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 #define pr_fmt(fmt) "UDP: " fmt
82 #include <asm/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/bootmem.h>
85 #include <linux/highmem.h>
86 #include <linux/swap.h>
87 #include <linux/types.h>
88 #include <linux/fcntl.h>
89 #include <linux/module.h>
90 #include <linux/socket.h>
91 #include <linux/sockios.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
95 #include <linux/errno.h>
96 #include <linux/timer.h>
98 #include <linux/inet.h>
99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/route.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <trace/events/udp.h>
112 #include <linux/static_key.h>
113 #include <trace/events/skb.h>
114 #include <net/busy_poll.h>
115 #include "udp_impl.h"
116 #include <net/sock_reuseport.h>
118 struct udp_table udp_table __read_mostly;
119 EXPORT_SYMBOL(udp_table);
121 long sysctl_udp_mem[3] __read_mostly;
122 EXPORT_SYMBOL(sysctl_udp_mem);
124 int sysctl_udp_rmem_min __read_mostly;
125 EXPORT_SYMBOL(sysctl_udp_rmem_min);
127 int sysctl_udp_wmem_min __read_mostly;
128 EXPORT_SYMBOL(sysctl_udp_wmem_min);
130 atomic_long_t udp_memory_allocated;
131 EXPORT_SYMBOL(udp_memory_allocated);
133 #define MAX_UDP_PORTS 65536
134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
136 static int udp_lib_lport_inuse(struct net *net, __u16 num,
137 const struct udp_hslot *hslot,
138 unsigned long *bitmap,
140 int (*saddr_comp)(const struct sock *sk1,
141 const struct sock *sk2,
142 bool match_wildcard),
146 struct hlist_nulls_node *node;
147 kuid_t uid = sock_i_uid(sk);
149 sk_nulls_for_each(sk2, node, &hslot->head) {
150 if (net_eq(sock_net(sk2), net) &&
152 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
153 (!sk2->sk_reuse || !sk->sk_reuse) &&
154 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
155 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
156 (!sk2->sk_reuseport || !sk->sk_reuseport ||
157 rcu_access_pointer(sk->sk_reuseport_cb) ||
158 !uid_eq(uid, sock_i_uid(sk2))) &&
159 saddr_comp(sk, sk2, true)) {
162 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap);
169 * Note: we still hold spinlock of primary hash chain, so no other writer
170 * can insert/delete a socket with local_port == num
172 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
173 struct udp_hslot *hslot2,
175 int (*saddr_comp)(const struct sock *sk1,
176 const struct sock *sk2,
177 bool match_wildcard))
180 struct hlist_nulls_node *node;
181 kuid_t uid = sock_i_uid(sk);
184 spin_lock(&hslot2->lock);
185 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) {
186 if (net_eq(sock_net(sk2), net) &&
188 (udp_sk(sk2)->udp_port_hash == num) &&
189 (!sk2->sk_reuse || !sk->sk_reuse) &&
190 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
191 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
192 (!sk2->sk_reuseport || !sk->sk_reuseport ||
193 rcu_access_pointer(sk->sk_reuseport_cb) ||
194 !uid_eq(uid, sock_i_uid(sk2))) &&
195 saddr_comp(sk, sk2, true)) {
200 spin_unlock(&hslot2->lock);
204 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot,
205 int (*saddr_same)(const struct sock *sk1,
206 const struct sock *sk2,
207 bool match_wildcard))
209 struct net *net = sock_net(sk);
210 struct hlist_nulls_node *node;
211 kuid_t uid = sock_i_uid(sk);
214 sk_nulls_for_each(sk2, node, &hslot->head) {
215 if (net_eq(sock_net(sk2), net) &&
217 sk2->sk_family == sk->sk_family &&
218 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
219 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
220 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
221 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
222 (*saddr_same)(sk, sk2, false)) {
223 return reuseport_add_sock(sk, sk2);
227 /* Initial allocation may have already happened via setsockopt */
228 if (!rcu_access_pointer(sk->sk_reuseport_cb))
229 return reuseport_alloc(sk);
234 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
236 * @sk: socket struct in question
237 * @snum: port number to look up
238 * @saddr_comp: AF-dependent comparison of bound local IP addresses
239 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
242 int udp_lib_get_port(struct sock *sk, unsigned short snum,
243 int (*saddr_comp)(const struct sock *sk1,
244 const struct sock *sk2,
245 bool match_wildcard),
246 unsigned int hash2_nulladdr)
248 struct udp_hslot *hslot, *hslot2;
249 struct udp_table *udptable = sk->sk_prot->h.udp_table;
251 struct net *net = sock_net(sk);
254 int low, high, remaining;
256 unsigned short first, last;
257 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
259 inet_get_local_port_range(net, &low, &high);
260 remaining = (high - low) + 1;
262 rand = prandom_u32();
263 first = reciprocal_scale(rand, remaining) + low;
265 * force rand to be an odd multiple of UDP_HTABLE_SIZE
267 rand = (rand | 1) * (udptable->mask + 1);
268 last = first + udptable->mask + 1;
270 hslot = udp_hashslot(udptable, net, first);
271 bitmap_zero(bitmap, PORTS_PER_CHAIN);
272 spin_lock_bh(&hslot->lock);
273 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
274 saddr_comp, udptable->log);
278 * Iterate on all possible values of snum for this hash.
279 * Using steps of an odd multiple of UDP_HTABLE_SIZE
280 * give us randomization and full range coverage.
283 if (low <= snum && snum <= high &&
284 !test_bit(snum >> udptable->log, bitmap) &&
285 !inet_is_local_reserved_port(net, snum))
288 } while (snum != first);
289 spin_unlock_bh(&hslot->lock);
290 } while (++first != last);
293 hslot = udp_hashslot(udptable, net, snum);
294 spin_lock_bh(&hslot->lock);
295 if (hslot->count > 10) {
297 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
299 slot2 &= udptable->mask;
300 hash2_nulladdr &= udptable->mask;
302 hslot2 = udp_hashslot2(udptable, slot2);
303 if (hslot->count < hslot2->count)
304 goto scan_primary_hash;
306 exist = udp_lib_lport_inuse2(net, snum, hslot2,
308 if (!exist && (hash2_nulladdr != slot2)) {
309 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
310 exist = udp_lib_lport_inuse2(net, snum, hslot2,
319 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
324 inet_sk(sk)->inet_num = snum;
325 udp_sk(sk)->udp_port_hash = snum;
326 udp_sk(sk)->udp_portaddr_hash ^= snum;
327 if (sk_unhashed(sk)) {
328 if (sk->sk_reuseport &&
329 udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
330 inet_sk(sk)->inet_num = 0;
331 udp_sk(sk)->udp_port_hash = 0;
332 udp_sk(sk)->udp_portaddr_hash ^= snum;
336 sk_nulls_add_node_rcu(sk, &hslot->head);
338 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
340 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
341 spin_lock(&hslot2->lock);
342 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
345 spin_unlock(&hslot2->lock);
349 spin_unlock_bh(&hslot->lock);
353 EXPORT_SYMBOL(udp_lib_get_port);
355 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
356 * match_wildcard == false: addresses must be exactly the same, i.e.
357 * 0.0.0.0 only equals to 0.0.0.0
359 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
362 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
364 if (!ipv6_only_sock(sk2)) {
365 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
367 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
368 return match_wildcard;
373 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
376 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
379 int udp_v4_get_port(struct sock *sk, unsigned short snum)
381 unsigned int hash2_nulladdr =
382 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
383 unsigned int hash2_partial =
384 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
386 /* precompute partial secondary hash */
387 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
388 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
391 static inline int compute_score(struct sock *sk, struct net *net,
392 __be32 saddr, unsigned short hnum, __be16 sport,
393 __be32 daddr, __be16 dport, int dif)
396 struct inet_sock *inet;
398 if (!net_eq(sock_net(sk), net) ||
399 udp_sk(sk)->udp_port_hash != hnum ||
403 score = (sk->sk_family == PF_INET) ? 2 : 1;
406 if (inet->inet_rcv_saddr) {
407 if (inet->inet_rcv_saddr != daddr)
412 if (inet->inet_daddr) {
413 if (inet->inet_daddr != saddr)
418 if (inet->inet_dport) {
419 if (inet->inet_dport != sport)
424 if (sk->sk_bound_dev_if) {
425 if (sk->sk_bound_dev_if != dif)
429 if (sk->sk_incoming_cpu == raw_smp_processor_id())
435 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
437 static inline int compute_score2(struct sock *sk, struct net *net,
438 __be32 saddr, __be16 sport,
439 __be32 daddr, unsigned int hnum, int dif)
442 struct inet_sock *inet;
444 if (!net_eq(sock_net(sk), net) ||
450 if (inet->inet_rcv_saddr != daddr ||
451 inet->inet_num != hnum)
454 score = (sk->sk_family == PF_INET) ? 2 : 1;
456 if (inet->inet_daddr) {
457 if (inet->inet_daddr != saddr)
462 if (inet->inet_dport) {
463 if (inet->inet_dport != sport)
468 if (sk->sk_bound_dev_if) {
469 if (sk->sk_bound_dev_if != dif)
474 if (sk->sk_incoming_cpu == raw_smp_processor_id())
480 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
481 const __u16 lport, const __be32 faddr,
484 static u32 udp_ehash_secret __read_mostly;
486 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
488 return __inet_ehashfn(laddr, lport, faddr, fport,
489 udp_ehash_secret + net_hash_mix(net));
492 /* called with read_rcu_lock() */
493 static struct sock *udp4_lib_lookup2(struct net *net,
494 __be32 saddr, __be16 sport,
495 __be32 daddr, unsigned int hnum, int dif,
496 struct udp_hslot *hslot2, unsigned int slot2,
499 struct sock *sk, *result;
500 struct hlist_nulls_node *node;
501 int score, badness, matches = 0, reuseport = 0;
507 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
508 score = compute_score2(sk, net, saddr, sport,
510 if (score > badness) {
513 reuseport = sk->sk_reuseport;
516 hash = udp_ehashfn(net, daddr, hnum,
518 sk2 = reuseport_select_sock(sk, hash, skb,
519 sizeof(struct udphdr));
526 } else if (score == badness && reuseport) {
528 if (reciprocal_scale(hash, matches) == 0)
530 hash = next_pseudo_random32(hash);
534 * if the nulls value we got at the end of this lookup is
535 * not the expected one, we must restart lookup.
536 * We probably met an item that was moved to another chain.
538 if (get_nulls_value(node) != slot2)
542 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
544 else if (unlikely(compute_score2(result, net, saddr, sport,
545 daddr, hnum, dif) < badness)) {
553 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
554 * harder than this. -DaveM
556 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
557 __be16 sport, __be32 daddr, __be16 dport,
558 int dif, struct udp_table *udptable, struct sk_buff *skb)
560 struct sock *sk, *result;
561 struct hlist_nulls_node *node;
562 unsigned short hnum = ntohs(dport);
563 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
564 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
565 int score, badness, matches = 0, reuseport = 0;
569 if (hslot->count > 10) {
570 hash2 = udp4_portaddr_hash(net, daddr, hnum);
571 slot2 = hash2 & udptable->mask;
572 hslot2 = &udptable->hash2[slot2];
573 if (hslot->count < hslot2->count)
576 result = udp4_lib_lookup2(net, saddr, sport,
580 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
581 slot2 = hash2 & udptable->mask;
582 hslot2 = &udptable->hash2[slot2];
583 if (hslot->count < hslot2->count)
586 result = udp4_lib_lookup2(net, saddr, sport,
587 htonl(INADDR_ANY), hnum, dif,
596 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
597 score = compute_score(sk, net, saddr, hnum, sport,
599 if (score > badness) {
602 reuseport = sk->sk_reuseport;
605 hash = udp_ehashfn(net, daddr, hnum,
607 sk2 = reuseport_select_sock(sk, hash, skb,
608 sizeof(struct udphdr));
615 } else if (score == badness && reuseport) {
617 if (reciprocal_scale(hash, matches) == 0)
619 hash = next_pseudo_random32(hash);
623 * if the nulls value we got at the end of this lookup is
624 * not the expected one, we must restart lookup.
625 * We probably met an item that was moved to another chain.
627 if (get_nulls_value(node) != slot)
632 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
634 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
635 daddr, dport, dif) < badness)) {
643 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
645 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
646 __be16 sport, __be16 dport,
647 struct udp_table *udptable)
649 const struct iphdr *iph = ip_hdr(skb);
651 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
652 iph->daddr, dport, inet_iif(skb),
656 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
657 __be32 daddr, __be16 dport, int dif)
659 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif,
662 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
664 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
665 __be16 loc_port, __be32 loc_addr,
666 __be16 rmt_port, __be32 rmt_addr,
667 int dif, unsigned short hnum)
669 struct inet_sock *inet = inet_sk(sk);
671 if (!net_eq(sock_net(sk), net) ||
672 udp_sk(sk)->udp_port_hash != hnum ||
673 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
674 (inet->inet_dport != rmt_port && inet->inet_dport) ||
675 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
676 ipv6_only_sock(sk) ||
677 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
679 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
685 * This routine is called by the ICMP module when it gets some
686 * sort of error condition. If err < 0 then the socket should
687 * be closed and the error returned to the user. If err > 0
688 * it's just the icmp type << 8 | icmp code.
689 * Header points to the ip header of the error packet. We move
690 * on past this. Then (as it used to claim before adjustment)
691 * header points to the first 8 bytes of the udp header. We need
692 * to find the appropriate port.
695 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
697 struct inet_sock *inet;
698 const struct iphdr *iph = (const struct iphdr *)skb->data;
699 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
700 const int type = icmp_hdr(skb)->type;
701 const int code = icmp_hdr(skb)->code;
705 struct net *net = dev_net(skb->dev);
707 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
708 iph->saddr, uh->source, skb->dev->ifindex, udptable,
711 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
712 return; /* No socket for error */
721 case ICMP_TIME_EXCEEDED:
724 case ICMP_SOURCE_QUENCH:
726 case ICMP_PARAMETERPROB:
730 case ICMP_DEST_UNREACH:
731 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
732 ipv4_sk_update_pmtu(skb, sk, info);
733 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
741 if (code <= NR_ICMP_UNREACH) {
742 harderr = icmp_err_convert[code].fatal;
743 err = icmp_err_convert[code].errno;
747 ipv4_sk_redirect(skb, sk);
752 * RFC1122: OK. Passes ICMP errors back to application, as per
755 if (!inet->recverr) {
756 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
759 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
762 sk->sk_error_report(sk);
767 void udp_err(struct sk_buff *skb, u32 info)
769 __udp4_lib_err(skb, info, &udp_table);
773 * Throw away all pending data and cancel the corking. Socket is locked.
775 void udp_flush_pending_frames(struct sock *sk)
777 struct udp_sock *up = udp_sk(sk);
782 ip_flush_pending_frames(sk);
785 EXPORT_SYMBOL(udp_flush_pending_frames);
788 * udp4_hwcsum - handle outgoing HW checksumming
789 * @skb: sk_buff containing the filled-in UDP header
790 * (checksum field must be zeroed out)
791 * @src: source IP address
792 * @dst: destination IP address
794 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
796 struct udphdr *uh = udp_hdr(skb);
797 int offset = skb_transport_offset(skb);
798 int len = skb->len - offset;
802 if (!skb_has_frag_list(skb)) {
804 * Only one fragment on the socket.
806 skb->csum_start = skb_transport_header(skb) - skb->head;
807 skb->csum_offset = offsetof(struct udphdr, check);
808 uh->check = ~csum_tcpudp_magic(src, dst, len,
811 struct sk_buff *frags;
814 * HW-checksum won't work as there are two or more
815 * fragments on the socket so that all csums of sk_buffs
818 skb_walk_frags(skb, frags) {
819 csum = csum_add(csum, frags->csum);
823 csum = skb_checksum(skb, offset, hlen, csum);
824 skb->ip_summed = CHECKSUM_NONE;
826 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
828 uh->check = CSUM_MANGLED_0;
831 EXPORT_SYMBOL_GPL(udp4_hwcsum);
833 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
834 * for the simple case like when setting the checksum for a UDP tunnel.
836 void udp_set_csum(bool nocheck, struct sk_buff *skb,
837 __be32 saddr, __be32 daddr, int len)
839 struct udphdr *uh = udp_hdr(skb);
843 else if (skb_is_gso(skb))
844 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
845 else if (skb_dst(skb) && skb_dst(skb)->dev &&
846 (skb_dst(skb)->dev->features &
847 (NETIF_F_IP_CSUM | NETIF_F_HW_CSUM))) {
849 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
851 skb->ip_summed = CHECKSUM_PARTIAL;
852 skb->csum_start = skb_transport_header(skb) - skb->head;
853 skb->csum_offset = offsetof(struct udphdr, check);
854 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
858 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
861 csum = skb_checksum(skb, 0, len, 0);
862 uh->check = udp_v4_check(len, saddr, daddr, csum);
864 uh->check = CSUM_MANGLED_0;
866 skb->ip_summed = CHECKSUM_UNNECESSARY;
869 EXPORT_SYMBOL(udp_set_csum);
871 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
873 struct sock *sk = skb->sk;
874 struct inet_sock *inet = inet_sk(sk);
877 int is_udplite = IS_UDPLITE(sk);
878 int offset = skb_transport_offset(skb);
879 int len = skb->len - offset;
883 * Create a UDP header
886 uh->source = inet->inet_sport;
887 uh->dest = fl4->fl4_dport;
888 uh->len = htons(len);
891 if (is_udplite) /* UDP-Lite */
892 csum = udplite_csum(skb);
894 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
896 skb->ip_summed = CHECKSUM_NONE;
899 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
901 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
905 csum = udp_csum(skb);
907 /* add protocol-dependent pseudo-header */
908 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
909 sk->sk_protocol, csum);
911 uh->check = CSUM_MANGLED_0;
914 err = ip_send_skb(sock_net(sk), skb);
916 if (err == -ENOBUFS && !inet->recverr) {
917 UDP_INC_STATS_USER(sock_net(sk),
918 UDP_MIB_SNDBUFERRORS, is_udplite);
922 UDP_INC_STATS_USER(sock_net(sk),
923 UDP_MIB_OUTDATAGRAMS, is_udplite);
928 * Push out all pending data as one UDP datagram. Socket is locked.
930 int udp_push_pending_frames(struct sock *sk)
932 struct udp_sock *up = udp_sk(sk);
933 struct inet_sock *inet = inet_sk(sk);
934 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
938 skb = ip_finish_skb(sk, fl4);
942 err = udp_send_skb(skb, fl4);
949 EXPORT_SYMBOL(udp_push_pending_frames);
951 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
953 struct inet_sock *inet = inet_sk(sk);
954 struct udp_sock *up = udp_sk(sk);
955 struct flowi4 fl4_stack;
958 struct ipcm_cookie ipc;
959 struct rtable *rt = NULL;
962 __be32 daddr, faddr, saddr;
965 int err, is_udplite = IS_UDPLITE(sk);
966 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
967 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
969 struct ip_options_data opt_copy;
978 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
986 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
988 fl4 = &inet->cork.fl.u.ip4;
991 * There are pending frames.
992 * The socket lock must be held while it's corked.
995 if (likely(up->pending)) {
996 if (unlikely(up->pending != AF_INET)) {
1000 goto do_append_data;
1004 ulen += sizeof(struct udphdr);
1007 * Get and verify the address.
1009 if (msg->msg_name) {
1010 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1011 if (msg->msg_namelen < sizeof(*usin))
1013 if (usin->sin_family != AF_INET) {
1014 if (usin->sin_family != AF_UNSPEC)
1015 return -EAFNOSUPPORT;
1018 daddr = usin->sin_addr.s_addr;
1019 dport = usin->sin_port;
1023 if (sk->sk_state != TCP_ESTABLISHED)
1024 return -EDESTADDRREQ;
1025 daddr = inet->inet_daddr;
1026 dport = inet->inet_dport;
1027 /* Open fast path for connected socket.
1028 Route will not be used, if at least one option is set.
1032 ipc.addr = inet->inet_saddr;
1034 ipc.oif = sk->sk_bound_dev_if;
1036 sock_tx_timestamp(sk, &ipc.tx_flags);
1038 if (msg->msg_controllen) {
1039 err = ip_cmsg_send(sock_net(sk), msg, &ipc,
1040 sk->sk_family == AF_INET6);
1048 struct ip_options_rcu *inet_opt;
1051 inet_opt = rcu_dereference(inet->inet_opt);
1053 memcpy(&opt_copy, inet_opt,
1054 sizeof(*inet_opt) + inet_opt->opt.optlen);
1055 ipc.opt = &opt_copy.opt;
1061 ipc.addr = faddr = daddr;
1063 if (ipc.opt && ipc.opt->opt.srr) {
1066 faddr = ipc.opt->opt.faddr;
1069 tos = get_rttos(&ipc, inet);
1070 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1071 (msg->msg_flags & MSG_DONTROUTE) ||
1072 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1077 if (ipv4_is_multicast(daddr)) {
1079 ipc.oif = inet->mc_index;
1081 saddr = inet->mc_addr;
1083 } else if (!ipc.oif)
1084 ipc.oif = inet->uc_index;
1087 rt = (struct rtable *)sk_dst_check(sk, 0);
1090 struct net *net = sock_net(sk);
1091 __u8 flow_flags = inet_sk_flowi_flags(sk);
1095 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1096 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1098 faddr, saddr, dport, inet->inet_sport);
1100 if (!saddr && ipc.oif)
1101 l3mdev_get_saddr(net, ipc.oif, fl4);
1103 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1104 rt = ip_route_output_flow(net, fl4, sk);
1108 if (err == -ENETUNREACH)
1109 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1114 if ((rt->rt_flags & RTCF_BROADCAST) &&
1115 !sock_flag(sk, SOCK_BROADCAST))
1118 sk_dst_set(sk, dst_clone(&rt->dst));
1121 if (msg->msg_flags&MSG_CONFIRM)
1127 daddr = ipc.addr = fl4->daddr;
1129 /* Lockless fast path for the non-corking case. */
1131 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1132 sizeof(struct udphdr), &ipc, &rt,
1135 if (!IS_ERR_OR_NULL(skb))
1136 err = udp_send_skb(skb, fl4);
1141 if (unlikely(up->pending)) {
1142 /* The socket is already corked while preparing it. */
1143 /* ... which is an evident application bug. --ANK */
1146 net_dbg_ratelimited("cork app bug 2\n");
1151 * Now cork the socket to pend data.
1153 fl4 = &inet->cork.fl.u.ip4;
1156 fl4->fl4_dport = dport;
1157 fl4->fl4_sport = inet->inet_sport;
1158 up->pending = AF_INET;
1162 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1163 sizeof(struct udphdr), &ipc, &rt,
1164 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1166 udp_flush_pending_frames(sk);
1168 err = udp_push_pending_frames(sk);
1169 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1180 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1181 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1182 * we don't have a good statistic (IpOutDiscards but it can be too many
1183 * things). We could add another new stat but at least for now that
1184 * seems like overkill.
1186 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1187 UDP_INC_STATS_USER(sock_net(sk),
1188 UDP_MIB_SNDBUFERRORS, is_udplite);
1193 dst_confirm(&rt->dst);
1194 if (!(msg->msg_flags&MSG_PROBE) || len)
1195 goto back_from_confirm;
1199 EXPORT_SYMBOL(udp_sendmsg);
1201 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1202 size_t size, int flags)
1204 struct inet_sock *inet = inet_sk(sk);
1205 struct udp_sock *up = udp_sk(sk);
1208 if (flags & MSG_SENDPAGE_NOTLAST)
1212 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1214 /* Call udp_sendmsg to specify destination address which
1215 * sendpage interface can't pass.
1216 * This will succeed only when the socket is connected.
1218 ret = udp_sendmsg(sk, &msg, 0);
1225 if (unlikely(!up->pending)) {
1228 net_dbg_ratelimited("udp cork app bug 3\n");
1232 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1233 page, offset, size, flags);
1234 if (ret == -EOPNOTSUPP) {
1236 return sock_no_sendpage(sk->sk_socket, page, offset,
1240 udp_flush_pending_frames(sk);
1245 if (!(up->corkflag || (flags&MSG_MORE)))
1246 ret = udp_push_pending_frames(sk);
1255 * first_packet_length - return length of first packet in receive queue
1258 * Drops all bad checksum frames, until a valid one is found.
1259 * Returns the length of found skb, or 0 if none is found.
1261 static unsigned int first_packet_length(struct sock *sk)
1263 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1264 struct sk_buff *skb;
1267 __skb_queue_head_init(&list_kill);
1269 spin_lock_bh(&rcvq->lock);
1270 while ((skb = skb_peek(rcvq)) != NULL &&
1271 udp_lib_checksum_complete(skb)) {
1272 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1274 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1276 atomic_inc(&sk->sk_drops);
1277 __skb_unlink(skb, rcvq);
1278 __skb_queue_tail(&list_kill, skb);
1280 res = skb ? skb->len : 0;
1281 spin_unlock_bh(&rcvq->lock);
1283 if (!skb_queue_empty(&list_kill)) {
1284 bool slow = lock_sock_fast(sk);
1286 __skb_queue_purge(&list_kill);
1287 sk_mem_reclaim_partial(sk);
1288 unlock_sock_fast(sk, slow);
1294 * IOCTL requests applicable to the UDP protocol
1297 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1302 int amount = sk_wmem_alloc_get(sk);
1304 return put_user(amount, (int __user *)arg);
1309 unsigned int amount = first_packet_length(sk);
1313 * We will only return the amount
1314 * of this packet since that is all
1315 * that will be read.
1317 amount -= sizeof(struct udphdr);
1319 return put_user(amount, (int __user *)arg);
1323 return -ENOIOCTLCMD;
1328 EXPORT_SYMBOL(udp_ioctl);
1331 * This should be easy, if there is something there we
1332 * return it, otherwise we block.
1335 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1336 int flags, int *addr_len)
1338 struct inet_sock *inet = inet_sk(sk);
1339 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1340 struct sk_buff *skb;
1341 unsigned int ulen, copied;
1342 int peeked, off = 0;
1344 int is_udplite = IS_UDPLITE(sk);
1345 bool checksum_valid = false;
1348 if (flags & MSG_ERRQUEUE)
1349 return ip_recv_error(sk, msg, len, addr_len);
1352 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1353 &peeked, &off, &err);
1357 ulen = skb->len - sizeof(struct udphdr);
1361 else if (copied < ulen)
1362 msg->msg_flags |= MSG_TRUNC;
1365 * If checksum is needed at all, try to do it while copying the
1366 * data. If the data is truncated, or if we only want a partial
1367 * coverage checksum (UDP-Lite), do it before the copy.
1370 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1371 checksum_valid = !udp_lib_checksum_complete(skb);
1372 if (!checksum_valid)
1376 if (checksum_valid || skb_csum_unnecessary(skb))
1377 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
1380 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr),
1387 if (unlikely(err)) {
1388 trace_kfree_skb(skb, udp_recvmsg);
1390 atomic_inc(&sk->sk_drops);
1391 UDP_INC_STATS_USER(sock_net(sk),
1392 UDP_MIB_INERRORS, is_udplite);
1398 UDP_INC_STATS_USER(sock_net(sk),
1399 UDP_MIB_INDATAGRAMS, is_udplite);
1401 sock_recv_ts_and_drops(msg, sk, skb);
1403 /* Copy the address. */
1405 sin->sin_family = AF_INET;
1406 sin->sin_port = udp_hdr(skb)->source;
1407 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1408 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1409 *addr_len = sizeof(*sin);
1411 if (inet->cmsg_flags)
1412 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr));
1415 if (flags & MSG_TRUNC)
1419 skb_free_datagram_locked(sk, skb);
1424 slow = lock_sock_fast(sk);
1425 if (!skb_kill_datagram(sk, skb, flags)) {
1426 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1427 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1429 unlock_sock_fast(sk, slow);
1431 /* starting over for a new packet, but check if we need to yield */
1433 msg->msg_flags &= ~MSG_TRUNC;
1437 int udp_disconnect(struct sock *sk, int flags)
1439 struct inet_sock *inet = inet_sk(sk);
1441 * 1003.1g - break association.
1444 sk->sk_state = TCP_CLOSE;
1445 inet->inet_daddr = 0;
1446 inet->inet_dport = 0;
1447 sock_rps_reset_rxhash(sk);
1448 sk->sk_bound_dev_if = 0;
1449 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1450 inet_reset_saddr(sk);
1452 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1453 sk->sk_prot->unhash(sk);
1454 inet->inet_sport = 0;
1459 EXPORT_SYMBOL(udp_disconnect);
1461 void udp_lib_unhash(struct sock *sk)
1463 if (sk_hashed(sk)) {
1464 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1465 struct udp_hslot *hslot, *hslot2;
1467 hslot = udp_hashslot(udptable, sock_net(sk),
1468 udp_sk(sk)->udp_port_hash);
1469 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1471 spin_lock_bh(&hslot->lock);
1472 if (rcu_access_pointer(sk->sk_reuseport_cb))
1473 reuseport_detach_sock(sk);
1474 if (sk_nulls_del_node_init_rcu(sk)) {
1476 inet_sk(sk)->inet_num = 0;
1477 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1479 spin_lock(&hslot2->lock);
1480 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1482 spin_unlock(&hslot2->lock);
1484 spin_unlock_bh(&hslot->lock);
1487 EXPORT_SYMBOL(udp_lib_unhash);
1490 * inet_rcv_saddr was changed, we must rehash secondary hash
1492 void udp_lib_rehash(struct sock *sk, u16 newhash)
1494 if (sk_hashed(sk)) {
1495 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1496 struct udp_hslot *hslot, *hslot2, *nhslot2;
1498 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1499 nhslot2 = udp_hashslot2(udptable, newhash);
1500 udp_sk(sk)->udp_portaddr_hash = newhash;
1502 if (hslot2 != nhslot2 ||
1503 rcu_access_pointer(sk->sk_reuseport_cb)) {
1504 hslot = udp_hashslot(udptable, sock_net(sk),
1505 udp_sk(sk)->udp_port_hash);
1506 /* we must lock primary chain too */
1507 spin_lock_bh(&hslot->lock);
1508 if (rcu_access_pointer(sk->sk_reuseport_cb))
1509 reuseport_detach_sock(sk);
1511 if (hslot2 != nhslot2) {
1512 spin_lock(&hslot2->lock);
1513 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1515 spin_unlock(&hslot2->lock);
1517 spin_lock(&nhslot2->lock);
1518 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1521 spin_unlock(&nhslot2->lock);
1524 spin_unlock_bh(&hslot->lock);
1528 EXPORT_SYMBOL(udp_lib_rehash);
1530 static void udp_v4_rehash(struct sock *sk)
1532 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1533 inet_sk(sk)->inet_rcv_saddr,
1534 inet_sk(sk)->inet_num);
1535 udp_lib_rehash(sk, new_hash);
1538 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1542 if (inet_sk(sk)->inet_daddr) {
1543 sock_rps_save_rxhash(sk, skb);
1544 sk_mark_napi_id(sk, skb);
1545 sk_incoming_cpu_update(sk);
1548 rc = sock_queue_rcv_skb(sk, skb);
1550 int is_udplite = IS_UDPLITE(sk);
1552 /* Note that an ENOMEM error is charged twice */
1554 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1556 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1558 trace_udp_fail_queue_rcv_skb(rc, sk);
1566 static struct static_key udp_encap_needed __read_mostly;
1567 void udp_encap_enable(void)
1569 if (!static_key_enabled(&udp_encap_needed))
1570 static_key_slow_inc(&udp_encap_needed);
1572 EXPORT_SYMBOL(udp_encap_enable);
1577 * >0: "udp encap" protocol resubmission
1579 * Note that in the success and error cases, the skb is assumed to
1580 * have either been requeued or freed.
1582 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1584 struct udp_sock *up = udp_sk(sk);
1586 int is_udplite = IS_UDPLITE(sk);
1589 * Charge it to the socket, dropping if the queue is full.
1591 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1595 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1596 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1599 * This is an encapsulation socket so pass the skb to
1600 * the socket's udp_encap_rcv() hook. Otherwise, just
1601 * fall through and pass this up the UDP socket.
1602 * up->encap_rcv() returns the following value:
1603 * =0 if skb was successfully passed to the encap
1604 * handler or was discarded by it.
1605 * >0 if skb should be passed on to UDP.
1606 * <0 if skb should be resubmitted as proto -N
1609 /* if we're overly short, let UDP handle it */
1610 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1611 if (skb->len > sizeof(struct udphdr) && encap_rcv) {
1614 /* Verify checksum before giving to encap */
1615 if (udp_lib_checksum_complete(skb))
1618 ret = encap_rcv(sk, skb);
1620 UDP_INC_STATS_BH(sock_net(sk),
1621 UDP_MIB_INDATAGRAMS,
1627 /* FALLTHROUGH -- it's a UDP Packet */
1631 * UDP-Lite specific tests, ignored on UDP sockets
1633 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1636 * MIB statistics other than incrementing the error count are
1637 * disabled for the following two types of errors: these depend
1638 * on the application settings, not on the functioning of the
1639 * protocol stack as such.
1641 * RFC 3828 here recommends (sec 3.3): "There should also be a
1642 * way ... to ... at least let the receiving application block
1643 * delivery of packets with coverage values less than a value
1644 * provided by the application."
1646 if (up->pcrlen == 0) { /* full coverage was set */
1647 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1648 UDP_SKB_CB(skb)->cscov, skb->len);
1651 /* The next case involves violating the min. coverage requested
1652 * by the receiver. This is subtle: if receiver wants x and x is
1653 * greater than the buffersize/MTU then receiver will complain
1654 * that it wants x while sender emits packets of smaller size y.
1655 * Therefore the above ...()->partial_cov statement is essential.
1657 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1658 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1659 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1664 if (rcu_access_pointer(sk->sk_filter) &&
1665 udp_lib_checksum_complete(skb))
1668 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1669 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1676 ipv4_pktinfo_prepare(sk, skb);
1678 if (!sock_owned_by_user(sk))
1679 rc = __udp_queue_rcv_skb(sk, skb);
1680 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1689 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1691 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1692 atomic_inc(&sk->sk_drops);
1697 static void flush_stack(struct sock **stack, unsigned int count,
1698 struct sk_buff *skb, unsigned int final)
1701 struct sk_buff *skb1 = NULL;
1704 for (i = 0; i < count; i++) {
1707 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1710 atomic_inc(&sk->sk_drops);
1711 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1713 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1717 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1726 /* For TCP sockets, sk_rx_dst is protected by socket lock
1727 * For UDP, we use xchg() to guard against concurrent changes.
1729 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1731 struct dst_entry *old;
1734 old = xchg(&sk->sk_rx_dst, dst);
1739 * Multicasts and broadcasts go to each listener.
1741 * Note: called only from the BH handler context.
1743 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1745 __be32 saddr, __be32 daddr,
1746 struct udp_table *udptable,
1749 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1750 struct hlist_nulls_node *node;
1751 unsigned short hnum = ntohs(uh->dest);
1752 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1753 int dif = skb->dev->ifindex;
1754 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
1755 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1756 bool inner_flushed = false;
1759 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1761 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1763 hslot = &udp_table.hash2[hash2];
1764 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1767 spin_lock(&hslot->lock);
1768 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
1769 if (__udp_is_mcast_sock(net, sk,
1773 if (unlikely(count == ARRAY_SIZE(stack))) {
1774 flush_stack(stack, count, skb, ~0);
1775 inner_flushed = true;
1778 stack[count++] = sk;
1783 spin_unlock(&hslot->lock);
1785 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1786 if (use_hash2 && hash2 != hash2_any) {
1792 * do the slow work with no lock held
1795 flush_stack(stack, count, skb, count - 1);
1798 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
1799 proto == IPPROTO_UDPLITE);
1805 /* Initialize UDP checksum. If exited with zero value (success),
1806 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1807 * Otherwise, csum completion requires chacksumming packet body,
1808 * including udp header and folding it to skb->csum.
1810 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1815 UDP_SKB_CB(skb)->partial_cov = 0;
1816 UDP_SKB_CB(skb)->cscov = skb->len;
1818 if (proto == IPPROTO_UDPLITE) {
1819 err = udplite_checksum_init(skb, uh);
1824 return skb_checksum_init_zero_check(skb, proto, uh->check,
1825 inet_compute_pseudo);
1829 * All we need to do is get the socket, and then do a checksum.
1832 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1837 unsigned short ulen;
1838 struct rtable *rt = skb_rtable(skb);
1839 __be32 saddr, daddr;
1840 struct net *net = dev_net(skb->dev);
1843 * Validate the packet.
1845 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1846 goto drop; /* No space for header. */
1849 ulen = ntohs(uh->len);
1850 saddr = ip_hdr(skb)->saddr;
1851 daddr = ip_hdr(skb)->daddr;
1853 if (ulen > skb->len)
1856 if (proto == IPPROTO_UDP) {
1857 /* UDP validates ulen. */
1858 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1863 if (udp4_csum_init(skb, uh, proto))
1866 sk = skb_steal_sock(skb);
1868 struct dst_entry *dst = skb_dst(skb);
1871 if (unlikely(sk->sk_rx_dst != dst))
1872 udp_sk_rx_dst_set(sk, dst);
1874 ret = udp_queue_rcv_skb(sk, skb);
1876 /* a return value > 0 means to resubmit the input, but
1877 * it wants the return to be -protocol, or 0
1884 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1885 return __udp4_lib_mcast_deliver(net, skb, uh,
1886 saddr, daddr, udptable, proto);
1888 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1892 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1893 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1894 inet_compute_pseudo);
1896 ret = udp_queue_rcv_skb(sk, skb);
1899 /* a return value > 0 means to resubmit the input, but
1900 * it wants the return to be -protocol, or 0
1907 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1911 /* No socket. Drop packet silently, if checksum is wrong */
1912 if (udp_lib_checksum_complete(skb))
1915 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1916 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1919 * Hmm. We got an UDP packet to a port to which we
1920 * don't wanna listen. Ignore it.
1926 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1927 proto == IPPROTO_UDPLITE ? "Lite" : "",
1928 &saddr, ntohs(uh->source),
1930 &daddr, ntohs(uh->dest));
1935 * RFC1122: OK. Discards the bad packet silently (as far as
1936 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1938 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1939 proto == IPPROTO_UDPLITE ? "Lite" : "",
1940 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1942 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1944 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1949 /* We can only early demux multicast if there is a single matching socket.
1950 * If more than one socket found returns NULL
1952 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1953 __be16 loc_port, __be32 loc_addr,
1954 __be16 rmt_port, __be32 rmt_addr,
1957 struct sock *sk, *result;
1958 struct hlist_nulls_node *node;
1959 unsigned short hnum = ntohs(loc_port);
1960 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1961 struct udp_hslot *hslot = &udp_table.hash[slot];
1963 /* Do not bother scanning a too big list */
1964 if (hslot->count > 10)
1971 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1972 if (__udp_is_mcast_sock(net, sk,
1981 * if the nulls value we got at the end of this lookup is
1982 * not the expected one, we must restart lookup.
1983 * We probably met an item that was moved to another chain.
1985 if (get_nulls_value(node) != slot)
1990 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1992 else if (unlikely(!__udp_is_mcast_sock(net, result,
2004 /* For unicast we should only early demux connected sockets or we can
2005 * break forwarding setups. The chains here can be long so only check
2006 * if the first socket is an exact match and if not move on.
2008 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2009 __be16 loc_port, __be32 loc_addr,
2010 __be16 rmt_port, __be32 rmt_addr,
2013 struct sock *sk, *result;
2014 struct hlist_nulls_node *node;
2015 unsigned short hnum = ntohs(loc_port);
2016 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2017 unsigned int slot2 = hash2 & udp_table.mask;
2018 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2019 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2020 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2024 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
2025 if (INET_MATCH(sk, net, acookie,
2026 rmt_addr, loc_addr, ports, dif))
2028 /* Only check first socket in chain */
2033 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
2035 else if (unlikely(!INET_MATCH(sk, net, acookie,
2046 void udp_v4_early_demux(struct sk_buff *skb)
2048 struct net *net = dev_net(skb->dev);
2049 const struct iphdr *iph;
2050 const struct udphdr *uh;
2052 struct dst_entry *dst;
2053 int dif = skb->dev->ifindex;
2056 /* validate the packet */
2057 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2063 if (skb->pkt_type == PACKET_BROADCAST ||
2064 skb->pkt_type == PACKET_MULTICAST) {
2065 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2070 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2074 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2075 uh->source, iph->saddr, dif);
2076 } else if (skb->pkt_type == PACKET_HOST) {
2077 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2078 uh->source, iph->saddr, dif);
2087 skb->destructor = sock_efree;
2088 dst = READ_ONCE(sk->sk_rx_dst);
2091 dst = dst_check(dst, 0);
2093 /* DST_NOCACHE can not be used without taking a reference */
2094 if (dst->flags & DST_NOCACHE) {
2095 if (likely(atomic_inc_not_zero(&dst->__refcnt)))
2096 skb_dst_set(skb, dst);
2098 skb_dst_set_noref(skb, dst);
2103 int udp_rcv(struct sk_buff *skb)
2105 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2108 void udp_destroy_sock(struct sock *sk)
2110 struct udp_sock *up = udp_sk(sk);
2111 bool slow = lock_sock_fast(sk);
2112 udp_flush_pending_frames(sk);
2113 unlock_sock_fast(sk, slow);
2114 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2115 void (*encap_destroy)(struct sock *sk);
2116 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2123 * Socket option code for UDP
2125 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2126 char __user *optval, unsigned int optlen,
2127 int (*push_pending_frames)(struct sock *))
2129 struct udp_sock *up = udp_sk(sk);
2132 int is_udplite = IS_UDPLITE(sk);
2134 if (optlen < sizeof(int))
2137 if (get_user(val, (int __user *)optval))
2140 valbool = val ? 1 : 0;
2149 push_pending_frames(sk);
2157 case UDP_ENCAP_ESPINUDP:
2158 case UDP_ENCAP_ESPINUDP_NON_IKE:
2159 up->encap_rcv = xfrm4_udp_encap_rcv;
2161 case UDP_ENCAP_L2TPINUDP:
2162 up->encap_type = val;
2171 case UDP_NO_CHECK6_TX:
2172 up->no_check6_tx = valbool;
2175 case UDP_NO_CHECK6_RX:
2176 up->no_check6_rx = valbool;
2180 * UDP-Lite's partial checksum coverage (RFC 3828).
2182 /* The sender sets actual checksum coverage length via this option.
2183 * The case coverage > packet length is handled by send module. */
2184 case UDPLITE_SEND_CSCOV:
2185 if (!is_udplite) /* Disable the option on UDP sockets */
2186 return -ENOPROTOOPT;
2187 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2189 else if (val > USHRT_MAX)
2192 up->pcflag |= UDPLITE_SEND_CC;
2195 /* The receiver specifies a minimum checksum coverage value. To make
2196 * sense, this should be set to at least 8 (as done below). If zero is
2197 * used, this again means full checksum coverage. */
2198 case UDPLITE_RECV_CSCOV:
2199 if (!is_udplite) /* Disable the option on UDP sockets */
2200 return -ENOPROTOOPT;
2201 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2203 else if (val > USHRT_MAX)
2206 up->pcflag |= UDPLITE_RECV_CC;
2216 EXPORT_SYMBOL(udp_lib_setsockopt);
2218 int udp_setsockopt(struct sock *sk, int level, int optname,
2219 char __user *optval, unsigned int optlen)
2221 if (level == SOL_UDP || level == SOL_UDPLITE)
2222 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2223 udp_push_pending_frames);
2224 return ip_setsockopt(sk, level, optname, optval, optlen);
2227 #ifdef CONFIG_COMPAT
2228 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2229 char __user *optval, unsigned int optlen)
2231 if (level == SOL_UDP || level == SOL_UDPLITE)
2232 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2233 udp_push_pending_frames);
2234 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2238 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2239 char __user *optval, int __user *optlen)
2241 struct udp_sock *up = udp_sk(sk);
2244 if (get_user(len, optlen))
2247 len = min_t(unsigned int, len, sizeof(int));
2258 val = up->encap_type;
2261 case UDP_NO_CHECK6_TX:
2262 val = up->no_check6_tx;
2265 case UDP_NO_CHECK6_RX:
2266 val = up->no_check6_rx;
2269 /* The following two cannot be changed on UDP sockets, the return is
2270 * always 0 (which corresponds to the full checksum coverage of UDP). */
2271 case UDPLITE_SEND_CSCOV:
2275 case UDPLITE_RECV_CSCOV:
2280 return -ENOPROTOOPT;
2283 if (put_user(len, optlen))
2285 if (copy_to_user(optval, &val, len))
2289 EXPORT_SYMBOL(udp_lib_getsockopt);
2291 int udp_getsockopt(struct sock *sk, int level, int optname,
2292 char __user *optval, int __user *optlen)
2294 if (level == SOL_UDP || level == SOL_UDPLITE)
2295 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2296 return ip_getsockopt(sk, level, optname, optval, optlen);
2299 #ifdef CONFIG_COMPAT
2300 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2301 char __user *optval, int __user *optlen)
2303 if (level == SOL_UDP || level == SOL_UDPLITE)
2304 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2305 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2309 * udp_poll - wait for a UDP event.
2310 * @file - file struct
2312 * @wait - poll table
2314 * This is same as datagram poll, except for the special case of
2315 * blocking sockets. If application is using a blocking fd
2316 * and a packet with checksum error is in the queue;
2317 * then it could get return from select indicating data available
2318 * but then block when reading it. Add special case code
2319 * to work around these arguably broken applications.
2321 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2323 unsigned int mask = datagram_poll(file, sock, wait);
2324 struct sock *sk = sock->sk;
2326 sock_rps_record_flow(sk);
2328 /* Check for false positives due to checksum errors */
2329 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2330 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2331 mask &= ~(POLLIN | POLLRDNORM);
2336 EXPORT_SYMBOL(udp_poll);
2338 struct proto udp_prot = {
2340 .owner = THIS_MODULE,
2341 .close = udp_lib_close,
2342 .connect = ip4_datagram_connect,
2343 .disconnect = udp_disconnect,
2345 .destroy = udp_destroy_sock,
2346 .setsockopt = udp_setsockopt,
2347 .getsockopt = udp_getsockopt,
2348 .sendmsg = udp_sendmsg,
2349 .recvmsg = udp_recvmsg,
2350 .sendpage = udp_sendpage,
2351 .backlog_rcv = __udp_queue_rcv_skb,
2352 .release_cb = ip4_datagram_release_cb,
2353 .hash = udp_lib_hash,
2354 .unhash = udp_lib_unhash,
2355 .rehash = udp_v4_rehash,
2356 .get_port = udp_v4_get_port,
2357 .memory_allocated = &udp_memory_allocated,
2358 .sysctl_mem = sysctl_udp_mem,
2359 .sysctl_wmem = &sysctl_udp_wmem_min,
2360 .sysctl_rmem = &sysctl_udp_rmem_min,
2361 .obj_size = sizeof(struct udp_sock),
2362 .slab_flags = SLAB_DESTROY_BY_RCU,
2363 .h.udp_table = &udp_table,
2364 #ifdef CONFIG_COMPAT
2365 .compat_setsockopt = compat_udp_setsockopt,
2366 .compat_getsockopt = compat_udp_getsockopt,
2368 .clear_sk = sk_prot_clear_portaddr_nulls,
2370 EXPORT_SYMBOL(udp_prot);
2372 /* ------------------------------------------------------------------------ */
2373 #ifdef CONFIG_PROC_FS
2375 static struct sock *udp_get_first(struct seq_file *seq, int start)
2378 struct udp_iter_state *state = seq->private;
2379 struct net *net = seq_file_net(seq);
2381 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2383 struct hlist_nulls_node *node;
2384 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2386 if (hlist_nulls_empty(&hslot->head))
2389 spin_lock_bh(&hslot->lock);
2390 sk_nulls_for_each(sk, node, &hslot->head) {
2391 if (!net_eq(sock_net(sk), net))
2393 if (sk->sk_family == state->family)
2396 spin_unlock_bh(&hslot->lock);
2403 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2405 struct udp_iter_state *state = seq->private;
2406 struct net *net = seq_file_net(seq);
2409 sk = sk_nulls_next(sk);
2410 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2413 if (state->bucket <= state->udp_table->mask)
2414 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2415 return udp_get_first(seq, state->bucket + 1);
2420 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2422 struct sock *sk = udp_get_first(seq, 0);
2425 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2427 return pos ? NULL : sk;
2430 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2432 struct udp_iter_state *state = seq->private;
2433 state->bucket = MAX_UDP_PORTS;
2435 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2438 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2442 if (v == SEQ_START_TOKEN)
2443 sk = udp_get_idx(seq, 0);
2445 sk = udp_get_next(seq, v);
2451 static void udp_seq_stop(struct seq_file *seq, void *v)
2453 struct udp_iter_state *state = seq->private;
2455 if (state->bucket <= state->udp_table->mask)
2456 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2459 int udp_seq_open(struct inode *inode, struct file *file)
2461 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2462 struct udp_iter_state *s;
2465 err = seq_open_net(inode, file, &afinfo->seq_ops,
2466 sizeof(struct udp_iter_state));
2470 s = ((struct seq_file *)file->private_data)->private;
2471 s->family = afinfo->family;
2472 s->udp_table = afinfo->udp_table;
2475 EXPORT_SYMBOL(udp_seq_open);
2477 /* ------------------------------------------------------------------------ */
2478 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2480 struct proc_dir_entry *p;
2483 afinfo->seq_ops.start = udp_seq_start;
2484 afinfo->seq_ops.next = udp_seq_next;
2485 afinfo->seq_ops.stop = udp_seq_stop;
2487 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2488 afinfo->seq_fops, afinfo);
2493 EXPORT_SYMBOL(udp_proc_register);
2495 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2497 remove_proc_entry(afinfo->name, net->proc_net);
2499 EXPORT_SYMBOL(udp_proc_unregister);
2501 /* ------------------------------------------------------------------------ */
2502 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2505 struct inet_sock *inet = inet_sk(sp);
2506 __be32 dest = inet->inet_daddr;
2507 __be32 src = inet->inet_rcv_saddr;
2508 __u16 destp = ntohs(inet->inet_dport);
2509 __u16 srcp = ntohs(inet->inet_sport);
2511 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2512 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2513 bucket, src, srcp, dest, destp, sp->sk_state,
2514 sk_wmem_alloc_get(sp),
2515 sk_rmem_alloc_get(sp),
2517 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2519 atomic_read(&sp->sk_refcnt), sp,
2520 atomic_read(&sp->sk_drops));
2523 int udp4_seq_show(struct seq_file *seq, void *v)
2525 seq_setwidth(seq, 127);
2526 if (v == SEQ_START_TOKEN)
2527 seq_puts(seq, " sl local_address rem_address st tx_queue "
2528 "rx_queue tr tm->when retrnsmt uid timeout "
2529 "inode ref pointer drops");
2531 struct udp_iter_state *state = seq->private;
2533 udp4_format_sock(v, seq, state->bucket);
2539 static const struct file_operations udp_afinfo_seq_fops = {
2540 .owner = THIS_MODULE,
2541 .open = udp_seq_open,
2543 .llseek = seq_lseek,
2544 .release = seq_release_net
2547 /* ------------------------------------------------------------------------ */
2548 static struct udp_seq_afinfo udp4_seq_afinfo = {
2551 .udp_table = &udp_table,
2552 .seq_fops = &udp_afinfo_seq_fops,
2554 .show = udp4_seq_show,
2558 static int __net_init udp4_proc_init_net(struct net *net)
2560 return udp_proc_register(net, &udp4_seq_afinfo);
2563 static void __net_exit udp4_proc_exit_net(struct net *net)
2565 udp_proc_unregister(net, &udp4_seq_afinfo);
2568 static struct pernet_operations udp4_net_ops = {
2569 .init = udp4_proc_init_net,
2570 .exit = udp4_proc_exit_net,
2573 int __init udp4_proc_init(void)
2575 return register_pernet_subsys(&udp4_net_ops);
2578 void udp4_proc_exit(void)
2580 unregister_pernet_subsys(&udp4_net_ops);
2582 #endif /* CONFIG_PROC_FS */
2584 static __initdata unsigned long uhash_entries;
2585 static int __init set_uhash_entries(char *str)
2592 ret = kstrtoul(str, 0, &uhash_entries);
2596 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2597 uhash_entries = UDP_HTABLE_SIZE_MIN;
2600 __setup("uhash_entries=", set_uhash_entries);
2602 void __init udp_table_init(struct udp_table *table, const char *name)
2606 table->hash = alloc_large_system_hash(name,
2607 2 * sizeof(struct udp_hslot),
2609 21, /* one slot per 2 MB */
2613 UDP_HTABLE_SIZE_MIN,
2616 table->hash2 = table->hash + (table->mask + 1);
2617 for (i = 0; i <= table->mask; i++) {
2618 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2619 table->hash[i].count = 0;
2620 spin_lock_init(&table->hash[i].lock);
2622 for (i = 0; i <= table->mask; i++) {
2623 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2624 table->hash2[i].count = 0;
2625 spin_lock_init(&table->hash2[i].lock);
2629 u32 udp_flow_hashrnd(void)
2631 static u32 hashrnd __read_mostly;
2633 net_get_random_once(&hashrnd, sizeof(hashrnd));
2637 EXPORT_SYMBOL(udp_flow_hashrnd);
2639 void __init udp_init(void)
2641 unsigned long limit;
2643 udp_table_init(&udp_table, "UDP");
2644 limit = nr_free_buffer_pages() / 8;
2645 limit = max(limit, 128UL);
2646 sysctl_udp_mem[0] = limit / 4 * 3;
2647 sysctl_udp_mem[1] = limit;
2648 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2650 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2651 sysctl_udp_wmem_min = SK_MEM_QUANTUM;