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 <linux/slab.h> 99 #include <net/tcp_states.h> 100 #include <linux/skbuff.h> 101 #include <linux/proc_fs.h> 102 #include <linux/seq_file.h> 103 #include <net/net_namespace.h> 104 #include <net/icmp.h> 105 #include <net/route.h> 106 #include <net/checksum.h> 107 #include <net/xfrm.h> 108 #include "udp_impl.h" 109 110 struct udp_table udp_table __read_mostly; 111 EXPORT_SYMBOL(udp_table); 112 113 int sysctl_udp_mem[3] __read_mostly; 114 EXPORT_SYMBOL(sysctl_udp_mem); 115 116 int sysctl_udp_rmem_min __read_mostly; 117 EXPORT_SYMBOL(sysctl_udp_rmem_min); 118 119 int sysctl_udp_wmem_min __read_mostly; 120 EXPORT_SYMBOL(sysctl_udp_wmem_min); 121 122 atomic_t udp_memory_allocated; 123 EXPORT_SYMBOL(udp_memory_allocated); 124 125 #define MAX_UDP_PORTS 65536 126 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 127 128 static int udp_lib_lport_inuse(struct net *net, __u16 num, 129 const struct udp_hslot *hslot, 130 unsigned long *bitmap, 131 struct sock *sk, 132 int (*saddr_comp)(const struct sock *sk1, 133 const struct sock *sk2), 134 unsigned int log) 135 { 136 struct sock *sk2; 137 struct hlist_nulls_node *node; 138 139 sk_nulls_for_each(sk2, node, &hslot->head) 140 if (net_eq(sock_net(sk2), net) && 141 sk2 != sk && 142 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 143 (!sk2->sk_reuse || !sk->sk_reuse) && 144 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 145 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 146 (*saddr_comp)(sk, sk2)) { 147 if (bitmap) 148 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 149 bitmap); 150 else 151 return 1; 152 } 153 return 0; 154 } 155 156 /* 157 * Note: we still hold spinlock of primary hash chain, so no other writer 158 * can insert/delete a socket with local_port == num 159 */ 160 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 161 struct udp_hslot *hslot2, 162 struct sock *sk, 163 int (*saddr_comp)(const struct sock *sk1, 164 const struct sock *sk2)) 165 { 166 struct sock *sk2; 167 struct hlist_nulls_node *node; 168 int res = 0; 169 170 spin_lock(&hslot2->lock); 171 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) 172 if (net_eq(sock_net(sk2), net) && 173 sk2 != sk && 174 (udp_sk(sk2)->udp_port_hash == num) && 175 (!sk2->sk_reuse || !sk->sk_reuse) && 176 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 177 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 178 (*saddr_comp)(sk, sk2)) { 179 res = 1; 180 break; 181 } 182 spin_unlock(&hslot2->lock); 183 return res; 184 } 185 186 /** 187 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 188 * 189 * @sk: socket struct in question 190 * @snum: port number to look up 191 * @saddr_comp: AF-dependent comparison of bound local IP addresses 192 * @hash2_nulladdr: AF-dependant hash value in secondary hash chains, 193 * with NULL address 194 */ 195 int udp_lib_get_port(struct sock *sk, unsigned short snum, 196 int (*saddr_comp)(const struct sock *sk1, 197 const struct sock *sk2), 198 unsigned int hash2_nulladdr) 199 { 200 struct udp_hslot *hslot, *hslot2; 201 struct udp_table *udptable = sk->sk_prot->h.udp_table; 202 int error = 1; 203 struct net *net = sock_net(sk); 204 205 if (!snum) { 206 int low, high, remaining; 207 unsigned rand; 208 unsigned short first, last; 209 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 210 211 inet_get_local_port_range(&low, &high); 212 remaining = (high - low) + 1; 213 214 rand = net_random(); 215 first = (((u64)rand * remaining) >> 32) + low; 216 /* 217 * force rand to be an odd multiple of UDP_HTABLE_SIZE 218 */ 219 rand = (rand | 1) * (udptable->mask + 1); 220 last = first + udptable->mask + 1; 221 do { 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 !inet_is_reserved_local_port(snum)) 238 goto found; 239 snum += rand; 240 } while (snum != first); 241 spin_unlock_bh(&hslot->lock); 242 } while (++first != last); 243 goto fail; 244 } else { 245 hslot = udp_hashslot(udptable, net, snum); 246 spin_lock_bh(&hslot->lock); 247 if (hslot->count > 10) { 248 int exist; 249 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 250 251 slot2 &= udptable->mask; 252 hash2_nulladdr &= udptable->mask; 253 254 hslot2 = udp_hashslot2(udptable, slot2); 255 if (hslot->count < hslot2->count) 256 goto scan_primary_hash; 257 258 exist = udp_lib_lport_inuse2(net, snum, hslot2, 259 sk, saddr_comp); 260 if (!exist && (hash2_nulladdr != slot2)) { 261 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 262 exist = udp_lib_lport_inuse2(net, snum, hslot2, 263 sk, saddr_comp); 264 } 265 if (exist) 266 goto fail_unlock; 267 else 268 goto found; 269 } 270 scan_primary_hash: 271 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 272 saddr_comp, 0)) 273 goto fail_unlock; 274 } 275 found: 276 inet_sk(sk)->inet_num = snum; 277 udp_sk(sk)->udp_port_hash = snum; 278 udp_sk(sk)->udp_portaddr_hash ^= snum; 279 if (sk_unhashed(sk)) { 280 sk_nulls_add_node_rcu(sk, &hslot->head); 281 hslot->count++; 282 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 283 284 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 285 spin_lock(&hslot2->lock); 286 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 287 &hslot2->head); 288 hslot2->count++; 289 spin_unlock(&hslot2->lock); 290 } 291 error = 0; 292 fail_unlock: 293 spin_unlock_bh(&hslot->lock); 294 fail: 295 return error; 296 } 297 EXPORT_SYMBOL(udp_lib_get_port); 298 299 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 300 { 301 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 302 303 return (!ipv6_only_sock(sk2) && 304 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 305 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 306 } 307 308 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr, 309 unsigned int port) 310 { 311 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 312 } 313 314 int udp_v4_get_port(struct sock *sk, unsigned short snum) 315 { 316 unsigned int hash2_nulladdr = 317 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 318 unsigned int hash2_partial = 319 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 320 321 /* precompute partial secondary hash */ 322 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 323 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 324 } 325 326 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, 327 unsigned short hnum, 328 __be16 sport, __be32 daddr, __be16 dport, int dif) 329 { 330 int score = -1; 331 332 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 333 !ipv6_only_sock(sk)) { 334 struct inet_sock *inet = inet_sk(sk); 335 336 score = (sk->sk_family == PF_INET ? 1 : 0); 337 if (inet->inet_rcv_saddr) { 338 if (inet->inet_rcv_saddr != daddr) 339 return -1; 340 score += 2; 341 } 342 if (inet->inet_daddr) { 343 if (inet->inet_daddr != saddr) 344 return -1; 345 score += 2; 346 } 347 if (inet->inet_dport) { 348 if (inet->inet_dport != sport) 349 return -1; 350 score += 2; 351 } 352 if (sk->sk_bound_dev_if) { 353 if (sk->sk_bound_dev_if != dif) 354 return -1; 355 score += 2; 356 } 357 } 358 return score; 359 } 360 361 /* 362 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 363 */ 364 #define SCORE2_MAX (1 + 2 + 2 + 2) 365 static inline int compute_score2(struct sock *sk, struct net *net, 366 __be32 saddr, __be16 sport, 367 __be32 daddr, unsigned int hnum, int dif) 368 { 369 int score = -1; 370 371 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) { 372 struct inet_sock *inet = inet_sk(sk); 373 374 if (inet->inet_rcv_saddr != daddr) 375 return -1; 376 if (inet->inet_num != hnum) 377 return -1; 378 379 score = (sk->sk_family == PF_INET ? 1 : 0); 380 if (inet->inet_daddr) { 381 if (inet->inet_daddr != saddr) 382 return -1; 383 score += 2; 384 } 385 if (inet->inet_dport) { 386 if (inet->inet_dport != sport) 387 return -1; 388 score += 2; 389 } 390 if (sk->sk_bound_dev_if) { 391 if (sk->sk_bound_dev_if != dif) 392 return -1; 393 score += 2; 394 } 395 } 396 return score; 397 } 398 399 400 /* called with read_rcu_lock() */ 401 static struct sock *udp4_lib_lookup2(struct net *net, 402 __be32 saddr, __be16 sport, 403 __be32 daddr, unsigned int hnum, int dif, 404 struct udp_hslot *hslot2, unsigned int slot2) 405 { 406 struct sock *sk, *result; 407 struct hlist_nulls_node *node; 408 int score, badness; 409 410 begin: 411 result = NULL; 412 badness = -1; 413 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 414 score = compute_score2(sk, net, saddr, sport, 415 daddr, hnum, dif); 416 if (score > badness) { 417 result = sk; 418 badness = score; 419 if (score == SCORE2_MAX) 420 goto exact_match; 421 } 422 } 423 /* 424 * if the nulls value we got at the end of this lookup is 425 * not the expected one, we must restart lookup. 426 * We probably met an item that was moved to another chain. 427 */ 428 if (get_nulls_value(node) != slot2) 429 goto begin; 430 431 if (result) { 432 exact_match: 433 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) 434 result = NULL; 435 else if (unlikely(compute_score2(result, net, saddr, sport, 436 daddr, hnum, dif) < badness)) { 437 sock_put(result); 438 goto begin; 439 } 440 } 441 return result; 442 } 443 444 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 445 * harder than this. -DaveM 446 */ 447 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 448 __be16 sport, __be32 daddr, __be16 dport, 449 int dif, struct udp_table *udptable) 450 { 451 struct sock *sk, *result; 452 struct hlist_nulls_node *node; 453 unsigned short hnum = ntohs(dport); 454 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 455 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 456 int score, badness; 457 458 rcu_read_lock(); 459 if (hslot->count > 10) { 460 hash2 = udp4_portaddr_hash(net, daddr, hnum); 461 slot2 = hash2 & udptable->mask; 462 hslot2 = &udptable->hash2[slot2]; 463 if (hslot->count < hslot2->count) 464 goto begin; 465 466 result = udp4_lib_lookup2(net, saddr, sport, 467 daddr, hnum, dif, 468 hslot2, slot2); 469 if (!result) { 470 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 471 slot2 = hash2 & udptable->mask; 472 hslot2 = &udptable->hash2[slot2]; 473 if (hslot->count < hslot2->count) 474 goto begin; 475 476 result = udp4_lib_lookup2(net, saddr, sport, 477 htonl(INADDR_ANY), hnum, dif, 478 hslot2, slot2); 479 } 480 rcu_read_unlock(); 481 return result; 482 } 483 begin: 484 result = NULL; 485 badness = -1; 486 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 487 score = compute_score(sk, net, saddr, hnum, sport, 488 daddr, dport, dif); 489 if (score > badness) { 490 result = sk; 491 badness = score; 492 } 493 } 494 /* 495 * if the nulls value we got at the end of this lookup is 496 * not the expected one, we must restart lookup. 497 * We probably met an item that was moved to another chain. 498 */ 499 if (get_nulls_value(node) != slot) 500 goto begin; 501 502 if (result) { 503 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) 504 result = NULL; 505 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 506 daddr, dport, dif) < badness)) { 507 sock_put(result); 508 goto begin; 509 } 510 } 511 rcu_read_unlock(); 512 return result; 513 } 514 515 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 516 __be16 sport, __be16 dport, 517 struct udp_table *udptable) 518 { 519 struct sock *sk; 520 const struct iphdr *iph = ip_hdr(skb); 521 522 if (unlikely(sk = skb_steal_sock(skb))) 523 return sk; 524 else 525 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 526 iph->daddr, dport, inet_iif(skb), 527 udptable); 528 } 529 530 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 531 __be32 daddr, __be16 dport, int dif) 532 { 533 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 534 } 535 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 536 537 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 538 __be16 loc_port, __be32 loc_addr, 539 __be16 rmt_port, __be32 rmt_addr, 540 int dif) 541 { 542 struct hlist_nulls_node *node; 543 struct sock *s = sk; 544 unsigned short hnum = ntohs(loc_port); 545 546 sk_nulls_for_each_from(s, node) { 547 struct inet_sock *inet = inet_sk(s); 548 549 if (!net_eq(sock_net(s), net) || 550 udp_sk(s)->udp_port_hash != hnum || 551 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 552 (inet->inet_dport != rmt_port && inet->inet_dport) || 553 (inet->inet_rcv_saddr && 554 inet->inet_rcv_saddr != loc_addr) || 555 ipv6_only_sock(s) || 556 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 557 continue; 558 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 559 continue; 560 goto found; 561 } 562 s = NULL; 563 found: 564 return s; 565 } 566 567 /* 568 * This routine is called by the ICMP module when it gets some 569 * sort of error condition. If err < 0 then the socket should 570 * be closed and the error returned to the user. If err > 0 571 * it's just the icmp type << 8 | icmp code. 572 * Header points to the ip header of the error packet. We move 573 * on past this. Then (as it used to claim before adjustment) 574 * header points to the first 8 bytes of the udp header. We need 575 * to find the appropriate port. 576 */ 577 578 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 579 { 580 struct inet_sock *inet; 581 struct iphdr *iph = (struct iphdr *)skb->data; 582 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 583 const int type = icmp_hdr(skb)->type; 584 const int code = icmp_hdr(skb)->code; 585 struct sock *sk; 586 int harderr; 587 int err; 588 struct net *net = dev_net(skb->dev); 589 590 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 591 iph->saddr, uh->source, skb->dev->ifindex, udptable); 592 if (sk == NULL) { 593 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 594 return; /* No socket for error */ 595 } 596 597 err = 0; 598 harderr = 0; 599 inet = inet_sk(sk); 600 601 switch (type) { 602 default: 603 case ICMP_TIME_EXCEEDED: 604 err = EHOSTUNREACH; 605 break; 606 case ICMP_SOURCE_QUENCH: 607 goto out; 608 case ICMP_PARAMETERPROB: 609 err = EPROTO; 610 harderr = 1; 611 break; 612 case ICMP_DEST_UNREACH: 613 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 614 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 615 err = EMSGSIZE; 616 harderr = 1; 617 break; 618 } 619 goto out; 620 } 621 err = EHOSTUNREACH; 622 if (code <= NR_ICMP_UNREACH) { 623 harderr = icmp_err_convert[code].fatal; 624 err = icmp_err_convert[code].errno; 625 } 626 break; 627 } 628 629 /* 630 * RFC1122: OK. Passes ICMP errors back to application, as per 631 * 4.1.3.3. 632 */ 633 if (!inet->recverr) { 634 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 635 goto out; 636 } else 637 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 638 639 sk->sk_err = err; 640 sk->sk_error_report(sk); 641 out: 642 sock_put(sk); 643 } 644 645 void udp_err(struct sk_buff *skb, u32 info) 646 { 647 __udp4_lib_err(skb, info, &udp_table); 648 } 649 650 /* 651 * Throw away all pending data and cancel the corking. Socket is locked. 652 */ 653 void udp_flush_pending_frames(struct sock *sk) 654 { 655 struct udp_sock *up = udp_sk(sk); 656 657 if (up->pending) { 658 up->len = 0; 659 up->pending = 0; 660 ip_flush_pending_frames(sk); 661 } 662 } 663 EXPORT_SYMBOL(udp_flush_pending_frames); 664 665 /** 666 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 667 * @sk: socket we are sending on 668 * @skb: sk_buff containing the filled-in UDP header 669 * (checksum field must be zeroed out) 670 */ 671 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 672 __be32 src, __be32 dst, int len) 673 { 674 unsigned int offset; 675 struct udphdr *uh = udp_hdr(skb); 676 __wsum csum = 0; 677 678 if (skb_queue_len(&sk->sk_write_queue) == 1) { 679 /* 680 * Only one fragment on the socket. 681 */ 682 skb->csum_start = skb_transport_header(skb) - skb->head; 683 skb->csum_offset = offsetof(struct udphdr, check); 684 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 685 } else { 686 /* 687 * HW-checksum won't work as there are two or more 688 * fragments on the socket so that all csums of sk_buffs 689 * should be together 690 */ 691 offset = skb_transport_offset(skb); 692 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 693 694 skb->ip_summed = CHECKSUM_NONE; 695 696 skb_queue_walk(&sk->sk_write_queue, skb) { 697 csum = csum_add(csum, skb->csum); 698 } 699 700 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 701 if (uh->check == 0) 702 uh->check = CSUM_MANGLED_0; 703 } 704 } 705 706 /* 707 * Push out all pending data as one UDP datagram. Socket is locked. 708 */ 709 static int udp_push_pending_frames(struct sock *sk) 710 { 711 struct udp_sock *up = udp_sk(sk); 712 struct inet_sock *inet = inet_sk(sk); 713 struct flowi *fl = &inet->cork.fl; 714 struct sk_buff *skb; 715 struct udphdr *uh; 716 int err = 0; 717 int is_udplite = IS_UDPLITE(sk); 718 __wsum csum = 0; 719 720 /* Grab the skbuff where UDP header space exists. */ 721 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 722 goto out; 723 724 /* 725 * Create a UDP header 726 */ 727 uh = udp_hdr(skb); 728 uh->source = fl->fl_ip_sport; 729 uh->dest = fl->fl_ip_dport; 730 uh->len = htons(up->len); 731 uh->check = 0; 732 733 if (is_udplite) /* UDP-Lite */ 734 csum = udplite_csum_outgoing(sk, skb); 735 736 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 737 738 skb->ip_summed = CHECKSUM_NONE; 739 goto send; 740 741 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 742 743 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len); 744 goto send; 745 746 } else /* `normal' UDP */ 747 csum = udp_csum_outgoing(sk, skb); 748 749 /* add protocol-dependent pseudo-header */ 750 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 751 sk->sk_protocol, csum); 752 if (uh->check == 0) 753 uh->check = CSUM_MANGLED_0; 754 755 send: 756 err = ip_push_pending_frames(sk); 757 if (err) { 758 if (err == -ENOBUFS && !inet->recverr) { 759 UDP_INC_STATS_USER(sock_net(sk), 760 UDP_MIB_SNDBUFERRORS, is_udplite); 761 err = 0; 762 } 763 } else 764 UDP_INC_STATS_USER(sock_net(sk), 765 UDP_MIB_OUTDATAGRAMS, is_udplite); 766 out: 767 up->len = 0; 768 up->pending = 0; 769 return err; 770 } 771 772 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 773 size_t len) 774 { 775 struct inet_sock *inet = inet_sk(sk); 776 struct udp_sock *up = udp_sk(sk); 777 int ulen = len; 778 struct ipcm_cookie ipc; 779 struct rtable *rt = NULL; 780 int free = 0; 781 int connected = 0; 782 __be32 daddr, faddr, saddr; 783 __be16 dport; 784 u8 tos; 785 int err, is_udplite = IS_UDPLITE(sk); 786 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 787 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 788 789 if (len > 0xFFFF) 790 return -EMSGSIZE; 791 792 /* 793 * Check the flags. 794 */ 795 796 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 797 return -EOPNOTSUPP; 798 799 ipc.opt = NULL; 800 ipc.tx_flags = 0; 801 802 if (up->pending) { 803 /* 804 * There are pending frames. 805 * The socket lock must be held while it's corked. 806 */ 807 lock_sock(sk); 808 if (likely(up->pending)) { 809 if (unlikely(up->pending != AF_INET)) { 810 release_sock(sk); 811 return -EINVAL; 812 } 813 goto do_append_data; 814 } 815 release_sock(sk); 816 } 817 ulen += sizeof(struct udphdr); 818 819 /* 820 * Get and verify the address. 821 */ 822 if (msg->msg_name) { 823 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name; 824 if (msg->msg_namelen < sizeof(*usin)) 825 return -EINVAL; 826 if (usin->sin_family != AF_INET) { 827 if (usin->sin_family != AF_UNSPEC) 828 return -EAFNOSUPPORT; 829 } 830 831 daddr = usin->sin_addr.s_addr; 832 dport = usin->sin_port; 833 if (dport == 0) 834 return -EINVAL; 835 } else { 836 if (sk->sk_state != TCP_ESTABLISHED) 837 return -EDESTADDRREQ; 838 daddr = inet->inet_daddr; 839 dport = inet->inet_dport; 840 /* Open fast path for connected socket. 841 Route will not be used, if at least one option is set. 842 */ 843 connected = 1; 844 } 845 ipc.addr = inet->inet_saddr; 846 847 ipc.oif = sk->sk_bound_dev_if; 848 err = sock_tx_timestamp(sk, &ipc.tx_flags); 849 if (err) 850 return err; 851 if (msg->msg_controllen) { 852 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 853 if (err) 854 return err; 855 if (ipc.opt) 856 free = 1; 857 connected = 0; 858 } 859 if (!ipc.opt) 860 ipc.opt = inet->opt; 861 862 saddr = ipc.addr; 863 ipc.addr = faddr = daddr; 864 865 if (ipc.opt && ipc.opt->srr) { 866 if (!daddr) 867 return -EINVAL; 868 faddr = ipc.opt->faddr; 869 connected = 0; 870 } 871 tos = RT_TOS(inet->tos); 872 if (sock_flag(sk, SOCK_LOCALROUTE) || 873 (msg->msg_flags & MSG_DONTROUTE) || 874 (ipc.opt && ipc.opt->is_strictroute)) { 875 tos |= RTO_ONLINK; 876 connected = 0; 877 } 878 879 if (ipv4_is_multicast(daddr)) { 880 if (!ipc.oif) 881 ipc.oif = inet->mc_index; 882 if (!saddr) 883 saddr = inet->mc_addr; 884 connected = 0; 885 } 886 887 if (connected) 888 rt = (struct rtable *)sk_dst_check(sk, 0); 889 890 if (rt == NULL) { 891 struct flowi fl = { .oif = ipc.oif, 892 .mark = sk->sk_mark, 893 .nl_u = { .ip4_u = 894 { .daddr = faddr, 895 .saddr = saddr, 896 .tos = tos } }, 897 .proto = sk->sk_protocol, 898 .flags = inet_sk_flowi_flags(sk), 899 .uli_u = { .ports = 900 { .sport = inet->inet_sport, 901 .dport = dport } } }; 902 struct net *net = sock_net(sk); 903 904 security_sk_classify_flow(sk, &fl); 905 err = ip_route_output_flow(net, &rt, &fl, sk, 1); 906 if (err) { 907 if (err == -ENETUNREACH) 908 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 909 goto out; 910 } 911 912 err = -EACCES; 913 if ((rt->rt_flags & RTCF_BROADCAST) && 914 !sock_flag(sk, SOCK_BROADCAST)) 915 goto out; 916 if (connected) 917 sk_dst_set(sk, dst_clone(&rt->dst)); 918 } 919 920 if (msg->msg_flags&MSG_CONFIRM) 921 goto do_confirm; 922 back_from_confirm: 923 924 saddr = rt->rt_src; 925 if (!ipc.addr) 926 daddr = ipc.addr = rt->rt_dst; 927 928 lock_sock(sk); 929 if (unlikely(up->pending)) { 930 /* The socket is already corked while preparing it. */ 931 /* ... which is an evident application bug. --ANK */ 932 release_sock(sk); 933 934 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 935 err = -EINVAL; 936 goto out; 937 } 938 /* 939 * Now cork the socket to pend data. 940 */ 941 inet->cork.fl.fl4_dst = daddr; 942 inet->cork.fl.fl_ip_dport = dport; 943 inet->cork.fl.fl4_src = saddr; 944 inet->cork.fl.fl_ip_sport = inet->inet_sport; 945 up->pending = AF_INET; 946 947 do_append_data: 948 up->len += ulen; 949 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 950 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 951 sizeof(struct udphdr), &ipc, &rt, 952 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 953 if (err) 954 udp_flush_pending_frames(sk); 955 else if (!corkreq) 956 err = udp_push_pending_frames(sk); 957 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 958 up->pending = 0; 959 release_sock(sk); 960 961 out: 962 ip_rt_put(rt); 963 if (free) 964 kfree(ipc.opt); 965 if (!err) 966 return len; 967 /* 968 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 969 * ENOBUFS might not be good (it's not tunable per se), but otherwise 970 * we don't have a good statistic (IpOutDiscards but it can be too many 971 * things). We could add another new stat but at least for now that 972 * seems like overkill. 973 */ 974 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 975 UDP_INC_STATS_USER(sock_net(sk), 976 UDP_MIB_SNDBUFERRORS, is_udplite); 977 } 978 return err; 979 980 do_confirm: 981 dst_confirm(&rt->dst); 982 if (!(msg->msg_flags&MSG_PROBE) || len) 983 goto back_from_confirm; 984 err = 0; 985 goto out; 986 } 987 EXPORT_SYMBOL(udp_sendmsg); 988 989 int udp_sendpage(struct sock *sk, struct page *page, int offset, 990 size_t size, int flags) 991 { 992 struct udp_sock *up = udp_sk(sk); 993 int ret; 994 995 if (!up->pending) { 996 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 997 998 /* Call udp_sendmsg to specify destination address which 999 * sendpage interface can't pass. 1000 * This will succeed only when the socket is connected. 1001 */ 1002 ret = udp_sendmsg(NULL, sk, &msg, 0); 1003 if (ret < 0) 1004 return ret; 1005 } 1006 1007 lock_sock(sk); 1008 1009 if (unlikely(!up->pending)) { 1010 release_sock(sk); 1011 1012 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 1013 return -EINVAL; 1014 } 1015 1016 ret = ip_append_page(sk, page, offset, size, flags); 1017 if (ret == -EOPNOTSUPP) { 1018 release_sock(sk); 1019 return sock_no_sendpage(sk->sk_socket, page, offset, 1020 size, flags); 1021 } 1022 if (ret < 0) { 1023 udp_flush_pending_frames(sk); 1024 goto out; 1025 } 1026 1027 up->len += size; 1028 if (!(up->corkflag || (flags&MSG_MORE))) 1029 ret = udp_push_pending_frames(sk); 1030 if (!ret) 1031 ret = size; 1032 out: 1033 release_sock(sk); 1034 return ret; 1035 } 1036 1037 1038 /** 1039 * first_packet_length - return length of first packet in receive queue 1040 * @sk: socket 1041 * 1042 * Drops all bad checksum frames, until a valid one is found. 1043 * Returns the length of found skb, or 0 if none is found. 1044 */ 1045 static unsigned int first_packet_length(struct sock *sk) 1046 { 1047 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1048 struct sk_buff *skb; 1049 unsigned int res; 1050 1051 __skb_queue_head_init(&list_kill); 1052 1053 spin_lock_bh(&rcvq->lock); 1054 while ((skb = skb_peek(rcvq)) != NULL && 1055 udp_lib_checksum_complete(skb)) { 1056 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1057 IS_UDPLITE(sk)); 1058 atomic_inc(&sk->sk_drops); 1059 __skb_unlink(skb, rcvq); 1060 __skb_queue_tail(&list_kill, skb); 1061 } 1062 res = skb ? skb->len : 0; 1063 spin_unlock_bh(&rcvq->lock); 1064 1065 if (!skb_queue_empty(&list_kill)) { 1066 bool slow = lock_sock_fast(sk); 1067 1068 __skb_queue_purge(&list_kill); 1069 sk_mem_reclaim_partial(sk); 1070 unlock_sock_fast(sk, slow); 1071 } 1072 return res; 1073 } 1074 1075 /* 1076 * IOCTL requests applicable to the UDP protocol 1077 */ 1078 1079 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1080 { 1081 switch (cmd) { 1082 case SIOCOUTQ: 1083 { 1084 int amount = sk_wmem_alloc_get(sk); 1085 1086 return put_user(amount, (int __user *)arg); 1087 } 1088 1089 case SIOCINQ: 1090 { 1091 unsigned int amount = first_packet_length(sk); 1092 1093 if (amount) 1094 /* 1095 * We will only return the amount 1096 * of this packet since that is all 1097 * that will be read. 1098 */ 1099 amount -= sizeof(struct udphdr); 1100 1101 return put_user(amount, (int __user *)arg); 1102 } 1103 1104 default: 1105 return -ENOIOCTLCMD; 1106 } 1107 1108 return 0; 1109 } 1110 EXPORT_SYMBOL(udp_ioctl); 1111 1112 /* 1113 * This should be easy, if there is something there we 1114 * return it, otherwise we block. 1115 */ 1116 1117 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1118 size_t len, int noblock, int flags, int *addr_len) 1119 { 1120 struct inet_sock *inet = inet_sk(sk); 1121 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 1122 struct sk_buff *skb; 1123 unsigned int ulen; 1124 int peeked; 1125 int err; 1126 int is_udplite = IS_UDPLITE(sk); 1127 bool slow; 1128 1129 /* 1130 * Check any passed addresses 1131 */ 1132 if (addr_len) 1133 *addr_len = sizeof(*sin); 1134 1135 if (flags & MSG_ERRQUEUE) 1136 return ip_recv_error(sk, msg, len); 1137 1138 try_again: 1139 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1140 &peeked, &err); 1141 if (!skb) 1142 goto out; 1143 1144 ulen = skb->len - sizeof(struct udphdr); 1145 if (len > ulen) 1146 len = ulen; 1147 else if (len < ulen) 1148 msg->msg_flags |= MSG_TRUNC; 1149 1150 /* 1151 * If checksum is needed at all, try to do it while copying the 1152 * data. If the data is truncated, or if we only want a partial 1153 * coverage checksum (UDP-Lite), do it before the copy. 1154 */ 1155 1156 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) { 1157 if (udp_lib_checksum_complete(skb)) 1158 goto csum_copy_err; 1159 } 1160 1161 if (skb_csum_unnecessary(skb)) 1162 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 1163 msg->msg_iov, len); 1164 else { 1165 err = skb_copy_and_csum_datagram_iovec(skb, 1166 sizeof(struct udphdr), 1167 msg->msg_iov); 1168 1169 if (err == -EINVAL) 1170 goto csum_copy_err; 1171 } 1172 1173 if (err) 1174 goto out_free; 1175 1176 if (!peeked) 1177 UDP_INC_STATS_USER(sock_net(sk), 1178 UDP_MIB_INDATAGRAMS, is_udplite); 1179 1180 sock_recv_ts_and_drops(msg, sk, skb); 1181 1182 /* Copy the address. */ 1183 if (sin) { 1184 sin->sin_family = AF_INET; 1185 sin->sin_port = udp_hdr(skb)->source; 1186 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1187 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1188 } 1189 if (inet->cmsg_flags) 1190 ip_cmsg_recv(msg, skb); 1191 1192 err = len; 1193 if (flags & MSG_TRUNC) 1194 err = ulen; 1195 1196 out_free: 1197 skb_free_datagram_locked(sk, skb); 1198 out: 1199 return err; 1200 1201 csum_copy_err: 1202 slow = lock_sock_fast(sk); 1203 if (!skb_kill_datagram(sk, skb, flags)) 1204 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1205 unlock_sock_fast(sk, slow); 1206 1207 if (noblock) 1208 return -EAGAIN; 1209 goto try_again; 1210 } 1211 1212 1213 int udp_disconnect(struct sock *sk, int flags) 1214 { 1215 struct inet_sock *inet = inet_sk(sk); 1216 /* 1217 * 1003.1g - break association. 1218 */ 1219 1220 sk->sk_state = TCP_CLOSE; 1221 inet->inet_daddr = 0; 1222 inet->inet_dport = 0; 1223 sock_rps_save_rxhash(sk, 0); 1224 sk->sk_bound_dev_if = 0; 1225 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1226 inet_reset_saddr(sk); 1227 1228 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1229 sk->sk_prot->unhash(sk); 1230 inet->inet_sport = 0; 1231 } 1232 sk_dst_reset(sk); 1233 return 0; 1234 } 1235 EXPORT_SYMBOL(udp_disconnect); 1236 1237 void udp_lib_unhash(struct sock *sk) 1238 { 1239 if (sk_hashed(sk)) { 1240 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1241 struct udp_hslot *hslot, *hslot2; 1242 1243 hslot = udp_hashslot(udptable, sock_net(sk), 1244 udp_sk(sk)->udp_port_hash); 1245 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1246 1247 spin_lock_bh(&hslot->lock); 1248 if (sk_nulls_del_node_init_rcu(sk)) { 1249 hslot->count--; 1250 inet_sk(sk)->inet_num = 0; 1251 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1252 1253 spin_lock(&hslot2->lock); 1254 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1255 hslot2->count--; 1256 spin_unlock(&hslot2->lock); 1257 } 1258 spin_unlock_bh(&hslot->lock); 1259 } 1260 } 1261 EXPORT_SYMBOL(udp_lib_unhash); 1262 1263 /* 1264 * inet_rcv_saddr was changed, we must rehash secondary hash 1265 */ 1266 void udp_lib_rehash(struct sock *sk, u16 newhash) 1267 { 1268 if (sk_hashed(sk)) { 1269 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1270 struct udp_hslot *hslot, *hslot2, *nhslot2; 1271 1272 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1273 nhslot2 = udp_hashslot2(udptable, newhash); 1274 udp_sk(sk)->udp_portaddr_hash = newhash; 1275 if (hslot2 != nhslot2) { 1276 hslot = udp_hashslot(udptable, sock_net(sk), 1277 udp_sk(sk)->udp_port_hash); 1278 /* we must lock primary chain too */ 1279 spin_lock_bh(&hslot->lock); 1280 1281 spin_lock(&hslot2->lock); 1282 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1283 hslot2->count--; 1284 spin_unlock(&hslot2->lock); 1285 1286 spin_lock(&nhslot2->lock); 1287 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1288 &nhslot2->head); 1289 nhslot2->count++; 1290 spin_unlock(&nhslot2->lock); 1291 1292 spin_unlock_bh(&hslot->lock); 1293 } 1294 } 1295 } 1296 EXPORT_SYMBOL(udp_lib_rehash); 1297 1298 static void udp_v4_rehash(struct sock *sk) 1299 { 1300 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1301 inet_sk(sk)->inet_rcv_saddr, 1302 inet_sk(sk)->inet_num); 1303 udp_lib_rehash(sk, new_hash); 1304 } 1305 1306 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1307 { 1308 int rc; 1309 1310 if (inet_sk(sk)->inet_daddr) 1311 sock_rps_save_rxhash(sk, skb->rxhash); 1312 1313 rc = ip_queue_rcv_skb(sk, skb); 1314 if (rc < 0) { 1315 int is_udplite = IS_UDPLITE(sk); 1316 1317 /* Note that an ENOMEM error is charged twice */ 1318 if (rc == -ENOMEM) 1319 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1320 is_udplite); 1321 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1322 kfree_skb(skb); 1323 return -1; 1324 } 1325 1326 return 0; 1327 1328 } 1329 1330 /* returns: 1331 * -1: error 1332 * 0: success 1333 * >0: "udp encap" protocol resubmission 1334 * 1335 * Note that in the success and error cases, the skb is assumed to 1336 * have either been requeued or freed. 1337 */ 1338 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1339 { 1340 struct udp_sock *up = udp_sk(sk); 1341 int rc; 1342 int is_udplite = IS_UDPLITE(sk); 1343 1344 /* 1345 * Charge it to the socket, dropping if the queue is full. 1346 */ 1347 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1348 goto drop; 1349 nf_reset(skb); 1350 1351 if (up->encap_type) { 1352 /* 1353 * This is an encapsulation socket so pass the skb to 1354 * the socket's udp_encap_rcv() hook. Otherwise, just 1355 * fall through and pass this up the UDP socket. 1356 * up->encap_rcv() returns the following value: 1357 * =0 if skb was successfully passed to the encap 1358 * handler or was discarded by it. 1359 * >0 if skb should be passed on to UDP. 1360 * <0 if skb should be resubmitted as proto -N 1361 */ 1362 1363 /* if we're overly short, let UDP handle it */ 1364 if (skb->len > sizeof(struct udphdr) && 1365 up->encap_rcv != NULL) { 1366 int ret; 1367 1368 ret = (*up->encap_rcv)(sk, skb); 1369 if (ret <= 0) { 1370 UDP_INC_STATS_BH(sock_net(sk), 1371 UDP_MIB_INDATAGRAMS, 1372 is_udplite); 1373 return -ret; 1374 } 1375 } 1376 1377 /* FALLTHROUGH -- it's a UDP Packet */ 1378 } 1379 1380 /* 1381 * UDP-Lite specific tests, ignored on UDP sockets 1382 */ 1383 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1384 1385 /* 1386 * MIB statistics other than incrementing the error count are 1387 * disabled for the following two types of errors: these depend 1388 * on the application settings, not on the functioning of the 1389 * protocol stack as such. 1390 * 1391 * RFC 3828 here recommends (sec 3.3): "There should also be a 1392 * way ... to ... at least let the receiving application block 1393 * delivery of packets with coverage values less than a value 1394 * provided by the application." 1395 */ 1396 if (up->pcrlen == 0) { /* full coverage was set */ 1397 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1398 "%d while full coverage %d requested\n", 1399 UDP_SKB_CB(skb)->cscov, skb->len); 1400 goto drop; 1401 } 1402 /* The next case involves violating the min. coverage requested 1403 * by the receiver. This is subtle: if receiver wants x and x is 1404 * greater than the buffersize/MTU then receiver will complain 1405 * that it wants x while sender emits packets of smaller size y. 1406 * Therefore the above ...()->partial_cov statement is essential. 1407 */ 1408 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1409 LIMIT_NETDEBUG(KERN_WARNING 1410 "UDPLITE: coverage %d too small, need min %d\n", 1411 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1412 goto drop; 1413 } 1414 } 1415 1416 if (sk->sk_filter) { 1417 if (udp_lib_checksum_complete(skb)) 1418 goto drop; 1419 } 1420 1421 1422 if (sk_rcvqueues_full(sk, skb)) 1423 goto drop; 1424 1425 rc = 0; 1426 1427 bh_lock_sock(sk); 1428 if (!sock_owned_by_user(sk)) 1429 rc = __udp_queue_rcv_skb(sk, skb); 1430 else if (sk_add_backlog(sk, skb)) { 1431 bh_unlock_sock(sk); 1432 goto drop; 1433 } 1434 bh_unlock_sock(sk); 1435 1436 return rc; 1437 1438 drop: 1439 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1440 atomic_inc(&sk->sk_drops); 1441 kfree_skb(skb); 1442 return -1; 1443 } 1444 1445 1446 static void flush_stack(struct sock **stack, unsigned int count, 1447 struct sk_buff *skb, unsigned int final) 1448 { 1449 unsigned int i; 1450 struct sk_buff *skb1 = NULL; 1451 struct sock *sk; 1452 1453 for (i = 0; i < count; i++) { 1454 sk = stack[i]; 1455 if (likely(skb1 == NULL)) 1456 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1457 1458 if (!skb1) { 1459 atomic_inc(&sk->sk_drops); 1460 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1461 IS_UDPLITE(sk)); 1462 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1463 IS_UDPLITE(sk)); 1464 } 1465 1466 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1467 skb1 = NULL; 1468 } 1469 if (unlikely(skb1)) 1470 kfree_skb(skb1); 1471 } 1472 1473 /* 1474 * Multicasts and broadcasts go to each listener. 1475 * 1476 * Note: called only from the BH handler context. 1477 */ 1478 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1479 struct udphdr *uh, 1480 __be32 saddr, __be32 daddr, 1481 struct udp_table *udptable) 1482 { 1483 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1484 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1485 int dif; 1486 unsigned int i, count = 0; 1487 1488 spin_lock(&hslot->lock); 1489 sk = sk_nulls_head(&hslot->head); 1490 dif = skb->dev->ifindex; 1491 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1492 while (sk) { 1493 stack[count++] = sk; 1494 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1495 daddr, uh->source, saddr, dif); 1496 if (unlikely(count == ARRAY_SIZE(stack))) { 1497 if (!sk) 1498 break; 1499 flush_stack(stack, count, skb, ~0); 1500 count = 0; 1501 } 1502 } 1503 /* 1504 * before releasing chain lock, we must take a reference on sockets 1505 */ 1506 for (i = 0; i < count; i++) 1507 sock_hold(stack[i]); 1508 1509 spin_unlock(&hslot->lock); 1510 1511 /* 1512 * do the slow work with no lock held 1513 */ 1514 if (count) { 1515 flush_stack(stack, count, skb, count - 1); 1516 1517 for (i = 0; i < count; i++) 1518 sock_put(stack[i]); 1519 } else { 1520 kfree_skb(skb); 1521 } 1522 return 0; 1523 } 1524 1525 /* Initialize UDP checksum. If exited with zero value (success), 1526 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1527 * Otherwise, csum completion requires chacksumming packet body, 1528 * including udp header and folding it to skb->csum. 1529 */ 1530 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1531 int proto) 1532 { 1533 const struct iphdr *iph; 1534 int err; 1535 1536 UDP_SKB_CB(skb)->partial_cov = 0; 1537 UDP_SKB_CB(skb)->cscov = skb->len; 1538 1539 if (proto == IPPROTO_UDPLITE) { 1540 err = udplite_checksum_init(skb, uh); 1541 if (err) 1542 return err; 1543 } 1544 1545 iph = ip_hdr(skb); 1546 if (uh->check == 0) { 1547 skb->ip_summed = CHECKSUM_UNNECESSARY; 1548 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1549 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1550 proto, skb->csum)) 1551 skb->ip_summed = CHECKSUM_UNNECESSARY; 1552 } 1553 if (!skb_csum_unnecessary(skb)) 1554 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1555 skb->len, proto, 0); 1556 /* Probably, we should checksum udp header (it should be in cache 1557 * in any case) and data in tiny packets (< rx copybreak). 1558 */ 1559 1560 return 0; 1561 } 1562 1563 /* 1564 * All we need to do is get the socket, and then do a checksum. 1565 */ 1566 1567 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1568 int proto) 1569 { 1570 struct sock *sk; 1571 struct udphdr *uh; 1572 unsigned short ulen; 1573 struct rtable *rt = skb_rtable(skb); 1574 __be32 saddr, daddr; 1575 struct net *net = dev_net(skb->dev); 1576 1577 /* 1578 * Validate the packet. 1579 */ 1580 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1581 goto drop; /* No space for header. */ 1582 1583 uh = udp_hdr(skb); 1584 ulen = ntohs(uh->len); 1585 saddr = ip_hdr(skb)->saddr; 1586 daddr = ip_hdr(skb)->daddr; 1587 1588 if (ulen > skb->len) 1589 goto short_packet; 1590 1591 if (proto == IPPROTO_UDP) { 1592 /* UDP validates ulen. */ 1593 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1594 goto short_packet; 1595 uh = udp_hdr(skb); 1596 } 1597 1598 if (udp4_csum_init(skb, uh, proto)) 1599 goto csum_error; 1600 1601 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1602 return __udp4_lib_mcast_deliver(net, skb, uh, 1603 saddr, daddr, udptable); 1604 1605 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1606 1607 if (sk != NULL) { 1608 int ret = udp_queue_rcv_skb(sk, skb); 1609 sock_put(sk); 1610 1611 /* a return value > 0 means to resubmit the input, but 1612 * it wants the return to be -protocol, or 0 1613 */ 1614 if (ret > 0) 1615 return -ret; 1616 return 0; 1617 } 1618 1619 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1620 goto drop; 1621 nf_reset(skb); 1622 1623 /* No socket. Drop packet silently, if checksum is wrong */ 1624 if (udp_lib_checksum_complete(skb)) 1625 goto csum_error; 1626 1627 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1628 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1629 1630 /* 1631 * Hmm. We got an UDP packet to a port to which we 1632 * don't wanna listen. Ignore it. 1633 */ 1634 kfree_skb(skb); 1635 return 0; 1636 1637 short_packet: 1638 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1639 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1640 &saddr, 1641 ntohs(uh->source), 1642 ulen, 1643 skb->len, 1644 &daddr, 1645 ntohs(uh->dest)); 1646 goto drop; 1647 1648 csum_error: 1649 /* 1650 * RFC1122: OK. Discards the bad packet silently (as far as 1651 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1652 */ 1653 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1654 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1655 &saddr, 1656 ntohs(uh->source), 1657 &daddr, 1658 ntohs(uh->dest), 1659 ulen); 1660 drop: 1661 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1662 kfree_skb(skb); 1663 return 0; 1664 } 1665 1666 int udp_rcv(struct sk_buff *skb) 1667 { 1668 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1669 } 1670 1671 void udp_destroy_sock(struct sock *sk) 1672 { 1673 bool slow = lock_sock_fast(sk); 1674 udp_flush_pending_frames(sk); 1675 unlock_sock_fast(sk, slow); 1676 } 1677 1678 /* 1679 * Socket option code for UDP 1680 */ 1681 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1682 char __user *optval, unsigned int optlen, 1683 int (*push_pending_frames)(struct sock *)) 1684 { 1685 struct udp_sock *up = udp_sk(sk); 1686 int val; 1687 int err = 0; 1688 int is_udplite = IS_UDPLITE(sk); 1689 1690 if (optlen < sizeof(int)) 1691 return -EINVAL; 1692 1693 if (get_user(val, (int __user *)optval)) 1694 return -EFAULT; 1695 1696 switch (optname) { 1697 case UDP_CORK: 1698 if (val != 0) { 1699 up->corkflag = 1; 1700 } else { 1701 up->corkflag = 0; 1702 lock_sock(sk); 1703 (*push_pending_frames)(sk); 1704 release_sock(sk); 1705 } 1706 break; 1707 1708 case UDP_ENCAP: 1709 switch (val) { 1710 case 0: 1711 case UDP_ENCAP_ESPINUDP: 1712 case UDP_ENCAP_ESPINUDP_NON_IKE: 1713 up->encap_rcv = xfrm4_udp_encap_rcv; 1714 /* FALLTHROUGH */ 1715 case UDP_ENCAP_L2TPINUDP: 1716 up->encap_type = val; 1717 break; 1718 default: 1719 err = -ENOPROTOOPT; 1720 break; 1721 } 1722 break; 1723 1724 /* 1725 * UDP-Lite's partial checksum coverage (RFC 3828). 1726 */ 1727 /* The sender sets actual checksum coverage length via this option. 1728 * The case coverage > packet length is handled by send module. */ 1729 case UDPLITE_SEND_CSCOV: 1730 if (!is_udplite) /* Disable the option on UDP sockets */ 1731 return -ENOPROTOOPT; 1732 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1733 val = 8; 1734 else if (val > USHRT_MAX) 1735 val = USHRT_MAX; 1736 up->pcslen = val; 1737 up->pcflag |= UDPLITE_SEND_CC; 1738 break; 1739 1740 /* The receiver specifies a minimum checksum coverage value. To make 1741 * sense, this should be set to at least 8 (as done below). If zero is 1742 * used, this again means full checksum coverage. */ 1743 case UDPLITE_RECV_CSCOV: 1744 if (!is_udplite) /* Disable the option on UDP sockets */ 1745 return -ENOPROTOOPT; 1746 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1747 val = 8; 1748 else if (val > USHRT_MAX) 1749 val = USHRT_MAX; 1750 up->pcrlen = val; 1751 up->pcflag |= UDPLITE_RECV_CC; 1752 break; 1753 1754 default: 1755 err = -ENOPROTOOPT; 1756 break; 1757 } 1758 1759 return err; 1760 } 1761 EXPORT_SYMBOL(udp_lib_setsockopt); 1762 1763 int udp_setsockopt(struct sock *sk, int level, int optname, 1764 char __user *optval, unsigned int optlen) 1765 { 1766 if (level == SOL_UDP || level == SOL_UDPLITE) 1767 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1768 udp_push_pending_frames); 1769 return ip_setsockopt(sk, level, optname, optval, optlen); 1770 } 1771 1772 #ifdef CONFIG_COMPAT 1773 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1774 char __user *optval, unsigned int optlen) 1775 { 1776 if (level == SOL_UDP || level == SOL_UDPLITE) 1777 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1778 udp_push_pending_frames); 1779 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1780 } 1781 #endif 1782 1783 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1784 char __user *optval, int __user *optlen) 1785 { 1786 struct udp_sock *up = udp_sk(sk); 1787 int val, len; 1788 1789 if (get_user(len, optlen)) 1790 return -EFAULT; 1791 1792 len = min_t(unsigned int, len, sizeof(int)); 1793 1794 if (len < 0) 1795 return -EINVAL; 1796 1797 switch (optname) { 1798 case UDP_CORK: 1799 val = up->corkflag; 1800 break; 1801 1802 case UDP_ENCAP: 1803 val = up->encap_type; 1804 break; 1805 1806 /* The following two cannot be changed on UDP sockets, the return is 1807 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1808 case UDPLITE_SEND_CSCOV: 1809 val = up->pcslen; 1810 break; 1811 1812 case UDPLITE_RECV_CSCOV: 1813 val = up->pcrlen; 1814 break; 1815 1816 default: 1817 return -ENOPROTOOPT; 1818 } 1819 1820 if (put_user(len, optlen)) 1821 return -EFAULT; 1822 if (copy_to_user(optval, &val, len)) 1823 return -EFAULT; 1824 return 0; 1825 } 1826 EXPORT_SYMBOL(udp_lib_getsockopt); 1827 1828 int udp_getsockopt(struct sock *sk, int level, int optname, 1829 char __user *optval, int __user *optlen) 1830 { 1831 if (level == SOL_UDP || level == SOL_UDPLITE) 1832 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1833 return ip_getsockopt(sk, level, optname, optval, optlen); 1834 } 1835 1836 #ifdef CONFIG_COMPAT 1837 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1838 char __user *optval, int __user *optlen) 1839 { 1840 if (level == SOL_UDP || level == SOL_UDPLITE) 1841 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1842 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1843 } 1844 #endif 1845 /** 1846 * udp_poll - wait for a UDP event. 1847 * @file - file struct 1848 * @sock - socket 1849 * @wait - poll table 1850 * 1851 * This is same as datagram poll, except for the special case of 1852 * blocking sockets. If application is using a blocking fd 1853 * and a packet with checksum error is in the queue; 1854 * then it could get return from select indicating data available 1855 * but then block when reading it. Add special case code 1856 * to work around these arguably broken applications. 1857 */ 1858 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1859 { 1860 unsigned int mask = datagram_poll(file, sock, wait); 1861 struct sock *sk = sock->sk; 1862 1863 /* Check for false positives due to checksum errors */ 1864 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 1865 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 1866 mask &= ~(POLLIN | POLLRDNORM); 1867 1868 return mask; 1869 1870 } 1871 EXPORT_SYMBOL(udp_poll); 1872 1873 struct proto udp_prot = { 1874 .name = "UDP", 1875 .owner = THIS_MODULE, 1876 .close = udp_lib_close, 1877 .connect = ip4_datagram_connect, 1878 .disconnect = udp_disconnect, 1879 .ioctl = udp_ioctl, 1880 .destroy = udp_destroy_sock, 1881 .setsockopt = udp_setsockopt, 1882 .getsockopt = udp_getsockopt, 1883 .sendmsg = udp_sendmsg, 1884 .recvmsg = udp_recvmsg, 1885 .sendpage = udp_sendpage, 1886 .backlog_rcv = __udp_queue_rcv_skb, 1887 .hash = udp_lib_hash, 1888 .unhash = udp_lib_unhash, 1889 .rehash = udp_v4_rehash, 1890 .get_port = udp_v4_get_port, 1891 .memory_allocated = &udp_memory_allocated, 1892 .sysctl_mem = sysctl_udp_mem, 1893 .sysctl_wmem = &sysctl_udp_wmem_min, 1894 .sysctl_rmem = &sysctl_udp_rmem_min, 1895 .obj_size = sizeof(struct udp_sock), 1896 .slab_flags = SLAB_DESTROY_BY_RCU, 1897 .h.udp_table = &udp_table, 1898 #ifdef CONFIG_COMPAT 1899 .compat_setsockopt = compat_udp_setsockopt, 1900 .compat_getsockopt = compat_udp_getsockopt, 1901 #endif 1902 }; 1903 EXPORT_SYMBOL(udp_prot); 1904 1905 /* ------------------------------------------------------------------------ */ 1906 #ifdef CONFIG_PROC_FS 1907 1908 static struct sock *udp_get_first(struct seq_file *seq, int start) 1909 { 1910 struct sock *sk; 1911 struct udp_iter_state *state = seq->private; 1912 struct net *net = seq_file_net(seq); 1913 1914 for (state->bucket = start; state->bucket <= state->udp_table->mask; 1915 ++state->bucket) { 1916 struct hlist_nulls_node *node; 1917 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 1918 1919 if (hlist_nulls_empty(&hslot->head)) 1920 continue; 1921 1922 spin_lock_bh(&hslot->lock); 1923 sk_nulls_for_each(sk, node, &hslot->head) { 1924 if (!net_eq(sock_net(sk), net)) 1925 continue; 1926 if (sk->sk_family == state->family) 1927 goto found; 1928 } 1929 spin_unlock_bh(&hslot->lock); 1930 } 1931 sk = NULL; 1932 found: 1933 return sk; 1934 } 1935 1936 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1937 { 1938 struct udp_iter_state *state = seq->private; 1939 struct net *net = seq_file_net(seq); 1940 1941 do { 1942 sk = sk_nulls_next(sk); 1943 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1944 1945 if (!sk) { 1946 if (state->bucket <= state->udp_table->mask) 1947 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1948 return udp_get_first(seq, state->bucket + 1); 1949 } 1950 return sk; 1951 } 1952 1953 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1954 { 1955 struct sock *sk = udp_get_first(seq, 0); 1956 1957 if (sk) 1958 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1959 --pos; 1960 return pos ? NULL : sk; 1961 } 1962 1963 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1964 { 1965 struct udp_iter_state *state = seq->private; 1966 state->bucket = MAX_UDP_PORTS; 1967 1968 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 1969 } 1970 1971 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1972 { 1973 struct sock *sk; 1974 1975 if (v == SEQ_START_TOKEN) 1976 sk = udp_get_idx(seq, 0); 1977 else 1978 sk = udp_get_next(seq, v); 1979 1980 ++*pos; 1981 return sk; 1982 } 1983 1984 static void udp_seq_stop(struct seq_file *seq, void *v) 1985 { 1986 struct udp_iter_state *state = seq->private; 1987 1988 if (state->bucket <= state->udp_table->mask) 1989 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1990 } 1991 1992 static int udp_seq_open(struct inode *inode, struct file *file) 1993 { 1994 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1995 struct udp_iter_state *s; 1996 int err; 1997 1998 err = seq_open_net(inode, file, &afinfo->seq_ops, 1999 sizeof(struct udp_iter_state)); 2000 if (err < 0) 2001 return err; 2002 2003 s = ((struct seq_file *)file->private_data)->private; 2004 s->family = afinfo->family; 2005 s->udp_table = afinfo->udp_table; 2006 return err; 2007 } 2008 2009 /* ------------------------------------------------------------------------ */ 2010 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2011 { 2012 struct proc_dir_entry *p; 2013 int rc = 0; 2014 2015 afinfo->seq_fops.open = udp_seq_open; 2016 afinfo->seq_fops.read = seq_read; 2017 afinfo->seq_fops.llseek = seq_lseek; 2018 afinfo->seq_fops.release = seq_release_net; 2019 2020 afinfo->seq_ops.start = udp_seq_start; 2021 afinfo->seq_ops.next = udp_seq_next; 2022 afinfo->seq_ops.stop = udp_seq_stop; 2023 2024 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2025 &afinfo->seq_fops, afinfo); 2026 if (!p) 2027 rc = -ENOMEM; 2028 return rc; 2029 } 2030 EXPORT_SYMBOL(udp_proc_register); 2031 2032 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2033 { 2034 proc_net_remove(net, afinfo->name); 2035 } 2036 EXPORT_SYMBOL(udp_proc_unregister); 2037 2038 /* ------------------------------------------------------------------------ */ 2039 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2040 int bucket, int *len) 2041 { 2042 struct inet_sock *inet = inet_sk(sp); 2043 __be32 dest = inet->inet_daddr; 2044 __be32 src = inet->inet_rcv_saddr; 2045 __u16 destp = ntohs(inet->inet_dport); 2046 __u16 srcp = ntohs(inet->inet_sport); 2047 2048 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2049 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 2050 bucket, src, srcp, dest, destp, sp->sk_state, 2051 sk_wmem_alloc_get(sp), 2052 sk_rmem_alloc_get(sp), 2053 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 2054 atomic_read(&sp->sk_refcnt), sp, 2055 atomic_read(&sp->sk_drops), len); 2056 } 2057 2058 int udp4_seq_show(struct seq_file *seq, void *v) 2059 { 2060 if (v == SEQ_START_TOKEN) 2061 seq_printf(seq, "%-127s\n", 2062 " sl local_address rem_address st tx_queue " 2063 "rx_queue tr tm->when retrnsmt uid timeout " 2064 "inode ref pointer drops"); 2065 else { 2066 struct udp_iter_state *state = seq->private; 2067 int len; 2068 2069 udp4_format_sock(v, seq, state->bucket, &len); 2070 seq_printf(seq, "%*s\n", 127 - len, ""); 2071 } 2072 return 0; 2073 } 2074 2075 /* ------------------------------------------------------------------------ */ 2076 static struct udp_seq_afinfo udp4_seq_afinfo = { 2077 .name = "udp", 2078 .family = AF_INET, 2079 .udp_table = &udp_table, 2080 .seq_fops = { 2081 .owner = THIS_MODULE, 2082 }, 2083 .seq_ops = { 2084 .show = udp4_seq_show, 2085 }, 2086 }; 2087 2088 static int __net_init udp4_proc_init_net(struct net *net) 2089 { 2090 return udp_proc_register(net, &udp4_seq_afinfo); 2091 } 2092 2093 static void __net_exit udp4_proc_exit_net(struct net *net) 2094 { 2095 udp_proc_unregister(net, &udp4_seq_afinfo); 2096 } 2097 2098 static struct pernet_operations udp4_net_ops = { 2099 .init = udp4_proc_init_net, 2100 .exit = udp4_proc_exit_net, 2101 }; 2102 2103 int __init udp4_proc_init(void) 2104 { 2105 return register_pernet_subsys(&udp4_net_ops); 2106 } 2107 2108 void udp4_proc_exit(void) 2109 { 2110 unregister_pernet_subsys(&udp4_net_ops); 2111 } 2112 #endif /* CONFIG_PROC_FS */ 2113 2114 static __initdata unsigned long uhash_entries; 2115 static int __init set_uhash_entries(char *str) 2116 { 2117 if (!str) 2118 return 0; 2119 uhash_entries = simple_strtoul(str, &str, 0); 2120 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2121 uhash_entries = UDP_HTABLE_SIZE_MIN; 2122 return 1; 2123 } 2124 __setup("uhash_entries=", set_uhash_entries); 2125 2126 void __init udp_table_init(struct udp_table *table, const char *name) 2127 { 2128 unsigned int i; 2129 2130 if (!CONFIG_BASE_SMALL) 2131 table->hash = alloc_large_system_hash(name, 2132 2 * sizeof(struct udp_hslot), 2133 uhash_entries, 2134 21, /* one slot per 2 MB */ 2135 0, 2136 &table->log, 2137 &table->mask, 2138 64 * 1024); 2139 /* 2140 * Make sure hash table has the minimum size 2141 */ 2142 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) { 2143 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN * 2144 2 * sizeof(struct udp_hslot), GFP_KERNEL); 2145 if (!table->hash) 2146 panic(name); 2147 table->log = ilog2(UDP_HTABLE_SIZE_MIN); 2148 table->mask = UDP_HTABLE_SIZE_MIN - 1; 2149 } 2150 table->hash2 = table->hash + (table->mask + 1); 2151 for (i = 0; i <= table->mask; i++) { 2152 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2153 table->hash[i].count = 0; 2154 spin_lock_init(&table->hash[i].lock); 2155 } 2156 for (i = 0; i <= table->mask; i++) { 2157 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2158 table->hash2[i].count = 0; 2159 spin_lock_init(&table->hash2[i].lock); 2160 } 2161 } 2162 2163 void __init udp_init(void) 2164 { 2165 unsigned long nr_pages, limit; 2166 2167 udp_table_init(&udp_table, "UDP"); 2168 /* Set the pressure threshold up by the same strategy of TCP. It is a 2169 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 2170 * toward zero with the amount of memory, with a floor of 128 pages. 2171 */ 2172 nr_pages = totalram_pages - totalhigh_pages; 2173 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 2174 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 2175 limit = max(limit, 128UL); 2176 sysctl_udp_mem[0] = limit / 4 * 3; 2177 sysctl_udp_mem[1] = limit; 2178 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2179 2180 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2181 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2182 } 2183 2184 int udp4_ufo_send_check(struct sk_buff *skb) 2185 { 2186 const struct iphdr *iph; 2187 struct udphdr *uh; 2188 2189 if (!pskb_may_pull(skb, sizeof(*uh))) 2190 return -EINVAL; 2191 2192 iph = ip_hdr(skb); 2193 uh = udp_hdr(skb); 2194 2195 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 2196 IPPROTO_UDP, 0); 2197 skb->csum_start = skb_transport_header(skb) - skb->head; 2198 skb->csum_offset = offsetof(struct udphdr, check); 2199 skb->ip_summed = CHECKSUM_PARTIAL; 2200 return 0; 2201 } 2202 2203 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features) 2204 { 2205 struct sk_buff *segs = ERR_PTR(-EINVAL); 2206 unsigned int mss; 2207 int offset; 2208 __wsum csum; 2209 2210 mss = skb_shinfo(skb)->gso_size; 2211 if (unlikely(skb->len <= mss)) 2212 goto out; 2213 2214 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 2215 /* Packet is from an untrusted source, reset gso_segs. */ 2216 int type = skb_shinfo(skb)->gso_type; 2217 2218 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || 2219 !(type & (SKB_GSO_UDP)))) 2220 goto out; 2221 2222 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 2223 2224 segs = NULL; 2225 goto out; 2226 } 2227 2228 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot 2229 * do checksum of UDP packets sent as multiple IP fragments. 2230 */ 2231 offset = skb->csum_start - skb_headroom(skb); 2232 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2233 offset += skb->csum_offset; 2234 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2235 skb->ip_summed = CHECKSUM_NONE; 2236 2237 /* Fragment the skb. IP headers of the fragments are updated in 2238 * inet_gso_segment() 2239 */ 2240 segs = skb_segment(skb, features); 2241 out: 2242 return segs; 2243 } 2244 2245