = udp_seq_next; 1965 afinfo->seq_ops.stop = udp_seq_stop; 1966 1967 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 1968 &afinfo->seq_fops, afinfo); 1969 if (!p) 1970 rc = -ENOMEM; 1971 return rc; 1972 } 1973 EXPORT_SYMBOL(udp_proc_register); 1974 1975 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 1976 { 1977 proc_net_remove(net, afinfo->name); 1978 } 1979 EXPORT_SYMBOL(udp_proc_unregister); 1980 1981 /* ------------------------------------------------------------------------ */ 1982 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 1983 int bucket, int *len) 1984 { 1985 struct inet_sock *inet = inet_sk(sp); 1986 __be32 dest = inet->inet_daddr; 1987 __be32 src = inet->inet_rcv_saddr; 1988 __u16 destp = ntohs(inet->inet_dport); 1989 __u16 srcp = ntohs(inet->inet_sport); 1990 1991 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 1992 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 1993 bucket, src, srcp, dest, destp, sp->sk_state, 1994 sk_wmem_alloc_get(sp), 1995 sk_rmem_alloc_get(sp), 1996 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1997 atomic_read(&sp->sk_refcnt), sp, 1998 atomic_read(&sp->sk_drops), len); 1999 } 2000 2001 int udp4_seq_show(struct seq_file *seq, void *v) 2002 { 2003 if (v == SEQ_START_TOKEN) 2004 seq_printf(seq, "%-127s\n", 2005 " sl local_address rem_address st tx_queue " 2006 "rx_queue tr tm->when retrnsmt uid timeout " 2007 "inode ref pointer drops"); 2008 else { 2009 struct udp_iter_state *state = seq->private; 2010 int len; 2011 2012 udp4_format_sock(v, seq, state->bucket, &len); 2013 seq_printf(seq, "%*s\n", 127 - len, ""); 2014 } 2015 return 0; 2016 } 2017 2018 /* ------------------------------------------------------------------------ */ 2019 static struct udp_seq_afinfo udp4_seq_afinfo = { 2020 .name = "udp", 2021 .family = AF_INET, 2022 .udp_table = &udp_table, 2023 .seq_fops = { 2024 .owner = THIS_MODULE, 2025 }, 2026 .seq_ops = { 2027 .show = udp4_seq_show, 2028 }, 2029 }; 2030 2031 static int udp4_proc_init_net(struct net *net) 2032 { 2033 return udp_proc_register(net, &udp4_seq_afinfo); 2034 } 2035 2036 static void udp4_proc_exit_net(struct net *net) 2037 { 2038 udp_proc_unregister(net, &udp4_seq_afinfo); 2039 } 2040 2041 static struct pernet_operations udp4_net_ops = { 2042 .init = udp4_proc_init_net, 2043 .exit = udp4_proc_exit_net, 2044 }; 2045 2046 int __init udp4_proc_init(void) 2047 { 2048 return register_pernet_subsys(&udp4_net_ops); 2049 } 2050 2051 void udp4_proc_exit(void) 2052 { 2053 unregister_pernet_subsys(&udp4_net_ops); 2054 } 2055 #endif /* CONFIG_PROC_FS */ 2056 2057 static __initdata unsigned long uhash_entries; 2058 static int __init set_uhash_entries(char *str) 2059 { 2060 if (!str) 2061 return 0; 2062 uhash_entries = simple_strtoul(str, &str, 0); 2063 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2064 uhash_entries = UDP_HTABLE_SIZE_MIN; 2065 return 1; 2066 } 2067 __setup("uhash_entries=", set_uhash_entries); 2068 2069 void __init udp_table_init(struct udp_table *table, const char *name) 2070 { 2071 unsigned int i; 2072 2073 if (!CONFIG_BASE_SMALL) 2074 table->hash = alloc_large_system_hash(name, 2075 2 * sizeof(struct udp_hslot), 2076 uhash_entries, 2077 21, /* one slot per 2 MB */ 2078 0, 2079 &table->log, 2080 &table->mask, 2081 64 * 1024); 2082 /* 2083 * Make sure hash table has the minimum size 2084 */ 2085 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) { 2086 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN * 2087 2 * sizeof(struct udp_hslot), GFP_KERNEL); 2088 if (!table->hash) 2089 panic(name); 2090 table->log = ilog2(UDP_HTABLE_SIZE_MIN); 2091 table->mask = UDP_HTABLE_SIZE_MIN - 1; 2092 } 2093 table->hash2 = table->hash + (table->mask + 1); 2094 for (i = 0; i <= table->mask; i++) { 2095 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2096 table->hash[i].count = 0; 2097 spin_lock_init(&table->hash[i].lock); 2098 } 2099 for (i = 0; i <= table->mask; i++) { 2100 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2101 table->hash2[i].count = 0; 2102 spin_lock_init(&table->hash2[i].lock); 2103 } 2104 } 2105 2106 void __init udp_init(void) 2107 { 2108 unsigned long nr_pages, limit; 2109 2110 udp_table_init(&udp_table, "UDP"); 2111 /* Set the pressure threshold up by the same strategy of TCP. It is a 2112 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 2113 * toward zero with the amount of memory, with a floor of 128 pages. 2114 */ 2115 nr_pages = totalram_pages - totalhigh_pages; 2116 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 2117 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 2118 limit = max(limit, 128UL); 2119 sysctl_udp_mem[0] = limit / 4 * 3; 2120 sysctl_udp_mem[1] = limit; 2121 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2122 2123 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2124 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2125 } 2126 2127 int udp4_ufo_send_check(struct sk_buff *skb) 2128 { 2129 const struct iphdr *iph; 2130 struct udphdr *uh; 2131 2132 if (!pskb_may_pull(skb, sizeof(*uh))) 2133 return -EINVAL; 2134 2135 iph = ip_hdr(skb); 2136 uh = udp_hdr(skb); 2137 2138 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 2139 IPPROTO_UDP, 0); 2140 skb->csum_start = skb_transport_header(skb) - skb->head; 2141 skb->csum_offset = offsetof(struct udphdr, check); 2142 skb->ip_summed = CHECKSUM_PARTIAL; 2143 return 0; 2144 } 2145 2146 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features) 2147 { 2148 struct sk_buff *segs = ERR_PTR(-EINVAL); 2149 unsigned int mss; 2150 int offset; 2151 __wsum csum; 2152 2153 mss = skb_shinfo(skb)->gso_size; 2154 if (unlikely(skb->len <= mss)) 2155 goto out; 2156 2157 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 2158 /* Packet is from an untrusted source, reset gso_segs. */ 2159 int type = skb_shinfo(skb)->gso_type; 2160 2161 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || 2162 !(type & (SKB_GSO_UDP)))) 2163 goto out; 2164 2165 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 2166 2167 segs = NULL; 2168 goto out; 2169 } 2170 2171 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot 2172 * do checksum of UDP packets sent as multiple IP fragments. 2173 */ 2174 offset = skb->csum_start - skb_headroom(skb); 2175 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2176 offset += skb->csum_offset; 2177 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2178 skb->ip_summed = CHECKSUM_NONE; 2179 2180 /* Fragment the skb. IP headers of the fragments are updated in 2181 * inet_gso_segment() 2182 */ 2183 segs = skb_segment(skb, features); 2184 out: 2185 return segs; 2186 } 2187 2188