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).
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.
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;
143 sk_nulls_for_each(sk2, node, &hslot->head)
144 if (net_eq(sock_net(sk2), net) &&
146 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
147 (!sk2->sk_reuse || !sk->sk_reuse) &&
148 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
149 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
150 (*saddr_comp)(sk, sk2)) {
152 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
161 * Note: we still hold spinlock of primary hash chain, so no other writer
162 * can insert/delete a socket with local_port == num
164 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
165 struct udp_hslot *hslot2,
167 int (*saddr_comp)(const struct sock *sk1,
168 const struct sock *sk2))
171 struct hlist_nulls_node *node;
174 spin_lock(&hslot2->lock);
175 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
176 if (net_eq(sock_net(sk2), net) &&
178 (udp_sk(sk2)->udp_port_hash == num) &&
179 (!sk2->sk_reuse || !sk->sk_reuse) &&
180 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
181 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
182 (*saddr_comp)(sk, sk2)) {
186 spin_unlock(&hslot2->lock);
191 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
193 * @sk: socket struct in question
194 * @snum: port number to look up
195 * @saddr_comp: AF-dependent comparison of bound local IP addresses
196 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
199 int udp_lib_get_port(struct sock *sk, unsigned short snum,
200 int (*saddr_comp)(const struct sock *sk1,
201 const struct sock *sk2),
202 unsigned int hash2_nulladdr)
204 struct udp_hslot *hslot, *hslot2;
205 struct udp_table *udptable = sk->sk_prot->h.udp_table;
207 struct net *net = sock_net(sk);
210 int low, high, remaining;
212 unsigned short first, last;
213 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
215 inet_get_local_port_range(&low, &high);
216 remaining = (high - low) + 1;
219 first = (((u64)rand * remaining) >> 32) + low;
221 * force rand to be an odd multiple of UDP_HTABLE_SIZE
223 rand = (rand | 1) * (udptable->mask + 1);
224 last = first + udptable->mask + 1;
226 hslot = udp_hashslot(udptable, net, first);
227 bitmap_zero(bitmap, PORTS_PER_CHAIN);
228 spin_lock_bh(&hslot->lock);
229 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
230 saddr_comp, udptable->log);
234 * Iterate on all possible values of snum for this hash.
235 * Using steps of an odd multiple of UDP_HTABLE_SIZE
236 * give us randomization and full range coverage.
239 if (low <= snum && snum <= high &&
240 !test_bit(snum >> udptable->log, bitmap) &&
241 !inet_is_reserved_local_port(snum))
244 } while (snum != first);
245 spin_unlock_bh(&hslot->lock);
246 } while (++first != last);
249 hslot = udp_hashslot(udptable, net, snum);
250 spin_lock_bh(&hslot->lock);
251 if (hslot->count > 10) {
253 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
255 slot2 &= udptable->mask;
256 hash2_nulladdr &= udptable->mask;
258 hslot2 = udp_hashslot2(udptable, slot2);
259 if (hslot->count < hslot2->count)
260 goto scan_primary_hash;
262 exist = udp_lib_lport_inuse2(net, snum, hslot2,
264 if (!exist && (hash2_nulladdr != slot2)) {
265 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
266 exist = udp_lib_lport_inuse2(net, snum, hslot2,
275 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
280 inet_sk(sk)->inet_num = snum;
281 udp_sk(sk)->udp_port_hash = snum;
282 udp_sk(sk)->udp_portaddr_hash ^= snum;
283 if (sk_unhashed(sk)) {
284 sk_nulls_add_node_rcu(sk, &hslot->head);
286 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
288 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
289 spin_lock(&hslot2->lock);
290 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
293 spin_unlock(&hslot2->lock);
297 spin_unlock_bh(&hslot->lock);
301 EXPORT_SYMBOL(udp_lib_get_port);
303 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
305 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
307 return (!ipv6_only_sock(sk2) &&
308 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
309 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
312 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
315 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
318 int udp_v4_get_port(struct sock *sk, unsigned short snum)
320 unsigned int hash2_nulladdr =
321 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
322 unsigned int hash2_partial =
323 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
325 /* precompute partial secondary hash */
326 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
327 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
330 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
332 __be16 sport, __be32 daddr, __be16 dport, int dif)
336 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
337 !ipv6_only_sock(sk)) {
338 struct inet_sock *inet = inet_sk(sk);
340 score = (sk->sk_family == PF_INET ? 1 : 0);
341 if (inet->inet_rcv_saddr) {
342 if (inet->inet_rcv_saddr != daddr)
346 if (inet->inet_daddr) {
347 if (inet->inet_daddr != saddr)
351 if (inet->inet_dport) {
352 if (inet->inet_dport != sport)
356 if (sk->sk_bound_dev_if) {
357 if (sk->sk_bound_dev_if != dif)
366 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
368 #define SCORE2_MAX (1 + 2 + 2 + 2)
369 static inline int compute_score2(struct sock *sk, struct net *net,
370 __be32 saddr, __be16 sport,
371 __be32 daddr, unsigned int hnum, int dif)
375 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
376 struct inet_sock *inet = inet_sk(sk);
378 if (inet->inet_rcv_saddr != daddr)
380 if (inet->inet_num != hnum)
383 score = (sk->sk_family == PF_INET ? 1 : 0);
384 if (inet->inet_daddr) {
385 if (inet->inet_daddr != saddr)
389 if (inet->inet_dport) {
390 if (inet->inet_dport != sport)
394 if (sk->sk_bound_dev_if) {
395 if (sk->sk_bound_dev_if != dif)
404 /* called with read_rcu_lock() */
405 static struct sock *udp4_lib_lookup2(struct net *net,
406 __be32 saddr, __be16 sport,
407 __be32 daddr, unsigned int hnum, int dif,
408 struct udp_hslot *hslot2, unsigned int slot2)
410 struct sock *sk, *result;
411 struct hlist_nulls_node *node;
417 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
418 score = compute_score2(sk, net, saddr, sport,
420 if (score > badness) {
423 if (score == SCORE2_MAX)
428 * if the nulls value we got at the end of this lookup is
429 * not the expected one, we must restart lookup.
430 * We probably met an item that was moved to another chain.
432 if (get_nulls_value(node) != slot2)
437 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
439 else if (unlikely(compute_score2(result, net, saddr, sport,
440 daddr, hnum, dif) < badness)) {
448 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
449 * harder than this. -DaveM
451 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
452 __be16 sport, __be32 daddr, __be16 dport,
453 int dif, struct udp_table *udptable)
455 struct sock *sk, *result;
456 struct hlist_nulls_node *node;
457 unsigned short hnum = ntohs(dport);
458 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
459 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
463 if (hslot->count > 10) {
464 hash2 = udp4_portaddr_hash(net, daddr, hnum);
465 slot2 = hash2 & udptable->mask;
466 hslot2 = &udptable->hash2[slot2];
467 if (hslot->count < hslot2->count)
470 result = udp4_lib_lookup2(net, saddr, sport,
474 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
475 slot2 = hash2 & udptable->mask;
476 hslot2 = &udptable->hash2[slot2];
477 if (hslot->count < hslot2->count)
480 result = udp4_lib_lookup2(net, saddr, sport,
481 htonl(INADDR_ANY), hnum, dif,
490 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
491 score = compute_score(sk, net, saddr, hnum, sport,
493 if (score > badness) {
499 * if the nulls value we got at the end of this lookup is
500 * not the expected one, we must restart lookup.
501 * We probably met an item that was moved to another chain.
503 if (get_nulls_value(node) != slot)
507 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
509 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
510 daddr, dport, dif) < badness)) {
518 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
520 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
521 __be16 sport, __be16 dport,
522 struct udp_table *udptable)
525 const struct iphdr *iph = ip_hdr(skb);
527 if (unlikely(sk = skb_steal_sock(skb)))
530 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
531 iph->daddr, dport, inet_iif(skb),
535 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
536 __be32 daddr, __be16 dport, int dif)
538 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
540 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
542 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
543 __be16 loc_port, __be32 loc_addr,
544 __be16 rmt_port, __be32 rmt_addr,
547 struct hlist_nulls_node *node;
549 unsigned short hnum = ntohs(loc_port);
551 sk_nulls_for_each_from(s, node) {
552 struct inet_sock *inet = inet_sk(s);
554 if (!net_eq(sock_net(s), net) ||
555 udp_sk(s)->udp_port_hash != hnum ||
556 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
557 (inet->inet_dport != rmt_port && inet->inet_dport) ||
558 (inet->inet_rcv_saddr &&
559 inet->inet_rcv_saddr != loc_addr) ||
561 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
563 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
573 * This routine is called by the ICMP module when it gets some
574 * sort of error condition. If err < 0 then the socket should
575 * be closed and the error returned to the user. If err > 0
576 * it's just the icmp type << 8 | icmp code.
577 * Header points to the ip header of the error packet. We move
578 * on past this. Then (as it used to claim before adjustment)
579 * header points to the first 8 bytes of the udp header. We need
580 * to find the appropriate port.
583 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
585 struct inet_sock *inet;
586 const struct iphdr *iph = (const struct iphdr *)skb->data;
587 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
588 const int type = icmp_hdr(skb)->type;
589 const int code = icmp_hdr(skb)->code;
593 struct net *net = dev_net(skb->dev);
595 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
596 iph->saddr, uh->source, skb->dev->ifindex, udptable);
598 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
599 return; /* No socket for error */
608 case ICMP_TIME_EXCEEDED:
611 case ICMP_SOURCE_QUENCH:
613 case ICMP_PARAMETERPROB:
617 case ICMP_DEST_UNREACH:
618 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
619 ipv4_sk_update_pmtu(skb, sk, info);
620 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
628 if (code <= NR_ICMP_UNREACH) {
629 harderr = icmp_err_convert[code].fatal;
630 err = icmp_err_convert[code].errno;
634 ipv4_sk_redirect(skb, sk);
639 * RFC1122: OK. Passes ICMP errors back to application, as per
642 if (!inet->recverr) {
643 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
646 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
649 sk->sk_error_report(sk);
654 void udp_err(struct sk_buff *skb, u32 info)
656 __udp4_lib_err(skb, info, &udp_table);
660 * Throw away all pending data and cancel the corking. Socket is locked.
662 void udp_flush_pending_frames(struct sock *sk)
664 struct udp_sock *up = udp_sk(sk);
669 ip_flush_pending_frames(sk);
672 EXPORT_SYMBOL(udp_flush_pending_frames);
675 * udp4_hwcsum - handle outgoing HW checksumming
676 * @skb: sk_buff containing the filled-in UDP header
677 * (checksum field must be zeroed out)
678 * @src: source IP address
679 * @dst: destination IP address
681 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
683 struct udphdr *uh = udp_hdr(skb);
684 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
685 int offset = skb_transport_offset(skb);
686 int len = skb->len - offset;
692 * Only one fragment on the socket.
694 skb->csum_start = skb_transport_header(skb) - skb->head;
695 skb->csum_offset = offsetof(struct udphdr, check);
696 uh->check = ~csum_tcpudp_magic(src, dst, len,
700 * HW-checksum won't work as there are two or more
701 * fragments on the socket so that all csums of sk_buffs
705 csum = csum_add(csum, frags->csum);
707 } while ((frags = frags->next));
709 csum = skb_checksum(skb, offset, hlen, csum);
710 skb->ip_summed = CHECKSUM_NONE;
712 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
714 uh->check = CSUM_MANGLED_0;
718 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
720 struct sock *sk = skb->sk;
721 struct inet_sock *inet = inet_sk(sk);
724 int is_udplite = IS_UDPLITE(sk);
725 int offset = skb_transport_offset(skb);
726 int len = skb->len - offset;
730 * Create a UDP header
733 uh->source = inet->inet_sport;
734 uh->dest = fl4->fl4_dport;
735 uh->len = htons(len);
738 if (is_udplite) /* UDP-Lite */
739 csum = udplite_csum(skb);
741 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
743 skb->ip_summed = CHECKSUM_NONE;
746 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
748 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
752 csum = udp_csum(skb);
754 /* add protocol-dependent pseudo-header */
755 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
756 sk->sk_protocol, csum);
758 uh->check = CSUM_MANGLED_0;
761 err = ip_send_skb(sock_net(sk), skb);
763 if (err == -ENOBUFS && !inet->recverr) {
764 UDP_INC_STATS_USER(sock_net(sk),
765 UDP_MIB_SNDBUFERRORS, is_udplite);
769 UDP_INC_STATS_USER(sock_net(sk),
770 UDP_MIB_OUTDATAGRAMS, is_udplite);
775 * Push out all pending data as one UDP datagram. Socket is locked.
777 static int udp_push_pending_frames(struct sock *sk)
779 struct udp_sock *up = udp_sk(sk);
780 struct inet_sock *inet = inet_sk(sk);
781 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
785 skb = ip_finish_skb(sk, fl4);
789 err = udp_send_skb(skb, fl4);
797 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
800 struct inet_sock *inet = inet_sk(sk);
801 struct udp_sock *up = udp_sk(sk);
802 struct flowi4 fl4_stack;
805 struct ipcm_cookie ipc;
806 struct rtable *rt = NULL;
809 __be32 daddr, faddr, saddr;
812 int err, is_udplite = IS_UDPLITE(sk);
813 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
814 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
816 struct ip_options_data opt_copy;
825 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
831 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
833 fl4 = &inet->cork.fl.u.ip4;
836 * There are pending frames.
837 * The socket lock must be held while it's corked.
840 if (likely(up->pending)) {
841 if (unlikely(up->pending != AF_INET)) {
849 ulen += sizeof(struct udphdr);
852 * Get and verify the address.
855 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
856 if (msg->msg_namelen < sizeof(*usin))
858 if (usin->sin_family != AF_INET) {
859 if (usin->sin_family != AF_UNSPEC)
860 return -EAFNOSUPPORT;
863 daddr = usin->sin_addr.s_addr;
864 dport = usin->sin_port;
868 if (sk->sk_state != TCP_ESTABLISHED)
869 return -EDESTADDRREQ;
870 daddr = inet->inet_daddr;
871 dport = inet->inet_dport;
872 /* Open fast path for connected socket.
873 Route will not be used, if at least one option is set.
877 ipc.addr = inet->inet_saddr;
879 ipc.oif = sk->sk_bound_dev_if;
880 err = sock_tx_timestamp(sk, &ipc.tx_flags);
883 if (msg->msg_controllen) {
884 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
892 struct ip_options_rcu *inet_opt;
895 inet_opt = rcu_dereference(inet->inet_opt);
897 memcpy(&opt_copy, inet_opt,
898 sizeof(*inet_opt) + inet_opt->opt.optlen);
899 ipc.opt = &opt_copy.opt;
905 ipc.addr = faddr = daddr;
907 if (ipc.opt && ipc.opt->opt.srr) {
910 faddr = ipc.opt->opt.faddr;
913 tos = RT_TOS(inet->tos);
914 if (sock_flag(sk, SOCK_LOCALROUTE) ||
915 (msg->msg_flags & MSG_DONTROUTE) ||
916 (ipc.opt && ipc.opt->opt.is_strictroute)) {
921 if (ipv4_is_multicast(daddr)) {
923 ipc.oif = inet->mc_index;
925 saddr = inet->mc_addr;
928 ipc.oif = inet->uc_index;
931 rt = (struct rtable *)sk_dst_check(sk, 0);
934 struct net *net = sock_net(sk);
937 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
938 RT_SCOPE_UNIVERSE, sk->sk_protocol,
939 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
940 faddr, saddr, dport, inet->inet_sport);
942 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
943 rt = ip_route_output_flow(net, fl4, sk);
947 if (err == -ENETUNREACH)
948 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
953 if ((rt->rt_flags & RTCF_BROADCAST) &&
954 !sock_flag(sk, SOCK_BROADCAST))
957 sk_dst_set(sk, dst_clone(&rt->dst));
960 if (msg->msg_flags&MSG_CONFIRM)
966 daddr = ipc.addr = fl4->daddr;
968 /* Lockless fast path for the non-corking case. */
970 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
971 sizeof(struct udphdr), &ipc, &rt,
974 if (skb && !IS_ERR(skb))
975 err = udp_send_skb(skb, fl4);
980 if (unlikely(up->pending)) {
981 /* The socket is already corked while preparing it. */
982 /* ... which is an evident application bug. --ANK */
985 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
990 * Now cork the socket to pend data.
992 fl4 = &inet->cork.fl.u.ip4;
995 fl4->fl4_dport = dport;
996 fl4->fl4_sport = inet->inet_sport;
997 up->pending = AF_INET;
1001 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1002 sizeof(struct udphdr), &ipc, &rt,
1003 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1005 udp_flush_pending_frames(sk);
1007 err = udp_push_pending_frames(sk);
1008 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1019 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1020 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1021 * we don't have a good statistic (IpOutDiscards but it can be too many
1022 * things). We could add another new stat but at least for now that
1023 * seems like overkill.
1025 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1026 UDP_INC_STATS_USER(sock_net(sk),
1027 UDP_MIB_SNDBUFERRORS, is_udplite);
1032 dst_confirm(&rt->dst);
1033 if (!(msg->msg_flags&MSG_PROBE) || len)
1034 goto back_from_confirm;
1038 EXPORT_SYMBOL(udp_sendmsg);
1040 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1041 size_t size, int flags)
1043 struct inet_sock *inet = inet_sk(sk);
1044 struct udp_sock *up = udp_sk(sk);
1048 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1050 /* Call udp_sendmsg to specify destination address which
1051 * sendpage interface can't pass.
1052 * This will succeed only when the socket is connected.
1054 ret = udp_sendmsg(NULL, sk, &msg, 0);
1061 if (unlikely(!up->pending)) {
1064 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1068 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1069 page, offset, size, flags);
1070 if (ret == -EOPNOTSUPP) {
1072 return sock_no_sendpage(sk->sk_socket, page, offset,
1076 udp_flush_pending_frames(sk);
1081 if (!(up->corkflag || (flags&MSG_MORE)))
1082 ret = udp_push_pending_frames(sk);
1092 * first_packet_length - return length of first packet in receive queue
1095 * Drops all bad checksum frames, until a valid one is found.
1096 * Returns the length of found skb, or 0 if none is found.
1098 static unsigned int first_packet_length(struct sock *sk)
1100 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1101 struct sk_buff *skb;
1104 __skb_queue_head_init(&list_kill);
1106 spin_lock_bh(&rcvq->lock);
1107 while ((skb = skb_peek(rcvq)) != NULL &&
1108 udp_lib_checksum_complete(skb)) {
1109 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1111 atomic_inc(&sk->sk_drops);
1112 __skb_unlink(skb, rcvq);
1113 __skb_queue_tail(&list_kill, skb);
1115 res = skb ? skb->len : 0;
1116 spin_unlock_bh(&rcvq->lock);
1118 if (!skb_queue_empty(&list_kill)) {
1119 bool slow = lock_sock_fast(sk);
1121 __skb_queue_purge(&list_kill);
1122 sk_mem_reclaim_partial(sk);
1123 unlock_sock_fast(sk, slow);
1129 * IOCTL requests applicable to the UDP protocol
1132 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1137 int amount = sk_wmem_alloc_get(sk);
1139 return put_user(amount, (int __user *)arg);
1144 unsigned int amount = first_packet_length(sk);
1148 * We will only return the amount
1149 * of this packet since that is all
1150 * that will be read.
1152 amount -= sizeof(struct udphdr);
1154 return put_user(amount, (int __user *)arg);
1158 return -ENOIOCTLCMD;
1163 EXPORT_SYMBOL(udp_ioctl);
1166 * This should be easy, if there is something there we
1167 * return it, otherwise we block.
1170 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1171 size_t len, int noblock, int flags, int *addr_len)
1173 struct inet_sock *inet = inet_sk(sk);
1174 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1175 struct sk_buff *skb;
1176 unsigned int ulen, copied;
1177 int peeked, off = 0;
1179 int is_udplite = IS_UDPLITE(sk);
1183 * Check any passed addresses
1186 *addr_len = sizeof(*sin);
1188 if (flags & MSG_ERRQUEUE)
1189 return ip_recv_error(sk, msg, len);
1192 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1193 &peeked, &off, &err);
1197 ulen = skb->len - sizeof(struct udphdr);
1201 else if (copied < ulen)
1202 msg->msg_flags |= MSG_TRUNC;
1205 * If checksum is needed at all, try to do it while copying the
1206 * data. If the data is truncated, or if we only want a partial
1207 * coverage checksum (UDP-Lite), do it before the copy.
1210 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1211 if (udp_lib_checksum_complete(skb))
1215 if (skb_csum_unnecessary(skb))
1216 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1217 msg->msg_iov, copied);
1219 err = skb_copy_and_csum_datagram_iovec(skb,
1220 sizeof(struct udphdr),
1227 if (unlikely(err)) {
1228 trace_kfree_skb(skb, udp_recvmsg);
1230 atomic_inc(&sk->sk_drops);
1231 UDP_INC_STATS_USER(sock_net(sk),
1232 UDP_MIB_INERRORS, is_udplite);
1238 UDP_INC_STATS_USER(sock_net(sk),
1239 UDP_MIB_INDATAGRAMS, is_udplite);
1241 sock_recv_ts_and_drops(msg, sk, skb);
1243 /* Copy the address. */
1245 sin->sin_family = AF_INET;
1246 sin->sin_port = udp_hdr(skb)->source;
1247 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1248 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1250 if (inet->cmsg_flags)
1251 ip_cmsg_recv(msg, skb);
1254 if (flags & MSG_TRUNC)
1258 skb_free_datagram_locked(sk, skb);
1263 slow = lock_sock_fast(sk);
1264 if (!skb_kill_datagram(sk, skb, flags))
1265 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1266 unlock_sock_fast(sk, slow);
1271 /* starting over for a new packet */
1272 msg->msg_flags &= ~MSG_TRUNC;
1277 int udp_disconnect(struct sock *sk, int flags)
1279 struct inet_sock *inet = inet_sk(sk);
1281 * 1003.1g - break association.
1284 sk->sk_state = TCP_CLOSE;
1285 inet->inet_daddr = 0;
1286 inet->inet_dport = 0;
1287 sock_rps_reset_rxhash(sk);
1288 sk->sk_bound_dev_if = 0;
1289 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1290 inet_reset_saddr(sk);
1292 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1293 sk->sk_prot->unhash(sk);
1294 inet->inet_sport = 0;
1299 EXPORT_SYMBOL(udp_disconnect);
1301 void udp_lib_unhash(struct sock *sk)
1303 if (sk_hashed(sk)) {
1304 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1305 struct udp_hslot *hslot, *hslot2;
1307 hslot = udp_hashslot(udptable, sock_net(sk),
1308 udp_sk(sk)->udp_port_hash);
1309 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1311 spin_lock_bh(&hslot->lock);
1312 if (sk_nulls_del_node_init_rcu(sk)) {
1314 inet_sk(sk)->inet_num = 0;
1315 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1317 spin_lock(&hslot2->lock);
1318 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1320 spin_unlock(&hslot2->lock);
1322 spin_unlock_bh(&hslot->lock);
1325 EXPORT_SYMBOL(udp_lib_unhash);
1328 * inet_rcv_saddr was changed, we must rehash secondary hash
1330 void udp_lib_rehash(struct sock *sk, u16 newhash)
1332 if (sk_hashed(sk)) {
1333 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1334 struct udp_hslot *hslot, *hslot2, *nhslot2;
1336 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1337 nhslot2 = udp_hashslot2(udptable, newhash);
1338 udp_sk(sk)->udp_portaddr_hash = newhash;
1339 if (hslot2 != nhslot2) {
1340 hslot = udp_hashslot(udptable, sock_net(sk),
1341 udp_sk(sk)->udp_port_hash);
1342 /* we must lock primary chain too */
1343 spin_lock_bh(&hslot->lock);
1345 spin_lock(&hslot2->lock);
1346 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1348 spin_unlock(&hslot2->lock);
1350 spin_lock(&nhslot2->lock);
1351 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1354 spin_unlock(&nhslot2->lock);
1356 spin_unlock_bh(&hslot->lock);
1360 EXPORT_SYMBOL(udp_lib_rehash);
1362 static void udp_v4_rehash(struct sock *sk)
1364 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1365 inet_sk(sk)->inet_rcv_saddr,
1366 inet_sk(sk)->inet_num);
1367 udp_lib_rehash(sk, new_hash);
1370 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1374 if (inet_sk(sk)->inet_daddr)
1375 sock_rps_save_rxhash(sk, skb);
1377 rc = sock_queue_rcv_skb(sk, skb);
1379 int is_udplite = IS_UDPLITE(sk);
1381 /* Note that an ENOMEM error is charged twice */
1383 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1385 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1387 trace_udp_fail_queue_rcv_skb(rc, sk);
1395 static struct static_key udp_encap_needed __read_mostly;
1396 void udp_encap_enable(void)
1398 if (!static_key_enabled(&udp_encap_needed))
1399 static_key_slow_inc(&udp_encap_needed);
1401 EXPORT_SYMBOL(udp_encap_enable);
1406 * >0: "udp encap" protocol resubmission
1408 * Note that in the success and error cases, the skb is assumed to
1409 * have either been requeued or freed.
1411 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1413 struct udp_sock *up = udp_sk(sk);
1415 int is_udplite = IS_UDPLITE(sk);
1418 * Charge it to the socket, dropping if the queue is full.
1420 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1424 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1425 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1428 * This is an encapsulation socket so pass the skb to
1429 * the socket's udp_encap_rcv() hook. Otherwise, just
1430 * fall through and pass this up the UDP socket.
1431 * up->encap_rcv() returns the following value:
1432 * =0 if skb was successfully passed to the encap
1433 * handler or was discarded by it.
1434 * >0 if skb should be passed on to UDP.
1435 * <0 if skb should be resubmitted as proto -N
1438 /* if we're overly short, let UDP handle it */
1439 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1440 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1443 ret = encap_rcv(sk, skb);
1445 UDP_INC_STATS_BH(sock_net(sk),
1446 UDP_MIB_INDATAGRAMS,
1452 /* FALLTHROUGH -- it's a UDP Packet */
1456 * UDP-Lite specific tests, ignored on UDP sockets
1458 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1461 * MIB statistics other than incrementing the error count are
1462 * disabled for the following two types of errors: these depend
1463 * on the application settings, not on the functioning of the
1464 * protocol stack as such.
1466 * RFC 3828 here recommends (sec 3.3): "There should also be a
1467 * way ... to ... at least let the receiving application block
1468 * delivery of packets with coverage values less than a value
1469 * provided by the application."
1471 if (up->pcrlen == 0) { /* full coverage was set */
1472 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1473 UDP_SKB_CB(skb)->cscov, skb->len);
1476 /* The next case involves violating the min. coverage requested
1477 * by the receiver. This is subtle: if receiver wants x and x is
1478 * greater than the buffersize/MTU then receiver will complain
1479 * that it wants x while sender emits packets of smaller size y.
1480 * Therefore the above ...()->partial_cov statement is essential.
1482 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1483 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1484 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1489 if (rcu_access_pointer(sk->sk_filter) &&
1490 udp_lib_checksum_complete(skb))
1494 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1499 ipv4_pktinfo_prepare(skb);
1501 if (!sock_owned_by_user(sk))
1502 rc = __udp_queue_rcv_skb(sk, skb);
1503 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1512 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1513 atomic_inc(&sk->sk_drops);
1519 static void flush_stack(struct sock **stack, unsigned int count,
1520 struct sk_buff *skb, unsigned int final)
1523 struct sk_buff *skb1 = NULL;
1526 for (i = 0; i < count; i++) {
1528 if (likely(skb1 == NULL))
1529 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1532 atomic_inc(&sk->sk_drops);
1533 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1535 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1539 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1547 * Multicasts and broadcasts go to each listener.
1549 * Note: called only from the BH handler context.
1551 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1553 __be32 saddr, __be32 daddr,
1554 struct udp_table *udptable)
1556 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1557 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1559 unsigned int i, count = 0;
1561 spin_lock(&hslot->lock);
1562 sk = sk_nulls_head(&hslot->head);
1563 dif = skb->dev->ifindex;
1564 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1566 stack[count++] = sk;
1567 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1568 daddr, uh->source, saddr, dif);
1569 if (unlikely(count == ARRAY_SIZE(stack))) {
1572 flush_stack(stack, count, skb, ~0);
1577 * before releasing chain lock, we must take a reference on sockets
1579 for (i = 0; i < count; i++)
1580 sock_hold(stack[i]);
1582 spin_unlock(&hslot->lock);
1585 * do the slow work with no lock held
1588 flush_stack(stack, count, skb, count - 1);
1590 for (i = 0; i < count; i++)
1598 /* Initialize UDP checksum. If exited with zero value (success),
1599 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1600 * Otherwise, csum completion requires chacksumming packet body,
1601 * including udp header and folding it to skb->csum.
1603 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1606 const struct iphdr *iph;
1609 UDP_SKB_CB(skb)->partial_cov = 0;
1610 UDP_SKB_CB(skb)->cscov = skb->len;
1612 if (proto == IPPROTO_UDPLITE) {
1613 err = udplite_checksum_init(skb, uh);
1619 if (uh->check == 0) {
1620 skb->ip_summed = CHECKSUM_UNNECESSARY;
1621 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1622 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1624 skb->ip_summed = CHECKSUM_UNNECESSARY;
1626 if (!skb_csum_unnecessary(skb))
1627 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1628 skb->len, proto, 0);
1629 /* Probably, we should checksum udp header (it should be in cache
1630 * in any case) and data in tiny packets (< rx copybreak).
1637 * All we need to do is get the socket, and then do a checksum.
1640 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1645 unsigned short ulen;
1646 struct rtable *rt = skb_rtable(skb);
1647 __be32 saddr, daddr;
1648 struct net *net = dev_net(skb->dev);
1651 * Validate the packet.
1653 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1654 goto drop; /* No space for header. */
1657 ulen = ntohs(uh->len);
1658 saddr = ip_hdr(skb)->saddr;
1659 daddr = ip_hdr(skb)->daddr;
1661 if (ulen > skb->len)
1664 if (proto == IPPROTO_UDP) {
1665 /* UDP validates ulen. */
1666 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1671 if (udp4_csum_init(skb, uh, proto))
1674 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1675 return __udp4_lib_mcast_deliver(net, skb, uh,
1676 saddr, daddr, udptable);
1678 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1681 int ret = udp_queue_rcv_skb(sk, skb);
1684 /* a return value > 0 means to resubmit the input, but
1685 * it wants the return to be -protocol, or 0
1692 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1696 /* No socket. Drop packet silently, if checksum is wrong */
1697 if (udp_lib_checksum_complete(skb))
1700 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1701 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1704 * Hmm. We got an UDP packet to a port to which we
1705 * don't wanna listen. Ignore it.
1711 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1712 proto == IPPROTO_UDPLITE ? "Lite" : "",
1713 &saddr, ntohs(uh->source),
1715 &daddr, ntohs(uh->dest));
1720 * RFC1122: OK. Discards the bad packet silently (as far as
1721 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1723 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1724 proto == IPPROTO_UDPLITE ? "Lite" : "",
1725 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1728 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1733 int udp_rcv(struct sk_buff *skb)
1735 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1738 void udp_destroy_sock(struct sock *sk)
1740 bool slow = lock_sock_fast(sk);
1741 udp_flush_pending_frames(sk);
1742 unlock_sock_fast(sk, slow);
1746 * Socket option code for UDP
1748 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1749 char __user *optval, unsigned int optlen,
1750 int (*push_pending_frames)(struct sock *))
1752 struct udp_sock *up = udp_sk(sk);
1755 int is_udplite = IS_UDPLITE(sk);
1757 if (optlen < sizeof(int))
1760 if (get_user(val, (int __user *)optval))
1770 (*push_pending_frames)(sk);
1778 case UDP_ENCAP_ESPINUDP:
1779 case UDP_ENCAP_ESPINUDP_NON_IKE:
1780 up->encap_rcv = xfrm4_udp_encap_rcv;
1782 case UDP_ENCAP_L2TPINUDP:
1783 up->encap_type = val;
1793 * UDP-Lite's partial checksum coverage (RFC 3828).
1795 /* The sender sets actual checksum coverage length via this option.
1796 * The case coverage > packet length is handled by send module. */
1797 case UDPLITE_SEND_CSCOV:
1798 if (!is_udplite) /* Disable the option on UDP sockets */
1799 return -ENOPROTOOPT;
1800 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1802 else if (val > USHRT_MAX)
1805 up->pcflag |= UDPLITE_SEND_CC;
1808 /* The receiver specifies a minimum checksum coverage value. To make
1809 * sense, this should be set to at least 8 (as done below). If zero is
1810 * used, this again means full checksum coverage. */
1811 case UDPLITE_RECV_CSCOV:
1812 if (!is_udplite) /* Disable the option on UDP sockets */
1813 return -ENOPROTOOPT;
1814 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1816 else if (val > USHRT_MAX)
1819 up->pcflag |= UDPLITE_RECV_CC;
1829 EXPORT_SYMBOL(udp_lib_setsockopt);
1831 int udp_setsockopt(struct sock *sk, int level, int optname,
1832 char __user *optval, unsigned int optlen)
1834 if (level == SOL_UDP || level == SOL_UDPLITE)
1835 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1836 udp_push_pending_frames);
1837 return ip_setsockopt(sk, level, optname, optval, optlen);
1840 #ifdef CONFIG_COMPAT
1841 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1842 char __user *optval, unsigned int optlen)
1844 if (level == SOL_UDP || level == SOL_UDPLITE)
1845 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1846 udp_push_pending_frames);
1847 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1851 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1852 char __user *optval, int __user *optlen)
1854 struct udp_sock *up = udp_sk(sk);
1857 if (get_user(len, optlen))
1860 len = min_t(unsigned int, len, sizeof(int));
1871 val = up->encap_type;
1874 /* The following two cannot be changed on UDP sockets, the return is
1875 * always 0 (which corresponds to the full checksum coverage of UDP). */
1876 case UDPLITE_SEND_CSCOV:
1880 case UDPLITE_RECV_CSCOV:
1885 return -ENOPROTOOPT;
1888 if (put_user(len, optlen))
1890 if (copy_to_user(optval, &val, len))
1894 EXPORT_SYMBOL(udp_lib_getsockopt);
1896 int udp_getsockopt(struct sock *sk, int level, int optname,
1897 char __user *optval, int __user *optlen)
1899 if (level == SOL_UDP || level == SOL_UDPLITE)
1900 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1901 return ip_getsockopt(sk, level, optname, optval, optlen);
1904 #ifdef CONFIG_COMPAT
1905 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1906 char __user *optval, int __user *optlen)
1908 if (level == SOL_UDP || level == SOL_UDPLITE)
1909 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1910 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1914 * udp_poll - wait for a UDP event.
1915 * @file - file struct
1917 * @wait - poll table
1919 * This is same as datagram poll, except for the special case of
1920 * blocking sockets. If application is using a blocking fd
1921 * and a packet with checksum error is in the queue;
1922 * then it could get return from select indicating data available
1923 * but then block when reading it. Add special case code
1924 * to work around these arguably broken applications.
1926 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1928 unsigned int mask = datagram_poll(file, sock, wait);
1929 struct sock *sk = sock->sk;
1931 /* Check for false positives due to checksum errors */
1932 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1933 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1934 mask &= ~(POLLIN | POLLRDNORM);
1939 EXPORT_SYMBOL(udp_poll);
1941 struct proto udp_prot = {
1943 .owner = THIS_MODULE,
1944 .close = udp_lib_close,
1945 .connect = ip4_datagram_connect,
1946 .disconnect = udp_disconnect,
1948 .destroy = udp_destroy_sock,
1949 .setsockopt = udp_setsockopt,
1950 .getsockopt = udp_getsockopt,
1951 .sendmsg = udp_sendmsg,
1952 .recvmsg = udp_recvmsg,
1953 .sendpage = udp_sendpage,
1954 .backlog_rcv = __udp_queue_rcv_skb,
1955 .hash = udp_lib_hash,
1956 .unhash = udp_lib_unhash,
1957 .rehash = udp_v4_rehash,
1958 .get_port = udp_v4_get_port,
1959 .memory_allocated = &udp_memory_allocated,
1960 .sysctl_mem = sysctl_udp_mem,
1961 .sysctl_wmem = &sysctl_udp_wmem_min,
1962 .sysctl_rmem = &sysctl_udp_rmem_min,
1963 .obj_size = sizeof(struct udp_sock),
1964 .slab_flags = SLAB_DESTROY_BY_RCU,
1965 .h.udp_table = &udp_table,
1966 #ifdef CONFIG_COMPAT
1967 .compat_setsockopt = compat_udp_setsockopt,
1968 .compat_getsockopt = compat_udp_getsockopt,
1970 .clear_sk = sk_prot_clear_portaddr_nulls,
1972 EXPORT_SYMBOL(udp_prot);
1974 /* ------------------------------------------------------------------------ */
1975 #ifdef CONFIG_PROC_FS
1977 static struct sock *udp_get_first(struct seq_file *seq, int start)
1980 struct udp_iter_state *state = seq->private;
1981 struct net *net = seq_file_net(seq);
1983 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1985 struct hlist_nulls_node *node;
1986 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1988 if (hlist_nulls_empty(&hslot->head))
1991 spin_lock_bh(&hslot->lock);
1992 sk_nulls_for_each(sk, node, &hslot->head) {
1993 if (!net_eq(sock_net(sk), net))
1995 if (sk->sk_family == state->family)
1998 spin_unlock_bh(&hslot->lock);
2005 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2007 struct udp_iter_state *state = seq->private;
2008 struct net *net = seq_file_net(seq);
2011 sk = sk_nulls_next(sk);
2012 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2015 if (state->bucket <= state->udp_table->mask)
2016 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2017 return udp_get_first(seq, state->bucket + 1);
2022 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2024 struct sock *sk = udp_get_first(seq, 0);
2027 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2029 return pos ? NULL : sk;
2032 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2034 struct udp_iter_state *state = seq->private;
2035 state->bucket = MAX_UDP_PORTS;
2037 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2040 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2044 if (v == SEQ_START_TOKEN)
2045 sk = udp_get_idx(seq, 0);
2047 sk = udp_get_next(seq, v);
2053 static void udp_seq_stop(struct seq_file *seq, void *v)
2055 struct udp_iter_state *state = seq->private;
2057 if (state->bucket <= state->udp_table->mask)
2058 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2061 int udp_seq_open(struct inode *inode, struct file *file)
2063 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
2064 struct udp_iter_state *s;
2067 err = seq_open_net(inode, file, &afinfo->seq_ops,
2068 sizeof(struct udp_iter_state));
2072 s = ((struct seq_file *)file->private_data)->private;
2073 s->family = afinfo->family;
2074 s->udp_table = afinfo->udp_table;
2077 EXPORT_SYMBOL(udp_seq_open);
2079 /* ------------------------------------------------------------------------ */
2080 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2082 struct proc_dir_entry *p;
2085 afinfo->seq_ops.start = udp_seq_start;
2086 afinfo->seq_ops.next = udp_seq_next;
2087 afinfo->seq_ops.stop = udp_seq_stop;
2089 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2090 afinfo->seq_fops, afinfo);
2095 EXPORT_SYMBOL(udp_proc_register);
2097 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2099 proc_net_remove(net, afinfo->name);
2101 EXPORT_SYMBOL(udp_proc_unregister);
2103 /* ------------------------------------------------------------------------ */
2104 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2105 int bucket, int *len)
2107 struct inet_sock *inet = inet_sk(sp);
2108 __be32 dest = inet->inet_daddr;
2109 __be32 src = inet->inet_rcv_saddr;
2110 __u16 destp = ntohs(inet->inet_dport);
2111 __u16 srcp = ntohs(inet->inet_sport);
2113 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2114 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2115 bucket, src, srcp, dest, destp, sp->sk_state,
2116 sk_wmem_alloc_get(sp),
2117 sk_rmem_alloc_get(sp),
2119 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2121 atomic_read(&sp->sk_refcnt), sp,
2122 atomic_read(&sp->sk_drops), len);
2125 int udp4_seq_show(struct seq_file *seq, void *v)
2127 if (v == SEQ_START_TOKEN)
2128 seq_printf(seq, "%-127s\n",
2129 " sl local_address rem_address st tx_queue "
2130 "rx_queue tr tm->when retrnsmt uid timeout "
2131 "inode ref pointer drops");
2133 struct udp_iter_state *state = seq->private;
2136 udp4_format_sock(v, seq, state->bucket, &len);
2137 seq_printf(seq, "%*s\n", 127 - len, "");
2142 static const struct file_operations udp_afinfo_seq_fops = {
2143 .owner = THIS_MODULE,
2144 .open = udp_seq_open,
2146 .llseek = seq_lseek,
2147 .release = seq_release_net
2150 /* ------------------------------------------------------------------------ */
2151 static struct udp_seq_afinfo udp4_seq_afinfo = {
2154 .udp_table = &udp_table,
2155 .seq_fops = &udp_afinfo_seq_fops,
2157 .show = udp4_seq_show,
2161 static int __net_init udp4_proc_init_net(struct net *net)
2163 return udp_proc_register(net, &udp4_seq_afinfo);
2166 static void __net_exit udp4_proc_exit_net(struct net *net)
2168 udp_proc_unregister(net, &udp4_seq_afinfo);
2171 static struct pernet_operations udp4_net_ops = {
2172 .init = udp4_proc_init_net,
2173 .exit = udp4_proc_exit_net,
2176 int __init udp4_proc_init(void)
2178 return register_pernet_subsys(&udp4_net_ops);
2181 void udp4_proc_exit(void)
2183 unregister_pernet_subsys(&udp4_net_ops);
2185 #endif /* CONFIG_PROC_FS */
2187 static __initdata unsigned long uhash_entries;
2188 static int __init set_uhash_entries(char *str)
2195 ret = kstrtoul(str, 0, &uhash_entries);
2199 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2200 uhash_entries = UDP_HTABLE_SIZE_MIN;
2203 __setup("uhash_entries=", set_uhash_entries);
2205 void __init udp_table_init(struct udp_table *table, const char *name)
2209 table->hash = alloc_large_system_hash(name,
2210 2 * sizeof(struct udp_hslot),
2212 21, /* one slot per 2 MB */
2216 UDP_HTABLE_SIZE_MIN,
2219 table->hash2 = table->hash + (table->mask + 1);
2220 for (i = 0; i <= table->mask; i++) {
2221 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2222 table->hash[i].count = 0;
2223 spin_lock_init(&table->hash[i].lock);
2225 for (i = 0; i <= table->mask; i++) {
2226 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2227 table->hash2[i].count = 0;
2228 spin_lock_init(&table->hash2[i].lock);
2232 void __init udp_init(void)
2234 unsigned long limit;
2236 udp_table_init(&udp_table, "UDP");
2237 limit = nr_free_buffer_pages() / 8;
2238 limit = max(limit, 128UL);
2239 sysctl_udp_mem[0] = limit / 4 * 3;
2240 sysctl_udp_mem[1] = limit;
2241 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2243 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2244 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2247 int udp4_ufo_send_check(struct sk_buff *skb)
2249 const struct iphdr *iph;
2252 if (!pskb_may_pull(skb, sizeof(*uh)))
2258 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2260 skb->csum_start = skb_transport_header(skb) - skb->head;
2261 skb->csum_offset = offsetof(struct udphdr, check);
2262 skb->ip_summed = CHECKSUM_PARTIAL;
2266 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
2267 netdev_features_t features)
2269 struct sk_buff *segs = ERR_PTR(-EINVAL);
2274 mss = skb_shinfo(skb)->gso_size;
2275 if (unlikely(skb->len <= mss))
2278 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2279 /* Packet is from an untrusted source, reset gso_segs. */
2280 int type = skb_shinfo(skb)->gso_type;
2282 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2283 !(type & (SKB_GSO_UDP))))
2286 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2292 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2293 * do checksum of UDP packets sent as multiple IP fragments.
2295 offset = skb_checksum_start_offset(skb);
2296 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2297 offset += skb->csum_offset;
2298 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2299 skb->ip_summed = CHECKSUM_NONE;
2301 /* Fragment the skb. IP headers of the fragments are updated in
2302 * inet_gso_segment()
2304 segs = skb_segment(skb, features);