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> 985a0e3ad6STejun Heo #include <linux/slab.h> 99c752f073SArnaldo Carvalho de Melo #include <net/tcp_states.h> 1001da177e4SLinus Torvalds #include <linux/skbuff.h> 1011da177e4SLinus Torvalds #include <linux/proc_fs.h> 1021da177e4SLinus Torvalds #include <linux/seq_file.h> 103457c4cbcSEric W. Biederman #include <net/net_namespace.h> 1041da177e4SLinus Torvalds #include <net/icmp.h> 1051da177e4SLinus Torvalds #include <net/route.h> 1061da177e4SLinus Torvalds #include <net/checksum.h> 1071da177e4SLinus Torvalds #include <net/xfrm.h> 108ba4e58ecSGerrit Renker #include "udp_impl.h" 1091da177e4SLinus Torvalds 110f86dcc5aSEric Dumazet struct udp_table udp_table __read_mostly; 111645ca708SEric Dumazet EXPORT_SYMBOL(udp_table); 1121da177e4SLinus Torvalds 11395766fffSHideo Aoki int sysctl_udp_mem[3] __read_mostly; 11495766fffSHideo Aoki EXPORT_SYMBOL(sysctl_udp_mem); 115c482c568SEric Dumazet 116c482c568SEric Dumazet int sysctl_udp_rmem_min __read_mostly; 11795766fffSHideo Aoki EXPORT_SYMBOL(sysctl_udp_rmem_min); 118c482c568SEric Dumazet 119c482c568SEric Dumazet int sysctl_udp_wmem_min __read_mostly; 12095766fffSHideo Aoki EXPORT_SYMBOL(sysctl_udp_wmem_min); 12195766fffSHideo Aoki 12295766fffSHideo Aoki atomic_t udp_memory_allocated; 12395766fffSHideo Aoki EXPORT_SYMBOL(udp_memory_allocated); 12495766fffSHideo Aoki 125f86dcc5aSEric Dumazet #define MAX_UDP_PORTS 65536 126f86dcc5aSEric Dumazet #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 12798322f22SEric Dumazet 128f24d43c0SEric Dumazet static int udp_lib_lport_inuse(struct net *net, __u16 num, 129645ca708SEric Dumazet const struct udp_hslot *hslot, 13098322f22SEric Dumazet unsigned long *bitmap, 131f24d43c0SEric Dumazet struct sock *sk, 132f24d43c0SEric Dumazet int (*saddr_comp)(const struct sock *sk1, 133f86dcc5aSEric Dumazet const struct sock *sk2), 134f86dcc5aSEric Dumazet unsigned int log) 13525030a7fSGerrit Renker { 136f24d43c0SEric Dumazet struct sock *sk2; 13788ab1932SEric Dumazet struct hlist_nulls_node *node; 13825030a7fSGerrit Renker 13988ab1932SEric Dumazet sk_nulls_for_each(sk2, node, &hslot->head) 140f24d43c0SEric Dumazet if (net_eq(sock_net(sk2), net) && 141f24d43c0SEric Dumazet sk2 != sk && 142d4cada4aSEric Dumazet (bitmap || udp_sk(sk2)->udp_port_hash == num) && 143f24d43c0SEric Dumazet (!sk2->sk_reuse || !sk->sk_reuse) && 1449d4fb27dSJoe Perches (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 1459d4fb27dSJoe Perches sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 14698322f22SEric Dumazet (*saddr_comp)(sk, sk2)) { 14798322f22SEric Dumazet if (bitmap) 148d4cada4aSEric Dumazet __set_bit(udp_sk(sk2)->udp_port_hash >> log, 149d4cada4aSEric Dumazet bitmap); 15098322f22SEric Dumazet else 151fc038410SDavid S. Miller return 1; 15298322f22SEric Dumazet } 15325030a7fSGerrit Renker return 0; 15425030a7fSGerrit Renker } 15525030a7fSGerrit Renker 15630fff923SEric Dumazet /* 15730fff923SEric Dumazet * Note: we still hold spinlock of primary hash chain, so no other writer 15830fff923SEric Dumazet * can insert/delete a socket with local_port == num 15930fff923SEric Dumazet */ 16030fff923SEric Dumazet static int udp_lib_lport_inuse2(struct net *net, __u16 num, 16130fff923SEric Dumazet struct udp_hslot *hslot2, 16230fff923SEric Dumazet struct sock *sk, 16330fff923SEric Dumazet int (*saddr_comp)(const struct sock *sk1, 16430fff923SEric Dumazet const struct sock *sk2)) 16530fff923SEric Dumazet { 16630fff923SEric Dumazet struct sock *sk2; 16730fff923SEric Dumazet struct hlist_nulls_node *node; 16830fff923SEric Dumazet int res = 0; 16930fff923SEric Dumazet 17030fff923SEric Dumazet spin_lock(&hslot2->lock); 17130fff923SEric Dumazet udp_portaddr_for_each_entry(sk2, node, &hslot2->head) 17230fff923SEric Dumazet if (net_eq(sock_net(sk2), net) && 17330fff923SEric Dumazet sk2 != sk && 17430fff923SEric Dumazet (udp_sk(sk2)->udp_port_hash == num) && 17530fff923SEric Dumazet (!sk2->sk_reuse || !sk->sk_reuse) && 1769d4fb27dSJoe Perches (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 1779d4fb27dSJoe Perches sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 17830fff923SEric Dumazet (*saddr_comp)(sk, sk2)) { 17930fff923SEric Dumazet res = 1; 18030fff923SEric Dumazet break; 18130fff923SEric Dumazet } 18230fff923SEric Dumazet spin_unlock(&hslot2->lock); 18330fff923SEric Dumazet return res; 18430fff923SEric Dumazet } 18530fff923SEric Dumazet 18625030a7fSGerrit Renker /** 1876ba5a3c5SPavel Emelyanov * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 18825030a7fSGerrit Renker * 18925030a7fSGerrit Renker * @sk: socket struct in question 19025030a7fSGerrit Renker * @snum: port number to look up 191df2bc459SDavid S. Miller * @saddr_comp: AF-dependent comparison of bound local IP addresses 19230fff923SEric Dumazet * @hash2_nulladdr: AF-dependant hash value in secondary hash chains, 19330fff923SEric Dumazet * with NULL address 19425030a7fSGerrit Renker */ 1956ba5a3c5SPavel Emelyanov int udp_lib_get_port(struct sock *sk, unsigned short snum, 196df2bc459SDavid S. Miller int (*saddr_comp)(const struct sock *sk1, 19730fff923SEric Dumazet const struct sock *sk2), 19830fff923SEric Dumazet unsigned int hash2_nulladdr) 1991da177e4SLinus Torvalds { 200512615b6SEric Dumazet struct udp_hslot *hslot, *hslot2; 201645ca708SEric Dumazet struct udp_table *udptable = sk->sk_prot->h.udp_table; 20225030a7fSGerrit Renker int error = 1; 2033b1e0a65SYOSHIFUJI Hideaki struct net *net = sock_net(sk); 2041da177e4SLinus Torvalds 20532c1da70SStephen Hemminger if (!snum) { 2069088c560SEric Dumazet int low, high, remaining; 2079088c560SEric Dumazet unsigned rand; 20898322f22SEric Dumazet unsigned short first, last; 20998322f22SEric Dumazet DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 2101da177e4SLinus Torvalds 211227b60f5SStephen Hemminger inet_get_local_port_range(&low, &high); 212a25de534SAnton Arapov remaining = (high - low) + 1; 213227b60f5SStephen Hemminger 2149088c560SEric Dumazet rand = net_random(); 21598322f22SEric Dumazet first = (((u64)rand * remaining) >> 32) + low; 21698322f22SEric Dumazet /* 21798322f22SEric Dumazet * force rand to be an odd multiple of UDP_HTABLE_SIZE 21898322f22SEric Dumazet */ 219f86dcc5aSEric Dumazet rand = (rand | 1) * (udptable->mask + 1); 2205781b235SEric Dumazet last = first + udptable->mask + 1; 2215781b235SEric Dumazet do { 222f86dcc5aSEric Dumazet hslot = udp_hashslot(udptable, net, first); 22398322f22SEric Dumazet bitmap_zero(bitmap, PORTS_PER_CHAIN); 224645ca708SEric Dumazet spin_lock_bh(&hslot->lock); 22598322f22SEric Dumazet udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 226f86dcc5aSEric Dumazet saddr_comp, udptable->log); 22798322f22SEric Dumazet 22898322f22SEric Dumazet snum = first; 22998322f22SEric Dumazet /* 23098322f22SEric Dumazet * Iterate on all possible values of snum for this hash. 23198322f22SEric Dumazet * Using steps of an odd multiple of UDP_HTABLE_SIZE 23298322f22SEric Dumazet * give us randomization and full range coverage. 23398322f22SEric Dumazet */ 2349088c560SEric Dumazet do { 23598322f22SEric Dumazet if (low <= snum && snum <= high && 236e3826f1eSAmerigo Wang !test_bit(snum >> udptable->log, bitmap) && 237e3826f1eSAmerigo Wang !inet_is_reserved_local_port(snum)) 23898322f22SEric Dumazet goto found; 23998322f22SEric Dumazet snum += rand; 24098322f22SEric Dumazet } while (snum != first); 24198322f22SEric Dumazet spin_unlock_bh(&hslot->lock); 2425781b235SEric Dumazet } while (++first != last); 24398322f22SEric Dumazet goto fail; 244645ca708SEric Dumazet } else { 245f86dcc5aSEric Dumazet hslot = udp_hashslot(udptable, net, snum); 246645ca708SEric Dumazet spin_lock_bh(&hslot->lock); 24730fff923SEric Dumazet if (hslot->count > 10) { 24830fff923SEric Dumazet int exist; 24930fff923SEric Dumazet unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 25030fff923SEric Dumazet 25130fff923SEric Dumazet slot2 &= udptable->mask; 25230fff923SEric Dumazet hash2_nulladdr &= udptable->mask; 25330fff923SEric Dumazet 25430fff923SEric Dumazet hslot2 = udp_hashslot2(udptable, slot2); 25530fff923SEric Dumazet if (hslot->count < hslot2->count) 25630fff923SEric Dumazet goto scan_primary_hash; 25730fff923SEric Dumazet 25830fff923SEric Dumazet exist = udp_lib_lport_inuse2(net, snum, hslot2, 25930fff923SEric Dumazet sk, saddr_comp); 26030fff923SEric Dumazet if (!exist && (hash2_nulladdr != slot2)) { 26130fff923SEric Dumazet hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 26230fff923SEric Dumazet exist = udp_lib_lport_inuse2(net, snum, hslot2, 26330fff923SEric Dumazet sk, saddr_comp); 26430fff923SEric Dumazet } 26530fff923SEric Dumazet if (exist) 26630fff923SEric Dumazet goto fail_unlock; 26730fff923SEric Dumazet else 26830fff923SEric Dumazet goto found; 26930fff923SEric Dumazet } 27030fff923SEric Dumazet scan_primary_hash: 271f86dcc5aSEric Dumazet if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 272f86dcc5aSEric Dumazet saddr_comp, 0)) 273645ca708SEric Dumazet goto fail_unlock; 274645ca708SEric Dumazet } 27598322f22SEric Dumazet found: 276c720c7e8SEric Dumazet inet_sk(sk)->inet_num = snum; 277d4cada4aSEric Dumazet udp_sk(sk)->udp_port_hash = snum; 278d4cada4aSEric Dumazet udp_sk(sk)->udp_portaddr_hash ^= snum; 2791da177e4SLinus Torvalds if (sk_unhashed(sk)) { 28088ab1932SEric Dumazet sk_nulls_add_node_rcu(sk, &hslot->head); 281fdcc8aa9SEric Dumazet hslot->count++; 282c29a0bc4SPavel Emelyanov sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 283512615b6SEric Dumazet 284512615b6SEric Dumazet hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 285512615b6SEric Dumazet spin_lock(&hslot2->lock); 286512615b6SEric Dumazet hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 287512615b6SEric Dumazet &hslot2->head); 288512615b6SEric Dumazet hslot2->count++; 289512615b6SEric Dumazet spin_unlock(&hslot2->lock); 2901da177e4SLinus Torvalds } 29125030a7fSGerrit Renker error = 0; 292645ca708SEric Dumazet fail_unlock: 293645ca708SEric Dumazet spin_unlock_bh(&hslot->lock); 2941da177e4SLinus Torvalds fail: 29525030a7fSGerrit Renker return error; 2961da177e4SLinus Torvalds } 297c482c568SEric Dumazet EXPORT_SYMBOL(udp_lib_get_port); 2981da177e4SLinus Torvalds 299499923c7SVlad Yasevich static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 300db8dac20SDavid S. Miller { 301db8dac20SDavid S. Miller struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 302db8dac20SDavid S. Miller 303db8dac20SDavid S. Miller return (!ipv6_only_sock(sk2) && 304c720c7e8SEric Dumazet (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 305c720c7e8SEric Dumazet inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 306db8dac20SDavid S. Miller } 307db8dac20SDavid S. Miller 308d4cada4aSEric Dumazet static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr, 309d4cada4aSEric Dumazet unsigned int port) 310d4cada4aSEric Dumazet { 3110eae88f3SEric Dumazet return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 312d4cada4aSEric Dumazet } 313d4cada4aSEric Dumazet 3146ba5a3c5SPavel Emelyanov int udp_v4_get_port(struct sock *sk, unsigned short snum) 315db8dac20SDavid S. Miller { 31630fff923SEric Dumazet unsigned int hash2_nulladdr = 3170eae88f3SEric Dumazet udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 31830fff923SEric Dumazet unsigned int hash2_partial = 31930fff923SEric Dumazet udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 32030fff923SEric Dumazet 321d4cada4aSEric Dumazet /* precompute partial secondary hash */ 32230fff923SEric Dumazet udp_sk(sk)->udp_portaddr_hash = hash2_partial; 32330fff923SEric Dumazet return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 324db8dac20SDavid S. Miller } 325db8dac20SDavid S. Miller 326645ca708SEric Dumazet static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, 327645ca708SEric Dumazet unsigned short hnum, 328645ca708SEric Dumazet __be16 sport, __be32 daddr, __be16 dport, int dif) 329645ca708SEric Dumazet { 330645ca708SEric Dumazet int score = -1; 331645ca708SEric Dumazet 332d4cada4aSEric Dumazet if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 333645ca708SEric Dumazet !ipv6_only_sock(sk)) { 334645ca708SEric Dumazet struct inet_sock *inet = inet_sk(sk); 335645ca708SEric Dumazet 336645ca708SEric Dumazet score = (sk->sk_family == PF_INET ? 1 : 0); 337c720c7e8SEric Dumazet if (inet->inet_rcv_saddr) { 338c720c7e8SEric Dumazet if (inet->inet_rcv_saddr != daddr) 339645ca708SEric Dumazet return -1; 340645ca708SEric Dumazet score += 2; 341645ca708SEric Dumazet } 342c720c7e8SEric Dumazet if (inet->inet_daddr) { 343c720c7e8SEric Dumazet if (inet->inet_daddr != saddr) 344645ca708SEric Dumazet return -1; 345645ca708SEric Dumazet score += 2; 346645ca708SEric Dumazet } 347c720c7e8SEric Dumazet if (inet->inet_dport) { 348c720c7e8SEric Dumazet if (inet->inet_dport != sport) 349645ca708SEric Dumazet return -1; 350645ca708SEric Dumazet score += 2; 351645ca708SEric Dumazet } 352645ca708SEric Dumazet if (sk->sk_bound_dev_if) { 353645ca708SEric Dumazet if (sk->sk_bound_dev_if != dif) 354645ca708SEric Dumazet return -1; 355645ca708SEric Dumazet score += 2; 356645ca708SEric Dumazet } 357645ca708SEric Dumazet } 358645ca708SEric Dumazet return score; 359645ca708SEric Dumazet } 360645ca708SEric Dumazet 3615051ebd2SEric Dumazet /* 3625051ebd2SEric Dumazet * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 3635051ebd2SEric Dumazet */ 3645051ebd2SEric Dumazet #define SCORE2_MAX (1 + 2 + 2 + 2) 3655051ebd2SEric Dumazet static inline int compute_score2(struct sock *sk, struct net *net, 3665051ebd2SEric Dumazet __be32 saddr, __be16 sport, 3675051ebd2SEric Dumazet __be32 daddr, unsigned int hnum, int dif) 3685051ebd2SEric Dumazet { 3695051ebd2SEric Dumazet int score = -1; 3705051ebd2SEric Dumazet 3715051ebd2SEric Dumazet if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) { 3725051ebd2SEric Dumazet struct inet_sock *inet = inet_sk(sk); 3735051ebd2SEric Dumazet 3745051ebd2SEric Dumazet if (inet->inet_rcv_saddr != daddr) 3755051ebd2SEric Dumazet return -1; 3765051ebd2SEric Dumazet if (inet->inet_num != hnum) 3775051ebd2SEric Dumazet return -1; 3785051ebd2SEric Dumazet 3795051ebd2SEric Dumazet score = (sk->sk_family == PF_INET ? 1 : 0); 3805051ebd2SEric Dumazet if (inet->inet_daddr) { 3815051ebd2SEric Dumazet if (inet->inet_daddr != saddr) 3825051ebd2SEric Dumazet return -1; 3835051ebd2SEric Dumazet score += 2; 3845051ebd2SEric Dumazet } 3855051ebd2SEric Dumazet if (inet->inet_dport) { 3865051ebd2SEric Dumazet if (inet->inet_dport != sport) 3875051ebd2SEric Dumazet return -1; 3885051ebd2SEric Dumazet score += 2; 3895051ebd2SEric Dumazet } 3905051ebd2SEric Dumazet if (sk->sk_bound_dev_if) { 3915051ebd2SEric Dumazet if (sk->sk_bound_dev_if != dif) 3925051ebd2SEric Dumazet return -1; 3935051ebd2SEric Dumazet score += 2; 3945051ebd2SEric Dumazet } 3955051ebd2SEric Dumazet } 3965051ebd2SEric Dumazet return score; 3975051ebd2SEric Dumazet } 3985051ebd2SEric Dumazet 3995051ebd2SEric Dumazet 4005051ebd2SEric Dumazet /* called with read_rcu_lock() */ 4015051ebd2SEric Dumazet static struct sock *udp4_lib_lookup2(struct net *net, 4025051ebd2SEric Dumazet __be32 saddr, __be16 sport, 4035051ebd2SEric Dumazet __be32 daddr, unsigned int hnum, int dif, 4045051ebd2SEric Dumazet struct udp_hslot *hslot2, unsigned int slot2) 4055051ebd2SEric Dumazet { 4065051ebd2SEric Dumazet struct sock *sk, *result; 4075051ebd2SEric Dumazet struct hlist_nulls_node *node; 4085051ebd2SEric Dumazet int score, badness; 4095051ebd2SEric Dumazet 4105051ebd2SEric Dumazet begin: 4115051ebd2SEric Dumazet result = NULL; 4125051ebd2SEric Dumazet badness = -1; 4135051ebd2SEric Dumazet udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 4145051ebd2SEric Dumazet score = compute_score2(sk, net, saddr, sport, 4155051ebd2SEric Dumazet daddr, hnum, dif); 4165051ebd2SEric Dumazet if (score > badness) { 4175051ebd2SEric Dumazet result = sk; 4185051ebd2SEric Dumazet badness = score; 4195051ebd2SEric Dumazet if (score == SCORE2_MAX) 4205051ebd2SEric Dumazet goto exact_match; 4215051ebd2SEric Dumazet } 4225051ebd2SEric Dumazet } 4235051ebd2SEric Dumazet /* 4245051ebd2SEric Dumazet * if the nulls value we got at the end of this lookup is 4255051ebd2SEric Dumazet * not the expected one, we must restart lookup. 4265051ebd2SEric Dumazet * We probably met an item that was moved to another chain. 4275051ebd2SEric Dumazet */ 4285051ebd2SEric Dumazet if (get_nulls_value(node) != slot2) 4295051ebd2SEric Dumazet goto begin; 4305051ebd2SEric Dumazet 4315051ebd2SEric Dumazet if (result) { 4325051ebd2SEric Dumazet exact_match: 4335051ebd2SEric Dumazet if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) 4345051ebd2SEric Dumazet result = NULL; 4355051ebd2SEric Dumazet else if (unlikely(compute_score2(result, net, saddr, sport, 4365051ebd2SEric Dumazet daddr, hnum, dif) < badness)) { 4375051ebd2SEric Dumazet sock_put(result); 4385051ebd2SEric Dumazet goto begin; 4395051ebd2SEric Dumazet } 4405051ebd2SEric Dumazet } 4415051ebd2SEric Dumazet return result; 4425051ebd2SEric Dumazet } 4435051ebd2SEric Dumazet 444db8dac20SDavid S. Miller /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 445db8dac20SDavid S. Miller * harder than this. -DaveM 446db8dac20SDavid S. Miller */ 447db8dac20SDavid S. Miller static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 448db8dac20SDavid S. Miller __be16 sport, __be32 daddr, __be16 dport, 449645ca708SEric Dumazet int dif, struct udp_table *udptable) 450db8dac20SDavid S. Miller { 451271b72c7SEric Dumazet struct sock *sk, *result; 45288ab1932SEric Dumazet struct hlist_nulls_node *node; 453db8dac20SDavid S. Miller unsigned short hnum = ntohs(dport); 4545051ebd2SEric Dumazet unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 4555051ebd2SEric Dumazet struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 456271b72c7SEric Dumazet int score, badness; 457db8dac20SDavid S. Miller 458271b72c7SEric Dumazet rcu_read_lock(); 4595051ebd2SEric Dumazet if (hslot->count > 10) { 4605051ebd2SEric Dumazet hash2 = udp4_portaddr_hash(net, daddr, hnum); 4615051ebd2SEric Dumazet slot2 = hash2 & udptable->mask; 4625051ebd2SEric Dumazet hslot2 = &udptable->hash2[slot2]; 4635051ebd2SEric Dumazet if (hslot->count < hslot2->count) 4645051ebd2SEric Dumazet goto begin; 4655051ebd2SEric Dumazet 4665051ebd2SEric Dumazet result = udp4_lib_lookup2(net, saddr, sport, 4675051ebd2SEric Dumazet daddr, hnum, dif, 4685051ebd2SEric Dumazet hslot2, slot2); 4695051ebd2SEric Dumazet if (!result) { 4700eae88f3SEric Dumazet hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 4715051ebd2SEric Dumazet slot2 = hash2 & udptable->mask; 4725051ebd2SEric Dumazet hslot2 = &udptable->hash2[slot2]; 4735051ebd2SEric Dumazet if (hslot->count < hslot2->count) 4745051ebd2SEric Dumazet goto begin; 4755051ebd2SEric Dumazet 4761223c67cSJorge Boncompte [DTI2] result = udp4_lib_lookup2(net, saddr, sport, 4770eae88f3SEric Dumazet htonl(INADDR_ANY), hnum, dif, 4785051ebd2SEric Dumazet hslot2, slot2); 4795051ebd2SEric Dumazet } 4805051ebd2SEric Dumazet rcu_read_unlock(); 4815051ebd2SEric Dumazet return result; 4825051ebd2SEric Dumazet } 483271b72c7SEric Dumazet begin: 484271b72c7SEric Dumazet result = NULL; 485271b72c7SEric Dumazet badness = -1; 48688ab1932SEric Dumazet sk_nulls_for_each_rcu(sk, node, &hslot->head) { 487645ca708SEric Dumazet score = compute_score(sk, net, saddr, hnum, sport, 488645ca708SEric Dumazet daddr, dport, dif); 489645ca708SEric Dumazet if (score > badness) { 490db8dac20SDavid S. Miller result = sk; 491db8dac20SDavid S. Miller badness = score; 492db8dac20SDavid S. Miller } 493db8dac20SDavid S. Miller } 49488ab1932SEric Dumazet /* 49588ab1932SEric Dumazet * if the nulls value we got at the end of this lookup is 49688ab1932SEric Dumazet * not the expected one, we must restart lookup. 49788ab1932SEric Dumazet * We probably met an item that was moved to another chain. 49888ab1932SEric Dumazet */ 4995051ebd2SEric Dumazet if (get_nulls_value(node) != slot) 50088ab1932SEric Dumazet goto begin; 50188ab1932SEric Dumazet 502271b72c7SEric Dumazet if (result) { 503271b72c7SEric Dumazet if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) 504271b72c7SEric Dumazet result = NULL; 505271b72c7SEric Dumazet else if (unlikely(compute_score(result, net, saddr, hnum, sport, 506271b72c7SEric Dumazet daddr, dport, dif) < badness)) { 507271b72c7SEric Dumazet sock_put(result); 508271b72c7SEric Dumazet goto begin; 509271b72c7SEric Dumazet } 510271b72c7SEric Dumazet } 511271b72c7SEric Dumazet rcu_read_unlock(); 512db8dac20SDavid S. Miller return result; 513db8dac20SDavid S. Miller } 514db8dac20SDavid S. Miller 515607c4aafSKOVACS Krisztian static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 516607c4aafSKOVACS Krisztian __be16 sport, __be16 dport, 517645ca708SEric Dumazet struct udp_table *udptable) 518607c4aafSKOVACS Krisztian { 51923542618SKOVACS Krisztian struct sock *sk; 520607c4aafSKOVACS Krisztian const struct iphdr *iph = ip_hdr(skb); 521607c4aafSKOVACS Krisztian 52223542618SKOVACS Krisztian if (unlikely(sk = skb_steal_sock(skb))) 52323542618SKOVACS Krisztian return sk; 52423542618SKOVACS Krisztian else 525adf30907SEric Dumazet return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 526607c4aafSKOVACS Krisztian iph->daddr, dport, inet_iif(skb), 527607c4aafSKOVACS Krisztian udptable); 528607c4aafSKOVACS Krisztian } 529607c4aafSKOVACS Krisztian 530bcd41303SKOVACS Krisztian struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 531bcd41303SKOVACS Krisztian __be32 daddr, __be16 dport, int dif) 532bcd41303SKOVACS Krisztian { 533645ca708SEric Dumazet return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 534bcd41303SKOVACS Krisztian } 535bcd41303SKOVACS Krisztian EXPORT_SYMBOL_GPL(udp4_lib_lookup); 536bcd41303SKOVACS Krisztian 537920a4611SEric Dumazet static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 538db8dac20SDavid S. Miller __be16 loc_port, __be32 loc_addr, 539db8dac20SDavid S. Miller __be16 rmt_port, __be32 rmt_addr, 540db8dac20SDavid S. Miller int dif) 541db8dac20SDavid S. Miller { 54288ab1932SEric Dumazet struct hlist_nulls_node *node; 543db8dac20SDavid S. Miller struct sock *s = sk; 544db8dac20SDavid S. Miller unsigned short hnum = ntohs(loc_port); 545db8dac20SDavid S. Miller 54688ab1932SEric Dumazet sk_nulls_for_each_from(s, node) { 547db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(s); 548db8dac20SDavid S. Miller 549920a4611SEric Dumazet if (!net_eq(sock_net(s), net) || 550d4cada4aSEric Dumazet udp_sk(s)->udp_port_hash != hnum || 551c720c7e8SEric Dumazet (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 552c720c7e8SEric Dumazet (inet->inet_dport != rmt_port && inet->inet_dport) || 553c720c7e8SEric Dumazet (inet->inet_rcv_saddr && 554c720c7e8SEric Dumazet inet->inet_rcv_saddr != loc_addr) || 555db8dac20SDavid S. Miller ipv6_only_sock(s) || 556db8dac20SDavid S. Miller (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 557db8dac20SDavid S. Miller continue; 558db8dac20SDavid S. Miller if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 559db8dac20SDavid S. Miller continue; 560db8dac20SDavid S. Miller goto found; 561db8dac20SDavid S. Miller } 562db8dac20SDavid S. Miller s = NULL; 563db8dac20SDavid S. Miller found: 564db8dac20SDavid S. Miller return s; 565db8dac20SDavid S. Miller } 566db8dac20SDavid S. Miller 567db8dac20SDavid S. Miller /* 568db8dac20SDavid S. Miller * This routine is called by the ICMP module when it gets some 569db8dac20SDavid S. Miller * sort of error condition. If err < 0 then the socket should 570db8dac20SDavid S. Miller * be closed and the error returned to the user. If err > 0 571db8dac20SDavid S. Miller * it's just the icmp type << 8 | icmp code. 572db8dac20SDavid S. Miller * Header points to the ip header of the error packet. We move 573db8dac20SDavid S. Miller * on past this. Then (as it used to claim before adjustment) 574db8dac20SDavid S. Miller * header points to the first 8 bytes of the udp header. We need 575db8dac20SDavid S. Miller * to find the appropriate port. 576db8dac20SDavid S. Miller */ 577db8dac20SDavid S. Miller 578645ca708SEric Dumazet void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 579db8dac20SDavid S. Miller { 580db8dac20SDavid S. Miller struct inet_sock *inet; 581db8dac20SDavid S. Miller struct iphdr *iph = (struct iphdr *)skb->data; 582db8dac20SDavid S. Miller struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 583db8dac20SDavid S. Miller const int type = icmp_hdr(skb)->type; 584db8dac20SDavid S. Miller const int code = icmp_hdr(skb)->code; 585db8dac20SDavid S. Miller struct sock *sk; 586db8dac20SDavid S. Miller int harderr; 587db8dac20SDavid S. Miller int err; 588fd54d716SPavel Emelyanov struct net *net = dev_net(skb->dev); 589db8dac20SDavid S. Miller 590fd54d716SPavel Emelyanov sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 591db8dac20SDavid S. Miller iph->saddr, uh->source, skb->dev->ifindex, udptable); 592db8dac20SDavid S. Miller if (sk == NULL) { 593dcfc23caSPavel Emelyanov ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 594db8dac20SDavid S. Miller return; /* No socket for error */ 595db8dac20SDavid S. Miller } 596db8dac20SDavid S. Miller 597db8dac20SDavid S. Miller err = 0; 598db8dac20SDavid S. Miller harderr = 0; 599db8dac20SDavid S. Miller inet = inet_sk(sk); 600db8dac20SDavid S. Miller 601db8dac20SDavid S. Miller switch (type) { 602db8dac20SDavid S. Miller default: 603db8dac20SDavid S. Miller case ICMP_TIME_EXCEEDED: 604db8dac20SDavid S. Miller err = EHOSTUNREACH; 605db8dac20SDavid S. Miller break; 606db8dac20SDavid S. Miller case ICMP_SOURCE_QUENCH: 607db8dac20SDavid S. Miller goto out; 608db8dac20SDavid S. Miller case ICMP_PARAMETERPROB: 609db8dac20SDavid S. Miller err = EPROTO; 610db8dac20SDavid S. Miller harderr = 1; 611db8dac20SDavid S. Miller break; 612db8dac20SDavid S. Miller case ICMP_DEST_UNREACH: 613db8dac20SDavid S. Miller if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 614db8dac20SDavid S. Miller if (inet->pmtudisc != IP_PMTUDISC_DONT) { 615db8dac20SDavid S. Miller err = EMSGSIZE; 616db8dac20SDavid S. Miller harderr = 1; 617db8dac20SDavid S. Miller break; 618db8dac20SDavid S. Miller } 619db8dac20SDavid S. Miller goto out; 620db8dac20SDavid S. Miller } 621db8dac20SDavid S. Miller err = EHOSTUNREACH; 622db8dac20SDavid S. Miller if (code <= NR_ICMP_UNREACH) { 623db8dac20SDavid S. Miller harderr = icmp_err_convert[code].fatal; 624db8dac20SDavid S. Miller err = icmp_err_convert[code].errno; 625db8dac20SDavid S. Miller } 626db8dac20SDavid S. Miller break; 627db8dac20SDavid S. Miller } 628db8dac20SDavid S. Miller 629db8dac20SDavid S. Miller /* 630db8dac20SDavid S. Miller * RFC1122: OK. Passes ICMP errors back to application, as per 631db8dac20SDavid S. Miller * 4.1.3.3. 632db8dac20SDavid S. Miller */ 633db8dac20SDavid S. Miller if (!inet->recverr) { 634db8dac20SDavid S. Miller if (!harderr || sk->sk_state != TCP_ESTABLISHED) 635db8dac20SDavid S. Miller goto out; 636b1faf566SEric Dumazet } else 637db8dac20SDavid S. Miller ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 638b1faf566SEric Dumazet 639db8dac20SDavid S. Miller sk->sk_err = err; 640db8dac20SDavid S. Miller sk->sk_error_report(sk); 641db8dac20SDavid S. Miller out: 642db8dac20SDavid S. Miller sock_put(sk); 643db8dac20SDavid S. Miller } 644db8dac20SDavid S. Miller 645db8dac20SDavid S. Miller void udp_err(struct sk_buff *skb, u32 info) 646db8dac20SDavid S. Miller { 647645ca708SEric Dumazet __udp4_lib_err(skb, info, &udp_table); 648db8dac20SDavid S. Miller } 649db8dac20SDavid S. Miller 650db8dac20SDavid S. Miller /* 651db8dac20SDavid S. Miller * Throw away all pending data and cancel the corking. Socket is locked. 652db8dac20SDavid S. Miller */ 65336d926b9SDenis V. Lunev void udp_flush_pending_frames(struct sock *sk) 654db8dac20SDavid S. Miller { 655db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 656db8dac20SDavid S. Miller 657db8dac20SDavid S. Miller if (up->pending) { 658db8dac20SDavid S. Miller up->len = 0; 659db8dac20SDavid S. Miller up->pending = 0; 660db8dac20SDavid S. Miller ip_flush_pending_frames(sk); 661db8dac20SDavid S. Miller } 662db8dac20SDavid S. Miller } 66336d926b9SDenis V. Lunev EXPORT_SYMBOL(udp_flush_pending_frames); 664db8dac20SDavid S. Miller 665db8dac20SDavid S. Miller /** 666db8dac20SDavid S. Miller * udp4_hwcsum_outgoing - handle outgoing HW checksumming 667db8dac20SDavid S. Miller * @sk: socket we are sending on 668db8dac20SDavid S. Miller * @skb: sk_buff containing the filled-in UDP header 669db8dac20SDavid S. Miller * (checksum field must be zeroed out) 670db8dac20SDavid S. Miller */ 671db8dac20SDavid S. Miller static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 672db8dac20SDavid S. Miller __be32 src, __be32 dst, int len) 673db8dac20SDavid S. Miller { 674db8dac20SDavid S. Miller unsigned int offset; 675db8dac20SDavid S. Miller struct udphdr *uh = udp_hdr(skb); 676db8dac20SDavid S. Miller __wsum csum = 0; 677db8dac20SDavid S. Miller 678db8dac20SDavid S. Miller if (skb_queue_len(&sk->sk_write_queue) == 1) { 679db8dac20SDavid S. Miller /* 680db8dac20SDavid S. Miller * Only one fragment on the socket. 681db8dac20SDavid S. Miller */ 682db8dac20SDavid S. Miller skb->csum_start = skb_transport_header(skb) - skb->head; 683db8dac20SDavid S. Miller skb->csum_offset = offsetof(struct udphdr, check); 684db8dac20SDavid S. Miller uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 685db8dac20SDavid S. Miller } else { 686db8dac20SDavid S. Miller /* 687db8dac20SDavid S. Miller * HW-checksum won't work as there are two or more 688db8dac20SDavid S. Miller * fragments on the socket so that all csums of sk_buffs 689db8dac20SDavid S. Miller * should be together 690db8dac20SDavid S. Miller */ 691db8dac20SDavid S. Miller offset = skb_transport_offset(skb); 692db8dac20SDavid S. Miller skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 693db8dac20SDavid S. Miller 694db8dac20SDavid S. Miller skb->ip_summed = CHECKSUM_NONE; 695db8dac20SDavid S. Miller 696db8dac20SDavid S. Miller skb_queue_walk(&sk->sk_write_queue, skb) { 697db8dac20SDavid S. Miller csum = csum_add(csum, skb->csum); 698db8dac20SDavid S. Miller } 699db8dac20SDavid S. Miller 700db8dac20SDavid S. Miller uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 701db8dac20SDavid S. Miller if (uh->check == 0) 702db8dac20SDavid S. Miller uh->check = CSUM_MANGLED_0; 703db8dac20SDavid S. Miller } 704db8dac20SDavid S. Miller } 705db8dac20SDavid S. Miller 706db8dac20SDavid S. Miller /* 707db8dac20SDavid S. Miller * Push out all pending data as one UDP datagram. Socket is locked. 708db8dac20SDavid S. Miller */ 709db8dac20SDavid S. Miller static int udp_push_pending_frames(struct sock *sk) 710db8dac20SDavid S. Miller { 711db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 712db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sk); 713db8dac20SDavid S. Miller struct flowi *fl = &inet->cork.fl; 714db8dac20SDavid S. Miller struct sk_buff *skb; 715db8dac20SDavid S. Miller struct udphdr *uh; 716db8dac20SDavid S. Miller int err = 0; 717db8dac20SDavid S. Miller int is_udplite = IS_UDPLITE(sk); 718db8dac20SDavid S. Miller __wsum csum = 0; 719db8dac20SDavid S. Miller 720db8dac20SDavid S. Miller /* Grab the skbuff where UDP header space exists. */ 721db8dac20SDavid S. Miller if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 722db8dac20SDavid S. Miller goto out; 723db8dac20SDavid S. Miller 724db8dac20SDavid S. Miller /* 725db8dac20SDavid S. Miller * Create a UDP header 726db8dac20SDavid S. Miller */ 727db8dac20SDavid S. Miller uh = udp_hdr(skb); 728db8dac20SDavid S. Miller uh->source = fl->fl_ip_sport; 729db8dac20SDavid S. Miller uh->dest = fl->fl_ip_dport; 730db8dac20SDavid S. Miller uh->len = htons(up->len); 731db8dac20SDavid S. Miller uh->check = 0; 732db8dac20SDavid S. Miller 733db8dac20SDavid S. Miller if (is_udplite) /* UDP-Lite */ 734db8dac20SDavid S. Miller csum = udplite_csum_outgoing(sk, skb); 735db8dac20SDavid S. Miller 736db8dac20SDavid S. Miller else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 737db8dac20SDavid S. Miller 738db8dac20SDavid S. Miller skb->ip_summed = CHECKSUM_NONE; 739db8dac20SDavid S. Miller goto send; 740db8dac20SDavid S. Miller 741db8dac20SDavid S. Miller } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 742db8dac20SDavid S. Miller 743db8dac20SDavid S. Miller udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len); 744db8dac20SDavid S. Miller goto send; 745db8dac20SDavid S. Miller 746db8dac20SDavid S. Miller } else /* `normal' UDP */ 747db8dac20SDavid S. Miller csum = udp_csum_outgoing(sk, skb); 748db8dac20SDavid S. Miller 749db8dac20SDavid S. Miller /* add protocol-dependent pseudo-header */ 750db8dac20SDavid S. Miller uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 751db8dac20SDavid S. Miller sk->sk_protocol, csum); 752db8dac20SDavid S. Miller if (uh->check == 0) 753db8dac20SDavid S. Miller uh->check = CSUM_MANGLED_0; 754db8dac20SDavid S. Miller 755db8dac20SDavid S. Miller send: 756db8dac20SDavid S. Miller err = ip_push_pending_frames(sk); 7576ce9e7b5SEric Dumazet if (err) { 7586ce9e7b5SEric Dumazet if (err == -ENOBUFS && !inet->recverr) { 7596ce9e7b5SEric Dumazet UDP_INC_STATS_USER(sock_net(sk), 7606ce9e7b5SEric Dumazet UDP_MIB_SNDBUFERRORS, is_udplite); 7616ce9e7b5SEric Dumazet err = 0; 7626ce9e7b5SEric Dumazet } 7636ce9e7b5SEric Dumazet } else 7646ce9e7b5SEric Dumazet UDP_INC_STATS_USER(sock_net(sk), 7656ce9e7b5SEric Dumazet UDP_MIB_OUTDATAGRAMS, is_udplite); 766db8dac20SDavid S. Miller out: 767db8dac20SDavid S. Miller up->len = 0; 768db8dac20SDavid S. Miller up->pending = 0; 769db8dac20SDavid S. Miller return err; 770db8dac20SDavid S. Miller } 771db8dac20SDavid S. Miller 772db8dac20SDavid S. Miller int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 773db8dac20SDavid S. Miller size_t len) 774db8dac20SDavid S. Miller { 775db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sk); 776db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 777db8dac20SDavid S. Miller int ulen = len; 778db8dac20SDavid S. Miller struct ipcm_cookie ipc; 779db8dac20SDavid S. Miller struct rtable *rt = NULL; 780db8dac20SDavid S. Miller int free = 0; 781db8dac20SDavid S. Miller int connected = 0; 782db8dac20SDavid S. Miller __be32 daddr, faddr, saddr; 783db8dac20SDavid S. Miller __be16 dport; 784db8dac20SDavid S. Miller u8 tos; 785db8dac20SDavid S. Miller int err, is_udplite = IS_UDPLITE(sk); 786db8dac20SDavid S. Miller int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 787db8dac20SDavid S. Miller int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 788db8dac20SDavid S. Miller 789db8dac20SDavid S. Miller if (len > 0xFFFF) 790db8dac20SDavid S. Miller return -EMSGSIZE; 791db8dac20SDavid S. Miller 792db8dac20SDavid S. Miller /* 793db8dac20SDavid S. Miller * Check the flags. 794db8dac20SDavid S. Miller */ 795db8dac20SDavid S. Miller 796db8dac20SDavid S. Miller if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 797db8dac20SDavid S. Miller return -EOPNOTSUPP; 798db8dac20SDavid S. Miller 799db8dac20SDavid S. Miller ipc.opt = NULL; 8002244d07bSOliver Hartkopp ipc.tx_flags = 0; 801db8dac20SDavid S. Miller 802db8dac20SDavid S. Miller if (up->pending) { 803db8dac20SDavid S. Miller /* 804db8dac20SDavid S. Miller * There are pending frames. 805db8dac20SDavid S. Miller * The socket lock must be held while it's corked. 806db8dac20SDavid S. Miller */ 807db8dac20SDavid S. Miller lock_sock(sk); 808db8dac20SDavid S. Miller if (likely(up->pending)) { 809db8dac20SDavid S. Miller if (unlikely(up->pending != AF_INET)) { 810db8dac20SDavid S. Miller release_sock(sk); 811db8dac20SDavid S. Miller return -EINVAL; 812db8dac20SDavid S. Miller } 813db8dac20SDavid S. Miller goto do_append_data; 814db8dac20SDavid S. Miller } 815db8dac20SDavid S. Miller release_sock(sk); 816db8dac20SDavid S. Miller } 817db8dac20SDavid S. Miller ulen += sizeof(struct udphdr); 818db8dac20SDavid S. Miller 819db8dac20SDavid S. Miller /* 820db8dac20SDavid S. Miller * Get and verify the address. 821db8dac20SDavid S. Miller */ 822db8dac20SDavid S. Miller if (msg->msg_name) { 823db8dac20SDavid S. Miller struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name; 824db8dac20SDavid S. Miller if (msg->msg_namelen < sizeof(*usin)) 825db8dac20SDavid S. Miller return -EINVAL; 826db8dac20SDavid S. Miller if (usin->sin_family != AF_INET) { 827db8dac20SDavid S. Miller if (usin->sin_family != AF_UNSPEC) 828db8dac20SDavid S. Miller return -EAFNOSUPPORT; 829db8dac20SDavid S. Miller } 830db8dac20SDavid S. Miller 831db8dac20SDavid S. Miller daddr = usin->sin_addr.s_addr; 832db8dac20SDavid S. Miller dport = usin->sin_port; 833db8dac20SDavid S. Miller if (dport == 0) 834db8dac20SDavid S. Miller return -EINVAL; 835db8dac20SDavid S. Miller } else { 836db8dac20SDavid S. Miller if (sk->sk_state != TCP_ESTABLISHED) 837db8dac20SDavid S. Miller return -EDESTADDRREQ; 838c720c7e8SEric Dumazet daddr = inet->inet_daddr; 839c720c7e8SEric Dumazet dport = inet->inet_dport; 840db8dac20SDavid S. Miller /* Open fast path for connected socket. 841db8dac20SDavid S. Miller Route will not be used, if at least one option is set. 842db8dac20SDavid S. Miller */ 843db8dac20SDavid S. Miller connected = 1; 844db8dac20SDavid S. Miller } 845c720c7e8SEric Dumazet ipc.addr = inet->inet_saddr; 846db8dac20SDavid S. Miller 847db8dac20SDavid S. Miller ipc.oif = sk->sk_bound_dev_if; 8482244d07bSOliver Hartkopp err = sock_tx_timestamp(sk, &ipc.tx_flags); 84951f31cabSPatrick Ohly if (err) 85051f31cabSPatrick Ohly return err; 851db8dac20SDavid S. Miller if (msg->msg_controllen) { 8523b1e0a65SYOSHIFUJI Hideaki err = ip_cmsg_send(sock_net(sk), msg, &ipc); 853db8dac20SDavid S. Miller if (err) 854db8dac20SDavid S. Miller return err; 855db8dac20SDavid S. Miller if (ipc.opt) 856db8dac20SDavid S. Miller free = 1; 857db8dac20SDavid S. Miller connected = 0; 858db8dac20SDavid S. Miller } 859db8dac20SDavid S. Miller if (!ipc.opt) 860db8dac20SDavid S. Miller ipc.opt = inet->opt; 861db8dac20SDavid S. Miller 862db8dac20SDavid S. Miller saddr = ipc.addr; 863db8dac20SDavid S. Miller ipc.addr = faddr = daddr; 864db8dac20SDavid S. Miller 865db8dac20SDavid S. Miller if (ipc.opt && ipc.opt->srr) { 866db8dac20SDavid S. Miller if (!daddr) 867db8dac20SDavid S. Miller return -EINVAL; 868db8dac20SDavid S. Miller faddr = ipc.opt->faddr; 869db8dac20SDavid S. Miller connected = 0; 870db8dac20SDavid S. Miller } 871db8dac20SDavid S. Miller tos = RT_TOS(inet->tos); 872db8dac20SDavid S. Miller if (sock_flag(sk, SOCK_LOCALROUTE) || 873db8dac20SDavid S. Miller (msg->msg_flags & MSG_DONTROUTE) || 874db8dac20SDavid S. Miller (ipc.opt && ipc.opt->is_strictroute)) { 875db8dac20SDavid S. Miller tos |= RTO_ONLINK; 876db8dac20SDavid S. Miller connected = 0; 877db8dac20SDavid S. Miller } 878db8dac20SDavid S. Miller 879db8dac20SDavid S. Miller if (ipv4_is_multicast(daddr)) { 880db8dac20SDavid S. Miller if (!ipc.oif) 881db8dac20SDavid S. Miller ipc.oif = inet->mc_index; 882db8dac20SDavid S. Miller if (!saddr) 883db8dac20SDavid S. Miller saddr = inet->mc_addr; 884db8dac20SDavid S. Miller connected = 0; 885db8dac20SDavid S. Miller } 886db8dac20SDavid S. Miller 887db8dac20SDavid S. Miller if (connected) 888db8dac20SDavid S. Miller rt = (struct rtable *)sk_dst_check(sk, 0); 889db8dac20SDavid S. Miller 890db8dac20SDavid S. Miller if (rt == NULL) { 891db8dac20SDavid S. Miller struct flowi fl = { .oif = ipc.oif, 892914a9ab3SAtis Elsts .mark = sk->sk_mark, 893db8dac20SDavid S. Miller .nl_u = { .ip4_u = 894db8dac20SDavid S. Miller { .daddr = faddr, 895db8dac20SDavid S. Miller .saddr = saddr, 896db8dac20SDavid S. Miller .tos = tos } }, 897db8dac20SDavid S. Miller .proto = sk->sk_protocol, 898a134f85cSBalazs Scheidler .flags = inet_sk_flowi_flags(sk), 899db8dac20SDavid S. Miller .uli_u = { .ports = 900c720c7e8SEric Dumazet { .sport = inet->inet_sport, 901db8dac20SDavid S. Miller .dport = dport } } }; 90284a3aa00SPavel Emelyanov struct net *net = sock_net(sk); 90384a3aa00SPavel Emelyanov 904db8dac20SDavid S. Miller security_sk_classify_flow(sk, &fl); 90584a3aa00SPavel Emelyanov err = ip_route_output_flow(net, &rt, &fl, sk, 1); 906db8dac20SDavid S. Miller if (err) { 907db8dac20SDavid S. Miller if (err == -ENETUNREACH) 9087c73a6faSPavel Emelyanov IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 909db8dac20SDavid S. Miller goto out; 910db8dac20SDavid S. Miller } 911db8dac20SDavid S. Miller 912db8dac20SDavid S. Miller err = -EACCES; 913db8dac20SDavid S. Miller if ((rt->rt_flags & RTCF_BROADCAST) && 914db8dac20SDavid S. Miller !sock_flag(sk, SOCK_BROADCAST)) 915db8dac20SDavid S. Miller goto out; 916db8dac20SDavid S. Miller if (connected) 917d8d1f30bSChangli Gao sk_dst_set(sk, dst_clone(&rt->dst)); 918db8dac20SDavid S. Miller } 919db8dac20SDavid S. Miller 920db8dac20SDavid S. Miller if (msg->msg_flags&MSG_CONFIRM) 921db8dac20SDavid S. Miller goto do_confirm; 922db8dac20SDavid S. Miller back_from_confirm: 923db8dac20SDavid S. Miller 924db8dac20SDavid S. Miller saddr = rt->rt_src; 925db8dac20SDavid S. Miller if (!ipc.addr) 926db8dac20SDavid S. Miller daddr = ipc.addr = rt->rt_dst; 927db8dac20SDavid S. Miller 928db8dac20SDavid S. Miller lock_sock(sk); 929db8dac20SDavid S. Miller if (unlikely(up->pending)) { 930db8dac20SDavid S. Miller /* The socket is already corked while preparing it. */ 931db8dac20SDavid S. Miller /* ... which is an evident application bug. --ANK */ 932db8dac20SDavid S. Miller release_sock(sk); 933db8dac20SDavid S. Miller 934db8dac20SDavid S. Miller LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 935db8dac20SDavid S. Miller err = -EINVAL; 936db8dac20SDavid S. Miller goto out; 937db8dac20SDavid S. Miller } 938db8dac20SDavid S. Miller /* 939db8dac20SDavid S. Miller * Now cork the socket to pend data. 940db8dac20SDavid S. Miller */ 941db8dac20SDavid S. Miller inet->cork.fl.fl4_dst = daddr; 942db8dac20SDavid S. Miller inet->cork.fl.fl_ip_dport = dport; 943db8dac20SDavid S. Miller inet->cork.fl.fl4_src = saddr; 944c720c7e8SEric Dumazet inet->cork.fl.fl_ip_sport = inet->inet_sport; 945db8dac20SDavid S. Miller up->pending = AF_INET; 946db8dac20SDavid S. Miller 947db8dac20SDavid S. Miller do_append_data: 948db8dac20SDavid S. Miller up->len += ulen; 949db8dac20SDavid S. Miller getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 950db8dac20SDavid S. Miller err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 9512e77d89bSEric Dumazet sizeof(struct udphdr), &ipc, &rt, 952db8dac20SDavid S. Miller corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 953db8dac20SDavid S. Miller if (err) 954db8dac20SDavid S. Miller udp_flush_pending_frames(sk); 955db8dac20SDavid S. Miller else if (!corkreq) 956db8dac20SDavid S. Miller err = udp_push_pending_frames(sk); 957db8dac20SDavid S. Miller else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 958db8dac20SDavid S. Miller up->pending = 0; 959db8dac20SDavid S. Miller release_sock(sk); 960db8dac20SDavid S. Miller 961db8dac20SDavid S. Miller out: 962db8dac20SDavid S. Miller ip_rt_put(rt); 963db8dac20SDavid S. Miller if (free) 964db8dac20SDavid S. Miller kfree(ipc.opt); 965db8dac20SDavid S. Miller if (!err) 966db8dac20SDavid S. Miller return len; 967db8dac20SDavid S. Miller /* 968db8dac20SDavid S. Miller * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 969db8dac20SDavid S. Miller * ENOBUFS might not be good (it's not tunable per se), but otherwise 970db8dac20SDavid S. Miller * we don't have a good statistic (IpOutDiscards but it can be too many 971db8dac20SDavid S. Miller * things). We could add another new stat but at least for now that 972db8dac20SDavid S. Miller * seems like overkill. 973db8dac20SDavid S. Miller */ 974db8dac20SDavid S. Miller if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 975629ca23cSPavel Emelyanov UDP_INC_STATS_USER(sock_net(sk), 976629ca23cSPavel Emelyanov UDP_MIB_SNDBUFERRORS, is_udplite); 977db8dac20SDavid S. Miller } 978db8dac20SDavid S. Miller return err; 979db8dac20SDavid S. Miller 980db8dac20SDavid S. Miller do_confirm: 981d8d1f30bSChangli Gao dst_confirm(&rt->dst); 982db8dac20SDavid S. Miller if (!(msg->msg_flags&MSG_PROBE) || len) 983db8dac20SDavid S. Miller goto back_from_confirm; 984db8dac20SDavid S. Miller err = 0; 985db8dac20SDavid S. Miller goto out; 986db8dac20SDavid S. Miller } 987c482c568SEric Dumazet EXPORT_SYMBOL(udp_sendmsg); 988db8dac20SDavid S. Miller 989db8dac20SDavid S. Miller int udp_sendpage(struct sock *sk, struct page *page, int offset, 990db8dac20SDavid S. Miller size_t size, int flags) 991db8dac20SDavid S. Miller { 992db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 993db8dac20SDavid S. Miller int ret; 994db8dac20SDavid S. Miller 995db8dac20SDavid S. Miller if (!up->pending) { 996db8dac20SDavid S. Miller struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 997db8dac20SDavid S. Miller 998db8dac20SDavid S. Miller /* Call udp_sendmsg to specify destination address which 999db8dac20SDavid S. Miller * sendpage interface can't pass. 1000db8dac20SDavid S. Miller * This will succeed only when the socket is connected. 1001db8dac20SDavid S. Miller */ 1002db8dac20SDavid S. Miller ret = udp_sendmsg(NULL, sk, &msg, 0); 1003db8dac20SDavid S. Miller if (ret < 0) 1004db8dac20SDavid S. Miller return ret; 1005db8dac20SDavid S. Miller } 1006db8dac20SDavid S. Miller 1007db8dac20SDavid S. Miller lock_sock(sk); 1008db8dac20SDavid S. Miller 1009db8dac20SDavid S. Miller if (unlikely(!up->pending)) { 1010db8dac20SDavid S. Miller release_sock(sk); 1011db8dac20SDavid S. Miller 1012db8dac20SDavid S. Miller LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 1013db8dac20SDavid S. Miller return -EINVAL; 1014db8dac20SDavid S. Miller } 1015db8dac20SDavid S. Miller 1016db8dac20SDavid S. Miller ret = ip_append_page(sk, page, offset, size, flags); 1017db8dac20SDavid S. Miller if (ret == -EOPNOTSUPP) { 1018db8dac20SDavid S. Miller release_sock(sk); 1019db8dac20SDavid S. Miller return sock_no_sendpage(sk->sk_socket, page, offset, 1020db8dac20SDavid S. Miller size, flags); 1021db8dac20SDavid S. Miller } 1022db8dac20SDavid S. Miller if (ret < 0) { 1023db8dac20SDavid S. Miller udp_flush_pending_frames(sk); 1024db8dac20SDavid S. Miller goto out; 1025db8dac20SDavid S. Miller } 1026db8dac20SDavid S. Miller 1027db8dac20SDavid S. Miller up->len += size; 1028db8dac20SDavid S. Miller if (!(up->corkflag || (flags&MSG_MORE))) 1029db8dac20SDavid S. Miller ret = udp_push_pending_frames(sk); 1030db8dac20SDavid S. Miller if (!ret) 1031db8dac20SDavid S. Miller ret = size; 1032db8dac20SDavid S. Miller out: 1033db8dac20SDavid S. Miller release_sock(sk); 1034db8dac20SDavid S. Miller return ret; 1035db8dac20SDavid S. Miller } 1036db8dac20SDavid S. Miller 103785584672SEric Dumazet 103885584672SEric Dumazet /** 103985584672SEric Dumazet * first_packet_length - return length of first packet in receive queue 104085584672SEric Dumazet * @sk: socket 104185584672SEric Dumazet * 104285584672SEric Dumazet * Drops all bad checksum frames, until a valid one is found. 104385584672SEric Dumazet * Returns the length of found skb, or 0 if none is found. 104485584672SEric Dumazet */ 104585584672SEric Dumazet static unsigned int first_packet_length(struct sock *sk) 104685584672SEric Dumazet { 104785584672SEric Dumazet struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 104885584672SEric Dumazet struct sk_buff *skb; 104985584672SEric Dumazet unsigned int res; 105085584672SEric Dumazet 105185584672SEric Dumazet __skb_queue_head_init(&list_kill); 105285584672SEric Dumazet 105385584672SEric Dumazet spin_lock_bh(&rcvq->lock); 105485584672SEric Dumazet while ((skb = skb_peek(rcvq)) != NULL && 105585584672SEric Dumazet udp_lib_checksum_complete(skb)) { 105685584672SEric Dumazet UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 105785584672SEric Dumazet IS_UDPLITE(sk)); 10588edf19c2SEric Dumazet atomic_inc(&sk->sk_drops); 105985584672SEric Dumazet __skb_unlink(skb, rcvq); 106085584672SEric Dumazet __skb_queue_tail(&list_kill, skb); 106185584672SEric Dumazet } 106285584672SEric Dumazet res = skb ? skb->len : 0; 106385584672SEric Dumazet spin_unlock_bh(&rcvq->lock); 106485584672SEric Dumazet 106585584672SEric Dumazet if (!skb_queue_empty(&list_kill)) { 10668a74ad60SEric Dumazet bool slow = lock_sock_fast(sk); 10678a74ad60SEric Dumazet 106885584672SEric Dumazet __skb_queue_purge(&list_kill); 106985584672SEric Dumazet sk_mem_reclaim_partial(sk); 10708a74ad60SEric Dumazet unlock_sock_fast(sk, slow); 107185584672SEric Dumazet } 107285584672SEric Dumazet return res; 107385584672SEric Dumazet } 107485584672SEric Dumazet 10751da177e4SLinus Torvalds /* 10761da177e4SLinus Torvalds * IOCTL requests applicable to the UDP protocol 10771da177e4SLinus Torvalds */ 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 10801da177e4SLinus Torvalds { 10816516c655SStephen Hemminger switch (cmd) { 10821da177e4SLinus Torvalds case SIOCOUTQ: 10831da177e4SLinus Torvalds { 108431e6d363SEric Dumazet int amount = sk_wmem_alloc_get(sk); 108531e6d363SEric Dumazet 10861da177e4SLinus Torvalds return put_user(amount, (int __user *)arg); 10871da177e4SLinus Torvalds } 10881da177e4SLinus Torvalds 10891da177e4SLinus Torvalds case SIOCINQ: 10901da177e4SLinus Torvalds { 109185584672SEric Dumazet unsigned int amount = first_packet_length(sk); 10921da177e4SLinus Torvalds 109385584672SEric Dumazet if (amount) 10941da177e4SLinus Torvalds /* 10951da177e4SLinus Torvalds * We will only return the amount 10961da177e4SLinus Torvalds * of this packet since that is all 10971da177e4SLinus Torvalds * that will be read. 10981da177e4SLinus Torvalds */ 109985584672SEric Dumazet amount -= sizeof(struct udphdr); 110085584672SEric Dumazet 11011da177e4SLinus Torvalds return put_user(amount, (int __user *)arg); 11021da177e4SLinus Torvalds } 11031da177e4SLinus Torvalds 11041da177e4SLinus Torvalds default: 11051da177e4SLinus Torvalds return -ENOIOCTLCMD; 11061da177e4SLinus Torvalds } 11076516c655SStephen Hemminger 11086516c655SStephen Hemminger return 0; 11091da177e4SLinus Torvalds } 1110c482c568SEric Dumazet EXPORT_SYMBOL(udp_ioctl); 11111da177e4SLinus Torvalds 1112db8dac20SDavid S. Miller /* 1113db8dac20SDavid S. Miller * This should be easy, if there is something there we 1114db8dac20SDavid S. Miller * return it, otherwise we block. 1115db8dac20SDavid S. Miller */ 1116db8dac20SDavid S. Miller 1117db8dac20SDavid S. Miller int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1118db8dac20SDavid S. Miller size_t len, int noblock, int flags, int *addr_len) 1119db8dac20SDavid S. Miller { 1120db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sk); 1121db8dac20SDavid S. Miller struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 1122db8dac20SDavid S. Miller struct sk_buff *skb; 112381d54ec8SGerrit Renker unsigned int ulen; 1124db8dac20SDavid S. Miller int peeked; 1125db8dac20SDavid S. Miller int err; 1126db8dac20SDavid S. Miller int is_udplite = IS_UDPLITE(sk); 11278a74ad60SEric Dumazet bool slow; 1128db8dac20SDavid S. Miller 1129db8dac20SDavid S. Miller /* 1130db8dac20SDavid S. Miller * Check any passed addresses 1131db8dac20SDavid S. Miller */ 1132db8dac20SDavid S. Miller if (addr_len) 1133db8dac20SDavid S. Miller *addr_len = sizeof(*sin); 1134db8dac20SDavid S. Miller 1135db8dac20SDavid S. Miller if (flags & MSG_ERRQUEUE) 1136db8dac20SDavid S. Miller return ip_recv_error(sk, msg, len); 1137db8dac20SDavid S. Miller 1138db8dac20SDavid S. Miller try_again: 1139db8dac20SDavid S. Miller skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1140db8dac20SDavid S. Miller &peeked, &err); 1141db8dac20SDavid S. Miller if (!skb) 1142db8dac20SDavid S. Miller goto out; 1143db8dac20SDavid S. Miller 1144db8dac20SDavid S. Miller ulen = skb->len - sizeof(struct udphdr); 114581d54ec8SGerrit Renker if (len > ulen) 114681d54ec8SGerrit Renker len = ulen; 114781d54ec8SGerrit Renker else if (len < ulen) 1148db8dac20SDavid S. Miller msg->msg_flags |= MSG_TRUNC; 1149db8dac20SDavid S. Miller 1150db8dac20SDavid S. Miller /* 1151db8dac20SDavid S. Miller * If checksum is needed at all, try to do it while copying the 1152db8dac20SDavid S. Miller * data. If the data is truncated, or if we only want a partial 1153db8dac20SDavid S. Miller * coverage checksum (UDP-Lite), do it before the copy. 1154db8dac20SDavid S. Miller */ 1155db8dac20SDavid S. Miller 115681d54ec8SGerrit Renker if (len < ulen || UDP_SKB_CB(skb)->partial_cov) { 1157db8dac20SDavid S. Miller if (udp_lib_checksum_complete(skb)) 1158db8dac20SDavid S. Miller goto csum_copy_err; 1159db8dac20SDavid S. Miller } 1160db8dac20SDavid S. Miller 1161db8dac20SDavid S. Miller if (skb_csum_unnecessary(skb)) 1162db8dac20SDavid S. Miller err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 116381d54ec8SGerrit Renker msg->msg_iov, len); 1164db8dac20SDavid S. Miller else { 1165c482c568SEric Dumazet err = skb_copy_and_csum_datagram_iovec(skb, 1166c482c568SEric Dumazet sizeof(struct udphdr), 1167c482c568SEric Dumazet msg->msg_iov); 1168db8dac20SDavid S. Miller 1169db8dac20SDavid S. Miller if (err == -EINVAL) 1170db8dac20SDavid S. Miller goto csum_copy_err; 1171db8dac20SDavid S. Miller } 1172db8dac20SDavid S. Miller 1173db8dac20SDavid S. Miller if (err) 1174db8dac20SDavid S. Miller goto out_free; 1175db8dac20SDavid S. Miller 1176db8dac20SDavid S. Miller if (!peeked) 1177629ca23cSPavel Emelyanov UDP_INC_STATS_USER(sock_net(sk), 1178629ca23cSPavel Emelyanov UDP_MIB_INDATAGRAMS, is_udplite); 1179db8dac20SDavid S. Miller 11803b885787SNeil Horman sock_recv_ts_and_drops(msg, sk, skb); 1181db8dac20SDavid S. Miller 1182db8dac20SDavid S. Miller /* Copy the address. */ 1183c482c568SEric Dumazet if (sin) { 1184db8dac20SDavid S. Miller sin->sin_family = AF_INET; 1185db8dac20SDavid S. Miller sin->sin_port = udp_hdr(skb)->source; 1186db8dac20SDavid S. Miller sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1187db8dac20SDavid S. Miller memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1188db8dac20SDavid S. Miller } 1189db8dac20SDavid S. Miller if (inet->cmsg_flags) 1190db8dac20SDavid S. Miller ip_cmsg_recv(msg, skb); 1191db8dac20SDavid S. Miller 119281d54ec8SGerrit Renker err = len; 1193db8dac20SDavid S. Miller if (flags & MSG_TRUNC) 1194db8dac20SDavid S. Miller err = ulen; 1195db8dac20SDavid S. Miller 1196db8dac20SDavid S. Miller out_free: 11979d410c79SEric Dumazet skb_free_datagram_locked(sk, skb); 1198db8dac20SDavid S. Miller out: 1199db8dac20SDavid S. Miller return err; 1200db8dac20SDavid S. Miller 1201db8dac20SDavid S. Miller csum_copy_err: 12028a74ad60SEric Dumazet slow = lock_sock_fast(sk); 1203db8dac20SDavid S. Miller if (!skb_kill_datagram(sk, skb, flags)) 1204629ca23cSPavel Emelyanov UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 12058a74ad60SEric Dumazet unlock_sock_fast(sk, slow); 1206db8dac20SDavid S. Miller 1207db8dac20SDavid S. Miller if (noblock) 1208db8dac20SDavid S. Miller return -EAGAIN; 1209db8dac20SDavid S. Miller goto try_again; 1210db8dac20SDavid S. Miller } 1211db8dac20SDavid S. Miller 1212db8dac20SDavid S. Miller 12131da177e4SLinus Torvalds int udp_disconnect(struct sock *sk, int flags) 12141da177e4SLinus Torvalds { 12151da177e4SLinus Torvalds struct inet_sock *inet = inet_sk(sk); 12161da177e4SLinus Torvalds /* 12171da177e4SLinus Torvalds * 1003.1g - break association. 12181da177e4SLinus Torvalds */ 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds sk->sk_state = TCP_CLOSE; 1221c720c7e8SEric Dumazet inet->inet_daddr = 0; 1222c720c7e8SEric Dumazet inet->inet_dport = 0; 1223c58dc01bSDavid S. Miller sock_rps_save_rxhash(sk, 0); 12241da177e4SLinus Torvalds sk->sk_bound_dev_if = 0; 12251da177e4SLinus Torvalds if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 12261da177e4SLinus Torvalds inet_reset_saddr(sk); 12271da177e4SLinus Torvalds 12281da177e4SLinus Torvalds if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 12291da177e4SLinus Torvalds sk->sk_prot->unhash(sk); 1230c720c7e8SEric Dumazet inet->inet_sport = 0; 12311da177e4SLinus Torvalds } 12321da177e4SLinus Torvalds sk_dst_reset(sk); 12331da177e4SLinus Torvalds return 0; 12341da177e4SLinus Torvalds } 1235c482c568SEric Dumazet EXPORT_SYMBOL(udp_disconnect); 12361da177e4SLinus Torvalds 1237645ca708SEric Dumazet void udp_lib_unhash(struct sock *sk) 1238645ca708SEric Dumazet { 1239723b4610SEric Dumazet if (sk_hashed(sk)) { 1240645ca708SEric Dumazet struct udp_table *udptable = sk->sk_prot->h.udp_table; 1241512615b6SEric Dumazet struct udp_hslot *hslot, *hslot2; 1242512615b6SEric Dumazet 1243512615b6SEric Dumazet hslot = udp_hashslot(udptable, sock_net(sk), 1244d4cada4aSEric Dumazet udp_sk(sk)->udp_port_hash); 1245512615b6SEric Dumazet hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1246645ca708SEric Dumazet 1247c8db3fecSEric Dumazet spin_lock_bh(&hslot->lock); 124888ab1932SEric Dumazet if (sk_nulls_del_node_init_rcu(sk)) { 1249fdcc8aa9SEric Dumazet hslot->count--; 1250c720c7e8SEric Dumazet inet_sk(sk)->inet_num = 0; 1251645ca708SEric Dumazet sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1252512615b6SEric Dumazet 1253512615b6SEric Dumazet spin_lock(&hslot2->lock); 1254512615b6SEric Dumazet hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1255512615b6SEric Dumazet hslot2->count--; 1256512615b6SEric Dumazet spin_unlock(&hslot2->lock); 1257645ca708SEric Dumazet } 1258c8db3fecSEric Dumazet spin_unlock_bh(&hslot->lock); 1259645ca708SEric Dumazet } 1260723b4610SEric Dumazet } 1261645ca708SEric Dumazet EXPORT_SYMBOL(udp_lib_unhash); 1262645ca708SEric Dumazet 1263719f8358SEric Dumazet /* 1264719f8358SEric Dumazet * inet_rcv_saddr was changed, we must rehash secondary hash 1265719f8358SEric Dumazet */ 1266719f8358SEric Dumazet void udp_lib_rehash(struct sock *sk, u16 newhash) 1267719f8358SEric Dumazet { 1268719f8358SEric Dumazet if (sk_hashed(sk)) { 1269719f8358SEric Dumazet struct udp_table *udptable = sk->sk_prot->h.udp_table; 1270719f8358SEric Dumazet struct udp_hslot *hslot, *hslot2, *nhslot2; 1271719f8358SEric Dumazet 1272719f8358SEric Dumazet hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1273719f8358SEric Dumazet nhslot2 = udp_hashslot2(udptable, newhash); 1274719f8358SEric Dumazet udp_sk(sk)->udp_portaddr_hash = newhash; 1275719f8358SEric Dumazet if (hslot2 != nhslot2) { 1276719f8358SEric Dumazet hslot = udp_hashslot(udptable, sock_net(sk), 1277719f8358SEric Dumazet udp_sk(sk)->udp_port_hash); 1278719f8358SEric Dumazet /* we must lock primary chain too */ 1279719f8358SEric Dumazet spin_lock_bh(&hslot->lock); 1280719f8358SEric Dumazet 1281719f8358SEric Dumazet spin_lock(&hslot2->lock); 1282719f8358SEric Dumazet hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1283719f8358SEric Dumazet hslot2->count--; 1284719f8358SEric Dumazet spin_unlock(&hslot2->lock); 1285719f8358SEric Dumazet 1286719f8358SEric Dumazet spin_lock(&nhslot2->lock); 1287719f8358SEric Dumazet hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1288719f8358SEric Dumazet &nhslot2->head); 1289719f8358SEric Dumazet nhslot2->count++; 1290719f8358SEric Dumazet spin_unlock(&nhslot2->lock); 1291719f8358SEric Dumazet 1292719f8358SEric Dumazet spin_unlock_bh(&hslot->lock); 1293719f8358SEric Dumazet } 1294719f8358SEric Dumazet } 1295719f8358SEric Dumazet } 1296719f8358SEric Dumazet EXPORT_SYMBOL(udp_lib_rehash); 1297719f8358SEric Dumazet 1298719f8358SEric Dumazet static void udp_v4_rehash(struct sock *sk) 1299719f8358SEric Dumazet { 1300719f8358SEric Dumazet u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1301719f8358SEric Dumazet inet_sk(sk)->inet_rcv_saddr, 1302719f8358SEric Dumazet inet_sk(sk)->inet_num); 1303719f8358SEric Dumazet udp_lib_rehash(sk, new_hash); 1304719f8358SEric Dumazet } 1305719f8358SEric Dumazet 130693821778SHerbert Xu static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 130793821778SHerbert Xu { 1308fec5e652STom Herbert int rc; 130993821778SHerbert Xu 1310fec5e652STom Herbert if (inet_sk(sk)->inet_daddr) 1311c58dc01bSDavid S. Miller sock_rps_save_rxhash(sk, skb->rxhash); 1312fec5e652STom Herbert 1313f84af32cSEric Dumazet rc = ip_queue_rcv_skb(sk, skb); 1314766e9037SEric Dumazet if (rc < 0) { 1315766e9037SEric Dumazet int is_udplite = IS_UDPLITE(sk); 1316766e9037SEric Dumazet 131793821778SHerbert Xu /* Note that an ENOMEM error is charged twice */ 1318766e9037SEric Dumazet if (rc == -ENOMEM) 131993821778SHerbert Xu UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 132093821778SHerbert Xu is_udplite); 1321766e9037SEric Dumazet UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1322766e9037SEric Dumazet kfree_skb(skb); 1323766e9037SEric Dumazet return -1; 132493821778SHerbert Xu } 132593821778SHerbert Xu 132693821778SHerbert Xu return 0; 132793821778SHerbert Xu 132893821778SHerbert Xu } 132993821778SHerbert Xu 1330db8dac20SDavid S. Miller /* returns: 1331db8dac20SDavid S. Miller * -1: error 1332db8dac20SDavid S. Miller * 0: success 1333db8dac20SDavid S. Miller * >0: "udp encap" protocol resubmission 1334db8dac20SDavid S. Miller * 1335db8dac20SDavid S. Miller * Note that in the success and error cases, the skb is assumed to 1336db8dac20SDavid S. Miller * have either been requeued or freed. 1337db8dac20SDavid S. Miller */ 1338db8dac20SDavid S. Miller int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1339db8dac20SDavid S. Miller { 1340db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 1341db8dac20SDavid S. Miller int rc; 1342db8dac20SDavid S. Miller int is_udplite = IS_UDPLITE(sk); 1343db8dac20SDavid S. Miller 1344db8dac20SDavid S. Miller /* 1345db8dac20SDavid S. Miller * Charge it to the socket, dropping if the queue is full. 1346db8dac20SDavid S. Miller */ 1347db8dac20SDavid S. Miller if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1348db8dac20SDavid S. Miller goto drop; 1349db8dac20SDavid S. Miller nf_reset(skb); 1350db8dac20SDavid S. Miller 1351db8dac20SDavid S. Miller if (up->encap_type) { 1352db8dac20SDavid S. Miller /* 1353db8dac20SDavid S. Miller * This is an encapsulation socket so pass the skb to 1354db8dac20SDavid S. Miller * the socket's udp_encap_rcv() hook. Otherwise, just 1355db8dac20SDavid S. Miller * fall through and pass this up the UDP socket. 1356db8dac20SDavid S. Miller * up->encap_rcv() returns the following value: 1357db8dac20SDavid S. Miller * =0 if skb was successfully passed to the encap 1358db8dac20SDavid S. Miller * handler or was discarded by it. 1359db8dac20SDavid S. Miller * >0 if skb should be passed on to UDP. 1360db8dac20SDavid S. Miller * <0 if skb should be resubmitted as proto -N 1361db8dac20SDavid S. Miller */ 1362db8dac20SDavid S. Miller 1363db8dac20SDavid S. Miller /* if we're overly short, let UDP handle it */ 1364db8dac20SDavid S. Miller if (skb->len > sizeof(struct udphdr) && 1365db8dac20SDavid S. Miller up->encap_rcv != NULL) { 1366db8dac20SDavid S. Miller int ret; 1367db8dac20SDavid S. Miller 1368db8dac20SDavid S. Miller ret = (*up->encap_rcv)(sk, skb); 1369db8dac20SDavid S. Miller if (ret <= 0) { 13700283328eSPavel Emelyanov UDP_INC_STATS_BH(sock_net(sk), 13710283328eSPavel Emelyanov UDP_MIB_INDATAGRAMS, 1372db8dac20SDavid S. Miller is_udplite); 1373db8dac20SDavid S. Miller return -ret; 1374db8dac20SDavid S. Miller } 1375db8dac20SDavid S. Miller } 1376db8dac20SDavid S. Miller 1377db8dac20SDavid S. Miller /* FALLTHROUGH -- it's a UDP Packet */ 1378db8dac20SDavid S. Miller } 1379db8dac20SDavid S. Miller 1380db8dac20SDavid S. Miller /* 1381db8dac20SDavid S. Miller * UDP-Lite specific tests, ignored on UDP sockets 1382db8dac20SDavid S. Miller */ 1383db8dac20SDavid S. Miller if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1384db8dac20SDavid S. Miller 1385db8dac20SDavid S. Miller /* 1386db8dac20SDavid S. Miller * MIB statistics other than incrementing the error count are 1387db8dac20SDavid S. Miller * disabled for the following two types of errors: these depend 1388db8dac20SDavid S. Miller * on the application settings, not on the functioning of the 1389db8dac20SDavid S. Miller * protocol stack as such. 1390db8dac20SDavid S. Miller * 1391db8dac20SDavid S. Miller * RFC 3828 here recommends (sec 3.3): "There should also be a 1392db8dac20SDavid S. Miller * way ... to ... at least let the receiving application block 1393db8dac20SDavid S. Miller * delivery of packets with coverage values less than a value 1394db8dac20SDavid S. Miller * provided by the application." 1395db8dac20SDavid S. Miller */ 1396db8dac20SDavid S. Miller if (up->pcrlen == 0) { /* full coverage was set */ 1397db8dac20SDavid S. Miller LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1398db8dac20SDavid S. Miller "%d while full coverage %d requested\n", 1399db8dac20SDavid S. Miller UDP_SKB_CB(skb)->cscov, skb->len); 1400db8dac20SDavid S. Miller goto drop; 1401db8dac20SDavid S. Miller } 1402db8dac20SDavid S. Miller /* The next case involves violating the min. coverage requested 1403db8dac20SDavid S. Miller * by the receiver. This is subtle: if receiver wants x and x is 1404db8dac20SDavid S. Miller * greater than the buffersize/MTU then receiver will complain 1405db8dac20SDavid S. Miller * that it wants x while sender emits packets of smaller size y. 1406db8dac20SDavid S. Miller * Therefore the above ...()->partial_cov statement is essential. 1407db8dac20SDavid S. Miller */ 1408db8dac20SDavid S. Miller if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1409db8dac20SDavid S. Miller LIMIT_NETDEBUG(KERN_WARNING 1410db8dac20SDavid S. Miller "UDPLITE: coverage %d too small, need min %d\n", 1411db8dac20SDavid S. Miller UDP_SKB_CB(skb)->cscov, up->pcrlen); 1412db8dac20SDavid S. Miller goto drop; 1413db8dac20SDavid S. Miller } 1414db8dac20SDavid S. Miller } 1415db8dac20SDavid S. Miller 1416*0d7da9ddSEric Dumazet if (rcu_dereference_raw(sk->sk_filter)) { 1417db8dac20SDavid S. Miller if (udp_lib_checksum_complete(skb)) 1418db8dac20SDavid S. Miller goto drop; 1419db8dac20SDavid S. Miller } 1420db8dac20SDavid S. Miller 1421c377411fSEric Dumazet 1422c377411fSEric Dumazet if (sk_rcvqueues_full(sk, skb)) 1423c377411fSEric Dumazet goto drop; 1424c377411fSEric Dumazet 142593821778SHerbert Xu rc = 0; 1426db8dac20SDavid S. Miller 142793821778SHerbert Xu bh_lock_sock(sk); 142893821778SHerbert Xu if (!sock_owned_by_user(sk)) 142993821778SHerbert Xu rc = __udp_queue_rcv_skb(sk, skb); 1430a3a858ffSZhu Yi else if (sk_add_backlog(sk, skb)) { 143155349790SZhu Yi bh_unlock_sock(sk); 143255349790SZhu Yi goto drop; 143355349790SZhu Yi } 143493821778SHerbert Xu bh_unlock_sock(sk); 143593821778SHerbert Xu 143693821778SHerbert Xu return rc; 1437db8dac20SDavid S. Miller 1438db8dac20SDavid S. Miller drop: 14390283328eSPavel Emelyanov UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 14408edf19c2SEric Dumazet atomic_inc(&sk->sk_drops); 1441db8dac20SDavid S. Miller kfree_skb(skb); 1442db8dac20SDavid S. Miller return -1; 1443db8dac20SDavid S. Miller } 1444db8dac20SDavid S. Miller 14451240d137SEric Dumazet 14461240d137SEric Dumazet static void flush_stack(struct sock **stack, unsigned int count, 14471240d137SEric Dumazet struct sk_buff *skb, unsigned int final) 14481240d137SEric Dumazet { 14491240d137SEric Dumazet unsigned int i; 14501240d137SEric Dumazet struct sk_buff *skb1 = NULL; 1451f6b8f32cSEric Dumazet struct sock *sk; 14521240d137SEric Dumazet 14531240d137SEric Dumazet for (i = 0; i < count; i++) { 1454f6b8f32cSEric Dumazet sk = stack[i]; 14551240d137SEric Dumazet if (likely(skb1 == NULL)) 14561240d137SEric Dumazet skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 14571240d137SEric Dumazet 1458f6b8f32cSEric Dumazet if (!skb1) { 1459f6b8f32cSEric Dumazet atomic_inc(&sk->sk_drops); 1460f6b8f32cSEric Dumazet UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1461f6b8f32cSEric Dumazet IS_UDPLITE(sk)); 1462f6b8f32cSEric Dumazet UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1463f6b8f32cSEric Dumazet IS_UDPLITE(sk)); 1464f6b8f32cSEric Dumazet } 1465f6b8f32cSEric Dumazet 1466f6b8f32cSEric Dumazet if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 14671240d137SEric Dumazet skb1 = NULL; 14681240d137SEric Dumazet } 14691240d137SEric Dumazet if (unlikely(skb1)) 14701240d137SEric Dumazet kfree_skb(skb1); 14711240d137SEric Dumazet } 14721240d137SEric Dumazet 1473db8dac20SDavid S. Miller /* 1474db8dac20SDavid S. Miller * Multicasts and broadcasts go to each listener. 1475db8dac20SDavid S. Miller * 14761240d137SEric Dumazet * Note: called only from the BH handler context. 1477db8dac20SDavid S. Miller */ 1478e3163493SPavel Emelyanov static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1479db8dac20SDavid S. Miller struct udphdr *uh, 1480db8dac20SDavid S. Miller __be32 saddr, __be32 daddr, 1481645ca708SEric Dumazet struct udp_table *udptable) 1482db8dac20SDavid S. Miller { 14831240d137SEric Dumazet struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1484f86dcc5aSEric Dumazet struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1485db8dac20SDavid S. Miller int dif; 14861240d137SEric Dumazet unsigned int i, count = 0; 1487db8dac20SDavid S. Miller 1488645ca708SEric Dumazet spin_lock(&hslot->lock); 148988ab1932SEric Dumazet sk = sk_nulls_head(&hslot->head); 1490db8dac20SDavid S. Miller dif = skb->dev->ifindex; 1491920a4611SEric Dumazet sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 14921240d137SEric Dumazet while (sk) { 14931240d137SEric Dumazet stack[count++] = sk; 14941240d137SEric Dumazet sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 14951240d137SEric Dumazet daddr, uh->source, saddr, dif); 14961240d137SEric Dumazet if (unlikely(count == ARRAY_SIZE(stack))) { 14971240d137SEric Dumazet if (!sk) 14981240d137SEric Dumazet break; 14991240d137SEric Dumazet flush_stack(stack, count, skb, ~0); 15001240d137SEric Dumazet count = 0; 1501db8dac20SDavid S. Miller } 15021240d137SEric Dumazet } 15031240d137SEric Dumazet /* 15041240d137SEric Dumazet * before releasing chain lock, we must take a reference on sockets 15051240d137SEric Dumazet */ 15061240d137SEric Dumazet for (i = 0; i < count; i++) 15071240d137SEric Dumazet sock_hold(stack[i]); 15081240d137SEric Dumazet 1509645ca708SEric Dumazet spin_unlock(&hslot->lock); 15101240d137SEric Dumazet 15111240d137SEric Dumazet /* 15121240d137SEric Dumazet * do the slow work with no lock held 15131240d137SEric Dumazet */ 15141240d137SEric Dumazet if (count) { 15151240d137SEric Dumazet flush_stack(stack, count, skb, count - 1); 15161240d137SEric Dumazet 15171240d137SEric Dumazet for (i = 0; i < count; i++) 15181240d137SEric Dumazet sock_put(stack[i]); 15191240d137SEric Dumazet } else { 15201240d137SEric Dumazet kfree_skb(skb); 15211240d137SEric Dumazet } 1522db8dac20SDavid S. Miller return 0; 1523db8dac20SDavid S. Miller } 1524db8dac20SDavid S. Miller 1525db8dac20SDavid S. Miller /* Initialize UDP checksum. If exited with zero value (success), 1526db8dac20SDavid S. Miller * CHECKSUM_UNNECESSARY means, that no more checks are required. 1527db8dac20SDavid S. Miller * Otherwise, csum completion requires chacksumming packet body, 1528db8dac20SDavid S. Miller * including udp header and folding it to skb->csum. 1529db8dac20SDavid S. Miller */ 1530db8dac20SDavid S. Miller static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1531db8dac20SDavid S. Miller int proto) 1532db8dac20SDavid S. Miller { 1533db8dac20SDavid S. Miller const struct iphdr *iph; 1534db8dac20SDavid S. Miller int err; 1535db8dac20SDavid S. Miller 1536db8dac20SDavid S. Miller UDP_SKB_CB(skb)->partial_cov = 0; 1537db8dac20SDavid S. Miller UDP_SKB_CB(skb)->cscov = skb->len; 1538db8dac20SDavid S. Miller 1539db8dac20SDavid S. Miller if (proto == IPPROTO_UDPLITE) { 1540db8dac20SDavid S. Miller err = udplite_checksum_init(skb, uh); 1541db8dac20SDavid S. Miller if (err) 1542db8dac20SDavid S. Miller return err; 1543db8dac20SDavid S. Miller } 1544db8dac20SDavid S. Miller 1545db8dac20SDavid S. Miller iph = ip_hdr(skb); 1546db8dac20SDavid S. Miller if (uh->check == 0) { 1547db8dac20SDavid S. Miller skb->ip_summed = CHECKSUM_UNNECESSARY; 1548db8dac20SDavid S. Miller } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1549db8dac20SDavid S. Miller if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1550db8dac20SDavid S. Miller proto, skb->csum)) 1551db8dac20SDavid S. Miller skb->ip_summed = CHECKSUM_UNNECESSARY; 1552db8dac20SDavid S. Miller } 1553db8dac20SDavid S. Miller if (!skb_csum_unnecessary(skb)) 1554db8dac20SDavid S. Miller skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1555db8dac20SDavid S. Miller skb->len, proto, 0); 1556db8dac20SDavid S. Miller /* Probably, we should checksum udp header (it should be in cache 1557db8dac20SDavid S. Miller * in any case) and data in tiny packets (< rx copybreak). 1558db8dac20SDavid S. Miller */ 1559db8dac20SDavid S. Miller 1560db8dac20SDavid S. Miller return 0; 1561db8dac20SDavid S. Miller } 1562db8dac20SDavid S. Miller 1563db8dac20SDavid S. Miller /* 1564db8dac20SDavid S. Miller * All we need to do is get the socket, and then do a checksum. 1565db8dac20SDavid S. Miller */ 1566db8dac20SDavid S. Miller 1567645ca708SEric Dumazet int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1568db8dac20SDavid S. Miller int proto) 1569db8dac20SDavid S. Miller { 1570db8dac20SDavid S. Miller struct sock *sk; 15717b5e56f9SJesper Dangaard Brouer struct udphdr *uh; 1572db8dac20SDavid S. Miller unsigned short ulen; 1573adf30907SEric Dumazet struct rtable *rt = skb_rtable(skb); 15742783ef23SJesper Dangaard Brouer __be32 saddr, daddr; 15750283328eSPavel Emelyanov struct net *net = dev_net(skb->dev); 1576db8dac20SDavid S. Miller 1577db8dac20SDavid S. Miller /* 1578db8dac20SDavid S. Miller * Validate the packet. 1579db8dac20SDavid S. Miller */ 1580db8dac20SDavid S. Miller if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1581db8dac20SDavid S. Miller goto drop; /* No space for header. */ 1582db8dac20SDavid S. Miller 15837b5e56f9SJesper Dangaard Brouer uh = udp_hdr(skb); 1584db8dac20SDavid S. Miller ulen = ntohs(uh->len); 1585ccc2d97cSBjørn Mork saddr = ip_hdr(skb)->saddr; 1586ccc2d97cSBjørn Mork daddr = ip_hdr(skb)->daddr; 1587ccc2d97cSBjørn Mork 1588db8dac20SDavid S. Miller if (ulen > skb->len) 1589db8dac20SDavid S. Miller goto short_packet; 1590db8dac20SDavid S. Miller 1591db8dac20SDavid S. Miller if (proto == IPPROTO_UDP) { 1592db8dac20SDavid S. Miller /* UDP validates ulen. */ 1593db8dac20SDavid S. Miller if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1594db8dac20SDavid S. Miller goto short_packet; 1595db8dac20SDavid S. Miller uh = udp_hdr(skb); 1596db8dac20SDavid S. Miller } 1597db8dac20SDavid S. Miller 1598db8dac20SDavid S. Miller if (udp4_csum_init(skb, uh, proto)) 1599db8dac20SDavid S. Miller goto csum_error; 1600db8dac20SDavid S. Miller 1601db8dac20SDavid S. Miller if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1602e3163493SPavel Emelyanov return __udp4_lib_mcast_deliver(net, skb, uh, 1603e3163493SPavel Emelyanov saddr, daddr, udptable); 1604db8dac20SDavid S. Miller 1605607c4aafSKOVACS Krisztian sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1606db8dac20SDavid S. Miller 1607db8dac20SDavid S. Miller if (sk != NULL) { 160893821778SHerbert Xu int ret = udp_queue_rcv_skb(sk, skb); 1609db8dac20SDavid S. Miller sock_put(sk); 1610db8dac20SDavid S. Miller 1611db8dac20SDavid S. Miller /* a return value > 0 means to resubmit the input, but 1612db8dac20SDavid S. Miller * it wants the return to be -protocol, or 0 1613db8dac20SDavid S. Miller */ 1614db8dac20SDavid S. Miller if (ret > 0) 1615db8dac20SDavid S. Miller return -ret; 1616db8dac20SDavid S. Miller return 0; 1617db8dac20SDavid S. Miller } 1618db8dac20SDavid S. Miller 1619db8dac20SDavid S. Miller if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1620db8dac20SDavid S. Miller goto drop; 1621db8dac20SDavid S. Miller nf_reset(skb); 1622db8dac20SDavid S. Miller 1623db8dac20SDavid S. Miller /* No socket. Drop packet silently, if checksum is wrong */ 1624db8dac20SDavid S. Miller if (udp_lib_checksum_complete(skb)) 1625db8dac20SDavid S. Miller goto csum_error; 1626db8dac20SDavid S. Miller 16270283328eSPavel Emelyanov UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1628db8dac20SDavid S. Miller icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1629db8dac20SDavid S. Miller 1630db8dac20SDavid S. Miller /* 1631db8dac20SDavid S. Miller * Hmm. We got an UDP packet to a port to which we 1632db8dac20SDavid S. Miller * don't wanna listen. Ignore it. 1633db8dac20SDavid S. Miller */ 1634db8dac20SDavid S. Miller kfree_skb(skb); 1635db8dac20SDavid S. Miller return 0; 1636db8dac20SDavid S. Miller 1637db8dac20SDavid S. Miller short_packet: 1638673d57e7SHarvey Harrison LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1639db8dac20SDavid S. Miller proto == IPPROTO_UDPLITE ? "-Lite" : "", 1640673d57e7SHarvey Harrison &saddr, 1641db8dac20SDavid S. Miller ntohs(uh->source), 1642db8dac20SDavid S. Miller ulen, 1643db8dac20SDavid S. Miller skb->len, 1644673d57e7SHarvey Harrison &daddr, 1645db8dac20SDavid S. Miller ntohs(uh->dest)); 1646db8dac20SDavid S. Miller goto drop; 1647db8dac20SDavid S. Miller 1648db8dac20SDavid S. Miller csum_error: 1649db8dac20SDavid S. Miller /* 1650db8dac20SDavid S. Miller * RFC1122: OK. Discards the bad packet silently (as far as 1651db8dac20SDavid S. Miller * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1652db8dac20SDavid S. Miller */ 1653673d57e7SHarvey Harrison LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1654db8dac20SDavid S. Miller proto == IPPROTO_UDPLITE ? "-Lite" : "", 1655673d57e7SHarvey Harrison &saddr, 1656db8dac20SDavid S. Miller ntohs(uh->source), 1657673d57e7SHarvey Harrison &daddr, 1658db8dac20SDavid S. Miller ntohs(uh->dest), 1659db8dac20SDavid S. Miller ulen); 1660db8dac20SDavid S. Miller drop: 16610283328eSPavel Emelyanov UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1662db8dac20SDavid S. Miller kfree_skb(skb); 1663db8dac20SDavid S. Miller return 0; 1664db8dac20SDavid S. Miller } 1665db8dac20SDavid S. Miller 1666db8dac20SDavid S. Miller int udp_rcv(struct sk_buff *skb) 1667db8dac20SDavid S. Miller { 1668645ca708SEric Dumazet return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1669db8dac20SDavid S. Miller } 1670db8dac20SDavid S. Miller 16717d06b2e0SBrian Haley void udp_destroy_sock(struct sock *sk) 1672db8dac20SDavid S. Miller { 16738a74ad60SEric Dumazet bool slow = lock_sock_fast(sk); 1674db8dac20SDavid S. Miller udp_flush_pending_frames(sk); 16758a74ad60SEric Dumazet unlock_sock_fast(sk, slow); 1676db8dac20SDavid S. Miller } 1677db8dac20SDavid S. Miller 16781da177e4SLinus Torvalds /* 16791da177e4SLinus Torvalds * Socket option code for UDP 16801da177e4SLinus Torvalds */ 16814c0a6cb0SGerrit Renker int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1682b7058842SDavid S. Miller char __user *optval, unsigned int optlen, 16834c0a6cb0SGerrit Renker int (*push_pending_frames)(struct sock *)) 16841da177e4SLinus Torvalds { 16851da177e4SLinus Torvalds struct udp_sock *up = udp_sk(sk); 16861da177e4SLinus Torvalds int val; 16871da177e4SLinus Torvalds int err = 0; 1688b2bf1e26SWang Chen int is_udplite = IS_UDPLITE(sk); 16891da177e4SLinus Torvalds 16901da177e4SLinus Torvalds if (optlen < sizeof(int)) 16911da177e4SLinus Torvalds return -EINVAL; 16921da177e4SLinus Torvalds 16931da177e4SLinus Torvalds if (get_user(val, (int __user *)optval)) 16941da177e4SLinus Torvalds return -EFAULT; 16951da177e4SLinus Torvalds 16961da177e4SLinus Torvalds switch (optname) { 16971da177e4SLinus Torvalds case UDP_CORK: 16981da177e4SLinus Torvalds if (val != 0) { 16991da177e4SLinus Torvalds up->corkflag = 1; 17001da177e4SLinus Torvalds } else { 17011da177e4SLinus Torvalds up->corkflag = 0; 17021da177e4SLinus Torvalds lock_sock(sk); 17034c0a6cb0SGerrit Renker (*push_pending_frames)(sk); 17041da177e4SLinus Torvalds release_sock(sk); 17051da177e4SLinus Torvalds } 17061da177e4SLinus Torvalds break; 17071da177e4SLinus Torvalds 17081da177e4SLinus Torvalds case UDP_ENCAP: 17091da177e4SLinus Torvalds switch (val) { 17101da177e4SLinus Torvalds case 0: 17111da177e4SLinus Torvalds case UDP_ENCAP_ESPINUDP: 17121da177e4SLinus Torvalds case UDP_ENCAP_ESPINUDP_NON_IKE: 1713067b207bSJames Chapman up->encap_rcv = xfrm4_udp_encap_rcv; 1714067b207bSJames Chapman /* FALLTHROUGH */ 1715342f0234SJames Chapman case UDP_ENCAP_L2TPINUDP: 17161da177e4SLinus Torvalds up->encap_type = val; 17171da177e4SLinus Torvalds break; 17181da177e4SLinus Torvalds default: 17191da177e4SLinus Torvalds err = -ENOPROTOOPT; 17201da177e4SLinus Torvalds break; 17211da177e4SLinus Torvalds } 17221da177e4SLinus Torvalds break; 17231da177e4SLinus Torvalds 1724ba4e58ecSGerrit Renker /* 1725ba4e58ecSGerrit Renker * UDP-Lite's partial checksum coverage (RFC 3828). 1726ba4e58ecSGerrit Renker */ 1727ba4e58ecSGerrit Renker /* The sender sets actual checksum coverage length via this option. 1728ba4e58ecSGerrit Renker * The case coverage > packet length is handled by send module. */ 1729ba4e58ecSGerrit Renker case UDPLITE_SEND_CSCOV: 1730b2bf1e26SWang Chen if (!is_udplite) /* Disable the option on UDP sockets */ 1731ba4e58ecSGerrit Renker return -ENOPROTOOPT; 1732ba4e58ecSGerrit Renker if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1733ba4e58ecSGerrit Renker val = 8; 17344be929beSAlexey Dobriyan else if (val > USHRT_MAX) 17354be929beSAlexey Dobriyan val = USHRT_MAX; 1736ba4e58ecSGerrit Renker up->pcslen = val; 1737ba4e58ecSGerrit Renker up->pcflag |= UDPLITE_SEND_CC; 1738ba4e58ecSGerrit Renker break; 1739ba4e58ecSGerrit Renker 1740ba4e58ecSGerrit Renker /* The receiver specifies a minimum checksum coverage value. To make 1741ba4e58ecSGerrit Renker * sense, this should be set to at least 8 (as done below). If zero is 1742ba4e58ecSGerrit Renker * used, this again means full checksum coverage. */ 1743ba4e58ecSGerrit Renker case UDPLITE_RECV_CSCOV: 1744b2bf1e26SWang Chen if (!is_udplite) /* Disable the option on UDP sockets */ 1745ba4e58ecSGerrit Renker return -ENOPROTOOPT; 1746ba4e58ecSGerrit Renker if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1747ba4e58ecSGerrit Renker val = 8; 17484be929beSAlexey Dobriyan else if (val > USHRT_MAX) 17494be929beSAlexey Dobriyan val = USHRT_MAX; 1750ba4e58ecSGerrit Renker up->pcrlen = val; 1751ba4e58ecSGerrit Renker up->pcflag |= UDPLITE_RECV_CC; 1752ba4e58ecSGerrit Renker break; 1753ba4e58ecSGerrit Renker 17541da177e4SLinus Torvalds default: 17551da177e4SLinus Torvalds err = -ENOPROTOOPT; 17561da177e4SLinus Torvalds break; 17576516c655SStephen Hemminger } 17581da177e4SLinus Torvalds 17591da177e4SLinus Torvalds return err; 17601da177e4SLinus Torvalds } 1761c482c568SEric Dumazet EXPORT_SYMBOL(udp_lib_setsockopt); 17621da177e4SLinus Torvalds 1763db8dac20SDavid S. Miller int udp_setsockopt(struct sock *sk, int level, int optname, 1764b7058842SDavid S. Miller char __user *optval, unsigned int optlen) 1765db8dac20SDavid S. Miller { 1766db8dac20SDavid S. Miller if (level == SOL_UDP || level == SOL_UDPLITE) 1767db8dac20SDavid S. Miller return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1768db8dac20SDavid S. Miller udp_push_pending_frames); 1769db8dac20SDavid S. Miller return ip_setsockopt(sk, level, optname, optval, optlen); 1770db8dac20SDavid S. Miller } 1771db8dac20SDavid S. Miller 1772db8dac20SDavid S. Miller #ifdef CONFIG_COMPAT 1773db8dac20SDavid S. Miller int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1774b7058842SDavid S. Miller char __user *optval, unsigned int optlen) 1775db8dac20SDavid S. Miller { 1776db8dac20SDavid S. Miller if (level == SOL_UDP || level == SOL_UDPLITE) 1777db8dac20SDavid S. Miller return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1778db8dac20SDavid S. Miller udp_push_pending_frames); 1779db8dac20SDavid S. Miller return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1780db8dac20SDavid S. Miller } 1781db8dac20SDavid S. Miller #endif 1782db8dac20SDavid S. Miller 17834c0a6cb0SGerrit Renker int udp_lib_getsockopt(struct sock *sk, int level, int optname, 17841da177e4SLinus Torvalds char __user *optval, int __user *optlen) 17851da177e4SLinus Torvalds { 17861da177e4SLinus Torvalds struct udp_sock *up = udp_sk(sk); 17871da177e4SLinus Torvalds int val, len; 17881da177e4SLinus Torvalds 17891da177e4SLinus Torvalds if (get_user(len, optlen)) 17901da177e4SLinus Torvalds return -EFAULT; 17911da177e4SLinus Torvalds 17921da177e4SLinus Torvalds len = min_t(unsigned int, len, sizeof(int)); 17931da177e4SLinus Torvalds 17941da177e4SLinus Torvalds if (len < 0) 17951da177e4SLinus Torvalds return -EINVAL; 17961da177e4SLinus Torvalds 17971da177e4SLinus Torvalds switch (optname) { 17981da177e4SLinus Torvalds case UDP_CORK: 17991da177e4SLinus Torvalds val = up->corkflag; 18001da177e4SLinus Torvalds break; 18011da177e4SLinus Torvalds 18021da177e4SLinus Torvalds case UDP_ENCAP: 18031da177e4SLinus Torvalds val = up->encap_type; 18041da177e4SLinus Torvalds break; 18051da177e4SLinus Torvalds 1806ba4e58ecSGerrit Renker /* The following two cannot be changed on UDP sockets, the return is 1807ba4e58ecSGerrit Renker * always 0 (which corresponds to the full checksum coverage of UDP). */ 1808ba4e58ecSGerrit Renker case UDPLITE_SEND_CSCOV: 1809ba4e58ecSGerrit Renker val = up->pcslen; 1810ba4e58ecSGerrit Renker break; 1811ba4e58ecSGerrit Renker 1812ba4e58ecSGerrit Renker case UDPLITE_RECV_CSCOV: 1813ba4e58ecSGerrit Renker val = up->pcrlen; 1814ba4e58ecSGerrit Renker break; 1815ba4e58ecSGerrit Renker 18161da177e4SLinus Torvalds default: 18171da177e4SLinus Torvalds return -ENOPROTOOPT; 18186516c655SStephen Hemminger } 18191da177e4SLinus Torvalds 18201da177e4SLinus Torvalds if (put_user(len, optlen)) 18211da177e4SLinus Torvalds return -EFAULT; 18221da177e4SLinus Torvalds if (copy_to_user(optval, &val, len)) 18231da177e4SLinus Torvalds return -EFAULT; 18241da177e4SLinus Torvalds return 0; 18251da177e4SLinus Torvalds } 1826c482c568SEric Dumazet EXPORT_SYMBOL(udp_lib_getsockopt); 18271da177e4SLinus Torvalds 1828db8dac20SDavid S. Miller int udp_getsockopt(struct sock *sk, int level, int optname, 1829db8dac20SDavid S. Miller char __user *optval, int __user *optlen) 1830db8dac20SDavid S. Miller { 1831db8dac20SDavid S. Miller if (level == SOL_UDP || level == SOL_UDPLITE) 1832db8dac20SDavid S. Miller return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1833db8dac20SDavid S. Miller return ip_getsockopt(sk, level, optname, optval, optlen); 1834db8dac20SDavid S. Miller } 1835db8dac20SDavid S. Miller 1836db8dac20SDavid S. Miller #ifdef CONFIG_COMPAT 1837db8dac20SDavid S. Miller int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1838db8dac20SDavid S. Miller char __user *optval, int __user *optlen) 1839db8dac20SDavid S. Miller { 1840db8dac20SDavid S. Miller if (level == SOL_UDP || level == SOL_UDPLITE) 1841db8dac20SDavid S. Miller return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1842db8dac20SDavid S. Miller return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1843db8dac20SDavid S. Miller } 1844db8dac20SDavid S. Miller #endif 18451da177e4SLinus Torvalds /** 18461da177e4SLinus Torvalds * udp_poll - wait for a UDP event. 18471da177e4SLinus Torvalds * @file - file struct 18481da177e4SLinus Torvalds * @sock - socket 18491da177e4SLinus Torvalds * @wait - poll table 18501da177e4SLinus Torvalds * 18511da177e4SLinus Torvalds * This is same as datagram poll, except for the special case of 18521da177e4SLinus Torvalds * blocking sockets. If application is using a blocking fd 18531da177e4SLinus Torvalds * and a packet with checksum error is in the queue; 18541da177e4SLinus Torvalds * then it could get return from select indicating data available 18551da177e4SLinus Torvalds * but then block when reading it. Add special case code 18561da177e4SLinus Torvalds * to work around these arguably broken applications. 18571da177e4SLinus Torvalds */ 18581da177e4SLinus Torvalds unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 18591da177e4SLinus Torvalds { 18601da177e4SLinus Torvalds unsigned int mask = datagram_poll(file, sock, wait); 18611da177e4SLinus Torvalds struct sock *sk = sock->sk; 18621da177e4SLinus Torvalds 18631da177e4SLinus Torvalds /* Check for false positives due to checksum errors */ 186485584672SEric Dumazet if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 186585584672SEric Dumazet !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 18661da177e4SLinus Torvalds mask &= ~(POLLIN | POLLRDNORM); 18671da177e4SLinus Torvalds 18681da177e4SLinus Torvalds return mask; 18691da177e4SLinus Torvalds 18701da177e4SLinus Torvalds } 1871c482c568SEric Dumazet EXPORT_SYMBOL(udp_poll); 18721da177e4SLinus Torvalds 1873db8dac20SDavid S. Miller struct proto udp_prot = { 1874db8dac20SDavid S. Miller .name = "UDP", 1875db8dac20SDavid S. Miller .owner = THIS_MODULE, 1876db8dac20SDavid S. Miller .close = udp_lib_close, 1877db8dac20SDavid S. Miller .connect = ip4_datagram_connect, 1878db8dac20SDavid S. Miller .disconnect = udp_disconnect, 1879db8dac20SDavid S. Miller .ioctl = udp_ioctl, 1880db8dac20SDavid S. Miller .destroy = udp_destroy_sock, 1881db8dac20SDavid S. Miller .setsockopt = udp_setsockopt, 1882db8dac20SDavid S. Miller .getsockopt = udp_getsockopt, 1883db8dac20SDavid S. Miller .sendmsg = udp_sendmsg, 1884db8dac20SDavid S. Miller .recvmsg = udp_recvmsg, 1885db8dac20SDavid S. Miller .sendpage = udp_sendpage, 188693821778SHerbert Xu .backlog_rcv = __udp_queue_rcv_skb, 1887db8dac20SDavid S. Miller .hash = udp_lib_hash, 1888db8dac20SDavid S. Miller .unhash = udp_lib_unhash, 1889719f8358SEric Dumazet .rehash = udp_v4_rehash, 1890db8dac20SDavid S. Miller .get_port = udp_v4_get_port, 1891db8dac20SDavid S. Miller .memory_allocated = &udp_memory_allocated, 1892db8dac20SDavid S. Miller .sysctl_mem = sysctl_udp_mem, 1893db8dac20SDavid S. Miller .sysctl_wmem = &sysctl_udp_wmem_min, 1894db8dac20SDavid S. Miller .sysctl_rmem = &sysctl_udp_rmem_min, 1895db8dac20SDavid S. Miller .obj_size = sizeof(struct udp_sock), 1896271b72c7SEric Dumazet .slab_flags = SLAB_DESTROY_BY_RCU, 1897645ca708SEric Dumazet .h.udp_table = &udp_table, 1898db8dac20SDavid S. Miller #ifdef CONFIG_COMPAT 1899db8dac20SDavid S. Miller .compat_setsockopt = compat_udp_setsockopt, 1900db8dac20SDavid S. Miller .compat_getsockopt = compat_udp_getsockopt, 1901db8dac20SDavid S. Miller #endif 1902db8dac20SDavid S. Miller }; 1903c482c568SEric Dumazet EXPORT_SYMBOL(udp_prot); 19041da177e4SLinus Torvalds 19051da177e4SLinus Torvalds /* ------------------------------------------------------------------------ */ 19061da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 19071da177e4SLinus Torvalds 1908645ca708SEric Dumazet static struct sock *udp_get_first(struct seq_file *seq, int start) 19091da177e4SLinus Torvalds { 19101da177e4SLinus Torvalds struct sock *sk; 19111da177e4SLinus Torvalds struct udp_iter_state *state = seq->private; 19126f191efeSDenis V. Lunev struct net *net = seq_file_net(seq); 19131da177e4SLinus Torvalds 1914f86dcc5aSEric Dumazet for (state->bucket = start; state->bucket <= state->udp_table->mask; 1915f86dcc5aSEric Dumazet ++state->bucket) { 191688ab1932SEric Dumazet struct hlist_nulls_node *node; 1917645ca708SEric Dumazet struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 1918f86dcc5aSEric Dumazet 1919f86dcc5aSEric Dumazet if (hlist_nulls_empty(&hslot->head)) 1920f86dcc5aSEric Dumazet continue; 1921f86dcc5aSEric Dumazet 1922645ca708SEric Dumazet spin_lock_bh(&hslot->lock); 192388ab1932SEric Dumazet sk_nulls_for_each(sk, node, &hslot->head) { 1924878628fbSYOSHIFUJI Hideaki if (!net_eq(sock_net(sk), net)) 1925a91275efSDaniel Lezcano continue; 19261da177e4SLinus Torvalds if (sk->sk_family == state->family) 19271da177e4SLinus Torvalds goto found; 19281da177e4SLinus Torvalds } 1929645ca708SEric Dumazet spin_unlock_bh(&hslot->lock); 19301da177e4SLinus Torvalds } 19311da177e4SLinus Torvalds sk = NULL; 19321da177e4SLinus Torvalds found: 19331da177e4SLinus Torvalds return sk; 19341da177e4SLinus Torvalds } 19351da177e4SLinus Torvalds 19361da177e4SLinus Torvalds static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 19371da177e4SLinus Torvalds { 19381da177e4SLinus Torvalds struct udp_iter_state *state = seq->private; 19396f191efeSDenis V. Lunev struct net *net = seq_file_net(seq); 19401da177e4SLinus Torvalds 19411da177e4SLinus Torvalds do { 194288ab1932SEric Dumazet sk = sk_nulls_next(sk); 1943878628fbSYOSHIFUJI Hideaki } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 19441da177e4SLinus Torvalds 1945645ca708SEric Dumazet if (!sk) { 1946f86dcc5aSEric Dumazet if (state->bucket <= state->udp_table->mask) 1947f52b5054SEric Dumazet spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1948645ca708SEric Dumazet return udp_get_first(seq, state->bucket + 1); 19491da177e4SLinus Torvalds } 19501da177e4SLinus Torvalds return sk; 19511da177e4SLinus Torvalds } 19521da177e4SLinus Torvalds 19531da177e4SLinus Torvalds static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 19541da177e4SLinus Torvalds { 1955645ca708SEric Dumazet struct sock *sk = udp_get_first(seq, 0); 19561da177e4SLinus Torvalds 19571da177e4SLinus Torvalds if (sk) 19581da177e4SLinus Torvalds while (pos && (sk = udp_get_next(seq, sk)) != NULL) 19591da177e4SLinus Torvalds --pos; 19601da177e4SLinus Torvalds return pos ? NULL : sk; 19611da177e4SLinus Torvalds } 19621da177e4SLinus Torvalds 19631da177e4SLinus Torvalds static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 19641da177e4SLinus Torvalds { 196530842f29SVitaly Mayatskikh struct udp_iter_state *state = seq->private; 1966f86dcc5aSEric Dumazet state->bucket = MAX_UDP_PORTS; 196730842f29SVitaly Mayatskikh 1968b50660f1SYOSHIFUJI Hideaki return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 19691da177e4SLinus Torvalds } 19701da177e4SLinus Torvalds 19711da177e4SLinus Torvalds static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 19721da177e4SLinus Torvalds { 19731da177e4SLinus Torvalds struct sock *sk; 19741da177e4SLinus Torvalds 1975b50660f1SYOSHIFUJI Hideaki if (v == SEQ_START_TOKEN) 19761da177e4SLinus Torvalds sk = udp_get_idx(seq, 0); 19771da177e4SLinus Torvalds else 19781da177e4SLinus Torvalds sk = udp_get_next(seq, v); 19791da177e4SLinus Torvalds 19801da177e4SLinus Torvalds ++*pos; 19811da177e4SLinus Torvalds return sk; 19821da177e4SLinus Torvalds } 19831da177e4SLinus Torvalds 19841da177e4SLinus Torvalds static void udp_seq_stop(struct seq_file *seq, void *v) 19851da177e4SLinus Torvalds { 1986645ca708SEric Dumazet struct udp_iter_state *state = seq->private; 1987645ca708SEric Dumazet 1988f86dcc5aSEric Dumazet if (state->bucket <= state->udp_table->mask) 1989645ca708SEric Dumazet spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 19901da177e4SLinus Torvalds } 19911da177e4SLinus Torvalds 19921da177e4SLinus Torvalds static int udp_seq_open(struct inode *inode, struct file *file) 19931da177e4SLinus Torvalds { 19941da177e4SLinus Torvalds struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1995a2be75c1SDenis V. Lunev struct udp_iter_state *s; 1996a2be75c1SDenis V. Lunev int err; 19971da177e4SLinus Torvalds 1998a2be75c1SDenis V. Lunev err = seq_open_net(inode, file, &afinfo->seq_ops, 1999a2be75c1SDenis V. Lunev sizeof(struct udp_iter_state)); 2000a2be75c1SDenis V. Lunev if (err < 0) 2001a2be75c1SDenis V. Lunev return err; 2002a91275efSDaniel Lezcano 2003a2be75c1SDenis V. Lunev s = ((struct seq_file *)file->private_data)->private; 20041da177e4SLinus Torvalds s->family = afinfo->family; 2005645ca708SEric Dumazet s->udp_table = afinfo->udp_table; 2006a2be75c1SDenis V. Lunev return err; 2007a91275efSDaniel Lezcano } 2008a91275efSDaniel Lezcano 20091da177e4SLinus Torvalds /* ------------------------------------------------------------------------ */ 20100c96d8c5SDaniel Lezcano int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 20111da177e4SLinus Torvalds { 20121da177e4SLinus Torvalds struct proc_dir_entry *p; 20131da177e4SLinus Torvalds int rc = 0; 20141da177e4SLinus Torvalds 20153ba9441bSDenis V. Lunev afinfo->seq_fops.open = udp_seq_open; 20163ba9441bSDenis V. Lunev afinfo->seq_fops.read = seq_read; 20173ba9441bSDenis V. Lunev afinfo->seq_fops.llseek = seq_lseek; 20183ba9441bSDenis V. Lunev afinfo->seq_fops.release = seq_release_net; 20191da177e4SLinus Torvalds 2020dda61925SDenis V. Lunev afinfo->seq_ops.start = udp_seq_start; 2021dda61925SDenis V. Lunev afinfo->seq_ops.next = udp_seq_next; 2022dda61925SDenis V. Lunev afinfo->seq_ops.stop = udp_seq_stop; 2023dda61925SDenis V. Lunev 202484841c3cSDenis V. Lunev p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 202584841c3cSDenis V. Lunev &afinfo->seq_fops, afinfo); 202684841c3cSDenis V. Lunev if (!p) 20271da177e4SLinus Torvalds rc = -ENOMEM; 20281da177e4SLinus Torvalds return rc; 20291da177e4SLinus Torvalds } 2030c482c568SEric Dumazet EXPORT_SYMBOL(udp_proc_register); 20311da177e4SLinus Torvalds 20320c96d8c5SDaniel Lezcano void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 20331da177e4SLinus Torvalds { 20340c96d8c5SDaniel Lezcano proc_net_remove(net, afinfo->name); 20351da177e4SLinus Torvalds } 2036c482c568SEric Dumazet EXPORT_SYMBOL(udp_proc_unregister); 2037db8dac20SDavid S. Miller 2038db8dac20SDavid S. Miller /* ------------------------------------------------------------------------ */ 20395e659e4cSPavel Emelyanov static void udp4_format_sock(struct sock *sp, struct seq_file *f, 20405e659e4cSPavel Emelyanov int bucket, int *len) 2041db8dac20SDavid S. Miller { 2042db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sp); 2043c720c7e8SEric Dumazet __be32 dest = inet->inet_daddr; 2044c720c7e8SEric Dumazet __be32 src = inet->inet_rcv_saddr; 2045c720c7e8SEric Dumazet __u16 destp = ntohs(inet->inet_dport); 2046c720c7e8SEric Dumazet __u16 srcp = ntohs(inet->inet_sport); 2047db8dac20SDavid S. Miller 2048f86dcc5aSEric Dumazet seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2049cb61cb9bSEric Dumazet " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 2050db8dac20SDavid S. Miller bucket, src, srcp, dest, destp, sp->sk_state, 205131e6d363SEric Dumazet sk_wmem_alloc_get(sp), 205231e6d363SEric Dumazet sk_rmem_alloc_get(sp), 2053db8dac20SDavid S. Miller 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 2054cb61cb9bSEric Dumazet atomic_read(&sp->sk_refcnt), sp, 2055cb61cb9bSEric Dumazet atomic_read(&sp->sk_drops), len); 2056db8dac20SDavid S. Miller } 2057db8dac20SDavid S. Miller 2058db8dac20SDavid S. Miller int udp4_seq_show(struct seq_file *seq, void *v) 2059db8dac20SDavid S. Miller { 2060db8dac20SDavid S. Miller if (v == SEQ_START_TOKEN) 2061db8dac20SDavid S. Miller seq_printf(seq, "%-127s\n", 2062db8dac20SDavid S. Miller " sl local_address rem_address st tx_queue " 2063db8dac20SDavid S. Miller "rx_queue tr tm->when retrnsmt uid timeout " 2064cb61cb9bSEric Dumazet "inode ref pointer drops"); 2065db8dac20SDavid S. Miller else { 2066db8dac20SDavid S. Miller struct udp_iter_state *state = seq->private; 20675e659e4cSPavel Emelyanov int len; 2068db8dac20SDavid S. Miller 20695e659e4cSPavel Emelyanov udp4_format_sock(v, seq, state->bucket, &len); 20705e659e4cSPavel Emelyanov seq_printf(seq, "%*s\n", 127 - len, ""); 2071db8dac20SDavid S. Miller } 2072db8dac20SDavid S. Miller return 0; 2073db8dac20SDavid S. Miller } 2074db8dac20SDavid S. Miller 2075db8dac20SDavid S. Miller /* ------------------------------------------------------------------------ */ 2076db8dac20SDavid S. Miller static struct udp_seq_afinfo udp4_seq_afinfo = { 2077db8dac20SDavid S. Miller .name = "udp", 2078db8dac20SDavid S. Miller .family = AF_INET, 2079645ca708SEric Dumazet .udp_table = &udp_table, 20804ad96d39SDenis V. Lunev .seq_fops = { 20814ad96d39SDenis V. Lunev .owner = THIS_MODULE, 20824ad96d39SDenis V. Lunev }, 2083dda61925SDenis V. Lunev .seq_ops = { 2084dda61925SDenis V. Lunev .show = udp4_seq_show, 2085dda61925SDenis V. Lunev }, 2086db8dac20SDavid S. Miller }; 2087db8dac20SDavid S. Miller 20882c8c1e72SAlexey Dobriyan static int __net_init udp4_proc_init_net(struct net *net) 208915439febSPavel Emelyanov { 209015439febSPavel Emelyanov return udp_proc_register(net, &udp4_seq_afinfo); 209115439febSPavel Emelyanov } 209215439febSPavel Emelyanov 20932c8c1e72SAlexey Dobriyan static void __net_exit udp4_proc_exit_net(struct net *net) 209415439febSPavel Emelyanov { 209515439febSPavel Emelyanov udp_proc_unregister(net, &udp4_seq_afinfo); 209615439febSPavel Emelyanov } 209715439febSPavel Emelyanov 209815439febSPavel Emelyanov static struct pernet_operations udp4_net_ops = { 209915439febSPavel Emelyanov .init = udp4_proc_init_net, 210015439febSPavel Emelyanov .exit = udp4_proc_exit_net, 210115439febSPavel Emelyanov }; 210215439febSPavel Emelyanov 2103db8dac20SDavid S. Miller int __init udp4_proc_init(void) 2104db8dac20SDavid S. Miller { 210515439febSPavel Emelyanov return register_pernet_subsys(&udp4_net_ops); 2106db8dac20SDavid S. Miller } 2107db8dac20SDavid S. Miller 2108db8dac20SDavid S. Miller void udp4_proc_exit(void) 2109db8dac20SDavid S. Miller { 211015439febSPavel Emelyanov unregister_pernet_subsys(&udp4_net_ops); 2111db8dac20SDavid S. Miller } 21121da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 21131da177e4SLinus Torvalds 2114f86dcc5aSEric Dumazet static __initdata unsigned long uhash_entries; 2115f86dcc5aSEric Dumazet static int __init set_uhash_entries(char *str) 2116645ca708SEric Dumazet { 2117f86dcc5aSEric Dumazet if (!str) 2118f86dcc5aSEric Dumazet return 0; 2119f86dcc5aSEric Dumazet uhash_entries = simple_strtoul(str, &str, 0); 2120f86dcc5aSEric Dumazet if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2121f86dcc5aSEric Dumazet uhash_entries = UDP_HTABLE_SIZE_MIN; 2122f86dcc5aSEric Dumazet return 1; 2123f86dcc5aSEric Dumazet } 2124f86dcc5aSEric Dumazet __setup("uhash_entries=", set_uhash_entries); 2125645ca708SEric Dumazet 2126f86dcc5aSEric Dumazet void __init udp_table_init(struct udp_table *table, const char *name) 2127f86dcc5aSEric Dumazet { 2128f86dcc5aSEric Dumazet unsigned int i; 2129f86dcc5aSEric Dumazet 2130f86dcc5aSEric Dumazet if (!CONFIG_BASE_SMALL) 2131f86dcc5aSEric Dumazet table->hash = alloc_large_system_hash(name, 2132512615b6SEric Dumazet 2 * sizeof(struct udp_hslot), 2133f86dcc5aSEric Dumazet uhash_entries, 2134f86dcc5aSEric Dumazet 21, /* one slot per 2 MB */ 2135f86dcc5aSEric Dumazet 0, 2136f86dcc5aSEric Dumazet &table->log, 2137f86dcc5aSEric Dumazet &table->mask, 2138f86dcc5aSEric Dumazet 64 * 1024); 2139f86dcc5aSEric Dumazet /* 2140f86dcc5aSEric Dumazet * Make sure hash table has the minimum size 2141f86dcc5aSEric Dumazet */ 2142f86dcc5aSEric Dumazet if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) { 2143f86dcc5aSEric Dumazet table->hash = kmalloc(UDP_HTABLE_SIZE_MIN * 2144512615b6SEric Dumazet 2 * sizeof(struct udp_hslot), GFP_KERNEL); 2145f86dcc5aSEric Dumazet if (!table->hash) 2146f86dcc5aSEric Dumazet panic(name); 2147f86dcc5aSEric Dumazet table->log = ilog2(UDP_HTABLE_SIZE_MIN); 2148f86dcc5aSEric Dumazet table->mask = UDP_HTABLE_SIZE_MIN - 1; 2149f86dcc5aSEric Dumazet } 2150512615b6SEric Dumazet table->hash2 = table->hash + (table->mask + 1); 2151f86dcc5aSEric Dumazet for (i = 0; i <= table->mask; i++) { 215288ab1932SEric Dumazet INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2153fdcc8aa9SEric Dumazet table->hash[i].count = 0; 2154645ca708SEric Dumazet spin_lock_init(&table->hash[i].lock); 2155645ca708SEric Dumazet } 2156512615b6SEric Dumazet for (i = 0; i <= table->mask; i++) { 2157512615b6SEric Dumazet INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2158512615b6SEric Dumazet table->hash2[i].count = 0; 2159512615b6SEric Dumazet spin_lock_init(&table->hash2[i].lock); 2160512615b6SEric Dumazet } 2161645ca708SEric Dumazet } 2162645ca708SEric Dumazet 216395766fffSHideo Aoki void __init udp_init(void) 216495766fffSHideo Aoki { 21658203efb3SEric Dumazet unsigned long nr_pages, limit; 216695766fffSHideo Aoki 2167f86dcc5aSEric Dumazet udp_table_init(&udp_table, "UDP"); 216895766fffSHideo Aoki /* Set the pressure threshold up by the same strategy of TCP. It is a 216995766fffSHideo Aoki * fraction of global memory that is up to 1/2 at 256 MB, decreasing 217095766fffSHideo Aoki * toward zero with the amount of memory, with a floor of 128 pages. 217195766fffSHideo Aoki */ 21728203efb3SEric Dumazet nr_pages = totalram_pages - totalhigh_pages; 21738203efb3SEric Dumazet limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 21748203efb3SEric Dumazet limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 217595766fffSHideo Aoki limit = max(limit, 128UL); 217695766fffSHideo Aoki sysctl_udp_mem[0] = limit / 4 * 3; 217795766fffSHideo Aoki sysctl_udp_mem[1] = limit; 217895766fffSHideo Aoki sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 217995766fffSHideo Aoki 218095766fffSHideo Aoki sysctl_udp_rmem_min = SK_MEM_QUANTUM; 218195766fffSHideo Aoki sysctl_udp_wmem_min = SK_MEM_QUANTUM; 218295766fffSHideo Aoki } 218395766fffSHideo Aoki 2184d7ca4cc0SSridhar Samudrala int udp4_ufo_send_check(struct sk_buff *skb) 2185d7ca4cc0SSridhar Samudrala { 2186d7ca4cc0SSridhar Samudrala const struct iphdr *iph; 2187d7ca4cc0SSridhar Samudrala struct udphdr *uh; 2188d7ca4cc0SSridhar Samudrala 2189d7ca4cc0SSridhar Samudrala if (!pskb_may_pull(skb, sizeof(*uh))) 2190d7ca4cc0SSridhar Samudrala return -EINVAL; 2191d7ca4cc0SSridhar Samudrala 2192d7ca4cc0SSridhar Samudrala iph = ip_hdr(skb); 2193d7ca4cc0SSridhar Samudrala uh = udp_hdr(skb); 2194d7ca4cc0SSridhar Samudrala 2195d7ca4cc0SSridhar Samudrala uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 2196d7ca4cc0SSridhar Samudrala IPPROTO_UDP, 0); 2197d7ca4cc0SSridhar Samudrala skb->csum_start = skb_transport_header(skb) - skb->head; 2198d7ca4cc0SSridhar Samudrala skb->csum_offset = offsetof(struct udphdr, check); 2199d7ca4cc0SSridhar Samudrala skb->ip_summed = CHECKSUM_PARTIAL; 2200d7ca4cc0SSridhar Samudrala return 0; 2201d7ca4cc0SSridhar Samudrala } 2202d7ca4cc0SSridhar Samudrala 2203d7ca4cc0SSridhar Samudrala struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features) 2204d7ca4cc0SSridhar Samudrala { 2205d7ca4cc0SSridhar Samudrala struct sk_buff *segs = ERR_PTR(-EINVAL); 2206d7ca4cc0SSridhar Samudrala unsigned int mss; 2207d7ca4cc0SSridhar Samudrala int offset; 2208d7ca4cc0SSridhar Samudrala __wsum csum; 2209d7ca4cc0SSridhar Samudrala 2210d7ca4cc0SSridhar Samudrala mss = skb_shinfo(skb)->gso_size; 2211d7ca4cc0SSridhar Samudrala if (unlikely(skb->len <= mss)) 2212d7ca4cc0SSridhar Samudrala goto out; 2213d7ca4cc0SSridhar Samudrala 2214d7ca4cc0SSridhar Samudrala if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 2215d7ca4cc0SSridhar Samudrala /* Packet is from an untrusted source, reset gso_segs. */ 2216d7ca4cc0SSridhar Samudrala int type = skb_shinfo(skb)->gso_type; 2217d7ca4cc0SSridhar Samudrala 2218d7ca4cc0SSridhar Samudrala if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || 2219d7ca4cc0SSridhar Samudrala !(type & (SKB_GSO_UDP)))) 2220d7ca4cc0SSridhar Samudrala goto out; 2221d7ca4cc0SSridhar Samudrala 2222d7ca4cc0SSridhar Samudrala skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 2223d7ca4cc0SSridhar Samudrala 2224d7ca4cc0SSridhar Samudrala segs = NULL; 2225d7ca4cc0SSridhar Samudrala goto out; 2226d7ca4cc0SSridhar Samudrala } 2227d7ca4cc0SSridhar Samudrala 2228d7ca4cc0SSridhar Samudrala /* Do software UFO. Complete and fill in the UDP checksum as HW cannot 2229d7ca4cc0SSridhar Samudrala * do checksum of UDP packets sent as multiple IP fragments. 2230d7ca4cc0SSridhar Samudrala */ 2231d7ca4cc0SSridhar Samudrala offset = skb->csum_start - skb_headroom(skb); 2232d7ca4cc0SSridhar Samudrala csum = skb_checksum(skb, offset, skb->len - offset, 0); 2233d7ca4cc0SSridhar Samudrala offset += skb->csum_offset; 2234d7ca4cc0SSridhar Samudrala *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2235d7ca4cc0SSridhar Samudrala skb->ip_summed = CHECKSUM_NONE; 2236d7ca4cc0SSridhar Samudrala 2237d7ca4cc0SSridhar Samudrala /* Fragment the skb. IP headers of the fragments are updated in 2238d7ca4cc0SSridhar Samudrala * inet_gso_segment() 2239d7ca4cc0SSridhar Samudrala */ 2240d7ca4cc0SSridhar Samudrala segs = skb_segment(skb, features); 2241d7ca4cc0SSridhar Samudrala out: 2242d7ca4cc0SSridhar Samudrala return segs; 2243d7ca4cc0SSridhar Samudrala } 2244d7ca4cc0SSridhar Samudrala 2245