12874c5fdSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * INET An implementation of the TCP/IP protocol suite for the LINUX 41da177e4SLinus Torvalds * operating system. INET is implemented using the BSD Socket 51da177e4SLinus Torvalds * interface as the means of communication with the user level. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * The User Datagram Protocol (UDP). 81da177e4SLinus Torvalds * 902c30a84SJesper Juhl * Authors: Ross Biro 101da177e4SLinus Torvalds * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 111da177e4SLinus Torvalds * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 12113aa838SAlan Cox * Alan Cox, <alan@lxorguk.ukuu.org.uk> 131da177e4SLinus Torvalds * Hirokazu Takahashi, <taka@valinux.co.jp> 141da177e4SLinus Torvalds * 151da177e4SLinus Torvalds * Fixes: 161da177e4SLinus Torvalds * Alan Cox : verify_area() calls 171da177e4SLinus Torvalds * Alan Cox : stopped close while in use off icmp 181da177e4SLinus Torvalds * messages. Not a fix but a botch that 191da177e4SLinus Torvalds * for udp at least is 'valid'. 201da177e4SLinus Torvalds * Alan Cox : Fixed icmp handling properly 211da177e4SLinus Torvalds * Alan Cox : Correct error for oversized datagrams 221da177e4SLinus Torvalds * Alan Cox : Tidied select() semantics. 231da177e4SLinus Torvalds * Alan Cox : udp_err() fixed properly, also now 241da177e4SLinus Torvalds * select and read wake correctly on errors 251da177e4SLinus Torvalds * Alan Cox : udp_send verify_area moved to avoid mem leak 261da177e4SLinus Torvalds * Alan Cox : UDP can count its memory 271da177e4SLinus Torvalds * Alan Cox : send to an unknown connection causes 281da177e4SLinus Torvalds * an ECONNREFUSED off the icmp, but 291da177e4SLinus Torvalds * does NOT close. 301da177e4SLinus Torvalds * Alan Cox : Switched to new sk_buff handlers. No more backlog! 311da177e4SLinus Torvalds * Alan Cox : Using generic datagram code. Even smaller and the PEEK 321da177e4SLinus Torvalds * bug no longer crashes it. 331da177e4SLinus Torvalds * Fred Van Kempen : Net2e support for sk->broadcast. 341da177e4SLinus Torvalds * Alan Cox : Uses skb_free_datagram 351da177e4SLinus Torvalds * Alan Cox : Added get/set sockopt support. 361da177e4SLinus Torvalds * Alan Cox : Broadcasting without option set returns EACCES. 371da177e4SLinus Torvalds * Alan Cox : No wakeup calls. Instead we now use the callbacks. 381da177e4SLinus Torvalds * Alan Cox : Use ip_tos and ip_ttl 391da177e4SLinus Torvalds * Alan Cox : SNMP Mibs 401da177e4SLinus Torvalds * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 411da177e4SLinus Torvalds * Matt Dillon : UDP length checks. 421da177e4SLinus Torvalds * Alan Cox : Smarter af_inet used properly. 431da177e4SLinus Torvalds * Alan Cox : Use new kernel side addressing. 441da177e4SLinus Torvalds * Alan Cox : Incorrect return on truncated datagram receive. 451da177e4SLinus Torvalds * Arnt Gulbrandsen : New udp_send and stuff 461da177e4SLinus Torvalds * Alan Cox : Cache last socket 471da177e4SLinus Torvalds * Alan Cox : Route cache 481da177e4SLinus Torvalds * Jon Peatfield : Minor efficiency fix to sendto(). 491da177e4SLinus Torvalds * Mike Shaver : RFC1122 checks. 501da177e4SLinus Torvalds * Alan Cox : Nonblocking error fix. 511da177e4SLinus Torvalds * Willy Konynenberg : Transparent proxying support. 521da177e4SLinus Torvalds * Mike McLagan : Routing by source 531da177e4SLinus Torvalds * David S. Miller : New socket lookup architecture. 541da177e4SLinus Torvalds * Last socket cache retained as it 551da177e4SLinus Torvalds * does have a high hit rate. 561da177e4SLinus Torvalds * Olaf Kirch : Don't linearise iovec on sendmsg. 571da177e4SLinus Torvalds * Andi Kleen : Some cleanups, cache destination entry 581da177e4SLinus Torvalds * for connect. 591da177e4SLinus Torvalds * Vitaly E. Lavrov : Transparent proxy revived after year coma. 601da177e4SLinus Torvalds * Melvin Smith : Check msg_name not msg_namelen in sendto(), 611da177e4SLinus Torvalds * return ENOTCONN for unconnected sockets (POSIX) 621da177e4SLinus Torvalds * Janos Farkas : don't deliver multi/broadcasts to a different 631da177e4SLinus Torvalds * bound-to-device socket 641da177e4SLinus Torvalds * Hirokazu Takahashi : HW checksumming for outgoing UDP 651da177e4SLinus Torvalds * datagrams. 661da177e4SLinus Torvalds * Hirokazu Takahashi : sendfile() on UDP works now. 671da177e4SLinus Torvalds * Arnaldo C. Melo : convert /proc/net/udp to seq_file 681da177e4SLinus Torvalds * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 691da177e4SLinus Torvalds * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 701da177e4SLinus Torvalds * a single port at the same time. 711da177e4SLinus Torvalds * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 72342f0234SJames Chapman * James Chapman : Add L2TP encapsulation type. 731da177e4SLinus Torvalds */ 741da177e4SLinus Torvalds 75afd46503SJoe Perches #define pr_fmt(fmt) "UDP: " fmt 76afd46503SJoe Perches 77aef2fedaSJakub Kicinski #include <linux/bpf-cgroup.h> 787c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 791da177e4SLinus Torvalds #include <asm/ioctls.h> 8057c8a661SMike Rapoport #include <linux/memblock.h> 818203efb3SEric Dumazet #include <linux/highmem.h> 821da177e4SLinus Torvalds #include <linux/types.h> 831da177e4SLinus Torvalds #include <linux/fcntl.h> 841da177e4SLinus Torvalds #include <linux/module.h> 851da177e4SLinus Torvalds #include <linux/socket.h> 861da177e4SLinus Torvalds #include <linux/sockios.h> 8714c85021SArnaldo Carvalho de Melo #include <linux/igmp.h> 886e540309SShawn Bohrer #include <linux/inetdevice.h> 891da177e4SLinus Torvalds #include <linux/in.h> 901da177e4SLinus Torvalds #include <linux/errno.h> 911da177e4SLinus Torvalds #include <linux/timer.h> 921da177e4SLinus Torvalds #include <linux/mm.h> 931da177e4SLinus Torvalds #include <linux/inet.h> 941da177e4SLinus Torvalds #include <linux/netdevice.h> 955a0e3ad6STejun Heo #include <linux/slab.h> 96c752f073SArnaldo Carvalho de Melo #include <net/tcp_states.h> 971da177e4SLinus Torvalds #include <linux/skbuff.h> 981da177e4SLinus Torvalds #include <linux/proc_fs.h> 991da177e4SLinus Torvalds #include <linux/seq_file.h> 100457c4cbcSEric W. Biederman #include <net/net_namespace.h> 1011da177e4SLinus Torvalds #include <net/icmp.h> 102421b3885SShawn Bohrer #include <net/inet_hashtables.h> 103e7cc0824SStefano Brivio #include <net/ip_tunnels.h> 1041da177e4SLinus Torvalds #include <net/route.h> 1051da177e4SLinus Torvalds #include <net/checksum.h> 106d457a0e3SEric Dumazet #include <net/gso.h> 1071da177e4SLinus Torvalds #include <net/xfrm.h> 108296f7ea7SSatoru Moriya #include <trace/events/udp.h> 109447167bfSEric Dumazet #include <linux/static_key.h> 110951cf368SYonghong Song #include <linux/btf_ids.h> 11122911fc5SEric Dumazet #include <trace/events/skb.h> 112076bb0c8SEliezer Tamir #include <net/busy_poll.h> 113ba4e58ecSGerrit Renker #include "udp_impl.h" 114e32ea7e7SCraig Gallek #include <net/sock_reuseport.h> 115217375a0SEric Dumazet #include <net/addrconf.h> 11660fb9567SPaolo Abeni #include <net/udp_tunnel.h> 1177938cd15SRichard Gobert #include <net/gro.h> 1180146dca7SSabrina Dubroca #if IS_ENABLED(CONFIG_IPV6) 1190146dca7SSabrina Dubroca #include <net/ipv6_stubs.h> 1200146dca7SSabrina Dubroca #endif 1211da177e4SLinus Torvalds 122f86dcc5aSEric Dumazet struct udp_table udp_table __read_mostly; 123645ca708SEric Dumazet EXPORT_SYMBOL(udp_table); 1241da177e4SLinus Torvalds 1258d987e5cSEric Dumazet long sysctl_udp_mem[3] __read_mostly; 12695766fffSHideo Aoki EXPORT_SYMBOL(sysctl_udp_mem); 127c482c568SEric Dumazet 12891b6d325SEric Dumazet atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp; 12995766fffSHideo Aoki EXPORT_SYMBOL(udp_memory_allocated); 1300defbb0aSEric Dumazet DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc); 1310defbb0aSEric Dumazet EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc); 13295766fffSHideo Aoki 133f86dcc5aSEric Dumazet #define MAX_UDP_PORTS 65536 1349804985bSKuniyuki Iwashima #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET) 13598322f22SEric Dumazet 13667fb4330SKuniyuki Iwashima static struct udp_table *udp_get_table_prot(struct sock *sk) 13767fb4330SKuniyuki Iwashima { 13867fb4330SKuniyuki Iwashima return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table; 13967fb4330SKuniyuki Iwashima } 14067fb4330SKuniyuki Iwashima 141f24d43c0SEric Dumazet static int udp_lib_lport_inuse(struct net *net, __u16 num, 142645ca708SEric Dumazet const struct udp_hslot *hslot, 14398322f22SEric Dumazet unsigned long *bitmap, 144fe38d2a1SJosef Bacik struct sock *sk, unsigned int log) 14525030a7fSGerrit Renker { 146f24d43c0SEric Dumazet struct sock *sk2; 147ba418fa3STom Herbert kuid_t uid = sock_i_uid(sk); 14825030a7fSGerrit Renker 149ca065d0cSEric Dumazet sk_for_each(sk2, &hslot->head) { 150f24d43c0SEric Dumazet if (net_eq(sock_net(sk2), net) && 151f24d43c0SEric Dumazet sk2 != sk && 152d4cada4aSEric Dumazet (bitmap || udp_sk(sk2)->udp_port_hash == num) && 153f24d43c0SEric Dumazet (!sk2->sk_reuse || !sk->sk_reuse) && 1549d4fb27dSJoe Perches (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 1559d4fb27dSJoe Perches sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 156fe38d2a1SJosef Bacik inet_rcv_saddr_equal(sk, sk2, true)) { 157df560056SEric Garver if (sk2->sk_reuseport && sk->sk_reuseport && 158df560056SEric Garver !rcu_access_pointer(sk->sk_reuseport_cb) && 159df560056SEric Garver uid_eq(uid, sock_i_uid(sk2))) { 160df560056SEric Garver if (!bitmap) 161df560056SEric Garver return 0; 162df560056SEric Garver } else { 1634243cdc2SJoe Perches if (!bitmap) 164fc038410SDavid S. Miller return 1; 165df560056SEric Garver __set_bit(udp_sk(sk2)->udp_port_hash >> log, 166df560056SEric Garver bitmap); 167df560056SEric Garver } 1684243cdc2SJoe Perches } 16998322f22SEric Dumazet } 17025030a7fSGerrit Renker return 0; 17125030a7fSGerrit Renker } 17225030a7fSGerrit Renker 17330fff923SEric Dumazet /* 17430fff923SEric Dumazet * Note: we still hold spinlock of primary hash chain, so no other writer 17530fff923SEric Dumazet * can insert/delete a socket with local_port == num 17630fff923SEric Dumazet */ 17730fff923SEric Dumazet static int udp_lib_lport_inuse2(struct net *net, __u16 num, 17830fff923SEric Dumazet struct udp_hslot *hslot2, 179fe38d2a1SJosef Bacik struct sock *sk) 18030fff923SEric Dumazet { 18130fff923SEric Dumazet struct sock *sk2; 182ba418fa3STom Herbert kuid_t uid = sock_i_uid(sk); 18330fff923SEric Dumazet int res = 0; 18430fff923SEric Dumazet 18530fff923SEric Dumazet spin_lock(&hslot2->lock); 186ca065d0cSEric Dumazet udp_portaddr_for_each_entry(sk2, &hslot2->head) { 18730fff923SEric Dumazet if (net_eq(sock_net(sk2), net) && 18830fff923SEric Dumazet sk2 != sk && 18930fff923SEric Dumazet (udp_sk(sk2)->udp_port_hash == num) && 19030fff923SEric Dumazet (!sk2->sk_reuse || !sk->sk_reuse) && 1919d4fb27dSJoe Perches (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 1929d4fb27dSJoe Perches sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 193fe38d2a1SJosef Bacik inet_rcv_saddr_equal(sk, sk2, true)) { 194df560056SEric Garver if (sk2->sk_reuseport && sk->sk_reuseport && 195df560056SEric Garver !rcu_access_pointer(sk->sk_reuseport_cb) && 196df560056SEric Garver uid_eq(uid, sock_i_uid(sk2))) { 197df560056SEric Garver res = 0; 198df560056SEric Garver } else { 19930fff923SEric Dumazet res = 1; 200df560056SEric Garver } 20130fff923SEric Dumazet break; 20230fff923SEric Dumazet } 2034243cdc2SJoe Perches } 20430fff923SEric Dumazet spin_unlock(&hslot2->lock); 20530fff923SEric Dumazet return res; 20630fff923SEric Dumazet } 20730fff923SEric Dumazet 208fe38d2a1SJosef Bacik static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 209e32ea7e7SCraig Gallek { 210e32ea7e7SCraig Gallek struct net *net = sock_net(sk); 211e32ea7e7SCraig Gallek kuid_t uid = sock_i_uid(sk); 212e32ea7e7SCraig Gallek struct sock *sk2; 213e32ea7e7SCraig Gallek 214ca065d0cSEric Dumazet sk_for_each(sk2, &hslot->head) { 215e32ea7e7SCraig Gallek if (net_eq(sock_net(sk2), net) && 216e32ea7e7SCraig Gallek sk2 != sk && 217e32ea7e7SCraig Gallek sk2->sk_family == sk->sk_family && 218e32ea7e7SCraig Gallek ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 219e32ea7e7SCraig Gallek (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 220e32ea7e7SCraig Gallek (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 221e32ea7e7SCraig Gallek sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 222fe38d2a1SJosef Bacik inet_rcv_saddr_equal(sk, sk2, false)) { 2232dbb9b9eSMartin KaFai Lau return reuseport_add_sock(sk, sk2, 2242dbb9b9eSMartin KaFai Lau inet_rcv_saddr_any(sk)); 225e32ea7e7SCraig Gallek } 226e32ea7e7SCraig Gallek } 227e32ea7e7SCraig Gallek 2282dbb9b9eSMartin KaFai Lau return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 229e32ea7e7SCraig Gallek } 230e32ea7e7SCraig Gallek 23125030a7fSGerrit Renker /** 2326ba5a3c5SPavel Emelyanov * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 23325030a7fSGerrit Renker * 23425030a7fSGerrit Renker * @sk: socket struct in question 23525030a7fSGerrit Renker * @snum: port number to look up 23625985edcSLucas De Marchi * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 23730fff923SEric Dumazet * with NULL address 23825030a7fSGerrit Renker */ 2396ba5a3c5SPavel Emelyanov int udp_lib_get_port(struct sock *sk, unsigned short snum, 24030fff923SEric Dumazet unsigned int hash2_nulladdr) 2411da177e4SLinus Torvalds { 24267fb4330SKuniyuki Iwashima struct udp_table *udptable = udp_get_table_prot(sk); 243919dfa0bSKuniyuki Iwashima struct udp_hslot *hslot, *hslot2; 2443b1e0a65SYOSHIFUJI Hideaki struct net *net = sock_net(sk); 2457a7160edSKuniyuki Iwashima int error = -EADDRINUSE; 2461da177e4SLinus Torvalds 24732c1da70SStephen Hemminger if (!snum) { 248919dfa0bSKuniyuki Iwashima DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 249919dfa0bSKuniyuki Iwashima unsigned short first, last; 2509088c560SEric Dumazet int low, high, remaining; 25195c96174SEric Dumazet unsigned int rand; 2521da177e4SLinus Torvalds 25391d0b78cSJakub Sitnicki inet_sk_get_local_port_range(sk, &low, &high); 254a25de534SAnton Arapov remaining = (high - low) + 1; 255227b60f5SStephen Hemminger 256a251c17aSJason A. Donenfeld rand = get_random_u32(); 2578fc54f68SDaniel Borkmann first = reciprocal_scale(rand, remaining) + low; 25898322f22SEric Dumazet /* 25998322f22SEric Dumazet * force rand to be an odd multiple of UDP_HTABLE_SIZE 26098322f22SEric Dumazet */ 261f86dcc5aSEric Dumazet rand = (rand | 1) * (udptable->mask + 1); 2625781b235SEric Dumazet last = first + udptable->mask + 1; 2635781b235SEric Dumazet do { 264f86dcc5aSEric Dumazet hslot = udp_hashslot(udptable, net, first); 26598322f22SEric Dumazet bitmap_zero(bitmap, PORTS_PER_CHAIN); 266645ca708SEric Dumazet spin_lock_bh(&hslot->lock); 26798322f22SEric Dumazet udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 268fe38d2a1SJosef Bacik udptable->log); 26998322f22SEric Dumazet 27098322f22SEric Dumazet snum = first; 27198322f22SEric Dumazet /* 27298322f22SEric Dumazet * Iterate on all possible values of snum for this hash. 27398322f22SEric Dumazet * Using steps of an odd multiple of UDP_HTABLE_SIZE 27498322f22SEric Dumazet * give us randomization and full range coverage. 27598322f22SEric Dumazet */ 2769088c560SEric Dumazet do { 27798322f22SEric Dumazet if (low <= snum && snum <= high && 278e3826f1eSAmerigo Wang !test_bit(snum >> udptable->log, bitmap) && 279122ff243SWANG Cong !inet_is_local_reserved_port(net, snum)) 28098322f22SEric Dumazet goto found; 28198322f22SEric Dumazet snum += rand; 28298322f22SEric Dumazet } while (snum != first); 28398322f22SEric Dumazet spin_unlock_bh(&hslot->lock); 284df560056SEric Garver cond_resched(); 2855781b235SEric Dumazet } while (++first != last); 28698322f22SEric Dumazet goto fail; 287645ca708SEric Dumazet } else { 288f86dcc5aSEric Dumazet hslot = udp_hashslot(udptable, net, snum); 289645ca708SEric Dumazet spin_lock_bh(&hslot->lock); 29030fff923SEric Dumazet if (hslot->count > 10) { 29130fff923SEric Dumazet int exist; 29230fff923SEric Dumazet unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 29330fff923SEric Dumazet 29430fff923SEric Dumazet slot2 &= udptable->mask; 29530fff923SEric Dumazet hash2_nulladdr &= udptable->mask; 29630fff923SEric Dumazet 29730fff923SEric Dumazet hslot2 = udp_hashslot2(udptable, slot2); 29830fff923SEric Dumazet if (hslot->count < hslot2->count) 29930fff923SEric Dumazet goto scan_primary_hash; 30030fff923SEric Dumazet 301fe38d2a1SJosef Bacik exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 30230fff923SEric Dumazet if (!exist && (hash2_nulladdr != slot2)) { 30330fff923SEric Dumazet hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 30430fff923SEric Dumazet exist = udp_lib_lport_inuse2(net, snum, hslot2, 305fe38d2a1SJosef Bacik sk); 30630fff923SEric Dumazet } 30730fff923SEric Dumazet if (exist) 30830fff923SEric Dumazet goto fail_unlock; 30930fff923SEric Dumazet else 31030fff923SEric Dumazet goto found; 31130fff923SEric Dumazet } 31230fff923SEric Dumazet scan_primary_hash: 313fe38d2a1SJosef Bacik if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 314645ca708SEric Dumazet goto fail_unlock; 315645ca708SEric Dumazet } 31698322f22SEric Dumazet found: 317c720c7e8SEric Dumazet inet_sk(sk)->inet_num = snum; 318d4cada4aSEric Dumazet udp_sk(sk)->udp_port_hash = snum; 319d4cada4aSEric Dumazet udp_sk(sk)->udp_portaddr_hash ^= snum; 3201da177e4SLinus Torvalds if (sk_unhashed(sk)) { 321e32ea7e7SCraig Gallek if (sk->sk_reuseport && 322fe38d2a1SJosef Bacik udp_reuseport_add_sock(sk, hslot)) { 323e32ea7e7SCraig Gallek inet_sk(sk)->inet_num = 0; 324e32ea7e7SCraig Gallek udp_sk(sk)->udp_port_hash = 0; 325e32ea7e7SCraig Gallek udp_sk(sk)->udp_portaddr_hash ^= snum; 326e32ea7e7SCraig Gallek goto fail_unlock; 327e32ea7e7SCraig Gallek } 328e32ea7e7SCraig Gallek 329ca065d0cSEric Dumazet sk_add_node_rcu(sk, &hslot->head); 330fdcc8aa9SEric Dumazet hslot->count++; 331c29a0bc4SPavel Emelyanov sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 332512615b6SEric Dumazet 333512615b6SEric Dumazet hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 334512615b6SEric Dumazet spin_lock(&hslot2->lock); 335d894ba18SCraig Gallek if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 336d894ba18SCraig Gallek sk->sk_family == AF_INET6) 3371602f49bSDavid S. Miller hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 338d894ba18SCraig Gallek &hslot2->head); 339d894ba18SCraig Gallek else 340ca065d0cSEric Dumazet hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 341512615b6SEric Dumazet &hslot2->head); 342512615b6SEric Dumazet hslot2->count++; 343512615b6SEric Dumazet spin_unlock(&hslot2->lock); 3441da177e4SLinus Torvalds } 345ca065d0cSEric Dumazet sock_set_flag(sk, SOCK_RCU_FREE); 34625030a7fSGerrit Renker error = 0; 347645ca708SEric Dumazet fail_unlock: 348645ca708SEric Dumazet spin_unlock_bh(&hslot->lock); 3491da177e4SLinus Torvalds fail: 35025030a7fSGerrit Renker return error; 3511da177e4SLinus Torvalds } 352c482c568SEric Dumazet EXPORT_SYMBOL(udp_lib_get_port); 3531da177e4SLinus Torvalds 3546ba5a3c5SPavel Emelyanov int udp_v4_get_port(struct sock *sk, unsigned short snum) 355db8dac20SDavid S. Miller { 35630fff923SEric Dumazet unsigned int hash2_nulladdr = 357f0b1e64cSMartin KaFai Lau ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 35830fff923SEric Dumazet unsigned int hash2_partial = 359f0b1e64cSMartin KaFai Lau ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 36030fff923SEric Dumazet 361d4cada4aSEric Dumazet /* precompute partial secondary hash */ 36230fff923SEric Dumazet udp_sk(sk)->udp_portaddr_hash = hash2_partial; 363fe38d2a1SJosef Bacik return udp_lib_get_port(sk, snum, hash2_nulladdr); 364db8dac20SDavid S. Miller } 365db8dac20SDavid S. Miller 366d1e37288SSu, Xuemin static int compute_score(struct sock *sk, struct net *net, 367d1e37288SSu, Xuemin __be32 saddr, __be16 sport, 368fb74c277SDavid Ahern __be32 daddr, unsigned short hnum, 36973545373STim Beale int dif, int sdif) 370645ca708SEric Dumazet { 37160c04aecSJoe Perches int score; 37260c04aecSJoe Perches struct inet_sock *inet; 3736da5b0f0SMike Manning bool dev_match; 374645ca708SEric Dumazet 37560c04aecSJoe Perches if (!net_eq(sock_net(sk), net) || 37660c04aecSJoe Perches udp_sk(sk)->udp_port_hash != hnum || 37760c04aecSJoe Perches ipv6_only_sock(sk)) 37860c04aecSJoe Perches return -1; 379645ca708SEric Dumazet 3804cdeeee9SPeter Oskolkov if (sk->sk_rcv_saddr != daddr) 381645ca708SEric Dumazet return -1; 38260c04aecSJoe Perches 3834cdeeee9SPeter Oskolkov score = (sk->sk_family == PF_INET) ? 2 : 1; 3844cdeeee9SPeter Oskolkov 3854cdeeee9SPeter Oskolkov inet = inet_sk(sk); 386c720c7e8SEric Dumazet if (inet->inet_daddr) { 387c720c7e8SEric Dumazet if (inet->inet_daddr != saddr) 388645ca708SEric Dumazet return -1; 389ba418fa3STom Herbert score += 4; 390645ca708SEric Dumazet } 39160c04aecSJoe Perches 392c720c7e8SEric Dumazet if (inet->inet_dport) { 393c720c7e8SEric Dumazet if (inet->inet_dport != sport) 394645ca708SEric Dumazet return -1; 395ba418fa3STom Herbert score += 4; 396645ca708SEric Dumazet } 39760c04aecSJoe Perches 3986da5b0f0SMike Manning dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 3996da5b0f0SMike Manning dif, sdif); 40069678bcdSPaolo Abeni if (!dev_match) 401645ca708SEric Dumazet return -1; 4028d6c414cSMike Manning if (sk->sk_bound_dev_if) 403ba418fa3STom Herbert score += 4; 404fb74c277SDavid Ahern 4057170a977SEric Dumazet if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 40670da268bSEric Dumazet score++; 407645ca708SEric Dumazet return score; 408645ca708SEric Dumazet } 409645ca708SEric Dumazet 4100f495f76SLorenz Bauer INDIRECT_CALLABLE_SCOPE 4110f495f76SLorenz Bauer u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport, 4120f495f76SLorenz Bauer const __be32 faddr, const __be16 fport) 41365cd8033SHannes Frederic Sowa { 4141bbdceefSHannes Frederic Sowa static u32 udp_ehash_secret __read_mostly; 4151bbdceefSHannes Frederic Sowa 4161bbdceefSHannes Frederic Sowa net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 4171bbdceefSHannes Frederic Sowa 41865cd8033SHannes Frederic Sowa return __inet_ehashfn(laddr, lport, faddr, fport, 4191bbdceefSHannes Frederic Sowa udp_ehash_secret + net_hash_mix(net)); 42065cd8033SHannes Frederic Sowa } 42165cd8033SHannes Frederic Sowa 422d1e37288SSu, Xuemin /* called with rcu_read_lock() */ 4235051ebd2SEric Dumazet static struct sock *udp4_lib_lookup2(struct net *net, 4245051ebd2SEric Dumazet __be32 saddr, __be16 sport, 425fb74c277SDavid Ahern __be32 daddr, unsigned int hnum, 42673545373STim Beale int dif, int sdif, 427d1e37288SSu, Xuemin struct udp_hslot *hslot2, 4281134158bSCraig Gallek struct sk_buff *skb) 4295051ebd2SEric Dumazet { 4305051ebd2SEric Dumazet struct sock *sk, *result; 431e94a62f5SPaolo Abeni int score, badness; 4325051ebd2SEric Dumazet 4335051ebd2SEric Dumazet result = NULL; 434ba418fa3STom Herbert badness = 0; 435ca065d0cSEric Dumazet udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 436d1e37288SSu, Xuemin score = compute_score(sk, net, saddr, sport, 43773545373STim Beale daddr, hnum, dif, sdif); 4385051ebd2SEric Dumazet if (score > badness) { 439ca065d0cSEric Dumazet badness = score; 4400f495f76SLorenz Bauer 4410f495f76SLorenz Bauer if (sk->sk_state == TCP_ESTABLISHED) { 4420f495f76SLorenz Bauer result = sk; 4430f495f76SLorenz Bauer continue; 4440f495f76SLorenz Bauer } 4450f495f76SLorenz Bauer 4460f495f76SLorenz Bauer result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr), 4470f495f76SLorenz Bauer saddr, sport, daddr, hnum, udp_ehashfn); 448f0ea27e7SLorenz Bauer if (!result) { 449f0ea27e7SLorenz Bauer result = sk; 450f0ea27e7SLorenz Bauer continue; 451f0ea27e7SLorenz Bauer } 452f0ea27e7SLorenz Bauer 453e32ea7e7SCraig Gallek /* Fall back to scoring if group has connections */ 454f0ea27e7SLorenz Bauer if (!reuseport_has_conns(sk)) 455ba418fa3STom Herbert return result; 456ba418fa3STom Herbert 457f0ea27e7SLorenz Bauer /* Reuseport logic returned an error, keep original score. */ 458f0ea27e7SLorenz Bauer if (IS_ERR(result)) 459f0ea27e7SLorenz Bauer continue; 460f0ea27e7SLorenz Bauer 461f0ea27e7SLorenz Bauer badness = compute_score(result, net, saddr, sport, 462f0ea27e7SLorenz Bauer daddr, hnum, dif, sdif); 463f0ea27e7SLorenz Bauer 4645051ebd2SEric Dumazet } 4655051ebd2SEric Dumazet } 4665051ebd2SEric Dumazet return result; 4675051ebd2SEric Dumazet } 4685051ebd2SEric Dumazet 469db8dac20SDavid S. Miller /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 470db8dac20SDavid S. Miller * harder than this. -DaveM 471db8dac20SDavid S. Miller */ 472fce82338SPavel Emelyanov struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 473fb74c277SDavid Ahern __be16 sport, __be32 daddr, __be16 dport, int dif, 474fb74c277SDavid Ahern int sdif, struct udp_table *udptable, struct sk_buff *skb) 475db8dac20SDavid S. Miller { 476db8dac20SDavid S. Miller unsigned short hnum = ntohs(dport); 4774cdeeee9SPeter Oskolkov unsigned int hash2, slot2; 4784cdeeee9SPeter Oskolkov struct udp_hslot *hslot2; 47972f7e944SJakub Sitnicki struct sock *result, *sk; 480db8dac20SDavid S. Miller 481f0b1e64cSMartin KaFai Lau hash2 = ipv4_portaddr_hash(net, daddr, hnum); 4825051ebd2SEric Dumazet slot2 = hash2 & udptable->mask; 4835051ebd2SEric Dumazet hslot2 = &udptable->hash2[slot2]; 4845051ebd2SEric Dumazet 48572f7e944SJakub Sitnicki /* Lookup connected or non-wildcard socket */ 4865051ebd2SEric Dumazet result = udp4_lib_lookup2(net, saddr, sport, 487fb74c277SDavid Ahern daddr, hnum, dif, sdif, 48873545373STim Beale hslot2, skb); 48972f7e944SJakub Sitnicki if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 49072f7e944SJakub Sitnicki goto done; 49172f7e944SJakub Sitnicki 49272f7e944SJakub Sitnicki /* Lookup redirect from BPF */ 4936c886db2SLorenz Bauer if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 4946c886db2SLorenz Bauer udptable == net->ipv4.udp_table) { 4956c886db2SLorenz Bauer sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr), 4966c886db2SLorenz Bauer saddr, sport, daddr, hnum, dif, 4976c886db2SLorenz Bauer udp_ehashfn); 49872f7e944SJakub Sitnicki if (sk) { 49972f7e944SJakub Sitnicki result = sk; 50072f7e944SJakub Sitnicki goto done; 50172f7e944SJakub Sitnicki } 50272f7e944SJakub Sitnicki } 50372f7e944SJakub Sitnicki 50472f7e944SJakub Sitnicki /* Got non-wildcard socket or error on first lookup */ 50572f7e944SJakub Sitnicki if (result) 50672f7e944SJakub Sitnicki goto done; 50772f7e944SJakub Sitnicki 50872f7e944SJakub Sitnicki /* Lookup wildcard sockets */ 509f0b1e64cSMartin KaFai Lau hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 5105051ebd2SEric Dumazet slot2 = hash2 & udptable->mask; 5115051ebd2SEric Dumazet hslot2 = &udptable->hash2[slot2]; 5125051ebd2SEric Dumazet 5131223c67cSJorge Boncompte [DTI2] result = udp4_lib_lookup2(net, saddr, sport, 5144cdeeee9SPeter Oskolkov htonl(INADDR_ANY), hnum, dif, sdif, 51573545373STim Beale hslot2, skb); 51672f7e944SJakub Sitnicki done: 51788e235b8SEnrico Weigelt if (IS_ERR(result)) 5188217ca65SMartin KaFai Lau return NULL; 5195051ebd2SEric Dumazet return result; 5205051ebd2SEric Dumazet } 521fce82338SPavel Emelyanov EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 522db8dac20SDavid S. Miller 523607c4aafSKOVACS Krisztian static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 524607c4aafSKOVACS Krisztian __be16 sport, __be16 dport, 525645ca708SEric Dumazet struct udp_table *udptable) 526607c4aafSKOVACS Krisztian { 527607c4aafSKOVACS Krisztian const struct iphdr *iph = ip_hdr(skb); 528607c4aafSKOVACS Krisztian 529ed7cbbceSAlexander Duyck return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 530607c4aafSKOVACS Krisztian iph->daddr, dport, inet_iif(skb), 531fb74c277SDavid Ahern inet_sdif(skb), udptable, skb); 532607c4aafSKOVACS Krisztian } 533607c4aafSKOVACS Krisztian 5347b58e63eSEric Dumazet struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 53563058308STom Herbert __be16 sport, __be16 dport) 53663058308STom Herbert { 537257a525fSMartin KaFai Lau const struct iphdr *iph = ip_hdr(skb); 538ba6aac15SKuniyuki Iwashima struct net *net = dev_net(skb->dev); 5397938cd15SRichard Gobert int iif, sdif; 5407938cd15SRichard Gobert 5417938cd15SRichard Gobert inet_get_iif_sdif(skb, &iif, &sdif); 542257a525fSMartin KaFai Lau 543ba6aac15SKuniyuki Iwashima return __udp4_lib_lookup(net, iph->saddr, sport, 5447938cd15SRichard Gobert iph->daddr, dport, iif, 5457938cd15SRichard Gobert sdif, net->ipv4.udp_table, NULL); 54663058308STom Herbert } 54763058308STom Herbert 548ca065d0cSEric Dumazet /* Must be called under rcu_read_lock(). 549ca065d0cSEric Dumazet * Does increment socket refcount. 550ca065d0cSEric Dumazet */ 5516e86000cSArnd Bergmann #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 552bcd41303SKOVACS Krisztian struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 553bcd41303SKOVACS Krisztian __be32 daddr, __be16 dport, int dif) 554bcd41303SKOVACS Krisztian { 555ca065d0cSEric Dumazet struct sock *sk; 556ca065d0cSEric Dumazet 557ca065d0cSEric Dumazet sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 558ba6aac15SKuniyuki Iwashima dif, 0, net->ipv4.udp_table, NULL); 55941c6d650SReshetova, Elena if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 560ca065d0cSEric Dumazet sk = NULL; 561ca065d0cSEric Dumazet return sk; 562bcd41303SKOVACS Krisztian } 563bcd41303SKOVACS Krisztian EXPORT_SYMBOL_GPL(udp4_lib_lookup); 564ca065d0cSEric Dumazet #endif 565bcd41303SKOVACS Krisztian 566a0a989d3SEric Dumazet static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk, 567421b3885SShawn Bohrer __be16 loc_port, __be32 loc_addr, 568421b3885SShawn Bohrer __be16 rmt_port, __be32 rmt_addr, 569fb74c277SDavid Ahern int dif, int sdif, unsigned short hnum) 570421b3885SShawn Bohrer { 571a0a989d3SEric Dumazet const struct inet_sock *inet = inet_sk(sk); 572421b3885SShawn Bohrer 573421b3885SShawn Bohrer if (!net_eq(sock_net(sk), net) || 574421b3885SShawn Bohrer udp_sk(sk)->udp_port_hash != hnum || 575421b3885SShawn Bohrer (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 576421b3885SShawn Bohrer (inet->inet_dport != rmt_port && inet->inet_dport) || 577421b3885SShawn Bohrer (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 578421b3885SShawn Bohrer ipv6_only_sock(sk) || 57982ba25c6STim Beale !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 580421b3885SShawn Bohrer return false; 58160d9b031SDavid Ahern if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 582421b3885SShawn Bohrer return false; 583421b3885SShawn Bohrer return true; 584421b3885SShawn Bohrer } 585421b3885SShawn Bohrer 586a36e185eSStefano Brivio DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 587a36e185eSStefano Brivio void udp_encap_enable(void) 588a36e185eSStefano Brivio { 5899c480601SPaolo Abeni static_branch_inc(&udp_encap_needed_key); 590a36e185eSStefano Brivio } 591a36e185eSStefano Brivio EXPORT_SYMBOL(udp_encap_enable); 592a36e185eSStefano Brivio 593a4a600ddSXin Long void udp_encap_disable(void) 594a4a600ddSXin Long { 595a4a600ddSXin Long static_branch_dec(&udp_encap_needed_key); 596a4a600ddSXin Long } 597a4a600ddSXin Long EXPORT_SYMBOL(udp_encap_disable); 598a4a600ddSXin Long 599e7cc0824SStefano Brivio /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 600e7cc0824SStefano Brivio * through error handlers in encapsulations looking for a match. 601e7cc0824SStefano Brivio */ 602e7cc0824SStefano Brivio static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 603e7cc0824SStefano Brivio { 604e7cc0824SStefano Brivio int i; 605e7cc0824SStefano Brivio 606e7cc0824SStefano Brivio for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 607e7cc0824SStefano Brivio int (*handler)(struct sk_buff *skb, u32 info); 60892b95364SPaolo Abeni const struct ip_tunnel_encap_ops *encap; 609e7cc0824SStefano Brivio 61092b95364SPaolo Abeni encap = rcu_dereference(iptun_encaps[i]); 61192b95364SPaolo Abeni if (!encap) 612e7cc0824SStefano Brivio continue; 61392b95364SPaolo Abeni handler = encap->err_handler; 614e7cc0824SStefano Brivio if (handler && !handler(skb, info)) 615e7cc0824SStefano Brivio return 0; 616e7cc0824SStefano Brivio } 617e7cc0824SStefano Brivio 618e7cc0824SStefano Brivio return -ENOENT; 619e7cc0824SStefano Brivio } 620e7cc0824SStefano Brivio 621a36e185eSStefano Brivio /* Try to match ICMP errors to UDP tunnels by looking up a socket without 622a36e185eSStefano Brivio * reversing source and destination port: this will match tunnels that force the 623a36e185eSStefano Brivio * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 624a36e185eSStefano Brivio * lwtunnels might actually break this assumption by being configured with 625a36e185eSStefano Brivio * different destination ports on endpoints, in this case we won't be able to 626a36e185eSStefano Brivio * trace ICMP messages back to them. 627a36e185eSStefano Brivio * 628e7cc0824SStefano Brivio * If this doesn't match any socket, probe tunnels with arbitrary destination 629e7cc0824SStefano Brivio * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 630e7cc0824SStefano Brivio * we've sent packets to won't necessarily match the local destination port. 631e7cc0824SStefano Brivio * 632a36e185eSStefano Brivio * Then ask the tunnel implementation to match the error against a valid 633a36e185eSStefano Brivio * association. 634a36e185eSStefano Brivio * 635e7cc0824SStefano Brivio * Return an error if we can't find a match, the socket if we need further 636e7cc0824SStefano Brivio * processing, zero otherwise. 637a36e185eSStefano Brivio */ 638a36e185eSStefano Brivio static struct sock *__udp4_lib_err_encap(struct net *net, 639a36e185eSStefano Brivio const struct iphdr *iph, 640a36e185eSStefano Brivio struct udphdr *uh, 641a36e185eSStefano Brivio struct udp_table *udptable, 6429bfce73cSVadim Fedorenko struct sock *sk, 643e7cc0824SStefano Brivio struct sk_buff *skb, u32 info) 644a36e185eSStefano Brivio { 6459bfce73cSVadim Fedorenko int (*lookup)(struct sock *sk, struct sk_buff *skb); 646a36e185eSStefano Brivio int network_offset, transport_offset; 6479bfce73cSVadim Fedorenko struct udp_sock *up; 648a36e185eSStefano Brivio 649a36e185eSStefano Brivio network_offset = skb_network_offset(skb); 650a36e185eSStefano Brivio transport_offset = skb_transport_offset(skb); 651a36e185eSStefano Brivio 652a36e185eSStefano Brivio /* Network header needs to point to the outer IPv4 header inside ICMP */ 653a36e185eSStefano Brivio skb_reset_network_header(skb); 654a36e185eSStefano Brivio 655a36e185eSStefano Brivio /* Transport header needs to point to the UDP header */ 656a36e185eSStefano Brivio skb_set_transport_header(skb, iph->ihl << 2); 657a36e185eSStefano Brivio 6589bfce73cSVadim Fedorenko if (sk) { 6599bfce73cSVadim Fedorenko up = udp_sk(sk); 6609bfce73cSVadim Fedorenko 6619bfce73cSVadim Fedorenko lookup = READ_ONCE(up->encap_err_lookup); 6629bfce73cSVadim Fedorenko if (lookup && lookup(sk, skb)) 6639bfce73cSVadim Fedorenko sk = NULL; 6649bfce73cSVadim Fedorenko 6659bfce73cSVadim Fedorenko goto out; 6669bfce73cSVadim Fedorenko } 6679bfce73cSVadim Fedorenko 668e7cc0824SStefano Brivio sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 669e7cc0824SStefano Brivio iph->saddr, uh->dest, skb->dev->ifindex, 0, 670e7cc0824SStefano Brivio udptable, NULL); 671e7cc0824SStefano Brivio if (sk) { 6729bfce73cSVadim Fedorenko up = udp_sk(sk); 673e7cc0824SStefano Brivio 674a36e185eSStefano Brivio lookup = READ_ONCE(up->encap_err_lookup); 675a36e185eSStefano Brivio if (!lookup || lookup(sk, skb)) 676a36e185eSStefano Brivio sk = NULL; 677e7cc0824SStefano Brivio } 678e7cc0824SStefano Brivio 6799bfce73cSVadim Fedorenko out: 680e7cc0824SStefano Brivio if (!sk) 681e7cc0824SStefano Brivio sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 682a36e185eSStefano Brivio 683a36e185eSStefano Brivio skb_set_transport_header(skb, transport_offset); 684a36e185eSStefano Brivio skb_set_network_header(skb, network_offset); 685a36e185eSStefano Brivio 686a36e185eSStefano Brivio return sk; 687a36e185eSStefano Brivio } 688a36e185eSStefano Brivio 689db8dac20SDavid S. Miller /* 690db8dac20SDavid S. Miller * This routine is called by the ICMP module when it gets some 691db8dac20SDavid S. Miller * sort of error condition. If err < 0 then the socket should 692db8dac20SDavid S. Miller * be closed and the error returned to the user. If err > 0 693db8dac20SDavid S. Miller * it's just the icmp type << 8 | icmp code. 694db8dac20SDavid S. Miller * Header points to the ip header of the error packet. We move 695db8dac20SDavid S. Miller * on past this. Then (as it used to claim before adjustment) 696db8dac20SDavid S. Miller * header points to the first 8 bytes of the udp header. We need 697db8dac20SDavid S. Miller * to find the appropriate port. 698db8dac20SDavid S. Miller */ 699db8dac20SDavid S. Miller 70032bbd879SStefano Brivio int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 701db8dac20SDavid S. Miller { 702db8dac20SDavid S. Miller struct inet_sock *inet; 703b71d1d42SEric Dumazet const struct iphdr *iph = (const struct iphdr *)skb->data; 704db8dac20SDavid S. Miller struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 705db8dac20SDavid S. Miller const int type = icmp_hdr(skb)->type; 706db8dac20SDavid S. Miller const int code = icmp_hdr(skb)->code; 707a36e185eSStefano Brivio bool tunnel = false; 708db8dac20SDavid S. Miller struct sock *sk; 709db8dac20SDavid S. Miller int harderr; 710db8dac20SDavid S. Miller int err; 711fd54d716SPavel Emelyanov struct net *net = dev_net(skb->dev); 712db8dac20SDavid S. Miller 713fd54d716SPavel Emelyanov sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 714f64bf6b8SMike Manning iph->saddr, uh->source, skb->dev->ifindex, 715f64bf6b8SMike Manning inet_sdif(skb), udptable, NULL); 7169bfce73cSVadim Fedorenko 717d26796aeSXin Long if (!sk || udp_sk(sk)->encap_type) { 718a36e185eSStefano Brivio /* No socket for error: try tunnels before discarding */ 719e7cc0824SStefano Brivio if (static_branch_unlikely(&udp_encap_needed_key)) { 7209bfce73cSVadim Fedorenko sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 721e7cc0824SStefano Brivio info); 722e7cc0824SStefano Brivio if (!sk) 723e7cc0824SStefano Brivio return 0; 7249bfce73cSVadim Fedorenko } else 7259bfce73cSVadim Fedorenko sk = ERR_PTR(-ENOENT); 726e7cc0824SStefano Brivio 727e7cc0824SStefano Brivio if (IS_ERR(sk)) { 728e7cc0824SStefano Brivio __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 729e7cc0824SStefano Brivio return PTR_ERR(sk); 730e7cc0824SStefano Brivio } 731e7cc0824SStefano Brivio 732a36e185eSStefano Brivio tunnel = true; 733db8dac20SDavid S. Miller } 734db8dac20SDavid S. Miller 735db8dac20SDavid S. Miller err = 0; 736db8dac20SDavid S. Miller harderr = 0; 737db8dac20SDavid S. Miller inet = inet_sk(sk); 738db8dac20SDavid S. Miller 739db8dac20SDavid S. Miller switch (type) { 740db8dac20SDavid S. Miller default: 741db8dac20SDavid S. Miller case ICMP_TIME_EXCEEDED: 742db8dac20SDavid S. Miller err = EHOSTUNREACH; 743db8dac20SDavid S. Miller break; 744db8dac20SDavid S. Miller case ICMP_SOURCE_QUENCH: 745db8dac20SDavid S. Miller goto out; 746db8dac20SDavid S. Miller case ICMP_PARAMETERPROB: 747db8dac20SDavid S. Miller err = EPROTO; 748db8dac20SDavid S. Miller harderr = 1; 749db8dac20SDavid S. Miller break; 750db8dac20SDavid S. Miller case ICMP_DEST_UNREACH: 751db8dac20SDavid S. Miller if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 75236393395SDavid S. Miller ipv4_sk_update_pmtu(skb, sk, info); 753db8dac20SDavid S. Miller if (inet->pmtudisc != IP_PMTUDISC_DONT) { 754db8dac20SDavid S. Miller err = EMSGSIZE; 755db8dac20SDavid S. Miller harderr = 1; 756db8dac20SDavid S. Miller break; 757db8dac20SDavid S. Miller } 758db8dac20SDavid S. Miller goto out; 759db8dac20SDavid S. Miller } 760db8dac20SDavid S. Miller err = EHOSTUNREACH; 761db8dac20SDavid S. Miller if (code <= NR_ICMP_UNREACH) { 762db8dac20SDavid S. Miller harderr = icmp_err_convert[code].fatal; 763db8dac20SDavid S. Miller err = icmp_err_convert[code].errno; 764db8dac20SDavid S. Miller } 765db8dac20SDavid S. Miller break; 76655be7a9cSDavid S. Miller case ICMP_REDIRECT: 76755be7a9cSDavid S. Miller ipv4_sk_redirect(skb, sk); 7681a462d18SDuan Jiong goto out; 769db8dac20SDavid S. Miller } 770db8dac20SDavid S. Miller 771db8dac20SDavid S. Miller /* 772db8dac20SDavid S. Miller * RFC1122: OK. Passes ICMP errors back to application, as per 773db8dac20SDavid S. Miller * 4.1.3.3. 774db8dac20SDavid S. Miller */ 775a36e185eSStefano Brivio if (tunnel) { 776a36e185eSStefano Brivio /* ...not for tunnels though: we don't have a sending socket */ 777ac56a0b4SDavid Howells if (udp_sk(sk)->encap_err_rcv) 77842fb06b3SDavid Howells udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info, 77942fb06b3SDavid Howells (u8 *)(uh+1)); 780a36e185eSStefano Brivio goto out; 781a36e185eSStefano Brivio } 782*6b5f43eaSEric Dumazet if (!inet_test_bit(RECVERR, sk)) { 783db8dac20SDavid S. Miller if (!harderr || sk->sk_state != TCP_ESTABLISHED) 784db8dac20SDavid S. Miller goto out; 785b1faf566SEric Dumazet } else 786db8dac20SDavid S. Miller ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 787b1faf566SEric Dumazet 788db8dac20SDavid S. Miller sk->sk_err = err; 789e3ae2365SAlexander Aring sk_error_report(sk); 790db8dac20SDavid S. Miller out: 79132bbd879SStefano Brivio return 0; 792db8dac20SDavid S. Miller } 793db8dac20SDavid S. Miller 79432bbd879SStefano Brivio int udp_err(struct sk_buff *skb, u32 info) 795db8dac20SDavid S. Miller { 796ba6aac15SKuniyuki Iwashima return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table); 797db8dac20SDavid S. Miller } 798db8dac20SDavid S. Miller 799db8dac20SDavid S. Miller /* 800db8dac20SDavid S. Miller * Throw away all pending data and cancel the corking. Socket is locked. 801db8dac20SDavid S. Miller */ 80236d926b9SDenis V. Lunev void udp_flush_pending_frames(struct sock *sk) 803db8dac20SDavid S. Miller { 804db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 805db8dac20SDavid S. Miller 806db8dac20SDavid S. Miller if (up->pending) { 807db8dac20SDavid S. Miller up->len = 0; 808db8dac20SDavid S. Miller up->pending = 0; 809db8dac20SDavid S. Miller ip_flush_pending_frames(sk); 810db8dac20SDavid S. Miller } 811db8dac20SDavid S. Miller } 81236d926b9SDenis V. Lunev EXPORT_SYMBOL(udp_flush_pending_frames); 813db8dac20SDavid S. Miller 814db8dac20SDavid S. Miller /** 815f6b9664fSHerbert Xu * udp4_hwcsum - handle outgoing HW checksumming 816db8dac20SDavid S. Miller * @skb: sk_buff containing the filled-in UDP header 817db8dac20SDavid S. Miller * (checksum field must be zeroed out) 818f6b9664fSHerbert Xu * @src: source IP address 819f6b9664fSHerbert Xu * @dst: destination IP address 820db8dac20SDavid S. Miller */ 821c26bf4a5SThomas Graf void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 822db8dac20SDavid S. Miller { 823db8dac20SDavid S. Miller struct udphdr *uh = udp_hdr(skb); 824f6b9664fSHerbert Xu int offset = skb_transport_offset(skb); 825f6b9664fSHerbert Xu int len = skb->len - offset; 826f6b9664fSHerbert Xu int hlen = len; 827db8dac20SDavid S. Miller __wsum csum = 0; 828db8dac20SDavid S. Miller 829ebbe495fSWANG Cong if (!skb_has_frag_list(skb)) { 830db8dac20SDavid S. Miller /* 831db8dac20SDavid S. Miller * Only one fragment on the socket. 832db8dac20SDavid S. Miller */ 833db8dac20SDavid S. Miller skb->csum_start = skb_transport_header(skb) - skb->head; 834db8dac20SDavid S. Miller skb->csum_offset = offsetof(struct udphdr, check); 835f6b9664fSHerbert Xu uh->check = ~csum_tcpudp_magic(src, dst, len, 836f6b9664fSHerbert Xu IPPROTO_UDP, 0); 837db8dac20SDavid S. Miller } else { 838ebbe495fSWANG Cong struct sk_buff *frags; 839ebbe495fSWANG Cong 840db8dac20SDavid S. Miller /* 841db8dac20SDavid S. Miller * HW-checksum won't work as there are two or more 842db8dac20SDavid S. Miller * fragments on the socket so that all csums of sk_buffs 843db8dac20SDavid S. Miller * should be together 844db8dac20SDavid S. Miller */ 845ebbe495fSWANG Cong skb_walk_frags(skb, frags) { 846f6b9664fSHerbert Xu csum = csum_add(csum, frags->csum); 847f6b9664fSHerbert Xu hlen -= frags->len; 848ebbe495fSWANG Cong } 849db8dac20SDavid S. Miller 850f6b9664fSHerbert Xu csum = skb_checksum(skb, offset, hlen, csum); 851db8dac20SDavid S. Miller skb->ip_summed = CHECKSUM_NONE; 852db8dac20SDavid S. Miller 853db8dac20SDavid S. Miller uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 854db8dac20SDavid S. Miller if (uh->check == 0) 855db8dac20SDavid S. Miller uh->check = CSUM_MANGLED_0; 856db8dac20SDavid S. Miller } 857db8dac20SDavid S. Miller } 858c26bf4a5SThomas Graf EXPORT_SYMBOL_GPL(udp4_hwcsum); 859db8dac20SDavid S. Miller 860af5fcba7STom Herbert /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 861af5fcba7STom Herbert * for the simple case like when setting the checksum for a UDP tunnel. 862af5fcba7STom Herbert */ 863af5fcba7STom Herbert void udp_set_csum(bool nocheck, struct sk_buff *skb, 864af5fcba7STom Herbert __be32 saddr, __be32 daddr, int len) 865af5fcba7STom Herbert { 866af5fcba7STom Herbert struct udphdr *uh = udp_hdr(skb); 867af5fcba7STom Herbert 868179bc67fSEdward Cree if (nocheck) { 869af5fcba7STom Herbert uh->check = 0; 870179bc67fSEdward Cree } else if (skb_is_gso(skb)) { 871af5fcba7STom Herbert uh->check = ~udp_v4_check(len, saddr, daddr, 0); 872179bc67fSEdward Cree } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 873179bc67fSEdward Cree uh->check = 0; 874179bc67fSEdward Cree uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 875179bc67fSEdward Cree if (uh->check == 0) 876179bc67fSEdward Cree uh->check = CSUM_MANGLED_0; 877d75f1306SEdward Cree } else { 878af5fcba7STom Herbert skb->ip_summed = CHECKSUM_PARTIAL; 879af5fcba7STom Herbert skb->csum_start = skb_transport_header(skb) - skb->head; 880af5fcba7STom Herbert skb->csum_offset = offsetof(struct udphdr, check); 881af5fcba7STom Herbert uh->check = ~udp_v4_check(len, saddr, daddr, 0); 882af5fcba7STom Herbert } 883af5fcba7STom Herbert } 884af5fcba7STom Herbert EXPORT_SYMBOL(udp_set_csum); 885af5fcba7STom Herbert 886bec1f6f6SWillem de Bruijn static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 887bec1f6f6SWillem de Bruijn struct inet_cork *cork) 888f6b9664fSHerbert Xu { 889f6b9664fSHerbert Xu struct sock *sk = skb->sk; 890f6b9664fSHerbert Xu struct inet_sock *inet = inet_sk(sk); 891f6b9664fSHerbert Xu struct udphdr *uh; 892cffb8f61SMenglong Dong int err; 893f6b9664fSHerbert Xu int is_udplite = IS_UDPLITE(sk); 894f6b9664fSHerbert Xu int offset = skb_transport_offset(skb); 895f6b9664fSHerbert Xu int len = skb->len - offset; 8964094871dSJosh Hunt int datalen = len - sizeof(*uh); 897f6b9664fSHerbert Xu __wsum csum = 0; 898f6b9664fSHerbert Xu 899f6b9664fSHerbert Xu /* 900f6b9664fSHerbert Xu * Create a UDP header 901f6b9664fSHerbert Xu */ 902f6b9664fSHerbert Xu uh = udp_hdr(skb); 903f6b9664fSHerbert Xu uh->source = inet->inet_sport; 90479ab0531SDavid S. Miller uh->dest = fl4->fl4_dport; 905f6b9664fSHerbert Xu uh->len = htons(len); 906f6b9664fSHerbert Xu uh->check = 0; 907f6b9664fSHerbert Xu 908bec1f6f6SWillem de Bruijn if (cork->gso_size) { 909bec1f6f6SWillem de Bruijn const int hlen = skb_network_header_len(skb) + 910bec1f6f6SWillem de Bruijn sizeof(struct udphdr); 911bec1f6f6SWillem de Bruijn 9120f149c9fSWillem de Bruijn if (hlen + cork->gso_size > cork->fragsize) { 9130f149c9fSWillem de Bruijn kfree_skb(skb); 914bec1f6f6SWillem de Bruijn return -EINVAL; 9150f149c9fSWillem de Bruijn } 916158390e4SJianguo Wu if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 9170f149c9fSWillem de Bruijn kfree_skb(skb); 918bec1f6f6SWillem de Bruijn return -EINVAL; 9190f149c9fSWillem de Bruijn } 9200f149c9fSWillem de Bruijn if (sk->sk_no_check_tx) { 9210f149c9fSWillem de Bruijn kfree_skb(skb); 922a8c744a8SWillem de Bruijn return -EINVAL; 9230f149c9fSWillem de Bruijn } 924ff06342cSWillem de Bruijn if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 9250f149c9fSWillem de Bruijn dst_xfrm(skb_dst(skb))) { 9260f149c9fSWillem de Bruijn kfree_skb(skb); 927bec1f6f6SWillem de Bruijn return -EIO; 9280f149c9fSWillem de Bruijn } 929bec1f6f6SWillem de Bruijn 9304094871dSJosh Hunt if (datalen > cork->gso_size) { 931bec1f6f6SWillem de Bruijn skb_shinfo(skb)->gso_size = cork->gso_size; 932bec1f6f6SWillem de Bruijn skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 9334094871dSJosh Hunt skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 934dfec0ee2SAlexander Duyck cork->gso_size); 9354094871dSJosh Hunt } 936a8c744a8SWillem de Bruijn goto csum_partial; 937bec1f6f6SWillem de Bruijn } 938bec1f6f6SWillem de Bruijn 939f6b9664fSHerbert Xu if (is_udplite) /* UDP-Lite */ 940f6b9664fSHerbert Xu csum = udplite_csum(skb); 941f6b9664fSHerbert Xu 942ab2fb7e3SWillem de Bruijn else if (sk->sk_no_check_tx) { /* UDP csum off */ 943f6b9664fSHerbert Xu 944f6b9664fSHerbert Xu skb->ip_summed = CHECKSUM_NONE; 945f6b9664fSHerbert Xu goto send; 946f6b9664fSHerbert Xu 947f6b9664fSHerbert Xu } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 948a8c744a8SWillem de Bruijn csum_partial: 949f6b9664fSHerbert Xu 95079ab0531SDavid S. Miller udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 951f6b9664fSHerbert Xu goto send; 952f6b9664fSHerbert Xu 953f6b9664fSHerbert Xu } else 954f6b9664fSHerbert Xu csum = udp_csum(skb); 955f6b9664fSHerbert Xu 956f6b9664fSHerbert Xu /* add protocol-dependent pseudo-header */ 95779ab0531SDavid S. Miller uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 958f6b9664fSHerbert Xu sk->sk_protocol, csum); 959f6b9664fSHerbert Xu if (uh->check == 0) 960f6b9664fSHerbert Xu uh->check = CSUM_MANGLED_0; 961f6b9664fSHerbert Xu 962f6b9664fSHerbert Xu send: 963b5ec8eeaSEric Dumazet err = ip_send_skb(sock_net(sk), skb); 964f6b9664fSHerbert Xu if (err) { 965*6b5f43eaSEric Dumazet if (err == -ENOBUFS && 966*6b5f43eaSEric Dumazet !inet_test_bit(RECVERR, sk)) { 9676aef70a8SEric Dumazet UDP_INC_STATS(sock_net(sk), 968f6b9664fSHerbert Xu UDP_MIB_SNDBUFERRORS, is_udplite); 969f6b9664fSHerbert Xu err = 0; 970f6b9664fSHerbert Xu } 971f6b9664fSHerbert Xu } else 9726aef70a8SEric Dumazet UDP_INC_STATS(sock_net(sk), 973f6b9664fSHerbert Xu UDP_MIB_OUTDATAGRAMS, is_udplite); 974f6b9664fSHerbert Xu return err; 975f6b9664fSHerbert Xu } 976f6b9664fSHerbert Xu 977db8dac20SDavid S. Miller /* 978db8dac20SDavid S. Miller * Push out all pending data as one UDP datagram. Socket is locked. 979db8dac20SDavid S. Miller */ 9808822b64aSHannes Frederic Sowa int udp_push_pending_frames(struct sock *sk) 981db8dac20SDavid S. Miller { 982db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 983db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sk); 984b6f21b26SDavid S. Miller struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 985db8dac20SDavid S. Miller struct sk_buff *skb; 986db8dac20SDavid S. Miller int err = 0; 987db8dac20SDavid S. Miller 98877968b78SDavid S. Miller skb = ip_finish_skb(sk, fl4); 989f6b9664fSHerbert Xu if (!skb) 990db8dac20SDavid S. Miller goto out; 991db8dac20SDavid S. Miller 992bec1f6f6SWillem de Bruijn err = udp_send_skb(skb, fl4, &inet->cork.base); 993db8dac20SDavid S. Miller 994db8dac20SDavid S. Miller out: 995db8dac20SDavid S. Miller up->len = 0; 996db8dac20SDavid S. Miller up->pending = 0; 997db8dac20SDavid S. Miller return err; 998db8dac20SDavid S. Miller } 9998822b64aSHannes Frederic Sowa EXPORT_SYMBOL(udp_push_pending_frames); 1000db8dac20SDavid S. Miller 10012e8de857SWillem de Bruijn static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 10022e8de857SWillem de Bruijn { 10032e8de857SWillem de Bruijn switch (cmsg->cmsg_type) { 10042e8de857SWillem de Bruijn case UDP_SEGMENT: 10052e8de857SWillem de Bruijn if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 10062e8de857SWillem de Bruijn return -EINVAL; 10072e8de857SWillem de Bruijn *gso_size = *(__u16 *)CMSG_DATA(cmsg); 10082e8de857SWillem de Bruijn return 0; 10092e8de857SWillem de Bruijn default: 10102e8de857SWillem de Bruijn return -EINVAL; 10112e8de857SWillem de Bruijn } 10122e8de857SWillem de Bruijn } 10132e8de857SWillem de Bruijn 10142e8de857SWillem de Bruijn int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 10152e8de857SWillem de Bruijn { 10162e8de857SWillem de Bruijn struct cmsghdr *cmsg; 10172e8de857SWillem de Bruijn bool need_ip = false; 10182e8de857SWillem de Bruijn int err; 10192e8de857SWillem de Bruijn 10202e8de857SWillem de Bruijn for_each_cmsghdr(cmsg, msg) { 10212e8de857SWillem de Bruijn if (!CMSG_OK(msg, cmsg)) 10222e8de857SWillem de Bruijn return -EINVAL; 10232e8de857SWillem de Bruijn 10242e8de857SWillem de Bruijn if (cmsg->cmsg_level != SOL_UDP) { 10252e8de857SWillem de Bruijn need_ip = true; 10262e8de857SWillem de Bruijn continue; 10272e8de857SWillem de Bruijn } 10282e8de857SWillem de Bruijn 10292e8de857SWillem de Bruijn err = __udp_cmsg_send(cmsg, gso_size); 10302e8de857SWillem de Bruijn if (err) 10312e8de857SWillem de Bruijn return err; 10322e8de857SWillem de Bruijn } 10332e8de857SWillem de Bruijn 10342e8de857SWillem de Bruijn return need_ip; 10352e8de857SWillem de Bruijn } 10362e8de857SWillem de Bruijn EXPORT_SYMBOL_GPL(udp_cmsg_send); 10372e8de857SWillem de Bruijn 10381b784140SYing Xue int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1039db8dac20SDavid S. Miller { 1040db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sk); 1041db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 10421cedee13SAndrey Ignatov DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1043e474995fSDavid S. Miller struct flowi4 fl4_stack; 1044b6f21b26SDavid S. Miller struct flowi4 *fl4; 1045db8dac20SDavid S. Miller int ulen = len; 1046db8dac20SDavid S. Miller struct ipcm_cookie ipc; 1047db8dac20SDavid S. Miller struct rtable *rt = NULL; 1048db8dac20SDavid S. Miller int free = 0; 1049db8dac20SDavid S. Miller int connected = 0; 1050db8dac20SDavid S. Miller __be32 daddr, faddr, saddr; 10510e26371dSGuillaume Nault u8 tos, scope; 1052db8dac20SDavid S. Miller __be16 dport; 1053db8dac20SDavid S. Miller int err, is_udplite = IS_UDPLITE(sk); 1054a9f59707SEric Dumazet int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1055db8dac20SDavid S. Miller int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1056903ab86dSHerbert Xu struct sk_buff *skb; 1057f6d8bd05SEric Dumazet struct ip_options_data opt_copy; 1058db8dac20SDavid S. Miller 1059db8dac20SDavid S. Miller if (len > 0xFFFF) 1060db8dac20SDavid S. Miller return -EMSGSIZE; 1061db8dac20SDavid S. Miller 1062db8dac20SDavid S. Miller /* 1063db8dac20SDavid S. Miller * Check the flags. 1064db8dac20SDavid S. Miller */ 1065db8dac20SDavid S. Miller 1066db8dac20SDavid S. Miller if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1067db8dac20SDavid S. Miller return -EOPNOTSUPP; 1068db8dac20SDavid S. Miller 1069903ab86dSHerbert Xu getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1070903ab86dSHerbert Xu 1071f5fca608SDavid S. Miller fl4 = &inet->cork.fl.u.ip4; 1072db8dac20SDavid S. Miller if (up->pending) { 1073db8dac20SDavid S. Miller /* 1074db8dac20SDavid S. Miller * There are pending frames. 1075db8dac20SDavid S. Miller * The socket lock must be held while it's corked. 1076db8dac20SDavid S. Miller */ 1077db8dac20SDavid S. Miller lock_sock(sk); 1078db8dac20SDavid S. Miller if (likely(up->pending)) { 1079db8dac20SDavid S. Miller if (unlikely(up->pending != AF_INET)) { 1080db8dac20SDavid S. Miller release_sock(sk); 1081db8dac20SDavid S. Miller return -EINVAL; 1082db8dac20SDavid S. Miller } 1083db8dac20SDavid S. Miller goto do_append_data; 1084db8dac20SDavid S. Miller } 1085db8dac20SDavid S. Miller release_sock(sk); 1086db8dac20SDavid S. Miller } 1087db8dac20SDavid S. Miller ulen += sizeof(struct udphdr); 1088db8dac20SDavid S. Miller 1089db8dac20SDavid S. Miller /* 1090db8dac20SDavid S. Miller * Get and verify the address. 1091db8dac20SDavid S. Miller */ 10921cedee13SAndrey Ignatov if (usin) { 1093db8dac20SDavid S. Miller if (msg->msg_namelen < sizeof(*usin)) 1094db8dac20SDavid S. Miller return -EINVAL; 1095db8dac20SDavid S. Miller if (usin->sin_family != AF_INET) { 1096db8dac20SDavid S. Miller if (usin->sin_family != AF_UNSPEC) 1097db8dac20SDavid S. Miller return -EAFNOSUPPORT; 1098db8dac20SDavid S. Miller } 1099db8dac20SDavid S. Miller 1100db8dac20SDavid S. Miller daddr = usin->sin_addr.s_addr; 1101db8dac20SDavid S. Miller dport = usin->sin_port; 1102db8dac20SDavid S. Miller if (dport == 0) 1103db8dac20SDavid S. Miller return -EINVAL; 1104db8dac20SDavid S. Miller } else { 1105db8dac20SDavid S. Miller if (sk->sk_state != TCP_ESTABLISHED) 1106db8dac20SDavid S. Miller return -EDESTADDRREQ; 1107c720c7e8SEric Dumazet daddr = inet->inet_daddr; 1108c720c7e8SEric Dumazet dport = inet->inet_dport; 1109db8dac20SDavid S. Miller /* Open fast path for connected socket. 1110db8dac20SDavid S. Miller Route will not be used, if at least one option is set. 1111db8dac20SDavid S. Miller */ 1112db8dac20SDavid S. Miller connected = 1; 1113db8dac20SDavid S. Miller } 1114c14ac945SSoheil Hassas Yeganeh 111535178206SWillem de Bruijn ipcm_init_sk(&ipc, inet); 111618a419baSEric Dumazet ipc.gso_size = READ_ONCE(up->gso_size); 1117bf84a010SDaniel Borkmann 1118db8dac20SDavid S. Miller if (msg->msg_controllen) { 11192e8de857SWillem de Bruijn err = udp_cmsg_send(sk, msg, &ipc.gso_size); 11202e8de857SWillem de Bruijn if (err > 0) 11212e8de857SWillem de Bruijn err = ip_cmsg_send(sk, msg, &ipc, 11222e8de857SWillem de Bruijn sk->sk_family == AF_INET6); 11232e8de857SWillem de Bruijn if (unlikely(err < 0)) { 112491948309SEric Dumazet kfree(ipc.opt); 1125db8dac20SDavid S. Miller return err; 112691948309SEric Dumazet } 1127db8dac20SDavid S. Miller if (ipc.opt) 1128db8dac20SDavid S. Miller free = 1; 1129db8dac20SDavid S. Miller connected = 0; 1130db8dac20SDavid S. Miller } 1131f6d8bd05SEric Dumazet if (!ipc.opt) { 1132f6d8bd05SEric Dumazet struct ip_options_rcu *inet_opt; 1133f6d8bd05SEric Dumazet 1134f6d8bd05SEric Dumazet rcu_read_lock(); 1135f6d8bd05SEric Dumazet inet_opt = rcu_dereference(inet->inet_opt); 1136f6d8bd05SEric Dumazet if (inet_opt) { 1137f6d8bd05SEric Dumazet memcpy(&opt_copy, inet_opt, 1138f6d8bd05SEric Dumazet sizeof(*inet_opt) + inet_opt->opt.optlen); 1139f6d8bd05SEric Dumazet ipc.opt = &opt_copy.opt; 1140f6d8bd05SEric Dumazet } 1141f6d8bd05SEric Dumazet rcu_read_unlock(); 1142f6d8bd05SEric Dumazet } 1143db8dac20SDavid S. Miller 11446fc88c35SDave Marchevsky if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 11451cedee13SAndrey Ignatov err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 11461cedee13SAndrey Ignatov (struct sockaddr *)usin, &ipc.addr); 11471cedee13SAndrey Ignatov if (err) 11481cedee13SAndrey Ignatov goto out_free; 11491cedee13SAndrey Ignatov if (usin) { 11501cedee13SAndrey Ignatov if (usin->sin_port == 0) { 11511cedee13SAndrey Ignatov /* BPF program set invalid port. Reject it. */ 11521cedee13SAndrey Ignatov err = -EINVAL; 11531cedee13SAndrey Ignatov goto out_free; 11541cedee13SAndrey Ignatov } 11551cedee13SAndrey Ignatov daddr = usin->sin_addr.s_addr; 11561cedee13SAndrey Ignatov dport = usin->sin_port; 11571cedee13SAndrey Ignatov } 11581cedee13SAndrey Ignatov } 11591cedee13SAndrey Ignatov 1160db8dac20SDavid S. Miller saddr = ipc.addr; 1161db8dac20SDavid S. Miller ipc.addr = faddr = daddr; 1162db8dac20SDavid S. Miller 1163f6d8bd05SEric Dumazet if (ipc.opt && ipc.opt->opt.srr) { 11641b97013bSAndrey Ignatov if (!daddr) { 11651b97013bSAndrey Ignatov err = -EINVAL; 11661b97013bSAndrey Ignatov goto out_free; 11671b97013bSAndrey Ignatov } 1168f6d8bd05SEric Dumazet faddr = ipc.opt->opt.faddr; 1169db8dac20SDavid S. Miller connected = 0; 1170db8dac20SDavid S. Miller } 1171aa661581SFrancesco Fusco tos = get_rttos(&ipc, inet); 11720e26371dSGuillaume Nault scope = ip_sendmsg_scope(inet, &ipc, msg); 11730e26371dSGuillaume Nault if (scope == RT_SCOPE_LINK) 1174db8dac20SDavid S. Miller connected = 0; 1175db8dac20SDavid S. Miller 1176db8dac20SDavid S. Miller if (ipv4_is_multicast(daddr)) { 1177854da991SRobert Shearman if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1178db8dac20SDavid S. Miller ipc.oif = inet->mc_index; 1179db8dac20SDavid S. Miller if (!saddr) 1180db8dac20SDavid S. Miller saddr = inet->mc_addr; 1181db8dac20SDavid S. Miller connected = 0; 11829515a2e0SDavid Ahern } else if (!ipc.oif) { 118376e21053SErich E. Hoover ipc.oif = inet->uc_index; 11849515a2e0SDavid Ahern } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 11859515a2e0SDavid Ahern /* oif is set, packet is to local broadcast and 11862bdcc73cSRandy Dunlap * uc_index is set. oif is most likely set 11879515a2e0SDavid Ahern * by sk_bound_dev_if. If uc_index != oif check if the 11889515a2e0SDavid Ahern * oif is an L3 master and uc_index is an L3 slave. 11899515a2e0SDavid Ahern * If so, we want to allow the send using the uc_index. 11909515a2e0SDavid Ahern */ 11919515a2e0SDavid Ahern if (ipc.oif != inet->uc_index && 11929515a2e0SDavid Ahern ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 11939515a2e0SDavid Ahern inet->uc_index)) { 11949515a2e0SDavid Ahern ipc.oif = inet->uc_index; 11959515a2e0SDavid Ahern } 11969515a2e0SDavid Ahern } 1197db8dac20SDavid S. Miller 1198db8dac20SDavid S. Miller if (connected) 1199db8dac20SDavid S. Miller rt = (struct rtable *)sk_dst_check(sk, 0); 1200db8dac20SDavid S. Miller 120151456b29SIan Morris if (!rt) { 120284a3aa00SPavel Emelyanov struct net *net = sock_net(sk); 12039a24abfaSDavid Ahern __u8 flow_flags = inet_sk_flowi_flags(sk); 120484a3aa00SPavel Emelyanov 1205e474995fSDavid S. Miller fl4 = &fl4_stack; 12069a24abfaSDavid Ahern 12070e26371dSGuillaume Nault flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope, 12080e26371dSGuillaume Nault sk->sk_protocol, flow_flags, faddr, saddr, 12090e26371dSGuillaume Nault dport, inet->inet_sport, sk->sk_uid); 1210c0951cbcSDavid S. Miller 12113df98d79SPaul Moore security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1212e474995fSDavid S. Miller rt = ip_route_output_flow(net, fl4, sk); 1213b23dd4feSDavid S. Miller if (IS_ERR(rt)) { 1214b23dd4feSDavid S. Miller err = PTR_ERR(rt); 121506dc94b1SDavid S. Miller rt = NULL; 1216db8dac20SDavid S. Miller if (err == -ENETUNREACH) 1217f1d8cba6SEric Dumazet IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1218db8dac20SDavid S. Miller goto out; 1219db8dac20SDavid S. Miller } 1220db8dac20SDavid S. Miller 1221db8dac20SDavid S. Miller err = -EACCES; 1222db8dac20SDavid S. Miller if ((rt->rt_flags & RTCF_BROADCAST) && 1223db8dac20SDavid S. Miller !sock_flag(sk, SOCK_BROADCAST)) 1224db8dac20SDavid S. Miller goto out; 1225db8dac20SDavid S. Miller if (connected) 1226d8d1f30bSChangli Gao sk_dst_set(sk, dst_clone(&rt->dst)); 1227db8dac20SDavid S. Miller } 1228db8dac20SDavid S. Miller 1229db8dac20SDavid S. Miller if (msg->msg_flags&MSG_CONFIRM) 1230db8dac20SDavid S. Miller goto do_confirm; 1231db8dac20SDavid S. Miller back_from_confirm: 1232db8dac20SDavid S. Miller 1233e474995fSDavid S. Miller saddr = fl4->saddr; 1234db8dac20SDavid S. Miller if (!ipc.addr) 1235e474995fSDavid S. Miller daddr = ipc.addr = fl4->daddr; 1236db8dac20SDavid S. Miller 1237903ab86dSHerbert Xu /* Lockless fast path for the non-corking case. */ 1238903ab86dSHerbert Xu if (!corkreq) { 12391cd7884dSWillem de Bruijn struct inet_cork cork; 12401cd7884dSWillem de Bruijn 1241f69e6d13SAl Viro skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1242903ab86dSHerbert Xu sizeof(struct udphdr), &ipc, &rt, 12431cd7884dSWillem de Bruijn &cork, msg->msg_flags); 1244903ab86dSHerbert Xu err = PTR_ERR(skb); 124550c3a487SYOSHIFUJI Hideaki / 吉藤英明 if (!IS_ERR_OR_NULL(skb)) 1246bec1f6f6SWillem de Bruijn err = udp_send_skb(skb, fl4, &cork); 1247903ab86dSHerbert Xu goto out; 1248903ab86dSHerbert Xu } 1249903ab86dSHerbert Xu 1250db8dac20SDavid S. Miller lock_sock(sk); 1251db8dac20SDavid S. Miller if (unlikely(up->pending)) { 1252db8dac20SDavid S. Miller /* The socket is already corked while preparing it. */ 1253db8dac20SDavid S. Miller /* ... which is an evident application bug. --ANK */ 1254db8dac20SDavid S. Miller release_sock(sk); 1255db8dac20SDavid S. Miller 1256197df02cSMatteo Croce net_dbg_ratelimited("socket already corked\n"); 1257db8dac20SDavid S. Miller err = -EINVAL; 1258db8dac20SDavid S. Miller goto out; 1259db8dac20SDavid S. Miller } 1260db8dac20SDavid S. Miller /* 1261db8dac20SDavid S. Miller * Now cork the socket to pend data. 1262db8dac20SDavid S. Miller */ 1263b6f21b26SDavid S. Miller fl4 = &inet->cork.fl.u.ip4; 1264b6f21b26SDavid S. Miller fl4->daddr = daddr; 1265b6f21b26SDavid S. Miller fl4->saddr = saddr; 12669cce96dfSDavid S. Miller fl4->fl4_dport = dport; 12679cce96dfSDavid S. Miller fl4->fl4_sport = inet->inet_sport; 1268db8dac20SDavid S. Miller up->pending = AF_INET; 1269db8dac20SDavid S. Miller 1270db8dac20SDavid S. Miller do_append_data: 1271db8dac20SDavid S. Miller up->len += ulen; 1272f69e6d13SAl Viro err = ip_append_data(sk, fl4, getfrag, msg, ulen, 12732e77d89bSEric Dumazet sizeof(struct udphdr), &ipc, &rt, 1274db8dac20SDavid S. Miller corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1275db8dac20SDavid S. Miller if (err) 1276db8dac20SDavid S. Miller udp_flush_pending_frames(sk); 1277db8dac20SDavid S. Miller else if (!corkreq) 1278db8dac20SDavid S. Miller err = udp_push_pending_frames(sk); 1279db8dac20SDavid S. Miller else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1280db8dac20SDavid S. Miller up->pending = 0; 1281db8dac20SDavid S. Miller release_sock(sk); 1282db8dac20SDavid S. Miller 1283db8dac20SDavid S. Miller out: 1284db8dac20SDavid S. Miller ip_rt_put(rt); 12851b97013bSAndrey Ignatov out_free: 1286db8dac20SDavid S. Miller if (free) 1287db8dac20SDavid S. Miller kfree(ipc.opt); 1288db8dac20SDavid S. Miller if (!err) 1289db8dac20SDavid S. Miller return len; 1290db8dac20SDavid S. Miller /* 1291db8dac20SDavid S. Miller * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1292db8dac20SDavid S. Miller * ENOBUFS might not be good (it's not tunable per se), but otherwise 1293db8dac20SDavid S. Miller * we don't have a good statistic (IpOutDiscards but it can be too many 1294db8dac20SDavid S. Miller * things). We could add another new stat but at least for now that 1295db8dac20SDavid S. Miller * seems like overkill. 1296db8dac20SDavid S. Miller */ 1297db8dac20SDavid S. Miller if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 12986aef70a8SEric Dumazet UDP_INC_STATS(sock_net(sk), 1299629ca23cSPavel Emelyanov UDP_MIB_SNDBUFERRORS, is_udplite); 1300db8dac20SDavid S. Miller } 1301db8dac20SDavid S. Miller return err; 1302db8dac20SDavid S. Miller 1303db8dac20SDavid S. Miller do_confirm: 13040dec879fSJulian Anastasov if (msg->msg_flags & MSG_PROBE) 13050dec879fSJulian Anastasov dst_confirm_neigh(&rt->dst, &fl4->daddr); 1306db8dac20SDavid S. Miller if (!(msg->msg_flags&MSG_PROBE) || len) 1307db8dac20SDavid S. Miller goto back_from_confirm; 1308db8dac20SDavid S. Miller err = 0; 1309db8dac20SDavid S. Miller goto out; 1310db8dac20SDavid S. Miller } 1311c482c568SEric Dumazet EXPORT_SYMBOL(udp_sendmsg); 1312db8dac20SDavid S. Miller 13131d7e4538SDavid Howells void udp_splice_eof(struct socket *sock) 13141d7e4538SDavid Howells { 13151d7e4538SDavid Howells struct sock *sk = sock->sk; 13161d7e4538SDavid Howells struct udp_sock *up = udp_sk(sk); 13171d7e4538SDavid Howells 13181d7e4538SDavid Howells if (!up->pending || READ_ONCE(up->corkflag)) 13191d7e4538SDavid Howells return; 13201d7e4538SDavid Howells 13211d7e4538SDavid Howells lock_sock(sk); 13221d7e4538SDavid Howells if (up->pending && !READ_ONCE(up->corkflag)) 13231d7e4538SDavid Howells udp_push_pending_frames(sk); 13241d7e4538SDavid Howells release_sock(sk); 13251d7e4538SDavid Howells } 13261d7e4538SDavid Howells EXPORT_SYMBOL_GPL(udp_splice_eof); 13271d7e4538SDavid Howells 1328dce4551cSPaolo Abeni #define UDP_SKB_IS_STATELESS 0x80000000 1329dce4551cSPaolo Abeni 1330677bf08cSFlorian Westphal /* all head states (dst, sk, nf conntrack) except skb extensions are 1331677bf08cSFlorian Westphal * cleared by udp_rcv(). 1332677bf08cSFlorian Westphal * 1333677bf08cSFlorian Westphal * We need to preserve secpath, if present, to eventually process 1334677bf08cSFlorian Westphal * IP_CMSG_PASSSEC at recvmsg() time. 1335677bf08cSFlorian Westphal * 1336677bf08cSFlorian Westphal * Other extensions can be cleared. 1337677bf08cSFlorian Westphal */ 1338677bf08cSFlorian Westphal static bool udp_try_make_stateless(struct sk_buff *skb) 1339677bf08cSFlorian Westphal { 1340677bf08cSFlorian Westphal if (!skb_has_extensions(skb)) 1341677bf08cSFlorian Westphal return true; 1342677bf08cSFlorian Westphal 1343677bf08cSFlorian Westphal if (!secpath_exists(skb)) { 1344677bf08cSFlorian Westphal skb_ext_reset(skb); 1345677bf08cSFlorian Westphal return true; 1346677bf08cSFlorian Westphal } 1347677bf08cSFlorian Westphal 1348677bf08cSFlorian Westphal return false; 1349677bf08cSFlorian Westphal } 1350677bf08cSFlorian Westphal 1351b65ac446SPaolo Abeni static void udp_set_dev_scratch(struct sk_buff *skb) 1352b65ac446SPaolo Abeni { 1353dce4551cSPaolo Abeni struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1354b65ac446SPaolo Abeni 1355b65ac446SPaolo Abeni BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1356dce4551cSPaolo Abeni scratch->_tsize_state = skb->truesize; 1357dce4551cSPaolo Abeni #if BITS_PER_LONG == 64 1358b65ac446SPaolo Abeni scratch->len = skb->len; 1359b65ac446SPaolo Abeni scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1360b65ac446SPaolo Abeni scratch->is_linear = !skb_is_nonlinear(skb); 1361b65ac446SPaolo Abeni #endif 1362677bf08cSFlorian Westphal if (udp_try_make_stateless(skb)) 1363dce4551cSPaolo Abeni scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1364dce4551cSPaolo Abeni } 1365dce4551cSPaolo Abeni 1366a793183cSEric Dumazet static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1367a793183cSEric Dumazet { 1368a793183cSEric Dumazet /* We come here after udp_lib_checksum_complete() returned 0. 1369a793183cSEric Dumazet * This means that __skb_checksum_complete() might have 1370a793183cSEric Dumazet * set skb->csum_valid to 1. 1371a793183cSEric Dumazet * On 64bit platforms, we can set csum_unnecessary 1372a793183cSEric Dumazet * to true, but only if the skb is not shared. 1373a793183cSEric Dumazet */ 1374a793183cSEric Dumazet #if BITS_PER_LONG == 64 1375a793183cSEric Dumazet if (!skb_shared(skb)) 1376a793183cSEric Dumazet udp_skb_scratch(skb)->csum_unnecessary = true; 1377a793183cSEric Dumazet #endif 1378a793183cSEric Dumazet } 1379a793183cSEric Dumazet 1380dce4551cSPaolo Abeni static int udp_skb_truesize(struct sk_buff *skb) 1381dce4551cSPaolo Abeni { 1382dce4551cSPaolo Abeni return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1383dce4551cSPaolo Abeni } 1384dce4551cSPaolo Abeni 1385dce4551cSPaolo Abeni static bool udp_skb_has_head_state(struct sk_buff *skb) 1386dce4551cSPaolo Abeni { 1387dce4551cSPaolo Abeni return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1388dce4551cSPaolo Abeni } 1389b65ac446SPaolo Abeni 13907c13f97fSPaolo Abeni /* fully reclaim rmem/fwd memory allocated for skb */ 13916dfb4367SPaolo Abeni static void udp_rmem_release(struct sock *sk, int size, int partial, 13926dfb4367SPaolo Abeni bool rx_queue_lock_held) 1393f970bd9eSPaolo Abeni { 13946b229cf7SEric Dumazet struct udp_sock *up = udp_sk(sk); 13952276f58aSPaolo Abeni struct sk_buff_head *sk_queue; 1396f970bd9eSPaolo Abeni int amt; 1397f970bd9eSPaolo Abeni 13986b229cf7SEric Dumazet if (likely(partial)) { 13996b229cf7SEric Dumazet up->forward_deficit += size; 14006b229cf7SEric Dumazet size = up->forward_deficit; 14018a3854c7SPaolo Abeni if (size < READ_ONCE(up->forward_threshold) && 1402d39ca259SPaolo Abeni !skb_queue_empty(&up->reader_queue)) 14036b229cf7SEric Dumazet return; 14046b229cf7SEric Dumazet } else { 14056b229cf7SEric Dumazet size += up->forward_deficit; 14066b229cf7SEric Dumazet } 14076b229cf7SEric Dumazet up->forward_deficit = 0; 14086b229cf7SEric Dumazet 14096dfb4367SPaolo Abeni /* acquire the sk_receive_queue for fwd allocated memory scheduling, 14106dfb4367SPaolo Abeni * if the called don't held it already 14116dfb4367SPaolo Abeni */ 14122276f58aSPaolo Abeni sk_queue = &sk->sk_receive_queue; 14136dfb4367SPaolo Abeni if (!rx_queue_lock_held) 14142276f58aSPaolo Abeni spin_lock(&sk_queue->lock); 14152276f58aSPaolo Abeni 14166dfb4367SPaolo Abeni 1417f970bd9eSPaolo Abeni sk->sk_forward_alloc += size; 1418100fdd1fSEric Dumazet amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); 1419f970bd9eSPaolo Abeni sk->sk_forward_alloc -= amt; 1420f970bd9eSPaolo Abeni 1421f970bd9eSPaolo Abeni if (amt) 1422100fdd1fSEric Dumazet __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); 142302ab0d13SEric Dumazet 142402ab0d13SEric Dumazet atomic_sub(size, &sk->sk_rmem_alloc); 14252276f58aSPaolo Abeni 14262276f58aSPaolo Abeni /* this can save us from acquiring the rx queue lock on next receive */ 14272276f58aSPaolo Abeni skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 14282276f58aSPaolo Abeni 14296dfb4367SPaolo Abeni if (!rx_queue_lock_held) 14302276f58aSPaolo Abeni spin_unlock(&sk_queue->lock); 1431f970bd9eSPaolo Abeni } 1432f970bd9eSPaolo Abeni 14332276f58aSPaolo Abeni /* Note: called with reader_queue.lock held. 1434c84d9490SEric Dumazet * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1435c84d9490SEric Dumazet * This avoids a cache line miss while receive_queue lock is held. 1436c84d9490SEric Dumazet * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1437c84d9490SEric Dumazet */ 14387c13f97fSPaolo Abeni void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1439f970bd9eSPaolo Abeni { 1440b65ac446SPaolo Abeni prefetch(&skb->data); 1441b65ac446SPaolo Abeni udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1442f970bd9eSPaolo Abeni } 14437c13f97fSPaolo Abeni EXPORT_SYMBOL(udp_skb_destructor); 1444f970bd9eSPaolo Abeni 14456dfb4367SPaolo Abeni /* as above, but the caller held the rx queue lock, too */ 144664f5102dSColin Ian King static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 14476dfb4367SPaolo Abeni { 1448b65ac446SPaolo Abeni prefetch(&skb->data); 1449b65ac446SPaolo Abeni udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 14506dfb4367SPaolo Abeni } 14516dfb4367SPaolo Abeni 14524b272750SEric Dumazet /* Idea of busylocks is to let producers grab an extra spinlock 14534b272750SEric Dumazet * to relieve pressure on the receive_queue spinlock shared by consumer. 14544b272750SEric Dumazet * Under flood, this means that only one producer can be in line 14554b272750SEric Dumazet * trying to acquire the receive_queue spinlock. 14564b272750SEric Dumazet * These busylock can be allocated on a per cpu manner, instead of a 14574b272750SEric Dumazet * per socket one (that would consume a cache line per socket) 14584b272750SEric Dumazet */ 14594b272750SEric Dumazet static int udp_busylocks_log __read_mostly; 14604b272750SEric Dumazet static spinlock_t *udp_busylocks __read_mostly; 14614b272750SEric Dumazet 14624b272750SEric Dumazet static spinlock_t *busylock_acquire(void *ptr) 14634b272750SEric Dumazet { 14644b272750SEric Dumazet spinlock_t *busy; 14654b272750SEric Dumazet 14664b272750SEric Dumazet busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 14674b272750SEric Dumazet spin_lock(busy); 14684b272750SEric Dumazet return busy; 14694b272750SEric Dumazet } 14704b272750SEric Dumazet 14714b272750SEric Dumazet static void busylock_release(spinlock_t *busy) 14724b272750SEric Dumazet { 14734b272750SEric Dumazet if (busy) 14744b272750SEric Dumazet spin_unlock(busy); 14754b272750SEric Dumazet } 14764b272750SEric Dumazet 1477fd9c31f8SJason Xing static int udp_rmem_schedule(struct sock *sk, int size) 1478fd9c31f8SJason Xing { 1479fd9c31f8SJason Xing int delta; 1480fd9c31f8SJason Xing 1481fd9c31f8SJason Xing delta = size - sk->sk_forward_alloc; 1482fd9c31f8SJason Xing if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV)) 1483fd9c31f8SJason Xing return -ENOBUFS; 1484fd9c31f8SJason Xing 1485fd9c31f8SJason Xing return 0; 1486fd9c31f8SJason Xing } 1487fd9c31f8SJason Xing 1488f970bd9eSPaolo Abeni int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1489f970bd9eSPaolo Abeni { 1490f970bd9eSPaolo Abeni struct sk_buff_head *list = &sk->sk_receive_queue; 1491fd9c31f8SJason Xing int rmem, err = -ENOMEM; 14924b272750SEric Dumazet spinlock_t *busy = NULL; 1493c8c8b127SEric Dumazet int size; 1494f970bd9eSPaolo Abeni 1495f970bd9eSPaolo Abeni /* try to avoid the costly atomic add/sub pair when the receive 1496f970bd9eSPaolo Abeni * queue is full; always allow at least a packet 1497f970bd9eSPaolo Abeni */ 1498f970bd9eSPaolo Abeni rmem = atomic_read(&sk->sk_rmem_alloc); 1499363dc73aSPaolo Abeni if (rmem > sk->sk_rcvbuf) 1500f970bd9eSPaolo Abeni goto drop; 1501f970bd9eSPaolo Abeni 1502c8c8b127SEric Dumazet /* Under mem pressure, it might be helpful to help udp_recvmsg() 1503c8c8b127SEric Dumazet * having linear skbs : 1504c8c8b127SEric Dumazet * - Reduce memory overhead and thus increase receive queue capacity 1505c8c8b127SEric Dumazet * - Less cache line misses at copyout() time 1506c8c8b127SEric Dumazet * - Less work at consume_skb() (less alien page frag freeing) 1507c8c8b127SEric Dumazet */ 15084b272750SEric Dumazet if (rmem > (sk->sk_rcvbuf >> 1)) { 1509c8c8b127SEric Dumazet skb_condense(skb); 15104b272750SEric Dumazet 15114b272750SEric Dumazet busy = busylock_acquire(sk); 15124b272750SEric Dumazet } 1513c8c8b127SEric Dumazet size = skb->truesize; 1514b65ac446SPaolo Abeni udp_set_dev_scratch(skb); 1515c8c8b127SEric Dumazet 1516f970bd9eSPaolo Abeni /* we drop only if the receive buf is full and the receive 1517f970bd9eSPaolo Abeni * queue contains some other skb 1518f970bd9eSPaolo Abeni */ 1519f970bd9eSPaolo Abeni rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1520feed8a4fSAntonio Messina if (rmem > (size + (unsigned int)sk->sk_rcvbuf)) 1521f970bd9eSPaolo Abeni goto uncharge_drop; 1522f970bd9eSPaolo Abeni 1523f970bd9eSPaolo Abeni spin_lock(&list->lock); 1524fd9c31f8SJason Xing err = udp_rmem_schedule(sk, size); 1525fd9c31f8SJason Xing if (err) { 1526f970bd9eSPaolo Abeni spin_unlock(&list->lock); 1527f970bd9eSPaolo Abeni goto uncharge_drop; 1528f970bd9eSPaolo Abeni } 1529f970bd9eSPaolo Abeni 1530f970bd9eSPaolo Abeni sk->sk_forward_alloc -= size; 1531f970bd9eSPaolo Abeni 15327c13f97fSPaolo Abeni /* no need to setup a destructor, we will explicitly release the 15337c13f97fSPaolo Abeni * forward allocated memory on dequeue 15347c13f97fSPaolo Abeni */ 1535f970bd9eSPaolo Abeni sock_skb_set_dropcount(sk, skb); 1536f970bd9eSPaolo Abeni 1537f970bd9eSPaolo Abeni __skb_queue_tail(list, skb); 1538f970bd9eSPaolo Abeni spin_unlock(&list->lock); 1539f970bd9eSPaolo Abeni 1540f970bd9eSPaolo Abeni if (!sock_flag(sk, SOCK_DEAD)) 15410558e167SPaolo Abeni INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk); 1542f970bd9eSPaolo Abeni 15434b272750SEric Dumazet busylock_release(busy); 1544f970bd9eSPaolo Abeni return 0; 1545f970bd9eSPaolo Abeni 1546f970bd9eSPaolo Abeni uncharge_drop: 1547f970bd9eSPaolo Abeni atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1548f970bd9eSPaolo Abeni 1549f970bd9eSPaolo Abeni drop: 1550f970bd9eSPaolo Abeni atomic_inc(&sk->sk_drops); 15514b272750SEric Dumazet busylock_release(busy); 1552f970bd9eSPaolo Abeni return err; 1553f970bd9eSPaolo Abeni } 1554f970bd9eSPaolo Abeni EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1555f970bd9eSPaolo Abeni 1556d38afeecSKuniyuki Iwashima void udp_destruct_common(struct sock *sk) 1557f970bd9eSPaolo Abeni { 1558f970bd9eSPaolo Abeni /* reclaim completely the forward allocated memory */ 15592276f58aSPaolo Abeni struct udp_sock *up = udp_sk(sk); 15607c13f97fSPaolo Abeni unsigned int total = 0; 15617c13f97fSPaolo Abeni struct sk_buff *skb; 15627c13f97fSPaolo Abeni 15632276f58aSPaolo Abeni skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 15642276f58aSPaolo Abeni while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 15657c13f97fSPaolo Abeni total += skb->truesize; 15667c13f97fSPaolo Abeni kfree_skb(skb); 15677c13f97fSPaolo Abeni } 15686dfb4367SPaolo Abeni udp_rmem_release(sk, total, 0, true); 1569d38afeecSKuniyuki Iwashima } 1570d38afeecSKuniyuki Iwashima EXPORT_SYMBOL_GPL(udp_destruct_common); 15717c13f97fSPaolo Abeni 1572d38afeecSKuniyuki Iwashima static void udp_destruct_sock(struct sock *sk) 1573d38afeecSKuniyuki Iwashima { 1574d38afeecSKuniyuki Iwashima udp_destruct_common(sk); 1575f970bd9eSPaolo Abeni inet_sock_destruct(sk); 1576f970bd9eSPaolo Abeni } 1577f970bd9eSPaolo Abeni 1578f970bd9eSPaolo Abeni int udp_init_sock(struct sock *sk) 1579f970bd9eSPaolo Abeni { 15808a3854c7SPaolo Abeni udp_lib_init_sock(sk); 1581f970bd9eSPaolo Abeni sk->sk_destruct = udp_destruct_sock; 1582e993ffe3SPavel Begunkov set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1583f970bd9eSPaolo Abeni return 0; 1584f970bd9eSPaolo Abeni } 1585f970bd9eSPaolo Abeni 1586f970bd9eSPaolo Abeni void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1587f970bd9eSPaolo Abeni { 1588f970bd9eSPaolo Abeni if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1589f970bd9eSPaolo Abeni bool slow = lock_sock_fast(sk); 1590f970bd9eSPaolo Abeni 1591f970bd9eSPaolo Abeni sk_peek_offset_bwd(sk, len); 1592f970bd9eSPaolo Abeni unlock_sock_fast(sk, slow); 1593f970bd9eSPaolo Abeni } 15940a463c78SPaolo Abeni 1595ca2c1418SPaolo Abeni if (!skb_unref(skb)) 1596ca2c1418SPaolo Abeni return; 1597ca2c1418SPaolo Abeni 1598dce4551cSPaolo Abeni /* In the more common cases we cleared the head states previously, 1599dce4551cSPaolo Abeni * see __udp_queue_rcv_skb(). 16000ddf3fb2SPaolo Abeni */ 1601dce4551cSPaolo Abeni if (unlikely(udp_skb_has_head_state(skb))) 16020ddf3fb2SPaolo Abeni skb_release_head_state(skb); 1603ca2c1418SPaolo Abeni __consume_stateless_skb(skb); 1604f970bd9eSPaolo Abeni } 1605f970bd9eSPaolo Abeni EXPORT_SYMBOL_GPL(skb_consume_udp); 1606f970bd9eSPaolo Abeni 16072276f58aSPaolo Abeni static struct sk_buff *__first_packet_length(struct sock *sk, 16082276f58aSPaolo Abeni struct sk_buff_head *rcvq, 16092276f58aSPaolo Abeni int *total) 16102276f58aSPaolo Abeni { 16112276f58aSPaolo Abeni struct sk_buff *skb; 16122276f58aSPaolo Abeni 16139bd780f5SPaolo Abeni while ((skb = skb_peek(rcvq)) != NULL) { 16149bd780f5SPaolo Abeni if (udp_lib_checksum_complete(skb)) { 16152276f58aSPaolo Abeni __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 16162276f58aSPaolo Abeni IS_UDPLITE(sk)); 16172276f58aSPaolo Abeni __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 16182276f58aSPaolo Abeni IS_UDPLITE(sk)); 16192276f58aSPaolo Abeni atomic_inc(&sk->sk_drops); 16202276f58aSPaolo Abeni __skb_unlink(skb, rcvq); 16212276f58aSPaolo Abeni *total += skb->truesize; 16222276f58aSPaolo Abeni kfree_skb(skb); 16239bd780f5SPaolo Abeni } else { 1624a793183cSEric Dumazet udp_skb_csum_unnecessary_set(skb); 16259bd780f5SPaolo Abeni break; 16269bd780f5SPaolo Abeni } 16272276f58aSPaolo Abeni } 16282276f58aSPaolo Abeni return skb; 16292276f58aSPaolo Abeni } 16302276f58aSPaolo Abeni 163185584672SEric Dumazet /** 163285584672SEric Dumazet * first_packet_length - return length of first packet in receive queue 163385584672SEric Dumazet * @sk: socket 163485584672SEric Dumazet * 163585584672SEric Dumazet * Drops all bad checksum frames, until a valid one is found. 1636e83c6744SEric Dumazet * Returns the length of found skb, or -1 if none is found. 163785584672SEric Dumazet */ 1638e83c6744SEric Dumazet static int first_packet_length(struct sock *sk) 163985584672SEric Dumazet { 16402276f58aSPaolo Abeni struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 16412276f58aSPaolo Abeni struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 164285584672SEric Dumazet struct sk_buff *skb; 16437c13f97fSPaolo Abeni int total = 0; 1644e83c6744SEric Dumazet int res; 164585584672SEric Dumazet 164685584672SEric Dumazet spin_lock_bh(&rcvq->lock); 16472276f58aSPaolo Abeni skb = __first_packet_length(sk, rcvq, &total); 1648137a0dbeSEric Dumazet if (!skb && !skb_queue_empty_lockless(sk_queue)) { 16492276f58aSPaolo Abeni spin_lock(&sk_queue->lock); 16502276f58aSPaolo Abeni skb_queue_splice_tail_init(sk_queue, rcvq); 16512276f58aSPaolo Abeni spin_unlock(&sk_queue->lock); 16522276f58aSPaolo Abeni 16532276f58aSPaolo Abeni skb = __first_packet_length(sk, rcvq, &total); 165485584672SEric Dumazet } 1655e83c6744SEric Dumazet res = skb ? skb->len : -1; 16567c13f97fSPaolo Abeni if (total) 16576dfb4367SPaolo Abeni udp_rmem_release(sk, total, 1, false); 165885584672SEric Dumazet spin_unlock_bh(&rcvq->lock); 165985584672SEric Dumazet return res; 166085584672SEric Dumazet } 166185584672SEric Dumazet 16621da177e4SLinus Torvalds /* 16631da177e4SLinus Torvalds * IOCTL requests applicable to the UDP protocol 16641da177e4SLinus Torvalds */ 16651da177e4SLinus Torvalds 1666e1d001faSBreno Leitao int udp_ioctl(struct sock *sk, int cmd, int *karg) 16671da177e4SLinus Torvalds { 16686516c655SStephen Hemminger switch (cmd) { 16691da177e4SLinus Torvalds case SIOCOUTQ: 16701da177e4SLinus Torvalds { 1671e1d001faSBreno Leitao *karg = sk_wmem_alloc_get(sk); 1672e1d001faSBreno Leitao return 0; 16731da177e4SLinus Torvalds } 16741da177e4SLinus Torvalds 16751da177e4SLinus Torvalds case SIOCINQ: 16761da177e4SLinus Torvalds { 1677e1d001faSBreno Leitao *karg = max_t(int, 0, first_packet_length(sk)); 1678e1d001faSBreno Leitao return 0; 16791da177e4SLinus Torvalds } 16801da177e4SLinus Torvalds 16811da177e4SLinus Torvalds default: 16821da177e4SLinus Torvalds return -ENOIOCTLCMD; 16831da177e4SLinus Torvalds } 16846516c655SStephen Hemminger 16856516c655SStephen Hemminger return 0; 16861da177e4SLinus Torvalds } 1687c482c568SEric Dumazet EXPORT_SYMBOL(udp_ioctl); 16881da177e4SLinus Torvalds 16892276f58aSPaolo Abeni struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1690ec095263SOliver Hartkopp int *off, int *err) 16912276f58aSPaolo Abeni { 16922276f58aSPaolo Abeni struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 16932276f58aSPaolo Abeni struct sk_buff_head *queue; 16942276f58aSPaolo Abeni struct sk_buff *last; 16952276f58aSPaolo Abeni long timeo; 16962276f58aSPaolo Abeni int error; 16972276f58aSPaolo Abeni 16982276f58aSPaolo Abeni queue = &udp_sk(sk)->reader_queue; 16992276f58aSPaolo Abeni timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 17002276f58aSPaolo Abeni do { 17012276f58aSPaolo Abeni struct sk_buff *skb; 17022276f58aSPaolo Abeni 17032276f58aSPaolo Abeni error = sock_error(sk); 17042276f58aSPaolo Abeni if (error) 17052276f58aSPaolo Abeni break; 17062276f58aSPaolo Abeni 17072276f58aSPaolo Abeni error = -EAGAIN; 17082276f58aSPaolo Abeni do { 17092276f58aSPaolo Abeni spin_lock_bh(&queue->lock); 1710e427cad6SPaolo Abeni skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1711e427cad6SPaolo Abeni err, &last); 17122276f58aSPaolo Abeni if (skb) { 1713e427cad6SPaolo Abeni if (!(flags & MSG_PEEK)) 1714e427cad6SPaolo Abeni udp_skb_destructor(sk, skb); 17152276f58aSPaolo Abeni spin_unlock_bh(&queue->lock); 17162276f58aSPaolo Abeni return skb; 17172276f58aSPaolo Abeni } 17182276f58aSPaolo Abeni 1719137a0dbeSEric Dumazet if (skb_queue_empty_lockless(sk_queue)) { 17202276f58aSPaolo Abeni spin_unlock_bh(&queue->lock); 17212276f58aSPaolo Abeni goto busy_check; 17222276f58aSPaolo Abeni } 17232276f58aSPaolo Abeni 17246dfb4367SPaolo Abeni /* refill the reader queue and walk it again 17256dfb4367SPaolo Abeni * keep both queues locked to avoid re-acquiring 17266dfb4367SPaolo Abeni * the sk_receive_queue lock if fwd memory scheduling 17276dfb4367SPaolo Abeni * is needed. 17286dfb4367SPaolo Abeni */ 17292276f58aSPaolo Abeni spin_lock(&sk_queue->lock); 17302276f58aSPaolo Abeni skb_queue_splice_tail_init(sk_queue, queue); 17312276f58aSPaolo Abeni 1732e427cad6SPaolo Abeni skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1733e427cad6SPaolo Abeni err, &last); 1734e427cad6SPaolo Abeni if (skb && !(flags & MSG_PEEK)) 1735e427cad6SPaolo Abeni udp_skb_dtor_locked(sk, skb); 17366dfb4367SPaolo Abeni spin_unlock(&sk_queue->lock); 17372276f58aSPaolo Abeni spin_unlock_bh(&queue->lock); 1738de321ed3SAndrey Vagin if (skb) 17392276f58aSPaolo Abeni return skb; 17402276f58aSPaolo Abeni 17412276f58aSPaolo Abeni busy_check: 17422276f58aSPaolo Abeni if (!sk_can_busy_loop(sk)) 17432276f58aSPaolo Abeni break; 17442276f58aSPaolo Abeni 17452276f58aSPaolo Abeni sk_busy_loop(sk, flags & MSG_DONTWAIT); 1746137a0dbeSEric Dumazet } while (!skb_queue_empty_lockless(sk_queue)); 17472276f58aSPaolo Abeni 17482276f58aSPaolo Abeni /* sk_queue is empty, reader_queue may contain peeked packets */ 17492276f58aSPaolo Abeni } while (timeo && 1750b50b0580SSabrina Dubroca !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1751b50b0580SSabrina Dubroca &error, &timeo, 17522276f58aSPaolo Abeni (struct sk_buff *)sk_queue)); 17532276f58aSPaolo Abeni 17542276f58aSPaolo Abeni *err = error; 17552276f58aSPaolo Abeni return NULL; 17562276f58aSPaolo Abeni } 17577e823644SJiri Kosina EXPORT_SYMBOL(__skb_recv_udp); 17582276f58aSPaolo Abeni 1759965b57b4SCong Wang int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) 1760d7f57118SCong Wang { 1761d7f57118SCong Wang struct sk_buff *skb; 176278fa0d61SJohn Fastabend int err; 1763d7f57118SCong Wang 176431f1fbcbSPeilin Ye try_again: 1765ec095263SOliver Hartkopp skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1766d7f57118SCong Wang if (!skb) 1767d7f57118SCong Wang return err; 1768099f896fSCong Wang 1769099f896fSCong Wang if (udp_lib_checksum_complete(skb)) { 177031f1fbcbSPeilin Ye int is_udplite = IS_UDPLITE(sk); 177131f1fbcbSPeilin Ye struct net *net = sock_net(sk); 177231f1fbcbSPeilin Ye 177331f1fbcbSPeilin Ye __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite); 177431f1fbcbSPeilin Ye __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite); 1775099f896fSCong Wang atomic_inc(&sk->sk_drops); 1776099f896fSCong Wang kfree_skb(skb); 177731f1fbcbSPeilin Ye goto try_again; 1778099f896fSCong Wang } 1779099f896fSCong Wang 1780db39dfdcSPeilin Ye WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk)); 178178fa0d61SJohn Fastabend return recv_actor(sk, skb); 1782d7f57118SCong Wang } 1783965b57b4SCong Wang EXPORT_SYMBOL(udp_read_skb); 1784d7f57118SCong Wang 1785db8dac20SDavid S. Miller /* 1786db8dac20SDavid S. Miller * This should be easy, if there is something there we 1787db8dac20SDavid S. Miller * return it, otherwise we block. 1788db8dac20SDavid S. Miller */ 1789db8dac20SDavid S. Miller 1790ec095263SOliver Hartkopp int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 1791ec095263SOliver Hartkopp int *addr_len) 1792db8dac20SDavid S. Miller { 1793db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sk); 1794342dfc30SSteffen Hurrle DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1795db8dac20SDavid S. Miller struct sk_buff *skb; 179659c2cdaeSDavid S. Miller unsigned int ulen, copied; 1797fd69c399SPaolo Abeni int off, err, peeking = flags & MSG_PEEK; 1798db8dac20SDavid S. Miller int is_udplite = IS_UDPLITE(sk); 1799197c949eSEric Dumazet bool checksum_valid = false; 1800db8dac20SDavid S. Miller 1801db8dac20SDavid S. Miller if (flags & MSG_ERRQUEUE) 180285fbaa75SHannes Frederic Sowa return ip_recv_error(sk, msg, len, addr_len); 1803db8dac20SDavid S. Miller 1804db8dac20SDavid S. Miller try_again: 1805a0917e0bSMatthew Dawson off = sk_peek_offset(sk, flags); 1806ec095263SOliver Hartkopp skb = __skb_recv_udp(sk, flags, &off, &err); 1807db8dac20SDavid S. Miller if (!skb) 1808627d2d6bSsamanthakumar return err; 1809db8dac20SDavid S. Miller 1810b65ac446SPaolo Abeni ulen = udp_skb_len(skb); 181159c2cdaeSDavid S. Miller copied = len; 1812627d2d6bSsamanthakumar if (copied > ulen - off) 1813627d2d6bSsamanthakumar copied = ulen - off; 181459c2cdaeSDavid S. Miller else if (copied < ulen) 1815db8dac20SDavid S. Miller msg->msg_flags |= MSG_TRUNC; 1816db8dac20SDavid S. Miller 1817db8dac20SDavid S. Miller /* 1818db8dac20SDavid S. Miller * If checksum is needed at all, try to do it while copying the 1819db8dac20SDavid S. Miller * data. If the data is truncated, or if we only want a partial 1820db8dac20SDavid S. Miller * coverage checksum (UDP-Lite), do it before the copy. 1821db8dac20SDavid S. Miller */ 1822db8dac20SDavid S. Miller 1823d21dbdfeSEric Dumazet if (copied < ulen || peeking || 1824d21dbdfeSEric Dumazet (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1825b65ac446SPaolo Abeni checksum_valid = udp_skb_csum_unnecessary(skb) || 1826b65ac446SPaolo Abeni !__udp_lib_checksum_complete(skb); 1827197c949eSEric Dumazet if (!checksum_valid) 1828db8dac20SDavid S. Miller goto csum_copy_err; 1829db8dac20SDavid S. Miller } 1830db8dac20SDavid S. Miller 1831b65ac446SPaolo Abeni if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1832b65ac446SPaolo Abeni if (udp_skb_is_linear(skb)) 1833b65ac446SPaolo Abeni err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1834b65ac446SPaolo Abeni else 1835627d2d6bSsamanthakumar err = skb_copy_datagram_msg(skb, off, msg, copied); 1836b65ac446SPaolo Abeni } else { 1837627d2d6bSsamanthakumar err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1838db8dac20SDavid S. Miller 1839db8dac20SDavid S. Miller if (err == -EINVAL) 1840db8dac20SDavid S. Miller goto csum_copy_err; 1841db8dac20SDavid S. Miller } 1842db8dac20SDavid S. Miller 184322911fc5SEric Dumazet if (unlikely(err)) { 1844fd69c399SPaolo Abeni if (!peeking) { 1845979402b1SEric Dumazet atomic_inc(&sk->sk_drops); 18466aef70a8SEric Dumazet UDP_INC_STATS(sock_net(sk), 1847979402b1SEric Dumazet UDP_MIB_INERRORS, is_udplite); 1848979402b1SEric Dumazet } 1849850cbaddSPaolo Abeni kfree_skb(skb); 1850627d2d6bSsamanthakumar return err; 185122911fc5SEric Dumazet } 1852db8dac20SDavid S. Miller 1853fd69c399SPaolo Abeni if (!peeking) 18546aef70a8SEric Dumazet UDP_INC_STATS(sock_net(sk), 1855629ca23cSPavel Emelyanov UDP_MIB_INDATAGRAMS, is_udplite); 1856db8dac20SDavid S. Miller 18576fd1d51cSErin MacNeil sock_recv_cmsgs(msg, sk, skb); 1858db8dac20SDavid S. Miller 1859db8dac20SDavid S. Miller /* Copy the address. */ 1860c482c568SEric Dumazet if (sin) { 1861db8dac20SDavid S. Miller sin->sin_family = AF_INET; 1862db8dac20SDavid S. Miller sin->sin_port = udp_hdr(skb)->source; 1863db8dac20SDavid S. Miller sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1864db8dac20SDavid S. Miller memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1865bceaa902SHannes Frederic Sowa *addr_len = sizeof(*sin); 1866983695faSDaniel Borkmann 1867983695faSDaniel Borkmann BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1868983695faSDaniel Borkmann (struct sockaddr *)sin); 1869db8dac20SDavid S. Miller } 1870bcd1665eSPaolo Abeni 1871bcd1665eSPaolo Abeni if (udp_sk(sk)->gro_enabled) 1872bcd1665eSPaolo Abeni udp_cmsg_recv(msg, sk, skb); 1873bcd1665eSPaolo Abeni 1874c274af22SEric Dumazet if (inet_cmsg_flags(inet)) 1875ad959036SPaolo Abeni ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1876db8dac20SDavid S. Miller 187759c2cdaeSDavid S. Miller err = copied; 1878db8dac20SDavid S. Miller if (flags & MSG_TRUNC) 1879db8dac20SDavid S. Miller err = ulen; 1880db8dac20SDavid S. Miller 1881850cbaddSPaolo Abeni skb_consume_udp(sk, skb, peeking ? -err : err); 1882db8dac20SDavid S. Miller return err; 1883db8dac20SDavid S. Miller 1884db8dac20SDavid S. Miller csum_copy_err: 18852276f58aSPaolo Abeni if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 18862276f58aSPaolo Abeni udp_skb_destructor)) { 18876aef70a8SEric Dumazet UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 18886aef70a8SEric Dumazet UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 18896a5dc9e5SEric Dumazet } 1890850cbaddSPaolo Abeni kfree_skb(skb); 1891db8dac20SDavid S. Miller 1892beb39db5SEric Dumazet /* starting over for a new packet, but check if we need to yield */ 1893beb39db5SEric Dumazet cond_resched(); 18949cfaa8deSXufeng Zhang msg->msg_flags &= ~MSG_TRUNC; 1895db8dac20SDavid S. Miller goto try_again; 1896db8dac20SDavid S. Miller } 1897db8dac20SDavid S. Miller 1898d74bad4eSAndrey Ignatov int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1899d74bad4eSAndrey Ignatov { 1900d74bad4eSAndrey Ignatov /* This check is replicated from __ip4_datagram_connect() and 1901d74bad4eSAndrey Ignatov * intended to prevent BPF program called below from accessing bytes 1902d74bad4eSAndrey Ignatov * that are out of the bound specified by user in addr_len. 1903d74bad4eSAndrey Ignatov */ 1904d74bad4eSAndrey Ignatov if (addr_len < sizeof(struct sockaddr_in)) 1905d74bad4eSAndrey Ignatov return -EINVAL; 1906d74bad4eSAndrey Ignatov 1907d74bad4eSAndrey Ignatov return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1908d74bad4eSAndrey Ignatov } 1909d74bad4eSAndrey Ignatov EXPORT_SYMBOL(udp_pre_connect); 1910d74bad4eSAndrey Ignatov 1911286c72deSEric Dumazet int __udp_disconnect(struct sock *sk, int flags) 19121da177e4SLinus Torvalds { 19131da177e4SLinus Torvalds struct inet_sock *inet = inet_sk(sk); 19141da177e4SLinus Torvalds /* 19151da177e4SLinus Torvalds * 1003.1g - break association. 19161da177e4SLinus Torvalds */ 19171da177e4SLinus Torvalds 19181da177e4SLinus Torvalds sk->sk_state = TCP_CLOSE; 1919c720c7e8SEric Dumazet inet->inet_daddr = 0; 1920c720c7e8SEric Dumazet inet->inet_dport = 0; 1921bdeab991STom Herbert sock_rps_reset_rxhash(sk); 19221da177e4SLinus Torvalds sk->sk_bound_dev_if = 0; 1923303d0403SWillem de Bruijn if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 19241da177e4SLinus Torvalds inet_reset_saddr(sk); 1925303d0403SWillem de Bruijn if (sk->sk_prot->rehash && 1926303d0403SWillem de Bruijn (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1927303d0403SWillem de Bruijn sk->sk_prot->rehash(sk); 1928303d0403SWillem de Bruijn } 19291da177e4SLinus Torvalds 19301da177e4SLinus Torvalds if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 19311da177e4SLinus Torvalds sk->sk_prot->unhash(sk); 1932c720c7e8SEric Dumazet inet->inet_sport = 0; 19331da177e4SLinus Torvalds } 19341da177e4SLinus Torvalds sk_dst_reset(sk); 19351da177e4SLinus Torvalds return 0; 19361da177e4SLinus Torvalds } 1937286c72deSEric Dumazet EXPORT_SYMBOL(__udp_disconnect); 1938286c72deSEric Dumazet 1939286c72deSEric Dumazet int udp_disconnect(struct sock *sk, int flags) 1940286c72deSEric Dumazet { 1941286c72deSEric Dumazet lock_sock(sk); 1942286c72deSEric Dumazet __udp_disconnect(sk, flags); 1943286c72deSEric Dumazet release_sock(sk); 1944286c72deSEric Dumazet return 0; 1945286c72deSEric Dumazet } 1946c482c568SEric Dumazet EXPORT_SYMBOL(udp_disconnect); 19471da177e4SLinus Torvalds 1948645ca708SEric Dumazet void udp_lib_unhash(struct sock *sk) 1949645ca708SEric Dumazet { 1950723b4610SEric Dumazet if (sk_hashed(sk)) { 195167fb4330SKuniyuki Iwashima struct udp_table *udptable = udp_get_table_prot(sk); 1952512615b6SEric Dumazet struct udp_hslot *hslot, *hslot2; 1953512615b6SEric Dumazet 1954512615b6SEric Dumazet hslot = udp_hashslot(udptable, sock_net(sk), 1955d4cada4aSEric Dumazet udp_sk(sk)->udp_port_hash); 1956512615b6SEric Dumazet hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1957645ca708SEric Dumazet 1958c8db3fecSEric Dumazet spin_lock_bh(&hslot->lock); 1959e32ea7e7SCraig Gallek if (rcu_access_pointer(sk->sk_reuseport_cb)) 1960e32ea7e7SCraig Gallek reuseport_detach_sock(sk); 1961ca065d0cSEric Dumazet if (sk_del_node_init_rcu(sk)) { 1962fdcc8aa9SEric Dumazet hslot->count--; 1963c720c7e8SEric Dumazet inet_sk(sk)->inet_num = 0; 1964645ca708SEric Dumazet sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1965512615b6SEric Dumazet 1966512615b6SEric Dumazet spin_lock(&hslot2->lock); 1967ca065d0cSEric Dumazet hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1968512615b6SEric Dumazet hslot2->count--; 1969512615b6SEric Dumazet spin_unlock(&hslot2->lock); 1970645ca708SEric Dumazet } 1971c8db3fecSEric Dumazet spin_unlock_bh(&hslot->lock); 1972645ca708SEric Dumazet } 1973723b4610SEric Dumazet } 1974645ca708SEric Dumazet EXPORT_SYMBOL(udp_lib_unhash); 1975645ca708SEric Dumazet 1976719f8358SEric Dumazet /* 1977719f8358SEric Dumazet * inet_rcv_saddr was changed, we must rehash secondary hash 1978719f8358SEric Dumazet */ 1979719f8358SEric Dumazet void udp_lib_rehash(struct sock *sk, u16 newhash) 1980719f8358SEric Dumazet { 1981719f8358SEric Dumazet if (sk_hashed(sk)) { 198267fb4330SKuniyuki Iwashima struct udp_table *udptable = udp_get_table_prot(sk); 1983719f8358SEric Dumazet struct udp_hslot *hslot, *hslot2, *nhslot2; 1984719f8358SEric Dumazet 1985719f8358SEric Dumazet hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1986719f8358SEric Dumazet nhslot2 = udp_hashslot2(udptable, newhash); 1987719f8358SEric Dumazet udp_sk(sk)->udp_portaddr_hash = newhash; 1988e32ea7e7SCraig Gallek 1989e32ea7e7SCraig Gallek if (hslot2 != nhslot2 || 1990e32ea7e7SCraig Gallek rcu_access_pointer(sk->sk_reuseport_cb)) { 1991719f8358SEric Dumazet hslot = udp_hashslot(udptable, sock_net(sk), 1992719f8358SEric Dumazet udp_sk(sk)->udp_port_hash); 1993719f8358SEric Dumazet /* we must lock primary chain too */ 1994719f8358SEric Dumazet spin_lock_bh(&hslot->lock); 1995e32ea7e7SCraig Gallek if (rcu_access_pointer(sk->sk_reuseport_cb)) 1996e32ea7e7SCraig Gallek reuseport_detach_sock(sk); 1997719f8358SEric Dumazet 1998e32ea7e7SCraig Gallek if (hslot2 != nhslot2) { 1999719f8358SEric Dumazet spin_lock(&hslot2->lock); 2000ca065d0cSEric Dumazet hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2001719f8358SEric Dumazet hslot2->count--; 2002719f8358SEric Dumazet spin_unlock(&hslot2->lock); 2003719f8358SEric Dumazet 2004719f8358SEric Dumazet spin_lock(&nhslot2->lock); 2005ca065d0cSEric Dumazet hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2006719f8358SEric Dumazet &nhslot2->head); 2007719f8358SEric Dumazet nhslot2->count++; 2008719f8358SEric Dumazet spin_unlock(&nhslot2->lock); 2009e32ea7e7SCraig Gallek } 2010719f8358SEric Dumazet 2011719f8358SEric Dumazet spin_unlock_bh(&hslot->lock); 2012719f8358SEric Dumazet } 2013719f8358SEric Dumazet } 2014719f8358SEric Dumazet } 2015719f8358SEric Dumazet EXPORT_SYMBOL(udp_lib_rehash); 2016719f8358SEric Dumazet 20178f6b5392SAlexey Kodanev void udp_v4_rehash(struct sock *sk) 2018719f8358SEric Dumazet { 2019f0b1e64cSMartin KaFai Lau u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2020719f8358SEric Dumazet inet_sk(sk)->inet_rcv_saddr, 2021719f8358SEric Dumazet inet_sk(sk)->inet_num); 2022719f8358SEric Dumazet udp_lib_rehash(sk, new_hash); 2023719f8358SEric Dumazet } 2024719f8358SEric Dumazet 2025a3f96c47SPaolo Abeni static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 202693821778SHerbert Xu { 2027fec5e652STom Herbert int rc; 202893821778SHerbert Xu 2029005ec974SShawn Bohrer if (inet_sk(sk)->inet_daddr) { 2030bdeab991STom Herbert sock_rps_save_rxhash(sk, skb); 2031005ec974SShawn Bohrer sk_mark_napi_id(sk, skb); 20322c8c56e1SEric Dumazet sk_incoming_cpu_update(sk); 2033e68b6e50SEric Dumazet } else { 2034e68b6e50SEric Dumazet sk_mark_napi_id_once(sk, skb); 2035005ec974SShawn Bohrer } 2036fec5e652STom Herbert 2037850cbaddSPaolo Abeni rc = __udp_enqueue_schedule_skb(sk, skb); 2038766e9037SEric Dumazet if (rc < 0) { 2039766e9037SEric Dumazet int is_udplite = IS_UDPLITE(sk); 204008d4c037SMenglong Dong int drop_reason; 2041766e9037SEric Dumazet 204293821778SHerbert Xu /* Note that an ENOMEM error is charged twice */ 204308d4c037SMenglong Dong if (rc == -ENOMEM) { 2044e61da9e2SEric Dumazet UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 204593821778SHerbert Xu is_udplite); 204608d4c037SMenglong Dong drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 204708d4c037SMenglong Dong } else { 2048a3ce2b10SMenglong Dong UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2049a3ce2b10SMenglong Dong is_udplite); 205008d4c037SMenglong Dong drop_reason = SKB_DROP_REASON_PROTO_MEM; 205108d4c037SMenglong Dong } 2052e61da9e2SEric Dumazet UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 205308d4c037SMenglong Dong kfree_skb_reason(skb, drop_reason); 2054296f7ea7SSatoru Moriya trace_udp_fail_queue_rcv_skb(rc, sk); 2055766e9037SEric Dumazet return -1; 205693821778SHerbert Xu } 205793821778SHerbert Xu 205893821778SHerbert Xu return 0; 205993821778SHerbert Xu } 206093821778SHerbert Xu 2061db8dac20SDavid S. Miller /* returns: 2062db8dac20SDavid S. Miller * -1: error 2063db8dac20SDavid S. Miller * 0: success 2064db8dac20SDavid S. Miller * >0: "udp encap" protocol resubmission 2065db8dac20SDavid S. Miller * 2066db8dac20SDavid S. Miller * Note that in the success and error cases, the skb is assumed to 2067db8dac20SDavid S. Miller * have either been requeued or freed. 2068db8dac20SDavid S. Miller */ 2069cf329aa4SPaolo Abeni static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2070db8dac20SDavid S. Miller { 20711379a92dSMenglong Dong int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2072db8dac20SDavid S. Miller struct udp_sock *up = udp_sk(sk); 2073db8dac20SDavid S. Miller int is_udplite = IS_UDPLITE(sk); 2074db8dac20SDavid S. Miller 2075db8dac20SDavid S. Miller /* 2076db8dac20SDavid S. Miller * Charge it to the socket, dropping if the queue is full. 2077db8dac20SDavid S. Miller */ 20781379a92dSMenglong Dong if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { 20791379a92dSMenglong Dong drop_reason = SKB_DROP_REASON_XFRM_POLICY; 2080db8dac20SDavid S. Miller goto drop; 20811379a92dSMenglong Dong } 2082895b5c9fSFlorian Westphal nf_reset_ct(skb); 2083db8dac20SDavid S. Miller 208488ab3108SDavidlohr Bueso if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 20850ad92ad0SEric Dumazet int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 20860ad92ad0SEric Dumazet 2087db8dac20SDavid S. Miller /* 2088db8dac20SDavid S. Miller * This is an encapsulation socket so pass the skb to 2089db8dac20SDavid S. Miller * the socket's udp_encap_rcv() hook. Otherwise, just 2090db8dac20SDavid S. Miller * fall through and pass this up the UDP socket. 2091db8dac20SDavid S. Miller * up->encap_rcv() returns the following value: 2092db8dac20SDavid S. Miller * =0 if skb was successfully passed to the encap 2093db8dac20SDavid S. Miller * handler or was discarded by it. 2094db8dac20SDavid S. Miller * >0 if skb should be passed on to UDP. 2095db8dac20SDavid S. Miller * <0 if skb should be resubmitted as proto -N 2096db8dac20SDavid S. Miller */ 2097db8dac20SDavid S. Miller 2098db8dac20SDavid S. Miller /* if we're overly short, let UDP handle it */ 20996aa7de05SMark Rutland encap_rcv = READ_ONCE(up->encap_rcv); 2100e5aed006SHannes Frederic Sowa if (encap_rcv) { 2101db8dac20SDavid S. Miller int ret; 2102db8dac20SDavid S. Miller 21030a80966bSTom Herbert /* Verify checksum before giving to encap */ 21040a80966bSTom Herbert if (udp_lib_checksum_complete(skb)) 21050a80966bSTom Herbert goto csum_error; 21060a80966bSTom Herbert 21070ad92ad0SEric Dumazet ret = encap_rcv(sk, skb); 2108db8dac20SDavid S. Miller if (ret <= 0) { 210902c22347SEric Dumazet __UDP_INC_STATS(sock_net(sk), 21100283328eSPavel Emelyanov UDP_MIB_INDATAGRAMS, 2111db8dac20SDavid S. Miller is_udplite); 2112db8dac20SDavid S. Miller return -ret; 2113db8dac20SDavid S. Miller } 2114db8dac20SDavid S. Miller } 2115db8dac20SDavid S. Miller 2116db8dac20SDavid S. Miller /* FALLTHROUGH -- it's a UDP Packet */ 2117db8dac20SDavid S. Miller } 2118db8dac20SDavid S. Miller 2119db8dac20SDavid S. Miller /* 2120db8dac20SDavid S. Miller * UDP-Lite specific tests, ignored on UDP sockets 2121db8dac20SDavid S. Miller */ 2122b0a42277SMiaohe Lin if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2123db8dac20SDavid S. Miller 2124db8dac20SDavid S. Miller /* 2125db8dac20SDavid S. Miller * MIB statistics other than incrementing the error count are 2126db8dac20SDavid S. Miller * disabled for the following two types of errors: these depend 2127db8dac20SDavid S. Miller * on the application settings, not on the functioning of the 2128db8dac20SDavid S. Miller * protocol stack as such. 2129db8dac20SDavid S. Miller * 2130db8dac20SDavid S. Miller * RFC 3828 here recommends (sec 3.3): "There should also be a 2131db8dac20SDavid S. Miller * way ... to ... at least let the receiving application block 2132db8dac20SDavid S. Miller * delivery of packets with coverage values less than a value 2133db8dac20SDavid S. Miller * provided by the application." 2134db8dac20SDavid S. Miller */ 2135db8dac20SDavid S. Miller if (up->pcrlen == 0) { /* full coverage was set */ 2136ba7a46f1SJoe Perches net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2137db8dac20SDavid S. Miller UDP_SKB_CB(skb)->cscov, skb->len); 2138db8dac20SDavid S. Miller goto drop; 2139db8dac20SDavid S. Miller } 2140db8dac20SDavid S. Miller /* The next case involves violating the min. coverage requested 2141db8dac20SDavid S. Miller * by the receiver. This is subtle: if receiver wants x and x is 2142db8dac20SDavid S. Miller * greater than the buffersize/MTU then receiver will complain 2143db8dac20SDavid S. Miller * that it wants x while sender emits packets of smaller size y. 2144db8dac20SDavid S. Miller * Therefore the above ...()->partial_cov statement is essential. 2145db8dac20SDavid S. Miller */ 2146db8dac20SDavid S. Miller if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2147ba7a46f1SJoe Perches net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2148db8dac20SDavid S. Miller UDP_SKB_CB(skb)->cscov, up->pcrlen); 2149db8dac20SDavid S. Miller goto drop; 2150db8dac20SDavid S. Miller } 2151db8dac20SDavid S. Miller } 2152db8dac20SDavid S. Miller 2153dd99e425SPaolo Abeni prefetch(&sk->sk_rmem_alloc); 2154ce25d66aSEric Dumazet if (rcu_access_pointer(sk->sk_filter) && 2155ce25d66aSEric Dumazet udp_lib_checksum_complete(skb)) 21566a5dc9e5SEric Dumazet goto csum_error; 2157ce25d66aSEric Dumazet 21581379a92dSMenglong Dong if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 21591379a92dSMenglong Dong drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 2160a6127697SMichal Kubeček goto drop; 21611379a92dSMenglong Dong } 2162db8dac20SDavid S. Miller 2163e6afc8acSsamanthakumar udp_csum_pull_header(skb); 2164db8dac20SDavid S. Miller 2165fbf8866dSShawn Bohrer ipv4_pktinfo_prepare(sk, skb); 2166850cbaddSPaolo Abeni return __udp_queue_rcv_skb(sk, skb); 2167db8dac20SDavid S. Miller 21686a5dc9e5SEric Dumazet csum_error: 21691379a92dSMenglong Dong drop_reason = SKB_DROP_REASON_UDP_CSUM; 217002c22347SEric Dumazet __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2171db8dac20SDavid S. Miller drop: 217202c22347SEric Dumazet __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 21738edf19c2SEric Dumazet atomic_inc(&sk->sk_drops); 21741379a92dSMenglong Dong kfree_skb_reason(skb, drop_reason); 2175db8dac20SDavid S. Miller return -1; 2176db8dac20SDavid S. Miller } 2177db8dac20SDavid S. Miller 2178cf329aa4SPaolo Abeni static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2179cf329aa4SPaolo Abeni { 2180cf329aa4SPaolo Abeni struct sk_buff *next, *segs; 2181cf329aa4SPaolo Abeni int ret; 2182cf329aa4SPaolo Abeni 2183cf329aa4SPaolo Abeni if (likely(!udp_unexpected_gso(sk, skb))) 2184cf329aa4SPaolo Abeni return udp_queue_rcv_one_skb(sk, skb); 2185cf329aa4SPaolo Abeni 2186a08e7fd9SCambda Zhu BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2187cf329aa4SPaolo Abeni __skb_push(skb, -skb_mac_offset(skb)); 2188cf329aa4SPaolo Abeni segs = udp_rcv_segment(sk, skb, true); 21891a186c14SJason A. Donenfeld skb_list_walk_safe(segs, skb, next) { 2190cf329aa4SPaolo Abeni __skb_pull(skb, skb_transport_offset(skb)); 2191000ac44dSPaolo Abeni 2192000ac44dSPaolo Abeni udp_post_segment_fix_csum(skb); 2193cf329aa4SPaolo Abeni ret = udp_queue_rcv_one_skb(sk, skb); 2194cf329aa4SPaolo Abeni if (ret > 0) 219510c678bdSXin Long ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2196cf329aa4SPaolo Abeni } 2197cf329aa4SPaolo Abeni return 0; 2198cf329aa4SPaolo Abeni } 2199cf329aa4SPaolo Abeni 220097502231SEric Dumazet /* For TCP sockets, sk_rx_dst is protected by socket lock 2201e47eb5dfSEric Dumazet * For UDP, we use xchg() to guard against concurrent changes. 220297502231SEric Dumazet */ 220364f0f5d1SPaolo Abeni bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2204421b3885SShawn Bohrer { 220597502231SEric Dumazet struct dst_entry *old; 2206421b3885SShawn Bohrer 2207d24406c8SWei Wang if (dst_hold_safe(dst)) { 22088f905c0eSEric Dumazet old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst); 220997502231SEric Dumazet dst_release(old); 221064f0f5d1SPaolo Abeni return old != dst; 221197502231SEric Dumazet } 221264f0f5d1SPaolo Abeni return false; 2213d24406c8SWei Wang } 2214c9f2c1aeSPaolo Abeni EXPORT_SYMBOL(udp_sk_rx_dst_set); 2215421b3885SShawn Bohrer 2216db8dac20SDavid S. Miller /* 2217db8dac20SDavid S. Miller * Multicasts and broadcasts go to each listener. 2218db8dac20SDavid S. Miller * 22191240d137SEric Dumazet * Note: called only from the BH handler context. 2220db8dac20SDavid S. Miller */ 2221e3163493SPavel Emelyanov static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2222db8dac20SDavid S. Miller struct udphdr *uh, 2223db8dac20SDavid S. Miller __be32 saddr, __be32 daddr, 222436cbb245SRick Jones struct udp_table *udptable, 222536cbb245SRick Jones int proto) 2226db8dac20SDavid S. Miller { 2227ca065d0cSEric Dumazet struct sock *sk, *first = NULL; 22285cf3d461SDavid Held unsigned short hnum = ntohs(uh->dest); 22295cf3d461SDavid Held struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 22302dc41cffSDavid Held unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2231ca065d0cSEric Dumazet unsigned int offset = offsetof(typeof(*sk), sk_node); 2232ca065d0cSEric Dumazet int dif = skb->dev->ifindex; 2233fb74c277SDavid Ahern int sdif = inet_sdif(skb); 2234ca065d0cSEric Dumazet struct hlist_node *node; 2235ca065d0cSEric Dumazet struct sk_buff *nskb; 22362dc41cffSDavid Held 22372dc41cffSDavid Held if (use_hash2) { 2238f0b1e64cSMartin KaFai Lau hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 223973e2d5e3SPablo Neira udptable->mask; 2240f0b1e64cSMartin KaFai Lau hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 22412dc41cffSDavid Held start_lookup: 224273e2d5e3SPablo Neira hslot = &udptable->hash2[hash2]; 22432dc41cffSDavid Held offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 22442dc41cffSDavid Held } 2245db8dac20SDavid S. Miller 2246ca065d0cSEric Dumazet sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2247ca065d0cSEric Dumazet if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2248fb74c277SDavid Ahern uh->source, saddr, dif, sdif, hnum)) 2249ca065d0cSEric Dumazet continue; 22501240d137SEric Dumazet 2251ca065d0cSEric Dumazet if (!first) { 2252ca065d0cSEric Dumazet first = sk; 2253ca065d0cSEric Dumazet continue; 2254ca065d0cSEric Dumazet } 2255ca065d0cSEric Dumazet nskb = skb_clone(skb, GFP_ATOMIC); 2256ca065d0cSEric Dumazet 2257ca065d0cSEric Dumazet if (unlikely(!nskb)) { 2258ca065d0cSEric Dumazet atomic_inc(&sk->sk_drops); 225902c22347SEric Dumazet __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2260ca065d0cSEric Dumazet IS_UDPLITE(sk)); 226102c22347SEric Dumazet __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2262ca065d0cSEric Dumazet IS_UDPLITE(sk)); 2263ca065d0cSEric Dumazet continue; 2264ca065d0cSEric Dumazet } 2265ca065d0cSEric Dumazet if (udp_queue_rcv_skb(sk, nskb) > 0) 2266ca065d0cSEric Dumazet consume_skb(nskb); 2267ca065d0cSEric Dumazet } 22681240d137SEric Dumazet 22692dc41cffSDavid Held /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 22702dc41cffSDavid Held if (use_hash2 && hash2 != hash2_any) { 22712dc41cffSDavid Held hash2 = hash2_any; 22722dc41cffSDavid Held goto start_lookup; 22732dc41cffSDavid Held } 22742dc41cffSDavid Held 2275ca065d0cSEric Dumazet if (first) { 2276ca065d0cSEric Dumazet if (udp_queue_rcv_skb(first, skb) > 0) 2277ca065d0cSEric Dumazet consume_skb(skb); 22781240d137SEric Dumazet } else { 2279ca065d0cSEric Dumazet kfree_skb(skb); 228002c22347SEric Dumazet __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 228136cbb245SRick Jones proto == IPPROTO_UDPLITE); 22821240d137SEric Dumazet } 2283db8dac20SDavid S. Miller return 0; 2284db8dac20SDavid S. Miller } 2285db8dac20SDavid S. Miller 2286db8dac20SDavid S. Miller /* Initialize UDP checksum. If exited with zero value (success), 2287db8dac20SDavid S. Miller * CHECKSUM_UNNECESSARY means, that no more checks are required. 2288666a3d6eSSu Yanjun * Otherwise, csum completion requires checksumming packet body, 2289db8dac20SDavid S. Miller * including udp header and folding it to skb->csum. 2290db8dac20SDavid S. Miller */ 2291db8dac20SDavid S. Miller static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2292db8dac20SDavid S. Miller int proto) 2293db8dac20SDavid S. Miller { 2294db8dac20SDavid S. Miller int err; 2295db8dac20SDavid S. Miller 2296db8dac20SDavid S. Miller UDP_SKB_CB(skb)->partial_cov = 0; 2297db8dac20SDavid S. Miller UDP_SKB_CB(skb)->cscov = skb->len; 2298db8dac20SDavid S. Miller 2299db8dac20SDavid S. Miller if (proto == IPPROTO_UDPLITE) { 2300db8dac20SDavid S. Miller err = udplite_checksum_init(skb, uh); 2301db8dac20SDavid S. Miller if (err) 2302db8dac20SDavid S. Miller return err; 230315f35d49SAlexey Kodanev 230415f35d49SAlexey Kodanev if (UDP_SKB_CB(skb)->partial_cov) { 230515f35d49SAlexey Kodanev skb->csum = inet_compute_pseudo(skb, proto); 230615f35d49SAlexey Kodanev return 0; 230715f35d49SAlexey Kodanev } 2308db8dac20SDavid S. Miller } 2309db8dac20SDavid S. Miller 2310b46d9f62SHannes Frederic Sowa /* Note, we are only interested in != 0 or == 0, thus the 2311b46d9f62SHannes Frederic Sowa * force to int. 2312b46d9f62SHannes Frederic Sowa */ 2313db4f1be3SSean Tranchetti err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2314ed70fcfcSTom Herbert inet_compute_pseudo); 2315db4f1be3SSean Tranchetti if (err) 2316db4f1be3SSean Tranchetti return err; 2317db4f1be3SSean Tranchetti 2318db4f1be3SSean Tranchetti if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2319db4f1be3SSean Tranchetti /* If SW calculated the value, we know it's bad */ 2320db4f1be3SSean Tranchetti if (skb->csum_complete_sw) 2321db4f1be3SSean Tranchetti return 1; 2322db4f1be3SSean Tranchetti 2323db4f1be3SSean Tranchetti /* HW says the value is bad. Let's validate that. 2324db4f1be3SSean Tranchetti * skb->csum is no longer the full packet checksum, 2325db4f1be3SSean Tranchetti * so don't treat it as such. 2326db4f1be3SSean Tranchetti */ 2327db4f1be3SSean Tranchetti skb_checksum_complete_unset(skb); 2328db4f1be3SSean Tranchetti } 2329db4f1be3SSean Tranchetti 2330db4f1be3SSean Tranchetti return 0; 2331db8dac20SDavid S. Miller } 2332db8dac20SDavid S. Miller 23332b5a9217SPaolo Abeni /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 23342b5a9217SPaolo Abeni * return code conversion for ip layer consumption 23352b5a9217SPaolo Abeni */ 23362b5a9217SPaolo Abeni static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 23372b5a9217SPaolo Abeni struct udphdr *uh) 23382b5a9217SPaolo Abeni { 23392b5a9217SPaolo Abeni int ret; 23402b5a9217SPaolo Abeni 23412b5a9217SPaolo Abeni if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2342e4aa33adSLi RongQing skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 23432b5a9217SPaolo Abeni 23442b5a9217SPaolo Abeni ret = udp_queue_rcv_skb(sk, skb); 23452b5a9217SPaolo Abeni 23462b5a9217SPaolo Abeni /* a return value > 0 means to resubmit the input, but 23472b5a9217SPaolo Abeni * it wants the return to be -protocol, or 0 23482b5a9217SPaolo Abeni */ 23492b5a9217SPaolo Abeni if (ret > 0) 23502b5a9217SPaolo Abeni return -ret; 23512b5a9217SPaolo Abeni return 0; 23522b5a9217SPaolo Abeni } 23532b5a9217SPaolo Abeni 2354db8dac20SDavid S. Miller /* 2355db8dac20SDavid S. Miller * All we need to do is get the socket, and then do a checksum. 2356db8dac20SDavid S. Miller */ 2357db8dac20SDavid S. Miller 2358645ca708SEric Dumazet int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2359db8dac20SDavid S. Miller int proto) 2360db8dac20SDavid S. Miller { 2361db8dac20SDavid S. Miller struct sock *sk; 23627b5e56f9SJesper Dangaard Brouer struct udphdr *uh; 2363db8dac20SDavid S. Miller unsigned short ulen; 2364adf30907SEric Dumazet struct rtable *rt = skb_rtable(skb); 23652783ef23SJesper Dangaard Brouer __be32 saddr, daddr; 23660283328eSPavel Emelyanov struct net *net = dev_net(skb->dev); 236771489e21SJoe Stringer bool refcounted; 23681c7fab70SMenglong Dong int drop_reason; 23691c7fab70SMenglong Dong 23701c7fab70SMenglong Dong drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2371db8dac20SDavid S. Miller 2372db8dac20SDavid S. Miller /* 2373db8dac20SDavid S. Miller * Validate the packet. 2374db8dac20SDavid S. Miller */ 2375db8dac20SDavid S. Miller if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2376db8dac20SDavid S. Miller goto drop; /* No space for header. */ 2377db8dac20SDavid S. Miller 23787b5e56f9SJesper Dangaard Brouer uh = udp_hdr(skb); 2379db8dac20SDavid S. Miller ulen = ntohs(uh->len); 2380ccc2d97cSBjørn Mork saddr = ip_hdr(skb)->saddr; 2381ccc2d97cSBjørn Mork daddr = ip_hdr(skb)->daddr; 2382ccc2d97cSBjørn Mork 2383db8dac20SDavid S. Miller if (ulen > skb->len) 2384db8dac20SDavid S. Miller goto short_packet; 2385db8dac20SDavid S. Miller 2386db8dac20SDavid S. Miller if (proto == IPPROTO_UDP) { 2387db8dac20SDavid S. Miller /* UDP validates ulen. */ 2388db8dac20SDavid S. Miller if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2389db8dac20SDavid S. Miller goto short_packet; 2390db8dac20SDavid S. Miller uh = udp_hdr(skb); 2391db8dac20SDavid S. Miller } 2392db8dac20SDavid S. Miller 2393db8dac20SDavid S. Miller if (udp4_csum_init(skb, uh, proto)) 2394db8dac20SDavid S. Miller goto csum_error; 2395db8dac20SDavid S. Miller 23969c02bec9SLorenz Bauer sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest, 23979c02bec9SLorenz Bauer &refcounted, udp_ehashfn); 23989c02bec9SLorenz Bauer if (IS_ERR(sk)) 23999c02bec9SLorenz Bauer goto no_sk; 24009c02bec9SLorenz Bauer 24018afdd99aSEric Dumazet if (sk) { 240297502231SEric Dumazet struct dst_entry *dst = skb_dst(skb); 2403421b3885SShawn Bohrer int ret; 2404421b3885SShawn Bohrer 24058f905c0eSEric Dumazet if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 240697502231SEric Dumazet udp_sk_rx_dst_set(sk, dst); 2407421b3885SShawn Bohrer 24082b5a9217SPaolo Abeni ret = udp_unicast_rcv_skb(sk, skb, uh); 240971489e21SJoe Stringer if (refcounted) 24108afdd99aSEric Dumazet sock_put(sk); 24112b5a9217SPaolo Abeni return ret; 2412c18450a5SFabian Frederick } 2413c18450a5SFabian Frederick 2414db8dac20SDavid S. Miller if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2415e3163493SPavel Emelyanov return __udp4_lib_mcast_deliver(net, skb, uh, 241636cbb245SRick Jones saddr, daddr, udptable, proto); 2417db8dac20SDavid S. Miller 2418607c4aafSKOVACS Krisztian sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 24192b5a9217SPaolo Abeni if (sk) 24202b5a9217SPaolo Abeni return udp_unicast_rcv_skb(sk, skb, uh); 24219c02bec9SLorenz Bauer no_sk: 2422db8dac20SDavid S. Miller if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2423db8dac20SDavid S. Miller goto drop; 2424895b5c9fSFlorian Westphal nf_reset_ct(skb); 2425db8dac20SDavid S. Miller 2426db8dac20SDavid S. Miller /* No socket. Drop packet silently, if checksum is wrong */ 2427db8dac20SDavid S. Miller if (udp_lib_checksum_complete(skb)) 2428db8dac20SDavid S. Miller goto csum_error; 2429db8dac20SDavid S. Miller 24301c7fab70SMenglong Dong drop_reason = SKB_DROP_REASON_NO_SOCKET; 243102c22347SEric Dumazet __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2432db8dac20SDavid S. Miller icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2433db8dac20SDavid S. Miller 2434db8dac20SDavid S. Miller /* 2435db8dac20SDavid S. Miller * Hmm. We got an UDP packet to a port to which we 2436db8dac20SDavid S. Miller * don't wanna listen. Ignore it. 2437db8dac20SDavid S. Miller */ 24381c7fab70SMenglong Dong kfree_skb_reason(skb, drop_reason); 2439db8dac20SDavid S. Miller return 0; 2440db8dac20SDavid S. Miller 2441db8dac20SDavid S. Miller short_packet: 24421c7fab70SMenglong Dong drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2443ba7a46f1SJoe Perches net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2444afd46503SJoe Perches proto == IPPROTO_UDPLITE ? "Lite" : "", 2445afd46503SJoe Perches &saddr, ntohs(uh->source), 2446afd46503SJoe Perches ulen, skb->len, 2447afd46503SJoe Perches &daddr, ntohs(uh->dest)); 2448db8dac20SDavid S. Miller goto drop; 2449db8dac20SDavid S. Miller 2450db8dac20SDavid S. Miller csum_error: 2451db8dac20SDavid S. Miller /* 2452db8dac20SDavid S. Miller * RFC1122: OK. Discards the bad packet silently (as far as 2453db8dac20SDavid S. Miller * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2454db8dac20SDavid S. Miller */ 24551c7fab70SMenglong Dong drop_reason = SKB_DROP_REASON_UDP_CSUM; 2456ba7a46f1SJoe Perches net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2457afd46503SJoe Perches proto == IPPROTO_UDPLITE ? "Lite" : "", 2458afd46503SJoe Perches &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2459db8dac20SDavid S. Miller ulen); 246002c22347SEric Dumazet __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2461db8dac20SDavid S. Miller drop: 246202c22347SEric Dumazet __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 24631c7fab70SMenglong Dong kfree_skb_reason(skb, drop_reason); 2464db8dac20SDavid S. Miller return 0; 2465db8dac20SDavid S. Miller } 2466db8dac20SDavid S. Miller 2467421b3885SShawn Bohrer /* We can only early demux multicast if there is a single matching socket. 2468421b3885SShawn Bohrer * If more than one socket found returns NULL 2469421b3885SShawn Bohrer */ 2470421b3885SShawn Bohrer static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2471421b3885SShawn Bohrer __be16 loc_port, __be32 loc_addr, 2472421b3885SShawn Bohrer __be16 rmt_port, __be32 rmt_addr, 2473fb74c277SDavid Ahern int dif, int sdif) 2474421b3885SShawn Bohrer { 2475ba6aac15SKuniyuki Iwashima struct udp_table *udptable = net->ipv4.udp_table; 2476421b3885SShawn Bohrer unsigned short hnum = ntohs(loc_port); 2477919dfa0bSKuniyuki Iwashima struct sock *sk, *result; 2478919dfa0bSKuniyuki Iwashima struct udp_hslot *hslot; 2479919dfa0bSKuniyuki Iwashima unsigned int slot; 2480919dfa0bSKuniyuki Iwashima 2481ba6aac15SKuniyuki Iwashima slot = udp_hashfn(net, hnum, udptable->mask); 2482ba6aac15SKuniyuki Iwashima hslot = &udptable->hash[slot]; 2483421b3885SShawn Bohrer 248463c6f81cSEric Dumazet /* Do not bother scanning a too big list */ 248563c6f81cSEric Dumazet if (hslot->count > 10) 248663c6f81cSEric Dumazet return NULL; 248763c6f81cSEric Dumazet 2488421b3885SShawn Bohrer result = NULL; 2489ca065d0cSEric Dumazet sk_for_each_rcu(sk, &hslot->head) { 2490ca065d0cSEric Dumazet if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2491fb74c277SDavid Ahern rmt_port, rmt_addr, dif, sdif, hnum)) { 2492ca065d0cSEric Dumazet if (result) 2493ca065d0cSEric Dumazet return NULL; 2494421b3885SShawn Bohrer result = sk; 2495421b3885SShawn Bohrer } 2496421b3885SShawn Bohrer } 2497421b3885SShawn Bohrer 2498421b3885SShawn Bohrer return result; 2499421b3885SShawn Bohrer } 2500421b3885SShawn Bohrer 2501421b3885SShawn Bohrer /* For unicast we should only early demux connected sockets or we can 2502421b3885SShawn Bohrer * break forwarding setups. The chains here can be long so only check 2503421b3885SShawn Bohrer * if the first socket is an exact match and if not move on. 2504421b3885SShawn Bohrer */ 2505421b3885SShawn Bohrer static struct sock *__udp4_lib_demux_lookup(struct net *net, 2506421b3885SShawn Bohrer __be16 loc_port, __be32 loc_addr, 2507421b3885SShawn Bohrer __be16 rmt_port, __be32 rmt_addr, 25083fa6f616SDavid Ahern int dif, int sdif) 2509421b3885SShawn Bohrer { 2510ba6aac15SKuniyuki Iwashima struct udp_table *udptable = net->ipv4.udp_table; 2511c7228317SJoe Perches INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2512919dfa0bSKuniyuki Iwashima unsigned short hnum = ntohs(loc_port); 2513919dfa0bSKuniyuki Iwashima unsigned int hash2, slot2; 2514919dfa0bSKuniyuki Iwashima struct udp_hslot *hslot2; 2515919dfa0bSKuniyuki Iwashima __portpair ports; 2516ca065d0cSEric Dumazet struct sock *sk; 2517421b3885SShawn Bohrer 2518919dfa0bSKuniyuki Iwashima hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2519ba6aac15SKuniyuki Iwashima slot2 = hash2 & udptable->mask; 2520ba6aac15SKuniyuki Iwashima hslot2 = &udptable->hash2[slot2]; 2521919dfa0bSKuniyuki Iwashima ports = INET_COMBINED_PORTS(rmt_port, hnum); 2522919dfa0bSKuniyuki Iwashima 2523ca065d0cSEric Dumazet udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2524eda090c3SEric Dumazet if (inet_match(net, sk, acookie, ports, dif, sdif)) 2525ca065d0cSEric Dumazet return sk; 2526421b3885SShawn Bohrer /* Only check first socket in chain */ 2527421b3885SShawn Bohrer break; 2528421b3885SShawn Bohrer } 2529ca065d0cSEric Dumazet return NULL; 2530421b3885SShawn Bohrer } 2531421b3885SShawn Bohrer 25327487449cSPaolo Abeni int udp_v4_early_demux(struct sk_buff *skb) 2533421b3885SShawn Bohrer { 2534610438b7SEric Dumazet struct net *net = dev_net(skb->dev); 2535bc044e8dSPaolo Abeni struct in_device *in_dev = NULL; 2536610438b7SEric Dumazet const struct iphdr *iph; 2537610438b7SEric Dumazet const struct udphdr *uh; 2538ca065d0cSEric Dumazet struct sock *sk = NULL; 2539421b3885SShawn Bohrer struct dst_entry *dst; 2540421b3885SShawn Bohrer int dif = skb->dev->ifindex; 2541fb74c277SDavid Ahern int sdif = inet_sdif(skb); 25426e540309SShawn Bohrer int ours; 2543421b3885SShawn Bohrer 2544421b3885SShawn Bohrer /* validate the packet */ 2545421b3885SShawn Bohrer if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 25467487449cSPaolo Abeni return 0; 2547421b3885SShawn Bohrer 2548610438b7SEric Dumazet iph = ip_hdr(skb); 2549610438b7SEric Dumazet uh = udp_hdr(skb); 2550610438b7SEric Dumazet 2551996b44fcSPaolo Abeni if (skb->pkt_type == PACKET_MULTICAST) { 2552bc044e8dSPaolo Abeni in_dev = __in_dev_get_rcu(skb->dev); 25536e540309SShawn Bohrer 25546e540309SShawn Bohrer if (!in_dev) 25557487449cSPaolo Abeni return 0; 25566e540309SShawn Bohrer 25576e540309SShawn Bohrer ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 25586e540309SShawn Bohrer iph->protocol); 25596e540309SShawn Bohrer if (!ours) 25607487449cSPaolo Abeni return 0; 2561ad0ea198SPaolo Abeni 2562421b3885SShawn Bohrer sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2563fb74c277SDavid Ahern uh->source, iph->saddr, 2564fb74c277SDavid Ahern dif, sdif); 25656e540309SShawn Bohrer } else if (skb->pkt_type == PACKET_HOST) { 2566421b3885SShawn Bohrer sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 25673fa6f616SDavid Ahern uh->source, iph->saddr, dif, sdif); 25686e540309SShawn Bohrer } 2569421b3885SShawn Bohrer 257041c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 25717487449cSPaolo Abeni return 0; 2572421b3885SShawn Bohrer 2573421b3885SShawn Bohrer skb->sk = sk; 257482eabd9eSAlexander Duyck skb->destructor = sock_efree; 25758f905c0eSEric Dumazet dst = rcu_dereference(sk->sk_rx_dst); 2576421b3885SShawn Bohrer 2577421b3885SShawn Bohrer if (dst) 2578421b3885SShawn Bohrer dst = dst_check(dst, 0); 257910e2eb87SEric Dumazet if (dst) { 2580bc044e8dSPaolo Abeni u32 itag = 0; 2581bc044e8dSPaolo Abeni 2582d24406c8SWei Wang /* set noref for now. 2583d24406c8SWei Wang * any place which wants to hold dst has to call 2584d24406c8SWei Wang * dst_hold_safe() 2585d24406c8SWei Wang */ 2586421b3885SShawn Bohrer skb_dst_set_noref(skb, dst); 2587bc044e8dSPaolo Abeni 2588bc044e8dSPaolo Abeni /* for unconnected multicast sockets we need to validate 2589bc044e8dSPaolo Abeni * the source on each packet 2590bc044e8dSPaolo Abeni */ 2591bc044e8dSPaolo Abeni if (!inet_sk(sk)->inet_daddr && in_dev) 2592bc044e8dSPaolo Abeni return ip_mc_validate_source(skb, iph->daddr, 25938d2b51b0SGuillaume Nault iph->saddr, 25948d2b51b0SGuillaume Nault iph->tos & IPTOS_RT_MASK, 2595bc044e8dSPaolo Abeni skb->dev, in_dev, &itag); 2596421b3885SShawn Bohrer } 25977487449cSPaolo Abeni return 0; 259810e2eb87SEric Dumazet } 2599421b3885SShawn Bohrer 2600db8dac20SDavid S. Miller int udp_rcv(struct sk_buff *skb) 2601db8dac20SDavid S. Miller { 2602ba6aac15SKuniyuki Iwashima return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 2603db8dac20SDavid S. Miller } 2604db8dac20SDavid S. Miller 26057d06b2e0SBrian Haley void udp_destroy_sock(struct sock *sk) 2606db8dac20SDavid S. Miller { 260744046a59STom Parkin struct udp_sock *up = udp_sk(sk); 26088a74ad60SEric Dumazet bool slow = lock_sock_fast(sk); 2609a8b897c7SPaolo Abeni 2610a8b897c7SPaolo Abeni /* protects from races with udp_abort() */ 2611a8b897c7SPaolo Abeni sock_set_flag(sk, SOCK_DEAD); 2612db8dac20SDavid S. Miller udp_flush_pending_frames(sk); 26138a74ad60SEric Dumazet unlock_sock_fast(sk, slow); 261460fb9567SPaolo Abeni if (static_branch_unlikely(&udp_encap_needed_key)) { 261560fb9567SPaolo Abeni if (up->encap_type) { 261644046a59STom Parkin void (*encap_destroy)(struct sock *sk); 26176aa7de05SMark Rutland encap_destroy = READ_ONCE(up->encap_destroy); 261844046a59STom Parkin if (encap_destroy) 261944046a59STom Parkin encap_destroy(sk); 262044046a59STom Parkin } 262160fb9567SPaolo Abeni if (up->encap_enabled) 26229c480601SPaolo Abeni static_branch_dec(&udp_encap_needed_key); 262360fb9567SPaolo Abeni } 2624db8dac20SDavid S. Miller } 2625db8dac20SDavid S. Miller 26261da177e4SLinus Torvalds /* 26271da177e4SLinus Torvalds * Socket option code for UDP 26281da177e4SLinus Torvalds */ 26294c0a6cb0SGerrit Renker int udp_lib_setsockopt(struct sock *sk, int level, int optname, 263091ac1ccaSChristoph Hellwig sockptr_t optval, unsigned int optlen, 26314c0a6cb0SGerrit Renker int (*push_pending_frames)(struct sock *)) 26321da177e4SLinus Torvalds { 26331da177e4SLinus Torvalds struct udp_sock *up = udp_sk(sk); 26341c19448cSTom Herbert int val, valbool; 26351da177e4SLinus Torvalds int err = 0; 2636b2bf1e26SWang Chen int is_udplite = IS_UDPLITE(sk); 26371da177e4SLinus Torvalds 26388a3854c7SPaolo Abeni if (level == SOL_SOCKET) { 26398a3854c7SPaolo Abeni err = sk_setsockopt(sk, level, optname, optval, optlen); 26408a3854c7SPaolo Abeni 26418a3854c7SPaolo Abeni if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) { 26428a3854c7SPaolo Abeni sockopt_lock_sock(sk); 26438a3854c7SPaolo Abeni /* paired with READ_ONCE in udp_rmem_release() */ 26448a3854c7SPaolo Abeni WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2); 26458a3854c7SPaolo Abeni sockopt_release_sock(sk); 26468a3854c7SPaolo Abeni } 26478a3854c7SPaolo Abeni return err; 26488a3854c7SPaolo Abeni } 26498a3854c7SPaolo Abeni 26501da177e4SLinus Torvalds if (optlen < sizeof(int)) 26511da177e4SLinus Torvalds return -EINVAL; 26521da177e4SLinus Torvalds 265391ac1ccaSChristoph Hellwig if (copy_from_sockptr(&val, optval, sizeof(val))) 26541da177e4SLinus Torvalds return -EFAULT; 26551da177e4SLinus Torvalds 26561c19448cSTom Herbert valbool = val ? 1 : 0; 26571c19448cSTom Herbert 26581da177e4SLinus Torvalds switch (optname) { 26591da177e4SLinus Torvalds case UDP_CORK: 26601da177e4SLinus Torvalds if (val != 0) { 2661a9f59707SEric Dumazet WRITE_ONCE(up->corkflag, 1); 26621da177e4SLinus Torvalds } else { 2663a9f59707SEric Dumazet WRITE_ONCE(up->corkflag, 0); 26641da177e4SLinus Torvalds lock_sock(sk); 26654243cdc2SJoe Perches push_pending_frames(sk); 26661da177e4SLinus Torvalds release_sock(sk); 26671da177e4SLinus Torvalds } 26681da177e4SLinus Torvalds break; 26691da177e4SLinus Torvalds 26701da177e4SLinus Torvalds case UDP_ENCAP: 26711da177e4SLinus Torvalds switch (val) { 26721da177e4SLinus Torvalds case 0: 2673fd1ac07fSAlexey Dobriyan #ifdef CONFIG_XFRM 26741da177e4SLinus Torvalds case UDP_ENCAP_ESPINUDP: 26751da177e4SLinus Torvalds case UDP_ENCAP_ESPINUDP_NON_IKE: 26760146dca7SSabrina Dubroca #if IS_ENABLED(CONFIG_IPV6) 26770146dca7SSabrina Dubroca if (sk->sk_family == AF_INET6) 26780146dca7SSabrina Dubroca up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv; 26790146dca7SSabrina Dubroca else 26800146dca7SSabrina Dubroca #endif 2681067b207bSJames Chapman up->encap_rcv = xfrm4_udp_encap_rcv; 2682fd1ac07fSAlexey Dobriyan #endif 2683a8eceea8SJoe Perches fallthrough; 2684342f0234SJames Chapman case UDP_ENCAP_L2TPINUDP: 26851da177e4SLinus Torvalds up->encap_type = val; 268660fb9567SPaolo Abeni lock_sock(sk); 268760fb9567SPaolo Abeni udp_tunnel_encap_enable(sk->sk_socket); 268860fb9567SPaolo Abeni release_sock(sk); 26891da177e4SLinus Torvalds break; 26901da177e4SLinus Torvalds default: 26911da177e4SLinus Torvalds err = -ENOPROTOOPT; 26921da177e4SLinus Torvalds break; 26931da177e4SLinus Torvalds } 26941da177e4SLinus Torvalds break; 26951da177e4SLinus Torvalds 26961c19448cSTom Herbert case UDP_NO_CHECK6_TX: 26971c19448cSTom Herbert up->no_check6_tx = valbool; 26981c19448cSTom Herbert break; 26991c19448cSTom Herbert 27001c19448cSTom Herbert case UDP_NO_CHECK6_RX: 27011c19448cSTom Herbert up->no_check6_rx = valbool; 27021c19448cSTom Herbert break; 27031c19448cSTom Herbert 2704bec1f6f6SWillem de Bruijn case UDP_SEGMENT: 2705bec1f6f6SWillem de Bruijn if (val < 0 || val > USHRT_MAX) 2706bec1f6f6SWillem de Bruijn return -EINVAL; 270718a419baSEric Dumazet WRITE_ONCE(up->gso_size, val); 2708bec1f6f6SWillem de Bruijn break; 2709bec1f6f6SWillem de Bruijn 2710e20cf8d3SPaolo Abeni case UDP_GRO: 2711e20cf8d3SPaolo Abeni lock_sock(sk); 271278352f73SPaolo Abeni 271378352f73SPaolo Abeni /* when enabling GRO, accept the related GSO packet type */ 2714e20cf8d3SPaolo Abeni if (valbool) 2715e20cf8d3SPaolo Abeni udp_tunnel_encap_enable(sk->sk_socket); 2716e20cf8d3SPaolo Abeni up->gro_enabled = valbool; 271778352f73SPaolo Abeni up->accept_udp_l4 = valbool; 2718e20cf8d3SPaolo Abeni release_sock(sk); 2719e20cf8d3SPaolo Abeni break; 2720e20cf8d3SPaolo Abeni 2721ba4e58ecSGerrit Renker /* 2722ba4e58ecSGerrit Renker * UDP-Lite's partial checksum coverage (RFC 3828). 2723ba4e58ecSGerrit Renker */ 2724ba4e58ecSGerrit Renker /* The sender sets actual checksum coverage length via this option. 2725ba4e58ecSGerrit Renker * The case coverage > packet length is handled by send module. */ 2726ba4e58ecSGerrit Renker case UDPLITE_SEND_CSCOV: 2727b2bf1e26SWang Chen if (!is_udplite) /* Disable the option on UDP sockets */ 2728ba4e58ecSGerrit Renker return -ENOPROTOOPT; 2729ba4e58ecSGerrit Renker if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2730ba4e58ecSGerrit Renker val = 8; 27314be929beSAlexey Dobriyan else if (val > USHRT_MAX) 27324be929beSAlexey Dobriyan val = USHRT_MAX; 2733ba4e58ecSGerrit Renker up->pcslen = val; 2734ba4e58ecSGerrit Renker up->pcflag |= UDPLITE_SEND_CC; 2735ba4e58ecSGerrit Renker break; 2736ba4e58ecSGerrit Renker 2737ba4e58ecSGerrit Renker /* The receiver specifies a minimum checksum coverage value. To make 2738ba4e58ecSGerrit Renker * sense, this should be set to at least 8 (as done below). If zero is 2739ba4e58ecSGerrit Renker * used, this again means full checksum coverage. */ 2740ba4e58ecSGerrit Renker case UDPLITE_RECV_CSCOV: 2741b2bf1e26SWang Chen if (!is_udplite) /* Disable the option on UDP sockets */ 2742ba4e58ecSGerrit Renker return -ENOPROTOOPT; 2743ba4e58ecSGerrit Renker if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2744ba4e58ecSGerrit Renker val = 8; 27454be929beSAlexey Dobriyan else if (val > USHRT_MAX) 27464be929beSAlexey Dobriyan val = USHRT_MAX; 2747ba4e58ecSGerrit Renker up->pcrlen = val; 2748ba4e58ecSGerrit Renker up->pcflag |= UDPLITE_RECV_CC; 2749ba4e58ecSGerrit Renker break; 2750ba4e58ecSGerrit Renker 27511da177e4SLinus Torvalds default: 27521da177e4SLinus Torvalds err = -ENOPROTOOPT; 27531da177e4SLinus Torvalds break; 27546516c655SStephen Hemminger } 27551da177e4SLinus Torvalds 27561da177e4SLinus Torvalds return err; 27571da177e4SLinus Torvalds } 2758c482c568SEric Dumazet EXPORT_SYMBOL(udp_lib_setsockopt); 27591da177e4SLinus Torvalds 2760a7b75c5aSChristoph Hellwig int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2761a7b75c5aSChristoph Hellwig unsigned int optlen) 2762db8dac20SDavid S. Miller { 27638a3854c7SPaolo Abeni if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 276491ac1ccaSChristoph Hellwig return udp_lib_setsockopt(sk, level, optname, 2765a7b75c5aSChristoph Hellwig optval, optlen, 2766db8dac20SDavid S. Miller udp_push_pending_frames); 2767db8dac20SDavid S. Miller return ip_setsockopt(sk, level, optname, optval, optlen); 2768db8dac20SDavid S. Miller } 2769db8dac20SDavid S. Miller 27704c0a6cb0SGerrit Renker int udp_lib_getsockopt(struct sock *sk, int level, int optname, 27711da177e4SLinus Torvalds char __user *optval, int __user *optlen) 27721da177e4SLinus Torvalds { 27731da177e4SLinus Torvalds struct udp_sock *up = udp_sk(sk); 27741da177e4SLinus Torvalds int val, len; 27751da177e4SLinus Torvalds 27761da177e4SLinus Torvalds if (get_user(len, optlen)) 27771da177e4SLinus Torvalds return -EFAULT; 27781da177e4SLinus Torvalds 27791da177e4SLinus Torvalds len = min_t(unsigned int, len, sizeof(int)); 27801da177e4SLinus Torvalds 27811da177e4SLinus Torvalds if (len < 0) 27821da177e4SLinus Torvalds return -EINVAL; 27831da177e4SLinus Torvalds 27841da177e4SLinus Torvalds switch (optname) { 27851da177e4SLinus Torvalds case UDP_CORK: 2786a9f59707SEric Dumazet val = READ_ONCE(up->corkflag); 27871da177e4SLinus Torvalds break; 27881da177e4SLinus Torvalds 27891da177e4SLinus Torvalds case UDP_ENCAP: 27901da177e4SLinus Torvalds val = up->encap_type; 27911da177e4SLinus Torvalds break; 27921da177e4SLinus Torvalds 27931c19448cSTom Herbert case UDP_NO_CHECK6_TX: 27941c19448cSTom Herbert val = up->no_check6_tx; 27951c19448cSTom Herbert break; 27961c19448cSTom Herbert 27971c19448cSTom Herbert case UDP_NO_CHECK6_RX: 27981c19448cSTom Herbert val = up->no_check6_rx; 27991c19448cSTom Herbert break; 28001c19448cSTom Herbert 2801bec1f6f6SWillem de Bruijn case UDP_SEGMENT: 280218a419baSEric Dumazet val = READ_ONCE(up->gso_size); 2803bec1f6f6SWillem de Bruijn break; 2804bec1f6f6SWillem de Bruijn 280598184612SNorman Maurer case UDP_GRO: 280698184612SNorman Maurer val = up->gro_enabled; 280798184612SNorman Maurer break; 280898184612SNorman Maurer 2809ba4e58ecSGerrit Renker /* The following two cannot be changed on UDP sockets, the return is 2810ba4e58ecSGerrit Renker * always 0 (which corresponds to the full checksum coverage of UDP). */ 2811ba4e58ecSGerrit Renker case UDPLITE_SEND_CSCOV: 2812ba4e58ecSGerrit Renker val = up->pcslen; 2813ba4e58ecSGerrit Renker break; 2814ba4e58ecSGerrit Renker 2815ba4e58ecSGerrit Renker case UDPLITE_RECV_CSCOV: 2816ba4e58ecSGerrit Renker val = up->pcrlen; 2817ba4e58ecSGerrit Renker break; 2818ba4e58ecSGerrit Renker 28191da177e4SLinus Torvalds default: 28201da177e4SLinus Torvalds return -ENOPROTOOPT; 28216516c655SStephen Hemminger } 28221da177e4SLinus Torvalds 28231da177e4SLinus Torvalds if (put_user(len, optlen)) 28241da177e4SLinus Torvalds return -EFAULT; 28251da177e4SLinus Torvalds if (copy_to_user(optval, &val, len)) 28261da177e4SLinus Torvalds return -EFAULT; 28271da177e4SLinus Torvalds return 0; 28281da177e4SLinus Torvalds } 2829c482c568SEric Dumazet EXPORT_SYMBOL(udp_lib_getsockopt); 28301da177e4SLinus Torvalds 2831db8dac20SDavid S. Miller int udp_getsockopt(struct sock *sk, int level, int optname, 2832db8dac20SDavid S. Miller char __user *optval, int __user *optlen) 2833db8dac20SDavid S. Miller { 2834db8dac20SDavid S. Miller if (level == SOL_UDP || level == SOL_UDPLITE) 2835db8dac20SDavid S. Miller return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2836db8dac20SDavid S. Miller return ip_getsockopt(sk, level, optname, optval, optlen); 2837db8dac20SDavid S. Miller } 2838db8dac20SDavid S. Miller 28391da177e4SLinus Torvalds /** 28401da177e4SLinus Torvalds * udp_poll - wait for a UDP event. 28413628e3cbSAndrew Lunn * @file: - file struct 28423628e3cbSAndrew Lunn * @sock: - socket 28433628e3cbSAndrew Lunn * @wait: - poll table 28441da177e4SLinus Torvalds * 28451da177e4SLinus Torvalds * This is same as datagram poll, except for the special case of 28461da177e4SLinus Torvalds * blocking sockets. If application is using a blocking fd 28471da177e4SLinus Torvalds * and a packet with checksum error is in the queue; 28481da177e4SLinus Torvalds * then it could get return from select indicating data available 28491da177e4SLinus Torvalds * but then block when reading it. Add special case code 28501da177e4SLinus Torvalds * to work around these arguably broken applications. 28511da177e4SLinus Torvalds */ 2852a11e1d43SLinus Torvalds __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 28531da177e4SLinus Torvalds { 2854a11e1d43SLinus Torvalds __poll_t mask = datagram_poll(file, sock, wait); 28551da177e4SLinus Torvalds struct sock *sk = sock->sk; 28561da177e4SLinus Torvalds 28573ef7cf57SEric Dumazet if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2858a9a08845SLinus Torvalds mask |= EPOLLIN | EPOLLRDNORM; 28592276f58aSPaolo Abeni 28601da177e4SLinus Torvalds /* Check for false positives due to checksum errors */ 2861a11e1d43SLinus Torvalds if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2862e83c6744SEric Dumazet !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2863a9a08845SLinus Torvalds mask &= ~(EPOLLIN | EPOLLRDNORM); 28641da177e4SLinus Torvalds 2865af493388SCong Wang /* psock ingress_msg queue should not contain any bad checksum frames */ 2866af493388SCong Wang if (sk_is_readable(sk)) 2867af493388SCong Wang mask |= EPOLLIN | EPOLLRDNORM; 28681da177e4SLinus Torvalds return mask; 28691da177e4SLinus Torvalds 28701da177e4SLinus Torvalds } 2871a11e1d43SLinus Torvalds EXPORT_SYMBOL(udp_poll); 28721da177e4SLinus Torvalds 28735d77dca8SDavid Ahern int udp_abort(struct sock *sk, int err) 28745d77dca8SDavid Ahern { 28754ddbcb88SAditi Ghag if (!has_current_bpf_ctx()) 28765d77dca8SDavid Ahern lock_sock(sk); 28775d77dca8SDavid Ahern 2878a8b897c7SPaolo Abeni /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2879a8b897c7SPaolo Abeni * with close() 2880a8b897c7SPaolo Abeni */ 2881a8b897c7SPaolo Abeni if (sock_flag(sk, SOCK_DEAD)) 2882a8b897c7SPaolo Abeni goto out; 2883a8b897c7SPaolo Abeni 28845d77dca8SDavid Ahern sk->sk_err = err; 2885e3ae2365SAlexander Aring sk_error_report(sk); 2886286c72deSEric Dumazet __udp_disconnect(sk, 0); 28875d77dca8SDavid Ahern 2888a8b897c7SPaolo Abeni out: 28894ddbcb88SAditi Ghag if (!has_current_bpf_ctx()) 28905d77dca8SDavid Ahern release_sock(sk); 28915d77dca8SDavid Ahern 28925d77dca8SDavid Ahern return 0; 28935d77dca8SDavid Ahern } 28945d77dca8SDavid Ahern EXPORT_SYMBOL_GPL(udp_abort); 28955d77dca8SDavid Ahern 2896db8dac20SDavid S. Miller struct proto udp_prot = { 2897db8dac20SDavid S. Miller .name = "UDP", 2898db8dac20SDavid S. Miller .owner = THIS_MODULE, 2899db8dac20SDavid S. Miller .close = udp_lib_close, 2900d74bad4eSAndrey Ignatov .pre_connect = udp_pre_connect, 2901db8dac20SDavid S. Miller .connect = ip4_datagram_connect, 2902db8dac20SDavid S. Miller .disconnect = udp_disconnect, 2903db8dac20SDavid S. Miller .ioctl = udp_ioctl, 2904850cbaddSPaolo Abeni .init = udp_init_sock, 2905db8dac20SDavid S. Miller .destroy = udp_destroy_sock, 2906db8dac20SDavid S. Miller .setsockopt = udp_setsockopt, 2907db8dac20SDavid S. Miller .getsockopt = udp_getsockopt, 2908db8dac20SDavid S. Miller .sendmsg = udp_sendmsg, 2909db8dac20SDavid S. Miller .recvmsg = udp_recvmsg, 29101d7e4538SDavid Howells .splice_eof = udp_splice_eof, 29118141ed9fSSteffen Klassert .release_cb = ip4_datagram_release_cb, 2912db8dac20SDavid S. Miller .hash = udp_lib_hash, 2913db8dac20SDavid S. Miller .unhash = udp_lib_unhash, 2914719f8358SEric Dumazet .rehash = udp_v4_rehash, 2915db8dac20SDavid S. Miller .get_port = udp_v4_get_port, 291691a760b2SMenglong Dong .put_port = udp_lib_unhash, 29178a59f9d1SCong Wang #ifdef CONFIG_BPF_SYSCALL 29188a59f9d1SCong Wang .psock_update_sk_prot = udp_bpf_update_proto, 29198a59f9d1SCong Wang #endif 2920db8dac20SDavid S. Miller .memory_allocated = &udp_memory_allocated, 29210defbb0aSEric Dumazet .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 29220defbb0aSEric Dumazet 2923db8dac20SDavid S. Miller .sysctl_mem = sysctl_udp_mem, 29241e802951STonghao Zhang .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 29251e802951STonghao Zhang .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2926db8dac20SDavid S. Miller .obj_size = sizeof(struct udp_sock), 292767fb4330SKuniyuki Iwashima .h.udp_table = NULL, 29285d77dca8SDavid Ahern .diag_destroy = udp_abort, 2929db8dac20SDavid S. Miller }; 2930c482c568SEric Dumazet EXPORT_SYMBOL(udp_prot); 29311da177e4SLinus Torvalds 29321da177e4SLinus Torvalds /* ------------------------------------------------------------------------ */ 29331da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 29341da177e4SLinus Torvalds 2935f44b1c51SAditi Ghag static unsigned short seq_file_family(const struct seq_file *seq); 2936f44b1c51SAditi Ghag static bool seq_sk_match(struct seq_file *seq, const struct sock *sk) 2937f44b1c51SAditi Ghag { 2938f44b1c51SAditi Ghag unsigned short family = seq_file_family(seq); 2939f44b1c51SAditi Ghag 2940f44b1c51SAditi Ghag /* AF_UNSPEC is used as a match all */ 2941f44b1c51SAditi Ghag return ((family == AF_UNSPEC || family == sk->sk_family) && 2942f44b1c51SAditi Ghag net_eq(sock_net(sk), seq_file_net(seq))); 2943f44b1c51SAditi Ghag } 2944f44b1c51SAditi Ghag 2945e4fe1bf1SAditi Ghag #ifdef CONFIG_BPF_SYSCALL 2946e4fe1bf1SAditi Ghag static const struct seq_operations bpf_iter_udp_seq_ops; 2947e4fe1bf1SAditi Ghag #endif 29487625d2e9SAditi Ghag static struct udp_table *udp_get_table_seq(struct seq_file *seq, 2949478aee5dSKuniyuki Iwashima struct net *net) 2950478aee5dSKuniyuki Iwashima { 29517625d2e9SAditi Ghag const struct udp_seq_afinfo *afinfo; 29527625d2e9SAditi Ghag 2953e4fe1bf1SAditi Ghag #ifdef CONFIG_BPF_SYSCALL 2954e4fe1bf1SAditi Ghag if (seq->op == &bpf_iter_udp_seq_ops) 29557625d2e9SAditi Ghag return net->ipv4.udp_table; 2956e4fe1bf1SAditi Ghag #endif 29577625d2e9SAditi Ghag 29587625d2e9SAditi Ghag afinfo = pde_data(file_inode(seq->file)); 2959478aee5dSKuniyuki Iwashima return afinfo->udp_table ? : net->ipv4.udp_table; 2960478aee5dSKuniyuki Iwashima } 2961478aee5dSKuniyuki Iwashima 2962645ca708SEric Dumazet static struct sock *udp_get_first(struct seq_file *seq, int start) 29631da177e4SLinus Torvalds { 29641da177e4SLinus Torvalds struct udp_iter_state *state = seq->private; 29656f191efeSDenis V. Lunev struct net *net = seq_file_net(seq); 2966478aee5dSKuniyuki Iwashima struct udp_table *udptable; 2967919dfa0bSKuniyuki Iwashima struct sock *sk; 29681da177e4SLinus Torvalds 29697625d2e9SAditi Ghag udptable = udp_get_table_seq(seq, net); 2970478aee5dSKuniyuki Iwashima 2971478aee5dSKuniyuki Iwashima for (state->bucket = start; state->bucket <= udptable->mask; 2972f86dcc5aSEric Dumazet ++state->bucket) { 2973478aee5dSKuniyuki Iwashima struct udp_hslot *hslot = &udptable->hash[state->bucket]; 2974f86dcc5aSEric Dumazet 2975ca065d0cSEric Dumazet if (hlist_empty(&hslot->head)) 2976f86dcc5aSEric Dumazet continue; 2977f86dcc5aSEric Dumazet 2978645ca708SEric Dumazet spin_lock_bh(&hslot->lock); 2979ca065d0cSEric Dumazet sk_for_each(sk, &hslot->head) { 2980f44b1c51SAditi Ghag if (seq_sk_match(seq, sk)) 29811da177e4SLinus Torvalds goto found; 29821da177e4SLinus Torvalds } 2983645ca708SEric Dumazet spin_unlock_bh(&hslot->lock); 29841da177e4SLinus Torvalds } 29851da177e4SLinus Torvalds sk = NULL; 29861da177e4SLinus Torvalds found: 29871da177e4SLinus Torvalds return sk; 29881da177e4SLinus Torvalds } 29891da177e4SLinus Torvalds 29901da177e4SLinus Torvalds static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 29911da177e4SLinus Torvalds { 29921da177e4SLinus Torvalds struct udp_iter_state *state = seq->private; 29936f191efeSDenis V. Lunev struct net *net = seq_file_net(seq); 2994478aee5dSKuniyuki Iwashima struct udp_table *udptable; 29951da177e4SLinus Torvalds 29961da177e4SLinus Torvalds do { 2997ca065d0cSEric Dumazet sk = sk_next(sk); 2998f44b1c51SAditi Ghag } while (sk && !seq_sk_match(seq, sk)); 29991da177e4SLinus Torvalds 3000645ca708SEric Dumazet if (!sk) { 30017625d2e9SAditi Ghag udptable = udp_get_table_seq(seq, net); 3002478aee5dSKuniyuki Iwashima 3003478aee5dSKuniyuki Iwashima if (state->bucket <= udptable->mask) 3004478aee5dSKuniyuki Iwashima spin_unlock_bh(&udptable->hash[state->bucket].lock); 3005478aee5dSKuniyuki Iwashima 3006645ca708SEric Dumazet return udp_get_first(seq, state->bucket + 1); 30071da177e4SLinus Torvalds } 30081da177e4SLinus Torvalds return sk; 30091da177e4SLinus Torvalds } 30101da177e4SLinus Torvalds 30111da177e4SLinus Torvalds static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 30121da177e4SLinus Torvalds { 3013645ca708SEric Dumazet struct sock *sk = udp_get_first(seq, 0); 30141da177e4SLinus Torvalds 30151da177e4SLinus Torvalds if (sk) 30161da177e4SLinus Torvalds while (pos && (sk = udp_get_next(seq, sk)) != NULL) 30171da177e4SLinus Torvalds --pos; 30181da177e4SLinus Torvalds return pos ? NULL : sk; 30191da177e4SLinus Torvalds } 30201da177e4SLinus Torvalds 3021a3d2599bSChristoph Hellwig void *udp_seq_start(struct seq_file *seq, loff_t *pos) 30221da177e4SLinus Torvalds { 302330842f29SVitaly Mayatskikh struct udp_iter_state *state = seq->private; 3024f86dcc5aSEric Dumazet state->bucket = MAX_UDP_PORTS; 302530842f29SVitaly Mayatskikh 3026b50660f1SYOSHIFUJI Hideaki return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 30271da177e4SLinus Torvalds } 3028a3d2599bSChristoph Hellwig EXPORT_SYMBOL(udp_seq_start); 30291da177e4SLinus Torvalds 3030a3d2599bSChristoph Hellwig void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 30311da177e4SLinus Torvalds { 30321da177e4SLinus Torvalds struct sock *sk; 30331da177e4SLinus Torvalds 3034b50660f1SYOSHIFUJI Hideaki if (v == SEQ_START_TOKEN) 30351da177e4SLinus Torvalds sk = udp_get_idx(seq, 0); 30361da177e4SLinus Torvalds else 30371da177e4SLinus Torvalds sk = udp_get_next(seq, v); 30381da177e4SLinus Torvalds 30391da177e4SLinus Torvalds ++*pos; 30401da177e4SLinus Torvalds return sk; 30411da177e4SLinus Torvalds } 3042a3d2599bSChristoph Hellwig EXPORT_SYMBOL(udp_seq_next); 30431da177e4SLinus Torvalds 3044a3d2599bSChristoph Hellwig void udp_seq_stop(struct seq_file *seq, void *v) 30451da177e4SLinus Torvalds { 3046645ca708SEric Dumazet struct udp_iter_state *state = seq->private; 3047478aee5dSKuniyuki Iwashima struct udp_table *udptable; 3048645ca708SEric Dumazet 30497625d2e9SAditi Ghag udptable = udp_get_table_seq(seq, seq_file_net(seq)); 3050478aee5dSKuniyuki Iwashima 3051478aee5dSKuniyuki Iwashima if (state->bucket <= udptable->mask) 3052478aee5dSKuniyuki Iwashima spin_unlock_bh(&udptable->hash[state->bucket].lock); 30531da177e4SLinus Torvalds } 3054a3d2599bSChristoph Hellwig EXPORT_SYMBOL(udp_seq_stop); 3055db8dac20SDavid S. Miller 3056db8dac20SDavid S. Miller /* ------------------------------------------------------------------------ */ 30575e659e4cSPavel Emelyanov static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3058652586dfSTetsuo Handa int bucket) 3059db8dac20SDavid S. Miller { 3060db8dac20SDavid S. Miller struct inet_sock *inet = inet_sk(sp); 3061c720c7e8SEric Dumazet __be32 dest = inet->inet_daddr; 3062c720c7e8SEric Dumazet __be32 src = inet->inet_rcv_saddr; 3063c720c7e8SEric Dumazet __u16 destp = ntohs(inet->inet_dport); 3064c720c7e8SEric Dumazet __u16 srcp = ntohs(inet->inet_sport); 3065db8dac20SDavid S. Miller 3066f86dcc5aSEric Dumazet seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3067ea9a0379SPatrick Talbert " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3068db8dac20SDavid S. Miller bucket, src, srcp, dest, destp, sp->sk_state, 306931e6d363SEric Dumazet sk_wmem_alloc_get(sp), 30706c206b20SPaolo Abeni udp_rqueue_get(sp), 3071a7cb5a49SEric W. Biederman 0, 0L, 0, 3072a7cb5a49SEric W. Biederman from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3073a7cb5a49SEric W. Biederman 0, sock_i_ino(sp), 307441c6d650SReshetova, Elena refcount_read(&sp->sk_refcnt), sp, 3075652586dfSTetsuo Handa atomic_read(&sp->sk_drops)); 3076db8dac20SDavid S. Miller } 3077db8dac20SDavid S. Miller 3078db8dac20SDavid S. Miller int udp4_seq_show(struct seq_file *seq, void *v) 3079db8dac20SDavid S. Miller { 3080652586dfSTetsuo Handa seq_setwidth(seq, 127); 3081db8dac20SDavid S. Miller if (v == SEQ_START_TOKEN) 3082652586dfSTetsuo Handa seq_puts(seq, " sl local_address rem_address st tx_queue " 3083db8dac20SDavid S. Miller "rx_queue tr tm->when retrnsmt uid timeout " 3084cb61cb9bSEric Dumazet "inode ref pointer drops"); 3085db8dac20SDavid S. Miller else { 3086db8dac20SDavid S. Miller struct udp_iter_state *state = seq->private; 3087db8dac20SDavid S. Miller 3088652586dfSTetsuo Handa udp4_format_sock(v, seq, state->bucket); 3089db8dac20SDavid S. Miller } 3090652586dfSTetsuo Handa seq_pad(seq, '\n'); 3091db8dac20SDavid S. Miller return 0; 3092db8dac20SDavid S. Miller } 3093db8dac20SDavid S. Miller 30945788b3a0SYonghong Song #ifdef CONFIG_BPF_SYSCALL 30955788b3a0SYonghong Song struct bpf_iter__udp { 30965788b3a0SYonghong Song __bpf_md_ptr(struct bpf_iter_meta *, meta); 30975788b3a0SYonghong Song __bpf_md_ptr(struct udp_sock *, udp_sk); 30985788b3a0SYonghong Song uid_t uid __aligned(8); 30995788b3a0SYonghong Song int bucket __aligned(8); 31005788b3a0SYonghong Song }; 31015788b3a0SYonghong Song 3102c96dac8dSAditi Ghag struct bpf_udp_iter_state { 3103c96dac8dSAditi Ghag struct udp_iter_state state; 3104c96dac8dSAditi Ghag unsigned int cur_sk; 3105c96dac8dSAditi Ghag unsigned int end_sk; 3106c96dac8dSAditi Ghag unsigned int max_sk; 3107c96dac8dSAditi Ghag int offset; 3108c96dac8dSAditi Ghag struct sock **batch; 3109c96dac8dSAditi Ghag bool st_bucket_done; 3110c96dac8dSAditi Ghag }; 3111c96dac8dSAditi Ghag 3112c96dac8dSAditi Ghag static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3113c96dac8dSAditi Ghag unsigned int new_batch_sz); 3114c96dac8dSAditi Ghag static struct sock *bpf_iter_udp_batch(struct seq_file *seq) 3115c96dac8dSAditi Ghag { 3116c96dac8dSAditi Ghag struct bpf_udp_iter_state *iter = seq->private; 3117c96dac8dSAditi Ghag struct udp_iter_state *state = &iter->state; 3118c96dac8dSAditi Ghag struct net *net = seq_file_net(seq); 3119c96dac8dSAditi Ghag struct udp_table *udptable; 3120c96dac8dSAditi Ghag unsigned int batch_sks = 0; 3121c96dac8dSAditi Ghag bool resized = false; 3122c96dac8dSAditi Ghag struct sock *sk; 3123c96dac8dSAditi Ghag 3124c96dac8dSAditi Ghag /* The current batch is done, so advance the bucket. */ 3125c96dac8dSAditi Ghag if (iter->st_bucket_done) { 3126c96dac8dSAditi Ghag state->bucket++; 3127c96dac8dSAditi Ghag iter->offset = 0; 3128c96dac8dSAditi Ghag } 3129c96dac8dSAditi Ghag 3130c96dac8dSAditi Ghag udptable = udp_get_table_seq(seq, net); 3131c96dac8dSAditi Ghag 3132c96dac8dSAditi Ghag again: 3133c96dac8dSAditi Ghag /* New batch for the next bucket. 3134c96dac8dSAditi Ghag * Iterate over the hash table to find a bucket with sockets matching 3135c96dac8dSAditi Ghag * the iterator attributes, and return the first matching socket from 3136c96dac8dSAditi Ghag * the bucket. The remaining matched sockets from the bucket are batched 3137c96dac8dSAditi Ghag * before releasing the bucket lock. This allows BPF programs that are 3138c96dac8dSAditi Ghag * called in seq_show to acquire the bucket lock if needed. 3139c96dac8dSAditi Ghag */ 3140c96dac8dSAditi Ghag iter->cur_sk = 0; 3141c96dac8dSAditi Ghag iter->end_sk = 0; 3142c96dac8dSAditi Ghag iter->st_bucket_done = false; 3143c96dac8dSAditi Ghag batch_sks = 0; 3144c96dac8dSAditi Ghag 3145c96dac8dSAditi Ghag for (; state->bucket <= udptable->mask; state->bucket++) { 3146c96dac8dSAditi Ghag struct udp_hslot *hslot2 = &udptable->hash2[state->bucket]; 3147c96dac8dSAditi Ghag 3148c96dac8dSAditi Ghag if (hlist_empty(&hslot2->head)) { 3149c96dac8dSAditi Ghag iter->offset = 0; 3150c96dac8dSAditi Ghag continue; 3151c96dac8dSAditi Ghag } 3152c96dac8dSAditi Ghag 3153c96dac8dSAditi Ghag spin_lock_bh(&hslot2->lock); 3154c96dac8dSAditi Ghag udp_portaddr_for_each_entry(sk, &hslot2->head) { 3155c96dac8dSAditi Ghag if (seq_sk_match(seq, sk)) { 3156c96dac8dSAditi Ghag /* Resume from the last iterated socket at the 3157c96dac8dSAditi Ghag * offset in the bucket before iterator was stopped. 3158c96dac8dSAditi Ghag */ 3159c96dac8dSAditi Ghag if (iter->offset) { 3160c96dac8dSAditi Ghag --iter->offset; 3161c96dac8dSAditi Ghag continue; 3162c96dac8dSAditi Ghag } 3163c96dac8dSAditi Ghag if (iter->end_sk < iter->max_sk) { 3164c96dac8dSAditi Ghag sock_hold(sk); 3165c96dac8dSAditi Ghag iter->batch[iter->end_sk++] = sk; 3166c96dac8dSAditi Ghag } 3167c96dac8dSAditi Ghag batch_sks++; 3168c96dac8dSAditi Ghag } 3169c96dac8dSAditi Ghag } 3170c96dac8dSAditi Ghag spin_unlock_bh(&hslot2->lock); 3171c96dac8dSAditi Ghag 3172c96dac8dSAditi Ghag if (iter->end_sk) 3173c96dac8dSAditi Ghag break; 3174c96dac8dSAditi Ghag 3175c96dac8dSAditi Ghag /* Reset the current bucket's offset before moving to the next bucket. */ 3176c96dac8dSAditi Ghag iter->offset = 0; 3177c96dac8dSAditi Ghag } 3178c96dac8dSAditi Ghag 3179c96dac8dSAditi Ghag /* All done: no batch made. */ 3180c96dac8dSAditi Ghag if (!iter->end_sk) 3181c96dac8dSAditi Ghag return NULL; 3182c96dac8dSAditi Ghag 3183c96dac8dSAditi Ghag if (iter->end_sk == batch_sks) { 3184c96dac8dSAditi Ghag /* Batching is done for the current bucket; return the first 3185c96dac8dSAditi Ghag * socket to be iterated from the batch. 3186c96dac8dSAditi Ghag */ 3187c96dac8dSAditi Ghag iter->st_bucket_done = true; 3188c96dac8dSAditi Ghag goto done; 3189c96dac8dSAditi Ghag } 3190c96dac8dSAditi Ghag if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) { 3191c96dac8dSAditi Ghag resized = true; 3192c96dac8dSAditi Ghag /* After allocating a larger batch, retry one more time to grab 3193c96dac8dSAditi Ghag * the whole bucket. 3194c96dac8dSAditi Ghag */ 3195c96dac8dSAditi Ghag state->bucket--; 3196c96dac8dSAditi Ghag goto again; 3197c96dac8dSAditi Ghag } 3198c96dac8dSAditi Ghag done: 3199c96dac8dSAditi Ghag return iter->batch[0]; 3200c96dac8dSAditi Ghag } 3201c96dac8dSAditi Ghag 3202c96dac8dSAditi Ghag static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3203c96dac8dSAditi Ghag { 3204c96dac8dSAditi Ghag struct bpf_udp_iter_state *iter = seq->private; 3205c96dac8dSAditi Ghag struct sock *sk; 3206c96dac8dSAditi Ghag 3207c96dac8dSAditi Ghag /* Whenever seq_next() is called, the iter->cur_sk is 3208c96dac8dSAditi Ghag * done with seq_show(), so unref the iter->cur_sk. 3209c96dac8dSAditi Ghag */ 3210c96dac8dSAditi Ghag if (iter->cur_sk < iter->end_sk) { 3211c96dac8dSAditi Ghag sock_put(iter->batch[iter->cur_sk++]); 3212c96dac8dSAditi Ghag ++iter->offset; 3213c96dac8dSAditi Ghag } 3214c96dac8dSAditi Ghag 3215c96dac8dSAditi Ghag /* After updating iter->cur_sk, check if there are more sockets 3216c96dac8dSAditi Ghag * available in the current bucket batch. 3217c96dac8dSAditi Ghag */ 3218c96dac8dSAditi Ghag if (iter->cur_sk < iter->end_sk) 3219c96dac8dSAditi Ghag sk = iter->batch[iter->cur_sk]; 3220c96dac8dSAditi Ghag else 3221c96dac8dSAditi Ghag /* Prepare a new batch. */ 3222c96dac8dSAditi Ghag sk = bpf_iter_udp_batch(seq); 3223c96dac8dSAditi Ghag 3224c96dac8dSAditi Ghag ++*pos; 3225c96dac8dSAditi Ghag return sk; 3226c96dac8dSAditi Ghag } 3227c96dac8dSAditi Ghag 3228c96dac8dSAditi Ghag static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos) 3229c96dac8dSAditi Ghag { 3230c96dac8dSAditi Ghag /* bpf iter does not support lseek, so it always 3231c96dac8dSAditi Ghag * continue from where it was stop()-ped. 3232c96dac8dSAditi Ghag */ 3233c96dac8dSAditi Ghag if (*pos) 3234c96dac8dSAditi Ghag return bpf_iter_udp_batch(seq); 3235c96dac8dSAditi Ghag 3236c96dac8dSAditi Ghag return SEQ_START_TOKEN; 3237c96dac8dSAditi Ghag } 3238c96dac8dSAditi Ghag 32395788b3a0SYonghong Song static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 32405788b3a0SYonghong Song struct udp_sock *udp_sk, uid_t uid, int bucket) 32415788b3a0SYonghong Song { 32425788b3a0SYonghong Song struct bpf_iter__udp ctx; 32435788b3a0SYonghong Song 32445788b3a0SYonghong Song meta->seq_num--; /* skip SEQ_START_TOKEN */ 32455788b3a0SYonghong Song ctx.meta = meta; 32465788b3a0SYonghong Song ctx.udp_sk = udp_sk; 32475788b3a0SYonghong Song ctx.uid = uid; 32485788b3a0SYonghong Song ctx.bucket = bucket; 32495788b3a0SYonghong Song return bpf_iter_run_prog(prog, &ctx); 32505788b3a0SYonghong Song } 32515788b3a0SYonghong Song 32525788b3a0SYonghong Song static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 32535788b3a0SYonghong Song { 32545788b3a0SYonghong Song struct udp_iter_state *state = seq->private; 32555788b3a0SYonghong Song struct bpf_iter_meta meta; 32565788b3a0SYonghong Song struct bpf_prog *prog; 32575788b3a0SYonghong Song struct sock *sk = v; 32585788b3a0SYonghong Song uid_t uid; 3259c96dac8dSAditi Ghag int ret; 32605788b3a0SYonghong Song 32615788b3a0SYonghong Song if (v == SEQ_START_TOKEN) 32625788b3a0SYonghong Song return 0; 32635788b3a0SYonghong Song 3264c96dac8dSAditi Ghag lock_sock(sk); 3265c96dac8dSAditi Ghag 3266c96dac8dSAditi Ghag if (unlikely(sk_unhashed(sk))) { 3267c96dac8dSAditi Ghag ret = SEQ_SKIP; 3268c96dac8dSAditi Ghag goto unlock; 3269c96dac8dSAditi Ghag } 3270c96dac8dSAditi Ghag 32715788b3a0SYonghong Song uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 32725788b3a0SYonghong Song meta.seq = seq; 32735788b3a0SYonghong Song prog = bpf_iter_get_info(&meta, false); 3274c96dac8dSAditi Ghag ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3275c96dac8dSAditi Ghag 3276c96dac8dSAditi Ghag unlock: 3277c96dac8dSAditi Ghag release_sock(sk); 3278c96dac8dSAditi Ghag return ret; 3279c96dac8dSAditi Ghag } 3280c96dac8dSAditi Ghag 3281c96dac8dSAditi Ghag static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter) 3282c96dac8dSAditi Ghag { 3283c96dac8dSAditi Ghag while (iter->cur_sk < iter->end_sk) 3284c96dac8dSAditi Ghag sock_put(iter->batch[iter->cur_sk++]); 32855788b3a0SYonghong Song } 32865788b3a0SYonghong Song 32875788b3a0SYonghong Song static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 32885788b3a0SYonghong Song { 3289c96dac8dSAditi Ghag struct bpf_udp_iter_state *iter = seq->private; 32905788b3a0SYonghong Song struct bpf_iter_meta meta; 32915788b3a0SYonghong Song struct bpf_prog *prog; 32925788b3a0SYonghong Song 32935788b3a0SYonghong Song if (!v) { 32945788b3a0SYonghong Song meta.seq = seq; 32955788b3a0SYonghong Song prog = bpf_iter_get_info(&meta, true); 32965788b3a0SYonghong Song if (prog) 32975788b3a0SYonghong Song (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 32985788b3a0SYonghong Song } 32995788b3a0SYonghong Song 3300c96dac8dSAditi Ghag if (iter->cur_sk < iter->end_sk) { 3301c96dac8dSAditi Ghag bpf_iter_udp_put_batch(iter); 3302c96dac8dSAditi Ghag iter->st_bucket_done = false; 3303c96dac8dSAditi Ghag } 33045788b3a0SYonghong Song } 33055788b3a0SYonghong Song 33065788b3a0SYonghong Song static const struct seq_operations bpf_iter_udp_seq_ops = { 3307c96dac8dSAditi Ghag .start = bpf_iter_udp_seq_start, 3308c96dac8dSAditi Ghag .next = bpf_iter_udp_seq_next, 33095788b3a0SYonghong Song .stop = bpf_iter_udp_seq_stop, 33105788b3a0SYonghong Song .show = bpf_iter_udp_seq_show, 33115788b3a0SYonghong Song }; 33125788b3a0SYonghong Song #endif 33135788b3a0SYonghong Song 3314f44b1c51SAditi Ghag static unsigned short seq_file_family(const struct seq_file *seq) 3315f44b1c51SAditi Ghag { 3316f44b1c51SAditi Ghag const struct udp_seq_afinfo *afinfo; 3317f44b1c51SAditi Ghag 3318f44b1c51SAditi Ghag #ifdef CONFIG_BPF_SYSCALL 3319f44b1c51SAditi Ghag /* BPF iterator: bpf programs to filter sockets. */ 3320f44b1c51SAditi Ghag if (seq->op == &bpf_iter_udp_seq_ops) 3321f44b1c51SAditi Ghag return AF_UNSPEC; 3322f44b1c51SAditi Ghag #endif 3323f44b1c51SAditi Ghag 3324f44b1c51SAditi Ghag /* Proc fs iterator */ 3325f44b1c51SAditi Ghag afinfo = pde_data(file_inode(seq->file)); 3326f44b1c51SAditi Ghag return afinfo->family; 3327f44b1c51SAditi Ghag } 3328f44b1c51SAditi Ghag 3329c3506372SChristoph Hellwig const struct seq_operations udp_seq_ops = { 3330a3d2599bSChristoph Hellwig .start = udp_seq_start, 3331a3d2599bSChristoph Hellwig .next = udp_seq_next, 3332a3d2599bSChristoph Hellwig .stop = udp_seq_stop, 3333a3d2599bSChristoph Hellwig .show = udp4_seq_show, 3334a3d2599bSChristoph Hellwig }; 3335c3506372SChristoph Hellwig EXPORT_SYMBOL(udp_seq_ops); 333673cb88ecSArjan van de Ven 3337db8dac20SDavid S. Miller static struct udp_seq_afinfo udp4_seq_afinfo = { 3338db8dac20SDavid S. Miller .family = AF_INET, 3339478aee5dSKuniyuki Iwashima .udp_table = NULL, 3340db8dac20SDavid S. Miller }; 3341db8dac20SDavid S. Miller 33422c8c1e72SAlexey Dobriyan static int __net_init udp4_proc_init_net(struct net *net) 334315439febSPavel Emelyanov { 3344c3506372SChristoph Hellwig if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3345c3506372SChristoph Hellwig sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3346a3d2599bSChristoph Hellwig return -ENOMEM; 3347a3d2599bSChristoph Hellwig return 0; 334815439febSPavel Emelyanov } 334915439febSPavel Emelyanov 33502c8c1e72SAlexey Dobriyan static void __net_exit udp4_proc_exit_net(struct net *net) 335115439febSPavel Emelyanov { 3352a3d2599bSChristoph Hellwig remove_proc_entry("udp", net->proc_net); 335315439febSPavel Emelyanov } 335415439febSPavel Emelyanov 335515439febSPavel Emelyanov static struct pernet_operations udp4_net_ops = { 335615439febSPavel Emelyanov .init = udp4_proc_init_net, 335715439febSPavel Emelyanov .exit = udp4_proc_exit_net, 335815439febSPavel Emelyanov }; 335915439febSPavel Emelyanov 3360db8dac20SDavid S. Miller int __init udp4_proc_init(void) 3361db8dac20SDavid S. Miller { 336215439febSPavel Emelyanov return register_pernet_subsys(&udp4_net_ops); 3363db8dac20SDavid S. Miller } 3364db8dac20SDavid S. Miller 3365db8dac20SDavid S. Miller void udp4_proc_exit(void) 3366db8dac20SDavid S. Miller { 336715439febSPavel Emelyanov unregister_pernet_subsys(&udp4_net_ops); 3368db8dac20SDavid S. Miller } 33691da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 33701da177e4SLinus Torvalds 3371f86dcc5aSEric Dumazet static __initdata unsigned long uhash_entries; 3372f86dcc5aSEric Dumazet static int __init set_uhash_entries(char *str) 3373645ca708SEric Dumazet { 3374413c27d8SEldad Zack ssize_t ret; 3375413c27d8SEldad Zack 3376f86dcc5aSEric Dumazet if (!str) 3377f86dcc5aSEric Dumazet return 0; 3378413c27d8SEldad Zack 3379413c27d8SEldad Zack ret = kstrtoul(str, 0, &uhash_entries); 3380413c27d8SEldad Zack if (ret) 3381413c27d8SEldad Zack return 0; 3382413c27d8SEldad Zack 3383f86dcc5aSEric Dumazet if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3384f86dcc5aSEric Dumazet uhash_entries = UDP_HTABLE_SIZE_MIN; 3385f86dcc5aSEric Dumazet return 1; 3386f86dcc5aSEric Dumazet } 3387f86dcc5aSEric Dumazet __setup("uhash_entries=", set_uhash_entries); 3388645ca708SEric Dumazet 3389f86dcc5aSEric Dumazet void __init udp_table_init(struct udp_table *table, const char *name) 3390f86dcc5aSEric Dumazet { 3391f86dcc5aSEric Dumazet unsigned int i; 3392f86dcc5aSEric Dumazet 3393f86dcc5aSEric Dumazet table->hash = alloc_large_system_hash(name, 3394512615b6SEric Dumazet 2 * sizeof(struct udp_hslot), 3395f86dcc5aSEric Dumazet uhash_entries, 3396f86dcc5aSEric Dumazet 21, /* one slot per 2 MB */ 3397f86dcc5aSEric Dumazet 0, 3398f86dcc5aSEric Dumazet &table->log, 3399f86dcc5aSEric Dumazet &table->mask, 340031fe62b9STim Bird UDP_HTABLE_SIZE_MIN, 34019804985bSKuniyuki Iwashima UDP_HTABLE_SIZE_MAX); 340231fe62b9STim Bird 3403512615b6SEric Dumazet table->hash2 = table->hash + (table->mask + 1); 3404f86dcc5aSEric Dumazet for (i = 0; i <= table->mask; i++) { 3405ca065d0cSEric Dumazet INIT_HLIST_HEAD(&table->hash[i].head); 3406fdcc8aa9SEric Dumazet table->hash[i].count = 0; 3407645ca708SEric Dumazet spin_lock_init(&table->hash[i].lock); 3408645ca708SEric Dumazet } 3409512615b6SEric Dumazet for (i = 0; i <= table->mask; i++) { 3410ca065d0cSEric Dumazet INIT_HLIST_HEAD(&table->hash2[i].head); 3411512615b6SEric Dumazet table->hash2[i].count = 0; 3412512615b6SEric Dumazet spin_lock_init(&table->hash2[i].lock); 3413512615b6SEric Dumazet } 3414645ca708SEric Dumazet } 3415645ca708SEric Dumazet 3416723b8e46STom Herbert u32 udp_flow_hashrnd(void) 3417723b8e46STom Herbert { 3418723b8e46STom Herbert static u32 hashrnd __read_mostly; 3419723b8e46STom Herbert 3420723b8e46STom Herbert net_get_random_once(&hashrnd, sizeof(hashrnd)); 3421723b8e46STom Herbert 3422723b8e46STom Herbert return hashrnd; 3423723b8e46STom Herbert } 3424723b8e46STom Herbert EXPORT_SYMBOL(udp_flow_hashrnd); 3425723b8e46STom Herbert 34269804985bSKuniyuki Iwashima static void __net_init udp_sysctl_init(struct net *net) 34271e802951STonghao Zhang { 3428100fdd1fSEric Dumazet net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; 3429100fdd1fSEric Dumazet net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; 34301e802951STonghao Zhang 34311e802951STonghao Zhang #ifdef CONFIG_NET_L3_MASTER_DEV 34321e802951STonghao Zhang net->ipv4.sysctl_udp_l3mdev_accept = 0; 34331e802951STonghao Zhang #endif 34349804985bSKuniyuki Iwashima } 34359804985bSKuniyuki Iwashima 34369804985bSKuniyuki Iwashima static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries) 34379804985bSKuniyuki Iwashima { 34389804985bSKuniyuki Iwashima struct udp_table *udptable; 34399804985bSKuniyuki Iwashima int i; 34409804985bSKuniyuki Iwashima 34419804985bSKuniyuki Iwashima udptable = kmalloc(sizeof(*udptable), GFP_KERNEL); 34429804985bSKuniyuki Iwashima if (!udptable) 34439804985bSKuniyuki Iwashima goto out; 34449804985bSKuniyuki Iwashima 34459804985bSKuniyuki Iwashima udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot), 34469804985bSKuniyuki Iwashima GFP_KERNEL_ACCOUNT); 34479804985bSKuniyuki Iwashima if (!udptable->hash) 34489804985bSKuniyuki Iwashima goto free_table; 34499804985bSKuniyuki Iwashima 34509804985bSKuniyuki Iwashima udptable->hash2 = udptable->hash + hash_entries; 34519804985bSKuniyuki Iwashima udptable->mask = hash_entries - 1; 34529804985bSKuniyuki Iwashima udptable->log = ilog2(hash_entries); 34539804985bSKuniyuki Iwashima 34549804985bSKuniyuki Iwashima for (i = 0; i < hash_entries; i++) { 34559804985bSKuniyuki Iwashima INIT_HLIST_HEAD(&udptable->hash[i].head); 34569804985bSKuniyuki Iwashima udptable->hash[i].count = 0; 34579804985bSKuniyuki Iwashima spin_lock_init(&udptable->hash[i].lock); 34589804985bSKuniyuki Iwashima 34599804985bSKuniyuki Iwashima INIT_HLIST_HEAD(&udptable->hash2[i].head); 34609804985bSKuniyuki Iwashima udptable->hash2[i].count = 0; 34619804985bSKuniyuki Iwashima spin_lock_init(&udptable->hash2[i].lock); 34629804985bSKuniyuki Iwashima } 34639804985bSKuniyuki Iwashima 34649804985bSKuniyuki Iwashima return udptable; 34659804985bSKuniyuki Iwashima 34669804985bSKuniyuki Iwashima free_table: 34679804985bSKuniyuki Iwashima kfree(udptable); 34689804985bSKuniyuki Iwashima out: 34699804985bSKuniyuki Iwashima return NULL; 34709804985bSKuniyuki Iwashima } 34719804985bSKuniyuki Iwashima 34729804985bSKuniyuki Iwashima static void __net_exit udp_pernet_table_free(struct net *net) 34739804985bSKuniyuki Iwashima { 34749804985bSKuniyuki Iwashima struct udp_table *udptable = net->ipv4.udp_table; 34759804985bSKuniyuki Iwashima 34769804985bSKuniyuki Iwashima if (udptable == &udp_table) 34779804985bSKuniyuki Iwashima return; 34789804985bSKuniyuki Iwashima 34799804985bSKuniyuki Iwashima kvfree(udptable->hash); 34809804985bSKuniyuki Iwashima kfree(udptable); 34819804985bSKuniyuki Iwashima } 34829804985bSKuniyuki Iwashima 34839804985bSKuniyuki Iwashima static void __net_init udp_set_table(struct net *net) 34849804985bSKuniyuki Iwashima { 34859804985bSKuniyuki Iwashima struct udp_table *udptable; 34869804985bSKuniyuki Iwashima unsigned int hash_entries; 34879804985bSKuniyuki Iwashima struct net *old_net; 34889804985bSKuniyuki Iwashima 34899804985bSKuniyuki Iwashima if (net_eq(net, &init_net)) 34909804985bSKuniyuki Iwashima goto fallback; 34919804985bSKuniyuki Iwashima 34929804985bSKuniyuki Iwashima old_net = current->nsproxy->net_ns; 34939804985bSKuniyuki Iwashima hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries); 34949804985bSKuniyuki Iwashima if (!hash_entries) 34959804985bSKuniyuki Iwashima goto fallback; 34969804985bSKuniyuki Iwashima 34979804985bSKuniyuki Iwashima /* Set min to keep the bitmap on stack in udp_lib_get_port() */ 34989804985bSKuniyuki Iwashima if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET) 34999804985bSKuniyuki Iwashima hash_entries = UDP_HTABLE_SIZE_MIN_PERNET; 35009804985bSKuniyuki Iwashima else 35019804985bSKuniyuki Iwashima hash_entries = roundup_pow_of_two(hash_entries); 35029804985bSKuniyuki Iwashima 35039804985bSKuniyuki Iwashima udptable = udp_pernet_table_alloc(hash_entries); 35049804985bSKuniyuki Iwashima if (udptable) { 35059804985bSKuniyuki Iwashima net->ipv4.udp_table = udptable; 35069804985bSKuniyuki Iwashima } else { 35079804985bSKuniyuki Iwashima pr_warn("Failed to allocate UDP hash table (entries: %u) " 35089804985bSKuniyuki Iwashima "for a netns, fallback to the global one\n", 35099804985bSKuniyuki Iwashima hash_entries); 35109804985bSKuniyuki Iwashima fallback: 35119804985bSKuniyuki Iwashima net->ipv4.udp_table = &udp_table; 35129804985bSKuniyuki Iwashima } 35139804985bSKuniyuki Iwashima } 35149804985bSKuniyuki Iwashima 35159804985bSKuniyuki Iwashima static int __net_init udp_pernet_init(struct net *net) 35169804985bSKuniyuki Iwashima { 35179804985bSKuniyuki Iwashima udp_sysctl_init(net); 35189804985bSKuniyuki Iwashima udp_set_table(net); 35191e802951STonghao Zhang 35201e802951STonghao Zhang return 0; 35211e802951STonghao Zhang } 35221e802951STonghao Zhang 35239804985bSKuniyuki Iwashima static void __net_exit udp_pernet_exit(struct net *net) 35249804985bSKuniyuki Iwashima { 35259804985bSKuniyuki Iwashima udp_pernet_table_free(net); 35269804985bSKuniyuki Iwashima } 35279804985bSKuniyuki Iwashima 35281e802951STonghao Zhang static struct pernet_operations __net_initdata udp_sysctl_ops = { 35299804985bSKuniyuki Iwashima .init = udp_pernet_init, 35309804985bSKuniyuki Iwashima .exit = udp_pernet_exit, 35311e802951STonghao Zhang }; 35321e802951STonghao Zhang 35335788b3a0SYonghong Song #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 35345788b3a0SYonghong Song DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 35355788b3a0SYonghong Song struct udp_sock *udp_sk, uid_t uid, int bucket) 35365788b3a0SYonghong Song 3537c96dac8dSAditi Ghag static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3538c96dac8dSAditi Ghag unsigned int new_batch_sz) 35395788b3a0SYonghong Song { 3540c96dac8dSAditi Ghag struct sock **new_batch; 35415788b3a0SYonghong Song 3542c96dac8dSAditi Ghag new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch), 3543c96dac8dSAditi Ghag GFP_USER | __GFP_NOWARN); 3544c96dac8dSAditi Ghag if (!new_batch) 35455788b3a0SYonghong Song return -ENOMEM; 35465788b3a0SYonghong Song 3547c96dac8dSAditi Ghag bpf_iter_udp_put_batch(iter); 3548c96dac8dSAditi Ghag kvfree(iter->batch); 3549c96dac8dSAditi Ghag iter->batch = new_batch; 3550c96dac8dSAditi Ghag iter->max_sk = new_batch_sz; 3551c96dac8dSAditi Ghag 3552c96dac8dSAditi Ghag return 0; 3553c96dac8dSAditi Ghag } 3554c96dac8dSAditi Ghag 3555c96dac8dSAditi Ghag #define INIT_BATCH_SZ 16 3556c96dac8dSAditi Ghag 35575788b3a0SYonghong Song static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 35585788b3a0SYonghong Song { 3559c96dac8dSAditi Ghag struct bpf_udp_iter_state *iter = priv_data; 3560c96dac8dSAditi Ghag int ret; 3561c96dac8dSAditi Ghag 3562f9c79272SYonghong Song ret = bpf_iter_init_seq_net(priv_data, aux); 35635788b3a0SYonghong Song if (ret) 3564c96dac8dSAditi Ghag return ret; 3565c96dac8dSAditi Ghag 3566c96dac8dSAditi Ghag ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ); 3567c96dac8dSAditi Ghag if (ret) 3568c96dac8dSAditi Ghag bpf_iter_fini_seq_net(priv_data); 3569c96dac8dSAditi Ghag 35705788b3a0SYonghong Song return ret; 35715788b3a0SYonghong Song } 35725788b3a0SYonghong Song 35735788b3a0SYonghong Song static void bpf_iter_fini_udp(void *priv_data) 35745788b3a0SYonghong Song { 3575c96dac8dSAditi Ghag struct bpf_udp_iter_state *iter = priv_data; 35765788b3a0SYonghong Song 35775788b3a0SYonghong Song bpf_iter_fini_seq_net(priv_data); 3578c96dac8dSAditi Ghag kvfree(iter->batch); 35795788b3a0SYonghong Song } 35805788b3a0SYonghong Song 358114fc6bd6SYonghong Song static const struct bpf_iter_seq_info udp_seq_info = { 35825788b3a0SYonghong Song .seq_ops = &bpf_iter_udp_seq_ops, 35835788b3a0SYonghong Song .init_seq_private = bpf_iter_init_udp, 35845788b3a0SYonghong Song .fini_seq_private = bpf_iter_fini_udp, 3585c96dac8dSAditi Ghag .seq_priv_size = sizeof(struct bpf_udp_iter_state), 358614fc6bd6SYonghong Song }; 358714fc6bd6SYonghong Song 358814fc6bd6SYonghong Song static struct bpf_iter_reg udp_reg_info = { 358914fc6bd6SYonghong Song .target = "udp", 35905788b3a0SYonghong Song .ctx_arg_info_size = 1, 35915788b3a0SYonghong Song .ctx_arg_info = { 35925788b3a0SYonghong Song { offsetof(struct bpf_iter__udp, udp_sk), 35934ddbcb88SAditi Ghag PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED }, 35945788b3a0SYonghong Song }, 359514fc6bd6SYonghong Song .seq_info = &udp_seq_info, 35965788b3a0SYonghong Song }; 35975788b3a0SYonghong Song 35985788b3a0SYonghong Song static void __init bpf_iter_register(void) 35995788b3a0SYonghong Song { 3600951cf368SYonghong Song udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 36015788b3a0SYonghong Song if (bpf_iter_reg_target(&udp_reg_info)) 36025788b3a0SYonghong Song pr_warn("Warning: could not register bpf iterator udp\n"); 36035788b3a0SYonghong Song } 36045788b3a0SYonghong Song #endif 36055788b3a0SYonghong Song 360695766fffSHideo Aoki void __init udp_init(void) 360795766fffSHideo Aoki { 3608f03d78dbSEric Dumazet unsigned long limit; 36094b272750SEric Dumazet unsigned int i; 361095766fffSHideo Aoki 3611f86dcc5aSEric Dumazet udp_table_init(&udp_table, "UDP"); 3612f03d78dbSEric Dumazet limit = nr_free_buffer_pages() / 8; 361395766fffSHideo Aoki limit = max(limit, 128UL); 361495766fffSHideo Aoki sysctl_udp_mem[0] = limit / 4 * 3; 361595766fffSHideo Aoki sysctl_udp_mem[1] = limit; 361695766fffSHideo Aoki sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 361795766fffSHideo Aoki 36184b272750SEric Dumazet /* 16 spinlocks per cpu */ 36194b272750SEric Dumazet udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 36204b272750SEric Dumazet udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 36214b272750SEric Dumazet GFP_KERNEL); 36224b272750SEric Dumazet if (!udp_busylocks) 36234b272750SEric Dumazet panic("UDP: failed to alloc udp_busylocks\n"); 36244b272750SEric Dumazet for (i = 0; i < (1U << udp_busylocks_log); i++) 36254b272750SEric Dumazet spin_lock_init(udp_busylocks + i); 36261e802951STonghao Zhang 36271e802951STonghao Zhang if (register_pernet_subsys(&udp_sysctl_ops)) 36281e802951STonghao Zhang panic("UDP: failed to init sysctl parameters.\n"); 36295788b3a0SYonghong Song 36305788b3a0SYonghong Song #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 36315788b3a0SYonghong Song bpf_iter_register(); 36325788b3a0SYonghong Song #endif 363395766fffSHideo Aoki } 3634