11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * INET An implementation of the TCP/IP protocol suite for the LINUX 31da177e4SLinus Torvalds * operating system. INET is implemented using the BSD Socket 41da177e4SLinus Torvalds * interface as the means of communication with the user level. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * The User Datagram Protocol (UDP). 71da177e4SLinus Torvalds * 802c30a84SJesper Juhl * Authors: Ross Biro 91da177e4SLinus Torvalds * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 101da177e4SLinus Torvalds * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 11113aa838SAlan Cox * Alan Cox, <alan@lxorguk.ukuu.org.uk> 121da177e4SLinus Torvalds * Hirokazu Takahashi, <taka@valinux.co.jp> 131da177e4SLinus Torvalds * 141da177e4SLinus Torvalds * Fixes: 151da177e4SLinus Torvalds * Alan Cox : verify_area() calls 161da177e4SLinus Torvalds * Alan Cox : stopped close while in use off icmp 171da177e4SLinus Torvalds * messages. Not a fix but a botch that 181da177e4SLinus Torvalds * for udp at least is 'valid'. 191da177e4SLinus Torvalds * Alan Cox : Fixed icmp handling properly 201da177e4SLinus Torvalds * Alan Cox : Correct error for oversized datagrams 211da177e4SLinus Torvalds * Alan Cox : Tidied select() semantics. 221da177e4SLinus Torvalds * Alan Cox : udp_err() fixed properly, also now 231da177e4SLinus Torvalds * select and read wake correctly on errors 241da177e4SLinus Torvalds * Alan Cox : udp_send verify_area moved to avoid mem leak 251da177e4SLinus Torvalds * Alan Cox : UDP can count its memory 261da177e4SLinus Torvalds * Alan Cox : send to an unknown connection causes 271da177e4SLinus Torvalds * an ECONNREFUSED off the icmp, but 281da177e4SLinus Torvalds * does NOT close. 291da177e4SLinus Torvalds * Alan Cox : Switched to new sk_buff handlers. No more backlog! 301da177e4SLinus Torvalds * Alan Cox : Using generic datagram code. Even smaller and the PEEK 311da177e4SLinus Torvalds * bug no longer crashes it. 321da177e4SLinus Torvalds * Fred Van Kempen : Net2e support for sk->broadcast. 331da177e4SLinus Torvalds * Alan Cox : Uses skb_free_datagram 341da177e4SLinus Torvalds * Alan Cox : Added get/set sockopt support. 351da177e4SLinus Torvalds * Alan Cox : Broadcasting without option set returns EACCES. 361da177e4SLinus Torvalds * Alan Cox : No wakeup calls. Instead we now use the callbacks. 371da177e4SLinus Torvalds * Alan Cox : Use ip_tos and ip_ttl 381da177e4SLinus Torvalds * Alan Cox : SNMP Mibs 391da177e4SLinus Torvalds * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 401da177e4SLinus Torvalds * Matt Dillon : UDP length checks. 411da177e4SLinus Torvalds * Alan Cox : Smarter af_inet used properly. 421da177e4SLinus Torvalds * Alan Cox : Use new kernel side addressing. 431da177e4SLinus Torvalds * Alan Cox : Incorrect return on truncated datagram receive. 441da177e4SLinus Torvalds * Arnt Gulbrandsen : New udp_send and stuff 451da177e4SLinus Torvalds * Alan Cox : Cache last socket 461da177e4SLinus Torvalds * Alan Cox : Route cache 471da177e4SLinus Torvalds * Jon Peatfield : Minor efficiency fix to sendto(). 481da177e4SLinus Torvalds * Mike Shaver : RFC1122 checks. 491da177e4SLinus Torvalds * Alan Cox : Nonblocking error fix. 501da177e4SLinus Torvalds * Willy Konynenberg : Transparent proxying support. 511da177e4SLinus Torvalds * Mike McLagan : Routing by source 521da177e4SLinus Torvalds * David S. Miller : New socket lookup architecture. 531da177e4SLinus Torvalds * Last socket cache retained as it 541da177e4SLinus Torvalds * does have a high hit rate. 551da177e4SLinus Torvalds * Olaf Kirch : Don't linearise iovec on sendmsg. 561da177e4SLinus Torvalds * Andi Kleen : Some cleanups, cache destination entry 571da177e4SLinus Torvalds * for connect. 581da177e4SLinus Torvalds * Vitaly E. Lavrov : Transparent proxy revived after year coma. 591da177e4SLinus Torvalds * Melvin Smith : Check msg_name not msg_namelen in sendto(), 601da177e4SLinus Torvalds * return ENOTCONN for unconnected sockets (POSIX) 611da177e4SLinus Torvalds * Janos Farkas : don't deliver multi/broadcasts to a different 621da177e4SLinus Torvalds * bound-to-device socket 631da177e4SLinus Torvalds * Hirokazu Takahashi : HW checksumming for outgoing UDP 641da177e4SLinus Torvalds * datagrams. 651da177e4SLinus Torvalds * Hirokazu Takahashi : sendfile() on UDP works now. 661da177e4SLinus Torvalds * Arnaldo C. Melo : convert /proc/net/udp to seq_file 671da177e4SLinus Torvalds * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 681da177e4SLinus Torvalds * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 691da177e4SLinus Torvalds * a single port at the same time. 701da177e4SLinus Torvalds * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 71342f0234SJames Chapman * James Chapman : Add L2TP encapsulation type. 721da177e4SLinus Torvalds * 731da177e4SLinus Torvalds * 741da177e4SLinus Torvalds * This program is free software; you can redistribute it and/or 751da177e4SLinus Torvalds * modify it under the terms of the GNU General Public License 761da177e4SLinus Torvalds * as published by the Free Software Foundation; either version 771da177e4SLinus Torvalds * 2 of the License, or (at your option) any later version. 781da177e4SLinus Torvalds */ 791da177e4SLinus Torvalds 801da177e4SLinus Torvalds #include <asm/system.h> 811da177e4SLinus Torvalds #include <asm/uaccess.h> 821da177e4SLinus Torvalds #include <asm/ioctls.h> 8395766fffSHideo Aoki #include <linux/bootmem.h> 848203efb3SEric Dumazet #include <linux/highmem.h> 858203efb3SEric Dumazet #include <linux/swap.h> 861da177e4SLinus Torvalds #include <linux/types.h> 871da177e4SLinus Torvalds #include <linux/fcntl.h> 881da177e4SLinus Torvalds #include <linux/module.h> 891da177e4SLinus Torvalds #include <linux/socket.h> 901da177e4SLinus Torvalds #include <linux/sockios.h> 9114c85021SArnaldo Carvalho de Melo #include <linux/igmp.h> 921da177e4SLinus Torvalds #include <linux/in.h> 931da177e4SLinus Torvalds #include <linux/errno.h> 941da177e4SLinus Torvalds #include <linux/timer.h> 951da177e4SLinus Torvalds #include <linux/mm.h> 961da177e4SLinus Torvalds #include <linux/inet.h> 971da177e4SLinus Torvalds #include <linux/netdevice.h> 98c752f073SArnaldo Carvalho de Melo #include <net/tcp_states.h> 991da177e4SLinus Torvalds #include <linux/skbuff.h> 1001da177e4SLinus Torvalds #include <linux/proc_fs.h> 1011da177e4SLinus Torvalds #include <linux/seq_file.h> 102457c4cbcSEric W. Biederman #include <net/net_namespace.h> 1031da177e4SLinus Torvalds #include <net/icmp.h> 1041da177e4SLinus Torvalds #include <net/route.h> 1051da177e4SLinus Torvalds #include <net/checksum.h> 1061da177e4SLinus Torvalds #include <net/xfrm.h> 107ba4e58ecSGerrit Renker #include "udp_impl.h" 1081da177e4SLinus Torvalds 109f86dcc5aSEric Dumazet struct udp_table udp_table __read_mostly; 110645ca708SEric Dumazet EXPORT_SYMBOL(udp_table); 1111da177e4SLinus Torvalds 11295766fffSHideo Aoki int sysctl_udp_mem[3] __read_mostly; 11395766fffSHideo Aoki EXPORT_SYMBOL(sysctl_udp_mem); 114c482c568SEric Dumazet 115c482c568SEric Dumazet int sysctl_udp_rmem_min __read_mostly; 11695766fffSHideo Aoki EXPORT_SYMBOL(sysctl_udp_rmem_min); 117c482c568SEric Dumazet 118c482c568SEric Dumazet int sysctl_udp_wmem_min __read_mostly; 11995766fffSHideo Aoki EXPORT_SYMBOL(sysctl_udp_wmem_min); 12095766fffSHideo Aoki 12195766fffSHideo Aoki atomic_t udp_memory_allocated; 12295766fffSHideo Aoki EXPORT_SYMBOL(udp_memory_allocated); 12395766fffSHideo Aoki 124f86dcc5aSEric Dumazet #define MAX_UDP_PORTS 65536 125f86dcc5aSEric Dumazet #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 12698322f22SEric Dumazet 127f24d43c0SEric Dumazet static int udp_lib_lport_inuse(struct net *net, __u16 num, 128645ca708SEric Dumazet const struct udp_hslot *hslot, 12998322f22SEric Dumazet unsigned long *bitmap, 130f24d43c0SEric Dumazet struct sock *sk, 131f24d43c0SEric Dumazet int (*saddr_comp)(const struct sock *sk1, 132f86dcc5aSEric Dumazet const struct sock *sk2), 133f86dcc5aSEric Dumazet unsigned int log) 13425030a7fSGerrit Renker { 135f24d43c0SEric Dumazet struct sock *sk2; 13688ab1932SEric Dumazet struct hlist_nulls_node *node; 13725030a7fSGerrit Renker 13888ab1932SEric Dumazet sk_nulls_for_each(sk2, node, &hslot->head) 139f24d43c0SEric Dumazet if (net_eq(sock_net(sk2), net) && 140f24d43c0SEric Dumazet sk2 != sk && 141d4cada4aSEric Dumazet (bitmap || udp_sk(sk2)->udp_port_hash == num) && 142f24d43c0SEric Dumazet (!sk2->sk_reuse || !sk->sk_reuse) && 143f24d43c0SEric Dumazet (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if 144f24d43c0SEric Dumazet || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 14598322f22SEric Dumazet (*saddr_comp)(sk, sk2)) { 14698322f22SEric Dumazet if (bitmap) 147d4cada4aSEric Dumazet __set_bit(udp_sk(sk2)->udp_port_hash >> log, 148d4cada4aSEric Dumazet bitmap); 14998322f22SEric Dumazet else 150fc038410SDavid S. Miller return 1; 15198322f22SEric Dumazet } 15225030a7fSGerrit Renker return 0; 15325030a7fSGerrit Renker } 15425030a7fSGerrit Renker 15525030a7fSGerrit Renker /** 1566ba5a3c5SPavel Emelyanov * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 15725030a7fSGerrit Renker * 15825030a7fSGerrit Renker * @sk: socket struct in question 15925030a7fSGerrit Renker * @snum: port number to look up 160df2bc459SDavid S. Miller * @saddr_comp: AF-dependent comparison of bound local IP addresses 16125030a7fSGerrit Renker */ 1626ba5a3c5SPavel Emelyanov int udp_lib_get_port(struct sock *sk, unsigned short snum, 163df2bc459SDavid S. Miller int (*saddr_comp)(const struct sock *sk1, 164df2bc459SDavid S. Miller const struct sock *sk2)) 1651da177e4SLinus Torvalds { 166512615b6SEric Dumazet struct udp_hslot *hslot, *hslot2; 167645ca708SEric Dumazet struct udp_table *udptable = sk->sk_prot->h.udp_table; 16825030a7fSGerrit Renker int error = 1; 1693b1e0a65SYOSHIFUJI Hideaki struct net *net = sock_net(sk); 1701da177e4SLinus Torvalds 17132c1da70SStephen Hemminger if (!snum) { 1729088c560SEric Dumazet int low, high, remaining; 1739088c560SEric Dumazet unsigned rand; 17498322f22SEric Dumazet unsigned short first, last; 17598322f22SEric Dumazet DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 1761da177e4SLinus Torvalds 177227b60f5SStephen Hemminger inet_get_local_port_range(&low, &high); 178a25de534SAnton Arapov remaining = (high - low) + 1; 179227b60f5SStephen Hemminger 1809088c560SEric Dumazet rand = net_random(); 18198322f22SEric Dumazet first = (((u64)rand * remaining) >> 32) + low; 18298322f22SEric Dumazet /* 18398322f22SEric Dumazet * force rand to be an odd multiple of UDP_HTABLE_SIZE 18498322f22SEric Dumazet */ 185f86dcc5aSEric Dumazet rand = (rand | 1) * (udptable->mask + 1); 186f86dcc5aSEric Dumazet for (last = first + udptable->mask + 1; 187f86dcc5aSEric Dumazet first != last; 188f86dcc5aSEric Dumazet first++) { 189f86dcc5aSEric Dumazet hslot = udp_hashslot(udptable, net, first); 19098322f22SEric Dumazet bitmap_zero(bitmap, PORTS_PER_CHAIN); 191645ca708SEric Dumazet spin_lock_bh(&hslot->lock); 19298322f22SEric Dumazet udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 193f86dcc5aSEric Dumazet saddr_comp, udptable->log); 19498322f22SEric Dumazet 19598322f22SEric Dumazet snum = first; 19698322f22SEric Dumazet /* 19798322f22SEric Dumazet * Iterate on all possible values of snum for this hash. 19898322f22SEric Dumazet * Using steps of an odd multiple of UDP_HTABLE_SIZE 19998322f22SEric Dumazet * give us randomization and full range coverage. 20098322f22SEric Dumazet */ 2019088c560SEric Dumazet do { 20298322f22SEric Dumazet if (low <= snum && snum <= high && 203f86dcc5aSEric Dumazet !test_bit(snum >> udptable->log, bitmap)) 20498322f22SEric Dumazet goto found; 20598322f22SEric Dumazet snum += rand; 20698322f22SEric Dumazet } while (snum != first); 20798322f22SEric Dumazet spin_unlock_bh(&hslot->lock); 2089088c560SEric Dumazet } 20998322f22SEric Dumazet goto fail; 210645ca708SEric Dumazet } else { 211f86dcc5aSEric Dumazet hslot = udp_hashslot(udptable, net, snum); 212645ca708SEric Dumazet spin_lock_bh(&hslot->lock); 213f86dcc5aSEric Dumazet if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 214f86dcc5aSEric Dumazet saddr_comp, 0)) 215645ca708SEric Dumazet goto fail_unlock; 216645ca708SEric Dumazet } 21798322f22SEric Dumazet found: 218c720c7e8SEric Dumazet inet_sk(sk)->inet_num = snum; 219d4cada4aSEric Dumazet udp_sk(sk)->udp_port_hash = snum; 220d4cada4aSEric Dumazet udp_sk(sk)->udp_portaddr_hash ^= snum; 2211da177e4SLinus Torvalds if (sk_unhashed(sk)) { 22288ab1932SEric Dumazet sk_nulls_add_node_rcu(sk, &hslot->head); 223fdcc8aa9SEric Dumazet hslot->count++; 224c29a0bc4SPavel Emelyanov sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 225512615b6SEric Dumazet 226512615b6SEric Dumazet hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 227512615b6SEric Dumazet spin_lock(&hslot2->lock); 228512615b6SEric Dumazet hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 229512615b6SEric Dumazet &hslot2->head); 230512615b6SEric Dumazet hslot2->count++; 231512615b6SEric Dumazet spin_unlock(&hslot2->lock); 2321da177e4SLinus Torvalds } 23325030a7fSGerrit Renker error = 0; 234645ca708SEric Dumazet fail_unlock: 235645ca708SEric Dumazet spin_unlock_bh(&hslot->lock); 2361da177e4SLinus Torvalds fail: 23725030a7fSGerrit Renker return error; 2381da177e4SLinus Torvalds } 239c482c568SEric Dumazet EXPORT_SYMBOL(udp_lib_get_port); 2401da177e4SLinus Torvalds 241499923c7SVlad Yasevich static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 242db8dac20SDavid S. Miller { 243db8dac20SDavid S. Miller struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 244db8dac20SDavid S. Miller 245db8dac20SDavid S. Miller return (!ipv6_only_sock(sk2) && 246c720c7e8SEric Dumazet (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 247c720c7e8SEric Dumazet inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 248db8dac20SDavid S. Miller } 249db8dac20SDavid S. Miller 250d4cada4aSEric Dumazet static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr, 251d4cada4aSEric Dumazet unsigned int port) 252d4cada4aSEric Dumazet { 253d4cada4aSEric Dumazet return jhash_1word(saddr, net_hash_mix(net)) ^ port; 254d4cada4aSEric Dumazet } 255d4cada4aSEric Dumazet 2566ba5a3c5SPavel Emelyanov int udp_v4_get_port(struct sock *sk, unsigned short snum) 257db8dac20SDavid S. Miller { 258d4cada4aSEric Dumazet /* precompute partial secondary hash */ 259d4cada4aSEric Dumazet udp_sk(sk)->udp_portaddr_hash = 260d4cada4aSEric Dumazet udp4_portaddr_hash(sock_net(sk), 261d4cada4aSEric Dumazet inet_sk(sk)->inet_rcv_saddr, 262d4cada4aSEric Dumazet 0); 2636ba5a3c5SPavel Emelyanov return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal); 264db8dac20SDavid S. Miller } 265db8dac20SDavid S. Miller 266645ca708SEric Dumazet static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, 267645ca708SEric Dumazet unsigned short hnum, 268645ca708SEric Dumazet __be16 sport, __be32 daddr, __be16 dport, int dif) 269645ca708SEric Dumazet { 270645ca708SEric Dumazet int score = -1; 271645ca708SEric Dumazet 272d4cada4aSEric Dumazet if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 273645ca708SEric Dumazet !ipv6_only_sock(sk)) { 274645ca708SEric Dumazet struct inet_sock *inet = inet_sk(sk); 275645ca708SEric Dumazet 276645ca708SEric Dumazet score = (sk->sk_family == PF_INET ? 1 : 0); 277c720c7e8SEric Dumazet if (inet->inet_rcv_saddr) { 278c720c7e8SEric Dumazet if (inet->inet_rcv_saddr != daddr) 279645ca708SEric Dumazet return -1; 280645ca708SEric Dumazet score += 2; 281645ca708SEric Dumazet } 282c720c7e8SEric Dumazet if (inet->inet_daddr) { 283c720c7e8SEric Dumazet if (inet->inet_daddr != saddr) 284645ca708SEric Dumazet return -1; 285645ca708SEric Dumazet score += 2; 286645ca708SEric Dumazet } 287c720c7e8SEric Dumazet if (inet->inet_dport) { 288c720c7e8SEric Dumazet if (inet->inet_dport != sport) 289645ca708SEric Dumazet return -1; 290645ca708SEric Dumazet score += 2; 291645ca708SEric Dumazet } 292645ca708SEric Dumazet if (sk->sk_bound_dev_if) { 293645ca708SEric Dumazet if (sk->sk_bound_dev_if != dif) 294645ca708SEric Dumazet return -1; 295645ca708SEric Dumazet score += 2; 296645ca708SEric Dumazet } 297645ca708SEric Dumazet } 298645ca708SEric Dumazet return score; 299645ca708SEric Dumazet } 300645ca708SEric Dumazet 301*5051ebd2SEric Dumazet /* 302*5051ebd2SEric Dumazet * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 303*5051ebd2SEric Dumazet */ 304*5051ebd2SEric Dumazet #define SCORE2_MAX (1 + 2 + 2 + 2) 305*5051ebd2SEric Dumazet static inline int compute_score2(struct sock *sk, struct net *net, 306*5051ebd2SEric Dumazet __be32 saddr, __be16 sport, 307*5051ebd2SEric Dumazet __be32 daddr, unsigned int hnum, int dif) 308*5051ebd2SEric Dumazet { 309*5051ebd2SEric Dumazet int score = -1; 310*5051ebd2SEric Dumazet 311*5051ebd2SEric Dumazet if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) { 312*5051ebd2SEric Dumazet struct inet_sock *inet = inet_sk(sk); 313*5051ebd2SEric Dumazet 314*5051ebd2SEric Dumazet if (inet->inet_rcv_saddr != daddr) 315*5051ebd2SEric Dumazet return -1; 316*5051ebd2SEric Dumazet if (inet->inet_num != hnum) 317*5051ebd2SEric Dumazet return -1; 318*5051ebd2SEric Dumazet 319*5051ebd2SEric Dumazet score = (sk->sk_family == PF_INET ? 1 : 0); 320*5051ebd2SEric Dumazet if (inet->inet_daddr) { 321*5051ebd2SEric Dumazet if (inet->inet_daddr != saddr) 322*5051ebd2SEric Dumazet return -1; 323*5051ebd2SEric Dumazet score += 2; 324*5051ebd2SEric Dumazet } 325*5051ebd2SEric Dumazet if (inet->inet_dport) { 326*5051ebd2SEric Dumazet if (inet->inet_dport != sport) 327*5051ebd2SEric Dumazet return -1; 328*5051ebd2SEric Dumazet score += 2; 329*5051ebd2SEric Dumazet } 330*5051ebd2SEric Dumazet if (sk->sk_bound_dev_if) { 331*5051ebd2SEric Dumazet if (sk->sk_bound_dev_if != dif) 332*5051ebd2SEric Dumazet return -1; 333*5051ebd2SEric Dumazet score += 2; 334*5051ebd2SEric Dumazet } 335*5051ebd2SEric Dumazet } 336*5051ebd2SEric Dumazet return score; 337*5051ebd2SEric Dumazet } 338*5051ebd2SEric Dumazet 339*5051ebd2SEric Dumazet #define udp_portaddr_for_each_entry_rcu(__sk, node, list) \ 340*5051ebd2SEric Dumazet hlist_nulls_for_each_entry_rcu(__sk, node, list, __sk_common.skc_portaddr_node) 341*5051ebd2SEric Dumazet 342*5051ebd2SEric Dumazet /* called with read_rcu_lock() */ 343*5051ebd2SEric Dumazet static struct sock *udp4_lib_lookup2(struct net *net, 344*5051ebd2SEric Dumazet __be32 saddr, __be16 sport, 345*5051ebd2SEric Dumazet __be32 daddr, unsigned int hnum, int dif, 346*5051ebd2SEric Dumazet struct udp_hslot *hslot2, unsigned int slot2) 347*5051ebd2SEric Dumazet { 348*5051ebd2SEric Dumazet struct sock *sk, *result; 349*5051ebd2SEric Dumazet struct hlist_nulls_node *node; 350*5051ebd2SEric Dumazet int score, badness; 351*5051ebd2SEric Dumazet 352*5051ebd2SEric Dumazet begin: 353*5051ebd2SEric Dumazet result = NULL; 354*5051ebd2SEric Dumazet badness = -1; 355*5051ebd2SEric Dumazet udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 356*5051ebd2SEric Dumazet score = compute_score2(sk, net, saddr, sport, 357*5051ebd2SEric Dumazet daddr, hnum, dif); 358*5051ebd2SEric Dumazet if (score > badness) { 359*5051ebd2SEric Dumazet result = sk; 360*5051ebd2SEric Dumazet badness = score; 361*5051ebd2SEric Dumazet if (score == SCORE2_MAX) 362*5051ebd2SEric Dumazet goto exact_match; 363*5051ebd2SEric Dumazet } 364*5051ebd2SEric Dumazet } 365*5051ebd2SEric Dumazet /* 366*5051ebd2SEric Dumazet * if the nulls value we got at the end of this lookup is 367*5051ebd2SEric Dumazet * not the expected one, we must restart lookup. 368*5051ebd2SEric Dumazet * We probably met an item that was moved to another chain. 369*5051ebd2SEric Dumazet */ 370*5051ebd2SEric Dumazet if (get_nulls_value(node) != slot2) 371*5051ebd2SEric Dumazet goto begin; 372*5051ebd2SEric Dumazet 373*5051ebd2SEric Dumazet if (result) { 374*5051ebd2SEric Dumazet exact_match: 375*5051ebd2SEric Dumazet if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) 376*5051ebd2SEric Dumazet result = NULL; 377*5051ebd2SEric Dumazet else if (unlikely(compute_score2(result, net, saddr, sport, 378*5051ebd2SEric Dumazet daddr, hnum, dif) < badness)) { 379*5051ebd2SEric Dumazet sock_put(result); 380*5051ebd2SEric Dumazet goto begin; 381*5051ebd2SEric Dumazet } 382*5051ebd2SEric Dumazet } 383*5051ebd2SEric Dumazet return result; 384*5051ebd2SEric Dumazet } 385*5051ebd2SEric Dumazet 386db8dac20SDavid S. Miller /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 387db8dac20SDavid S. Miller * harder than this. -DaveM 388db8dac20SDavid S. Miller */ 389db8dac20SDavid S. Miller static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 390db8dac20SDavid S. Miller __be16 sport, __be32 daddr, __be16 dport, 391645ca708SEric Dumazet int dif, struct udp_table *udptable) 392db8dac20SDavid S. Miller { 393271b72c7SEric Dumazet struct sock *sk, *result; 39488ab1932SEric Dumazet struct hlist_nulls_node *node; 395db8dac20SDavid S. Miller unsigned short hnum = ntohs(dport); 396*5051ebd2SEric Dumazet unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 397*5051ebd2SEric Dumazet struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 398271b72c7SEric Dumazet int score, badness; 399db8dac20SDavid S. Miller 400271b72c7SEric Dumazet rcu_read_lock(); 401*5051ebd2SEric Dumazet if (hslot->count > 10) { 402*5051ebd2SEric Dumazet hash2 = udp4_portaddr_hash(net, daddr, hnum); 403*5051ebd2SEric Dumazet slot2 = hash2 & udptable->mask; 404*5051ebd2SEric Dumazet hslot2 = &udptable->hash2[slot2]; 405*5051ebd2SEric Dumazet if (hslot->count < hslot2->count) 406*5051ebd2SEric Dumazet goto begin; 407*5051ebd2SEric Dumazet 408*5051ebd2SEric Dumazet result = udp4_lib_lookup2(net, saddr, sport, 409*5051ebd2SEric Dumazet daddr, hnum, dif, 410*5051ebd2SEric Dumazet hslot2, slot2); 411*5051ebd2SEric Dumazet if (!result) { 412*5051ebd2SEric Dumazet hash2 = udp4_portaddr_hash(net, INADDR_ANY, hnum); 413*5051ebd2SEric Dumazet slot2 = hash2 & udptable->mask; 414*5051ebd2SEric Dumazet hslot2 = &udptable->hash2[slot2]; 415*5051ebd2SEric Dumazet if (hslot->count < hslot2->count) 416*5051ebd2SEric Dumazet goto begin; 417*5051ebd2SEric Dumazet 418*5051ebd2SEric Dumazet result = udp4_lib_lookup2(net, INADDR_ANY, sport, 419*5051ebd2SEric Dumazet daddr, hnum, dif, 420*5051ebd2SEric Dumazet hslot2, slot2); 421*5051ebd2SEric Dumazet } 422*5051ebd2SEric Dumazet rcu_read_unlock(); 423*5051ebd2SEric Dumazet return result; 424*5051ebd2SEric Dumazet } 425271b72c7SEric Dumazet begin: 426271b72c7SEric Dumazet result = NULL; 427271b72c7SEric Dumazet badness = -1; 42888ab1932SEric Dumazet sk_nulls_for_each_rcu(sk, node, &hslot->head) { 429645ca708SEric Dumazet score = compute_score(sk, net, saddr, hnum, sport, 430645ca708SEric Dumazet daddr, dport, dif); 431645ca708SEric Dumazet if (score > badness) { 432db8dac20SDavid S. Miller result = sk; 433db8dac20SDavid S. Miller badness = score; 434db8dac20SDavid S. Miller } 435db8dac20SDavid S. Miller } 43688ab1932SEric Dumazet /* 43788ab1932SEric Dumazet * if the nulls value we got at the end of this lookup is 43888ab1932SEric Dumazet * not the expected one, we must restart lookup. 43988ab1932SEric Dumazet * We probably met an item that was moved to another chain. 44088ab1932SEric Dumazet */ 441*5051ebd2SEric Dumazet if (get_nulls_value(node) != slot) 44288ab1932SEric Dumazet goto begin; 44388ab1932SEric Dumazet 444271b72c7SEric Dumazet if (result) { 445271b72c7SEric Dumazet if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) 446271b72c7SEric Dumazet result = NULL; 447271b72c7SEric Dumazet else if (unlikely(compute_score(result, net, saddr, hnum, sport, 448271b72c7SEric Dumazet daddr, dport, dif) < badness)) { 449271b72c7SEric Dumazet sock_put(result); 450271b72c7SEric Dumazet goto begin; 451271b72c7SEric Dumazet } 452271b72c7SEric Dumazet } 453271b72c7SEric Dumazet rcu_read_unlock(); 454db8dac20SDavid S. Miller return result; 455db8dac20SDavid S. Miller } 456db8dac20SDavid S. Miller 457607c4aafSKOVACS Krisztian static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 458607c4aafSKOVACS Krisztian __be16 sport, __be16 dport, 459645ca708SEric Dumazet struct udp_table *udptable) 460607c4aafSKOVACS Krisztian { 46123542618SKOVACS Krisztian struct sock *sk; 462607c4aafSKOVACS Krisztian const struct iphdr *iph = ip_hdr(skb); 463607c4aafSKOVACS Krisztian 46423542618SKOVACS Krisztian if (unlikely(sk = skb_steal_sock(skb))) 46523542618SKOVACS Krisztian return sk; 46623542618SKOVACS Krisztian else 467adf30907SEric Dumazet return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 468607c4aafSKOVACS Krisztian iph->daddr, dport, inet_iif(skb), 469607c4aafSKOVACS Krisztian udptable); 470607c4aafSKOVACS Krisztian } 471607c4aafSKOVACS Krisztian 472bcd41303SKOVACS Krisztian struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 473bcd41303SKOVACS Krisztian __be32 daddr, __be16 dport, int dif) 474bcd41303SKOVACS Krisztian { 475645ca708SEric Dumazet return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 476bcd41303SKOVACS Krisztian } 477bcd41303SKOVACS Krisztian EXPORT_SYMBOL_GPL(udp4_lib_lookup); 478bcd41303SKOVACS Krisztian 479920a4611SEric Dumazet static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 480db8dac20SDavid S. Miller __be16 loc_port, __be32 loc_addr, 481db8dac20SDavid S. Miller __be16 rmt_port, __be32 rmt_addr, 482db8dac20SDavid S. Miller int dif) 483db8dac20SDavid S. Miller { 48488ab1932SEric Dumazet struct hlist_nulls_node *node; 485db8dac20SDavid S. Miller struct sock *s = sk; 486db8dac20SDavid S. Miller unsigned short hnum = ntohs(loc_port); 487db8dac20SDavid S. Miller 48888ab1932SEric Dumazet sk_nulls_for_each_from(s, node) { 489db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(s); 490db8dac20SDavid S. Miller 491920a4611SEric Dumazet if (!net_eq(sock_net(s), net) || 492d4cada4aSEric Dumazet udp_sk(s)->udp_port_hash != hnum || 493c720c7e8SEric Dumazet (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 494c720c7e8SEric Dumazet (inet->inet_dport != rmt_port && inet->inet_dport) || 495c720c7e8SEric Dumazet (inet->inet_rcv_saddr && 496c720c7e8SEric Dumazet inet->inet_rcv_saddr != loc_addr) || 497db8dac20SDavid S. Miller ipv6_only_sock(s) || 498db8dac20SDavid S. Miller (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 499db8dac20SDavid S. Miller continue; 500db8dac20SDavid S. Miller if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 501db8dac20SDavid S. Miller continue; 502db8dac20SDavid S. Miller goto found; 503db8dac20SDavid S. Miller } 504db8dac20SDavid S. Miller s = NULL; 505db8dac20SDavid S. Miller found: 506db8dac20SDavid S. Miller return s; 507db8dac20SDavid S. Miller } 508db8dac20SDavid S. Miller 509db8dac20SDavid S. Miller /* 510db8dac20SDavid S. Miller * This routine is called by the ICMP module when it gets some 511db8dac20SDavid S. Miller * sort of error condition. If err < 0 then the socket should 512db8dac20SDavid S. Miller * be closed and the error returned to the user. If err > 0 513db8dac20SDavid S. Miller * it's just the icmp type << 8 | icmp code. 514db8dac20SDavid S. Miller * Header points to the ip header of the error packet. We move 515db8dac20SDavid S. Miller * on past this. Then (as it used to claim before adjustment) 516db8dac20SDavid S. Miller * header points to the first 8 bytes of the udp header. We need 517db8dac20SDavid S. Miller * to find the appropriate port. 518db8dac20SDavid S. Miller */ 519db8dac20SDavid S. Miller 520645ca708SEric Dumazet void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 521db8dac20SDavid S. Miller { 522db8dac20SDavid S. Miller struct inet_sock *inet; 523db8dac20SDavid S. Miller struct iphdr *iph = (struct iphdr *)skb->data; 524db8dac20SDavid S. Miller struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 525db8dac20SDavid S. Miller const int type = icmp_hdr(skb)->type; 526db8dac20SDavid S. Miller const int code = icmp_hdr(skb)->code; 527db8dac20SDavid S. Miller struct sock *sk; 528db8dac20SDavid S. Miller int harderr; 529db8dac20SDavid S. Miller int err; 530fd54d716SPavel Emelyanov struct net *net = dev_net(skb->dev); 531db8dac20SDavid S. Miller 532fd54d716SPavel Emelyanov sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 533db8dac20SDavid S. Miller iph->saddr, uh->source, skb->dev->ifindex, udptable); 534db8dac20SDavid S. Miller if (sk == NULL) { 535dcfc23caSPavel Emelyanov ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 536db8dac20SDavid S. Miller return; /* No socket for error */ 537db8dac20SDavid S. Miller } 538db8dac20SDavid S. Miller 539db8dac20SDavid S. Miller err = 0; 540db8dac20SDavid S. Miller harderr = 0; 541db8dac20SDavid S. Miller inet = inet_sk(sk); 542db8dac20SDavid S. Miller 543db8dac20SDavid S. Miller switch (type) { 544db8dac20SDavid S. Miller default: 545db8dac20SDavid S. Miller case ICMP_TIME_EXCEEDED: 546db8dac20SDavid S. Miller err = EHOSTUNREACH; 547db8dac20SDavid S. Miller break; 548db8dac20SDavid S. Miller case ICMP_SOURCE_QUENCH: 549db8dac20SDavid S. Miller goto out; 550db8dac20SDavid S. Miller case ICMP_PARAMETERPROB: 551db8dac20SDavid S. Miller err = EPROTO; 552db8dac20SDavid S. Miller harderr = 1; 553db8dac20SDavid S. Miller break; 554db8dac20SDavid S. Miller case ICMP_DEST_UNREACH: 555db8dac20SDavid S. Miller if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 556db8dac20SDavid S. Miller if (inet->pmtudisc != IP_PMTUDISC_DONT) { 557db8dac20SDavid S. Miller err = EMSGSIZE; 558db8dac20SDavid S. Miller harderr = 1; 559db8dac20SDavid S. Miller break; 560db8dac20SDavid S. Miller } 561db8dac20SDavid S. Miller goto out; 562db8dac20SDavid S. Miller } 563db8dac20SDavid S. Miller err = EHOSTUNREACH; 564db8dac20SDavid S. Miller if (code <= NR_ICMP_UNREACH) { 565db8dac20SDavid S. Miller harderr = icmp_err_convert[code].fatal; 566db8dac20SDavid S. Miller err = icmp_err_convert[code].errno; 567db8dac20SDavid S. Miller } 568db8dac20SDavid S. Miller break; 569db8dac20SDavid S. Miller } 570db8dac20SDavid S. Miller 571db8dac20SDavid S. Miller /* 572db8dac20SDavid S. Miller * RFC1122: OK. Passes ICMP errors back to application, as per 573db8dac20SDavid S. Miller * 4.1.3.3. 574db8dac20SDavid S. Miller */ 575db8dac20SDavid S. Miller if (!inet->recverr) { 576db8dac20SDavid S. Miller if (!harderr || sk->sk_state != TCP_ESTABLISHED) 577db8dac20SDavid S. Miller goto out; 578db8dac20SDavid S. Miller } else { 579db8dac20SDavid S. Miller ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 580db8dac20SDavid S. Miller } 581db8dac20SDavid S. Miller sk->sk_err = err; 582db8dac20SDavid S. Miller sk->sk_error_report(sk); 583db8dac20SDavid S. Miller out: 584db8dac20SDavid S. Miller sock_put(sk); 585db8dac20SDavid S. Miller } 586db8dac20SDavid S. Miller 587db8dac20SDavid S. Miller void udp_err(struct sk_buff *skb, u32 info) 588db8dac20SDavid S. Miller { 589645ca708SEric Dumazet __udp4_lib_err(skb, info, &udp_table); 590db8dac20SDavid S. Miller } 591db8dac20SDavid S. Miller 592db8dac20SDavid S. Miller /* 593db8dac20SDavid S. Miller * Throw away all pending data and cancel the corking. Socket is locked. 594db8dac20SDavid S. Miller */ 59536d926b9SDenis V. Lunev void udp_flush_pending_frames(struct sock *sk) 596db8dac20SDavid S. Miller { 597db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 598db8dac20SDavid S. Miller 599db8dac20SDavid S. Miller if (up->pending) { 600db8dac20SDavid S. Miller up->len = 0; 601db8dac20SDavid S. Miller up->pending = 0; 602db8dac20SDavid S. Miller ip_flush_pending_frames(sk); 603db8dac20SDavid S. Miller } 604db8dac20SDavid S. Miller } 60536d926b9SDenis V. Lunev EXPORT_SYMBOL(udp_flush_pending_frames); 606db8dac20SDavid S. Miller 607db8dac20SDavid S. Miller /** 608db8dac20SDavid S. Miller * udp4_hwcsum_outgoing - handle outgoing HW checksumming 609db8dac20SDavid S. Miller * @sk: socket we are sending on 610db8dac20SDavid S. Miller * @skb: sk_buff containing the filled-in UDP header 611db8dac20SDavid S. Miller * (checksum field must be zeroed out) 612db8dac20SDavid S. Miller */ 613db8dac20SDavid S. Miller static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 614db8dac20SDavid S. Miller __be32 src, __be32 dst, int len) 615db8dac20SDavid S. Miller { 616db8dac20SDavid S. Miller unsigned int offset; 617db8dac20SDavid S. Miller struct udphdr *uh = udp_hdr(skb); 618db8dac20SDavid S. Miller __wsum csum = 0; 619db8dac20SDavid S. Miller 620db8dac20SDavid S. Miller if (skb_queue_len(&sk->sk_write_queue) == 1) { 621db8dac20SDavid S. Miller /* 622db8dac20SDavid S. Miller * Only one fragment on the socket. 623db8dac20SDavid S. Miller */ 624db8dac20SDavid S. Miller skb->csum_start = skb_transport_header(skb) - skb->head; 625db8dac20SDavid S. Miller skb->csum_offset = offsetof(struct udphdr, check); 626db8dac20SDavid S. Miller uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 627db8dac20SDavid S. Miller } else { 628db8dac20SDavid S. Miller /* 629db8dac20SDavid S. Miller * HW-checksum won't work as there are two or more 630db8dac20SDavid S. Miller * fragments on the socket so that all csums of sk_buffs 631db8dac20SDavid S. Miller * should be together 632db8dac20SDavid S. Miller */ 633db8dac20SDavid S. Miller offset = skb_transport_offset(skb); 634db8dac20SDavid S. Miller skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 635db8dac20SDavid S. Miller 636db8dac20SDavid S. Miller skb->ip_summed = CHECKSUM_NONE; 637db8dac20SDavid S. Miller 638db8dac20SDavid S. Miller skb_queue_walk(&sk->sk_write_queue, skb) { 639db8dac20SDavid S. Miller csum = csum_add(csum, skb->csum); 640db8dac20SDavid S. Miller } 641db8dac20SDavid S. Miller 642db8dac20SDavid S. Miller uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 643db8dac20SDavid S. Miller if (uh->check == 0) 644db8dac20SDavid S. Miller uh->check = CSUM_MANGLED_0; 645db8dac20SDavid S. Miller } 646db8dac20SDavid S. Miller } 647db8dac20SDavid S. Miller 648db8dac20SDavid S. Miller /* 649db8dac20SDavid S. Miller * Push out all pending data as one UDP datagram. Socket is locked. 650db8dac20SDavid S. Miller */ 651db8dac20SDavid S. Miller static int udp_push_pending_frames(struct sock *sk) 652db8dac20SDavid S. Miller { 653db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 654db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sk); 655db8dac20SDavid S. Miller struct flowi *fl = &inet->cork.fl; 656db8dac20SDavid S. Miller struct sk_buff *skb; 657db8dac20SDavid S. Miller struct udphdr *uh; 658db8dac20SDavid S. Miller int err = 0; 659db8dac20SDavid S. Miller int is_udplite = IS_UDPLITE(sk); 660db8dac20SDavid S. Miller __wsum csum = 0; 661db8dac20SDavid S. Miller 662db8dac20SDavid S. Miller /* Grab the skbuff where UDP header space exists. */ 663db8dac20SDavid S. Miller if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 664db8dac20SDavid S. Miller goto out; 665db8dac20SDavid S. Miller 666db8dac20SDavid S. Miller /* 667db8dac20SDavid S. Miller * Create a UDP header 668db8dac20SDavid S. Miller */ 669db8dac20SDavid S. Miller uh = udp_hdr(skb); 670db8dac20SDavid S. Miller uh->source = fl->fl_ip_sport; 671db8dac20SDavid S. Miller uh->dest = fl->fl_ip_dport; 672db8dac20SDavid S. Miller uh->len = htons(up->len); 673db8dac20SDavid S. Miller uh->check = 0; 674db8dac20SDavid S. Miller 675db8dac20SDavid S. Miller if (is_udplite) /* UDP-Lite */ 676db8dac20SDavid S. Miller csum = udplite_csum_outgoing(sk, skb); 677db8dac20SDavid S. Miller 678db8dac20SDavid S. Miller else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 679db8dac20SDavid S. Miller 680db8dac20SDavid S. Miller skb->ip_summed = CHECKSUM_NONE; 681db8dac20SDavid S. Miller goto send; 682db8dac20SDavid S. Miller 683db8dac20SDavid S. Miller } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 684db8dac20SDavid S. Miller 685db8dac20SDavid S. Miller udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len); 686db8dac20SDavid S. Miller goto send; 687db8dac20SDavid S. Miller 688db8dac20SDavid S. Miller } else /* `normal' UDP */ 689db8dac20SDavid S. Miller csum = udp_csum_outgoing(sk, skb); 690db8dac20SDavid S. Miller 691db8dac20SDavid S. Miller /* add protocol-dependent pseudo-header */ 692db8dac20SDavid S. Miller uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 693db8dac20SDavid S. Miller sk->sk_protocol, csum); 694db8dac20SDavid S. Miller if (uh->check == 0) 695db8dac20SDavid S. Miller uh->check = CSUM_MANGLED_0; 696db8dac20SDavid S. Miller 697db8dac20SDavid S. Miller send: 698db8dac20SDavid S. Miller err = ip_push_pending_frames(sk); 6996ce9e7b5SEric Dumazet if (err) { 7006ce9e7b5SEric Dumazet if (err == -ENOBUFS && !inet->recverr) { 7016ce9e7b5SEric Dumazet UDP_INC_STATS_USER(sock_net(sk), 7026ce9e7b5SEric Dumazet UDP_MIB_SNDBUFERRORS, is_udplite); 7036ce9e7b5SEric Dumazet err = 0; 7046ce9e7b5SEric Dumazet } 7056ce9e7b5SEric Dumazet } else 7066ce9e7b5SEric Dumazet UDP_INC_STATS_USER(sock_net(sk), 7076ce9e7b5SEric Dumazet UDP_MIB_OUTDATAGRAMS, is_udplite); 708db8dac20SDavid S. Miller out: 709db8dac20SDavid S. Miller up->len = 0; 710db8dac20SDavid S. Miller up->pending = 0; 711db8dac20SDavid S. Miller return err; 712db8dac20SDavid S. Miller } 713db8dac20SDavid S. Miller 714db8dac20SDavid S. Miller int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 715db8dac20SDavid S. Miller size_t len) 716db8dac20SDavid S. Miller { 717db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sk); 718db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 719db8dac20SDavid S. Miller int ulen = len; 720db8dac20SDavid S. Miller struct ipcm_cookie ipc; 721db8dac20SDavid S. Miller struct rtable *rt = NULL; 722db8dac20SDavid S. Miller int free = 0; 723db8dac20SDavid S. Miller int connected = 0; 724db8dac20SDavid S. Miller __be32 daddr, faddr, saddr; 725db8dac20SDavid S. Miller __be16 dport; 726db8dac20SDavid S. Miller u8 tos; 727db8dac20SDavid S. Miller int err, is_udplite = IS_UDPLITE(sk); 728db8dac20SDavid S. Miller int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 729db8dac20SDavid S. Miller int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 730db8dac20SDavid S. Miller 731db8dac20SDavid S. Miller if (len > 0xFFFF) 732db8dac20SDavid S. Miller return -EMSGSIZE; 733db8dac20SDavid S. Miller 734db8dac20SDavid S. Miller /* 735db8dac20SDavid S. Miller * Check the flags. 736db8dac20SDavid S. Miller */ 737db8dac20SDavid S. Miller 738db8dac20SDavid S. Miller if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 739db8dac20SDavid S. Miller return -EOPNOTSUPP; 740db8dac20SDavid S. Miller 741db8dac20SDavid S. Miller ipc.opt = NULL; 74251f31cabSPatrick Ohly ipc.shtx.flags = 0; 743db8dac20SDavid S. Miller 744db8dac20SDavid S. Miller if (up->pending) { 745db8dac20SDavid S. Miller /* 746db8dac20SDavid S. Miller * There are pending frames. 747db8dac20SDavid S. Miller * The socket lock must be held while it's corked. 748db8dac20SDavid S. Miller */ 749db8dac20SDavid S. Miller lock_sock(sk); 750db8dac20SDavid S. Miller if (likely(up->pending)) { 751db8dac20SDavid S. Miller if (unlikely(up->pending != AF_INET)) { 752db8dac20SDavid S. Miller release_sock(sk); 753db8dac20SDavid S. Miller return -EINVAL; 754db8dac20SDavid S. Miller } 755db8dac20SDavid S. Miller goto do_append_data; 756db8dac20SDavid S. Miller } 757db8dac20SDavid S. Miller release_sock(sk); 758db8dac20SDavid S. Miller } 759db8dac20SDavid S. Miller ulen += sizeof(struct udphdr); 760db8dac20SDavid S. Miller 761db8dac20SDavid S. Miller /* 762db8dac20SDavid S. Miller * Get and verify the address. 763db8dac20SDavid S. Miller */ 764db8dac20SDavid S. Miller if (msg->msg_name) { 765db8dac20SDavid S. Miller struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name; 766db8dac20SDavid S. Miller if (msg->msg_namelen < sizeof(*usin)) 767db8dac20SDavid S. Miller return -EINVAL; 768db8dac20SDavid S. Miller if (usin->sin_family != AF_INET) { 769db8dac20SDavid S. Miller if (usin->sin_family != AF_UNSPEC) 770db8dac20SDavid S. Miller return -EAFNOSUPPORT; 771db8dac20SDavid S. Miller } 772db8dac20SDavid S. Miller 773db8dac20SDavid S. Miller daddr = usin->sin_addr.s_addr; 774db8dac20SDavid S. Miller dport = usin->sin_port; 775db8dac20SDavid S. Miller if (dport == 0) 776db8dac20SDavid S. Miller return -EINVAL; 777db8dac20SDavid S. Miller } else { 778db8dac20SDavid S. Miller if (sk->sk_state != TCP_ESTABLISHED) 779db8dac20SDavid S. Miller return -EDESTADDRREQ; 780c720c7e8SEric Dumazet daddr = inet->inet_daddr; 781c720c7e8SEric Dumazet dport = inet->inet_dport; 782db8dac20SDavid S. Miller /* Open fast path for connected socket. 783db8dac20SDavid S. Miller Route will not be used, if at least one option is set. 784db8dac20SDavid S. Miller */ 785db8dac20SDavid S. Miller connected = 1; 786db8dac20SDavid S. Miller } 787c720c7e8SEric Dumazet ipc.addr = inet->inet_saddr; 788db8dac20SDavid S. Miller 789db8dac20SDavid S. Miller ipc.oif = sk->sk_bound_dev_if; 79051f31cabSPatrick Ohly err = sock_tx_timestamp(msg, sk, &ipc.shtx); 79151f31cabSPatrick Ohly if (err) 79251f31cabSPatrick Ohly return err; 793db8dac20SDavid S. Miller if (msg->msg_controllen) { 7943b1e0a65SYOSHIFUJI Hideaki err = ip_cmsg_send(sock_net(sk), msg, &ipc); 795db8dac20SDavid S. Miller if (err) 796db8dac20SDavid S. Miller return err; 797db8dac20SDavid S. Miller if (ipc.opt) 798db8dac20SDavid S. Miller free = 1; 799db8dac20SDavid S. Miller connected = 0; 800db8dac20SDavid S. Miller } 801db8dac20SDavid S. Miller if (!ipc.opt) 802db8dac20SDavid S. Miller ipc.opt = inet->opt; 803db8dac20SDavid S. Miller 804db8dac20SDavid S. Miller saddr = ipc.addr; 805db8dac20SDavid S. Miller ipc.addr = faddr = daddr; 806db8dac20SDavid S. Miller 807db8dac20SDavid S. Miller if (ipc.opt && ipc.opt->srr) { 808db8dac20SDavid S. Miller if (!daddr) 809db8dac20SDavid S. Miller return -EINVAL; 810db8dac20SDavid S. Miller faddr = ipc.opt->faddr; 811db8dac20SDavid S. Miller connected = 0; 812db8dac20SDavid S. Miller } 813db8dac20SDavid S. Miller tos = RT_TOS(inet->tos); 814db8dac20SDavid S. Miller if (sock_flag(sk, SOCK_LOCALROUTE) || 815db8dac20SDavid S. Miller (msg->msg_flags & MSG_DONTROUTE) || 816db8dac20SDavid S. Miller (ipc.opt && ipc.opt->is_strictroute)) { 817db8dac20SDavid S. Miller tos |= RTO_ONLINK; 818db8dac20SDavid S. Miller connected = 0; 819db8dac20SDavid S. Miller } 820db8dac20SDavid S. Miller 821db8dac20SDavid S. Miller if (ipv4_is_multicast(daddr)) { 822db8dac20SDavid S. Miller if (!ipc.oif) 823db8dac20SDavid S. Miller ipc.oif = inet->mc_index; 824db8dac20SDavid S. Miller if (!saddr) 825db8dac20SDavid S. Miller saddr = inet->mc_addr; 826db8dac20SDavid S. Miller connected = 0; 827db8dac20SDavid S. Miller } 828db8dac20SDavid S. Miller 829db8dac20SDavid S. Miller if (connected) 830db8dac20SDavid S. Miller rt = (struct rtable *)sk_dst_check(sk, 0); 831db8dac20SDavid S. Miller 832db8dac20SDavid S. Miller if (rt == NULL) { 833db8dac20SDavid S. Miller struct flowi fl = { .oif = ipc.oif, 834914a9ab3SAtis Elsts .mark = sk->sk_mark, 835db8dac20SDavid S. Miller .nl_u = { .ip4_u = 836db8dac20SDavid S. Miller { .daddr = faddr, 837db8dac20SDavid S. Miller .saddr = saddr, 838db8dac20SDavid S. Miller .tos = tos } }, 839db8dac20SDavid S. Miller .proto = sk->sk_protocol, 840a134f85cSBalazs Scheidler .flags = inet_sk_flowi_flags(sk), 841db8dac20SDavid S. Miller .uli_u = { .ports = 842c720c7e8SEric Dumazet { .sport = inet->inet_sport, 843db8dac20SDavid S. Miller .dport = dport } } }; 84484a3aa00SPavel Emelyanov struct net *net = sock_net(sk); 84584a3aa00SPavel Emelyanov 846db8dac20SDavid S. Miller security_sk_classify_flow(sk, &fl); 84784a3aa00SPavel Emelyanov err = ip_route_output_flow(net, &rt, &fl, sk, 1); 848db8dac20SDavid S. Miller if (err) { 849db8dac20SDavid S. Miller if (err == -ENETUNREACH) 8507c73a6faSPavel Emelyanov IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 851db8dac20SDavid S. Miller goto out; 852db8dac20SDavid S. Miller } 853db8dac20SDavid S. Miller 854db8dac20SDavid S. Miller err = -EACCES; 855db8dac20SDavid S. Miller if ((rt->rt_flags & RTCF_BROADCAST) && 856db8dac20SDavid S. Miller !sock_flag(sk, SOCK_BROADCAST)) 857db8dac20SDavid S. Miller goto out; 858db8dac20SDavid S. Miller if (connected) 859db8dac20SDavid S. Miller sk_dst_set(sk, dst_clone(&rt->u.dst)); 860db8dac20SDavid S. Miller } 861db8dac20SDavid S. Miller 862db8dac20SDavid S. Miller if (msg->msg_flags&MSG_CONFIRM) 863db8dac20SDavid S. Miller goto do_confirm; 864db8dac20SDavid S. Miller back_from_confirm: 865db8dac20SDavid S. Miller 866db8dac20SDavid S. Miller saddr = rt->rt_src; 867db8dac20SDavid S. Miller if (!ipc.addr) 868db8dac20SDavid S. Miller daddr = ipc.addr = rt->rt_dst; 869db8dac20SDavid S. Miller 870db8dac20SDavid S. Miller lock_sock(sk); 871db8dac20SDavid S. Miller if (unlikely(up->pending)) { 872db8dac20SDavid S. Miller /* The socket is already corked while preparing it. */ 873db8dac20SDavid S. Miller /* ... which is an evident application bug. --ANK */ 874db8dac20SDavid S. Miller release_sock(sk); 875db8dac20SDavid S. Miller 876db8dac20SDavid S. Miller LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 877db8dac20SDavid S. Miller err = -EINVAL; 878db8dac20SDavid S. Miller goto out; 879db8dac20SDavid S. Miller } 880db8dac20SDavid S. Miller /* 881db8dac20SDavid S. Miller * Now cork the socket to pend data. 882db8dac20SDavid S. Miller */ 883db8dac20SDavid S. Miller inet->cork.fl.fl4_dst = daddr; 884db8dac20SDavid S. Miller inet->cork.fl.fl_ip_dport = dport; 885db8dac20SDavid S. Miller inet->cork.fl.fl4_src = saddr; 886c720c7e8SEric Dumazet inet->cork.fl.fl_ip_sport = inet->inet_sport; 887db8dac20SDavid S. Miller up->pending = AF_INET; 888db8dac20SDavid S. Miller 889db8dac20SDavid S. Miller do_append_data: 890db8dac20SDavid S. Miller up->len += ulen; 891db8dac20SDavid S. Miller getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 892db8dac20SDavid S. Miller err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 8932e77d89bSEric Dumazet sizeof(struct udphdr), &ipc, &rt, 894db8dac20SDavid S. Miller corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 895db8dac20SDavid S. Miller if (err) 896db8dac20SDavid S. Miller udp_flush_pending_frames(sk); 897db8dac20SDavid S. Miller else if (!corkreq) 898db8dac20SDavid S. Miller err = udp_push_pending_frames(sk); 899db8dac20SDavid S. Miller else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 900db8dac20SDavid S. Miller up->pending = 0; 901db8dac20SDavid S. Miller release_sock(sk); 902db8dac20SDavid S. Miller 903db8dac20SDavid S. Miller out: 904db8dac20SDavid S. Miller ip_rt_put(rt); 905db8dac20SDavid S. Miller if (free) 906db8dac20SDavid S. Miller kfree(ipc.opt); 907db8dac20SDavid S. Miller if (!err) 908db8dac20SDavid S. Miller return len; 909db8dac20SDavid S. Miller /* 910db8dac20SDavid S. Miller * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 911db8dac20SDavid S. Miller * ENOBUFS might not be good (it's not tunable per se), but otherwise 912db8dac20SDavid S. Miller * we don't have a good statistic (IpOutDiscards but it can be too many 913db8dac20SDavid S. Miller * things). We could add another new stat but at least for now that 914db8dac20SDavid S. Miller * seems like overkill. 915db8dac20SDavid S. Miller */ 916db8dac20SDavid S. Miller if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 917629ca23cSPavel Emelyanov UDP_INC_STATS_USER(sock_net(sk), 918629ca23cSPavel Emelyanov UDP_MIB_SNDBUFERRORS, is_udplite); 919db8dac20SDavid S. Miller } 920db8dac20SDavid S. Miller return err; 921db8dac20SDavid S. Miller 922db8dac20SDavid S. Miller do_confirm: 923db8dac20SDavid S. Miller dst_confirm(&rt->u.dst); 924db8dac20SDavid S. Miller if (!(msg->msg_flags&MSG_PROBE) || len) 925db8dac20SDavid S. Miller goto back_from_confirm; 926db8dac20SDavid S. Miller err = 0; 927db8dac20SDavid S. Miller goto out; 928db8dac20SDavid S. Miller } 929c482c568SEric Dumazet EXPORT_SYMBOL(udp_sendmsg); 930db8dac20SDavid S. Miller 931db8dac20SDavid S. Miller int udp_sendpage(struct sock *sk, struct page *page, int offset, 932db8dac20SDavid S. Miller size_t size, int flags) 933db8dac20SDavid S. Miller { 934db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 935db8dac20SDavid S. Miller int ret; 936db8dac20SDavid S. Miller 937db8dac20SDavid S. Miller if (!up->pending) { 938db8dac20SDavid S. Miller struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 939db8dac20SDavid S. Miller 940db8dac20SDavid S. Miller /* Call udp_sendmsg to specify destination address which 941db8dac20SDavid S. Miller * sendpage interface can't pass. 942db8dac20SDavid S. Miller * This will succeed only when the socket is connected. 943db8dac20SDavid S. Miller */ 944db8dac20SDavid S. Miller ret = udp_sendmsg(NULL, sk, &msg, 0); 945db8dac20SDavid S. Miller if (ret < 0) 946db8dac20SDavid S. Miller return ret; 947db8dac20SDavid S. Miller } 948db8dac20SDavid S. Miller 949db8dac20SDavid S. Miller lock_sock(sk); 950db8dac20SDavid S. Miller 951db8dac20SDavid S. Miller if (unlikely(!up->pending)) { 952db8dac20SDavid S. Miller release_sock(sk); 953db8dac20SDavid S. Miller 954db8dac20SDavid S. Miller LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 955db8dac20SDavid S. Miller return -EINVAL; 956db8dac20SDavid S. Miller } 957db8dac20SDavid S. Miller 958db8dac20SDavid S. Miller ret = ip_append_page(sk, page, offset, size, flags); 959db8dac20SDavid S. Miller if (ret == -EOPNOTSUPP) { 960db8dac20SDavid S. Miller release_sock(sk); 961db8dac20SDavid S. Miller return sock_no_sendpage(sk->sk_socket, page, offset, 962db8dac20SDavid S. Miller size, flags); 963db8dac20SDavid S. Miller } 964db8dac20SDavid S. Miller if (ret < 0) { 965db8dac20SDavid S. Miller udp_flush_pending_frames(sk); 966db8dac20SDavid S. Miller goto out; 967db8dac20SDavid S. Miller } 968db8dac20SDavid S. Miller 969db8dac20SDavid S. Miller up->len += size; 970db8dac20SDavid S. Miller if (!(up->corkflag || (flags&MSG_MORE))) 971db8dac20SDavid S. Miller ret = udp_push_pending_frames(sk); 972db8dac20SDavid S. Miller if (!ret) 973db8dac20SDavid S. Miller ret = size; 974db8dac20SDavid S. Miller out: 975db8dac20SDavid S. Miller release_sock(sk); 976db8dac20SDavid S. Miller return ret; 977db8dac20SDavid S. Miller } 978db8dac20SDavid S. Miller 97985584672SEric Dumazet 98085584672SEric Dumazet /** 98185584672SEric Dumazet * first_packet_length - return length of first packet in receive queue 98285584672SEric Dumazet * @sk: socket 98385584672SEric Dumazet * 98485584672SEric Dumazet * Drops all bad checksum frames, until a valid one is found. 98585584672SEric Dumazet * Returns the length of found skb, or 0 if none is found. 98685584672SEric Dumazet */ 98785584672SEric Dumazet static unsigned int first_packet_length(struct sock *sk) 98885584672SEric Dumazet { 98985584672SEric Dumazet struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 99085584672SEric Dumazet struct sk_buff *skb; 99185584672SEric Dumazet unsigned int res; 99285584672SEric Dumazet 99385584672SEric Dumazet __skb_queue_head_init(&list_kill); 99485584672SEric Dumazet 99585584672SEric Dumazet spin_lock_bh(&rcvq->lock); 99685584672SEric Dumazet while ((skb = skb_peek(rcvq)) != NULL && 99785584672SEric Dumazet udp_lib_checksum_complete(skb)) { 99885584672SEric Dumazet UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 99985584672SEric Dumazet IS_UDPLITE(sk)); 10008edf19c2SEric Dumazet atomic_inc(&sk->sk_drops); 100185584672SEric Dumazet __skb_unlink(skb, rcvq); 100285584672SEric Dumazet __skb_queue_tail(&list_kill, skb); 100385584672SEric Dumazet } 100485584672SEric Dumazet res = skb ? skb->len : 0; 100585584672SEric Dumazet spin_unlock_bh(&rcvq->lock); 100685584672SEric Dumazet 100785584672SEric Dumazet if (!skb_queue_empty(&list_kill)) { 100885584672SEric Dumazet lock_sock(sk); 100985584672SEric Dumazet __skb_queue_purge(&list_kill); 101085584672SEric Dumazet sk_mem_reclaim_partial(sk); 101185584672SEric Dumazet release_sock(sk); 101285584672SEric Dumazet } 101385584672SEric Dumazet return res; 101485584672SEric Dumazet } 101585584672SEric Dumazet 10161da177e4SLinus Torvalds /* 10171da177e4SLinus Torvalds * IOCTL requests applicable to the UDP protocol 10181da177e4SLinus Torvalds */ 10191da177e4SLinus Torvalds 10201da177e4SLinus Torvalds int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 10211da177e4SLinus Torvalds { 10226516c655SStephen Hemminger switch (cmd) { 10231da177e4SLinus Torvalds case SIOCOUTQ: 10241da177e4SLinus Torvalds { 102531e6d363SEric Dumazet int amount = sk_wmem_alloc_get(sk); 102631e6d363SEric Dumazet 10271da177e4SLinus Torvalds return put_user(amount, (int __user *)arg); 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds 10301da177e4SLinus Torvalds case SIOCINQ: 10311da177e4SLinus Torvalds { 103285584672SEric Dumazet unsigned int amount = first_packet_length(sk); 10331da177e4SLinus Torvalds 103485584672SEric Dumazet if (amount) 10351da177e4SLinus Torvalds /* 10361da177e4SLinus Torvalds * We will only return the amount 10371da177e4SLinus Torvalds * of this packet since that is all 10381da177e4SLinus Torvalds * that will be read. 10391da177e4SLinus Torvalds */ 104085584672SEric Dumazet amount -= sizeof(struct udphdr); 104185584672SEric Dumazet 10421da177e4SLinus Torvalds return put_user(amount, (int __user *)arg); 10431da177e4SLinus Torvalds } 10441da177e4SLinus Torvalds 10451da177e4SLinus Torvalds default: 10461da177e4SLinus Torvalds return -ENOIOCTLCMD; 10471da177e4SLinus Torvalds } 10486516c655SStephen Hemminger 10496516c655SStephen Hemminger return 0; 10501da177e4SLinus Torvalds } 1051c482c568SEric Dumazet EXPORT_SYMBOL(udp_ioctl); 10521da177e4SLinus Torvalds 1053db8dac20SDavid S. Miller /* 1054db8dac20SDavid S. Miller * This should be easy, if there is something there we 1055db8dac20SDavid S. Miller * return it, otherwise we block. 1056db8dac20SDavid S. Miller */ 1057db8dac20SDavid S. Miller 1058db8dac20SDavid S. Miller int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1059db8dac20SDavid S. Miller size_t len, int noblock, int flags, int *addr_len) 1060db8dac20SDavid S. Miller { 1061db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sk); 1062db8dac20SDavid S. Miller struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 1063db8dac20SDavid S. Miller struct sk_buff *skb; 1064db8dac20SDavid S. Miller unsigned int ulen, copied; 1065db8dac20SDavid S. Miller int peeked; 1066db8dac20SDavid S. Miller int err; 1067db8dac20SDavid S. Miller int is_udplite = IS_UDPLITE(sk); 1068db8dac20SDavid S. Miller 1069db8dac20SDavid S. Miller /* 1070db8dac20SDavid S. Miller * Check any passed addresses 1071db8dac20SDavid S. Miller */ 1072db8dac20SDavid S. Miller if (addr_len) 1073db8dac20SDavid S. Miller *addr_len = sizeof(*sin); 1074db8dac20SDavid S. Miller 1075db8dac20SDavid S. Miller if (flags & MSG_ERRQUEUE) 1076db8dac20SDavid S. Miller return ip_recv_error(sk, msg, len); 1077db8dac20SDavid S. Miller 1078db8dac20SDavid S. Miller try_again: 1079db8dac20SDavid S. Miller skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1080db8dac20SDavid S. Miller &peeked, &err); 1081db8dac20SDavid S. Miller if (!skb) 1082db8dac20SDavid S. Miller goto out; 1083db8dac20SDavid S. Miller 1084db8dac20SDavid S. Miller ulen = skb->len - sizeof(struct udphdr); 1085db8dac20SDavid S. Miller copied = len; 1086db8dac20SDavid S. Miller if (copied > ulen) 1087db8dac20SDavid S. Miller copied = ulen; 1088db8dac20SDavid S. Miller else if (copied < ulen) 1089db8dac20SDavid S. Miller msg->msg_flags |= MSG_TRUNC; 1090db8dac20SDavid S. Miller 1091db8dac20SDavid S. Miller /* 1092db8dac20SDavid S. Miller * If checksum is needed at all, try to do it while copying the 1093db8dac20SDavid S. Miller * data. If the data is truncated, or if we only want a partial 1094db8dac20SDavid S. Miller * coverage checksum (UDP-Lite), do it before the copy. 1095db8dac20SDavid S. Miller */ 1096db8dac20SDavid S. Miller 1097db8dac20SDavid S. Miller if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1098db8dac20SDavid S. Miller if (udp_lib_checksum_complete(skb)) 1099db8dac20SDavid S. Miller goto csum_copy_err; 1100db8dac20SDavid S. Miller } 1101db8dac20SDavid S. Miller 1102db8dac20SDavid S. Miller if (skb_csum_unnecessary(skb)) 1103db8dac20SDavid S. Miller err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 1104db8dac20SDavid S. Miller msg->msg_iov, copied); 1105db8dac20SDavid S. Miller else { 1106c482c568SEric Dumazet err = skb_copy_and_csum_datagram_iovec(skb, 1107c482c568SEric Dumazet sizeof(struct udphdr), 1108c482c568SEric Dumazet msg->msg_iov); 1109db8dac20SDavid S. Miller 1110db8dac20SDavid S. Miller if (err == -EINVAL) 1111db8dac20SDavid S. Miller goto csum_copy_err; 1112db8dac20SDavid S. Miller } 1113db8dac20SDavid S. Miller 1114db8dac20SDavid S. Miller if (err) 1115db8dac20SDavid S. Miller goto out_free; 1116db8dac20SDavid S. Miller 1117db8dac20SDavid S. Miller if (!peeked) 1118629ca23cSPavel Emelyanov UDP_INC_STATS_USER(sock_net(sk), 1119629ca23cSPavel Emelyanov UDP_MIB_INDATAGRAMS, is_udplite); 1120db8dac20SDavid S. Miller 11213b885787SNeil Horman sock_recv_ts_and_drops(msg, sk, skb); 1122db8dac20SDavid S. Miller 1123db8dac20SDavid S. Miller /* Copy the address. */ 1124c482c568SEric Dumazet if (sin) { 1125db8dac20SDavid S. Miller sin->sin_family = AF_INET; 1126db8dac20SDavid S. Miller sin->sin_port = udp_hdr(skb)->source; 1127db8dac20SDavid S. Miller sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1128db8dac20SDavid S. Miller memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1129db8dac20SDavid S. Miller } 1130db8dac20SDavid S. Miller if (inet->cmsg_flags) 1131db8dac20SDavid S. Miller ip_cmsg_recv(msg, skb); 1132db8dac20SDavid S. Miller 1133db8dac20SDavid S. Miller err = copied; 1134db8dac20SDavid S. Miller if (flags & MSG_TRUNC) 1135db8dac20SDavid S. Miller err = ulen; 1136db8dac20SDavid S. Miller 1137db8dac20SDavid S. Miller out_free: 11389d410c79SEric Dumazet skb_free_datagram_locked(sk, skb); 1139db8dac20SDavid S. Miller out: 1140db8dac20SDavid S. Miller return err; 1141db8dac20SDavid S. Miller 1142db8dac20SDavid S. Miller csum_copy_err: 1143db8dac20SDavid S. Miller lock_sock(sk); 1144db8dac20SDavid S. Miller if (!skb_kill_datagram(sk, skb, flags)) 1145629ca23cSPavel Emelyanov UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1146db8dac20SDavid S. Miller release_sock(sk); 1147db8dac20SDavid S. Miller 1148db8dac20SDavid S. Miller if (noblock) 1149db8dac20SDavid S. Miller return -EAGAIN; 1150db8dac20SDavid S. Miller goto try_again; 1151db8dac20SDavid S. Miller } 1152db8dac20SDavid S. Miller 1153db8dac20SDavid S. Miller 11541da177e4SLinus Torvalds int udp_disconnect(struct sock *sk, int flags) 11551da177e4SLinus Torvalds { 11561da177e4SLinus Torvalds struct inet_sock *inet = inet_sk(sk); 11571da177e4SLinus Torvalds /* 11581da177e4SLinus Torvalds * 1003.1g - break association. 11591da177e4SLinus Torvalds */ 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds sk->sk_state = TCP_CLOSE; 1162c720c7e8SEric Dumazet inet->inet_daddr = 0; 1163c720c7e8SEric Dumazet inet->inet_dport = 0; 11641da177e4SLinus Torvalds sk->sk_bound_dev_if = 0; 11651da177e4SLinus Torvalds if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 11661da177e4SLinus Torvalds inet_reset_saddr(sk); 11671da177e4SLinus Torvalds 11681da177e4SLinus Torvalds if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 11691da177e4SLinus Torvalds sk->sk_prot->unhash(sk); 1170c720c7e8SEric Dumazet inet->inet_sport = 0; 11711da177e4SLinus Torvalds } 11721da177e4SLinus Torvalds sk_dst_reset(sk); 11731da177e4SLinus Torvalds return 0; 11741da177e4SLinus Torvalds } 1175c482c568SEric Dumazet EXPORT_SYMBOL(udp_disconnect); 11761da177e4SLinus Torvalds 1177645ca708SEric Dumazet void udp_lib_unhash(struct sock *sk) 1178645ca708SEric Dumazet { 1179723b4610SEric Dumazet if (sk_hashed(sk)) { 1180645ca708SEric Dumazet struct udp_table *udptable = sk->sk_prot->h.udp_table; 1181512615b6SEric Dumazet struct udp_hslot *hslot, *hslot2; 1182512615b6SEric Dumazet 1183512615b6SEric Dumazet hslot = udp_hashslot(udptable, sock_net(sk), 1184d4cada4aSEric Dumazet udp_sk(sk)->udp_port_hash); 1185512615b6SEric Dumazet hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1186645ca708SEric Dumazet 1187c8db3fecSEric Dumazet spin_lock_bh(&hslot->lock); 118888ab1932SEric Dumazet if (sk_nulls_del_node_init_rcu(sk)) { 1189fdcc8aa9SEric Dumazet hslot->count--; 1190c720c7e8SEric Dumazet inet_sk(sk)->inet_num = 0; 1191645ca708SEric Dumazet sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1192512615b6SEric Dumazet 1193512615b6SEric Dumazet spin_lock(&hslot2->lock); 1194512615b6SEric Dumazet hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1195512615b6SEric Dumazet hslot2->count--; 1196512615b6SEric Dumazet spin_unlock(&hslot2->lock); 1197645ca708SEric Dumazet } 1198c8db3fecSEric Dumazet spin_unlock_bh(&hslot->lock); 1199645ca708SEric Dumazet } 1200723b4610SEric Dumazet } 1201645ca708SEric Dumazet EXPORT_SYMBOL(udp_lib_unhash); 1202645ca708SEric Dumazet 120393821778SHerbert Xu static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 120493821778SHerbert Xu { 1205766e9037SEric Dumazet int rc = sock_queue_rcv_skb(sk, skb); 120693821778SHerbert Xu 1207766e9037SEric Dumazet if (rc < 0) { 1208766e9037SEric Dumazet int is_udplite = IS_UDPLITE(sk); 1209766e9037SEric Dumazet 121093821778SHerbert Xu /* Note that an ENOMEM error is charged twice */ 1211766e9037SEric Dumazet if (rc == -ENOMEM) 121293821778SHerbert Xu UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 121393821778SHerbert Xu is_udplite); 1214766e9037SEric Dumazet UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1215766e9037SEric Dumazet kfree_skb(skb); 1216766e9037SEric Dumazet return -1; 121793821778SHerbert Xu } 121893821778SHerbert Xu 121993821778SHerbert Xu return 0; 122093821778SHerbert Xu 122193821778SHerbert Xu } 122293821778SHerbert Xu 1223db8dac20SDavid S. Miller /* returns: 1224db8dac20SDavid S. Miller * -1: error 1225db8dac20SDavid S. Miller * 0: success 1226db8dac20SDavid S. Miller * >0: "udp encap" protocol resubmission 1227db8dac20SDavid S. Miller * 1228db8dac20SDavid S. Miller * Note that in the success and error cases, the skb is assumed to 1229db8dac20SDavid S. Miller * have either been requeued or freed. 1230db8dac20SDavid S. Miller */ 1231db8dac20SDavid S. Miller int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1232db8dac20SDavid S. Miller { 1233db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 1234db8dac20SDavid S. Miller int rc; 1235db8dac20SDavid S. Miller int is_udplite = IS_UDPLITE(sk); 1236db8dac20SDavid S. Miller 1237db8dac20SDavid S. Miller /* 1238db8dac20SDavid S. Miller * Charge it to the socket, dropping if the queue is full. 1239db8dac20SDavid S. Miller */ 1240db8dac20SDavid S. Miller if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1241db8dac20SDavid S. Miller goto drop; 1242db8dac20SDavid S. Miller nf_reset(skb); 1243db8dac20SDavid S. Miller 1244db8dac20SDavid S. Miller if (up->encap_type) { 1245db8dac20SDavid S. Miller /* 1246db8dac20SDavid S. Miller * This is an encapsulation socket so pass the skb to 1247db8dac20SDavid S. Miller * the socket's udp_encap_rcv() hook. Otherwise, just 1248db8dac20SDavid S. Miller * fall through and pass this up the UDP socket. 1249db8dac20SDavid S. Miller * up->encap_rcv() returns the following value: 1250db8dac20SDavid S. Miller * =0 if skb was successfully passed to the encap 1251db8dac20SDavid S. Miller * handler or was discarded by it. 1252db8dac20SDavid S. Miller * >0 if skb should be passed on to UDP. 1253db8dac20SDavid S. Miller * <0 if skb should be resubmitted as proto -N 1254db8dac20SDavid S. Miller */ 1255db8dac20SDavid S. Miller 1256db8dac20SDavid S. Miller /* if we're overly short, let UDP handle it */ 1257db8dac20SDavid S. Miller if (skb->len > sizeof(struct udphdr) && 1258db8dac20SDavid S. Miller up->encap_rcv != NULL) { 1259db8dac20SDavid S. Miller int ret; 1260db8dac20SDavid S. Miller 1261db8dac20SDavid S. Miller ret = (*up->encap_rcv)(sk, skb); 1262db8dac20SDavid S. Miller if (ret <= 0) { 12630283328eSPavel Emelyanov UDP_INC_STATS_BH(sock_net(sk), 12640283328eSPavel Emelyanov UDP_MIB_INDATAGRAMS, 1265db8dac20SDavid S. Miller is_udplite); 1266db8dac20SDavid S. Miller return -ret; 1267db8dac20SDavid S. Miller } 1268db8dac20SDavid S. Miller } 1269db8dac20SDavid S. Miller 1270db8dac20SDavid S. Miller /* FALLTHROUGH -- it's a UDP Packet */ 1271db8dac20SDavid S. Miller } 1272db8dac20SDavid S. Miller 1273db8dac20SDavid S. Miller /* 1274db8dac20SDavid S. Miller * UDP-Lite specific tests, ignored on UDP sockets 1275db8dac20SDavid S. Miller */ 1276db8dac20SDavid S. Miller if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1277db8dac20SDavid S. Miller 1278db8dac20SDavid S. Miller /* 1279db8dac20SDavid S. Miller * MIB statistics other than incrementing the error count are 1280db8dac20SDavid S. Miller * disabled for the following two types of errors: these depend 1281db8dac20SDavid S. Miller * on the application settings, not on the functioning of the 1282db8dac20SDavid S. Miller * protocol stack as such. 1283db8dac20SDavid S. Miller * 1284db8dac20SDavid S. Miller * RFC 3828 here recommends (sec 3.3): "There should also be a 1285db8dac20SDavid S. Miller * way ... to ... at least let the receiving application block 1286db8dac20SDavid S. Miller * delivery of packets with coverage values less than a value 1287db8dac20SDavid S. Miller * provided by the application." 1288db8dac20SDavid S. Miller */ 1289db8dac20SDavid S. Miller if (up->pcrlen == 0) { /* full coverage was set */ 1290db8dac20SDavid S. Miller LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1291db8dac20SDavid S. Miller "%d while full coverage %d requested\n", 1292db8dac20SDavid S. Miller UDP_SKB_CB(skb)->cscov, skb->len); 1293db8dac20SDavid S. Miller goto drop; 1294db8dac20SDavid S. Miller } 1295db8dac20SDavid S. Miller /* The next case involves violating the min. coverage requested 1296db8dac20SDavid S. Miller * by the receiver. This is subtle: if receiver wants x and x is 1297db8dac20SDavid S. Miller * greater than the buffersize/MTU then receiver will complain 1298db8dac20SDavid S. Miller * that it wants x while sender emits packets of smaller size y. 1299db8dac20SDavid S. Miller * Therefore the above ...()->partial_cov statement is essential. 1300db8dac20SDavid S. Miller */ 1301db8dac20SDavid S. Miller if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1302db8dac20SDavid S. Miller LIMIT_NETDEBUG(KERN_WARNING 1303db8dac20SDavid S. Miller "UDPLITE: coverage %d too small, need min %d\n", 1304db8dac20SDavid S. Miller UDP_SKB_CB(skb)->cscov, up->pcrlen); 1305db8dac20SDavid S. Miller goto drop; 1306db8dac20SDavid S. Miller } 1307db8dac20SDavid S. Miller } 1308db8dac20SDavid S. Miller 1309db8dac20SDavid S. Miller if (sk->sk_filter) { 1310db8dac20SDavid S. Miller if (udp_lib_checksum_complete(skb)) 1311db8dac20SDavid S. Miller goto drop; 1312db8dac20SDavid S. Miller } 1313db8dac20SDavid S. Miller 131493821778SHerbert Xu rc = 0; 1315db8dac20SDavid S. Miller 131693821778SHerbert Xu bh_lock_sock(sk); 131793821778SHerbert Xu if (!sock_owned_by_user(sk)) 131893821778SHerbert Xu rc = __udp_queue_rcv_skb(sk, skb); 131993821778SHerbert Xu else 132093821778SHerbert Xu sk_add_backlog(sk, skb); 132193821778SHerbert Xu bh_unlock_sock(sk); 132293821778SHerbert Xu 132393821778SHerbert Xu return rc; 1324db8dac20SDavid S. Miller 1325db8dac20SDavid S. Miller drop: 13260283328eSPavel Emelyanov UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 13278edf19c2SEric Dumazet atomic_inc(&sk->sk_drops); 1328db8dac20SDavid S. Miller kfree_skb(skb); 1329db8dac20SDavid S. Miller return -1; 1330db8dac20SDavid S. Miller } 1331db8dac20SDavid S. Miller 1332db8dac20SDavid S. Miller /* 1333db8dac20SDavid S. Miller * Multicasts and broadcasts go to each listener. 1334db8dac20SDavid S. Miller * 1335db8dac20SDavid S. Miller * Note: called only from the BH handler context, 1336db8dac20SDavid S. Miller * so we don't need to lock the hashes. 1337db8dac20SDavid S. Miller */ 1338e3163493SPavel Emelyanov static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1339db8dac20SDavid S. Miller struct udphdr *uh, 1340db8dac20SDavid S. Miller __be32 saddr, __be32 daddr, 1341645ca708SEric Dumazet struct udp_table *udptable) 1342db8dac20SDavid S. Miller { 1343db8dac20SDavid S. Miller struct sock *sk; 1344f86dcc5aSEric Dumazet struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1345db8dac20SDavid S. Miller int dif; 1346db8dac20SDavid S. Miller 1347645ca708SEric Dumazet spin_lock(&hslot->lock); 134888ab1932SEric Dumazet sk = sk_nulls_head(&hslot->head); 1349db8dac20SDavid S. Miller dif = skb->dev->ifindex; 1350920a4611SEric Dumazet sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1351db8dac20SDavid S. Miller if (sk) { 1352db8dac20SDavid S. Miller struct sock *sknext = NULL; 1353db8dac20SDavid S. Miller 1354db8dac20SDavid S. Miller do { 1355db8dac20SDavid S. Miller struct sk_buff *skb1 = skb; 1356db8dac20SDavid S. Miller 135788ab1932SEric Dumazet sknext = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1358920a4611SEric Dumazet daddr, uh->source, saddr, 1359920a4611SEric Dumazet dif); 1360db8dac20SDavid S. Miller if (sknext) 1361db8dac20SDavid S. Miller skb1 = skb_clone(skb, GFP_ATOMIC); 1362db8dac20SDavid S. Miller 1363db8dac20SDavid S. Miller if (skb1) { 136493821778SHerbert Xu int ret = udp_queue_rcv_skb(sk, skb1); 1365db8dac20SDavid S. Miller if (ret > 0) 1366db8dac20SDavid S. Miller /* we should probably re-process instead 1367db8dac20SDavid S. Miller * of dropping packets here. */ 1368db8dac20SDavid S. Miller kfree_skb(skb1); 1369db8dac20SDavid S. Miller } 1370db8dac20SDavid S. Miller sk = sknext; 1371db8dac20SDavid S. Miller } while (sknext); 1372db8dac20SDavid S. Miller } else 1373ead2ceb0SNeil Horman consume_skb(skb); 1374645ca708SEric Dumazet spin_unlock(&hslot->lock); 1375db8dac20SDavid S. Miller return 0; 1376db8dac20SDavid S. Miller } 1377db8dac20SDavid S. Miller 1378db8dac20SDavid S. Miller /* Initialize UDP checksum. If exited with zero value (success), 1379db8dac20SDavid S. Miller * CHECKSUM_UNNECESSARY means, that no more checks are required. 1380db8dac20SDavid S. Miller * Otherwise, csum completion requires chacksumming packet body, 1381db8dac20SDavid S. Miller * including udp header and folding it to skb->csum. 1382db8dac20SDavid S. Miller */ 1383db8dac20SDavid S. Miller static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1384db8dac20SDavid S. Miller int proto) 1385db8dac20SDavid S. Miller { 1386db8dac20SDavid S. Miller const struct iphdr *iph; 1387db8dac20SDavid S. Miller int err; 1388db8dac20SDavid S. Miller 1389db8dac20SDavid S. Miller UDP_SKB_CB(skb)->partial_cov = 0; 1390db8dac20SDavid S. Miller UDP_SKB_CB(skb)->cscov = skb->len; 1391db8dac20SDavid S. Miller 1392db8dac20SDavid S. Miller if (proto == IPPROTO_UDPLITE) { 1393db8dac20SDavid S. Miller err = udplite_checksum_init(skb, uh); 1394db8dac20SDavid S. Miller if (err) 1395db8dac20SDavid S. Miller return err; 1396db8dac20SDavid S. Miller } 1397db8dac20SDavid S. Miller 1398db8dac20SDavid S. Miller iph = ip_hdr(skb); 1399db8dac20SDavid S. Miller if (uh->check == 0) { 1400db8dac20SDavid S. Miller skb->ip_summed = CHECKSUM_UNNECESSARY; 1401db8dac20SDavid S. Miller } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1402db8dac20SDavid S. Miller if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1403db8dac20SDavid S. Miller proto, skb->csum)) 1404db8dac20SDavid S. Miller skb->ip_summed = CHECKSUM_UNNECESSARY; 1405db8dac20SDavid S. Miller } 1406db8dac20SDavid S. Miller if (!skb_csum_unnecessary(skb)) 1407db8dac20SDavid S. Miller skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1408db8dac20SDavid S. Miller skb->len, proto, 0); 1409db8dac20SDavid S. Miller /* Probably, we should checksum udp header (it should be in cache 1410db8dac20SDavid S. Miller * in any case) and data in tiny packets (< rx copybreak). 1411db8dac20SDavid S. Miller */ 1412db8dac20SDavid S. Miller 1413db8dac20SDavid S. Miller return 0; 1414db8dac20SDavid S. Miller } 1415db8dac20SDavid S. Miller 1416db8dac20SDavid S. Miller /* 1417db8dac20SDavid S. Miller * All we need to do is get the socket, and then do a checksum. 1418db8dac20SDavid S. Miller */ 1419db8dac20SDavid S. Miller 1420645ca708SEric Dumazet int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1421db8dac20SDavid S. Miller int proto) 1422db8dac20SDavid S. Miller { 1423db8dac20SDavid S. Miller struct sock *sk; 14247b5e56f9SJesper Dangaard Brouer struct udphdr *uh; 1425db8dac20SDavid S. Miller unsigned short ulen; 1426adf30907SEric Dumazet struct rtable *rt = skb_rtable(skb); 14272783ef23SJesper Dangaard Brouer __be32 saddr, daddr; 14280283328eSPavel Emelyanov struct net *net = dev_net(skb->dev); 1429db8dac20SDavid S. Miller 1430db8dac20SDavid S. Miller /* 1431db8dac20SDavid S. Miller * Validate the packet. 1432db8dac20SDavid S. Miller */ 1433db8dac20SDavid S. Miller if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1434db8dac20SDavid S. Miller goto drop; /* No space for header. */ 1435db8dac20SDavid S. Miller 14367b5e56f9SJesper Dangaard Brouer uh = udp_hdr(skb); 1437db8dac20SDavid S. Miller ulen = ntohs(uh->len); 1438db8dac20SDavid S. Miller if (ulen > skb->len) 1439db8dac20SDavid S. Miller goto short_packet; 1440db8dac20SDavid S. Miller 1441db8dac20SDavid S. Miller if (proto == IPPROTO_UDP) { 1442db8dac20SDavid S. Miller /* UDP validates ulen. */ 1443db8dac20SDavid S. Miller if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1444db8dac20SDavid S. Miller goto short_packet; 1445db8dac20SDavid S. Miller uh = udp_hdr(skb); 1446db8dac20SDavid S. Miller } 1447db8dac20SDavid S. Miller 1448db8dac20SDavid S. Miller if (udp4_csum_init(skb, uh, proto)) 1449db8dac20SDavid S. Miller goto csum_error; 1450db8dac20SDavid S. Miller 14512783ef23SJesper Dangaard Brouer saddr = ip_hdr(skb)->saddr; 14522783ef23SJesper Dangaard Brouer daddr = ip_hdr(skb)->daddr; 14532783ef23SJesper Dangaard Brouer 1454db8dac20SDavid S. Miller if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1455e3163493SPavel Emelyanov return __udp4_lib_mcast_deliver(net, skb, uh, 1456e3163493SPavel Emelyanov saddr, daddr, udptable); 1457db8dac20SDavid S. Miller 1458607c4aafSKOVACS Krisztian sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1459db8dac20SDavid S. Miller 1460db8dac20SDavid S. Miller if (sk != NULL) { 146193821778SHerbert Xu int ret = udp_queue_rcv_skb(sk, skb); 1462db8dac20SDavid S. Miller sock_put(sk); 1463db8dac20SDavid S. Miller 1464db8dac20SDavid S. Miller /* a return value > 0 means to resubmit the input, but 1465db8dac20SDavid S. Miller * it wants the return to be -protocol, or 0 1466db8dac20SDavid S. Miller */ 1467db8dac20SDavid S. Miller if (ret > 0) 1468db8dac20SDavid S. Miller return -ret; 1469db8dac20SDavid S. Miller return 0; 1470db8dac20SDavid S. Miller } 1471db8dac20SDavid S. Miller 1472db8dac20SDavid S. Miller if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1473db8dac20SDavid S. Miller goto drop; 1474db8dac20SDavid S. Miller nf_reset(skb); 1475db8dac20SDavid S. Miller 1476db8dac20SDavid S. Miller /* No socket. Drop packet silently, if checksum is wrong */ 1477db8dac20SDavid S. Miller if (udp_lib_checksum_complete(skb)) 1478db8dac20SDavid S. Miller goto csum_error; 1479db8dac20SDavid S. Miller 14800283328eSPavel Emelyanov UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1481db8dac20SDavid S. Miller icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1482db8dac20SDavid S. Miller 1483db8dac20SDavid S. Miller /* 1484db8dac20SDavid S. Miller * Hmm. We got an UDP packet to a port to which we 1485db8dac20SDavid S. Miller * don't wanna listen. Ignore it. 1486db8dac20SDavid S. Miller */ 1487db8dac20SDavid S. Miller kfree_skb(skb); 1488db8dac20SDavid S. Miller return 0; 1489db8dac20SDavid S. Miller 1490db8dac20SDavid S. Miller short_packet: 1491673d57e7SHarvey Harrison LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1492db8dac20SDavid S. Miller proto == IPPROTO_UDPLITE ? "-Lite" : "", 1493673d57e7SHarvey Harrison &saddr, 1494db8dac20SDavid S. Miller ntohs(uh->source), 1495db8dac20SDavid S. Miller ulen, 1496db8dac20SDavid S. Miller skb->len, 1497673d57e7SHarvey Harrison &daddr, 1498db8dac20SDavid S. Miller ntohs(uh->dest)); 1499db8dac20SDavid S. Miller goto drop; 1500db8dac20SDavid S. Miller 1501db8dac20SDavid S. Miller csum_error: 1502db8dac20SDavid S. Miller /* 1503db8dac20SDavid S. Miller * RFC1122: OK. Discards the bad packet silently (as far as 1504db8dac20SDavid S. Miller * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1505db8dac20SDavid S. Miller */ 1506673d57e7SHarvey Harrison LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1507db8dac20SDavid S. Miller proto == IPPROTO_UDPLITE ? "-Lite" : "", 1508673d57e7SHarvey Harrison &saddr, 1509db8dac20SDavid S. Miller ntohs(uh->source), 1510673d57e7SHarvey Harrison &daddr, 1511db8dac20SDavid S. Miller ntohs(uh->dest), 1512db8dac20SDavid S. Miller ulen); 1513db8dac20SDavid S. Miller drop: 15140283328eSPavel Emelyanov UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1515db8dac20SDavid S. Miller kfree_skb(skb); 1516db8dac20SDavid S. Miller return 0; 1517db8dac20SDavid S. Miller } 1518db8dac20SDavid S. Miller 1519db8dac20SDavid S. Miller int udp_rcv(struct sk_buff *skb) 1520db8dac20SDavid S. Miller { 1521645ca708SEric Dumazet return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1522db8dac20SDavid S. Miller } 1523db8dac20SDavid S. Miller 15247d06b2e0SBrian Haley void udp_destroy_sock(struct sock *sk) 1525db8dac20SDavid S. Miller { 1526db8dac20SDavid S. Miller lock_sock(sk); 1527db8dac20SDavid S. Miller udp_flush_pending_frames(sk); 1528db8dac20SDavid S. Miller release_sock(sk); 1529db8dac20SDavid S. Miller } 1530db8dac20SDavid S. Miller 15311da177e4SLinus Torvalds /* 15321da177e4SLinus Torvalds * Socket option code for UDP 15331da177e4SLinus Torvalds */ 15344c0a6cb0SGerrit Renker int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1535b7058842SDavid S. Miller char __user *optval, unsigned int optlen, 15364c0a6cb0SGerrit Renker int (*push_pending_frames)(struct sock *)) 15371da177e4SLinus Torvalds { 15381da177e4SLinus Torvalds struct udp_sock *up = udp_sk(sk); 15391da177e4SLinus Torvalds int val; 15401da177e4SLinus Torvalds int err = 0; 1541b2bf1e26SWang Chen int is_udplite = IS_UDPLITE(sk); 15421da177e4SLinus Torvalds 15431da177e4SLinus Torvalds if (optlen < sizeof(int)) 15441da177e4SLinus Torvalds return -EINVAL; 15451da177e4SLinus Torvalds 15461da177e4SLinus Torvalds if (get_user(val, (int __user *)optval)) 15471da177e4SLinus Torvalds return -EFAULT; 15481da177e4SLinus Torvalds 15491da177e4SLinus Torvalds switch (optname) { 15501da177e4SLinus Torvalds case UDP_CORK: 15511da177e4SLinus Torvalds if (val != 0) { 15521da177e4SLinus Torvalds up->corkflag = 1; 15531da177e4SLinus Torvalds } else { 15541da177e4SLinus Torvalds up->corkflag = 0; 15551da177e4SLinus Torvalds lock_sock(sk); 15564c0a6cb0SGerrit Renker (*push_pending_frames)(sk); 15571da177e4SLinus Torvalds release_sock(sk); 15581da177e4SLinus Torvalds } 15591da177e4SLinus Torvalds break; 15601da177e4SLinus Torvalds 15611da177e4SLinus Torvalds case UDP_ENCAP: 15621da177e4SLinus Torvalds switch (val) { 15631da177e4SLinus Torvalds case 0: 15641da177e4SLinus Torvalds case UDP_ENCAP_ESPINUDP: 15651da177e4SLinus Torvalds case UDP_ENCAP_ESPINUDP_NON_IKE: 1566067b207bSJames Chapman up->encap_rcv = xfrm4_udp_encap_rcv; 1567067b207bSJames Chapman /* FALLTHROUGH */ 1568342f0234SJames Chapman case UDP_ENCAP_L2TPINUDP: 15691da177e4SLinus Torvalds up->encap_type = val; 15701da177e4SLinus Torvalds break; 15711da177e4SLinus Torvalds default: 15721da177e4SLinus Torvalds err = -ENOPROTOOPT; 15731da177e4SLinus Torvalds break; 15741da177e4SLinus Torvalds } 15751da177e4SLinus Torvalds break; 15761da177e4SLinus Torvalds 1577ba4e58ecSGerrit Renker /* 1578ba4e58ecSGerrit Renker * UDP-Lite's partial checksum coverage (RFC 3828). 1579ba4e58ecSGerrit Renker */ 1580ba4e58ecSGerrit Renker /* The sender sets actual checksum coverage length via this option. 1581ba4e58ecSGerrit Renker * The case coverage > packet length is handled by send module. */ 1582ba4e58ecSGerrit Renker case UDPLITE_SEND_CSCOV: 1583b2bf1e26SWang Chen if (!is_udplite) /* Disable the option on UDP sockets */ 1584ba4e58ecSGerrit Renker return -ENOPROTOOPT; 1585ba4e58ecSGerrit Renker if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1586ba4e58ecSGerrit Renker val = 8; 158747112e25SGerrit Renker else if (val > USHORT_MAX) 158847112e25SGerrit Renker val = USHORT_MAX; 1589ba4e58ecSGerrit Renker up->pcslen = val; 1590ba4e58ecSGerrit Renker up->pcflag |= UDPLITE_SEND_CC; 1591ba4e58ecSGerrit Renker break; 1592ba4e58ecSGerrit Renker 1593ba4e58ecSGerrit Renker /* The receiver specifies a minimum checksum coverage value. To make 1594ba4e58ecSGerrit Renker * sense, this should be set to at least 8 (as done below). If zero is 1595ba4e58ecSGerrit Renker * used, this again means full checksum coverage. */ 1596ba4e58ecSGerrit Renker case UDPLITE_RECV_CSCOV: 1597b2bf1e26SWang Chen if (!is_udplite) /* Disable the option on UDP sockets */ 1598ba4e58ecSGerrit Renker return -ENOPROTOOPT; 1599ba4e58ecSGerrit Renker if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1600ba4e58ecSGerrit Renker val = 8; 160147112e25SGerrit Renker else if (val > USHORT_MAX) 160247112e25SGerrit Renker val = USHORT_MAX; 1603ba4e58ecSGerrit Renker up->pcrlen = val; 1604ba4e58ecSGerrit Renker up->pcflag |= UDPLITE_RECV_CC; 1605ba4e58ecSGerrit Renker break; 1606ba4e58ecSGerrit Renker 16071da177e4SLinus Torvalds default: 16081da177e4SLinus Torvalds err = -ENOPROTOOPT; 16091da177e4SLinus Torvalds break; 16106516c655SStephen Hemminger } 16111da177e4SLinus Torvalds 16121da177e4SLinus Torvalds return err; 16131da177e4SLinus Torvalds } 1614c482c568SEric Dumazet EXPORT_SYMBOL(udp_lib_setsockopt); 16151da177e4SLinus Torvalds 1616db8dac20SDavid S. Miller int udp_setsockopt(struct sock *sk, int level, int optname, 1617b7058842SDavid S. Miller char __user *optval, unsigned int optlen) 1618db8dac20SDavid S. Miller { 1619db8dac20SDavid S. Miller if (level == SOL_UDP || level == SOL_UDPLITE) 1620db8dac20SDavid S. Miller return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1621db8dac20SDavid S. Miller udp_push_pending_frames); 1622db8dac20SDavid S. Miller return ip_setsockopt(sk, level, optname, optval, optlen); 1623db8dac20SDavid S. Miller } 1624db8dac20SDavid S. Miller 1625db8dac20SDavid S. Miller #ifdef CONFIG_COMPAT 1626db8dac20SDavid S. Miller int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1627b7058842SDavid S. Miller char __user *optval, unsigned int optlen) 1628db8dac20SDavid S. Miller { 1629db8dac20SDavid S. Miller if (level == SOL_UDP || level == SOL_UDPLITE) 1630db8dac20SDavid S. Miller return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1631db8dac20SDavid S. Miller udp_push_pending_frames); 1632db8dac20SDavid S. Miller return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1633db8dac20SDavid S. Miller } 1634db8dac20SDavid S. Miller #endif 1635db8dac20SDavid S. Miller 16364c0a6cb0SGerrit Renker int udp_lib_getsockopt(struct sock *sk, int level, int optname, 16371da177e4SLinus Torvalds char __user *optval, int __user *optlen) 16381da177e4SLinus Torvalds { 16391da177e4SLinus Torvalds struct udp_sock *up = udp_sk(sk); 16401da177e4SLinus Torvalds int val, len; 16411da177e4SLinus Torvalds 16421da177e4SLinus Torvalds if (get_user(len, optlen)) 16431da177e4SLinus Torvalds return -EFAULT; 16441da177e4SLinus Torvalds 16451da177e4SLinus Torvalds len = min_t(unsigned int, len, sizeof(int)); 16461da177e4SLinus Torvalds 16471da177e4SLinus Torvalds if (len < 0) 16481da177e4SLinus Torvalds return -EINVAL; 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds switch (optname) { 16511da177e4SLinus Torvalds case UDP_CORK: 16521da177e4SLinus Torvalds val = up->corkflag; 16531da177e4SLinus Torvalds break; 16541da177e4SLinus Torvalds 16551da177e4SLinus Torvalds case UDP_ENCAP: 16561da177e4SLinus Torvalds val = up->encap_type; 16571da177e4SLinus Torvalds break; 16581da177e4SLinus Torvalds 1659ba4e58ecSGerrit Renker /* The following two cannot be changed on UDP sockets, the return is 1660ba4e58ecSGerrit Renker * always 0 (which corresponds to the full checksum coverage of UDP). */ 1661ba4e58ecSGerrit Renker case UDPLITE_SEND_CSCOV: 1662ba4e58ecSGerrit Renker val = up->pcslen; 1663ba4e58ecSGerrit Renker break; 1664ba4e58ecSGerrit Renker 1665ba4e58ecSGerrit Renker case UDPLITE_RECV_CSCOV: 1666ba4e58ecSGerrit Renker val = up->pcrlen; 1667ba4e58ecSGerrit Renker break; 1668ba4e58ecSGerrit Renker 16691da177e4SLinus Torvalds default: 16701da177e4SLinus Torvalds return -ENOPROTOOPT; 16716516c655SStephen Hemminger } 16721da177e4SLinus Torvalds 16731da177e4SLinus Torvalds if (put_user(len, optlen)) 16741da177e4SLinus Torvalds return -EFAULT; 16751da177e4SLinus Torvalds if (copy_to_user(optval, &val, len)) 16761da177e4SLinus Torvalds return -EFAULT; 16771da177e4SLinus Torvalds return 0; 16781da177e4SLinus Torvalds } 1679c482c568SEric Dumazet EXPORT_SYMBOL(udp_lib_getsockopt); 16801da177e4SLinus Torvalds 1681db8dac20SDavid S. Miller int udp_getsockopt(struct sock *sk, int level, int optname, 1682db8dac20SDavid S. Miller char __user *optval, int __user *optlen) 1683db8dac20SDavid S. Miller { 1684db8dac20SDavid S. Miller if (level == SOL_UDP || level == SOL_UDPLITE) 1685db8dac20SDavid S. Miller return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1686db8dac20SDavid S. Miller return ip_getsockopt(sk, level, optname, optval, optlen); 1687db8dac20SDavid S. Miller } 1688db8dac20SDavid S. Miller 1689db8dac20SDavid S. Miller #ifdef CONFIG_COMPAT 1690db8dac20SDavid S. Miller int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1691db8dac20SDavid S. Miller char __user *optval, int __user *optlen) 1692db8dac20SDavid S. Miller { 1693db8dac20SDavid S. Miller if (level == SOL_UDP || level == SOL_UDPLITE) 1694db8dac20SDavid S. Miller return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1695db8dac20SDavid S. Miller return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1696db8dac20SDavid S. Miller } 1697db8dac20SDavid S. Miller #endif 16981da177e4SLinus Torvalds /** 16991da177e4SLinus Torvalds * udp_poll - wait for a UDP event. 17001da177e4SLinus Torvalds * @file - file struct 17011da177e4SLinus Torvalds * @sock - socket 17021da177e4SLinus Torvalds * @wait - poll table 17031da177e4SLinus Torvalds * 17041da177e4SLinus Torvalds * This is same as datagram poll, except for the special case of 17051da177e4SLinus Torvalds * blocking sockets. If application is using a blocking fd 17061da177e4SLinus Torvalds * and a packet with checksum error is in the queue; 17071da177e4SLinus Torvalds * then it could get return from select indicating data available 17081da177e4SLinus Torvalds * but then block when reading it. Add special case code 17091da177e4SLinus Torvalds * to work around these arguably broken applications. 17101da177e4SLinus Torvalds */ 17111da177e4SLinus Torvalds unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 17121da177e4SLinus Torvalds { 17131da177e4SLinus Torvalds unsigned int mask = datagram_poll(file, sock, wait); 17141da177e4SLinus Torvalds struct sock *sk = sock->sk; 17151da177e4SLinus Torvalds 17161da177e4SLinus Torvalds /* Check for false positives due to checksum errors */ 171785584672SEric Dumazet if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 171885584672SEric Dumazet !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 17191da177e4SLinus Torvalds mask &= ~(POLLIN | POLLRDNORM); 17201da177e4SLinus Torvalds 17211da177e4SLinus Torvalds return mask; 17221da177e4SLinus Torvalds 17231da177e4SLinus Torvalds } 1724c482c568SEric Dumazet EXPORT_SYMBOL(udp_poll); 17251da177e4SLinus Torvalds 1726db8dac20SDavid S. Miller struct proto udp_prot = { 1727db8dac20SDavid S. Miller .name = "UDP", 1728db8dac20SDavid S. Miller .owner = THIS_MODULE, 1729db8dac20SDavid S. Miller .close = udp_lib_close, 1730db8dac20SDavid S. Miller .connect = ip4_datagram_connect, 1731db8dac20SDavid S. Miller .disconnect = udp_disconnect, 1732db8dac20SDavid S. Miller .ioctl = udp_ioctl, 1733db8dac20SDavid S. Miller .destroy = udp_destroy_sock, 1734db8dac20SDavid S. Miller .setsockopt = udp_setsockopt, 1735db8dac20SDavid S. Miller .getsockopt = udp_getsockopt, 1736db8dac20SDavid S. Miller .sendmsg = udp_sendmsg, 1737db8dac20SDavid S. Miller .recvmsg = udp_recvmsg, 1738db8dac20SDavid S. Miller .sendpage = udp_sendpage, 173993821778SHerbert Xu .backlog_rcv = __udp_queue_rcv_skb, 1740db8dac20SDavid S. Miller .hash = udp_lib_hash, 1741db8dac20SDavid S. Miller .unhash = udp_lib_unhash, 1742db8dac20SDavid S. Miller .get_port = udp_v4_get_port, 1743db8dac20SDavid S. Miller .memory_allocated = &udp_memory_allocated, 1744db8dac20SDavid S. Miller .sysctl_mem = sysctl_udp_mem, 1745db8dac20SDavid S. Miller .sysctl_wmem = &sysctl_udp_wmem_min, 1746db8dac20SDavid S. Miller .sysctl_rmem = &sysctl_udp_rmem_min, 1747db8dac20SDavid S. Miller .obj_size = sizeof(struct udp_sock), 1748271b72c7SEric Dumazet .slab_flags = SLAB_DESTROY_BY_RCU, 1749645ca708SEric Dumazet .h.udp_table = &udp_table, 1750db8dac20SDavid S. Miller #ifdef CONFIG_COMPAT 1751db8dac20SDavid S. Miller .compat_setsockopt = compat_udp_setsockopt, 1752db8dac20SDavid S. Miller .compat_getsockopt = compat_udp_getsockopt, 1753db8dac20SDavid S. Miller #endif 1754db8dac20SDavid S. Miller }; 1755c482c568SEric Dumazet EXPORT_SYMBOL(udp_prot); 17561da177e4SLinus Torvalds 17571da177e4SLinus Torvalds /* ------------------------------------------------------------------------ */ 17581da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 17591da177e4SLinus Torvalds 1760645ca708SEric Dumazet static struct sock *udp_get_first(struct seq_file *seq, int start) 17611da177e4SLinus Torvalds { 17621da177e4SLinus Torvalds struct sock *sk; 17631da177e4SLinus Torvalds struct udp_iter_state *state = seq->private; 17646f191efeSDenis V. Lunev struct net *net = seq_file_net(seq); 17651da177e4SLinus Torvalds 1766f86dcc5aSEric Dumazet for (state->bucket = start; state->bucket <= state->udp_table->mask; 1767f86dcc5aSEric Dumazet ++state->bucket) { 176888ab1932SEric Dumazet struct hlist_nulls_node *node; 1769645ca708SEric Dumazet struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 1770f86dcc5aSEric Dumazet 1771f86dcc5aSEric Dumazet if (hlist_nulls_empty(&hslot->head)) 1772f86dcc5aSEric Dumazet continue; 1773f86dcc5aSEric Dumazet 1774645ca708SEric Dumazet spin_lock_bh(&hslot->lock); 177588ab1932SEric Dumazet sk_nulls_for_each(sk, node, &hslot->head) { 1776878628fbSYOSHIFUJI Hideaki if (!net_eq(sock_net(sk), net)) 1777a91275efSDaniel Lezcano continue; 17781da177e4SLinus Torvalds if (sk->sk_family == state->family) 17791da177e4SLinus Torvalds goto found; 17801da177e4SLinus Torvalds } 1781645ca708SEric Dumazet spin_unlock_bh(&hslot->lock); 17821da177e4SLinus Torvalds } 17831da177e4SLinus Torvalds sk = NULL; 17841da177e4SLinus Torvalds found: 17851da177e4SLinus Torvalds return sk; 17861da177e4SLinus Torvalds } 17871da177e4SLinus Torvalds 17881da177e4SLinus Torvalds static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 17891da177e4SLinus Torvalds { 17901da177e4SLinus Torvalds struct udp_iter_state *state = seq->private; 17916f191efeSDenis V. Lunev struct net *net = seq_file_net(seq); 17921da177e4SLinus Torvalds 17931da177e4SLinus Torvalds do { 179488ab1932SEric Dumazet sk = sk_nulls_next(sk); 1795878628fbSYOSHIFUJI Hideaki } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 17961da177e4SLinus Torvalds 1797645ca708SEric Dumazet if (!sk) { 1798f86dcc5aSEric Dumazet if (state->bucket <= state->udp_table->mask) 1799f52b5054SEric Dumazet spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1800645ca708SEric Dumazet return udp_get_first(seq, state->bucket + 1); 18011da177e4SLinus Torvalds } 18021da177e4SLinus Torvalds return sk; 18031da177e4SLinus Torvalds } 18041da177e4SLinus Torvalds 18051da177e4SLinus Torvalds static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 18061da177e4SLinus Torvalds { 1807645ca708SEric Dumazet struct sock *sk = udp_get_first(seq, 0); 18081da177e4SLinus Torvalds 18091da177e4SLinus Torvalds if (sk) 18101da177e4SLinus Torvalds while (pos && (sk = udp_get_next(seq, sk)) != NULL) 18111da177e4SLinus Torvalds --pos; 18121da177e4SLinus Torvalds return pos ? NULL : sk; 18131da177e4SLinus Torvalds } 18141da177e4SLinus Torvalds 18151da177e4SLinus Torvalds static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 18161da177e4SLinus Torvalds { 181730842f29SVitaly Mayatskikh struct udp_iter_state *state = seq->private; 1818f86dcc5aSEric Dumazet state->bucket = MAX_UDP_PORTS; 181930842f29SVitaly Mayatskikh 1820b50660f1SYOSHIFUJI Hideaki return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 18211da177e4SLinus Torvalds } 18221da177e4SLinus Torvalds 18231da177e4SLinus Torvalds static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 18241da177e4SLinus Torvalds { 18251da177e4SLinus Torvalds struct sock *sk; 18261da177e4SLinus Torvalds 1827b50660f1SYOSHIFUJI Hideaki if (v == SEQ_START_TOKEN) 18281da177e4SLinus Torvalds sk = udp_get_idx(seq, 0); 18291da177e4SLinus Torvalds else 18301da177e4SLinus Torvalds sk = udp_get_next(seq, v); 18311da177e4SLinus Torvalds 18321da177e4SLinus Torvalds ++*pos; 18331da177e4SLinus Torvalds return sk; 18341da177e4SLinus Torvalds } 18351da177e4SLinus Torvalds 18361da177e4SLinus Torvalds static void udp_seq_stop(struct seq_file *seq, void *v) 18371da177e4SLinus Torvalds { 1838645ca708SEric Dumazet struct udp_iter_state *state = seq->private; 1839645ca708SEric Dumazet 1840f86dcc5aSEric Dumazet if (state->bucket <= state->udp_table->mask) 1841645ca708SEric Dumazet spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 18421da177e4SLinus Torvalds } 18431da177e4SLinus Torvalds 18441da177e4SLinus Torvalds static int udp_seq_open(struct inode *inode, struct file *file) 18451da177e4SLinus Torvalds { 18461da177e4SLinus Torvalds struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1847a2be75c1SDenis V. Lunev struct udp_iter_state *s; 1848a2be75c1SDenis V. Lunev int err; 18491da177e4SLinus Torvalds 1850a2be75c1SDenis V. Lunev err = seq_open_net(inode, file, &afinfo->seq_ops, 1851a2be75c1SDenis V. Lunev sizeof(struct udp_iter_state)); 1852a2be75c1SDenis V. Lunev if (err < 0) 1853a2be75c1SDenis V. Lunev return err; 1854a91275efSDaniel Lezcano 1855a2be75c1SDenis V. Lunev s = ((struct seq_file *)file->private_data)->private; 18561da177e4SLinus Torvalds s->family = afinfo->family; 1857645ca708SEric Dumazet s->udp_table = afinfo->udp_table; 1858a2be75c1SDenis V. Lunev return err; 1859a91275efSDaniel Lezcano } 1860a91275efSDaniel Lezcano 18611da177e4SLinus Torvalds /* ------------------------------------------------------------------------ */ 18620c96d8c5SDaniel Lezcano int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 18631da177e4SLinus Torvalds { 18641da177e4SLinus Torvalds struct proc_dir_entry *p; 18651da177e4SLinus Torvalds int rc = 0; 18661da177e4SLinus Torvalds 18673ba9441bSDenis V. Lunev afinfo->seq_fops.open = udp_seq_open; 18683ba9441bSDenis V. Lunev afinfo->seq_fops.read = seq_read; 18693ba9441bSDenis V. Lunev afinfo->seq_fops.llseek = seq_lseek; 18703ba9441bSDenis V. Lunev afinfo->seq_fops.release = seq_release_net; 18711da177e4SLinus Torvalds 1872dda61925SDenis V. Lunev afinfo->seq_ops.start = udp_seq_start; 1873dda61925SDenis V. Lunev afinfo->seq_ops.next = udp_seq_next; 1874dda61925SDenis V. Lunev afinfo->seq_ops.stop = udp_seq_stop; 1875dda61925SDenis V. Lunev 187684841c3cSDenis V. Lunev p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 187784841c3cSDenis V. Lunev &afinfo->seq_fops, afinfo); 187884841c3cSDenis V. Lunev if (!p) 18791da177e4SLinus Torvalds rc = -ENOMEM; 18801da177e4SLinus Torvalds return rc; 18811da177e4SLinus Torvalds } 1882c482c568SEric Dumazet EXPORT_SYMBOL(udp_proc_register); 18831da177e4SLinus Torvalds 18840c96d8c5SDaniel Lezcano void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 18851da177e4SLinus Torvalds { 18860c96d8c5SDaniel Lezcano proc_net_remove(net, afinfo->name); 18871da177e4SLinus Torvalds } 1888c482c568SEric Dumazet EXPORT_SYMBOL(udp_proc_unregister); 1889db8dac20SDavid S. Miller 1890db8dac20SDavid S. Miller /* ------------------------------------------------------------------------ */ 18915e659e4cSPavel Emelyanov static void udp4_format_sock(struct sock *sp, struct seq_file *f, 18925e659e4cSPavel Emelyanov int bucket, int *len) 1893db8dac20SDavid S. Miller { 1894db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sp); 1895c720c7e8SEric Dumazet __be32 dest = inet->inet_daddr; 1896c720c7e8SEric Dumazet __be32 src = inet->inet_rcv_saddr; 1897c720c7e8SEric Dumazet __u16 destp = ntohs(inet->inet_dport); 1898c720c7e8SEric Dumazet __u16 srcp = ntohs(inet->inet_sport); 1899db8dac20SDavid S. Miller 1900f86dcc5aSEric Dumazet seq_printf(f, "%5d: %08X:%04X %08X:%04X" 1901cb61cb9bSEric Dumazet " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 1902db8dac20SDavid S. Miller bucket, src, srcp, dest, destp, sp->sk_state, 190331e6d363SEric Dumazet sk_wmem_alloc_get(sp), 190431e6d363SEric Dumazet sk_rmem_alloc_get(sp), 1905db8dac20SDavid S. Miller 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1906cb61cb9bSEric Dumazet atomic_read(&sp->sk_refcnt), sp, 1907cb61cb9bSEric Dumazet atomic_read(&sp->sk_drops), len); 1908db8dac20SDavid S. Miller } 1909db8dac20SDavid S. Miller 1910db8dac20SDavid S. Miller int udp4_seq_show(struct seq_file *seq, void *v) 1911db8dac20SDavid S. Miller { 1912db8dac20SDavid S. Miller if (v == SEQ_START_TOKEN) 1913db8dac20SDavid S. Miller seq_printf(seq, "%-127s\n", 1914db8dac20SDavid S. Miller " sl local_address rem_address st tx_queue " 1915db8dac20SDavid S. Miller "rx_queue tr tm->when retrnsmt uid timeout " 1916cb61cb9bSEric Dumazet "inode ref pointer drops"); 1917db8dac20SDavid S. Miller else { 1918db8dac20SDavid S. Miller struct udp_iter_state *state = seq->private; 19195e659e4cSPavel Emelyanov int len; 1920db8dac20SDavid S. Miller 19215e659e4cSPavel Emelyanov udp4_format_sock(v, seq, state->bucket, &len); 19225e659e4cSPavel Emelyanov seq_printf(seq, "%*s\n", 127 - len, ""); 1923db8dac20SDavid S. Miller } 1924db8dac20SDavid S. Miller return 0; 1925db8dac20SDavid S. Miller } 1926db8dac20SDavid S. Miller 1927db8dac20SDavid S. Miller /* ------------------------------------------------------------------------ */ 1928db8dac20SDavid S. Miller static struct udp_seq_afinfo udp4_seq_afinfo = { 1929db8dac20SDavid S. Miller .name = "udp", 1930db8dac20SDavid S. Miller .family = AF_INET, 1931645ca708SEric Dumazet .udp_table = &udp_table, 19324ad96d39SDenis V. Lunev .seq_fops = { 19334ad96d39SDenis V. Lunev .owner = THIS_MODULE, 19344ad96d39SDenis V. Lunev }, 1935dda61925SDenis V. Lunev .seq_ops = { 1936dda61925SDenis V. Lunev .show = udp4_seq_show, 1937dda61925SDenis V. Lunev }, 1938db8dac20SDavid S. Miller }; 1939db8dac20SDavid S. Miller 194015439febSPavel Emelyanov static int udp4_proc_init_net(struct net *net) 194115439febSPavel Emelyanov { 194215439febSPavel Emelyanov return udp_proc_register(net, &udp4_seq_afinfo); 194315439febSPavel Emelyanov } 194415439febSPavel Emelyanov 194515439febSPavel Emelyanov static void udp4_proc_exit_net(struct net *net) 194615439febSPavel Emelyanov { 194715439febSPavel Emelyanov udp_proc_unregister(net, &udp4_seq_afinfo); 194815439febSPavel Emelyanov } 194915439febSPavel Emelyanov 195015439febSPavel Emelyanov static struct pernet_operations udp4_net_ops = { 195115439febSPavel Emelyanov .init = udp4_proc_init_net, 195215439febSPavel Emelyanov .exit = udp4_proc_exit_net, 195315439febSPavel Emelyanov }; 195415439febSPavel Emelyanov 1955db8dac20SDavid S. Miller int __init udp4_proc_init(void) 1956db8dac20SDavid S. Miller { 195715439febSPavel Emelyanov return register_pernet_subsys(&udp4_net_ops); 1958db8dac20SDavid S. Miller } 1959db8dac20SDavid S. Miller 1960db8dac20SDavid S. Miller void udp4_proc_exit(void) 1961db8dac20SDavid S. Miller { 196215439febSPavel Emelyanov unregister_pernet_subsys(&udp4_net_ops); 1963db8dac20SDavid S. Miller } 19641da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 19651da177e4SLinus Torvalds 1966f86dcc5aSEric Dumazet static __initdata unsigned long uhash_entries; 1967f86dcc5aSEric Dumazet static int __init set_uhash_entries(char *str) 1968645ca708SEric Dumazet { 1969f86dcc5aSEric Dumazet if (!str) 1970f86dcc5aSEric Dumazet return 0; 1971f86dcc5aSEric Dumazet uhash_entries = simple_strtoul(str, &str, 0); 1972f86dcc5aSEric Dumazet if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 1973f86dcc5aSEric Dumazet uhash_entries = UDP_HTABLE_SIZE_MIN; 1974f86dcc5aSEric Dumazet return 1; 1975f86dcc5aSEric Dumazet } 1976f86dcc5aSEric Dumazet __setup("uhash_entries=", set_uhash_entries); 1977645ca708SEric Dumazet 1978f86dcc5aSEric Dumazet void __init udp_table_init(struct udp_table *table, const char *name) 1979f86dcc5aSEric Dumazet { 1980f86dcc5aSEric Dumazet unsigned int i; 1981f86dcc5aSEric Dumazet 1982f86dcc5aSEric Dumazet if (!CONFIG_BASE_SMALL) 1983f86dcc5aSEric Dumazet table->hash = alloc_large_system_hash(name, 1984512615b6SEric Dumazet 2 * sizeof(struct udp_hslot), 1985f86dcc5aSEric Dumazet uhash_entries, 1986f86dcc5aSEric Dumazet 21, /* one slot per 2 MB */ 1987f86dcc5aSEric Dumazet 0, 1988f86dcc5aSEric Dumazet &table->log, 1989f86dcc5aSEric Dumazet &table->mask, 1990f86dcc5aSEric Dumazet 64 * 1024); 1991f86dcc5aSEric Dumazet /* 1992f86dcc5aSEric Dumazet * Make sure hash table has the minimum size 1993f86dcc5aSEric Dumazet */ 1994f86dcc5aSEric Dumazet if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) { 1995f86dcc5aSEric Dumazet table->hash = kmalloc(UDP_HTABLE_SIZE_MIN * 1996512615b6SEric Dumazet 2 * sizeof(struct udp_hslot), GFP_KERNEL); 1997f86dcc5aSEric Dumazet if (!table->hash) 1998f86dcc5aSEric Dumazet panic(name); 1999f86dcc5aSEric Dumazet table->log = ilog2(UDP_HTABLE_SIZE_MIN); 2000f86dcc5aSEric Dumazet table->mask = UDP_HTABLE_SIZE_MIN - 1; 2001f86dcc5aSEric Dumazet } 2002512615b6SEric Dumazet table->hash2 = table->hash + (table->mask + 1); 2003f86dcc5aSEric Dumazet for (i = 0; i <= table->mask; i++) { 200488ab1932SEric Dumazet INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2005fdcc8aa9SEric Dumazet table->hash[i].count = 0; 2006645ca708SEric Dumazet spin_lock_init(&table->hash[i].lock); 2007645ca708SEric Dumazet } 2008512615b6SEric Dumazet for (i = 0; i <= table->mask; i++) { 2009512615b6SEric Dumazet INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2010512615b6SEric Dumazet table->hash2[i].count = 0; 2011512615b6SEric Dumazet spin_lock_init(&table->hash2[i].lock); 2012512615b6SEric Dumazet } 2013645ca708SEric Dumazet } 2014645ca708SEric Dumazet 201595766fffSHideo Aoki void __init udp_init(void) 201695766fffSHideo Aoki { 20178203efb3SEric Dumazet unsigned long nr_pages, limit; 201895766fffSHideo Aoki 2019f86dcc5aSEric Dumazet udp_table_init(&udp_table, "UDP"); 202095766fffSHideo Aoki /* Set the pressure threshold up by the same strategy of TCP. It is a 202195766fffSHideo Aoki * fraction of global memory that is up to 1/2 at 256 MB, decreasing 202295766fffSHideo Aoki * toward zero with the amount of memory, with a floor of 128 pages. 202395766fffSHideo Aoki */ 20248203efb3SEric Dumazet nr_pages = totalram_pages - totalhigh_pages; 20258203efb3SEric Dumazet limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 20268203efb3SEric Dumazet limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 202795766fffSHideo Aoki limit = max(limit, 128UL); 202895766fffSHideo Aoki sysctl_udp_mem[0] = limit / 4 * 3; 202995766fffSHideo Aoki sysctl_udp_mem[1] = limit; 203095766fffSHideo Aoki sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 203195766fffSHideo Aoki 203295766fffSHideo Aoki sysctl_udp_rmem_min = SK_MEM_QUANTUM; 203395766fffSHideo Aoki sysctl_udp_wmem_min = SK_MEM_QUANTUM; 203495766fffSHideo Aoki } 203595766fffSHideo Aoki 2036d7ca4cc0SSridhar Samudrala int udp4_ufo_send_check(struct sk_buff *skb) 2037d7ca4cc0SSridhar Samudrala { 2038d7ca4cc0SSridhar Samudrala const struct iphdr *iph; 2039d7ca4cc0SSridhar Samudrala struct udphdr *uh; 2040d7ca4cc0SSridhar Samudrala 2041d7ca4cc0SSridhar Samudrala if (!pskb_may_pull(skb, sizeof(*uh))) 2042d7ca4cc0SSridhar Samudrala return -EINVAL; 2043d7ca4cc0SSridhar Samudrala 2044d7ca4cc0SSridhar Samudrala iph = ip_hdr(skb); 2045d7ca4cc0SSridhar Samudrala uh = udp_hdr(skb); 2046d7ca4cc0SSridhar Samudrala 2047d7ca4cc0SSridhar Samudrala uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 2048d7ca4cc0SSridhar Samudrala IPPROTO_UDP, 0); 2049d7ca4cc0SSridhar Samudrala skb->csum_start = skb_transport_header(skb) - skb->head; 2050d7ca4cc0SSridhar Samudrala skb->csum_offset = offsetof(struct udphdr, check); 2051d7ca4cc0SSridhar Samudrala skb->ip_summed = CHECKSUM_PARTIAL; 2052d7ca4cc0SSridhar Samudrala return 0; 2053d7ca4cc0SSridhar Samudrala } 2054d7ca4cc0SSridhar Samudrala 2055d7ca4cc0SSridhar Samudrala struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features) 2056d7ca4cc0SSridhar Samudrala { 2057d7ca4cc0SSridhar Samudrala struct sk_buff *segs = ERR_PTR(-EINVAL); 2058d7ca4cc0SSridhar Samudrala unsigned int mss; 2059d7ca4cc0SSridhar Samudrala int offset; 2060d7ca4cc0SSridhar Samudrala __wsum csum; 2061d7ca4cc0SSridhar Samudrala 2062d7ca4cc0SSridhar Samudrala mss = skb_shinfo(skb)->gso_size; 2063d7ca4cc0SSridhar Samudrala if (unlikely(skb->len <= mss)) 2064d7ca4cc0SSridhar Samudrala goto out; 2065d7ca4cc0SSridhar Samudrala 2066d7ca4cc0SSridhar Samudrala if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 2067d7ca4cc0SSridhar Samudrala /* Packet is from an untrusted source, reset gso_segs. */ 2068d7ca4cc0SSridhar Samudrala int type = skb_shinfo(skb)->gso_type; 2069d7ca4cc0SSridhar Samudrala 2070d7ca4cc0SSridhar Samudrala if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || 2071d7ca4cc0SSridhar Samudrala !(type & (SKB_GSO_UDP)))) 2072d7ca4cc0SSridhar Samudrala goto out; 2073d7ca4cc0SSridhar Samudrala 2074d7ca4cc0SSridhar Samudrala skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 2075d7ca4cc0SSridhar Samudrala 2076d7ca4cc0SSridhar Samudrala segs = NULL; 2077d7ca4cc0SSridhar Samudrala goto out; 2078d7ca4cc0SSridhar Samudrala } 2079d7ca4cc0SSridhar Samudrala 2080d7ca4cc0SSridhar Samudrala /* Do software UFO. Complete and fill in the UDP checksum as HW cannot 2081d7ca4cc0SSridhar Samudrala * do checksum of UDP packets sent as multiple IP fragments. 2082d7ca4cc0SSridhar Samudrala */ 2083d7ca4cc0SSridhar Samudrala offset = skb->csum_start - skb_headroom(skb); 2084d7ca4cc0SSridhar Samudrala csum = skb_checksum(skb, offset, skb->len - offset, 0); 2085d7ca4cc0SSridhar Samudrala offset += skb->csum_offset; 2086d7ca4cc0SSridhar Samudrala *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2087d7ca4cc0SSridhar Samudrala skb->ip_summed = CHECKSUM_NONE; 2088d7ca4cc0SSridhar Samudrala 2089d7ca4cc0SSridhar Samudrala /* Fragment the skb. IP headers of the fragments are updated in 2090d7ca4cc0SSridhar Samudrala * inet_gso_segment() 2091d7ca4cc0SSridhar Samudrala */ 2092d7ca4cc0SSridhar Samudrala segs = skb_segment(skb, features); 2093d7ca4cc0SSridhar Samudrala out: 2094d7ca4cc0SSridhar Samudrala return segs; 2095d7ca4cc0SSridhar Samudrala } 2096d7ca4cc0SSridhar Samudrala 2097