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 EXPORT_SYMBOL(sysctl_udp_mem);
115 int sysctl_udp_rmem_min __read_mostly;
116 EXPORT_SYMBOL(sysctl_udp_rmem_min);
118 int sysctl_udp_wmem_min __read_mostly;
119 EXPORT_SYMBOL(sysctl_udp_wmem_min);
121 atomic_t udp_memory_allocated;
122 EXPORT_SYMBOL(udp_memory_allocated);
124 #define PORTS_PER_CHAIN (65536 / UDP_HTABLE_SIZE)
126 static int udp_lib_lport_inuse(struct net *net, __u16 num,
127 const struct udp_hslot *hslot,
128 unsigned long *bitmap,
130 int (*saddr_comp)(const struct sock *sk1,
131 const struct sock *sk2))
134 struct hlist_nulls_node *node;
136 sk_nulls_for_each(sk2, node, &hslot->head)
137 if (net_eq(sock_net(sk2), net) &&
139 (bitmap || sk2->sk_hash == num) &&
140 (!sk2->sk_reuse || !sk->sk_reuse) &&
141 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
142 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
143 (*saddr_comp)(sk, sk2)) {
145 __set_bit(sk2->sk_hash / UDP_HTABLE_SIZE,
154 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
156 * @sk: socket struct in question
157 * @snum: port number to look up
158 * @saddr_comp: AF-dependent comparison of bound local IP addresses
160 int udp_lib_get_port(struct sock *sk, unsigned short snum,
161 int (*saddr_comp)(const struct sock *sk1,
162 const struct sock *sk2))
164 struct udp_hslot *hslot;
165 struct udp_table *udptable = sk->sk_prot->h.udp_table;
167 struct net *net = sock_net(sk);
170 int low, high, remaining;
172 unsigned short first, last;
173 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
175 inet_get_local_port_range(&low, &high);
176 remaining = (high - low) + 1;
179 first = (((u64)rand * remaining) >> 32) + low;
181 * force rand to be an odd multiple of UDP_HTABLE_SIZE
183 rand = (rand | 1) * UDP_HTABLE_SIZE;
184 for (last = first + UDP_HTABLE_SIZE; first != last; first++) {
185 hslot = &udptable->hash[udp_hashfn(net, first)];
186 bitmap_zero(bitmap, PORTS_PER_CHAIN);
187 spin_lock_bh(&hslot->lock);
188 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
193 * Iterate on all possible values of snum for this hash.
194 * Using steps of an odd multiple of UDP_HTABLE_SIZE
195 * give us randomization and full range coverage.
198 if (low <= snum && snum <= high &&
199 !test_bit(snum / UDP_HTABLE_SIZE, bitmap))
202 } while (snum != first);
203 spin_unlock_bh(&hslot->lock);
207 hslot = &udptable->hash[udp_hashfn(net, snum)];
208 spin_lock_bh(&hslot->lock);
209 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, saddr_comp))
213 inet_sk(sk)->num = snum;
215 if (sk_unhashed(sk)) {
216 sk_nulls_add_node_rcu(sk, &hslot->head);
217 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
221 spin_unlock_bh(&hslot->lock);
225 EXPORT_SYMBOL(udp_lib_get_port);
227 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
229 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
231 return (!ipv6_only_sock(sk2) &&
232 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
233 inet1->rcv_saddr == inet2->rcv_saddr));
236 int udp_v4_get_port(struct sock *sk, unsigned short snum)
238 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
241 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
243 __be16 sport, __be32 daddr, __be16 dport, int dif)
247 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
248 !ipv6_only_sock(sk)) {
249 struct inet_sock *inet = inet_sk(sk);
251 score = (sk->sk_family == PF_INET ? 1 : 0);
252 if (inet->rcv_saddr) {
253 if (inet->rcv_saddr != daddr)
258 if (inet->daddr != saddr)
263 if (inet->dport != sport)
267 if (sk->sk_bound_dev_if) {
268 if (sk->sk_bound_dev_if != dif)
276 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
277 * harder than this. -DaveM
279 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
280 __be16 sport, __be32 daddr, __be16 dport,
281 int dif, struct udp_table *udptable)
283 struct sock *sk, *result;
284 struct hlist_nulls_node *node;
285 unsigned short hnum = ntohs(dport);
286 unsigned int hash = udp_hashfn(net, hnum);
287 struct udp_hslot *hslot = &udptable->hash[hash];
294 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
295 score = compute_score(sk, net, saddr, hnum, sport,
297 if (score > badness) {
303 * if the nulls value we got at the end of this lookup is
304 * not the expected one, we must restart lookup.
305 * We probably met an item that was moved to another chain.
307 if (get_nulls_value(node) != hash)
311 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
313 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
314 daddr, dport, dif) < badness)) {
323 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
324 __be16 sport, __be16 dport,
325 struct udp_table *udptable)
328 const struct iphdr *iph = ip_hdr(skb);
330 if (unlikely(sk = skb_steal_sock(skb)))
333 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
334 iph->daddr, dport, inet_iif(skb),
338 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
339 __be32 daddr, __be16 dport, int dif)
341 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
343 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
345 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
346 __be16 loc_port, __be32 loc_addr,
347 __be16 rmt_port, __be32 rmt_addr,
350 struct hlist_nulls_node *node;
352 unsigned short hnum = ntohs(loc_port);
354 sk_nulls_for_each_from(s, node) {
355 struct inet_sock *inet = inet_sk(s);
357 if (!net_eq(sock_net(s), net) ||
358 s->sk_hash != hnum ||
359 (inet->daddr && inet->daddr != rmt_addr) ||
360 (inet->dport != rmt_port && inet->dport) ||
361 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
363 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
365 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
375 * This routine is called by the ICMP module when it gets some
376 * sort of error condition. If err < 0 then the socket should
377 * be closed and the error returned to the user. If err > 0
378 * it's just the icmp type << 8 | icmp code.
379 * Header points to the ip header of the error packet. We move
380 * on past this. Then (as it used to claim before adjustment)
381 * header points to the first 8 bytes of the udp header. We need
382 * to find the appropriate port.
385 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
387 struct inet_sock *inet;
388 struct iphdr *iph = (struct iphdr *)skb->data;
389 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
390 const int type = icmp_hdr(skb)->type;
391 const int code = icmp_hdr(skb)->code;
395 struct net *net = dev_net(skb->dev);
397 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
398 iph->saddr, uh->source, skb->dev->ifindex, udptable);
400 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
401 return; /* No socket for error */
410 case ICMP_TIME_EXCEEDED:
413 case ICMP_SOURCE_QUENCH:
415 case ICMP_PARAMETERPROB:
419 case ICMP_DEST_UNREACH:
420 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
421 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
429 if (code <= NR_ICMP_UNREACH) {
430 harderr = icmp_err_convert[code].fatal;
431 err = icmp_err_convert[code].errno;
437 * RFC1122: OK. Passes ICMP errors back to application, as per
440 if (!inet->recverr) {
441 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
444 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
447 sk->sk_error_report(sk);
452 void udp_err(struct sk_buff *skb, u32 info)
454 __udp4_lib_err(skb, info, &udp_table);
458 * Throw away all pending data and cancel the corking. Socket is locked.
460 void udp_flush_pending_frames(struct sock *sk)
462 struct udp_sock *up = udp_sk(sk);
467 ip_flush_pending_frames(sk);
470 EXPORT_SYMBOL(udp_flush_pending_frames);
473 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
474 * @sk: socket we are sending on
475 * @skb: sk_buff containing the filled-in UDP header
476 * (checksum field must be zeroed out)
478 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
479 __be32 src, __be32 dst, int len)
482 struct udphdr *uh = udp_hdr(skb);
485 if (skb_queue_len(&sk->sk_write_queue) == 1) {
487 * Only one fragment on the socket.
489 skb->csum_start = skb_transport_header(skb) - skb->head;
490 skb->csum_offset = offsetof(struct udphdr, check);
491 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
494 * HW-checksum won't work as there are two or more
495 * fragments on the socket so that all csums of sk_buffs
498 offset = skb_transport_offset(skb);
499 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
501 skb->ip_summed = CHECKSUM_NONE;
503 skb_queue_walk(&sk->sk_write_queue, skb) {
504 csum = csum_add(csum, skb->csum);
507 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
509 uh->check = CSUM_MANGLED_0;
514 * Push out all pending data as one UDP datagram. Socket is locked.
516 static int udp_push_pending_frames(struct sock *sk)
518 struct udp_sock *up = udp_sk(sk);
519 struct inet_sock *inet = inet_sk(sk);
520 struct flowi *fl = &inet->cork.fl;
524 int is_udplite = IS_UDPLITE(sk);
527 /* Grab the skbuff where UDP header space exists. */
528 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
532 * Create a UDP header
535 uh->source = fl->fl_ip_sport;
536 uh->dest = fl->fl_ip_dport;
537 uh->len = htons(up->len);
540 if (is_udplite) /* UDP-Lite */
541 csum = udplite_csum_outgoing(sk, skb);
543 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
545 skb->ip_summed = CHECKSUM_NONE;
548 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
550 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len);
553 } else /* `normal' UDP */
554 csum = udp_csum_outgoing(sk, skb);
556 /* add protocol-dependent pseudo-header */
557 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
558 sk->sk_protocol, csum);
560 uh->check = CSUM_MANGLED_0;
563 err = ip_push_pending_frames(sk);
565 if (err == -ENOBUFS && !inet->recverr) {
566 UDP_INC_STATS_USER(sock_net(sk),
567 UDP_MIB_SNDBUFERRORS, is_udplite);
571 UDP_INC_STATS_USER(sock_net(sk),
572 UDP_MIB_OUTDATAGRAMS, is_udplite);
579 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
582 struct inet_sock *inet = inet_sk(sk);
583 struct udp_sock *up = udp_sk(sk);
585 struct ipcm_cookie ipc;
586 struct rtable *rt = NULL;
589 __be32 daddr, faddr, saddr;
592 int err, is_udplite = IS_UDPLITE(sk);
593 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
594 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
603 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
611 * There are pending frames.
612 * The socket lock must be held while it's corked.
615 if (likely(up->pending)) {
616 if (unlikely(up->pending != AF_INET)) {
624 ulen += sizeof(struct udphdr);
627 * Get and verify the address.
630 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
631 if (msg->msg_namelen < sizeof(*usin))
633 if (usin->sin_family != AF_INET) {
634 if (usin->sin_family != AF_UNSPEC)
635 return -EAFNOSUPPORT;
638 daddr = usin->sin_addr.s_addr;
639 dport = usin->sin_port;
643 if (sk->sk_state != TCP_ESTABLISHED)
644 return -EDESTADDRREQ;
647 /* Open fast path for connected socket.
648 Route will not be used, if at least one option is set.
652 ipc.addr = inet->saddr;
654 ipc.oif = sk->sk_bound_dev_if;
655 err = sock_tx_timestamp(msg, sk, &ipc.shtx);
658 if (msg->msg_controllen) {
659 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
670 ipc.addr = faddr = daddr;
672 if (ipc.opt && ipc.opt->srr) {
675 faddr = ipc.opt->faddr;
678 tos = RT_TOS(inet->tos);
679 if (sock_flag(sk, SOCK_LOCALROUTE) ||
680 (msg->msg_flags & MSG_DONTROUTE) ||
681 (ipc.opt && ipc.opt->is_strictroute)) {
686 if (ipv4_is_multicast(daddr)) {
688 ipc.oif = inet->mc_index;
690 saddr = inet->mc_addr;
695 rt = (struct rtable *)sk_dst_check(sk, 0);
698 struct flowi fl = { .oif = ipc.oif,
704 .proto = sk->sk_protocol,
705 .flags = inet_sk_flowi_flags(sk),
707 { .sport = inet->sport,
708 .dport = dport } } };
709 struct net *net = sock_net(sk);
711 security_sk_classify_flow(sk, &fl);
712 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
714 if (err == -ENETUNREACH)
715 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
720 if ((rt->rt_flags & RTCF_BROADCAST) &&
721 !sock_flag(sk, SOCK_BROADCAST))
724 sk_dst_set(sk, dst_clone(&rt->u.dst));
727 if (msg->msg_flags&MSG_CONFIRM)
733 daddr = ipc.addr = rt->rt_dst;
736 if (unlikely(up->pending)) {
737 /* The socket is already corked while preparing it. */
738 /* ... which is an evident application bug. --ANK */
741 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
746 * Now cork the socket to pend data.
748 inet->cork.fl.fl4_dst = daddr;
749 inet->cork.fl.fl_ip_dport = dport;
750 inet->cork.fl.fl4_src = saddr;
751 inet->cork.fl.fl_ip_sport = inet->sport;
752 up->pending = AF_INET;
756 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
757 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
758 sizeof(struct udphdr), &ipc, &rt,
759 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
761 udp_flush_pending_frames(sk);
763 err = udp_push_pending_frames(sk);
764 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
775 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
776 * ENOBUFS might not be good (it's not tunable per se), but otherwise
777 * we don't have a good statistic (IpOutDiscards but it can be too many
778 * things). We could add another new stat but at least for now that
779 * seems like overkill.
781 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
782 UDP_INC_STATS_USER(sock_net(sk),
783 UDP_MIB_SNDBUFERRORS, is_udplite);
788 dst_confirm(&rt->u.dst);
789 if (!(msg->msg_flags&MSG_PROBE) || len)
790 goto back_from_confirm;
794 EXPORT_SYMBOL(udp_sendmsg);
796 int udp_sendpage(struct sock *sk, struct page *page, int offset,
797 size_t size, int flags)
799 struct udp_sock *up = udp_sk(sk);
803 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
805 /* Call udp_sendmsg to specify destination address which
806 * sendpage interface can't pass.
807 * This will succeed only when the socket is connected.
809 ret = udp_sendmsg(NULL, sk, &msg, 0);
816 if (unlikely(!up->pending)) {
819 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
823 ret = ip_append_page(sk, page, offset, size, flags);
824 if (ret == -EOPNOTSUPP) {
826 return sock_no_sendpage(sk->sk_socket, page, offset,
830 udp_flush_pending_frames(sk);
835 if (!(up->corkflag || (flags&MSG_MORE)))
836 ret = udp_push_pending_frames(sk);
845 * IOCTL requests applicable to the UDP protocol
848 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
853 int amount = sk_wmem_alloc_get(sk);
855 return put_user(amount, (int __user *)arg);
861 unsigned long amount;
864 spin_lock_bh(&sk->sk_receive_queue.lock);
865 skb = skb_peek(&sk->sk_receive_queue);
868 * We will only return the amount
869 * of this packet since that is all
872 amount = skb->len - sizeof(struct udphdr);
874 spin_unlock_bh(&sk->sk_receive_queue.lock);
875 return put_user(amount, (int __user *)arg);
884 EXPORT_SYMBOL(udp_ioctl);
887 * This should be easy, if there is something there we
888 * return it, otherwise we block.
891 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
892 size_t len, int noblock, int flags, int *addr_len)
894 struct inet_sock *inet = inet_sk(sk);
895 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
897 unsigned int ulen, copied;
900 int is_udplite = IS_UDPLITE(sk);
903 * Check any passed addresses
906 *addr_len = sizeof(*sin);
908 if (flags & MSG_ERRQUEUE)
909 return ip_recv_error(sk, msg, len);
912 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
917 ulen = skb->len - sizeof(struct udphdr);
921 else if (copied < ulen)
922 msg->msg_flags |= MSG_TRUNC;
925 * If checksum is needed at all, try to do it while copying the
926 * data. If the data is truncated, or if we only want a partial
927 * coverage checksum (UDP-Lite), do it before the copy.
930 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
931 if (udp_lib_checksum_complete(skb))
935 if (skb_csum_unnecessary(skb))
936 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
937 msg->msg_iov, copied);
939 err = skb_copy_and_csum_datagram_iovec(skb,
940 sizeof(struct udphdr),
951 UDP_INC_STATS_USER(sock_net(sk),
952 UDP_MIB_INDATAGRAMS, is_udplite);
954 sock_recv_timestamp(msg, sk, skb);
956 /* Copy the address. */
958 sin->sin_family = AF_INET;
959 sin->sin_port = udp_hdr(skb)->source;
960 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
961 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
963 if (inet->cmsg_flags)
964 ip_cmsg_recv(msg, skb);
967 if (flags & MSG_TRUNC)
972 skb_free_datagram(sk, skb);
979 if (!skb_kill_datagram(sk, skb, flags))
980 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
989 int udp_disconnect(struct sock *sk, int flags)
991 struct inet_sock *inet = inet_sk(sk);
993 * 1003.1g - break association.
996 sk->sk_state = TCP_CLOSE;
999 sk->sk_bound_dev_if = 0;
1000 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1001 inet_reset_saddr(sk);
1003 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1004 sk->sk_prot->unhash(sk);
1010 EXPORT_SYMBOL(udp_disconnect);
1012 void udp_lib_unhash(struct sock *sk)
1014 if (sk_hashed(sk)) {
1015 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1016 unsigned int hash = udp_hashfn(sock_net(sk), sk->sk_hash);
1017 struct udp_hslot *hslot = &udptable->hash[hash];
1019 spin_lock_bh(&hslot->lock);
1020 if (sk_nulls_del_node_init_rcu(sk)) {
1021 inet_sk(sk)->num = 0;
1022 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1024 spin_unlock_bh(&hslot->lock);
1027 EXPORT_SYMBOL(udp_lib_unhash);
1029 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1031 int is_udplite = IS_UDPLITE(sk);
1034 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
1035 /* Note that an ENOMEM error is charged twice */
1036 if (rc == -ENOMEM) {
1037 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1039 atomic_inc(&sk->sk_drops);
1047 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1055 * >0: "udp encap" protocol resubmission
1057 * Note that in the success and error cases, the skb is assumed to
1058 * have either been requeued or freed.
1060 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1062 struct udp_sock *up = udp_sk(sk);
1064 int is_udplite = IS_UDPLITE(sk);
1067 * Charge it to the socket, dropping if the queue is full.
1069 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1073 if (up->encap_type) {
1075 * This is an encapsulation socket so pass the skb to
1076 * the socket's udp_encap_rcv() hook. Otherwise, just
1077 * fall through and pass this up the UDP socket.
1078 * up->encap_rcv() returns the following value:
1079 * =0 if skb was successfully passed to the encap
1080 * handler or was discarded by it.
1081 * >0 if skb should be passed on to UDP.
1082 * <0 if skb should be resubmitted as proto -N
1085 /* if we're overly short, let UDP handle it */
1086 if (skb->len > sizeof(struct udphdr) &&
1087 up->encap_rcv != NULL) {
1090 ret = (*up->encap_rcv)(sk, skb);
1092 UDP_INC_STATS_BH(sock_net(sk),
1093 UDP_MIB_INDATAGRAMS,
1099 /* FALLTHROUGH -- it's a UDP Packet */
1103 * UDP-Lite specific tests, ignored on UDP sockets
1105 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1108 * MIB statistics other than incrementing the error count are
1109 * disabled for the following two types of errors: these depend
1110 * on the application settings, not on the functioning of the
1111 * protocol stack as such.
1113 * RFC 3828 here recommends (sec 3.3): "There should also be a
1114 * way ... to ... at least let the receiving application block
1115 * delivery of packets with coverage values less than a value
1116 * provided by the application."
1118 if (up->pcrlen == 0) { /* full coverage was set */
1119 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1120 "%d while full coverage %d requested\n",
1121 UDP_SKB_CB(skb)->cscov, skb->len);
1124 /* The next case involves violating the min. coverage requested
1125 * by the receiver. This is subtle: if receiver wants x and x is
1126 * greater than the buffersize/MTU then receiver will complain
1127 * that it wants x while sender emits packets of smaller size y.
1128 * Therefore the above ...()->partial_cov statement is essential.
1130 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1131 LIMIT_NETDEBUG(KERN_WARNING
1132 "UDPLITE: coverage %d too small, need min %d\n",
1133 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1138 if (sk->sk_filter) {
1139 if (udp_lib_checksum_complete(skb))
1146 if (!sock_owned_by_user(sk))
1147 rc = __udp_queue_rcv_skb(sk, skb);
1149 sk_add_backlog(sk, skb);
1155 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1161 * Multicasts and broadcasts go to each listener.
1163 * Note: called only from the BH handler context,
1164 * so we don't need to lock the hashes.
1166 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1168 __be32 saddr, __be32 daddr,
1169 struct udp_table *udptable)
1172 struct udp_hslot *hslot = &udptable->hash[udp_hashfn(net, ntohs(uh->dest))];
1175 spin_lock(&hslot->lock);
1176 sk = sk_nulls_head(&hslot->head);
1177 dif = skb->dev->ifindex;
1178 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1180 struct sock *sknext = NULL;
1183 struct sk_buff *skb1 = skb;
1185 sknext = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1186 daddr, uh->source, saddr,
1189 skb1 = skb_clone(skb, GFP_ATOMIC);
1192 int ret = udp_queue_rcv_skb(sk, skb1);
1194 /* we should probably re-process instead
1195 * of dropping packets here. */
1202 spin_unlock(&hslot->lock);
1206 /* Initialize UDP checksum. If exited with zero value (success),
1207 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1208 * Otherwise, csum completion requires chacksumming packet body,
1209 * including udp header and folding it to skb->csum.
1211 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1214 const struct iphdr *iph;
1217 UDP_SKB_CB(skb)->partial_cov = 0;
1218 UDP_SKB_CB(skb)->cscov = skb->len;
1220 if (proto == IPPROTO_UDPLITE) {
1221 err = udplite_checksum_init(skb, uh);
1227 if (uh->check == 0) {
1228 skb->ip_summed = CHECKSUM_UNNECESSARY;
1229 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1230 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1232 skb->ip_summed = CHECKSUM_UNNECESSARY;
1234 if (!skb_csum_unnecessary(skb))
1235 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1236 skb->len, proto, 0);
1237 /* Probably, we should checksum udp header (it should be in cache
1238 * in any case) and data in tiny packets (< rx copybreak).
1245 * All we need to do is get the socket, and then do a checksum.
1248 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1253 unsigned short ulen;
1254 struct rtable *rt = skb_rtable(skb);
1255 __be32 saddr, daddr;
1256 struct net *net = dev_net(skb->dev);
1259 * Validate the packet.
1261 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1262 goto drop; /* No space for header. */
1265 ulen = ntohs(uh->len);
1266 if (ulen > skb->len)
1269 if (proto == IPPROTO_UDP) {
1270 /* UDP validates ulen. */
1271 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1276 if (udp4_csum_init(skb, uh, proto))
1279 saddr = ip_hdr(skb)->saddr;
1280 daddr = ip_hdr(skb)->daddr;
1282 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1283 return __udp4_lib_mcast_deliver(net, skb, uh,
1284 saddr, daddr, udptable);
1286 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1289 int ret = udp_queue_rcv_skb(sk, skb);
1292 /* a return value > 0 means to resubmit the input, but
1293 * it wants the return to be -protocol, or 0
1300 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1304 /* No socket. Drop packet silently, if checksum is wrong */
1305 if (udp_lib_checksum_complete(skb))
1308 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1309 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1312 * Hmm. We got an UDP packet to a port to which we
1313 * don't wanna listen. Ignore it.
1319 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1320 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1331 * RFC1122: OK. Discards the bad packet silently (as far as
1332 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1334 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1335 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1342 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1347 int udp_rcv(struct sk_buff *skb)
1349 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1352 void udp_destroy_sock(struct sock *sk)
1355 udp_flush_pending_frames(sk);
1360 * Socket option code for UDP
1362 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1363 char __user *optval, unsigned int optlen,
1364 int (*push_pending_frames)(struct sock *))
1366 struct udp_sock *up = udp_sk(sk);
1369 int is_udplite = IS_UDPLITE(sk);
1371 if (optlen < sizeof(int))
1374 if (get_user(val, (int __user *)optval))
1384 (*push_pending_frames)(sk);
1392 case UDP_ENCAP_ESPINUDP:
1393 case UDP_ENCAP_ESPINUDP_NON_IKE:
1394 up->encap_rcv = xfrm4_udp_encap_rcv;
1396 case UDP_ENCAP_L2TPINUDP:
1397 up->encap_type = val;
1406 * UDP-Lite's partial checksum coverage (RFC 3828).
1408 /* The sender sets actual checksum coverage length via this option.
1409 * The case coverage > packet length is handled by send module. */
1410 case UDPLITE_SEND_CSCOV:
1411 if (!is_udplite) /* Disable the option on UDP sockets */
1412 return -ENOPROTOOPT;
1413 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1415 else if (val > USHORT_MAX)
1418 up->pcflag |= UDPLITE_SEND_CC;
1421 /* The receiver specifies a minimum checksum coverage value. To make
1422 * sense, this should be set to at least 8 (as done below). If zero is
1423 * used, this again means full checksum coverage. */
1424 case UDPLITE_RECV_CSCOV:
1425 if (!is_udplite) /* Disable the option on UDP sockets */
1426 return -ENOPROTOOPT;
1427 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1429 else if (val > USHORT_MAX)
1432 up->pcflag |= UDPLITE_RECV_CC;
1442 EXPORT_SYMBOL(udp_lib_setsockopt);
1444 int udp_setsockopt(struct sock *sk, int level, int optname,
1445 char __user *optval, unsigned int optlen)
1447 if (level == SOL_UDP || level == SOL_UDPLITE)
1448 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1449 udp_push_pending_frames);
1450 return ip_setsockopt(sk, level, optname, optval, optlen);
1453 #ifdef CONFIG_COMPAT
1454 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1455 char __user *optval, unsigned int optlen)
1457 if (level == SOL_UDP || level == SOL_UDPLITE)
1458 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1459 udp_push_pending_frames);
1460 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1464 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1465 char __user *optval, int __user *optlen)
1467 struct udp_sock *up = udp_sk(sk);
1470 if (get_user(len, optlen))
1473 len = min_t(unsigned int, len, sizeof(int));
1484 val = up->encap_type;
1487 /* The following two cannot be changed on UDP sockets, the return is
1488 * always 0 (which corresponds to the full checksum coverage of UDP). */
1489 case UDPLITE_SEND_CSCOV:
1493 case UDPLITE_RECV_CSCOV:
1498 return -ENOPROTOOPT;
1501 if (put_user(len, optlen))
1503 if (copy_to_user(optval, &val, len))
1507 EXPORT_SYMBOL(udp_lib_getsockopt);
1509 int udp_getsockopt(struct sock *sk, int level, int optname,
1510 char __user *optval, int __user *optlen)
1512 if (level == SOL_UDP || level == SOL_UDPLITE)
1513 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1514 return ip_getsockopt(sk, level, optname, optval, optlen);
1517 #ifdef CONFIG_COMPAT
1518 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1519 char __user *optval, int __user *optlen)
1521 if (level == SOL_UDP || level == SOL_UDPLITE)
1522 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1523 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1527 * udp_poll - wait for a UDP event.
1528 * @file - file struct
1530 * @wait - poll table
1532 * This is same as datagram poll, except for the special case of
1533 * blocking sockets. If application is using a blocking fd
1534 * and a packet with checksum error is in the queue;
1535 * then it could get return from select indicating data available
1536 * but then block when reading it. Add special case code
1537 * to work around these arguably broken applications.
1539 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1541 unsigned int mask = datagram_poll(file, sock, wait);
1542 struct sock *sk = sock->sk;
1543 int is_lite = IS_UDPLITE(sk);
1545 /* Check for false positives due to checksum errors */
1546 if ((mask & POLLRDNORM) &&
1547 !(file->f_flags & O_NONBLOCK) &&
1548 !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1549 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1550 struct sk_buff *skb;
1552 spin_lock_bh(&rcvq->lock);
1553 while ((skb = skb_peek(rcvq)) != NULL &&
1554 udp_lib_checksum_complete(skb)) {
1555 UDP_INC_STATS_BH(sock_net(sk),
1556 UDP_MIB_INERRORS, is_lite);
1557 __skb_unlink(skb, rcvq);
1560 spin_unlock_bh(&rcvq->lock);
1562 /* nothing to see, move along */
1564 mask &= ~(POLLIN | POLLRDNORM);
1570 EXPORT_SYMBOL(udp_poll);
1572 struct proto udp_prot = {
1574 .owner = THIS_MODULE,
1575 .close = udp_lib_close,
1576 .connect = ip4_datagram_connect,
1577 .disconnect = udp_disconnect,
1579 .destroy = udp_destroy_sock,
1580 .setsockopt = udp_setsockopt,
1581 .getsockopt = udp_getsockopt,
1582 .sendmsg = udp_sendmsg,
1583 .recvmsg = udp_recvmsg,
1584 .sendpage = udp_sendpage,
1585 .backlog_rcv = __udp_queue_rcv_skb,
1586 .hash = udp_lib_hash,
1587 .unhash = udp_lib_unhash,
1588 .get_port = udp_v4_get_port,
1589 .memory_allocated = &udp_memory_allocated,
1590 .sysctl_mem = sysctl_udp_mem,
1591 .sysctl_wmem = &sysctl_udp_wmem_min,
1592 .sysctl_rmem = &sysctl_udp_rmem_min,
1593 .obj_size = sizeof(struct udp_sock),
1594 .slab_flags = SLAB_DESTROY_BY_RCU,
1595 .h.udp_table = &udp_table,
1596 #ifdef CONFIG_COMPAT
1597 .compat_setsockopt = compat_udp_setsockopt,
1598 .compat_getsockopt = compat_udp_getsockopt,
1601 EXPORT_SYMBOL(udp_prot);
1603 /* ------------------------------------------------------------------------ */
1604 #ifdef CONFIG_PROC_FS
1606 static struct sock *udp_get_first(struct seq_file *seq, int start)
1609 struct udp_iter_state *state = seq->private;
1610 struct net *net = seq_file_net(seq);
1612 for (state->bucket = start; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1613 struct hlist_nulls_node *node;
1614 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1615 spin_lock_bh(&hslot->lock);
1616 sk_nulls_for_each(sk, node, &hslot->head) {
1617 if (!net_eq(sock_net(sk), net))
1619 if (sk->sk_family == state->family)
1622 spin_unlock_bh(&hslot->lock);
1629 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1631 struct udp_iter_state *state = seq->private;
1632 struct net *net = seq_file_net(seq);
1635 sk = sk_nulls_next(sk);
1636 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1639 if (state->bucket < UDP_HTABLE_SIZE)
1640 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1641 return udp_get_first(seq, state->bucket + 1);
1646 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1648 struct sock *sk = udp_get_first(seq, 0);
1651 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1653 return pos ? NULL : sk;
1656 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1658 struct udp_iter_state *state = seq->private;
1659 state->bucket = UDP_HTABLE_SIZE;
1661 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1664 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1668 if (v == SEQ_START_TOKEN)
1669 sk = udp_get_idx(seq, 0);
1671 sk = udp_get_next(seq, v);
1677 static void udp_seq_stop(struct seq_file *seq, void *v)
1679 struct udp_iter_state *state = seq->private;
1681 if (state->bucket < UDP_HTABLE_SIZE)
1682 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1685 static int udp_seq_open(struct inode *inode, struct file *file)
1687 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1688 struct udp_iter_state *s;
1691 err = seq_open_net(inode, file, &afinfo->seq_ops,
1692 sizeof(struct udp_iter_state));
1696 s = ((struct seq_file *)file->private_data)->private;
1697 s->family = afinfo->family;
1698 s->udp_table = afinfo->udp_table;
1702 /* ------------------------------------------------------------------------ */
1703 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1705 struct proc_dir_entry *p;
1708 afinfo->seq_fops.open = udp_seq_open;
1709 afinfo->seq_fops.read = seq_read;
1710 afinfo->seq_fops.llseek = seq_lseek;
1711 afinfo->seq_fops.release = seq_release_net;
1713 afinfo->seq_ops.start = udp_seq_start;
1714 afinfo->seq_ops.next = udp_seq_next;
1715 afinfo->seq_ops.stop = udp_seq_stop;
1717 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1718 &afinfo->seq_fops, afinfo);
1723 EXPORT_SYMBOL(udp_proc_register);
1725 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1727 proc_net_remove(net, afinfo->name);
1729 EXPORT_SYMBOL(udp_proc_unregister);
1731 /* ------------------------------------------------------------------------ */
1732 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1733 int bucket, int *len)
1735 struct inet_sock *inet = inet_sk(sp);
1736 __be32 dest = inet->daddr;
1737 __be32 src = inet->rcv_saddr;
1738 __u16 destp = ntohs(inet->dport);
1739 __u16 srcp = ntohs(inet->sport);
1741 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
1742 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1743 bucket, src, srcp, dest, destp, sp->sk_state,
1744 sk_wmem_alloc_get(sp),
1745 sk_rmem_alloc_get(sp),
1746 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1747 atomic_read(&sp->sk_refcnt), sp,
1748 atomic_read(&sp->sk_drops), len);
1751 int udp4_seq_show(struct seq_file *seq, void *v)
1753 if (v == SEQ_START_TOKEN)
1754 seq_printf(seq, "%-127s\n",
1755 " sl local_address rem_address st tx_queue "
1756 "rx_queue tr tm->when retrnsmt uid timeout "
1757 "inode ref pointer drops");
1759 struct udp_iter_state *state = seq->private;
1762 udp4_format_sock(v, seq, state->bucket, &len);
1763 seq_printf(seq, "%*s\n", 127 - len, "");
1768 /* ------------------------------------------------------------------------ */
1769 static struct udp_seq_afinfo udp4_seq_afinfo = {
1772 .udp_table = &udp_table,
1774 .owner = THIS_MODULE,
1777 .show = udp4_seq_show,
1781 static int udp4_proc_init_net(struct net *net)
1783 return udp_proc_register(net, &udp4_seq_afinfo);
1786 static void udp4_proc_exit_net(struct net *net)
1788 udp_proc_unregister(net, &udp4_seq_afinfo);
1791 static struct pernet_operations udp4_net_ops = {
1792 .init = udp4_proc_init_net,
1793 .exit = udp4_proc_exit_net,
1796 int __init udp4_proc_init(void)
1798 return register_pernet_subsys(&udp4_net_ops);
1801 void udp4_proc_exit(void)
1803 unregister_pernet_subsys(&udp4_net_ops);
1805 #endif /* CONFIG_PROC_FS */
1807 void __init udp_table_init(struct udp_table *table)
1811 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
1812 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
1813 spin_lock_init(&table->hash[i].lock);
1817 void __init udp_init(void)
1819 unsigned long nr_pages, limit;
1821 udp_table_init(&udp_table);
1822 /* Set the pressure threshold up by the same strategy of TCP. It is a
1823 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1824 * toward zero with the amount of memory, with a floor of 128 pages.
1826 nr_pages = totalram_pages - totalhigh_pages;
1827 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1828 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1829 limit = max(limit, 128UL);
1830 sysctl_udp_mem[0] = limit / 4 * 3;
1831 sysctl_udp_mem[1] = limit;
1832 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1834 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1835 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1838 int udp4_ufo_send_check(struct sk_buff *skb)
1840 const struct iphdr *iph;
1843 if (!pskb_may_pull(skb, sizeof(*uh)))
1849 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1851 skb->csum_start = skb_transport_header(skb) - skb->head;
1852 skb->csum_offset = offsetof(struct udphdr, check);
1853 skb->ip_summed = CHECKSUM_PARTIAL;
1857 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features)
1859 struct sk_buff *segs = ERR_PTR(-EINVAL);
1864 mss = skb_shinfo(skb)->gso_size;
1865 if (unlikely(skb->len <= mss))
1868 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1869 /* Packet is from an untrusted source, reset gso_segs. */
1870 int type = skb_shinfo(skb)->gso_type;
1872 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1873 !(type & (SKB_GSO_UDP))))
1876 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1882 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1883 * do checksum of UDP packets sent as multiple IP fragments.
1885 offset = skb->csum_start - skb_headroom(skb);
1886 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1887 offset += skb->csum_offset;
1888 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1889 skb->ip_summed = CHECKSUM_NONE;
1891 /* Fragment the skb. IP headers of the fragments are updated in
1892 * inet_gso_segment()
1894 segs = skb_segment(skb, features);