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 <trace/events/udp.h> 109 #include "udp_impl.h" 110 111 struct udp_table udp_table __read_mostly; 112 EXPORT_SYMBOL(udp_table); 113 114 long sysctl_udp_mem[3] __read_mostly; 115 EXPORT_SYMBOL(sysctl_udp_mem); 116 117 int sysctl_udp_rmem_min __read_mostly; 118 EXPORT_SYMBOL(sysctl_udp_rmem_min); 119 120 int sysctl_udp_wmem_min __read_mostly; 121 EXPORT_SYMBOL(sysctl_udp_wmem_min); 122 123 atomic_long_t udp_memory_allocated; 124 EXPORT_SYMBOL(udp_memory_allocated); 125 126 #define MAX_UDP_PORTS 65536 127 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 128 129 static int udp_lib_lport_inuse(struct net *net, __u16 num, 130 const struct udp_hslot *hslot, 131 unsigned long *bitmap, 132 struct sock *sk, 133 int (*saddr_comp)(const struct sock *sk1, 134 const struct sock *sk2), 135 unsigned int log) 136 { 137 struct sock *sk2; 138 struct hlist_nulls_node *node; 139 140 sk_nulls_for_each(sk2, node, &hslot->head) 141 if (net_eq(sock_net(sk2), net) && 142 sk2 != sk && 143 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 144 (!sk2->sk_reuse || !sk->sk_reuse) && 145 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 146 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 147 (*saddr_comp)(sk, sk2)) { 148 if (bitmap) 149 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 150 bitmap); 151 else 152 return 1; 153 } 154 return 0; 155 } 156 157 /* 158 * Note: we still hold spinlock of primary hash chain, so no other writer 159 * can insert/delete a socket with local_port == num 160 */ 161 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 162 struct udp_hslot *hslot2, 163 struct sock *sk, 164 int (*saddr_comp)(const struct sock *sk1, 165 const struct sock *sk2)) 166 { 167 struct sock *sk2; 168 struct hlist_nulls_node *node; 169 int res = 0; 170 171 spin_lock(&hslot2->lock); 172 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) 173 if (net_eq(sock_net(sk2), net) && 174 sk2 != sk && 175 (udp_sk(sk2)->udp_port_hash == num) && 176 (!sk2->sk_reuse || !sk->sk_reuse) && 177 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 178 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 179 (*saddr_comp)(sk, sk2)) { 180 res = 1; 181 break; 182 } 183 spin_unlock(&hslot2->lock); 184 return res; 185 } 186 187 /** 188 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 189 * 190 * @sk: socket struct in question 191 * @snum: port number to look up 192 * @saddr_comp: AF-dependent comparison of bound local IP addresses 193 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 194 * with NULL address 195 */ 196 int udp_lib_get_port(struct sock *sk, unsigned short snum, 197 int (*saddr_comp)(const struct sock *sk1, 198 const struct sock *sk2), 199 unsigned int hash2_nulladdr) 200 { 201 struct udp_hslot *hslot, *hslot2; 202 struct udp_table *udptable = sk->sk_prot->h.udp_table; 203 int error = 1; 204 struct net *net = sock_net(sk); 205 206 if (!snum) { 207 int low, high, remaining; 208 unsigned rand; 209 unsigned short first, last; 210 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 211 212 inet_get_local_port_range(&low, &high); 213 remaining = (high - low) + 1; 214 215 rand = net_random(); 216 first = (((u64)rand * remaining) >> 32) + low; 217 /* 218 * force rand to be an odd multiple of UDP_HTABLE_SIZE 219 */ 220 rand = (rand | 1) * (udptable->mask + 1); 221 last = first + udptable->mask + 1; 222 do { 223 hslot = udp_hashslot(udptable, net, first); 224 bitmap_zero(bitmap, PORTS_PER_CHAIN); 225 spin_lock_bh(&hslot->lock); 226 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 227 saddr_comp, udptable->log); 228 229 snum = first; 230 /* 231 * Iterate on all possible values of snum for this hash. 232 * Using steps of an odd multiple of UDP_HTABLE_SIZE 233 * give us randomization and full range coverage. 234 */ 235 do { 236 if (low <= snum && snum <= high && 237 !test_bit(snum >> udptable->log, bitmap) && 238 !inet_is_reserved_local_port(snum)) 239 goto found; 240 snum += rand; 241 } while (snum != first); 242 spin_unlock_bh(&hslot->lock); 243 } while (++first != last); 244 goto fail; 245 } else { 246 hslot = udp_hashslot(udptable, net, snum); 247 spin_lock_bh(&hslot->lock); 248 if (hslot->count > 10) { 249 int exist; 250 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 251 252 slot2 &= udptable->mask; 253 hash2_nulladdr &= udptable->mask; 254 255 hslot2 = udp_hashslot2(udptable, slot2); 256 if (hslot->count < hslot2->count) 257 goto scan_primary_hash; 258 259 exist = udp_lib_lport_inuse2(net, snum, hslot2, 260 sk, saddr_comp); 261 if (!exist && (hash2_nulladdr != slot2)) { 262 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 263 exist = udp_lib_lport_inuse2(net, snum, hslot2, 264 sk, saddr_comp); 265 } 266 if (exist) 267 goto fail_unlock; 268 else 269 goto found; 270 } 271 scan_primary_hash: 272 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 273 saddr_comp, 0)) 274 goto fail_unlock; 275 } 276 found: 277 inet_sk(sk)->inet_num = snum; 278 udp_sk(sk)->udp_port_hash = snum; 279 udp_sk(sk)->udp_portaddr_hash ^= snum; 280 if (sk_unhashed(sk)) { 281 sk_nulls_add_node_rcu(sk, &hslot->head); 282 hslot->count++; 283 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 284 285 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 286 spin_lock(&hslot2->lock); 287 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 288 &hslot2->head); 289 hslot2->count++; 290 spin_unlock(&hslot2->lock); 291 } 292 error = 0; 293 fail_unlock: 294 spin_unlock_bh(&hslot->lock); 295 fail: 296 return error; 297 } 298 EXPORT_SYMBOL(udp_lib_get_port); 299 300 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 301 { 302 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 303 304 return (!ipv6_only_sock(sk2) && 305 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 306 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 307 } 308 309 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr, 310 unsigned int port) 311 { 312 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 313 } 314 315 int udp_v4_get_port(struct sock *sk, unsigned short snum) 316 { 317 unsigned int hash2_nulladdr = 318 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 319 unsigned int hash2_partial = 320 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 321 322 /* precompute partial secondary hash */ 323 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 324 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 325 } 326 327 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, 328 unsigned short hnum, 329 __be16 sport, __be32 daddr, __be16 dport, int dif) 330 { 331 int score = -1; 332 333 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 334 !ipv6_only_sock(sk)) { 335 struct inet_sock *inet = inet_sk(sk); 336 337 score = (sk->sk_family == PF_INET ? 1 : 0); 338 if (inet->inet_rcv_saddr) { 339 if (inet->inet_rcv_saddr != daddr) 340 return -1; 341 score += 2; 342 } 343 if (inet->inet_daddr) { 344 if (inet->inet_daddr != saddr) 345 return -1; 346 score += 2; 347 } 348 if (inet->inet_dport) { 349 if (inet->inet_dport != sport) 350 return -1; 351 score += 2; 352 } 353 if (sk->sk_bound_dev_if) { 354 if (sk->sk_bound_dev_if != dif) 355 return -1; 356 score += 2; 357 } 358 } 359 return score; 360 } 361 362 /* 363 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 364 */ 365 #define SCORE2_MAX (1 + 2 + 2 + 2) 366 static inline int compute_score2(struct sock *sk, struct net *net, 367 __be32 saddr, __be16 sport, 368 __be32 daddr, unsigned int hnum, int dif) 369 { 370 int score = -1; 371 372 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) { 373 struct inet_sock *inet = inet_sk(sk); 374 375 if (inet->inet_rcv_saddr != daddr) 376 return -1; 377 if (inet->inet_num != hnum) 378 return -1; 379 380 score = (sk->sk_family == PF_INET ? 1 : 0); 381 if (inet->inet_daddr) { 382 if (inet->inet_daddr != saddr) 383 return -1; 384 score += 2; 385 } 386 if (inet->inet_dport) { 387 if (inet->inet_dport != sport) 388 return -1; 389 score += 2; 390 } 391 if (sk->sk_bound_dev_if) { 392 if (sk->sk_bound_dev_if != dif) 393 return -1; 394 score += 2; 395 } 396 } 397 return score; 398 } 399 400 401 /* called with read_rcu_lock() */ 402 static struct sock *udp4_lib_lookup2(struct net *net, 403 __be32 saddr, __be16 sport, 404 __be32 daddr, unsigned int hnum, int dif, 405 struct udp_hslot *hslot2, unsigned int slot2) 406 { 407 struct sock *sk, *result; 408 struct hlist_nulls_node *node; 409 int score, badness; 410 411 begin: 412 result = NULL; 413 badness = -1; 414 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 415 score = compute_score2(sk, net, saddr, sport, 416 daddr, hnum, dif); 417 if (score > badness) { 418 result = sk; 419 badness = score; 420 if (score == SCORE2_MAX) 421 goto exact_match; 422 } 423 } 424 /* 425 * if the nulls value we got at the end of this lookup is 426 * not the expected one, we must restart lookup. 427 * We probably met an item that was moved to another chain. 428 */ 429 if (get_nulls_value(node) != slot2) 430 goto begin; 431 432 if (result) { 433 exact_match: 434 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 435 result = NULL; 436 else if (unlikely(compute_score2(result, net, saddr, sport, 437 daddr, hnum, dif) < badness)) { 438 sock_put(result); 439 goto begin; 440 } 441 } 442 return result; 443 } 444 445 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 446 * harder than this. -DaveM 447 */ 448 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 449 __be16 sport, __be32 daddr, __be16 dport, 450 int dif, struct udp_table *udptable) 451 { 452 struct sock *sk, *result; 453 struct hlist_nulls_node *node; 454 unsigned short hnum = ntohs(dport); 455 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 456 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 457 int score, badness; 458 459 rcu_read_lock(); 460 if (hslot->count > 10) { 461 hash2 = udp4_portaddr_hash(net, daddr, hnum); 462 slot2 = hash2 & udptable->mask; 463 hslot2 = &udptable->hash2[slot2]; 464 if (hslot->count < hslot2->count) 465 goto begin; 466 467 result = udp4_lib_lookup2(net, saddr, sport, 468 daddr, hnum, dif, 469 hslot2, slot2); 470 if (!result) { 471 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 472 slot2 = hash2 & udptable->mask; 473 hslot2 = &udptable->hash2[slot2]; 474 if (hslot->count < hslot2->count) 475 goto begin; 476 477 result = udp4_lib_lookup2(net, saddr, sport, 478 htonl(INADDR_ANY), hnum, dif, 479 hslot2, slot2); 480 } 481 rcu_read_unlock(); 482 return result; 483 } 484 begin: 485 result = NULL; 486 badness = -1; 487 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 488 score = compute_score(sk, net, saddr, hnum, sport, 489 daddr, dport, dif); 490 if (score > badness) { 491 result = sk; 492 badness = score; 493 } 494 } 495 /* 496 * if the nulls value we got at the end of this lookup is 497 * not the expected one, we must restart lookup. 498 * We probably met an item that was moved to another chain. 499 */ 500 if (get_nulls_value(node) != slot) 501 goto begin; 502 503 if (result) { 504 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 505 result = NULL; 506 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 507 daddr, dport, dif) < badness)) { 508 sock_put(result); 509 goto begin; 510 } 511 } 512 rcu_read_unlock(); 513 return result; 514 } 515 516 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 517 __be16 sport, __be16 dport, 518 struct udp_table *udptable) 519 { 520 struct sock *sk; 521 const struct iphdr *iph = ip_hdr(skb); 522 523 if (unlikely(sk = skb_steal_sock(skb))) 524 return sk; 525 else 526 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 527 iph->daddr, dport, inet_iif(skb), 528 udptable); 529 } 530 531 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 532 __be32 daddr, __be16 dport, int dif) 533 { 534 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 535 } 536 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 537 538 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 539 __be16 loc_port, __be32 loc_addr, 540 __be16 rmt_port, __be32 rmt_addr, 541 int dif) 542 { 543 struct hlist_nulls_node *node; 544 struct sock *s = sk; 545 unsigned short hnum = ntohs(loc_port); 546 547 sk_nulls_for_each_from(s, node) { 548 struct inet_sock *inet = inet_sk(s); 549 550 if (!net_eq(sock_net(s), net) || 551 udp_sk(s)->udp_port_hash != hnum || 552 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 553 (inet->inet_dport != rmt_port && inet->inet_dport) || 554 (inet->inet_rcv_saddr && 555 inet->inet_rcv_saddr != loc_addr) || 556 ipv6_only_sock(s) || 557 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 558 continue; 559 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 560 continue; 561 goto found; 562 } 563 s = NULL; 564 found: 565 return s; 566 } 567 568 /* 569 * This routine is called by the ICMP module when it gets some 570 * sort of error condition. If err < 0 then the socket should 571 * be closed and the error returned to the user. If err > 0 572 * it's just the icmp type << 8 | icmp code. 573 * Header points to the ip header of the error packet. We move 574 * on past this. Then (as it used to claim before adjustment) 575 * header points to the first 8 bytes of the udp header. We need 576 * to find the appropriate port. 577 */ 578 579 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 580 { 581 struct inet_sock *inet; 582 const struct iphdr *iph = (const struct iphdr *)skb->data; 583 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 584 const int type = icmp_hdr(skb)->type; 585 const int code = icmp_hdr(skb)->code; 586 struct sock *sk; 587 int harderr; 588 int err; 589 struct net *net = dev_net(skb->dev); 590 591 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 592 iph->saddr, uh->source, skb->dev->ifindex, udptable); 593 if (sk == NULL) { 594 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 595 return; /* No socket for error */ 596 } 597 598 err = 0; 599 harderr = 0; 600 inet = inet_sk(sk); 601 602 switch (type) { 603 default: 604 case ICMP_TIME_EXCEEDED: 605 err = EHOSTUNREACH; 606 break; 607 case ICMP_SOURCE_QUENCH: 608 goto out; 609 case ICMP_PARAMETERPROB: 610 err = EPROTO; 611 harderr = 1; 612 break; 613 case ICMP_DEST_UNREACH: 614 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 615 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 616 err = EMSGSIZE; 617 harderr = 1; 618 break; 619 } 620 goto out; 621 } 622 err = EHOSTUNREACH; 623 if (code <= NR_ICMP_UNREACH) { 624 harderr = icmp_err_convert[code].fatal; 625 err = icmp_err_convert[code].errno; 626 } 627 break; 628 } 629 630 /* 631 * RFC1122: OK. Passes ICMP errors back to application, as per 632 * 4.1.3.3. 633 */ 634 if (!inet->recverr) { 635 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 636 goto out; 637 } else 638 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 639 640 sk->sk_err = err; 641 sk->sk_error_report(sk); 642 out: 643 sock_put(sk); 644 } 645 646 void udp_err(struct sk_buff *skb, u32 info) 647 { 648 __udp4_lib_err(skb, info, &udp_table); 649 } 650 651 /* 652 * Throw away all pending data and cancel the corking. Socket is locked. 653 */ 654 void udp_flush_pending_frames(struct sock *sk) 655 { 656 struct udp_sock *up = udp_sk(sk); 657 658 if (up->pending) { 659 up->len = 0; 660 up->pending = 0; 661 ip_flush_pending_frames(sk); 662 } 663 } 664 EXPORT_SYMBOL(udp_flush_pending_frames); 665 666 /** 667 * udp4_hwcsum - handle outgoing HW checksumming 668 * @skb: sk_buff containing the filled-in UDP header 669 * (checksum field must be zeroed out) 670 * @src: source IP address 671 * @dst: destination IP address 672 */ 673 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 674 { 675 struct udphdr *uh = udp_hdr(skb); 676 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 677 int offset = skb_transport_offset(skb); 678 int len = skb->len - offset; 679 int hlen = len; 680 __wsum csum = 0; 681 682 if (!frags) { 683 /* 684 * Only one fragment on the socket. 685 */ 686 skb->csum_start = skb_transport_header(skb) - skb->head; 687 skb->csum_offset = offsetof(struct udphdr, check); 688 uh->check = ~csum_tcpudp_magic(src, dst, len, 689 IPPROTO_UDP, 0); 690 } else { 691 /* 692 * HW-checksum won't work as there are two or more 693 * fragments on the socket so that all csums of sk_buffs 694 * should be together 695 */ 696 do { 697 csum = csum_add(csum, frags->csum); 698 hlen -= frags->len; 699 } while ((frags = frags->next)); 700 701 csum = skb_checksum(skb, offset, hlen, csum); 702 skb->ip_summed = CHECKSUM_NONE; 703 704 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 705 if (uh->check == 0) 706 uh->check = CSUM_MANGLED_0; 707 } 708 } 709 710 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 711 { 712 struct sock *sk = skb->sk; 713 struct inet_sock *inet = inet_sk(sk); 714 struct udphdr *uh; 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 = fl4->fl4_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, fl4->saddr, fl4->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(fl4->saddr, fl4->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, fl4); 778 if (!skb) 779 goto out; 780 781 err = udp_send_skb(skb, fl4); 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_stack; 795 struct flowi4 *fl4; 796 int ulen = len; 797 struct ipcm_cookie ipc; 798 struct rtable *rt = NULL; 799 int free = 0; 800 int connected = 0; 801 __be32 daddr, faddr, saddr; 802 __be16 dport; 803 u8 tos; 804 int err, is_udplite = IS_UDPLITE(sk); 805 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 806 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 807 struct sk_buff *skb; 808 struct ip_options_data opt_copy; 809 810 if (len > 0xFFFF) 811 return -EMSGSIZE; 812 813 /* 814 * Check the flags. 815 */ 816 817 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 818 return -EOPNOTSUPP; 819 820 ipc.opt = NULL; 821 ipc.tx_flags = 0; 822 823 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 824 825 fl4 = &inet->cork.fl.u.ip4; 826 if (up->pending) { 827 /* 828 * There are pending frames. 829 * The socket lock must be held while it's corked. 830 */ 831 lock_sock(sk); 832 if (likely(up->pending)) { 833 if (unlikely(up->pending != AF_INET)) { 834 release_sock(sk); 835 return -EINVAL; 836 } 837 goto do_append_data; 838 } 839 release_sock(sk); 840 } 841 ulen += sizeof(struct udphdr); 842 843 /* 844 * Get and verify the address. 845 */ 846 if (msg->msg_name) { 847 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name; 848 if (msg->msg_namelen < sizeof(*usin)) 849 return -EINVAL; 850 if (usin->sin_family != AF_INET) { 851 if (usin->sin_family != AF_UNSPEC) 852 return -EAFNOSUPPORT; 853 } 854 855 daddr = usin->sin_addr.s_addr; 856 dport = usin->sin_port; 857 if (dport == 0) 858 return -EINVAL; 859 } else { 860 if (sk->sk_state != TCP_ESTABLISHED) 861 return -EDESTADDRREQ; 862 daddr = inet->inet_daddr; 863 dport = inet->inet_dport; 864 /* Open fast path for connected socket. 865 Route will not be used, if at least one option is set. 866 */ 867 connected = 1; 868 } 869 ipc.addr = inet->inet_saddr; 870 871 ipc.oif = sk->sk_bound_dev_if; 872 err = sock_tx_timestamp(sk, &ipc.tx_flags); 873 if (err) 874 return err; 875 if (msg->msg_controllen) { 876 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 877 if (err) 878 return err; 879 if (ipc.opt) 880 free = 1; 881 connected = 0; 882 } 883 if (!ipc.opt) { 884 struct ip_options_rcu *inet_opt; 885 886 rcu_read_lock(); 887 inet_opt = rcu_dereference(inet->inet_opt); 888 if (inet_opt) { 889 memcpy(&opt_copy, inet_opt, 890 sizeof(*inet_opt) + inet_opt->opt.optlen); 891 ipc.opt = &opt_copy.opt; 892 } 893 rcu_read_unlock(); 894 } 895 896 saddr = ipc.addr; 897 ipc.addr = faddr = daddr; 898 899 if (ipc.opt && ipc.opt->opt.srr) { 900 if (!daddr) 901 return -EINVAL; 902 faddr = ipc.opt->opt.faddr; 903 connected = 0; 904 } 905 tos = RT_TOS(inet->tos); 906 if (sock_flag(sk, SOCK_LOCALROUTE) || 907 (msg->msg_flags & MSG_DONTROUTE) || 908 (ipc.opt && ipc.opt->opt.is_strictroute)) { 909 tos |= RTO_ONLINK; 910 connected = 0; 911 } 912 913 if (ipv4_is_multicast(daddr)) { 914 if (!ipc.oif) 915 ipc.oif = inet->mc_index; 916 if (!saddr) 917 saddr = inet->mc_addr; 918 connected = 0; 919 } 920 921 if (connected) 922 rt = (struct rtable *)sk_dst_check(sk, 0); 923 924 if (rt == NULL) { 925 struct net *net = sock_net(sk); 926 927 fl4 = &fl4_stack; 928 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 929 RT_SCOPE_UNIVERSE, sk->sk_protocol, 930 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP, 931 faddr, saddr, dport, inet->inet_sport); 932 933 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 934 rt = ip_route_output_flow(net, fl4, sk); 935 if (IS_ERR(rt)) { 936 err = PTR_ERR(rt); 937 rt = NULL; 938 if (err == -ENETUNREACH) 939 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 940 goto out; 941 } 942 943 err = -EACCES; 944 if ((rt->rt_flags & RTCF_BROADCAST) && 945 !sock_flag(sk, SOCK_BROADCAST)) 946 goto out; 947 if (connected) 948 sk_dst_set(sk, dst_clone(&rt->dst)); 949 } 950 951 if (msg->msg_flags&MSG_CONFIRM) 952 goto do_confirm; 953 back_from_confirm: 954 955 saddr = fl4->saddr; 956 if (!ipc.addr) 957 daddr = ipc.addr = fl4->daddr; 958 959 /* Lockless fast path for the non-corking case. */ 960 if (!corkreq) { 961 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen, 962 sizeof(struct udphdr), &ipc, &rt, 963 msg->msg_flags); 964 err = PTR_ERR(skb); 965 if (skb && !IS_ERR(skb)) 966 err = udp_send_skb(skb, fl4); 967 goto out; 968 } 969 970 lock_sock(sk); 971 if (unlikely(up->pending)) { 972 /* The socket is already corked while preparing it. */ 973 /* ... which is an evident application bug. --ANK */ 974 release_sock(sk); 975 976 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 977 err = -EINVAL; 978 goto out; 979 } 980 /* 981 * Now cork the socket to pend data. 982 */ 983 fl4 = &inet->cork.fl.u.ip4; 984 fl4->daddr = daddr; 985 fl4->saddr = saddr; 986 fl4->fl4_dport = dport; 987 fl4->fl4_sport = inet->inet_sport; 988 up->pending = AF_INET; 989 990 do_append_data: 991 up->len += ulen; 992 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen, 993 sizeof(struct udphdr), &ipc, &rt, 994 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 995 if (err) 996 udp_flush_pending_frames(sk); 997 else if (!corkreq) 998 err = udp_push_pending_frames(sk); 999 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1000 up->pending = 0; 1001 release_sock(sk); 1002 1003 out: 1004 ip_rt_put(rt); 1005 if (free) 1006 kfree(ipc.opt); 1007 if (!err) 1008 return len; 1009 /* 1010 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1011 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1012 * we don't have a good statistic (IpOutDiscards but it can be too many 1013 * things). We could add another new stat but at least for now that 1014 * seems like overkill. 1015 */ 1016 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1017 UDP_INC_STATS_USER(sock_net(sk), 1018 UDP_MIB_SNDBUFERRORS, is_udplite); 1019 } 1020 return err; 1021 1022 do_confirm: 1023 dst_confirm(&rt->dst); 1024 if (!(msg->msg_flags&MSG_PROBE) || len) 1025 goto back_from_confirm; 1026 err = 0; 1027 goto out; 1028 } 1029 EXPORT_SYMBOL(udp_sendmsg); 1030 1031 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1032 size_t size, int flags) 1033 { 1034 struct inet_sock *inet = inet_sk(sk); 1035 struct udp_sock *up = udp_sk(sk); 1036 int ret; 1037 1038 if (!up->pending) { 1039 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1040 1041 /* Call udp_sendmsg to specify destination address which 1042 * sendpage interface can't pass. 1043 * This will succeed only when the socket is connected. 1044 */ 1045 ret = udp_sendmsg(NULL, sk, &msg, 0); 1046 if (ret < 0) 1047 return ret; 1048 } 1049 1050 lock_sock(sk); 1051 1052 if (unlikely(!up->pending)) { 1053 release_sock(sk); 1054 1055 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 1056 return -EINVAL; 1057 } 1058 1059 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1060 page, offset, size, flags); 1061 if (ret == -EOPNOTSUPP) { 1062 release_sock(sk); 1063 return sock_no_sendpage(sk->sk_socket, page, offset, 1064 size, flags); 1065 } 1066 if (ret < 0) { 1067 udp_flush_pending_frames(sk); 1068 goto out; 1069 } 1070 1071 up->len += size; 1072 if (!(up->corkflag || (flags&MSG_MORE))) 1073 ret = udp_push_pending_frames(sk); 1074 if (!ret) 1075 ret = size; 1076 out: 1077 release_sock(sk); 1078 return ret; 1079 } 1080 1081 1082 /** 1083 * first_packet_length - return length of first packet in receive queue 1084 * @sk: socket 1085 * 1086 * Drops all bad checksum frames, until a valid one is found. 1087 * Returns the length of found skb, or 0 if none is found. 1088 */ 1089 static unsigned int first_packet_length(struct sock *sk) 1090 { 1091 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1092 struct sk_buff *skb; 1093 unsigned int res; 1094 1095 __skb_queue_head_init(&list_kill); 1096 1097 spin_lock_bh(&rcvq->lock); 1098 while ((skb = skb_peek(rcvq)) != NULL && 1099 udp_lib_checksum_complete(skb)) { 1100 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1101 IS_UDPLITE(sk)); 1102 atomic_inc(&sk->sk_drops); 1103 __skb_unlink(skb, rcvq); 1104 __skb_queue_tail(&list_kill, skb); 1105 } 1106 res = skb ? skb->len : 0; 1107 spin_unlock_bh(&rcvq->lock); 1108 1109 if (!skb_queue_empty(&list_kill)) { 1110 bool slow = lock_sock_fast(sk); 1111 1112 __skb_queue_purge(&list_kill); 1113 sk_mem_reclaim_partial(sk); 1114 unlock_sock_fast(sk, slow); 1115 } 1116 return res; 1117 } 1118 1119 /* 1120 * IOCTL requests applicable to the UDP protocol 1121 */ 1122 1123 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1124 { 1125 switch (cmd) { 1126 case SIOCOUTQ: 1127 { 1128 int amount = sk_wmem_alloc_get(sk); 1129 1130 return put_user(amount, (int __user *)arg); 1131 } 1132 1133 case SIOCINQ: 1134 { 1135 unsigned int amount = first_packet_length(sk); 1136 1137 if (amount) 1138 /* 1139 * We will only return the amount 1140 * of this packet since that is all 1141 * that will be read. 1142 */ 1143 amount -= sizeof(struct udphdr); 1144 1145 return put_user(amount, (int __user *)arg); 1146 } 1147 1148 default: 1149 return -ENOIOCTLCMD; 1150 } 1151 1152 return 0; 1153 } 1154 EXPORT_SYMBOL(udp_ioctl); 1155 1156 /* 1157 * This should be easy, if there is something there we 1158 * return it, otherwise we block. 1159 */ 1160 1161 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1162 size_t len, int noblock, int flags, int *addr_len) 1163 { 1164 struct inet_sock *inet = inet_sk(sk); 1165 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 1166 struct sk_buff *skb; 1167 unsigned int ulen; 1168 int peeked; 1169 int err; 1170 int is_udplite = IS_UDPLITE(sk); 1171 bool slow; 1172 1173 /* 1174 * Check any passed addresses 1175 */ 1176 if (addr_len) 1177 *addr_len = sizeof(*sin); 1178 1179 if (flags & MSG_ERRQUEUE) 1180 return ip_recv_error(sk, msg, len); 1181 1182 try_again: 1183 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1184 &peeked, &err); 1185 if (!skb) 1186 goto out; 1187 1188 ulen = skb->len - sizeof(struct udphdr); 1189 if (len > ulen) 1190 len = ulen; 1191 else if (len < ulen) 1192 msg->msg_flags |= MSG_TRUNC; 1193 1194 /* 1195 * If checksum is needed at all, try to do it while copying the 1196 * data. If the data is truncated, or if we only want a partial 1197 * coverage checksum (UDP-Lite), do it before the copy. 1198 */ 1199 1200 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) { 1201 if (udp_lib_checksum_complete(skb)) 1202 goto csum_copy_err; 1203 } 1204 1205 if (skb_csum_unnecessary(skb)) 1206 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 1207 msg->msg_iov, len); 1208 else { 1209 err = skb_copy_and_csum_datagram_iovec(skb, 1210 sizeof(struct udphdr), 1211 msg->msg_iov); 1212 1213 if (err == -EINVAL) 1214 goto csum_copy_err; 1215 } 1216 1217 if (err) 1218 goto out_free; 1219 1220 if (!peeked) 1221 UDP_INC_STATS_USER(sock_net(sk), 1222 UDP_MIB_INDATAGRAMS, is_udplite); 1223 1224 sock_recv_ts_and_drops(msg, sk, skb); 1225 1226 /* Copy the address. */ 1227 if (sin) { 1228 sin->sin_family = AF_INET; 1229 sin->sin_port = udp_hdr(skb)->source; 1230 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1231 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1232 } 1233 if (inet->cmsg_flags) 1234 ip_cmsg_recv(msg, skb); 1235 1236 err = len; 1237 if (flags & MSG_TRUNC) 1238 err = ulen; 1239 1240 out_free: 1241 skb_free_datagram_locked(sk, skb); 1242 out: 1243 return err; 1244 1245 csum_copy_err: 1246 slow = lock_sock_fast(sk); 1247 if (!skb_kill_datagram(sk, skb, flags)) 1248 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1249 unlock_sock_fast(sk, slow); 1250 1251 if (noblock) 1252 return -EAGAIN; 1253 1254 /* starting over for a new packet */ 1255 msg->msg_flags &= ~MSG_TRUNC; 1256 goto try_again; 1257 } 1258 1259 1260 int udp_disconnect(struct sock *sk, int flags) 1261 { 1262 struct inet_sock *inet = inet_sk(sk); 1263 /* 1264 * 1003.1g - break association. 1265 */ 1266 1267 sk->sk_state = TCP_CLOSE; 1268 inet->inet_daddr = 0; 1269 inet->inet_dport = 0; 1270 sock_rps_reset_rxhash(sk); 1271 sk->sk_bound_dev_if = 0; 1272 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1273 inet_reset_saddr(sk); 1274 1275 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1276 sk->sk_prot->unhash(sk); 1277 inet->inet_sport = 0; 1278 } 1279 sk_dst_reset(sk); 1280 return 0; 1281 } 1282 EXPORT_SYMBOL(udp_disconnect); 1283 1284 void udp_lib_unhash(struct sock *sk) 1285 { 1286 if (sk_hashed(sk)) { 1287 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1288 struct udp_hslot *hslot, *hslot2; 1289 1290 hslot = udp_hashslot(udptable, sock_net(sk), 1291 udp_sk(sk)->udp_port_hash); 1292 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1293 1294 spin_lock_bh(&hslot->lock); 1295 if (sk_nulls_del_node_init_rcu(sk)) { 1296 hslot->count--; 1297 inet_sk(sk)->inet_num = 0; 1298 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1299 1300 spin_lock(&hslot2->lock); 1301 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1302 hslot2->count--; 1303 spin_unlock(&hslot2->lock); 1304 } 1305 spin_unlock_bh(&hslot->lock); 1306 } 1307 } 1308 EXPORT_SYMBOL(udp_lib_unhash); 1309 1310 /* 1311 * inet_rcv_saddr was changed, we must rehash secondary hash 1312 */ 1313 void udp_lib_rehash(struct sock *sk, u16 newhash) 1314 { 1315 if (sk_hashed(sk)) { 1316 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1317 struct udp_hslot *hslot, *hslot2, *nhslot2; 1318 1319 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1320 nhslot2 = udp_hashslot2(udptable, newhash); 1321 udp_sk(sk)->udp_portaddr_hash = newhash; 1322 if (hslot2 != nhslot2) { 1323 hslot = udp_hashslot(udptable, sock_net(sk), 1324 udp_sk(sk)->udp_port_hash); 1325 /* we must lock primary chain too */ 1326 spin_lock_bh(&hslot->lock); 1327 1328 spin_lock(&hslot2->lock); 1329 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1330 hslot2->count--; 1331 spin_unlock(&hslot2->lock); 1332 1333 spin_lock(&nhslot2->lock); 1334 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1335 &nhslot2->head); 1336 nhslot2->count++; 1337 spin_unlock(&nhslot2->lock); 1338 1339 spin_unlock_bh(&hslot->lock); 1340 } 1341 } 1342 } 1343 EXPORT_SYMBOL(udp_lib_rehash); 1344 1345 static void udp_v4_rehash(struct sock *sk) 1346 { 1347 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1348 inet_sk(sk)->inet_rcv_saddr, 1349 inet_sk(sk)->inet_num); 1350 udp_lib_rehash(sk, new_hash); 1351 } 1352 1353 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1354 { 1355 int rc; 1356 1357 if (inet_sk(sk)->inet_daddr) 1358 sock_rps_save_rxhash(sk, skb); 1359 1360 rc = ip_queue_rcv_skb(sk, skb); 1361 if (rc < 0) { 1362 int is_udplite = IS_UDPLITE(sk); 1363 1364 /* Note that an ENOMEM error is charged twice */ 1365 if (rc == -ENOMEM) 1366 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1367 is_udplite); 1368 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1369 kfree_skb(skb); 1370 trace_udp_fail_queue_rcv_skb(rc, sk); 1371 return -1; 1372 } 1373 1374 return 0; 1375 1376 } 1377 1378 /* returns: 1379 * -1: error 1380 * 0: success 1381 * >0: "udp encap" protocol resubmission 1382 * 1383 * Note that in the success and error cases, the skb is assumed to 1384 * have either been requeued or freed. 1385 */ 1386 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1387 { 1388 struct udp_sock *up = udp_sk(sk); 1389 int rc; 1390 int is_udplite = IS_UDPLITE(sk); 1391 1392 /* 1393 * Charge it to the socket, dropping if the queue is full. 1394 */ 1395 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1396 goto drop; 1397 nf_reset(skb); 1398 1399 if (up->encap_type) { 1400 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1401 1402 /* 1403 * This is an encapsulation socket so pass the skb to 1404 * the socket's udp_encap_rcv() hook. Otherwise, just 1405 * fall through and pass this up the UDP socket. 1406 * up->encap_rcv() returns the following value: 1407 * =0 if skb was successfully passed to the encap 1408 * handler or was discarded by it. 1409 * >0 if skb should be passed on to UDP. 1410 * <0 if skb should be resubmitted as proto -N 1411 */ 1412 1413 /* if we're overly short, let UDP handle it */ 1414 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1415 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) { 1416 int ret; 1417 1418 ret = encap_rcv(sk, skb); 1419 if (ret <= 0) { 1420 UDP_INC_STATS_BH(sock_net(sk), 1421 UDP_MIB_INDATAGRAMS, 1422 is_udplite); 1423 return -ret; 1424 } 1425 } 1426 1427 /* FALLTHROUGH -- it's a UDP Packet */ 1428 } 1429 1430 /* 1431 * UDP-Lite specific tests, ignored on UDP sockets 1432 */ 1433 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1434 1435 /* 1436 * MIB statistics other than incrementing the error count are 1437 * disabled for the following two types of errors: these depend 1438 * on the application settings, not on the functioning of the 1439 * protocol stack as such. 1440 * 1441 * RFC 3828 here recommends (sec 3.3): "There should also be a 1442 * way ... to ... at least let the receiving application block 1443 * delivery of packets with coverage values less than a value 1444 * provided by the application." 1445 */ 1446 if (up->pcrlen == 0) { /* full coverage was set */ 1447 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1448 "%d while full coverage %d requested\n", 1449 UDP_SKB_CB(skb)->cscov, skb->len); 1450 goto drop; 1451 } 1452 /* The next case involves violating the min. coverage requested 1453 * by the receiver. This is subtle: if receiver wants x and x is 1454 * greater than the buffersize/MTU then receiver will complain 1455 * that it wants x while sender emits packets of smaller size y. 1456 * Therefore the above ...()->partial_cov statement is essential. 1457 */ 1458 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1459 LIMIT_NETDEBUG(KERN_WARNING 1460 "UDPLITE: coverage %d too small, need min %d\n", 1461 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1462 goto drop; 1463 } 1464 } 1465 1466 if (rcu_access_pointer(sk->sk_filter) && 1467 udp_lib_checksum_complete(skb)) 1468 goto drop; 1469 1470 1471 if (sk_rcvqueues_full(sk, skb)) 1472 goto drop; 1473 1474 rc = 0; 1475 1476 bh_lock_sock(sk); 1477 if (!sock_owned_by_user(sk)) 1478 rc = __udp_queue_rcv_skb(sk, skb); 1479 else if (sk_add_backlog(sk, skb)) { 1480 bh_unlock_sock(sk); 1481 goto drop; 1482 } 1483 bh_unlock_sock(sk); 1484 1485 return rc; 1486 1487 drop: 1488 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1489 atomic_inc(&sk->sk_drops); 1490 kfree_skb(skb); 1491 return -1; 1492 } 1493 1494 1495 static void flush_stack(struct sock **stack, unsigned int count, 1496 struct sk_buff *skb, unsigned int final) 1497 { 1498 unsigned int i; 1499 struct sk_buff *skb1 = NULL; 1500 struct sock *sk; 1501 1502 for (i = 0; i < count; i++) { 1503 sk = stack[i]; 1504 if (likely(skb1 == NULL)) 1505 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1506 1507 if (!skb1) { 1508 atomic_inc(&sk->sk_drops); 1509 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1510 IS_UDPLITE(sk)); 1511 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1512 IS_UDPLITE(sk)); 1513 } 1514 1515 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1516 skb1 = NULL; 1517 } 1518 if (unlikely(skb1)) 1519 kfree_skb(skb1); 1520 } 1521 1522 /* 1523 * Multicasts and broadcasts go to each listener. 1524 * 1525 * Note: called only from the BH handler context. 1526 */ 1527 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1528 struct udphdr *uh, 1529 __be32 saddr, __be32 daddr, 1530 struct udp_table *udptable) 1531 { 1532 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1533 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1534 int dif; 1535 unsigned int i, count = 0; 1536 1537 spin_lock(&hslot->lock); 1538 sk = sk_nulls_head(&hslot->head); 1539 dif = skb->dev->ifindex; 1540 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1541 while (sk) { 1542 stack[count++] = sk; 1543 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1544 daddr, uh->source, saddr, dif); 1545 if (unlikely(count == ARRAY_SIZE(stack))) { 1546 if (!sk) 1547 break; 1548 flush_stack(stack, count, skb, ~0); 1549 count = 0; 1550 } 1551 } 1552 /* 1553 * before releasing chain lock, we must take a reference on sockets 1554 */ 1555 for (i = 0; i < count; i++) 1556 sock_hold(stack[i]); 1557 1558 spin_unlock(&hslot->lock); 1559 1560 /* 1561 * do the slow work with no lock held 1562 */ 1563 if (count) { 1564 flush_stack(stack, count, skb, count - 1); 1565 1566 for (i = 0; i < count; i++) 1567 sock_put(stack[i]); 1568 } else { 1569 kfree_skb(skb); 1570 } 1571 return 0; 1572 } 1573 1574 /* Initialize UDP checksum. If exited with zero value (success), 1575 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1576 * Otherwise, csum completion requires chacksumming packet body, 1577 * including udp header and folding it to skb->csum. 1578 */ 1579 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1580 int proto) 1581 { 1582 const struct iphdr *iph; 1583 int err; 1584 1585 UDP_SKB_CB(skb)->partial_cov = 0; 1586 UDP_SKB_CB(skb)->cscov = skb->len; 1587 1588 if (proto == IPPROTO_UDPLITE) { 1589 err = udplite_checksum_init(skb, uh); 1590 if (err) 1591 return err; 1592 } 1593 1594 iph = ip_hdr(skb); 1595 if (uh->check == 0) { 1596 skb->ip_summed = CHECKSUM_UNNECESSARY; 1597 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1598 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1599 proto, skb->csum)) 1600 skb->ip_summed = CHECKSUM_UNNECESSARY; 1601 } 1602 if (!skb_csum_unnecessary(skb)) 1603 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1604 skb->len, proto, 0); 1605 /* Probably, we should checksum udp header (it should be in cache 1606 * in any case) and data in tiny packets (< rx copybreak). 1607 */ 1608 1609 return 0; 1610 } 1611 1612 /* 1613 * All we need to do is get the socket, and then do a checksum. 1614 */ 1615 1616 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1617 int proto) 1618 { 1619 struct sock *sk; 1620 struct udphdr *uh; 1621 unsigned short ulen; 1622 struct rtable *rt = skb_rtable(skb); 1623 __be32 saddr, daddr; 1624 struct net *net = dev_net(skb->dev); 1625 1626 /* 1627 * Validate the packet. 1628 */ 1629 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1630 goto drop; /* No space for header. */ 1631 1632 uh = udp_hdr(skb); 1633 ulen = ntohs(uh->len); 1634 saddr = ip_hdr(skb)->saddr; 1635 daddr = ip_hdr(skb)->daddr; 1636 1637 if (ulen > skb->len) 1638 goto short_packet; 1639 1640 if (proto == IPPROTO_UDP) { 1641 /* UDP validates ulen. */ 1642 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1643 goto short_packet; 1644 uh = udp_hdr(skb); 1645 } 1646 1647 if (udp4_csum_init(skb, uh, proto)) 1648 goto csum_error; 1649 1650 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1651 return __udp4_lib_mcast_deliver(net, skb, uh, 1652 saddr, daddr, udptable); 1653 1654 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1655 1656 if (sk != NULL) { 1657 int ret = udp_queue_rcv_skb(sk, skb); 1658 sock_put(sk); 1659 1660 /* a return value > 0 means to resubmit the input, but 1661 * it wants the return to be -protocol, or 0 1662 */ 1663 if (ret > 0) 1664 return -ret; 1665 return 0; 1666 } 1667 1668 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1669 goto drop; 1670 nf_reset(skb); 1671 1672 /* No socket. Drop packet silently, if checksum is wrong */ 1673 if (udp_lib_checksum_complete(skb)) 1674 goto csum_error; 1675 1676 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1677 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1678 1679 /* 1680 * Hmm. We got an UDP packet to a port to which we 1681 * don't wanna listen. Ignore it. 1682 */ 1683 kfree_skb(skb); 1684 return 0; 1685 1686 short_packet: 1687 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1688 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1689 &saddr, 1690 ntohs(uh->source), 1691 ulen, 1692 skb->len, 1693 &daddr, 1694 ntohs(uh->dest)); 1695 goto drop; 1696 1697 csum_error: 1698 /* 1699 * RFC1122: OK. Discards the bad packet silently (as far as 1700 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1701 */ 1702 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1703 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1704 &saddr, 1705 ntohs(uh->source), 1706 &daddr, 1707 ntohs(uh->dest), 1708 ulen); 1709 drop: 1710 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1711 kfree_skb(skb); 1712 return 0; 1713 } 1714 1715 int udp_rcv(struct sk_buff *skb) 1716 { 1717 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1718 } 1719 1720 void udp_destroy_sock(struct sock *sk) 1721 { 1722 bool slow = lock_sock_fast(sk); 1723 udp_flush_pending_frames(sk); 1724 unlock_sock_fast(sk, slow); 1725 } 1726 1727 /* 1728 * Socket option code for UDP 1729 */ 1730 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1731 char __user *optval, unsigned int optlen, 1732 int (*push_pending_frames)(struct sock *)) 1733 { 1734 struct udp_sock *up = udp_sk(sk); 1735 int val; 1736 int err = 0; 1737 int is_udplite = IS_UDPLITE(sk); 1738 1739 if (optlen < sizeof(int)) 1740 return -EINVAL; 1741 1742 if (get_user(val, (int __user *)optval)) 1743 return -EFAULT; 1744 1745 switch (optname) { 1746 case UDP_CORK: 1747 if (val != 0) { 1748 up->corkflag = 1; 1749 } else { 1750 up->corkflag = 0; 1751 lock_sock(sk); 1752 (*push_pending_frames)(sk); 1753 release_sock(sk); 1754 } 1755 break; 1756 1757 case UDP_ENCAP: 1758 switch (val) { 1759 case 0: 1760 case UDP_ENCAP_ESPINUDP: 1761 case UDP_ENCAP_ESPINUDP_NON_IKE: 1762 up->encap_rcv = xfrm4_udp_encap_rcv; 1763 /* FALLTHROUGH */ 1764 case UDP_ENCAP_L2TPINUDP: 1765 up->encap_type = val; 1766 break; 1767 default: 1768 err = -ENOPROTOOPT; 1769 break; 1770 } 1771 break; 1772 1773 /* 1774 * UDP-Lite's partial checksum coverage (RFC 3828). 1775 */ 1776 /* The sender sets actual checksum coverage length via this option. 1777 * The case coverage > packet length is handled by send module. */ 1778 case UDPLITE_SEND_CSCOV: 1779 if (!is_udplite) /* Disable the option on UDP sockets */ 1780 return -ENOPROTOOPT; 1781 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1782 val = 8; 1783 else if (val > USHRT_MAX) 1784 val = USHRT_MAX; 1785 up->pcslen = val; 1786 up->pcflag |= UDPLITE_SEND_CC; 1787 break; 1788 1789 /* The receiver specifies a minimum checksum coverage value. To make 1790 * sense, this should be set to at least 8 (as done below). If zero is 1791 * used, this again means full checksum coverage. */ 1792 case UDPLITE_RECV_CSCOV: 1793 if (!is_udplite) /* Disable the option on UDP sockets */ 1794 return -ENOPROTOOPT; 1795 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1796 val = 8; 1797 else if (val > USHRT_MAX) 1798 val = USHRT_MAX; 1799 up->pcrlen = val; 1800 up->pcflag |= UDPLITE_RECV_CC; 1801 break; 1802 1803 default: 1804 err = -ENOPROTOOPT; 1805 break; 1806 } 1807 1808 return err; 1809 } 1810 EXPORT_SYMBOL(udp_lib_setsockopt); 1811 1812 int udp_setsockopt(struct sock *sk, int level, int optname, 1813 char __user *optval, unsigned int optlen) 1814 { 1815 if (level == SOL_UDP || level == SOL_UDPLITE) 1816 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1817 udp_push_pending_frames); 1818 return ip_setsockopt(sk, level, optname, optval, optlen); 1819 } 1820 1821 #ifdef CONFIG_COMPAT 1822 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1823 char __user *optval, unsigned int optlen) 1824 { 1825 if (level == SOL_UDP || level == SOL_UDPLITE) 1826 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1827 udp_push_pending_frames); 1828 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1829 } 1830 #endif 1831 1832 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1833 char __user *optval, int __user *optlen) 1834 { 1835 struct udp_sock *up = udp_sk(sk); 1836 int val, len; 1837 1838 if (get_user(len, optlen)) 1839 return -EFAULT; 1840 1841 len = min_t(unsigned int, len, sizeof(int)); 1842 1843 if (len < 0) 1844 return -EINVAL; 1845 1846 switch (optname) { 1847 case UDP_CORK: 1848 val = up->corkflag; 1849 break; 1850 1851 case UDP_ENCAP: 1852 val = up->encap_type; 1853 break; 1854 1855 /* The following two cannot be changed on UDP sockets, the return is 1856 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1857 case UDPLITE_SEND_CSCOV: 1858 val = up->pcslen; 1859 break; 1860 1861 case UDPLITE_RECV_CSCOV: 1862 val = up->pcrlen; 1863 break; 1864 1865 default: 1866 return -ENOPROTOOPT; 1867 } 1868 1869 if (put_user(len, optlen)) 1870 return -EFAULT; 1871 if (copy_to_user(optval, &val, len)) 1872 return -EFAULT; 1873 return 0; 1874 } 1875 EXPORT_SYMBOL(udp_lib_getsockopt); 1876 1877 int udp_getsockopt(struct sock *sk, int level, int optname, 1878 char __user *optval, int __user *optlen) 1879 { 1880 if (level == SOL_UDP || level == SOL_UDPLITE) 1881 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1882 return ip_getsockopt(sk, level, optname, optval, optlen); 1883 } 1884 1885 #ifdef CONFIG_COMPAT 1886 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1887 char __user *optval, int __user *optlen) 1888 { 1889 if (level == SOL_UDP || level == SOL_UDPLITE) 1890 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1891 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1892 } 1893 #endif 1894 /** 1895 * udp_poll - wait for a UDP event. 1896 * @file - file struct 1897 * @sock - socket 1898 * @wait - poll table 1899 * 1900 * This is same as datagram poll, except for the special case of 1901 * blocking sockets. If application is using a blocking fd 1902 * and a packet with checksum error is in the queue; 1903 * then it could get return from select indicating data available 1904 * but then block when reading it. Add special case code 1905 * to work around these arguably broken applications. 1906 */ 1907 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1908 { 1909 unsigned int mask = datagram_poll(file, sock, wait); 1910 struct sock *sk = sock->sk; 1911 1912 /* Check for false positives due to checksum errors */ 1913 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 1914 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 1915 mask &= ~(POLLIN | POLLRDNORM); 1916 1917 return mask; 1918 1919 } 1920 EXPORT_SYMBOL(udp_poll); 1921 1922 struct proto udp_prot = { 1923 .name = "UDP", 1924 .owner = THIS_MODULE, 1925 .close = udp_lib_close, 1926 .connect = ip4_datagram_connect, 1927 .disconnect = udp_disconnect, 1928 .ioctl = udp_ioctl, 1929 .destroy = udp_destroy_sock, 1930 .setsockopt = udp_setsockopt, 1931 .getsockopt = udp_getsockopt, 1932 .sendmsg = udp_sendmsg, 1933 .recvmsg = udp_recvmsg, 1934 .sendpage = udp_sendpage, 1935 .backlog_rcv = __udp_queue_rcv_skb, 1936 .hash = udp_lib_hash, 1937 .unhash = udp_lib_unhash, 1938 .rehash = udp_v4_rehash, 1939 .get_port = udp_v4_get_port, 1940 .memory_allocated = &udp_memory_allocated, 1941 .sysctl_mem = sysctl_udp_mem, 1942 .sysctl_wmem = &sysctl_udp_wmem_min, 1943 .sysctl_rmem = &sysctl_udp_rmem_min, 1944 .obj_size = sizeof(struct udp_sock), 1945 .slab_flags = SLAB_DESTROY_BY_RCU, 1946 .h.udp_table = &udp_table, 1947 #ifdef CONFIG_COMPAT 1948 .compat_setsockopt = compat_udp_setsockopt, 1949 .compat_getsockopt = compat_udp_getsockopt, 1950 #endif 1951 .clear_sk = sk_prot_clear_portaddr_nulls, 1952 }; 1953 EXPORT_SYMBOL(udp_prot); 1954 1955 /* ------------------------------------------------------------------------ */ 1956 #ifdef CONFIG_PROC_FS 1957 1958 static struct sock *udp_get_first(struct seq_file *seq, int start) 1959 { 1960 struct sock *sk; 1961 struct udp_iter_state *state = seq->private; 1962 struct net *net = seq_file_net(seq); 1963 1964 for (state->bucket = start; state->bucket <= state->udp_table->mask; 1965 ++state->bucket) { 1966 struct hlist_nulls_node *node; 1967 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 1968 1969 if (hlist_nulls_empty(&hslot->head)) 1970 continue; 1971 1972 spin_lock_bh(&hslot->lock); 1973 sk_nulls_for_each(sk, node, &hslot->head) { 1974 if (!net_eq(sock_net(sk), net)) 1975 continue; 1976 if (sk->sk_family == state->family) 1977 goto found; 1978 } 1979 spin_unlock_bh(&hslot->lock); 1980 } 1981 sk = NULL; 1982 found: 1983 return sk; 1984 } 1985 1986 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1987 { 1988 struct udp_iter_state *state = seq->private; 1989 struct net *net = seq_file_net(seq); 1990 1991 do { 1992 sk = sk_nulls_next(sk); 1993 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1994 1995 if (!sk) { 1996 if (state->bucket <= state->udp_table->mask) 1997 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1998 return udp_get_first(seq, state->bucket + 1); 1999 } 2000 return sk; 2001 } 2002 2003 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2004 { 2005 struct sock *sk = udp_get_first(seq, 0); 2006 2007 if (sk) 2008 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2009 --pos; 2010 return pos ? NULL : sk; 2011 } 2012 2013 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2014 { 2015 struct udp_iter_state *state = seq->private; 2016 state->bucket = MAX_UDP_PORTS; 2017 2018 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2019 } 2020 2021 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2022 { 2023 struct sock *sk; 2024 2025 if (v == SEQ_START_TOKEN) 2026 sk = udp_get_idx(seq, 0); 2027 else 2028 sk = udp_get_next(seq, v); 2029 2030 ++*pos; 2031 return sk; 2032 } 2033 2034 static void udp_seq_stop(struct seq_file *seq, void *v) 2035 { 2036 struct udp_iter_state *state = seq->private; 2037 2038 if (state->bucket <= state->udp_table->mask) 2039 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2040 } 2041 2042 int udp_seq_open(struct inode *inode, struct file *file) 2043 { 2044 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 2045 struct udp_iter_state *s; 2046 int err; 2047 2048 err = seq_open_net(inode, file, &afinfo->seq_ops, 2049 sizeof(struct udp_iter_state)); 2050 if (err < 0) 2051 return err; 2052 2053 s = ((struct seq_file *)file->private_data)->private; 2054 s->family = afinfo->family; 2055 s->udp_table = afinfo->udp_table; 2056 return err; 2057 } 2058 EXPORT_SYMBOL(udp_seq_open); 2059 2060 /* ------------------------------------------------------------------------ */ 2061 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2062 { 2063 struct proc_dir_entry *p; 2064 int rc = 0; 2065 2066 afinfo->seq_ops.start = udp_seq_start; 2067 afinfo->seq_ops.next = udp_seq_next; 2068 afinfo->seq_ops.stop = udp_seq_stop; 2069 2070 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2071 afinfo->seq_fops, afinfo); 2072 if (!p) 2073 rc = -ENOMEM; 2074 return rc; 2075 } 2076 EXPORT_SYMBOL(udp_proc_register); 2077 2078 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2079 { 2080 proc_net_remove(net, afinfo->name); 2081 } 2082 EXPORT_SYMBOL(udp_proc_unregister); 2083 2084 /* ------------------------------------------------------------------------ */ 2085 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2086 int bucket, int *len) 2087 { 2088 struct inet_sock *inet = inet_sk(sp); 2089 __be32 dest = inet->inet_daddr; 2090 __be32 src = inet->inet_rcv_saddr; 2091 __u16 destp = ntohs(inet->inet_dport); 2092 __u16 srcp = ntohs(inet->inet_sport); 2093 2094 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2095 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n", 2096 bucket, src, srcp, dest, destp, sp->sk_state, 2097 sk_wmem_alloc_get(sp), 2098 sk_rmem_alloc_get(sp), 2099 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 2100 atomic_read(&sp->sk_refcnt), sp, 2101 atomic_read(&sp->sk_drops), len); 2102 } 2103 2104 int udp4_seq_show(struct seq_file *seq, void *v) 2105 { 2106 if (v == SEQ_START_TOKEN) 2107 seq_printf(seq, "%-127s\n", 2108 " sl local_address rem_address st tx_queue " 2109 "rx_queue tr tm->when retrnsmt uid timeout " 2110 "inode ref pointer drops"); 2111 else { 2112 struct udp_iter_state *state = seq->private; 2113 int len; 2114 2115 udp4_format_sock(v, seq, state->bucket, &len); 2116 seq_printf(seq, "%*s\n", 127 - len, ""); 2117 } 2118 return 0; 2119 } 2120 2121 static const struct file_operations udp_afinfo_seq_fops = { 2122 .owner = THIS_MODULE, 2123 .open = udp_seq_open, 2124 .read = seq_read, 2125 .llseek = seq_lseek, 2126 .release = seq_release_net 2127 }; 2128 2129 /* ------------------------------------------------------------------------ */ 2130 static struct udp_seq_afinfo udp4_seq_afinfo = { 2131 .name = "udp", 2132 .family = AF_INET, 2133 .udp_table = &udp_table, 2134 .seq_fops = &udp_afinfo_seq_fops, 2135 .seq_ops = { 2136 .show = udp4_seq_show, 2137 }, 2138 }; 2139 2140 static int __net_init udp4_proc_init_net(struct net *net) 2141 { 2142 return udp_proc_register(net, &udp4_seq_afinfo); 2143 } 2144 2145 static void __net_exit udp4_proc_exit_net(struct net *net) 2146 { 2147 udp_proc_unregister(net, &udp4_seq_afinfo); 2148 } 2149 2150 static struct pernet_operations udp4_net_ops = { 2151 .init = udp4_proc_init_net, 2152 .exit = udp4_proc_exit_net, 2153 }; 2154 2155 int __init udp4_proc_init(void) 2156 { 2157 return register_pernet_subsys(&udp4_net_ops); 2158 } 2159 2160 void udp4_proc_exit(void) 2161 { 2162 unregister_pernet_subsys(&udp4_net_ops); 2163 } 2164 #endif /* CONFIG_PROC_FS */ 2165 2166 static __initdata unsigned long uhash_entries; 2167 static int __init set_uhash_entries(char *str) 2168 { 2169 if (!str) 2170 return 0; 2171 uhash_entries = simple_strtoul(str, &str, 0); 2172 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2173 uhash_entries = UDP_HTABLE_SIZE_MIN; 2174 return 1; 2175 } 2176 __setup("uhash_entries=", set_uhash_entries); 2177 2178 void __init udp_table_init(struct udp_table *table, const char *name) 2179 { 2180 unsigned int i; 2181 2182 if (!CONFIG_BASE_SMALL) 2183 table->hash = alloc_large_system_hash(name, 2184 2 * sizeof(struct udp_hslot), 2185 uhash_entries, 2186 21, /* one slot per 2 MB */ 2187 0, 2188 &table->log, 2189 &table->mask, 2190 64 * 1024); 2191 /* 2192 * Make sure hash table has the minimum size 2193 */ 2194 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) { 2195 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN * 2196 2 * sizeof(struct udp_hslot), GFP_KERNEL); 2197 if (!table->hash) 2198 panic(name); 2199 table->log = ilog2(UDP_HTABLE_SIZE_MIN); 2200 table->mask = UDP_HTABLE_SIZE_MIN - 1; 2201 } 2202 table->hash2 = table->hash + (table->mask + 1); 2203 for (i = 0; i <= table->mask; i++) { 2204 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2205 table->hash[i].count = 0; 2206 spin_lock_init(&table->hash[i].lock); 2207 } 2208 for (i = 0; i <= table->mask; i++) { 2209 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2210 table->hash2[i].count = 0; 2211 spin_lock_init(&table->hash2[i].lock); 2212 } 2213 } 2214 2215 void __init udp_init(void) 2216 { 2217 unsigned long limit; 2218 2219 udp_table_init(&udp_table, "UDP"); 2220 limit = nr_free_buffer_pages() / 8; 2221 limit = max(limit, 128UL); 2222 sysctl_udp_mem[0] = limit / 4 * 3; 2223 sysctl_udp_mem[1] = limit; 2224 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2225 2226 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2227 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2228 } 2229 2230 int udp4_ufo_send_check(struct sk_buff *skb) 2231 { 2232 const struct iphdr *iph; 2233 struct udphdr *uh; 2234 2235 if (!pskb_may_pull(skb, sizeof(*uh))) 2236 return -EINVAL; 2237 2238 iph = ip_hdr(skb); 2239 uh = udp_hdr(skb); 2240 2241 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 2242 IPPROTO_UDP, 0); 2243 skb->csum_start = skb_transport_header(skb) - skb->head; 2244 skb->csum_offset = offsetof(struct udphdr, check); 2245 skb->ip_summed = CHECKSUM_PARTIAL; 2246 return 0; 2247 } 2248 2249 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, u32 features) 2250 { 2251 struct sk_buff *segs = ERR_PTR(-EINVAL); 2252 unsigned int mss; 2253 int offset; 2254 __wsum csum; 2255 2256 mss = skb_shinfo(skb)->gso_size; 2257 if (unlikely(skb->len <= mss)) 2258 goto out; 2259 2260 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 2261 /* Packet is from an untrusted source, reset gso_segs. */ 2262 int type = skb_shinfo(skb)->gso_type; 2263 2264 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || 2265 !(type & (SKB_GSO_UDP)))) 2266 goto out; 2267 2268 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 2269 2270 segs = NULL; 2271 goto out; 2272 } 2273 2274 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot 2275 * do checksum of UDP packets sent as multiple IP fragments. 2276 */ 2277 offset = skb_checksum_start_offset(skb); 2278 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2279 offset += skb->csum_offset; 2280 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2281 skb->ip_summed = CHECKSUM_NONE; 2282 2283 /* Fragment the skb. IP headers of the fragments are updated in 2284 * inet_gso_segment() 2285 */ 2286 segs = skb_segment(skb, features); 2287 out: 2288 return segs; 2289 } 2290 2291