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 <linux/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>
117 #include <net/addrconf.h>
119 struct udp_table udp_table __read_mostly;
120 EXPORT_SYMBOL(udp_table);
122 long sysctl_udp_mem[3] __read_mostly;
123 EXPORT_SYMBOL(sysctl_udp_mem);
125 int sysctl_udp_rmem_min __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_rmem_min);
128 int sysctl_udp_wmem_min __read_mostly;
129 EXPORT_SYMBOL(sysctl_udp_wmem_min);
131 atomic_long_t udp_memory_allocated;
132 EXPORT_SYMBOL(udp_memory_allocated);
134 #define MAX_UDP_PORTS 65536
135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
137 static int udp_lib_lport_inuse(struct net *net, __u16 num,
138 const struct udp_hslot *hslot,
139 unsigned long *bitmap,
140 struct sock *sk, unsigned int log)
143 kuid_t uid = sock_i_uid(sk);
145 sk_for_each(sk2, &hslot->head) {
146 if (net_eq(sock_net(sk2), net) &&
148 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
149 (!sk2->sk_reuse || !sk->sk_reuse) &&
150 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
151 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
152 inet_rcv_saddr_equal(sk, sk2, true)) {
153 if (sk2->sk_reuseport && sk->sk_reuseport &&
154 !rcu_access_pointer(sk->sk_reuseport_cb) &&
155 uid_eq(uid, sock_i_uid(sk2))) {
161 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
170 * Note: we still hold spinlock of primary hash chain, so no other writer
171 * can insert/delete a socket with local_port == num
173 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
174 struct udp_hslot *hslot2,
178 kuid_t uid = sock_i_uid(sk);
181 spin_lock(&hslot2->lock);
182 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
183 if (net_eq(sock_net(sk2), net) &&
185 (udp_sk(sk2)->udp_port_hash == num) &&
186 (!sk2->sk_reuse || !sk->sk_reuse) &&
187 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
188 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
189 inet_rcv_saddr_equal(sk, sk2, true)) {
190 if (sk2->sk_reuseport && sk->sk_reuseport &&
191 !rcu_access_pointer(sk->sk_reuseport_cb) &&
192 uid_eq(uid, sock_i_uid(sk2))) {
200 spin_unlock(&hslot2->lock);
204 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
206 struct net *net = sock_net(sk);
207 kuid_t uid = sock_i_uid(sk);
210 sk_for_each(sk2, &hslot->head) {
211 if (net_eq(sock_net(sk2), net) &&
213 sk2->sk_family == sk->sk_family &&
214 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
215 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
216 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
217 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
218 inet_rcv_saddr_equal(sk, sk2, false)) {
219 return reuseport_add_sock(sk, sk2);
223 /* Initial allocation may have already happened via setsockopt */
224 if (!rcu_access_pointer(sk->sk_reuseport_cb))
225 return reuseport_alloc(sk);
230 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
232 * @sk: socket struct in question
233 * @snum: port number to look up
234 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237 int udp_lib_get_port(struct sock *sk, unsigned short snum,
238 unsigned int hash2_nulladdr)
240 struct udp_hslot *hslot, *hslot2;
241 struct udp_table *udptable = sk->sk_prot->h.udp_table;
243 struct net *net = sock_net(sk);
246 int low, high, remaining;
248 unsigned short first, last;
249 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
251 inet_get_local_port_range(net, &low, &high);
252 remaining = (high - low) + 1;
254 rand = prandom_u32();
255 first = reciprocal_scale(rand, remaining) + low;
257 * force rand to be an odd multiple of UDP_HTABLE_SIZE
259 rand = (rand | 1) * (udptable->mask + 1);
260 last = first + udptable->mask + 1;
262 hslot = udp_hashslot(udptable, net, first);
263 bitmap_zero(bitmap, PORTS_PER_CHAIN);
264 spin_lock_bh(&hslot->lock);
265 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
270 * Iterate on all possible values of snum for this hash.
271 * Using steps of an odd multiple of UDP_HTABLE_SIZE
272 * give us randomization and full range coverage.
275 if (low <= snum && snum <= high &&
276 !test_bit(snum >> udptable->log, bitmap) &&
277 !inet_is_local_reserved_port(net, snum))
280 } while (snum != first);
281 spin_unlock_bh(&hslot->lock);
283 } while (++first != last);
286 hslot = udp_hashslot(udptable, net, snum);
287 spin_lock_bh(&hslot->lock);
288 if (hslot->count > 10) {
290 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
292 slot2 &= udptable->mask;
293 hash2_nulladdr &= udptable->mask;
295 hslot2 = udp_hashslot2(udptable, slot2);
296 if (hslot->count < hslot2->count)
297 goto scan_primary_hash;
299 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
300 if (!exist && (hash2_nulladdr != slot2)) {
301 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
302 exist = udp_lib_lport_inuse2(net, snum, hslot2,
311 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
315 inet_sk(sk)->inet_num = snum;
316 udp_sk(sk)->udp_port_hash = snum;
317 udp_sk(sk)->udp_portaddr_hash ^= snum;
318 if (sk_unhashed(sk)) {
319 if (sk->sk_reuseport &&
320 udp_reuseport_add_sock(sk, hslot)) {
321 inet_sk(sk)->inet_num = 0;
322 udp_sk(sk)->udp_port_hash = 0;
323 udp_sk(sk)->udp_portaddr_hash ^= snum;
327 sk_add_node_rcu(sk, &hslot->head);
329 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
331 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
332 spin_lock(&hslot2->lock);
333 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
334 sk->sk_family == AF_INET6)
335 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
338 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
341 spin_unlock(&hslot2->lock);
343 sock_set_flag(sk, SOCK_RCU_FREE);
346 spin_unlock_bh(&hslot->lock);
350 EXPORT_SYMBOL(udp_lib_get_port);
352 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
355 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
358 int udp_v4_get_port(struct sock *sk, unsigned short snum)
360 unsigned int hash2_nulladdr =
361 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
362 unsigned int hash2_partial =
363 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
365 /* precompute partial secondary hash */
366 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
367 return udp_lib_get_port(sk, snum, hash2_nulladdr);
370 static int compute_score(struct sock *sk, struct net *net,
371 __be32 saddr, __be16 sport,
372 __be32 daddr, unsigned short hnum, int dif)
375 struct inet_sock *inet;
377 if (!net_eq(sock_net(sk), net) ||
378 udp_sk(sk)->udp_port_hash != hnum ||
382 score = (sk->sk_family == PF_INET) ? 2 : 1;
385 if (inet->inet_rcv_saddr) {
386 if (inet->inet_rcv_saddr != daddr)
391 if (inet->inet_daddr) {
392 if (inet->inet_daddr != saddr)
397 if (inet->inet_dport) {
398 if (inet->inet_dport != sport)
403 if (sk->sk_bound_dev_if) {
404 if (sk->sk_bound_dev_if != dif)
408 if (sk->sk_incoming_cpu == raw_smp_processor_id())
413 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
414 const __u16 lport, const __be32 faddr,
417 static u32 udp_ehash_secret __read_mostly;
419 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
421 return __inet_ehashfn(laddr, lport, faddr, fport,
422 udp_ehash_secret + net_hash_mix(net));
425 /* called with rcu_read_lock() */
426 static struct sock *udp4_lib_lookup2(struct net *net,
427 __be32 saddr, __be16 sport,
428 __be32 daddr, unsigned int hnum, int dif,
429 struct udp_hslot *hslot2,
432 struct sock *sk, *result;
433 int score, badness, matches = 0, reuseport = 0;
438 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
439 score = compute_score(sk, net, saddr, sport,
441 if (score > badness) {
442 reuseport = sk->sk_reuseport;
444 hash = udp_ehashfn(net, daddr, hnum,
446 result = reuseport_select_sock(sk, hash, skb,
447 sizeof(struct udphdr));
454 } else if (score == badness && reuseport) {
456 if (reciprocal_scale(hash, matches) == 0)
458 hash = next_pseudo_random32(hash);
464 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
465 * harder than this. -DaveM
467 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
468 __be16 sport, __be32 daddr, __be16 dport,
469 int dif, struct udp_table *udptable, struct sk_buff *skb)
471 struct sock *sk, *result;
472 unsigned short hnum = ntohs(dport);
473 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
474 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
475 int score, badness, matches = 0, reuseport = 0;
478 if (hslot->count > 10) {
479 hash2 = udp4_portaddr_hash(net, daddr, hnum);
480 slot2 = hash2 & udptable->mask;
481 hslot2 = &udptable->hash2[slot2];
482 if (hslot->count < hslot2->count)
485 result = udp4_lib_lookup2(net, saddr, sport,
489 unsigned int old_slot2 = slot2;
490 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
491 slot2 = hash2 & udptable->mask;
492 /* avoid searching the same slot again. */
493 if (unlikely(slot2 == old_slot2))
496 hslot2 = &udptable->hash2[slot2];
497 if (hslot->count < hslot2->count)
500 result = udp4_lib_lookup2(net, saddr, sport,
509 sk_for_each_rcu(sk, &hslot->head) {
510 score = compute_score(sk, net, saddr, sport,
512 if (score > badness) {
513 reuseport = sk->sk_reuseport;
515 hash = udp_ehashfn(net, daddr, hnum,
517 result = reuseport_select_sock(sk, hash, skb,
518 sizeof(struct udphdr));
525 } else if (score == badness && reuseport) {
527 if (reciprocal_scale(hash, matches) == 0)
529 hash = next_pseudo_random32(hash);
534 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
536 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
537 __be16 sport, __be16 dport,
538 struct udp_table *udptable)
540 const struct iphdr *iph = ip_hdr(skb);
542 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
543 iph->daddr, dport, inet_iif(skb),
547 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
548 __be16 sport, __be16 dport)
550 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
552 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
554 /* Must be called under rcu_read_lock().
555 * Does increment socket refcount.
557 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
558 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
559 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
560 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
561 __be32 daddr, __be16 dport, int dif)
565 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
566 dif, &udp_table, NULL);
567 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
571 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
574 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
575 __be16 loc_port, __be32 loc_addr,
576 __be16 rmt_port, __be32 rmt_addr,
577 int dif, unsigned short hnum)
579 struct inet_sock *inet = inet_sk(sk);
581 if (!net_eq(sock_net(sk), net) ||
582 udp_sk(sk)->udp_port_hash != hnum ||
583 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
584 (inet->inet_dport != rmt_port && inet->inet_dport) ||
585 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
586 ipv6_only_sock(sk) ||
587 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
589 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
595 * This routine is called by the ICMP module when it gets some
596 * sort of error condition. If err < 0 then the socket should
597 * be closed and the error returned to the user. If err > 0
598 * it's just the icmp type << 8 | icmp code.
599 * Header points to the ip header of the error packet. We move
600 * on past this. Then (as it used to claim before adjustment)
601 * header points to the first 8 bytes of the udp header. We need
602 * to find the appropriate port.
605 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
607 struct inet_sock *inet;
608 const struct iphdr *iph = (const struct iphdr *)skb->data;
609 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
610 const int type = icmp_hdr(skb)->type;
611 const int code = icmp_hdr(skb)->code;
615 struct net *net = dev_net(skb->dev);
617 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
618 iph->saddr, uh->source, skb->dev->ifindex, udptable,
621 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
622 return; /* No socket for error */
631 case ICMP_TIME_EXCEEDED:
634 case ICMP_SOURCE_QUENCH:
636 case ICMP_PARAMETERPROB:
640 case ICMP_DEST_UNREACH:
641 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
642 ipv4_sk_update_pmtu(skb, sk, info);
643 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
651 if (code <= NR_ICMP_UNREACH) {
652 harderr = icmp_err_convert[code].fatal;
653 err = icmp_err_convert[code].errno;
657 ipv4_sk_redirect(skb, sk);
662 * RFC1122: OK. Passes ICMP errors back to application, as per
665 if (!inet->recverr) {
666 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
669 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
672 sk->sk_error_report(sk);
677 void udp_err(struct sk_buff *skb, u32 info)
679 __udp4_lib_err(skb, info, &udp_table);
683 * Throw away all pending data and cancel the corking. Socket is locked.
685 void udp_flush_pending_frames(struct sock *sk)
687 struct udp_sock *up = udp_sk(sk);
692 ip_flush_pending_frames(sk);
695 EXPORT_SYMBOL(udp_flush_pending_frames);
698 * udp4_hwcsum - handle outgoing HW checksumming
699 * @skb: sk_buff containing the filled-in UDP header
700 * (checksum field must be zeroed out)
701 * @src: source IP address
702 * @dst: destination IP address
704 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
706 struct udphdr *uh = udp_hdr(skb);
707 int offset = skb_transport_offset(skb);
708 int len = skb->len - offset;
712 if (!skb_has_frag_list(skb)) {
714 * Only one fragment on the socket.
716 skb->csum_start = skb_transport_header(skb) - skb->head;
717 skb->csum_offset = offsetof(struct udphdr, check);
718 uh->check = ~csum_tcpudp_magic(src, dst, len,
721 struct sk_buff *frags;
724 * HW-checksum won't work as there are two or more
725 * fragments on the socket so that all csums of sk_buffs
728 skb_walk_frags(skb, frags) {
729 csum = csum_add(csum, frags->csum);
733 csum = skb_checksum(skb, offset, hlen, csum);
734 skb->ip_summed = CHECKSUM_NONE;
736 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
738 uh->check = CSUM_MANGLED_0;
741 EXPORT_SYMBOL_GPL(udp4_hwcsum);
743 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
744 * for the simple case like when setting the checksum for a UDP tunnel.
746 void udp_set_csum(bool nocheck, struct sk_buff *skb,
747 __be32 saddr, __be32 daddr, int len)
749 struct udphdr *uh = udp_hdr(skb);
753 } else if (skb_is_gso(skb)) {
754 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
755 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
757 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
759 uh->check = CSUM_MANGLED_0;
761 skb->ip_summed = CHECKSUM_PARTIAL;
762 skb->csum_start = skb_transport_header(skb) - skb->head;
763 skb->csum_offset = offsetof(struct udphdr, check);
764 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
767 EXPORT_SYMBOL(udp_set_csum);
769 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
771 struct sock *sk = skb->sk;
772 struct inet_sock *inet = inet_sk(sk);
775 int is_udplite = IS_UDPLITE(sk);
776 int offset = skb_transport_offset(skb);
777 int len = skb->len - offset;
781 * Create a UDP header
784 uh->source = inet->inet_sport;
785 uh->dest = fl4->fl4_dport;
786 uh->len = htons(len);
789 if (is_udplite) /* UDP-Lite */
790 csum = udplite_csum(skb);
792 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
794 skb->ip_summed = CHECKSUM_NONE;
797 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
799 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
803 csum = udp_csum(skb);
805 /* add protocol-dependent pseudo-header */
806 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
807 sk->sk_protocol, csum);
809 uh->check = CSUM_MANGLED_0;
812 err = ip_send_skb(sock_net(sk), skb);
814 if (err == -ENOBUFS && !inet->recverr) {
815 UDP_INC_STATS(sock_net(sk),
816 UDP_MIB_SNDBUFERRORS, is_udplite);
820 UDP_INC_STATS(sock_net(sk),
821 UDP_MIB_OUTDATAGRAMS, is_udplite);
826 * Push out all pending data as one UDP datagram. Socket is locked.
828 int udp_push_pending_frames(struct sock *sk)
830 struct udp_sock *up = udp_sk(sk);
831 struct inet_sock *inet = inet_sk(sk);
832 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
836 skb = ip_finish_skb(sk, fl4);
840 err = udp_send_skb(skb, fl4);
847 EXPORT_SYMBOL(udp_push_pending_frames);
849 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
851 struct inet_sock *inet = inet_sk(sk);
852 struct udp_sock *up = udp_sk(sk);
853 struct flowi4 fl4_stack;
856 struct ipcm_cookie ipc;
857 struct rtable *rt = NULL;
860 __be32 daddr, faddr, saddr;
863 int err, is_udplite = IS_UDPLITE(sk);
864 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
865 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
867 struct ip_options_data opt_copy;
876 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
884 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
886 fl4 = &inet->cork.fl.u.ip4;
889 * There are pending frames.
890 * The socket lock must be held while it's corked.
893 if (likely(up->pending)) {
894 if (unlikely(up->pending != AF_INET)) {
902 ulen += sizeof(struct udphdr);
905 * Get and verify the address.
908 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
909 if (msg->msg_namelen < sizeof(*usin))
911 if (usin->sin_family != AF_INET) {
912 if (usin->sin_family != AF_UNSPEC)
913 return -EAFNOSUPPORT;
916 daddr = usin->sin_addr.s_addr;
917 dport = usin->sin_port;
921 if (sk->sk_state != TCP_ESTABLISHED)
922 return -EDESTADDRREQ;
923 daddr = inet->inet_daddr;
924 dport = inet->inet_dport;
925 /* Open fast path for connected socket.
926 Route will not be used, if at least one option is set.
931 ipc.sockc.tsflags = sk->sk_tsflags;
932 ipc.addr = inet->inet_saddr;
933 ipc.oif = sk->sk_bound_dev_if;
935 if (msg->msg_controllen) {
936 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
946 struct ip_options_rcu *inet_opt;
949 inet_opt = rcu_dereference(inet->inet_opt);
951 memcpy(&opt_copy, inet_opt,
952 sizeof(*inet_opt) + inet_opt->opt.optlen);
953 ipc.opt = &opt_copy.opt;
959 ipc.addr = faddr = daddr;
961 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
963 if (ipc.opt && ipc.opt->opt.srr) {
966 faddr = ipc.opt->opt.faddr;
969 tos = get_rttos(&ipc, inet);
970 if (sock_flag(sk, SOCK_LOCALROUTE) ||
971 (msg->msg_flags & MSG_DONTROUTE) ||
972 (ipc.opt && ipc.opt->opt.is_strictroute)) {
977 if (ipv4_is_multicast(daddr)) {
979 ipc.oif = inet->mc_index;
981 saddr = inet->mc_addr;
984 ipc.oif = inet->uc_index;
987 rt = (struct rtable *)sk_dst_check(sk, 0);
990 struct net *net = sock_net(sk);
991 __u8 flow_flags = inet_sk_flowi_flags(sk);
995 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
996 RT_SCOPE_UNIVERSE, sk->sk_protocol,
998 faddr, saddr, dport, inet->inet_sport,
1001 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1002 rt = ip_route_output_flow(net, fl4, sk);
1006 if (err == -ENETUNREACH)
1007 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1012 if ((rt->rt_flags & RTCF_BROADCAST) &&
1013 !sock_flag(sk, SOCK_BROADCAST))
1016 sk_dst_set(sk, dst_clone(&rt->dst));
1019 if (msg->msg_flags&MSG_CONFIRM)
1025 daddr = ipc.addr = fl4->daddr;
1027 /* Lockless fast path for the non-corking case. */
1029 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1030 sizeof(struct udphdr), &ipc, &rt,
1033 if (!IS_ERR_OR_NULL(skb))
1034 err = udp_send_skb(skb, fl4);
1039 if (unlikely(up->pending)) {
1040 /* The socket is already corked while preparing it. */
1041 /* ... which is an evident application bug. --ANK */
1044 net_dbg_ratelimited("cork app bug 2\n");
1049 * Now cork the socket to pend data.
1051 fl4 = &inet->cork.fl.u.ip4;
1054 fl4->fl4_dport = dport;
1055 fl4->fl4_sport = inet->inet_sport;
1056 up->pending = AF_INET;
1060 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1061 sizeof(struct udphdr), &ipc, &rt,
1062 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1064 udp_flush_pending_frames(sk);
1066 err = udp_push_pending_frames(sk);
1067 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1078 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1079 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1080 * we don't have a good statistic (IpOutDiscards but it can be too many
1081 * things). We could add another new stat but at least for now that
1082 * seems like overkill.
1084 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1085 UDP_INC_STATS(sock_net(sk),
1086 UDP_MIB_SNDBUFERRORS, is_udplite);
1091 dst_confirm(&rt->dst);
1092 if (!(msg->msg_flags&MSG_PROBE) || len)
1093 goto back_from_confirm;
1097 EXPORT_SYMBOL(udp_sendmsg);
1099 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1100 size_t size, int flags)
1102 struct inet_sock *inet = inet_sk(sk);
1103 struct udp_sock *up = udp_sk(sk);
1106 if (flags & MSG_SENDPAGE_NOTLAST)
1110 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1112 /* Call udp_sendmsg to specify destination address which
1113 * sendpage interface can't pass.
1114 * This will succeed only when the socket is connected.
1116 ret = udp_sendmsg(sk, &msg, 0);
1123 if (unlikely(!up->pending)) {
1126 net_dbg_ratelimited("udp cork app bug 3\n");
1130 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1131 page, offset, size, flags);
1132 if (ret == -EOPNOTSUPP) {
1134 return sock_no_sendpage(sk->sk_socket, page, offset,
1138 udp_flush_pending_frames(sk);
1143 if (!(up->corkflag || (flags&MSG_MORE)))
1144 ret = udp_push_pending_frames(sk);
1152 /* fully reclaim rmem/fwd memory allocated for skb */
1153 static void udp_rmem_release(struct sock *sk, int size, int partial)
1155 struct udp_sock *up = udp_sk(sk);
1158 if (likely(partial)) {
1159 up->forward_deficit += size;
1160 size = up->forward_deficit;
1161 if (size < (sk->sk_rcvbuf >> 2) &&
1162 !skb_queue_empty(&sk->sk_receive_queue))
1165 size += up->forward_deficit;
1167 up->forward_deficit = 0;
1169 sk->sk_forward_alloc += size;
1170 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1171 sk->sk_forward_alloc -= amt;
1174 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1176 atomic_sub(size, &sk->sk_rmem_alloc);
1179 /* Note: called with sk_receive_queue.lock held.
1180 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1181 * This avoids a cache line miss while receive_queue lock is held.
1182 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1184 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1186 udp_rmem_release(sk, skb->dev_scratch, 1);
1188 EXPORT_SYMBOL(udp_skb_destructor);
1190 /* Idea of busylocks is to let producers grab an extra spinlock
1191 * to relieve pressure on the receive_queue spinlock shared by consumer.
1192 * Under flood, this means that only one producer can be in line
1193 * trying to acquire the receive_queue spinlock.
1194 * These busylock can be allocated on a per cpu manner, instead of a
1195 * per socket one (that would consume a cache line per socket)
1197 static int udp_busylocks_log __read_mostly;
1198 static spinlock_t *udp_busylocks __read_mostly;
1200 static spinlock_t *busylock_acquire(void *ptr)
1204 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1209 static void busylock_release(spinlock_t *busy)
1215 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1217 struct sk_buff_head *list = &sk->sk_receive_queue;
1218 int rmem, delta, amt, err = -ENOMEM;
1219 spinlock_t *busy = NULL;
1222 /* try to avoid the costly atomic add/sub pair when the receive
1223 * queue is full; always allow at least a packet
1225 rmem = atomic_read(&sk->sk_rmem_alloc);
1226 if (rmem > sk->sk_rcvbuf)
1229 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1230 * having linear skbs :
1231 * - Reduce memory overhead and thus increase receive queue capacity
1232 * - Less cache line misses at copyout() time
1233 * - Less work at consume_skb() (less alien page frag freeing)
1235 if (rmem > (sk->sk_rcvbuf >> 1)) {
1238 busy = busylock_acquire(sk);
1240 size = skb->truesize;
1241 /* Copy skb->truesize into skb->dev_scratch to avoid a cache line miss
1242 * in udp_skb_destructor()
1244 skb->dev_scratch = size;
1246 /* we drop only if the receive buf is full and the receive
1247 * queue contains some other skb
1249 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1250 if (rmem > (size + sk->sk_rcvbuf))
1253 spin_lock(&list->lock);
1254 if (size >= sk->sk_forward_alloc) {
1255 amt = sk_mem_pages(size);
1256 delta = amt << SK_MEM_QUANTUM_SHIFT;
1257 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1259 spin_unlock(&list->lock);
1263 sk->sk_forward_alloc += delta;
1266 sk->sk_forward_alloc -= size;
1268 /* no need to setup a destructor, we will explicitly release the
1269 * forward allocated memory on dequeue
1271 sock_skb_set_dropcount(sk, skb);
1273 __skb_queue_tail(list, skb);
1274 spin_unlock(&list->lock);
1276 if (!sock_flag(sk, SOCK_DEAD))
1277 sk->sk_data_ready(sk);
1279 busylock_release(busy);
1283 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1286 atomic_inc(&sk->sk_drops);
1287 busylock_release(busy);
1290 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1292 void udp_destruct_sock(struct sock *sk)
1294 /* reclaim completely the forward allocated memory */
1295 unsigned int total = 0;
1296 struct sk_buff *skb;
1298 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1299 total += skb->truesize;
1302 udp_rmem_release(sk, total, 0);
1304 inet_sock_destruct(sk);
1306 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1308 int udp_init_sock(struct sock *sk)
1310 sk->sk_destruct = udp_destruct_sock;
1313 EXPORT_SYMBOL_GPL(udp_init_sock);
1315 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1317 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1318 bool slow = lock_sock_fast(sk);
1320 sk_peek_offset_bwd(sk, len);
1321 unlock_sock_fast(sk, slow);
1325 EXPORT_SYMBOL_GPL(skb_consume_udp);
1328 * first_packet_length - return length of first packet in receive queue
1331 * Drops all bad checksum frames, until a valid one is found.
1332 * Returns the length of found skb, or -1 if none is found.
1334 static int first_packet_length(struct sock *sk)
1336 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1337 struct sk_buff *skb;
1341 spin_lock_bh(&rcvq->lock);
1342 while ((skb = skb_peek(rcvq)) != NULL &&
1343 udp_lib_checksum_complete(skb)) {
1344 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1346 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1348 atomic_inc(&sk->sk_drops);
1349 __skb_unlink(skb, rcvq);
1350 total += skb->truesize;
1353 res = skb ? skb->len : -1;
1355 udp_rmem_release(sk, total, 1);
1356 spin_unlock_bh(&rcvq->lock);
1361 * IOCTL requests applicable to the UDP protocol
1364 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1369 int amount = sk_wmem_alloc_get(sk);
1371 return put_user(amount, (int __user *)arg);
1376 int amount = max_t(int, 0, first_packet_length(sk));
1378 return put_user(amount, (int __user *)arg);
1382 return -ENOIOCTLCMD;
1387 EXPORT_SYMBOL(udp_ioctl);
1390 * This should be easy, if there is something there we
1391 * return it, otherwise we block.
1394 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1395 int flags, int *addr_len)
1397 struct inet_sock *inet = inet_sk(sk);
1398 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1399 struct sk_buff *skb;
1400 unsigned int ulen, copied;
1401 int peeked, peeking, off;
1403 int is_udplite = IS_UDPLITE(sk);
1404 bool checksum_valid = false;
1406 if (flags & MSG_ERRQUEUE)
1407 return ip_recv_error(sk, msg, len, addr_len);
1410 peeking = off = sk_peek_offset(sk, flags);
1411 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1417 if (copied > ulen - off)
1418 copied = ulen - off;
1419 else if (copied < ulen)
1420 msg->msg_flags |= MSG_TRUNC;
1423 * If checksum is needed at all, try to do it while copying the
1424 * data. If the data is truncated, or if we only want a partial
1425 * coverage checksum (UDP-Lite), do it before the copy.
1428 if (copied < ulen || peeking ||
1429 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1430 checksum_valid = !udp_lib_checksum_complete(skb);
1431 if (!checksum_valid)
1435 if (checksum_valid || skb_csum_unnecessary(skb))
1436 err = skb_copy_datagram_msg(skb, off, msg, copied);
1438 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1444 if (unlikely(err)) {
1446 atomic_inc(&sk->sk_drops);
1447 UDP_INC_STATS(sock_net(sk),
1448 UDP_MIB_INERRORS, is_udplite);
1455 UDP_INC_STATS(sock_net(sk),
1456 UDP_MIB_INDATAGRAMS, is_udplite);
1458 sock_recv_ts_and_drops(msg, sk, skb);
1460 /* Copy the address. */
1462 sin->sin_family = AF_INET;
1463 sin->sin_port = udp_hdr(skb)->source;
1464 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1465 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1466 *addr_len = sizeof(*sin);
1468 if (inet->cmsg_flags)
1469 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1472 if (flags & MSG_TRUNC)
1475 skb_consume_udp(sk, skb, peeking ? -err : err);
1479 if (!__sk_queue_drop_skb(sk, skb, flags)) {
1480 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1481 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1485 /* starting over for a new packet, but check if we need to yield */
1487 msg->msg_flags &= ~MSG_TRUNC;
1491 int __udp_disconnect(struct sock *sk, int flags)
1493 struct inet_sock *inet = inet_sk(sk);
1495 * 1003.1g - break association.
1498 sk->sk_state = TCP_CLOSE;
1499 inet->inet_daddr = 0;
1500 inet->inet_dport = 0;
1501 sock_rps_reset_rxhash(sk);
1502 sk->sk_bound_dev_if = 0;
1503 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1504 inet_reset_saddr(sk);
1506 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1507 sk->sk_prot->unhash(sk);
1508 inet->inet_sport = 0;
1513 EXPORT_SYMBOL(__udp_disconnect);
1515 int udp_disconnect(struct sock *sk, int flags)
1518 __udp_disconnect(sk, flags);
1522 EXPORT_SYMBOL(udp_disconnect);
1524 void udp_lib_unhash(struct sock *sk)
1526 if (sk_hashed(sk)) {
1527 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1528 struct udp_hslot *hslot, *hslot2;
1530 hslot = udp_hashslot(udptable, sock_net(sk),
1531 udp_sk(sk)->udp_port_hash);
1532 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1534 spin_lock_bh(&hslot->lock);
1535 if (rcu_access_pointer(sk->sk_reuseport_cb))
1536 reuseport_detach_sock(sk);
1537 if (sk_del_node_init_rcu(sk)) {
1539 inet_sk(sk)->inet_num = 0;
1540 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1542 spin_lock(&hslot2->lock);
1543 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1545 spin_unlock(&hslot2->lock);
1547 spin_unlock_bh(&hslot->lock);
1550 EXPORT_SYMBOL(udp_lib_unhash);
1553 * inet_rcv_saddr was changed, we must rehash secondary hash
1555 void udp_lib_rehash(struct sock *sk, u16 newhash)
1557 if (sk_hashed(sk)) {
1558 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1559 struct udp_hslot *hslot, *hslot2, *nhslot2;
1561 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1562 nhslot2 = udp_hashslot2(udptable, newhash);
1563 udp_sk(sk)->udp_portaddr_hash = newhash;
1565 if (hslot2 != nhslot2 ||
1566 rcu_access_pointer(sk->sk_reuseport_cb)) {
1567 hslot = udp_hashslot(udptable, sock_net(sk),
1568 udp_sk(sk)->udp_port_hash);
1569 /* we must lock primary chain too */
1570 spin_lock_bh(&hslot->lock);
1571 if (rcu_access_pointer(sk->sk_reuseport_cb))
1572 reuseport_detach_sock(sk);
1574 if (hslot2 != nhslot2) {
1575 spin_lock(&hslot2->lock);
1576 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1578 spin_unlock(&hslot2->lock);
1580 spin_lock(&nhslot2->lock);
1581 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1584 spin_unlock(&nhslot2->lock);
1587 spin_unlock_bh(&hslot->lock);
1591 EXPORT_SYMBOL(udp_lib_rehash);
1593 static void udp_v4_rehash(struct sock *sk)
1595 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1596 inet_sk(sk)->inet_rcv_saddr,
1597 inet_sk(sk)->inet_num);
1598 udp_lib_rehash(sk, new_hash);
1601 int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1605 if (inet_sk(sk)->inet_daddr) {
1606 sock_rps_save_rxhash(sk, skb);
1607 sk_mark_napi_id(sk, skb);
1608 sk_incoming_cpu_update(sk);
1610 sk_mark_napi_id_once(sk, skb);
1613 rc = __udp_enqueue_schedule_skb(sk, skb);
1615 int is_udplite = IS_UDPLITE(sk);
1617 /* Note that an ENOMEM error is charged twice */
1619 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1621 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1623 trace_udp_fail_queue_rcv_skb(rc, sk);
1630 static struct static_key udp_encap_needed __read_mostly;
1631 void udp_encap_enable(void)
1633 if (!static_key_enabled(&udp_encap_needed))
1634 static_key_slow_inc(&udp_encap_needed);
1636 EXPORT_SYMBOL(udp_encap_enable);
1641 * >0: "udp encap" protocol resubmission
1643 * Note that in the success and error cases, the skb is assumed to
1644 * have either been requeued or freed.
1646 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1648 struct udp_sock *up = udp_sk(sk);
1649 int is_udplite = IS_UDPLITE(sk);
1652 * Charge it to the socket, dropping if the queue is full.
1654 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1658 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1659 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1662 * This is an encapsulation socket so pass the skb to
1663 * the socket's udp_encap_rcv() hook. Otherwise, just
1664 * fall through and pass this up the UDP socket.
1665 * up->encap_rcv() returns the following value:
1666 * =0 if skb was successfully passed to the encap
1667 * handler or was discarded by it.
1668 * >0 if skb should be passed on to UDP.
1669 * <0 if skb should be resubmitted as proto -N
1672 /* if we're overly short, let UDP handle it */
1673 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1677 /* Verify checksum before giving to encap */
1678 if (udp_lib_checksum_complete(skb))
1681 ret = encap_rcv(sk, skb);
1683 __UDP_INC_STATS(sock_net(sk),
1684 UDP_MIB_INDATAGRAMS,
1690 /* FALLTHROUGH -- it's a UDP Packet */
1694 * UDP-Lite specific tests, ignored on UDP sockets
1696 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1699 * MIB statistics other than incrementing the error count are
1700 * disabled for the following two types of errors: these depend
1701 * on the application settings, not on the functioning of the
1702 * protocol stack as such.
1704 * RFC 3828 here recommends (sec 3.3): "There should also be a
1705 * way ... to ... at least let the receiving application block
1706 * delivery of packets with coverage values less than a value
1707 * provided by the application."
1709 if (up->pcrlen == 0) { /* full coverage was set */
1710 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1711 UDP_SKB_CB(skb)->cscov, skb->len);
1714 /* The next case involves violating the min. coverage requested
1715 * by the receiver. This is subtle: if receiver wants x and x is
1716 * greater than the buffersize/MTU then receiver will complain
1717 * that it wants x while sender emits packets of smaller size y.
1718 * Therefore the above ...()->partial_cov statement is essential.
1720 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1721 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1722 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1727 if (rcu_access_pointer(sk->sk_filter) &&
1728 udp_lib_checksum_complete(skb))
1731 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1734 udp_csum_pull_header(skb);
1736 ipv4_pktinfo_prepare(sk, skb);
1737 return __udp_queue_rcv_skb(sk, skb);
1740 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1742 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1743 atomic_inc(&sk->sk_drops);
1748 /* For TCP sockets, sk_rx_dst is protected by socket lock
1749 * For UDP, we use xchg() to guard against concurrent changes.
1751 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1753 struct dst_entry *old;
1756 old = xchg(&sk->sk_rx_dst, dst);
1761 * Multicasts and broadcasts go to each listener.
1763 * Note: called only from the BH handler context.
1765 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1767 __be32 saddr, __be32 daddr,
1768 struct udp_table *udptable,
1771 struct sock *sk, *first = NULL;
1772 unsigned short hnum = ntohs(uh->dest);
1773 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1774 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1775 unsigned int offset = offsetof(typeof(*sk), sk_node);
1776 int dif = skb->dev->ifindex;
1777 struct hlist_node *node;
1778 struct sk_buff *nskb;
1781 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1783 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1785 hslot = &udptable->hash2[hash2];
1786 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1789 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1790 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1791 uh->source, saddr, dif, hnum))
1798 nskb = skb_clone(skb, GFP_ATOMIC);
1800 if (unlikely(!nskb)) {
1801 atomic_inc(&sk->sk_drops);
1802 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1804 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1808 if (udp_queue_rcv_skb(sk, nskb) > 0)
1812 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1813 if (use_hash2 && hash2 != hash2_any) {
1819 if (udp_queue_rcv_skb(first, skb) > 0)
1823 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
1824 proto == IPPROTO_UDPLITE);
1829 /* Initialize UDP checksum. If exited with zero value (success),
1830 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1831 * Otherwise, csum completion requires chacksumming packet body,
1832 * including udp header and folding it to skb->csum.
1834 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1839 UDP_SKB_CB(skb)->partial_cov = 0;
1840 UDP_SKB_CB(skb)->cscov = skb->len;
1842 if (proto == IPPROTO_UDPLITE) {
1843 err = udplite_checksum_init(skb, uh);
1848 /* Note, we are only interested in != 0 or == 0, thus the
1851 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
1852 inet_compute_pseudo);
1856 * All we need to do is get the socket, and then do a checksum.
1859 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1864 unsigned short ulen;
1865 struct rtable *rt = skb_rtable(skb);
1866 __be32 saddr, daddr;
1867 struct net *net = dev_net(skb->dev);
1870 * Validate the packet.
1872 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1873 goto drop; /* No space for header. */
1876 ulen = ntohs(uh->len);
1877 saddr = ip_hdr(skb)->saddr;
1878 daddr = ip_hdr(skb)->daddr;
1880 if (ulen > skb->len)
1883 if (proto == IPPROTO_UDP) {
1884 /* UDP validates ulen. */
1885 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1890 if (udp4_csum_init(skb, uh, proto))
1893 sk = skb_steal_sock(skb);
1895 struct dst_entry *dst = skb_dst(skb);
1898 if (unlikely(sk->sk_rx_dst != dst))
1899 udp_sk_rx_dst_set(sk, dst);
1901 ret = udp_queue_rcv_skb(sk, skb);
1903 /* a return value > 0 means to resubmit the input, but
1904 * it wants the return to be -protocol, or 0
1911 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1912 return __udp4_lib_mcast_deliver(net, skb, uh,
1913 saddr, daddr, udptable, proto);
1915 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1919 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1920 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1921 inet_compute_pseudo);
1923 ret = udp_queue_rcv_skb(sk, skb);
1925 /* a return value > 0 means to resubmit the input, but
1926 * it wants the return to be -protocol, or 0
1933 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1937 /* No socket. Drop packet silently, if checksum is wrong */
1938 if (udp_lib_checksum_complete(skb))
1941 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1942 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1945 * Hmm. We got an UDP packet to a port to which we
1946 * don't wanna listen. Ignore it.
1952 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1953 proto == IPPROTO_UDPLITE ? "Lite" : "",
1954 &saddr, ntohs(uh->source),
1956 &daddr, ntohs(uh->dest));
1961 * RFC1122: OK. Discards the bad packet silently (as far as
1962 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1964 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1965 proto == IPPROTO_UDPLITE ? "Lite" : "",
1966 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1968 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1970 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1975 /* We can only early demux multicast if there is a single matching socket.
1976 * If more than one socket found returns NULL
1978 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1979 __be16 loc_port, __be32 loc_addr,
1980 __be16 rmt_port, __be32 rmt_addr,
1983 struct sock *sk, *result;
1984 unsigned short hnum = ntohs(loc_port);
1985 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
1986 struct udp_hslot *hslot = &udp_table.hash[slot];
1988 /* Do not bother scanning a too big list */
1989 if (hslot->count > 10)
1993 sk_for_each_rcu(sk, &hslot->head) {
1994 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
1995 rmt_port, rmt_addr, dif, hnum)) {
2005 /* For unicast we should only early demux connected sockets or we can
2006 * break forwarding setups. The chains here can be long so only check
2007 * if the first socket is an exact match and if not move on.
2009 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2010 __be16 loc_port, __be32 loc_addr,
2011 __be16 rmt_port, __be32 rmt_addr,
2014 unsigned short hnum = ntohs(loc_port);
2015 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2016 unsigned int slot2 = hash2 & udp_table.mask;
2017 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2018 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2019 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2022 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2023 if (INET_MATCH(sk, net, acookie, rmt_addr,
2024 loc_addr, ports, dif))
2026 /* Only check first socket in chain */
2032 void udp_v4_early_demux(struct sk_buff *skb)
2034 struct net *net = dev_net(skb->dev);
2035 const struct iphdr *iph;
2036 const struct udphdr *uh;
2037 struct sock *sk = NULL;
2038 struct dst_entry *dst;
2039 int dif = skb->dev->ifindex;
2042 /* validate the packet */
2043 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2049 if (skb->pkt_type == PACKET_BROADCAST ||
2050 skb->pkt_type == PACKET_MULTICAST) {
2051 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2056 /* we are supposed to accept bcast packets */
2057 if (skb->pkt_type == PACKET_MULTICAST) {
2058 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2064 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2065 uh->source, iph->saddr, dif);
2066 } else if (skb->pkt_type == PACKET_HOST) {
2067 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2068 uh->source, iph->saddr, dif);
2071 if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2))
2075 skb->destructor = sock_efree;
2076 dst = READ_ONCE(sk->sk_rx_dst);
2079 dst = dst_check(dst, 0);
2081 /* DST_NOCACHE can not be used without taking a reference */
2082 if (dst->flags & DST_NOCACHE) {
2083 if (likely(atomic_inc_not_zero(&dst->__refcnt)))
2084 skb_dst_set(skb, dst);
2086 skb_dst_set_noref(skb, dst);
2091 int udp_rcv(struct sk_buff *skb)
2093 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2096 void udp_destroy_sock(struct sock *sk)
2098 struct udp_sock *up = udp_sk(sk);
2099 bool slow = lock_sock_fast(sk);
2100 udp_flush_pending_frames(sk);
2101 unlock_sock_fast(sk, slow);
2102 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2103 void (*encap_destroy)(struct sock *sk);
2104 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2111 * Socket option code for UDP
2113 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2114 char __user *optval, unsigned int optlen,
2115 int (*push_pending_frames)(struct sock *))
2117 struct udp_sock *up = udp_sk(sk);
2120 int is_udplite = IS_UDPLITE(sk);
2122 if (optlen < sizeof(int))
2125 if (get_user(val, (int __user *)optval))
2128 valbool = val ? 1 : 0;
2137 push_pending_frames(sk);
2145 case UDP_ENCAP_ESPINUDP:
2146 case UDP_ENCAP_ESPINUDP_NON_IKE:
2147 up->encap_rcv = xfrm4_udp_encap_rcv;
2149 case UDP_ENCAP_L2TPINUDP:
2150 up->encap_type = val;
2159 case UDP_NO_CHECK6_TX:
2160 up->no_check6_tx = valbool;
2163 case UDP_NO_CHECK6_RX:
2164 up->no_check6_rx = valbool;
2168 * UDP-Lite's partial checksum coverage (RFC 3828).
2170 /* The sender sets actual checksum coverage length via this option.
2171 * The case coverage > packet length is handled by send module. */
2172 case UDPLITE_SEND_CSCOV:
2173 if (!is_udplite) /* Disable the option on UDP sockets */
2174 return -ENOPROTOOPT;
2175 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2177 else if (val > USHRT_MAX)
2180 up->pcflag |= UDPLITE_SEND_CC;
2183 /* The receiver specifies a minimum checksum coverage value. To make
2184 * sense, this should be set to at least 8 (as done below). If zero is
2185 * used, this again means full checksum coverage. */
2186 case UDPLITE_RECV_CSCOV:
2187 if (!is_udplite) /* Disable the option on UDP sockets */
2188 return -ENOPROTOOPT;
2189 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2191 else if (val > USHRT_MAX)
2194 up->pcflag |= UDPLITE_RECV_CC;
2204 EXPORT_SYMBOL(udp_lib_setsockopt);
2206 int udp_setsockopt(struct sock *sk, int level, int optname,
2207 char __user *optval, unsigned int optlen)
2209 if (level == SOL_UDP || level == SOL_UDPLITE)
2210 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2211 udp_push_pending_frames);
2212 return ip_setsockopt(sk, level, optname, optval, optlen);
2215 #ifdef CONFIG_COMPAT
2216 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2217 char __user *optval, unsigned int optlen)
2219 if (level == SOL_UDP || level == SOL_UDPLITE)
2220 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2221 udp_push_pending_frames);
2222 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2226 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2227 char __user *optval, int __user *optlen)
2229 struct udp_sock *up = udp_sk(sk);
2232 if (get_user(len, optlen))
2235 len = min_t(unsigned int, len, sizeof(int));
2246 val = up->encap_type;
2249 case UDP_NO_CHECK6_TX:
2250 val = up->no_check6_tx;
2253 case UDP_NO_CHECK6_RX:
2254 val = up->no_check6_rx;
2257 /* The following two cannot be changed on UDP sockets, the return is
2258 * always 0 (which corresponds to the full checksum coverage of UDP). */
2259 case UDPLITE_SEND_CSCOV:
2263 case UDPLITE_RECV_CSCOV:
2268 return -ENOPROTOOPT;
2271 if (put_user(len, optlen))
2273 if (copy_to_user(optval, &val, len))
2277 EXPORT_SYMBOL(udp_lib_getsockopt);
2279 int udp_getsockopt(struct sock *sk, int level, int optname,
2280 char __user *optval, int __user *optlen)
2282 if (level == SOL_UDP || level == SOL_UDPLITE)
2283 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2284 return ip_getsockopt(sk, level, optname, optval, optlen);
2287 #ifdef CONFIG_COMPAT
2288 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2289 char __user *optval, int __user *optlen)
2291 if (level == SOL_UDP || level == SOL_UDPLITE)
2292 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2293 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2297 * udp_poll - wait for a UDP event.
2298 * @file - file struct
2300 * @wait - poll table
2302 * This is same as datagram poll, except for the special case of
2303 * blocking sockets. If application is using a blocking fd
2304 * and a packet with checksum error is in the queue;
2305 * then it could get return from select indicating data available
2306 * but then block when reading it. Add special case code
2307 * to work around these arguably broken applications.
2309 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2311 unsigned int mask = datagram_poll(file, sock, wait);
2312 struct sock *sk = sock->sk;
2314 sock_rps_record_flow(sk);
2316 /* Check for false positives due to checksum errors */
2317 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2318 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2319 mask &= ~(POLLIN | POLLRDNORM);
2324 EXPORT_SYMBOL(udp_poll);
2326 int udp_abort(struct sock *sk, int err)
2331 sk->sk_error_report(sk);
2332 __udp_disconnect(sk, 0);
2338 EXPORT_SYMBOL_GPL(udp_abort);
2340 struct proto udp_prot = {
2342 .owner = THIS_MODULE,
2343 .close = udp_lib_close,
2344 .connect = ip4_datagram_connect,
2345 .disconnect = udp_disconnect,
2347 .init = udp_init_sock,
2348 .destroy = udp_destroy_sock,
2349 .setsockopt = udp_setsockopt,
2350 .getsockopt = udp_getsockopt,
2351 .sendmsg = udp_sendmsg,
2352 .recvmsg = udp_recvmsg,
2353 .sendpage = udp_sendpage,
2354 .release_cb = ip4_datagram_release_cb,
2355 .hash = udp_lib_hash,
2356 .unhash = udp_lib_unhash,
2357 .rehash = udp_v4_rehash,
2358 .get_port = udp_v4_get_port,
2359 .memory_allocated = &udp_memory_allocated,
2360 .sysctl_mem = sysctl_udp_mem,
2361 .sysctl_wmem = &sysctl_udp_wmem_min,
2362 .sysctl_rmem = &sysctl_udp_rmem_min,
2363 .obj_size = sizeof(struct udp_sock),
2364 .h.udp_table = &udp_table,
2365 #ifdef CONFIG_COMPAT
2366 .compat_setsockopt = compat_udp_setsockopt,
2367 .compat_getsockopt = compat_udp_getsockopt,
2369 .diag_destroy = udp_abort,
2371 EXPORT_SYMBOL(udp_prot);
2373 /* ------------------------------------------------------------------------ */
2374 #ifdef CONFIG_PROC_FS
2376 static struct sock *udp_get_first(struct seq_file *seq, int start)
2379 struct udp_iter_state *state = seq->private;
2380 struct net *net = seq_file_net(seq);
2382 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2384 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2386 if (hlist_empty(&hslot->head))
2389 spin_lock_bh(&hslot->lock);
2390 sk_for_each(sk, &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);
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_HEAD(&table->hash[i].head);
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_HEAD(&table->hash2[i].head);
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;
2644 udp_table_init(&udp_table, "UDP");
2645 limit = nr_free_buffer_pages() / 8;
2646 limit = max(limit, 128UL);
2647 sysctl_udp_mem[0] = limit / 4 * 3;
2648 sysctl_udp_mem[1] = limit;
2649 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2651 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2652 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2654 /* 16 spinlocks per cpu */
2655 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2656 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2659 panic("UDP: failed to alloc udp_busylocks\n");
2660 for (i = 0; i < (1U << udp_busylocks_log); i++)
2661 spin_lock_init(udp_busylocks + i);