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/route.h>
107 #include <net/checksum.h>
108 #include <net/xfrm.h>
109 #include <trace/events/udp.h>
110 #include <linux/static_key.h>
111 #include <trace/events/skb.h>
112 #include <net/busy_poll.h>
113 #include "udp_impl.h"
115 struct udp_table udp_table __read_mostly;
116 EXPORT_SYMBOL(udp_table);
118 long sysctl_udp_mem[3] __read_mostly;
119 EXPORT_SYMBOL(sysctl_udp_mem);
121 int sysctl_udp_rmem_min __read_mostly;
122 EXPORT_SYMBOL(sysctl_udp_rmem_min);
124 int sysctl_udp_wmem_min __read_mostly;
125 EXPORT_SYMBOL(sysctl_udp_wmem_min);
127 atomic_long_t udp_memory_allocated;
128 EXPORT_SYMBOL(udp_memory_allocated);
130 #define MAX_UDP_PORTS 65536
131 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
133 static int udp_lib_lport_inuse(struct net *net, __u16 num,
134 const struct udp_hslot *hslot,
135 unsigned long *bitmap,
137 int (*saddr_comp)(const struct sock *sk1,
138 const struct sock *sk2),
142 struct hlist_nulls_node *node;
143 kuid_t uid = sock_i_uid(sk);
145 sk_nulls_for_each(sk2, node, &hslot->head)
146 if (net_eq(sock_net(sk2), net) &&
148 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
149 (!sk2->sk_reuse || !sk->sk_reuse) &&
150 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
151 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
152 (!sk2->sk_reuseport || !sk->sk_reuseport ||
153 !uid_eq(uid, sock_i_uid(sk2))) &&
154 (*saddr_comp)(sk, sk2)) {
156 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
165 * Note: we still hold spinlock of primary hash chain, so no other writer
166 * can insert/delete a socket with local_port == num
168 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
169 struct udp_hslot *hslot2,
171 int (*saddr_comp)(const struct sock *sk1,
172 const struct sock *sk2))
175 struct hlist_nulls_node *node;
176 kuid_t uid = sock_i_uid(sk);
179 spin_lock(&hslot2->lock);
180 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
181 if (net_eq(sock_net(sk2), net) &&
183 (udp_sk(sk2)->udp_port_hash == num) &&
184 (!sk2->sk_reuse || !sk->sk_reuse) &&
185 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
186 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
187 (!sk2->sk_reuseport || !sk->sk_reuseport ||
188 !uid_eq(uid, sock_i_uid(sk2))) &&
189 (*saddr_comp)(sk, sk2)) {
193 spin_unlock(&hslot2->lock);
198 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
200 * @sk: socket struct in question
201 * @snum: port number to look up
202 * @saddr_comp: AF-dependent comparison of bound local IP addresses
203 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
206 int udp_lib_get_port(struct sock *sk, unsigned short snum,
207 int (*saddr_comp)(const struct sock *sk1,
208 const struct sock *sk2),
209 unsigned int hash2_nulladdr)
211 struct udp_hslot *hslot, *hslot2;
212 struct udp_table *udptable = sk->sk_prot->h.udp_table;
214 struct net *net = sock_net(sk);
217 int low, high, remaining;
219 unsigned short first, last;
220 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
222 inet_get_local_port_range(&low, &high);
223 remaining = (high - low) + 1;
226 first = (((u64)rand * remaining) >> 32) + low;
228 * force rand to be an odd multiple of UDP_HTABLE_SIZE
230 rand = (rand | 1) * (udptable->mask + 1);
231 last = first + udptable->mask + 1;
233 hslot = udp_hashslot(udptable, net, first);
234 bitmap_zero(bitmap, PORTS_PER_CHAIN);
235 spin_lock_bh(&hslot->lock);
236 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
237 saddr_comp, udptable->log);
241 * Iterate on all possible values of snum for this hash.
242 * Using steps of an odd multiple of UDP_HTABLE_SIZE
243 * give us randomization and full range coverage.
246 if (low <= snum && snum <= high &&
247 !test_bit(snum >> udptable->log, bitmap) &&
248 !inet_is_reserved_local_port(snum))
251 } while (snum != first);
252 spin_unlock_bh(&hslot->lock);
253 } while (++first != last);
256 hslot = udp_hashslot(udptable, net, snum);
257 spin_lock_bh(&hslot->lock);
258 if (hslot->count > 10) {
260 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
262 slot2 &= udptable->mask;
263 hash2_nulladdr &= udptable->mask;
265 hslot2 = udp_hashslot2(udptable, slot2);
266 if (hslot->count < hslot2->count)
267 goto scan_primary_hash;
269 exist = udp_lib_lport_inuse2(net, snum, hslot2,
271 if (!exist && (hash2_nulladdr != slot2)) {
272 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
273 exist = udp_lib_lport_inuse2(net, snum, hslot2,
282 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
287 inet_sk(sk)->inet_num = snum;
288 udp_sk(sk)->udp_port_hash = snum;
289 udp_sk(sk)->udp_portaddr_hash ^= snum;
290 if (sk_unhashed(sk)) {
291 sk_nulls_add_node_rcu(sk, &hslot->head);
293 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
295 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
296 spin_lock(&hslot2->lock);
297 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
300 spin_unlock(&hslot2->lock);
304 spin_unlock_bh(&hslot->lock);
308 EXPORT_SYMBOL(udp_lib_get_port);
310 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
312 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
314 return (!ipv6_only_sock(sk2) &&
315 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
316 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
319 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
322 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
325 int udp_v4_get_port(struct sock *sk, unsigned short snum)
327 unsigned int hash2_nulladdr =
328 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
329 unsigned int hash2_partial =
330 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
332 /* precompute partial secondary hash */
333 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
334 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
337 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
339 __be16 sport, __be32 daddr, __be16 dport, int dif)
343 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
344 !ipv6_only_sock(sk)) {
345 struct inet_sock *inet = inet_sk(sk);
347 score = (sk->sk_family == PF_INET ? 2 : 1);
348 if (inet->inet_rcv_saddr) {
349 if (inet->inet_rcv_saddr != daddr)
353 if (inet->inet_daddr) {
354 if (inet->inet_daddr != saddr)
358 if (inet->inet_dport) {
359 if (inet->inet_dport != sport)
363 if (sk->sk_bound_dev_if) {
364 if (sk->sk_bound_dev_if != dif)
373 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
375 static inline int compute_score2(struct sock *sk, struct net *net,
376 __be32 saddr, __be16 sport,
377 __be32 daddr, unsigned int hnum, int dif)
381 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
382 struct inet_sock *inet = inet_sk(sk);
384 if (inet->inet_rcv_saddr != daddr)
386 if (inet->inet_num != hnum)
389 score = (sk->sk_family == PF_INET ? 2 : 1);
390 if (inet->inet_daddr) {
391 if (inet->inet_daddr != saddr)
395 if (inet->inet_dport) {
396 if (inet->inet_dport != sport)
400 if (sk->sk_bound_dev_if) {
401 if (sk->sk_bound_dev_if != dif)
410 /* called with read_rcu_lock() */
411 static struct sock *udp4_lib_lookup2(struct net *net,
412 __be32 saddr, __be16 sport,
413 __be32 daddr, unsigned int hnum, int dif,
414 struct udp_hslot *hslot2, unsigned int slot2)
416 struct sock *sk, *result;
417 struct hlist_nulls_node *node;
418 int score, badness, matches = 0, reuseport = 0;
424 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
425 score = compute_score2(sk, net, saddr, sport,
427 if (score > badness) {
430 reuseport = sk->sk_reuseport;
432 hash = inet_ehashfn(net, daddr, hnum,
436 } else if (score == badness && reuseport) {
438 if (((u64)hash * matches) >> 32 == 0)
440 hash = next_pseudo_random32(hash);
444 * if the nulls value we got at the end of this lookup is
445 * not the expected one, we must restart lookup.
446 * We probably met an item that was moved to another chain.
448 if (get_nulls_value(node) != slot2)
451 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
453 else if (unlikely(compute_score2(result, net, saddr, sport,
454 daddr, hnum, dif) < badness)) {
462 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
463 * harder than this. -DaveM
465 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
466 __be16 sport, __be32 daddr, __be16 dport,
467 int dif, struct udp_table *udptable)
469 struct sock *sk, *result;
470 struct hlist_nulls_node *node;
471 unsigned short hnum = ntohs(dport);
472 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
473 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
474 int score, badness, matches = 0, reuseport = 0;
478 if (hslot->count > 10) {
479 hash2 = udp4_portaddr_hash(net, daddr, hnum);
480 slot2 = hash2 & udptable->mask;
481 hslot2 = &udptable->hash2[slot2];
482 if (hslot->count < hslot2->count)
485 result = udp4_lib_lookup2(net, saddr, sport,
489 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
490 slot2 = hash2 & udptable->mask;
491 hslot2 = &udptable->hash2[slot2];
492 if (hslot->count < hslot2->count)
495 result = udp4_lib_lookup2(net, saddr, sport,
496 htonl(INADDR_ANY), hnum, dif,
505 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
506 score = compute_score(sk, net, saddr, hnum, sport,
508 if (score > badness) {
511 reuseport = sk->sk_reuseport;
513 hash = inet_ehashfn(net, daddr, hnum,
517 } else if (score == badness && reuseport) {
519 if (((u64)hash * matches) >> 32 == 0)
521 hash = next_pseudo_random32(hash);
525 * if the nulls value we got at the end of this lookup is
526 * not the expected one, we must restart lookup.
527 * We probably met an item that was moved to another chain.
529 if (get_nulls_value(node) != slot)
533 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
535 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
536 daddr, dport, dif) < badness)) {
544 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
546 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
547 __be16 sport, __be16 dport,
548 struct udp_table *udptable)
551 const struct iphdr *iph = ip_hdr(skb);
553 if (unlikely(sk = skb_steal_sock(skb)))
556 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
557 iph->daddr, dport, inet_iif(skb),
561 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
562 __be32 daddr, __be16 dport, int dif)
564 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
566 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
568 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
569 __be16 loc_port, __be32 loc_addr,
570 __be16 rmt_port, __be32 rmt_addr,
573 struct hlist_nulls_node *node;
575 unsigned short hnum = ntohs(loc_port);
577 sk_nulls_for_each_from(s, node) {
578 struct inet_sock *inet = inet_sk(s);
580 if (!net_eq(sock_net(s), net) ||
581 udp_sk(s)->udp_port_hash != hnum ||
582 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
583 (inet->inet_dport != rmt_port && inet->inet_dport) ||
584 (inet->inet_rcv_saddr &&
585 inet->inet_rcv_saddr != loc_addr) ||
587 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
589 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
599 * This routine is called by the ICMP module when it gets some
600 * sort of error condition. If err < 0 then the socket should
601 * be closed and the error returned to the user. If err > 0
602 * it's just the icmp type << 8 | icmp code.
603 * Header points to the ip header of the error packet. We move
604 * on past this. Then (as it used to claim before adjustment)
605 * header points to the first 8 bytes of the udp header. We need
606 * to find the appropriate port.
609 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
611 struct inet_sock *inet;
612 const struct iphdr *iph = (const struct iphdr *)skb->data;
613 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
614 const int type = icmp_hdr(skb)->type;
615 const int code = icmp_hdr(skb)->code;
619 struct net *net = dev_net(skb->dev);
621 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
622 iph->saddr, uh->source, skb->dev->ifindex, udptable);
624 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
625 return; /* No socket for error */
634 case ICMP_TIME_EXCEEDED:
637 case ICMP_SOURCE_QUENCH:
639 case ICMP_PARAMETERPROB:
643 case ICMP_DEST_UNREACH:
644 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
645 ipv4_sk_update_pmtu(skb, sk, info);
646 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
654 if (code <= NR_ICMP_UNREACH) {
655 harderr = icmp_err_convert[code].fatal;
656 err = icmp_err_convert[code].errno;
660 ipv4_sk_redirect(skb, sk);
665 * RFC1122: OK. Passes ICMP errors back to application, as per
668 if (!inet->recverr) {
669 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
672 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
675 sk->sk_error_report(sk);
680 void udp_err(struct sk_buff *skb, u32 info)
682 __udp4_lib_err(skb, info, &udp_table);
686 * Throw away all pending data and cancel the corking. Socket is locked.
688 void udp_flush_pending_frames(struct sock *sk)
690 struct udp_sock *up = udp_sk(sk);
695 ip_flush_pending_frames(sk);
698 EXPORT_SYMBOL(udp_flush_pending_frames);
701 * udp4_hwcsum - handle outgoing HW checksumming
702 * @skb: sk_buff containing the filled-in UDP header
703 * (checksum field must be zeroed out)
704 * @src: source IP address
705 * @dst: destination IP address
707 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
709 struct udphdr *uh = udp_hdr(skb);
710 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
711 int offset = skb_transport_offset(skb);
712 int len = skb->len - offset;
718 * Only one fragment on the socket.
720 skb->csum_start = skb_transport_header(skb) - skb->head;
721 skb->csum_offset = offsetof(struct udphdr, check);
722 uh->check = ~csum_tcpudp_magic(src, dst, len,
726 * HW-checksum won't work as there are two or more
727 * fragments on the socket so that all csums of sk_buffs
731 csum = csum_add(csum, frags->csum);
733 } while ((frags = frags->next));
735 csum = skb_checksum(skb, offset, hlen, csum);
736 skb->ip_summed = CHECKSUM_NONE;
738 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
740 uh->check = CSUM_MANGLED_0;
743 EXPORT_SYMBOL_GPL(udp4_hwcsum);
745 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
747 struct sock *sk = skb->sk;
748 struct inet_sock *inet = inet_sk(sk);
751 int is_udplite = IS_UDPLITE(sk);
752 int offset = skb_transport_offset(skb);
753 int len = skb->len - offset;
757 * Create a UDP header
760 uh->source = inet->inet_sport;
761 uh->dest = fl4->fl4_dport;
762 uh->len = htons(len);
765 if (is_udplite) /* UDP-Lite */
766 csum = udplite_csum(skb);
768 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
770 skb->ip_summed = CHECKSUM_NONE;
773 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
775 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
779 csum = udp_csum(skb);
781 /* add protocol-dependent pseudo-header */
782 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
783 sk->sk_protocol, csum);
785 uh->check = CSUM_MANGLED_0;
788 err = ip_send_skb(sock_net(sk), skb);
790 if (err == -ENOBUFS && !inet->recverr) {
791 UDP_INC_STATS_USER(sock_net(sk),
792 UDP_MIB_SNDBUFERRORS, is_udplite);
796 UDP_INC_STATS_USER(sock_net(sk),
797 UDP_MIB_OUTDATAGRAMS, is_udplite);
802 * Push out all pending data as one UDP datagram. Socket is locked.
804 int udp_push_pending_frames(struct sock *sk)
806 struct udp_sock *up = udp_sk(sk);
807 struct inet_sock *inet = inet_sk(sk);
808 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
812 skb = ip_finish_skb(sk, fl4);
816 err = udp_send_skb(skb, fl4);
823 EXPORT_SYMBOL(udp_push_pending_frames);
825 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
828 struct inet_sock *inet = inet_sk(sk);
829 struct udp_sock *up = udp_sk(sk);
830 struct flowi4 fl4_stack;
833 struct ipcm_cookie ipc;
834 struct rtable *rt = NULL;
837 __be32 daddr, faddr, saddr;
840 int err, is_udplite = IS_UDPLITE(sk);
841 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
842 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
844 struct ip_options_data opt_copy;
853 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
861 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
863 fl4 = &inet->cork.fl.u.ip4;
866 * There are pending frames.
867 * The socket lock must be held while it's corked.
870 if (likely(up->pending)) {
871 if (unlikely(up->pending != AF_INET)) {
879 ulen += sizeof(struct udphdr);
882 * Get and verify the address.
885 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
886 if (msg->msg_namelen < sizeof(*usin))
888 if (usin->sin_family != AF_INET) {
889 if (usin->sin_family != AF_UNSPEC)
890 return -EAFNOSUPPORT;
893 daddr = usin->sin_addr.s_addr;
894 dport = usin->sin_port;
898 if (sk->sk_state != TCP_ESTABLISHED)
899 return -EDESTADDRREQ;
900 daddr = inet->inet_daddr;
901 dport = inet->inet_dport;
902 /* Open fast path for connected socket.
903 Route will not be used, if at least one option is set.
907 ipc.addr = inet->inet_saddr;
909 ipc.oif = sk->sk_bound_dev_if;
911 sock_tx_timestamp(sk, &ipc.tx_flags);
913 if (msg->msg_controllen) {
914 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
922 struct ip_options_rcu *inet_opt;
925 inet_opt = rcu_dereference(inet->inet_opt);
927 memcpy(&opt_copy, inet_opt,
928 sizeof(*inet_opt) + inet_opt->opt.optlen);
929 ipc.opt = &opt_copy.opt;
935 ipc.addr = faddr = daddr;
937 if (ipc.opt && ipc.opt->opt.srr) {
940 faddr = ipc.opt->opt.faddr;
943 tos = get_rttos(&ipc, inet);
944 if (sock_flag(sk, SOCK_LOCALROUTE) ||
945 (msg->msg_flags & MSG_DONTROUTE) ||
946 (ipc.opt && ipc.opt->opt.is_strictroute)) {
951 if (ipv4_is_multicast(daddr)) {
953 ipc.oif = inet->mc_index;
955 saddr = inet->mc_addr;
958 ipc.oif = inet->uc_index;
961 rt = (struct rtable *)sk_dst_check(sk, 0);
964 struct net *net = sock_net(sk);
967 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
968 RT_SCOPE_UNIVERSE, sk->sk_protocol,
969 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
970 faddr, saddr, dport, inet->inet_sport);
972 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
973 rt = ip_route_output_flow(net, fl4, sk);
977 if (err == -ENETUNREACH)
978 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
983 if ((rt->rt_flags & RTCF_BROADCAST) &&
984 !sock_flag(sk, SOCK_BROADCAST))
987 sk_dst_set(sk, dst_clone(&rt->dst));
990 if (msg->msg_flags&MSG_CONFIRM)
996 daddr = ipc.addr = fl4->daddr;
998 /* Lockless fast path for the non-corking case. */
1000 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
1001 sizeof(struct udphdr), &ipc, &rt,
1004 if (!IS_ERR_OR_NULL(skb))
1005 err = udp_send_skb(skb, fl4);
1010 if (unlikely(up->pending)) {
1011 /* The socket is already corked while preparing it. */
1012 /* ... which is an evident application bug. --ANK */
1015 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1020 * Now cork the socket to pend data.
1022 fl4 = &inet->cork.fl.u.ip4;
1025 fl4->fl4_dport = dport;
1026 fl4->fl4_sport = inet->inet_sport;
1027 up->pending = AF_INET;
1031 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1032 sizeof(struct udphdr), &ipc, &rt,
1033 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1035 udp_flush_pending_frames(sk);
1037 err = udp_push_pending_frames(sk);
1038 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1049 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1050 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1051 * we don't have a good statistic (IpOutDiscards but it can be too many
1052 * things). We could add another new stat but at least for now that
1053 * seems like overkill.
1055 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1056 UDP_INC_STATS_USER(sock_net(sk),
1057 UDP_MIB_SNDBUFERRORS, is_udplite);
1062 dst_confirm(&rt->dst);
1063 if (!(msg->msg_flags&MSG_PROBE) || len)
1064 goto back_from_confirm;
1068 EXPORT_SYMBOL(udp_sendmsg);
1070 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1071 size_t size, int flags)
1073 struct inet_sock *inet = inet_sk(sk);
1074 struct udp_sock *up = udp_sk(sk);
1078 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1080 /* Call udp_sendmsg to specify destination address which
1081 * sendpage interface can't pass.
1082 * This will succeed only when the socket is connected.
1084 ret = udp_sendmsg(NULL, sk, &msg, 0);
1091 if (unlikely(!up->pending)) {
1094 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1098 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1099 page, offset, size, flags);
1100 if (ret == -EOPNOTSUPP) {
1102 return sock_no_sendpage(sk->sk_socket, page, offset,
1106 udp_flush_pending_frames(sk);
1111 if (!(up->corkflag || (flags&MSG_MORE)))
1112 ret = udp_push_pending_frames(sk);
1122 * first_packet_length - return length of first packet in receive queue
1125 * Drops all bad checksum frames, until a valid one is found.
1126 * Returns the length of found skb, or 0 if none is found.
1128 static unsigned int first_packet_length(struct sock *sk)
1130 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1131 struct sk_buff *skb;
1134 __skb_queue_head_init(&list_kill);
1136 spin_lock_bh(&rcvq->lock);
1137 while ((skb = skb_peek(rcvq)) != NULL &&
1138 udp_lib_checksum_complete(skb)) {
1139 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1141 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1143 atomic_inc(&sk->sk_drops);
1144 __skb_unlink(skb, rcvq);
1145 __skb_queue_tail(&list_kill, skb);
1147 res = skb ? skb->len : 0;
1148 spin_unlock_bh(&rcvq->lock);
1150 if (!skb_queue_empty(&list_kill)) {
1151 bool slow = lock_sock_fast(sk);
1153 __skb_queue_purge(&list_kill);
1154 sk_mem_reclaim_partial(sk);
1155 unlock_sock_fast(sk, slow);
1161 * IOCTL requests applicable to the UDP protocol
1164 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1169 int amount = sk_wmem_alloc_get(sk);
1171 return put_user(amount, (int __user *)arg);
1176 unsigned int amount = first_packet_length(sk);
1180 * We will only return the amount
1181 * of this packet since that is all
1182 * that will be read.
1184 amount -= sizeof(struct udphdr);
1186 return put_user(amount, (int __user *)arg);
1190 return -ENOIOCTLCMD;
1195 EXPORT_SYMBOL(udp_ioctl);
1198 * This should be easy, if there is something there we
1199 * return it, otherwise we block.
1202 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1203 size_t len, int noblock, int flags, int *addr_len)
1205 struct inet_sock *inet = inet_sk(sk);
1206 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1207 struct sk_buff *skb;
1208 unsigned int ulen, copied;
1209 int peeked, off = 0;
1211 int is_udplite = IS_UDPLITE(sk);
1215 * Check any passed addresses
1218 *addr_len = sizeof(*sin);
1220 if (flags & MSG_ERRQUEUE)
1221 return ip_recv_error(sk, msg, len);
1224 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1225 &peeked, &off, &err);
1229 ulen = skb->len - sizeof(struct udphdr);
1233 else if (copied < ulen)
1234 msg->msg_flags |= MSG_TRUNC;
1237 * If checksum is needed at all, try to do it while copying the
1238 * data. If the data is truncated, or if we only want a partial
1239 * coverage checksum (UDP-Lite), do it before the copy.
1242 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1243 if (udp_lib_checksum_complete(skb))
1247 if (skb_csum_unnecessary(skb))
1248 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1249 msg->msg_iov, copied);
1251 err = skb_copy_and_csum_datagram_iovec(skb,
1252 sizeof(struct udphdr),
1259 if (unlikely(err)) {
1260 trace_kfree_skb(skb, udp_recvmsg);
1262 atomic_inc(&sk->sk_drops);
1263 UDP_INC_STATS_USER(sock_net(sk),
1264 UDP_MIB_INERRORS, is_udplite);
1270 UDP_INC_STATS_USER(sock_net(sk),
1271 UDP_MIB_INDATAGRAMS, is_udplite);
1273 sock_recv_ts_and_drops(msg, sk, skb);
1275 /* Copy the address. */
1277 sin->sin_family = AF_INET;
1278 sin->sin_port = udp_hdr(skb)->source;
1279 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1280 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1282 if (inet->cmsg_flags)
1283 ip_cmsg_recv(msg, skb);
1286 if (flags & MSG_TRUNC)
1290 skb_free_datagram_locked(sk, skb);
1295 slow = lock_sock_fast(sk);
1296 if (!skb_kill_datagram(sk, skb, flags)) {
1297 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1298 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1300 unlock_sock_fast(sk, slow);
1305 /* starting over for a new packet */
1306 msg->msg_flags &= ~MSG_TRUNC;
1311 int udp_disconnect(struct sock *sk, int flags)
1313 struct inet_sock *inet = inet_sk(sk);
1315 * 1003.1g - break association.
1318 sk->sk_state = TCP_CLOSE;
1319 inet->inet_daddr = 0;
1320 inet->inet_dport = 0;
1321 sock_rps_reset_rxhash(sk);
1322 sk->sk_bound_dev_if = 0;
1323 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1324 inet_reset_saddr(sk);
1326 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1327 sk->sk_prot->unhash(sk);
1328 inet->inet_sport = 0;
1333 EXPORT_SYMBOL(udp_disconnect);
1335 void udp_lib_unhash(struct sock *sk)
1337 if (sk_hashed(sk)) {
1338 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1339 struct udp_hslot *hslot, *hslot2;
1341 hslot = udp_hashslot(udptable, sock_net(sk),
1342 udp_sk(sk)->udp_port_hash);
1343 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1345 spin_lock_bh(&hslot->lock);
1346 if (sk_nulls_del_node_init_rcu(sk)) {
1348 inet_sk(sk)->inet_num = 0;
1349 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1351 spin_lock(&hslot2->lock);
1352 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1354 spin_unlock(&hslot2->lock);
1356 spin_unlock_bh(&hslot->lock);
1359 EXPORT_SYMBOL(udp_lib_unhash);
1362 * inet_rcv_saddr was changed, we must rehash secondary hash
1364 void udp_lib_rehash(struct sock *sk, u16 newhash)
1366 if (sk_hashed(sk)) {
1367 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1368 struct udp_hslot *hslot, *hslot2, *nhslot2;
1370 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1371 nhslot2 = udp_hashslot2(udptable, newhash);
1372 udp_sk(sk)->udp_portaddr_hash = newhash;
1373 if (hslot2 != nhslot2) {
1374 hslot = udp_hashslot(udptable, sock_net(sk),
1375 udp_sk(sk)->udp_port_hash);
1376 /* we must lock primary chain too */
1377 spin_lock_bh(&hslot->lock);
1379 spin_lock(&hslot2->lock);
1380 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1382 spin_unlock(&hslot2->lock);
1384 spin_lock(&nhslot2->lock);
1385 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1388 spin_unlock(&nhslot2->lock);
1390 spin_unlock_bh(&hslot->lock);
1394 EXPORT_SYMBOL(udp_lib_rehash);
1396 static void udp_v4_rehash(struct sock *sk)
1398 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1399 inet_sk(sk)->inet_rcv_saddr,
1400 inet_sk(sk)->inet_num);
1401 udp_lib_rehash(sk, new_hash);
1404 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1408 if (inet_sk(sk)->inet_daddr)
1409 sock_rps_save_rxhash(sk, skb);
1411 rc = sock_queue_rcv_skb(sk, skb);
1413 int is_udplite = IS_UDPLITE(sk);
1415 /* Note that an ENOMEM error is charged twice */
1417 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1419 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1421 trace_udp_fail_queue_rcv_skb(rc, sk);
1429 static struct static_key udp_encap_needed __read_mostly;
1430 void udp_encap_enable(void)
1432 if (!static_key_enabled(&udp_encap_needed))
1433 static_key_slow_inc(&udp_encap_needed);
1435 EXPORT_SYMBOL(udp_encap_enable);
1440 * >0: "udp encap" protocol resubmission
1442 * Note that in the success and error cases, the skb is assumed to
1443 * have either been requeued or freed.
1445 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1447 struct udp_sock *up = udp_sk(sk);
1449 int is_udplite = IS_UDPLITE(sk);
1452 * Charge it to the socket, dropping if the queue is full.
1454 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1458 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1459 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1462 * This is an encapsulation socket so pass the skb to
1463 * the socket's udp_encap_rcv() hook. Otherwise, just
1464 * fall through and pass this up the UDP socket.
1465 * up->encap_rcv() returns the following value:
1466 * =0 if skb was successfully passed to the encap
1467 * handler or was discarded by it.
1468 * >0 if skb should be passed on to UDP.
1469 * <0 if skb should be resubmitted as proto -N
1472 /* if we're overly short, let UDP handle it */
1473 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1474 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1477 ret = encap_rcv(sk, skb);
1479 UDP_INC_STATS_BH(sock_net(sk),
1480 UDP_MIB_INDATAGRAMS,
1486 /* FALLTHROUGH -- it's a UDP Packet */
1490 * UDP-Lite specific tests, ignored on UDP sockets
1492 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1495 * MIB statistics other than incrementing the error count are
1496 * disabled for the following two types of errors: these depend
1497 * on the application settings, not on the functioning of the
1498 * protocol stack as such.
1500 * RFC 3828 here recommends (sec 3.3): "There should also be a
1501 * way ... to ... at least let the receiving application block
1502 * delivery of packets with coverage values less than a value
1503 * provided by the application."
1505 if (up->pcrlen == 0) { /* full coverage was set */
1506 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1507 UDP_SKB_CB(skb)->cscov, skb->len);
1510 /* The next case involves violating the min. coverage requested
1511 * by the receiver. This is subtle: if receiver wants x and x is
1512 * greater than the buffersize/MTU then receiver will complain
1513 * that it wants x while sender emits packets of smaller size y.
1514 * Therefore the above ...()->partial_cov statement is essential.
1516 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1517 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1518 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1523 if (rcu_access_pointer(sk->sk_filter) &&
1524 udp_lib_checksum_complete(skb))
1528 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1533 ipv4_pktinfo_prepare(skb);
1535 if (!sock_owned_by_user(sk))
1536 rc = __udp_queue_rcv_skb(sk, skb);
1537 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1546 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1548 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1549 atomic_inc(&sk->sk_drops);
1555 static void flush_stack(struct sock **stack, unsigned int count,
1556 struct sk_buff *skb, unsigned int final)
1559 struct sk_buff *skb1 = NULL;
1562 for (i = 0; i < count; i++) {
1564 if (likely(skb1 == NULL))
1565 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1568 atomic_inc(&sk->sk_drops);
1569 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1571 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1575 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1583 * Multicasts and broadcasts go to each listener.
1585 * Note: called only from the BH handler context.
1587 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1589 __be32 saddr, __be32 daddr,
1590 struct udp_table *udptable)
1592 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1593 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1595 unsigned int i, count = 0;
1597 spin_lock(&hslot->lock);
1598 sk = sk_nulls_head(&hslot->head);
1599 dif = skb->dev->ifindex;
1600 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1602 stack[count++] = sk;
1603 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1604 daddr, uh->source, saddr, dif);
1605 if (unlikely(count == ARRAY_SIZE(stack))) {
1608 flush_stack(stack, count, skb, ~0);
1613 * before releasing chain lock, we must take a reference on sockets
1615 for (i = 0; i < count; i++)
1616 sock_hold(stack[i]);
1618 spin_unlock(&hslot->lock);
1621 * do the slow work with no lock held
1624 flush_stack(stack, count, skb, count - 1);
1626 for (i = 0; i < count; i++)
1634 /* Initialize UDP checksum. If exited with zero value (success),
1635 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1636 * Otherwise, csum completion requires chacksumming packet body,
1637 * including udp header and folding it to skb->csum.
1639 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1642 const struct iphdr *iph;
1645 UDP_SKB_CB(skb)->partial_cov = 0;
1646 UDP_SKB_CB(skb)->cscov = skb->len;
1648 if (proto == IPPROTO_UDPLITE) {
1649 err = udplite_checksum_init(skb, uh);
1655 if (uh->check == 0) {
1656 skb->ip_summed = CHECKSUM_UNNECESSARY;
1657 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1658 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1660 skb->ip_summed = CHECKSUM_UNNECESSARY;
1662 if (!skb_csum_unnecessary(skb))
1663 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1664 skb->len, proto, 0);
1665 /* Probably, we should checksum udp header (it should be in cache
1666 * in any case) and data in tiny packets (< rx copybreak).
1673 * All we need to do is get the socket, and then do a checksum.
1676 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1681 unsigned short ulen;
1682 struct rtable *rt = skb_rtable(skb);
1683 __be32 saddr, daddr;
1684 struct net *net = dev_net(skb->dev);
1687 * Validate the packet.
1689 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1690 goto drop; /* No space for header. */
1693 ulen = ntohs(uh->len);
1694 saddr = ip_hdr(skb)->saddr;
1695 daddr = ip_hdr(skb)->daddr;
1697 if (ulen > skb->len)
1700 if (proto == IPPROTO_UDP) {
1701 /* UDP validates ulen. */
1702 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1707 if (udp4_csum_init(skb, uh, proto))
1710 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1711 return __udp4_lib_mcast_deliver(net, skb, uh,
1712 saddr, daddr, udptable);
1714 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1719 sk_mark_napi_id(sk, skb);
1720 ret = udp_queue_rcv_skb(sk, skb);
1723 /* a return value > 0 means to resubmit the input, but
1724 * it wants the return to be -protocol, or 0
1731 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1735 /* No socket. Drop packet silently, if checksum is wrong */
1736 if (udp_lib_checksum_complete(skb))
1739 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1740 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1743 * Hmm. We got an UDP packet to a port to which we
1744 * don't wanna listen. Ignore it.
1750 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1751 proto == IPPROTO_UDPLITE ? "Lite" : "",
1752 &saddr, ntohs(uh->source),
1754 &daddr, ntohs(uh->dest));
1759 * RFC1122: OK. Discards the bad packet silently (as far as
1760 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1762 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1763 proto == IPPROTO_UDPLITE ? "Lite" : "",
1764 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1766 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1768 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1773 int udp_rcv(struct sk_buff *skb)
1775 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1778 void udp_destroy_sock(struct sock *sk)
1780 struct udp_sock *up = udp_sk(sk);
1781 bool slow = lock_sock_fast(sk);
1782 udp_flush_pending_frames(sk);
1783 unlock_sock_fast(sk, slow);
1784 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1785 void (*encap_destroy)(struct sock *sk);
1786 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1793 * Socket option code for UDP
1795 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1796 char __user *optval, unsigned int optlen,
1797 int (*push_pending_frames)(struct sock *))
1799 struct udp_sock *up = udp_sk(sk);
1802 int is_udplite = IS_UDPLITE(sk);
1804 if (optlen < sizeof(int))
1807 if (get_user(val, (int __user *)optval))
1817 (*push_pending_frames)(sk);
1825 case UDP_ENCAP_ESPINUDP:
1826 case UDP_ENCAP_ESPINUDP_NON_IKE:
1827 up->encap_rcv = xfrm4_udp_encap_rcv;
1829 case UDP_ENCAP_L2TPINUDP:
1830 up->encap_type = val;
1840 * UDP-Lite's partial checksum coverage (RFC 3828).
1842 /* The sender sets actual checksum coverage length via this option.
1843 * The case coverage > packet length is handled by send module. */
1844 case UDPLITE_SEND_CSCOV:
1845 if (!is_udplite) /* Disable the option on UDP sockets */
1846 return -ENOPROTOOPT;
1847 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1849 else if (val > USHRT_MAX)
1852 up->pcflag |= UDPLITE_SEND_CC;
1855 /* The receiver specifies a minimum checksum coverage value. To make
1856 * sense, this should be set to at least 8 (as done below). If zero is
1857 * used, this again means full checksum coverage. */
1858 case UDPLITE_RECV_CSCOV:
1859 if (!is_udplite) /* Disable the option on UDP sockets */
1860 return -ENOPROTOOPT;
1861 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1863 else if (val > USHRT_MAX)
1866 up->pcflag |= UDPLITE_RECV_CC;
1876 EXPORT_SYMBOL(udp_lib_setsockopt);
1878 int udp_setsockopt(struct sock *sk, int level, int optname,
1879 char __user *optval, unsigned int optlen)
1881 if (level == SOL_UDP || level == SOL_UDPLITE)
1882 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1883 udp_push_pending_frames);
1884 return ip_setsockopt(sk, level, optname, optval, optlen);
1887 #ifdef CONFIG_COMPAT
1888 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1889 char __user *optval, unsigned int optlen)
1891 if (level == SOL_UDP || level == SOL_UDPLITE)
1892 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1893 udp_push_pending_frames);
1894 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1898 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1899 char __user *optval, int __user *optlen)
1901 struct udp_sock *up = udp_sk(sk);
1904 if (get_user(len, optlen))
1907 len = min_t(unsigned int, len, sizeof(int));
1918 val = up->encap_type;
1921 /* The following two cannot be changed on UDP sockets, the return is
1922 * always 0 (which corresponds to the full checksum coverage of UDP). */
1923 case UDPLITE_SEND_CSCOV:
1927 case UDPLITE_RECV_CSCOV:
1932 return -ENOPROTOOPT;
1935 if (put_user(len, optlen))
1937 if (copy_to_user(optval, &val, len))
1941 EXPORT_SYMBOL(udp_lib_getsockopt);
1943 int udp_getsockopt(struct sock *sk, int level, int optname,
1944 char __user *optval, int __user *optlen)
1946 if (level == SOL_UDP || level == SOL_UDPLITE)
1947 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1948 return ip_getsockopt(sk, level, optname, optval, optlen);
1951 #ifdef CONFIG_COMPAT
1952 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1953 char __user *optval, int __user *optlen)
1955 if (level == SOL_UDP || level == SOL_UDPLITE)
1956 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1957 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1961 * udp_poll - wait for a UDP event.
1962 * @file - file struct
1964 * @wait - poll table
1966 * This is same as datagram poll, except for the special case of
1967 * blocking sockets. If application is using a blocking fd
1968 * and a packet with checksum error is in the queue;
1969 * then it could get return from select indicating data available
1970 * but then block when reading it. Add special case code
1971 * to work around these arguably broken applications.
1973 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1975 unsigned int mask = datagram_poll(file, sock, wait);
1976 struct sock *sk = sock->sk;
1978 sock_rps_record_flow(sk);
1980 /* Check for false positives due to checksum errors */
1981 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1982 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1983 mask &= ~(POLLIN | POLLRDNORM);
1988 EXPORT_SYMBOL(udp_poll);
1990 struct proto udp_prot = {
1992 .owner = THIS_MODULE,
1993 .close = udp_lib_close,
1994 .connect = ip4_datagram_connect,
1995 .disconnect = udp_disconnect,
1997 .destroy = udp_destroy_sock,
1998 .setsockopt = udp_setsockopt,
1999 .getsockopt = udp_getsockopt,
2000 .sendmsg = udp_sendmsg,
2001 .recvmsg = udp_recvmsg,
2002 .sendpage = udp_sendpage,
2003 .backlog_rcv = __udp_queue_rcv_skb,
2004 .release_cb = ip4_datagram_release_cb,
2005 .hash = udp_lib_hash,
2006 .unhash = udp_lib_unhash,
2007 .rehash = udp_v4_rehash,
2008 .get_port = udp_v4_get_port,
2009 .memory_allocated = &udp_memory_allocated,
2010 .sysctl_mem = sysctl_udp_mem,
2011 .sysctl_wmem = &sysctl_udp_wmem_min,
2012 .sysctl_rmem = &sysctl_udp_rmem_min,
2013 .obj_size = sizeof(struct udp_sock),
2014 .slab_flags = SLAB_DESTROY_BY_RCU,
2015 .h.udp_table = &udp_table,
2016 #ifdef CONFIG_COMPAT
2017 .compat_setsockopt = compat_udp_setsockopt,
2018 .compat_getsockopt = compat_udp_getsockopt,
2020 .clear_sk = sk_prot_clear_portaddr_nulls,
2022 EXPORT_SYMBOL(udp_prot);
2024 /* ------------------------------------------------------------------------ */
2025 #ifdef CONFIG_PROC_FS
2027 static struct sock *udp_get_first(struct seq_file *seq, int start)
2030 struct udp_iter_state *state = seq->private;
2031 struct net *net = seq_file_net(seq);
2033 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2035 struct hlist_nulls_node *node;
2036 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2038 if (hlist_nulls_empty(&hslot->head))
2041 spin_lock_bh(&hslot->lock);
2042 sk_nulls_for_each(sk, node, &hslot->head) {
2043 if (!net_eq(sock_net(sk), net))
2045 if (sk->sk_family == state->family)
2048 spin_unlock_bh(&hslot->lock);
2055 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2057 struct udp_iter_state *state = seq->private;
2058 struct net *net = seq_file_net(seq);
2061 sk = sk_nulls_next(sk);
2062 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2065 if (state->bucket <= state->udp_table->mask)
2066 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2067 return udp_get_first(seq, state->bucket + 1);
2072 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2074 struct sock *sk = udp_get_first(seq, 0);
2077 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2079 return pos ? NULL : sk;
2082 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2084 struct udp_iter_state *state = seq->private;
2085 state->bucket = MAX_UDP_PORTS;
2087 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2090 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2094 if (v == SEQ_START_TOKEN)
2095 sk = udp_get_idx(seq, 0);
2097 sk = udp_get_next(seq, v);
2103 static void udp_seq_stop(struct seq_file *seq, void *v)
2105 struct udp_iter_state *state = seq->private;
2107 if (state->bucket <= state->udp_table->mask)
2108 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2111 int udp_seq_open(struct inode *inode, struct file *file)
2113 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2114 struct udp_iter_state *s;
2117 err = seq_open_net(inode, file, &afinfo->seq_ops,
2118 sizeof(struct udp_iter_state));
2122 s = ((struct seq_file *)file->private_data)->private;
2123 s->family = afinfo->family;
2124 s->udp_table = afinfo->udp_table;
2127 EXPORT_SYMBOL(udp_seq_open);
2129 /* ------------------------------------------------------------------------ */
2130 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2132 struct proc_dir_entry *p;
2135 afinfo->seq_ops.start = udp_seq_start;
2136 afinfo->seq_ops.next = udp_seq_next;
2137 afinfo->seq_ops.stop = udp_seq_stop;
2139 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2140 afinfo->seq_fops, afinfo);
2145 EXPORT_SYMBOL(udp_proc_register);
2147 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2149 remove_proc_entry(afinfo->name, net->proc_net);
2151 EXPORT_SYMBOL(udp_proc_unregister);
2153 /* ------------------------------------------------------------------------ */
2154 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2155 int bucket, int *len)
2157 struct inet_sock *inet = inet_sk(sp);
2158 __be32 dest = inet->inet_daddr;
2159 __be32 src = inet->inet_rcv_saddr;
2160 __u16 destp = ntohs(inet->inet_dport);
2161 __u16 srcp = ntohs(inet->inet_sport);
2163 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2164 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d%n",
2165 bucket, src, srcp, dest, destp, sp->sk_state,
2166 sk_wmem_alloc_get(sp),
2167 sk_rmem_alloc_get(sp),
2169 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2171 atomic_read(&sp->sk_refcnt), sp,
2172 atomic_read(&sp->sk_drops), len);
2175 int udp4_seq_show(struct seq_file *seq, void *v)
2177 if (v == SEQ_START_TOKEN)
2178 seq_printf(seq, "%-127s\n",
2179 " sl local_address rem_address st tx_queue "
2180 "rx_queue tr tm->when retrnsmt uid timeout "
2181 "inode ref pointer drops");
2183 struct udp_iter_state *state = seq->private;
2186 udp4_format_sock(v, seq, state->bucket, &len);
2187 seq_printf(seq, "%*s\n", 127 - len, "");
2192 static const struct file_operations udp_afinfo_seq_fops = {
2193 .owner = THIS_MODULE,
2194 .open = udp_seq_open,
2196 .llseek = seq_lseek,
2197 .release = seq_release_net
2200 /* ------------------------------------------------------------------------ */
2201 static struct udp_seq_afinfo udp4_seq_afinfo = {
2204 .udp_table = &udp_table,
2205 .seq_fops = &udp_afinfo_seq_fops,
2207 .show = udp4_seq_show,
2211 static int __net_init udp4_proc_init_net(struct net *net)
2213 return udp_proc_register(net, &udp4_seq_afinfo);
2216 static void __net_exit udp4_proc_exit_net(struct net *net)
2218 udp_proc_unregister(net, &udp4_seq_afinfo);
2221 static struct pernet_operations udp4_net_ops = {
2222 .init = udp4_proc_init_net,
2223 .exit = udp4_proc_exit_net,
2226 int __init udp4_proc_init(void)
2228 return register_pernet_subsys(&udp4_net_ops);
2231 void udp4_proc_exit(void)
2233 unregister_pernet_subsys(&udp4_net_ops);
2235 #endif /* CONFIG_PROC_FS */
2237 static __initdata unsigned long uhash_entries;
2238 static int __init set_uhash_entries(char *str)
2245 ret = kstrtoul(str, 0, &uhash_entries);
2249 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2250 uhash_entries = UDP_HTABLE_SIZE_MIN;
2253 __setup("uhash_entries=", set_uhash_entries);
2255 void __init udp_table_init(struct udp_table *table, const char *name)
2259 table->hash = alloc_large_system_hash(name,
2260 2 * sizeof(struct udp_hslot),
2262 21, /* one slot per 2 MB */
2266 UDP_HTABLE_SIZE_MIN,
2269 table->hash2 = table->hash + (table->mask + 1);
2270 for (i = 0; i <= table->mask; i++) {
2271 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2272 table->hash[i].count = 0;
2273 spin_lock_init(&table->hash[i].lock);
2275 for (i = 0; i <= table->mask; i++) {
2276 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2277 table->hash2[i].count = 0;
2278 spin_lock_init(&table->hash2[i].lock);
2282 void __init udp_init(void)
2284 unsigned long limit;
2286 udp_table_init(&udp_table, "UDP");
2287 limit = nr_free_buffer_pages() / 8;
2288 limit = max(limit, 128UL);
2289 sysctl_udp_mem[0] = limit / 4 * 3;
2290 sysctl_udp_mem[1] = limit;
2291 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2293 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2294 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2297 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
2298 netdev_features_t features)
2300 struct sk_buff *segs = ERR_PTR(-EINVAL);
2301 int mac_len = skb->mac_len;
2302 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
2303 __be16 protocol = skb->protocol;
2304 netdev_features_t enc_features;
2307 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
2310 skb->encapsulation = 0;
2311 __skb_pull(skb, tnl_hlen);
2312 skb_reset_mac_header(skb);
2313 skb_set_network_header(skb, skb_inner_network_offset(skb));
2314 skb->mac_len = skb_inner_network_offset(skb);
2315 skb->protocol = htons(ETH_P_TEB);
2317 /* segment inner packet. */
2318 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb);
2319 segs = skb_mac_gso_segment(skb, enc_features);
2320 if (!segs || IS_ERR(segs))
2323 outer_hlen = skb_tnl_header_len(skb);
2327 int udp_offset = outer_hlen - tnl_hlen;
2329 skb_reset_inner_headers(skb);
2330 skb->encapsulation = 1;
2332 skb->mac_len = mac_len;
2334 skb_push(skb, outer_hlen);
2335 skb_reset_mac_header(skb);
2336 skb_set_network_header(skb, mac_len);
2337 skb_set_transport_header(skb, udp_offset);
2339 uh->len = htons(skb->len - udp_offset);
2341 /* csum segment if tunnel sets skb with csum. */
2342 if (protocol == htons(ETH_P_IP) && unlikely(uh->check)) {
2343 struct iphdr *iph = ip_hdr(skb);
2345 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
2346 skb->len - udp_offset,
2348 uh->check = csum_fold(skb_checksum(skb, udp_offset,
2349 skb->len - udp_offset, 0));
2351 uh->check = CSUM_MANGLED_0;
2353 } else if (protocol == htons(ETH_P_IPV6)) {
2354 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
2355 u32 len = skb->len - udp_offset;
2357 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
2358 len, IPPROTO_UDP, 0);
2359 uh->check = csum_fold(skb_checksum(skb, udp_offset, len, 0));
2361 uh->check = CSUM_MANGLED_0;
2362 skb->ip_summed = CHECKSUM_NONE;
2365 skb->protocol = protocol;
2366 } while ((skb = skb->next));