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 <linux/slab.h>
99 #include <net/tcp_states.h>
100 #include <linux/skbuff.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <net/net_namespace.h>
104 #include <net/icmp.h>
105 #include <net/route.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include "udp_impl.h"
110 struct udp_table udp_table __read_mostly;
111 EXPORT_SYMBOL(udp_table);
113 int sysctl_udp_mem[3] __read_mostly;
114 EXPORT_SYMBOL(sysctl_udp_mem);
116 int sysctl_udp_rmem_min __read_mostly;
117 EXPORT_SYMBOL(sysctl_udp_rmem_min);
119 int sysctl_udp_wmem_min __read_mostly;
120 EXPORT_SYMBOL(sysctl_udp_wmem_min);
122 atomic_t udp_memory_allocated;
123 EXPORT_SYMBOL(udp_memory_allocated);
125 #define MAX_UDP_PORTS 65536
126 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
128 static int udp_lib_lport_inuse(struct net *net, __u16 num,
129 const struct udp_hslot *hslot,
130 unsigned long *bitmap,
132 int (*saddr_comp)(const struct sock *sk1,
133 const struct sock *sk2),
137 struct hlist_nulls_node *node;
139 sk_nulls_for_each(sk2, node, &hslot->head)
140 if (net_eq(sock_net(sk2), net) &&
142 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
143 (!sk2->sk_reuse || !sk->sk_reuse) &&
144 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
145 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
146 (*saddr_comp)(sk, sk2)) {
148 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
157 * Note: we still hold spinlock of primary hash chain, so no other writer
158 * can insert/delete a socket with local_port == num
160 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
161 struct udp_hslot *hslot2,
163 int (*saddr_comp)(const struct sock *sk1,
164 const struct sock *sk2))
167 struct hlist_nulls_node *node;
170 spin_lock(&hslot2->lock);
171 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
172 if (net_eq(sock_net(sk2), net) &&
174 (udp_sk(sk2)->udp_port_hash == num) &&
175 (!sk2->sk_reuse || !sk->sk_reuse) &&
176 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
177 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
178 (*saddr_comp)(sk, sk2)) {
182 spin_unlock(&hslot2->lock);
187 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
189 * @sk: socket struct in question
190 * @snum: port number to look up
191 * @saddr_comp: AF-dependent comparison of bound local IP addresses
192 * @hash2_nulladdr: AF-dependant hash value in secondary hash chains,
195 int udp_lib_get_port(struct sock *sk, unsigned short snum,
196 int (*saddr_comp)(const struct sock *sk1,
197 const struct sock *sk2),
198 unsigned int hash2_nulladdr)
200 struct udp_hslot *hslot, *hslot2;
201 struct udp_table *udptable = sk->sk_prot->h.udp_table;
203 struct net *net = sock_net(sk);
206 int low, high, remaining;
208 unsigned short first, last;
209 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
211 inet_get_local_port_range(&low, &high);
212 remaining = (high - low) + 1;
215 first = (((u64)rand * remaining) >> 32) + low;
217 * force rand to be an odd multiple of UDP_HTABLE_SIZE
219 rand = (rand | 1) * (udptable->mask + 1);
220 last = first + udptable->mask + 1;
222 hslot = udp_hashslot(udptable, net, first);
223 bitmap_zero(bitmap, PORTS_PER_CHAIN);
224 spin_lock_bh(&hslot->lock);
225 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
226 saddr_comp, udptable->log);
230 * Iterate on all possible values of snum for this hash.
231 * Using steps of an odd multiple of UDP_HTABLE_SIZE
232 * give us randomization and full range coverage.
235 if (low <= snum && snum <= high &&
236 !test_bit(snum >> udptable->log, bitmap))
239 } while (snum != first);
240 spin_unlock_bh(&hslot->lock);
241 } while (++first != last);
244 hslot = udp_hashslot(udptable, net, snum);
245 spin_lock_bh(&hslot->lock);
246 if (hslot->count > 10) {
248 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
250 slot2 &= udptable->mask;
251 hash2_nulladdr &= udptable->mask;
253 hslot2 = udp_hashslot2(udptable, slot2);
254 if (hslot->count < hslot2->count)
255 goto scan_primary_hash;
257 exist = udp_lib_lport_inuse2(net, snum, hslot2,
259 if (!exist && (hash2_nulladdr != slot2)) {
260 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
261 exist = udp_lib_lport_inuse2(net, snum, hslot2,
270 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
275 inet_sk(sk)->inet_num = snum;
276 udp_sk(sk)->udp_port_hash = snum;
277 udp_sk(sk)->udp_portaddr_hash ^= snum;
278 if (sk_unhashed(sk)) {
279 sk_nulls_add_node_rcu(sk, &hslot->head);
281 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
283 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
284 spin_lock(&hslot2->lock);
285 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
288 spin_unlock(&hslot2->lock);
292 spin_unlock_bh(&hslot->lock);
296 EXPORT_SYMBOL(udp_lib_get_port);
298 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
300 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
302 return (!ipv6_only_sock(sk2) &&
303 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
304 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
307 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
310 return jhash_1word(saddr, net_hash_mix(net)) ^ port;
313 int udp_v4_get_port(struct sock *sk, unsigned short snum)
315 unsigned int hash2_nulladdr =
316 udp4_portaddr_hash(sock_net(sk), INADDR_ANY, snum);
317 unsigned int hash2_partial =
318 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
320 /* precompute partial secondary hash */
321 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
322 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
325 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
327 __be16 sport, __be32 daddr, __be16 dport, int dif)
331 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
332 !ipv6_only_sock(sk)) {
333 struct inet_sock *inet = inet_sk(sk);
335 score = (sk->sk_family == PF_INET ? 1 : 0);
336 if (inet->inet_rcv_saddr) {
337 if (inet->inet_rcv_saddr != daddr)
341 if (inet->inet_daddr) {
342 if (inet->inet_daddr != saddr)
346 if (inet->inet_dport) {
347 if (inet->inet_dport != sport)
351 if (sk->sk_bound_dev_if) {
352 if (sk->sk_bound_dev_if != dif)
361 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
363 #define SCORE2_MAX (1 + 2 + 2 + 2)
364 static inline int compute_score2(struct sock *sk, struct net *net,
365 __be32 saddr, __be16 sport,
366 __be32 daddr, unsigned int hnum, int dif)
370 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
371 struct inet_sock *inet = inet_sk(sk);
373 if (inet->inet_rcv_saddr != daddr)
375 if (inet->inet_num != hnum)
378 score = (sk->sk_family == PF_INET ? 1 : 0);
379 if (inet->inet_daddr) {
380 if (inet->inet_daddr != saddr)
384 if (inet->inet_dport) {
385 if (inet->inet_dport != sport)
389 if (sk->sk_bound_dev_if) {
390 if (sk->sk_bound_dev_if != dif)
399 /* called with read_rcu_lock() */
400 static struct sock *udp4_lib_lookup2(struct net *net,
401 __be32 saddr, __be16 sport,
402 __be32 daddr, unsigned int hnum, int dif,
403 struct udp_hslot *hslot2, unsigned int slot2)
405 struct sock *sk, *result;
406 struct hlist_nulls_node *node;
412 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
413 score = compute_score2(sk, net, saddr, sport,
415 if (score > badness) {
418 if (score == SCORE2_MAX)
423 * if the nulls value we got at the end of this lookup is
424 * not the expected one, we must restart lookup.
425 * We probably met an item that was moved to another chain.
427 if (get_nulls_value(node) != slot2)
432 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
434 else if (unlikely(compute_score2(result, net, saddr, sport,
435 daddr, hnum, dif) < badness)) {
443 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
444 * harder than this. -DaveM
446 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
447 __be16 sport, __be32 daddr, __be16 dport,
448 int dif, struct udp_table *udptable)
450 struct sock *sk, *result;
451 struct hlist_nulls_node *node;
452 unsigned short hnum = ntohs(dport);
453 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
454 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
458 if (hslot->count > 10) {
459 hash2 = udp4_portaddr_hash(net, daddr, hnum);
460 slot2 = hash2 & udptable->mask;
461 hslot2 = &udptable->hash2[slot2];
462 if (hslot->count < hslot2->count)
465 result = udp4_lib_lookup2(net, saddr, sport,
469 hash2 = udp4_portaddr_hash(net, INADDR_ANY, hnum);
470 slot2 = hash2 & udptable->mask;
471 hslot2 = &udptable->hash2[slot2];
472 if (hslot->count < hslot2->count)
475 result = udp4_lib_lookup2(net, saddr, sport,
476 INADDR_ANY, hnum, dif,
485 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
486 score = compute_score(sk, net, saddr, hnum, sport,
488 if (score > badness) {
494 * if the nulls value we got at the end of this lookup is
495 * not the expected one, we must restart lookup.
496 * We probably met an item that was moved to another chain.
498 if (get_nulls_value(node) != slot)
502 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
504 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
505 daddr, dport, dif) < badness)) {
514 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
515 __be16 sport, __be16 dport,
516 struct udp_table *udptable)
519 const struct iphdr *iph = ip_hdr(skb);
521 if (unlikely(sk = skb_steal_sock(skb)))
524 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
525 iph->daddr, dport, inet_iif(skb),
529 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
530 __be32 daddr, __be16 dport, int dif)
532 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
534 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
536 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
537 __be16 loc_port, __be32 loc_addr,
538 __be16 rmt_port, __be32 rmt_addr,
541 struct hlist_nulls_node *node;
543 unsigned short hnum = ntohs(loc_port);
545 sk_nulls_for_each_from(s, node) {
546 struct inet_sock *inet = inet_sk(s);
548 if (!net_eq(sock_net(s), net) ||
549 udp_sk(s)->udp_port_hash != hnum ||
550 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
551 (inet->inet_dport != rmt_port && inet->inet_dport) ||
552 (inet->inet_rcv_saddr &&
553 inet->inet_rcv_saddr != loc_addr) ||
555 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
557 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
567 * This routine is called by the ICMP module when it gets some
568 * sort of error condition. If err < 0 then the socket should
569 * be closed and the error returned to the user. If err > 0
570 * it's just the icmp type << 8 | icmp code.
571 * Header points to the ip header of the error packet. We move
572 * on past this. Then (as it used to claim before adjustment)
573 * header points to the first 8 bytes of the udp header. We need
574 * to find the appropriate port.
577 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
579 struct inet_sock *inet;
580 struct iphdr *iph = (struct iphdr *)skb->data;
581 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
582 const int type = icmp_hdr(skb)->type;
583 const int code = icmp_hdr(skb)->code;
587 struct net *net = dev_net(skb->dev);
589 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
590 iph->saddr, uh->source, skb->dev->ifindex, udptable);
592 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
593 return; /* No socket for error */
602 case ICMP_TIME_EXCEEDED:
605 case ICMP_SOURCE_QUENCH:
607 case ICMP_PARAMETERPROB:
611 case ICMP_DEST_UNREACH:
612 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
613 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
621 if (code <= NR_ICMP_UNREACH) {
622 harderr = icmp_err_convert[code].fatal;
623 err = icmp_err_convert[code].errno;
629 * RFC1122: OK. Passes ICMP errors back to application, as per
632 if (!inet->recverr) {
633 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
636 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
639 sk->sk_error_report(sk);
644 void udp_err(struct sk_buff *skb, u32 info)
646 __udp4_lib_err(skb, info, &udp_table);
650 * Throw away all pending data and cancel the corking. Socket is locked.
652 void udp_flush_pending_frames(struct sock *sk)
654 struct udp_sock *up = udp_sk(sk);
659 ip_flush_pending_frames(sk);
662 EXPORT_SYMBOL(udp_flush_pending_frames);
665 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
666 * @sk: socket we are sending on
667 * @skb: sk_buff containing the filled-in UDP header
668 * (checksum field must be zeroed out)
670 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
671 __be32 src, __be32 dst, int len)
674 struct udphdr *uh = udp_hdr(skb);
677 if (skb_queue_len(&sk->sk_write_queue) == 1) {
679 * Only one fragment on the socket.
681 skb->csum_start = skb_transport_header(skb) - skb->head;
682 skb->csum_offset = offsetof(struct udphdr, check);
683 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
686 * HW-checksum won't work as there are two or more
687 * fragments on the socket so that all csums of sk_buffs
690 offset = skb_transport_offset(skb);
691 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
693 skb->ip_summed = CHECKSUM_NONE;
695 skb_queue_walk(&sk->sk_write_queue, skb) {
696 csum = csum_add(csum, skb->csum);
699 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
701 uh->check = CSUM_MANGLED_0;
706 * Push out all pending data as one UDP datagram. Socket is locked.
708 static int udp_push_pending_frames(struct sock *sk)
710 struct udp_sock *up = udp_sk(sk);
711 struct inet_sock *inet = inet_sk(sk);
712 struct flowi *fl = &inet->cork.fl;
716 int is_udplite = IS_UDPLITE(sk);
719 /* Grab the skbuff where UDP header space exists. */
720 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
724 * Create a UDP header
727 uh->source = fl->fl_ip_sport;
728 uh->dest = fl->fl_ip_dport;
729 uh->len = htons(up->len);
732 if (is_udplite) /* UDP-Lite */
733 csum = udplite_csum_outgoing(sk, skb);
735 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
737 skb->ip_summed = CHECKSUM_NONE;
740 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
742 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len);
745 } else /* `normal' UDP */
746 csum = udp_csum_outgoing(sk, skb);
748 /* add protocol-dependent pseudo-header */
749 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
750 sk->sk_protocol, csum);
752 uh->check = CSUM_MANGLED_0;
755 err = ip_push_pending_frames(sk);
757 if (err == -ENOBUFS && !inet->recverr) {
758 UDP_INC_STATS_USER(sock_net(sk),
759 UDP_MIB_SNDBUFERRORS, is_udplite);
763 UDP_INC_STATS_USER(sock_net(sk),
764 UDP_MIB_OUTDATAGRAMS, is_udplite);
771 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
774 struct inet_sock *inet = inet_sk(sk);
775 struct udp_sock *up = udp_sk(sk);
777 struct ipcm_cookie ipc;
778 struct rtable *rt = NULL;
781 __be32 daddr, faddr, saddr;
784 int err, is_udplite = IS_UDPLITE(sk);
785 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
786 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
795 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
803 * There are pending frames.
804 * The socket lock must be held while it's corked.
807 if (likely(up->pending)) {
808 if (unlikely(up->pending != AF_INET)) {
816 ulen += sizeof(struct udphdr);
819 * Get and verify the address.
822 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
823 if (msg->msg_namelen < sizeof(*usin))
825 if (usin->sin_family != AF_INET) {
826 if (usin->sin_family != AF_UNSPEC)
827 return -EAFNOSUPPORT;
830 daddr = usin->sin_addr.s_addr;
831 dport = usin->sin_port;
835 if (sk->sk_state != TCP_ESTABLISHED)
836 return -EDESTADDRREQ;
837 daddr = inet->inet_daddr;
838 dport = inet->inet_dport;
839 /* Open fast path for connected socket.
840 Route will not be used, if at least one option is set.
844 ipc.addr = inet->inet_saddr;
846 ipc.oif = sk->sk_bound_dev_if;
847 err = sock_tx_timestamp(msg, sk, &ipc.shtx);
850 if (msg->msg_controllen) {
851 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
862 ipc.addr = faddr = daddr;
864 if (ipc.opt && ipc.opt->srr) {
867 faddr = ipc.opt->faddr;
870 tos = RT_TOS(inet->tos);
871 if (sock_flag(sk, SOCK_LOCALROUTE) ||
872 (msg->msg_flags & MSG_DONTROUTE) ||
873 (ipc.opt && ipc.opt->is_strictroute)) {
878 if (ipv4_is_multicast(daddr)) {
880 ipc.oif = inet->mc_index;
882 saddr = inet->mc_addr;
887 rt = (struct rtable *)sk_dst_check(sk, 0);
890 struct flowi fl = { .oif = ipc.oif,
896 .proto = sk->sk_protocol,
897 .flags = inet_sk_flowi_flags(sk),
899 { .sport = inet->inet_sport,
900 .dport = dport } } };
901 struct net *net = sock_net(sk);
903 security_sk_classify_flow(sk, &fl);
904 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
906 if (err == -ENETUNREACH)
907 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
912 if ((rt->rt_flags & RTCF_BROADCAST) &&
913 !sock_flag(sk, SOCK_BROADCAST))
916 sk_dst_set(sk, dst_clone(&rt->u.dst));
919 if (msg->msg_flags&MSG_CONFIRM)
925 daddr = ipc.addr = rt->rt_dst;
928 if (unlikely(up->pending)) {
929 /* The socket is already corked while preparing it. */
930 /* ... which is an evident application bug. --ANK */
933 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
938 * Now cork the socket to pend data.
940 inet->cork.fl.fl4_dst = daddr;
941 inet->cork.fl.fl_ip_dport = dport;
942 inet->cork.fl.fl4_src = saddr;
943 inet->cork.fl.fl_ip_sport = inet->inet_sport;
944 up->pending = AF_INET;
948 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
949 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
950 sizeof(struct udphdr), &ipc, &rt,
951 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
953 udp_flush_pending_frames(sk);
955 err = udp_push_pending_frames(sk);
956 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
967 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
968 * ENOBUFS might not be good (it's not tunable per se), but otherwise
969 * we don't have a good statistic (IpOutDiscards but it can be too many
970 * things). We could add another new stat but at least for now that
971 * seems like overkill.
973 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
974 UDP_INC_STATS_USER(sock_net(sk),
975 UDP_MIB_SNDBUFERRORS, is_udplite);
980 dst_confirm(&rt->u.dst);
981 if (!(msg->msg_flags&MSG_PROBE) || len)
982 goto back_from_confirm;
986 EXPORT_SYMBOL(udp_sendmsg);
988 int udp_sendpage(struct sock *sk, struct page *page, int offset,
989 size_t size, int flags)
991 struct udp_sock *up = udp_sk(sk);
995 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
997 /* Call udp_sendmsg to specify destination address which
998 * sendpage interface can't pass.
999 * This will succeed only when the socket is connected.
1001 ret = udp_sendmsg(NULL, sk, &msg, 0);
1008 if (unlikely(!up->pending)) {
1011 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
1015 ret = ip_append_page(sk, page, offset, size, flags);
1016 if (ret == -EOPNOTSUPP) {
1018 return sock_no_sendpage(sk->sk_socket, page, offset,
1022 udp_flush_pending_frames(sk);
1027 if (!(up->corkflag || (flags&MSG_MORE)))
1028 ret = udp_push_pending_frames(sk);
1038 * first_packet_length - return length of first packet in receive queue
1041 * Drops all bad checksum frames, until a valid one is found.
1042 * Returns the length of found skb, or 0 if none is found.
1044 static unsigned int first_packet_length(struct sock *sk)
1046 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1047 struct sk_buff *skb;
1050 __skb_queue_head_init(&list_kill);
1052 spin_lock_bh(&rcvq->lock);
1053 while ((skb = skb_peek(rcvq)) != NULL &&
1054 udp_lib_checksum_complete(skb)) {
1055 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1057 atomic_inc(&sk->sk_drops);
1058 __skb_unlink(skb, rcvq);
1059 __skb_queue_tail(&list_kill, skb);
1061 res = skb ? skb->len : 0;
1062 spin_unlock_bh(&rcvq->lock);
1064 if (!skb_queue_empty(&list_kill)) {
1066 __skb_queue_purge(&list_kill);
1067 sk_mem_reclaim_partial(sk);
1074 * IOCTL requests applicable to the UDP protocol
1077 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1082 int amount = sk_wmem_alloc_get(sk);
1084 return put_user(amount, (int __user *)arg);
1089 unsigned int amount = first_packet_length(sk);
1093 * We will only return the amount
1094 * of this packet since that is all
1095 * that will be read.
1097 amount -= sizeof(struct udphdr);
1099 return put_user(amount, (int __user *)arg);
1103 return -ENOIOCTLCMD;
1108 EXPORT_SYMBOL(udp_ioctl);
1111 * This should be easy, if there is something there we
1112 * return it, otherwise we block.
1115 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1116 size_t len, int noblock, int flags, int *addr_len)
1118 struct inet_sock *inet = inet_sk(sk);
1119 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1120 struct sk_buff *skb;
1124 int is_udplite = IS_UDPLITE(sk);
1127 * Check any passed addresses
1130 *addr_len = sizeof(*sin);
1132 if (flags & MSG_ERRQUEUE)
1133 return ip_recv_error(sk, msg, len);
1136 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1141 ulen = skb->len - sizeof(struct udphdr);
1144 else if (len < ulen)
1145 msg->msg_flags |= MSG_TRUNC;
1148 * If checksum is needed at all, try to do it while copying the
1149 * data. If the data is truncated, or if we only want a partial
1150 * coverage checksum (UDP-Lite), do it before the copy.
1153 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
1154 if (udp_lib_checksum_complete(skb))
1158 if (skb_csum_unnecessary(skb))
1159 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1162 err = skb_copy_and_csum_datagram_iovec(skb,
1163 sizeof(struct udphdr),
1174 UDP_INC_STATS_USER(sock_net(sk),
1175 UDP_MIB_INDATAGRAMS, is_udplite);
1177 sock_recv_ts_and_drops(msg, sk, skb);
1179 /* Copy the address. */
1181 sin->sin_family = AF_INET;
1182 sin->sin_port = udp_hdr(skb)->source;
1183 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1184 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1186 if (inet->cmsg_flags)
1187 ip_cmsg_recv(msg, skb);
1190 if (flags & MSG_TRUNC)
1194 skb_free_datagram_locked(sk, skb);
1200 if (!skb_kill_datagram(sk, skb, flags))
1201 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1210 int udp_disconnect(struct sock *sk, int flags)
1212 struct inet_sock *inet = inet_sk(sk);
1214 * 1003.1g - break association.
1217 sk->sk_state = TCP_CLOSE;
1218 inet->inet_daddr = 0;
1219 inet->inet_dport = 0;
1220 sk->sk_bound_dev_if = 0;
1221 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1222 inet_reset_saddr(sk);
1224 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1225 sk->sk_prot->unhash(sk);
1226 inet->inet_sport = 0;
1231 EXPORT_SYMBOL(udp_disconnect);
1233 void udp_lib_unhash(struct sock *sk)
1235 if (sk_hashed(sk)) {
1236 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1237 struct udp_hslot *hslot, *hslot2;
1239 hslot = udp_hashslot(udptable, sock_net(sk),
1240 udp_sk(sk)->udp_port_hash);
1241 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1243 spin_lock_bh(&hslot->lock);
1244 if (sk_nulls_del_node_init_rcu(sk)) {
1246 inet_sk(sk)->inet_num = 0;
1247 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1249 spin_lock(&hslot2->lock);
1250 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1252 spin_unlock(&hslot2->lock);
1254 spin_unlock_bh(&hslot->lock);
1257 EXPORT_SYMBOL(udp_lib_unhash);
1259 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1261 int rc = sock_queue_rcv_skb(sk, skb);
1264 int is_udplite = IS_UDPLITE(sk);
1266 /* Note that an ENOMEM error is charged twice */
1268 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1270 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1282 * >0: "udp encap" protocol resubmission
1284 * Note that in the success and error cases, the skb is assumed to
1285 * have either been requeued or freed.
1287 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1289 struct udp_sock *up = udp_sk(sk);
1291 int is_udplite = IS_UDPLITE(sk);
1294 * Charge it to the socket, dropping if the queue is full.
1296 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1300 if (up->encap_type) {
1302 * This is an encapsulation socket so pass the skb to
1303 * the socket's udp_encap_rcv() hook. Otherwise, just
1304 * fall through and pass this up the UDP socket.
1305 * up->encap_rcv() returns the following value:
1306 * =0 if skb was successfully passed to the encap
1307 * handler or was discarded by it.
1308 * >0 if skb should be passed on to UDP.
1309 * <0 if skb should be resubmitted as proto -N
1312 /* if we're overly short, let UDP handle it */
1313 if (skb->len > sizeof(struct udphdr) &&
1314 up->encap_rcv != NULL) {
1317 ret = (*up->encap_rcv)(sk, skb);
1319 UDP_INC_STATS_BH(sock_net(sk),
1320 UDP_MIB_INDATAGRAMS,
1326 /* FALLTHROUGH -- it's a UDP Packet */
1330 * UDP-Lite specific tests, ignored on UDP sockets
1332 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1335 * MIB statistics other than incrementing the error count are
1336 * disabled for the following two types of errors: these depend
1337 * on the application settings, not on the functioning of the
1338 * protocol stack as such.
1340 * RFC 3828 here recommends (sec 3.3): "There should also be a
1341 * way ... to ... at least let the receiving application block
1342 * delivery of packets with coverage values less than a value
1343 * provided by the application."
1345 if (up->pcrlen == 0) { /* full coverage was set */
1346 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1347 "%d while full coverage %d requested\n",
1348 UDP_SKB_CB(skb)->cscov, skb->len);
1351 /* The next case involves violating the min. coverage requested
1352 * by the receiver. This is subtle: if receiver wants x and x is
1353 * greater than the buffersize/MTU then receiver will complain
1354 * that it wants x while sender emits packets of smaller size y.
1355 * Therefore the above ...()->partial_cov statement is essential.
1357 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1358 LIMIT_NETDEBUG(KERN_WARNING
1359 "UDPLITE: coverage %d too small, need min %d\n",
1360 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1365 if (sk->sk_filter) {
1366 if (udp_lib_checksum_complete(skb))
1373 if (!sock_owned_by_user(sk))
1374 rc = __udp_queue_rcv_skb(sk, skb);
1375 else if (sk_add_backlog(sk, skb)) {
1384 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1385 atomic_inc(&sk->sk_drops);
1391 static void flush_stack(struct sock **stack, unsigned int count,
1392 struct sk_buff *skb, unsigned int final)
1395 struct sk_buff *skb1 = NULL;
1398 for (i = 0; i < count; i++) {
1400 if (likely(skb1 == NULL))
1401 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1404 atomic_inc(&sk->sk_drops);
1405 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1407 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1411 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1419 * Multicasts and broadcasts go to each listener.
1421 * Note: called only from the BH handler context.
1423 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1425 __be32 saddr, __be32 daddr,
1426 struct udp_table *udptable)
1428 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1429 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1431 unsigned int i, count = 0;
1433 spin_lock(&hslot->lock);
1434 sk = sk_nulls_head(&hslot->head);
1435 dif = skb->dev->ifindex;
1436 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1438 stack[count++] = sk;
1439 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1440 daddr, uh->source, saddr, dif);
1441 if (unlikely(count == ARRAY_SIZE(stack))) {
1444 flush_stack(stack, count, skb, ~0);
1449 * before releasing chain lock, we must take a reference on sockets
1451 for (i = 0; i < count; i++)
1452 sock_hold(stack[i]);
1454 spin_unlock(&hslot->lock);
1457 * do the slow work with no lock held
1460 flush_stack(stack, count, skb, count - 1);
1462 for (i = 0; i < count; i++)
1470 /* Initialize UDP checksum. If exited with zero value (success),
1471 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1472 * Otherwise, csum completion requires chacksumming packet body,
1473 * including udp header and folding it to skb->csum.
1475 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1478 const struct iphdr *iph;
1481 UDP_SKB_CB(skb)->partial_cov = 0;
1482 UDP_SKB_CB(skb)->cscov = skb->len;
1484 if (proto == IPPROTO_UDPLITE) {
1485 err = udplite_checksum_init(skb, uh);
1491 if (uh->check == 0) {
1492 skb->ip_summed = CHECKSUM_UNNECESSARY;
1493 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1494 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1496 skb->ip_summed = CHECKSUM_UNNECESSARY;
1498 if (!skb_csum_unnecessary(skb))
1499 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1500 skb->len, proto, 0);
1501 /* Probably, we should checksum udp header (it should be in cache
1502 * in any case) and data in tiny packets (< rx copybreak).
1509 * All we need to do is get the socket, and then do a checksum.
1512 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1517 unsigned short ulen;
1518 struct rtable *rt = skb_rtable(skb);
1519 __be32 saddr, daddr;
1520 struct net *net = dev_net(skb->dev);
1523 * Validate the packet.
1525 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1526 goto drop; /* No space for header. */
1529 ulen = ntohs(uh->len);
1530 saddr = ip_hdr(skb)->saddr;
1531 daddr = ip_hdr(skb)->daddr;
1533 if (ulen > skb->len)
1536 if (proto == IPPROTO_UDP) {
1537 /* UDP validates ulen. */
1538 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1543 if (udp4_csum_init(skb, uh, proto))
1546 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1547 return __udp4_lib_mcast_deliver(net, skb, uh,
1548 saddr, daddr, udptable);
1550 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1553 int ret = udp_queue_rcv_skb(sk, skb);
1556 /* a return value > 0 means to resubmit the input, but
1557 * it wants the return to be -protocol, or 0
1564 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1568 /* No socket. Drop packet silently, if checksum is wrong */
1569 if (udp_lib_checksum_complete(skb))
1572 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1573 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1576 * Hmm. We got an UDP packet to a port to which we
1577 * don't wanna listen. Ignore it.
1583 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1584 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1595 * RFC1122: OK. Discards the bad packet silently (as far as
1596 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1598 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1599 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1606 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1611 int udp_rcv(struct sk_buff *skb)
1613 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1616 void udp_destroy_sock(struct sock *sk)
1619 udp_flush_pending_frames(sk);
1624 * Socket option code for UDP
1626 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1627 char __user *optval, unsigned int optlen,
1628 int (*push_pending_frames)(struct sock *))
1630 struct udp_sock *up = udp_sk(sk);
1633 int is_udplite = IS_UDPLITE(sk);
1635 if (optlen < sizeof(int))
1638 if (get_user(val, (int __user *)optval))
1648 (*push_pending_frames)(sk);
1656 case UDP_ENCAP_ESPINUDP:
1657 case UDP_ENCAP_ESPINUDP_NON_IKE:
1658 up->encap_rcv = xfrm4_udp_encap_rcv;
1660 case UDP_ENCAP_L2TPINUDP:
1661 up->encap_type = val;
1670 * UDP-Lite's partial checksum coverage (RFC 3828).
1672 /* The sender sets actual checksum coverage length via this option.
1673 * The case coverage > packet length is handled by send module. */
1674 case UDPLITE_SEND_CSCOV:
1675 if (!is_udplite) /* Disable the option on UDP sockets */
1676 return -ENOPROTOOPT;
1677 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1679 else if (val > USHORT_MAX)
1682 up->pcflag |= UDPLITE_SEND_CC;
1685 /* The receiver specifies a minimum checksum coverage value. To make
1686 * sense, this should be set to at least 8 (as done below). If zero is
1687 * used, this again means full checksum coverage. */
1688 case UDPLITE_RECV_CSCOV:
1689 if (!is_udplite) /* Disable the option on UDP sockets */
1690 return -ENOPROTOOPT;
1691 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1693 else if (val > USHORT_MAX)
1696 up->pcflag |= UDPLITE_RECV_CC;
1706 EXPORT_SYMBOL(udp_lib_setsockopt);
1708 int udp_setsockopt(struct sock *sk, int level, int optname,
1709 char __user *optval, unsigned int optlen)
1711 if (level == SOL_UDP || level == SOL_UDPLITE)
1712 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1713 udp_push_pending_frames);
1714 return ip_setsockopt(sk, level, optname, optval, optlen);
1717 #ifdef CONFIG_COMPAT
1718 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1719 char __user *optval, unsigned int optlen)
1721 if (level == SOL_UDP || level == SOL_UDPLITE)
1722 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1723 udp_push_pending_frames);
1724 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1728 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1729 char __user *optval, int __user *optlen)
1731 struct udp_sock *up = udp_sk(sk);
1734 if (get_user(len, optlen))
1737 len = min_t(unsigned int, len, sizeof(int));
1748 val = up->encap_type;
1751 /* The following two cannot be changed on UDP sockets, the return is
1752 * always 0 (which corresponds to the full checksum coverage of UDP). */
1753 case UDPLITE_SEND_CSCOV:
1757 case UDPLITE_RECV_CSCOV:
1762 return -ENOPROTOOPT;
1765 if (put_user(len, optlen))
1767 if (copy_to_user(optval, &val, len))
1771 EXPORT_SYMBOL(udp_lib_getsockopt);
1773 int udp_getsockopt(struct sock *sk, int level, int optname,
1774 char __user *optval, int __user *optlen)
1776 if (level == SOL_UDP || level == SOL_UDPLITE)
1777 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1778 return ip_getsockopt(sk, level, optname, optval, optlen);
1781 #ifdef CONFIG_COMPAT
1782 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1783 char __user *optval, int __user *optlen)
1785 if (level == SOL_UDP || level == SOL_UDPLITE)
1786 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1787 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1791 * udp_poll - wait for a UDP event.
1792 * @file - file struct
1794 * @wait - poll table
1796 * This is same as datagram poll, except for the special case of
1797 * blocking sockets. If application is using a blocking fd
1798 * and a packet with checksum error is in the queue;
1799 * then it could get return from select indicating data available
1800 * but then block when reading it. Add special case code
1801 * to work around these arguably broken applications.
1803 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1805 unsigned int mask = datagram_poll(file, sock, wait);
1806 struct sock *sk = sock->sk;
1808 /* Check for false positives due to checksum errors */
1809 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1810 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1811 mask &= ~(POLLIN | POLLRDNORM);
1816 EXPORT_SYMBOL(udp_poll);
1818 struct proto udp_prot = {
1820 .owner = THIS_MODULE,
1821 .close = udp_lib_close,
1822 .connect = ip4_datagram_connect,
1823 .disconnect = udp_disconnect,
1825 .destroy = udp_destroy_sock,
1826 .setsockopt = udp_setsockopt,
1827 .getsockopt = udp_getsockopt,
1828 .sendmsg = udp_sendmsg,
1829 .recvmsg = udp_recvmsg,
1830 .sendpage = udp_sendpage,
1831 .backlog_rcv = __udp_queue_rcv_skb,
1832 .hash = udp_lib_hash,
1833 .unhash = udp_lib_unhash,
1834 .get_port = udp_v4_get_port,
1835 .memory_allocated = &udp_memory_allocated,
1836 .sysctl_mem = sysctl_udp_mem,
1837 .sysctl_wmem = &sysctl_udp_wmem_min,
1838 .sysctl_rmem = &sysctl_udp_rmem_min,
1839 .obj_size = sizeof(struct udp_sock),
1840 .slab_flags = SLAB_DESTROY_BY_RCU,
1841 .h.udp_table = &udp_table,
1842 #ifdef CONFIG_COMPAT
1843 .compat_setsockopt = compat_udp_setsockopt,
1844 .compat_getsockopt = compat_udp_getsockopt,
1847 EXPORT_SYMBOL(udp_prot);
1849 /* ------------------------------------------------------------------------ */
1850 #ifdef CONFIG_PROC_FS
1852 static struct sock *udp_get_first(struct seq_file *seq, int start)
1855 struct udp_iter_state *state = seq->private;
1856 struct net *net = seq_file_net(seq);
1858 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1860 struct hlist_nulls_node *node;
1861 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1863 if (hlist_nulls_empty(&hslot->head))
1866 spin_lock_bh(&hslot->lock);
1867 sk_nulls_for_each(sk, node, &hslot->head) {
1868 if (!net_eq(sock_net(sk), net))
1870 if (sk->sk_family == state->family)
1873 spin_unlock_bh(&hslot->lock);
1880 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1882 struct udp_iter_state *state = seq->private;
1883 struct net *net = seq_file_net(seq);
1886 sk = sk_nulls_next(sk);
1887 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1890 if (state->bucket <= state->udp_table->mask)
1891 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1892 return udp_get_first(seq, state->bucket + 1);
1897 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1899 struct sock *sk = udp_get_first(seq, 0);
1902 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1904 return pos ? NULL : sk;
1907 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1909 struct udp_iter_state *state = seq->private;
1910 state->bucket = MAX_UDP_PORTS;
1912 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1915 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1919 if (v == SEQ_START_TOKEN)
1920 sk = udp_get_idx(seq, 0);
1922 sk = udp_get_next(seq, v);
1928 static void udp_seq_stop(struct seq_file *seq, void *v)
1930 struct udp_iter_state *state = seq->private;
1932 if (state->bucket <= state->udp_table->mask)
1933 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1936 static int udp_seq_open(struct inode *inode, struct file *file)
1938 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1939 struct udp_iter_state *s;
1942 err = seq_open_net(inode, file, &afinfo->seq_ops,
1943 sizeof(struct udp_iter_state));
1947 s = ((struct seq_file *)file->private_data)->private;
1948 s->family = afinfo->family;
1949 s->udp_table = afinfo->udp_table;
1953 /* ------------------------------------------------------------------------ */
1954 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1956 struct proc_dir_entry *p;
1959 afinfo->seq_fops.open = udp_seq_open;
1960 afinfo->seq_fops.read = seq_read;
1961 afinfo->seq_fops.llseek = seq_lseek;
1962 afinfo->seq_fops.release = seq_release_net;
1964 afinfo->seq_ops.start = udp_seq_start;
1965 afinfo->seq_ops.next = udp_seq_next;
1966 afinfo->seq_ops.stop = udp_seq_stop;
1968 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1969 &afinfo->seq_fops, afinfo);
1974 EXPORT_SYMBOL(udp_proc_register);
1976 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1978 proc_net_remove(net, afinfo->name);
1980 EXPORT_SYMBOL(udp_proc_unregister);
1982 /* ------------------------------------------------------------------------ */
1983 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1984 int bucket, int *len)
1986 struct inet_sock *inet = inet_sk(sp);
1987 __be32 dest = inet->inet_daddr;
1988 __be32 src = inet->inet_rcv_saddr;
1989 __u16 destp = ntohs(inet->inet_dport);
1990 __u16 srcp = ntohs(inet->inet_sport);
1992 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
1993 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1994 bucket, src, srcp, dest, destp, sp->sk_state,
1995 sk_wmem_alloc_get(sp),
1996 sk_rmem_alloc_get(sp),
1997 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1998 atomic_read(&sp->sk_refcnt), sp,
1999 atomic_read(&sp->sk_drops), len);
2002 int udp4_seq_show(struct seq_file *seq, void *v)
2004 if (v == SEQ_START_TOKEN)
2005 seq_printf(seq, "%-127s\n",
2006 " sl local_address rem_address st tx_queue "
2007 "rx_queue tr tm->when retrnsmt uid timeout "
2008 "inode ref pointer drops");
2010 struct udp_iter_state *state = seq->private;
2013 udp4_format_sock(v, seq, state->bucket, &len);
2014 seq_printf(seq, "%*s\n", 127 - len, "");
2019 /* ------------------------------------------------------------------------ */
2020 static struct udp_seq_afinfo udp4_seq_afinfo = {
2023 .udp_table = &udp_table,
2025 .owner = THIS_MODULE,
2028 .show = udp4_seq_show,
2032 static int __net_init udp4_proc_init_net(struct net *net)
2034 return udp_proc_register(net, &udp4_seq_afinfo);
2037 static void __net_exit udp4_proc_exit_net(struct net *net)
2039 udp_proc_unregister(net, &udp4_seq_afinfo);
2042 static struct pernet_operations udp4_net_ops = {
2043 .init = udp4_proc_init_net,
2044 .exit = udp4_proc_exit_net,
2047 int __init udp4_proc_init(void)
2049 return register_pernet_subsys(&udp4_net_ops);
2052 void udp4_proc_exit(void)
2054 unregister_pernet_subsys(&udp4_net_ops);
2056 #endif /* CONFIG_PROC_FS */
2058 static __initdata unsigned long uhash_entries;
2059 static int __init set_uhash_entries(char *str)
2063 uhash_entries = simple_strtoul(str, &str, 0);
2064 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2065 uhash_entries = UDP_HTABLE_SIZE_MIN;
2068 __setup("uhash_entries=", set_uhash_entries);
2070 void __init udp_table_init(struct udp_table *table, const char *name)
2074 if (!CONFIG_BASE_SMALL)
2075 table->hash = alloc_large_system_hash(name,
2076 2 * sizeof(struct udp_hslot),
2078 21, /* one slot per 2 MB */
2084 * Make sure hash table has the minimum size
2086 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2087 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
2088 2 * sizeof(struct udp_hslot), GFP_KERNEL);
2091 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2092 table->mask = UDP_HTABLE_SIZE_MIN - 1;
2094 table->hash2 = table->hash + (table->mask + 1);
2095 for (i = 0; i <= table->mask; i++) {
2096 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2097 table->hash[i].count = 0;
2098 spin_lock_init(&table->hash[i].lock);
2100 for (i = 0; i <= table->mask; i++) {
2101 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2102 table->hash2[i].count = 0;
2103 spin_lock_init(&table->hash2[i].lock);
2107 void __init udp_init(void)
2109 unsigned long nr_pages, limit;
2111 udp_table_init(&udp_table, "UDP");
2112 /* Set the pressure threshold up by the same strategy of TCP. It is a
2113 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
2114 * toward zero with the amount of memory, with a floor of 128 pages.
2116 nr_pages = totalram_pages - totalhigh_pages;
2117 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
2118 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
2119 limit = max(limit, 128UL);
2120 sysctl_udp_mem[0] = limit / 4 * 3;
2121 sysctl_udp_mem[1] = limit;
2122 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2124 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2125 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2128 int udp4_ufo_send_check(struct sk_buff *skb)
2130 const struct iphdr *iph;
2133 if (!pskb_may_pull(skb, sizeof(*uh)))
2139 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2141 skb->csum_start = skb_transport_header(skb) - skb->head;
2142 skb->csum_offset = offsetof(struct udphdr, check);
2143 skb->ip_summed = CHECKSUM_PARTIAL;
2147 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features)
2149 struct sk_buff *segs = ERR_PTR(-EINVAL);
2154 mss = skb_shinfo(skb)->gso_size;
2155 if (unlikely(skb->len <= mss))
2158 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2159 /* Packet is from an untrusted source, reset gso_segs. */
2160 int type = skb_shinfo(skb)->gso_type;
2162 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2163 !(type & (SKB_GSO_UDP))))
2166 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2172 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2173 * do checksum of UDP packets sent as multiple IP fragments.
2175 offset = skb->csum_start - skb_headroom(skb);
2176 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2177 offset += skb->csum_offset;
2178 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2179 skb->ip_summed = CHECKSUM_NONE;
2181 /* Fragment the skb. IP headers of the fragments are updated in
2182 * inet_gso_segment()
2184 segs = skb_segment(skb, features);