2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The User Datagram Protocol (UDP).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
80 #include <asm/system.h>
81 #include <asm/uaccess.h>
82 #include <asm/ioctls.h>
83 #include <linux/bootmem.h>
84 #include <linux/highmem.h>
85 #include <linux/swap.h>
86 #include <linux/types.h>
87 #include <linux/fcntl.h>
88 #include <linux/module.h>
89 #include <linux/socket.h>
90 #include <linux/sockios.h>
91 #include <linux/igmp.h>
93 #include <linux/errno.h>
94 #include <linux/timer.h>
96 #include <linux/inet.h>
97 #include <linux/netdevice.h>
98 #include <net/tcp_states.h>
99 #include <linux/skbuff.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <net/net_namespace.h>
103 #include <net/icmp.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include "udp_impl.h"
109 struct udp_table udp_table __read_mostly;
110 EXPORT_SYMBOL(udp_table);
112 int sysctl_udp_mem[3] __read_mostly;
113 EXPORT_SYMBOL(sysctl_udp_mem);
115 int sysctl_udp_rmem_min __read_mostly;
116 EXPORT_SYMBOL(sysctl_udp_rmem_min);
118 int sysctl_udp_wmem_min __read_mostly;
119 EXPORT_SYMBOL(sysctl_udp_wmem_min);
121 atomic_t udp_memory_allocated;
122 EXPORT_SYMBOL(udp_memory_allocated);
124 #define MAX_UDP_PORTS 65536
125 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
127 static int udp_lib_lport_inuse(struct net *net, __u16 num,
128 const struct udp_hslot *hslot,
129 unsigned long *bitmap,
131 int (*saddr_comp)(const struct sock *sk1,
132 const struct sock *sk2),
136 struct hlist_nulls_node *node;
138 sk_nulls_for_each(sk2, node, &hslot->head)
139 if (net_eq(sock_net(sk2), net) &&
141 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
142 (!sk2->sk_reuse || !sk->sk_reuse) &&
143 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
144 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
145 (*saddr_comp)(sk, sk2)) {
147 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
156 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
158 * @sk: socket struct in question
159 * @snum: port number to look up
160 * @saddr_comp: AF-dependent comparison of bound local IP addresses
162 int udp_lib_get_port(struct sock *sk, unsigned short snum,
163 int (*saddr_comp)(const struct sock *sk1,
164 const struct sock *sk2))
166 struct udp_hslot *hslot, *hslot2;
167 struct udp_table *udptable = sk->sk_prot->h.udp_table;
169 struct net *net = sock_net(sk);
172 int low, high, remaining;
174 unsigned short first, last;
175 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
177 inet_get_local_port_range(&low, &high);
178 remaining = (high - low) + 1;
181 first = (((u64)rand * remaining) >> 32) + low;
183 * force rand to be an odd multiple of UDP_HTABLE_SIZE
185 rand = (rand | 1) * (udptable->mask + 1);
186 for (last = first + udptable->mask + 1;
189 hslot = udp_hashslot(udptable, net, first);
190 bitmap_zero(bitmap, PORTS_PER_CHAIN);
191 spin_lock_bh(&hslot->lock);
192 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
193 saddr_comp, udptable->log);
197 * Iterate on all possible values of snum for this hash.
198 * Using steps of an odd multiple of UDP_HTABLE_SIZE
199 * give us randomization and full range coverage.
202 if (low <= snum && snum <= high &&
203 !test_bit(snum >> udptable->log, bitmap))
206 } while (snum != first);
207 spin_unlock_bh(&hslot->lock);
211 hslot = udp_hashslot(udptable, net, snum);
212 spin_lock_bh(&hslot->lock);
213 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
218 inet_sk(sk)->inet_num = snum;
219 udp_sk(sk)->udp_port_hash = snum;
220 udp_sk(sk)->udp_portaddr_hash ^= snum;
221 if (sk_unhashed(sk)) {
222 sk_nulls_add_node_rcu(sk, &hslot->head);
224 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
226 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
227 spin_lock(&hslot2->lock);
228 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
231 spin_unlock(&hslot2->lock);
235 spin_unlock_bh(&hslot->lock);
239 EXPORT_SYMBOL(udp_lib_get_port);
241 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
243 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
245 return (!ipv6_only_sock(sk2) &&
246 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
247 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
250 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
253 return jhash_1word(saddr, net_hash_mix(net)) ^ port;
256 int udp_v4_get_port(struct sock *sk, unsigned short snum)
258 /* precompute partial secondary hash */
259 udp_sk(sk)->udp_portaddr_hash =
260 udp4_portaddr_hash(sock_net(sk),
261 inet_sk(sk)->inet_rcv_saddr,
263 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
266 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
268 __be16 sport, __be32 daddr, __be16 dport, int dif)
272 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
273 !ipv6_only_sock(sk)) {
274 struct inet_sock *inet = inet_sk(sk);
276 score = (sk->sk_family == PF_INET ? 1 : 0);
277 if (inet->inet_rcv_saddr) {
278 if (inet->inet_rcv_saddr != daddr)
282 if (inet->inet_daddr) {
283 if (inet->inet_daddr != saddr)
287 if (inet->inet_dport) {
288 if (inet->inet_dport != sport)
292 if (sk->sk_bound_dev_if) {
293 if (sk->sk_bound_dev_if != dif)
302 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
304 #define SCORE2_MAX (1 + 2 + 2 + 2)
305 static inline int compute_score2(struct sock *sk, struct net *net,
306 __be32 saddr, __be16 sport,
307 __be32 daddr, unsigned int hnum, int dif)
311 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
312 struct inet_sock *inet = inet_sk(sk);
314 if (inet->inet_rcv_saddr != daddr)
316 if (inet->inet_num != hnum)
319 score = (sk->sk_family == PF_INET ? 1 : 0);
320 if (inet->inet_daddr) {
321 if (inet->inet_daddr != saddr)
325 if (inet->inet_dport) {
326 if (inet->inet_dport != sport)
330 if (sk->sk_bound_dev_if) {
331 if (sk->sk_bound_dev_if != dif)
339 #define udp_portaddr_for_each_entry_rcu(__sk, node, list) \
340 hlist_nulls_for_each_entry_rcu(__sk, node, list, __sk_common.skc_portaddr_node)
342 /* called with read_rcu_lock() */
343 static struct sock *udp4_lib_lookup2(struct net *net,
344 __be32 saddr, __be16 sport,
345 __be32 daddr, unsigned int hnum, int dif,
346 struct udp_hslot *hslot2, unsigned int slot2)
348 struct sock *sk, *result;
349 struct hlist_nulls_node *node;
355 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
356 score = compute_score2(sk, net, saddr, sport,
358 if (score > badness) {
361 if (score == SCORE2_MAX)
366 * if the nulls value we got at the end of this lookup is
367 * not the expected one, we must restart lookup.
368 * We probably met an item that was moved to another chain.
370 if (get_nulls_value(node) != slot2)
375 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
377 else if (unlikely(compute_score2(result, net, saddr, sport,
378 daddr, hnum, dif) < badness)) {
386 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
387 * harder than this. -DaveM
389 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
390 __be16 sport, __be32 daddr, __be16 dport,
391 int dif, struct udp_table *udptable)
393 struct sock *sk, *result;
394 struct hlist_nulls_node *node;
395 unsigned short hnum = ntohs(dport);
396 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
397 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
401 if (hslot->count > 10) {
402 hash2 = udp4_portaddr_hash(net, daddr, hnum);
403 slot2 = hash2 & udptable->mask;
404 hslot2 = &udptable->hash2[slot2];
405 if (hslot->count < hslot2->count)
408 result = udp4_lib_lookup2(net, saddr, sport,
412 hash2 = udp4_portaddr_hash(net, INADDR_ANY, hnum);
413 slot2 = hash2 & udptable->mask;
414 hslot2 = &udptable->hash2[slot2];
415 if (hslot->count < hslot2->count)
418 result = udp4_lib_lookup2(net, INADDR_ANY, sport,
428 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
429 score = compute_score(sk, net, saddr, hnum, sport,
431 if (score > badness) {
437 * if the nulls value we got at the end of this lookup is
438 * not the expected one, we must restart lookup.
439 * We probably met an item that was moved to another chain.
441 if (get_nulls_value(node) != slot)
445 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
447 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
448 daddr, dport, dif) < badness)) {
457 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
458 __be16 sport, __be16 dport,
459 struct udp_table *udptable)
462 const struct iphdr *iph = ip_hdr(skb);
464 if (unlikely(sk = skb_steal_sock(skb)))
467 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
468 iph->daddr, dport, inet_iif(skb),
472 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
473 __be32 daddr, __be16 dport, int dif)
475 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
477 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
479 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
480 __be16 loc_port, __be32 loc_addr,
481 __be16 rmt_port, __be32 rmt_addr,
484 struct hlist_nulls_node *node;
486 unsigned short hnum = ntohs(loc_port);
488 sk_nulls_for_each_from(s, node) {
489 struct inet_sock *inet = inet_sk(s);
491 if (!net_eq(sock_net(s), net) ||
492 udp_sk(s)->udp_port_hash != hnum ||
493 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
494 (inet->inet_dport != rmt_port && inet->inet_dport) ||
495 (inet->inet_rcv_saddr &&
496 inet->inet_rcv_saddr != loc_addr) ||
498 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
500 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
510 * This routine is called by the ICMP module when it gets some
511 * sort of error condition. If err < 0 then the socket should
512 * be closed and the error returned to the user. If err > 0
513 * it's just the icmp type << 8 | icmp code.
514 * Header points to the ip header of the error packet. We move
515 * on past this. Then (as it used to claim before adjustment)
516 * header points to the first 8 bytes of the udp header. We need
517 * to find the appropriate port.
520 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
522 struct inet_sock *inet;
523 struct iphdr *iph = (struct iphdr *)skb->data;
524 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
525 const int type = icmp_hdr(skb)->type;
526 const int code = icmp_hdr(skb)->code;
530 struct net *net = dev_net(skb->dev);
532 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
533 iph->saddr, uh->source, skb->dev->ifindex, udptable);
535 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
536 return; /* No socket for error */
545 case ICMP_TIME_EXCEEDED:
548 case ICMP_SOURCE_QUENCH:
550 case ICMP_PARAMETERPROB:
554 case ICMP_DEST_UNREACH:
555 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
556 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
564 if (code <= NR_ICMP_UNREACH) {
565 harderr = icmp_err_convert[code].fatal;
566 err = icmp_err_convert[code].errno;
572 * RFC1122: OK. Passes ICMP errors back to application, as per
575 if (!inet->recverr) {
576 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
579 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
582 sk->sk_error_report(sk);
587 void udp_err(struct sk_buff *skb, u32 info)
589 __udp4_lib_err(skb, info, &udp_table);
593 * Throw away all pending data and cancel the corking. Socket is locked.
595 void udp_flush_pending_frames(struct sock *sk)
597 struct udp_sock *up = udp_sk(sk);
602 ip_flush_pending_frames(sk);
605 EXPORT_SYMBOL(udp_flush_pending_frames);
608 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
609 * @sk: socket we are sending on
610 * @skb: sk_buff containing the filled-in UDP header
611 * (checksum field must be zeroed out)
613 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
614 __be32 src, __be32 dst, int len)
617 struct udphdr *uh = udp_hdr(skb);
620 if (skb_queue_len(&sk->sk_write_queue) == 1) {
622 * Only one fragment on the socket.
624 skb->csum_start = skb_transport_header(skb) - skb->head;
625 skb->csum_offset = offsetof(struct udphdr, check);
626 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
629 * HW-checksum won't work as there are two or more
630 * fragments on the socket so that all csums of sk_buffs
633 offset = skb_transport_offset(skb);
634 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
636 skb->ip_summed = CHECKSUM_NONE;
638 skb_queue_walk(&sk->sk_write_queue, skb) {
639 csum = csum_add(csum, skb->csum);
642 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
644 uh->check = CSUM_MANGLED_0;
649 * Push out all pending data as one UDP datagram. Socket is locked.
651 static int udp_push_pending_frames(struct sock *sk)
653 struct udp_sock *up = udp_sk(sk);
654 struct inet_sock *inet = inet_sk(sk);
655 struct flowi *fl = &inet->cork.fl;
659 int is_udplite = IS_UDPLITE(sk);
662 /* Grab the skbuff where UDP header space exists. */
663 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
667 * Create a UDP header
670 uh->source = fl->fl_ip_sport;
671 uh->dest = fl->fl_ip_dport;
672 uh->len = htons(up->len);
675 if (is_udplite) /* UDP-Lite */
676 csum = udplite_csum_outgoing(sk, skb);
678 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
680 skb->ip_summed = CHECKSUM_NONE;
683 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
685 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len);
688 } else /* `normal' UDP */
689 csum = udp_csum_outgoing(sk, skb);
691 /* add protocol-dependent pseudo-header */
692 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
693 sk->sk_protocol, csum);
695 uh->check = CSUM_MANGLED_0;
698 err = ip_push_pending_frames(sk);
700 if (err == -ENOBUFS && !inet->recverr) {
701 UDP_INC_STATS_USER(sock_net(sk),
702 UDP_MIB_SNDBUFERRORS, is_udplite);
706 UDP_INC_STATS_USER(sock_net(sk),
707 UDP_MIB_OUTDATAGRAMS, is_udplite);
714 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
717 struct inet_sock *inet = inet_sk(sk);
718 struct udp_sock *up = udp_sk(sk);
720 struct ipcm_cookie ipc;
721 struct rtable *rt = NULL;
724 __be32 daddr, faddr, saddr;
727 int err, is_udplite = IS_UDPLITE(sk);
728 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
729 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
738 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
746 * There are pending frames.
747 * The socket lock must be held while it's corked.
750 if (likely(up->pending)) {
751 if (unlikely(up->pending != AF_INET)) {
759 ulen += sizeof(struct udphdr);
762 * Get and verify the address.
765 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
766 if (msg->msg_namelen < sizeof(*usin))
768 if (usin->sin_family != AF_INET) {
769 if (usin->sin_family != AF_UNSPEC)
770 return -EAFNOSUPPORT;
773 daddr = usin->sin_addr.s_addr;
774 dport = usin->sin_port;
778 if (sk->sk_state != TCP_ESTABLISHED)
779 return -EDESTADDRREQ;
780 daddr = inet->inet_daddr;
781 dport = inet->inet_dport;
782 /* Open fast path for connected socket.
783 Route will not be used, if at least one option is set.
787 ipc.addr = inet->inet_saddr;
789 ipc.oif = sk->sk_bound_dev_if;
790 err = sock_tx_timestamp(msg, sk, &ipc.shtx);
793 if (msg->msg_controllen) {
794 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
805 ipc.addr = faddr = daddr;
807 if (ipc.opt && ipc.opt->srr) {
810 faddr = ipc.opt->faddr;
813 tos = RT_TOS(inet->tos);
814 if (sock_flag(sk, SOCK_LOCALROUTE) ||
815 (msg->msg_flags & MSG_DONTROUTE) ||
816 (ipc.opt && ipc.opt->is_strictroute)) {
821 if (ipv4_is_multicast(daddr)) {
823 ipc.oif = inet->mc_index;
825 saddr = inet->mc_addr;
830 rt = (struct rtable *)sk_dst_check(sk, 0);
833 struct flowi fl = { .oif = ipc.oif,
839 .proto = sk->sk_protocol,
840 .flags = inet_sk_flowi_flags(sk),
842 { .sport = inet->inet_sport,
843 .dport = dport } } };
844 struct net *net = sock_net(sk);
846 security_sk_classify_flow(sk, &fl);
847 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
849 if (err == -ENETUNREACH)
850 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
855 if ((rt->rt_flags & RTCF_BROADCAST) &&
856 !sock_flag(sk, SOCK_BROADCAST))
859 sk_dst_set(sk, dst_clone(&rt->u.dst));
862 if (msg->msg_flags&MSG_CONFIRM)
868 daddr = ipc.addr = rt->rt_dst;
871 if (unlikely(up->pending)) {
872 /* The socket is already corked while preparing it. */
873 /* ... which is an evident application bug. --ANK */
876 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
881 * Now cork the socket to pend data.
883 inet->cork.fl.fl4_dst = daddr;
884 inet->cork.fl.fl_ip_dport = dport;
885 inet->cork.fl.fl4_src = saddr;
886 inet->cork.fl.fl_ip_sport = inet->inet_sport;
887 up->pending = AF_INET;
891 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
892 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
893 sizeof(struct udphdr), &ipc, &rt,
894 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
896 udp_flush_pending_frames(sk);
898 err = udp_push_pending_frames(sk);
899 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
910 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
911 * ENOBUFS might not be good (it's not tunable per se), but otherwise
912 * we don't have a good statistic (IpOutDiscards but it can be too many
913 * things). We could add another new stat but at least for now that
914 * seems like overkill.
916 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
917 UDP_INC_STATS_USER(sock_net(sk),
918 UDP_MIB_SNDBUFERRORS, is_udplite);
923 dst_confirm(&rt->u.dst);
924 if (!(msg->msg_flags&MSG_PROBE) || len)
925 goto back_from_confirm;
929 EXPORT_SYMBOL(udp_sendmsg);
931 int udp_sendpage(struct sock *sk, struct page *page, int offset,
932 size_t size, int flags)
934 struct udp_sock *up = udp_sk(sk);
938 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
940 /* Call udp_sendmsg to specify destination address which
941 * sendpage interface can't pass.
942 * This will succeed only when the socket is connected.
944 ret = udp_sendmsg(NULL, sk, &msg, 0);
951 if (unlikely(!up->pending)) {
954 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
958 ret = ip_append_page(sk, page, offset, size, flags);
959 if (ret == -EOPNOTSUPP) {
961 return sock_no_sendpage(sk->sk_socket, page, offset,
965 udp_flush_pending_frames(sk);
970 if (!(up->corkflag || (flags&MSG_MORE)))
971 ret = udp_push_pending_frames(sk);
981 * first_packet_length - return length of first packet in receive queue
984 * Drops all bad checksum frames, until a valid one is found.
985 * Returns the length of found skb, or 0 if none is found.
987 static unsigned int first_packet_length(struct sock *sk)
989 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
993 __skb_queue_head_init(&list_kill);
995 spin_lock_bh(&rcvq->lock);
996 while ((skb = skb_peek(rcvq)) != NULL &&
997 udp_lib_checksum_complete(skb)) {
998 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1000 atomic_inc(&sk->sk_drops);
1001 __skb_unlink(skb, rcvq);
1002 __skb_queue_tail(&list_kill, skb);
1004 res = skb ? skb->len : 0;
1005 spin_unlock_bh(&rcvq->lock);
1007 if (!skb_queue_empty(&list_kill)) {
1009 __skb_queue_purge(&list_kill);
1010 sk_mem_reclaim_partial(sk);
1017 * IOCTL requests applicable to the UDP protocol
1020 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1025 int amount = sk_wmem_alloc_get(sk);
1027 return put_user(amount, (int __user *)arg);
1032 unsigned int amount = first_packet_length(sk);
1036 * We will only return the amount
1037 * of this packet since that is all
1038 * that will be read.
1040 amount -= sizeof(struct udphdr);
1042 return put_user(amount, (int __user *)arg);
1046 return -ENOIOCTLCMD;
1051 EXPORT_SYMBOL(udp_ioctl);
1054 * This should be easy, if there is something there we
1055 * return it, otherwise we block.
1058 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1059 size_t len, int noblock, int flags, int *addr_len)
1061 struct inet_sock *inet = inet_sk(sk);
1062 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1063 struct sk_buff *skb;
1064 unsigned int ulen, copied;
1067 int is_udplite = IS_UDPLITE(sk);
1070 * Check any passed addresses
1073 *addr_len = sizeof(*sin);
1075 if (flags & MSG_ERRQUEUE)
1076 return ip_recv_error(sk, msg, len);
1079 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1084 ulen = skb->len - sizeof(struct udphdr);
1088 else if (copied < ulen)
1089 msg->msg_flags |= MSG_TRUNC;
1092 * If checksum is needed at all, try to do it while copying the
1093 * data. If the data is truncated, or if we only want a partial
1094 * coverage checksum (UDP-Lite), do it before the copy.
1097 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1098 if (udp_lib_checksum_complete(skb))
1102 if (skb_csum_unnecessary(skb))
1103 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1104 msg->msg_iov, copied);
1106 err = skb_copy_and_csum_datagram_iovec(skb,
1107 sizeof(struct udphdr),
1118 UDP_INC_STATS_USER(sock_net(sk),
1119 UDP_MIB_INDATAGRAMS, is_udplite);
1121 sock_recv_ts_and_drops(msg, sk, skb);
1123 /* Copy the address. */
1125 sin->sin_family = AF_INET;
1126 sin->sin_port = udp_hdr(skb)->source;
1127 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1128 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1130 if (inet->cmsg_flags)
1131 ip_cmsg_recv(msg, skb);
1134 if (flags & MSG_TRUNC)
1138 skb_free_datagram_locked(sk, skb);
1144 if (!skb_kill_datagram(sk, skb, flags))
1145 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1154 int udp_disconnect(struct sock *sk, int flags)
1156 struct inet_sock *inet = inet_sk(sk);
1158 * 1003.1g - break association.
1161 sk->sk_state = TCP_CLOSE;
1162 inet->inet_daddr = 0;
1163 inet->inet_dport = 0;
1164 sk->sk_bound_dev_if = 0;
1165 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1166 inet_reset_saddr(sk);
1168 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1169 sk->sk_prot->unhash(sk);
1170 inet->inet_sport = 0;
1175 EXPORT_SYMBOL(udp_disconnect);
1177 void udp_lib_unhash(struct sock *sk)
1179 if (sk_hashed(sk)) {
1180 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1181 struct udp_hslot *hslot, *hslot2;
1183 hslot = udp_hashslot(udptable, sock_net(sk),
1184 udp_sk(sk)->udp_port_hash);
1185 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1187 spin_lock_bh(&hslot->lock);
1188 if (sk_nulls_del_node_init_rcu(sk)) {
1190 inet_sk(sk)->inet_num = 0;
1191 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1193 spin_lock(&hslot2->lock);
1194 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1196 spin_unlock(&hslot2->lock);
1198 spin_unlock_bh(&hslot->lock);
1201 EXPORT_SYMBOL(udp_lib_unhash);
1203 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1205 int rc = sock_queue_rcv_skb(sk, skb);
1208 int is_udplite = IS_UDPLITE(sk);
1210 /* Note that an ENOMEM error is charged twice */
1212 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1214 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1226 * >0: "udp encap" protocol resubmission
1228 * Note that in the success and error cases, the skb is assumed to
1229 * have either been requeued or freed.
1231 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1233 struct udp_sock *up = udp_sk(sk);
1235 int is_udplite = IS_UDPLITE(sk);
1238 * Charge it to the socket, dropping if the queue is full.
1240 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1244 if (up->encap_type) {
1246 * This is an encapsulation socket so pass the skb to
1247 * the socket's udp_encap_rcv() hook. Otherwise, just
1248 * fall through and pass this up the UDP socket.
1249 * up->encap_rcv() returns the following value:
1250 * =0 if skb was successfully passed to the encap
1251 * handler or was discarded by it.
1252 * >0 if skb should be passed on to UDP.
1253 * <0 if skb should be resubmitted as proto -N
1256 /* if we're overly short, let UDP handle it */
1257 if (skb->len > sizeof(struct udphdr) &&
1258 up->encap_rcv != NULL) {
1261 ret = (*up->encap_rcv)(sk, skb);
1263 UDP_INC_STATS_BH(sock_net(sk),
1264 UDP_MIB_INDATAGRAMS,
1270 /* FALLTHROUGH -- it's a UDP Packet */
1274 * UDP-Lite specific tests, ignored on UDP sockets
1276 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1279 * MIB statistics other than incrementing the error count are
1280 * disabled for the following two types of errors: these depend
1281 * on the application settings, not on the functioning of the
1282 * protocol stack as such.
1284 * RFC 3828 here recommends (sec 3.3): "There should also be a
1285 * way ... to ... at least let the receiving application block
1286 * delivery of packets with coverage values less than a value
1287 * provided by the application."
1289 if (up->pcrlen == 0) { /* full coverage was set */
1290 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1291 "%d while full coverage %d requested\n",
1292 UDP_SKB_CB(skb)->cscov, skb->len);
1295 /* The next case involves violating the min. coverage requested
1296 * by the receiver. This is subtle: if receiver wants x and x is
1297 * greater than the buffersize/MTU then receiver will complain
1298 * that it wants x while sender emits packets of smaller size y.
1299 * Therefore the above ...()->partial_cov statement is essential.
1301 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1302 LIMIT_NETDEBUG(KERN_WARNING
1303 "UDPLITE: coverage %d too small, need min %d\n",
1304 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1309 if (sk->sk_filter) {
1310 if (udp_lib_checksum_complete(skb))
1317 if (!sock_owned_by_user(sk))
1318 rc = __udp_queue_rcv_skb(sk, skb);
1320 sk_add_backlog(sk, skb);
1326 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1327 atomic_inc(&sk->sk_drops);
1333 static void flush_stack(struct sock **stack, unsigned int count,
1334 struct sk_buff *skb, unsigned int final)
1337 struct sk_buff *skb1 = NULL;
1339 for (i = 0; i < count; i++) {
1340 if (likely(skb1 == NULL))
1341 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1343 if (skb1 && udp_queue_rcv_skb(stack[i], skb1) <= 0)
1351 * Multicasts and broadcasts go to each listener.
1353 * Note: called only from the BH handler context.
1355 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1357 __be32 saddr, __be32 daddr,
1358 struct udp_table *udptable)
1360 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1361 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1363 unsigned int i, count = 0;
1365 spin_lock(&hslot->lock);
1366 sk = sk_nulls_head(&hslot->head);
1367 dif = skb->dev->ifindex;
1368 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1370 stack[count++] = sk;
1371 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1372 daddr, uh->source, saddr, dif);
1373 if (unlikely(count == ARRAY_SIZE(stack))) {
1376 flush_stack(stack, count, skb, ~0);
1381 * before releasing chain lock, we must take a reference on sockets
1383 for (i = 0; i < count; i++)
1384 sock_hold(stack[i]);
1386 spin_unlock(&hslot->lock);
1389 * do the slow work with no lock held
1392 flush_stack(stack, count, skb, count - 1);
1394 for (i = 0; i < count; i++)
1402 /* Initialize UDP checksum. If exited with zero value (success),
1403 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1404 * Otherwise, csum completion requires chacksumming packet body,
1405 * including udp header and folding it to skb->csum.
1407 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1410 const struct iphdr *iph;
1413 UDP_SKB_CB(skb)->partial_cov = 0;
1414 UDP_SKB_CB(skb)->cscov = skb->len;
1416 if (proto == IPPROTO_UDPLITE) {
1417 err = udplite_checksum_init(skb, uh);
1423 if (uh->check == 0) {
1424 skb->ip_summed = CHECKSUM_UNNECESSARY;
1425 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1426 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1428 skb->ip_summed = CHECKSUM_UNNECESSARY;
1430 if (!skb_csum_unnecessary(skb))
1431 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1432 skb->len, proto, 0);
1433 /* Probably, we should checksum udp header (it should be in cache
1434 * in any case) and data in tiny packets (< rx copybreak).
1441 * All we need to do is get the socket, and then do a checksum.
1444 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1449 unsigned short ulen;
1450 struct rtable *rt = skb_rtable(skb);
1451 __be32 saddr, daddr;
1452 struct net *net = dev_net(skb->dev);
1455 * Validate the packet.
1457 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1458 goto drop; /* No space for header. */
1461 ulen = ntohs(uh->len);
1462 if (ulen > skb->len)
1465 if (proto == IPPROTO_UDP) {
1466 /* UDP validates ulen. */
1467 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1472 if (udp4_csum_init(skb, uh, proto))
1475 saddr = ip_hdr(skb)->saddr;
1476 daddr = ip_hdr(skb)->daddr;
1478 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1479 return __udp4_lib_mcast_deliver(net, skb, uh,
1480 saddr, daddr, udptable);
1482 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1485 int ret = udp_queue_rcv_skb(sk, skb);
1488 /* a return value > 0 means to resubmit the input, but
1489 * it wants the return to be -protocol, or 0
1496 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1500 /* No socket. Drop packet silently, if checksum is wrong */
1501 if (udp_lib_checksum_complete(skb))
1504 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1505 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1508 * Hmm. We got an UDP packet to a port to which we
1509 * don't wanna listen. Ignore it.
1515 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1516 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1527 * RFC1122: OK. Discards the bad packet silently (as far as
1528 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1530 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1531 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1538 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1543 int udp_rcv(struct sk_buff *skb)
1545 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1548 void udp_destroy_sock(struct sock *sk)
1551 udp_flush_pending_frames(sk);
1556 * Socket option code for UDP
1558 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1559 char __user *optval, unsigned int optlen,
1560 int (*push_pending_frames)(struct sock *))
1562 struct udp_sock *up = udp_sk(sk);
1565 int is_udplite = IS_UDPLITE(sk);
1567 if (optlen < sizeof(int))
1570 if (get_user(val, (int __user *)optval))
1580 (*push_pending_frames)(sk);
1588 case UDP_ENCAP_ESPINUDP:
1589 case UDP_ENCAP_ESPINUDP_NON_IKE:
1590 up->encap_rcv = xfrm4_udp_encap_rcv;
1592 case UDP_ENCAP_L2TPINUDP:
1593 up->encap_type = val;
1602 * UDP-Lite's partial checksum coverage (RFC 3828).
1604 /* The sender sets actual checksum coverage length via this option.
1605 * The case coverage > packet length is handled by send module. */
1606 case UDPLITE_SEND_CSCOV:
1607 if (!is_udplite) /* Disable the option on UDP sockets */
1608 return -ENOPROTOOPT;
1609 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1611 else if (val > USHORT_MAX)
1614 up->pcflag |= UDPLITE_SEND_CC;
1617 /* The receiver specifies a minimum checksum coverage value. To make
1618 * sense, this should be set to at least 8 (as done below). If zero is
1619 * used, this again means full checksum coverage. */
1620 case UDPLITE_RECV_CSCOV:
1621 if (!is_udplite) /* Disable the option on UDP sockets */
1622 return -ENOPROTOOPT;
1623 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1625 else if (val > USHORT_MAX)
1628 up->pcflag |= UDPLITE_RECV_CC;
1638 EXPORT_SYMBOL(udp_lib_setsockopt);
1640 int udp_setsockopt(struct sock *sk, int level, int optname,
1641 char __user *optval, unsigned int optlen)
1643 if (level == SOL_UDP || level == SOL_UDPLITE)
1644 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1645 udp_push_pending_frames);
1646 return ip_setsockopt(sk, level, optname, optval, optlen);
1649 #ifdef CONFIG_COMPAT
1650 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1651 char __user *optval, unsigned int optlen)
1653 if (level == SOL_UDP || level == SOL_UDPLITE)
1654 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1655 udp_push_pending_frames);
1656 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1660 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1661 char __user *optval, int __user *optlen)
1663 struct udp_sock *up = udp_sk(sk);
1666 if (get_user(len, optlen))
1669 len = min_t(unsigned int, len, sizeof(int));
1680 val = up->encap_type;
1683 /* The following two cannot be changed on UDP sockets, the return is
1684 * always 0 (which corresponds to the full checksum coverage of UDP). */
1685 case UDPLITE_SEND_CSCOV:
1689 case UDPLITE_RECV_CSCOV:
1694 return -ENOPROTOOPT;
1697 if (put_user(len, optlen))
1699 if (copy_to_user(optval, &val, len))
1703 EXPORT_SYMBOL(udp_lib_getsockopt);
1705 int udp_getsockopt(struct sock *sk, int level, int optname,
1706 char __user *optval, int __user *optlen)
1708 if (level == SOL_UDP || level == SOL_UDPLITE)
1709 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1710 return ip_getsockopt(sk, level, optname, optval, optlen);
1713 #ifdef CONFIG_COMPAT
1714 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1715 char __user *optval, int __user *optlen)
1717 if (level == SOL_UDP || level == SOL_UDPLITE)
1718 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1719 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1723 * udp_poll - wait for a UDP event.
1724 * @file - file struct
1726 * @wait - poll table
1728 * This is same as datagram poll, except for the special case of
1729 * blocking sockets. If application is using a blocking fd
1730 * and a packet with checksum error is in the queue;
1731 * then it could get return from select indicating data available
1732 * but then block when reading it. Add special case code
1733 * to work around these arguably broken applications.
1735 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1737 unsigned int mask = datagram_poll(file, sock, wait);
1738 struct sock *sk = sock->sk;
1740 /* Check for false positives due to checksum errors */
1741 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1742 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1743 mask &= ~(POLLIN | POLLRDNORM);
1748 EXPORT_SYMBOL(udp_poll);
1750 struct proto udp_prot = {
1752 .owner = THIS_MODULE,
1753 .close = udp_lib_close,
1754 .connect = ip4_datagram_connect,
1755 .disconnect = udp_disconnect,
1757 .destroy = udp_destroy_sock,
1758 .setsockopt = udp_setsockopt,
1759 .getsockopt = udp_getsockopt,
1760 .sendmsg = udp_sendmsg,
1761 .recvmsg = udp_recvmsg,
1762 .sendpage = udp_sendpage,
1763 .backlog_rcv = __udp_queue_rcv_skb,
1764 .hash = udp_lib_hash,
1765 .unhash = udp_lib_unhash,
1766 .get_port = udp_v4_get_port,
1767 .memory_allocated = &udp_memory_allocated,
1768 .sysctl_mem = sysctl_udp_mem,
1769 .sysctl_wmem = &sysctl_udp_wmem_min,
1770 .sysctl_rmem = &sysctl_udp_rmem_min,
1771 .obj_size = sizeof(struct udp_sock),
1772 .slab_flags = SLAB_DESTROY_BY_RCU,
1773 .h.udp_table = &udp_table,
1774 #ifdef CONFIG_COMPAT
1775 .compat_setsockopt = compat_udp_setsockopt,
1776 .compat_getsockopt = compat_udp_getsockopt,
1779 EXPORT_SYMBOL(udp_prot);
1781 /* ------------------------------------------------------------------------ */
1782 #ifdef CONFIG_PROC_FS
1784 static struct sock *udp_get_first(struct seq_file *seq, int start)
1787 struct udp_iter_state *state = seq->private;
1788 struct net *net = seq_file_net(seq);
1790 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1792 struct hlist_nulls_node *node;
1793 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1795 if (hlist_nulls_empty(&hslot->head))
1798 spin_lock_bh(&hslot->lock);
1799 sk_nulls_for_each(sk, node, &hslot->head) {
1800 if (!net_eq(sock_net(sk), net))
1802 if (sk->sk_family == state->family)
1805 spin_unlock_bh(&hslot->lock);
1812 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1814 struct udp_iter_state *state = seq->private;
1815 struct net *net = seq_file_net(seq);
1818 sk = sk_nulls_next(sk);
1819 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1822 if (state->bucket <= state->udp_table->mask)
1823 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1824 return udp_get_first(seq, state->bucket + 1);
1829 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1831 struct sock *sk = udp_get_first(seq, 0);
1834 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1836 return pos ? NULL : sk;
1839 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1841 struct udp_iter_state *state = seq->private;
1842 state->bucket = MAX_UDP_PORTS;
1844 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1847 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1851 if (v == SEQ_START_TOKEN)
1852 sk = udp_get_idx(seq, 0);
1854 sk = udp_get_next(seq, v);
1860 static void udp_seq_stop(struct seq_file *seq, void *v)
1862 struct udp_iter_state *state = seq->private;
1864 if (state->bucket <= state->udp_table->mask)
1865 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1868 static int udp_seq_open(struct inode *inode, struct file *file)
1870 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1871 struct udp_iter_state *s;
1874 err = seq_open_net(inode, file, &afinfo->seq_ops,
1875 sizeof(struct udp_iter_state));
1879 s = ((struct seq_file *)file->private_data)->private;
1880 s->family = afinfo->family;
1881 s->udp_table = afinfo->udp_table;
1885 /* ------------------------------------------------------------------------ */
1886 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1888 struct proc_dir_entry *p;
1891 afinfo->seq_fops.open = udp_seq_open;
1892 afinfo->seq_fops.read = seq_read;
1893 afinfo->seq_fops.llseek = seq_lseek;
1894 afinfo->seq_fops.release = seq_release_net;
1896 afinfo->seq_ops.start = udp_seq_start;
1897 afinfo->seq_ops.next = udp_seq_next;
1898 afinfo->seq_ops.stop = udp_seq_stop;
1900 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1901 &afinfo->seq_fops, afinfo);
1906 EXPORT_SYMBOL(udp_proc_register);
1908 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1910 proc_net_remove(net, afinfo->name);
1912 EXPORT_SYMBOL(udp_proc_unregister);
1914 /* ------------------------------------------------------------------------ */
1915 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1916 int bucket, int *len)
1918 struct inet_sock *inet = inet_sk(sp);
1919 __be32 dest = inet->inet_daddr;
1920 __be32 src = inet->inet_rcv_saddr;
1921 __u16 destp = ntohs(inet->inet_dport);
1922 __u16 srcp = ntohs(inet->inet_sport);
1924 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
1925 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1926 bucket, src, srcp, dest, destp, sp->sk_state,
1927 sk_wmem_alloc_get(sp),
1928 sk_rmem_alloc_get(sp),
1929 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1930 atomic_read(&sp->sk_refcnt), sp,
1931 atomic_read(&sp->sk_drops), len);
1934 int udp4_seq_show(struct seq_file *seq, void *v)
1936 if (v == SEQ_START_TOKEN)
1937 seq_printf(seq, "%-127s\n",
1938 " sl local_address rem_address st tx_queue "
1939 "rx_queue tr tm->when retrnsmt uid timeout "
1940 "inode ref pointer drops");
1942 struct udp_iter_state *state = seq->private;
1945 udp4_format_sock(v, seq, state->bucket, &len);
1946 seq_printf(seq, "%*s\n", 127 - len, "");
1951 /* ------------------------------------------------------------------------ */
1952 static struct udp_seq_afinfo udp4_seq_afinfo = {
1955 .udp_table = &udp_table,
1957 .owner = THIS_MODULE,
1960 .show = udp4_seq_show,
1964 static int udp4_proc_init_net(struct net *net)
1966 return udp_proc_register(net, &udp4_seq_afinfo);
1969 static void udp4_proc_exit_net(struct net *net)
1971 udp_proc_unregister(net, &udp4_seq_afinfo);
1974 static struct pernet_operations udp4_net_ops = {
1975 .init = udp4_proc_init_net,
1976 .exit = udp4_proc_exit_net,
1979 int __init udp4_proc_init(void)
1981 return register_pernet_subsys(&udp4_net_ops);
1984 void udp4_proc_exit(void)
1986 unregister_pernet_subsys(&udp4_net_ops);
1988 #endif /* CONFIG_PROC_FS */
1990 static __initdata unsigned long uhash_entries;
1991 static int __init set_uhash_entries(char *str)
1995 uhash_entries = simple_strtoul(str, &str, 0);
1996 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
1997 uhash_entries = UDP_HTABLE_SIZE_MIN;
2000 __setup("uhash_entries=", set_uhash_entries);
2002 void __init udp_table_init(struct udp_table *table, const char *name)
2006 if (!CONFIG_BASE_SMALL)
2007 table->hash = alloc_large_system_hash(name,
2008 2 * sizeof(struct udp_hslot),
2010 21, /* one slot per 2 MB */
2016 * Make sure hash table has the minimum size
2018 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2019 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
2020 2 * sizeof(struct udp_hslot), GFP_KERNEL);
2023 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2024 table->mask = UDP_HTABLE_SIZE_MIN - 1;
2026 table->hash2 = table->hash + (table->mask + 1);
2027 for (i = 0; i <= table->mask; i++) {
2028 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2029 table->hash[i].count = 0;
2030 spin_lock_init(&table->hash[i].lock);
2032 for (i = 0; i <= table->mask; i++) {
2033 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2034 table->hash2[i].count = 0;
2035 spin_lock_init(&table->hash2[i].lock);
2039 void __init udp_init(void)
2041 unsigned long nr_pages, limit;
2043 udp_table_init(&udp_table, "UDP");
2044 /* Set the pressure threshold up by the same strategy of TCP. It is a
2045 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
2046 * toward zero with the amount of memory, with a floor of 128 pages.
2048 nr_pages = totalram_pages - totalhigh_pages;
2049 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
2050 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
2051 limit = max(limit, 128UL);
2052 sysctl_udp_mem[0] = limit / 4 * 3;
2053 sysctl_udp_mem[1] = limit;
2054 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2056 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2057 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2060 int udp4_ufo_send_check(struct sk_buff *skb)
2062 const struct iphdr *iph;
2065 if (!pskb_may_pull(skb, sizeof(*uh)))
2071 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2073 skb->csum_start = skb_transport_header(skb) - skb->head;
2074 skb->csum_offset = offsetof(struct udphdr, check);
2075 skb->ip_summed = CHECKSUM_PARTIAL;
2079 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features)
2081 struct sk_buff *segs = ERR_PTR(-EINVAL);
2086 mss = skb_shinfo(skb)->gso_size;
2087 if (unlikely(skb->len <= mss))
2090 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2091 /* Packet is from an untrusted source, reset gso_segs. */
2092 int type = skb_shinfo(skb)->gso_type;
2094 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2095 !(type & (SKB_GSO_UDP))))
2098 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2104 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2105 * do checksum of UDP packets sent as multiple IP fragments.
2107 offset = skb->csum_start - skb_headroom(skb);
2108 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2109 offset += skb->csum_offset;
2110 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2111 skb->ip_summed = CHECKSUM_NONE;
2113 /* Fragment the skb. IP headers of the fragments are updated in
2114 * inet_gso_segment()
2116 segs = skb_segment(skb, features);