2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The User Datagram Protocol (UDP).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
80 #define pr_fmt(fmt) "UDP: " fmt
82 #include <linux/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/memblock.h>
85 #include <linux/highmem.h>
86 #include <linux/swap.h>
87 #include <linux/types.h>
88 #include <linux/fcntl.h>
89 #include <linux/module.h>
90 #include <linux/socket.h>
91 #include <linux/sockios.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
95 #include <linux/errno.h>
96 #include <linux/timer.h>
98 #include <linux/inet.h>
99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/route.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <trace/events/udp.h>
112 #include <linux/static_key.h>
113 #include <trace/events/skb.h>
114 #include <net/busy_poll.h>
115 #include "udp_impl.h"
116 #include <net/sock_reuseport.h>
117 #include <net/addrconf.h>
118 #include <net/udp_tunnel.h>
120 struct udp_table udp_table __read_mostly;
121 EXPORT_SYMBOL(udp_table);
123 long sysctl_udp_mem[3] __read_mostly;
124 EXPORT_SYMBOL(sysctl_udp_mem);
126 atomic_long_t udp_memory_allocated;
127 EXPORT_SYMBOL(udp_memory_allocated);
129 #define MAX_UDP_PORTS 65536
130 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
132 /* IPCB reference means this can not be used from early demux */
133 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
135 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
136 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
137 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
143 static int udp_lib_lport_inuse(struct net *net, __u16 num,
144 const struct udp_hslot *hslot,
145 unsigned long *bitmap,
146 struct sock *sk, unsigned int log)
149 kuid_t uid = sock_i_uid(sk);
151 sk_for_each(sk2, &hslot->head) {
152 if (net_eq(sock_net(sk2), net) &&
154 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
155 (!sk2->sk_reuse || !sk->sk_reuse) &&
156 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
157 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
158 inet_rcv_saddr_equal(sk, sk2, true)) {
159 if (sk2->sk_reuseport && sk->sk_reuseport &&
160 !rcu_access_pointer(sk->sk_reuseport_cb) &&
161 uid_eq(uid, sock_i_uid(sk2))) {
167 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
176 * Note: we still hold spinlock of primary hash chain, so no other writer
177 * can insert/delete a socket with local_port == num
179 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
180 struct udp_hslot *hslot2,
184 kuid_t uid = sock_i_uid(sk);
187 spin_lock(&hslot2->lock);
188 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
189 if (net_eq(sock_net(sk2), net) &&
191 (udp_sk(sk2)->udp_port_hash == num) &&
192 (!sk2->sk_reuse || !sk->sk_reuse) &&
193 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
194 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
195 inet_rcv_saddr_equal(sk, sk2, true)) {
196 if (sk2->sk_reuseport && sk->sk_reuseport &&
197 !rcu_access_pointer(sk->sk_reuseport_cb) &&
198 uid_eq(uid, sock_i_uid(sk2))) {
206 spin_unlock(&hslot2->lock);
210 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
212 struct net *net = sock_net(sk);
213 kuid_t uid = sock_i_uid(sk);
216 sk_for_each(sk2, &hslot->head) {
217 if (net_eq(sock_net(sk2), net) &&
219 sk2->sk_family == sk->sk_family &&
220 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
221 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
222 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
223 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
224 inet_rcv_saddr_equal(sk, sk2, false)) {
225 return reuseport_add_sock(sk, sk2,
226 inet_rcv_saddr_any(sk));
230 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
234 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
236 * @sk: socket struct in question
237 * @snum: port number to look up
238 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
241 int udp_lib_get_port(struct sock *sk, unsigned short snum,
242 unsigned int hash2_nulladdr)
244 struct udp_hslot *hslot, *hslot2;
245 struct udp_table *udptable = sk->sk_prot->h.udp_table;
247 struct net *net = sock_net(sk);
250 int low, high, remaining;
252 unsigned short first, last;
253 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
255 inet_get_local_port_range(net, &low, &high);
256 remaining = (high - low) + 1;
258 rand = prandom_u32();
259 first = reciprocal_scale(rand, remaining) + low;
261 * force rand to be an odd multiple of UDP_HTABLE_SIZE
263 rand = (rand | 1) * (udptable->mask + 1);
264 last = first + udptable->mask + 1;
266 hslot = udp_hashslot(udptable, net, first);
267 bitmap_zero(bitmap, PORTS_PER_CHAIN);
268 spin_lock_bh(&hslot->lock);
269 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
274 * Iterate on all possible values of snum for this hash.
275 * Using steps of an odd multiple of UDP_HTABLE_SIZE
276 * give us randomization and full range coverage.
279 if (low <= snum && snum <= high &&
280 !test_bit(snum >> udptable->log, bitmap) &&
281 !inet_is_local_reserved_port(net, snum))
284 } while (snum != first);
285 spin_unlock_bh(&hslot->lock);
287 } while (++first != last);
290 hslot = udp_hashslot(udptable, net, snum);
291 spin_lock_bh(&hslot->lock);
292 if (hslot->count > 10) {
294 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
296 slot2 &= udptable->mask;
297 hash2_nulladdr &= udptable->mask;
299 hslot2 = udp_hashslot2(udptable, slot2);
300 if (hslot->count < hslot2->count)
301 goto scan_primary_hash;
303 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
304 if (!exist && (hash2_nulladdr != slot2)) {
305 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
306 exist = udp_lib_lport_inuse2(net, snum, hslot2,
315 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
319 inet_sk(sk)->inet_num = snum;
320 udp_sk(sk)->udp_port_hash = snum;
321 udp_sk(sk)->udp_portaddr_hash ^= snum;
322 if (sk_unhashed(sk)) {
323 if (sk->sk_reuseport &&
324 udp_reuseport_add_sock(sk, hslot)) {
325 inet_sk(sk)->inet_num = 0;
326 udp_sk(sk)->udp_port_hash = 0;
327 udp_sk(sk)->udp_portaddr_hash ^= snum;
331 sk_add_node_rcu(sk, &hslot->head);
333 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
335 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
336 spin_lock(&hslot2->lock);
337 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
338 sk->sk_family == AF_INET6)
339 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
342 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
345 spin_unlock(&hslot2->lock);
347 sock_set_flag(sk, SOCK_RCU_FREE);
350 spin_unlock_bh(&hslot->lock);
354 EXPORT_SYMBOL(udp_lib_get_port);
356 int udp_v4_get_port(struct sock *sk, unsigned short snum)
358 unsigned int hash2_nulladdr =
359 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
360 unsigned int hash2_partial =
361 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
363 /* precompute partial secondary hash */
364 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
365 return udp_lib_get_port(sk, snum, hash2_nulladdr);
368 static int compute_score(struct sock *sk, struct net *net,
369 __be32 saddr, __be16 sport,
370 __be32 daddr, unsigned short hnum,
371 int dif, int sdif, bool exact_dif)
374 struct inet_sock *inet;
377 if (!net_eq(sock_net(sk), net) ||
378 udp_sk(sk)->udp_port_hash != hnum ||
382 score = (sk->sk_family == PF_INET) ? 2 : 1;
385 if (inet->inet_rcv_saddr) {
386 if (inet->inet_rcv_saddr != daddr)
391 if (inet->inet_daddr) {
392 if (inet->inet_daddr != saddr)
397 if (inet->inet_dport) {
398 if (inet->inet_dport != sport)
403 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
409 if (sk->sk_incoming_cpu == raw_smp_processor_id())
414 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
415 const __u16 lport, const __be32 faddr,
418 static u32 udp_ehash_secret __read_mostly;
420 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
422 return __inet_ehashfn(laddr, lport, faddr, fport,
423 udp_ehash_secret + net_hash_mix(net));
426 /* called with rcu_read_lock() */
427 static struct sock *udp4_lib_lookup2(struct net *net,
428 __be32 saddr, __be16 sport,
429 __be32 daddr, unsigned int hnum,
430 int dif, int sdif, bool exact_dif,
431 struct udp_hslot *hslot2,
434 struct sock *sk, *result;
440 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
441 score = compute_score(sk, net, saddr, sport,
442 daddr, hnum, dif, sdif, exact_dif);
443 if (score > badness) {
444 if (sk->sk_reuseport) {
445 hash = udp_ehashfn(net, daddr, hnum,
447 result = reuseport_select_sock(sk, hash, skb,
448 sizeof(struct udphdr));
459 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
460 * harder than this. -DaveM
462 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
463 __be16 sport, __be32 daddr, __be16 dport, int dif,
464 int sdif, struct udp_table *udptable, struct sk_buff *skb)
466 struct sock *sk, *result;
467 unsigned short hnum = ntohs(dport);
468 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
469 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
470 bool exact_dif = udp_lib_exact_dif_match(net, skb);
474 if (hslot->count > 10) {
475 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
476 slot2 = hash2 & udptable->mask;
477 hslot2 = &udptable->hash2[slot2];
478 if (hslot->count < hslot2->count)
481 result = udp4_lib_lookup2(net, saddr, sport,
482 daddr, hnum, dif, sdif,
483 exact_dif, hslot2, skb);
485 unsigned int old_slot2 = slot2;
486 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
487 slot2 = hash2 & udptable->mask;
488 /* avoid searching the same slot again. */
489 if (unlikely(slot2 == old_slot2))
492 hslot2 = &udptable->hash2[slot2];
493 if (hslot->count < hslot2->count)
496 result = udp4_lib_lookup2(net, saddr, sport,
497 daddr, hnum, dif, sdif,
498 exact_dif, hslot2, skb);
500 if (unlikely(IS_ERR(result)))
507 sk_for_each_rcu(sk, &hslot->head) {
508 score = compute_score(sk, net, saddr, sport,
509 daddr, hnum, dif, sdif, exact_dif);
510 if (score > badness) {
511 if (sk->sk_reuseport) {
512 hash = udp_ehashfn(net, daddr, hnum,
514 result = reuseport_select_sock(sk, hash, skb,
515 sizeof(struct udphdr));
516 if (unlikely(IS_ERR(result)))
527 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
529 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
530 __be16 sport, __be16 dport,
531 struct udp_table *udptable)
533 const struct iphdr *iph = ip_hdr(skb);
535 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
536 iph->daddr, dport, inet_iif(skb),
537 inet_sdif(skb), udptable, skb);
540 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
541 __be16 sport, __be16 dport)
543 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
545 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
547 /* Must be called under rcu_read_lock().
548 * Does increment socket refcount.
550 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
551 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
552 __be32 daddr, __be16 dport, int dif)
556 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
557 dif, 0, &udp_table, NULL);
558 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
562 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
565 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
566 __be16 loc_port, __be32 loc_addr,
567 __be16 rmt_port, __be32 rmt_addr,
568 int dif, int sdif, unsigned short hnum)
570 struct inet_sock *inet = inet_sk(sk);
572 if (!net_eq(sock_net(sk), net) ||
573 udp_sk(sk)->udp_port_hash != hnum ||
574 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
575 (inet->inet_dport != rmt_port && inet->inet_dport) ||
576 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
577 ipv6_only_sock(sk) ||
578 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
579 sk->sk_bound_dev_if != sdif))
581 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
586 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
587 void udp_encap_enable(void)
589 static_branch_enable(&udp_encap_needed_key);
591 EXPORT_SYMBOL(udp_encap_enable);
593 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
594 * reversing source and destination port: this will match tunnels that force the
595 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
596 * lwtunnels might actually break this assumption by being configured with
597 * different destination ports on endpoints, in this case we won't be able to
598 * trace ICMP messages back to them.
600 * Then ask the tunnel implementation to match the error against a valid
603 * Return the socket if we have a match.
605 static struct sock *__udp4_lib_err_encap(struct net *net,
606 const struct iphdr *iph,
608 struct udp_table *udptable,
611 int (*lookup)(struct sock *sk, struct sk_buff *skb);
612 int network_offset, transport_offset;
616 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
617 iph->saddr, uh->dest, skb->dev->ifindex, 0,
622 network_offset = skb_network_offset(skb);
623 transport_offset = skb_transport_offset(skb);
625 /* Network header needs to point to the outer IPv4 header inside ICMP */
626 skb_reset_network_header(skb);
628 /* Transport header needs to point to the UDP header */
629 skb_set_transport_header(skb, iph->ihl << 2);
632 lookup = READ_ONCE(up->encap_err_lookup);
633 if (!lookup || lookup(sk, skb))
636 skb_set_transport_header(skb, transport_offset);
637 skb_set_network_header(skb, network_offset);
643 * This routine is called by the ICMP module when it gets some
644 * sort of error condition. If err < 0 then the socket should
645 * be closed and the error returned to the user. If err > 0
646 * it's just the icmp type << 8 | icmp code.
647 * Header points to the ip header of the error packet. We move
648 * on past this. Then (as it used to claim before adjustment)
649 * header points to the first 8 bytes of the udp header. We need
650 * to find the appropriate port.
653 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
655 struct inet_sock *inet;
656 const struct iphdr *iph = (const struct iphdr *)skb->data;
657 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
658 const int type = icmp_hdr(skb)->type;
659 const int code = icmp_hdr(skb)->code;
664 struct net *net = dev_net(skb->dev);
666 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
667 iph->saddr, uh->source, skb->dev->ifindex,
668 inet_sdif(skb), udptable, NULL);
670 /* No socket for error: try tunnels before discarding */
671 if (static_branch_unlikely(&udp_encap_needed_key))
672 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb);
675 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
687 case ICMP_TIME_EXCEEDED:
690 case ICMP_SOURCE_QUENCH:
692 case ICMP_PARAMETERPROB:
696 case ICMP_DEST_UNREACH:
697 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
698 ipv4_sk_update_pmtu(skb, sk, info);
699 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
707 if (code <= NR_ICMP_UNREACH) {
708 harderr = icmp_err_convert[code].fatal;
709 err = icmp_err_convert[code].errno;
713 ipv4_sk_redirect(skb, sk);
718 * RFC1122: OK. Passes ICMP errors back to application, as per
722 /* ...not for tunnels though: we don't have a sending socket */
725 if (!inet->recverr) {
726 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
729 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
732 sk->sk_error_report(sk);
737 int udp_err(struct sk_buff *skb, u32 info)
739 return __udp4_lib_err(skb, info, &udp_table);
743 * Throw away all pending data and cancel the corking. Socket is locked.
745 void udp_flush_pending_frames(struct sock *sk)
747 struct udp_sock *up = udp_sk(sk);
752 ip_flush_pending_frames(sk);
755 EXPORT_SYMBOL(udp_flush_pending_frames);
758 * udp4_hwcsum - handle outgoing HW checksumming
759 * @skb: sk_buff containing the filled-in UDP header
760 * (checksum field must be zeroed out)
761 * @src: source IP address
762 * @dst: destination IP address
764 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
766 struct udphdr *uh = udp_hdr(skb);
767 int offset = skb_transport_offset(skb);
768 int len = skb->len - offset;
772 if (!skb_has_frag_list(skb)) {
774 * Only one fragment on the socket.
776 skb->csum_start = skb_transport_header(skb) - skb->head;
777 skb->csum_offset = offsetof(struct udphdr, check);
778 uh->check = ~csum_tcpudp_magic(src, dst, len,
781 struct sk_buff *frags;
784 * HW-checksum won't work as there are two or more
785 * fragments on the socket so that all csums of sk_buffs
788 skb_walk_frags(skb, frags) {
789 csum = csum_add(csum, frags->csum);
793 csum = skb_checksum(skb, offset, hlen, csum);
794 skb->ip_summed = CHECKSUM_NONE;
796 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
798 uh->check = CSUM_MANGLED_0;
801 EXPORT_SYMBOL_GPL(udp4_hwcsum);
803 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
804 * for the simple case like when setting the checksum for a UDP tunnel.
806 void udp_set_csum(bool nocheck, struct sk_buff *skb,
807 __be32 saddr, __be32 daddr, int len)
809 struct udphdr *uh = udp_hdr(skb);
813 } else if (skb_is_gso(skb)) {
814 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
815 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
817 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
819 uh->check = CSUM_MANGLED_0;
821 skb->ip_summed = CHECKSUM_PARTIAL;
822 skb->csum_start = skb_transport_header(skb) - skb->head;
823 skb->csum_offset = offsetof(struct udphdr, check);
824 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
827 EXPORT_SYMBOL(udp_set_csum);
829 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
830 struct inet_cork *cork)
832 struct sock *sk = skb->sk;
833 struct inet_sock *inet = inet_sk(sk);
836 int is_udplite = IS_UDPLITE(sk);
837 int offset = skb_transport_offset(skb);
838 int len = skb->len - offset;
842 * Create a UDP header
845 uh->source = inet->inet_sport;
846 uh->dest = fl4->fl4_dport;
847 uh->len = htons(len);
850 if (cork->gso_size) {
851 const int hlen = skb_network_header_len(skb) +
852 sizeof(struct udphdr);
854 if (hlen + cork->gso_size > cork->fragsize)
856 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS)
858 if (sk->sk_no_check_tx)
860 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
861 dst_xfrm(skb_dst(skb)))
864 skb_shinfo(skb)->gso_size = cork->gso_size;
865 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
866 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh),
871 if (is_udplite) /* UDP-Lite */
872 csum = udplite_csum(skb);
874 else if (sk->sk_no_check_tx) { /* UDP csum off */
876 skb->ip_summed = CHECKSUM_NONE;
879 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
882 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
886 csum = udp_csum(skb);
888 /* add protocol-dependent pseudo-header */
889 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
890 sk->sk_protocol, csum);
892 uh->check = CSUM_MANGLED_0;
895 err = ip_send_skb(sock_net(sk), skb);
897 if (err == -ENOBUFS && !inet->recverr) {
898 UDP_INC_STATS(sock_net(sk),
899 UDP_MIB_SNDBUFERRORS, is_udplite);
903 UDP_INC_STATS(sock_net(sk),
904 UDP_MIB_OUTDATAGRAMS, is_udplite);
909 * Push out all pending data as one UDP datagram. Socket is locked.
911 int udp_push_pending_frames(struct sock *sk)
913 struct udp_sock *up = udp_sk(sk);
914 struct inet_sock *inet = inet_sk(sk);
915 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
919 skb = ip_finish_skb(sk, fl4);
923 err = udp_send_skb(skb, fl4, &inet->cork.base);
930 EXPORT_SYMBOL(udp_push_pending_frames);
932 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
934 switch (cmsg->cmsg_type) {
936 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
938 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
945 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
947 struct cmsghdr *cmsg;
948 bool need_ip = false;
951 for_each_cmsghdr(cmsg, msg) {
952 if (!CMSG_OK(msg, cmsg))
955 if (cmsg->cmsg_level != SOL_UDP) {
960 err = __udp_cmsg_send(cmsg, gso_size);
967 EXPORT_SYMBOL_GPL(udp_cmsg_send);
969 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
971 struct inet_sock *inet = inet_sk(sk);
972 struct udp_sock *up = udp_sk(sk);
973 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
974 struct flowi4 fl4_stack;
977 struct ipcm_cookie ipc;
978 struct rtable *rt = NULL;
981 __be32 daddr, faddr, saddr;
984 int err, is_udplite = IS_UDPLITE(sk);
985 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
986 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
988 struct ip_options_data opt_copy;
997 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1000 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1002 fl4 = &inet->cork.fl.u.ip4;
1005 * There are pending frames.
1006 * The socket lock must be held while it's corked.
1009 if (likely(up->pending)) {
1010 if (unlikely(up->pending != AF_INET)) {
1014 goto do_append_data;
1018 ulen += sizeof(struct udphdr);
1021 * Get and verify the address.
1024 if (msg->msg_namelen < sizeof(*usin))
1026 if (usin->sin_family != AF_INET) {
1027 if (usin->sin_family != AF_UNSPEC)
1028 return -EAFNOSUPPORT;
1031 daddr = usin->sin_addr.s_addr;
1032 dport = usin->sin_port;
1036 if (sk->sk_state != TCP_ESTABLISHED)
1037 return -EDESTADDRREQ;
1038 daddr = inet->inet_daddr;
1039 dport = inet->inet_dport;
1040 /* Open fast path for connected socket.
1041 Route will not be used, if at least one option is set.
1046 ipcm_init_sk(&ipc, inet);
1047 ipc.gso_size = up->gso_size;
1049 if (msg->msg_controllen) {
1050 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1052 err = ip_cmsg_send(sk, msg, &ipc,
1053 sk->sk_family == AF_INET6);
1054 if (unlikely(err < 0)) {
1063 struct ip_options_rcu *inet_opt;
1066 inet_opt = rcu_dereference(inet->inet_opt);
1068 memcpy(&opt_copy, inet_opt,
1069 sizeof(*inet_opt) + inet_opt->opt.optlen);
1070 ipc.opt = &opt_copy.opt;
1075 if (cgroup_bpf_enabled && !connected) {
1076 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1077 (struct sockaddr *)usin, &ipc.addr);
1081 if (usin->sin_port == 0) {
1082 /* BPF program set invalid port. Reject it. */
1086 daddr = usin->sin_addr.s_addr;
1087 dport = usin->sin_port;
1092 ipc.addr = faddr = daddr;
1094 if (ipc.opt && ipc.opt->opt.srr) {
1099 faddr = ipc.opt->opt.faddr;
1102 tos = get_rttos(&ipc, inet);
1103 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1104 (msg->msg_flags & MSG_DONTROUTE) ||
1105 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1110 if (ipv4_is_multicast(daddr)) {
1111 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1112 ipc.oif = inet->mc_index;
1114 saddr = inet->mc_addr;
1116 } else if (!ipc.oif) {
1117 ipc.oif = inet->uc_index;
1118 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1119 /* oif is set, packet is to local broadcast and
1120 * and uc_index is set. oif is most likely set
1121 * by sk_bound_dev_if. If uc_index != oif check if the
1122 * oif is an L3 master and uc_index is an L3 slave.
1123 * If so, we want to allow the send using the uc_index.
1125 if (ipc.oif != inet->uc_index &&
1126 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1128 ipc.oif = inet->uc_index;
1133 rt = (struct rtable *)sk_dst_check(sk, 0);
1136 struct net *net = sock_net(sk);
1137 __u8 flow_flags = inet_sk_flowi_flags(sk);
1141 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1142 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1144 faddr, saddr, dport, inet->inet_sport,
1147 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1148 rt = ip_route_output_flow(net, fl4, sk);
1152 if (err == -ENETUNREACH)
1153 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1158 if ((rt->rt_flags & RTCF_BROADCAST) &&
1159 !sock_flag(sk, SOCK_BROADCAST))
1162 sk_dst_set(sk, dst_clone(&rt->dst));
1165 if (msg->msg_flags&MSG_CONFIRM)
1171 daddr = ipc.addr = fl4->daddr;
1173 /* Lockless fast path for the non-corking case. */
1175 struct inet_cork cork;
1177 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1178 sizeof(struct udphdr), &ipc, &rt,
1179 &cork, msg->msg_flags);
1181 if (!IS_ERR_OR_NULL(skb))
1182 err = udp_send_skb(skb, fl4, &cork);
1187 if (unlikely(up->pending)) {
1188 /* The socket is already corked while preparing it. */
1189 /* ... which is an evident application bug. --ANK */
1192 net_dbg_ratelimited("socket already corked\n");
1197 * Now cork the socket to pend data.
1199 fl4 = &inet->cork.fl.u.ip4;
1202 fl4->fl4_dport = dport;
1203 fl4->fl4_sport = inet->inet_sport;
1204 up->pending = AF_INET;
1208 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1209 sizeof(struct udphdr), &ipc, &rt,
1210 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1212 udp_flush_pending_frames(sk);
1214 err = udp_push_pending_frames(sk);
1215 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1227 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1228 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1229 * we don't have a good statistic (IpOutDiscards but it can be too many
1230 * things). We could add another new stat but at least for now that
1231 * seems like overkill.
1233 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1234 UDP_INC_STATS(sock_net(sk),
1235 UDP_MIB_SNDBUFERRORS, is_udplite);
1240 if (msg->msg_flags & MSG_PROBE)
1241 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1242 if (!(msg->msg_flags&MSG_PROBE) || len)
1243 goto back_from_confirm;
1247 EXPORT_SYMBOL(udp_sendmsg);
1249 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1250 size_t size, int flags)
1252 struct inet_sock *inet = inet_sk(sk);
1253 struct udp_sock *up = udp_sk(sk);
1256 if (flags & MSG_SENDPAGE_NOTLAST)
1260 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1262 /* Call udp_sendmsg to specify destination address which
1263 * sendpage interface can't pass.
1264 * This will succeed only when the socket is connected.
1266 ret = udp_sendmsg(sk, &msg, 0);
1273 if (unlikely(!up->pending)) {
1276 net_dbg_ratelimited("cork failed\n");
1280 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1281 page, offset, size, flags);
1282 if (ret == -EOPNOTSUPP) {
1284 return sock_no_sendpage(sk->sk_socket, page, offset,
1288 udp_flush_pending_frames(sk);
1293 if (!(up->corkflag || (flags&MSG_MORE)))
1294 ret = udp_push_pending_frames(sk);
1302 #define UDP_SKB_IS_STATELESS 0x80000000
1304 static void udp_set_dev_scratch(struct sk_buff *skb)
1306 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1308 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1309 scratch->_tsize_state = skb->truesize;
1310 #if BITS_PER_LONG == 64
1311 scratch->len = skb->len;
1312 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1313 scratch->is_linear = !skb_is_nonlinear(skb);
1315 /* all head states execept sp (dst, sk, nf) are always cleared by
1316 * udp_rcv() and we need to preserve secpath, if present, to eventually
1317 * process IP_CMSG_PASSSEC at recvmsg() time
1319 if (likely(!skb_sec_path(skb)))
1320 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1323 static int udp_skb_truesize(struct sk_buff *skb)
1325 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1328 static bool udp_skb_has_head_state(struct sk_buff *skb)
1330 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1333 /* fully reclaim rmem/fwd memory allocated for skb */
1334 static void udp_rmem_release(struct sock *sk, int size, int partial,
1335 bool rx_queue_lock_held)
1337 struct udp_sock *up = udp_sk(sk);
1338 struct sk_buff_head *sk_queue;
1341 if (likely(partial)) {
1342 up->forward_deficit += size;
1343 size = up->forward_deficit;
1344 if (size < (sk->sk_rcvbuf >> 2))
1347 size += up->forward_deficit;
1349 up->forward_deficit = 0;
1351 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1352 * if the called don't held it already
1354 sk_queue = &sk->sk_receive_queue;
1355 if (!rx_queue_lock_held)
1356 spin_lock(&sk_queue->lock);
1359 sk->sk_forward_alloc += size;
1360 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1361 sk->sk_forward_alloc -= amt;
1364 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1366 atomic_sub(size, &sk->sk_rmem_alloc);
1368 /* this can save us from acquiring the rx queue lock on next receive */
1369 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1371 if (!rx_queue_lock_held)
1372 spin_unlock(&sk_queue->lock);
1375 /* Note: called with reader_queue.lock held.
1376 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1377 * This avoids a cache line miss while receive_queue lock is held.
1378 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1380 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1382 prefetch(&skb->data);
1383 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1385 EXPORT_SYMBOL(udp_skb_destructor);
1387 /* as above, but the caller held the rx queue lock, too */
1388 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1390 prefetch(&skb->data);
1391 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1394 /* Idea of busylocks is to let producers grab an extra spinlock
1395 * to relieve pressure on the receive_queue spinlock shared by consumer.
1396 * Under flood, this means that only one producer can be in line
1397 * trying to acquire the receive_queue spinlock.
1398 * These busylock can be allocated on a per cpu manner, instead of a
1399 * per socket one (that would consume a cache line per socket)
1401 static int udp_busylocks_log __read_mostly;
1402 static spinlock_t *udp_busylocks __read_mostly;
1404 static spinlock_t *busylock_acquire(void *ptr)
1408 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1413 static void busylock_release(spinlock_t *busy)
1419 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1421 struct sk_buff_head *list = &sk->sk_receive_queue;
1422 int rmem, delta, amt, err = -ENOMEM;
1423 spinlock_t *busy = NULL;
1426 /* try to avoid the costly atomic add/sub pair when the receive
1427 * queue is full; always allow at least a packet
1429 rmem = atomic_read(&sk->sk_rmem_alloc);
1430 if (rmem > sk->sk_rcvbuf)
1433 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1434 * having linear skbs :
1435 * - Reduce memory overhead and thus increase receive queue capacity
1436 * - Less cache line misses at copyout() time
1437 * - Less work at consume_skb() (less alien page frag freeing)
1439 if (rmem > (sk->sk_rcvbuf >> 1)) {
1442 busy = busylock_acquire(sk);
1444 size = skb->truesize;
1445 udp_set_dev_scratch(skb);
1447 /* we drop only if the receive buf is full and the receive
1448 * queue contains some other skb
1450 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1451 if (rmem > (size + sk->sk_rcvbuf))
1454 spin_lock(&list->lock);
1455 if (size >= sk->sk_forward_alloc) {
1456 amt = sk_mem_pages(size);
1457 delta = amt << SK_MEM_QUANTUM_SHIFT;
1458 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1460 spin_unlock(&list->lock);
1464 sk->sk_forward_alloc += delta;
1467 sk->sk_forward_alloc -= size;
1469 /* no need to setup a destructor, we will explicitly release the
1470 * forward allocated memory on dequeue
1472 sock_skb_set_dropcount(sk, skb);
1474 __skb_queue_tail(list, skb);
1475 spin_unlock(&list->lock);
1477 if (!sock_flag(sk, SOCK_DEAD))
1478 sk->sk_data_ready(sk);
1480 busylock_release(busy);
1484 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1487 atomic_inc(&sk->sk_drops);
1488 busylock_release(busy);
1491 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1493 void udp_destruct_sock(struct sock *sk)
1495 /* reclaim completely the forward allocated memory */
1496 struct udp_sock *up = udp_sk(sk);
1497 unsigned int total = 0;
1498 struct sk_buff *skb;
1500 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1501 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1502 total += skb->truesize;
1505 udp_rmem_release(sk, total, 0, true);
1507 inet_sock_destruct(sk);
1509 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1511 int udp_init_sock(struct sock *sk)
1513 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1514 sk->sk_destruct = udp_destruct_sock;
1517 EXPORT_SYMBOL_GPL(udp_init_sock);
1519 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1521 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1522 bool slow = lock_sock_fast(sk);
1524 sk_peek_offset_bwd(sk, len);
1525 unlock_sock_fast(sk, slow);
1528 if (!skb_unref(skb))
1531 /* In the more common cases we cleared the head states previously,
1532 * see __udp_queue_rcv_skb().
1534 if (unlikely(udp_skb_has_head_state(skb)))
1535 skb_release_head_state(skb);
1536 __consume_stateless_skb(skb);
1538 EXPORT_SYMBOL_GPL(skb_consume_udp);
1540 static struct sk_buff *__first_packet_length(struct sock *sk,
1541 struct sk_buff_head *rcvq,
1544 struct sk_buff *skb;
1546 while ((skb = skb_peek(rcvq)) != NULL) {
1547 if (udp_lib_checksum_complete(skb)) {
1548 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1550 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1552 atomic_inc(&sk->sk_drops);
1553 __skb_unlink(skb, rcvq);
1554 *total += skb->truesize;
1557 /* the csum related bits could be changed, refresh
1560 udp_set_dev_scratch(skb);
1568 * first_packet_length - return length of first packet in receive queue
1571 * Drops all bad checksum frames, until a valid one is found.
1572 * Returns the length of found skb, or -1 if none is found.
1574 static int first_packet_length(struct sock *sk)
1576 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1577 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1578 struct sk_buff *skb;
1582 spin_lock_bh(&rcvq->lock);
1583 skb = __first_packet_length(sk, rcvq, &total);
1584 if (!skb && !skb_queue_empty(sk_queue)) {
1585 spin_lock(&sk_queue->lock);
1586 skb_queue_splice_tail_init(sk_queue, rcvq);
1587 spin_unlock(&sk_queue->lock);
1589 skb = __first_packet_length(sk, rcvq, &total);
1591 res = skb ? skb->len : -1;
1593 udp_rmem_release(sk, total, 1, false);
1594 spin_unlock_bh(&rcvq->lock);
1599 * IOCTL requests applicable to the UDP protocol
1602 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1607 int amount = sk_wmem_alloc_get(sk);
1609 return put_user(amount, (int __user *)arg);
1614 int amount = max_t(int, 0, first_packet_length(sk));
1616 return put_user(amount, (int __user *)arg);
1620 return -ENOIOCTLCMD;
1625 EXPORT_SYMBOL(udp_ioctl);
1627 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1628 int noblock, int *peeked, int *off, int *err)
1630 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1631 struct sk_buff_head *queue;
1632 struct sk_buff *last;
1636 queue = &udp_sk(sk)->reader_queue;
1637 flags |= noblock ? MSG_DONTWAIT : 0;
1638 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1640 struct sk_buff *skb;
1642 error = sock_error(sk);
1649 spin_lock_bh(&queue->lock);
1650 skb = __skb_try_recv_from_queue(sk, queue, flags,
1655 spin_unlock_bh(&queue->lock);
1659 if (skb_queue_empty(sk_queue)) {
1660 spin_unlock_bh(&queue->lock);
1664 /* refill the reader queue and walk it again
1665 * keep both queues locked to avoid re-acquiring
1666 * the sk_receive_queue lock if fwd memory scheduling
1669 spin_lock(&sk_queue->lock);
1670 skb_queue_splice_tail_init(sk_queue, queue);
1672 skb = __skb_try_recv_from_queue(sk, queue, flags,
1673 udp_skb_dtor_locked,
1676 spin_unlock(&sk_queue->lock);
1677 spin_unlock_bh(&queue->lock);
1682 if (!sk_can_busy_loop(sk))
1685 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1686 } while (!skb_queue_empty(sk_queue));
1688 /* sk_queue is empty, reader_queue may contain peeked packets */
1690 !__skb_wait_for_more_packets(sk, &error, &timeo,
1691 (struct sk_buff *)sk_queue));
1696 EXPORT_SYMBOL(__skb_recv_udp);
1699 * This should be easy, if there is something there we
1700 * return it, otherwise we block.
1703 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1704 int flags, int *addr_len)
1706 struct inet_sock *inet = inet_sk(sk);
1707 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1708 struct sk_buff *skb;
1709 unsigned int ulen, copied;
1710 int peeked, peeking, off;
1712 int is_udplite = IS_UDPLITE(sk);
1713 bool checksum_valid = false;
1715 if (flags & MSG_ERRQUEUE)
1716 return ip_recv_error(sk, msg, len, addr_len);
1719 peeking = flags & MSG_PEEK;
1720 off = sk_peek_offset(sk, flags);
1721 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1725 ulen = udp_skb_len(skb);
1727 if (copied > ulen - off)
1728 copied = ulen - off;
1729 else if (copied < ulen)
1730 msg->msg_flags |= MSG_TRUNC;
1733 * If checksum is needed at all, try to do it while copying the
1734 * data. If the data is truncated, or if we only want a partial
1735 * coverage checksum (UDP-Lite), do it before the copy.
1738 if (copied < ulen || peeking ||
1739 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1740 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1741 !__udp_lib_checksum_complete(skb);
1742 if (!checksum_valid)
1746 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1747 if (udp_skb_is_linear(skb))
1748 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1750 err = skb_copy_datagram_msg(skb, off, msg, copied);
1752 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1758 if (unlikely(err)) {
1760 atomic_inc(&sk->sk_drops);
1761 UDP_INC_STATS(sock_net(sk),
1762 UDP_MIB_INERRORS, is_udplite);
1769 UDP_INC_STATS(sock_net(sk),
1770 UDP_MIB_INDATAGRAMS, is_udplite);
1772 sock_recv_ts_and_drops(msg, sk, skb);
1774 /* Copy the address. */
1776 sin->sin_family = AF_INET;
1777 sin->sin_port = udp_hdr(skb)->source;
1778 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1779 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1780 *addr_len = sizeof(*sin);
1783 if (udp_sk(sk)->gro_enabled)
1784 udp_cmsg_recv(msg, sk, skb);
1786 if (inet->cmsg_flags)
1787 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1790 if (flags & MSG_TRUNC)
1793 skb_consume_udp(sk, skb, peeking ? -err : err);
1797 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1798 udp_skb_destructor)) {
1799 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1800 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1804 /* starting over for a new packet, but check if we need to yield */
1806 msg->msg_flags &= ~MSG_TRUNC;
1810 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1812 /* This check is replicated from __ip4_datagram_connect() and
1813 * intended to prevent BPF program called below from accessing bytes
1814 * that are out of the bound specified by user in addr_len.
1816 if (addr_len < sizeof(struct sockaddr_in))
1819 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1821 EXPORT_SYMBOL(udp_pre_connect);
1823 int __udp_disconnect(struct sock *sk, int flags)
1825 struct inet_sock *inet = inet_sk(sk);
1827 * 1003.1g - break association.
1830 sk->sk_state = TCP_CLOSE;
1831 inet->inet_daddr = 0;
1832 inet->inet_dport = 0;
1833 sock_rps_reset_rxhash(sk);
1834 sk->sk_bound_dev_if = 0;
1835 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1836 inet_reset_saddr(sk);
1838 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1839 sk->sk_prot->unhash(sk);
1840 inet->inet_sport = 0;
1845 EXPORT_SYMBOL(__udp_disconnect);
1847 int udp_disconnect(struct sock *sk, int flags)
1850 __udp_disconnect(sk, flags);
1854 EXPORT_SYMBOL(udp_disconnect);
1856 void udp_lib_unhash(struct sock *sk)
1858 if (sk_hashed(sk)) {
1859 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1860 struct udp_hslot *hslot, *hslot2;
1862 hslot = udp_hashslot(udptable, sock_net(sk),
1863 udp_sk(sk)->udp_port_hash);
1864 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1866 spin_lock_bh(&hslot->lock);
1867 if (rcu_access_pointer(sk->sk_reuseport_cb))
1868 reuseport_detach_sock(sk);
1869 if (sk_del_node_init_rcu(sk)) {
1871 inet_sk(sk)->inet_num = 0;
1872 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1874 spin_lock(&hslot2->lock);
1875 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1877 spin_unlock(&hslot2->lock);
1879 spin_unlock_bh(&hslot->lock);
1882 EXPORT_SYMBOL(udp_lib_unhash);
1885 * inet_rcv_saddr was changed, we must rehash secondary hash
1887 void udp_lib_rehash(struct sock *sk, u16 newhash)
1889 if (sk_hashed(sk)) {
1890 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1891 struct udp_hslot *hslot, *hslot2, *nhslot2;
1893 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1894 nhslot2 = udp_hashslot2(udptable, newhash);
1895 udp_sk(sk)->udp_portaddr_hash = newhash;
1897 if (hslot2 != nhslot2 ||
1898 rcu_access_pointer(sk->sk_reuseport_cb)) {
1899 hslot = udp_hashslot(udptable, sock_net(sk),
1900 udp_sk(sk)->udp_port_hash);
1901 /* we must lock primary chain too */
1902 spin_lock_bh(&hslot->lock);
1903 if (rcu_access_pointer(sk->sk_reuseport_cb))
1904 reuseport_detach_sock(sk);
1906 if (hslot2 != nhslot2) {
1907 spin_lock(&hslot2->lock);
1908 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1910 spin_unlock(&hslot2->lock);
1912 spin_lock(&nhslot2->lock);
1913 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1916 spin_unlock(&nhslot2->lock);
1919 spin_unlock_bh(&hslot->lock);
1923 EXPORT_SYMBOL(udp_lib_rehash);
1925 static void udp_v4_rehash(struct sock *sk)
1927 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1928 inet_sk(sk)->inet_rcv_saddr,
1929 inet_sk(sk)->inet_num);
1930 udp_lib_rehash(sk, new_hash);
1933 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1937 if (inet_sk(sk)->inet_daddr) {
1938 sock_rps_save_rxhash(sk, skb);
1939 sk_mark_napi_id(sk, skb);
1940 sk_incoming_cpu_update(sk);
1942 sk_mark_napi_id_once(sk, skb);
1945 rc = __udp_enqueue_schedule_skb(sk, skb);
1947 int is_udplite = IS_UDPLITE(sk);
1949 /* Note that an ENOMEM error is charged twice */
1951 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1953 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1955 trace_udp_fail_queue_rcv_skb(rc, sk);
1965 * >0: "udp encap" protocol resubmission
1967 * Note that in the success and error cases, the skb is assumed to
1968 * have either been requeued or freed.
1970 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
1972 struct udp_sock *up = udp_sk(sk);
1973 int is_udplite = IS_UDPLITE(sk);
1976 * Charge it to the socket, dropping if the queue is full.
1978 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1982 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
1983 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1986 * This is an encapsulation socket so pass the skb to
1987 * the socket's udp_encap_rcv() hook. Otherwise, just
1988 * fall through and pass this up the UDP socket.
1989 * up->encap_rcv() returns the following value:
1990 * =0 if skb was successfully passed to the encap
1991 * handler or was discarded by it.
1992 * >0 if skb should be passed on to UDP.
1993 * <0 if skb should be resubmitted as proto -N
1996 /* if we're overly short, let UDP handle it */
1997 encap_rcv = READ_ONCE(up->encap_rcv);
2001 /* Verify checksum before giving to encap */
2002 if (udp_lib_checksum_complete(skb))
2005 ret = encap_rcv(sk, skb);
2007 __UDP_INC_STATS(sock_net(sk),
2008 UDP_MIB_INDATAGRAMS,
2014 /* FALLTHROUGH -- it's a UDP Packet */
2018 * UDP-Lite specific tests, ignored on UDP sockets
2020 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2023 * MIB statistics other than incrementing the error count are
2024 * disabled for the following two types of errors: these depend
2025 * on the application settings, not on the functioning of the
2026 * protocol stack as such.
2028 * RFC 3828 here recommends (sec 3.3): "There should also be a
2029 * way ... to ... at least let the receiving application block
2030 * delivery of packets with coverage values less than a value
2031 * provided by the application."
2033 if (up->pcrlen == 0) { /* full coverage was set */
2034 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2035 UDP_SKB_CB(skb)->cscov, skb->len);
2038 /* The next case involves violating the min. coverage requested
2039 * by the receiver. This is subtle: if receiver wants x and x is
2040 * greater than the buffersize/MTU then receiver will complain
2041 * that it wants x while sender emits packets of smaller size y.
2042 * Therefore the above ...()->partial_cov statement is essential.
2044 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2045 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2046 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2051 prefetch(&sk->sk_rmem_alloc);
2052 if (rcu_access_pointer(sk->sk_filter) &&
2053 udp_lib_checksum_complete(skb))
2056 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2059 udp_csum_pull_header(skb);
2061 ipv4_pktinfo_prepare(sk, skb);
2062 return __udp_queue_rcv_skb(sk, skb);
2065 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2067 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2068 atomic_inc(&sk->sk_drops);
2073 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2075 struct sk_buff *next, *segs;
2078 if (likely(!udp_unexpected_gso(sk, skb)))
2079 return udp_queue_rcv_one_skb(sk, skb);
2081 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET);
2082 __skb_push(skb, -skb_mac_offset(skb));
2083 segs = udp_rcv_segment(sk, skb, true);
2084 for (skb = segs; skb; skb = next) {
2086 __skb_pull(skb, skb_transport_offset(skb));
2087 ret = udp_queue_rcv_one_skb(sk, skb);
2089 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2094 /* For TCP sockets, sk_rx_dst is protected by socket lock
2095 * For UDP, we use xchg() to guard against concurrent changes.
2097 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2099 struct dst_entry *old;
2101 if (dst_hold_safe(dst)) {
2102 old = xchg(&sk->sk_rx_dst, dst);
2108 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2111 * Multicasts and broadcasts go to each listener.
2113 * Note: called only from the BH handler context.
2115 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2117 __be32 saddr, __be32 daddr,
2118 struct udp_table *udptable,
2121 struct sock *sk, *first = NULL;
2122 unsigned short hnum = ntohs(uh->dest);
2123 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2124 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2125 unsigned int offset = offsetof(typeof(*sk), sk_node);
2126 int dif = skb->dev->ifindex;
2127 int sdif = inet_sdif(skb);
2128 struct hlist_node *node;
2129 struct sk_buff *nskb;
2132 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2134 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2136 hslot = &udptable->hash2[hash2];
2137 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2140 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2141 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2142 uh->source, saddr, dif, sdif, hnum))
2149 nskb = skb_clone(skb, GFP_ATOMIC);
2151 if (unlikely(!nskb)) {
2152 atomic_inc(&sk->sk_drops);
2153 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2155 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2159 if (udp_queue_rcv_skb(sk, nskb) > 0)
2163 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2164 if (use_hash2 && hash2 != hash2_any) {
2170 if (udp_queue_rcv_skb(first, skb) > 0)
2174 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2175 proto == IPPROTO_UDPLITE);
2180 /* Initialize UDP checksum. If exited with zero value (success),
2181 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2182 * Otherwise, csum completion requires chacksumming packet body,
2183 * including udp header and folding it to skb->csum.
2185 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2190 UDP_SKB_CB(skb)->partial_cov = 0;
2191 UDP_SKB_CB(skb)->cscov = skb->len;
2193 if (proto == IPPROTO_UDPLITE) {
2194 err = udplite_checksum_init(skb, uh);
2198 if (UDP_SKB_CB(skb)->partial_cov) {
2199 skb->csum = inet_compute_pseudo(skb, proto);
2204 /* Note, we are only interested in != 0 or == 0, thus the
2207 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2208 inet_compute_pseudo);
2212 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2213 /* If SW calculated the value, we know it's bad */
2214 if (skb->csum_complete_sw)
2217 /* HW says the value is bad. Let's validate that.
2218 * skb->csum is no longer the full packet checksum,
2219 * so don't treat it as such.
2221 skb_checksum_complete_unset(skb);
2227 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2228 * return code conversion for ip layer consumption
2230 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2235 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2236 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2237 inet_compute_pseudo);
2239 ret = udp_queue_rcv_skb(sk, skb);
2241 /* a return value > 0 means to resubmit the input, but
2242 * it wants the return to be -protocol, or 0
2250 * All we need to do is get the socket, and then do a checksum.
2253 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2258 unsigned short ulen;
2259 struct rtable *rt = skb_rtable(skb);
2260 __be32 saddr, daddr;
2261 struct net *net = dev_net(skb->dev);
2264 * Validate the packet.
2266 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2267 goto drop; /* No space for header. */
2270 ulen = ntohs(uh->len);
2271 saddr = ip_hdr(skb)->saddr;
2272 daddr = ip_hdr(skb)->daddr;
2274 if (ulen > skb->len)
2277 if (proto == IPPROTO_UDP) {
2278 /* UDP validates ulen. */
2279 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2284 if (udp4_csum_init(skb, uh, proto))
2287 sk = skb_steal_sock(skb);
2289 struct dst_entry *dst = skb_dst(skb);
2292 if (unlikely(sk->sk_rx_dst != dst))
2293 udp_sk_rx_dst_set(sk, dst);
2295 ret = udp_unicast_rcv_skb(sk, skb, uh);
2300 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2301 return __udp4_lib_mcast_deliver(net, skb, uh,
2302 saddr, daddr, udptable, proto);
2304 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2306 return udp_unicast_rcv_skb(sk, skb, uh);
2308 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2312 /* No socket. Drop packet silently, if checksum is wrong */
2313 if (udp_lib_checksum_complete(skb))
2316 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2317 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2320 * Hmm. We got an UDP packet to a port to which we
2321 * don't wanna listen. Ignore it.
2327 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2328 proto == IPPROTO_UDPLITE ? "Lite" : "",
2329 &saddr, ntohs(uh->source),
2331 &daddr, ntohs(uh->dest));
2336 * RFC1122: OK. Discards the bad packet silently (as far as
2337 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2339 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2340 proto == IPPROTO_UDPLITE ? "Lite" : "",
2341 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2343 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2345 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2350 /* We can only early demux multicast if there is a single matching socket.
2351 * If more than one socket found returns NULL
2353 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2354 __be16 loc_port, __be32 loc_addr,
2355 __be16 rmt_port, __be32 rmt_addr,
2358 struct sock *sk, *result;
2359 unsigned short hnum = ntohs(loc_port);
2360 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2361 struct udp_hslot *hslot = &udp_table.hash[slot];
2363 /* Do not bother scanning a too big list */
2364 if (hslot->count > 10)
2368 sk_for_each_rcu(sk, &hslot->head) {
2369 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2370 rmt_port, rmt_addr, dif, sdif, hnum)) {
2380 /* For unicast we should only early demux connected sockets or we can
2381 * break forwarding setups. The chains here can be long so only check
2382 * if the first socket is an exact match and if not move on.
2384 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2385 __be16 loc_port, __be32 loc_addr,
2386 __be16 rmt_port, __be32 rmt_addr,
2389 unsigned short hnum = ntohs(loc_port);
2390 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2391 unsigned int slot2 = hash2 & udp_table.mask;
2392 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2393 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2394 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2397 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2398 if (INET_MATCH(sk, net, acookie, rmt_addr,
2399 loc_addr, ports, dif, sdif))
2401 /* Only check first socket in chain */
2407 int udp_v4_early_demux(struct sk_buff *skb)
2409 struct net *net = dev_net(skb->dev);
2410 struct in_device *in_dev = NULL;
2411 const struct iphdr *iph;
2412 const struct udphdr *uh;
2413 struct sock *sk = NULL;
2414 struct dst_entry *dst;
2415 int dif = skb->dev->ifindex;
2416 int sdif = inet_sdif(skb);
2419 /* validate the packet */
2420 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2426 if (skb->pkt_type == PACKET_MULTICAST) {
2427 in_dev = __in_dev_get_rcu(skb->dev);
2432 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2437 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2438 uh->source, iph->saddr,
2440 } else if (skb->pkt_type == PACKET_HOST) {
2441 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2442 uh->source, iph->saddr, dif, sdif);
2445 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2449 skb->destructor = sock_efree;
2450 dst = READ_ONCE(sk->sk_rx_dst);
2453 dst = dst_check(dst, 0);
2457 /* set noref for now.
2458 * any place which wants to hold dst has to call
2461 skb_dst_set_noref(skb, dst);
2463 /* for unconnected multicast sockets we need to validate
2464 * the source on each packet
2466 if (!inet_sk(sk)->inet_daddr && in_dev)
2467 return ip_mc_validate_source(skb, iph->daddr,
2468 iph->saddr, iph->tos,
2469 skb->dev, in_dev, &itag);
2474 int udp_rcv(struct sk_buff *skb)
2476 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2479 void udp_destroy_sock(struct sock *sk)
2481 struct udp_sock *up = udp_sk(sk);
2482 bool slow = lock_sock_fast(sk);
2483 udp_flush_pending_frames(sk);
2484 unlock_sock_fast(sk, slow);
2485 if (static_branch_unlikely(&udp_encap_needed_key)) {
2486 if (up->encap_type) {
2487 void (*encap_destroy)(struct sock *sk);
2488 encap_destroy = READ_ONCE(up->encap_destroy);
2492 if (up->encap_enabled)
2493 static_branch_disable(&udp_encap_needed_key);
2498 * Socket option code for UDP
2500 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2501 char __user *optval, unsigned int optlen,
2502 int (*push_pending_frames)(struct sock *))
2504 struct udp_sock *up = udp_sk(sk);
2507 int is_udplite = IS_UDPLITE(sk);
2509 if (optlen < sizeof(int))
2512 if (get_user(val, (int __user *)optval))
2515 valbool = val ? 1 : 0;
2524 push_pending_frames(sk);
2532 case UDP_ENCAP_ESPINUDP:
2533 case UDP_ENCAP_ESPINUDP_NON_IKE:
2534 up->encap_rcv = xfrm4_udp_encap_rcv;
2536 case UDP_ENCAP_L2TPINUDP:
2537 up->encap_type = val;
2539 udp_tunnel_encap_enable(sk->sk_socket);
2548 case UDP_NO_CHECK6_TX:
2549 up->no_check6_tx = valbool;
2552 case UDP_NO_CHECK6_RX:
2553 up->no_check6_rx = valbool;
2557 if (val < 0 || val > USHRT_MAX)
2565 udp_tunnel_encap_enable(sk->sk_socket);
2566 up->gro_enabled = valbool;
2571 * UDP-Lite's partial checksum coverage (RFC 3828).
2573 /* The sender sets actual checksum coverage length via this option.
2574 * The case coverage > packet length is handled by send module. */
2575 case UDPLITE_SEND_CSCOV:
2576 if (!is_udplite) /* Disable the option on UDP sockets */
2577 return -ENOPROTOOPT;
2578 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2580 else if (val > USHRT_MAX)
2583 up->pcflag |= UDPLITE_SEND_CC;
2586 /* The receiver specifies a minimum checksum coverage value. To make
2587 * sense, this should be set to at least 8 (as done below). If zero is
2588 * used, this again means full checksum coverage. */
2589 case UDPLITE_RECV_CSCOV:
2590 if (!is_udplite) /* Disable the option on UDP sockets */
2591 return -ENOPROTOOPT;
2592 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2594 else if (val > USHRT_MAX)
2597 up->pcflag |= UDPLITE_RECV_CC;
2607 EXPORT_SYMBOL(udp_lib_setsockopt);
2609 int udp_setsockopt(struct sock *sk, int level, int optname,
2610 char __user *optval, unsigned int optlen)
2612 if (level == SOL_UDP || level == SOL_UDPLITE)
2613 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2614 udp_push_pending_frames);
2615 return ip_setsockopt(sk, level, optname, optval, optlen);
2618 #ifdef CONFIG_COMPAT
2619 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2620 char __user *optval, unsigned int optlen)
2622 if (level == SOL_UDP || level == SOL_UDPLITE)
2623 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2624 udp_push_pending_frames);
2625 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2629 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2630 char __user *optval, int __user *optlen)
2632 struct udp_sock *up = udp_sk(sk);
2635 if (get_user(len, optlen))
2638 len = min_t(unsigned int, len, sizeof(int));
2649 val = up->encap_type;
2652 case UDP_NO_CHECK6_TX:
2653 val = up->no_check6_tx;
2656 case UDP_NO_CHECK6_RX:
2657 val = up->no_check6_rx;
2664 /* The following two cannot be changed on UDP sockets, the return is
2665 * always 0 (which corresponds to the full checksum coverage of UDP). */
2666 case UDPLITE_SEND_CSCOV:
2670 case UDPLITE_RECV_CSCOV:
2675 return -ENOPROTOOPT;
2678 if (put_user(len, optlen))
2680 if (copy_to_user(optval, &val, len))
2684 EXPORT_SYMBOL(udp_lib_getsockopt);
2686 int udp_getsockopt(struct sock *sk, int level, int optname,
2687 char __user *optval, int __user *optlen)
2689 if (level == SOL_UDP || level == SOL_UDPLITE)
2690 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2691 return ip_getsockopt(sk, level, optname, optval, optlen);
2694 #ifdef CONFIG_COMPAT
2695 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2696 char __user *optval, int __user *optlen)
2698 if (level == SOL_UDP || level == SOL_UDPLITE)
2699 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2700 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2704 * udp_poll - wait for a UDP event.
2705 * @file - file struct
2707 * @wait - poll table
2709 * This is same as datagram poll, except for the special case of
2710 * blocking sockets. If application is using a blocking fd
2711 * and a packet with checksum error is in the queue;
2712 * then it could get return from select indicating data available
2713 * but then block when reading it. Add special case code
2714 * to work around these arguably broken applications.
2716 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2718 __poll_t mask = datagram_poll(file, sock, wait);
2719 struct sock *sk = sock->sk;
2721 if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2722 mask |= EPOLLIN | EPOLLRDNORM;
2724 /* Check for false positives due to checksum errors */
2725 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2726 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2727 mask &= ~(EPOLLIN | EPOLLRDNORM);
2732 EXPORT_SYMBOL(udp_poll);
2734 int udp_abort(struct sock *sk, int err)
2739 sk->sk_error_report(sk);
2740 __udp_disconnect(sk, 0);
2746 EXPORT_SYMBOL_GPL(udp_abort);
2748 struct proto udp_prot = {
2750 .owner = THIS_MODULE,
2751 .close = udp_lib_close,
2752 .pre_connect = udp_pre_connect,
2753 .connect = ip4_datagram_connect,
2754 .disconnect = udp_disconnect,
2756 .init = udp_init_sock,
2757 .destroy = udp_destroy_sock,
2758 .setsockopt = udp_setsockopt,
2759 .getsockopt = udp_getsockopt,
2760 .sendmsg = udp_sendmsg,
2761 .recvmsg = udp_recvmsg,
2762 .sendpage = udp_sendpage,
2763 .release_cb = ip4_datagram_release_cb,
2764 .hash = udp_lib_hash,
2765 .unhash = udp_lib_unhash,
2766 .rehash = udp_v4_rehash,
2767 .get_port = udp_v4_get_port,
2768 .memory_allocated = &udp_memory_allocated,
2769 .sysctl_mem = sysctl_udp_mem,
2770 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2771 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2772 .obj_size = sizeof(struct udp_sock),
2773 .h.udp_table = &udp_table,
2774 #ifdef CONFIG_COMPAT
2775 .compat_setsockopt = compat_udp_setsockopt,
2776 .compat_getsockopt = compat_udp_getsockopt,
2778 .diag_destroy = udp_abort,
2780 EXPORT_SYMBOL(udp_prot);
2782 /* ------------------------------------------------------------------------ */
2783 #ifdef CONFIG_PROC_FS
2785 static struct sock *udp_get_first(struct seq_file *seq, int start)
2788 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2789 struct udp_iter_state *state = seq->private;
2790 struct net *net = seq_file_net(seq);
2792 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2794 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2796 if (hlist_empty(&hslot->head))
2799 spin_lock_bh(&hslot->lock);
2800 sk_for_each(sk, &hslot->head) {
2801 if (!net_eq(sock_net(sk), net))
2803 if (sk->sk_family == afinfo->family)
2806 spin_unlock_bh(&hslot->lock);
2813 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2815 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2816 struct udp_iter_state *state = seq->private;
2817 struct net *net = seq_file_net(seq);
2821 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2824 if (state->bucket <= afinfo->udp_table->mask)
2825 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2826 return udp_get_first(seq, state->bucket + 1);
2831 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2833 struct sock *sk = udp_get_first(seq, 0);
2836 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2838 return pos ? NULL : sk;
2841 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2843 struct udp_iter_state *state = seq->private;
2844 state->bucket = MAX_UDP_PORTS;
2846 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2848 EXPORT_SYMBOL(udp_seq_start);
2850 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2854 if (v == SEQ_START_TOKEN)
2855 sk = udp_get_idx(seq, 0);
2857 sk = udp_get_next(seq, v);
2862 EXPORT_SYMBOL(udp_seq_next);
2864 void udp_seq_stop(struct seq_file *seq, void *v)
2866 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2867 struct udp_iter_state *state = seq->private;
2869 if (state->bucket <= afinfo->udp_table->mask)
2870 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2872 EXPORT_SYMBOL(udp_seq_stop);
2874 /* ------------------------------------------------------------------------ */
2875 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2878 struct inet_sock *inet = inet_sk(sp);
2879 __be32 dest = inet->inet_daddr;
2880 __be32 src = inet->inet_rcv_saddr;
2881 __u16 destp = ntohs(inet->inet_dport);
2882 __u16 srcp = ntohs(inet->inet_sport);
2884 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2885 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2886 bucket, src, srcp, dest, destp, sp->sk_state,
2887 sk_wmem_alloc_get(sp),
2890 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2892 refcount_read(&sp->sk_refcnt), sp,
2893 atomic_read(&sp->sk_drops));
2896 int udp4_seq_show(struct seq_file *seq, void *v)
2898 seq_setwidth(seq, 127);
2899 if (v == SEQ_START_TOKEN)
2900 seq_puts(seq, " sl local_address rem_address st tx_queue "
2901 "rx_queue tr tm->when retrnsmt uid timeout "
2902 "inode ref pointer drops");
2904 struct udp_iter_state *state = seq->private;
2906 udp4_format_sock(v, seq, state->bucket);
2912 const struct seq_operations udp_seq_ops = {
2913 .start = udp_seq_start,
2914 .next = udp_seq_next,
2915 .stop = udp_seq_stop,
2916 .show = udp4_seq_show,
2918 EXPORT_SYMBOL(udp_seq_ops);
2920 static struct udp_seq_afinfo udp4_seq_afinfo = {
2922 .udp_table = &udp_table,
2925 static int __net_init udp4_proc_init_net(struct net *net)
2927 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2928 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2933 static void __net_exit udp4_proc_exit_net(struct net *net)
2935 remove_proc_entry("udp", net->proc_net);
2938 static struct pernet_operations udp4_net_ops = {
2939 .init = udp4_proc_init_net,
2940 .exit = udp4_proc_exit_net,
2943 int __init udp4_proc_init(void)
2945 return register_pernet_subsys(&udp4_net_ops);
2948 void udp4_proc_exit(void)
2950 unregister_pernet_subsys(&udp4_net_ops);
2952 #endif /* CONFIG_PROC_FS */
2954 static __initdata unsigned long uhash_entries;
2955 static int __init set_uhash_entries(char *str)
2962 ret = kstrtoul(str, 0, &uhash_entries);
2966 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2967 uhash_entries = UDP_HTABLE_SIZE_MIN;
2970 __setup("uhash_entries=", set_uhash_entries);
2972 void __init udp_table_init(struct udp_table *table, const char *name)
2976 table->hash = alloc_large_system_hash(name,
2977 2 * sizeof(struct udp_hslot),
2979 21, /* one slot per 2 MB */
2983 UDP_HTABLE_SIZE_MIN,
2986 table->hash2 = table->hash + (table->mask + 1);
2987 for (i = 0; i <= table->mask; i++) {
2988 INIT_HLIST_HEAD(&table->hash[i].head);
2989 table->hash[i].count = 0;
2990 spin_lock_init(&table->hash[i].lock);
2992 for (i = 0; i <= table->mask; i++) {
2993 INIT_HLIST_HEAD(&table->hash2[i].head);
2994 table->hash2[i].count = 0;
2995 spin_lock_init(&table->hash2[i].lock);
2999 u32 udp_flow_hashrnd(void)
3001 static u32 hashrnd __read_mostly;
3003 net_get_random_once(&hashrnd, sizeof(hashrnd));
3007 EXPORT_SYMBOL(udp_flow_hashrnd);
3009 static void __udp_sysctl_init(struct net *net)
3011 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3012 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3014 #ifdef CONFIG_NET_L3_MASTER_DEV
3015 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3019 static int __net_init udp_sysctl_init(struct net *net)
3021 __udp_sysctl_init(net);
3025 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3026 .init = udp_sysctl_init,
3029 void __init udp_init(void)
3031 unsigned long limit;
3034 udp_table_init(&udp_table, "UDP");
3035 limit = nr_free_buffer_pages() / 8;
3036 limit = max(limit, 128UL);
3037 sysctl_udp_mem[0] = limit / 4 * 3;
3038 sysctl_udp_mem[1] = limit;
3039 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3041 __udp_sysctl_init(&init_net);
3043 /* 16 spinlocks per cpu */
3044 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3045 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3048 panic("UDP: failed to alloc udp_busylocks\n");
3049 for (i = 0; i < (1U << udp_busylocks_log); i++)
3050 spin_lock_init(udp_busylocks + i);
3052 if (register_pernet_subsys(&udp_sysctl_ops))
3053 panic("UDP: failed to init sysctl parameters.\n");