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 long 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_long_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_hint(&result->sk_refcnt, 2))) 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_hint(&result->sk_refcnt, 2))) 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 - handle outgoing HW checksumming 667 * @skb: sk_buff containing the filled-in UDP header 668 * (checksum field must be zeroed out) 669 * @src: source IP address 670 * @dst: destination IP address 671 */ 672 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 673 { 674 struct udphdr *uh = udp_hdr(skb); 675 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 676 int offset = skb_transport_offset(skb); 677 int len = skb->len - offset; 678 int hlen = len; 679 __wsum csum = 0; 680 681 if (!frags) { 682 /* 683 * Only one fragment on the socket. 684 */ 685 skb->csum_start = skb_transport_header(skb) - skb->head; 686 skb->csum_offset = offsetof(struct udphdr, check); 687 uh->check = ~csum_tcpudp_magic(src, dst, len, 688 IPPROTO_UDP, 0); 689 } else { 690 /* 691 * HW-checksum won't work as there are two or more 692 * fragments on the socket so that all csums of sk_buffs 693 * should be together 694 */ 695 do { 696 csum = csum_add(csum, frags->csum); 697 hlen -= frags->len; 698 } while ((frags = frags->next)); 699 700 csum = skb_checksum(skb, offset, hlen, csum); 701 skb->ip_summed = CHECKSUM_NONE; 702 703 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 704 if (uh->check == 0) 705 uh->check = CSUM_MANGLED_0; 706 } 707 } 708 709 static int udp_send_skb(struct sk_buff *skb, __be32 daddr, __be32 dport) 710 { 711 struct sock *sk = skb->sk; 712 struct inet_sock *inet = inet_sk(sk); 713 struct udphdr *uh; 714 struct rtable *rt = (struct rtable *)skb_dst(skb); 715 int err = 0; 716 int is_udplite = IS_UDPLITE(sk); 717 int offset = skb_transport_offset(skb); 718 int len = skb->len - offset; 719 __wsum csum = 0; 720 721 /* 722 * Create a UDP header 723 */ 724 uh = udp_hdr(skb); 725 uh->source = inet->inet_sport; 726 uh->dest = dport; 727 uh->len = htons(len); 728 uh->check = 0; 729 730 if (is_udplite) /* UDP-Lite */ 731 csum = udplite_csum(skb); 732 733 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 734 735 skb->ip_summed = CHECKSUM_NONE; 736 goto send; 737 738 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 739 740 udp4_hwcsum(skb, rt->rt_src, daddr); 741 goto send; 742 743 } else 744 csum = udp_csum(skb); 745 746 /* add protocol-dependent pseudo-header */ 747 uh->check = csum_tcpudp_magic(rt->rt_src, daddr, len, 748 sk->sk_protocol, csum); 749 if (uh->check == 0) 750 uh->check = CSUM_MANGLED_0; 751 752 send: 753 err = ip_send_skb(skb); 754 if (err) { 755 if (err == -ENOBUFS && !inet->recverr) { 756 UDP_INC_STATS_USER(sock_net(sk), 757 UDP_MIB_SNDBUFERRORS, is_udplite); 758 err = 0; 759 } 760 } else 761 UDP_INC_STATS_USER(sock_net(sk), 762 UDP_MIB_OUTDATAGRAMS, is_udplite); 763 return err; 764 } 765 766 /* 767 * Push out all pending data as one UDP datagram. Socket is locked. 768 */ 769 static int udp_push_pending_frames(struct sock *sk) 770 { 771 struct udp_sock *up = udp_sk(sk); 772 struct inet_sock *inet = inet_sk(sk); 773 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 774 struct sk_buff *skb; 775 int err = 0; 776 777 skb = ip_finish_skb(sk); 778 if (!skb) 779 goto out; 780 781 err = udp_send_skb(skb, fl4->daddr, fl4->fl4_dport); 782 783 out: 784 up->len = 0; 785 up->pending = 0; 786 return err; 787 } 788 789 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 790 size_t len) 791 { 792 struct inet_sock *inet = inet_sk(sk); 793 struct udp_sock *up = udp_sk(sk); 794 struct flowi4 *fl4; 795 int ulen = len; 796 struct ipcm_cookie ipc; 797 struct rtable *rt = NULL; 798 int free = 0; 799 int connected = 0; 800 __be32 daddr, faddr, saddr; 801 __be16 dport; 802 u8 tos; 803 int err, is_udplite = IS_UDPLITE(sk); 804 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 805 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 806 struct sk_buff *skb; 807 808 if (len > 0xFFFF) 809 return -EMSGSIZE; 810 811 /* 812 * Check the flags. 813 */ 814 815 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 816 return -EOPNOTSUPP; 817 818 ipc.opt = NULL; 819 ipc.tx_flags = 0; 820 821 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 822 823 if (up->pending) { 824 /* 825 * There are pending frames. 826 * The socket lock must be held while it's corked. 827 */ 828 lock_sock(sk); 829 if (likely(up->pending)) { 830 if (unlikely(up->pending != AF_INET)) { 831 release_sock(sk); 832 return -EINVAL; 833 } 834 goto do_append_data; 835 } 836 release_sock(sk); 837 } 838 ulen += sizeof(struct udphdr); 839 840 /* 841 * Get and verify the address. 842 */ 843 if (msg->msg_name) { 844 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name; 845 if (msg->msg_namelen < sizeof(*usin)) 846 return -EINVAL; 847 if (usin->sin_family != AF_INET) { 848 if (usin->sin_family != AF_UNSPEC) 849 return -EAFNOSUPPORT; 850 } 851 852 daddr = usin->sin_addr.s_addr; 853 dport = usin->sin_port; 854 if (dport == 0) 855 return -EINVAL; 856 } else { 857 if (sk->sk_state != TCP_ESTABLISHED) 858 return -EDESTADDRREQ; 859 daddr = inet->inet_daddr; 860 dport = inet->inet_dport; 861 /* Open fast path for connected socket. 862 Route will not be used, if at least one option is set. 863 */ 864 connected = 1; 865 } 866 ipc.addr = inet->inet_saddr; 867 868 ipc.oif = sk->sk_bound_dev_if; 869 err = sock_tx_timestamp(sk, &ipc.tx_flags); 870 if (err) 871 return err; 872 if (msg->msg_controllen) { 873 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 874 if (err) 875 return err; 876 if (ipc.opt) 877 free = 1; 878 connected = 0; 879 } 880 if (!ipc.opt) 881 ipc.opt = inet->opt; 882 883 saddr = ipc.addr; 884 ipc.addr = faddr = daddr; 885 886 if (ipc.opt && ipc.opt->srr) { 887 if (!daddr) 888 return -EINVAL; 889 faddr = ipc.opt->faddr; 890 connected = 0; 891 } 892 tos = RT_TOS(inet->tos); 893 if (sock_flag(sk, SOCK_LOCALROUTE) || 894 (msg->msg_flags & MSG_DONTROUTE) || 895 (ipc.opt && ipc.opt->is_strictroute)) { 896 tos |= RTO_ONLINK; 897 connected = 0; 898 } 899 900 if (ipv4_is_multicast(daddr)) { 901 if (!ipc.oif) 902 ipc.oif = inet->mc_index; 903 if (!saddr) 904 saddr = inet->mc_addr; 905 connected = 0; 906 } 907 908 if (connected) 909 rt = (struct rtable *)sk_dst_check(sk, 0); 910 911 if (rt == NULL) { 912 struct flowi4 fl4 = { 913 .flowi4_oif = ipc.oif, 914 .flowi4_mark = sk->sk_mark, 915 .daddr = faddr, 916 .saddr = saddr, 917 .flowi4_tos = tos, 918 .flowi4_proto = sk->sk_protocol, 919 .flowi4_flags = (inet_sk_flowi_flags(sk) | 920 FLOWI_FLAG_CAN_SLEEP), 921 .fl4_sport = inet->inet_sport, 922 .fl4_dport = dport, 923 }; 924 struct net *net = sock_net(sk); 925 926 security_sk_classify_flow(sk, flowi4_to_flowi(&fl4)); 927 rt = ip_route_output_flow(net, &fl4, sk); 928 if (IS_ERR(rt)) { 929 err = PTR_ERR(rt); 930 rt = NULL; 931 if (err == -ENETUNREACH) 932 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 933 goto out; 934 } 935 936 err = -EACCES; 937 if ((rt->rt_flags & RTCF_BROADCAST) && 938 !sock_flag(sk, SOCK_BROADCAST)) 939 goto out; 940 if (connected) 941 sk_dst_set(sk, dst_clone(&rt->dst)); 942 } 943 944 if (msg->msg_flags&MSG_CONFIRM) 945 goto do_confirm; 946 back_from_confirm: 947 948 saddr = rt->rt_src; 949 if (!ipc.addr) 950 daddr = ipc.addr = rt->rt_dst; 951 952 /* Lockless fast path for the non-corking case. */ 953 if (!corkreq) { 954 skb = ip_make_skb(sk, getfrag, msg->msg_iov, ulen, 955 sizeof(struct udphdr), &ipc, &rt, 956 msg->msg_flags); 957 err = PTR_ERR(skb); 958 if (skb && !IS_ERR(skb)) 959 err = udp_send_skb(skb, daddr, dport); 960 goto out; 961 } 962 963 lock_sock(sk); 964 if (unlikely(up->pending)) { 965 /* The socket is already corked while preparing it. */ 966 /* ... which is an evident application bug. --ANK */ 967 release_sock(sk); 968 969 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 970 err = -EINVAL; 971 goto out; 972 } 973 /* 974 * Now cork the socket to pend data. 975 */ 976 fl4 = &inet->cork.fl.u.ip4; 977 fl4->daddr = daddr; 978 fl4->saddr = saddr; 979 fl4->fl4_dport = dport; 980 fl4->fl4_sport = inet->inet_sport; 981 up->pending = AF_INET; 982 983 do_append_data: 984 up->len += ulen; 985 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 986 sizeof(struct udphdr), &ipc, &rt, 987 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 988 if (err) 989 udp_flush_pending_frames(sk); 990 else if (!corkreq) 991 err = udp_push_pending_frames(sk); 992 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 993 up->pending = 0; 994 release_sock(sk); 995 996 out: 997 ip_rt_put(rt); 998 if (free) 999 kfree(ipc.opt); 1000 if (!err) 1001 return len; 1002 /* 1003 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1004 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1005 * we don't have a good statistic (IpOutDiscards but it can be too many 1006 * things). We could add another new stat but at least for now that 1007 * seems like overkill. 1008 */ 1009 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1010 UDP_INC_STATS_USER(sock_net(sk), 1011 UDP_MIB_SNDBUFERRORS, is_udplite); 1012 } 1013 return err; 1014 1015 do_confirm: 1016 dst_confirm(&rt->dst); 1017 if (!(msg->msg_flags&MSG_PROBE) || len) 1018 goto back_from_confirm; 1019 err = 0; 1020 goto out; 1021 } 1022 EXPORT_SYMBOL(udp_sendmsg); 1023 1024 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1025 size_t size, int flags) 1026 { 1027 struct udp_sock *up = udp_sk(sk); 1028 int ret; 1029 1030 if (!up->pending) { 1031 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1032 1033 /* Call udp_sendmsg to specify destination address which 1034 * sendpage interface can't pass. 1035 * This will succeed only when the socket is connected. 1036 */ 1037 ret = udp_sendmsg(NULL, sk, &msg, 0); 1038 if (ret < 0) 1039 return ret; 1040 } 1041 1042 lock_sock(sk); 1043 1044 if (unlikely(!up->pending)) { 1045 release_sock(sk); 1046 1047 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 1048 return -EINVAL; 1049 } 1050 1051 ret = ip_append_page(sk, page, offset, size, flags); 1052 if (ret == -EOPNOTSUPP) { 1053 release_sock(sk); 1054 return sock_no_sendpage(sk->sk_socket, page, offset, 1055 size, flags); 1056 } 1057 if (ret < 0) { 1058 udp_flush_pending_frames(sk); 1059 goto out; 1060 } 1061 1062 up->len += size; 1063 if (!(up->corkflag || (flags&MSG_MORE))) 1064 ret = udp_push_pending_frames(sk); 1065 if (!ret) 1066 ret = size; 1067 out: 1068 release_sock(sk); 1069 return ret; 1070 } 1071 1072 1073 /** 1074 * first_packet_length - return length of first packet in receive queue 1075 * @sk: socket 1076 * 1077 * Drops all bad checksum frames, until a valid one is found. 1078 * Returns the length of found skb, or 0 if none is found. 1079 */ 1080 static unsigned int first_packet_length(struct sock *sk) 1081 { 1082 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1083 struct sk_buff *skb; 1084 unsigned int res; 1085 1086 __skb_queue_head_init(&list_kill); 1087 1088 spin_lock_bh(&rcvq->lock); 1089 while ((skb = skb_peek(rcvq)) != NULL && 1090 udp_lib_checksum_complete(skb)) { 1091 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1092 IS_UDPLITE(sk)); 1093 atomic_inc(&sk->sk_drops); 1094 __skb_unlink(skb, rcvq); 1095 __skb_queue_tail(&list_kill, skb); 1096 } 1097 res = skb ? skb->len : 0; 1098 spin_unlock_bh(&rcvq->lock); 1099 1100 if (!skb_queue_empty(&list_kill)) { 1101 bool slow = lock_sock_fast(sk); 1102 1103 __skb_queue_purge(&list_kill); 1104 sk_mem_reclaim_partial(sk); 1105 unlock_sock_fast(sk, slow); 1106 } 1107 return res; 1108 } 1109 1110 /* 1111 * IOCTL requests applicable to the UDP protocol 1112 */ 1113 1114 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1115 { 1116 switch (cmd) { 1117 case SIOCOUTQ: 1118 { 1119 int amount = sk_wmem_alloc_get(sk); 1120 1121 return put_user(amount, (int __user *)arg); 1122 } 1123 1124 case SIOCINQ: 1125 { 1126 unsigned int amount = first_packet_length(sk); 1127 1128 if (amount) 1129 /* 1130 * We will only return the amount 1131 * of this packet since that is all 1132 * that will be read. 1133 */ 1134 amount -= sizeof(struct udphdr); 1135 1136 return put_user(amount, (int __user *)arg); 1137 } 1138 1139 default: 1140 return -ENOIOCTLCMD; 1141 } 1142 1143 return 0; 1144 } 1145 EXPORT_SYMBOL(udp_ioctl); 1146 1147 /* 1148 * This should be easy, if there is something there we 1149 * return it, otherwise we block. 1150 */ 1151 1152 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1153 size_t len, int noblock, int flags, int *addr_len) 1154 { 1155 struct inet_sock *inet = inet_sk(sk); 1156 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 1157 struct sk_buff *skb; 1158 unsigned int ulen; 1159 int peeked; 1160 int err; 1161 int is_udplite = IS_UDPLITE(sk); 1162 bool slow; 1163 1164 /* 1165 * Check any passed addresses 1166 */ 1167 if (addr_len) 1168 *addr_len = sizeof(*sin); 1169 1170 if (flags & MSG_ERRQUEUE) 1171 return ip_recv_error(sk, msg, len); 1172 1173 try_again: 1174 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1175 &peeked, &err); 1176 if (!skb) 1177 goto out; 1178 1179 ulen = skb->len - sizeof(struct udphdr); 1180 if (len > ulen) 1181 len = ulen; 1182 else if (len < ulen) 1183 msg->msg_flags |= MSG_TRUNC; 1184 1185 /* 1186 * If checksum is needed at all, try to do it while copying the 1187 * data. If the data is truncated, or if we only want a partial 1188 * coverage checksum (UDP-Lite), do it before the copy. 1189 */ 1190 1191 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) { 1192 if (udp_lib_checksum_complete(skb)) 1193 goto csum_copy_err; 1194 } 1195 1196 if (skb_csum_unnecessary(skb)) 1197 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 1198 msg->msg_iov, len); 1199 else { 1200 err = skb_copy_and_csum_datagram_iovec(skb, 1201 sizeof(struct udphdr), 1202 msg->msg_iov); 1203 1204 if (err == -EINVAL) 1205 goto csum_copy_err; 1206 } 1207 1208 if (err) 1209 goto out_free; 1210 1211 if (!peeked) 1212 UDP_INC_STATS_USER(sock_net(sk), 1213 UDP_MIB_INDATAGRAMS, is_udplite); 1214 1215 sock_recv_ts_and_drops(msg, sk, skb); 1216 1217 /* Copy the address. */ 1218 if (sin) { 1219 sin->sin_family = AF_INET; 1220 sin->sin_port = udp_hdr(skb)->source; 1221 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1222 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1223 } 1224 if (inet->cmsg_flags) 1225 ip_cmsg_recv(msg, skb); 1226 1227 err = len; 1228 if (flags & MSG_TRUNC) 1229 err = ulen; 1230 1231 out_free: 1232 skb_free_datagram_locked(sk, skb); 1233 out: 1234 return err; 1235 1236 csum_copy_err: 1237 slow = lock_sock_fast(sk); 1238 if (!skb_kill_datagram(sk, skb, flags)) 1239 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1240 unlock_sock_fast(sk, slow); 1241 1242 if (noblock) 1243 return -EAGAIN; 1244 goto try_again; 1245 } 1246 1247 1248 int udp_disconnect(struct sock *sk, int flags) 1249 { 1250 struct inet_sock *inet = inet_sk(sk); 1251 /* 1252 * 1003.1g - break association. 1253 */ 1254 1255 sk->sk_state = TCP_CLOSE; 1256 inet->inet_daddr = 0; 1257 inet->inet_dport = 0; 1258 sock_rps_save_rxhash(sk, 0); 1259 sk->sk_bound_dev_if = 0; 1260 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1261 inet_reset_saddr(sk); 1262 1263 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1264 sk->sk_prot->unhash(sk); 1265 inet->inet_sport = 0; 1266 } 1267 sk_dst_reset(sk); 1268 return 0; 1269 } 1270 EXPORT_SYMBOL(udp_disconnect); 1271 1272 void udp_lib_unhash(struct sock *sk) 1273 { 1274 if (sk_hashed(sk)) { 1275 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1276 struct udp_hslot *hslot, *hslot2; 1277 1278 hslot = udp_hashslot(udptable, sock_net(sk), 1279 udp_sk(sk)->udp_port_hash); 1280 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1281 1282 spin_lock_bh(&hslot->lock); 1283 if (sk_nulls_del_node_init_rcu(sk)) { 1284 hslot->count--; 1285 inet_sk(sk)->inet_num = 0; 1286 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1287 1288 spin_lock(&hslot2->lock); 1289 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1290 hslot2->count--; 1291 spin_unlock(&hslot2->lock); 1292 } 1293 spin_unlock_bh(&hslot->lock); 1294 } 1295 } 1296 EXPORT_SYMBOL(udp_lib_unhash); 1297 1298 /* 1299 * inet_rcv_saddr was changed, we must rehash secondary hash 1300 */ 1301 void udp_lib_rehash(struct sock *sk, u16 newhash) 1302 { 1303 if (sk_hashed(sk)) { 1304 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1305 struct udp_hslot *hslot, *hslot2, *nhslot2; 1306 1307 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1308 nhslot2 = udp_hashslot2(udptable, newhash); 1309 udp_sk(sk)->udp_portaddr_hash = newhash; 1310 if (hslot2 != nhslot2) { 1311 hslot = udp_hashslot(udptable, sock_net(sk), 1312 udp_sk(sk)->udp_port_hash); 1313 /* we must lock primary chain too */ 1314 spin_lock_bh(&hslot->lock); 1315 1316 spin_lock(&hslot2->lock); 1317 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1318 hslot2->count--; 1319 spin_unlock(&hslot2->lock); 1320 1321 spin_lock(&nhslot2->lock); 1322 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1323 &nhslot2->head); 1324 nhslot2->count++; 1325 spin_unlock(&nhslot2->lock); 1326 1327 spin_unlock_bh(&hslot->lock); 1328 } 1329 } 1330 } 1331 EXPORT_SYMBOL(udp_lib_rehash); 1332 1333 static void udp_v4_rehash(struct sock *sk) 1334 { 1335 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1336 inet_sk(sk)->inet_rcv_saddr, 1337 inet_sk(sk)->inet_num); 1338 udp_lib_rehash(sk, new_hash); 1339 } 1340 1341 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1342 { 1343 int rc; 1344 1345 if (inet_sk(sk)->inet_daddr) 1346 sock_rps_save_rxhash(sk, skb->rxhash); 1347 1348 rc = ip_queue_rcv_skb(sk, skb); 1349 if (rc < 0) { 1350 int is_udplite = IS_UDPLITE(sk); 1351 1352 /* Note that an ENOMEM error is charged twice */ 1353 if (rc == -ENOMEM) 1354 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1355 is_udplite); 1356 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1357 kfree_skb(skb); 1358 return -1; 1359 } 1360 1361 return 0; 1362 1363 } 1364 1365 /* returns: 1366 * -1: error 1367 * 0: success 1368 * >0: "udp encap" protocol resubmission 1369 * 1370 * Note that in the success and error cases, the skb is assumed to 1371 * have either been requeued or freed. 1372 */ 1373 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1374 { 1375 struct udp_sock *up = udp_sk(sk); 1376 int rc; 1377 int is_udplite = IS_UDPLITE(sk); 1378 1379 /* 1380 * Charge it to the socket, dropping if the queue is full. 1381 */ 1382 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1383 goto drop; 1384 nf_reset(skb); 1385 1386 if (up->encap_type) { 1387 /* 1388 * This is an encapsulation socket so pass the skb to 1389 * the socket's udp_encap_rcv() hook. Otherwise, just 1390 * fall through and pass this up the UDP socket. 1391 * up->encap_rcv() returns the following value: 1392 * =0 if skb was successfully passed to the encap 1393 * handler or was discarded by it. 1394 * >0 if skb should be passed on to UDP. 1395 * <0 if skb should be resubmitted as proto -N 1396 */ 1397 1398 /* if we're overly short, let UDP handle it */ 1399 if (skb->len > sizeof(struct udphdr) && 1400 up->encap_rcv != NULL) { 1401 int ret; 1402 1403 ret = (*up->encap_rcv)(sk, skb); 1404 if (ret <= 0) { 1405 UDP_INC_STATS_BH(sock_net(sk), 1406 UDP_MIB_INDATAGRAMS, 1407 is_udplite); 1408 return -ret; 1409 } 1410 } 1411 1412 /* FALLTHROUGH -- it's a UDP Packet */ 1413 } 1414 1415 /* 1416 * UDP-Lite specific tests, ignored on UDP sockets 1417 */ 1418 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1419 1420 /* 1421 * MIB statistics other than incrementing the error count are 1422 * disabled for the following two types of errors: these depend 1423 * on the application settings, not on the functioning of the 1424 * protocol stack as such. 1425 * 1426 * RFC 3828 here recommends (sec 3.3): "There should also be a 1427 * way ... to ... at least let the receiving application block 1428 * delivery of packets with coverage values less than a value 1429 * provided by the application." 1430 */ 1431 if (up->pcrlen == 0) { /* full coverage was set */ 1432 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1433 "%d while full coverage %d requested\n", 1434 UDP_SKB_CB(skb)->cscov, skb->len); 1435 goto drop; 1436 } 1437 /* The next case involves violating the min. coverage requested 1438 * by the receiver. This is subtle: if receiver wants x and x is 1439 * greater than the buffersize/MTU then receiver will complain 1440 * that it wants x while sender emits packets of smaller size y. 1441 * Therefore the above ...()->partial_cov statement is essential. 1442 */ 1443 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1444 LIMIT_NETDEBUG(KERN_WARNING 1445 "UDPLITE: coverage %d too small, need min %d\n", 1446 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1447 goto drop; 1448 } 1449 } 1450 1451 if (rcu_dereference_raw(sk->sk_filter)) { 1452 if (udp_lib_checksum_complete(skb)) 1453 goto drop; 1454 } 1455 1456 1457 if (sk_rcvqueues_full(sk, skb)) 1458 goto drop; 1459 1460 rc = 0; 1461 1462 bh_lock_sock(sk); 1463 if (!sock_owned_by_user(sk)) 1464 rc = __udp_queue_rcv_skb(sk, skb); 1465 else if (sk_add_backlog(sk, skb)) { 1466 bh_unlock_sock(sk); 1467 goto drop; 1468 } 1469 bh_unlock_sock(sk); 1470 1471 return rc; 1472 1473 drop: 1474 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1475 atomic_inc(&sk->sk_drops); 1476 kfree_skb(skb); 1477 return -1; 1478 } 1479 1480 1481 static void flush_stack(struct sock **stack, unsigned int count, 1482 struct sk_buff *skb, unsigned int final) 1483 { 1484 unsigned int i; 1485 struct sk_buff *skb1 = NULL; 1486 struct sock *sk; 1487 1488 for (i = 0; i < count; i++) { 1489 sk = stack[i]; 1490 if (likely(skb1 == NULL)) 1491 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1492 1493 if (!skb1) { 1494 atomic_inc(&sk->sk_drops); 1495 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1496 IS_UDPLITE(sk)); 1497 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1498 IS_UDPLITE(sk)); 1499 } 1500 1501 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1502 skb1 = NULL; 1503 } 1504 if (unlikely(skb1)) 1505 kfree_skb(skb1); 1506 } 1507 1508 /* 1509 * Multicasts and broadcasts go to each listener. 1510 * 1511 * Note: called only from the BH handler context. 1512 */ 1513 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1514 struct udphdr *uh, 1515 __be32 saddr, __be32 daddr, 1516 struct udp_table *udptable) 1517 { 1518 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1519 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1520 int dif; 1521 unsigned int i, count = 0; 1522 1523 spin_lock(&hslot->lock); 1524 sk = sk_nulls_head(&hslot->head); 1525 dif = skb->dev->ifindex; 1526 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1527 while (sk) { 1528 stack[count++] = sk; 1529 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1530 daddr, uh->source, saddr, dif); 1531 if (unlikely(count == ARRAY_SIZE(stack))) { 1532 if (!sk) 1533 break; 1534 flush_stack(stack, count, skb, ~0); 1535 count = 0; 1536 } 1537 } 1538 /* 1539 * before releasing chain lock, we must take a reference on sockets 1540 */ 1541 for (i = 0; i < count; i++) 1542 sock_hold(stack[i]); 1543 1544 spin_unlock(&hslot->lock); 1545 1546 /* 1547 * do the slow work with no lock held 1548 */ 1549 if (count) { 1550 flush_stack(stack, count, skb, count - 1); 1551 1552 for (i = 0; i < count; i++) 1553 sock_put(stack[i]); 1554 } else { 1555 kfree_skb(skb); 1556 } 1557 return 0; 1558 } 1559 1560 /* Initialize UDP checksum. If exited with zero value (success), 1561 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1562 * Otherwise, csum completion requires chacksumming packet body, 1563 * including udp header and folding it to skb->csum. 1564 */ 1565 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1566 int proto) 1567 { 1568 const struct iphdr *iph; 1569 int err; 1570 1571 UDP_SKB_CB(skb)->partial_cov = 0; 1572 UDP_SKB_CB(skb)->cscov = skb->len; 1573 1574 if (proto == IPPROTO_UDPLITE) { 1575 err = udplite_checksum_init(skb, uh); 1576 if (err) 1577 return err; 1578 } 1579 1580 iph = ip_hdr(skb); 1581 if (uh->check == 0) { 1582 skb->ip_summed = CHECKSUM_UNNECESSARY; 1583 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1584 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1585 proto, skb->csum)) 1586 skb->ip_summed = CHECKSUM_UNNECESSARY; 1587 } 1588 if (!skb_csum_unnecessary(skb)) 1589 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1590 skb->len, proto, 0); 1591 /* Probably, we should checksum udp header (it should be in cache 1592 * in any case) and data in tiny packets (< rx copybreak). 1593 */ 1594 1595 return 0; 1596 } 1597 1598 /* 1599 * All we need to do is get the socket, and then do a checksum. 1600 */ 1601 1602 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1603 int proto) 1604 { 1605 struct sock *sk; 1606 struct udphdr *uh; 1607 unsigned short ulen; 1608 struct rtable *rt = skb_rtable(skb); 1609 __be32 saddr, daddr; 1610 struct net *net = dev_net(skb->dev); 1611 1612 /* 1613 * Validate the packet. 1614 */ 1615 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1616 goto drop; /* No space for header. */ 1617 1618 uh = udp_hdr(skb); 1619 ulen = ntohs(uh->len); 1620 saddr = ip_hdr(skb)->saddr; 1621 daddr = ip_hdr(skb)->daddr; 1622 1623 if (ulen > skb->len) 1624 goto short_packet; 1625 1626 if (proto == IPPROTO_UDP) { 1627 /* UDP validates ulen. */ 1628 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1629 goto short_packet; 1630 uh = udp_hdr(skb); 1631 } 1632 1633 if (udp4_csum_init(skb, uh, proto)) 1634 goto csum_error; 1635 1636 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1637 return __udp4_lib_mcast_deliver(net, skb, uh, 1638 saddr, daddr, udptable); 1639 1640 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1641 1642 if (sk != NULL) { 1643 int ret = udp_queue_rcv_skb(sk, skb); 1644 sock_put(sk); 1645 1646 /* a return value > 0 means to resubmit the input, but 1647 * it wants the return to be -protocol, or 0 1648 */ 1649 if (ret > 0) 1650 return -ret; 1651 return 0; 1652 } 1653 1654 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1655 goto drop; 1656 nf_reset(skb); 1657 1658 /* No socket. Drop packet silently, if checksum is wrong */ 1659 if (udp_lib_checksum_complete(skb)) 1660 goto csum_error; 1661 1662 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1663 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1664 1665 /* 1666 * Hmm. We got an UDP packet to a port to which we 1667 * don't wanna listen. Ignore it. 1668 */ 1669 kfree_skb(skb); 1670 return 0; 1671 1672 short_packet: 1673 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1674 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1675 &saddr, 1676 ntohs(uh->source), 1677 ulen, 1678 skb->len, 1679 &daddr, 1680 ntohs(uh->dest)); 1681 goto drop; 1682 1683 csum_error: 1684 /* 1685 * RFC1122: OK. Discards the bad packet silently (as far as 1686 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1687 */ 1688 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1689 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1690 &saddr, 1691 ntohs(uh->source), 1692 &daddr, 1693 ntohs(uh->dest), 1694 ulen); 1695 drop: 1696 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1697 kfree_skb(skb); 1698 return 0; 1699 } 1700 1701 int udp_rcv(struct sk_buff *skb) 1702 { 1703 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1704 } 1705 1706 void udp_destroy_sock(struct sock *sk) 1707 { 1708 bool slow = lock_sock_fast(sk); 1709 udp_flush_pending_frames(sk); 1710 unlock_sock_fast(sk, slow); 1711 } 1712 1713 /* 1714 * Socket option code for UDP 1715 */ 1716 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1717 char __user *optval, unsigned int optlen, 1718 int (*push_pending_frames)(struct sock *)) 1719 { 1720 struct udp_sock *up = udp_sk(sk); 1721 int val; 1722 int err = 0; 1723 int is_udplite = IS_UDPLITE(sk); 1724 1725 if (optlen < sizeof(int)) 1726 return -EINVAL; 1727 1728 if (get_user(val, (int __user *)optval)) 1729 return -EFAULT; 1730 1731 switch (optname) { 1732 case UDP_CORK: 1733 if (val != 0) { 1734 up->corkflag = 1; 1735 } else { 1736 up->corkflag = 0; 1737 lock_sock(sk); 1738 (*push_pending_frames)(sk); 1739 release_sock(sk); 1740 } 1741 break; 1742 1743 case UDP_ENCAP: 1744 switch (val) { 1745 case 0: 1746 case UDP_ENCAP_ESPINUDP: 1747 case UDP_ENCAP_ESPINUDP_NON_IKE: 1748 up->encap_rcv = xfrm4_udp_encap_rcv; 1749 /* FALLTHROUGH */ 1750 case UDP_ENCAP_L2TPINUDP: 1751 up->encap_type = val; 1752 break; 1753 default: 1754 err = -ENOPROTOOPT; 1755 break; 1756 } 1757 break; 1758 1759 /* 1760 * UDP-Lite's partial checksum coverage (RFC 3828). 1761 */ 1762 /* The sender sets actual checksum coverage length via this option. 1763 * The case coverage > packet length is handled by send module. */ 1764 case UDPLITE_SEND_CSCOV: 1765 if (!is_udplite) /* Disable the option on UDP sockets */ 1766 return -ENOPROTOOPT; 1767 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1768 val = 8; 1769 else if (val > USHRT_MAX) 1770 val = USHRT_MAX; 1771 up->pcslen = val; 1772 up->pcflag |= UDPLITE_SEND_CC; 1773 break; 1774 1775 /* The receiver specifies a minimum checksum coverage value. To make 1776 * sense, this should be set to at least 8 (as done below). If zero is 1777 * used, this again means full checksum coverage. */ 1778 case UDPLITE_RECV_CSCOV: 1779 if (!is_udplite) /* Disable the option on UDP sockets */ 1780 return -ENOPROTOOPT; 1781 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1782 val = 8; 1783 else if (val > USHRT_MAX) 1784 val = USHRT_MAX; 1785 up->pcrlen = val; 1786 up->pcflag |= UDPLITE_RECV_CC; 1787 break; 1788 1789 default: 1790 err = -ENOPROTOOPT; 1791 break; 1792 } 1793 1794 return err; 1795 } 1796 EXPORT_SYMBOL(udp_lib_setsockopt); 1797 1798 int udp_setsockopt(struct sock *sk, int level, int optname, 1799 char __user *optval, unsigned int optlen) 1800 { 1801 if (level == SOL_UDP || level == SOL_UDPLITE) 1802 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1803 udp_push_pending_frames); 1804 return ip_setsockopt(sk, level, optname, optval, optlen); 1805 } 1806 1807 #ifdef CONFIG_COMPAT 1808 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1809 char __user *optval, unsigned int optlen) 1810 { 1811 if (level == SOL_UDP || level == SOL_UDPLITE) 1812 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1813 udp_push_pending_frames); 1814 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1815 } 1816 #endif 1817 1818 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1819 char __user *optval, int __user *optlen) 1820 { 1821 struct udp_sock *up = udp_sk(sk); 1822 int val, len; 1823 1824 if (get_user(len, optlen)) 1825 return -EFAULT; 1826 1827 len = min_t(unsigned int, len, sizeof(int)); 1828 1829 if (len < 0) 1830 return -EINVAL; 1831 1832 switch (optname) { 1833 case UDP_CORK: 1834 val = up->corkflag; 1835 break; 1836 1837 case UDP_ENCAP: 1838 val = up->encap_type; 1839 break; 1840 1841 /* The following two cannot be changed on UDP sockets, the return is 1842 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1843 case UDPLITE_SEND_CSCOV: 1844 val = up->pcslen; 1845 break; 1846 1847 case UDPLITE_RECV_CSCOV: 1848 val = up->pcrlen; 1849 break; 1850 1851 default: 1852 return -ENOPROTOOPT; 1853 } 1854 1855 if (put_user(len, optlen)) 1856 return -EFAULT; 1857 if (copy_to_user(optval, &val, len)) 1858 return -EFAULT; 1859 return 0; 1860 } 1861 EXPORT_SYMBOL(udp_lib_getsockopt); 1862 1863 int udp_getsockopt(struct sock *sk, int level, int optname, 1864 char __user *optval, int __user *optlen) 1865 { 1866 if (level == SOL_UDP || level == SOL_UDPLITE) 1867 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1868 return ip_getsockopt(sk, level, optname, optval, optlen); 1869 } 1870 1871 #ifdef CONFIG_COMPAT 1872 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1873 char __user *optval, int __user *optlen) 1874 { 1875 if (level == SOL_UDP || level == SOL_UDPLITE) 1876 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1877 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1878 } 1879 #endif 1880 /** 1881 * udp_poll - wait for a UDP event. 1882 * @file - file struct 1883 * @sock - socket 1884 * @wait - poll table 1885 * 1886 * This is same as datagram poll, except for the special case of 1887 * blocking sockets. If application is using a blocking fd 1888 * and a packet with checksum error is in the queue; 1889 * then it could get return from select indicating data available 1890 * but then block when reading it. Add special case code 1891 * to work around these arguably broken applications. 1892 */ 1893 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1894 { 1895 unsigned int mask = datagram_poll(file, sock, wait); 1896 struct sock *sk = sock->sk; 1897 1898 /* Check for false positives due to checksum errors */ 1899 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 1900 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 1901 mask &= ~(POLLIN | POLLRDNORM); 1902 1903 return mask; 1904 1905 } 1906 EXPORT_SYMBOL(udp_poll); 1907 1908 struct proto udp_prot = { 1909 .name = "UDP", 1910 .owner = THIS_MODULE, 1911 .close = udp_lib_close, 1912 .connect = ip4_datagram_connect, 1913 .disconnect = udp_disconnect, 1914 .ioctl = udp_ioctl, 1915 .destroy = udp_destroy_sock, 1916 .setsockopt = udp_setsockopt, 1917 .getsockopt = udp_getsockopt, 1918 .sendmsg = udp_sendmsg, 1919 .recvmsg = udp_recvmsg, 1920 .sendpage = udp_sendpage, 1921 .backlog_rcv = __udp_queue_rcv_skb, 1922 .hash = udp_lib_hash, 1923 .unhash = udp_lib_unhash, 1924 .rehash = udp_v4_rehash, 1925 .get_port = udp_v4_get_port, 1926 .memory_allocated = &udp_memory_allocated, 1927 .sysctl_mem = sysctl_udp_mem, 1928 .sysctl_wmem = &sysctl_udp_wmem_min, 1929 .sysctl_rmem = &sysctl_udp_rmem_min, 1930 .obj_size = sizeof(struct udp_sock), 1931 .slab_flags = SLAB_DESTROY_BY_RCU, 1932 .h.udp_table = &udp_table, 1933 #ifdef CONFIG_COMPAT 1934 .compat_setsockopt = compat_udp_setsockopt, 1935 .compat_getsockopt = compat_udp_getsockopt, 1936 #endif 1937 .clear_sk = sk_prot_clear_portaddr_nulls, 1938 }; 1939 EXPORT_SYMBOL(udp_prot); 1940 1941 /* ------------------------------------------------------------------------ */ 1942 #ifdef CONFIG_PROC_FS 1943 1944 static struct sock *udp_get_first(struct seq_file *seq, int start) 1945 { 1946 struct sock *sk; 1947 struct udp_iter_state *state = seq->private; 1948 struct net *net = seq_file_net(seq); 1949 1950 for (state->bucket = start; state->bucket <= state->udp_table->mask; 1951 ++state->bucket) { 1952 struct hlist_nulls_node *node; 1953 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 1954 1955 if (hlist_nulls_empty(&hslot->head)) 1956 continue; 1957 1958 spin_lock_bh(&hslot->lock); 1959 sk_nulls_for_each(sk, node, &hslot->head) { 1960 if (!net_eq(sock_net(sk), net)) 1961 continue; 1962 if (sk->sk_family == state->family) 1963 goto found; 1964 } 1965 spin_unlock_bh(&hslot->lock); 1966 } 1967 sk = NULL; 1968 found: 1969 return sk; 1970 } 1971 1972 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1973 { 1974 struct udp_iter_state *state = seq->private; 1975 struct net *net = seq_file_net(seq); 1976 1977 do { 1978 sk = sk_nulls_next(sk); 1979 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1980 1981 if (!sk) { 1982 if (state->bucket <= state->udp_table->mask) 1983 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1984 return udp_get_first(seq, state->bucket + 1); 1985 } 1986 return sk; 1987 } 1988 1989 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1990 { 1991 struct sock *sk = udp_get_first(seq, 0); 1992 1993 if (sk) 1994 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1995 --pos; 1996 return pos ? NULL : sk; 1997 } 1998 1999 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2000 { 2001 struct udp_iter_state *state = seq->private; 2002 state->bucket = MAX_UDP_PORTS; 2003 2004 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2005 } 2006 2007 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2008 { 2009 struct sock *sk; 2010 2011 if (v == SEQ_START_TOKEN) 2012 sk = udp_get_idx(seq, 0); 2013 else 2014 sk = udp_get_next(seq, v); 2015 2016 ++*pos; 2017 return sk; 2018 } 2019 2020 static void udp_seq_stop(struct seq_file *seq, void *v) 2021 { 2022 struct udp_iter_state *state = seq->private; 2023 2024 if (state->bucket <= state->udp_table->mask) 2025 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2026 } 2027 2028 static int udp_seq_open(struct inode *inode, struct file *file) 2029 { 2030 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 2031 struct udp_iter_state *s; 2032 int err; 2033 2034 err = seq_open_net(inode, file, &afinfo->seq_ops, 2035 sizeof(struct udp_iter_state)); 2036 if (err < 0) 2037 return err; 2038 2039 s = ((struct seq_file *)file->private_data)->private; 2040 s->family = afinfo->family; 2041 s->udp_table = afinfo->udp_table; 2042 return err; 2043 } 2044 2045 /* ------------------------------------------------------------------------ */ 2046 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2047 { 2048 struct proc_dir_entry *p; 2049 int rc = 0; 2050 2051 afinfo->seq_fops.open = udp_seq_open; 2052 afinfo->seq_fops.read = seq_read; 2053 afinfo->seq_fops.llseek = seq_lseek; 2054 afinfo->seq_fops.release = seq_release_net; 2055 2056 afinfo->seq_ops.start = udp_seq_start; 2057 afinfo->seq_ops.next = udp_seq_next; 2058 afinfo->seq_ops.stop = udp_seq_stop; 2059 2060 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2061 &afinfo->seq_fops, afinfo); 2062 if (!p) 2063 rc = -ENOMEM; 2064 return rc; 2065 } 2066 EXPORT_SYMBOL(udp_proc_register); 2067 2068 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2069 { 2070 proc_net_remove(net, afinfo->name); 2071 } 2072 EXPORT_SYMBOL(udp_proc_unregister); 2073 2074 /* ------------------------------------------------------------------------ */ 2075 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2076 int bucket, int *len) 2077 { 2078 struct inet_sock *inet = inet_sk(sp); 2079 __be32 dest = inet->inet_daddr; 2080 __be32 src = inet->inet_rcv_saddr; 2081 __u16 destp = ntohs(inet->inet_dport); 2082 __u16 srcp = ntohs(inet->inet_sport); 2083 2084 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2085 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 2086 bucket, src, srcp, dest, destp, sp->sk_state, 2087 sk_wmem_alloc_get(sp), 2088 sk_rmem_alloc_get(sp), 2089 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 2090 atomic_read(&sp->sk_refcnt), sp, 2091 atomic_read(&sp->sk_drops), len); 2092 } 2093 2094 int udp4_seq_show(struct seq_file *seq, void *v) 2095 { 2096 if (v == SEQ_START_TOKEN) 2097 seq_printf(seq, "%-127s\n", 2098 " sl local_address rem_address st tx_queue " 2099 "rx_queue tr tm->when retrnsmt uid timeout " 2100 "inode ref pointer drops"); 2101 else { 2102 struct udp_iter_state *state = seq->private; 2103 int len; 2104 2105 udp4_format_sock(v, seq, state->bucket, &len); 2106 seq_printf(seq, "%*s\n", 127 - len, ""); 2107 } 2108 return 0; 2109 } 2110 2111 /* ------------------------------------------------------------------------ */ 2112 static struct udp_seq_afinfo udp4_seq_afinfo = { 2113 .name = "udp", 2114 .family = AF_INET, 2115 .udp_table = &udp_table, 2116 .seq_fops = { 2117 .owner = THIS_MODULE, 2118 }, 2119 .seq_ops = { 2120 .show = udp4_seq_show, 2121 }, 2122 }; 2123 2124 static int __net_init udp4_proc_init_net(struct net *net) 2125 { 2126 return udp_proc_register(net, &udp4_seq_afinfo); 2127 } 2128 2129 static void __net_exit udp4_proc_exit_net(struct net *net) 2130 { 2131 udp_proc_unregister(net, &udp4_seq_afinfo); 2132 } 2133 2134 static struct pernet_operations udp4_net_ops = { 2135 .init = udp4_proc_init_net, 2136 .exit = udp4_proc_exit_net, 2137 }; 2138 2139 int __init udp4_proc_init(void) 2140 { 2141 return register_pernet_subsys(&udp4_net_ops); 2142 } 2143 2144 void udp4_proc_exit(void) 2145 { 2146 unregister_pernet_subsys(&udp4_net_ops); 2147 } 2148 #endif /* CONFIG_PROC_FS */ 2149 2150 static __initdata unsigned long uhash_entries; 2151 static int __init set_uhash_entries(char *str) 2152 { 2153 if (!str) 2154 return 0; 2155 uhash_entries = simple_strtoul(str, &str, 0); 2156 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2157 uhash_entries = UDP_HTABLE_SIZE_MIN; 2158 return 1; 2159 } 2160 __setup("uhash_entries=", set_uhash_entries); 2161 2162 void __init udp_table_init(struct udp_table *table, const char *name) 2163 { 2164 unsigned int i; 2165 2166 if (!CONFIG_BASE_SMALL) 2167 table->hash = alloc_large_system_hash(name, 2168 2 * sizeof(struct udp_hslot), 2169 uhash_entries, 2170 21, /* one slot per 2 MB */ 2171 0, 2172 &table->log, 2173 &table->mask, 2174 64 * 1024); 2175 /* 2176 * Make sure hash table has the minimum size 2177 */ 2178 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) { 2179 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN * 2180 2 * sizeof(struct udp_hslot), GFP_KERNEL); 2181 if (!table->hash) 2182 panic(name); 2183 table->log = ilog2(UDP_HTABLE_SIZE_MIN); 2184 table->mask = UDP_HTABLE_SIZE_MIN - 1; 2185 } 2186 table->hash2 = table->hash + (table->mask + 1); 2187 for (i = 0; i <= table->mask; i++) { 2188 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2189 table->hash[i].count = 0; 2190 spin_lock_init(&table->hash[i].lock); 2191 } 2192 for (i = 0; i <= table->mask; i++) { 2193 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2194 table->hash2[i].count = 0; 2195 spin_lock_init(&table->hash2[i].lock); 2196 } 2197 } 2198 2199 void __init udp_init(void) 2200 { 2201 unsigned long nr_pages, limit; 2202 2203 udp_table_init(&udp_table, "UDP"); 2204 /* Set the pressure threshold up by the same strategy of TCP. It is a 2205 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 2206 * toward zero with the amount of memory, with a floor of 128 pages. 2207 */ 2208 nr_pages = totalram_pages - totalhigh_pages; 2209 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 2210 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 2211 limit = max(limit, 128UL); 2212 sysctl_udp_mem[0] = limit / 4 * 3; 2213 sysctl_udp_mem[1] = limit; 2214 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2215 2216 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2217 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2218 } 2219 2220 int udp4_ufo_send_check(struct sk_buff *skb) 2221 { 2222 const struct iphdr *iph; 2223 struct udphdr *uh; 2224 2225 if (!pskb_may_pull(skb, sizeof(*uh))) 2226 return -EINVAL; 2227 2228 iph = ip_hdr(skb); 2229 uh = udp_hdr(skb); 2230 2231 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 2232 IPPROTO_UDP, 0); 2233 skb->csum_start = skb_transport_header(skb) - skb->head; 2234 skb->csum_offset = offsetof(struct udphdr, check); 2235 skb->ip_summed = CHECKSUM_PARTIAL; 2236 return 0; 2237 } 2238 2239 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, u32 features) 2240 { 2241 struct sk_buff *segs = ERR_PTR(-EINVAL); 2242 unsigned int mss; 2243 int offset; 2244 __wsum csum; 2245 2246 mss = skb_shinfo(skb)->gso_size; 2247 if (unlikely(skb->len <= mss)) 2248 goto out; 2249 2250 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 2251 /* Packet is from an untrusted source, reset gso_segs. */ 2252 int type = skb_shinfo(skb)->gso_type; 2253 2254 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || 2255 !(type & (SKB_GSO_UDP)))) 2256 goto out; 2257 2258 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 2259 2260 segs = NULL; 2261 goto out; 2262 } 2263 2264 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot 2265 * do checksum of UDP packets sent as multiple IP fragments. 2266 */ 2267 offset = skb_checksum_start_offset(skb); 2268 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2269 offset += skb->csum_offset; 2270 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2271 skb->ip_summed = CHECKSUM_NONE; 2272 2273 /* Fragment the skb. IP headers of the fragments are updated in 2274 * inet_gso_segment() 2275 */ 2276 segs = skb_segment(skb, features); 2277 out: 2278 return segs; 2279 } 2280 2281