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 #include <asm/system.h>
81 #include <asm/uaccess.h>
82 #include <asm/ioctls.h>
83 #include <linux/bootmem.h>
84 #include <linux/highmem.h>
85 #include <linux/swap.h>
86 #include <linux/types.h>
87 #include <linux/fcntl.h>
88 #include <linux/module.h>
89 #include <linux/socket.h>
90 #include <linux/sockios.h>
91 #include <linux/igmp.h>
93 #include <linux/errno.h>
94 #include <linux/timer.h>
96 #include <linux/inet.h>
97 #include <linux/netdevice.h>
98 #include <net/tcp_states.h>
99 #include <linux/skbuff.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <net/net_namespace.h>
103 #include <net/icmp.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include "udp_impl.h"
109 struct udp_table udp_table;
110 EXPORT_SYMBOL(udp_table);
112 int sysctl_udp_mem[3] __read_mostly;
113 int sysctl_udp_rmem_min __read_mostly;
114 int sysctl_udp_wmem_min __read_mostly;
116 EXPORT_SYMBOL(sysctl_udp_mem);
117 EXPORT_SYMBOL(sysctl_udp_rmem_min);
118 EXPORT_SYMBOL(sysctl_udp_wmem_min);
120 atomic_t udp_memory_allocated;
121 EXPORT_SYMBOL(udp_memory_allocated);
123 #define PORTS_PER_CHAIN (65536 / UDP_HTABLE_SIZE)
125 static int udp_lib_lport_inuse(struct net *net, __u16 num,
126 const struct udp_hslot *hslot,
127 unsigned long *bitmap,
129 int (*saddr_comp)(const struct sock *sk1,
130 const struct sock *sk2))
133 struct hlist_nulls_node *node;
135 sk_nulls_for_each(sk2, node, &hslot->head)
136 if (net_eq(sock_net(sk2), net) &&
138 (bitmap || sk2->sk_hash == num) &&
139 (!sk2->sk_reuse || !sk->sk_reuse) &&
140 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
141 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
142 (*saddr_comp)(sk, sk2)) {
144 __set_bit(sk2->sk_hash / UDP_HTABLE_SIZE,
153 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
155 * @sk: socket struct in question
156 * @snum: port number to look up
157 * @saddr_comp: AF-dependent comparison of bound local IP addresses
159 int udp_lib_get_port(struct sock *sk, unsigned short snum,
160 int (*saddr_comp)(const struct sock *sk1,
161 const struct sock *sk2 ) )
163 struct udp_hslot *hslot;
164 struct udp_table *udptable = sk->sk_prot->h.udp_table;
166 struct net *net = sock_net(sk);
169 int low, high, remaining;
171 unsigned short first, last;
172 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
174 inet_get_local_port_range(&low, &high);
175 remaining = (high - low) + 1;
178 first = (((u64)rand * remaining) >> 32) + low;
180 * force rand to be an odd multiple of UDP_HTABLE_SIZE
182 rand = (rand | 1) * UDP_HTABLE_SIZE;
183 for (last = first + UDP_HTABLE_SIZE; first != last; first++) {
184 hslot = &udptable->hash[udp_hashfn(net, first)];
185 bitmap_zero(bitmap, PORTS_PER_CHAIN);
186 spin_lock_bh(&hslot->lock);
187 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
192 * Iterate on all possible values of snum for this hash.
193 * Using steps of an odd multiple of UDP_HTABLE_SIZE
194 * give us randomization and full range coverage.
197 if (low <= snum && snum <= high &&
198 !test_bit(snum / UDP_HTABLE_SIZE, bitmap))
201 } while (snum != first);
202 spin_unlock_bh(&hslot->lock);
206 hslot = &udptable->hash[udp_hashfn(net, snum)];
207 spin_lock_bh(&hslot->lock);
208 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, saddr_comp))
212 inet_sk(sk)->num = snum;
214 if (sk_unhashed(sk)) {
215 sk_nulls_add_node_rcu(sk, &hslot->head);
216 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
220 spin_unlock_bh(&hslot->lock);
225 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
227 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
229 return ( !ipv6_only_sock(sk2) &&
230 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
231 inet1->rcv_saddr == inet2->rcv_saddr ));
234 int udp_v4_get_port(struct sock *sk, unsigned short snum)
236 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
239 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
241 __be16 sport, __be32 daddr, __be16 dport, int dif)
245 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
246 !ipv6_only_sock(sk)) {
247 struct inet_sock *inet = inet_sk(sk);
249 score = (sk->sk_family == PF_INET ? 1 : 0);
250 if (inet->rcv_saddr) {
251 if (inet->rcv_saddr != daddr)
256 if (inet->daddr != saddr)
261 if (inet->dport != sport)
265 if (sk->sk_bound_dev_if) {
266 if (sk->sk_bound_dev_if != dif)
274 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
275 * harder than this. -DaveM
277 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
278 __be16 sport, __be32 daddr, __be16 dport,
279 int dif, struct udp_table *udptable)
281 struct sock *sk, *result;
282 struct hlist_nulls_node *node;
283 unsigned short hnum = ntohs(dport);
284 unsigned int hash = udp_hashfn(net, hnum);
285 struct udp_hslot *hslot = &udptable->hash[hash];
292 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
293 score = compute_score(sk, net, saddr, hnum, sport,
295 if (score > badness) {
301 * if the nulls value we got at the end of this lookup is
302 * not the expected one, we must restart lookup.
303 * We probably met an item that was moved to another chain.
305 if (get_nulls_value(node) != hash)
309 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
311 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
312 daddr, dport, dif) < badness)) {
321 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
322 __be16 sport, __be16 dport,
323 struct udp_table *udptable)
326 const struct iphdr *iph = ip_hdr(skb);
328 if (unlikely(sk = skb_steal_sock(skb)))
331 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
332 iph->daddr, dport, inet_iif(skb),
336 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
337 __be32 daddr, __be16 dport, int dif)
339 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
341 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
343 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
344 __be16 loc_port, __be32 loc_addr,
345 __be16 rmt_port, __be32 rmt_addr,
348 struct hlist_nulls_node *node;
350 unsigned short hnum = ntohs(loc_port);
352 sk_nulls_for_each_from(s, node) {
353 struct inet_sock *inet = inet_sk(s);
355 if (!net_eq(sock_net(s), net) ||
356 s->sk_hash != hnum ||
357 (inet->daddr && inet->daddr != rmt_addr) ||
358 (inet->dport != rmt_port && inet->dport) ||
359 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
361 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
363 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
373 * This routine is called by the ICMP module when it gets some
374 * sort of error condition. If err < 0 then the socket should
375 * be closed and the error returned to the user. If err > 0
376 * it's just the icmp type << 8 | icmp code.
377 * Header points to the ip header of the error packet. We move
378 * on past this. Then (as it used to claim before adjustment)
379 * header points to the first 8 bytes of the udp header. We need
380 * to find the appropriate port.
383 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
385 struct inet_sock *inet;
386 struct iphdr *iph = (struct iphdr*)skb->data;
387 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
388 const int type = icmp_hdr(skb)->type;
389 const int code = icmp_hdr(skb)->code;
393 struct net *net = dev_net(skb->dev);
395 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
396 iph->saddr, uh->source, skb->dev->ifindex, udptable);
398 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
399 return; /* No socket for error */
408 case ICMP_TIME_EXCEEDED:
411 case ICMP_SOURCE_QUENCH:
413 case ICMP_PARAMETERPROB:
417 case ICMP_DEST_UNREACH:
418 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
419 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
427 if (code <= NR_ICMP_UNREACH) {
428 harderr = icmp_err_convert[code].fatal;
429 err = icmp_err_convert[code].errno;
435 * RFC1122: OK. Passes ICMP errors back to application, as per
438 if (!inet->recverr) {
439 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
442 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
445 sk->sk_error_report(sk);
450 void udp_err(struct sk_buff *skb, u32 info)
452 __udp4_lib_err(skb, info, &udp_table);
456 * Throw away all pending data and cancel the corking. Socket is locked.
458 void udp_flush_pending_frames(struct sock *sk)
460 struct udp_sock *up = udp_sk(sk);
465 ip_flush_pending_frames(sk);
468 EXPORT_SYMBOL(udp_flush_pending_frames);
471 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
472 * @sk: socket we are sending on
473 * @skb: sk_buff containing the filled-in UDP header
474 * (checksum field must be zeroed out)
476 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
477 __be32 src, __be32 dst, int len )
480 struct udphdr *uh = udp_hdr(skb);
483 if (skb_queue_len(&sk->sk_write_queue) == 1) {
485 * Only one fragment on the socket.
487 skb->csum_start = skb_transport_header(skb) - skb->head;
488 skb->csum_offset = offsetof(struct udphdr, check);
489 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
492 * HW-checksum won't work as there are two or more
493 * fragments on the socket so that all csums of sk_buffs
496 offset = skb_transport_offset(skb);
497 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
499 skb->ip_summed = CHECKSUM_NONE;
501 skb_queue_walk(&sk->sk_write_queue, skb) {
502 csum = csum_add(csum, skb->csum);
505 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
507 uh->check = CSUM_MANGLED_0;
512 * Push out all pending data as one UDP datagram. Socket is locked.
514 static int udp_push_pending_frames(struct sock *sk)
516 struct udp_sock *up = udp_sk(sk);
517 struct inet_sock *inet = inet_sk(sk);
518 struct flowi *fl = &inet->cork.fl;
522 int is_udplite = IS_UDPLITE(sk);
525 /* Grab the skbuff where UDP header space exists. */
526 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
530 * Create a UDP header
533 uh->source = fl->fl_ip_sport;
534 uh->dest = fl->fl_ip_dport;
535 uh->len = htons(up->len);
538 if (is_udplite) /* UDP-Lite */
539 csum = udplite_csum_outgoing(sk, skb);
541 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
543 skb->ip_summed = CHECKSUM_NONE;
546 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
548 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
551 } else /* `normal' UDP */
552 csum = udp_csum_outgoing(sk, skb);
554 /* add protocol-dependent pseudo-header */
555 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
556 sk->sk_protocol, csum );
558 uh->check = CSUM_MANGLED_0;
561 err = ip_push_pending_frames(sk);
566 UDP_INC_STATS_USER(sock_net(sk),
567 UDP_MIB_OUTDATAGRAMS, is_udplite);
571 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
574 struct inet_sock *inet = inet_sk(sk);
575 struct udp_sock *up = udp_sk(sk);
577 struct ipcm_cookie ipc;
578 struct rtable *rt = NULL;
581 __be32 daddr, faddr, saddr;
584 int err, is_udplite = IS_UDPLITE(sk);
585 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
586 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
595 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
603 * There are pending frames.
604 * The socket lock must be held while it's corked.
607 if (likely(up->pending)) {
608 if (unlikely(up->pending != AF_INET)) {
616 ulen += sizeof(struct udphdr);
619 * Get and verify the address.
622 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
623 if (msg->msg_namelen < sizeof(*usin))
625 if (usin->sin_family != AF_INET) {
626 if (usin->sin_family != AF_UNSPEC)
627 return -EAFNOSUPPORT;
630 daddr = usin->sin_addr.s_addr;
631 dport = usin->sin_port;
635 if (sk->sk_state != TCP_ESTABLISHED)
636 return -EDESTADDRREQ;
639 /* Open fast path for connected socket.
640 Route will not be used, if at least one option is set.
644 ipc.addr = inet->saddr;
646 ipc.oif = sk->sk_bound_dev_if;
647 err = sock_tx_timestamp(msg, sk, &ipc.shtx);
650 if (msg->msg_controllen) {
651 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
662 ipc.addr = faddr = daddr;
664 if (ipc.opt && ipc.opt->srr) {
667 faddr = ipc.opt->faddr;
670 tos = RT_TOS(inet->tos);
671 if (sock_flag(sk, SOCK_LOCALROUTE) ||
672 (msg->msg_flags & MSG_DONTROUTE) ||
673 (ipc.opt && ipc.opt->is_strictroute)) {
678 if (ipv4_is_multicast(daddr)) {
680 ipc.oif = inet->mc_index;
682 saddr = inet->mc_addr;
687 rt = (struct rtable*)sk_dst_check(sk, 0);
690 struct flowi fl = { .oif = ipc.oif,
695 .proto = sk->sk_protocol,
696 .flags = inet_sk_flowi_flags(sk),
698 { .sport = inet->sport,
699 .dport = dport } } };
700 struct net *net = sock_net(sk);
702 security_sk_classify_flow(sk, &fl);
703 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
705 if (err == -ENETUNREACH)
706 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
711 if ((rt->rt_flags & RTCF_BROADCAST) &&
712 !sock_flag(sk, SOCK_BROADCAST))
715 sk_dst_set(sk, dst_clone(&rt->u.dst));
718 if (msg->msg_flags&MSG_CONFIRM)
724 daddr = ipc.addr = rt->rt_dst;
727 if (unlikely(up->pending)) {
728 /* The socket is already corked while preparing it. */
729 /* ... which is an evident application bug. --ANK */
732 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
737 * Now cork the socket to pend data.
739 inet->cork.fl.fl4_dst = daddr;
740 inet->cork.fl.fl_ip_dport = dport;
741 inet->cork.fl.fl4_src = saddr;
742 inet->cork.fl.fl_ip_sport = inet->sport;
743 up->pending = AF_INET;
747 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
748 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
749 sizeof(struct udphdr), &ipc, &rt,
750 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
752 udp_flush_pending_frames(sk);
754 err = udp_push_pending_frames(sk);
755 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
766 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
767 * ENOBUFS might not be good (it's not tunable per se), but otherwise
768 * we don't have a good statistic (IpOutDiscards but it can be too many
769 * things). We could add another new stat but at least for now that
770 * seems like overkill.
772 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
773 UDP_INC_STATS_USER(sock_net(sk),
774 UDP_MIB_SNDBUFERRORS, is_udplite);
779 dst_confirm(&rt->u.dst);
780 if (!(msg->msg_flags&MSG_PROBE) || len)
781 goto back_from_confirm;
786 int udp_sendpage(struct sock *sk, struct page *page, int offset,
787 size_t size, int flags)
789 struct udp_sock *up = udp_sk(sk);
793 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
795 /* Call udp_sendmsg to specify destination address which
796 * sendpage interface can't pass.
797 * This will succeed only when the socket is connected.
799 ret = udp_sendmsg(NULL, sk, &msg, 0);
806 if (unlikely(!up->pending)) {
809 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
813 ret = ip_append_page(sk, page, offset, size, flags);
814 if (ret == -EOPNOTSUPP) {
816 return sock_no_sendpage(sk->sk_socket, page, offset,
820 udp_flush_pending_frames(sk);
825 if (!(up->corkflag || (flags&MSG_MORE)))
826 ret = udp_push_pending_frames(sk);
835 * IOCTL requests applicable to the UDP protocol
838 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
843 int amount = atomic_read(&sk->sk_wmem_alloc);
844 return put_user(amount, (int __user *)arg);
850 unsigned long amount;
853 spin_lock_bh(&sk->sk_receive_queue.lock);
854 skb = skb_peek(&sk->sk_receive_queue);
857 * We will only return the amount
858 * of this packet since that is all
861 amount = skb->len - sizeof(struct udphdr);
863 spin_unlock_bh(&sk->sk_receive_queue.lock);
864 return put_user(amount, (int __user *)arg);
875 * This should be easy, if there is something there we
876 * return it, otherwise we block.
879 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
880 size_t len, int noblock, int flags, int *addr_len)
882 struct inet_sock *inet = inet_sk(sk);
883 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
885 unsigned int ulen, copied;
888 int is_udplite = IS_UDPLITE(sk);
891 * Check any passed addresses
894 *addr_len=sizeof(*sin);
896 if (flags & MSG_ERRQUEUE)
897 return ip_recv_error(sk, msg, len);
900 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
905 ulen = skb->len - sizeof(struct udphdr);
909 else if (copied < ulen)
910 msg->msg_flags |= MSG_TRUNC;
913 * If checksum is needed at all, try to do it while copying the
914 * data. If the data is truncated, or if we only want a partial
915 * coverage checksum (UDP-Lite), do it before the copy.
918 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
919 if (udp_lib_checksum_complete(skb))
923 if (skb_csum_unnecessary(skb))
924 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
925 msg->msg_iov, copied );
927 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
937 UDP_INC_STATS_USER(sock_net(sk),
938 UDP_MIB_INDATAGRAMS, is_udplite);
940 sock_recv_timestamp(msg, sk, skb);
942 /* Copy the address. */
945 sin->sin_family = AF_INET;
946 sin->sin_port = udp_hdr(skb)->source;
947 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
948 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
950 if (inet->cmsg_flags)
951 ip_cmsg_recv(msg, skb);
954 if (flags & MSG_TRUNC)
959 skb_free_datagram(sk, skb);
966 if (!skb_kill_datagram(sk, skb, flags))
967 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
976 int udp_disconnect(struct sock *sk, int flags)
978 struct inet_sock *inet = inet_sk(sk);
980 * 1003.1g - break association.
983 sk->sk_state = TCP_CLOSE;
986 sk->sk_bound_dev_if = 0;
987 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
988 inet_reset_saddr(sk);
990 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
991 sk->sk_prot->unhash(sk);
998 void udp_lib_unhash(struct sock *sk)
1000 if (sk_hashed(sk)) {
1001 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1002 unsigned int hash = udp_hashfn(sock_net(sk), sk->sk_hash);
1003 struct udp_hslot *hslot = &udptable->hash[hash];
1005 spin_lock_bh(&hslot->lock);
1006 if (sk_nulls_del_node_init_rcu(sk)) {
1007 inet_sk(sk)->num = 0;
1008 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1010 spin_unlock_bh(&hslot->lock);
1013 EXPORT_SYMBOL(udp_lib_unhash);
1015 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1017 int is_udplite = IS_UDPLITE(sk);
1020 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
1021 /* Note that an ENOMEM error is charged twice */
1022 if (rc == -ENOMEM) {
1023 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1025 atomic_inc(&sk->sk_drops);
1033 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1041 * >0: "udp encap" protocol resubmission
1043 * Note that in the success and error cases, the skb is assumed to
1044 * have either been requeued or freed.
1046 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
1048 struct udp_sock *up = udp_sk(sk);
1050 int is_udplite = IS_UDPLITE(sk);
1053 * Charge it to the socket, dropping if the queue is full.
1055 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1059 if (up->encap_type) {
1061 * This is an encapsulation socket so pass the skb to
1062 * the socket's udp_encap_rcv() hook. Otherwise, just
1063 * fall through and pass this up the UDP socket.
1064 * up->encap_rcv() returns the following value:
1065 * =0 if skb was successfully passed to the encap
1066 * handler or was discarded by it.
1067 * >0 if skb should be passed on to UDP.
1068 * <0 if skb should be resubmitted as proto -N
1071 /* if we're overly short, let UDP handle it */
1072 if (skb->len > sizeof(struct udphdr) &&
1073 up->encap_rcv != NULL) {
1076 ret = (*up->encap_rcv)(sk, skb);
1078 UDP_INC_STATS_BH(sock_net(sk),
1079 UDP_MIB_INDATAGRAMS,
1085 /* FALLTHROUGH -- it's a UDP Packet */
1089 * UDP-Lite specific tests, ignored on UDP sockets
1091 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1094 * MIB statistics other than incrementing the error count are
1095 * disabled for the following two types of errors: these depend
1096 * on the application settings, not on the functioning of the
1097 * protocol stack as such.
1099 * RFC 3828 here recommends (sec 3.3): "There should also be a
1100 * way ... to ... at least let the receiving application block
1101 * delivery of packets with coverage values less than a value
1102 * provided by the application."
1104 if (up->pcrlen == 0) { /* full coverage was set */
1105 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1106 "%d while full coverage %d requested\n",
1107 UDP_SKB_CB(skb)->cscov, skb->len);
1110 /* The next case involves violating the min. coverage requested
1111 * by the receiver. This is subtle: if receiver wants x and x is
1112 * greater than the buffersize/MTU then receiver will complain
1113 * that it wants x while sender emits packets of smaller size y.
1114 * Therefore the above ...()->partial_cov statement is essential.
1116 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1117 LIMIT_NETDEBUG(KERN_WARNING
1118 "UDPLITE: coverage %d too small, need min %d\n",
1119 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1124 if (sk->sk_filter) {
1125 if (udp_lib_checksum_complete(skb))
1132 if (!sock_owned_by_user(sk))
1133 rc = __udp_queue_rcv_skb(sk, skb);
1135 sk_add_backlog(sk, skb);
1141 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1147 * Multicasts and broadcasts go to each listener.
1149 * Note: called only from the BH handler context,
1150 * so we don't need to lock the hashes.
1152 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1154 __be32 saddr, __be32 daddr,
1155 struct udp_table *udptable)
1158 struct udp_hslot *hslot = &udptable->hash[udp_hashfn(net, ntohs(uh->dest))];
1161 spin_lock(&hslot->lock);
1162 sk = sk_nulls_head(&hslot->head);
1163 dif = skb->dev->ifindex;
1164 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1166 struct sock *sknext = NULL;
1169 struct sk_buff *skb1 = skb;
1171 sknext = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1172 daddr, uh->source, saddr,
1175 skb1 = skb_clone(skb, GFP_ATOMIC);
1178 int ret = udp_queue_rcv_skb(sk, skb1);
1180 /* we should probably re-process instead
1181 * of dropping packets here. */
1188 spin_unlock(&hslot->lock);
1192 /* Initialize UDP checksum. If exited with zero value (success),
1193 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1194 * Otherwise, csum completion requires chacksumming packet body,
1195 * including udp header and folding it to skb->csum.
1197 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1200 const struct iphdr *iph;
1203 UDP_SKB_CB(skb)->partial_cov = 0;
1204 UDP_SKB_CB(skb)->cscov = skb->len;
1206 if (proto == IPPROTO_UDPLITE) {
1207 err = udplite_checksum_init(skb, uh);
1213 if (uh->check == 0) {
1214 skb->ip_summed = CHECKSUM_UNNECESSARY;
1215 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1216 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1218 skb->ip_summed = CHECKSUM_UNNECESSARY;
1220 if (!skb_csum_unnecessary(skb))
1221 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1222 skb->len, proto, 0);
1223 /* Probably, we should checksum udp header (it should be in cache
1224 * in any case) and data in tiny packets (< rx copybreak).
1231 * All we need to do is get the socket, and then do a checksum.
1234 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1239 unsigned short ulen;
1240 struct rtable *rt = skb_rtable(skb);
1241 __be32 saddr, daddr;
1242 struct net *net = dev_net(skb->dev);
1245 * Validate the packet.
1247 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1248 goto drop; /* No space for header. */
1251 ulen = ntohs(uh->len);
1252 if (ulen > skb->len)
1255 if (proto == IPPROTO_UDP) {
1256 /* UDP validates ulen. */
1257 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1262 if (udp4_csum_init(skb, uh, proto))
1265 saddr = ip_hdr(skb)->saddr;
1266 daddr = ip_hdr(skb)->daddr;
1268 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1269 return __udp4_lib_mcast_deliver(net, skb, uh,
1270 saddr, daddr, udptable);
1272 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1275 int ret = udp_queue_rcv_skb(sk, skb);
1278 /* a return value > 0 means to resubmit the input, but
1279 * it wants the return to be -protocol, or 0
1286 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1290 /* No socket. Drop packet silently, if checksum is wrong */
1291 if (udp_lib_checksum_complete(skb))
1294 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1295 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1298 * Hmm. We got an UDP packet to a port to which we
1299 * don't wanna listen. Ignore it.
1305 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1306 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1317 * RFC1122: OK. Discards the bad packet silently (as far as
1318 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1320 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1321 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1328 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1333 int udp_rcv(struct sk_buff *skb)
1335 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1338 void udp_destroy_sock(struct sock *sk)
1341 udp_flush_pending_frames(sk);
1346 * Socket option code for UDP
1348 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1349 char __user *optval, int optlen,
1350 int (*push_pending_frames)(struct sock *))
1352 struct udp_sock *up = udp_sk(sk);
1355 int is_udplite = IS_UDPLITE(sk);
1357 if (optlen<sizeof(int))
1360 if (get_user(val, (int __user *)optval))
1370 (*push_pending_frames)(sk);
1378 case UDP_ENCAP_ESPINUDP:
1379 case UDP_ENCAP_ESPINUDP_NON_IKE:
1380 up->encap_rcv = xfrm4_udp_encap_rcv;
1382 case UDP_ENCAP_L2TPINUDP:
1383 up->encap_type = val;
1392 * UDP-Lite's partial checksum coverage (RFC 3828).
1394 /* The sender sets actual checksum coverage length via this option.
1395 * The case coverage > packet length is handled by send module. */
1396 case UDPLITE_SEND_CSCOV:
1397 if (!is_udplite) /* Disable the option on UDP sockets */
1398 return -ENOPROTOOPT;
1399 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1401 else if (val > USHORT_MAX)
1404 up->pcflag |= UDPLITE_SEND_CC;
1407 /* The receiver specifies a minimum checksum coverage value. To make
1408 * sense, this should be set to at least 8 (as done below). If zero is
1409 * used, this again means full checksum coverage. */
1410 case UDPLITE_RECV_CSCOV:
1411 if (!is_udplite) /* Disable the option on UDP sockets */
1412 return -ENOPROTOOPT;
1413 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1415 else if (val > USHORT_MAX)
1418 up->pcflag |= UDPLITE_RECV_CC;
1429 int udp_setsockopt(struct sock *sk, int level, int optname,
1430 char __user *optval, int optlen)
1432 if (level == SOL_UDP || level == SOL_UDPLITE)
1433 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1434 udp_push_pending_frames);
1435 return ip_setsockopt(sk, level, optname, optval, optlen);
1438 #ifdef CONFIG_COMPAT
1439 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1440 char __user *optval, int optlen)
1442 if (level == SOL_UDP || level == SOL_UDPLITE)
1443 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1444 udp_push_pending_frames);
1445 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1449 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1450 char __user *optval, int __user *optlen)
1452 struct udp_sock *up = udp_sk(sk);
1455 if (get_user(len,optlen))
1458 len = min_t(unsigned int, len, sizeof(int));
1469 val = up->encap_type;
1472 /* The following two cannot be changed on UDP sockets, the return is
1473 * always 0 (which corresponds to the full checksum coverage of UDP). */
1474 case UDPLITE_SEND_CSCOV:
1478 case UDPLITE_RECV_CSCOV:
1483 return -ENOPROTOOPT;
1486 if (put_user(len, optlen))
1488 if (copy_to_user(optval, &val,len))
1493 int udp_getsockopt(struct sock *sk, int level, int optname,
1494 char __user *optval, int __user *optlen)
1496 if (level == SOL_UDP || level == SOL_UDPLITE)
1497 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1498 return ip_getsockopt(sk, level, optname, optval, optlen);
1501 #ifdef CONFIG_COMPAT
1502 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1503 char __user *optval, int __user *optlen)
1505 if (level == SOL_UDP || level == SOL_UDPLITE)
1506 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1507 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1511 * udp_poll - wait for a UDP event.
1512 * @file - file struct
1514 * @wait - poll table
1516 * This is same as datagram poll, except for the special case of
1517 * blocking sockets. If application is using a blocking fd
1518 * and a packet with checksum error is in the queue;
1519 * then it could get return from select indicating data available
1520 * but then block when reading it. Add special case code
1521 * to work around these arguably broken applications.
1523 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1525 unsigned int mask = datagram_poll(file, sock, wait);
1526 struct sock *sk = sock->sk;
1527 int is_lite = IS_UDPLITE(sk);
1529 /* Check for false positives due to checksum errors */
1530 if ( (mask & POLLRDNORM) &&
1531 !(file->f_flags & O_NONBLOCK) &&
1532 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1533 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1534 struct sk_buff *skb;
1536 spin_lock_bh(&rcvq->lock);
1537 while ((skb = skb_peek(rcvq)) != NULL &&
1538 udp_lib_checksum_complete(skb)) {
1539 UDP_INC_STATS_BH(sock_net(sk),
1540 UDP_MIB_INERRORS, is_lite);
1541 __skb_unlink(skb, rcvq);
1544 spin_unlock_bh(&rcvq->lock);
1546 /* nothing to see, move along */
1548 mask &= ~(POLLIN | POLLRDNORM);
1555 struct proto udp_prot = {
1557 .owner = THIS_MODULE,
1558 .close = udp_lib_close,
1559 .connect = ip4_datagram_connect,
1560 .disconnect = udp_disconnect,
1562 .destroy = udp_destroy_sock,
1563 .setsockopt = udp_setsockopt,
1564 .getsockopt = udp_getsockopt,
1565 .sendmsg = udp_sendmsg,
1566 .recvmsg = udp_recvmsg,
1567 .sendpage = udp_sendpage,
1568 .backlog_rcv = __udp_queue_rcv_skb,
1569 .hash = udp_lib_hash,
1570 .unhash = udp_lib_unhash,
1571 .get_port = udp_v4_get_port,
1572 .memory_allocated = &udp_memory_allocated,
1573 .sysctl_mem = sysctl_udp_mem,
1574 .sysctl_wmem = &sysctl_udp_wmem_min,
1575 .sysctl_rmem = &sysctl_udp_rmem_min,
1576 .obj_size = sizeof(struct udp_sock),
1577 .slab_flags = SLAB_DESTROY_BY_RCU,
1578 .h.udp_table = &udp_table,
1579 #ifdef CONFIG_COMPAT
1580 .compat_setsockopt = compat_udp_setsockopt,
1581 .compat_getsockopt = compat_udp_getsockopt,
1585 /* ------------------------------------------------------------------------ */
1586 #ifdef CONFIG_PROC_FS
1588 static struct sock *udp_get_first(struct seq_file *seq, int start)
1591 struct udp_iter_state *state = seq->private;
1592 struct net *net = seq_file_net(seq);
1594 for (state->bucket = start; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1595 struct hlist_nulls_node *node;
1596 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1597 spin_lock_bh(&hslot->lock);
1598 sk_nulls_for_each(sk, node, &hslot->head) {
1599 if (!net_eq(sock_net(sk), net))
1601 if (sk->sk_family == state->family)
1604 spin_unlock_bh(&hslot->lock);
1611 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1613 struct udp_iter_state *state = seq->private;
1614 struct net *net = seq_file_net(seq);
1617 sk = sk_nulls_next(sk);
1618 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1621 if (state->bucket < UDP_HTABLE_SIZE)
1622 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1623 return udp_get_first(seq, state->bucket + 1);
1628 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1630 struct sock *sk = udp_get_first(seq, 0);
1633 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1635 return pos ? NULL : sk;
1638 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1640 struct udp_iter_state *state = seq->private;
1641 state->bucket = UDP_HTABLE_SIZE;
1643 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1646 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1650 if (v == SEQ_START_TOKEN)
1651 sk = udp_get_idx(seq, 0);
1653 sk = udp_get_next(seq, v);
1659 static void udp_seq_stop(struct seq_file *seq, void *v)
1661 struct udp_iter_state *state = seq->private;
1663 if (state->bucket < UDP_HTABLE_SIZE)
1664 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1667 static int udp_seq_open(struct inode *inode, struct file *file)
1669 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1670 struct udp_iter_state *s;
1673 err = seq_open_net(inode, file, &afinfo->seq_ops,
1674 sizeof(struct udp_iter_state));
1678 s = ((struct seq_file *)file->private_data)->private;
1679 s->family = afinfo->family;
1680 s->udp_table = afinfo->udp_table;
1684 /* ------------------------------------------------------------------------ */
1685 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1687 struct proc_dir_entry *p;
1690 afinfo->seq_fops.open = udp_seq_open;
1691 afinfo->seq_fops.read = seq_read;
1692 afinfo->seq_fops.llseek = seq_lseek;
1693 afinfo->seq_fops.release = seq_release_net;
1695 afinfo->seq_ops.start = udp_seq_start;
1696 afinfo->seq_ops.next = udp_seq_next;
1697 afinfo->seq_ops.stop = udp_seq_stop;
1699 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1700 &afinfo->seq_fops, afinfo);
1706 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1708 proc_net_remove(net, afinfo->name);
1711 /* ------------------------------------------------------------------------ */
1712 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1713 int bucket, int *len)
1715 struct inet_sock *inet = inet_sk(sp);
1716 __be32 dest = inet->daddr;
1717 __be32 src = inet->rcv_saddr;
1718 __u16 destp = ntohs(inet->dport);
1719 __u16 srcp = ntohs(inet->sport);
1721 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
1722 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1723 bucket, src, srcp, dest, destp, sp->sk_state,
1724 atomic_read(&sp->sk_wmem_alloc),
1725 atomic_read(&sp->sk_rmem_alloc),
1726 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1727 atomic_read(&sp->sk_refcnt), sp,
1728 atomic_read(&sp->sk_drops), len);
1731 int udp4_seq_show(struct seq_file *seq, void *v)
1733 if (v == SEQ_START_TOKEN)
1734 seq_printf(seq, "%-127s\n",
1735 " sl local_address rem_address st tx_queue "
1736 "rx_queue tr tm->when retrnsmt uid timeout "
1737 "inode ref pointer drops");
1739 struct udp_iter_state *state = seq->private;
1742 udp4_format_sock(v, seq, state->bucket, &len);
1743 seq_printf(seq, "%*s\n", 127 - len ,"");
1748 /* ------------------------------------------------------------------------ */
1749 static struct udp_seq_afinfo udp4_seq_afinfo = {
1752 .udp_table = &udp_table,
1754 .owner = THIS_MODULE,
1757 .show = udp4_seq_show,
1761 static int udp4_proc_init_net(struct net *net)
1763 return udp_proc_register(net, &udp4_seq_afinfo);
1766 static void udp4_proc_exit_net(struct net *net)
1768 udp_proc_unregister(net, &udp4_seq_afinfo);
1771 static struct pernet_operations udp4_net_ops = {
1772 .init = udp4_proc_init_net,
1773 .exit = udp4_proc_exit_net,
1776 int __init udp4_proc_init(void)
1778 return register_pernet_subsys(&udp4_net_ops);
1781 void udp4_proc_exit(void)
1783 unregister_pernet_subsys(&udp4_net_ops);
1785 #endif /* CONFIG_PROC_FS */
1787 void __init udp_table_init(struct udp_table *table)
1791 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
1792 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
1793 spin_lock_init(&table->hash[i].lock);
1797 void __init udp_init(void)
1799 unsigned long nr_pages, limit;
1801 udp_table_init(&udp_table);
1802 /* Set the pressure threshold up by the same strategy of TCP. It is a
1803 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1804 * toward zero with the amount of memory, with a floor of 128 pages.
1806 nr_pages = totalram_pages - totalhigh_pages;
1807 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1808 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1809 limit = max(limit, 128UL);
1810 sysctl_udp_mem[0] = limit / 4 * 3;
1811 sysctl_udp_mem[1] = limit;
1812 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1814 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1815 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1818 EXPORT_SYMBOL(udp_disconnect);
1819 EXPORT_SYMBOL(udp_ioctl);
1820 EXPORT_SYMBOL(udp_prot);
1821 EXPORT_SYMBOL(udp_sendmsg);
1822 EXPORT_SYMBOL(udp_lib_getsockopt);
1823 EXPORT_SYMBOL(udp_lib_setsockopt);
1824 EXPORT_SYMBOL(udp_poll);
1825 EXPORT_SYMBOL(udp_lib_get_port);
1827 #ifdef CONFIG_PROC_FS
1828 EXPORT_SYMBOL(udp_proc_register);
1829 EXPORT_SYMBOL(udp_proc_unregister);