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).
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.
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 int udp_v4_get_port(struct sock *sk, unsigned short snum)
362 unsigned int hash2_nulladdr =
363 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
364 unsigned int hash2_partial =
365 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
367 /* precompute partial secondary hash */
368 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
369 return udp_lib_get_port(sk, snum, hash2_nulladdr);
372 static int compute_score(struct sock *sk, struct net *net,
373 __be32 saddr, __be16 sport,
374 __be32 daddr, unsigned short hnum,
375 int dif, int sdif, bool exact_dif)
378 struct inet_sock *inet;
380 if (!net_eq(sock_net(sk), net) ||
381 udp_sk(sk)->udp_port_hash != hnum ||
385 score = (sk->sk_family == PF_INET) ? 2 : 1;
388 if (inet->inet_rcv_saddr) {
389 if (inet->inet_rcv_saddr != daddr)
394 if (inet->inet_daddr) {
395 if (inet->inet_daddr != saddr)
400 if (inet->inet_dport) {
401 if (inet->inet_dport != sport)
406 if (sk->sk_bound_dev_if || exact_dif) {
407 bool dev_match = (sk->sk_bound_dev_if == dif ||
408 sk->sk_bound_dev_if == sdif);
410 if (exact_dif && !dev_match)
412 if (sk->sk_bound_dev_if && dev_match)
416 if (sk->sk_incoming_cpu == raw_smp_processor_id())
421 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
422 const __u16 lport, const __be32 faddr,
425 static u32 udp_ehash_secret __read_mostly;
427 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
429 return __inet_ehashfn(laddr, lport, faddr, fport,
430 udp_ehash_secret + net_hash_mix(net));
433 /* called with rcu_read_lock() */
434 static struct sock *udp4_lib_lookup2(struct net *net,
435 __be32 saddr, __be16 sport,
436 __be32 daddr, unsigned int hnum,
437 int dif, int sdif, bool exact_dif,
438 struct udp_hslot *hslot2,
441 struct sock *sk, *result;
447 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
448 score = compute_score(sk, net, saddr, sport,
449 daddr, hnum, dif, sdif, exact_dif);
450 if (score > badness) {
451 if (sk->sk_reuseport) {
452 hash = udp_ehashfn(net, daddr, hnum,
454 result = reuseport_select_sock(sk, hash, skb,
455 sizeof(struct udphdr));
466 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
467 * harder than this. -DaveM
469 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
470 __be16 sport, __be32 daddr, __be16 dport, int dif,
471 int sdif, struct udp_table *udptable, struct sk_buff *skb)
473 struct sock *sk, *result;
474 unsigned short hnum = ntohs(dport);
475 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
476 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
477 bool exact_dif = udp_lib_exact_dif_match(net, skb);
481 if (hslot->count > 10) {
482 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
483 slot2 = hash2 & udptable->mask;
484 hslot2 = &udptable->hash2[slot2];
485 if (hslot->count < hslot2->count)
488 result = udp4_lib_lookup2(net, saddr, sport,
489 daddr, hnum, dif, sdif,
490 exact_dif, hslot2, skb);
492 unsigned int old_slot2 = slot2;
493 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
494 slot2 = hash2 & udptable->mask;
495 /* avoid searching the same slot again. */
496 if (unlikely(slot2 == old_slot2))
499 hslot2 = &udptable->hash2[slot2];
500 if (hslot->count < hslot2->count)
503 result = udp4_lib_lookup2(net, saddr, sport,
504 daddr, hnum, dif, sdif,
505 exact_dif, hslot2, skb);
512 sk_for_each_rcu(sk, &hslot->head) {
513 score = compute_score(sk, net, saddr, sport,
514 daddr, hnum, dif, sdif, exact_dif);
515 if (score > badness) {
516 if (sk->sk_reuseport) {
517 hash = udp_ehashfn(net, daddr, hnum,
519 result = reuseport_select_sock(sk, hash, skb,
520 sizeof(struct udphdr));
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 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
548 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
550 /* Must be called under rcu_read_lock().
551 * Does increment socket refcount.
553 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
554 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
555 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
556 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
557 __be32 daddr, __be16 dport, int dif)
561 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
562 dif, 0, &udp_table, NULL);
563 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
567 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
570 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
571 __be16 loc_port, __be32 loc_addr,
572 __be16 rmt_port, __be32 rmt_addr,
573 int dif, int sdif, unsigned short hnum)
575 struct inet_sock *inet = inet_sk(sk);
577 if (!net_eq(sock_net(sk), net) ||
578 udp_sk(sk)->udp_port_hash != hnum ||
579 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
580 (inet->inet_dport != rmt_port && inet->inet_dport) ||
581 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
582 ipv6_only_sock(sk) ||
583 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
584 sk->sk_bound_dev_if != sdif))
586 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
592 * This routine is called by the ICMP module when it gets some
593 * sort of error condition. If err < 0 then the socket should
594 * be closed and the error returned to the user. If err > 0
595 * it's just the icmp type << 8 | icmp code.
596 * Header points to the ip header of the error packet. We move
597 * on past this. Then (as it used to claim before adjustment)
598 * header points to the first 8 bytes of the udp header. We need
599 * to find the appropriate port.
602 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
604 struct inet_sock *inet;
605 const struct iphdr *iph = (const struct iphdr *)skb->data;
606 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
607 const int type = icmp_hdr(skb)->type;
608 const int code = icmp_hdr(skb)->code;
612 struct net *net = dev_net(skb->dev);
614 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
615 iph->saddr, uh->source, skb->dev->ifindex, 0,
618 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
619 return; /* No socket for error */
628 case ICMP_TIME_EXCEEDED:
631 case ICMP_SOURCE_QUENCH:
633 case ICMP_PARAMETERPROB:
637 case ICMP_DEST_UNREACH:
638 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
639 ipv4_sk_update_pmtu(skb, sk, info);
640 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
648 if (code <= NR_ICMP_UNREACH) {
649 harderr = icmp_err_convert[code].fatal;
650 err = icmp_err_convert[code].errno;
654 ipv4_sk_redirect(skb, sk);
659 * RFC1122: OK. Passes ICMP errors back to application, as per
662 if (!inet->recverr) {
663 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
666 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
669 sk->sk_error_report(sk);
674 void udp_err(struct sk_buff *skb, u32 info)
676 __udp4_lib_err(skb, info, &udp_table);
680 * Throw away all pending data and cancel the corking. Socket is locked.
682 void udp_flush_pending_frames(struct sock *sk)
684 struct udp_sock *up = udp_sk(sk);
689 ip_flush_pending_frames(sk);
692 EXPORT_SYMBOL(udp_flush_pending_frames);
695 * udp4_hwcsum - handle outgoing HW checksumming
696 * @skb: sk_buff containing the filled-in UDP header
697 * (checksum field must be zeroed out)
698 * @src: source IP address
699 * @dst: destination IP address
701 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
703 struct udphdr *uh = udp_hdr(skb);
704 int offset = skb_transport_offset(skb);
705 int len = skb->len - offset;
709 if (!skb_has_frag_list(skb)) {
711 * Only one fragment on the socket.
713 skb->csum_start = skb_transport_header(skb) - skb->head;
714 skb->csum_offset = offsetof(struct udphdr, check);
715 uh->check = ~csum_tcpudp_magic(src, dst, len,
718 struct sk_buff *frags;
721 * HW-checksum won't work as there are two or more
722 * fragments on the socket so that all csums of sk_buffs
725 skb_walk_frags(skb, frags) {
726 csum = csum_add(csum, frags->csum);
730 csum = skb_checksum(skb, offset, hlen, csum);
731 skb->ip_summed = CHECKSUM_NONE;
733 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
735 uh->check = CSUM_MANGLED_0;
738 EXPORT_SYMBOL_GPL(udp4_hwcsum);
740 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
741 * for the simple case like when setting the checksum for a UDP tunnel.
743 void udp_set_csum(bool nocheck, struct sk_buff *skb,
744 __be32 saddr, __be32 daddr, int len)
746 struct udphdr *uh = udp_hdr(skb);
750 } else if (skb_is_gso(skb)) {
751 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
752 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
754 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
756 uh->check = CSUM_MANGLED_0;
758 skb->ip_summed = CHECKSUM_PARTIAL;
759 skb->csum_start = skb_transport_header(skb) - skb->head;
760 skb->csum_offset = offsetof(struct udphdr, check);
761 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
764 EXPORT_SYMBOL(udp_set_csum);
766 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
768 struct sock *sk = skb->sk;
769 struct inet_sock *inet = inet_sk(sk);
772 int is_udplite = IS_UDPLITE(sk);
773 int offset = skb_transport_offset(skb);
774 int len = skb->len - offset;
778 * Create a UDP header
781 uh->source = inet->inet_sport;
782 uh->dest = fl4->fl4_dport;
783 uh->len = htons(len);
786 if (is_udplite) /* UDP-Lite */
787 csum = udplite_csum(skb);
789 else if (sk->sk_no_check_tx) { /* UDP csum off */
791 skb->ip_summed = CHECKSUM_NONE;
794 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
796 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
800 csum = udp_csum(skb);
802 /* add protocol-dependent pseudo-header */
803 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
804 sk->sk_protocol, csum);
806 uh->check = CSUM_MANGLED_0;
809 err = ip_send_skb(sock_net(sk), skb);
811 if (err == -ENOBUFS && !inet->recverr) {
812 UDP_INC_STATS(sock_net(sk),
813 UDP_MIB_SNDBUFERRORS, is_udplite);
817 UDP_INC_STATS(sock_net(sk),
818 UDP_MIB_OUTDATAGRAMS, is_udplite);
823 * Push out all pending data as one UDP datagram. Socket is locked.
825 int udp_push_pending_frames(struct sock *sk)
827 struct udp_sock *up = udp_sk(sk);
828 struct inet_sock *inet = inet_sk(sk);
829 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
833 skb = ip_finish_skb(sk, fl4);
837 err = udp_send_skb(skb, fl4);
844 EXPORT_SYMBOL(udp_push_pending_frames);
846 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
848 struct inet_sock *inet = inet_sk(sk);
849 struct udp_sock *up = udp_sk(sk);
850 struct flowi4 fl4_stack;
853 struct ipcm_cookie ipc;
854 struct rtable *rt = NULL;
857 __be32 daddr, faddr, saddr;
860 int err, is_udplite = IS_UDPLITE(sk);
861 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
862 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
864 struct ip_options_data opt_copy;
873 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
881 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
883 fl4 = &inet->cork.fl.u.ip4;
886 * There are pending frames.
887 * The socket lock must be held while it's corked.
890 if (likely(up->pending)) {
891 if (unlikely(up->pending != AF_INET)) {
899 ulen += sizeof(struct udphdr);
902 * Get and verify the address.
905 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
906 if (msg->msg_namelen < sizeof(*usin))
908 if (usin->sin_family != AF_INET) {
909 if (usin->sin_family != AF_UNSPEC)
910 return -EAFNOSUPPORT;
913 daddr = usin->sin_addr.s_addr;
914 dport = usin->sin_port;
918 if (sk->sk_state != TCP_ESTABLISHED)
919 return -EDESTADDRREQ;
920 daddr = inet->inet_daddr;
921 dport = inet->inet_dport;
922 /* Open fast path for connected socket.
923 Route will not be used, if at least one option is set.
928 ipc.sockc.tsflags = sk->sk_tsflags;
929 ipc.addr = inet->inet_saddr;
930 ipc.oif = sk->sk_bound_dev_if;
932 if (msg->msg_controllen) {
933 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
943 struct ip_options_rcu *inet_opt;
946 inet_opt = rcu_dereference(inet->inet_opt);
948 memcpy(&opt_copy, inet_opt,
949 sizeof(*inet_opt) + inet_opt->opt.optlen);
950 ipc.opt = &opt_copy.opt;
956 ipc.addr = faddr = daddr;
958 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
960 if (ipc.opt && ipc.opt->opt.srr) {
963 faddr = ipc.opt->opt.faddr;
966 tos = get_rttos(&ipc, inet);
967 if (sock_flag(sk, SOCK_LOCALROUTE) ||
968 (msg->msg_flags & MSG_DONTROUTE) ||
969 (ipc.opt && ipc.opt->opt.is_strictroute)) {
974 if (ipv4_is_multicast(daddr)) {
976 ipc.oif = inet->mc_index;
978 saddr = inet->mc_addr;
981 ipc.oif = inet->uc_index;
984 rt = (struct rtable *)sk_dst_check(sk, 0);
987 struct net *net = sock_net(sk);
988 __u8 flow_flags = inet_sk_flowi_flags(sk);
992 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
993 RT_SCOPE_UNIVERSE, sk->sk_protocol,
995 faddr, saddr, dport, inet->inet_sport,
998 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
999 rt = ip_route_output_flow(net, fl4, sk);
1003 if (err == -ENETUNREACH)
1004 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1009 if ((rt->rt_flags & RTCF_BROADCAST) &&
1010 !sock_flag(sk, SOCK_BROADCAST))
1013 sk_dst_set(sk, dst_clone(&rt->dst));
1016 if (msg->msg_flags&MSG_CONFIRM)
1022 daddr = ipc.addr = fl4->daddr;
1024 /* Lockless fast path for the non-corking case. */
1026 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1027 sizeof(struct udphdr), &ipc, &rt,
1030 if (!IS_ERR_OR_NULL(skb))
1031 err = udp_send_skb(skb, fl4);
1036 if (unlikely(up->pending)) {
1037 /* The socket is already corked while preparing it. */
1038 /* ... which is an evident application bug. --ANK */
1041 net_dbg_ratelimited("socket already corked\n");
1046 * Now cork the socket to pend data.
1048 fl4 = &inet->cork.fl.u.ip4;
1051 fl4->fl4_dport = dport;
1052 fl4->fl4_sport = inet->inet_sport;
1053 up->pending = AF_INET;
1057 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1058 sizeof(struct udphdr), &ipc, &rt,
1059 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1061 udp_flush_pending_frames(sk);
1063 err = udp_push_pending_frames(sk);
1064 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1075 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1076 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1077 * we don't have a good statistic (IpOutDiscards but it can be too many
1078 * things). We could add another new stat but at least for now that
1079 * seems like overkill.
1081 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1082 UDP_INC_STATS(sock_net(sk),
1083 UDP_MIB_SNDBUFERRORS, is_udplite);
1088 if (msg->msg_flags & MSG_PROBE)
1089 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1090 if (!(msg->msg_flags&MSG_PROBE) || len)
1091 goto back_from_confirm;
1095 EXPORT_SYMBOL(udp_sendmsg);
1097 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1098 size_t size, int flags)
1100 struct inet_sock *inet = inet_sk(sk);
1101 struct udp_sock *up = udp_sk(sk);
1104 if (flags & MSG_SENDPAGE_NOTLAST)
1108 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1110 /* Call udp_sendmsg to specify destination address which
1111 * sendpage interface can't pass.
1112 * This will succeed only when the socket is connected.
1114 ret = udp_sendmsg(sk, &msg, 0);
1121 if (unlikely(!up->pending)) {
1124 net_dbg_ratelimited("cork failed\n");
1128 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1129 page, offset, size, flags);
1130 if (ret == -EOPNOTSUPP) {
1132 return sock_no_sendpage(sk->sk_socket, page, offset,
1136 udp_flush_pending_frames(sk);
1141 if (!(up->corkflag || (flags&MSG_MORE)))
1142 ret = udp_push_pending_frames(sk);
1150 #define UDP_SKB_IS_STATELESS 0x80000000
1152 static void udp_set_dev_scratch(struct sk_buff *skb)
1154 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1156 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1157 scratch->_tsize_state = skb->truesize;
1158 #if BITS_PER_LONG == 64
1159 scratch->len = skb->len;
1160 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1161 scratch->is_linear = !skb_is_nonlinear(skb);
1163 /* all head states execept sp (dst, sk, nf) are always cleared by
1164 * udp_rcv() and we need to preserve secpath, if present, to eventually
1165 * process IP_CMSG_PASSSEC at recvmsg() time
1167 if (likely(!skb_sec_path(skb)))
1168 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1171 static int udp_skb_truesize(struct sk_buff *skb)
1173 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1176 static bool udp_skb_has_head_state(struct sk_buff *skb)
1178 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1181 /* fully reclaim rmem/fwd memory allocated for skb */
1182 static void udp_rmem_release(struct sock *sk, int size, int partial,
1183 bool rx_queue_lock_held)
1185 struct udp_sock *up = udp_sk(sk);
1186 struct sk_buff_head *sk_queue;
1189 if (likely(partial)) {
1190 up->forward_deficit += size;
1191 size = up->forward_deficit;
1192 if (size < (sk->sk_rcvbuf >> 2))
1195 size += up->forward_deficit;
1197 up->forward_deficit = 0;
1199 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1200 * if the called don't held it already
1202 sk_queue = &sk->sk_receive_queue;
1203 if (!rx_queue_lock_held)
1204 spin_lock(&sk_queue->lock);
1207 sk->sk_forward_alloc += size;
1208 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1209 sk->sk_forward_alloc -= amt;
1212 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1214 atomic_sub(size, &sk->sk_rmem_alloc);
1216 /* this can save us from acquiring the rx queue lock on next receive */
1217 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1219 if (!rx_queue_lock_held)
1220 spin_unlock(&sk_queue->lock);
1223 /* Note: called with reader_queue.lock held.
1224 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1225 * This avoids a cache line miss while receive_queue lock is held.
1226 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1228 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1230 prefetch(&skb->data);
1231 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1233 EXPORT_SYMBOL(udp_skb_destructor);
1235 /* as above, but the caller held the rx queue lock, too */
1236 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1238 prefetch(&skb->data);
1239 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1242 /* Idea of busylocks is to let producers grab an extra spinlock
1243 * to relieve pressure on the receive_queue spinlock shared by consumer.
1244 * Under flood, this means that only one producer can be in line
1245 * trying to acquire the receive_queue spinlock.
1246 * These busylock can be allocated on a per cpu manner, instead of a
1247 * per socket one (that would consume a cache line per socket)
1249 static int udp_busylocks_log __read_mostly;
1250 static spinlock_t *udp_busylocks __read_mostly;
1252 static spinlock_t *busylock_acquire(void *ptr)
1256 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1261 static void busylock_release(spinlock_t *busy)
1267 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1269 struct sk_buff_head *list = &sk->sk_receive_queue;
1270 int rmem, delta, amt, err = -ENOMEM;
1271 spinlock_t *busy = NULL;
1274 /* try to avoid the costly atomic add/sub pair when the receive
1275 * queue is full; always allow at least a packet
1277 rmem = atomic_read(&sk->sk_rmem_alloc);
1278 if (rmem > sk->sk_rcvbuf)
1281 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1282 * having linear skbs :
1283 * - Reduce memory overhead and thus increase receive queue capacity
1284 * - Less cache line misses at copyout() time
1285 * - Less work at consume_skb() (less alien page frag freeing)
1287 if (rmem > (sk->sk_rcvbuf >> 1)) {
1290 busy = busylock_acquire(sk);
1292 size = skb->truesize;
1293 udp_set_dev_scratch(skb);
1295 /* we drop only if the receive buf is full and the receive
1296 * queue contains some other skb
1298 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1299 if (rmem > (size + sk->sk_rcvbuf))
1302 spin_lock(&list->lock);
1303 if (size >= sk->sk_forward_alloc) {
1304 amt = sk_mem_pages(size);
1305 delta = amt << SK_MEM_QUANTUM_SHIFT;
1306 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1308 spin_unlock(&list->lock);
1312 sk->sk_forward_alloc += delta;
1315 sk->sk_forward_alloc -= size;
1317 /* no need to setup a destructor, we will explicitly release the
1318 * forward allocated memory on dequeue
1320 sock_skb_set_dropcount(sk, skb);
1322 __skb_queue_tail(list, skb);
1323 spin_unlock(&list->lock);
1325 if (!sock_flag(sk, SOCK_DEAD))
1326 sk->sk_data_ready(sk);
1328 busylock_release(busy);
1332 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1335 atomic_inc(&sk->sk_drops);
1336 busylock_release(busy);
1339 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1341 void udp_destruct_sock(struct sock *sk)
1343 /* reclaim completely the forward allocated memory */
1344 struct udp_sock *up = udp_sk(sk);
1345 unsigned int total = 0;
1346 struct sk_buff *skb;
1348 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1349 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1350 total += skb->truesize;
1353 udp_rmem_release(sk, total, 0, true);
1355 inet_sock_destruct(sk);
1357 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1359 int udp_init_sock(struct sock *sk)
1361 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1362 sk->sk_destruct = udp_destruct_sock;
1365 EXPORT_SYMBOL_GPL(udp_init_sock);
1367 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1369 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1370 bool slow = lock_sock_fast(sk);
1372 sk_peek_offset_bwd(sk, len);
1373 unlock_sock_fast(sk, slow);
1376 if (!skb_unref(skb))
1379 /* In the more common cases we cleared the head states previously,
1380 * see __udp_queue_rcv_skb().
1382 if (unlikely(udp_skb_has_head_state(skb)))
1383 skb_release_head_state(skb);
1384 __consume_stateless_skb(skb);
1386 EXPORT_SYMBOL_GPL(skb_consume_udp);
1388 static struct sk_buff *__first_packet_length(struct sock *sk,
1389 struct sk_buff_head *rcvq,
1392 struct sk_buff *skb;
1394 while ((skb = skb_peek(rcvq)) != NULL) {
1395 if (udp_lib_checksum_complete(skb)) {
1396 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1398 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1400 atomic_inc(&sk->sk_drops);
1401 __skb_unlink(skb, rcvq);
1402 *total += skb->truesize;
1405 /* the csum related bits could be changed, refresh
1408 udp_set_dev_scratch(skb);
1416 * first_packet_length - return length of first packet in receive queue
1419 * Drops all bad checksum frames, until a valid one is found.
1420 * Returns the length of found skb, or -1 if none is found.
1422 static int first_packet_length(struct sock *sk)
1424 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1425 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1426 struct sk_buff *skb;
1430 spin_lock_bh(&rcvq->lock);
1431 skb = __first_packet_length(sk, rcvq, &total);
1432 if (!skb && !skb_queue_empty(sk_queue)) {
1433 spin_lock(&sk_queue->lock);
1434 skb_queue_splice_tail_init(sk_queue, rcvq);
1435 spin_unlock(&sk_queue->lock);
1437 skb = __first_packet_length(sk, rcvq, &total);
1439 res = skb ? skb->len : -1;
1441 udp_rmem_release(sk, total, 1, false);
1442 spin_unlock_bh(&rcvq->lock);
1447 * IOCTL requests applicable to the UDP protocol
1450 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1455 int amount = sk_wmem_alloc_get(sk);
1457 return put_user(amount, (int __user *)arg);
1462 int amount = max_t(int, 0, first_packet_length(sk));
1464 return put_user(amount, (int __user *)arg);
1468 return -ENOIOCTLCMD;
1473 EXPORT_SYMBOL(udp_ioctl);
1475 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1476 int noblock, int *peeked, int *off, int *err)
1478 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1479 struct sk_buff_head *queue;
1480 struct sk_buff *last;
1484 queue = &udp_sk(sk)->reader_queue;
1485 flags |= noblock ? MSG_DONTWAIT : 0;
1486 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1488 struct sk_buff *skb;
1490 error = sock_error(sk);
1497 spin_lock_bh(&queue->lock);
1498 skb = __skb_try_recv_from_queue(sk, queue, flags,
1503 spin_unlock_bh(&queue->lock);
1507 if (skb_queue_empty(sk_queue)) {
1508 spin_unlock_bh(&queue->lock);
1512 /* refill the reader queue and walk it again
1513 * keep both queues locked to avoid re-acquiring
1514 * the sk_receive_queue lock if fwd memory scheduling
1517 spin_lock(&sk_queue->lock);
1518 skb_queue_splice_tail_init(sk_queue, queue);
1520 skb = __skb_try_recv_from_queue(sk, queue, flags,
1521 udp_skb_dtor_locked,
1524 spin_unlock(&sk_queue->lock);
1525 spin_unlock_bh(&queue->lock);
1530 if (!sk_can_busy_loop(sk))
1533 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1534 } while (!skb_queue_empty(sk_queue));
1536 /* sk_queue is empty, reader_queue may contain peeked packets */
1538 !__skb_wait_for_more_packets(sk, &error, &timeo,
1539 (struct sk_buff *)sk_queue));
1544 EXPORT_SYMBOL_GPL(__skb_recv_udp);
1547 * This should be easy, if there is something there we
1548 * return it, otherwise we block.
1551 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1552 int flags, int *addr_len)
1554 struct inet_sock *inet = inet_sk(sk);
1555 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1556 struct sk_buff *skb;
1557 unsigned int ulen, copied;
1558 int peeked, peeking, off;
1560 int is_udplite = IS_UDPLITE(sk);
1561 bool checksum_valid = false;
1563 if (flags & MSG_ERRQUEUE)
1564 return ip_recv_error(sk, msg, len, addr_len);
1567 peeking = flags & MSG_PEEK;
1568 off = sk_peek_offset(sk, flags);
1569 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1573 ulen = udp_skb_len(skb);
1575 if (copied > ulen - off)
1576 copied = ulen - off;
1577 else if (copied < ulen)
1578 msg->msg_flags |= MSG_TRUNC;
1581 * If checksum is needed at all, try to do it while copying the
1582 * data. If the data is truncated, or if we only want a partial
1583 * coverage checksum (UDP-Lite), do it before the copy.
1586 if (copied < ulen || peeking ||
1587 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1588 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1589 !__udp_lib_checksum_complete(skb);
1590 if (!checksum_valid)
1594 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1595 if (udp_skb_is_linear(skb))
1596 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1598 err = skb_copy_datagram_msg(skb, off, msg, copied);
1600 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1606 if (unlikely(err)) {
1608 atomic_inc(&sk->sk_drops);
1609 UDP_INC_STATS(sock_net(sk),
1610 UDP_MIB_INERRORS, is_udplite);
1617 UDP_INC_STATS(sock_net(sk),
1618 UDP_MIB_INDATAGRAMS, is_udplite);
1620 sock_recv_ts_and_drops(msg, sk, skb);
1622 /* Copy the address. */
1624 sin->sin_family = AF_INET;
1625 sin->sin_port = udp_hdr(skb)->source;
1626 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1627 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1628 *addr_len = sizeof(*sin);
1630 if (inet->cmsg_flags)
1631 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1634 if (flags & MSG_TRUNC)
1637 skb_consume_udp(sk, skb, peeking ? -err : err);
1641 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1642 udp_skb_destructor)) {
1643 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1644 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1648 /* starting over for a new packet, but check if we need to yield */
1650 msg->msg_flags &= ~MSG_TRUNC;
1654 int __udp_disconnect(struct sock *sk, int flags)
1656 struct inet_sock *inet = inet_sk(sk);
1658 * 1003.1g - break association.
1661 sk->sk_state = TCP_CLOSE;
1662 inet->inet_daddr = 0;
1663 inet->inet_dport = 0;
1664 sock_rps_reset_rxhash(sk);
1665 sk->sk_bound_dev_if = 0;
1666 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1667 inet_reset_saddr(sk);
1669 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1670 sk->sk_prot->unhash(sk);
1671 inet->inet_sport = 0;
1676 EXPORT_SYMBOL(__udp_disconnect);
1678 int udp_disconnect(struct sock *sk, int flags)
1681 __udp_disconnect(sk, flags);
1685 EXPORT_SYMBOL(udp_disconnect);
1687 void udp_lib_unhash(struct sock *sk)
1689 if (sk_hashed(sk)) {
1690 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1691 struct udp_hslot *hslot, *hslot2;
1693 hslot = udp_hashslot(udptable, sock_net(sk),
1694 udp_sk(sk)->udp_port_hash);
1695 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1697 spin_lock_bh(&hslot->lock);
1698 if (rcu_access_pointer(sk->sk_reuseport_cb))
1699 reuseport_detach_sock(sk);
1700 if (sk_del_node_init_rcu(sk)) {
1702 inet_sk(sk)->inet_num = 0;
1703 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1705 spin_lock(&hslot2->lock);
1706 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1708 spin_unlock(&hslot2->lock);
1710 spin_unlock_bh(&hslot->lock);
1713 EXPORT_SYMBOL(udp_lib_unhash);
1716 * inet_rcv_saddr was changed, we must rehash secondary hash
1718 void udp_lib_rehash(struct sock *sk, u16 newhash)
1720 if (sk_hashed(sk)) {
1721 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1722 struct udp_hslot *hslot, *hslot2, *nhslot2;
1724 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1725 nhslot2 = udp_hashslot2(udptable, newhash);
1726 udp_sk(sk)->udp_portaddr_hash = newhash;
1728 if (hslot2 != nhslot2 ||
1729 rcu_access_pointer(sk->sk_reuseport_cb)) {
1730 hslot = udp_hashslot(udptable, sock_net(sk),
1731 udp_sk(sk)->udp_port_hash);
1732 /* we must lock primary chain too */
1733 spin_lock_bh(&hslot->lock);
1734 if (rcu_access_pointer(sk->sk_reuseport_cb))
1735 reuseport_detach_sock(sk);
1737 if (hslot2 != nhslot2) {
1738 spin_lock(&hslot2->lock);
1739 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1741 spin_unlock(&hslot2->lock);
1743 spin_lock(&nhslot2->lock);
1744 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1747 spin_unlock(&nhslot2->lock);
1750 spin_unlock_bh(&hslot->lock);
1754 EXPORT_SYMBOL(udp_lib_rehash);
1756 static void udp_v4_rehash(struct sock *sk)
1758 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1759 inet_sk(sk)->inet_rcv_saddr,
1760 inet_sk(sk)->inet_num);
1761 udp_lib_rehash(sk, new_hash);
1764 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1768 if (inet_sk(sk)->inet_daddr) {
1769 sock_rps_save_rxhash(sk, skb);
1770 sk_mark_napi_id(sk, skb);
1771 sk_incoming_cpu_update(sk);
1773 sk_mark_napi_id_once(sk, skb);
1776 rc = __udp_enqueue_schedule_skb(sk, skb);
1778 int is_udplite = IS_UDPLITE(sk);
1780 /* Note that an ENOMEM error is charged twice */
1782 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1784 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1786 trace_udp_fail_queue_rcv_skb(rc, sk);
1793 static struct static_key udp_encap_needed __read_mostly;
1794 void udp_encap_enable(void)
1796 static_key_enable(&udp_encap_needed);
1798 EXPORT_SYMBOL(udp_encap_enable);
1803 * >0: "udp encap" protocol resubmission
1805 * Note that in the success and error cases, the skb is assumed to
1806 * have either been requeued or freed.
1808 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1810 struct udp_sock *up = udp_sk(sk);
1811 int is_udplite = IS_UDPLITE(sk);
1814 * Charge it to the socket, dropping if the queue is full.
1816 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1820 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1821 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1824 * This is an encapsulation socket so pass the skb to
1825 * the socket's udp_encap_rcv() hook. Otherwise, just
1826 * fall through and pass this up the UDP socket.
1827 * up->encap_rcv() returns the following value:
1828 * =0 if skb was successfully passed to the encap
1829 * handler or was discarded by it.
1830 * >0 if skb should be passed on to UDP.
1831 * <0 if skb should be resubmitted as proto -N
1834 /* if we're overly short, let UDP handle it */
1835 encap_rcv = READ_ONCE(up->encap_rcv);
1839 /* Verify checksum before giving to encap */
1840 if (udp_lib_checksum_complete(skb))
1843 ret = encap_rcv(sk, skb);
1845 __UDP_INC_STATS(sock_net(sk),
1846 UDP_MIB_INDATAGRAMS,
1852 /* FALLTHROUGH -- it's a UDP Packet */
1856 * UDP-Lite specific tests, ignored on UDP sockets
1858 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1861 * MIB statistics other than incrementing the error count are
1862 * disabled for the following two types of errors: these depend
1863 * on the application settings, not on the functioning of the
1864 * protocol stack as such.
1866 * RFC 3828 here recommends (sec 3.3): "There should also be a
1867 * way ... to ... at least let the receiving application block
1868 * delivery of packets with coverage values less than a value
1869 * provided by the application."
1871 if (up->pcrlen == 0) { /* full coverage was set */
1872 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1873 UDP_SKB_CB(skb)->cscov, skb->len);
1876 /* The next case involves violating the min. coverage requested
1877 * by the receiver. This is subtle: if receiver wants x and x is
1878 * greater than the buffersize/MTU then receiver will complain
1879 * that it wants x while sender emits packets of smaller size y.
1880 * Therefore the above ...()->partial_cov statement is essential.
1882 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1883 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1884 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1889 prefetch(&sk->sk_rmem_alloc);
1890 if (rcu_access_pointer(sk->sk_filter) &&
1891 udp_lib_checksum_complete(skb))
1894 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1897 udp_csum_pull_header(skb);
1899 ipv4_pktinfo_prepare(sk, skb);
1900 return __udp_queue_rcv_skb(sk, skb);
1903 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1905 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1906 atomic_inc(&sk->sk_drops);
1911 /* For TCP sockets, sk_rx_dst is protected by socket lock
1912 * For UDP, we use xchg() to guard against concurrent changes.
1914 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1916 struct dst_entry *old;
1918 if (dst_hold_safe(dst)) {
1919 old = xchg(&sk->sk_rx_dst, dst);
1925 EXPORT_SYMBOL(udp_sk_rx_dst_set);
1928 * Multicasts and broadcasts go to each listener.
1930 * Note: called only from the BH handler context.
1932 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1934 __be32 saddr, __be32 daddr,
1935 struct udp_table *udptable,
1938 struct sock *sk, *first = NULL;
1939 unsigned short hnum = ntohs(uh->dest);
1940 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1941 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1942 unsigned int offset = offsetof(typeof(*sk), sk_node);
1943 int dif = skb->dev->ifindex;
1944 int sdif = inet_sdif(skb);
1945 struct hlist_node *node;
1946 struct sk_buff *nskb;
1949 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1951 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1953 hslot = &udptable->hash2[hash2];
1954 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1957 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1958 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1959 uh->source, saddr, dif, sdif, hnum))
1966 nskb = skb_clone(skb, GFP_ATOMIC);
1968 if (unlikely(!nskb)) {
1969 atomic_inc(&sk->sk_drops);
1970 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1972 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1976 if (udp_queue_rcv_skb(sk, nskb) > 0)
1980 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1981 if (use_hash2 && hash2 != hash2_any) {
1987 if (udp_queue_rcv_skb(first, skb) > 0)
1991 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
1992 proto == IPPROTO_UDPLITE);
1997 /* Initialize UDP checksum. If exited with zero value (success),
1998 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1999 * Otherwise, csum completion requires chacksumming packet body,
2000 * including udp header and folding it to skb->csum.
2002 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2007 UDP_SKB_CB(skb)->partial_cov = 0;
2008 UDP_SKB_CB(skb)->cscov = skb->len;
2010 if (proto == IPPROTO_UDPLITE) {
2011 err = udplite_checksum_init(skb, uh);
2016 /* Note, we are only interested in != 0 or == 0, thus the
2019 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2020 inet_compute_pseudo);
2024 * All we need to do is get the socket, and then do a checksum.
2027 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2032 unsigned short ulen;
2033 struct rtable *rt = skb_rtable(skb);
2034 __be32 saddr, daddr;
2035 struct net *net = dev_net(skb->dev);
2038 * Validate the packet.
2040 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2041 goto drop; /* No space for header. */
2044 ulen = ntohs(uh->len);
2045 saddr = ip_hdr(skb)->saddr;
2046 daddr = ip_hdr(skb)->daddr;
2048 if (ulen > skb->len)
2051 if (proto == IPPROTO_UDP) {
2052 /* UDP validates ulen. */
2053 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2058 if (udp4_csum_init(skb, uh, proto))
2061 sk = skb_steal_sock(skb);
2063 struct dst_entry *dst = skb_dst(skb);
2066 if (unlikely(sk->sk_rx_dst != dst))
2067 udp_sk_rx_dst_set(sk, dst);
2069 ret = udp_queue_rcv_skb(sk, skb);
2071 /* a return value > 0 means to resubmit the input, but
2072 * it wants the return to be -protocol, or 0
2079 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2080 return __udp4_lib_mcast_deliver(net, skb, uh,
2081 saddr, daddr, udptable, proto);
2083 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2087 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2088 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2089 inet_compute_pseudo);
2091 ret = udp_queue_rcv_skb(sk, skb);
2093 /* a return value > 0 means to resubmit the input, but
2094 * it wants the return to be -protocol, or 0
2101 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2105 /* No socket. Drop packet silently, if checksum is wrong */
2106 if (udp_lib_checksum_complete(skb))
2109 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2110 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2113 * Hmm. We got an UDP packet to a port to which we
2114 * don't wanna listen. Ignore it.
2120 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2121 proto == IPPROTO_UDPLITE ? "Lite" : "",
2122 &saddr, ntohs(uh->source),
2124 &daddr, ntohs(uh->dest));
2129 * RFC1122: OK. Discards the bad packet silently (as far as
2130 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2132 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2133 proto == IPPROTO_UDPLITE ? "Lite" : "",
2134 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2136 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2138 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2143 /* We can only early demux multicast if there is a single matching socket.
2144 * If more than one socket found returns NULL
2146 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2147 __be16 loc_port, __be32 loc_addr,
2148 __be16 rmt_port, __be32 rmt_addr,
2151 struct sock *sk, *result;
2152 unsigned short hnum = ntohs(loc_port);
2153 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2154 struct udp_hslot *hslot = &udp_table.hash[slot];
2156 /* Do not bother scanning a too big list */
2157 if (hslot->count > 10)
2161 sk_for_each_rcu(sk, &hslot->head) {
2162 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2163 rmt_port, rmt_addr, dif, sdif, hnum)) {
2173 /* For unicast we should only early demux connected sockets or we can
2174 * break forwarding setups. The chains here can be long so only check
2175 * if the first socket is an exact match and if not move on.
2177 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2178 __be16 loc_port, __be32 loc_addr,
2179 __be16 rmt_port, __be32 rmt_addr,
2182 unsigned short hnum = ntohs(loc_port);
2183 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2184 unsigned int slot2 = hash2 & udp_table.mask;
2185 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2186 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2187 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2190 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2191 if (INET_MATCH(sk, net, acookie, rmt_addr,
2192 loc_addr, ports, dif, sdif))
2194 /* Only check first socket in chain */
2200 int udp_v4_early_demux(struct sk_buff *skb)
2202 struct net *net = dev_net(skb->dev);
2203 struct in_device *in_dev = NULL;
2204 const struct iphdr *iph;
2205 const struct udphdr *uh;
2206 struct sock *sk = NULL;
2207 struct dst_entry *dst;
2208 int dif = skb->dev->ifindex;
2209 int sdif = inet_sdif(skb);
2212 /* validate the packet */
2213 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2219 if (skb->pkt_type == PACKET_MULTICAST) {
2220 in_dev = __in_dev_get_rcu(skb->dev);
2225 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2230 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2231 uh->source, iph->saddr,
2233 } else if (skb->pkt_type == PACKET_HOST) {
2234 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2235 uh->source, iph->saddr, dif, sdif);
2238 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2242 skb->destructor = sock_efree;
2243 dst = READ_ONCE(sk->sk_rx_dst);
2246 dst = dst_check(dst, 0);
2250 /* set noref for now.
2251 * any place which wants to hold dst has to call
2254 skb_dst_set_noref(skb, dst);
2256 /* for unconnected multicast sockets we need to validate
2257 * the source on each packet
2259 if (!inet_sk(sk)->inet_daddr && in_dev)
2260 return ip_mc_validate_source(skb, iph->daddr,
2261 iph->saddr, iph->tos,
2262 skb->dev, in_dev, &itag);
2267 int udp_rcv(struct sk_buff *skb)
2269 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2272 void udp_destroy_sock(struct sock *sk)
2274 struct udp_sock *up = udp_sk(sk);
2275 bool slow = lock_sock_fast(sk);
2276 udp_flush_pending_frames(sk);
2277 unlock_sock_fast(sk, slow);
2278 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2279 void (*encap_destroy)(struct sock *sk);
2280 encap_destroy = READ_ONCE(up->encap_destroy);
2287 * Socket option code for UDP
2289 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2290 char __user *optval, unsigned int optlen,
2291 int (*push_pending_frames)(struct sock *))
2293 struct udp_sock *up = udp_sk(sk);
2296 int is_udplite = IS_UDPLITE(sk);
2298 if (optlen < sizeof(int))
2301 if (get_user(val, (int __user *)optval))
2304 valbool = val ? 1 : 0;
2313 push_pending_frames(sk);
2321 case UDP_ENCAP_ESPINUDP:
2322 case UDP_ENCAP_ESPINUDP_NON_IKE:
2323 up->encap_rcv = xfrm4_udp_encap_rcv;
2325 case UDP_ENCAP_L2TPINUDP:
2326 up->encap_type = val;
2335 case UDP_NO_CHECK6_TX:
2336 up->no_check6_tx = valbool;
2339 case UDP_NO_CHECK6_RX:
2340 up->no_check6_rx = valbool;
2344 * UDP-Lite's partial checksum coverage (RFC 3828).
2346 /* The sender sets actual checksum coverage length via this option.
2347 * The case coverage > packet length is handled by send module. */
2348 case UDPLITE_SEND_CSCOV:
2349 if (!is_udplite) /* Disable the option on UDP sockets */
2350 return -ENOPROTOOPT;
2351 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2353 else if (val > USHRT_MAX)
2356 up->pcflag |= UDPLITE_SEND_CC;
2359 /* The receiver specifies a minimum checksum coverage value. To make
2360 * sense, this should be set to at least 8 (as done below). If zero is
2361 * used, this again means full checksum coverage. */
2362 case UDPLITE_RECV_CSCOV:
2363 if (!is_udplite) /* Disable the option on UDP sockets */
2364 return -ENOPROTOOPT;
2365 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2367 else if (val > USHRT_MAX)
2370 up->pcflag |= UDPLITE_RECV_CC;
2380 EXPORT_SYMBOL(udp_lib_setsockopt);
2382 int udp_setsockopt(struct sock *sk, int level, int optname,
2383 char __user *optval, unsigned int optlen)
2385 if (level == SOL_UDP || level == SOL_UDPLITE)
2386 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2387 udp_push_pending_frames);
2388 return ip_setsockopt(sk, level, optname, optval, optlen);
2391 #ifdef CONFIG_COMPAT
2392 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2393 char __user *optval, unsigned int optlen)
2395 if (level == SOL_UDP || level == SOL_UDPLITE)
2396 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2397 udp_push_pending_frames);
2398 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2402 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2403 char __user *optval, int __user *optlen)
2405 struct udp_sock *up = udp_sk(sk);
2408 if (get_user(len, optlen))
2411 len = min_t(unsigned int, len, sizeof(int));
2422 val = up->encap_type;
2425 case UDP_NO_CHECK6_TX:
2426 val = up->no_check6_tx;
2429 case UDP_NO_CHECK6_RX:
2430 val = up->no_check6_rx;
2433 /* The following two cannot be changed on UDP sockets, the return is
2434 * always 0 (which corresponds to the full checksum coverage of UDP). */
2435 case UDPLITE_SEND_CSCOV:
2439 case UDPLITE_RECV_CSCOV:
2444 return -ENOPROTOOPT;
2447 if (put_user(len, optlen))
2449 if (copy_to_user(optval, &val, len))
2453 EXPORT_SYMBOL(udp_lib_getsockopt);
2455 int udp_getsockopt(struct sock *sk, int level, int optname,
2456 char __user *optval, int __user *optlen)
2458 if (level == SOL_UDP || level == SOL_UDPLITE)
2459 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2460 return ip_getsockopt(sk, level, optname, optval, optlen);
2463 #ifdef CONFIG_COMPAT
2464 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2465 char __user *optval, int __user *optlen)
2467 if (level == SOL_UDP || level == SOL_UDPLITE)
2468 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2469 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2473 * udp_poll - wait for a UDP event.
2474 * @file - file struct
2476 * @wait - poll table
2478 * This is same as datagram poll, except for the special case of
2479 * blocking sockets. If application is using a blocking fd
2480 * and a packet with checksum error is in the queue;
2481 * then it could get return from select indicating data available
2482 * but then block when reading it. Add special case code
2483 * to work around these arguably broken applications.
2485 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2487 unsigned int mask = datagram_poll(file, sock, wait);
2488 struct sock *sk = sock->sk;
2490 if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2491 mask |= POLLIN | POLLRDNORM;
2493 /* Check for false positives due to checksum errors */
2494 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2495 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2496 mask &= ~(POLLIN | POLLRDNORM);
2501 EXPORT_SYMBOL(udp_poll);
2503 int udp_abort(struct sock *sk, int err)
2508 sk->sk_error_report(sk);
2509 __udp_disconnect(sk, 0);
2515 EXPORT_SYMBOL_GPL(udp_abort);
2517 struct proto udp_prot = {
2519 .owner = THIS_MODULE,
2520 .close = udp_lib_close,
2521 .connect = ip4_datagram_connect,
2522 .disconnect = udp_disconnect,
2524 .init = udp_init_sock,
2525 .destroy = udp_destroy_sock,
2526 .setsockopt = udp_setsockopt,
2527 .getsockopt = udp_getsockopt,
2528 .sendmsg = udp_sendmsg,
2529 .recvmsg = udp_recvmsg,
2530 .sendpage = udp_sendpage,
2531 .release_cb = ip4_datagram_release_cb,
2532 .hash = udp_lib_hash,
2533 .unhash = udp_lib_unhash,
2534 .rehash = udp_v4_rehash,
2535 .get_port = udp_v4_get_port,
2536 .memory_allocated = &udp_memory_allocated,
2537 .sysctl_mem = sysctl_udp_mem,
2538 .sysctl_wmem = &sysctl_udp_wmem_min,
2539 .sysctl_rmem = &sysctl_udp_rmem_min,
2540 .obj_size = sizeof(struct udp_sock),
2541 .h.udp_table = &udp_table,
2542 #ifdef CONFIG_COMPAT
2543 .compat_setsockopt = compat_udp_setsockopt,
2544 .compat_getsockopt = compat_udp_getsockopt,
2546 .diag_destroy = udp_abort,
2548 EXPORT_SYMBOL(udp_prot);
2550 /* ------------------------------------------------------------------------ */
2551 #ifdef CONFIG_PROC_FS
2553 static struct sock *udp_get_first(struct seq_file *seq, int start)
2556 struct udp_iter_state *state = seq->private;
2557 struct net *net = seq_file_net(seq);
2559 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2561 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2563 if (hlist_empty(&hslot->head))
2566 spin_lock_bh(&hslot->lock);
2567 sk_for_each(sk, &hslot->head) {
2568 if (!net_eq(sock_net(sk), net))
2570 if (sk->sk_family == state->family)
2573 spin_unlock_bh(&hslot->lock);
2580 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2582 struct udp_iter_state *state = seq->private;
2583 struct net *net = seq_file_net(seq);
2587 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2590 if (state->bucket <= state->udp_table->mask)
2591 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2592 return udp_get_first(seq, state->bucket + 1);
2597 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2599 struct sock *sk = udp_get_first(seq, 0);
2602 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2604 return pos ? NULL : sk;
2607 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2609 struct udp_iter_state *state = seq->private;
2610 state->bucket = MAX_UDP_PORTS;
2612 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2615 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2619 if (v == SEQ_START_TOKEN)
2620 sk = udp_get_idx(seq, 0);
2622 sk = udp_get_next(seq, v);
2628 static void udp_seq_stop(struct seq_file *seq, void *v)
2630 struct udp_iter_state *state = seq->private;
2632 if (state->bucket <= state->udp_table->mask)
2633 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2636 int udp_seq_open(struct inode *inode, struct file *file)
2638 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2639 struct udp_iter_state *s;
2642 err = seq_open_net(inode, file, &afinfo->seq_ops,
2643 sizeof(struct udp_iter_state));
2647 s = ((struct seq_file *)file->private_data)->private;
2648 s->family = afinfo->family;
2649 s->udp_table = afinfo->udp_table;
2652 EXPORT_SYMBOL(udp_seq_open);
2654 /* ------------------------------------------------------------------------ */
2655 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2657 struct proc_dir_entry *p;
2660 afinfo->seq_ops.start = udp_seq_start;
2661 afinfo->seq_ops.next = udp_seq_next;
2662 afinfo->seq_ops.stop = udp_seq_stop;
2664 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2665 afinfo->seq_fops, afinfo);
2670 EXPORT_SYMBOL(udp_proc_register);
2672 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2674 remove_proc_entry(afinfo->name, net->proc_net);
2676 EXPORT_SYMBOL(udp_proc_unregister);
2678 /* ------------------------------------------------------------------------ */
2679 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2682 struct inet_sock *inet = inet_sk(sp);
2683 __be32 dest = inet->inet_daddr;
2684 __be32 src = inet->inet_rcv_saddr;
2685 __u16 destp = ntohs(inet->inet_dport);
2686 __u16 srcp = ntohs(inet->inet_sport);
2688 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2689 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2690 bucket, src, srcp, dest, destp, sp->sk_state,
2691 sk_wmem_alloc_get(sp),
2692 sk_rmem_alloc_get(sp),
2694 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2696 refcount_read(&sp->sk_refcnt), sp,
2697 atomic_read(&sp->sk_drops));
2700 int udp4_seq_show(struct seq_file *seq, void *v)
2702 seq_setwidth(seq, 127);
2703 if (v == SEQ_START_TOKEN)
2704 seq_puts(seq, " sl local_address rem_address st tx_queue "
2705 "rx_queue tr tm->when retrnsmt uid timeout "
2706 "inode ref pointer drops");
2708 struct udp_iter_state *state = seq->private;
2710 udp4_format_sock(v, seq, state->bucket);
2716 static const struct file_operations udp_afinfo_seq_fops = {
2717 .owner = THIS_MODULE,
2718 .open = udp_seq_open,
2720 .llseek = seq_lseek,
2721 .release = seq_release_net
2724 /* ------------------------------------------------------------------------ */
2725 static struct udp_seq_afinfo udp4_seq_afinfo = {
2728 .udp_table = &udp_table,
2729 .seq_fops = &udp_afinfo_seq_fops,
2731 .show = udp4_seq_show,
2735 static int __net_init udp4_proc_init_net(struct net *net)
2737 return udp_proc_register(net, &udp4_seq_afinfo);
2740 static void __net_exit udp4_proc_exit_net(struct net *net)
2742 udp_proc_unregister(net, &udp4_seq_afinfo);
2745 static struct pernet_operations udp4_net_ops = {
2746 .init = udp4_proc_init_net,
2747 .exit = udp4_proc_exit_net,
2750 int __init udp4_proc_init(void)
2752 return register_pernet_subsys(&udp4_net_ops);
2755 void udp4_proc_exit(void)
2757 unregister_pernet_subsys(&udp4_net_ops);
2759 #endif /* CONFIG_PROC_FS */
2761 static __initdata unsigned long uhash_entries;
2762 static int __init set_uhash_entries(char *str)
2769 ret = kstrtoul(str, 0, &uhash_entries);
2773 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2774 uhash_entries = UDP_HTABLE_SIZE_MIN;
2777 __setup("uhash_entries=", set_uhash_entries);
2779 void __init udp_table_init(struct udp_table *table, const char *name)
2783 table->hash = alloc_large_system_hash(name,
2784 2 * sizeof(struct udp_hslot),
2786 21, /* one slot per 2 MB */
2790 UDP_HTABLE_SIZE_MIN,
2793 table->hash2 = table->hash + (table->mask + 1);
2794 for (i = 0; i <= table->mask; i++) {
2795 INIT_HLIST_HEAD(&table->hash[i].head);
2796 table->hash[i].count = 0;
2797 spin_lock_init(&table->hash[i].lock);
2799 for (i = 0; i <= table->mask; i++) {
2800 INIT_HLIST_HEAD(&table->hash2[i].head);
2801 table->hash2[i].count = 0;
2802 spin_lock_init(&table->hash2[i].lock);
2806 u32 udp_flow_hashrnd(void)
2808 static u32 hashrnd __read_mostly;
2810 net_get_random_once(&hashrnd, sizeof(hashrnd));
2814 EXPORT_SYMBOL(udp_flow_hashrnd);
2816 void __init udp_init(void)
2818 unsigned long limit;
2821 udp_table_init(&udp_table, "UDP");
2822 limit = nr_free_buffer_pages() / 8;
2823 limit = max(limit, 128UL);
2824 sysctl_udp_mem[0] = limit / 4 * 3;
2825 sysctl_udp_mem[1] = limit;
2826 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2828 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2829 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2831 /* 16 spinlocks per cpu */
2832 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2833 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2836 panic("UDP: failed to alloc udp_busylocks\n");
2837 for (i = 0; i < (1U << udp_busylocks_log); i++)
2838 spin_lock_init(udp_busylocks + i);