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 <trace/events/udp.h>
109 #include "udp_impl.h"
111 struct udp_table udp_table __read_mostly;
112 EXPORT_SYMBOL(udp_table);
114 long sysctl_udp_mem[3] __read_mostly;
115 EXPORT_SYMBOL(sysctl_udp_mem);
117 int sysctl_udp_rmem_min __read_mostly;
118 EXPORT_SYMBOL(sysctl_udp_rmem_min);
120 int sysctl_udp_wmem_min __read_mostly;
121 EXPORT_SYMBOL(sysctl_udp_wmem_min);
123 atomic_long_t udp_memory_allocated;
124 EXPORT_SYMBOL(udp_memory_allocated);
126 #define MAX_UDP_PORTS 65536
127 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
129 static int udp_lib_lport_inuse(struct net *net, __u16 num,
130 const struct udp_hslot *hslot,
131 unsigned long *bitmap,
133 int (*saddr_comp)(const struct sock *sk1,
134 const struct sock *sk2),
138 struct hlist_nulls_node *node;
140 sk_nulls_for_each(sk2, node, &hslot->head)
141 if (net_eq(sock_net(sk2), net) &&
143 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
144 (!sk2->sk_reuse || !sk->sk_reuse) &&
145 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
146 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
147 (*saddr_comp)(sk, sk2)) {
149 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
158 * Note: we still hold spinlock of primary hash chain, so no other writer
159 * can insert/delete a socket with local_port == num
161 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
162 struct udp_hslot *hslot2,
164 int (*saddr_comp)(const struct sock *sk1,
165 const struct sock *sk2))
168 struct hlist_nulls_node *node;
171 spin_lock(&hslot2->lock);
172 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
173 if (net_eq(sock_net(sk2), net) &&
175 (udp_sk(sk2)->udp_port_hash == num) &&
176 (!sk2->sk_reuse || !sk->sk_reuse) &&
177 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
178 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
179 (*saddr_comp)(sk, sk2)) {
183 spin_unlock(&hslot2->lock);
188 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
190 * @sk: socket struct in question
191 * @snum: port number to look up
192 * @saddr_comp: AF-dependent comparison of bound local IP addresses
193 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
196 int udp_lib_get_port(struct sock *sk, unsigned short snum,
197 int (*saddr_comp)(const struct sock *sk1,
198 const struct sock *sk2),
199 unsigned int hash2_nulladdr)
201 struct udp_hslot *hslot, *hslot2;
202 struct udp_table *udptable = sk->sk_prot->h.udp_table;
204 struct net *net = sock_net(sk);
207 int low, high, remaining;
209 unsigned short first, last;
210 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
212 inet_get_local_port_range(&low, &high);
213 remaining = (high - low) + 1;
216 first = (((u64)rand * remaining) >> 32) + low;
218 * force rand to be an odd multiple of UDP_HTABLE_SIZE
220 rand = (rand | 1) * (udptable->mask + 1);
221 last = first + udptable->mask + 1;
223 hslot = udp_hashslot(udptable, net, first);
224 bitmap_zero(bitmap, PORTS_PER_CHAIN);
225 spin_lock_bh(&hslot->lock);
226 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
227 saddr_comp, udptable->log);
231 * Iterate on all possible values of snum for this hash.
232 * Using steps of an odd multiple of UDP_HTABLE_SIZE
233 * give us randomization and full range coverage.
236 if (low <= snum && snum <= high &&
237 !test_bit(snum >> udptable->log, bitmap) &&
238 !inet_is_reserved_local_port(snum))
241 } while (snum != first);
242 spin_unlock_bh(&hslot->lock);
243 } while (++first != last);
246 hslot = udp_hashslot(udptable, net, snum);
247 spin_lock_bh(&hslot->lock);
248 if (hslot->count > 10) {
250 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
252 slot2 &= udptable->mask;
253 hash2_nulladdr &= udptable->mask;
255 hslot2 = udp_hashslot2(udptable, slot2);
256 if (hslot->count < hslot2->count)
257 goto scan_primary_hash;
259 exist = udp_lib_lport_inuse2(net, snum, hslot2,
261 if (!exist && (hash2_nulladdr != slot2)) {
262 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
263 exist = udp_lib_lport_inuse2(net, snum, hslot2,
272 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
277 inet_sk(sk)->inet_num = snum;
278 udp_sk(sk)->udp_port_hash = snum;
279 udp_sk(sk)->udp_portaddr_hash ^= snum;
280 if (sk_unhashed(sk)) {
281 sk_nulls_add_node_rcu(sk, &hslot->head);
283 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
285 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
286 spin_lock(&hslot2->lock);
287 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
290 spin_unlock(&hslot2->lock);
294 spin_unlock_bh(&hslot->lock);
298 EXPORT_SYMBOL(udp_lib_get_port);
300 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
302 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
304 return (!ipv6_only_sock(sk2) &&
305 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
306 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
309 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
312 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
315 int udp_v4_get_port(struct sock *sk, unsigned short snum)
317 unsigned int hash2_nulladdr =
318 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
319 unsigned int hash2_partial =
320 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
322 /* precompute partial secondary hash */
323 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
324 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
327 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
329 __be16 sport, __be32 daddr, __be16 dport, int dif)
333 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
334 !ipv6_only_sock(sk)) {
335 struct inet_sock *inet = inet_sk(sk);
337 score = (sk->sk_family == PF_INET ? 1 : 0);
338 if (inet->inet_rcv_saddr) {
339 if (inet->inet_rcv_saddr != daddr)
343 if (inet->inet_daddr) {
344 if (inet->inet_daddr != saddr)
348 if (inet->inet_dport) {
349 if (inet->inet_dport != sport)
353 if (sk->sk_bound_dev_if) {
354 if (sk->sk_bound_dev_if != dif)
363 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
365 #define SCORE2_MAX (1 + 2 + 2 + 2)
366 static inline int compute_score2(struct sock *sk, struct net *net,
367 __be32 saddr, __be16 sport,
368 __be32 daddr, unsigned int hnum, int dif)
372 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
373 struct inet_sock *inet = inet_sk(sk);
375 if (inet->inet_rcv_saddr != daddr)
377 if (inet->inet_num != hnum)
380 score = (sk->sk_family == PF_INET ? 1 : 0);
381 if (inet->inet_daddr) {
382 if (inet->inet_daddr != saddr)
386 if (inet->inet_dport) {
387 if (inet->inet_dport != sport)
391 if (sk->sk_bound_dev_if) {
392 if (sk->sk_bound_dev_if != dif)
401 /* called with read_rcu_lock() */
402 static struct sock *udp4_lib_lookup2(struct net *net,
403 __be32 saddr, __be16 sport,
404 __be32 daddr, unsigned int hnum, int dif,
405 struct udp_hslot *hslot2, unsigned int slot2)
407 struct sock *sk, *result;
408 struct hlist_nulls_node *node;
414 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
415 score = compute_score2(sk, net, saddr, sport,
417 if (score > badness) {
420 if (score == SCORE2_MAX)
425 * if the nulls value we got at the end of this lookup is
426 * not the expected one, we must restart lookup.
427 * We probably met an item that was moved to another chain.
429 if (get_nulls_value(node) != slot2)
434 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
436 else if (unlikely(compute_score2(result, net, saddr, sport,
437 daddr, hnum, dif) < badness)) {
445 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
446 * harder than this. -DaveM
448 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
449 __be16 sport, __be32 daddr, __be16 dport,
450 int dif, struct udp_table *udptable)
452 struct sock *sk, *result;
453 struct hlist_nulls_node *node;
454 unsigned short hnum = ntohs(dport);
455 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
456 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
460 if (hslot->count > 10) {
461 hash2 = udp4_portaddr_hash(net, daddr, hnum);
462 slot2 = hash2 & udptable->mask;
463 hslot2 = &udptable->hash2[slot2];
464 if (hslot->count < hslot2->count)
467 result = udp4_lib_lookup2(net, saddr, sport,
471 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
472 slot2 = hash2 & udptable->mask;
473 hslot2 = &udptable->hash2[slot2];
474 if (hslot->count < hslot2->count)
477 result = udp4_lib_lookup2(net, saddr, sport,
478 htonl(INADDR_ANY), hnum, dif,
487 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
488 score = compute_score(sk, net, saddr, hnum, sport,
490 if (score > badness) {
496 * if the nulls value we got at the end of this lookup is
497 * not the expected one, we must restart lookup.
498 * We probably met an item that was moved to another chain.
500 if (get_nulls_value(node) != slot)
504 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
506 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
507 daddr, dport, dif) < badness)) {
516 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
517 __be16 sport, __be16 dport,
518 struct udp_table *udptable)
521 const struct iphdr *iph = ip_hdr(skb);
523 if (unlikely(sk = skb_steal_sock(skb)))
526 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
527 iph->daddr, dport, inet_iif(skb),
531 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
532 __be32 daddr, __be16 dport, int dif)
534 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
536 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
538 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
539 __be16 loc_port, __be32 loc_addr,
540 __be16 rmt_port, __be32 rmt_addr,
543 struct hlist_nulls_node *node;
545 unsigned short hnum = ntohs(loc_port);
547 sk_nulls_for_each_from(s, node) {
548 struct inet_sock *inet = inet_sk(s);
550 if (!net_eq(sock_net(s), net) ||
551 udp_sk(s)->udp_port_hash != hnum ||
552 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
553 (inet->inet_dport != rmt_port && inet->inet_dport) ||
554 (inet->inet_rcv_saddr &&
555 inet->inet_rcv_saddr != loc_addr) ||
557 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
559 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
569 * This routine is called by the ICMP module when it gets some
570 * sort of error condition. If err < 0 then the socket should
571 * be closed and the error returned to the user. If err > 0
572 * it's just the icmp type << 8 | icmp code.
573 * Header points to the ip header of the error packet. We move
574 * on past this. Then (as it used to claim before adjustment)
575 * header points to the first 8 bytes of the udp header. We need
576 * to find the appropriate port.
579 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
581 struct inet_sock *inet;
582 const struct iphdr *iph = (const struct iphdr *)skb->data;
583 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
584 const int type = icmp_hdr(skb)->type;
585 const int code = icmp_hdr(skb)->code;
589 struct net *net = dev_net(skb->dev);
591 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
592 iph->saddr, uh->source, skb->dev->ifindex, udptable);
594 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
595 return; /* No socket for error */
604 case ICMP_TIME_EXCEEDED:
607 case ICMP_SOURCE_QUENCH:
609 case ICMP_PARAMETERPROB:
613 case ICMP_DEST_UNREACH:
614 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
615 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
623 if (code <= NR_ICMP_UNREACH) {
624 harderr = icmp_err_convert[code].fatal;
625 err = icmp_err_convert[code].errno;
631 * RFC1122: OK. Passes ICMP errors back to application, as per
634 if (!inet->recverr) {
635 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
638 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
641 sk->sk_error_report(sk);
646 void udp_err(struct sk_buff *skb, u32 info)
648 __udp4_lib_err(skb, info, &udp_table);
652 * Throw away all pending data and cancel the corking. Socket is locked.
654 void udp_flush_pending_frames(struct sock *sk)
656 struct udp_sock *up = udp_sk(sk);
661 ip_flush_pending_frames(sk);
664 EXPORT_SYMBOL(udp_flush_pending_frames);
667 * udp4_hwcsum - handle outgoing HW checksumming
668 * @skb: sk_buff containing the filled-in UDP header
669 * (checksum field must be zeroed out)
670 * @src: source IP address
671 * @dst: destination IP address
673 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
675 struct udphdr *uh = udp_hdr(skb);
676 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
677 int offset = skb_transport_offset(skb);
678 int len = skb->len - offset;
684 * Only one fragment on the socket.
686 skb->csum_start = skb_transport_header(skb) - skb->head;
687 skb->csum_offset = offsetof(struct udphdr, check);
688 uh->check = ~csum_tcpudp_magic(src, dst, len,
692 * HW-checksum won't work as there are two or more
693 * fragments on the socket so that all csums of sk_buffs
697 csum = csum_add(csum, frags->csum);
699 } while ((frags = frags->next));
701 csum = skb_checksum(skb, offset, hlen, csum);
702 skb->ip_summed = CHECKSUM_NONE;
704 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
706 uh->check = CSUM_MANGLED_0;
710 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
712 struct sock *sk = skb->sk;
713 struct inet_sock *inet = inet_sk(sk);
716 int is_udplite = IS_UDPLITE(sk);
717 int offset = skb_transport_offset(skb);
718 int len = skb->len - offset;
722 * Create a UDP header
725 uh->source = inet->inet_sport;
726 uh->dest = fl4->fl4_dport;
727 uh->len = htons(len);
730 if (is_udplite) /* UDP-Lite */
731 csum = udplite_csum(skb);
733 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
735 skb->ip_summed = CHECKSUM_NONE;
738 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
740 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
744 csum = udp_csum(skb);
746 /* add protocol-dependent pseudo-header */
747 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
748 sk->sk_protocol, csum);
750 uh->check = CSUM_MANGLED_0;
753 err = ip_send_skb(skb);
755 if (err == -ENOBUFS && !inet->recverr) {
756 UDP_INC_STATS_USER(sock_net(sk),
757 UDP_MIB_SNDBUFERRORS, is_udplite);
761 UDP_INC_STATS_USER(sock_net(sk),
762 UDP_MIB_OUTDATAGRAMS, is_udplite);
767 * Push out all pending data as one UDP datagram. Socket is locked.
769 static int udp_push_pending_frames(struct sock *sk)
771 struct udp_sock *up = udp_sk(sk);
772 struct inet_sock *inet = inet_sk(sk);
773 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
777 skb = ip_finish_skb(sk, fl4);
781 err = udp_send_skb(skb, fl4);
789 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
792 struct inet_sock *inet = inet_sk(sk);
793 struct udp_sock *up = udp_sk(sk);
794 struct flowi4 fl4_stack;
797 struct ipcm_cookie ipc;
798 struct rtable *rt = NULL;
801 __be32 daddr, faddr, saddr;
804 int err, is_udplite = IS_UDPLITE(sk);
805 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
806 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
808 struct ip_options_data opt_copy;
817 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
823 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
825 fl4 = &inet->cork.fl.u.ip4;
828 * There are pending frames.
829 * The socket lock must be held while it's corked.
832 if (likely(up->pending)) {
833 if (unlikely(up->pending != AF_INET)) {
841 ulen += sizeof(struct udphdr);
844 * Get and verify the address.
847 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
848 if (msg->msg_namelen < sizeof(*usin))
850 if (usin->sin_family != AF_INET) {
851 if (usin->sin_family != AF_UNSPEC)
852 return -EAFNOSUPPORT;
855 daddr = usin->sin_addr.s_addr;
856 dport = usin->sin_port;
860 if (sk->sk_state != TCP_ESTABLISHED)
861 return -EDESTADDRREQ;
862 daddr = inet->inet_daddr;
863 dport = inet->inet_dport;
864 /* Open fast path for connected socket.
865 Route will not be used, if at least one option is set.
869 ipc.addr = inet->inet_saddr;
871 ipc.oif = sk->sk_bound_dev_if;
872 err = sock_tx_timestamp(sk, &ipc.tx_flags);
875 if (msg->msg_controllen) {
876 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
884 struct ip_options_rcu *inet_opt;
887 inet_opt = rcu_dereference(inet->inet_opt);
889 memcpy(&opt_copy, inet_opt,
890 sizeof(*inet_opt) + inet_opt->opt.optlen);
891 ipc.opt = &opt_copy.opt;
897 ipc.addr = faddr = daddr;
899 if (ipc.opt && ipc.opt->opt.srr) {
902 faddr = ipc.opt->opt.faddr;
905 tos = RT_TOS(inet->tos);
906 if (sock_flag(sk, SOCK_LOCALROUTE) ||
907 (msg->msg_flags & MSG_DONTROUTE) ||
908 (ipc.opt && ipc.opt->opt.is_strictroute)) {
913 if (ipv4_is_multicast(daddr)) {
915 ipc.oif = inet->mc_index;
917 saddr = inet->mc_addr;
922 rt = (struct rtable *)sk_dst_check(sk, 0);
925 struct net *net = sock_net(sk);
928 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
929 RT_SCOPE_UNIVERSE, sk->sk_protocol,
930 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
931 faddr, saddr, dport, inet->inet_sport);
933 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
934 rt = ip_route_output_flow(net, fl4, sk);
938 if (err == -ENETUNREACH)
939 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
944 if ((rt->rt_flags & RTCF_BROADCAST) &&
945 !sock_flag(sk, SOCK_BROADCAST))
948 sk_dst_set(sk, dst_clone(&rt->dst));
951 if (msg->msg_flags&MSG_CONFIRM)
957 daddr = ipc.addr = fl4->daddr;
959 /* Lockless fast path for the non-corking case. */
961 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
962 sizeof(struct udphdr), &ipc, &rt,
965 if (skb && !IS_ERR(skb))
966 err = udp_send_skb(skb, fl4);
971 if (unlikely(up->pending)) {
972 /* The socket is already corked while preparing it. */
973 /* ... which is an evident application bug. --ANK */
976 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
981 * Now cork the socket to pend data.
983 fl4 = &inet->cork.fl.u.ip4;
986 fl4->fl4_dport = dport;
987 fl4->fl4_sport = inet->inet_sport;
988 up->pending = AF_INET;
992 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
993 sizeof(struct udphdr), &ipc, &rt,
994 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
996 udp_flush_pending_frames(sk);
998 err = udp_push_pending_frames(sk);
999 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1010 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1011 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1012 * we don't have a good statistic (IpOutDiscards but it can be too many
1013 * things). We could add another new stat but at least for now that
1014 * seems like overkill.
1016 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1017 UDP_INC_STATS_USER(sock_net(sk),
1018 UDP_MIB_SNDBUFERRORS, is_udplite);
1023 dst_confirm(&rt->dst);
1024 if (!(msg->msg_flags&MSG_PROBE) || len)
1025 goto back_from_confirm;
1029 EXPORT_SYMBOL(udp_sendmsg);
1031 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1032 size_t size, int flags)
1034 struct inet_sock *inet = inet_sk(sk);
1035 struct udp_sock *up = udp_sk(sk);
1039 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1041 /* Call udp_sendmsg to specify destination address which
1042 * sendpage interface can't pass.
1043 * This will succeed only when the socket is connected.
1045 ret = udp_sendmsg(NULL, sk, &msg, 0);
1052 if (unlikely(!up->pending)) {
1055 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
1059 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1060 page, offset, size, flags);
1061 if (ret == -EOPNOTSUPP) {
1063 return sock_no_sendpage(sk->sk_socket, page, offset,
1067 udp_flush_pending_frames(sk);
1072 if (!(up->corkflag || (flags&MSG_MORE)))
1073 ret = udp_push_pending_frames(sk);
1083 * first_packet_length - return length of first packet in receive queue
1086 * Drops all bad checksum frames, until a valid one is found.
1087 * Returns the length of found skb, or 0 if none is found.
1089 static unsigned int first_packet_length(struct sock *sk)
1091 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1092 struct sk_buff *skb;
1095 __skb_queue_head_init(&list_kill);
1097 spin_lock_bh(&rcvq->lock);
1098 while ((skb = skb_peek(rcvq)) != NULL &&
1099 udp_lib_checksum_complete(skb)) {
1100 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1102 atomic_inc(&sk->sk_drops);
1103 __skb_unlink(skb, rcvq);
1104 __skb_queue_tail(&list_kill, skb);
1106 res = skb ? skb->len : 0;
1107 spin_unlock_bh(&rcvq->lock);
1109 if (!skb_queue_empty(&list_kill)) {
1110 bool slow = lock_sock_fast(sk);
1112 __skb_queue_purge(&list_kill);
1113 sk_mem_reclaim_partial(sk);
1114 unlock_sock_fast(sk, slow);
1120 * IOCTL requests applicable to the UDP protocol
1123 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1128 int amount = sk_wmem_alloc_get(sk);
1130 return put_user(amount, (int __user *)arg);
1135 unsigned int amount = first_packet_length(sk);
1139 * We will only return the amount
1140 * of this packet since that is all
1141 * that will be read.
1143 amount -= sizeof(struct udphdr);
1145 return put_user(amount, (int __user *)arg);
1149 return -ENOIOCTLCMD;
1154 EXPORT_SYMBOL(udp_ioctl);
1157 * This should be easy, if there is something there we
1158 * return it, otherwise we block.
1161 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1162 size_t len, int noblock, int flags, int *addr_len)
1164 struct inet_sock *inet = inet_sk(sk);
1165 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1166 struct sk_buff *skb;
1170 int is_udplite = IS_UDPLITE(sk);
1174 * Check any passed addresses
1177 *addr_len = sizeof(*sin);
1179 if (flags & MSG_ERRQUEUE)
1180 return ip_recv_error(sk, msg, len);
1183 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1188 ulen = skb->len - sizeof(struct udphdr);
1191 else if (len < ulen)
1192 msg->msg_flags |= MSG_TRUNC;
1195 * If checksum is needed at all, try to do it while copying the
1196 * data. If the data is truncated, or if we only want a partial
1197 * coverage checksum (UDP-Lite), do it before the copy.
1200 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
1201 if (udp_lib_checksum_complete(skb))
1205 if (skb_csum_unnecessary(skb))
1206 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1209 err = skb_copy_and_csum_datagram_iovec(skb,
1210 sizeof(struct udphdr),
1221 UDP_INC_STATS_USER(sock_net(sk),
1222 UDP_MIB_INDATAGRAMS, is_udplite);
1224 sock_recv_ts_and_drops(msg, sk, skb);
1226 /* Copy the address. */
1228 sin->sin_family = AF_INET;
1229 sin->sin_port = udp_hdr(skb)->source;
1230 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1231 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1233 if (inet->cmsg_flags)
1234 ip_cmsg_recv(msg, skb);
1237 if (flags & MSG_TRUNC)
1241 skb_free_datagram_locked(sk, skb);
1246 slow = lock_sock_fast(sk);
1247 if (!skb_kill_datagram(sk, skb, flags))
1248 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1249 unlock_sock_fast(sk, slow);
1254 /* starting over for a new packet */
1255 msg->msg_flags &= ~MSG_TRUNC;
1260 int udp_disconnect(struct sock *sk, int flags)
1262 struct inet_sock *inet = inet_sk(sk);
1264 * 1003.1g - break association.
1267 sk->sk_state = TCP_CLOSE;
1268 inet->inet_daddr = 0;
1269 inet->inet_dport = 0;
1270 sock_rps_save_rxhash(sk, 0);
1271 sk->sk_bound_dev_if = 0;
1272 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1273 inet_reset_saddr(sk);
1275 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1276 sk->sk_prot->unhash(sk);
1277 inet->inet_sport = 0;
1282 EXPORT_SYMBOL(udp_disconnect);
1284 void udp_lib_unhash(struct sock *sk)
1286 if (sk_hashed(sk)) {
1287 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1288 struct udp_hslot *hslot, *hslot2;
1290 hslot = udp_hashslot(udptable, sock_net(sk),
1291 udp_sk(sk)->udp_port_hash);
1292 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1294 spin_lock_bh(&hslot->lock);
1295 if (sk_nulls_del_node_init_rcu(sk)) {
1297 inet_sk(sk)->inet_num = 0;
1298 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1300 spin_lock(&hslot2->lock);
1301 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1303 spin_unlock(&hslot2->lock);
1305 spin_unlock_bh(&hslot->lock);
1308 EXPORT_SYMBOL(udp_lib_unhash);
1311 * inet_rcv_saddr was changed, we must rehash secondary hash
1313 void udp_lib_rehash(struct sock *sk, u16 newhash)
1315 if (sk_hashed(sk)) {
1316 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1317 struct udp_hslot *hslot, *hslot2, *nhslot2;
1319 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1320 nhslot2 = udp_hashslot2(udptable, newhash);
1321 udp_sk(sk)->udp_portaddr_hash = newhash;
1322 if (hslot2 != nhslot2) {
1323 hslot = udp_hashslot(udptable, sock_net(sk),
1324 udp_sk(sk)->udp_port_hash);
1325 /* we must lock primary chain too */
1326 spin_lock_bh(&hslot->lock);
1328 spin_lock(&hslot2->lock);
1329 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1331 spin_unlock(&hslot2->lock);
1333 spin_lock(&nhslot2->lock);
1334 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1337 spin_unlock(&nhslot2->lock);
1339 spin_unlock_bh(&hslot->lock);
1343 EXPORT_SYMBOL(udp_lib_rehash);
1345 static void udp_v4_rehash(struct sock *sk)
1347 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1348 inet_sk(sk)->inet_rcv_saddr,
1349 inet_sk(sk)->inet_num);
1350 udp_lib_rehash(sk, new_hash);
1353 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1357 if (inet_sk(sk)->inet_daddr)
1358 sock_rps_save_rxhash(sk, skb->rxhash);
1360 rc = ip_queue_rcv_skb(sk, skb);
1362 int is_udplite = IS_UDPLITE(sk);
1364 /* Note that an ENOMEM error is charged twice */
1366 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1368 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1370 trace_udp_fail_queue_rcv_skb(rc, sk);
1381 * >0: "udp encap" protocol resubmission
1383 * Note that in the success and error cases, the skb is assumed to
1384 * have either been requeued or freed.
1386 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1388 struct udp_sock *up = udp_sk(sk);
1390 int is_udplite = IS_UDPLITE(sk);
1393 * Charge it to the socket, dropping if the queue is full.
1395 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1399 if (up->encap_type) {
1401 * This is an encapsulation socket so pass the skb to
1402 * the socket's udp_encap_rcv() hook. Otherwise, just
1403 * fall through and pass this up the UDP socket.
1404 * up->encap_rcv() returns the following value:
1405 * =0 if skb was successfully passed to the encap
1406 * handler or was discarded by it.
1407 * >0 if skb should be passed on to UDP.
1408 * <0 if skb should be resubmitted as proto -N
1411 /* if we're overly short, let UDP handle it */
1412 if (skb->len > sizeof(struct udphdr) &&
1413 up->encap_rcv != NULL) {
1416 ret = (*up->encap_rcv)(sk, skb);
1418 UDP_INC_STATS_BH(sock_net(sk),
1419 UDP_MIB_INDATAGRAMS,
1425 /* FALLTHROUGH -- it's a UDP Packet */
1429 * UDP-Lite specific tests, ignored on UDP sockets
1431 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1434 * MIB statistics other than incrementing the error count are
1435 * disabled for the following two types of errors: these depend
1436 * on the application settings, not on the functioning of the
1437 * protocol stack as such.
1439 * RFC 3828 here recommends (sec 3.3): "There should also be a
1440 * way ... to ... at least let the receiving application block
1441 * delivery of packets with coverage values less than a value
1442 * provided by the application."
1444 if (up->pcrlen == 0) { /* full coverage was set */
1445 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1446 "%d while full coverage %d requested\n",
1447 UDP_SKB_CB(skb)->cscov, skb->len);
1450 /* The next case involves violating the min. coverage requested
1451 * by the receiver. This is subtle: if receiver wants x and x is
1452 * greater than the buffersize/MTU then receiver will complain
1453 * that it wants x while sender emits packets of smaller size y.
1454 * Therefore the above ...()->partial_cov statement is essential.
1456 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1457 LIMIT_NETDEBUG(KERN_WARNING
1458 "UDPLITE: coverage %d too small, need min %d\n",
1459 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1464 if (rcu_dereference_raw(sk->sk_filter)) {
1465 if (udp_lib_checksum_complete(skb))
1470 if (sk_rcvqueues_full(sk, skb))
1476 if (!sock_owned_by_user(sk))
1477 rc = __udp_queue_rcv_skb(sk, skb);
1478 else if (sk_add_backlog(sk, skb)) {
1487 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1488 atomic_inc(&sk->sk_drops);
1494 static void flush_stack(struct sock **stack, unsigned int count,
1495 struct sk_buff *skb, unsigned int final)
1498 struct sk_buff *skb1 = NULL;
1501 for (i = 0; i < count; i++) {
1503 if (likely(skb1 == NULL))
1504 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1507 atomic_inc(&sk->sk_drops);
1508 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1510 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1514 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1522 * Multicasts and broadcasts go to each listener.
1524 * Note: called only from the BH handler context.
1526 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1528 __be32 saddr, __be32 daddr,
1529 struct udp_table *udptable)
1531 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1532 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1534 unsigned int i, count = 0;
1536 spin_lock(&hslot->lock);
1537 sk = sk_nulls_head(&hslot->head);
1538 dif = skb->dev->ifindex;
1539 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1541 stack[count++] = sk;
1542 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1543 daddr, uh->source, saddr, dif);
1544 if (unlikely(count == ARRAY_SIZE(stack))) {
1547 flush_stack(stack, count, skb, ~0);
1552 * before releasing chain lock, we must take a reference on sockets
1554 for (i = 0; i < count; i++)
1555 sock_hold(stack[i]);
1557 spin_unlock(&hslot->lock);
1560 * do the slow work with no lock held
1563 flush_stack(stack, count, skb, count - 1);
1565 for (i = 0; i < count; i++)
1573 /* Initialize UDP checksum. If exited with zero value (success),
1574 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1575 * Otherwise, csum completion requires chacksumming packet body,
1576 * including udp header and folding it to skb->csum.
1578 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1581 const struct iphdr *iph;
1584 UDP_SKB_CB(skb)->partial_cov = 0;
1585 UDP_SKB_CB(skb)->cscov = skb->len;
1587 if (proto == IPPROTO_UDPLITE) {
1588 err = udplite_checksum_init(skb, uh);
1594 if (uh->check == 0) {
1595 skb->ip_summed = CHECKSUM_UNNECESSARY;
1596 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1597 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1599 skb->ip_summed = CHECKSUM_UNNECESSARY;
1601 if (!skb_csum_unnecessary(skb))
1602 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1603 skb->len, proto, 0);
1604 /* Probably, we should checksum udp header (it should be in cache
1605 * in any case) and data in tiny packets (< rx copybreak).
1612 * All we need to do is get the socket, and then do a checksum.
1615 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1620 unsigned short ulen;
1621 struct rtable *rt = skb_rtable(skb);
1622 __be32 saddr, daddr;
1623 struct net *net = dev_net(skb->dev);
1626 * Validate the packet.
1628 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1629 goto drop; /* No space for header. */
1632 ulen = ntohs(uh->len);
1633 saddr = ip_hdr(skb)->saddr;
1634 daddr = ip_hdr(skb)->daddr;
1636 if (ulen > skb->len)
1639 if (proto == IPPROTO_UDP) {
1640 /* UDP validates ulen. */
1641 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1646 if (udp4_csum_init(skb, uh, proto))
1649 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1650 return __udp4_lib_mcast_deliver(net, skb, uh,
1651 saddr, daddr, udptable);
1653 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1656 int ret = udp_queue_rcv_skb(sk, skb);
1659 /* a return value > 0 means to resubmit the input, but
1660 * it wants the return to be -protocol, or 0
1667 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1671 /* No socket. Drop packet silently, if checksum is wrong */
1672 if (udp_lib_checksum_complete(skb))
1675 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1676 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1679 * Hmm. We got an UDP packet to a port to which we
1680 * don't wanna listen. Ignore it.
1686 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1687 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1698 * RFC1122: OK. Discards the bad packet silently (as far as
1699 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1701 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1702 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1709 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1714 int udp_rcv(struct sk_buff *skb)
1716 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1719 void udp_destroy_sock(struct sock *sk)
1721 bool slow = lock_sock_fast(sk);
1722 udp_flush_pending_frames(sk);
1723 unlock_sock_fast(sk, slow);
1727 * Socket option code for UDP
1729 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1730 char __user *optval, unsigned int optlen,
1731 int (*push_pending_frames)(struct sock *))
1733 struct udp_sock *up = udp_sk(sk);
1736 int is_udplite = IS_UDPLITE(sk);
1738 if (optlen < sizeof(int))
1741 if (get_user(val, (int __user *)optval))
1751 (*push_pending_frames)(sk);
1759 case UDP_ENCAP_ESPINUDP:
1760 case UDP_ENCAP_ESPINUDP_NON_IKE:
1761 up->encap_rcv = xfrm4_udp_encap_rcv;
1763 case UDP_ENCAP_L2TPINUDP:
1764 up->encap_type = val;
1773 * UDP-Lite's partial checksum coverage (RFC 3828).
1775 /* The sender sets actual checksum coverage length via this option.
1776 * The case coverage > packet length is handled by send module. */
1777 case UDPLITE_SEND_CSCOV:
1778 if (!is_udplite) /* Disable the option on UDP sockets */
1779 return -ENOPROTOOPT;
1780 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1782 else if (val > USHRT_MAX)
1785 up->pcflag |= UDPLITE_SEND_CC;
1788 /* The receiver specifies a minimum checksum coverage value. To make
1789 * sense, this should be set to at least 8 (as done below). If zero is
1790 * used, this again means full checksum coverage. */
1791 case UDPLITE_RECV_CSCOV:
1792 if (!is_udplite) /* Disable the option on UDP sockets */
1793 return -ENOPROTOOPT;
1794 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1796 else if (val > USHRT_MAX)
1799 up->pcflag |= UDPLITE_RECV_CC;
1809 EXPORT_SYMBOL(udp_lib_setsockopt);
1811 int udp_setsockopt(struct sock *sk, int level, int optname,
1812 char __user *optval, unsigned int optlen)
1814 if (level == SOL_UDP || level == SOL_UDPLITE)
1815 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1816 udp_push_pending_frames);
1817 return ip_setsockopt(sk, level, optname, optval, optlen);
1820 #ifdef CONFIG_COMPAT
1821 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1822 char __user *optval, unsigned int optlen)
1824 if (level == SOL_UDP || level == SOL_UDPLITE)
1825 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1826 udp_push_pending_frames);
1827 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1831 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1832 char __user *optval, int __user *optlen)
1834 struct udp_sock *up = udp_sk(sk);
1837 if (get_user(len, optlen))
1840 len = min_t(unsigned int, len, sizeof(int));
1851 val = up->encap_type;
1854 /* The following two cannot be changed on UDP sockets, the return is
1855 * always 0 (which corresponds to the full checksum coverage of UDP). */
1856 case UDPLITE_SEND_CSCOV:
1860 case UDPLITE_RECV_CSCOV:
1865 return -ENOPROTOOPT;
1868 if (put_user(len, optlen))
1870 if (copy_to_user(optval, &val, len))
1874 EXPORT_SYMBOL(udp_lib_getsockopt);
1876 int udp_getsockopt(struct sock *sk, int level, int optname,
1877 char __user *optval, int __user *optlen)
1879 if (level == SOL_UDP || level == SOL_UDPLITE)
1880 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1881 return ip_getsockopt(sk, level, optname, optval, optlen);
1884 #ifdef CONFIG_COMPAT
1885 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1886 char __user *optval, int __user *optlen)
1888 if (level == SOL_UDP || level == SOL_UDPLITE)
1889 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1890 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1894 * udp_poll - wait for a UDP event.
1895 * @file - file struct
1897 * @wait - poll table
1899 * This is same as datagram poll, except for the special case of
1900 * blocking sockets. If application is using a blocking fd
1901 * and a packet with checksum error is in the queue;
1902 * then it could get return from select indicating data available
1903 * but then block when reading it. Add special case code
1904 * to work around these arguably broken applications.
1906 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1908 unsigned int mask = datagram_poll(file, sock, wait);
1909 struct sock *sk = sock->sk;
1911 /* Check for false positives due to checksum errors */
1912 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1913 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1914 mask &= ~(POLLIN | POLLRDNORM);
1919 EXPORT_SYMBOL(udp_poll);
1921 struct proto udp_prot = {
1923 .owner = THIS_MODULE,
1924 .close = udp_lib_close,
1925 .connect = ip4_datagram_connect,
1926 .disconnect = udp_disconnect,
1928 .destroy = udp_destroy_sock,
1929 .setsockopt = udp_setsockopt,
1930 .getsockopt = udp_getsockopt,
1931 .sendmsg = udp_sendmsg,
1932 .recvmsg = udp_recvmsg,
1933 .sendpage = udp_sendpage,
1934 .backlog_rcv = __udp_queue_rcv_skb,
1935 .hash = udp_lib_hash,
1936 .unhash = udp_lib_unhash,
1937 .rehash = udp_v4_rehash,
1938 .get_port = udp_v4_get_port,
1939 .memory_allocated = &udp_memory_allocated,
1940 .sysctl_mem = sysctl_udp_mem,
1941 .sysctl_wmem = &sysctl_udp_wmem_min,
1942 .sysctl_rmem = &sysctl_udp_rmem_min,
1943 .obj_size = sizeof(struct udp_sock),
1944 .slab_flags = SLAB_DESTROY_BY_RCU,
1945 .h.udp_table = &udp_table,
1946 #ifdef CONFIG_COMPAT
1947 .compat_setsockopt = compat_udp_setsockopt,
1948 .compat_getsockopt = compat_udp_getsockopt,
1950 .clear_sk = sk_prot_clear_portaddr_nulls,
1952 EXPORT_SYMBOL(udp_prot);
1954 /* ------------------------------------------------------------------------ */
1955 #ifdef CONFIG_PROC_FS
1957 static struct sock *udp_get_first(struct seq_file *seq, int start)
1960 struct udp_iter_state *state = seq->private;
1961 struct net *net = seq_file_net(seq);
1963 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1965 struct hlist_nulls_node *node;
1966 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1968 if (hlist_nulls_empty(&hslot->head))
1971 spin_lock_bh(&hslot->lock);
1972 sk_nulls_for_each(sk, node, &hslot->head) {
1973 if (!net_eq(sock_net(sk), net))
1975 if (sk->sk_family == state->family)
1978 spin_unlock_bh(&hslot->lock);
1985 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1987 struct udp_iter_state *state = seq->private;
1988 struct net *net = seq_file_net(seq);
1991 sk = sk_nulls_next(sk);
1992 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1995 if (state->bucket <= state->udp_table->mask)
1996 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1997 return udp_get_first(seq, state->bucket + 1);
2002 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2004 struct sock *sk = udp_get_first(seq, 0);
2007 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2009 return pos ? NULL : sk;
2012 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2014 struct udp_iter_state *state = seq->private;
2015 state->bucket = MAX_UDP_PORTS;
2017 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2020 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2024 if (v == SEQ_START_TOKEN)
2025 sk = udp_get_idx(seq, 0);
2027 sk = udp_get_next(seq, v);
2033 static void udp_seq_stop(struct seq_file *seq, void *v)
2035 struct udp_iter_state *state = seq->private;
2037 if (state->bucket <= state->udp_table->mask)
2038 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2041 static int udp_seq_open(struct inode *inode, struct file *file)
2043 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
2044 struct udp_iter_state *s;
2047 err = seq_open_net(inode, file, &afinfo->seq_ops,
2048 sizeof(struct udp_iter_state));
2052 s = ((struct seq_file *)file->private_data)->private;
2053 s->family = afinfo->family;
2054 s->udp_table = afinfo->udp_table;
2058 /* ------------------------------------------------------------------------ */
2059 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2061 struct proc_dir_entry *p;
2064 afinfo->seq_fops.open = udp_seq_open;
2065 afinfo->seq_fops.read = seq_read;
2066 afinfo->seq_fops.llseek = seq_lseek;
2067 afinfo->seq_fops.release = seq_release_net;
2069 afinfo->seq_ops.start = udp_seq_start;
2070 afinfo->seq_ops.next = udp_seq_next;
2071 afinfo->seq_ops.stop = udp_seq_stop;
2073 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2074 &afinfo->seq_fops, afinfo);
2079 EXPORT_SYMBOL(udp_proc_register);
2081 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2083 proc_net_remove(net, afinfo->name);
2085 EXPORT_SYMBOL(udp_proc_unregister);
2087 /* ------------------------------------------------------------------------ */
2088 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2089 int bucket, int *len)
2091 struct inet_sock *inet = inet_sk(sp);
2092 __be32 dest = inet->inet_daddr;
2093 __be32 src = inet->inet_rcv_saddr;
2094 __u16 destp = ntohs(inet->inet_dport);
2095 __u16 srcp = ntohs(inet->inet_sport);
2097 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2098 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2099 bucket, src, srcp, dest, destp, sp->sk_state,
2100 sk_wmem_alloc_get(sp),
2101 sk_rmem_alloc_get(sp),
2102 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
2103 atomic_read(&sp->sk_refcnt), sp,
2104 atomic_read(&sp->sk_drops), len);
2107 int udp4_seq_show(struct seq_file *seq, void *v)
2109 if (v == SEQ_START_TOKEN)
2110 seq_printf(seq, "%-127s\n",
2111 " sl local_address rem_address st tx_queue "
2112 "rx_queue tr tm->when retrnsmt uid timeout "
2113 "inode ref pointer drops");
2115 struct udp_iter_state *state = seq->private;
2118 udp4_format_sock(v, seq, state->bucket, &len);
2119 seq_printf(seq, "%*s\n", 127 - len, "");
2124 /* ------------------------------------------------------------------------ */
2125 static struct udp_seq_afinfo udp4_seq_afinfo = {
2128 .udp_table = &udp_table,
2130 .owner = THIS_MODULE,
2133 .show = udp4_seq_show,
2137 static int __net_init udp4_proc_init_net(struct net *net)
2139 return udp_proc_register(net, &udp4_seq_afinfo);
2142 static void __net_exit udp4_proc_exit_net(struct net *net)
2144 udp_proc_unregister(net, &udp4_seq_afinfo);
2147 static struct pernet_operations udp4_net_ops = {
2148 .init = udp4_proc_init_net,
2149 .exit = udp4_proc_exit_net,
2152 int __init udp4_proc_init(void)
2154 return register_pernet_subsys(&udp4_net_ops);
2157 void udp4_proc_exit(void)
2159 unregister_pernet_subsys(&udp4_net_ops);
2161 #endif /* CONFIG_PROC_FS */
2163 static __initdata unsigned long uhash_entries;
2164 static int __init set_uhash_entries(char *str)
2168 uhash_entries = simple_strtoul(str, &str, 0);
2169 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2170 uhash_entries = UDP_HTABLE_SIZE_MIN;
2173 __setup("uhash_entries=", set_uhash_entries);
2175 void __init udp_table_init(struct udp_table *table, const char *name)
2179 if (!CONFIG_BASE_SMALL)
2180 table->hash = alloc_large_system_hash(name,
2181 2 * sizeof(struct udp_hslot),
2183 21, /* one slot per 2 MB */
2189 * Make sure hash table has the minimum size
2191 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2192 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
2193 2 * sizeof(struct udp_hslot), GFP_KERNEL);
2196 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2197 table->mask = UDP_HTABLE_SIZE_MIN - 1;
2199 table->hash2 = table->hash + (table->mask + 1);
2200 for (i = 0; i <= table->mask; i++) {
2201 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2202 table->hash[i].count = 0;
2203 spin_lock_init(&table->hash[i].lock);
2205 for (i = 0; i <= table->mask; i++) {
2206 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2207 table->hash2[i].count = 0;
2208 spin_lock_init(&table->hash2[i].lock);
2212 void __init udp_init(void)
2214 unsigned long limit;
2216 udp_table_init(&udp_table, "UDP");
2217 limit = nr_free_buffer_pages() / 8;
2218 limit = max(limit, 128UL);
2219 sysctl_udp_mem[0] = limit / 4 * 3;
2220 sysctl_udp_mem[1] = limit;
2221 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2223 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2224 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2227 int udp4_ufo_send_check(struct sk_buff *skb)
2229 const struct iphdr *iph;
2232 if (!pskb_may_pull(skb, sizeof(*uh)))
2238 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2240 skb->csum_start = skb_transport_header(skb) - skb->head;
2241 skb->csum_offset = offsetof(struct udphdr, check);
2242 skb->ip_summed = CHECKSUM_PARTIAL;
2246 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, u32 features)
2248 struct sk_buff *segs = ERR_PTR(-EINVAL);
2253 mss = skb_shinfo(skb)->gso_size;
2254 if (unlikely(skb->len <= mss))
2257 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2258 /* Packet is from an untrusted source, reset gso_segs. */
2259 int type = skb_shinfo(skb)->gso_type;
2261 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2262 !(type & (SKB_GSO_UDP))))
2265 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2271 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2272 * do checksum of UDP packets sent as multiple IP fragments.
2274 offset = skb_checksum_start_offset(skb);
2275 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2276 offset += skb->csum_offset;
2277 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2278 skb->ip_summed = CHECKSUM_NONE;
2280 /* Fragment the skb. IP headers of the fragments are updated in
2281 * inet_gso_segment()
2283 segs = skb_segment(skb, features);