2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The User Datagram Protocol (UDP).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
80 #define pr_fmt(fmt) "UDP: " fmt
82 #include <asm/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/bootmem.h>
85 #include <linux/highmem.h>
86 #include <linux/swap.h>
87 #include <linux/types.h>
88 #include <linux/fcntl.h>
89 #include <linux/module.h>
90 #include <linux/socket.h>
91 #include <linux/sockios.h>
92 #include <linux/igmp.h>
94 #include <linux/errno.h>
95 #include <linux/timer.h>
97 #include <linux/inet.h>
98 #include <linux/netdevice.h>
99 #include <linux/slab.h>
100 #include <net/tcp_states.h>
101 #include <linux/skbuff.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <net/net_namespace.h>
105 #include <net/icmp.h>
106 #include <net/inet_hashtables.h>
107 #include <net/route.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <trace/events/udp.h>
111 #include <linux/static_key.h>
112 #include <trace/events/skb.h>
113 #include <net/busy_poll.h>
114 #include "udp_impl.h"
116 struct udp_table udp_table __read_mostly;
117 EXPORT_SYMBOL(udp_table);
119 long sysctl_udp_mem[3] __read_mostly;
120 EXPORT_SYMBOL(sysctl_udp_mem);
122 int sysctl_udp_rmem_min __read_mostly;
123 EXPORT_SYMBOL(sysctl_udp_rmem_min);
125 int sysctl_udp_wmem_min __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_wmem_min);
128 atomic_long_t udp_memory_allocated;
129 EXPORT_SYMBOL(udp_memory_allocated);
131 #define MAX_UDP_PORTS 65536
132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
134 static int udp_lib_lport_inuse(struct net *net, __u16 num,
135 const struct udp_hslot *hslot,
136 unsigned long *bitmap,
138 int (*saddr_comp)(const struct sock *sk1,
139 const struct sock *sk2),
143 struct hlist_nulls_node *node;
144 kuid_t uid = sock_i_uid(sk);
146 sk_nulls_for_each(sk2, node, &hslot->head)
147 if (net_eq(sock_net(sk2), net) &&
149 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
150 (!sk2->sk_reuse || !sk->sk_reuse) &&
151 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
152 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
153 (!sk2->sk_reuseport || !sk->sk_reuseport ||
154 !uid_eq(uid, sock_i_uid(sk2))) &&
155 (*saddr_comp)(sk, sk2)) {
157 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
166 * Note: we still hold spinlock of primary hash chain, so no other writer
167 * can insert/delete a socket with local_port == num
169 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
170 struct udp_hslot *hslot2,
172 int (*saddr_comp)(const struct sock *sk1,
173 const struct sock *sk2))
176 struct hlist_nulls_node *node;
177 kuid_t uid = sock_i_uid(sk);
180 spin_lock(&hslot2->lock);
181 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
182 if (net_eq(sock_net(sk2), net) &&
184 (udp_sk(sk2)->udp_port_hash == num) &&
185 (!sk2->sk_reuse || !sk->sk_reuse) &&
186 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
187 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
188 (!sk2->sk_reuseport || !sk->sk_reuseport ||
189 !uid_eq(uid, sock_i_uid(sk2))) &&
190 (*saddr_comp)(sk, sk2)) {
194 spin_unlock(&hslot2->lock);
199 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
201 * @sk: socket struct in question
202 * @snum: port number to look up
203 * @saddr_comp: AF-dependent comparison of bound local IP addresses
204 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
207 int udp_lib_get_port(struct sock *sk, unsigned short snum,
208 int (*saddr_comp)(const struct sock *sk1,
209 const struct sock *sk2),
210 unsigned int hash2_nulladdr)
212 struct udp_hslot *hslot, *hslot2;
213 struct udp_table *udptable = sk->sk_prot->h.udp_table;
215 struct net *net = sock_net(sk);
218 int low, high, remaining;
220 unsigned short first, last;
221 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
223 inet_get_local_port_range(net, &low, &high);
224 remaining = (high - low) + 1;
226 rand = prandom_u32();
227 first = (((u64)rand * remaining) >> 32) + low;
229 * force rand to be an odd multiple of UDP_HTABLE_SIZE
231 rand = (rand | 1) * (udptable->mask + 1);
232 last = first + udptable->mask + 1;
234 hslot = udp_hashslot(udptable, net, first);
235 bitmap_zero(bitmap, PORTS_PER_CHAIN);
236 spin_lock_bh(&hslot->lock);
237 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
238 saddr_comp, udptable->log);
242 * Iterate on all possible values of snum for this hash.
243 * Using steps of an odd multiple of UDP_HTABLE_SIZE
244 * give us randomization and full range coverage.
247 if (low <= snum && snum <= high &&
248 !test_bit(snum >> udptable->log, bitmap) &&
249 !inet_is_local_reserved_port(net, snum))
252 } while (snum != first);
253 spin_unlock_bh(&hslot->lock);
254 } while (++first != last);
257 hslot = udp_hashslot(udptable, net, snum);
258 spin_lock_bh(&hslot->lock);
259 if (hslot->count > 10) {
261 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
263 slot2 &= udptable->mask;
264 hash2_nulladdr &= udptable->mask;
266 hslot2 = udp_hashslot2(udptable, slot2);
267 if (hslot->count < hslot2->count)
268 goto scan_primary_hash;
270 exist = udp_lib_lport_inuse2(net, snum, hslot2,
272 if (!exist && (hash2_nulladdr != slot2)) {
273 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
274 exist = udp_lib_lport_inuse2(net, snum, hslot2,
283 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
288 inet_sk(sk)->inet_num = snum;
289 udp_sk(sk)->udp_port_hash = snum;
290 udp_sk(sk)->udp_portaddr_hash ^= snum;
291 if (sk_unhashed(sk)) {
292 sk_nulls_add_node_rcu(sk, &hslot->head);
294 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
296 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
297 spin_lock(&hslot2->lock);
298 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
301 spin_unlock(&hslot2->lock);
305 spin_unlock_bh(&hslot->lock);
309 EXPORT_SYMBOL(udp_lib_get_port);
311 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
313 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
315 return (!ipv6_only_sock(sk2) &&
316 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
317 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
320 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
323 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
326 int udp_v4_get_port(struct sock *sk, unsigned short snum)
328 unsigned int hash2_nulladdr =
329 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
330 unsigned int hash2_partial =
331 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
333 /* precompute partial secondary hash */
334 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
335 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
338 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
340 __be16 sport, __be32 daddr, __be16 dport, int dif)
344 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
345 !ipv6_only_sock(sk)) {
346 struct inet_sock *inet = inet_sk(sk);
348 score = (sk->sk_family == PF_INET ? 2 : 1);
349 if (inet->inet_rcv_saddr) {
350 if (inet->inet_rcv_saddr != daddr)
354 if (inet->inet_daddr) {
355 if (inet->inet_daddr != saddr)
359 if (inet->inet_dport) {
360 if (inet->inet_dport != sport)
364 if (sk->sk_bound_dev_if) {
365 if (sk->sk_bound_dev_if != dif)
374 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
376 static inline int compute_score2(struct sock *sk, struct net *net,
377 __be32 saddr, __be16 sport,
378 __be32 daddr, unsigned int hnum, int dif)
382 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
383 struct inet_sock *inet = inet_sk(sk);
385 if (inet->inet_rcv_saddr != daddr)
387 if (inet->inet_num != hnum)
390 score = (sk->sk_family == PF_INET ? 2 : 1);
391 if (inet->inet_daddr) {
392 if (inet->inet_daddr != saddr)
396 if (inet->inet_dport) {
397 if (inet->inet_dport != sport)
401 if (sk->sk_bound_dev_if) {
402 if (sk->sk_bound_dev_if != dif)
410 static unsigned int udp_ehashfn(struct net *net, const __be32 laddr,
411 const __u16 lport, const __be32 faddr,
414 static u32 udp_ehash_secret __read_mostly;
416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
418 return __inet_ehashfn(laddr, lport, faddr, fport,
419 udp_ehash_secret + net_hash_mix(net));
423 /* called with read_rcu_lock() */
424 static struct sock *udp4_lib_lookup2(struct net *net,
425 __be32 saddr, __be16 sport,
426 __be32 daddr, unsigned int hnum, int dif,
427 struct udp_hslot *hslot2, unsigned int slot2)
429 struct sock *sk, *result;
430 struct hlist_nulls_node *node;
431 int score, badness, matches = 0, reuseport = 0;
437 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
438 score = compute_score2(sk, net, saddr, sport,
440 if (score > badness) {
443 reuseport = sk->sk_reuseport;
445 hash = udp_ehashfn(net, daddr, hnum,
449 } else if (score == badness && reuseport) {
451 if (((u64)hash * matches) >> 32 == 0)
453 hash = next_pseudo_random32(hash);
457 * if the nulls value we got at the end of this lookup is
458 * not the expected one, we must restart lookup.
459 * We probably met an item that was moved to another chain.
461 if (get_nulls_value(node) != slot2)
464 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
466 else if (unlikely(compute_score2(result, net, saddr, sport,
467 daddr, hnum, dif) < badness)) {
475 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
476 * harder than this. -DaveM
478 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
479 __be16 sport, __be32 daddr, __be16 dport,
480 int dif, struct udp_table *udptable)
482 struct sock *sk, *result;
483 struct hlist_nulls_node *node;
484 unsigned short hnum = ntohs(dport);
485 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
486 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
487 int score, badness, matches = 0, reuseport = 0;
491 if (hslot->count > 10) {
492 hash2 = udp4_portaddr_hash(net, daddr, hnum);
493 slot2 = hash2 & udptable->mask;
494 hslot2 = &udptable->hash2[slot2];
495 if (hslot->count < hslot2->count)
498 result = udp4_lib_lookup2(net, saddr, sport,
502 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
503 slot2 = hash2 & udptable->mask;
504 hslot2 = &udptable->hash2[slot2];
505 if (hslot->count < hslot2->count)
508 result = udp4_lib_lookup2(net, saddr, sport,
509 htonl(INADDR_ANY), hnum, dif,
518 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
519 score = compute_score(sk, net, saddr, hnum, sport,
521 if (score > badness) {
524 reuseport = sk->sk_reuseport;
526 hash = udp_ehashfn(net, daddr, hnum,
530 } else if (score == badness && reuseport) {
532 if (((u64)hash * matches) >> 32 == 0)
534 hash = next_pseudo_random32(hash);
538 * if the nulls value we got at the end of this lookup is
539 * not the expected one, we must restart lookup.
540 * We probably met an item that was moved to another chain.
542 if (get_nulls_value(node) != slot)
546 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
548 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
549 daddr, dport, dif) < badness)) {
557 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
559 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
560 __be16 sport, __be16 dport,
561 struct udp_table *udptable)
563 const struct iphdr *iph = ip_hdr(skb);
565 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
566 iph->daddr, dport, inet_iif(skb),
570 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
571 __be32 daddr, __be16 dport, int dif)
573 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
575 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
577 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
578 __be16 loc_port, __be32 loc_addr,
579 __be16 rmt_port, __be32 rmt_addr,
580 int dif, unsigned short hnum)
582 struct inet_sock *inet = inet_sk(sk);
584 if (!net_eq(sock_net(sk), net) ||
585 udp_sk(sk)->udp_port_hash != hnum ||
586 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
587 (inet->inet_dport != rmt_port && inet->inet_dport) ||
588 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
589 ipv6_only_sock(sk) ||
590 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
592 if (!ip_mc_sf_allow(sk, 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 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
708 struct udphdr *uh = udp_hdr(skb);
709 int offset = skb_transport_offset(skb);
710 int len = skb->len - offset;
714 if (!skb_has_frag_list(skb)) {
716 * Only one fragment on the socket.
718 skb->csum_start = skb_transport_header(skb) - skb->head;
719 skb->csum_offset = offsetof(struct udphdr, check);
720 uh->check = ~csum_tcpudp_magic(src, dst, len,
723 struct sk_buff *frags;
726 * HW-checksum won't work as there are two or more
727 * fragments on the socket so that all csums of sk_buffs
730 skb_walk_frags(skb, frags) {
731 csum = csum_add(csum, frags->csum);
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 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
746 * for the simple case like when setting the checksum for a UDP tunnel.
748 void udp_set_csum(bool nocheck, struct sk_buff *skb,
749 __be32 saddr, __be32 daddr, int len)
751 struct udphdr *uh = udp_hdr(skb);
755 else if (skb_is_gso(skb))
756 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
757 else if (skb_dst(skb) && skb_dst(skb)->dev &&
758 (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) {
760 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
762 skb->ip_summed = CHECKSUM_PARTIAL;
763 skb->csum_start = skb_transport_header(skb) - skb->head;
764 skb->csum_offset = offsetof(struct udphdr, check);
765 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
769 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
772 csum = skb_checksum(skb, 0, len, 0);
773 uh->check = udp_v4_check(len, saddr, daddr, csum);
775 uh->check = CSUM_MANGLED_0;
777 skb->ip_summed = CHECKSUM_UNNECESSARY;
780 EXPORT_SYMBOL(udp_set_csum);
782 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
784 struct sock *sk = skb->sk;
785 struct inet_sock *inet = inet_sk(sk);
788 int is_udplite = IS_UDPLITE(sk);
789 int offset = skb_transport_offset(skb);
790 int len = skb->len - offset;
794 * Create a UDP header
797 uh->source = inet->inet_sport;
798 uh->dest = fl4->fl4_dport;
799 uh->len = htons(len);
802 if (is_udplite) /* UDP-Lite */
803 csum = udplite_csum(skb);
805 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
807 skb->ip_summed = CHECKSUM_NONE;
810 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
812 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
816 csum = udp_csum(skb);
818 /* add protocol-dependent pseudo-header */
819 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
820 sk->sk_protocol, csum);
822 uh->check = CSUM_MANGLED_0;
825 err = ip_send_skb(sock_net(sk), skb);
827 if (err == -ENOBUFS && !inet->recverr) {
828 UDP_INC_STATS_USER(sock_net(sk),
829 UDP_MIB_SNDBUFERRORS, is_udplite);
833 UDP_INC_STATS_USER(sock_net(sk),
834 UDP_MIB_OUTDATAGRAMS, is_udplite);
839 * Push out all pending data as one UDP datagram. Socket is locked.
841 int udp_push_pending_frames(struct sock *sk)
843 struct udp_sock *up = udp_sk(sk);
844 struct inet_sock *inet = inet_sk(sk);
845 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
849 skb = ip_finish_skb(sk, fl4);
853 err = udp_send_skb(skb, fl4);
860 EXPORT_SYMBOL(udp_push_pending_frames);
862 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
865 struct inet_sock *inet = inet_sk(sk);
866 struct udp_sock *up = udp_sk(sk);
867 struct flowi4 fl4_stack;
870 struct ipcm_cookie ipc;
871 struct rtable *rt = NULL;
874 __be32 daddr, faddr, saddr;
877 int err, is_udplite = IS_UDPLITE(sk);
878 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
879 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
881 struct ip_options_data opt_copy;
890 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
898 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
900 fl4 = &inet->cork.fl.u.ip4;
903 * There are pending frames.
904 * The socket lock must be held while it's corked.
907 if (likely(up->pending)) {
908 if (unlikely(up->pending != AF_INET)) {
916 ulen += sizeof(struct udphdr);
919 * Get and verify the address.
922 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
923 if (msg->msg_namelen < sizeof(*usin))
925 if (usin->sin_family != AF_INET) {
926 if (usin->sin_family != AF_UNSPEC)
927 return -EAFNOSUPPORT;
930 daddr = usin->sin_addr.s_addr;
931 dport = usin->sin_port;
935 if (sk->sk_state != TCP_ESTABLISHED)
936 return -EDESTADDRREQ;
937 daddr = inet->inet_daddr;
938 dport = inet->inet_dport;
939 /* Open fast path for connected socket.
940 Route will not be used, if at least one option is set.
944 ipc.addr = inet->inet_saddr;
946 ipc.oif = sk->sk_bound_dev_if;
948 sock_tx_timestamp(sk, &ipc.tx_flags);
950 if (msg->msg_controllen) {
951 err = ip_cmsg_send(sock_net(sk), msg, &ipc,
952 sk->sk_family == AF_INET6);
960 struct ip_options_rcu *inet_opt;
963 inet_opt = rcu_dereference(inet->inet_opt);
965 memcpy(&opt_copy, inet_opt,
966 sizeof(*inet_opt) + inet_opt->opt.optlen);
967 ipc.opt = &opt_copy.opt;
973 ipc.addr = faddr = daddr;
975 if (ipc.opt && ipc.opt->opt.srr) {
978 faddr = ipc.opt->opt.faddr;
981 tos = get_rttos(&ipc, inet);
982 if (sock_flag(sk, SOCK_LOCALROUTE) ||
983 (msg->msg_flags & MSG_DONTROUTE) ||
984 (ipc.opt && ipc.opt->opt.is_strictroute)) {
989 if (ipv4_is_multicast(daddr)) {
991 ipc.oif = inet->mc_index;
993 saddr = inet->mc_addr;
996 ipc.oif = inet->uc_index;
999 rt = (struct rtable *)sk_dst_check(sk, 0);
1002 struct net *net = sock_net(sk);
1005 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1006 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1007 inet_sk_flowi_flags(sk),
1008 faddr, saddr, dport, inet->inet_sport);
1010 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1011 rt = ip_route_output_flow(net, fl4, sk);
1015 if (err == -ENETUNREACH)
1016 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1021 if ((rt->rt_flags & RTCF_BROADCAST) &&
1022 !sock_flag(sk, SOCK_BROADCAST))
1025 sk_dst_set(sk, dst_clone(&rt->dst));
1028 if (msg->msg_flags&MSG_CONFIRM)
1034 daddr = ipc.addr = fl4->daddr;
1036 /* Lockless fast path for the non-corking case. */
1038 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
1039 sizeof(struct udphdr), &ipc, &rt,
1042 if (!IS_ERR_OR_NULL(skb))
1043 err = udp_send_skb(skb, fl4);
1048 if (unlikely(up->pending)) {
1049 /* The socket is already corked while preparing it. */
1050 /* ... which is an evident application bug. --ANK */
1053 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1058 * Now cork the socket to pend data.
1060 fl4 = &inet->cork.fl.u.ip4;
1063 fl4->fl4_dport = dport;
1064 fl4->fl4_sport = inet->inet_sport;
1065 up->pending = AF_INET;
1069 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1070 sizeof(struct udphdr), &ipc, &rt,
1071 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1073 udp_flush_pending_frames(sk);
1075 err = udp_push_pending_frames(sk);
1076 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1087 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1088 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1089 * we don't have a good statistic (IpOutDiscards but it can be too many
1090 * things). We could add another new stat but at least for now that
1091 * seems like overkill.
1093 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1094 UDP_INC_STATS_USER(sock_net(sk),
1095 UDP_MIB_SNDBUFERRORS, is_udplite);
1100 dst_confirm(&rt->dst);
1101 if (!(msg->msg_flags&MSG_PROBE) || len)
1102 goto back_from_confirm;
1106 EXPORT_SYMBOL(udp_sendmsg);
1108 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1109 size_t size, int flags)
1111 struct inet_sock *inet = inet_sk(sk);
1112 struct udp_sock *up = udp_sk(sk);
1115 if (flags & MSG_SENDPAGE_NOTLAST)
1119 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1121 /* Call udp_sendmsg to specify destination address which
1122 * sendpage interface can't pass.
1123 * This will succeed only when the socket is connected.
1125 ret = udp_sendmsg(NULL, sk, &msg, 0);
1132 if (unlikely(!up->pending)) {
1135 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1139 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1140 page, offset, size, flags);
1141 if (ret == -EOPNOTSUPP) {
1143 return sock_no_sendpage(sk->sk_socket, page, offset,
1147 udp_flush_pending_frames(sk);
1152 if (!(up->corkflag || (flags&MSG_MORE)))
1153 ret = udp_push_pending_frames(sk);
1163 * first_packet_length - return length of first packet in receive queue
1166 * Drops all bad checksum frames, until a valid one is found.
1167 * Returns the length of found skb, or 0 if none is found.
1169 static unsigned int first_packet_length(struct sock *sk)
1171 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1172 struct sk_buff *skb;
1175 __skb_queue_head_init(&list_kill);
1177 spin_lock_bh(&rcvq->lock);
1178 while ((skb = skb_peek(rcvq)) != NULL &&
1179 udp_lib_checksum_complete(skb)) {
1180 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1182 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1184 atomic_inc(&sk->sk_drops);
1185 __skb_unlink(skb, rcvq);
1186 __skb_queue_tail(&list_kill, skb);
1188 res = skb ? skb->len : 0;
1189 spin_unlock_bh(&rcvq->lock);
1191 if (!skb_queue_empty(&list_kill)) {
1192 bool slow = lock_sock_fast(sk);
1194 __skb_queue_purge(&list_kill);
1195 sk_mem_reclaim_partial(sk);
1196 unlock_sock_fast(sk, slow);
1202 * IOCTL requests applicable to the UDP protocol
1205 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1210 int amount = sk_wmem_alloc_get(sk);
1212 return put_user(amount, (int __user *)arg);
1217 unsigned int amount = first_packet_length(sk);
1221 * We will only return the amount
1222 * of this packet since that is all
1223 * that will be read.
1225 amount -= sizeof(struct udphdr);
1227 return put_user(amount, (int __user *)arg);
1231 return -ENOIOCTLCMD;
1236 EXPORT_SYMBOL(udp_ioctl);
1239 * This should be easy, if there is something there we
1240 * return it, otherwise we block.
1243 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1244 size_t len, int noblock, int flags, int *addr_len)
1246 struct inet_sock *inet = inet_sk(sk);
1247 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1248 struct sk_buff *skb;
1249 unsigned int ulen, copied;
1250 int peeked, off = 0;
1252 int is_udplite = IS_UDPLITE(sk);
1255 if (flags & MSG_ERRQUEUE)
1256 return ip_recv_error(sk, msg, len, addr_len);
1259 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1260 &peeked, &off, &err);
1264 ulen = skb->len - sizeof(struct udphdr);
1268 else if (copied < ulen)
1269 msg->msg_flags |= MSG_TRUNC;
1272 * If checksum is needed at all, try to do it while copying the
1273 * data. If the data is truncated, or if we only want a partial
1274 * coverage checksum (UDP-Lite), do it before the copy.
1277 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1278 if (udp_lib_checksum_complete(skb))
1282 if (skb_csum_unnecessary(skb))
1283 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1284 msg->msg_iov, copied);
1286 err = skb_copy_and_csum_datagram_iovec(skb,
1287 sizeof(struct udphdr),
1294 if (unlikely(err)) {
1295 trace_kfree_skb(skb, udp_recvmsg);
1297 atomic_inc(&sk->sk_drops);
1298 UDP_INC_STATS_USER(sock_net(sk),
1299 UDP_MIB_INERRORS, is_udplite);
1305 UDP_INC_STATS_USER(sock_net(sk),
1306 UDP_MIB_INDATAGRAMS, is_udplite);
1308 sock_recv_ts_and_drops(msg, sk, skb);
1310 /* Copy the address. */
1312 sin->sin_family = AF_INET;
1313 sin->sin_port = udp_hdr(skb)->source;
1314 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1315 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1316 *addr_len = sizeof(*sin);
1318 if (inet->cmsg_flags)
1319 ip_cmsg_recv(msg, skb);
1322 if (flags & MSG_TRUNC)
1326 skb_free_datagram_locked(sk, skb);
1331 slow = lock_sock_fast(sk);
1332 if (!skb_kill_datagram(sk, skb, flags)) {
1333 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1334 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1336 unlock_sock_fast(sk, slow);
1341 /* starting over for a new packet */
1342 msg->msg_flags &= ~MSG_TRUNC;
1347 int udp_disconnect(struct sock *sk, int flags)
1349 struct inet_sock *inet = inet_sk(sk);
1351 * 1003.1g - break association.
1354 sk->sk_state = TCP_CLOSE;
1355 inet->inet_daddr = 0;
1356 inet->inet_dport = 0;
1357 sock_rps_reset_rxhash(sk);
1358 sk->sk_bound_dev_if = 0;
1359 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1360 inet_reset_saddr(sk);
1362 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1363 sk->sk_prot->unhash(sk);
1364 inet->inet_sport = 0;
1369 EXPORT_SYMBOL(udp_disconnect);
1371 void udp_lib_unhash(struct sock *sk)
1373 if (sk_hashed(sk)) {
1374 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1375 struct udp_hslot *hslot, *hslot2;
1377 hslot = udp_hashslot(udptable, sock_net(sk),
1378 udp_sk(sk)->udp_port_hash);
1379 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1381 spin_lock_bh(&hslot->lock);
1382 if (sk_nulls_del_node_init_rcu(sk)) {
1384 inet_sk(sk)->inet_num = 0;
1385 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1387 spin_lock(&hslot2->lock);
1388 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1390 spin_unlock(&hslot2->lock);
1392 spin_unlock_bh(&hslot->lock);
1395 EXPORT_SYMBOL(udp_lib_unhash);
1398 * inet_rcv_saddr was changed, we must rehash secondary hash
1400 void udp_lib_rehash(struct sock *sk, u16 newhash)
1402 if (sk_hashed(sk)) {
1403 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1404 struct udp_hslot *hslot, *hslot2, *nhslot2;
1406 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1407 nhslot2 = udp_hashslot2(udptable, newhash);
1408 udp_sk(sk)->udp_portaddr_hash = newhash;
1409 if (hslot2 != nhslot2) {
1410 hslot = udp_hashslot(udptable, sock_net(sk),
1411 udp_sk(sk)->udp_port_hash);
1412 /* we must lock primary chain too */
1413 spin_lock_bh(&hslot->lock);
1415 spin_lock(&hslot2->lock);
1416 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1418 spin_unlock(&hslot2->lock);
1420 spin_lock(&nhslot2->lock);
1421 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1424 spin_unlock(&nhslot2->lock);
1426 spin_unlock_bh(&hslot->lock);
1430 EXPORT_SYMBOL(udp_lib_rehash);
1432 static void udp_v4_rehash(struct sock *sk)
1434 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1435 inet_sk(sk)->inet_rcv_saddr,
1436 inet_sk(sk)->inet_num);
1437 udp_lib_rehash(sk, new_hash);
1440 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1444 if (inet_sk(sk)->inet_daddr) {
1445 sock_rps_save_rxhash(sk, skb);
1446 sk_mark_napi_id(sk, skb);
1449 rc = sock_queue_rcv_skb(sk, skb);
1451 int is_udplite = IS_UDPLITE(sk);
1453 /* Note that an ENOMEM error is charged twice */
1455 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1457 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1459 trace_udp_fail_queue_rcv_skb(rc, sk);
1467 static struct static_key udp_encap_needed __read_mostly;
1468 void udp_encap_enable(void)
1470 if (!static_key_enabled(&udp_encap_needed))
1471 static_key_slow_inc(&udp_encap_needed);
1473 EXPORT_SYMBOL(udp_encap_enable);
1478 * >0: "udp encap" protocol resubmission
1480 * Note that in the success and error cases, the skb is assumed to
1481 * have either been requeued or freed.
1483 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1485 struct udp_sock *up = udp_sk(sk);
1487 int is_udplite = IS_UDPLITE(sk);
1490 * Charge it to the socket, dropping if the queue is full.
1492 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1496 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1497 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1500 * This is an encapsulation socket so pass the skb to
1501 * the socket's udp_encap_rcv() hook. Otherwise, just
1502 * fall through and pass this up the UDP socket.
1503 * up->encap_rcv() returns the following value:
1504 * =0 if skb was successfully passed to the encap
1505 * handler or was discarded by it.
1506 * >0 if skb should be passed on to UDP.
1507 * <0 if skb should be resubmitted as proto -N
1510 /* if we're overly short, let UDP handle it */
1511 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1512 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1515 /* Verify checksum before giving to encap */
1516 if (udp_lib_checksum_complete(skb))
1519 ret = encap_rcv(sk, skb);
1521 UDP_INC_STATS_BH(sock_net(sk),
1522 UDP_MIB_INDATAGRAMS,
1528 /* FALLTHROUGH -- it's a UDP Packet */
1532 * UDP-Lite specific tests, ignored on UDP sockets
1534 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1537 * MIB statistics other than incrementing the error count are
1538 * disabled for the following two types of errors: these depend
1539 * on the application settings, not on the functioning of the
1540 * protocol stack as such.
1542 * RFC 3828 here recommends (sec 3.3): "There should also be a
1543 * way ... to ... at least let the receiving application block
1544 * delivery of packets with coverage values less than a value
1545 * provided by the application."
1547 if (up->pcrlen == 0) { /* full coverage was set */
1548 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1549 UDP_SKB_CB(skb)->cscov, skb->len);
1552 /* The next case involves violating the min. coverage requested
1553 * by the receiver. This is subtle: if receiver wants x and x is
1554 * greater than the buffersize/MTU then receiver will complain
1555 * that it wants x while sender emits packets of smaller size y.
1556 * Therefore the above ...()->partial_cov statement is essential.
1558 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1559 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1560 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1565 if (rcu_access_pointer(sk->sk_filter) &&
1566 udp_lib_checksum_complete(skb))
1570 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
1571 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1578 ipv4_pktinfo_prepare(sk, skb);
1580 if (!sock_owned_by_user(sk))
1581 rc = __udp_queue_rcv_skb(sk, skb);
1582 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1591 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1593 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1594 atomic_inc(&sk->sk_drops);
1600 static void flush_stack(struct sock **stack, unsigned int count,
1601 struct sk_buff *skb, unsigned int final)
1604 struct sk_buff *skb1 = NULL;
1607 for (i = 0; i < count; i++) {
1609 if (likely(skb1 == NULL))
1610 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1613 atomic_inc(&sk->sk_drops);
1614 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1616 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1620 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1629 /* For TCP sockets, sk_rx_dst is protected by socket lock
1630 * For UDP, we use xchg() to guard against concurrent changes.
1632 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1634 struct dst_entry *old;
1637 old = xchg(&sk->sk_rx_dst, dst);
1642 * Multicasts and broadcasts go to each listener.
1644 * Note: called only from the BH handler context.
1646 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1648 __be32 saddr, __be32 daddr,
1649 struct udp_table *udptable)
1651 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1652 struct hlist_nulls_node *node;
1653 unsigned short hnum = ntohs(uh->dest);
1654 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1655 int dif = skb->dev->ifindex;
1656 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
1657 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1660 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1662 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1664 hslot = &udp_table.hash2[hash2];
1665 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1668 spin_lock(&hslot->lock);
1669 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
1670 if (__udp_is_mcast_sock(net, sk,
1674 if (unlikely(count == ARRAY_SIZE(stack))) {
1675 flush_stack(stack, count, skb, ~0);
1678 stack[count++] = sk;
1683 spin_unlock(&hslot->lock);
1685 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1686 if (use_hash2 && hash2 != hash2_any) {
1692 * do the slow work with no lock held
1695 flush_stack(stack, count, skb, count - 1);
1702 /* Initialize UDP checksum. If exited with zero value (success),
1703 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1704 * Otherwise, csum completion requires chacksumming packet body,
1705 * including udp header and folding it to skb->csum.
1707 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1712 UDP_SKB_CB(skb)->partial_cov = 0;
1713 UDP_SKB_CB(skb)->cscov = skb->len;
1715 if (proto == IPPROTO_UDPLITE) {
1716 err = udplite_checksum_init(skb, uh);
1721 return skb_checksum_init_zero_check(skb, proto, uh->check,
1722 inet_compute_pseudo);
1726 * All we need to do is get the socket, and then do a checksum.
1729 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1734 unsigned short ulen;
1735 struct rtable *rt = skb_rtable(skb);
1736 __be32 saddr, daddr;
1737 struct net *net = dev_net(skb->dev);
1740 * Validate the packet.
1742 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1743 goto drop; /* No space for header. */
1746 ulen = ntohs(uh->len);
1747 saddr = ip_hdr(skb)->saddr;
1748 daddr = ip_hdr(skb)->daddr;
1750 if (ulen > skb->len)
1753 if (proto == IPPROTO_UDP) {
1754 /* UDP validates ulen. */
1755 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1760 if (udp4_csum_init(skb, uh, proto))
1763 sk = skb_steal_sock(skb);
1765 struct dst_entry *dst = skb_dst(skb);
1768 if (unlikely(sk->sk_rx_dst != dst))
1769 udp_sk_rx_dst_set(sk, dst);
1771 ret = udp_queue_rcv_skb(sk, skb);
1773 /* a return value > 0 means to resubmit the input, but
1774 * it wants the return to be -protocol, or 0
1780 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1781 return __udp4_lib_mcast_deliver(net, skb, uh,
1782 saddr, daddr, udptable);
1784 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1790 ret = udp_queue_rcv_skb(sk, skb);
1793 /* a return value > 0 means to resubmit the input, but
1794 * it wants the return to be -protocol, or 0
1801 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1805 /* No socket. Drop packet silently, if checksum is wrong */
1806 if (udp_lib_checksum_complete(skb))
1809 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1810 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1813 * Hmm. We got an UDP packet to a port to which we
1814 * don't wanna listen. Ignore it.
1820 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1821 proto == IPPROTO_UDPLITE ? "Lite" : "",
1822 &saddr, ntohs(uh->source),
1824 &daddr, ntohs(uh->dest));
1829 * RFC1122: OK. Discards the bad packet silently (as far as
1830 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1832 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1833 proto == IPPROTO_UDPLITE ? "Lite" : "",
1834 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1836 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1838 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1843 /* We can only early demux multicast if there is a single matching socket.
1844 * If more than one socket found returns NULL
1846 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1847 __be16 loc_port, __be32 loc_addr,
1848 __be16 rmt_port, __be32 rmt_addr,
1851 struct sock *sk, *result;
1852 struct hlist_nulls_node *node;
1853 unsigned short hnum = ntohs(loc_port);
1854 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1855 struct udp_hslot *hslot = &udp_table.hash[slot];
1857 /* Do not bother scanning a too big list */
1858 if (hslot->count > 10)
1865 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1866 if (__udp_is_mcast_sock(net, sk,
1875 * if the nulls value we got at the end of this lookup is
1876 * not the expected one, we must restart lookup.
1877 * We probably met an item that was moved to another chain.
1879 if (get_nulls_value(node) != slot)
1884 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1886 else if (unlikely(!__udp_is_mcast_sock(net, result,
1898 /* For unicast we should only early demux connected sockets or we can
1899 * break forwarding setups. The chains here can be long so only check
1900 * if the first socket is an exact match and if not move on.
1902 static struct sock *__udp4_lib_demux_lookup(struct net *net,
1903 __be16 loc_port, __be32 loc_addr,
1904 __be16 rmt_port, __be32 rmt_addr,
1907 struct sock *sk, *result;
1908 struct hlist_nulls_node *node;
1909 unsigned short hnum = ntohs(loc_port);
1910 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1911 unsigned int slot2 = hash2 & udp_table.mask;
1912 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1913 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
1914 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1918 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
1919 if (INET_MATCH(sk, net, acookie,
1920 rmt_addr, loc_addr, ports, dif))
1922 /* Only check first socket in chain */
1927 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1929 else if (unlikely(!INET_MATCH(sk, net, acookie,
1940 void udp_v4_early_demux(struct sk_buff *skb)
1942 struct net *net = dev_net(skb->dev);
1943 const struct iphdr *iph;
1944 const struct udphdr *uh;
1946 struct dst_entry *dst;
1947 int dif = skb->dev->ifindex;
1949 /* validate the packet */
1950 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1956 if (skb->pkt_type == PACKET_BROADCAST ||
1957 skb->pkt_type == PACKET_MULTICAST)
1958 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1959 uh->source, iph->saddr, dif);
1960 else if (skb->pkt_type == PACKET_HOST)
1961 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1962 uh->source, iph->saddr, dif);
1970 skb->destructor = sock_edemux;
1971 dst = sk->sk_rx_dst;
1974 dst = dst_check(dst, 0);
1976 skb_dst_set_noref(skb, dst);
1979 int udp_rcv(struct sk_buff *skb)
1981 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1984 void udp_destroy_sock(struct sock *sk)
1986 struct udp_sock *up = udp_sk(sk);
1987 bool slow = lock_sock_fast(sk);
1988 udp_flush_pending_frames(sk);
1989 unlock_sock_fast(sk, slow);
1990 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1991 void (*encap_destroy)(struct sock *sk);
1992 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1999 * Socket option code for UDP
2001 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2002 char __user *optval, unsigned int optlen,
2003 int (*push_pending_frames)(struct sock *))
2005 struct udp_sock *up = udp_sk(sk);
2008 int is_udplite = IS_UDPLITE(sk);
2010 if (optlen < sizeof(int))
2013 if (get_user(val, (int __user *)optval))
2016 valbool = val ? 1 : 0;
2025 (*push_pending_frames)(sk);
2033 case UDP_ENCAP_ESPINUDP:
2034 case UDP_ENCAP_ESPINUDP_NON_IKE:
2035 up->encap_rcv = xfrm4_udp_encap_rcv;
2037 case UDP_ENCAP_L2TPINUDP:
2038 up->encap_type = val;
2047 case UDP_NO_CHECK6_TX:
2048 up->no_check6_tx = valbool;
2051 case UDP_NO_CHECK6_RX:
2052 up->no_check6_rx = valbool;
2056 * UDP-Lite's partial checksum coverage (RFC 3828).
2058 /* The sender sets actual checksum coverage length via this option.
2059 * The case coverage > packet length is handled by send module. */
2060 case UDPLITE_SEND_CSCOV:
2061 if (!is_udplite) /* Disable the option on UDP sockets */
2062 return -ENOPROTOOPT;
2063 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2065 else if (val > USHRT_MAX)
2068 up->pcflag |= UDPLITE_SEND_CC;
2071 /* The receiver specifies a minimum checksum coverage value. To make
2072 * sense, this should be set to at least 8 (as done below). If zero is
2073 * used, this again means full checksum coverage. */
2074 case UDPLITE_RECV_CSCOV:
2075 if (!is_udplite) /* Disable the option on UDP sockets */
2076 return -ENOPROTOOPT;
2077 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2079 else if (val > USHRT_MAX)
2082 up->pcflag |= UDPLITE_RECV_CC;
2092 EXPORT_SYMBOL(udp_lib_setsockopt);
2094 int udp_setsockopt(struct sock *sk, int level, int optname,
2095 char __user *optval, unsigned int optlen)
2097 if (level == SOL_UDP || level == SOL_UDPLITE)
2098 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2099 udp_push_pending_frames);
2100 return ip_setsockopt(sk, level, optname, optval, optlen);
2103 #ifdef CONFIG_COMPAT
2104 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2105 char __user *optval, unsigned int optlen)
2107 if (level == SOL_UDP || level == SOL_UDPLITE)
2108 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2109 udp_push_pending_frames);
2110 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2114 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2115 char __user *optval, int __user *optlen)
2117 struct udp_sock *up = udp_sk(sk);
2120 if (get_user(len, optlen))
2123 len = min_t(unsigned int, len, sizeof(int));
2134 val = up->encap_type;
2137 case UDP_NO_CHECK6_TX:
2138 val = up->no_check6_tx;
2141 case UDP_NO_CHECK6_RX:
2142 val = up->no_check6_rx;
2145 /* The following two cannot be changed on UDP sockets, the return is
2146 * always 0 (which corresponds to the full checksum coverage of UDP). */
2147 case UDPLITE_SEND_CSCOV:
2151 case UDPLITE_RECV_CSCOV:
2156 return -ENOPROTOOPT;
2159 if (put_user(len, optlen))
2161 if (copy_to_user(optval, &val, len))
2165 EXPORT_SYMBOL(udp_lib_getsockopt);
2167 int udp_getsockopt(struct sock *sk, int level, int optname,
2168 char __user *optval, int __user *optlen)
2170 if (level == SOL_UDP || level == SOL_UDPLITE)
2171 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2172 return ip_getsockopt(sk, level, optname, optval, optlen);
2175 #ifdef CONFIG_COMPAT
2176 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2177 char __user *optval, int __user *optlen)
2179 if (level == SOL_UDP || level == SOL_UDPLITE)
2180 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2181 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2185 * udp_poll - wait for a UDP event.
2186 * @file - file struct
2188 * @wait - poll table
2190 * This is same as datagram poll, except for the special case of
2191 * blocking sockets. If application is using a blocking fd
2192 * and a packet with checksum error is in the queue;
2193 * then it could get return from select indicating data available
2194 * but then block when reading it. Add special case code
2195 * to work around these arguably broken applications.
2197 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2199 unsigned int mask = datagram_poll(file, sock, wait);
2200 struct sock *sk = sock->sk;
2202 sock_rps_record_flow(sk);
2204 /* Check for false positives due to checksum errors */
2205 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2206 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2207 mask &= ~(POLLIN | POLLRDNORM);
2212 EXPORT_SYMBOL(udp_poll);
2214 struct proto udp_prot = {
2216 .owner = THIS_MODULE,
2217 .close = udp_lib_close,
2218 .connect = ip4_datagram_connect,
2219 .disconnect = udp_disconnect,
2221 .destroy = udp_destroy_sock,
2222 .setsockopt = udp_setsockopt,
2223 .getsockopt = udp_getsockopt,
2224 .sendmsg = udp_sendmsg,
2225 .recvmsg = udp_recvmsg,
2226 .sendpage = udp_sendpage,
2227 .backlog_rcv = __udp_queue_rcv_skb,
2228 .release_cb = ip4_datagram_release_cb,
2229 .hash = udp_lib_hash,
2230 .unhash = udp_lib_unhash,
2231 .rehash = udp_v4_rehash,
2232 .get_port = udp_v4_get_port,
2233 .memory_allocated = &udp_memory_allocated,
2234 .sysctl_mem = sysctl_udp_mem,
2235 .sysctl_wmem = &sysctl_udp_wmem_min,
2236 .sysctl_rmem = &sysctl_udp_rmem_min,
2237 .obj_size = sizeof(struct udp_sock),
2238 .slab_flags = SLAB_DESTROY_BY_RCU,
2239 .h.udp_table = &udp_table,
2240 #ifdef CONFIG_COMPAT
2241 .compat_setsockopt = compat_udp_setsockopt,
2242 .compat_getsockopt = compat_udp_getsockopt,
2244 .clear_sk = sk_prot_clear_portaddr_nulls,
2246 EXPORT_SYMBOL(udp_prot);
2248 /* ------------------------------------------------------------------------ */
2249 #ifdef CONFIG_PROC_FS
2251 static struct sock *udp_get_first(struct seq_file *seq, int start)
2254 struct udp_iter_state *state = seq->private;
2255 struct net *net = seq_file_net(seq);
2257 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2259 struct hlist_nulls_node *node;
2260 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2262 if (hlist_nulls_empty(&hslot->head))
2265 spin_lock_bh(&hslot->lock);
2266 sk_nulls_for_each(sk, node, &hslot->head) {
2267 if (!net_eq(sock_net(sk), net))
2269 if (sk->sk_family == state->family)
2272 spin_unlock_bh(&hslot->lock);
2279 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2281 struct udp_iter_state *state = seq->private;
2282 struct net *net = seq_file_net(seq);
2285 sk = sk_nulls_next(sk);
2286 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2289 if (state->bucket <= state->udp_table->mask)
2290 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2291 return udp_get_first(seq, state->bucket + 1);
2296 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2298 struct sock *sk = udp_get_first(seq, 0);
2301 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2303 return pos ? NULL : sk;
2306 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2308 struct udp_iter_state *state = seq->private;
2309 state->bucket = MAX_UDP_PORTS;
2311 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2314 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2318 if (v == SEQ_START_TOKEN)
2319 sk = udp_get_idx(seq, 0);
2321 sk = udp_get_next(seq, v);
2327 static void udp_seq_stop(struct seq_file *seq, void *v)
2329 struct udp_iter_state *state = seq->private;
2331 if (state->bucket <= state->udp_table->mask)
2332 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2335 int udp_seq_open(struct inode *inode, struct file *file)
2337 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2338 struct udp_iter_state *s;
2341 err = seq_open_net(inode, file, &afinfo->seq_ops,
2342 sizeof(struct udp_iter_state));
2346 s = ((struct seq_file *)file->private_data)->private;
2347 s->family = afinfo->family;
2348 s->udp_table = afinfo->udp_table;
2351 EXPORT_SYMBOL(udp_seq_open);
2353 /* ------------------------------------------------------------------------ */
2354 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2356 struct proc_dir_entry *p;
2359 afinfo->seq_ops.start = udp_seq_start;
2360 afinfo->seq_ops.next = udp_seq_next;
2361 afinfo->seq_ops.stop = udp_seq_stop;
2363 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2364 afinfo->seq_fops, afinfo);
2369 EXPORT_SYMBOL(udp_proc_register);
2371 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2373 remove_proc_entry(afinfo->name, net->proc_net);
2375 EXPORT_SYMBOL(udp_proc_unregister);
2377 /* ------------------------------------------------------------------------ */
2378 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2381 struct inet_sock *inet = inet_sk(sp);
2382 __be32 dest = inet->inet_daddr;
2383 __be32 src = inet->inet_rcv_saddr;
2384 __u16 destp = ntohs(inet->inet_dport);
2385 __u16 srcp = ntohs(inet->inet_sport);
2387 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2388 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2389 bucket, src, srcp, dest, destp, sp->sk_state,
2390 sk_wmem_alloc_get(sp),
2391 sk_rmem_alloc_get(sp),
2393 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2395 atomic_read(&sp->sk_refcnt), sp,
2396 atomic_read(&sp->sk_drops));
2399 int udp4_seq_show(struct seq_file *seq, void *v)
2401 seq_setwidth(seq, 127);
2402 if (v == SEQ_START_TOKEN)
2403 seq_puts(seq, " sl local_address rem_address st tx_queue "
2404 "rx_queue tr tm->when retrnsmt uid timeout "
2405 "inode ref pointer drops");
2407 struct udp_iter_state *state = seq->private;
2409 udp4_format_sock(v, seq, state->bucket);
2415 static const struct file_operations udp_afinfo_seq_fops = {
2416 .owner = THIS_MODULE,
2417 .open = udp_seq_open,
2419 .llseek = seq_lseek,
2420 .release = seq_release_net
2423 /* ------------------------------------------------------------------------ */
2424 static struct udp_seq_afinfo udp4_seq_afinfo = {
2427 .udp_table = &udp_table,
2428 .seq_fops = &udp_afinfo_seq_fops,
2430 .show = udp4_seq_show,
2434 static int __net_init udp4_proc_init_net(struct net *net)
2436 return udp_proc_register(net, &udp4_seq_afinfo);
2439 static void __net_exit udp4_proc_exit_net(struct net *net)
2441 udp_proc_unregister(net, &udp4_seq_afinfo);
2444 static struct pernet_operations udp4_net_ops = {
2445 .init = udp4_proc_init_net,
2446 .exit = udp4_proc_exit_net,
2449 int __init udp4_proc_init(void)
2451 return register_pernet_subsys(&udp4_net_ops);
2454 void udp4_proc_exit(void)
2456 unregister_pernet_subsys(&udp4_net_ops);
2458 #endif /* CONFIG_PROC_FS */
2460 static __initdata unsigned long uhash_entries;
2461 static int __init set_uhash_entries(char *str)
2468 ret = kstrtoul(str, 0, &uhash_entries);
2472 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2473 uhash_entries = UDP_HTABLE_SIZE_MIN;
2476 __setup("uhash_entries=", set_uhash_entries);
2478 void __init udp_table_init(struct udp_table *table, const char *name)
2482 table->hash = alloc_large_system_hash(name,
2483 2 * sizeof(struct udp_hslot),
2485 21, /* one slot per 2 MB */
2489 UDP_HTABLE_SIZE_MIN,
2492 table->hash2 = table->hash + (table->mask + 1);
2493 for (i = 0; i <= table->mask; i++) {
2494 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2495 table->hash[i].count = 0;
2496 spin_lock_init(&table->hash[i].lock);
2498 for (i = 0; i <= table->mask; i++) {
2499 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2500 table->hash2[i].count = 0;
2501 spin_lock_init(&table->hash2[i].lock);
2505 void __init udp_init(void)
2507 unsigned long limit;
2509 udp_table_init(&udp_table, "UDP");
2510 limit = nr_free_buffer_pages() / 8;
2511 limit = max(limit, 128UL);
2512 sysctl_udp_mem[0] = limit / 4 * 3;
2513 sysctl_udp_mem[1] = limit;
2514 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2516 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2517 sysctl_udp_wmem_min = SK_MEM_QUANTUM;