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 return reuseport_alloc(sk);
238 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
240 * @sk: socket struct in question
241 * @snum: port number to look up
242 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
245 int udp_lib_get_port(struct sock *sk, unsigned short snum,
246 unsigned int hash2_nulladdr)
248 struct udp_hslot *hslot, *hslot2;
249 struct udp_table *udptable = sk->sk_prot->h.udp_table;
251 struct net *net = sock_net(sk);
254 int low, high, remaining;
256 unsigned short first, last;
257 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
259 inet_get_local_port_range(net, &low, &high);
260 remaining = (high - low) + 1;
262 rand = prandom_u32();
263 first = reciprocal_scale(rand, remaining) + low;
265 * force rand to be an odd multiple of UDP_HTABLE_SIZE
267 rand = (rand | 1) * (udptable->mask + 1);
268 last = first + udptable->mask + 1;
270 hslot = udp_hashslot(udptable, net, first);
271 bitmap_zero(bitmap, PORTS_PER_CHAIN);
272 spin_lock_bh(&hslot->lock);
273 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
278 * Iterate on all possible values of snum for this hash.
279 * Using steps of an odd multiple of UDP_HTABLE_SIZE
280 * give us randomization and full range coverage.
283 if (low <= snum && snum <= high &&
284 !test_bit(snum >> udptable->log, bitmap) &&
285 !inet_is_local_reserved_port(net, snum))
288 } while (snum != first);
289 spin_unlock_bh(&hslot->lock);
291 } while (++first != last);
294 hslot = udp_hashslot(udptable, net, snum);
295 spin_lock_bh(&hslot->lock);
296 if (hslot->count > 10) {
298 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
300 slot2 &= udptable->mask;
301 hash2_nulladdr &= udptable->mask;
303 hslot2 = udp_hashslot2(udptable, slot2);
304 if (hslot->count < hslot2->count)
305 goto scan_primary_hash;
307 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
308 if (!exist && (hash2_nulladdr != slot2)) {
309 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
310 exist = udp_lib_lport_inuse2(net, snum, hslot2,
319 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
323 inet_sk(sk)->inet_num = snum;
324 udp_sk(sk)->udp_port_hash = snum;
325 udp_sk(sk)->udp_portaddr_hash ^= snum;
326 if (sk_unhashed(sk)) {
327 if (sk->sk_reuseport &&
328 udp_reuseport_add_sock(sk, hslot)) {
329 inet_sk(sk)->inet_num = 0;
330 udp_sk(sk)->udp_port_hash = 0;
331 udp_sk(sk)->udp_portaddr_hash ^= snum;
335 sk_add_node_rcu(sk, &hslot->head);
337 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
339 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
340 spin_lock(&hslot2->lock);
341 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
342 sk->sk_family == AF_INET6)
343 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
346 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
349 spin_unlock(&hslot2->lock);
351 sock_set_flag(sk, SOCK_RCU_FREE);
354 spin_unlock_bh(&hslot->lock);
358 EXPORT_SYMBOL(udp_lib_get_port);
360 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
363 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
366 int udp_v4_get_port(struct sock *sk, unsigned short snum)
368 unsigned int hash2_nulladdr =
369 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
370 unsigned int hash2_partial =
371 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
373 /* precompute partial secondary hash */
374 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
375 return udp_lib_get_port(sk, snum, hash2_nulladdr);
378 static int compute_score(struct sock *sk, struct net *net,
379 __be32 saddr, __be16 sport,
380 __be32 daddr, unsigned short hnum,
381 int dif, int sdif, bool exact_dif)
384 struct inet_sock *inet;
386 if (!net_eq(sock_net(sk), net) ||
387 udp_sk(sk)->udp_port_hash != hnum ||
391 score = (sk->sk_family == PF_INET) ? 2 : 1;
394 if (inet->inet_rcv_saddr) {
395 if (inet->inet_rcv_saddr != daddr)
400 if (inet->inet_daddr) {
401 if (inet->inet_daddr != saddr)
406 if (inet->inet_dport) {
407 if (inet->inet_dport != sport)
412 if (sk->sk_bound_dev_if || exact_dif) {
413 bool dev_match = (sk->sk_bound_dev_if == dif ||
414 sk->sk_bound_dev_if == sdif);
418 if (sk->sk_bound_dev_if)
422 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
427 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
428 const __u16 lport, const __be32 faddr,
431 static u32 udp_ehash_secret __read_mostly;
433 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
435 return __inet_ehashfn(laddr, lport, faddr, fport,
436 udp_ehash_secret + net_hash_mix(net));
439 /* called with rcu_read_lock() */
440 static struct sock *udp4_lib_lookup2(struct net *net,
441 __be32 saddr, __be16 sport,
442 __be32 daddr, unsigned int hnum,
443 int dif, int sdif, bool exact_dif,
444 struct udp_hslot *hslot2,
447 struct sock *sk, *result;
448 int score, badness, matches = 0, reuseport = 0;
453 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
454 score = compute_score(sk, net, saddr, sport,
455 daddr, hnum, dif, sdif, exact_dif);
456 if (score > badness) {
457 reuseport = sk->sk_reuseport;
459 hash = udp_ehashfn(net, daddr, hnum,
461 result = reuseport_select_sock(sk, hash, skb,
462 sizeof(struct udphdr));
469 } else if (score == badness && reuseport) {
471 if (reciprocal_scale(hash, matches) == 0)
473 hash = next_pseudo_random32(hash);
479 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
480 * harder than this. -DaveM
482 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
483 __be16 sport, __be32 daddr, __be16 dport, int dif,
484 int sdif, struct udp_table *udptable, struct sk_buff *skb)
486 struct sock *sk, *result;
487 unsigned short hnum = ntohs(dport);
488 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
489 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
490 bool exact_dif = udp_lib_exact_dif_match(net, skb);
491 int score, badness, matches = 0, reuseport = 0;
494 if (hslot->count > 10) {
495 hash2 = udp4_portaddr_hash(net, daddr, hnum);
496 slot2 = hash2 & udptable->mask;
497 hslot2 = &udptable->hash2[slot2];
498 if (hslot->count < hslot2->count)
501 result = udp4_lib_lookup2(net, saddr, sport,
502 daddr, hnum, dif, sdif,
503 exact_dif, hslot2, skb);
505 unsigned int old_slot2 = slot2;
506 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
507 slot2 = hash2 & udptable->mask;
508 /* avoid searching the same slot again. */
509 if (unlikely(slot2 == old_slot2))
512 hslot2 = &udptable->hash2[slot2];
513 if (hslot->count < hslot2->count)
516 result = udp4_lib_lookup2(net, saddr, sport,
517 daddr, hnum, dif, sdif,
518 exact_dif, hslot2, skb);
525 sk_for_each_rcu(sk, &hslot->head) {
526 score = compute_score(sk, net, saddr, sport,
527 daddr, hnum, dif, sdif, exact_dif);
528 if (score > badness) {
529 reuseport = sk->sk_reuseport;
531 hash = udp_ehashfn(net, daddr, hnum,
533 result = reuseport_select_sock(sk, hash, skb,
534 sizeof(struct udphdr));
541 } else if (score == badness && reuseport) {
543 if (reciprocal_scale(hash, matches) == 0)
545 hash = next_pseudo_random32(hash);
550 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
552 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
553 __be16 sport, __be16 dport,
554 struct udp_table *udptable)
556 const struct iphdr *iph = ip_hdr(skb);
558 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
559 iph->daddr, dport, inet_iif(skb),
560 inet_sdif(skb), udptable, skb);
563 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
564 __be16 sport, __be16 dport)
566 const struct iphdr *iph = ip_hdr(skb);
568 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
569 iph->daddr, dport, inet_iif(skb),
570 inet_sdif(skb), &udp_table, NULL);
572 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
574 /* Must be called under rcu_read_lock().
575 * Does increment socket refcount.
577 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
578 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
579 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
580 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
581 __be32 daddr, __be16 dport, int dif)
585 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
586 dif, 0, &udp_table, NULL);
587 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
591 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
594 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
595 __be16 loc_port, __be32 loc_addr,
596 __be16 rmt_port, __be32 rmt_addr,
597 int dif, int sdif, unsigned short hnum)
599 struct inet_sock *inet = inet_sk(sk);
601 if (!net_eq(sock_net(sk), net) ||
602 udp_sk(sk)->udp_port_hash != hnum ||
603 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
604 (inet->inet_dport != rmt_port && inet->inet_dport) ||
605 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
606 ipv6_only_sock(sk) ||
607 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
608 sk->sk_bound_dev_if != sdif))
610 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
616 * This routine is called by the ICMP module when it gets some
617 * sort of error condition. If err < 0 then the socket should
618 * be closed and the error returned to the user. If err > 0
619 * it's just the icmp type << 8 | icmp code.
620 * Header points to the ip header of the error packet. We move
621 * on past this. Then (as it used to claim before adjustment)
622 * header points to the first 8 bytes of the udp header. We need
623 * to find the appropriate port.
626 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
628 struct inet_sock *inet;
629 const struct iphdr *iph = (const struct iphdr *)skb->data;
630 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
631 const int type = icmp_hdr(skb)->type;
632 const int code = icmp_hdr(skb)->code;
636 struct net *net = dev_net(skb->dev);
638 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
639 iph->saddr, uh->source, skb->dev->ifindex, 0,
642 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
643 return; /* No socket for error */
652 case ICMP_TIME_EXCEEDED:
655 case ICMP_SOURCE_QUENCH:
657 case ICMP_PARAMETERPROB:
661 case ICMP_DEST_UNREACH:
662 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
663 ipv4_sk_update_pmtu(skb, sk, info);
664 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
672 if (code <= NR_ICMP_UNREACH) {
673 harderr = icmp_err_convert[code].fatal;
674 err = icmp_err_convert[code].errno;
678 ipv4_sk_redirect(skb, sk);
683 * RFC1122: OK. Passes ICMP errors back to application, as per
686 if (!inet->recverr) {
687 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
690 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
693 sk->sk_error_report(sk);
698 void udp_err(struct sk_buff *skb, u32 info)
700 __udp4_lib_err(skb, info, &udp_table);
704 * Throw away all pending data and cancel the corking. Socket is locked.
706 void udp_flush_pending_frames(struct sock *sk)
708 struct udp_sock *up = udp_sk(sk);
713 ip_flush_pending_frames(sk);
716 EXPORT_SYMBOL(udp_flush_pending_frames);
719 * udp4_hwcsum - handle outgoing HW checksumming
720 * @skb: sk_buff containing the filled-in UDP header
721 * (checksum field must be zeroed out)
722 * @src: source IP address
723 * @dst: destination IP address
725 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
727 struct udphdr *uh = udp_hdr(skb);
728 int offset = skb_transport_offset(skb);
729 int len = skb->len - offset;
733 if (!skb_has_frag_list(skb)) {
735 * Only one fragment on the socket.
737 skb->csum_start = skb_transport_header(skb) - skb->head;
738 skb->csum_offset = offsetof(struct udphdr, check);
739 uh->check = ~csum_tcpudp_magic(src, dst, len,
742 struct sk_buff *frags;
745 * HW-checksum won't work as there are two or more
746 * fragments on the socket so that all csums of sk_buffs
749 skb_walk_frags(skb, frags) {
750 csum = csum_add(csum, frags->csum);
754 csum = skb_checksum(skb, offset, hlen, csum);
755 skb->ip_summed = CHECKSUM_NONE;
757 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
759 uh->check = CSUM_MANGLED_0;
762 EXPORT_SYMBOL_GPL(udp4_hwcsum);
764 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
765 * for the simple case like when setting the checksum for a UDP tunnel.
767 void udp_set_csum(bool nocheck, struct sk_buff *skb,
768 __be32 saddr, __be32 daddr, int len)
770 struct udphdr *uh = udp_hdr(skb);
774 } else if (skb_is_gso(skb)) {
775 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
776 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
778 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
780 uh->check = CSUM_MANGLED_0;
782 skb->ip_summed = CHECKSUM_PARTIAL;
783 skb->csum_start = skb_transport_header(skb) - skb->head;
784 skb->csum_offset = offsetof(struct udphdr, check);
785 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
788 EXPORT_SYMBOL(udp_set_csum);
790 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
792 struct sock *sk = skb->sk;
793 struct inet_sock *inet = inet_sk(sk);
796 int is_udplite = IS_UDPLITE(sk);
797 int offset = skb_transport_offset(skb);
798 int len = skb->len - offset;
802 * Create a UDP header
805 uh->source = inet->inet_sport;
806 uh->dest = fl4->fl4_dport;
807 uh->len = htons(len);
810 if (is_udplite) /* UDP-Lite */
811 csum = udplite_csum(skb);
813 else if (sk->sk_no_check_tx) { /* UDP csum off */
815 skb->ip_summed = CHECKSUM_NONE;
818 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
820 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
824 csum = udp_csum(skb);
826 /* add protocol-dependent pseudo-header */
827 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
828 sk->sk_protocol, csum);
830 uh->check = CSUM_MANGLED_0;
833 err = ip_send_skb(sock_net(sk), skb);
835 if (err == -ENOBUFS && !inet->recverr) {
836 UDP_INC_STATS(sock_net(sk),
837 UDP_MIB_SNDBUFERRORS, is_udplite);
841 UDP_INC_STATS(sock_net(sk),
842 UDP_MIB_OUTDATAGRAMS, is_udplite);
847 * Push out all pending data as one UDP datagram. Socket is locked.
849 int udp_push_pending_frames(struct sock *sk)
851 struct udp_sock *up = udp_sk(sk);
852 struct inet_sock *inet = inet_sk(sk);
853 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
857 skb = ip_finish_skb(sk, fl4);
861 err = udp_send_skb(skb, fl4);
868 EXPORT_SYMBOL(udp_push_pending_frames);
870 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
872 struct inet_sock *inet = inet_sk(sk);
873 struct udp_sock *up = udp_sk(sk);
874 struct flowi4 fl4_stack;
877 struct ipcm_cookie ipc;
878 struct rtable *rt = NULL;
881 __be32 daddr, faddr, saddr;
884 int err, is_udplite = IS_UDPLITE(sk);
885 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
886 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
888 struct ip_options_data opt_copy;
897 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
905 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
907 fl4 = &inet->cork.fl.u.ip4;
910 * There are pending frames.
911 * The socket lock must be held while it's corked.
914 if (likely(up->pending)) {
915 if (unlikely(up->pending != AF_INET)) {
923 ulen += sizeof(struct udphdr);
926 * Get and verify the address.
929 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
930 if (msg->msg_namelen < sizeof(*usin))
932 if (usin->sin_family != AF_INET) {
933 if (usin->sin_family != AF_UNSPEC)
934 return -EAFNOSUPPORT;
937 daddr = usin->sin_addr.s_addr;
938 dport = usin->sin_port;
942 if (sk->sk_state != TCP_ESTABLISHED)
943 return -EDESTADDRREQ;
944 daddr = inet->inet_daddr;
945 dport = inet->inet_dport;
946 /* Open fast path for connected socket.
947 Route will not be used, if at least one option is set.
952 ipc.sockc.tsflags = sk->sk_tsflags;
953 ipc.addr = inet->inet_saddr;
954 ipc.oif = sk->sk_bound_dev_if;
956 if (msg->msg_controllen) {
957 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
967 struct ip_options_rcu *inet_opt;
970 inet_opt = rcu_dereference(inet->inet_opt);
972 memcpy(&opt_copy, inet_opt,
973 sizeof(*inet_opt) + inet_opt->opt.optlen);
974 ipc.opt = &opt_copy.opt;
980 ipc.addr = faddr = daddr;
982 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
984 if (ipc.opt && ipc.opt->opt.srr) {
989 faddr = ipc.opt->opt.faddr;
992 tos = get_rttos(&ipc, inet);
993 if (sock_flag(sk, SOCK_LOCALROUTE) ||
994 (msg->msg_flags & MSG_DONTROUTE) ||
995 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1000 if (ipv4_is_multicast(daddr)) {
1002 ipc.oif = inet->mc_index;
1004 saddr = inet->mc_addr;
1006 } else if (!ipc.oif)
1007 ipc.oif = inet->uc_index;
1010 rt = (struct rtable *)sk_dst_check(sk, 0);
1013 struct net *net = sock_net(sk);
1014 __u8 flow_flags = inet_sk_flowi_flags(sk);
1018 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1019 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1021 faddr, saddr, dport, inet->inet_sport,
1024 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1025 rt = ip_route_output_flow(net, fl4, sk);
1029 if (err == -ENETUNREACH)
1030 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1035 if ((rt->rt_flags & RTCF_BROADCAST) &&
1036 !sock_flag(sk, SOCK_BROADCAST))
1039 sk_dst_set(sk, dst_clone(&rt->dst));
1042 if (msg->msg_flags&MSG_CONFIRM)
1048 daddr = ipc.addr = fl4->daddr;
1050 /* Lockless fast path for the non-corking case. */
1052 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1053 sizeof(struct udphdr), &ipc, &rt,
1056 if (!IS_ERR_OR_NULL(skb))
1057 err = udp_send_skb(skb, fl4);
1062 if (unlikely(up->pending)) {
1063 /* The socket is already corked while preparing it. */
1064 /* ... which is an evident application bug. --ANK */
1067 net_dbg_ratelimited("socket already corked\n");
1072 * Now cork the socket to pend data.
1074 fl4 = &inet->cork.fl.u.ip4;
1077 fl4->fl4_dport = dport;
1078 fl4->fl4_sport = inet->inet_sport;
1079 up->pending = AF_INET;
1083 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1084 sizeof(struct udphdr), &ipc, &rt,
1085 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1087 udp_flush_pending_frames(sk);
1089 err = udp_push_pending_frames(sk);
1090 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1102 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1103 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1104 * we don't have a good statistic (IpOutDiscards but it can be too many
1105 * things). We could add another new stat but at least for now that
1106 * seems like overkill.
1108 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1109 UDP_INC_STATS(sock_net(sk),
1110 UDP_MIB_SNDBUFERRORS, is_udplite);
1115 if (msg->msg_flags & MSG_PROBE)
1116 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1117 if (!(msg->msg_flags&MSG_PROBE) || len)
1118 goto back_from_confirm;
1122 EXPORT_SYMBOL(udp_sendmsg);
1124 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1125 size_t size, int flags)
1127 struct inet_sock *inet = inet_sk(sk);
1128 struct udp_sock *up = udp_sk(sk);
1131 if (flags & MSG_SENDPAGE_NOTLAST)
1135 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1137 /* Call udp_sendmsg to specify destination address which
1138 * sendpage interface can't pass.
1139 * This will succeed only when the socket is connected.
1141 ret = udp_sendmsg(sk, &msg, 0);
1148 if (unlikely(!up->pending)) {
1151 net_dbg_ratelimited("cork failed\n");
1155 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1156 page, offset, size, flags);
1157 if (ret == -EOPNOTSUPP) {
1159 return sock_no_sendpage(sk->sk_socket, page, offset,
1163 udp_flush_pending_frames(sk);
1168 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1169 ret = udp_push_pending_frames(sk);
1177 #define UDP_SKB_IS_STATELESS 0x80000000
1179 static void udp_set_dev_scratch(struct sk_buff *skb)
1181 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1183 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1184 scratch->_tsize_state = skb->truesize;
1185 #if BITS_PER_LONG == 64
1186 scratch->len = skb->len;
1187 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1188 scratch->is_linear = !skb_is_nonlinear(skb);
1190 /* all head states execept sp (dst, sk, nf) are always cleared by
1191 * udp_rcv() and we need to preserve secpath, if present, to eventually
1192 * process IP_CMSG_PASSSEC at recvmsg() time
1194 if (likely(!skb_sec_path(skb)))
1195 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1198 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1200 /* We come here after udp_lib_checksum_complete() returned 0.
1201 * This means that __skb_checksum_complete() might have
1202 * set skb->csum_valid to 1.
1203 * On 64bit platforms, we can set csum_unnecessary
1204 * to true, but only if the skb is not shared.
1206 #if BITS_PER_LONG == 64
1207 if (!skb_shared(skb))
1208 udp_skb_scratch(skb)->csum_unnecessary = true;
1212 static int udp_skb_truesize(struct sk_buff *skb)
1214 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1217 static bool udp_skb_has_head_state(struct sk_buff *skb)
1219 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1222 /* fully reclaim rmem/fwd memory allocated for skb */
1223 static void udp_rmem_release(struct sock *sk, int size, int partial,
1224 bool rx_queue_lock_held)
1226 struct udp_sock *up = udp_sk(sk);
1227 struct sk_buff_head *sk_queue;
1230 if (likely(partial)) {
1231 up->forward_deficit += size;
1232 size = up->forward_deficit;
1233 if (size < (sk->sk_rcvbuf >> 2) &&
1234 !skb_queue_empty(&up->reader_queue))
1237 size += up->forward_deficit;
1239 up->forward_deficit = 0;
1241 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1242 * if the called don't held it already
1244 sk_queue = &sk->sk_receive_queue;
1245 if (!rx_queue_lock_held)
1246 spin_lock(&sk_queue->lock);
1249 sk->sk_forward_alloc += size;
1250 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1251 sk->sk_forward_alloc -= amt;
1254 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1256 atomic_sub(size, &sk->sk_rmem_alloc);
1258 /* this can save us from acquiring the rx queue lock on next receive */
1259 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1261 if (!rx_queue_lock_held)
1262 spin_unlock(&sk_queue->lock);
1265 /* Note: called with reader_queue.lock held.
1266 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1267 * This avoids a cache line miss while receive_queue lock is held.
1268 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1270 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1272 prefetch(&skb->data);
1273 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1275 EXPORT_SYMBOL(udp_skb_destructor);
1277 /* as above, but the caller held the rx queue lock, too */
1278 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1280 prefetch(&skb->data);
1281 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1284 /* Idea of busylocks is to let producers grab an extra spinlock
1285 * to relieve pressure on the receive_queue spinlock shared by consumer.
1286 * Under flood, this means that only one producer can be in line
1287 * trying to acquire the receive_queue spinlock.
1288 * These busylock can be allocated on a per cpu manner, instead of a
1289 * per socket one (that would consume a cache line per socket)
1291 static int udp_busylocks_log __read_mostly;
1292 static spinlock_t *udp_busylocks __read_mostly;
1294 static spinlock_t *busylock_acquire(void *ptr)
1298 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1303 static void busylock_release(spinlock_t *busy)
1309 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1311 struct sk_buff_head *list = &sk->sk_receive_queue;
1312 int rmem, delta, amt, err = -ENOMEM;
1313 spinlock_t *busy = NULL;
1316 /* try to avoid the costly atomic add/sub pair when the receive
1317 * queue is full; always allow at least a packet
1319 rmem = atomic_read(&sk->sk_rmem_alloc);
1320 if (rmem > sk->sk_rcvbuf)
1323 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1324 * having linear skbs :
1325 * - Reduce memory overhead and thus increase receive queue capacity
1326 * - Less cache line misses at copyout() time
1327 * - Less work at consume_skb() (less alien page frag freeing)
1329 if (rmem > (sk->sk_rcvbuf >> 1)) {
1332 busy = busylock_acquire(sk);
1334 size = skb->truesize;
1335 udp_set_dev_scratch(skb);
1337 /* we drop only if the receive buf is full and the receive
1338 * queue contains some other skb
1340 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1341 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1344 spin_lock(&list->lock);
1345 if (size >= sk->sk_forward_alloc) {
1346 amt = sk_mem_pages(size);
1347 delta = amt << SK_MEM_QUANTUM_SHIFT;
1348 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1350 spin_unlock(&list->lock);
1354 sk->sk_forward_alloc += delta;
1357 sk->sk_forward_alloc -= size;
1359 /* no need to setup a destructor, we will explicitly release the
1360 * forward allocated memory on dequeue
1362 sock_skb_set_dropcount(sk, skb);
1364 __skb_queue_tail(list, skb);
1365 spin_unlock(&list->lock);
1367 if (!sock_flag(sk, SOCK_DEAD))
1368 sk->sk_data_ready(sk);
1370 busylock_release(busy);
1374 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1377 atomic_inc(&sk->sk_drops);
1378 busylock_release(busy);
1381 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1383 void udp_destruct_common(struct sock *sk)
1385 /* reclaim completely the forward allocated memory */
1386 struct udp_sock *up = udp_sk(sk);
1387 unsigned int total = 0;
1388 struct sk_buff *skb;
1390 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1391 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1392 total += skb->truesize;
1395 udp_rmem_release(sk, total, 0, true);
1397 EXPORT_SYMBOL_GPL(udp_destruct_common);
1399 static void udp_destruct_sock(struct sock *sk)
1401 udp_destruct_common(sk);
1402 inet_sock_destruct(sk);
1405 int udp_init_sock(struct sock *sk)
1407 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1408 sk->sk_destruct = udp_destruct_sock;
1412 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1414 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1415 bool slow = lock_sock_fast(sk);
1417 sk_peek_offset_bwd(sk, len);
1418 unlock_sock_fast(sk, slow);
1421 if (!skb_unref(skb))
1424 /* In the more common cases we cleared the head states previously,
1425 * see __udp_queue_rcv_skb().
1427 if (unlikely(udp_skb_has_head_state(skb)))
1428 skb_release_head_state(skb);
1429 __consume_stateless_skb(skb);
1431 EXPORT_SYMBOL_GPL(skb_consume_udp);
1433 static struct sk_buff *__first_packet_length(struct sock *sk,
1434 struct sk_buff_head *rcvq,
1437 struct sk_buff *skb;
1439 while ((skb = skb_peek(rcvq)) != NULL) {
1440 if (udp_lib_checksum_complete(skb)) {
1441 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1443 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1445 atomic_inc(&sk->sk_drops);
1446 __skb_unlink(skb, rcvq);
1447 *total += skb->truesize;
1450 udp_skb_csum_unnecessary_set(skb);
1458 * first_packet_length - return length of first packet in receive queue
1461 * Drops all bad checksum frames, until a valid one is found.
1462 * Returns the length of found skb, or -1 if none is found.
1464 static int first_packet_length(struct sock *sk)
1466 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1467 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1468 struct sk_buff *skb;
1472 spin_lock_bh(&rcvq->lock);
1473 skb = __first_packet_length(sk, rcvq, &total);
1474 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1475 spin_lock(&sk_queue->lock);
1476 skb_queue_splice_tail_init(sk_queue, rcvq);
1477 spin_unlock(&sk_queue->lock);
1479 skb = __first_packet_length(sk, rcvq, &total);
1481 res = skb ? skb->len : -1;
1483 udp_rmem_release(sk, total, 1, false);
1484 spin_unlock_bh(&rcvq->lock);
1489 * IOCTL requests applicable to the UDP protocol
1492 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1497 int amount = sk_wmem_alloc_get(sk);
1499 return put_user(amount, (int __user *)arg);
1504 int amount = max_t(int, 0, first_packet_length(sk));
1506 return put_user(amount, (int __user *)arg);
1510 return -ENOIOCTLCMD;
1515 EXPORT_SYMBOL(udp_ioctl);
1517 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1518 int noblock, int *peeked, int *off, int *err)
1520 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1521 struct sk_buff_head *queue;
1522 struct sk_buff *last;
1526 queue = &udp_sk(sk)->reader_queue;
1527 flags |= noblock ? MSG_DONTWAIT : 0;
1528 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1530 struct sk_buff *skb;
1532 error = sock_error(sk);
1539 spin_lock_bh(&queue->lock);
1540 skb = __skb_try_recv_from_queue(sk, queue, flags,
1545 spin_unlock_bh(&queue->lock);
1549 if (skb_queue_empty_lockless(sk_queue)) {
1550 spin_unlock_bh(&queue->lock);
1554 /* refill the reader queue and walk it again
1555 * keep both queues locked to avoid re-acquiring
1556 * the sk_receive_queue lock if fwd memory scheduling
1559 spin_lock(&sk_queue->lock);
1560 skb_queue_splice_tail_init(sk_queue, queue);
1562 skb = __skb_try_recv_from_queue(sk, queue, flags,
1563 udp_skb_dtor_locked,
1566 spin_unlock(&sk_queue->lock);
1567 spin_unlock_bh(&queue->lock);
1572 if (!sk_can_busy_loop(sk))
1575 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1576 } while (!skb_queue_empty_lockless(sk_queue));
1578 /* sk_queue is empty, reader_queue may contain peeked packets */
1580 !__skb_wait_for_more_packets(sk, &error, &timeo,
1581 (struct sk_buff *)sk_queue));
1586 EXPORT_SYMBOL(__skb_recv_udp);
1589 * This should be easy, if there is something there we
1590 * return it, otherwise we block.
1593 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1594 int flags, int *addr_len)
1596 struct inet_sock *inet = inet_sk(sk);
1597 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1598 struct sk_buff *skb;
1599 unsigned int ulen, copied;
1600 int peeked, peeking, off;
1602 int is_udplite = IS_UDPLITE(sk);
1603 bool checksum_valid = false;
1605 if (flags & MSG_ERRQUEUE)
1606 return ip_recv_error(sk, msg, len, addr_len);
1609 peeking = flags & MSG_PEEK;
1610 off = sk_peek_offset(sk, flags);
1611 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1615 ulen = udp_skb_len(skb);
1617 if (copied > ulen - off)
1618 copied = ulen - off;
1619 else if (copied < ulen)
1620 msg->msg_flags |= MSG_TRUNC;
1623 * If checksum is needed at all, try to do it while copying the
1624 * data. If the data is truncated, or if we only want a partial
1625 * coverage checksum (UDP-Lite), do it before the copy.
1628 if (copied < ulen || peeking ||
1629 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1630 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1631 !__udp_lib_checksum_complete(skb);
1632 if (!checksum_valid)
1636 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1637 if (udp_skb_is_linear(skb))
1638 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1640 err = skb_copy_datagram_msg(skb, off, msg, copied);
1642 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1648 if (unlikely(err)) {
1650 atomic_inc(&sk->sk_drops);
1651 UDP_INC_STATS(sock_net(sk),
1652 UDP_MIB_INERRORS, is_udplite);
1659 UDP_INC_STATS(sock_net(sk),
1660 UDP_MIB_INDATAGRAMS, is_udplite);
1662 sock_recv_ts_and_drops(msg, sk, skb);
1664 /* Copy the address. */
1666 sin->sin_family = AF_INET;
1667 sin->sin_port = udp_hdr(skb)->source;
1668 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1669 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1670 *addr_len = sizeof(*sin);
1672 if (inet->cmsg_flags)
1673 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1676 if (flags & MSG_TRUNC)
1679 skb_consume_udp(sk, skb, peeking ? -err : err);
1683 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1684 udp_skb_destructor)) {
1685 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1686 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1690 /* starting over for a new packet, but check if we need to yield */
1692 msg->msg_flags &= ~MSG_TRUNC;
1696 int __udp_disconnect(struct sock *sk, int flags)
1698 struct inet_sock *inet = inet_sk(sk);
1700 * 1003.1g - break association.
1703 sk->sk_state = TCP_CLOSE;
1704 inet->inet_daddr = 0;
1705 inet->inet_dport = 0;
1706 sock_rps_reset_rxhash(sk);
1707 sk->sk_bound_dev_if = 0;
1708 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1709 inet_reset_saddr(sk);
1711 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1712 sk->sk_prot->unhash(sk);
1713 inet->inet_sport = 0;
1718 EXPORT_SYMBOL(__udp_disconnect);
1720 int udp_disconnect(struct sock *sk, int flags)
1723 __udp_disconnect(sk, flags);
1727 EXPORT_SYMBOL(udp_disconnect);
1729 void udp_lib_unhash(struct sock *sk)
1731 if (sk_hashed(sk)) {
1732 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1733 struct udp_hslot *hslot, *hslot2;
1735 hslot = udp_hashslot(udptable, sock_net(sk),
1736 udp_sk(sk)->udp_port_hash);
1737 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1739 spin_lock_bh(&hslot->lock);
1740 if (rcu_access_pointer(sk->sk_reuseport_cb))
1741 reuseport_detach_sock(sk);
1742 if (sk_del_node_init_rcu(sk)) {
1744 inet_sk(sk)->inet_num = 0;
1745 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1747 spin_lock(&hslot2->lock);
1748 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1750 spin_unlock(&hslot2->lock);
1752 spin_unlock_bh(&hslot->lock);
1755 EXPORT_SYMBOL(udp_lib_unhash);
1758 * inet_rcv_saddr was changed, we must rehash secondary hash
1760 void udp_lib_rehash(struct sock *sk, u16 newhash)
1762 if (sk_hashed(sk)) {
1763 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1764 struct udp_hslot *hslot, *hslot2, *nhslot2;
1766 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1767 nhslot2 = udp_hashslot2(udptable, newhash);
1768 udp_sk(sk)->udp_portaddr_hash = newhash;
1770 if (hslot2 != nhslot2 ||
1771 rcu_access_pointer(sk->sk_reuseport_cb)) {
1772 hslot = udp_hashslot(udptable, sock_net(sk),
1773 udp_sk(sk)->udp_port_hash);
1774 /* we must lock primary chain too */
1775 spin_lock_bh(&hslot->lock);
1776 if (rcu_access_pointer(sk->sk_reuseport_cb))
1777 reuseport_detach_sock(sk);
1779 if (hslot2 != nhslot2) {
1780 spin_lock(&hslot2->lock);
1781 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1783 spin_unlock(&hslot2->lock);
1785 spin_lock(&nhslot2->lock);
1786 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1789 spin_unlock(&nhslot2->lock);
1792 spin_unlock_bh(&hslot->lock);
1796 EXPORT_SYMBOL(udp_lib_rehash);
1798 static void udp_v4_rehash(struct sock *sk)
1800 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1801 inet_sk(sk)->inet_rcv_saddr,
1802 inet_sk(sk)->inet_num);
1803 udp_lib_rehash(sk, new_hash);
1806 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1810 if (inet_sk(sk)->inet_daddr) {
1811 sock_rps_save_rxhash(sk, skb);
1812 sk_mark_napi_id(sk, skb);
1813 sk_incoming_cpu_update(sk);
1815 sk_mark_napi_id_once(sk, skb);
1818 rc = __udp_enqueue_schedule_skb(sk, skb);
1820 int is_udplite = IS_UDPLITE(sk);
1822 /* Note that an ENOMEM error is charged twice */
1824 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1826 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1828 trace_udp_fail_queue_rcv_skb(rc, sk);
1835 static struct static_key udp_encap_needed __read_mostly;
1836 void udp_encap_enable(void)
1838 static_key_enable(&udp_encap_needed);
1840 EXPORT_SYMBOL(udp_encap_enable);
1845 * >0: "udp encap" protocol resubmission
1847 * Note that in the success and error cases, the skb is assumed to
1848 * have either been requeued or freed.
1850 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1852 struct udp_sock *up = udp_sk(sk);
1853 int is_udplite = IS_UDPLITE(sk);
1856 * Charge it to the socket, dropping if the queue is full.
1858 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1862 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1863 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1866 * This is an encapsulation socket so pass the skb to
1867 * the socket's udp_encap_rcv() hook. Otherwise, just
1868 * fall through and pass this up the UDP socket.
1869 * up->encap_rcv() returns the following value:
1870 * =0 if skb was successfully passed to the encap
1871 * handler or was discarded by it.
1872 * >0 if skb should be passed on to UDP.
1873 * <0 if skb should be resubmitted as proto -N
1876 /* if we're overly short, let UDP handle it */
1877 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1881 /* Verify checksum before giving to encap */
1882 if (udp_lib_checksum_complete(skb))
1885 ret = encap_rcv(sk, skb);
1887 __UDP_INC_STATS(sock_net(sk),
1888 UDP_MIB_INDATAGRAMS,
1894 /* FALLTHROUGH -- it's a UDP Packet */
1898 * UDP-Lite specific tests, ignored on UDP sockets
1900 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1903 * MIB statistics other than incrementing the error count are
1904 * disabled for the following two types of errors: these depend
1905 * on the application settings, not on the functioning of the
1906 * protocol stack as such.
1908 * RFC 3828 here recommends (sec 3.3): "There should also be a
1909 * way ... to ... at least let the receiving application block
1910 * delivery of packets with coverage values less than a value
1911 * provided by the application."
1913 if (up->pcrlen == 0) { /* full coverage was set */
1914 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1915 UDP_SKB_CB(skb)->cscov, skb->len);
1918 /* The next case involves violating the min. coverage requested
1919 * by the receiver. This is subtle: if receiver wants x and x is
1920 * greater than the buffersize/MTU then receiver will complain
1921 * that it wants x while sender emits packets of smaller size y.
1922 * Therefore the above ...()->partial_cov statement is essential.
1924 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1925 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1926 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1931 prefetch(&sk->sk_rmem_alloc);
1932 if (rcu_access_pointer(sk->sk_filter) &&
1933 udp_lib_checksum_complete(skb))
1936 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1939 udp_csum_pull_header(skb);
1941 ipv4_pktinfo_prepare(sk, skb);
1942 return __udp_queue_rcv_skb(sk, skb);
1945 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1947 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1948 atomic_inc(&sk->sk_drops);
1953 /* For TCP sockets, sk_rx_dst is protected by socket lock
1954 * For UDP, we use xchg() to guard against concurrent changes.
1956 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1958 struct dst_entry *old;
1960 if (dst_hold_safe(dst)) {
1961 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
1967 EXPORT_SYMBOL(udp_sk_rx_dst_set);
1970 * Multicasts and broadcasts go to each listener.
1972 * Note: called only from the BH handler context.
1974 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1976 __be32 saddr, __be32 daddr,
1977 struct udp_table *udptable,
1980 struct sock *sk, *first = NULL;
1981 unsigned short hnum = ntohs(uh->dest);
1982 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1983 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1984 unsigned int offset = offsetof(typeof(*sk), sk_node);
1985 int dif = skb->dev->ifindex;
1986 int sdif = inet_sdif(skb);
1987 struct hlist_node *node;
1988 struct sk_buff *nskb;
1991 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1993 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1995 hslot = &udptable->hash2[hash2];
1996 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1999 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2000 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2001 uh->source, saddr, dif, sdif, hnum))
2008 nskb = skb_clone(skb, GFP_ATOMIC);
2010 if (unlikely(!nskb)) {
2011 atomic_inc(&sk->sk_drops);
2012 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2014 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2018 if (udp_queue_rcv_skb(sk, nskb) > 0)
2022 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2023 if (use_hash2 && hash2 != hash2_any) {
2029 if (udp_queue_rcv_skb(first, skb) > 0)
2033 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2034 proto == IPPROTO_UDPLITE);
2039 /* Initialize UDP checksum. If exited with zero value (success),
2040 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2041 * Otherwise, csum completion requires chacksumming packet body,
2042 * including udp header and folding it to skb->csum.
2044 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2049 UDP_SKB_CB(skb)->partial_cov = 0;
2050 UDP_SKB_CB(skb)->cscov = skb->len;
2052 if (proto == IPPROTO_UDPLITE) {
2053 err = udplite_checksum_init(skb, uh);
2057 if (UDP_SKB_CB(skb)->partial_cov) {
2058 skb->csum = inet_compute_pseudo(skb, proto);
2063 /* Note, we are only interested in != 0 or == 0, thus the
2066 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2067 inet_compute_pseudo);
2071 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2072 /* If SW calculated the value, we know it's bad */
2073 if (skb->csum_complete_sw)
2076 /* HW says the value is bad. Let's validate that.
2077 * skb->csum is no longer the full packet checksum,
2078 * so don't treat it as such.
2080 skb_checksum_complete_unset(skb);
2086 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2087 * return code conversion for ip layer consumption
2089 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2094 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2095 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2096 inet_compute_pseudo);
2098 ret = udp_queue_rcv_skb(sk, skb);
2100 /* a return value > 0 means to resubmit the input, but
2101 * it wants the return to be -protocol, or 0
2109 * All we need to do is get the socket, and then do a checksum.
2112 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2117 unsigned short ulen;
2118 struct rtable *rt = skb_rtable(skb);
2119 __be32 saddr, daddr;
2120 struct net *net = dev_net(skb->dev);
2123 * Validate the packet.
2125 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2126 goto drop; /* No space for header. */
2129 ulen = ntohs(uh->len);
2130 saddr = ip_hdr(skb)->saddr;
2131 daddr = ip_hdr(skb)->daddr;
2133 if (ulen > skb->len)
2136 if (proto == IPPROTO_UDP) {
2137 /* UDP validates ulen. */
2138 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2143 if (udp4_csum_init(skb, uh, proto))
2146 sk = skb_steal_sock(skb);
2148 struct dst_entry *dst = skb_dst(skb);
2151 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2152 udp_sk_rx_dst_set(sk, dst);
2154 ret = udp_unicast_rcv_skb(sk, skb, uh);
2159 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2160 return __udp4_lib_mcast_deliver(net, skb, uh,
2161 saddr, daddr, udptable, proto);
2163 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2165 return udp_unicast_rcv_skb(sk, skb, uh);
2167 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2171 /* No socket. Drop packet silently, if checksum is wrong */
2172 if (udp_lib_checksum_complete(skb))
2175 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2176 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2179 * Hmm. We got an UDP packet to a port to which we
2180 * don't wanna listen. Ignore it.
2186 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2187 proto == IPPROTO_UDPLITE ? "Lite" : "",
2188 &saddr, ntohs(uh->source),
2190 &daddr, ntohs(uh->dest));
2195 * RFC1122: OK. Discards the bad packet silently (as far as
2196 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2198 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2199 proto == IPPROTO_UDPLITE ? "Lite" : "",
2200 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2202 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2204 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2209 /* We can only early demux multicast if there is a single matching socket.
2210 * If more than one socket found returns NULL
2212 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2213 __be16 loc_port, __be32 loc_addr,
2214 __be16 rmt_port, __be32 rmt_addr,
2217 struct sock *sk, *result;
2218 unsigned short hnum = ntohs(loc_port);
2219 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2220 struct udp_hslot *hslot = &udp_table.hash[slot];
2222 /* Do not bother scanning a too big list */
2223 if (hslot->count > 10)
2227 sk_for_each_rcu(sk, &hslot->head) {
2228 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2229 rmt_port, rmt_addr, dif, sdif, hnum)) {
2239 /* For unicast we should only early demux connected sockets or we can
2240 * break forwarding setups. The chains here can be long so only check
2241 * if the first socket is an exact match and if not move on.
2243 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2244 __be16 loc_port, __be32 loc_addr,
2245 __be16 rmt_port, __be32 rmt_addr,
2248 unsigned short hnum = ntohs(loc_port);
2249 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2250 unsigned int slot2 = hash2 & udp_table.mask;
2251 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2252 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2253 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2256 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2257 if (INET_MATCH(sk, net, acookie, rmt_addr,
2258 loc_addr, ports, dif, sdif))
2260 /* Only check first socket in chain */
2266 int udp_v4_early_demux(struct sk_buff *skb)
2268 struct net *net = dev_net(skb->dev);
2269 struct in_device *in_dev = NULL;
2270 const struct iphdr *iph;
2271 const struct udphdr *uh;
2272 struct sock *sk = NULL;
2273 struct dst_entry *dst;
2274 int dif = skb->dev->ifindex;
2275 int sdif = inet_sdif(skb);
2278 /* validate the packet */
2279 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2285 if (skb->pkt_type == PACKET_MULTICAST) {
2286 in_dev = __in_dev_get_rcu(skb->dev);
2291 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2296 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2297 uh->source, iph->saddr,
2299 } else if (skb->pkt_type == PACKET_HOST) {
2300 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2301 uh->source, iph->saddr, dif, sdif);
2304 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2308 skb->destructor = sock_efree;
2309 dst = rcu_dereference(sk->sk_rx_dst);
2312 dst = dst_check(dst, 0);
2316 /* set noref for now.
2317 * any place which wants to hold dst has to call
2320 skb_dst_set_noref(skb, dst);
2322 /* for unconnected multicast sockets we need to validate
2323 * the source on each packet
2325 if (!inet_sk(sk)->inet_daddr && in_dev)
2326 return ip_mc_validate_source(skb, iph->daddr,
2328 iph->tos & IPTOS_RT_MASK,
2329 skb->dev, in_dev, &itag);
2334 int udp_rcv(struct sk_buff *skb)
2336 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2339 void udp_destroy_sock(struct sock *sk)
2341 struct udp_sock *up = udp_sk(sk);
2342 bool slow = lock_sock_fast(sk);
2344 /* protects from races with udp_abort() */
2345 sock_set_flag(sk, SOCK_DEAD);
2346 udp_flush_pending_frames(sk);
2347 unlock_sock_fast(sk, slow);
2348 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2349 void (*encap_destroy)(struct sock *sk);
2350 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2357 * Socket option code for UDP
2359 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2360 char __user *optval, unsigned int optlen,
2361 int (*push_pending_frames)(struct sock *))
2363 struct udp_sock *up = udp_sk(sk);
2366 int is_udplite = IS_UDPLITE(sk);
2368 if (optlen < sizeof(int))
2371 if (get_user(val, (int __user *)optval))
2374 valbool = val ? 1 : 0;
2379 WRITE_ONCE(up->corkflag, 1);
2381 WRITE_ONCE(up->corkflag, 0);
2383 push_pending_frames(sk);
2391 case UDP_ENCAP_ESPINUDP:
2392 case UDP_ENCAP_ESPINUDP_NON_IKE:
2393 up->encap_rcv = xfrm4_udp_encap_rcv;
2395 case UDP_ENCAP_L2TPINUDP:
2396 up->encap_type = val;
2405 case UDP_NO_CHECK6_TX:
2406 up->no_check6_tx = valbool;
2409 case UDP_NO_CHECK6_RX:
2410 up->no_check6_rx = valbool;
2414 * UDP-Lite's partial checksum coverage (RFC 3828).
2416 /* The sender sets actual checksum coverage length via this option.
2417 * The case coverage > packet length is handled by send module. */
2418 case UDPLITE_SEND_CSCOV:
2419 if (!is_udplite) /* Disable the option on UDP sockets */
2420 return -ENOPROTOOPT;
2421 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2423 else if (val > USHRT_MAX)
2426 up->pcflag |= UDPLITE_SEND_CC;
2429 /* The receiver specifies a minimum checksum coverage value. To make
2430 * sense, this should be set to at least 8 (as done below). If zero is
2431 * used, this again means full checksum coverage. */
2432 case UDPLITE_RECV_CSCOV:
2433 if (!is_udplite) /* Disable the option on UDP sockets */
2434 return -ENOPROTOOPT;
2435 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2437 else if (val > USHRT_MAX)
2440 up->pcflag |= UDPLITE_RECV_CC;
2450 EXPORT_SYMBOL(udp_lib_setsockopt);
2452 int udp_setsockopt(struct sock *sk, int level, int optname,
2453 char __user *optval, unsigned int optlen)
2455 if (level == SOL_UDP || level == SOL_UDPLITE)
2456 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2457 udp_push_pending_frames);
2458 return ip_setsockopt(sk, level, optname, optval, optlen);
2461 #ifdef CONFIG_COMPAT
2462 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2463 char __user *optval, unsigned int optlen)
2465 if (level == SOL_UDP || level == SOL_UDPLITE)
2466 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2467 udp_push_pending_frames);
2468 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2472 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2473 char __user *optval, int __user *optlen)
2475 struct udp_sock *up = udp_sk(sk);
2478 if (get_user(len, optlen))
2481 len = min_t(unsigned int, len, sizeof(int));
2488 val = READ_ONCE(up->corkflag);
2492 val = up->encap_type;
2495 case UDP_NO_CHECK6_TX:
2496 val = up->no_check6_tx;
2499 case UDP_NO_CHECK6_RX:
2500 val = up->no_check6_rx;
2503 /* The following two cannot be changed on UDP sockets, the return is
2504 * always 0 (which corresponds to the full checksum coverage of UDP). */
2505 case UDPLITE_SEND_CSCOV:
2509 case UDPLITE_RECV_CSCOV:
2514 return -ENOPROTOOPT;
2517 if (put_user(len, optlen))
2519 if (copy_to_user(optval, &val, len))
2523 EXPORT_SYMBOL(udp_lib_getsockopt);
2525 int udp_getsockopt(struct sock *sk, int level, int optname,
2526 char __user *optval, int __user *optlen)
2528 if (level == SOL_UDP || level == SOL_UDPLITE)
2529 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2530 return ip_getsockopt(sk, level, optname, optval, optlen);
2533 #ifdef CONFIG_COMPAT
2534 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2535 char __user *optval, int __user *optlen)
2537 if (level == SOL_UDP || level == SOL_UDPLITE)
2538 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2539 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2543 * udp_poll - wait for a UDP event.
2544 * @file - file struct
2546 * @wait - poll table
2548 * This is same as datagram poll, except for the special case of
2549 * blocking sockets. If application is using a blocking fd
2550 * and a packet with checksum error is in the queue;
2551 * then it could get return from select indicating data available
2552 * but then block when reading it. Add special case code
2553 * to work around these arguably broken applications.
2555 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2557 unsigned int mask = datagram_poll(file, sock, wait);
2558 struct sock *sk = sock->sk;
2560 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2561 mask |= POLLIN | POLLRDNORM;
2563 sock_rps_record_flow(sk);
2565 /* Check for false positives due to checksum errors */
2566 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2567 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2568 mask &= ~(POLLIN | POLLRDNORM);
2573 EXPORT_SYMBOL(udp_poll);
2575 int udp_abort(struct sock *sk, int err)
2579 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2582 if (sock_flag(sk, SOCK_DEAD))
2586 sk->sk_error_report(sk);
2587 __udp_disconnect(sk, 0);
2594 EXPORT_SYMBOL_GPL(udp_abort);
2596 struct proto udp_prot = {
2598 .owner = THIS_MODULE,
2599 .close = udp_lib_close,
2600 .connect = ip4_datagram_connect,
2601 .disconnect = udp_disconnect,
2603 .init = udp_init_sock,
2604 .destroy = udp_destroy_sock,
2605 .setsockopt = udp_setsockopt,
2606 .getsockopt = udp_getsockopt,
2607 .sendmsg = udp_sendmsg,
2608 .recvmsg = udp_recvmsg,
2609 .sendpage = udp_sendpage,
2610 .release_cb = ip4_datagram_release_cb,
2611 .hash = udp_lib_hash,
2612 .unhash = udp_lib_unhash,
2613 .rehash = udp_v4_rehash,
2614 .get_port = udp_v4_get_port,
2615 .memory_allocated = &udp_memory_allocated,
2616 .sysctl_mem = sysctl_udp_mem,
2617 .sysctl_wmem = &sysctl_udp_wmem_min,
2618 .sysctl_rmem = &sysctl_udp_rmem_min,
2619 .obj_size = sizeof(struct udp_sock),
2620 .h.udp_table = &udp_table,
2621 #ifdef CONFIG_COMPAT
2622 .compat_setsockopt = compat_udp_setsockopt,
2623 .compat_getsockopt = compat_udp_getsockopt,
2625 .diag_destroy = udp_abort,
2627 EXPORT_SYMBOL(udp_prot);
2629 /* ------------------------------------------------------------------------ */
2630 #ifdef CONFIG_PROC_FS
2632 static struct sock *udp_get_first(struct seq_file *seq, int start)
2635 struct udp_iter_state *state = seq->private;
2636 struct net *net = seq_file_net(seq);
2638 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2640 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2642 if (hlist_empty(&hslot->head))
2645 spin_lock_bh(&hslot->lock);
2646 sk_for_each(sk, &hslot->head) {
2647 if (!net_eq(sock_net(sk), net))
2649 if (sk->sk_family == state->family)
2652 spin_unlock_bh(&hslot->lock);
2659 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2661 struct udp_iter_state *state = seq->private;
2662 struct net *net = seq_file_net(seq);
2666 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2669 if (state->bucket <= state->udp_table->mask)
2670 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2671 return udp_get_first(seq, state->bucket + 1);
2676 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2678 struct sock *sk = udp_get_first(seq, 0);
2681 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2683 return pos ? NULL : sk;
2686 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2688 struct udp_iter_state *state = seq->private;
2689 state->bucket = MAX_UDP_PORTS;
2691 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2694 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2698 if (v == SEQ_START_TOKEN)
2699 sk = udp_get_idx(seq, 0);
2701 sk = udp_get_next(seq, v);
2707 static void udp_seq_stop(struct seq_file *seq, void *v)
2709 struct udp_iter_state *state = seq->private;
2711 if (state->bucket <= state->udp_table->mask)
2712 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2715 int udp_seq_open(struct inode *inode, struct file *file)
2717 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2718 struct udp_iter_state *s;
2721 err = seq_open_net(inode, file, &afinfo->seq_ops,
2722 sizeof(struct udp_iter_state));
2726 s = ((struct seq_file *)file->private_data)->private;
2727 s->family = afinfo->family;
2728 s->udp_table = afinfo->udp_table;
2731 EXPORT_SYMBOL(udp_seq_open);
2733 /* ------------------------------------------------------------------------ */
2734 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2736 struct proc_dir_entry *p;
2739 afinfo->seq_ops.start = udp_seq_start;
2740 afinfo->seq_ops.next = udp_seq_next;
2741 afinfo->seq_ops.stop = udp_seq_stop;
2743 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2744 afinfo->seq_fops, afinfo);
2749 EXPORT_SYMBOL(udp_proc_register);
2751 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2753 remove_proc_entry(afinfo->name, net->proc_net);
2755 EXPORT_SYMBOL(udp_proc_unregister);
2757 /* ------------------------------------------------------------------------ */
2758 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2761 struct inet_sock *inet = inet_sk(sp);
2762 __be32 dest = inet->inet_daddr;
2763 __be32 src = inet->inet_rcv_saddr;
2764 __u16 destp = ntohs(inet->inet_dport);
2765 __u16 srcp = ntohs(inet->inet_sport);
2767 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2768 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2769 bucket, src, srcp, dest, destp, sp->sk_state,
2770 sk_wmem_alloc_get(sp),
2773 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2775 refcount_read(&sp->sk_refcnt), sp,
2776 atomic_read(&sp->sk_drops));
2779 int udp4_seq_show(struct seq_file *seq, void *v)
2781 seq_setwidth(seq, 127);
2782 if (v == SEQ_START_TOKEN)
2783 seq_puts(seq, " sl local_address rem_address st tx_queue "
2784 "rx_queue tr tm->when retrnsmt uid timeout "
2785 "inode ref pointer drops");
2787 struct udp_iter_state *state = seq->private;
2789 udp4_format_sock(v, seq, state->bucket);
2795 static const struct file_operations udp_afinfo_seq_fops = {
2796 .owner = THIS_MODULE,
2797 .open = udp_seq_open,
2799 .llseek = seq_lseek,
2800 .release = seq_release_net
2803 /* ------------------------------------------------------------------------ */
2804 static struct udp_seq_afinfo udp4_seq_afinfo = {
2807 .udp_table = &udp_table,
2808 .seq_fops = &udp_afinfo_seq_fops,
2810 .show = udp4_seq_show,
2814 static int __net_init udp4_proc_init_net(struct net *net)
2816 return udp_proc_register(net, &udp4_seq_afinfo);
2819 static void __net_exit udp4_proc_exit_net(struct net *net)
2821 udp_proc_unregister(net, &udp4_seq_afinfo);
2824 static struct pernet_operations udp4_net_ops = {
2825 .init = udp4_proc_init_net,
2826 .exit = udp4_proc_exit_net,
2829 int __init udp4_proc_init(void)
2831 return register_pernet_subsys(&udp4_net_ops);
2834 void udp4_proc_exit(void)
2836 unregister_pernet_subsys(&udp4_net_ops);
2838 #endif /* CONFIG_PROC_FS */
2840 static __initdata unsigned long uhash_entries;
2841 static int __init set_uhash_entries(char *str)
2848 ret = kstrtoul(str, 0, &uhash_entries);
2852 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2853 uhash_entries = UDP_HTABLE_SIZE_MIN;
2856 __setup("uhash_entries=", set_uhash_entries);
2858 void __init udp_table_init(struct udp_table *table, const char *name)
2862 table->hash = alloc_large_system_hash(name,
2863 2 * sizeof(struct udp_hslot),
2865 21, /* one slot per 2 MB */
2869 UDP_HTABLE_SIZE_MIN,
2872 table->hash2 = table->hash + (table->mask + 1);
2873 for (i = 0; i <= table->mask; i++) {
2874 INIT_HLIST_HEAD(&table->hash[i].head);
2875 table->hash[i].count = 0;
2876 spin_lock_init(&table->hash[i].lock);
2878 for (i = 0; i <= table->mask; i++) {
2879 INIT_HLIST_HEAD(&table->hash2[i].head);
2880 table->hash2[i].count = 0;
2881 spin_lock_init(&table->hash2[i].lock);
2885 u32 udp_flow_hashrnd(void)
2887 static u32 hashrnd __read_mostly;
2889 net_get_random_once(&hashrnd, sizeof(hashrnd));
2893 EXPORT_SYMBOL(udp_flow_hashrnd);
2895 void __init udp_init(void)
2897 unsigned long limit;
2900 udp_table_init(&udp_table, "UDP");
2901 limit = nr_free_buffer_pages() / 8;
2902 limit = max(limit, 128UL);
2903 sysctl_udp_mem[0] = limit / 4 * 3;
2904 sysctl_udp_mem[1] = limit;
2905 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2907 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2908 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2910 /* 16 spinlocks per cpu */
2911 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2912 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2915 panic("UDP: failed to alloc udp_busylocks\n");
2916 for (i = 0; i < (1U << udp_busylocks_log); i++)
2917 spin_lock_init(udp_busylocks + i);