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 atomic_long_t udp_memory_allocated;
126 EXPORT_SYMBOL(udp_memory_allocated);
128 #define MAX_UDP_PORTS 65536
129 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
131 /* IPCB reference means this can not be used from early demux */
132 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
134 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
135 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
136 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
142 static int udp_lib_lport_inuse(struct net *net, __u16 num,
143 const struct udp_hslot *hslot,
144 unsigned long *bitmap,
145 struct sock *sk, unsigned int log)
148 kuid_t uid = sock_i_uid(sk);
150 sk_for_each(sk2, &hslot->head) {
151 if (net_eq(sock_net(sk2), net) &&
153 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
154 (!sk2->sk_reuse || !sk->sk_reuse) &&
155 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
156 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
157 inet_rcv_saddr_equal(sk, sk2, true)) {
158 if (sk2->sk_reuseport && sk->sk_reuseport &&
159 !rcu_access_pointer(sk->sk_reuseport_cb) &&
160 uid_eq(uid, sock_i_uid(sk2))) {
166 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
175 * Note: we still hold spinlock of primary hash chain, so no other writer
176 * can insert/delete a socket with local_port == num
178 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
179 struct udp_hslot *hslot2,
183 kuid_t uid = sock_i_uid(sk);
186 spin_lock(&hslot2->lock);
187 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
188 if (net_eq(sock_net(sk2), net) &&
190 (udp_sk(sk2)->udp_port_hash == num) &&
191 (!sk2->sk_reuse || !sk->sk_reuse) &&
192 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
193 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
194 inet_rcv_saddr_equal(sk, sk2, true)) {
195 if (sk2->sk_reuseport && sk->sk_reuseport &&
196 !rcu_access_pointer(sk->sk_reuseport_cb) &&
197 uid_eq(uid, sock_i_uid(sk2))) {
205 spin_unlock(&hslot2->lock);
209 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
211 struct net *net = sock_net(sk);
212 kuid_t uid = sock_i_uid(sk);
215 sk_for_each(sk2, &hslot->head) {
216 if (net_eq(sock_net(sk2), net) &&
218 sk2->sk_family == sk->sk_family &&
219 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
220 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
221 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
222 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
223 inet_rcv_saddr_equal(sk, sk2, false)) {
224 return reuseport_add_sock(sk, sk2,
225 inet_rcv_saddr_any(sk));
229 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
233 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
235 * @sk: socket struct in question
236 * @snum: port number to look up
237 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
240 int udp_lib_get_port(struct sock *sk, unsigned short snum,
241 unsigned int hash2_nulladdr)
243 struct udp_hslot *hslot, *hslot2;
244 struct udp_table *udptable = sk->sk_prot->h.udp_table;
246 struct net *net = sock_net(sk);
249 int low, high, remaining;
251 unsigned short first, last;
252 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
254 inet_get_local_port_range(net, &low, &high);
255 remaining = (high - low) + 1;
257 rand = prandom_u32();
258 first = reciprocal_scale(rand, remaining) + low;
260 * force rand to be an odd multiple of UDP_HTABLE_SIZE
262 rand = (rand | 1) * (udptable->mask + 1);
263 last = first + udptable->mask + 1;
265 hslot = udp_hashslot(udptable, net, first);
266 bitmap_zero(bitmap, PORTS_PER_CHAIN);
267 spin_lock_bh(&hslot->lock);
268 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
273 * Iterate on all possible values of snum for this hash.
274 * Using steps of an odd multiple of UDP_HTABLE_SIZE
275 * give us randomization and full range coverage.
278 if (low <= snum && snum <= high &&
279 !test_bit(snum >> udptable->log, bitmap) &&
280 !inet_is_local_reserved_port(net, snum))
283 } while (snum != first);
284 spin_unlock_bh(&hslot->lock);
286 } while (++first != last);
289 hslot = udp_hashslot(udptable, net, snum);
290 spin_lock_bh(&hslot->lock);
291 if (hslot->count > 10) {
293 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
295 slot2 &= udptable->mask;
296 hash2_nulladdr &= udptable->mask;
298 hslot2 = udp_hashslot2(udptable, slot2);
299 if (hslot->count < hslot2->count)
300 goto scan_primary_hash;
302 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
303 if (!exist && (hash2_nulladdr != slot2)) {
304 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
305 exist = udp_lib_lport_inuse2(net, snum, hslot2,
314 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
318 inet_sk(sk)->inet_num = snum;
319 udp_sk(sk)->udp_port_hash = snum;
320 udp_sk(sk)->udp_portaddr_hash ^= snum;
321 if (sk_unhashed(sk)) {
322 if (sk->sk_reuseport &&
323 udp_reuseport_add_sock(sk, hslot)) {
324 inet_sk(sk)->inet_num = 0;
325 udp_sk(sk)->udp_port_hash = 0;
326 udp_sk(sk)->udp_portaddr_hash ^= snum;
330 sk_add_node_rcu(sk, &hslot->head);
332 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
334 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
335 spin_lock(&hslot2->lock);
336 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
337 sk->sk_family == AF_INET6)
338 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
341 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
344 spin_unlock(&hslot2->lock);
346 sock_set_flag(sk, SOCK_RCU_FREE);
349 spin_unlock_bh(&hslot->lock);
353 EXPORT_SYMBOL(udp_lib_get_port);
355 int udp_v4_get_port(struct sock *sk, unsigned short snum)
357 unsigned int hash2_nulladdr =
358 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
359 unsigned int hash2_partial =
360 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
362 /* precompute partial secondary hash */
363 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
364 return udp_lib_get_port(sk, snum, hash2_nulladdr);
367 static int compute_score(struct sock *sk, struct net *net,
368 __be32 saddr, __be16 sport,
369 __be32 daddr, unsigned short hnum,
370 int dif, int sdif, bool exact_dif)
373 struct inet_sock *inet;
375 if (!net_eq(sock_net(sk), net) ||
376 udp_sk(sk)->udp_port_hash != hnum ||
380 score = (sk->sk_family == PF_INET) ? 2 : 1;
383 if (inet->inet_rcv_saddr) {
384 if (inet->inet_rcv_saddr != daddr)
389 if (inet->inet_daddr) {
390 if (inet->inet_daddr != saddr)
395 if (inet->inet_dport) {
396 if (inet->inet_dport != sport)
401 if (sk->sk_bound_dev_if || exact_dif) {
402 bool dev_match = (sk->sk_bound_dev_if == dif ||
403 sk->sk_bound_dev_if == sdif);
407 if (sk->sk_bound_dev_if)
411 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
416 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
417 const __u16 lport, const __be32 faddr,
420 static u32 udp_ehash_secret __read_mostly;
422 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
424 return __inet_ehashfn(laddr, lport, faddr, fport,
425 udp_ehash_secret + net_hash_mix(net));
428 /* called with rcu_read_lock() */
429 static struct sock *udp4_lib_lookup2(struct net *net,
430 __be32 saddr, __be16 sport,
431 __be32 daddr, unsigned int hnum,
432 int dif, int sdif, bool exact_dif,
433 struct udp_hslot *hslot2,
436 struct sock *sk, *result;
442 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
443 score = compute_score(sk, net, saddr, sport,
444 daddr, hnum, dif, sdif, exact_dif);
445 if (score > badness) {
446 if (sk->sk_reuseport &&
447 sk->sk_state != TCP_ESTABLISHED) {
448 hash = udp_ehashfn(net, daddr, hnum,
450 result = reuseport_select_sock(sk, hash, skb,
451 sizeof(struct udphdr));
452 if (result && !reuseport_has_conns(sk, false))
462 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
463 * harder than this. -DaveM
465 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
466 __be16 sport, __be32 daddr, __be16 dport, int dif,
467 int sdif, struct udp_table *udptable, struct sk_buff *skb)
469 struct sock *sk, *result;
470 unsigned short hnum = ntohs(dport);
471 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
472 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
473 bool exact_dif = udp_lib_exact_dif_match(net, skb);
477 if (hslot->count > 10) {
478 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
479 slot2 = hash2 & udptable->mask;
480 hslot2 = &udptable->hash2[slot2];
481 if (hslot->count < hslot2->count)
484 result = udp4_lib_lookup2(net, saddr, sport,
485 daddr, hnum, dif, sdif,
486 exact_dif, hslot2, skb);
488 unsigned int old_slot2 = slot2;
489 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
490 slot2 = hash2 & udptable->mask;
491 /* avoid searching the same slot again. */
492 if (unlikely(slot2 == old_slot2))
495 hslot2 = &udptable->hash2[slot2];
496 if (hslot->count < hslot2->count)
499 result = udp4_lib_lookup2(net, saddr, sport,
500 daddr, hnum, dif, sdif,
501 exact_dif, hslot2, skb);
503 if (unlikely(IS_ERR(result)))
510 sk_for_each_rcu(sk, &hslot->head) {
511 score = compute_score(sk, net, saddr, sport,
512 daddr, hnum, dif, sdif, exact_dif);
513 if (score > badness) {
514 if (sk->sk_reuseport) {
515 hash = udp_ehashfn(net, daddr, hnum,
517 result = reuseport_select_sock(sk, hash, skb,
518 sizeof(struct udphdr));
519 if (unlikely(IS_ERR(result)))
530 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
532 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
533 __be16 sport, __be16 dport,
534 struct udp_table *udptable)
536 const struct iphdr *iph = ip_hdr(skb);
538 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
539 iph->daddr, dport, inet_iif(skb),
540 inet_sdif(skb), udptable, skb);
543 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
544 __be16 sport, __be16 dport)
546 const struct iphdr *iph = ip_hdr(skb);
548 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
549 iph->daddr, dport, inet_iif(skb),
550 inet_sdif(skb), &udp_table, NULL);
552 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
554 /* Must be called under rcu_read_lock().
555 * Does increment socket refcount.
557 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
558 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
559 __be32 daddr, __be16 dport, int dif)
563 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
564 dif, 0, &udp_table, NULL);
565 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
569 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
572 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
573 __be16 loc_port, __be32 loc_addr,
574 __be16 rmt_port, __be32 rmt_addr,
575 int dif, int sdif, unsigned short hnum)
577 struct inet_sock *inet = inet_sk(sk);
579 if (!net_eq(sock_net(sk), net) ||
580 udp_sk(sk)->udp_port_hash != hnum ||
581 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
582 (inet->inet_dport != rmt_port && inet->inet_dport) ||
583 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
584 ipv6_only_sock(sk) ||
585 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
586 sk->sk_bound_dev_if != sdif))
588 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
594 * This routine is called by the ICMP module when it gets some
595 * sort of error condition. If err < 0 then the socket should
596 * be closed and the error returned to the user. If err > 0
597 * it's just the icmp type << 8 | icmp code.
598 * Header points to the ip header of the error packet. We move
599 * on past this. Then (as it used to claim before adjustment)
600 * header points to the first 8 bytes of the udp header. We need
601 * to find the appropriate port.
604 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
606 struct inet_sock *inet;
607 const struct iphdr *iph = (const struct iphdr *)skb->data;
608 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
609 const int type = icmp_hdr(skb)->type;
610 const int code = icmp_hdr(skb)->code;
614 struct net *net = dev_net(skb->dev);
616 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
617 iph->saddr, uh->source, skb->dev->ifindex, 0,
620 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
621 return; /* No socket for error */
630 case ICMP_TIME_EXCEEDED:
633 case ICMP_SOURCE_QUENCH:
635 case ICMP_PARAMETERPROB:
639 case ICMP_DEST_UNREACH:
640 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
641 ipv4_sk_update_pmtu(skb, sk, info);
642 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
650 if (code <= NR_ICMP_UNREACH) {
651 harderr = icmp_err_convert[code].fatal;
652 err = icmp_err_convert[code].errno;
656 ipv4_sk_redirect(skb, sk);
661 * RFC1122: OK. Passes ICMP errors back to application, as per
664 if (!inet->recverr) {
665 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
668 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
671 sk->sk_error_report(sk);
676 void udp_err(struct sk_buff *skb, u32 info)
678 __udp4_lib_err(skb, info, &udp_table);
682 * Throw away all pending data and cancel the corking. Socket is locked.
684 void udp_flush_pending_frames(struct sock *sk)
686 struct udp_sock *up = udp_sk(sk);
691 ip_flush_pending_frames(sk);
694 EXPORT_SYMBOL(udp_flush_pending_frames);
697 * udp4_hwcsum - handle outgoing HW checksumming
698 * @skb: sk_buff containing the filled-in UDP header
699 * (checksum field must be zeroed out)
700 * @src: source IP address
701 * @dst: destination IP address
703 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
705 struct udphdr *uh = udp_hdr(skb);
706 int offset = skb_transport_offset(skb);
707 int len = skb->len - offset;
711 if (!skb_has_frag_list(skb)) {
713 * Only one fragment on the socket.
715 skb->csum_start = skb_transport_header(skb) - skb->head;
716 skb->csum_offset = offsetof(struct udphdr, check);
717 uh->check = ~csum_tcpudp_magic(src, dst, len,
720 struct sk_buff *frags;
723 * HW-checksum won't work as there are two or more
724 * fragments on the socket so that all csums of sk_buffs
727 skb_walk_frags(skb, frags) {
728 csum = csum_add(csum, frags->csum);
732 csum = skb_checksum(skb, offset, hlen, csum);
733 skb->ip_summed = CHECKSUM_NONE;
735 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
737 uh->check = CSUM_MANGLED_0;
740 EXPORT_SYMBOL_GPL(udp4_hwcsum);
742 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
743 * for the simple case like when setting the checksum for a UDP tunnel.
745 void udp_set_csum(bool nocheck, struct sk_buff *skb,
746 __be32 saddr, __be32 daddr, int len)
748 struct udphdr *uh = udp_hdr(skb);
752 } else if (skb_is_gso(skb)) {
753 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
754 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
756 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
758 uh->check = CSUM_MANGLED_0;
760 skb->ip_summed = CHECKSUM_PARTIAL;
761 skb->csum_start = skb_transport_header(skb) - skb->head;
762 skb->csum_offset = offsetof(struct udphdr, check);
763 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
766 EXPORT_SYMBOL(udp_set_csum);
768 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
769 struct inet_cork *cork)
771 struct sock *sk = skb->sk;
772 struct inet_sock *inet = inet_sk(sk);
775 int is_udplite = IS_UDPLITE(sk);
776 int offset = skb_transport_offset(skb);
777 int len = skb->len - offset;
778 int datalen = len - sizeof(*uh);
782 * Create a UDP header
785 uh->source = inet->inet_sport;
786 uh->dest = fl4->fl4_dport;
787 uh->len = htons(len);
790 if (cork->gso_size) {
791 const int hlen = skb_network_header_len(skb) +
792 sizeof(struct udphdr);
794 if (hlen + cork->gso_size > cork->fragsize) {
798 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
802 if (sk->sk_no_check_tx) {
806 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
807 dst_xfrm(skb_dst(skb))) {
812 if (datalen > cork->gso_size) {
813 skb_shinfo(skb)->gso_size = cork->gso_size;
814 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
815 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
821 if (is_udplite) /* UDP-Lite */
822 csum = udplite_csum(skb);
824 else if (sk->sk_no_check_tx) { /* UDP csum off */
826 skb->ip_summed = CHECKSUM_NONE;
829 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
832 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
836 csum = udp_csum(skb);
838 /* add protocol-dependent pseudo-header */
839 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
840 sk->sk_protocol, csum);
842 uh->check = CSUM_MANGLED_0;
845 err = ip_send_skb(sock_net(sk), skb);
847 if (err == -ENOBUFS && !inet->recverr) {
848 UDP_INC_STATS(sock_net(sk),
849 UDP_MIB_SNDBUFERRORS, is_udplite);
853 UDP_INC_STATS(sock_net(sk),
854 UDP_MIB_OUTDATAGRAMS, is_udplite);
859 * Push out all pending data as one UDP datagram. Socket is locked.
861 int udp_push_pending_frames(struct sock *sk)
863 struct udp_sock *up = udp_sk(sk);
864 struct inet_sock *inet = inet_sk(sk);
865 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
869 skb = ip_finish_skb(sk, fl4);
873 err = udp_send_skb(skb, fl4, &inet->cork.base);
880 EXPORT_SYMBOL(udp_push_pending_frames);
882 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
884 switch (cmsg->cmsg_type) {
886 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
888 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
895 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
897 struct cmsghdr *cmsg;
898 bool need_ip = false;
901 for_each_cmsghdr(cmsg, msg) {
902 if (!CMSG_OK(msg, cmsg))
905 if (cmsg->cmsg_level != SOL_UDP) {
910 err = __udp_cmsg_send(cmsg, gso_size);
917 EXPORT_SYMBOL_GPL(udp_cmsg_send);
919 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
921 struct inet_sock *inet = inet_sk(sk);
922 struct udp_sock *up = udp_sk(sk);
923 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
924 struct flowi4 fl4_stack;
927 struct ipcm_cookie ipc;
928 struct rtable *rt = NULL;
931 __be32 daddr, faddr, saddr;
934 int err, is_udplite = IS_UDPLITE(sk);
935 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
936 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
938 struct ip_options_data opt_copy;
947 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
950 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
952 fl4 = &inet->cork.fl.u.ip4;
955 * There are pending frames.
956 * The socket lock must be held while it's corked.
959 if (likely(up->pending)) {
960 if (unlikely(up->pending != AF_INET)) {
968 ulen += sizeof(struct udphdr);
971 * Get and verify the address.
974 if (msg->msg_namelen < sizeof(*usin))
976 if (usin->sin_family != AF_INET) {
977 if (usin->sin_family != AF_UNSPEC)
978 return -EAFNOSUPPORT;
981 daddr = usin->sin_addr.s_addr;
982 dport = usin->sin_port;
986 if (sk->sk_state != TCP_ESTABLISHED)
987 return -EDESTADDRREQ;
988 daddr = inet->inet_daddr;
989 dport = inet->inet_dport;
990 /* Open fast path for connected socket.
991 Route will not be used, if at least one option is set.
996 ipcm_init_sk(&ipc, inet);
997 ipc.gso_size = up->gso_size;
999 if (msg->msg_controllen) {
1000 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1002 err = ip_cmsg_send(sk, msg, &ipc,
1003 sk->sk_family == AF_INET6);
1004 if (unlikely(err < 0)) {
1013 struct ip_options_rcu *inet_opt;
1016 inet_opt = rcu_dereference(inet->inet_opt);
1018 memcpy(&opt_copy, inet_opt,
1019 sizeof(*inet_opt) + inet_opt->opt.optlen);
1020 ipc.opt = &opt_copy.opt;
1025 if (cgroup_bpf_enabled && !connected) {
1026 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1027 (struct sockaddr *)usin, &ipc.addr);
1031 if (usin->sin_port == 0) {
1032 /* BPF program set invalid port. Reject it. */
1036 daddr = usin->sin_addr.s_addr;
1037 dport = usin->sin_port;
1042 ipc.addr = faddr = daddr;
1044 if (ipc.opt && ipc.opt->opt.srr) {
1049 faddr = ipc.opt->opt.faddr;
1052 tos = get_rttos(&ipc, inet);
1053 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1054 (msg->msg_flags & MSG_DONTROUTE) ||
1055 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1060 if (ipv4_is_multicast(daddr)) {
1062 ipc.oif = inet->mc_index;
1064 saddr = inet->mc_addr;
1066 } else if (!ipc.oif) {
1067 ipc.oif = inet->uc_index;
1068 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1069 /* oif is set, packet is to local broadcast and
1070 * and uc_index is set. oif is most likely set
1071 * by sk_bound_dev_if. If uc_index != oif check if the
1072 * oif is an L3 master and uc_index is an L3 slave.
1073 * If so, we want to allow the send using the uc_index.
1075 if (ipc.oif != inet->uc_index &&
1076 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1078 ipc.oif = inet->uc_index;
1083 rt = (struct rtable *)sk_dst_check(sk, 0);
1086 struct net *net = sock_net(sk);
1087 __u8 flow_flags = inet_sk_flowi_flags(sk);
1091 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1092 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1094 faddr, saddr, dport, inet->inet_sport,
1097 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1098 rt = ip_route_output_flow(net, fl4, sk);
1102 if (err == -ENETUNREACH)
1103 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1108 if ((rt->rt_flags & RTCF_BROADCAST) &&
1109 !sock_flag(sk, SOCK_BROADCAST))
1112 sk_dst_set(sk, dst_clone(&rt->dst));
1115 if (msg->msg_flags&MSG_CONFIRM)
1121 daddr = ipc.addr = fl4->daddr;
1123 /* Lockless fast path for the non-corking case. */
1125 struct inet_cork cork;
1127 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1128 sizeof(struct udphdr), &ipc, &rt,
1129 &cork, msg->msg_flags);
1131 if (!IS_ERR_OR_NULL(skb))
1132 err = udp_send_skb(skb, fl4, &cork);
1137 if (unlikely(up->pending)) {
1138 /* The socket is already corked while preparing it. */
1139 /* ... which is an evident application bug. --ANK */
1142 net_dbg_ratelimited("socket already corked\n");
1147 * Now cork the socket to pend data.
1149 fl4 = &inet->cork.fl.u.ip4;
1152 fl4->fl4_dport = dport;
1153 fl4->fl4_sport = inet->inet_sport;
1154 up->pending = AF_INET;
1158 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1159 sizeof(struct udphdr), &ipc, &rt,
1160 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1162 udp_flush_pending_frames(sk);
1164 err = udp_push_pending_frames(sk);
1165 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1177 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1178 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1179 * we don't have a good statistic (IpOutDiscards but it can be too many
1180 * things). We could add another new stat but at least for now that
1181 * seems like overkill.
1183 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1184 UDP_INC_STATS(sock_net(sk),
1185 UDP_MIB_SNDBUFERRORS, is_udplite);
1190 if (msg->msg_flags & MSG_PROBE)
1191 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1192 if (!(msg->msg_flags&MSG_PROBE) || len)
1193 goto back_from_confirm;
1197 EXPORT_SYMBOL(udp_sendmsg);
1199 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1200 size_t size, int flags)
1202 struct inet_sock *inet = inet_sk(sk);
1203 struct udp_sock *up = udp_sk(sk);
1206 if (flags & MSG_SENDPAGE_NOTLAST)
1210 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1212 /* Call udp_sendmsg to specify destination address which
1213 * sendpage interface can't pass.
1214 * This will succeed only when the socket is connected.
1216 ret = udp_sendmsg(sk, &msg, 0);
1223 if (unlikely(!up->pending)) {
1226 net_dbg_ratelimited("cork failed\n");
1230 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1231 page, offset, size, flags);
1232 if (ret == -EOPNOTSUPP) {
1234 return sock_no_sendpage(sk->sk_socket, page, offset,
1238 udp_flush_pending_frames(sk);
1243 if (!(up->corkflag || (flags&MSG_MORE)))
1244 ret = udp_push_pending_frames(sk);
1252 #define UDP_SKB_IS_STATELESS 0x80000000
1254 static void udp_set_dev_scratch(struct sk_buff *skb)
1256 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1258 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1259 scratch->_tsize_state = skb->truesize;
1260 #if BITS_PER_LONG == 64
1261 scratch->len = skb->len;
1262 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1263 scratch->is_linear = !skb_is_nonlinear(skb);
1265 /* all head states execept sp (dst, sk, nf) are always cleared by
1266 * udp_rcv() and we need to preserve secpath, if present, to eventually
1267 * process IP_CMSG_PASSSEC at recvmsg() time
1269 if (likely(!skb_sec_path(skb)))
1270 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1273 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1275 /* We come here after udp_lib_checksum_complete() returned 0.
1276 * This means that __skb_checksum_complete() might have
1277 * set skb->csum_valid to 1.
1278 * On 64bit platforms, we can set csum_unnecessary
1279 * to true, but only if the skb is not shared.
1281 #if BITS_PER_LONG == 64
1282 if (!skb_shared(skb))
1283 udp_skb_scratch(skb)->csum_unnecessary = true;
1287 static int udp_skb_truesize(struct sk_buff *skb)
1289 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1292 static bool udp_skb_has_head_state(struct sk_buff *skb)
1294 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1297 /* fully reclaim rmem/fwd memory allocated for skb */
1298 static void udp_rmem_release(struct sock *sk, int size, int partial,
1299 bool rx_queue_lock_held)
1301 struct udp_sock *up = udp_sk(sk);
1302 struct sk_buff_head *sk_queue;
1305 if (likely(partial)) {
1306 up->forward_deficit += size;
1307 size = up->forward_deficit;
1308 if (size < (sk->sk_rcvbuf >> 2))
1311 size += up->forward_deficit;
1313 up->forward_deficit = 0;
1315 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1316 * if the called don't held it already
1318 sk_queue = &sk->sk_receive_queue;
1319 if (!rx_queue_lock_held)
1320 spin_lock(&sk_queue->lock);
1323 sk->sk_forward_alloc += size;
1324 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1325 sk->sk_forward_alloc -= amt;
1328 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1330 atomic_sub(size, &sk->sk_rmem_alloc);
1332 /* this can save us from acquiring the rx queue lock on next receive */
1333 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1335 if (!rx_queue_lock_held)
1336 spin_unlock(&sk_queue->lock);
1339 /* Note: called with reader_queue.lock held.
1340 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1341 * This avoids a cache line miss while receive_queue lock is held.
1342 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1344 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1346 prefetch(&skb->data);
1347 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1349 EXPORT_SYMBOL(udp_skb_destructor);
1351 /* as above, but the caller held the rx queue lock, too */
1352 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1354 prefetch(&skb->data);
1355 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1358 /* Idea of busylocks is to let producers grab an extra spinlock
1359 * to relieve pressure on the receive_queue spinlock shared by consumer.
1360 * Under flood, this means that only one producer can be in line
1361 * trying to acquire the receive_queue spinlock.
1362 * These busylock can be allocated on a per cpu manner, instead of a
1363 * per socket one (that would consume a cache line per socket)
1365 static int udp_busylocks_log __read_mostly;
1366 static spinlock_t *udp_busylocks __read_mostly;
1368 static spinlock_t *busylock_acquire(void *ptr)
1372 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1377 static void busylock_release(spinlock_t *busy)
1383 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1385 struct sk_buff_head *list = &sk->sk_receive_queue;
1386 int rmem, delta, amt, err = -ENOMEM;
1387 spinlock_t *busy = NULL;
1390 /* try to avoid the costly atomic add/sub pair when the receive
1391 * queue is full; always allow at least a packet
1393 rmem = atomic_read(&sk->sk_rmem_alloc);
1394 if (rmem > sk->sk_rcvbuf)
1397 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1398 * having linear skbs :
1399 * - Reduce memory overhead and thus increase receive queue capacity
1400 * - Less cache line misses at copyout() time
1401 * - Less work at consume_skb() (less alien page frag freeing)
1403 if (rmem > (sk->sk_rcvbuf >> 1)) {
1406 busy = busylock_acquire(sk);
1408 size = skb->truesize;
1409 udp_set_dev_scratch(skb);
1411 /* we drop only if the receive buf is full and the receive
1412 * queue contains some other skb
1414 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1415 if (rmem > (size + sk->sk_rcvbuf))
1418 spin_lock(&list->lock);
1419 if (size >= sk->sk_forward_alloc) {
1420 amt = sk_mem_pages(size);
1421 delta = amt << SK_MEM_QUANTUM_SHIFT;
1422 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1424 spin_unlock(&list->lock);
1428 sk->sk_forward_alloc += delta;
1431 sk->sk_forward_alloc -= size;
1433 /* no need to setup a destructor, we will explicitly release the
1434 * forward allocated memory on dequeue
1436 sock_skb_set_dropcount(sk, skb);
1438 __skb_queue_tail(list, skb);
1439 spin_unlock(&list->lock);
1441 if (!sock_flag(sk, SOCK_DEAD))
1442 sk->sk_data_ready(sk);
1444 busylock_release(busy);
1448 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1451 atomic_inc(&sk->sk_drops);
1452 busylock_release(busy);
1455 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1457 void udp_destruct_sock(struct sock *sk)
1459 /* reclaim completely the forward allocated memory */
1460 struct udp_sock *up = udp_sk(sk);
1461 unsigned int total = 0;
1462 struct sk_buff *skb;
1464 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1465 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1466 total += skb->truesize;
1469 udp_rmem_release(sk, total, 0, true);
1471 inet_sock_destruct(sk);
1473 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1475 int udp_init_sock(struct sock *sk)
1477 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1478 sk->sk_destruct = udp_destruct_sock;
1481 EXPORT_SYMBOL_GPL(udp_init_sock);
1483 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1485 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1486 bool slow = lock_sock_fast(sk);
1488 sk_peek_offset_bwd(sk, len);
1489 unlock_sock_fast(sk, slow);
1492 if (!skb_unref(skb))
1495 /* In the more common cases we cleared the head states previously,
1496 * see __udp_queue_rcv_skb().
1498 if (unlikely(udp_skb_has_head_state(skb)))
1499 skb_release_head_state(skb);
1500 __consume_stateless_skb(skb);
1502 EXPORT_SYMBOL_GPL(skb_consume_udp);
1504 static struct sk_buff *__first_packet_length(struct sock *sk,
1505 struct sk_buff_head *rcvq,
1508 struct sk_buff *skb;
1510 while ((skb = skb_peek(rcvq)) != NULL) {
1511 if (udp_lib_checksum_complete(skb)) {
1512 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1514 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1516 atomic_inc(&sk->sk_drops);
1517 __skb_unlink(skb, rcvq);
1518 *total += skb->truesize;
1521 udp_skb_csum_unnecessary_set(skb);
1529 * first_packet_length - return length of first packet in receive queue
1532 * Drops all bad checksum frames, until a valid one is found.
1533 * Returns the length of found skb, or -1 if none is found.
1535 static int first_packet_length(struct sock *sk)
1537 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1538 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1539 struct sk_buff *skb;
1543 spin_lock_bh(&rcvq->lock);
1544 skb = __first_packet_length(sk, rcvq, &total);
1545 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1546 spin_lock(&sk_queue->lock);
1547 skb_queue_splice_tail_init(sk_queue, rcvq);
1548 spin_unlock(&sk_queue->lock);
1550 skb = __first_packet_length(sk, rcvq, &total);
1552 res = skb ? skb->len : -1;
1554 udp_rmem_release(sk, total, 1, false);
1555 spin_unlock_bh(&rcvq->lock);
1560 * IOCTL requests applicable to the UDP protocol
1563 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1568 int amount = sk_wmem_alloc_get(sk);
1570 return put_user(amount, (int __user *)arg);
1575 int amount = max_t(int, 0, first_packet_length(sk));
1577 return put_user(amount, (int __user *)arg);
1581 return -ENOIOCTLCMD;
1586 EXPORT_SYMBOL(udp_ioctl);
1588 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1589 int noblock, int *peeked, int *off, int *err)
1591 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1592 struct sk_buff_head *queue;
1593 struct sk_buff *last;
1597 queue = &udp_sk(sk)->reader_queue;
1598 flags |= noblock ? MSG_DONTWAIT : 0;
1599 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1601 struct sk_buff *skb;
1603 error = sock_error(sk);
1610 spin_lock_bh(&queue->lock);
1611 skb = __skb_try_recv_from_queue(sk, queue, flags,
1616 spin_unlock_bh(&queue->lock);
1620 if (skb_queue_empty_lockless(sk_queue)) {
1621 spin_unlock_bh(&queue->lock);
1625 /* refill the reader queue and walk it again
1626 * keep both queues locked to avoid re-acquiring
1627 * the sk_receive_queue lock if fwd memory scheduling
1630 spin_lock(&sk_queue->lock);
1631 skb_queue_splice_tail_init(sk_queue, queue);
1633 skb = __skb_try_recv_from_queue(sk, queue, flags,
1634 udp_skb_dtor_locked,
1637 spin_unlock(&sk_queue->lock);
1638 spin_unlock_bh(&queue->lock);
1643 if (!sk_can_busy_loop(sk))
1646 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1647 } while (!skb_queue_empty_lockless(sk_queue));
1649 /* sk_queue is empty, reader_queue may contain peeked packets */
1651 !__skb_wait_for_more_packets(sk, &error, &timeo,
1652 (struct sk_buff *)sk_queue));
1657 EXPORT_SYMBOL(__skb_recv_udp);
1660 * This should be easy, if there is something there we
1661 * return it, otherwise we block.
1664 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1665 int flags, int *addr_len)
1667 struct inet_sock *inet = inet_sk(sk);
1668 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1669 struct sk_buff *skb;
1670 unsigned int ulen, copied;
1671 int peeked, peeking, off;
1673 int is_udplite = IS_UDPLITE(sk);
1674 bool checksum_valid = false;
1676 if (flags & MSG_ERRQUEUE)
1677 return ip_recv_error(sk, msg, len, addr_len);
1680 peeking = flags & MSG_PEEK;
1681 off = sk_peek_offset(sk, flags);
1682 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1686 ulen = udp_skb_len(skb);
1688 if (copied > ulen - off)
1689 copied = ulen - off;
1690 else if (copied < ulen)
1691 msg->msg_flags |= MSG_TRUNC;
1694 * If checksum is needed at all, try to do it while copying the
1695 * data. If the data is truncated, or if we only want a partial
1696 * coverage checksum (UDP-Lite), do it before the copy.
1699 if (copied < ulen || peeking ||
1700 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1701 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1702 !__udp_lib_checksum_complete(skb);
1703 if (!checksum_valid)
1707 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1708 if (udp_skb_is_linear(skb))
1709 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1711 err = skb_copy_datagram_msg(skb, off, msg, copied);
1713 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1719 if (unlikely(err)) {
1721 atomic_inc(&sk->sk_drops);
1722 UDP_INC_STATS(sock_net(sk),
1723 UDP_MIB_INERRORS, is_udplite);
1730 UDP_INC_STATS(sock_net(sk),
1731 UDP_MIB_INDATAGRAMS, is_udplite);
1733 sock_recv_ts_and_drops(msg, sk, skb);
1735 /* Copy the address. */
1737 sin->sin_family = AF_INET;
1738 sin->sin_port = udp_hdr(skb)->source;
1739 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1740 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1741 *addr_len = sizeof(*sin);
1743 if (cgroup_bpf_enabled)
1744 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1745 (struct sockaddr *)sin);
1747 if (inet->cmsg_flags)
1748 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1751 if (flags & MSG_TRUNC)
1754 skb_consume_udp(sk, skb, peeking ? -err : err);
1758 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1759 udp_skb_destructor)) {
1760 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1761 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1765 /* starting over for a new packet, but check if we need to yield */
1767 msg->msg_flags &= ~MSG_TRUNC;
1771 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1773 /* This check is replicated from __ip4_datagram_connect() and
1774 * intended to prevent BPF program called below from accessing bytes
1775 * that are out of the bound specified by user in addr_len.
1777 if (addr_len < sizeof(struct sockaddr_in))
1780 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1782 EXPORT_SYMBOL(udp_pre_connect);
1784 int __udp_disconnect(struct sock *sk, int flags)
1786 struct inet_sock *inet = inet_sk(sk);
1788 * 1003.1g - break association.
1791 sk->sk_state = TCP_CLOSE;
1792 inet->inet_daddr = 0;
1793 inet->inet_dport = 0;
1794 sock_rps_reset_rxhash(sk);
1795 sk->sk_bound_dev_if = 0;
1796 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1797 inet_reset_saddr(sk);
1799 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1800 sk->sk_prot->unhash(sk);
1801 inet->inet_sport = 0;
1806 EXPORT_SYMBOL(__udp_disconnect);
1808 int udp_disconnect(struct sock *sk, int flags)
1811 __udp_disconnect(sk, flags);
1815 EXPORT_SYMBOL(udp_disconnect);
1817 void udp_lib_unhash(struct sock *sk)
1819 if (sk_hashed(sk)) {
1820 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1821 struct udp_hslot *hslot, *hslot2;
1823 hslot = udp_hashslot(udptable, sock_net(sk),
1824 udp_sk(sk)->udp_port_hash);
1825 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1827 spin_lock_bh(&hslot->lock);
1828 if (rcu_access_pointer(sk->sk_reuseport_cb))
1829 reuseport_detach_sock(sk);
1830 if (sk_del_node_init_rcu(sk)) {
1832 inet_sk(sk)->inet_num = 0;
1833 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1835 spin_lock(&hslot2->lock);
1836 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1838 spin_unlock(&hslot2->lock);
1840 spin_unlock_bh(&hslot->lock);
1843 EXPORT_SYMBOL(udp_lib_unhash);
1846 * inet_rcv_saddr was changed, we must rehash secondary hash
1848 void udp_lib_rehash(struct sock *sk, u16 newhash)
1850 if (sk_hashed(sk)) {
1851 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1852 struct udp_hslot *hslot, *hslot2, *nhslot2;
1854 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1855 nhslot2 = udp_hashslot2(udptable, newhash);
1856 udp_sk(sk)->udp_portaddr_hash = newhash;
1858 if (hslot2 != nhslot2 ||
1859 rcu_access_pointer(sk->sk_reuseport_cb)) {
1860 hslot = udp_hashslot(udptable, sock_net(sk),
1861 udp_sk(sk)->udp_port_hash);
1862 /* we must lock primary chain too */
1863 spin_lock_bh(&hslot->lock);
1864 if (rcu_access_pointer(sk->sk_reuseport_cb))
1865 reuseport_detach_sock(sk);
1867 if (hslot2 != nhslot2) {
1868 spin_lock(&hslot2->lock);
1869 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1871 spin_unlock(&hslot2->lock);
1873 spin_lock(&nhslot2->lock);
1874 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1877 spin_unlock(&nhslot2->lock);
1880 spin_unlock_bh(&hslot->lock);
1884 EXPORT_SYMBOL(udp_lib_rehash);
1886 static void udp_v4_rehash(struct sock *sk)
1888 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1889 inet_sk(sk)->inet_rcv_saddr,
1890 inet_sk(sk)->inet_num);
1891 udp_lib_rehash(sk, new_hash);
1894 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1898 if (inet_sk(sk)->inet_daddr) {
1899 sock_rps_save_rxhash(sk, skb);
1900 sk_mark_napi_id(sk, skb);
1901 sk_incoming_cpu_update(sk);
1903 sk_mark_napi_id_once(sk, skb);
1906 rc = __udp_enqueue_schedule_skb(sk, skb);
1908 int is_udplite = IS_UDPLITE(sk);
1910 /* Note that an ENOMEM error is charged twice */
1912 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1914 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1916 trace_udp_fail_queue_rcv_skb(rc, sk);
1923 static DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
1924 void udp_encap_enable(void)
1926 static_branch_enable(&udp_encap_needed_key);
1928 EXPORT_SYMBOL(udp_encap_enable);
1933 * >0: "udp encap" protocol resubmission
1935 * Note that in the success and error cases, the skb is assumed to
1936 * have either been requeued or freed.
1938 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1940 struct udp_sock *up = udp_sk(sk);
1941 int is_udplite = IS_UDPLITE(sk);
1944 * Charge it to the socket, dropping if the queue is full.
1946 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1950 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
1951 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1954 * This is an encapsulation socket so pass the skb to
1955 * the socket's udp_encap_rcv() hook. Otherwise, just
1956 * fall through and pass this up the UDP socket.
1957 * up->encap_rcv() returns the following value:
1958 * =0 if skb was successfully passed to the encap
1959 * handler or was discarded by it.
1960 * >0 if skb should be passed on to UDP.
1961 * <0 if skb should be resubmitted as proto -N
1964 /* if we're overly short, let UDP handle it */
1965 encap_rcv = READ_ONCE(up->encap_rcv);
1969 /* Verify checksum before giving to encap */
1970 if (udp_lib_checksum_complete(skb))
1973 ret = encap_rcv(sk, skb);
1975 __UDP_INC_STATS(sock_net(sk),
1976 UDP_MIB_INDATAGRAMS,
1982 /* FALLTHROUGH -- it's a UDP Packet */
1986 * UDP-Lite specific tests, ignored on UDP sockets
1988 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1991 * MIB statistics other than incrementing the error count are
1992 * disabled for the following two types of errors: these depend
1993 * on the application settings, not on the functioning of the
1994 * protocol stack as such.
1996 * RFC 3828 here recommends (sec 3.3): "There should also be a
1997 * way ... to ... at least let the receiving application block
1998 * delivery of packets with coverage values less than a value
1999 * provided by the application."
2001 if (up->pcrlen == 0) { /* full coverage was set */
2002 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2003 UDP_SKB_CB(skb)->cscov, skb->len);
2006 /* The next case involves violating the min. coverage requested
2007 * by the receiver. This is subtle: if receiver wants x and x is
2008 * greater than the buffersize/MTU then receiver will complain
2009 * that it wants x while sender emits packets of smaller size y.
2010 * Therefore the above ...()->partial_cov statement is essential.
2012 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2013 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2014 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2019 prefetch(&sk->sk_rmem_alloc);
2020 if (rcu_access_pointer(sk->sk_filter) &&
2021 udp_lib_checksum_complete(skb))
2024 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2027 udp_csum_pull_header(skb);
2029 ipv4_pktinfo_prepare(sk, skb);
2030 return __udp_queue_rcv_skb(sk, skb);
2033 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2035 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2036 atomic_inc(&sk->sk_drops);
2041 /* For TCP sockets, sk_rx_dst is protected by socket lock
2042 * For UDP, we use xchg() to guard against concurrent changes.
2044 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2046 struct dst_entry *old;
2048 if (dst_hold_safe(dst)) {
2049 old = xchg(&sk->sk_rx_dst, dst);
2055 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2058 * Multicasts and broadcasts go to each listener.
2060 * Note: called only from the BH handler context.
2062 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2064 __be32 saddr, __be32 daddr,
2065 struct udp_table *udptable,
2068 struct sock *sk, *first = NULL;
2069 unsigned short hnum = ntohs(uh->dest);
2070 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2071 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2072 unsigned int offset = offsetof(typeof(*sk), sk_node);
2073 int dif = skb->dev->ifindex;
2074 int sdif = inet_sdif(skb);
2075 struct hlist_node *node;
2076 struct sk_buff *nskb;
2079 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2081 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2083 hslot = &udptable->hash2[hash2];
2084 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2087 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2088 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2089 uh->source, saddr, dif, sdif, hnum))
2096 nskb = skb_clone(skb, GFP_ATOMIC);
2098 if (unlikely(!nskb)) {
2099 atomic_inc(&sk->sk_drops);
2100 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2102 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2106 if (udp_queue_rcv_skb(sk, nskb) > 0)
2110 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2111 if (use_hash2 && hash2 != hash2_any) {
2117 if (udp_queue_rcv_skb(first, skb) > 0)
2121 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2122 proto == IPPROTO_UDPLITE);
2127 /* Initialize UDP checksum. If exited with zero value (success),
2128 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2129 * Otherwise, csum completion requires chacksumming packet body,
2130 * including udp header and folding it to skb->csum.
2132 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2137 UDP_SKB_CB(skb)->partial_cov = 0;
2138 UDP_SKB_CB(skb)->cscov = skb->len;
2140 if (proto == IPPROTO_UDPLITE) {
2141 err = udplite_checksum_init(skb, uh);
2145 if (UDP_SKB_CB(skb)->partial_cov) {
2146 skb->csum = inet_compute_pseudo(skb, proto);
2151 /* Note, we are only interested in != 0 or == 0, thus the
2154 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2155 inet_compute_pseudo);
2159 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2160 /* If SW calculated the value, we know it's bad */
2161 if (skb->csum_complete_sw)
2164 /* HW says the value is bad. Let's validate that.
2165 * skb->csum is no longer the full packet checksum,
2166 * so don't treat it as such.
2168 skb_checksum_complete_unset(skb);
2174 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2175 * return code conversion for ip layer consumption
2177 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2182 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2183 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2184 inet_compute_pseudo);
2186 ret = udp_queue_rcv_skb(sk, skb);
2188 /* a return value > 0 means to resubmit the input, but
2189 * it wants the return to be -protocol, or 0
2197 * All we need to do is get the socket, and then do a checksum.
2200 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2205 unsigned short ulen;
2206 struct rtable *rt = skb_rtable(skb);
2207 __be32 saddr, daddr;
2208 struct net *net = dev_net(skb->dev);
2211 * Validate the packet.
2213 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2214 goto drop; /* No space for header. */
2217 ulen = ntohs(uh->len);
2218 saddr = ip_hdr(skb)->saddr;
2219 daddr = ip_hdr(skb)->daddr;
2221 if (ulen > skb->len)
2224 if (proto == IPPROTO_UDP) {
2225 /* UDP validates ulen. */
2226 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2231 if (udp4_csum_init(skb, uh, proto))
2234 sk = skb_steal_sock(skb);
2236 struct dst_entry *dst = skb_dst(skb);
2239 if (unlikely(sk->sk_rx_dst != dst))
2240 udp_sk_rx_dst_set(sk, dst);
2242 ret = udp_unicast_rcv_skb(sk, skb, uh);
2247 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2248 return __udp4_lib_mcast_deliver(net, skb, uh,
2249 saddr, daddr, udptable, proto);
2251 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2253 return udp_unicast_rcv_skb(sk, skb, uh);
2255 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2259 /* No socket. Drop packet silently, if checksum is wrong */
2260 if (udp_lib_checksum_complete(skb))
2263 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2264 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2267 * Hmm. We got an UDP packet to a port to which we
2268 * don't wanna listen. Ignore it.
2274 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2275 proto == IPPROTO_UDPLITE ? "Lite" : "",
2276 &saddr, ntohs(uh->source),
2278 &daddr, ntohs(uh->dest));
2283 * RFC1122: OK. Discards the bad packet silently (as far as
2284 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2286 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2287 proto == IPPROTO_UDPLITE ? "Lite" : "",
2288 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2290 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2292 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2297 /* We can only early demux multicast if there is a single matching socket.
2298 * If more than one socket found returns NULL
2300 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2301 __be16 loc_port, __be32 loc_addr,
2302 __be16 rmt_port, __be32 rmt_addr,
2305 struct sock *sk, *result;
2306 unsigned short hnum = ntohs(loc_port);
2307 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2308 struct udp_hslot *hslot = &udp_table.hash[slot];
2310 /* Do not bother scanning a too big list */
2311 if (hslot->count > 10)
2315 sk_for_each_rcu(sk, &hslot->head) {
2316 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2317 rmt_port, rmt_addr, dif, sdif, hnum)) {
2327 /* For unicast we should only early demux connected sockets or we can
2328 * break forwarding setups. The chains here can be long so only check
2329 * if the first socket is an exact match and if not move on.
2331 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2332 __be16 loc_port, __be32 loc_addr,
2333 __be16 rmt_port, __be32 rmt_addr,
2336 unsigned short hnum = ntohs(loc_port);
2337 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2338 unsigned int slot2 = hash2 & udp_table.mask;
2339 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2340 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2341 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2344 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2345 if (INET_MATCH(sk, net, acookie, rmt_addr,
2346 loc_addr, ports, dif, sdif))
2348 /* Only check first socket in chain */
2354 int udp_v4_early_demux(struct sk_buff *skb)
2356 struct net *net = dev_net(skb->dev);
2357 struct in_device *in_dev = NULL;
2358 const struct iphdr *iph;
2359 const struct udphdr *uh;
2360 struct sock *sk = NULL;
2361 struct dst_entry *dst;
2362 int dif = skb->dev->ifindex;
2363 int sdif = inet_sdif(skb);
2366 /* validate the packet */
2367 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2373 if (skb->pkt_type == PACKET_MULTICAST) {
2374 in_dev = __in_dev_get_rcu(skb->dev);
2379 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2384 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2385 uh->source, iph->saddr,
2387 } else if (skb->pkt_type == PACKET_HOST) {
2388 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2389 uh->source, iph->saddr, dif, sdif);
2392 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2396 skb->destructor = sock_efree;
2397 dst = READ_ONCE(sk->sk_rx_dst);
2400 dst = dst_check(dst, 0);
2404 /* set noref for now.
2405 * any place which wants to hold dst has to call
2408 skb_dst_set_noref(skb, dst);
2410 /* for unconnected multicast sockets we need to validate
2411 * the source on each packet
2413 if (!inet_sk(sk)->inet_daddr && in_dev)
2414 return ip_mc_validate_source(skb, iph->daddr,
2415 iph->saddr, iph->tos,
2416 skb->dev, in_dev, &itag);
2421 int udp_rcv(struct sk_buff *skb)
2423 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2426 void udp_destroy_sock(struct sock *sk)
2428 struct udp_sock *up = udp_sk(sk);
2429 bool slow = lock_sock_fast(sk);
2430 udp_flush_pending_frames(sk);
2431 unlock_sock_fast(sk, slow);
2432 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2433 void (*encap_destroy)(struct sock *sk);
2434 encap_destroy = READ_ONCE(up->encap_destroy);
2441 * Socket option code for UDP
2443 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2444 char __user *optval, unsigned int optlen,
2445 int (*push_pending_frames)(struct sock *))
2447 struct udp_sock *up = udp_sk(sk);
2450 int is_udplite = IS_UDPLITE(sk);
2452 if (optlen < sizeof(int))
2455 if (get_user(val, (int __user *)optval))
2458 valbool = val ? 1 : 0;
2467 push_pending_frames(sk);
2475 case UDP_ENCAP_ESPINUDP:
2476 case UDP_ENCAP_ESPINUDP_NON_IKE:
2477 up->encap_rcv = xfrm4_udp_encap_rcv;
2479 case UDP_ENCAP_L2TPINUDP:
2480 up->encap_type = val;
2489 case UDP_NO_CHECK6_TX:
2490 up->no_check6_tx = valbool;
2493 case UDP_NO_CHECK6_RX:
2494 up->no_check6_rx = valbool;
2498 if (val < 0 || val > USHRT_MAX)
2504 * UDP-Lite's partial checksum coverage (RFC 3828).
2506 /* The sender sets actual checksum coverage length via this option.
2507 * The case coverage > packet length is handled by send module. */
2508 case UDPLITE_SEND_CSCOV:
2509 if (!is_udplite) /* Disable the option on UDP sockets */
2510 return -ENOPROTOOPT;
2511 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2513 else if (val > USHRT_MAX)
2516 up->pcflag |= UDPLITE_SEND_CC;
2519 /* The receiver specifies a minimum checksum coverage value. To make
2520 * sense, this should be set to at least 8 (as done below). If zero is
2521 * used, this again means full checksum coverage. */
2522 case UDPLITE_RECV_CSCOV:
2523 if (!is_udplite) /* Disable the option on UDP sockets */
2524 return -ENOPROTOOPT;
2525 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2527 else if (val > USHRT_MAX)
2530 up->pcflag |= UDPLITE_RECV_CC;
2540 EXPORT_SYMBOL(udp_lib_setsockopt);
2542 int udp_setsockopt(struct sock *sk, int level, int optname,
2543 char __user *optval, unsigned int optlen)
2545 if (level == SOL_UDP || level == SOL_UDPLITE)
2546 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2547 udp_push_pending_frames);
2548 return ip_setsockopt(sk, level, optname, optval, optlen);
2551 #ifdef CONFIG_COMPAT
2552 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2553 char __user *optval, unsigned int optlen)
2555 if (level == SOL_UDP || level == SOL_UDPLITE)
2556 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2557 udp_push_pending_frames);
2558 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2562 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2563 char __user *optval, int __user *optlen)
2565 struct udp_sock *up = udp_sk(sk);
2568 if (get_user(len, optlen))
2571 len = min_t(unsigned int, len, sizeof(int));
2582 val = up->encap_type;
2585 case UDP_NO_CHECK6_TX:
2586 val = up->no_check6_tx;
2589 case UDP_NO_CHECK6_RX:
2590 val = up->no_check6_rx;
2597 /* The following two cannot be changed on UDP sockets, the return is
2598 * always 0 (which corresponds to the full checksum coverage of UDP). */
2599 case UDPLITE_SEND_CSCOV:
2603 case UDPLITE_RECV_CSCOV:
2608 return -ENOPROTOOPT;
2611 if (put_user(len, optlen))
2613 if (copy_to_user(optval, &val, len))
2617 EXPORT_SYMBOL(udp_lib_getsockopt);
2619 int udp_getsockopt(struct sock *sk, int level, int optname,
2620 char __user *optval, int __user *optlen)
2622 if (level == SOL_UDP || level == SOL_UDPLITE)
2623 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2624 return ip_getsockopt(sk, level, optname, optval, optlen);
2627 #ifdef CONFIG_COMPAT
2628 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2629 char __user *optval, int __user *optlen)
2631 if (level == SOL_UDP || level == SOL_UDPLITE)
2632 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2633 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2637 * udp_poll - wait for a UDP event.
2638 * @file - file struct
2640 * @wait - poll table
2642 * This is same as datagram poll, except for the special case of
2643 * blocking sockets. If application is using a blocking fd
2644 * and a packet with checksum error is in the queue;
2645 * then it could get return from select indicating data available
2646 * but then block when reading it. Add special case code
2647 * to work around these arguably broken applications.
2649 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2651 __poll_t mask = datagram_poll(file, sock, wait);
2652 struct sock *sk = sock->sk;
2654 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2655 mask |= EPOLLIN | EPOLLRDNORM;
2657 /* Check for false positives due to checksum errors */
2658 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2659 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2660 mask &= ~(EPOLLIN | EPOLLRDNORM);
2665 EXPORT_SYMBOL(udp_poll);
2667 int udp_abort(struct sock *sk, int err)
2672 sk->sk_error_report(sk);
2673 __udp_disconnect(sk, 0);
2679 EXPORT_SYMBOL_GPL(udp_abort);
2681 struct proto udp_prot = {
2683 .owner = THIS_MODULE,
2684 .close = udp_lib_close,
2685 .pre_connect = udp_pre_connect,
2686 .connect = ip4_datagram_connect,
2687 .disconnect = udp_disconnect,
2689 .init = udp_init_sock,
2690 .destroy = udp_destroy_sock,
2691 .setsockopt = udp_setsockopt,
2692 .getsockopt = udp_getsockopt,
2693 .sendmsg = udp_sendmsg,
2694 .recvmsg = udp_recvmsg,
2695 .sendpage = udp_sendpage,
2696 .release_cb = ip4_datagram_release_cb,
2697 .hash = udp_lib_hash,
2698 .unhash = udp_lib_unhash,
2699 .rehash = udp_v4_rehash,
2700 .get_port = udp_v4_get_port,
2701 .memory_allocated = &udp_memory_allocated,
2702 .sysctl_mem = sysctl_udp_mem,
2703 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2704 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2705 .obj_size = sizeof(struct udp_sock),
2706 .h.udp_table = &udp_table,
2707 #ifdef CONFIG_COMPAT
2708 .compat_setsockopt = compat_udp_setsockopt,
2709 .compat_getsockopt = compat_udp_getsockopt,
2711 .diag_destroy = udp_abort,
2713 EXPORT_SYMBOL(udp_prot);
2715 /* ------------------------------------------------------------------------ */
2716 #ifdef CONFIG_PROC_FS
2718 static struct sock *udp_get_first(struct seq_file *seq, int start)
2721 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2722 struct udp_iter_state *state = seq->private;
2723 struct net *net = seq_file_net(seq);
2725 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2727 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2729 if (hlist_empty(&hslot->head))
2732 spin_lock_bh(&hslot->lock);
2733 sk_for_each(sk, &hslot->head) {
2734 if (!net_eq(sock_net(sk), net))
2736 if (sk->sk_family == afinfo->family)
2739 spin_unlock_bh(&hslot->lock);
2746 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2748 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2749 struct udp_iter_state *state = seq->private;
2750 struct net *net = seq_file_net(seq);
2754 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2757 if (state->bucket <= afinfo->udp_table->mask)
2758 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2759 return udp_get_first(seq, state->bucket + 1);
2764 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2766 struct sock *sk = udp_get_first(seq, 0);
2769 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2771 return pos ? NULL : sk;
2774 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2776 struct udp_iter_state *state = seq->private;
2777 state->bucket = MAX_UDP_PORTS;
2779 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2781 EXPORT_SYMBOL(udp_seq_start);
2783 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2787 if (v == SEQ_START_TOKEN)
2788 sk = udp_get_idx(seq, 0);
2790 sk = udp_get_next(seq, v);
2795 EXPORT_SYMBOL(udp_seq_next);
2797 void udp_seq_stop(struct seq_file *seq, void *v)
2799 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2800 struct udp_iter_state *state = seq->private;
2802 if (state->bucket <= afinfo->udp_table->mask)
2803 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2805 EXPORT_SYMBOL(udp_seq_stop);
2807 /* ------------------------------------------------------------------------ */
2808 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2811 struct inet_sock *inet = inet_sk(sp);
2812 __be32 dest = inet->inet_daddr;
2813 __be32 src = inet->inet_rcv_saddr;
2814 __u16 destp = ntohs(inet->inet_dport);
2815 __u16 srcp = ntohs(inet->inet_sport);
2817 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2818 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2819 bucket, src, srcp, dest, destp, sp->sk_state,
2820 sk_wmem_alloc_get(sp),
2823 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2825 refcount_read(&sp->sk_refcnt), sp,
2826 atomic_read(&sp->sk_drops));
2829 int udp4_seq_show(struct seq_file *seq, void *v)
2831 seq_setwidth(seq, 127);
2832 if (v == SEQ_START_TOKEN)
2833 seq_puts(seq, " sl local_address rem_address st tx_queue "
2834 "rx_queue tr tm->when retrnsmt uid timeout "
2835 "inode ref pointer drops");
2837 struct udp_iter_state *state = seq->private;
2839 udp4_format_sock(v, seq, state->bucket);
2845 const struct seq_operations udp_seq_ops = {
2846 .start = udp_seq_start,
2847 .next = udp_seq_next,
2848 .stop = udp_seq_stop,
2849 .show = udp4_seq_show,
2851 EXPORT_SYMBOL(udp_seq_ops);
2853 static struct udp_seq_afinfo udp4_seq_afinfo = {
2855 .udp_table = &udp_table,
2858 static int __net_init udp4_proc_init_net(struct net *net)
2860 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2861 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2866 static void __net_exit udp4_proc_exit_net(struct net *net)
2868 remove_proc_entry("udp", net->proc_net);
2871 static struct pernet_operations udp4_net_ops = {
2872 .init = udp4_proc_init_net,
2873 .exit = udp4_proc_exit_net,
2876 int __init udp4_proc_init(void)
2878 return register_pernet_subsys(&udp4_net_ops);
2881 void udp4_proc_exit(void)
2883 unregister_pernet_subsys(&udp4_net_ops);
2885 #endif /* CONFIG_PROC_FS */
2887 static __initdata unsigned long uhash_entries;
2888 static int __init set_uhash_entries(char *str)
2895 ret = kstrtoul(str, 0, &uhash_entries);
2899 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2900 uhash_entries = UDP_HTABLE_SIZE_MIN;
2903 __setup("uhash_entries=", set_uhash_entries);
2905 void __init udp_table_init(struct udp_table *table, const char *name)
2909 table->hash = alloc_large_system_hash(name,
2910 2 * sizeof(struct udp_hslot),
2912 21, /* one slot per 2 MB */
2916 UDP_HTABLE_SIZE_MIN,
2919 table->hash2 = table->hash + (table->mask + 1);
2920 for (i = 0; i <= table->mask; i++) {
2921 INIT_HLIST_HEAD(&table->hash[i].head);
2922 table->hash[i].count = 0;
2923 spin_lock_init(&table->hash[i].lock);
2925 for (i = 0; i <= table->mask; i++) {
2926 INIT_HLIST_HEAD(&table->hash2[i].head);
2927 table->hash2[i].count = 0;
2928 spin_lock_init(&table->hash2[i].lock);
2932 u32 udp_flow_hashrnd(void)
2934 static u32 hashrnd __read_mostly;
2936 net_get_random_once(&hashrnd, sizeof(hashrnd));
2940 EXPORT_SYMBOL(udp_flow_hashrnd);
2942 static void __udp_sysctl_init(struct net *net)
2944 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2945 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2947 #ifdef CONFIG_NET_L3_MASTER_DEV
2948 net->ipv4.sysctl_udp_l3mdev_accept = 0;
2952 static int __net_init udp_sysctl_init(struct net *net)
2954 __udp_sysctl_init(net);
2958 static struct pernet_operations __net_initdata udp_sysctl_ops = {
2959 .init = udp_sysctl_init,
2962 void __init udp_init(void)
2964 unsigned long limit;
2967 udp_table_init(&udp_table, "UDP");
2968 limit = nr_free_buffer_pages() / 8;
2969 limit = max(limit, 128UL);
2970 sysctl_udp_mem[0] = limit / 4 * 3;
2971 sysctl_udp_mem[1] = limit;
2972 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2974 __udp_sysctl_init(&init_net);
2976 /* 16 spinlocks per cpu */
2977 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2978 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2981 panic("UDP: failed to alloc udp_busylocks\n");
2982 for (i = 0; i < (1U << udp_busylocks_log); i++)
2983 spin_lock_init(udp_busylocks + i);
2985 if (register_pernet_subsys(&udp_sysctl_ops))
2986 panic("UDP: failed to init sysctl parameters.\n");