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 "udp_impl.h"
114 struct udp_table udp_table __read_mostly;
115 EXPORT_SYMBOL(udp_table);
117 long sysctl_udp_mem[3] __read_mostly;
118 EXPORT_SYMBOL(sysctl_udp_mem);
120 int sysctl_udp_rmem_min __read_mostly;
121 EXPORT_SYMBOL(sysctl_udp_rmem_min);
123 int sysctl_udp_wmem_min __read_mostly;
124 EXPORT_SYMBOL(sysctl_udp_wmem_min);
126 atomic_long_t udp_memory_allocated;
127 EXPORT_SYMBOL(udp_memory_allocated);
129 #define MAX_UDP_PORTS 65536
130 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
132 static int udp_lib_lport_inuse(struct net *net, __u16 num,
133 const struct udp_hslot *hslot,
134 unsigned long *bitmap,
136 int (*saddr_comp)(const struct sock *sk1,
137 const struct sock *sk2),
141 struct hlist_nulls_node *node;
142 kuid_t uid = sock_i_uid(sk);
144 sk_nulls_for_each(sk2, node, &hslot->head)
145 if (net_eq(sock_net(sk2), net) &&
147 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
148 (!sk2->sk_reuse || !sk->sk_reuse) &&
149 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
150 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
151 (!sk2->sk_reuseport || !sk->sk_reuseport ||
152 !uid_eq(uid, sock_i_uid(sk2))) &&
153 (*saddr_comp)(sk, sk2)) {
155 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
164 * Note: we still hold spinlock of primary hash chain, so no other writer
165 * can insert/delete a socket with local_port == num
167 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
168 struct udp_hslot *hslot2,
170 int (*saddr_comp)(const struct sock *sk1,
171 const struct sock *sk2))
174 struct hlist_nulls_node *node;
175 kuid_t uid = sock_i_uid(sk);
178 spin_lock(&hslot2->lock);
179 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
180 if (net_eq(sock_net(sk2), net) &&
182 (udp_sk(sk2)->udp_port_hash == num) &&
183 (!sk2->sk_reuse || !sk->sk_reuse) &&
184 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
185 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
186 (!sk2->sk_reuseport || !sk->sk_reuseport ||
187 !uid_eq(uid, sock_i_uid(sk2))) &&
188 (*saddr_comp)(sk, sk2)) {
192 spin_unlock(&hslot2->lock);
197 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
199 * @sk: socket struct in question
200 * @snum: port number to look up
201 * @saddr_comp: AF-dependent comparison of bound local IP addresses
202 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
205 int udp_lib_get_port(struct sock *sk, unsigned short snum,
206 int (*saddr_comp)(const struct sock *sk1,
207 const struct sock *sk2),
208 unsigned int hash2_nulladdr)
210 struct udp_hslot *hslot, *hslot2;
211 struct udp_table *udptable = sk->sk_prot->h.udp_table;
213 struct net *net = sock_net(sk);
216 int low, high, remaining;
218 unsigned short first, last;
219 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
221 inet_get_local_port_range(&low, &high);
222 remaining = (high - low) + 1;
225 first = (((u64)rand * remaining) >> 32) + low;
227 * force rand to be an odd multiple of UDP_HTABLE_SIZE
229 rand = (rand | 1) * (udptable->mask + 1);
230 last = first + udptable->mask + 1;
232 hslot = udp_hashslot(udptable, net, first);
233 bitmap_zero(bitmap, PORTS_PER_CHAIN);
234 spin_lock_bh(&hslot->lock);
235 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
236 saddr_comp, udptable->log);
240 * Iterate on all possible values of snum for this hash.
241 * Using steps of an odd multiple of UDP_HTABLE_SIZE
242 * give us randomization and full range coverage.
245 if (low <= snum && snum <= high &&
246 !test_bit(snum >> udptable->log, bitmap) &&
247 !inet_is_reserved_local_port(snum))
250 } while (snum != first);
251 spin_unlock_bh(&hslot->lock);
252 } while (++first != last);
255 hslot = udp_hashslot(udptable, net, snum);
256 spin_lock_bh(&hslot->lock);
257 if (hslot->count > 10) {
259 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
261 slot2 &= udptable->mask;
262 hash2_nulladdr &= udptable->mask;
264 hslot2 = udp_hashslot2(udptable, slot2);
265 if (hslot->count < hslot2->count)
266 goto scan_primary_hash;
268 exist = udp_lib_lport_inuse2(net, snum, hslot2,
270 if (!exist && (hash2_nulladdr != slot2)) {
271 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
272 exist = udp_lib_lport_inuse2(net, snum, hslot2,
281 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
286 inet_sk(sk)->inet_num = snum;
287 udp_sk(sk)->udp_port_hash = snum;
288 udp_sk(sk)->udp_portaddr_hash ^= snum;
289 if (sk_unhashed(sk)) {
290 sk_nulls_add_node_rcu(sk, &hslot->head);
292 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
294 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
295 spin_lock(&hslot2->lock);
296 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
299 spin_unlock(&hslot2->lock);
303 spin_unlock_bh(&hslot->lock);
307 EXPORT_SYMBOL(udp_lib_get_port);
309 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
311 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
313 return (!ipv6_only_sock(sk2) &&
314 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
315 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
318 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
321 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
324 int udp_v4_get_port(struct sock *sk, unsigned short snum)
326 unsigned int hash2_nulladdr =
327 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
328 unsigned int hash2_partial =
329 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
331 /* precompute partial secondary hash */
332 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
333 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
336 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
338 __be16 sport, __be32 daddr, __be16 dport, int dif)
342 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
343 !ipv6_only_sock(sk)) {
344 struct inet_sock *inet = inet_sk(sk);
346 score = (sk->sk_family == PF_INET ? 2 : 1);
347 if (inet->inet_rcv_saddr) {
348 if (inet->inet_rcv_saddr != daddr)
352 if (inet->inet_daddr) {
353 if (inet->inet_daddr != saddr)
357 if (inet->inet_dport) {
358 if (inet->inet_dport != sport)
362 if (sk->sk_bound_dev_if) {
363 if (sk->sk_bound_dev_if != dif)
372 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
374 static inline int compute_score2(struct sock *sk, struct net *net,
375 __be32 saddr, __be16 sport,
376 __be32 daddr, unsigned int hnum, int dif)
380 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
381 struct inet_sock *inet = inet_sk(sk);
383 if (inet->inet_rcv_saddr != daddr)
385 if (inet->inet_num != hnum)
388 score = (sk->sk_family == PF_INET ? 2 : 1);
389 if (inet->inet_daddr) {
390 if (inet->inet_daddr != saddr)
394 if (inet->inet_dport) {
395 if (inet->inet_dport != sport)
399 if (sk->sk_bound_dev_if) {
400 if (sk->sk_bound_dev_if != dif)
409 /* called with read_rcu_lock() */
410 static struct sock *udp4_lib_lookup2(struct net *net,
411 __be32 saddr, __be16 sport,
412 __be32 daddr, unsigned int hnum, int dif,
413 struct udp_hslot *hslot2, unsigned int slot2)
415 struct sock *sk, *result;
416 struct hlist_nulls_node *node;
417 int score, badness, matches = 0, reuseport = 0;
423 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
424 score = compute_score2(sk, net, saddr, sport,
426 if (score > badness) {
429 reuseport = sk->sk_reuseport;
431 hash = inet_ehashfn(net, daddr, hnum,
432 saddr, htons(sport));
435 } else if (score == badness && reuseport) {
437 if (((u64)hash * matches) >> 32 == 0)
439 hash = next_pseudo_random32(hash);
443 * if the nulls value we got at the end of this lookup is
444 * not the expected one, we must restart lookup.
445 * We probably met an item that was moved to another chain.
447 if (get_nulls_value(node) != slot2)
450 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
452 else if (unlikely(compute_score2(result, net, saddr, sport,
453 daddr, hnum, dif) < badness)) {
461 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
462 * harder than this. -DaveM
464 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
465 __be16 sport, __be32 daddr, __be16 dport,
466 int dif, struct udp_table *udptable)
468 struct sock *sk, *result;
469 struct hlist_nulls_node *node;
470 unsigned short hnum = ntohs(dport);
471 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
472 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
473 int score, badness, matches = 0, reuseport = 0;
477 if (hslot->count > 10) {
478 hash2 = udp4_portaddr_hash(net, daddr, hnum);
479 slot2 = hash2 & udptable->mask;
480 hslot2 = &udptable->hash2[slot2];
481 if (hslot->count < hslot2->count)
484 result = udp4_lib_lookup2(net, saddr, sport,
488 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
489 slot2 = hash2 & udptable->mask;
490 hslot2 = &udptable->hash2[slot2];
491 if (hslot->count < hslot2->count)
494 result = udp4_lib_lookup2(net, saddr, sport,
495 htonl(INADDR_ANY), hnum, dif,
504 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
505 score = compute_score(sk, net, saddr, hnum, sport,
507 if (score > badness) {
510 reuseport = sk->sk_reuseport;
512 hash = inet_ehashfn(net, daddr, hnum,
513 saddr, htons(sport));
516 } else if (score == badness && reuseport) {
518 if (((u64)hash * matches) >> 32 == 0)
520 hash = next_pseudo_random32(hash);
524 * if the nulls value we got at the end of this lookup is
525 * not the expected one, we must restart lookup.
526 * We probably met an item that was moved to another chain.
528 if (get_nulls_value(node) != slot)
532 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
534 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
535 daddr, dport, dif) < badness)) {
543 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
545 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
546 __be16 sport, __be16 dport,
547 struct udp_table *udptable)
550 const struct iphdr *iph = ip_hdr(skb);
552 if (unlikely(sk = skb_steal_sock(skb)))
555 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
556 iph->daddr, dport, inet_iif(skb),
560 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
561 __be32 daddr, __be16 dport, int dif)
563 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
565 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
567 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
568 __be16 loc_port, __be32 loc_addr,
569 __be16 rmt_port, __be32 rmt_addr,
572 struct hlist_nulls_node *node;
574 unsigned short hnum = ntohs(loc_port);
576 sk_nulls_for_each_from(s, node) {
577 struct inet_sock *inet = inet_sk(s);
579 if (!net_eq(sock_net(s), net) ||
580 udp_sk(s)->udp_port_hash != hnum ||
581 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
582 (inet->inet_dport != rmt_port && inet->inet_dport) ||
583 (inet->inet_rcv_saddr &&
584 inet->inet_rcv_saddr != loc_addr) ||
586 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
588 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
598 * This routine is called by the ICMP module when it gets some
599 * sort of error condition. If err < 0 then the socket should
600 * be closed and the error returned to the user. If err > 0
601 * it's just the icmp type << 8 | icmp code.
602 * Header points to the ip header of the error packet. We move
603 * on past this. Then (as it used to claim before adjustment)
604 * header points to the first 8 bytes of the udp header. We need
605 * to find the appropriate port.
608 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
610 struct inet_sock *inet;
611 const struct iphdr *iph = (const struct iphdr *)skb->data;
612 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
613 const int type = icmp_hdr(skb)->type;
614 const int code = icmp_hdr(skb)->code;
618 struct net *net = dev_net(skb->dev);
620 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
621 iph->saddr, uh->source, skb->dev->ifindex, udptable);
623 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
624 return; /* No socket for error */
633 case ICMP_TIME_EXCEEDED:
636 case ICMP_SOURCE_QUENCH:
638 case ICMP_PARAMETERPROB:
642 case ICMP_DEST_UNREACH:
643 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
644 ipv4_sk_update_pmtu(skb, sk, info);
645 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
653 if (code <= NR_ICMP_UNREACH) {
654 harderr = icmp_err_convert[code].fatal;
655 err = icmp_err_convert[code].errno;
659 ipv4_sk_redirect(skb, sk);
664 * RFC1122: OK. Passes ICMP errors back to application, as per
667 if (!inet->recverr) {
668 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
671 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
674 sk->sk_error_report(sk);
679 void udp_err(struct sk_buff *skb, u32 info)
681 __udp4_lib_err(skb, info, &udp_table);
685 * Throw away all pending data and cancel the corking. Socket is locked.
687 void udp_flush_pending_frames(struct sock *sk)
689 struct udp_sock *up = udp_sk(sk);
694 ip_flush_pending_frames(sk);
697 EXPORT_SYMBOL(udp_flush_pending_frames);
700 * udp4_hwcsum - handle outgoing HW checksumming
701 * @skb: sk_buff containing the filled-in UDP header
702 * (checksum field must be zeroed out)
703 * @src: source IP address
704 * @dst: destination IP address
706 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
708 struct udphdr *uh = udp_hdr(skb);
709 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
710 int offset = skb_transport_offset(skb);
711 int len = skb->len - offset;
717 * Only one fragment on the socket.
719 skb->csum_start = skb_transport_header(skb) - skb->head;
720 skb->csum_offset = offsetof(struct udphdr, check);
721 uh->check = ~csum_tcpudp_magic(src, dst, len,
725 * HW-checksum won't work as there are two or more
726 * fragments on the socket so that all csums of sk_buffs
730 csum = csum_add(csum, frags->csum);
732 } while ((frags = frags->next));
734 csum = skb_checksum(skb, offset, hlen, csum);
735 skb->ip_summed = CHECKSUM_NONE;
737 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
739 uh->check = CSUM_MANGLED_0;
743 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
745 struct sock *sk = skb->sk;
746 struct inet_sock *inet = inet_sk(sk);
749 int is_udplite = IS_UDPLITE(sk);
750 int offset = skb_transport_offset(skb);
751 int len = skb->len - offset;
755 * Create a UDP header
758 uh->source = inet->inet_sport;
759 uh->dest = fl4->fl4_dport;
760 uh->len = htons(len);
763 if (is_udplite) /* UDP-Lite */
764 csum = udplite_csum(skb);
766 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
768 skb->ip_summed = CHECKSUM_NONE;
771 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
773 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
777 csum = udp_csum(skb);
779 /* add protocol-dependent pseudo-header */
780 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
781 sk->sk_protocol, csum);
783 uh->check = CSUM_MANGLED_0;
786 err = ip_send_skb(sock_net(sk), skb);
788 if (err == -ENOBUFS && !inet->recverr) {
789 UDP_INC_STATS_USER(sock_net(sk),
790 UDP_MIB_SNDBUFERRORS, is_udplite);
794 UDP_INC_STATS_USER(sock_net(sk),
795 UDP_MIB_OUTDATAGRAMS, is_udplite);
800 * Push out all pending data as one UDP datagram. Socket is locked.
802 int udp_push_pending_frames(struct sock *sk)
804 struct udp_sock *up = udp_sk(sk);
805 struct inet_sock *inet = inet_sk(sk);
806 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
810 skb = ip_finish_skb(sk, fl4);
814 err = udp_send_skb(skb, fl4);
821 EXPORT_SYMBOL(udp_push_pending_frames);
823 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
826 struct inet_sock *inet = inet_sk(sk);
827 struct udp_sock *up = udp_sk(sk);
828 struct flowi4 fl4_stack;
831 struct ipcm_cookie ipc;
832 struct rtable *rt = NULL;
835 __be32 daddr, faddr, saddr;
838 int err, is_udplite = IS_UDPLITE(sk);
839 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
840 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
842 struct ip_options_data opt_copy;
851 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
857 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
859 fl4 = &inet->cork.fl.u.ip4;
862 * There are pending frames.
863 * The socket lock must be held while it's corked.
866 if (likely(up->pending)) {
867 if (unlikely(up->pending != AF_INET)) {
875 ulen += sizeof(struct udphdr);
878 * Get and verify the address.
881 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
882 if (msg->msg_namelen < sizeof(*usin))
884 if (usin->sin_family != AF_INET) {
885 if (usin->sin_family != AF_UNSPEC)
886 return -EAFNOSUPPORT;
889 daddr = usin->sin_addr.s_addr;
890 dport = usin->sin_port;
894 if (sk->sk_state != TCP_ESTABLISHED)
895 return -EDESTADDRREQ;
896 daddr = inet->inet_daddr;
897 dport = inet->inet_dport;
898 /* Open fast path for connected socket.
899 Route will not be used, if at least one option is set.
903 ipc.addr = inet->inet_saddr;
905 ipc.oif = sk->sk_bound_dev_if;
907 sock_tx_timestamp(sk, &ipc.tx_flags);
909 if (msg->msg_controllen) {
910 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
918 struct ip_options_rcu *inet_opt;
921 inet_opt = rcu_dereference(inet->inet_opt);
923 memcpy(&opt_copy, inet_opt,
924 sizeof(*inet_opt) + inet_opt->opt.optlen);
925 ipc.opt = &opt_copy.opt;
931 ipc.addr = faddr = daddr;
933 if (ipc.opt && ipc.opt->opt.srr) {
936 faddr = ipc.opt->opt.faddr;
939 tos = RT_TOS(inet->tos);
940 if (sock_flag(sk, SOCK_LOCALROUTE) ||
941 (msg->msg_flags & MSG_DONTROUTE) ||
942 (ipc.opt && ipc.opt->opt.is_strictroute)) {
947 if (ipv4_is_multicast(daddr)) {
949 ipc.oif = inet->mc_index;
951 saddr = inet->mc_addr;
954 ipc.oif = inet->uc_index;
957 rt = (struct rtable *)sk_dst_check(sk, 0);
960 struct net *net = sock_net(sk);
963 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
964 RT_SCOPE_UNIVERSE, sk->sk_protocol,
965 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
966 faddr, saddr, dport, inet->inet_sport);
968 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
969 rt = ip_route_output_flow(net, fl4, sk);
973 if (err == -ENETUNREACH)
974 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
979 if ((rt->rt_flags & RTCF_BROADCAST) &&
980 !sock_flag(sk, SOCK_BROADCAST))
983 sk_dst_set(sk, dst_clone(&rt->dst));
986 if (msg->msg_flags&MSG_CONFIRM)
992 daddr = ipc.addr = fl4->daddr;
994 /* Lockless fast path for the non-corking case. */
996 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
997 sizeof(struct udphdr), &ipc, &rt,
1000 if (!IS_ERR_OR_NULL(skb))
1001 err = udp_send_skb(skb, fl4);
1006 if (unlikely(up->pending)) {
1007 /* The socket is already corked while preparing it. */
1008 /* ... which is an evident application bug. --ANK */
1011 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1016 * Now cork the socket to pend data.
1018 fl4 = &inet->cork.fl.u.ip4;
1021 fl4->fl4_dport = dport;
1022 fl4->fl4_sport = inet->inet_sport;
1023 up->pending = AF_INET;
1027 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1028 sizeof(struct udphdr), &ipc, &rt,
1029 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1031 udp_flush_pending_frames(sk);
1033 err = udp_push_pending_frames(sk);
1034 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1045 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1046 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1047 * we don't have a good statistic (IpOutDiscards but it can be too many
1048 * things). We could add another new stat but at least for now that
1049 * seems like overkill.
1051 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1052 UDP_INC_STATS_USER(sock_net(sk),
1053 UDP_MIB_SNDBUFERRORS, is_udplite);
1058 dst_confirm(&rt->dst);
1059 if (!(msg->msg_flags&MSG_PROBE) || len)
1060 goto back_from_confirm;
1064 EXPORT_SYMBOL(udp_sendmsg);
1066 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1067 size_t size, int flags)
1069 struct inet_sock *inet = inet_sk(sk);
1070 struct udp_sock *up = udp_sk(sk);
1073 if (flags & MSG_SENDPAGE_NOTLAST)
1077 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1079 /* Call udp_sendmsg to specify destination address which
1080 * sendpage interface can't pass.
1081 * This will succeed only when the socket is connected.
1083 ret = udp_sendmsg(NULL, sk, &msg, 0);
1090 if (unlikely(!up->pending)) {
1093 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1097 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1098 page, offset, size, flags);
1099 if (ret == -EOPNOTSUPP) {
1101 return sock_no_sendpage(sk->sk_socket, page, offset,
1105 udp_flush_pending_frames(sk);
1110 if (!(up->corkflag || (flags&MSG_MORE)))
1111 ret = udp_push_pending_frames(sk);
1121 * first_packet_length - return length of first packet in receive queue
1124 * Drops all bad checksum frames, until a valid one is found.
1125 * Returns the length of found skb, or 0 if none is found.
1127 static unsigned int first_packet_length(struct sock *sk)
1129 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1130 struct sk_buff *skb;
1133 __skb_queue_head_init(&list_kill);
1135 spin_lock_bh(&rcvq->lock);
1136 while ((skb = skb_peek(rcvq)) != NULL &&
1137 udp_lib_checksum_complete(skb)) {
1138 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1140 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1142 atomic_inc(&sk->sk_drops);
1143 __skb_unlink(skb, rcvq);
1144 __skb_queue_tail(&list_kill, skb);
1146 res = skb ? skb->len : 0;
1147 spin_unlock_bh(&rcvq->lock);
1149 if (!skb_queue_empty(&list_kill)) {
1150 bool slow = lock_sock_fast(sk);
1152 __skb_queue_purge(&list_kill);
1153 sk_mem_reclaim_partial(sk);
1154 unlock_sock_fast(sk, slow);
1160 * IOCTL requests applicable to the UDP protocol
1163 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1168 int amount = sk_wmem_alloc_get(sk);
1170 return put_user(amount, (int __user *)arg);
1175 unsigned int amount = first_packet_length(sk);
1179 * We will only return the amount
1180 * of this packet since that is all
1181 * that will be read.
1183 amount -= sizeof(struct udphdr);
1185 return put_user(amount, (int __user *)arg);
1189 return -ENOIOCTLCMD;
1194 EXPORT_SYMBOL(udp_ioctl);
1197 * This should be easy, if there is something there we
1198 * return it, otherwise we block.
1201 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1202 size_t len, int noblock, int flags, int *addr_len)
1204 struct inet_sock *inet = inet_sk(sk);
1205 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1206 struct sk_buff *skb;
1207 unsigned int ulen, copied;
1208 int peeked, off = 0;
1210 int is_udplite = IS_UDPLITE(sk);
1213 if (flags & MSG_ERRQUEUE)
1214 return ip_recv_error(sk, msg, len, addr_len);
1217 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1218 &peeked, &off, &err);
1222 ulen = skb->len - sizeof(struct udphdr);
1226 else if (copied < ulen)
1227 msg->msg_flags |= MSG_TRUNC;
1230 * If checksum is needed at all, try to do it while copying the
1231 * data. If the data is truncated, or if we only want a partial
1232 * coverage checksum (UDP-Lite), do it before the copy.
1235 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1236 if (udp_lib_checksum_complete(skb))
1240 if (skb_csum_unnecessary(skb))
1241 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1242 msg->msg_iov, copied);
1244 err = skb_copy_and_csum_datagram_iovec(skb,
1245 sizeof(struct udphdr),
1252 if (unlikely(err)) {
1253 trace_kfree_skb(skb, udp_recvmsg);
1255 atomic_inc(&sk->sk_drops);
1256 UDP_INC_STATS_USER(sock_net(sk),
1257 UDP_MIB_INERRORS, is_udplite);
1263 UDP_INC_STATS_USER(sock_net(sk),
1264 UDP_MIB_INDATAGRAMS, is_udplite);
1266 sock_recv_ts_and_drops(msg, sk, skb);
1268 /* Copy the address. */
1270 sin->sin_family = AF_INET;
1271 sin->sin_port = udp_hdr(skb)->source;
1272 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1273 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1274 *addr_len = sizeof(*sin);
1276 if (inet->cmsg_flags)
1277 ip_cmsg_recv(msg, skb);
1280 if (flags & MSG_TRUNC)
1284 skb_free_datagram_locked(sk, skb);
1289 slow = lock_sock_fast(sk);
1290 if (!skb_kill_datagram(sk, skb, flags)) {
1291 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1292 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1294 unlock_sock_fast(sk, slow);
1299 /* starting over for a new packet */
1300 msg->msg_flags &= ~MSG_TRUNC;
1305 int udp_disconnect(struct sock *sk, int flags)
1307 struct inet_sock *inet = inet_sk(sk);
1309 * 1003.1g - break association.
1312 sk->sk_state = TCP_CLOSE;
1313 inet->inet_daddr = 0;
1314 inet->inet_dport = 0;
1315 sock_rps_reset_rxhash(sk);
1316 sk->sk_bound_dev_if = 0;
1317 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1318 inet_reset_saddr(sk);
1320 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1321 sk->sk_prot->unhash(sk);
1322 inet->inet_sport = 0;
1327 EXPORT_SYMBOL(udp_disconnect);
1329 void udp_lib_unhash(struct sock *sk)
1331 if (sk_hashed(sk)) {
1332 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1333 struct udp_hslot *hslot, *hslot2;
1335 hslot = udp_hashslot(udptable, sock_net(sk),
1336 udp_sk(sk)->udp_port_hash);
1337 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1339 spin_lock_bh(&hslot->lock);
1340 if (sk_nulls_del_node_init_rcu(sk)) {
1342 inet_sk(sk)->inet_num = 0;
1343 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1345 spin_lock(&hslot2->lock);
1346 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1348 spin_unlock(&hslot2->lock);
1350 spin_unlock_bh(&hslot->lock);
1353 EXPORT_SYMBOL(udp_lib_unhash);
1356 * inet_rcv_saddr was changed, we must rehash secondary hash
1358 void udp_lib_rehash(struct sock *sk, u16 newhash)
1360 if (sk_hashed(sk)) {
1361 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1362 struct udp_hslot *hslot, *hslot2, *nhslot2;
1364 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1365 nhslot2 = udp_hashslot2(udptable, newhash);
1366 udp_sk(sk)->udp_portaddr_hash = newhash;
1367 if (hslot2 != nhslot2) {
1368 hslot = udp_hashslot(udptable, sock_net(sk),
1369 udp_sk(sk)->udp_port_hash);
1370 /* we must lock primary chain too */
1371 spin_lock_bh(&hslot->lock);
1373 spin_lock(&hslot2->lock);
1374 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1376 spin_unlock(&hslot2->lock);
1378 spin_lock(&nhslot2->lock);
1379 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1382 spin_unlock(&nhslot2->lock);
1384 spin_unlock_bh(&hslot->lock);
1388 EXPORT_SYMBOL(udp_lib_rehash);
1390 static void udp_v4_rehash(struct sock *sk)
1392 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1393 inet_sk(sk)->inet_rcv_saddr,
1394 inet_sk(sk)->inet_num);
1395 udp_lib_rehash(sk, new_hash);
1398 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1402 if (inet_sk(sk)->inet_daddr)
1403 sock_rps_save_rxhash(sk, skb);
1405 rc = sock_queue_rcv_skb(sk, skb);
1407 int is_udplite = IS_UDPLITE(sk);
1409 /* Note that an ENOMEM error is charged twice */
1411 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1413 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1415 trace_udp_fail_queue_rcv_skb(rc, sk);
1423 static struct static_key udp_encap_needed __read_mostly;
1424 void udp_encap_enable(void)
1426 if (!static_key_enabled(&udp_encap_needed))
1427 static_key_slow_inc(&udp_encap_needed);
1429 EXPORT_SYMBOL(udp_encap_enable);
1434 * >0: "udp encap" protocol resubmission
1436 * Note that in the success and error cases, the skb is assumed to
1437 * have either been requeued or freed.
1439 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1441 struct udp_sock *up = udp_sk(sk);
1443 int is_udplite = IS_UDPLITE(sk);
1446 * Charge it to the socket, dropping if the queue is full.
1448 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1452 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1453 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1456 * This is an encapsulation socket so pass the skb to
1457 * the socket's udp_encap_rcv() hook. Otherwise, just
1458 * fall through and pass this up the UDP socket.
1459 * up->encap_rcv() returns the following value:
1460 * =0 if skb was successfully passed to the encap
1461 * handler or was discarded by it.
1462 * >0 if skb should be passed on to UDP.
1463 * <0 if skb should be resubmitted as proto -N
1466 /* if we're overly short, let UDP handle it */
1467 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1468 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1471 ret = encap_rcv(sk, skb);
1473 UDP_INC_STATS_BH(sock_net(sk),
1474 UDP_MIB_INDATAGRAMS,
1480 /* FALLTHROUGH -- it's a UDP Packet */
1484 * UDP-Lite specific tests, ignored on UDP sockets
1486 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1489 * MIB statistics other than incrementing the error count are
1490 * disabled for the following two types of errors: these depend
1491 * on the application settings, not on the functioning of the
1492 * protocol stack as such.
1494 * RFC 3828 here recommends (sec 3.3): "There should also be a
1495 * way ... to ... at least let the receiving application block
1496 * delivery of packets with coverage values less than a value
1497 * provided by the application."
1499 if (up->pcrlen == 0) { /* full coverage was set */
1500 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1501 UDP_SKB_CB(skb)->cscov, skb->len);
1504 /* The next case involves violating the min. coverage requested
1505 * by the receiver. This is subtle: if receiver wants x and x is
1506 * greater than the buffersize/MTU then receiver will complain
1507 * that it wants x while sender emits packets of smaller size y.
1508 * Therefore the above ...()->partial_cov statement is essential.
1510 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1511 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1512 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1517 if (rcu_access_pointer(sk->sk_filter) &&
1518 udp_lib_checksum_complete(skb))
1522 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1527 ipv4_pktinfo_prepare(skb);
1529 if (!sock_owned_by_user(sk))
1530 rc = __udp_queue_rcv_skb(sk, skb);
1531 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1540 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1542 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1543 atomic_inc(&sk->sk_drops);
1549 static void flush_stack(struct sock **stack, unsigned int count,
1550 struct sk_buff *skb, unsigned int final)
1553 struct sk_buff *skb1 = NULL;
1556 for (i = 0; i < count; i++) {
1558 if (likely(skb1 == NULL))
1559 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1562 atomic_inc(&sk->sk_drops);
1563 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1565 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1569 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1577 * Multicasts and broadcasts go to each listener.
1579 * Note: called only from the BH handler context.
1581 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1583 __be32 saddr, __be32 daddr,
1584 struct udp_table *udptable)
1586 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1587 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1589 unsigned int i, count = 0;
1591 spin_lock(&hslot->lock);
1592 sk = sk_nulls_head(&hslot->head);
1593 dif = skb->dev->ifindex;
1594 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1596 stack[count++] = sk;
1597 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1598 daddr, uh->source, saddr, dif);
1599 if (unlikely(count == ARRAY_SIZE(stack))) {
1602 flush_stack(stack, count, skb, ~0);
1607 * before releasing chain lock, we must take a reference on sockets
1609 for (i = 0; i < count; i++)
1610 sock_hold(stack[i]);
1612 spin_unlock(&hslot->lock);
1615 * do the slow work with no lock held
1618 flush_stack(stack, count, skb, count - 1);
1620 for (i = 0; i < count; i++)
1628 /* Initialize UDP checksum. If exited with zero value (success),
1629 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1630 * Otherwise, csum completion requires chacksumming packet body,
1631 * including udp header and folding it to skb->csum.
1633 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1636 const struct iphdr *iph;
1639 UDP_SKB_CB(skb)->partial_cov = 0;
1640 UDP_SKB_CB(skb)->cscov = skb->len;
1642 if (proto == IPPROTO_UDPLITE) {
1643 err = udplite_checksum_init(skb, uh);
1649 if (uh->check == 0) {
1650 skb->ip_summed = CHECKSUM_UNNECESSARY;
1651 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1652 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1654 skb->ip_summed = CHECKSUM_UNNECESSARY;
1656 if (!skb_csum_unnecessary(skb))
1657 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1658 skb->len, proto, 0);
1659 /* Probably, we should checksum udp header (it should be in cache
1660 * in any case) and data in tiny packets (< rx copybreak).
1667 * All we need to do is get the socket, and then do a checksum.
1670 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1675 unsigned short ulen;
1676 struct rtable *rt = skb_rtable(skb);
1677 __be32 saddr, daddr;
1678 struct net *net = dev_net(skb->dev);
1681 * Validate the packet.
1683 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1684 goto drop; /* No space for header. */
1687 ulen = ntohs(uh->len);
1688 saddr = ip_hdr(skb)->saddr;
1689 daddr = ip_hdr(skb)->daddr;
1691 if (ulen > skb->len)
1694 if (proto == IPPROTO_UDP) {
1695 /* UDP validates ulen. */
1696 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1701 if (udp4_csum_init(skb, uh, proto))
1704 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1705 return __udp4_lib_mcast_deliver(net, skb, uh,
1706 saddr, daddr, udptable);
1708 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1711 int ret = udp_queue_rcv_skb(sk, skb);
1714 /* a return value > 0 means to resubmit the input, but
1715 * it wants the return to be -protocol, or 0
1722 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1726 /* No socket. Drop packet silently, if checksum is wrong */
1727 if (udp_lib_checksum_complete(skb))
1730 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1731 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1734 * Hmm. We got an UDP packet to a port to which we
1735 * don't wanna listen. Ignore it.
1741 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1742 proto == IPPROTO_UDPLITE ? "Lite" : "",
1743 &saddr, ntohs(uh->source),
1745 &daddr, ntohs(uh->dest));
1750 * RFC1122: OK. Discards the bad packet silently (as far as
1751 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1753 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1754 proto == IPPROTO_UDPLITE ? "Lite" : "",
1755 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1757 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1759 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1764 int udp_rcv(struct sk_buff *skb)
1766 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1769 void udp_destroy_sock(struct sock *sk)
1771 struct udp_sock *up = udp_sk(sk);
1772 bool slow = lock_sock_fast(sk);
1773 udp_flush_pending_frames(sk);
1774 unlock_sock_fast(sk, slow);
1775 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1776 void (*encap_destroy)(struct sock *sk);
1777 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1784 * Socket option code for UDP
1786 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1787 char __user *optval, unsigned int optlen,
1788 int (*push_pending_frames)(struct sock *))
1790 struct udp_sock *up = udp_sk(sk);
1793 int is_udplite = IS_UDPLITE(sk);
1795 if (optlen < sizeof(int))
1798 if (get_user(val, (int __user *)optval))
1808 (*push_pending_frames)(sk);
1816 case UDP_ENCAP_ESPINUDP:
1817 case UDP_ENCAP_ESPINUDP_NON_IKE:
1818 up->encap_rcv = xfrm4_udp_encap_rcv;
1820 case UDP_ENCAP_L2TPINUDP:
1821 up->encap_type = val;
1831 * UDP-Lite's partial checksum coverage (RFC 3828).
1833 /* The sender sets actual checksum coverage length via this option.
1834 * The case coverage > packet length is handled by send module. */
1835 case UDPLITE_SEND_CSCOV:
1836 if (!is_udplite) /* Disable the option on UDP sockets */
1837 return -ENOPROTOOPT;
1838 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1840 else if (val > USHRT_MAX)
1843 up->pcflag |= UDPLITE_SEND_CC;
1846 /* The receiver specifies a minimum checksum coverage value. To make
1847 * sense, this should be set to at least 8 (as done below). If zero is
1848 * used, this again means full checksum coverage. */
1849 case UDPLITE_RECV_CSCOV:
1850 if (!is_udplite) /* Disable the option on UDP sockets */
1851 return -ENOPROTOOPT;
1852 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1854 else if (val > USHRT_MAX)
1857 up->pcflag |= UDPLITE_RECV_CC;
1867 EXPORT_SYMBOL(udp_lib_setsockopt);
1869 int udp_setsockopt(struct sock *sk, int level, int optname,
1870 char __user *optval, unsigned int optlen)
1872 if (level == SOL_UDP || level == SOL_UDPLITE)
1873 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1874 udp_push_pending_frames);
1875 return ip_setsockopt(sk, level, optname, optval, optlen);
1878 #ifdef CONFIG_COMPAT
1879 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1880 char __user *optval, unsigned int optlen)
1882 if (level == SOL_UDP || level == SOL_UDPLITE)
1883 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1884 udp_push_pending_frames);
1885 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1889 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1890 char __user *optval, int __user *optlen)
1892 struct udp_sock *up = udp_sk(sk);
1895 if (get_user(len, optlen))
1898 len = min_t(unsigned int, len, sizeof(int));
1909 val = up->encap_type;
1912 /* The following two cannot be changed on UDP sockets, the return is
1913 * always 0 (which corresponds to the full checksum coverage of UDP). */
1914 case UDPLITE_SEND_CSCOV:
1918 case UDPLITE_RECV_CSCOV:
1923 return -ENOPROTOOPT;
1926 if (put_user(len, optlen))
1928 if (copy_to_user(optval, &val, len))
1932 EXPORT_SYMBOL(udp_lib_getsockopt);
1934 int udp_getsockopt(struct sock *sk, int level, int optname,
1935 char __user *optval, int __user *optlen)
1937 if (level == SOL_UDP || level == SOL_UDPLITE)
1938 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1939 return ip_getsockopt(sk, level, optname, optval, optlen);
1942 #ifdef CONFIG_COMPAT
1943 int compat_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 compat_ip_getsockopt(sk, level, optname, optval, optlen);
1952 * udp_poll - wait for a UDP event.
1953 * @file - file struct
1955 * @wait - poll table
1957 * This is same as datagram poll, except for the special case of
1958 * blocking sockets. If application is using a blocking fd
1959 * and a packet with checksum error is in the queue;
1960 * then it could get return from select indicating data available
1961 * but then block when reading it. Add special case code
1962 * to work around these arguably broken applications.
1964 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1966 unsigned int mask = datagram_poll(file, sock, wait);
1967 struct sock *sk = sock->sk;
1969 /* Check for false positives due to checksum errors */
1970 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1971 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1972 mask &= ~(POLLIN | POLLRDNORM);
1977 EXPORT_SYMBOL(udp_poll);
1979 struct proto udp_prot = {
1981 .owner = THIS_MODULE,
1982 .close = udp_lib_close,
1983 .connect = ip4_datagram_connect,
1984 .disconnect = udp_disconnect,
1986 .destroy = udp_destroy_sock,
1987 .setsockopt = udp_setsockopt,
1988 .getsockopt = udp_getsockopt,
1989 .sendmsg = udp_sendmsg,
1990 .recvmsg = udp_recvmsg,
1991 .sendpage = udp_sendpage,
1992 .backlog_rcv = __udp_queue_rcv_skb,
1993 .release_cb = ip4_datagram_release_cb,
1994 .hash = udp_lib_hash,
1995 .unhash = udp_lib_unhash,
1996 .rehash = udp_v4_rehash,
1997 .get_port = udp_v4_get_port,
1998 .memory_allocated = &udp_memory_allocated,
1999 .sysctl_mem = sysctl_udp_mem,
2000 .sysctl_wmem = &sysctl_udp_wmem_min,
2001 .sysctl_rmem = &sysctl_udp_rmem_min,
2002 .obj_size = sizeof(struct udp_sock),
2003 .slab_flags = SLAB_DESTROY_BY_RCU,
2004 .h.udp_table = &udp_table,
2005 #ifdef CONFIG_COMPAT
2006 .compat_setsockopt = compat_udp_setsockopt,
2007 .compat_getsockopt = compat_udp_getsockopt,
2009 .clear_sk = sk_prot_clear_portaddr_nulls,
2011 EXPORT_SYMBOL(udp_prot);
2013 /* ------------------------------------------------------------------------ */
2014 #ifdef CONFIG_PROC_FS
2016 static struct sock *udp_get_first(struct seq_file *seq, int start)
2019 struct udp_iter_state *state = seq->private;
2020 struct net *net = seq_file_net(seq);
2022 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2024 struct hlist_nulls_node *node;
2025 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2027 if (hlist_nulls_empty(&hslot->head))
2030 spin_lock_bh(&hslot->lock);
2031 sk_nulls_for_each(sk, node, &hslot->head) {
2032 if (!net_eq(sock_net(sk), net))
2034 if (sk->sk_family == state->family)
2037 spin_unlock_bh(&hslot->lock);
2044 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2046 struct udp_iter_state *state = seq->private;
2047 struct net *net = seq_file_net(seq);
2050 sk = sk_nulls_next(sk);
2051 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2054 if (state->bucket <= state->udp_table->mask)
2055 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2056 return udp_get_first(seq, state->bucket + 1);
2061 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2063 struct sock *sk = udp_get_first(seq, 0);
2066 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2068 return pos ? NULL : sk;
2071 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2073 struct udp_iter_state *state = seq->private;
2074 state->bucket = MAX_UDP_PORTS;
2076 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2079 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2083 if (v == SEQ_START_TOKEN)
2084 sk = udp_get_idx(seq, 0);
2086 sk = udp_get_next(seq, v);
2092 static void udp_seq_stop(struct seq_file *seq, void *v)
2094 struct udp_iter_state *state = seq->private;
2096 if (state->bucket <= state->udp_table->mask)
2097 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2100 int udp_seq_open(struct inode *inode, struct file *file)
2102 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2103 struct udp_iter_state *s;
2106 err = seq_open_net(inode, file, &afinfo->seq_ops,
2107 sizeof(struct udp_iter_state));
2111 s = ((struct seq_file *)file->private_data)->private;
2112 s->family = afinfo->family;
2113 s->udp_table = afinfo->udp_table;
2116 EXPORT_SYMBOL(udp_seq_open);
2118 /* ------------------------------------------------------------------------ */
2119 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2121 struct proc_dir_entry *p;
2124 afinfo->seq_ops.start = udp_seq_start;
2125 afinfo->seq_ops.next = udp_seq_next;
2126 afinfo->seq_ops.stop = udp_seq_stop;
2128 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2129 afinfo->seq_fops, afinfo);
2134 EXPORT_SYMBOL(udp_proc_register);
2136 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2138 remove_proc_entry(afinfo->name, net->proc_net);
2140 EXPORT_SYMBOL(udp_proc_unregister);
2142 /* ------------------------------------------------------------------------ */
2143 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2144 int bucket, int *len)
2146 struct inet_sock *inet = inet_sk(sp);
2147 __be32 dest = inet->inet_daddr;
2148 __be32 src = inet->inet_rcv_saddr;
2149 __u16 destp = ntohs(inet->inet_dport);
2150 __u16 srcp = ntohs(inet->inet_sport);
2152 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2153 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2154 bucket, src, srcp, dest, destp, sp->sk_state,
2155 sk_wmem_alloc_get(sp),
2156 sk_rmem_alloc_get(sp),
2158 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2160 atomic_read(&sp->sk_refcnt), sp,
2161 atomic_read(&sp->sk_drops), len);
2164 int udp4_seq_show(struct seq_file *seq, void *v)
2166 if (v == SEQ_START_TOKEN)
2167 seq_printf(seq, "%-127s\n",
2168 " sl local_address rem_address st tx_queue "
2169 "rx_queue tr tm->when retrnsmt uid timeout "
2170 "inode ref pointer drops");
2172 struct udp_iter_state *state = seq->private;
2175 udp4_format_sock(v, seq, state->bucket, &len);
2176 seq_printf(seq, "%*s\n", 127 - len, "");
2181 static const struct file_operations udp_afinfo_seq_fops = {
2182 .owner = THIS_MODULE,
2183 .open = udp_seq_open,
2185 .llseek = seq_lseek,
2186 .release = seq_release_net
2189 /* ------------------------------------------------------------------------ */
2190 static struct udp_seq_afinfo udp4_seq_afinfo = {
2193 .udp_table = &udp_table,
2194 .seq_fops = &udp_afinfo_seq_fops,
2196 .show = udp4_seq_show,
2200 static int __net_init udp4_proc_init_net(struct net *net)
2202 return udp_proc_register(net, &udp4_seq_afinfo);
2205 static void __net_exit udp4_proc_exit_net(struct net *net)
2207 udp_proc_unregister(net, &udp4_seq_afinfo);
2210 static struct pernet_operations udp4_net_ops = {
2211 .init = udp4_proc_init_net,
2212 .exit = udp4_proc_exit_net,
2215 int __init udp4_proc_init(void)
2217 return register_pernet_subsys(&udp4_net_ops);
2220 void udp4_proc_exit(void)
2222 unregister_pernet_subsys(&udp4_net_ops);
2224 #endif /* CONFIG_PROC_FS */
2226 static __initdata unsigned long uhash_entries;
2227 static int __init set_uhash_entries(char *str)
2234 ret = kstrtoul(str, 0, &uhash_entries);
2238 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2239 uhash_entries = UDP_HTABLE_SIZE_MIN;
2242 __setup("uhash_entries=", set_uhash_entries);
2244 void __init udp_table_init(struct udp_table *table, const char *name)
2248 table->hash = alloc_large_system_hash(name,
2249 2 * sizeof(struct udp_hslot),
2251 21, /* one slot per 2 MB */
2255 UDP_HTABLE_SIZE_MIN,
2258 table->hash2 = table->hash + (table->mask + 1);
2259 for (i = 0; i <= table->mask; i++) {
2260 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2261 table->hash[i].count = 0;
2262 spin_lock_init(&table->hash[i].lock);
2264 for (i = 0; i <= table->mask; i++) {
2265 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2266 table->hash2[i].count = 0;
2267 spin_lock_init(&table->hash2[i].lock);
2271 void __init udp_init(void)
2273 unsigned long limit;
2275 udp_table_init(&udp_table, "UDP");
2276 limit = nr_free_buffer_pages() / 8;
2277 limit = max(limit, 128UL);
2278 sysctl_udp_mem[0] = limit / 4 * 3;
2279 sysctl_udp_mem[1] = limit;
2280 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2282 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2283 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2286 int udp4_ufo_send_check(struct sk_buff *skb)
2288 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2291 if (likely(!skb->encapsulation)) {
2292 const struct iphdr *iph;
2298 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2300 skb->csum_start = skb_transport_header(skb) - skb->head;
2301 skb->csum_offset = offsetof(struct udphdr, check);
2302 skb->ip_summed = CHECKSUM_PARTIAL;
2307 static struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
2308 netdev_features_t features)
2310 struct sk_buff *segs = ERR_PTR(-EINVAL);
2311 int mac_len = skb->mac_len;
2312 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
2313 __be16 protocol = skb->protocol;
2314 netdev_features_t enc_features;
2317 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
2320 skb->encapsulation = 0;
2321 __skb_pull(skb, tnl_hlen);
2322 skb_reset_mac_header(skb);
2323 skb_set_network_header(skb, skb_inner_network_offset(skb));
2324 skb->mac_len = skb_inner_network_offset(skb);
2325 skb->protocol = htons(ETH_P_TEB);
2327 /* segment inner packet. */
2328 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb);
2329 segs = skb_mac_gso_segment(skb, enc_features);
2330 if (!segs || IS_ERR(segs))
2333 outer_hlen = skb_tnl_header_len(skb);
2337 int udp_offset = outer_hlen - tnl_hlen;
2339 skb->mac_len = mac_len;
2341 skb_push(skb, outer_hlen);
2342 skb_reset_mac_header(skb);
2343 skb_set_network_header(skb, mac_len);
2344 skb_set_transport_header(skb, udp_offset);
2346 uh->len = htons(skb->len - udp_offset);
2348 /* csum segment if tunnel sets skb with csum. */
2349 if (unlikely(uh->check)) {
2350 struct iphdr *iph = ip_hdr(skb);
2352 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
2353 skb->len - udp_offset,
2355 uh->check = csum_fold(skb_checksum(skb, udp_offset,
2356 skb->len - udp_offset, 0));
2358 uh->check = CSUM_MANGLED_0;
2361 skb->ip_summed = CHECKSUM_NONE;
2362 skb->protocol = protocol;
2363 } while ((skb = skb->next));
2368 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
2369 netdev_features_t features)
2371 struct sk_buff *segs = ERR_PTR(-EINVAL);
2373 mss = skb_shinfo(skb)->gso_size;
2374 if (unlikely(skb->len <= mss))
2377 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2378 /* Packet is from an untrusted source, reset gso_segs. */
2379 int type = skb_shinfo(skb)->gso_type;
2381 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY |
2382 SKB_GSO_UDP_TUNNEL |
2384 !(type & (SKB_GSO_UDP))))
2387 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2393 /* Fragment the skb. IP headers of the fragments are updated in
2394 * inet_gso_segment()
2396 if (skb->encapsulation && skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL)
2397 segs = skb_udp_tunnel_segment(skb, features);
2402 /* Do software UFO. Complete and fill in the UDP checksum as
2403 * HW cannot do checksum of UDP packets sent as multiple
2406 offset = skb_checksum_start_offset(skb);
2407 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2408 offset += skb->csum_offset;
2409 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2410 skb->ip_summed = CHECKSUM_NONE;
2412 segs = skb_segment(skb, features);