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 <asm/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/bootmem.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>
94 #include <linux/errno.h>
95 #include <linux/timer.h>
97 #include <linux/inet.h>
98 #include <linux/netdevice.h>
99 #include <linux/slab.h>
100 #include <net/tcp_states.h>
101 #include <linux/skbuff.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <net/net_namespace.h>
105 #include <net/icmp.h>
106 #include <net/inet_hashtables.h>
107 #include <net/route.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <trace/events/udp.h>
111 #include <linux/static_key.h>
112 #include <trace/events/skb.h>
113 #include <net/busy_poll.h>
114 #include "udp_impl.h"
116 struct udp_table udp_table __read_mostly;
117 EXPORT_SYMBOL(udp_table);
119 long sysctl_udp_mem[3] __read_mostly;
120 EXPORT_SYMBOL(sysctl_udp_mem);
122 int sysctl_udp_rmem_min __read_mostly;
123 EXPORT_SYMBOL(sysctl_udp_rmem_min);
125 int sysctl_udp_wmem_min __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_wmem_min);
128 atomic_long_t udp_memory_allocated;
129 EXPORT_SYMBOL(udp_memory_allocated);
131 #define MAX_UDP_PORTS 65536
132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
134 static int udp_lib_lport_inuse(struct net *net, __u16 num,
135 const struct udp_hslot *hslot,
136 unsigned long *bitmap,
138 int (*saddr_comp)(const struct sock *sk1,
139 const struct sock *sk2),
143 struct hlist_nulls_node *node;
144 kuid_t uid = sock_i_uid(sk);
146 sk_nulls_for_each(sk2, node, &hslot->head)
147 if (net_eq(sock_net(sk2), net) &&
149 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
150 (!sk2->sk_reuse || !sk->sk_reuse) &&
151 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
152 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
153 (!sk2->sk_reuseport || !sk->sk_reuseport ||
154 !uid_eq(uid, sock_i_uid(sk2))) &&
155 (*saddr_comp)(sk, sk2)) {
157 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
166 * Note: we still hold spinlock of primary hash chain, so no other writer
167 * can insert/delete a socket with local_port == num
169 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
170 struct udp_hslot *hslot2,
172 int (*saddr_comp)(const struct sock *sk1,
173 const struct sock *sk2))
176 struct hlist_nulls_node *node;
177 kuid_t uid = sock_i_uid(sk);
180 spin_lock(&hslot2->lock);
181 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
182 if (net_eq(sock_net(sk2), net) &&
184 (udp_sk(sk2)->udp_port_hash == num) &&
185 (!sk2->sk_reuse || !sk->sk_reuse) &&
186 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
187 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
188 (!sk2->sk_reuseport || !sk->sk_reuseport ||
189 !uid_eq(uid, sock_i_uid(sk2))) &&
190 (*saddr_comp)(sk, sk2)) {
194 spin_unlock(&hslot2->lock);
199 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
201 * @sk: socket struct in question
202 * @snum: port number to look up
203 * @saddr_comp: AF-dependent comparison of bound local IP addresses
204 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
207 int udp_lib_get_port(struct sock *sk, unsigned short snum,
208 int (*saddr_comp)(const struct sock *sk1,
209 const struct sock *sk2),
210 unsigned int hash2_nulladdr)
212 struct udp_hslot *hslot, *hslot2;
213 struct udp_table *udptable = sk->sk_prot->h.udp_table;
215 struct net *net = sock_net(sk);
218 int low, high, remaining;
220 unsigned short first, last;
221 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
223 inet_get_local_port_range(net, &low, &high);
224 remaining = (high - low) + 1;
227 first = (((u64)rand * remaining) >> 32) + low;
229 * force rand to be an odd multiple of UDP_HTABLE_SIZE
231 rand = (rand | 1) * (udptable->mask + 1);
232 last = first + udptable->mask + 1;
234 hslot = udp_hashslot(udptable, net, first);
235 bitmap_zero(bitmap, PORTS_PER_CHAIN);
236 spin_lock_bh(&hslot->lock);
237 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
238 saddr_comp, udptable->log);
242 * Iterate on all possible values of snum for this hash.
243 * Using steps of an odd multiple of UDP_HTABLE_SIZE
244 * give us randomization and full range coverage.
247 if (low <= snum && snum <= high &&
248 !test_bit(snum >> udptable->log, bitmap) &&
249 !inet_is_reserved_local_port(snum))
252 } while (snum != first);
253 spin_unlock_bh(&hslot->lock);
254 } while (++first != last);
257 hslot = udp_hashslot(udptable, net, snum);
258 spin_lock_bh(&hslot->lock);
259 if (hslot->count > 10) {
261 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
263 slot2 &= udptable->mask;
264 hash2_nulladdr &= udptable->mask;
266 hslot2 = udp_hashslot2(udptable, slot2);
267 if (hslot->count < hslot2->count)
268 goto scan_primary_hash;
270 exist = udp_lib_lport_inuse2(net, snum, hslot2,
272 if (!exist && (hash2_nulladdr != slot2)) {
273 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
274 exist = udp_lib_lport_inuse2(net, snum, hslot2,
283 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
288 inet_sk(sk)->inet_num = snum;
289 udp_sk(sk)->udp_port_hash = snum;
290 udp_sk(sk)->udp_portaddr_hash ^= snum;
291 if (sk_unhashed(sk)) {
292 sk_nulls_add_node_rcu(sk, &hslot->head);
294 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
296 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
297 spin_lock(&hslot2->lock);
298 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
301 spin_unlock(&hslot2->lock);
305 spin_unlock_bh(&hslot->lock);
309 EXPORT_SYMBOL(udp_lib_get_port);
311 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
313 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
315 return (!ipv6_only_sock(sk2) &&
316 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
317 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
320 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
323 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
326 int udp_v4_get_port(struct sock *sk, unsigned short snum)
328 unsigned int hash2_nulladdr =
329 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
330 unsigned int hash2_partial =
331 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
333 /* precompute partial secondary hash */
334 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
335 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
338 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
340 __be16 sport, __be32 daddr, __be16 dport, int dif)
344 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
345 !ipv6_only_sock(sk)) {
346 struct inet_sock *inet = inet_sk(sk);
348 score = (sk->sk_family == PF_INET ? 2 : 1);
349 if (inet->inet_rcv_saddr) {
350 if (inet->inet_rcv_saddr != daddr)
354 if (inet->inet_daddr) {
355 if (inet->inet_daddr != saddr)
359 if (inet->inet_dport) {
360 if (inet->inet_dport != sport)
364 if (sk->sk_bound_dev_if) {
365 if (sk->sk_bound_dev_if != dif)
374 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
376 static inline int compute_score2(struct sock *sk, struct net *net,
377 __be32 saddr, __be16 sport,
378 __be32 daddr, unsigned int hnum, int dif)
382 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
383 struct inet_sock *inet = inet_sk(sk);
385 if (inet->inet_rcv_saddr != daddr)
387 if (inet->inet_num != hnum)
390 score = (sk->sk_family == PF_INET ? 2 : 1);
391 if (inet->inet_daddr) {
392 if (inet->inet_daddr != saddr)
396 if (inet->inet_dport) {
397 if (inet->inet_dport != sport)
401 if (sk->sk_bound_dev_if) {
402 if (sk->sk_bound_dev_if != dif)
410 static unsigned int udp_ehashfn(struct net *net, const __be32 laddr,
411 const __u16 lport, const __be32 faddr,
414 static u32 udp_ehash_secret __read_mostly;
416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
418 return __inet_ehashfn(laddr, lport, faddr, fport,
419 udp_ehash_secret + net_hash_mix(net));
423 /* called with read_rcu_lock() */
424 static struct sock *udp4_lib_lookup2(struct net *net,
425 __be32 saddr, __be16 sport,
426 __be32 daddr, unsigned int hnum, int dif,
427 struct udp_hslot *hslot2, unsigned int slot2)
429 struct sock *sk, *result;
430 struct hlist_nulls_node *node;
431 int score, badness, matches = 0, reuseport = 0;
437 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
438 score = compute_score2(sk, net, saddr, sport,
440 if (score > badness) {
443 reuseport = sk->sk_reuseport;
445 hash = udp_ehashfn(net, daddr, hnum,
449 } else if (score == badness && reuseport) {
451 if (((u64)hash * matches) >> 32 == 0)
453 hash = next_pseudo_random32(hash);
457 * if the nulls value we got at the end of this lookup is
458 * not the expected one, we must restart lookup.
459 * We probably met an item that was moved to another chain.
461 if (get_nulls_value(node) != slot2)
464 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
466 else if (unlikely(compute_score2(result, net, saddr, sport,
467 daddr, hnum, dif) < badness)) {
475 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
476 * harder than this. -DaveM
478 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
479 __be16 sport, __be32 daddr, __be16 dport,
480 int dif, struct udp_table *udptable)
482 struct sock *sk, *result;
483 struct hlist_nulls_node *node;
484 unsigned short hnum = ntohs(dport);
485 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
486 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
487 int score, badness, matches = 0, reuseport = 0;
491 if (hslot->count > 10) {
492 hash2 = udp4_portaddr_hash(net, daddr, hnum);
493 slot2 = hash2 & udptable->mask;
494 hslot2 = &udptable->hash2[slot2];
495 if (hslot->count < hslot2->count)
498 result = udp4_lib_lookup2(net, saddr, sport,
502 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
503 slot2 = hash2 & udptable->mask;
504 hslot2 = &udptable->hash2[slot2];
505 if (hslot->count < hslot2->count)
508 result = udp4_lib_lookup2(net, saddr, sport,
509 htonl(INADDR_ANY), hnum, dif,
518 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
519 score = compute_score(sk, net, saddr, hnum, sport,
521 if (score > badness) {
524 reuseport = sk->sk_reuseport;
526 hash = udp_ehashfn(net, daddr, hnum,
530 } else if (score == badness && reuseport) {
532 if (((u64)hash * matches) >> 32 == 0)
534 hash = next_pseudo_random32(hash);
538 * if the nulls value we got at the end of this lookup is
539 * not the expected one, we must restart lookup.
540 * We probably met an item that was moved to another chain.
542 if (get_nulls_value(node) != slot)
546 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
548 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
549 daddr, dport, dif) < badness)) {
557 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
559 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
560 __be16 sport, __be16 dport,
561 struct udp_table *udptable)
564 const struct iphdr *iph = ip_hdr(skb);
566 if (unlikely(sk = skb_steal_sock(skb)))
569 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
570 iph->daddr, dport, inet_iif(skb),
574 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
575 __be32 daddr, __be16 dport, int dif)
577 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
579 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
581 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
582 __be16 loc_port, __be32 loc_addr,
583 __be16 rmt_port, __be32 rmt_addr,
584 int dif, unsigned short hnum)
586 struct inet_sock *inet = inet_sk(sk);
588 if (!net_eq(sock_net(sk), net) ||
589 udp_sk(sk)->udp_port_hash != hnum ||
590 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
591 (inet->inet_dport != rmt_port && inet->inet_dport) ||
592 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
593 ipv6_only_sock(sk) ||
594 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
596 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
601 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
602 __be16 loc_port, __be32 loc_addr,
603 __be16 rmt_port, __be32 rmt_addr,
606 struct hlist_nulls_node *node;
608 unsigned short hnum = ntohs(loc_port);
610 sk_nulls_for_each_from(s, node) {
611 if (__udp_is_mcast_sock(net, s,
623 * This routine is called by the ICMP module when it gets some
624 * sort of error condition. If err < 0 then the socket should
625 * be closed and the error returned to the user. If err > 0
626 * it's just the icmp type << 8 | icmp code.
627 * Header points to the ip header of the error packet. We move
628 * on past this. Then (as it used to claim before adjustment)
629 * header points to the first 8 bytes of the udp header. We need
630 * to find the appropriate port.
633 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
635 struct inet_sock *inet;
636 const struct iphdr *iph = (const struct iphdr *)skb->data;
637 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
638 const int type = icmp_hdr(skb)->type;
639 const int code = icmp_hdr(skb)->code;
643 struct net *net = dev_net(skb->dev);
645 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
646 iph->saddr, uh->source, skb->dev->ifindex, udptable);
648 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
649 return; /* No socket for error */
658 case ICMP_TIME_EXCEEDED:
661 case ICMP_SOURCE_QUENCH:
663 case ICMP_PARAMETERPROB:
667 case ICMP_DEST_UNREACH:
668 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
669 ipv4_sk_update_pmtu(skb, sk, info);
670 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
678 if (code <= NR_ICMP_UNREACH) {
679 harderr = icmp_err_convert[code].fatal;
680 err = icmp_err_convert[code].errno;
684 ipv4_sk_redirect(skb, sk);
689 * RFC1122: OK. Passes ICMP errors back to application, as per
692 if (!inet->recverr) {
693 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
696 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
699 sk->sk_error_report(sk);
704 void udp_err(struct sk_buff *skb, u32 info)
706 __udp4_lib_err(skb, info, &udp_table);
710 * Throw away all pending data and cancel the corking. Socket is locked.
712 void udp_flush_pending_frames(struct sock *sk)
714 struct udp_sock *up = udp_sk(sk);
719 ip_flush_pending_frames(sk);
722 EXPORT_SYMBOL(udp_flush_pending_frames);
725 * udp4_hwcsum - handle outgoing HW checksumming
726 * @skb: sk_buff containing the filled-in UDP header
727 * (checksum field must be zeroed out)
728 * @src: source IP address
729 * @dst: destination IP address
731 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
733 struct udphdr *uh = udp_hdr(skb);
734 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
735 int offset = skb_transport_offset(skb);
736 int len = skb->len - offset;
742 * Only one fragment on the socket.
744 skb->csum_start = skb_transport_header(skb) - skb->head;
745 skb->csum_offset = offsetof(struct udphdr, check);
746 uh->check = ~csum_tcpudp_magic(src, dst, len,
750 * HW-checksum won't work as there are two or more
751 * fragments on the socket so that all csums of sk_buffs
755 csum = csum_add(csum, frags->csum);
757 } while ((frags = frags->next));
759 csum = skb_checksum(skb, offset, hlen, csum);
760 skb->ip_summed = CHECKSUM_NONE;
762 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
764 uh->check = CSUM_MANGLED_0;
767 EXPORT_SYMBOL_GPL(udp4_hwcsum);
769 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
771 struct sock *sk = skb->sk;
772 struct inet_sock *inet = inet_sk(sk);
775 int is_udplite = IS_UDPLITE(sk);
776 int offset = skb_transport_offset(skb);
777 int len = skb->len - offset;
781 * Create a UDP header
784 uh->source = inet->inet_sport;
785 uh->dest = fl4->fl4_dport;
786 uh->len = htons(len);
789 if (is_udplite) /* UDP-Lite */
790 csum = udplite_csum(skb);
792 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
794 skb->ip_summed = CHECKSUM_NONE;
797 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
799 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
803 csum = udp_csum(skb);
805 /* add protocol-dependent pseudo-header */
806 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
807 sk->sk_protocol, csum);
809 uh->check = CSUM_MANGLED_0;
812 err = ip_send_skb(sock_net(sk), skb);
814 if (err == -ENOBUFS && !inet->recverr) {
815 UDP_INC_STATS_USER(sock_net(sk),
816 UDP_MIB_SNDBUFERRORS, is_udplite);
820 UDP_INC_STATS_USER(sock_net(sk),
821 UDP_MIB_OUTDATAGRAMS, is_udplite);
826 * Push out all pending data as one UDP datagram. Socket is locked.
828 int udp_push_pending_frames(struct sock *sk)
830 struct udp_sock *up = udp_sk(sk);
831 struct inet_sock *inet = inet_sk(sk);
832 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
836 skb = ip_finish_skb(sk, fl4);
840 err = udp_send_skb(skb, fl4);
847 EXPORT_SYMBOL(udp_push_pending_frames);
849 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
852 struct inet_sock *inet = inet_sk(sk);
853 struct udp_sock *up = udp_sk(sk);
854 struct flowi4 fl4_stack;
857 struct ipcm_cookie ipc;
858 struct rtable *rt = NULL;
861 __be32 daddr, faddr, saddr;
864 int err, is_udplite = IS_UDPLITE(sk);
865 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
866 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
868 struct ip_options_data opt_copy;
877 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
885 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
887 fl4 = &inet->cork.fl.u.ip4;
890 * There are pending frames.
891 * The socket lock must be held while it's corked.
894 if (likely(up->pending)) {
895 if (unlikely(up->pending != AF_INET)) {
903 ulen += sizeof(struct udphdr);
906 * Get and verify the address.
909 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
910 if (msg->msg_namelen < sizeof(*usin))
912 if (usin->sin_family != AF_INET) {
913 if (usin->sin_family != AF_UNSPEC)
914 return -EAFNOSUPPORT;
917 daddr = usin->sin_addr.s_addr;
918 dport = usin->sin_port;
922 if (sk->sk_state != TCP_ESTABLISHED)
923 return -EDESTADDRREQ;
924 daddr = inet->inet_daddr;
925 dport = inet->inet_dport;
926 /* Open fast path for connected socket.
927 Route will not be used, if at least one option is set.
931 ipc.addr = inet->inet_saddr;
933 ipc.oif = sk->sk_bound_dev_if;
935 sock_tx_timestamp(sk, &ipc.tx_flags);
937 if (msg->msg_controllen) {
938 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
946 struct ip_options_rcu *inet_opt;
949 inet_opt = rcu_dereference(inet->inet_opt);
951 memcpy(&opt_copy, inet_opt,
952 sizeof(*inet_opt) + inet_opt->opt.optlen);
953 ipc.opt = &opt_copy.opt;
959 ipc.addr = faddr = daddr;
961 if (ipc.opt && ipc.opt->opt.srr) {
964 faddr = ipc.opt->opt.faddr;
967 tos = get_rttos(&ipc, inet);
968 if (sock_flag(sk, SOCK_LOCALROUTE) ||
969 (msg->msg_flags & MSG_DONTROUTE) ||
970 (ipc.opt && ipc.opt->opt.is_strictroute)) {
975 if (ipv4_is_multicast(daddr)) {
977 ipc.oif = inet->mc_index;
979 saddr = inet->mc_addr;
982 ipc.oif = inet->uc_index;
985 rt = (struct rtable *)sk_dst_check(sk, 0);
988 struct net *net = sock_net(sk);
991 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
992 RT_SCOPE_UNIVERSE, sk->sk_protocol,
993 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
994 faddr, saddr, dport, inet->inet_sport);
996 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
997 rt = ip_route_output_flow(net, fl4, sk);
1001 if (err == -ENETUNREACH)
1002 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
1007 if ((rt->rt_flags & RTCF_BROADCAST) &&
1008 !sock_flag(sk, SOCK_BROADCAST))
1011 sk_dst_set(sk, dst_clone(&rt->dst));
1014 if (msg->msg_flags&MSG_CONFIRM)
1020 daddr = ipc.addr = fl4->daddr;
1022 /* Lockless fast path for the non-corking case. */
1024 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
1025 sizeof(struct udphdr), &ipc, &rt,
1028 if (!IS_ERR_OR_NULL(skb))
1029 err = udp_send_skb(skb, fl4);
1034 if (unlikely(up->pending)) {
1035 /* The socket is already corked while preparing it. */
1036 /* ... which is an evident application bug. --ANK */
1039 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1044 * Now cork the socket to pend data.
1046 fl4 = &inet->cork.fl.u.ip4;
1049 fl4->fl4_dport = dport;
1050 fl4->fl4_sport = inet->inet_sport;
1051 up->pending = AF_INET;
1055 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1056 sizeof(struct udphdr), &ipc, &rt,
1057 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1059 udp_flush_pending_frames(sk);
1061 err = udp_push_pending_frames(sk);
1062 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1073 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1074 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1075 * we don't have a good statistic (IpOutDiscards but it can be too many
1076 * things). We could add another new stat but at least for now that
1077 * seems like overkill.
1079 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1080 UDP_INC_STATS_USER(sock_net(sk),
1081 UDP_MIB_SNDBUFERRORS, is_udplite);
1086 dst_confirm(&rt->dst);
1087 if (!(msg->msg_flags&MSG_PROBE) || len)
1088 goto back_from_confirm;
1092 EXPORT_SYMBOL(udp_sendmsg);
1094 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1095 size_t size, int flags)
1097 struct inet_sock *inet = inet_sk(sk);
1098 struct udp_sock *up = udp_sk(sk);
1101 if (flags & MSG_SENDPAGE_NOTLAST)
1105 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1107 /* Call udp_sendmsg to specify destination address which
1108 * sendpage interface can't pass.
1109 * This will succeed only when the socket is connected.
1111 ret = udp_sendmsg(NULL, sk, &msg, 0);
1118 if (unlikely(!up->pending)) {
1121 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1125 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1126 page, offset, size, flags);
1127 if (ret == -EOPNOTSUPP) {
1129 return sock_no_sendpage(sk->sk_socket, page, offset,
1133 udp_flush_pending_frames(sk);
1138 if (!(up->corkflag || (flags&MSG_MORE)))
1139 ret = udp_push_pending_frames(sk);
1149 * first_packet_length - return length of first packet in receive queue
1152 * Drops all bad checksum frames, until a valid one is found.
1153 * Returns the length of found skb, or 0 if none is found.
1155 static unsigned int first_packet_length(struct sock *sk)
1157 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1158 struct sk_buff *skb;
1161 __skb_queue_head_init(&list_kill);
1163 spin_lock_bh(&rcvq->lock);
1164 while ((skb = skb_peek(rcvq)) != NULL &&
1165 udp_lib_checksum_complete(skb)) {
1166 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1168 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1170 atomic_inc(&sk->sk_drops);
1171 __skb_unlink(skb, rcvq);
1172 __skb_queue_tail(&list_kill, skb);
1174 res = skb ? skb->len : 0;
1175 spin_unlock_bh(&rcvq->lock);
1177 if (!skb_queue_empty(&list_kill)) {
1178 bool slow = lock_sock_fast(sk);
1180 __skb_queue_purge(&list_kill);
1181 sk_mem_reclaim_partial(sk);
1182 unlock_sock_fast(sk, slow);
1188 * IOCTL requests applicable to the UDP protocol
1191 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1196 int amount = sk_wmem_alloc_get(sk);
1198 return put_user(amount, (int __user *)arg);
1203 unsigned int amount = first_packet_length(sk);
1207 * We will only return the amount
1208 * of this packet since that is all
1209 * that will be read.
1211 amount -= sizeof(struct udphdr);
1213 return put_user(amount, (int __user *)arg);
1217 return -ENOIOCTLCMD;
1222 EXPORT_SYMBOL(udp_ioctl);
1225 * This should be easy, if there is something there we
1226 * return it, otherwise we block.
1229 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1230 size_t len, int noblock, int flags, int *addr_len)
1232 struct inet_sock *inet = inet_sk(sk);
1233 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1234 struct sk_buff *skb;
1235 unsigned int ulen, copied;
1236 int peeked, off = 0;
1238 int is_udplite = IS_UDPLITE(sk);
1241 if (flags & MSG_ERRQUEUE)
1242 return ip_recv_error(sk, msg, len, addr_len);
1245 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1246 &peeked, &off, &err);
1250 ulen = skb->len - sizeof(struct udphdr);
1254 else if (copied < ulen)
1255 msg->msg_flags |= MSG_TRUNC;
1258 * If checksum is needed at all, try to do it while copying the
1259 * data. If the data is truncated, or if we only want a partial
1260 * coverage checksum (UDP-Lite), do it before the copy.
1263 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1264 if (udp_lib_checksum_complete(skb))
1268 if (skb_csum_unnecessary(skb))
1269 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1270 msg->msg_iov, copied);
1272 err = skb_copy_and_csum_datagram_iovec(skb,
1273 sizeof(struct udphdr),
1280 if (unlikely(err)) {
1281 trace_kfree_skb(skb, udp_recvmsg);
1283 atomic_inc(&sk->sk_drops);
1284 UDP_INC_STATS_USER(sock_net(sk),
1285 UDP_MIB_INERRORS, is_udplite);
1291 UDP_INC_STATS_USER(sock_net(sk),
1292 UDP_MIB_INDATAGRAMS, is_udplite);
1294 sock_recv_ts_and_drops(msg, sk, skb);
1296 /* Copy the address. */
1298 sin->sin_family = AF_INET;
1299 sin->sin_port = udp_hdr(skb)->source;
1300 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1301 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1302 *addr_len = sizeof(*sin);
1304 if (inet->cmsg_flags)
1305 ip_cmsg_recv(msg, skb);
1308 if (flags & MSG_TRUNC)
1312 skb_free_datagram_locked(sk, skb);
1317 slow = lock_sock_fast(sk);
1318 if (!skb_kill_datagram(sk, skb, flags)) {
1319 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1320 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1322 unlock_sock_fast(sk, slow);
1327 /* starting over for a new packet */
1328 msg->msg_flags &= ~MSG_TRUNC;
1333 int udp_disconnect(struct sock *sk, int flags)
1335 struct inet_sock *inet = inet_sk(sk);
1337 * 1003.1g - break association.
1340 sk->sk_state = TCP_CLOSE;
1341 inet->inet_daddr = 0;
1342 inet->inet_dport = 0;
1343 sock_rps_reset_rxhash(sk);
1344 sk->sk_bound_dev_if = 0;
1345 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1346 inet_reset_saddr(sk);
1348 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1349 sk->sk_prot->unhash(sk);
1350 inet->inet_sport = 0;
1355 EXPORT_SYMBOL(udp_disconnect);
1357 void udp_lib_unhash(struct sock *sk)
1359 if (sk_hashed(sk)) {
1360 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1361 struct udp_hslot *hslot, *hslot2;
1363 hslot = udp_hashslot(udptable, sock_net(sk),
1364 udp_sk(sk)->udp_port_hash);
1365 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1367 spin_lock_bh(&hslot->lock);
1368 if (sk_nulls_del_node_init_rcu(sk)) {
1370 inet_sk(sk)->inet_num = 0;
1371 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1373 spin_lock(&hslot2->lock);
1374 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1376 spin_unlock(&hslot2->lock);
1378 spin_unlock_bh(&hslot->lock);
1381 EXPORT_SYMBOL(udp_lib_unhash);
1384 * inet_rcv_saddr was changed, we must rehash secondary hash
1386 void udp_lib_rehash(struct sock *sk, u16 newhash)
1388 if (sk_hashed(sk)) {
1389 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1390 struct udp_hslot *hslot, *hslot2, *nhslot2;
1392 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1393 nhslot2 = udp_hashslot2(udptable, newhash);
1394 udp_sk(sk)->udp_portaddr_hash = newhash;
1395 if (hslot2 != nhslot2) {
1396 hslot = udp_hashslot(udptable, sock_net(sk),
1397 udp_sk(sk)->udp_port_hash);
1398 /* we must lock primary chain too */
1399 spin_lock_bh(&hslot->lock);
1401 spin_lock(&hslot2->lock);
1402 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1404 spin_unlock(&hslot2->lock);
1406 spin_lock(&nhslot2->lock);
1407 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1410 spin_unlock(&nhslot2->lock);
1412 spin_unlock_bh(&hslot->lock);
1416 EXPORT_SYMBOL(udp_lib_rehash);
1418 static void udp_v4_rehash(struct sock *sk)
1420 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1421 inet_sk(sk)->inet_rcv_saddr,
1422 inet_sk(sk)->inet_num);
1423 udp_lib_rehash(sk, new_hash);
1426 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1430 if (inet_sk(sk)->inet_daddr) {
1431 sock_rps_save_rxhash(sk, skb);
1432 sk_mark_napi_id(sk, skb);
1435 rc = sock_queue_rcv_skb(sk, skb);
1437 int is_udplite = IS_UDPLITE(sk);
1439 /* Note that an ENOMEM error is charged twice */
1441 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1443 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1445 trace_udp_fail_queue_rcv_skb(rc, sk);
1453 static struct static_key udp_encap_needed __read_mostly;
1454 void udp_encap_enable(void)
1456 if (!static_key_enabled(&udp_encap_needed))
1457 static_key_slow_inc(&udp_encap_needed);
1459 EXPORT_SYMBOL(udp_encap_enable);
1464 * >0: "udp encap" protocol resubmission
1466 * Note that in the success and error cases, the skb is assumed to
1467 * have either been requeued or freed.
1469 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1471 struct udp_sock *up = udp_sk(sk);
1473 int is_udplite = IS_UDPLITE(sk);
1476 * Charge it to the socket, dropping if the queue is full.
1478 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1482 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1483 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1486 * This is an encapsulation socket so pass the skb to
1487 * the socket's udp_encap_rcv() hook. Otherwise, just
1488 * fall through and pass this up the UDP socket.
1489 * up->encap_rcv() returns the following value:
1490 * =0 if skb was successfully passed to the encap
1491 * handler or was discarded by it.
1492 * >0 if skb should be passed on to UDP.
1493 * <0 if skb should be resubmitted as proto -N
1496 /* if we're overly short, let UDP handle it */
1497 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1498 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1501 ret = encap_rcv(sk, skb);
1503 UDP_INC_STATS_BH(sock_net(sk),
1504 UDP_MIB_INDATAGRAMS,
1510 /* FALLTHROUGH -- it's a UDP Packet */
1514 * UDP-Lite specific tests, ignored on UDP sockets
1516 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1519 * MIB statistics other than incrementing the error count are
1520 * disabled for the following two types of errors: these depend
1521 * on the application settings, not on the functioning of the
1522 * protocol stack as such.
1524 * RFC 3828 here recommends (sec 3.3): "There should also be a
1525 * way ... to ... at least let the receiving application block
1526 * delivery of packets with coverage values less than a value
1527 * provided by the application."
1529 if (up->pcrlen == 0) { /* full coverage was set */
1530 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1531 UDP_SKB_CB(skb)->cscov, skb->len);
1534 /* The next case involves violating the min. coverage requested
1535 * by the receiver. This is subtle: if receiver wants x and x is
1536 * greater than the buffersize/MTU then receiver will complain
1537 * that it wants x while sender emits packets of smaller size y.
1538 * Therefore the above ...()->partial_cov statement is essential.
1540 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1541 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1542 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1547 if (rcu_access_pointer(sk->sk_filter) &&
1548 udp_lib_checksum_complete(skb))
1552 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1557 ipv4_pktinfo_prepare(sk, skb);
1559 if (!sock_owned_by_user(sk))
1560 rc = __udp_queue_rcv_skb(sk, skb);
1561 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1570 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1572 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1573 atomic_inc(&sk->sk_drops);
1579 static void flush_stack(struct sock **stack, unsigned int count,
1580 struct sk_buff *skb, unsigned int final)
1583 struct sk_buff *skb1 = NULL;
1586 for (i = 0; i < count; i++) {
1588 if (likely(skb1 == NULL))
1589 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1592 atomic_inc(&sk->sk_drops);
1593 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1595 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1599 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1606 static void udp_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1608 struct dst_entry *dst = skb_dst(skb);
1611 sk->sk_rx_dst = dst;
1615 * Multicasts and broadcasts go to each listener.
1617 * Note: called only from the BH handler context.
1619 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1621 __be32 saddr, __be32 daddr,
1622 struct udp_table *udptable)
1624 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1625 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1627 unsigned int i, count = 0;
1629 spin_lock(&hslot->lock);
1630 sk = sk_nulls_head(&hslot->head);
1631 dif = skb->dev->ifindex;
1632 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1634 stack[count++] = sk;
1635 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1636 daddr, uh->source, saddr, dif);
1637 if (unlikely(count == ARRAY_SIZE(stack))) {
1640 flush_stack(stack, count, skb, ~0);
1645 * before releasing chain lock, we must take a reference on sockets
1647 for (i = 0; i < count; i++)
1648 sock_hold(stack[i]);
1650 spin_unlock(&hslot->lock);
1653 * do the slow work with no lock held
1656 flush_stack(stack, count, skb, count - 1);
1658 for (i = 0; i < count; i++)
1666 /* Initialize UDP checksum. If exited with zero value (success),
1667 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1668 * Otherwise, csum completion requires chacksumming packet body,
1669 * including udp header and folding it to skb->csum.
1671 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1674 const struct iphdr *iph;
1677 UDP_SKB_CB(skb)->partial_cov = 0;
1678 UDP_SKB_CB(skb)->cscov = skb->len;
1680 if (proto == IPPROTO_UDPLITE) {
1681 err = udplite_checksum_init(skb, uh);
1687 if (uh->check == 0) {
1688 skb->ip_summed = CHECKSUM_UNNECESSARY;
1689 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1690 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1692 skb->ip_summed = CHECKSUM_UNNECESSARY;
1694 if (!skb_csum_unnecessary(skb))
1695 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1696 skb->len, proto, 0);
1697 /* Probably, we should checksum udp header (it should be in cache
1698 * in any case) and data in tiny packets (< rx copybreak).
1705 * All we need to do is get the socket, and then do a checksum.
1708 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1713 unsigned short ulen;
1714 struct rtable *rt = skb_rtable(skb);
1715 __be32 saddr, daddr;
1716 struct net *net = dev_net(skb->dev);
1719 * Validate the packet.
1721 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1722 goto drop; /* No space for header. */
1725 ulen = ntohs(uh->len);
1726 saddr = ip_hdr(skb)->saddr;
1727 daddr = ip_hdr(skb)->daddr;
1729 if (ulen > skb->len)
1732 if (proto == IPPROTO_UDP) {
1733 /* UDP validates ulen. */
1734 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1739 if (udp4_csum_init(skb, uh, proto))
1746 if (unlikely(sk->sk_rx_dst == NULL))
1747 udp_sk_rx_dst_set(sk, skb);
1749 ret = udp_queue_rcv_skb(sk, skb);
1751 /* a return value > 0 means to resubmit the input, but
1752 * it wants the return to be -protocol, or 0
1758 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1759 return __udp4_lib_mcast_deliver(net, skb, uh,
1760 saddr, daddr, udptable);
1762 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1768 ret = udp_queue_rcv_skb(sk, skb);
1771 /* a return value > 0 means to resubmit the input, but
1772 * it wants the return to be -protocol, or 0
1779 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1783 /* No socket. Drop packet silently, if checksum is wrong */
1784 if (udp_lib_checksum_complete(skb))
1787 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1788 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1791 * Hmm. We got an UDP packet to a port to which we
1792 * don't wanna listen. Ignore it.
1798 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1799 proto == IPPROTO_UDPLITE ? "Lite" : "",
1800 &saddr, ntohs(uh->source),
1802 &daddr, ntohs(uh->dest));
1807 * RFC1122: OK. Discards the bad packet silently (as far as
1808 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1810 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1811 proto == IPPROTO_UDPLITE ? "Lite" : "",
1812 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1814 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1816 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1821 /* We can only early demux multicast if there is a single matching socket.
1822 * If more than one socket found returns NULL
1824 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1825 __be16 loc_port, __be32 loc_addr,
1826 __be16 rmt_port, __be32 rmt_addr,
1829 struct sock *sk, *result;
1830 struct hlist_nulls_node *node;
1831 unsigned short hnum = ntohs(loc_port);
1832 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1833 struct udp_hslot *hslot = &udp_table.hash[slot];
1839 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1840 if (__udp_is_mcast_sock(net, sk,
1849 * if the nulls value we got at the end of this lookup is
1850 * not the expected one, we must restart lookup.
1851 * We probably met an item that was moved to another chain.
1853 if (get_nulls_value(node) != slot)
1858 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1860 else if (unlikely(!__udp_is_mcast_sock(net, result,
1872 /* For unicast we should only early demux connected sockets or we can
1873 * break forwarding setups. The chains here can be long so only check
1874 * if the first socket is an exact match and if not move on.
1876 static struct sock *__udp4_lib_demux_lookup(struct net *net,
1877 __be16 loc_port, __be32 loc_addr,
1878 __be16 rmt_port, __be32 rmt_addr,
1881 struct sock *sk, *result;
1882 struct hlist_nulls_node *node;
1883 unsigned short hnum = ntohs(loc_port);
1884 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1885 unsigned int slot2 = hash2 & udp_table.mask;
1886 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1887 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr)
1888 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1892 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
1893 if (INET_MATCH(sk, net, acookie,
1894 rmt_addr, loc_addr, ports, dif))
1896 /* Only check first socket in chain */
1901 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1903 else if (unlikely(!INET_MATCH(sk, net, acookie,
1914 void udp_v4_early_demux(struct sk_buff *skb)
1916 const struct iphdr *iph = ip_hdr(skb);
1917 const struct udphdr *uh = udp_hdr(skb);
1919 struct dst_entry *dst;
1920 struct net *net = dev_net(skb->dev);
1921 int dif = skb->dev->ifindex;
1923 /* validate the packet */
1924 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1927 if (skb->pkt_type == PACKET_BROADCAST ||
1928 skb->pkt_type == PACKET_MULTICAST)
1929 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1930 uh->source, iph->saddr, dif);
1931 else if (skb->pkt_type == PACKET_HOST)
1932 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1933 uh->source, iph->saddr, dif);
1941 skb->destructor = sock_edemux;
1942 dst = sk->sk_rx_dst;
1945 dst = dst_check(dst, 0);
1947 skb_dst_set_noref(skb, dst);
1950 int udp_rcv(struct sk_buff *skb)
1952 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1955 void udp_destroy_sock(struct sock *sk)
1957 struct udp_sock *up = udp_sk(sk);
1958 bool slow = lock_sock_fast(sk);
1959 udp_flush_pending_frames(sk);
1960 unlock_sock_fast(sk, slow);
1961 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1962 void (*encap_destroy)(struct sock *sk);
1963 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1970 * Socket option code for UDP
1972 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1973 char __user *optval, unsigned int optlen,
1974 int (*push_pending_frames)(struct sock *))
1976 struct udp_sock *up = udp_sk(sk);
1979 int is_udplite = IS_UDPLITE(sk);
1981 if (optlen < sizeof(int))
1984 if (get_user(val, (int __user *)optval))
1994 (*push_pending_frames)(sk);
2002 case UDP_ENCAP_ESPINUDP:
2003 case UDP_ENCAP_ESPINUDP_NON_IKE:
2004 up->encap_rcv = xfrm4_udp_encap_rcv;
2006 case UDP_ENCAP_L2TPINUDP:
2007 up->encap_type = val;
2017 * UDP-Lite's partial checksum coverage (RFC 3828).
2019 /* The sender sets actual checksum coverage length via this option.
2020 * The case coverage > packet length is handled by send module. */
2021 case UDPLITE_SEND_CSCOV:
2022 if (!is_udplite) /* Disable the option on UDP sockets */
2023 return -ENOPROTOOPT;
2024 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2026 else if (val > USHRT_MAX)
2029 up->pcflag |= UDPLITE_SEND_CC;
2032 /* The receiver specifies a minimum checksum coverage value. To make
2033 * sense, this should be set to at least 8 (as done below). If zero is
2034 * used, this again means full checksum coverage. */
2035 case UDPLITE_RECV_CSCOV:
2036 if (!is_udplite) /* Disable the option on UDP sockets */
2037 return -ENOPROTOOPT;
2038 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2040 else if (val > USHRT_MAX)
2043 up->pcflag |= UDPLITE_RECV_CC;
2053 EXPORT_SYMBOL(udp_lib_setsockopt);
2055 int udp_setsockopt(struct sock *sk, int level, int optname,
2056 char __user *optval, unsigned int optlen)
2058 if (level == SOL_UDP || level == SOL_UDPLITE)
2059 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2060 udp_push_pending_frames);
2061 return ip_setsockopt(sk, level, optname, optval, optlen);
2064 #ifdef CONFIG_COMPAT
2065 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2066 char __user *optval, unsigned int optlen)
2068 if (level == SOL_UDP || level == SOL_UDPLITE)
2069 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2070 udp_push_pending_frames);
2071 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2075 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2076 char __user *optval, int __user *optlen)
2078 struct udp_sock *up = udp_sk(sk);
2081 if (get_user(len, optlen))
2084 len = min_t(unsigned int, len, sizeof(int));
2095 val = up->encap_type;
2098 /* The following two cannot be changed on UDP sockets, the return is
2099 * always 0 (which corresponds to the full checksum coverage of UDP). */
2100 case UDPLITE_SEND_CSCOV:
2104 case UDPLITE_RECV_CSCOV:
2109 return -ENOPROTOOPT;
2112 if (put_user(len, optlen))
2114 if (copy_to_user(optval, &val, len))
2118 EXPORT_SYMBOL(udp_lib_getsockopt);
2120 int udp_getsockopt(struct sock *sk, int level, int optname,
2121 char __user *optval, int __user *optlen)
2123 if (level == SOL_UDP || level == SOL_UDPLITE)
2124 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2125 return ip_getsockopt(sk, level, optname, optval, optlen);
2128 #ifdef CONFIG_COMPAT
2129 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2130 char __user *optval, int __user *optlen)
2132 if (level == SOL_UDP || level == SOL_UDPLITE)
2133 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2134 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2138 * udp_poll - wait for a UDP event.
2139 * @file - file struct
2141 * @wait - poll table
2143 * This is same as datagram poll, except for the special case of
2144 * blocking sockets. If application is using a blocking fd
2145 * and a packet with checksum error is in the queue;
2146 * then it could get return from select indicating data available
2147 * but then block when reading it. Add special case code
2148 * to work around these arguably broken applications.
2150 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2152 unsigned int mask = datagram_poll(file, sock, wait);
2153 struct sock *sk = sock->sk;
2155 sock_rps_record_flow(sk);
2157 /* Check for false positives due to checksum errors */
2158 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2159 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2160 mask &= ~(POLLIN | POLLRDNORM);
2165 EXPORT_SYMBOL(udp_poll);
2167 struct proto udp_prot = {
2169 .owner = THIS_MODULE,
2170 .close = udp_lib_close,
2171 .connect = ip4_datagram_connect,
2172 .disconnect = udp_disconnect,
2174 .destroy = udp_destroy_sock,
2175 .setsockopt = udp_setsockopt,
2176 .getsockopt = udp_getsockopt,
2177 .sendmsg = udp_sendmsg,
2178 .recvmsg = udp_recvmsg,
2179 .sendpage = udp_sendpage,
2180 .backlog_rcv = __udp_queue_rcv_skb,
2181 .release_cb = ip4_datagram_release_cb,
2182 .hash = udp_lib_hash,
2183 .unhash = udp_lib_unhash,
2184 .rehash = udp_v4_rehash,
2185 .get_port = udp_v4_get_port,
2186 .memory_allocated = &udp_memory_allocated,
2187 .sysctl_mem = sysctl_udp_mem,
2188 .sysctl_wmem = &sysctl_udp_wmem_min,
2189 .sysctl_rmem = &sysctl_udp_rmem_min,
2190 .obj_size = sizeof(struct udp_sock),
2191 .slab_flags = SLAB_DESTROY_BY_RCU,
2192 .h.udp_table = &udp_table,
2193 #ifdef CONFIG_COMPAT
2194 .compat_setsockopt = compat_udp_setsockopt,
2195 .compat_getsockopt = compat_udp_getsockopt,
2197 .clear_sk = sk_prot_clear_portaddr_nulls,
2199 EXPORT_SYMBOL(udp_prot);
2201 /* ------------------------------------------------------------------------ */
2202 #ifdef CONFIG_PROC_FS
2204 static struct sock *udp_get_first(struct seq_file *seq, int start)
2207 struct udp_iter_state *state = seq->private;
2208 struct net *net = seq_file_net(seq);
2210 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2212 struct hlist_nulls_node *node;
2213 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2215 if (hlist_nulls_empty(&hslot->head))
2218 spin_lock_bh(&hslot->lock);
2219 sk_nulls_for_each(sk, node, &hslot->head) {
2220 if (!net_eq(sock_net(sk), net))
2222 if (sk->sk_family == state->family)
2225 spin_unlock_bh(&hslot->lock);
2232 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2234 struct udp_iter_state *state = seq->private;
2235 struct net *net = seq_file_net(seq);
2238 sk = sk_nulls_next(sk);
2239 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2242 if (state->bucket <= state->udp_table->mask)
2243 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2244 return udp_get_first(seq, state->bucket + 1);
2249 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2251 struct sock *sk = udp_get_first(seq, 0);
2254 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2256 return pos ? NULL : sk;
2259 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2261 struct udp_iter_state *state = seq->private;
2262 state->bucket = MAX_UDP_PORTS;
2264 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2267 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2271 if (v == SEQ_START_TOKEN)
2272 sk = udp_get_idx(seq, 0);
2274 sk = udp_get_next(seq, v);
2280 static void udp_seq_stop(struct seq_file *seq, void *v)
2282 struct udp_iter_state *state = seq->private;
2284 if (state->bucket <= state->udp_table->mask)
2285 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2288 int udp_seq_open(struct inode *inode, struct file *file)
2290 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2291 struct udp_iter_state *s;
2294 err = seq_open_net(inode, file, &afinfo->seq_ops,
2295 sizeof(struct udp_iter_state));
2299 s = ((struct seq_file *)file->private_data)->private;
2300 s->family = afinfo->family;
2301 s->udp_table = afinfo->udp_table;
2304 EXPORT_SYMBOL(udp_seq_open);
2306 /* ------------------------------------------------------------------------ */
2307 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2309 struct proc_dir_entry *p;
2312 afinfo->seq_ops.start = udp_seq_start;
2313 afinfo->seq_ops.next = udp_seq_next;
2314 afinfo->seq_ops.stop = udp_seq_stop;
2316 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2317 afinfo->seq_fops, afinfo);
2322 EXPORT_SYMBOL(udp_proc_register);
2324 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2326 remove_proc_entry(afinfo->name, net->proc_net);
2328 EXPORT_SYMBOL(udp_proc_unregister);
2330 /* ------------------------------------------------------------------------ */
2331 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2334 struct inet_sock *inet = inet_sk(sp);
2335 __be32 dest = inet->inet_daddr;
2336 __be32 src = inet->inet_rcv_saddr;
2337 __u16 destp = ntohs(inet->inet_dport);
2338 __u16 srcp = ntohs(inet->inet_sport);
2340 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2341 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2342 bucket, src, srcp, dest, destp, sp->sk_state,
2343 sk_wmem_alloc_get(sp),
2344 sk_rmem_alloc_get(sp),
2346 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2348 atomic_read(&sp->sk_refcnt), sp,
2349 atomic_read(&sp->sk_drops));
2352 int udp4_seq_show(struct seq_file *seq, void *v)
2354 seq_setwidth(seq, 127);
2355 if (v == SEQ_START_TOKEN)
2356 seq_puts(seq, " sl local_address rem_address st tx_queue "
2357 "rx_queue tr tm->when retrnsmt uid timeout "
2358 "inode ref pointer drops");
2360 struct udp_iter_state *state = seq->private;
2362 udp4_format_sock(v, seq, state->bucket);
2368 static const struct file_operations udp_afinfo_seq_fops = {
2369 .owner = THIS_MODULE,
2370 .open = udp_seq_open,
2372 .llseek = seq_lseek,
2373 .release = seq_release_net
2376 /* ------------------------------------------------------------------------ */
2377 static struct udp_seq_afinfo udp4_seq_afinfo = {
2380 .udp_table = &udp_table,
2381 .seq_fops = &udp_afinfo_seq_fops,
2383 .show = udp4_seq_show,
2387 static int __net_init udp4_proc_init_net(struct net *net)
2389 return udp_proc_register(net, &udp4_seq_afinfo);
2392 static void __net_exit udp4_proc_exit_net(struct net *net)
2394 udp_proc_unregister(net, &udp4_seq_afinfo);
2397 static struct pernet_operations udp4_net_ops = {
2398 .init = udp4_proc_init_net,
2399 .exit = udp4_proc_exit_net,
2402 int __init udp4_proc_init(void)
2404 return register_pernet_subsys(&udp4_net_ops);
2407 void udp4_proc_exit(void)
2409 unregister_pernet_subsys(&udp4_net_ops);
2411 #endif /* CONFIG_PROC_FS */
2413 static __initdata unsigned long uhash_entries;
2414 static int __init set_uhash_entries(char *str)
2421 ret = kstrtoul(str, 0, &uhash_entries);
2425 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2426 uhash_entries = UDP_HTABLE_SIZE_MIN;
2429 __setup("uhash_entries=", set_uhash_entries);
2431 void __init udp_table_init(struct udp_table *table, const char *name)
2435 table->hash = alloc_large_system_hash(name,
2436 2 * sizeof(struct udp_hslot),
2438 21, /* one slot per 2 MB */
2442 UDP_HTABLE_SIZE_MIN,
2445 table->hash2 = table->hash + (table->mask + 1);
2446 for (i = 0; i <= table->mask; i++) {
2447 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2448 table->hash[i].count = 0;
2449 spin_lock_init(&table->hash[i].lock);
2451 for (i = 0; i <= table->mask; i++) {
2452 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2453 table->hash2[i].count = 0;
2454 spin_lock_init(&table->hash2[i].lock);
2458 void __init udp_init(void)
2460 unsigned long limit;
2462 udp_table_init(&udp_table, "UDP");
2463 limit = nr_free_buffer_pages() / 8;
2464 limit = max(limit, 128UL);
2465 sysctl_udp_mem[0] = limit / 4 * 3;
2466 sysctl_udp_mem[1] = limit;
2467 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2469 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2470 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2473 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
2474 netdev_features_t features)
2476 struct sk_buff *segs = ERR_PTR(-EINVAL);
2477 int mac_len = skb->mac_len;
2478 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
2479 __be16 protocol = skb->protocol;
2480 netdev_features_t enc_features;
2483 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
2486 skb->encapsulation = 0;
2487 __skb_pull(skb, tnl_hlen);
2488 skb_reset_mac_header(skb);
2489 skb_set_network_header(skb, skb_inner_network_offset(skb));
2490 skb->mac_len = skb_inner_network_offset(skb);
2491 skb->protocol = htons(ETH_P_TEB);
2493 /* segment inner packet. */
2494 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb);
2495 segs = skb_mac_gso_segment(skb, enc_features);
2496 if (!segs || IS_ERR(segs))
2499 outer_hlen = skb_tnl_header_len(skb);
2503 int udp_offset = outer_hlen - tnl_hlen;
2505 skb_reset_inner_headers(skb);
2506 skb->encapsulation = 1;
2508 skb->mac_len = mac_len;
2510 skb_push(skb, outer_hlen);
2511 skb_reset_mac_header(skb);
2512 skb_set_network_header(skb, mac_len);
2513 skb_set_transport_header(skb, udp_offset);
2515 uh->len = htons(skb->len - udp_offset);
2517 /* csum segment if tunnel sets skb with csum. */
2518 if (protocol == htons(ETH_P_IP) && unlikely(uh->check)) {
2519 struct iphdr *iph = ip_hdr(skb);
2521 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
2522 skb->len - udp_offset,
2524 uh->check = csum_fold(skb_checksum(skb, udp_offset,
2525 skb->len - udp_offset, 0));
2527 uh->check = CSUM_MANGLED_0;
2529 } else if (protocol == htons(ETH_P_IPV6)) {
2530 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
2531 u32 len = skb->len - udp_offset;
2533 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
2534 len, IPPROTO_UDP, 0);
2535 uh->check = csum_fold(skb_checksum(skb, udp_offset, len, 0));
2537 uh->check = CSUM_MANGLED_0;
2538 skb->ip_summed = CHECKSUM_NONE;
2541 skb->protocol = protocol;
2542 } while ((skb = skb->next));