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 /* IPCB reference means this can not be used from early demux */
138 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
140 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
141 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
142 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
148 static int udp_lib_lport_inuse(struct net *net, __u16 num,
149 const struct udp_hslot *hslot,
150 unsigned long *bitmap,
151 struct sock *sk, unsigned int log)
154 kuid_t uid = sock_i_uid(sk);
156 sk_for_each(sk2, &hslot->head) {
157 if (net_eq(sock_net(sk2), net) &&
159 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
160 (!sk2->sk_reuse || !sk->sk_reuse) &&
161 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
162 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
163 inet_rcv_saddr_equal(sk, sk2, true)) {
164 if (sk2->sk_reuseport && sk->sk_reuseport &&
165 !rcu_access_pointer(sk->sk_reuseport_cb) &&
166 uid_eq(uid, sock_i_uid(sk2))) {
172 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
181 * Note: we still hold spinlock of primary hash chain, so no other writer
182 * can insert/delete a socket with local_port == num
184 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
185 struct udp_hslot *hslot2,
189 kuid_t uid = sock_i_uid(sk);
192 spin_lock(&hslot2->lock);
193 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
194 if (net_eq(sock_net(sk2), net) &&
196 (udp_sk(sk2)->udp_port_hash == num) &&
197 (!sk2->sk_reuse || !sk->sk_reuse) &&
198 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
199 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
200 inet_rcv_saddr_equal(sk, sk2, true)) {
201 if (sk2->sk_reuseport && sk->sk_reuseport &&
202 !rcu_access_pointer(sk->sk_reuseport_cb) &&
203 uid_eq(uid, sock_i_uid(sk2))) {
211 spin_unlock(&hslot2->lock);
215 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
217 struct net *net = sock_net(sk);
218 kuid_t uid = sock_i_uid(sk);
221 sk_for_each(sk2, &hslot->head) {
222 if (net_eq(sock_net(sk2), net) &&
224 sk2->sk_family == sk->sk_family &&
225 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
226 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
227 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
228 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
229 inet_rcv_saddr_equal(sk, sk2, false)) {
230 return reuseport_add_sock(sk, sk2);
234 /* Initial allocation may have already happened via setsockopt */
235 if (!rcu_access_pointer(sk->sk_reuseport_cb))
236 return reuseport_alloc(sk);
241 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
243 * @sk: socket struct in question
244 * @snum: port number to look up
245 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
248 int udp_lib_get_port(struct sock *sk, unsigned short snum,
249 unsigned int hash2_nulladdr)
251 struct udp_hslot *hslot, *hslot2;
252 struct udp_table *udptable = sk->sk_prot->h.udp_table;
254 struct net *net = sock_net(sk);
257 int low, high, remaining;
259 unsigned short first, last;
260 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
262 inet_get_local_port_range(net, &low, &high);
263 remaining = (high - low) + 1;
265 rand = prandom_u32();
266 first = reciprocal_scale(rand, remaining) + low;
268 * force rand to be an odd multiple of UDP_HTABLE_SIZE
270 rand = (rand | 1) * (udptable->mask + 1);
271 last = first + udptable->mask + 1;
273 hslot = udp_hashslot(udptable, net, first);
274 bitmap_zero(bitmap, PORTS_PER_CHAIN);
275 spin_lock_bh(&hslot->lock);
276 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
281 * Iterate on all possible values of snum for this hash.
282 * Using steps of an odd multiple of UDP_HTABLE_SIZE
283 * give us randomization and full range coverage.
286 if (low <= snum && snum <= high &&
287 !test_bit(snum >> udptable->log, bitmap) &&
288 !inet_is_local_reserved_port(net, snum))
291 } while (snum != first);
292 spin_unlock_bh(&hslot->lock);
294 } while (++first != last);
297 hslot = udp_hashslot(udptable, net, snum);
298 spin_lock_bh(&hslot->lock);
299 if (hslot->count > 10) {
301 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
303 slot2 &= udptable->mask;
304 hash2_nulladdr &= udptable->mask;
306 hslot2 = udp_hashslot2(udptable, slot2);
307 if (hslot->count < hslot2->count)
308 goto scan_primary_hash;
310 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
311 if (!exist && (hash2_nulladdr != slot2)) {
312 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
313 exist = udp_lib_lport_inuse2(net, snum, hslot2,
322 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
326 inet_sk(sk)->inet_num = snum;
327 udp_sk(sk)->udp_port_hash = snum;
328 udp_sk(sk)->udp_portaddr_hash ^= snum;
329 if (sk_unhashed(sk)) {
330 if (sk->sk_reuseport &&
331 udp_reuseport_add_sock(sk, hslot)) {
332 inet_sk(sk)->inet_num = 0;
333 udp_sk(sk)->udp_port_hash = 0;
334 udp_sk(sk)->udp_portaddr_hash ^= snum;
338 sk_add_node_rcu(sk, &hslot->head);
340 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
342 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
343 spin_lock(&hslot2->lock);
344 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
345 sk->sk_family == AF_INET6)
346 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
349 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
352 spin_unlock(&hslot2->lock);
354 sock_set_flag(sk, SOCK_RCU_FREE);
357 spin_unlock_bh(&hslot->lock);
361 EXPORT_SYMBOL(udp_lib_get_port);
363 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
366 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
369 int udp_v4_get_port(struct sock *sk, unsigned short snum)
371 unsigned int hash2_nulladdr =
372 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
373 unsigned int hash2_partial =
374 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
376 /* precompute partial secondary hash */
377 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
378 return udp_lib_get_port(sk, snum, hash2_nulladdr);
381 static int compute_score(struct sock *sk, struct net *net,
382 __be32 saddr, __be16 sport,
383 __be32 daddr, unsigned short hnum, int dif,
387 struct inet_sock *inet;
389 if (!net_eq(sock_net(sk), net) ||
390 udp_sk(sk)->udp_port_hash != hnum ||
394 score = (sk->sk_family == PF_INET) ? 2 : 1;
397 if (inet->inet_rcv_saddr) {
398 if (inet->inet_rcv_saddr != daddr)
403 if (inet->inet_daddr) {
404 if (inet->inet_daddr != saddr)
409 if (inet->inet_dport) {
410 if (inet->inet_dport != sport)
415 if (sk->sk_bound_dev_if || exact_dif) {
416 if (sk->sk_bound_dev_if != dif)
420 if (sk->sk_incoming_cpu == raw_smp_processor_id())
425 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
426 const __u16 lport, const __be32 faddr,
429 static u32 udp_ehash_secret __read_mostly;
431 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
433 return __inet_ehashfn(laddr, lport, faddr, fport,
434 udp_ehash_secret + net_hash_mix(net));
437 /* called with rcu_read_lock() */
438 static struct sock *udp4_lib_lookup2(struct net *net,
439 __be32 saddr, __be16 sport,
440 __be32 daddr, unsigned int hnum, int dif, bool exact_dif,
441 struct udp_hslot *hslot2,
444 struct sock *sk, *result;
445 int score, badness, matches = 0, reuseport = 0;
450 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
451 score = compute_score(sk, net, saddr, sport,
452 daddr, hnum, dif, exact_dif);
453 if (score > badness) {
454 reuseport = sk->sk_reuseport;
456 hash = udp_ehashfn(net, daddr, hnum,
458 result = reuseport_select_sock(sk, hash, skb,
459 sizeof(struct udphdr));
466 } else if (score == badness && reuseport) {
468 if (reciprocal_scale(hash, matches) == 0)
470 hash = next_pseudo_random32(hash);
476 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
477 * harder than this. -DaveM
479 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
480 __be16 sport, __be32 daddr, __be16 dport,
481 int dif, struct udp_table *udptable, struct sk_buff *skb)
483 struct sock *sk, *result;
484 unsigned short hnum = ntohs(dport);
485 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
486 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
487 bool exact_dif = udp_lib_exact_dif_match(net, skb);
488 int score, badness, matches = 0, reuseport = 0;
491 if (hslot->count > 10) {
492 hash2 = udp4_portaddr_hash(net, daddr, hnum);
493 slot2 = hash2 & udptable->mask;
494 hslot2 = &udptable->hash2[slot2];
495 if (hslot->count < hslot2->count)
498 result = udp4_lib_lookup2(net, saddr, sport,
500 exact_dif, hslot2, skb);
502 unsigned int old_slot2 = slot2;
503 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
504 slot2 = hash2 & udptable->mask;
505 /* avoid searching the same slot again. */
506 if (unlikely(slot2 == old_slot2))
509 hslot2 = &udptable->hash2[slot2];
510 if (hslot->count < hslot2->count)
513 result = udp4_lib_lookup2(net, saddr, sport,
515 exact_dif, hslot2, skb);
522 sk_for_each_rcu(sk, &hslot->head) {
523 score = compute_score(sk, net, saddr, sport,
524 daddr, hnum, dif, exact_dif);
525 if (score > badness) {
526 reuseport = sk->sk_reuseport;
528 hash = udp_ehashfn(net, daddr, hnum,
530 result = reuseport_select_sock(sk, hash, skb,
531 sizeof(struct udphdr));
538 } else if (score == badness && reuseport) {
540 if (reciprocal_scale(hash, matches) == 0)
542 hash = next_pseudo_random32(hash);
547 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
549 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
550 __be16 sport, __be16 dport,
551 struct udp_table *udptable)
553 const struct iphdr *iph = ip_hdr(skb);
555 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
556 iph->daddr, dport, inet_iif(skb),
560 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
561 __be16 sport, __be16 dport)
563 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
565 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
567 /* Must be called under rcu_read_lock().
568 * Does increment socket refcount.
570 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
571 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
572 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
573 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
574 __be32 daddr, __be16 dport, int dif)
578 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
579 dif, &udp_table, NULL);
580 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
584 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
587 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
588 __be16 loc_port, __be32 loc_addr,
589 __be16 rmt_port, __be32 rmt_addr,
590 int dif, unsigned short hnum)
592 struct inet_sock *inet = inet_sk(sk);
594 if (!net_eq(sock_net(sk), net) ||
595 udp_sk(sk)->udp_port_hash != hnum ||
596 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
597 (inet->inet_dport != rmt_port && inet->inet_dport) ||
598 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
599 ipv6_only_sock(sk) ||
600 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
602 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
608 * This routine is called by the ICMP module when it gets some
609 * sort of error condition. If err < 0 then the socket should
610 * be closed and the error returned to the user. If err > 0
611 * it's just the icmp type << 8 | icmp code.
612 * Header points to the ip header of the error packet. We move
613 * on past this. Then (as it used to claim before adjustment)
614 * header points to the first 8 bytes of the udp header. We need
615 * to find the appropriate port.
618 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
620 struct inet_sock *inet;
621 const struct iphdr *iph = (const struct iphdr *)skb->data;
622 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
623 const int type = icmp_hdr(skb)->type;
624 const int code = icmp_hdr(skb)->code;
628 struct net *net = dev_net(skb->dev);
630 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
631 iph->saddr, uh->source, skb->dev->ifindex, udptable,
634 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
635 return; /* No socket for error */
644 case ICMP_TIME_EXCEEDED:
647 case ICMP_SOURCE_QUENCH:
649 case ICMP_PARAMETERPROB:
653 case ICMP_DEST_UNREACH:
654 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
655 ipv4_sk_update_pmtu(skb, sk, info);
656 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
664 if (code <= NR_ICMP_UNREACH) {
665 harderr = icmp_err_convert[code].fatal;
666 err = icmp_err_convert[code].errno;
670 ipv4_sk_redirect(skb, sk);
675 * RFC1122: OK. Passes ICMP errors back to application, as per
678 if (!inet->recverr) {
679 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
682 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
685 sk->sk_error_report(sk);
690 void udp_err(struct sk_buff *skb, u32 info)
692 __udp4_lib_err(skb, info, &udp_table);
696 * Throw away all pending data and cancel the corking. Socket is locked.
698 void udp_flush_pending_frames(struct sock *sk)
700 struct udp_sock *up = udp_sk(sk);
705 ip_flush_pending_frames(sk);
708 EXPORT_SYMBOL(udp_flush_pending_frames);
711 * udp4_hwcsum - handle outgoing HW checksumming
712 * @skb: sk_buff containing the filled-in UDP header
713 * (checksum field must be zeroed out)
714 * @src: source IP address
715 * @dst: destination IP address
717 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
719 struct udphdr *uh = udp_hdr(skb);
720 int offset = skb_transport_offset(skb);
721 int len = skb->len - offset;
725 if (!skb_has_frag_list(skb)) {
727 * Only one fragment on the socket.
729 skb->csum_start = skb_transport_header(skb) - skb->head;
730 skb->csum_offset = offsetof(struct udphdr, check);
731 uh->check = ~csum_tcpudp_magic(src, dst, len,
734 struct sk_buff *frags;
737 * HW-checksum won't work as there are two or more
738 * fragments on the socket so that all csums of sk_buffs
741 skb_walk_frags(skb, frags) {
742 csum = csum_add(csum, frags->csum);
746 csum = skb_checksum(skb, offset, hlen, csum);
747 skb->ip_summed = CHECKSUM_NONE;
749 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
751 uh->check = CSUM_MANGLED_0;
754 EXPORT_SYMBOL_GPL(udp4_hwcsum);
756 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
757 * for the simple case like when setting the checksum for a UDP tunnel.
759 void udp_set_csum(bool nocheck, struct sk_buff *skb,
760 __be32 saddr, __be32 daddr, int len)
762 struct udphdr *uh = udp_hdr(skb);
766 } else if (skb_is_gso(skb)) {
767 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
768 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
770 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
772 uh->check = CSUM_MANGLED_0;
774 skb->ip_summed = CHECKSUM_PARTIAL;
775 skb->csum_start = skb_transport_header(skb) - skb->head;
776 skb->csum_offset = offsetof(struct udphdr, check);
777 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
780 EXPORT_SYMBOL(udp_set_csum);
782 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
784 struct sock *sk = skb->sk;
785 struct inet_sock *inet = inet_sk(sk);
788 int is_udplite = IS_UDPLITE(sk);
789 int offset = skb_transport_offset(skb);
790 int len = skb->len - offset;
794 * Create a UDP header
797 uh->source = inet->inet_sport;
798 uh->dest = fl4->fl4_dport;
799 uh->len = htons(len);
802 if (is_udplite) /* UDP-Lite */
803 csum = udplite_csum(skb);
805 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
807 skb->ip_summed = CHECKSUM_NONE;
810 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
812 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
816 csum = udp_csum(skb);
818 /* add protocol-dependent pseudo-header */
819 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
820 sk->sk_protocol, csum);
822 uh->check = CSUM_MANGLED_0;
825 err = ip_send_skb(sock_net(sk), skb);
827 if (err == -ENOBUFS && !inet->recverr) {
828 UDP_INC_STATS(sock_net(sk),
829 UDP_MIB_SNDBUFERRORS, is_udplite);
833 UDP_INC_STATS(sock_net(sk),
834 UDP_MIB_OUTDATAGRAMS, is_udplite);
839 * Push out all pending data as one UDP datagram. Socket is locked.
841 int udp_push_pending_frames(struct sock *sk)
843 struct udp_sock *up = udp_sk(sk);
844 struct inet_sock *inet = inet_sk(sk);
845 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
849 skb = ip_finish_skb(sk, fl4);
853 err = udp_send_skb(skb, fl4);
860 EXPORT_SYMBOL(udp_push_pending_frames);
862 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
864 struct inet_sock *inet = inet_sk(sk);
865 struct udp_sock *up = udp_sk(sk);
866 struct flowi4 fl4_stack;
869 struct ipcm_cookie ipc;
870 struct rtable *rt = NULL;
873 __be32 daddr, faddr, saddr;
876 int err, is_udplite = IS_UDPLITE(sk);
877 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
878 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
880 struct ip_options_data opt_copy;
889 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
897 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
899 fl4 = &inet->cork.fl.u.ip4;
902 * There are pending frames.
903 * The socket lock must be held while it's corked.
906 if (likely(up->pending)) {
907 if (unlikely(up->pending != AF_INET)) {
915 ulen += sizeof(struct udphdr);
918 * Get and verify the address.
921 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
922 if (msg->msg_namelen < sizeof(*usin))
924 if (usin->sin_family != AF_INET) {
925 if (usin->sin_family != AF_UNSPEC)
926 return -EAFNOSUPPORT;
929 daddr = usin->sin_addr.s_addr;
930 dport = usin->sin_port;
934 if (sk->sk_state != TCP_ESTABLISHED)
935 return -EDESTADDRREQ;
936 daddr = inet->inet_daddr;
937 dport = inet->inet_dport;
938 /* Open fast path for connected socket.
939 Route will not be used, if at least one option is set.
944 ipc.sockc.tsflags = sk->sk_tsflags;
945 ipc.addr = inet->inet_saddr;
946 ipc.oif = sk->sk_bound_dev_if;
948 if (msg->msg_controllen) {
949 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
959 struct ip_options_rcu *inet_opt;
962 inet_opt = rcu_dereference(inet->inet_opt);
964 memcpy(&opt_copy, inet_opt,
965 sizeof(*inet_opt) + inet_opt->opt.optlen);
966 ipc.opt = &opt_copy.opt;
972 ipc.addr = faddr = daddr;
974 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
976 if (ipc.opt && ipc.opt->opt.srr) {
979 faddr = ipc.opt->opt.faddr;
982 tos = get_rttos(&ipc, inet);
983 if (sock_flag(sk, SOCK_LOCALROUTE) ||
984 (msg->msg_flags & MSG_DONTROUTE) ||
985 (ipc.opt && ipc.opt->opt.is_strictroute)) {
990 if (ipv4_is_multicast(daddr)) {
992 ipc.oif = inet->mc_index;
994 saddr = inet->mc_addr;
997 ipc.oif = inet->uc_index;
1000 rt = (struct rtable *)sk_dst_check(sk, 0);
1003 struct net *net = sock_net(sk);
1004 __u8 flow_flags = inet_sk_flowi_flags(sk);
1008 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1009 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1011 faddr, saddr, dport, inet->inet_sport,
1014 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1015 rt = ip_route_output_flow(net, fl4, sk);
1019 if (err == -ENETUNREACH)
1020 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1025 if ((rt->rt_flags & RTCF_BROADCAST) &&
1026 !sock_flag(sk, SOCK_BROADCAST))
1029 sk_dst_set(sk, dst_clone(&rt->dst));
1032 if (msg->msg_flags&MSG_CONFIRM)
1038 daddr = ipc.addr = fl4->daddr;
1040 /* Lockless fast path for the non-corking case. */
1042 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1043 sizeof(struct udphdr), &ipc, &rt,
1046 if (!IS_ERR_OR_NULL(skb))
1047 err = udp_send_skb(skb, fl4);
1052 if (unlikely(up->pending)) {
1053 /* The socket is already corked while preparing it. */
1054 /* ... which is an evident application bug. --ANK */
1057 net_dbg_ratelimited("cork app bug 2\n");
1062 * Now cork the socket to pend data.
1064 fl4 = &inet->cork.fl.u.ip4;
1067 fl4->fl4_dport = dport;
1068 fl4->fl4_sport = inet->inet_sport;
1069 up->pending = AF_INET;
1073 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1074 sizeof(struct udphdr), &ipc, &rt,
1075 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1077 udp_flush_pending_frames(sk);
1079 err = udp_push_pending_frames(sk);
1080 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1091 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1092 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1093 * we don't have a good statistic (IpOutDiscards but it can be too many
1094 * things). We could add another new stat but at least for now that
1095 * seems like overkill.
1097 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1098 UDP_INC_STATS(sock_net(sk),
1099 UDP_MIB_SNDBUFERRORS, is_udplite);
1104 dst_confirm(&rt->dst);
1105 if (!(msg->msg_flags&MSG_PROBE) || len)
1106 goto back_from_confirm;
1110 EXPORT_SYMBOL(udp_sendmsg);
1112 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1113 size_t size, int flags)
1115 struct inet_sock *inet = inet_sk(sk);
1116 struct udp_sock *up = udp_sk(sk);
1119 if (flags & MSG_SENDPAGE_NOTLAST)
1123 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1125 /* Call udp_sendmsg to specify destination address which
1126 * sendpage interface can't pass.
1127 * This will succeed only when the socket is connected.
1129 ret = udp_sendmsg(sk, &msg, 0);
1136 if (unlikely(!up->pending)) {
1139 net_dbg_ratelimited("udp cork app bug 3\n");
1143 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1144 page, offset, size, flags);
1145 if (ret == -EOPNOTSUPP) {
1147 return sock_no_sendpage(sk->sk_socket, page, offset,
1151 udp_flush_pending_frames(sk);
1156 if (!(up->corkflag || (flags&MSG_MORE)))
1157 ret = udp_push_pending_frames(sk);
1165 /* fully reclaim rmem/fwd memory allocated for skb */
1166 static void udp_rmem_release(struct sock *sk, int size, int partial)
1168 struct udp_sock *up = udp_sk(sk);
1171 if (likely(partial)) {
1172 up->forward_deficit += size;
1173 size = up->forward_deficit;
1174 if (size < (sk->sk_rcvbuf >> 2) &&
1175 !skb_queue_empty(&sk->sk_receive_queue))
1178 size += up->forward_deficit;
1180 up->forward_deficit = 0;
1182 sk->sk_forward_alloc += size;
1183 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1184 sk->sk_forward_alloc -= amt;
1187 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1189 atomic_sub(size, &sk->sk_rmem_alloc);
1192 /* Note: called with sk_receive_queue.lock held.
1193 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1194 * This avoids a cache line miss while receive_queue lock is held.
1195 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1197 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1199 udp_rmem_release(sk, skb->dev_scratch, 1);
1201 EXPORT_SYMBOL(udp_skb_destructor);
1203 /* Idea of busylocks is to let producers grab an extra spinlock
1204 * to relieve pressure on the receive_queue spinlock shared by consumer.
1205 * Under flood, this means that only one producer can be in line
1206 * trying to acquire the receive_queue spinlock.
1207 * These busylock can be allocated on a per cpu manner, instead of a
1208 * per socket one (that would consume a cache line per socket)
1210 static int udp_busylocks_log __read_mostly;
1211 static spinlock_t *udp_busylocks __read_mostly;
1213 static spinlock_t *busylock_acquire(void *ptr)
1217 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1222 static void busylock_release(spinlock_t *busy)
1228 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1230 struct sk_buff_head *list = &sk->sk_receive_queue;
1231 int rmem, delta, amt, err = -ENOMEM;
1232 spinlock_t *busy = NULL;
1235 /* try to avoid the costly atomic add/sub pair when the receive
1236 * queue is full; always allow at least a packet
1238 rmem = atomic_read(&sk->sk_rmem_alloc);
1239 if (rmem > sk->sk_rcvbuf)
1242 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1243 * having linear skbs :
1244 * - Reduce memory overhead and thus increase receive queue capacity
1245 * - Less cache line misses at copyout() time
1246 * - Less work at consume_skb() (less alien page frag freeing)
1248 if (rmem > (sk->sk_rcvbuf >> 1)) {
1251 busy = busylock_acquire(sk);
1253 size = skb->truesize;
1254 /* Copy skb->truesize into skb->dev_scratch to avoid a cache line miss
1255 * in udp_skb_destructor()
1257 skb->dev_scratch = size;
1259 /* we drop only if the receive buf is full and the receive
1260 * queue contains some other skb
1262 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1263 if (rmem > (size + sk->sk_rcvbuf))
1266 spin_lock(&list->lock);
1267 if (size >= sk->sk_forward_alloc) {
1268 amt = sk_mem_pages(size);
1269 delta = amt << SK_MEM_QUANTUM_SHIFT;
1270 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1272 spin_unlock(&list->lock);
1276 sk->sk_forward_alloc += delta;
1279 sk->sk_forward_alloc -= size;
1281 /* no need to setup a destructor, we will explicitly release the
1282 * forward allocated memory on dequeue
1284 sock_skb_set_dropcount(sk, skb);
1286 __skb_queue_tail(list, skb);
1287 spin_unlock(&list->lock);
1289 if (!sock_flag(sk, SOCK_DEAD))
1290 sk->sk_data_ready(sk);
1292 busylock_release(busy);
1296 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1299 atomic_inc(&sk->sk_drops);
1300 busylock_release(busy);
1303 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1305 void udp_destruct_sock(struct sock *sk)
1307 /* reclaim completely the forward allocated memory */
1308 unsigned int total = 0;
1309 struct sk_buff *skb;
1311 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1312 total += skb->truesize;
1315 udp_rmem_release(sk, total, 0);
1317 inet_sock_destruct(sk);
1319 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1321 int udp_init_sock(struct sock *sk)
1323 sk->sk_destruct = udp_destruct_sock;
1326 EXPORT_SYMBOL_GPL(udp_init_sock);
1328 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1330 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1331 bool slow = lock_sock_fast(sk);
1333 sk_peek_offset_bwd(sk, len);
1334 unlock_sock_fast(sk, slow);
1338 EXPORT_SYMBOL_GPL(skb_consume_udp);
1341 * first_packet_length - return length of first packet in receive queue
1344 * Drops all bad checksum frames, until a valid one is found.
1345 * Returns the length of found skb, or -1 if none is found.
1347 static int first_packet_length(struct sock *sk)
1349 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1350 struct sk_buff *skb;
1354 spin_lock_bh(&rcvq->lock);
1355 while ((skb = skb_peek(rcvq)) != NULL &&
1356 udp_lib_checksum_complete(skb)) {
1357 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1359 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1361 atomic_inc(&sk->sk_drops);
1362 __skb_unlink(skb, rcvq);
1363 total += skb->truesize;
1366 res = skb ? skb->len : -1;
1368 udp_rmem_release(sk, total, 1);
1369 spin_unlock_bh(&rcvq->lock);
1374 * IOCTL requests applicable to the UDP protocol
1377 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1382 int amount = sk_wmem_alloc_get(sk);
1384 return put_user(amount, (int __user *)arg);
1389 int amount = max_t(int, 0, first_packet_length(sk));
1391 return put_user(amount, (int __user *)arg);
1395 return -ENOIOCTLCMD;
1400 EXPORT_SYMBOL(udp_ioctl);
1403 * This should be easy, if there is something there we
1404 * return it, otherwise we block.
1407 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1408 int flags, int *addr_len)
1410 struct inet_sock *inet = inet_sk(sk);
1411 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1412 struct sk_buff *skb;
1413 unsigned int ulen, copied;
1414 int peeked, peeking, off;
1416 int is_udplite = IS_UDPLITE(sk);
1417 bool checksum_valid = false;
1419 if (flags & MSG_ERRQUEUE)
1420 return ip_recv_error(sk, msg, len, addr_len);
1423 peeking = off = sk_peek_offset(sk, flags);
1424 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1430 if (copied > ulen - off)
1431 copied = ulen - off;
1432 else if (copied < ulen)
1433 msg->msg_flags |= MSG_TRUNC;
1436 * If checksum is needed at all, try to do it while copying the
1437 * data. If the data is truncated, or if we only want a partial
1438 * coverage checksum (UDP-Lite), do it before the copy.
1441 if (copied < ulen || peeking ||
1442 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1443 checksum_valid = !udp_lib_checksum_complete(skb);
1444 if (!checksum_valid)
1448 if (checksum_valid || skb_csum_unnecessary(skb))
1449 err = skb_copy_datagram_msg(skb, off, msg, copied);
1451 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1457 if (unlikely(err)) {
1459 atomic_inc(&sk->sk_drops);
1460 UDP_INC_STATS(sock_net(sk),
1461 UDP_MIB_INERRORS, is_udplite);
1468 UDP_INC_STATS(sock_net(sk),
1469 UDP_MIB_INDATAGRAMS, is_udplite);
1471 sock_recv_ts_and_drops(msg, sk, skb);
1473 /* Copy the address. */
1475 sin->sin_family = AF_INET;
1476 sin->sin_port = udp_hdr(skb)->source;
1477 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1478 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1479 *addr_len = sizeof(*sin);
1481 if (inet->cmsg_flags)
1482 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1485 if (flags & MSG_TRUNC)
1488 skb_consume_udp(sk, skb, peeking ? -err : err);
1492 if (!__sk_queue_drop_skb(sk, skb, flags)) {
1493 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1494 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1498 /* starting over for a new packet, but check if we need to yield */
1500 msg->msg_flags &= ~MSG_TRUNC;
1504 int __udp_disconnect(struct sock *sk, int flags)
1506 struct inet_sock *inet = inet_sk(sk);
1508 * 1003.1g - break association.
1511 sk->sk_state = TCP_CLOSE;
1512 inet->inet_daddr = 0;
1513 inet->inet_dport = 0;
1514 sock_rps_reset_rxhash(sk);
1515 sk->sk_bound_dev_if = 0;
1516 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1517 inet_reset_saddr(sk);
1519 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1520 sk->sk_prot->unhash(sk);
1521 inet->inet_sport = 0;
1526 EXPORT_SYMBOL(__udp_disconnect);
1528 int udp_disconnect(struct sock *sk, int flags)
1531 __udp_disconnect(sk, flags);
1535 EXPORT_SYMBOL(udp_disconnect);
1537 void udp_lib_unhash(struct sock *sk)
1539 if (sk_hashed(sk)) {
1540 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1541 struct udp_hslot *hslot, *hslot2;
1543 hslot = udp_hashslot(udptable, sock_net(sk),
1544 udp_sk(sk)->udp_port_hash);
1545 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1547 spin_lock_bh(&hslot->lock);
1548 if (rcu_access_pointer(sk->sk_reuseport_cb))
1549 reuseport_detach_sock(sk);
1550 if (sk_del_node_init_rcu(sk)) {
1552 inet_sk(sk)->inet_num = 0;
1553 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1555 spin_lock(&hslot2->lock);
1556 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1558 spin_unlock(&hslot2->lock);
1560 spin_unlock_bh(&hslot->lock);
1563 EXPORT_SYMBOL(udp_lib_unhash);
1566 * inet_rcv_saddr was changed, we must rehash secondary hash
1568 void udp_lib_rehash(struct sock *sk, u16 newhash)
1570 if (sk_hashed(sk)) {
1571 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1572 struct udp_hslot *hslot, *hslot2, *nhslot2;
1574 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1575 nhslot2 = udp_hashslot2(udptable, newhash);
1576 udp_sk(sk)->udp_portaddr_hash = newhash;
1578 if (hslot2 != nhslot2 ||
1579 rcu_access_pointer(sk->sk_reuseport_cb)) {
1580 hslot = udp_hashslot(udptable, sock_net(sk),
1581 udp_sk(sk)->udp_port_hash);
1582 /* we must lock primary chain too */
1583 spin_lock_bh(&hslot->lock);
1584 if (rcu_access_pointer(sk->sk_reuseport_cb))
1585 reuseport_detach_sock(sk);
1587 if (hslot2 != nhslot2) {
1588 spin_lock(&hslot2->lock);
1589 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1591 spin_unlock(&hslot2->lock);
1593 spin_lock(&nhslot2->lock);
1594 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1597 spin_unlock(&nhslot2->lock);
1600 spin_unlock_bh(&hslot->lock);
1604 EXPORT_SYMBOL(udp_lib_rehash);
1606 static void udp_v4_rehash(struct sock *sk)
1608 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1609 inet_sk(sk)->inet_rcv_saddr,
1610 inet_sk(sk)->inet_num);
1611 udp_lib_rehash(sk, new_hash);
1614 int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1618 if (inet_sk(sk)->inet_daddr) {
1619 sock_rps_save_rxhash(sk, skb);
1620 sk_mark_napi_id(sk, skb);
1621 sk_incoming_cpu_update(sk);
1623 sk_mark_napi_id_once(sk, skb);
1626 rc = __udp_enqueue_schedule_skb(sk, skb);
1628 int is_udplite = IS_UDPLITE(sk);
1630 /* Note that an ENOMEM error is charged twice */
1632 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1634 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1636 trace_udp_fail_queue_rcv_skb(rc, sk);
1643 static struct static_key udp_encap_needed __read_mostly;
1644 void udp_encap_enable(void)
1646 if (!static_key_enabled(&udp_encap_needed))
1647 static_key_slow_inc(&udp_encap_needed);
1649 EXPORT_SYMBOL(udp_encap_enable);
1654 * >0: "udp encap" protocol resubmission
1656 * Note that in the success and error cases, the skb is assumed to
1657 * have either been requeued or freed.
1659 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1661 struct udp_sock *up = udp_sk(sk);
1662 int is_udplite = IS_UDPLITE(sk);
1665 * Charge it to the socket, dropping if the queue is full.
1667 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1671 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1672 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1675 * This is an encapsulation socket so pass the skb to
1676 * the socket's udp_encap_rcv() hook. Otherwise, just
1677 * fall through and pass this up the UDP socket.
1678 * up->encap_rcv() returns the following value:
1679 * =0 if skb was successfully passed to the encap
1680 * handler or was discarded by it.
1681 * >0 if skb should be passed on to UDP.
1682 * <0 if skb should be resubmitted as proto -N
1685 /* if we're overly short, let UDP handle it */
1686 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1690 /* Verify checksum before giving to encap */
1691 if (udp_lib_checksum_complete(skb))
1694 ret = encap_rcv(sk, skb);
1696 __UDP_INC_STATS(sock_net(sk),
1697 UDP_MIB_INDATAGRAMS,
1703 /* FALLTHROUGH -- it's a UDP Packet */
1707 * UDP-Lite specific tests, ignored on UDP sockets
1709 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1712 * MIB statistics other than incrementing the error count are
1713 * disabled for the following two types of errors: these depend
1714 * on the application settings, not on the functioning of the
1715 * protocol stack as such.
1717 * RFC 3828 here recommends (sec 3.3): "There should also be a
1718 * way ... to ... at least let the receiving application block
1719 * delivery of packets with coverage values less than a value
1720 * provided by the application."
1722 if (up->pcrlen == 0) { /* full coverage was set */
1723 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1724 UDP_SKB_CB(skb)->cscov, skb->len);
1727 /* The next case involves violating the min. coverage requested
1728 * by the receiver. This is subtle: if receiver wants x and x is
1729 * greater than the buffersize/MTU then receiver will complain
1730 * that it wants x while sender emits packets of smaller size y.
1731 * Therefore the above ...()->partial_cov statement is essential.
1733 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1734 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1735 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1740 if (rcu_access_pointer(sk->sk_filter) &&
1741 udp_lib_checksum_complete(skb))
1744 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1747 udp_csum_pull_header(skb);
1749 ipv4_pktinfo_prepare(sk, skb);
1750 return __udp_queue_rcv_skb(sk, skb);
1753 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1755 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1756 atomic_inc(&sk->sk_drops);
1761 /* For TCP sockets, sk_rx_dst is protected by socket lock
1762 * For UDP, we use xchg() to guard against concurrent changes.
1764 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1766 struct dst_entry *old;
1769 old = xchg(&sk->sk_rx_dst, dst);
1774 * Multicasts and broadcasts go to each listener.
1776 * Note: called only from the BH handler context.
1778 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1780 __be32 saddr, __be32 daddr,
1781 struct udp_table *udptable,
1784 struct sock *sk, *first = NULL;
1785 unsigned short hnum = ntohs(uh->dest);
1786 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1787 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1788 unsigned int offset = offsetof(typeof(*sk), sk_node);
1789 int dif = skb->dev->ifindex;
1790 struct hlist_node *node;
1791 struct sk_buff *nskb;
1794 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1796 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1798 hslot = &udptable->hash2[hash2];
1799 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1802 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1803 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1804 uh->source, saddr, dif, hnum))
1811 nskb = skb_clone(skb, GFP_ATOMIC);
1813 if (unlikely(!nskb)) {
1814 atomic_inc(&sk->sk_drops);
1815 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1817 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1821 if (udp_queue_rcv_skb(sk, nskb) > 0)
1825 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1826 if (use_hash2 && hash2 != hash2_any) {
1832 if (udp_queue_rcv_skb(first, skb) > 0)
1836 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
1837 proto == IPPROTO_UDPLITE);
1842 /* Initialize UDP checksum. If exited with zero value (success),
1843 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1844 * Otherwise, csum completion requires chacksumming packet body,
1845 * including udp header and folding it to skb->csum.
1847 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1852 UDP_SKB_CB(skb)->partial_cov = 0;
1853 UDP_SKB_CB(skb)->cscov = skb->len;
1855 if (proto == IPPROTO_UDPLITE) {
1856 err = udplite_checksum_init(skb, uh);
1861 /* Note, we are only interested in != 0 or == 0, thus the
1864 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
1865 inet_compute_pseudo);
1869 * All we need to do is get the socket, and then do a checksum.
1872 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1877 unsigned short ulen;
1878 struct rtable *rt = skb_rtable(skb);
1879 __be32 saddr, daddr;
1880 struct net *net = dev_net(skb->dev);
1883 * Validate the packet.
1885 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1886 goto drop; /* No space for header. */
1889 ulen = ntohs(uh->len);
1890 saddr = ip_hdr(skb)->saddr;
1891 daddr = ip_hdr(skb)->daddr;
1893 if (ulen > skb->len)
1896 if (proto == IPPROTO_UDP) {
1897 /* UDP validates ulen. */
1898 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1903 if (udp4_csum_init(skb, uh, proto))
1906 sk = skb_steal_sock(skb);
1908 struct dst_entry *dst = skb_dst(skb);
1911 if (unlikely(sk->sk_rx_dst != dst))
1912 udp_sk_rx_dst_set(sk, dst);
1914 ret = udp_queue_rcv_skb(sk, skb);
1916 /* a return value > 0 means to resubmit the input, but
1917 * it wants the return to be -protocol, or 0
1924 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1925 return __udp4_lib_mcast_deliver(net, skb, uh,
1926 saddr, daddr, udptable, proto);
1928 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1932 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1933 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1934 inet_compute_pseudo);
1936 ret = udp_queue_rcv_skb(sk, skb);
1938 /* a return value > 0 means to resubmit the input, but
1939 * it wants the return to be -protocol, or 0
1946 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1950 /* No socket. Drop packet silently, if checksum is wrong */
1951 if (udp_lib_checksum_complete(skb))
1954 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1955 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1958 * Hmm. We got an UDP packet to a port to which we
1959 * don't wanna listen. Ignore it.
1965 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1966 proto == IPPROTO_UDPLITE ? "Lite" : "",
1967 &saddr, ntohs(uh->source),
1969 &daddr, ntohs(uh->dest));
1974 * RFC1122: OK. Discards the bad packet silently (as far as
1975 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1977 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1978 proto == IPPROTO_UDPLITE ? "Lite" : "",
1979 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1981 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1983 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1988 /* We can only early demux multicast if there is a single matching socket.
1989 * If more than one socket found returns NULL
1991 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1992 __be16 loc_port, __be32 loc_addr,
1993 __be16 rmt_port, __be32 rmt_addr,
1996 struct sock *sk, *result;
1997 unsigned short hnum = ntohs(loc_port);
1998 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
1999 struct udp_hslot *hslot = &udp_table.hash[slot];
2001 /* Do not bother scanning a too big list */
2002 if (hslot->count > 10)
2006 sk_for_each_rcu(sk, &hslot->head) {
2007 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2008 rmt_port, rmt_addr, dif, hnum)) {
2018 /* For unicast we should only early demux connected sockets or we can
2019 * break forwarding setups. The chains here can be long so only check
2020 * if the first socket is an exact match and if not move on.
2022 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2023 __be16 loc_port, __be32 loc_addr,
2024 __be16 rmt_port, __be32 rmt_addr,
2027 unsigned short hnum = ntohs(loc_port);
2028 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2029 unsigned int slot2 = hash2 & udp_table.mask;
2030 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2031 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2032 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2035 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2036 if (INET_MATCH(sk, net, acookie, rmt_addr,
2037 loc_addr, ports, dif))
2039 /* Only check first socket in chain */
2045 void udp_v4_early_demux(struct sk_buff *skb)
2047 struct net *net = dev_net(skb->dev);
2048 const struct iphdr *iph;
2049 const struct udphdr *uh;
2050 struct sock *sk = NULL;
2051 struct dst_entry *dst;
2052 int dif = skb->dev->ifindex;
2055 /* validate the packet */
2056 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2062 if (skb->pkt_type == PACKET_BROADCAST ||
2063 skb->pkt_type == PACKET_MULTICAST) {
2064 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2069 /* we are supposed to accept bcast packets */
2070 if (skb->pkt_type == PACKET_MULTICAST) {
2071 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2077 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2078 uh->source, iph->saddr, dif);
2079 } else if (skb->pkt_type == PACKET_HOST) {
2080 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2081 uh->source, iph->saddr, dif);
2084 if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2))
2088 skb->destructor = sock_efree;
2089 dst = READ_ONCE(sk->sk_rx_dst);
2092 dst = dst_check(dst, 0);
2094 /* DST_NOCACHE can not be used without taking a reference */
2095 if (dst->flags & DST_NOCACHE) {
2096 if (likely(atomic_inc_not_zero(&dst->__refcnt)))
2097 skb_dst_set(skb, dst);
2099 skb_dst_set_noref(skb, dst);
2104 int udp_rcv(struct sk_buff *skb)
2106 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2109 void udp_destroy_sock(struct sock *sk)
2111 struct udp_sock *up = udp_sk(sk);
2112 bool slow = lock_sock_fast(sk);
2113 udp_flush_pending_frames(sk);
2114 unlock_sock_fast(sk, slow);
2115 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2116 void (*encap_destroy)(struct sock *sk);
2117 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2124 * Socket option code for UDP
2126 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2127 char __user *optval, unsigned int optlen,
2128 int (*push_pending_frames)(struct sock *))
2130 struct udp_sock *up = udp_sk(sk);
2133 int is_udplite = IS_UDPLITE(sk);
2135 if (optlen < sizeof(int))
2138 if (get_user(val, (int __user *)optval))
2141 valbool = val ? 1 : 0;
2150 push_pending_frames(sk);
2158 case UDP_ENCAP_ESPINUDP:
2159 case UDP_ENCAP_ESPINUDP_NON_IKE:
2160 up->encap_rcv = xfrm4_udp_encap_rcv;
2162 case UDP_ENCAP_L2TPINUDP:
2163 up->encap_type = val;
2172 case UDP_NO_CHECK6_TX:
2173 up->no_check6_tx = valbool;
2176 case UDP_NO_CHECK6_RX:
2177 up->no_check6_rx = valbool;
2181 * UDP-Lite's partial checksum coverage (RFC 3828).
2183 /* The sender sets actual checksum coverage length via this option.
2184 * The case coverage > packet length is handled by send module. */
2185 case UDPLITE_SEND_CSCOV:
2186 if (!is_udplite) /* Disable the option on UDP sockets */
2187 return -ENOPROTOOPT;
2188 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2190 else if (val > USHRT_MAX)
2193 up->pcflag |= UDPLITE_SEND_CC;
2196 /* The receiver specifies a minimum checksum coverage value. To make
2197 * sense, this should be set to at least 8 (as done below). If zero is
2198 * used, this again means full checksum coverage. */
2199 case UDPLITE_RECV_CSCOV:
2200 if (!is_udplite) /* Disable the option on UDP sockets */
2201 return -ENOPROTOOPT;
2202 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2204 else if (val > USHRT_MAX)
2207 up->pcflag |= UDPLITE_RECV_CC;
2217 EXPORT_SYMBOL(udp_lib_setsockopt);
2219 int udp_setsockopt(struct sock *sk, int level, int optname,
2220 char __user *optval, unsigned int optlen)
2222 if (level == SOL_UDP || level == SOL_UDPLITE)
2223 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2224 udp_push_pending_frames);
2225 return ip_setsockopt(sk, level, optname, optval, optlen);
2228 #ifdef CONFIG_COMPAT
2229 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2230 char __user *optval, unsigned int optlen)
2232 if (level == SOL_UDP || level == SOL_UDPLITE)
2233 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2234 udp_push_pending_frames);
2235 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2239 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2240 char __user *optval, int __user *optlen)
2242 struct udp_sock *up = udp_sk(sk);
2245 if (get_user(len, optlen))
2248 len = min_t(unsigned int, len, sizeof(int));
2259 val = up->encap_type;
2262 case UDP_NO_CHECK6_TX:
2263 val = up->no_check6_tx;
2266 case UDP_NO_CHECK6_RX:
2267 val = up->no_check6_rx;
2270 /* The following two cannot be changed on UDP sockets, the return is
2271 * always 0 (which corresponds to the full checksum coverage of UDP). */
2272 case UDPLITE_SEND_CSCOV:
2276 case UDPLITE_RECV_CSCOV:
2281 return -ENOPROTOOPT;
2284 if (put_user(len, optlen))
2286 if (copy_to_user(optval, &val, len))
2290 EXPORT_SYMBOL(udp_lib_getsockopt);
2292 int udp_getsockopt(struct sock *sk, int level, int optname,
2293 char __user *optval, int __user *optlen)
2295 if (level == SOL_UDP || level == SOL_UDPLITE)
2296 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2297 return ip_getsockopt(sk, level, optname, optval, optlen);
2300 #ifdef CONFIG_COMPAT
2301 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2302 char __user *optval, int __user *optlen)
2304 if (level == SOL_UDP || level == SOL_UDPLITE)
2305 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2306 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2310 * udp_poll - wait for a UDP event.
2311 * @file - file struct
2313 * @wait - poll table
2315 * This is same as datagram poll, except for the special case of
2316 * blocking sockets. If application is using a blocking fd
2317 * and a packet with checksum error is in the queue;
2318 * then it could get return from select indicating data available
2319 * but then block when reading it. Add special case code
2320 * to work around these arguably broken applications.
2322 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2324 unsigned int mask = datagram_poll(file, sock, wait);
2325 struct sock *sk = sock->sk;
2327 sock_rps_record_flow(sk);
2329 /* Check for false positives due to checksum errors */
2330 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2331 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2332 mask &= ~(POLLIN | POLLRDNORM);
2337 EXPORT_SYMBOL(udp_poll);
2339 int udp_abort(struct sock *sk, int err)
2344 sk->sk_error_report(sk);
2345 __udp_disconnect(sk, 0);
2351 EXPORT_SYMBOL_GPL(udp_abort);
2353 struct proto udp_prot = {
2355 .owner = THIS_MODULE,
2356 .close = udp_lib_close,
2357 .connect = ip4_datagram_connect,
2358 .disconnect = udp_disconnect,
2360 .init = udp_init_sock,
2361 .destroy = udp_destroy_sock,
2362 .setsockopt = udp_setsockopt,
2363 .getsockopt = udp_getsockopt,
2364 .sendmsg = udp_sendmsg,
2365 .recvmsg = udp_recvmsg,
2366 .sendpage = udp_sendpage,
2367 .release_cb = ip4_datagram_release_cb,
2368 .hash = udp_lib_hash,
2369 .unhash = udp_lib_unhash,
2370 .rehash = udp_v4_rehash,
2371 .get_port = udp_v4_get_port,
2372 .memory_allocated = &udp_memory_allocated,
2373 .sysctl_mem = sysctl_udp_mem,
2374 .sysctl_wmem = &sysctl_udp_wmem_min,
2375 .sysctl_rmem = &sysctl_udp_rmem_min,
2376 .obj_size = sizeof(struct udp_sock),
2377 .h.udp_table = &udp_table,
2378 #ifdef CONFIG_COMPAT
2379 .compat_setsockopt = compat_udp_setsockopt,
2380 .compat_getsockopt = compat_udp_getsockopt,
2382 .diag_destroy = udp_abort,
2384 EXPORT_SYMBOL(udp_prot);
2386 /* ------------------------------------------------------------------------ */
2387 #ifdef CONFIG_PROC_FS
2389 static struct sock *udp_get_first(struct seq_file *seq, int start)
2392 struct udp_iter_state *state = seq->private;
2393 struct net *net = seq_file_net(seq);
2395 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2397 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2399 if (hlist_empty(&hslot->head))
2402 spin_lock_bh(&hslot->lock);
2403 sk_for_each(sk, &hslot->head) {
2404 if (!net_eq(sock_net(sk), net))
2406 if (sk->sk_family == state->family)
2409 spin_unlock_bh(&hslot->lock);
2416 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2418 struct udp_iter_state *state = seq->private;
2419 struct net *net = seq_file_net(seq);
2423 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2426 if (state->bucket <= state->udp_table->mask)
2427 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2428 return udp_get_first(seq, state->bucket + 1);
2433 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2435 struct sock *sk = udp_get_first(seq, 0);
2438 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2440 return pos ? NULL : sk;
2443 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2445 struct udp_iter_state *state = seq->private;
2446 state->bucket = MAX_UDP_PORTS;
2448 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2451 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2455 if (v == SEQ_START_TOKEN)
2456 sk = udp_get_idx(seq, 0);
2458 sk = udp_get_next(seq, v);
2464 static void udp_seq_stop(struct seq_file *seq, void *v)
2466 struct udp_iter_state *state = seq->private;
2468 if (state->bucket <= state->udp_table->mask)
2469 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2472 int udp_seq_open(struct inode *inode, struct file *file)
2474 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2475 struct udp_iter_state *s;
2478 err = seq_open_net(inode, file, &afinfo->seq_ops,
2479 sizeof(struct udp_iter_state));
2483 s = ((struct seq_file *)file->private_data)->private;
2484 s->family = afinfo->family;
2485 s->udp_table = afinfo->udp_table;
2488 EXPORT_SYMBOL(udp_seq_open);
2490 /* ------------------------------------------------------------------------ */
2491 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2493 struct proc_dir_entry *p;
2496 afinfo->seq_ops.start = udp_seq_start;
2497 afinfo->seq_ops.next = udp_seq_next;
2498 afinfo->seq_ops.stop = udp_seq_stop;
2500 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2501 afinfo->seq_fops, afinfo);
2506 EXPORT_SYMBOL(udp_proc_register);
2508 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2510 remove_proc_entry(afinfo->name, net->proc_net);
2512 EXPORT_SYMBOL(udp_proc_unregister);
2514 /* ------------------------------------------------------------------------ */
2515 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2518 struct inet_sock *inet = inet_sk(sp);
2519 __be32 dest = inet->inet_daddr;
2520 __be32 src = inet->inet_rcv_saddr;
2521 __u16 destp = ntohs(inet->inet_dport);
2522 __u16 srcp = ntohs(inet->inet_sport);
2524 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2525 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2526 bucket, src, srcp, dest, destp, sp->sk_state,
2527 sk_wmem_alloc_get(sp),
2528 sk_rmem_alloc_get(sp),
2530 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2532 atomic_read(&sp->sk_refcnt), sp,
2533 atomic_read(&sp->sk_drops));
2536 int udp4_seq_show(struct seq_file *seq, void *v)
2538 seq_setwidth(seq, 127);
2539 if (v == SEQ_START_TOKEN)
2540 seq_puts(seq, " sl local_address rem_address st tx_queue "
2541 "rx_queue tr tm->when retrnsmt uid timeout "
2542 "inode ref pointer drops");
2544 struct udp_iter_state *state = seq->private;
2546 udp4_format_sock(v, seq, state->bucket);
2552 static const struct file_operations udp_afinfo_seq_fops = {
2553 .owner = THIS_MODULE,
2554 .open = udp_seq_open,
2556 .llseek = seq_lseek,
2557 .release = seq_release_net
2560 /* ------------------------------------------------------------------------ */
2561 static struct udp_seq_afinfo udp4_seq_afinfo = {
2564 .udp_table = &udp_table,
2565 .seq_fops = &udp_afinfo_seq_fops,
2567 .show = udp4_seq_show,
2571 static int __net_init udp4_proc_init_net(struct net *net)
2573 return udp_proc_register(net, &udp4_seq_afinfo);
2576 static void __net_exit udp4_proc_exit_net(struct net *net)
2578 udp_proc_unregister(net, &udp4_seq_afinfo);
2581 static struct pernet_operations udp4_net_ops = {
2582 .init = udp4_proc_init_net,
2583 .exit = udp4_proc_exit_net,
2586 int __init udp4_proc_init(void)
2588 return register_pernet_subsys(&udp4_net_ops);
2591 void udp4_proc_exit(void)
2593 unregister_pernet_subsys(&udp4_net_ops);
2595 #endif /* CONFIG_PROC_FS */
2597 static __initdata unsigned long uhash_entries;
2598 static int __init set_uhash_entries(char *str)
2605 ret = kstrtoul(str, 0, &uhash_entries);
2609 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2610 uhash_entries = UDP_HTABLE_SIZE_MIN;
2613 __setup("uhash_entries=", set_uhash_entries);
2615 void __init udp_table_init(struct udp_table *table, const char *name)
2619 table->hash = alloc_large_system_hash(name,
2620 2 * sizeof(struct udp_hslot),
2622 21, /* one slot per 2 MB */
2626 UDP_HTABLE_SIZE_MIN,
2629 table->hash2 = table->hash + (table->mask + 1);
2630 for (i = 0; i <= table->mask; i++) {
2631 INIT_HLIST_HEAD(&table->hash[i].head);
2632 table->hash[i].count = 0;
2633 spin_lock_init(&table->hash[i].lock);
2635 for (i = 0; i <= table->mask; i++) {
2636 INIT_HLIST_HEAD(&table->hash2[i].head);
2637 table->hash2[i].count = 0;
2638 spin_lock_init(&table->hash2[i].lock);
2642 u32 udp_flow_hashrnd(void)
2644 static u32 hashrnd __read_mostly;
2646 net_get_random_once(&hashrnd, sizeof(hashrnd));
2650 EXPORT_SYMBOL(udp_flow_hashrnd);
2652 void __init udp_init(void)
2654 unsigned long limit;
2657 udp_table_init(&udp_table, "UDP");
2658 limit = nr_free_buffer_pages() / 8;
2659 limit = max(limit, 128UL);
2660 sysctl_udp_mem[0] = limit / 4 * 3;
2661 sysctl_udp_mem[1] = limit;
2662 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2664 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2665 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2667 /* 16 spinlocks per cpu */
2668 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2669 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2672 panic("UDP: failed to alloc udp_busylocks\n");
2673 for (i = 0; i < (1U << udp_busylocks_log); i++)
2674 spin_lock_init(udp_busylocks + i);