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 /* Initial allocation may have already happened via setsockopt */
235 if (!rcu_access_pointer(sk->sk_reuseport_cb))
236 return reuseport_alloc(sk);
241 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
243 * @sk: socket struct in question
244 * @snum: port number to look up
245 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
248 int udp_lib_get_port(struct sock *sk, unsigned short snum,
249 unsigned int hash2_nulladdr)
251 struct udp_hslot *hslot, *hslot2;
252 struct udp_table *udptable = sk->sk_prot->h.udp_table;
254 struct net *net = sock_net(sk);
257 int low, high, remaining;
259 unsigned short first, last;
260 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
262 inet_get_local_port_range(net, &low, &high);
263 remaining = (high - low) + 1;
265 rand = prandom_u32();
266 first = reciprocal_scale(rand, remaining) + low;
268 * force rand to be an odd multiple of UDP_HTABLE_SIZE
270 rand = (rand | 1) * (udptable->mask + 1);
271 last = first + udptable->mask + 1;
273 hslot = udp_hashslot(udptable, net, first);
274 bitmap_zero(bitmap, PORTS_PER_CHAIN);
275 spin_lock_bh(&hslot->lock);
276 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
281 * Iterate on all possible values of snum for this hash.
282 * Using steps of an odd multiple of UDP_HTABLE_SIZE
283 * give us randomization and full range coverage.
286 if (low <= snum && snum <= high &&
287 !test_bit(snum >> udptable->log, bitmap) &&
288 !inet_is_local_reserved_port(net, snum))
291 } while (snum != first);
292 spin_unlock_bh(&hslot->lock);
294 } while (++first != last);
297 hslot = udp_hashslot(udptable, net, snum);
298 spin_lock_bh(&hslot->lock);
299 if (hslot->count > 10) {
301 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
303 slot2 &= udptable->mask;
304 hash2_nulladdr &= udptable->mask;
306 hslot2 = udp_hashslot2(udptable, slot2);
307 if (hslot->count < hslot2->count)
308 goto scan_primary_hash;
310 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
311 if (!exist && (hash2_nulladdr != slot2)) {
312 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
313 exist = udp_lib_lport_inuse2(net, snum, hslot2,
322 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
326 inet_sk(sk)->inet_num = snum;
327 udp_sk(sk)->udp_port_hash = snum;
328 udp_sk(sk)->udp_portaddr_hash ^= snum;
329 if (sk_unhashed(sk)) {
330 if (sk->sk_reuseport &&
331 udp_reuseport_add_sock(sk, hslot)) {
332 inet_sk(sk)->inet_num = 0;
333 udp_sk(sk)->udp_port_hash = 0;
334 udp_sk(sk)->udp_portaddr_hash ^= snum;
338 sk_add_node_rcu(sk, &hslot->head);
340 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
342 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
343 spin_lock(&hslot2->lock);
344 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
345 sk->sk_family == AF_INET6)
346 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
349 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
352 spin_unlock(&hslot2->lock);
354 sock_set_flag(sk, SOCK_RCU_FREE);
357 spin_unlock_bh(&hslot->lock);
361 EXPORT_SYMBOL(udp_lib_get_port);
363 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
366 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
369 int udp_v4_get_port(struct sock *sk, unsigned short snum)
371 unsigned int hash2_nulladdr =
372 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
373 unsigned int hash2_partial =
374 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
376 /* precompute partial secondary hash */
377 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
378 return udp_lib_get_port(sk, snum, hash2_nulladdr);
381 static int compute_score(struct sock *sk, struct net *net,
382 __be32 saddr, __be16 sport,
383 __be32 daddr, unsigned short hnum,
384 int dif, int sdif, bool exact_dif)
387 struct inet_sock *inet;
389 if (!net_eq(sock_net(sk), net) ||
390 udp_sk(sk)->udp_port_hash != hnum ||
394 score = (sk->sk_family == PF_INET) ? 2 : 1;
397 if (inet->inet_rcv_saddr) {
398 if (inet->inet_rcv_saddr != daddr)
403 if (inet->inet_daddr) {
404 if (inet->inet_daddr != saddr)
409 if (inet->inet_dport) {
410 if (inet->inet_dport != sport)
415 if (sk->sk_bound_dev_if || exact_dif) {
416 bool dev_match = (sk->sk_bound_dev_if == dif ||
417 sk->sk_bound_dev_if == sdif);
419 if (exact_dif && !dev_match)
421 if (sk->sk_bound_dev_if && dev_match)
425 if (sk->sk_incoming_cpu == raw_smp_processor_id())
430 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
431 const __u16 lport, const __be32 faddr,
434 static u32 udp_ehash_secret __read_mostly;
436 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
438 return __inet_ehashfn(laddr, lport, faddr, fport,
439 udp_ehash_secret + net_hash_mix(net));
442 /* called with rcu_read_lock() */
443 static struct sock *udp4_lib_lookup2(struct net *net,
444 __be32 saddr, __be16 sport,
445 __be32 daddr, unsigned int hnum,
446 int dif, int sdif, bool exact_dif,
447 struct udp_hslot *hslot2,
450 struct sock *sk, *result;
451 int score, badness, matches = 0, reuseport = 0;
456 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
457 score = compute_score(sk, net, saddr, sport,
458 daddr, hnum, dif, sdif, exact_dif);
459 if (score > badness) {
460 reuseport = sk->sk_reuseport;
462 hash = udp_ehashfn(net, daddr, hnum,
464 result = reuseport_select_sock(sk, hash, skb,
465 sizeof(struct udphdr));
472 } else if (score == badness && reuseport) {
474 if (reciprocal_scale(hash, matches) == 0)
476 hash = next_pseudo_random32(hash);
482 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
483 * harder than this. -DaveM
485 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
486 __be16 sport, __be32 daddr, __be16 dport, int dif,
487 int sdif, struct udp_table *udptable, struct sk_buff *skb)
489 struct sock *sk, *result;
490 unsigned short hnum = ntohs(dport);
491 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
492 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
493 bool exact_dif = udp_lib_exact_dif_match(net, skb);
494 int score, badness, matches = 0, reuseport = 0;
497 if (hslot->count > 10) {
498 hash2 = udp4_portaddr_hash(net, daddr, hnum);
499 slot2 = hash2 & udptable->mask;
500 hslot2 = &udptable->hash2[slot2];
501 if (hslot->count < hslot2->count)
504 result = udp4_lib_lookup2(net, saddr, sport,
505 daddr, hnum, dif, sdif,
506 exact_dif, hslot2, skb);
508 unsigned int old_slot2 = slot2;
509 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
510 slot2 = hash2 & udptable->mask;
511 /* avoid searching the same slot again. */
512 if (unlikely(slot2 == old_slot2))
515 hslot2 = &udptable->hash2[slot2];
516 if (hslot->count < hslot2->count)
519 result = udp4_lib_lookup2(net, saddr, sport,
520 daddr, hnum, dif, sdif,
521 exact_dif, hslot2, skb);
528 sk_for_each_rcu(sk, &hslot->head) {
529 score = compute_score(sk, net, saddr, sport,
530 daddr, hnum, dif, sdif, exact_dif);
531 if (score > badness) {
532 reuseport = sk->sk_reuseport;
534 hash = udp_ehashfn(net, daddr, hnum,
536 result = reuseport_select_sock(sk, hash, skb,
537 sizeof(struct udphdr));
544 } else if (score == badness && reuseport) {
546 if (reciprocal_scale(hash, matches) == 0)
548 hash = next_pseudo_random32(hash);
553 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
555 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
556 __be16 sport, __be16 dport,
557 struct udp_table *udptable)
559 const struct iphdr *iph = ip_hdr(skb);
561 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
562 iph->daddr, dport, inet_iif(skb),
563 inet_sdif(skb), udptable, skb);
566 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
567 __be16 sport, __be16 dport)
569 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
571 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
573 /* Must be called under rcu_read_lock().
574 * Does increment socket refcount.
576 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
577 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
578 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
579 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
580 __be32 daddr, __be16 dport, int dif)
584 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
585 dif, 0, &udp_table, NULL);
586 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
590 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
593 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
594 __be16 loc_port, __be32 loc_addr,
595 __be16 rmt_port, __be32 rmt_addr,
596 int dif, int sdif, unsigned short hnum)
598 struct inet_sock *inet = inet_sk(sk);
600 if (!net_eq(sock_net(sk), net) ||
601 udp_sk(sk)->udp_port_hash != hnum ||
602 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
603 (inet->inet_dport != rmt_port && inet->inet_dport) ||
604 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
605 ipv6_only_sock(sk) ||
606 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
607 sk->sk_bound_dev_if != sdif))
609 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
615 * This routine is called by the ICMP module when it gets some
616 * sort of error condition. If err < 0 then the socket should
617 * be closed and the error returned to the user. If err > 0
618 * it's just the icmp type << 8 | icmp code.
619 * Header points to the ip header of the error packet. We move
620 * on past this. Then (as it used to claim before adjustment)
621 * header points to the first 8 bytes of the udp header. We need
622 * to find the appropriate port.
625 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
627 struct inet_sock *inet;
628 const struct iphdr *iph = (const struct iphdr *)skb->data;
629 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
630 const int type = icmp_hdr(skb)->type;
631 const int code = icmp_hdr(skb)->code;
635 struct net *net = dev_net(skb->dev);
637 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
638 iph->saddr, uh->source, skb->dev->ifindex, 0,
641 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
642 return; /* No socket for error */
651 case ICMP_TIME_EXCEEDED:
654 case ICMP_SOURCE_QUENCH:
656 case ICMP_PARAMETERPROB:
660 case ICMP_DEST_UNREACH:
661 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
662 ipv4_sk_update_pmtu(skb, sk, info);
663 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
671 if (code <= NR_ICMP_UNREACH) {
672 harderr = icmp_err_convert[code].fatal;
673 err = icmp_err_convert[code].errno;
677 ipv4_sk_redirect(skb, sk);
682 * RFC1122: OK. Passes ICMP errors back to application, as per
685 if (!inet->recverr) {
686 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
689 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
692 sk->sk_error_report(sk);
697 void udp_err(struct sk_buff *skb, u32 info)
699 __udp4_lib_err(skb, info, &udp_table);
703 * Throw away all pending data and cancel the corking. Socket is locked.
705 void udp_flush_pending_frames(struct sock *sk)
707 struct udp_sock *up = udp_sk(sk);
712 ip_flush_pending_frames(sk);
715 EXPORT_SYMBOL(udp_flush_pending_frames);
718 * udp4_hwcsum - handle outgoing HW checksumming
719 * @skb: sk_buff containing the filled-in UDP header
720 * (checksum field must be zeroed out)
721 * @src: source IP address
722 * @dst: destination IP address
724 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
726 struct udphdr *uh = udp_hdr(skb);
727 int offset = skb_transport_offset(skb);
728 int len = skb->len - offset;
732 if (!skb_has_frag_list(skb)) {
734 * Only one fragment on the socket.
736 skb->csum_start = skb_transport_header(skb) - skb->head;
737 skb->csum_offset = offsetof(struct udphdr, check);
738 uh->check = ~csum_tcpudp_magic(src, dst, len,
741 struct sk_buff *frags;
744 * HW-checksum won't work as there are two or more
745 * fragments on the socket so that all csums of sk_buffs
748 skb_walk_frags(skb, frags) {
749 csum = csum_add(csum, frags->csum);
753 csum = skb_checksum(skb, offset, hlen, csum);
754 skb->ip_summed = CHECKSUM_NONE;
756 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
758 uh->check = CSUM_MANGLED_0;
761 EXPORT_SYMBOL_GPL(udp4_hwcsum);
763 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
764 * for the simple case like when setting the checksum for a UDP tunnel.
766 void udp_set_csum(bool nocheck, struct sk_buff *skb,
767 __be32 saddr, __be32 daddr, int len)
769 struct udphdr *uh = udp_hdr(skb);
773 } else if (skb_is_gso(skb)) {
774 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
775 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
777 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
779 uh->check = CSUM_MANGLED_0;
781 skb->ip_summed = CHECKSUM_PARTIAL;
782 skb->csum_start = skb_transport_header(skb) - skb->head;
783 skb->csum_offset = offsetof(struct udphdr, check);
784 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
787 EXPORT_SYMBOL(udp_set_csum);
789 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
791 struct sock *sk = skb->sk;
792 struct inet_sock *inet = inet_sk(sk);
795 int is_udplite = IS_UDPLITE(sk);
796 int offset = skb_transport_offset(skb);
797 int len = skb->len - offset;
801 * Create a UDP header
804 uh->source = inet->inet_sport;
805 uh->dest = fl4->fl4_dport;
806 uh->len = htons(len);
809 if (is_udplite) /* UDP-Lite */
810 csum = udplite_csum(skb);
812 else if (sk->sk_no_check_tx && !skb_is_gso(skb)) { /* UDP csum off */
814 skb->ip_summed = CHECKSUM_NONE;
817 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
819 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
823 csum = udp_csum(skb);
825 /* add protocol-dependent pseudo-header */
826 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
827 sk->sk_protocol, csum);
829 uh->check = CSUM_MANGLED_0;
832 err = ip_send_skb(sock_net(sk), skb);
834 if (err == -ENOBUFS && !inet->recverr) {
835 UDP_INC_STATS(sock_net(sk),
836 UDP_MIB_SNDBUFERRORS, is_udplite);
840 UDP_INC_STATS(sock_net(sk),
841 UDP_MIB_OUTDATAGRAMS, is_udplite);
846 * Push out all pending data as one UDP datagram. Socket is locked.
848 int udp_push_pending_frames(struct sock *sk)
850 struct udp_sock *up = udp_sk(sk);
851 struct inet_sock *inet = inet_sk(sk);
852 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
856 skb = ip_finish_skb(sk, fl4);
860 err = udp_send_skb(skb, fl4);
867 EXPORT_SYMBOL(udp_push_pending_frames);
869 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
871 struct inet_sock *inet = inet_sk(sk);
872 struct udp_sock *up = udp_sk(sk);
873 struct flowi4 fl4_stack;
876 struct ipcm_cookie ipc;
877 struct rtable *rt = NULL;
880 __be32 daddr, faddr, saddr;
883 int err, is_udplite = IS_UDPLITE(sk);
884 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
885 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
887 struct ip_options_data opt_copy;
896 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
904 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
906 fl4 = &inet->cork.fl.u.ip4;
909 * There are pending frames.
910 * The socket lock must be held while it's corked.
913 if (likely(up->pending)) {
914 if (unlikely(up->pending != AF_INET)) {
922 ulen += sizeof(struct udphdr);
925 * Get and verify the address.
928 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
929 if (msg->msg_namelen < sizeof(*usin))
931 if (usin->sin_family != AF_INET) {
932 if (usin->sin_family != AF_UNSPEC)
933 return -EAFNOSUPPORT;
936 daddr = usin->sin_addr.s_addr;
937 dport = usin->sin_port;
941 if (sk->sk_state != TCP_ESTABLISHED)
942 return -EDESTADDRREQ;
943 daddr = inet->inet_daddr;
944 dport = inet->inet_dport;
945 /* Open fast path for connected socket.
946 Route will not be used, if at least one option is set.
951 ipc.sockc.tsflags = sk->sk_tsflags;
952 ipc.addr = inet->inet_saddr;
953 ipc.oif = sk->sk_bound_dev_if;
955 if (msg->msg_controllen) {
956 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
966 struct ip_options_rcu *inet_opt;
969 inet_opt = rcu_dereference(inet->inet_opt);
971 memcpy(&opt_copy, inet_opt,
972 sizeof(*inet_opt) + inet_opt->opt.optlen);
973 ipc.opt = &opt_copy.opt;
979 ipc.addr = faddr = daddr;
981 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
983 if (ipc.opt && ipc.opt->opt.srr) {
986 faddr = ipc.opt->opt.faddr;
989 tos = get_rttos(&ipc, inet);
990 if (sock_flag(sk, SOCK_LOCALROUTE) ||
991 (msg->msg_flags & MSG_DONTROUTE) ||
992 (ipc.opt && ipc.opt->opt.is_strictroute)) {
997 if (ipv4_is_multicast(daddr)) {
999 ipc.oif = inet->mc_index;
1001 saddr = inet->mc_addr;
1003 } else if (!ipc.oif)
1004 ipc.oif = inet->uc_index;
1007 rt = (struct rtable *)sk_dst_check(sk, 0);
1010 struct net *net = sock_net(sk);
1011 __u8 flow_flags = inet_sk_flowi_flags(sk);
1015 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1016 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1018 faddr, saddr, dport, inet->inet_sport,
1021 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1022 rt = ip_route_output_flow(net, fl4, sk);
1026 if (err == -ENETUNREACH)
1027 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1032 if ((rt->rt_flags & RTCF_BROADCAST) &&
1033 !sock_flag(sk, SOCK_BROADCAST))
1036 sk_dst_set(sk, dst_clone(&rt->dst));
1039 if (msg->msg_flags&MSG_CONFIRM)
1045 daddr = ipc.addr = fl4->daddr;
1047 /* Lockless fast path for the non-corking case. */
1049 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1050 sizeof(struct udphdr), &ipc, &rt,
1053 if (!IS_ERR_OR_NULL(skb))
1054 err = udp_send_skb(skb, fl4);
1059 if (unlikely(up->pending)) {
1060 /* The socket is already corked while preparing it. */
1061 /* ... which is an evident application bug. --ANK */
1064 net_dbg_ratelimited("cork app bug 2\n");
1069 * Now cork the socket to pend data.
1071 fl4 = &inet->cork.fl.u.ip4;
1074 fl4->fl4_dport = dport;
1075 fl4->fl4_sport = inet->inet_sport;
1076 up->pending = AF_INET;
1080 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1081 sizeof(struct udphdr), &ipc, &rt,
1082 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1084 udp_flush_pending_frames(sk);
1086 err = udp_push_pending_frames(sk);
1087 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1098 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1099 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1100 * we don't have a good statistic (IpOutDiscards but it can be too many
1101 * things). We could add another new stat but at least for now that
1102 * seems like overkill.
1104 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1105 UDP_INC_STATS(sock_net(sk),
1106 UDP_MIB_SNDBUFERRORS, is_udplite);
1111 if (msg->msg_flags & MSG_PROBE)
1112 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1113 if (!(msg->msg_flags&MSG_PROBE) || len)
1114 goto back_from_confirm;
1118 EXPORT_SYMBOL(udp_sendmsg);
1120 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1121 size_t size, int flags)
1123 struct inet_sock *inet = inet_sk(sk);
1124 struct udp_sock *up = udp_sk(sk);
1127 if (flags & MSG_SENDPAGE_NOTLAST)
1131 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1133 /* Call udp_sendmsg to specify destination address which
1134 * sendpage interface can't pass.
1135 * This will succeed only when the socket is connected.
1137 ret = udp_sendmsg(sk, &msg, 0);
1144 if (unlikely(!up->pending)) {
1147 net_dbg_ratelimited("udp cork app bug 3\n");
1151 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1152 page, offset, size, flags);
1153 if (ret == -EOPNOTSUPP) {
1155 return sock_no_sendpage(sk->sk_socket, page, offset,
1159 udp_flush_pending_frames(sk);
1164 if (!(up->corkflag || (flags&MSG_MORE)))
1165 ret = udp_push_pending_frames(sk);
1173 #define UDP_SKB_IS_STATELESS 0x80000000
1175 static void udp_set_dev_scratch(struct sk_buff *skb)
1177 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1179 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1180 scratch->_tsize_state = skb->truesize;
1181 #if BITS_PER_LONG == 64
1182 scratch->len = skb->len;
1183 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1184 scratch->is_linear = !skb_is_nonlinear(skb);
1186 /* all head states execept sp (dst, sk, nf) are always cleared by
1187 * udp_rcv() and we need to preserve secpath, if present, to eventually
1188 * process IP_CMSG_PASSSEC at recvmsg() time
1190 if (likely(!skb_sec_path(skb)))
1191 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1194 static int udp_skb_truesize(struct sk_buff *skb)
1196 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1199 static bool udp_skb_has_head_state(struct sk_buff *skb)
1201 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1204 /* fully reclaim rmem/fwd memory allocated for skb */
1205 static void udp_rmem_release(struct sock *sk, int size, int partial,
1206 bool rx_queue_lock_held)
1208 struct udp_sock *up = udp_sk(sk);
1209 struct sk_buff_head *sk_queue;
1212 if (likely(partial)) {
1213 up->forward_deficit += size;
1214 size = up->forward_deficit;
1215 if (size < (sk->sk_rcvbuf >> 2) &&
1216 !skb_queue_empty(&up->reader_queue))
1219 size += up->forward_deficit;
1221 up->forward_deficit = 0;
1223 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1224 * if the called don't held it already
1226 sk_queue = &sk->sk_receive_queue;
1227 if (!rx_queue_lock_held)
1228 spin_lock(&sk_queue->lock);
1231 sk->sk_forward_alloc += size;
1232 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1233 sk->sk_forward_alloc -= amt;
1236 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1238 atomic_sub(size, &sk->sk_rmem_alloc);
1240 /* this can save us from acquiring the rx queue lock on next receive */
1241 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1243 if (!rx_queue_lock_held)
1244 spin_unlock(&sk_queue->lock);
1247 /* Note: called with reader_queue.lock held.
1248 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1249 * This avoids a cache line miss while receive_queue lock is held.
1250 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1252 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1254 prefetch(&skb->data);
1255 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1257 EXPORT_SYMBOL(udp_skb_destructor);
1259 /* as above, but the caller held the rx queue lock, too */
1260 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1262 prefetch(&skb->data);
1263 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1266 /* Idea of busylocks is to let producers grab an extra spinlock
1267 * to relieve pressure on the receive_queue spinlock shared by consumer.
1268 * Under flood, this means that only one producer can be in line
1269 * trying to acquire the receive_queue spinlock.
1270 * These busylock can be allocated on a per cpu manner, instead of a
1271 * per socket one (that would consume a cache line per socket)
1273 static int udp_busylocks_log __read_mostly;
1274 static spinlock_t *udp_busylocks __read_mostly;
1276 static spinlock_t *busylock_acquire(void *ptr)
1280 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1285 static void busylock_release(spinlock_t *busy)
1291 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1293 struct sk_buff_head *list = &sk->sk_receive_queue;
1294 int rmem, delta, amt, err = -ENOMEM;
1295 spinlock_t *busy = NULL;
1298 /* try to avoid the costly atomic add/sub pair when the receive
1299 * queue is full; always allow at least a packet
1301 rmem = atomic_read(&sk->sk_rmem_alloc);
1302 if (rmem > sk->sk_rcvbuf)
1305 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1306 * having linear skbs :
1307 * - Reduce memory overhead and thus increase receive queue capacity
1308 * - Less cache line misses at copyout() time
1309 * - Less work at consume_skb() (less alien page frag freeing)
1311 if (rmem > (sk->sk_rcvbuf >> 1)) {
1314 busy = busylock_acquire(sk);
1316 size = skb->truesize;
1317 udp_set_dev_scratch(skb);
1319 /* we drop only if the receive buf is full and the receive
1320 * queue contains some other skb
1322 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1323 if (rmem > (size + sk->sk_rcvbuf))
1326 spin_lock(&list->lock);
1327 if (size >= sk->sk_forward_alloc) {
1328 amt = sk_mem_pages(size);
1329 delta = amt << SK_MEM_QUANTUM_SHIFT;
1330 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1332 spin_unlock(&list->lock);
1336 sk->sk_forward_alloc += delta;
1339 sk->sk_forward_alloc -= size;
1341 /* no need to setup a destructor, we will explicitly release the
1342 * forward allocated memory on dequeue
1344 sock_skb_set_dropcount(sk, skb);
1346 __skb_queue_tail(list, skb);
1347 spin_unlock(&list->lock);
1349 if (!sock_flag(sk, SOCK_DEAD))
1350 sk->sk_data_ready(sk);
1352 busylock_release(busy);
1356 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1359 atomic_inc(&sk->sk_drops);
1360 busylock_release(busy);
1363 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1365 void udp_destruct_sock(struct sock *sk)
1367 /* reclaim completely the forward allocated memory */
1368 struct udp_sock *up = udp_sk(sk);
1369 unsigned int total = 0;
1370 struct sk_buff *skb;
1372 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1373 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1374 total += skb->truesize;
1377 udp_rmem_release(sk, total, 0, true);
1379 inet_sock_destruct(sk);
1381 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1383 int udp_init_sock(struct sock *sk)
1385 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1386 sk->sk_destruct = udp_destruct_sock;
1389 EXPORT_SYMBOL_GPL(udp_init_sock);
1391 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1393 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1394 bool slow = lock_sock_fast(sk);
1396 sk_peek_offset_bwd(sk, len);
1397 unlock_sock_fast(sk, slow);
1400 /* In the more common cases we cleared the head states previously,
1401 * see __udp_queue_rcv_skb().
1403 if (unlikely(udp_skb_has_head_state(skb)))
1404 skb_release_head_state(skb);
1405 consume_stateless_skb(skb);
1407 EXPORT_SYMBOL_GPL(skb_consume_udp);
1409 static struct sk_buff *__first_packet_length(struct sock *sk,
1410 struct sk_buff_head *rcvq,
1413 struct sk_buff *skb;
1415 while ((skb = skb_peek(rcvq)) != NULL) {
1416 if (udp_lib_checksum_complete(skb)) {
1417 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1419 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1421 atomic_inc(&sk->sk_drops);
1422 __skb_unlink(skb, rcvq);
1423 *total += skb->truesize;
1426 /* the csum related bits could be changed, refresh
1429 udp_set_dev_scratch(skb);
1437 * first_packet_length - return length of first packet in receive queue
1440 * Drops all bad checksum frames, until a valid one is found.
1441 * Returns the length of found skb, or -1 if none is found.
1443 static int first_packet_length(struct sock *sk)
1445 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1446 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1447 struct sk_buff *skb;
1451 spin_lock_bh(&rcvq->lock);
1452 skb = __first_packet_length(sk, rcvq, &total);
1453 if (!skb && !skb_queue_empty(sk_queue)) {
1454 spin_lock(&sk_queue->lock);
1455 skb_queue_splice_tail_init(sk_queue, rcvq);
1456 spin_unlock(&sk_queue->lock);
1458 skb = __first_packet_length(sk, rcvq, &total);
1460 res = skb ? skb->len : -1;
1462 udp_rmem_release(sk, total, 1, false);
1463 spin_unlock_bh(&rcvq->lock);
1468 * IOCTL requests applicable to the UDP protocol
1471 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1476 int amount = sk_wmem_alloc_get(sk);
1478 return put_user(amount, (int __user *)arg);
1483 int amount = max_t(int, 0, first_packet_length(sk));
1485 return put_user(amount, (int __user *)arg);
1489 return -ENOIOCTLCMD;
1494 EXPORT_SYMBOL(udp_ioctl);
1496 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1497 int noblock, int *peeked, int *off, int *err)
1499 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1500 struct sk_buff_head *queue;
1501 struct sk_buff *last;
1505 queue = &udp_sk(sk)->reader_queue;
1506 flags |= noblock ? MSG_DONTWAIT : 0;
1507 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1509 struct sk_buff *skb;
1511 error = sock_error(sk);
1518 spin_lock_bh(&queue->lock);
1519 skb = __skb_try_recv_from_queue(sk, queue, flags,
1524 spin_unlock_bh(&queue->lock);
1528 if (skb_queue_empty(sk_queue)) {
1529 spin_unlock_bh(&queue->lock);
1533 /* refill the reader queue and walk it again
1534 * keep both queues locked to avoid re-acquiring
1535 * the sk_receive_queue lock if fwd memory scheduling
1538 spin_lock(&sk_queue->lock);
1539 skb_queue_splice_tail_init(sk_queue, queue);
1541 skb = __skb_try_recv_from_queue(sk, queue, flags,
1542 udp_skb_dtor_locked,
1545 spin_unlock(&sk_queue->lock);
1546 spin_unlock_bh(&queue->lock);
1551 if (!sk_can_busy_loop(sk))
1554 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1555 } while (!skb_queue_empty(sk_queue));
1557 /* sk_queue is empty, reader_queue may contain peeked packets */
1559 !__skb_wait_for_more_packets(sk, &error, &timeo,
1560 (struct sk_buff *)sk_queue));
1565 EXPORT_SYMBOL_GPL(__skb_recv_udp);
1568 * This should be easy, if there is something there we
1569 * return it, otherwise we block.
1572 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1573 int flags, int *addr_len)
1575 struct inet_sock *inet = inet_sk(sk);
1576 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1577 struct sk_buff *skb;
1578 unsigned int ulen, copied;
1579 int peeked, peeking, off;
1581 int is_udplite = IS_UDPLITE(sk);
1582 bool checksum_valid = false;
1584 if (flags & MSG_ERRQUEUE)
1585 return ip_recv_error(sk, msg, len, addr_len);
1588 peeking = off = sk_peek_offset(sk, flags);
1589 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1593 ulen = udp_skb_len(skb);
1595 if (copied > ulen - off)
1596 copied = ulen - off;
1597 else if (copied < ulen)
1598 msg->msg_flags |= MSG_TRUNC;
1601 * If checksum is needed at all, try to do it while copying the
1602 * data. If the data is truncated, or if we only want a partial
1603 * coverage checksum (UDP-Lite), do it before the copy.
1606 if (copied < ulen || peeking ||
1607 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1608 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1609 !__udp_lib_checksum_complete(skb);
1610 if (!checksum_valid)
1614 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1615 if (udp_skb_is_linear(skb))
1616 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1618 err = skb_copy_datagram_msg(skb, off, msg, copied);
1620 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1626 if (unlikely(err)) {
1628 atomic_inc(&sk->sk_drops);
1629 UDP_INC_STATS(sock_net(sk),
1630 UDP_MIB_INERRORS, is_udplite);
1637 UDP_INC_STATS(sock_net(sk),
1638 UDP_MIB_INDATAGRAMS, is_udplite);
1640 sock_recv_ts_and_drops(msg, sk, skb);
1642 /* Copy the address. */
1644 sin->sin_family = AF_INET;
1645 sin->sin_port = udp_hdr(skb)->source;
1646 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1647 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1648 *addr_len = sizeof(*sin);
1650 if (inet->cmsg_flags)
1651 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1654 if (flags & MSG_TRUNC)
1657 skb_consume_udp(sk, skb, peeking ? -err : err);
1661 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1662 udp_skb_destructor)) {
1663 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1664 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1668 /* starting over for a new packet, but check if we need to yield */
1670 msg->msg_flags &= ~MSG_TRUNC;
1674 int __udp_disconnect(struct sock *sk, int flags)
1676 struct inet_sock *inet = inet_sk(sk);
1678 * 1003.1g - break association.
1681 sk->sk_state = TCP_CLOSE;
1682 inet->inet_daddr = 0;
1683 inet->inet_dport = 0;
1684 sock_rps_reset_rxhash(sk);
1685 sk->sk_bound_dev_if = 0;
1686 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1687 inet_reset_saddr(sk);
1689 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1690 sk->sk_prot->unhash(sk);
1691 inet->inet_sport = 0;
1696 EXPORT_SYMBOL(__udp_disconnect);
1698 int udp_disconnect(struct sock *sk, int flags)
1701 __udp_disconnect(sk, flags);
1705 EXPORT_SYMBOL(udp_disconnect);
1707 void udp_lib_unhash(struct sock *sk)
1709 if (sk_hashed(sk)) {
1710 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1711 struct udp_hslot *hslot, *hslot2;
1713 hslot = udp_hashslot(udptable, sock_net(sk),
1714 udp_sk(sk)->udp_port_hash);
1715 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1717 spin_lock_bh(&hslot->lock);
1718 if (rcu_access_pointer(sk->sk_reuseport_cb))
1719 reuseport_detach_sock(sk);
1720 if (sk_del_node_init_rcu(sk)) {
1722 inet_sk(sk)->inet_num = 0;
1723 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1725 spin_lock(&hslot2->lock);
1726 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1728 spin_unlock(&hslot2->lock);
1730 spin_unlock_bh(&hslot->lock);
1733 EXPORT_SYMBOL(udp_lib_unhash);
1736 * inet_rcv_saddr was changed, we must rehash secondary hash
1738 void udp_lib_rehash(struct sock *sk, u16 newhash)
1740 if (sk_hashed(sk)) {
1741 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1742 struct udp_hslot *hslot, *hslot2, *nhslot2;
1744 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1745 nhslot2 = udp_hashslot2(udptable, newhash);
1746 udp_sk(sk)->udp_portaddr_hash = newhash;
1748 if (hslot2 != nhslot2 ||
1749 rcu_access_pointer(sk->sk_reuseport_cb)) {
1750 hslot = udp_hashslot(udptable, sock_net(sk),
1751 udp_sk(sk)->udp_port_hash);
1752 /* we must lock primary chain too */
1753 spin_lock_bh(&hslot->lock);
1754 if (rcu_access_pointer(sk->sk_reuseport_cb))
1755 reuseport_detach_sock(sk);
1757 if (hslot2 != nhslot2) {
1758 spin_lock(&hslot2->lock);
1759 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1761 spin_unlock(&hslot2->lock);
1763 spin_lock(&nhslot2->lock);
1764 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1767 spin_unlock(&nhslot2->lock);
1770 spin_unlock_bh(&hslot->lock);
1774 EXPORT_SYMBOL(udp_lib_rehash);
1776 static void udp_v4_rehash(struct sock *sk)
1778 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1779 inet_sk(sk)->inet_rcv_saddr,
1780 inet_sk(sk)->inet_num);
1781 udp_lib_rehash(sk, new_hash);
1784 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1788 if (inet_sk(sk)->inet_daddr) {
1789 sock_rps_save_rxhash(sk, skb);
1790 sk_mark_napi_id(sk, skb);
1791 sk_incoming_cpu_update(sk);
1793 sk_mark_napi_id_once(sk, skb);
1796 rc = __udp_enqueue_schedule_skb(sk, skb);
1798 int is_udplite = IS_UDPLITE(sk);
1800 /* Note that an ENOMEM error is charged twice */
1802 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1804 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1806 trace_udp_fail_queue_rcv_skb(rc, sk);
1813 static struct static_key udp_encap_needed __read_mostly;
1814 void udp_encap_enable(void)
1816 if (!static_key_enabled(&udp_encap_needed))
1817 static_key_slow_inc(&udp_encap_needed);
1819 EXPORT_SYMBOL(udp_encap_enable);
1824 * >0: "udp encap" protocol resubmission
1826 * Note that in the success and error cases, the skb is assumed to
1827 * have either been requeued or freed.
1829 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1831 struct udp_sock *up = udp_sk(sk);
1832 int is_udplite = IS_UDPLITE(sk);
1835 * Charge it to the socket, dropping if the queue is full.
1837 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1841 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1842 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1845 * This is an encapsulation socket so pass the skb to
1846 * the socket's udp_encap_rcv() hook. Otherwise, just
1847 * fall through and pass this up the UDP socket.
1848 * up->encap_rcv() returns the following value:
1849 * =0 if skb was successfully passed to the encap
1850 * handler or was discarded by it.
1851 * >0 if skb should be passed on to UDP.
1852 * <0 if skb should be resubmitted as proto -N
1855 /* if we're overly short, let UDP handle it */
1856 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1860 /* Verify checksum before giving to encap */
1861 if (udp_lib_checksum_complete(skb))
1864 ret = encap_rcv(sk, skb);
1866 __UDP_INC_STATS(sock_net(sk),
1867 UDP_MIB_INDATAGRAMS,
1873 /* FALLTHROUGH -- it's a UDP Packet */
1877 * UDP-Lite specific tests, ignored on UDP sockets
1879 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1882 * MIB statistics other than incrementing the error count are
1883 * disabled for the following two types of errors: these depend
1884 * on the application settings, not on the functioning of the
1885 * protocol stack as such.
1887 * RFC 3828 here recommends (sec 3.3): "There should also be a
1888 * way ... to ... at least let the receiving application block
1889 * delivery of packets with coverage values less than a value
1890 * provided by the application."
1892 if (up->pcrlen == 0) { /* full coverage was set */
1893 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1894 UDP_SKB_CB(skb)->cscov, skb->len);
1897 /* The next case involves violating the min. coverage requested
1898 * by the receiver. This is subtle: if receiver wants x and x is
1899 * greater than the buffersize/MTU then receiver will complain
1900 * that it wants x while sender emits packets of smaller size y.
1901 * Therefore the above ...()->partial_cov statement is essential.
1903 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1904 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1905 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1910 prefetch(&sk->sk_rmem_alloc);
1911 if (rcu_access_pointer(sk->sk_filter) &&
1912 udp_lib_checksum_complete(skb))
1915 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1918 udp_csum_pull_header(skb);
1920 ipv4_pktinfo_prepare(sk, skb);
1921 return __udp_queue_rcv_skb(sk, skb);
1924 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1926 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1927 atomic_inc(&sk->sk_drops);
1932 /* For TCP sockets, sk_rx_dst is protected by socket lock
1933 * For UDP, we use xchg() to guard against concurrent changes.
1935 void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1937 struct dst_entry *old;
1939 if (dst_hold_safe(dst)) {
1940 old = xchg(&sk->sk_rx_dst, dst);
1944 EXPORT_SYMBOL(udp_sk_rx_dst_set);
1947 * Multicasts and broadcasts go to each listener.
1949 * Note: called only from the BH handler context.
1951 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1953 __be32 saddr, __be32 daddr,
1954 struct udp_table *udptable,
1957 struct sock *sk, *first = NULL;
1958 unsigned short hnum = ntohs(uh->dest);
1959 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1960 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1961 unsigned int offset = offsetof(typeof(*sk), sk_node);
1962 int dif = skb->dev->ifindex;
1963 int sdif = inet_sdif(skb);
1964 struct hlist_node *node;
1965 struct sk_buff *nskb;
1968 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1970 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1972 hslot = &udptable->hash2[hash2];
1973 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1976 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1977 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1978 uh->source, saddr, dif, sdif, hnum))
1985 nskb = skb_clone(skb, GFP_ATOMIC);
1987 if (unlikely(!nskb)) {
1988 atomic_inc(&sk->sk_drops);
1989 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1991 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1995 if (udp_queue_rcv_skb(sk, nskb) > 0)
1999 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2000 if (use_hash2 && hash2 != hash2_any) {
2006 if (udp_queue_rcv_skb(first, skb) > 0)
2010 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2011 proto == IPPROTO_UDPLITE);
2016 /* Initialize UDP checksum. If exited with zero value (success),
2017 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2018 * Otherwise, csum completion requires chacksumming packet body,
2019 * including udp header and folding it to skb->csum.
2021 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2026 UDP_SKB_CB(skb)->partial_cov = 0;
2027 UDP_SKB_CB(skb)->cscov = skb->len;
2029 if (proto == IPPROTO_UDPLITE) {
2030 err = udplite_checksum_init(skb, uh);
2035 /* Note, we are only interested in != 0 or == 0, thus the
2038 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2039 inet_compute_pseudo);
2043 * All we need to do is get the socket, and then do a checksum.
2046 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2051 unsigned short ulen;
2052 struct rtable *rt = skb_rtable(skb);
2053 __be32 saddr, daddr;
2054 struct net *net = dev_net(skb->dev);
2057 * Validate the packet.
2059 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2060 goto drop; /* No space for header. */
2063 ulen = ntohs(uh->len);
2064 saddr = ip_hdr(skb)->saddr;
2065 daddr = ip_hdr(skb)->daddr;
2067 if (ulen > skb->len)
2070 if (proto == IPPROTO_UDP) {
2071 /* UDP validates ulen. */
2072 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2077 if (udp4_csum_init(skb, uh, proto))
2080 sk = skb_steal_sock(skb);
2082 struct dst_entry *dst = skb_dst(skb);
2085 if (unlikely(sk->sk_rx_dst != dst))
2086 udp_sk_rx_dst_set(sk, dst);
2088 ret = udp_queue_rcv_skb(sk, skb);
2090 /* a return value > 0 means to resubmit the input, but
2091 * it wants the return to be -protocol, or 0
2098 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2099 return __udp4_lib_mcast_deliver(net, skb, uh,
2100 saddr, daddr, udptable, proto);
2102 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2106 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2107 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2108 inet_compute_pseudo);
2110 ret = udp_queue_rcv_skb(sk, skb);
2112 /* a return value > 0 means to resubmit the input, but
2113 * it wants the return to be -protocol, or 0
2120 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2124 /* No socket. Drop packet silently, if checksum is wrong */
2125 if (udp_lib_checksum_complete(skb))
2128 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2129 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2132 * Hmm. We got an UDP packet to a port to which we
2133 * don't wanna listen. Ignore it.
2139 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2140 proto == IPPROTO_UDPLITE ? "Lite" : "",
2141 &saddr, ntohs(uh->source),
2143 &daddr, ntohs(uh->dest));
2148 * RFC1122: OK. Discards the bad packet silently (as far as
2149 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2151 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2152 proto == IPPROTO_UDPLITE ? "Lite" : "",
2153 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2155 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2157 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2162 /* We can only early demux multicast if there is a single matching socket.
2163 * If more than one socket found returns NULL
2165 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2166 __be16 loc_port, __be32 loc_addr,
2167 __be16 rmt_port, __be32 rmt_addr,
2170 struct sock *sk, *result;
2171 unsigned short hnum = ntohs(loc_port);
2172 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2173 struct udp_hslot *hslot = &udp_table.hash[slot];
2175 /* Do not bother scanning a too big list */
2176 if (hslot->count > 10)
2180 sk_for_each_rcu(sk, &hslot->head) {
2181 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2182 rmt_port, rmt_addr, dif, sdif, hnum)) {
2192 /* For unicast we should only early demux connected sockets or we can
2193 * break forwarding setups. The chains here can be long so only check
2194 * if the first socket is an exact match and if not move on.
2196 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2197 __be16 loc_port, __be32 loc_addr,
2198 __be16 rmt_port, __be32 rmt_addr,
2201 unsigned short hnum = ntohs(loc_port);
2202 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2203 unsigned int slot2 = hash2 & udp_table.mask;
2204 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2205 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2206 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2209 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2210 if (INET_MATCH(sk, net, acookie, rmt_addr,
2211 loc_addr, ports, dif, sdif))
2213 /* Only check first socket in chain */
2219 void udp_v4_early_demux(struct sk_buff *skb)
2221 struct net *net = dev_net(skb->dev);
2222 const struct iphdr *iph;
2223 const struct udphdr *uh;
2224 struct sock *sk = NULL;
2225 struct dst_entry *dst;
2226 int dif = skb->dev->ifindex;
2227 int sdif = inet_sdif(skb);
2230 /* validate the packet */
2231 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2237 if (skb->pkt_type == PACKET_BROADCAST ||
2238 skb->pkt_type == PACKET_MULTICAST) {
2239 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2244 /* we are supposed to accept bcast packets */
2245 if (skb->pkt_type == PACKET_MULTICAST) {
2246 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2252 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2253 uh->source, iph->saddr,
2255 } else if (skb->pkt_type == PACKET_HOST) {
2256 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2257 uh->source, iph->saddr, dif, sdif);
2260 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2264 skb->destructor = sock_efree;
2265 dst = READ_ONCE(sk->sk_rx_dst);
2268 dst = dst_check(dst, 0);
2270 /* set noref for now.
2271 * any place which wants to hold dst has to call
2274 skb_dst_set_noref(skb, dst);
2278 int udp_rcv(struct sk_buff *skb)
2280 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2283 void udp_destroy_sock(struct sock *sk)
2285 struct udp_sock *up = udp_sk(sk);
2286 bool slow = lock_sock_fast(sk);
2287 udp_flush_pending_frames(sk);
2288 unlock_sock_fast(sk, slow);
2289 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2290 void (*encap_destroy)(struct sock *sk);
2291 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2298 * Socket option code for UDP
2300 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2301 char __user *optval, unsigned int optlen,
2302 int (*push_pending_frames)(struct sock *))
2304 struct udp_sock *up = udp_sk(sk);
2307 int is_udplite = IS_UDPLITE(sk);
2309 if (optlen < sizeof(int))
2312 if (get_user(val, (int __user *)optval))
2315 valbool = val ? 1 : 0;
2324 push_pending_frames(sk);
2332 case UDP_ENCAP_ESPINUDP:
2333 case UDP_ENCAP_ESPINUDP_NON_IKE:
2334 up->encap_rcv = xfrm4_udp_encap_rcv;
2336 case UDP_ENCAP_L2TPINUDP:
2337 up->encap_type = val;
2346 case UDP_NO_CHECK6_TX:
2347 up->no_check6_tx = valbool;
2350 case UDP_NO_CHECK6_RX:
2351 up->no_check6_rx = valbool;
2355 * UDP-Lite's partial checksum coverage (RFC 3828).
2357 /* The sender sets actual checksum coverage length via this option.
2358 * The case coverage > packet length is handled by send module. */
2359 case UDPLITE_SEND_CSCOV:
2360 if (!is_udplite) /* Disable the option on UDP sockets */
2361 return -ENOPROTOOPT;
2362 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2364 else if (val > USHRT_MAX)
2367 up->pcflag |= UDPLITE_SEND_CC;
2370 /* The receiver specifies a minimum checksum coverage value. To make
2371 * sense, this should be set to at least 8 (as done below). If zero is
2372 * used, this again means full checksum coverage. */
2373 case UDPLITE_RECV_CSCOV:
2374 if (!is_udplite) /* Disable the option on UDP sockets */
2375 return -ENOPROTOOPT;
2376 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2378 else if (val > USHRT_MAX)
2381 up->pcflag |= UDPLITE_RECV_CC;
2391 EXPORT_SYMBOL(udp_lib_setsockopt);
2393 int udp_setsockopt(struct sock *sk, int level, int optname,
2394 char __user *optval, unsigned int optlen)
2396 if (level == SOL_UDP || level == SOL_UDPLITE)
2397 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2398 udp_push_pending_frames);
2399 return ip_setsockopt(sk, level, optname, optval, optlen);
2402 #ifdef CONFIG_COMPAT
2403 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2404 char __user *optval, unsigned int optlen)
2406 if (level == SOL_UDP || level == SOL_UDPLITE)
2407 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2408 udp_push_pending_frames);
2409 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2413 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2414 char __user *optval, int __user *optlen)
2416 struct udp_sock *up = udp_sk(sk);
2419 if (get_user(len, optlen))
2422 len = min_t(unsigned int, len, sizeof(int));
2433 val = up->encap_type;
2436 case UDP_NO_CHECK6_TX:
2437 val = up->no_check6_tx;
2440 case UDP_NO_CHECK6_RX:
2441 val = up->no_check6_rx;
2444 /* The following two cannot be changed on UDP sockets, the return is
2445 * always 0 (which corresponds to the full checksum coverage of UDP). */
2446 case UDPLITE_SEND_CSCOV:
2450 case UDPLITE_RECV_CSCOV:
2455 return -ENOPROTOOPT;
2458 if (put_user(len, optlen))
2460 if (copy_to_user(optval, &val, len))
2464 EXPORT_SYMBOL(udp_lib_getsockopt);
2466 int udp_getsockopt(struct sock *sk, int level, int optname,
2467 char __user *optval, int __user *optlen)
2469 if (level == SOL_UDP || level == SOL_UDPLITE)
2470 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2471 return ip_getsockopt(sk, level, optname, optval, optlen);
2474 #ifdef CONFIG_COMPAT
2475 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2476 char __user *optval, int __user *optlen)
2478 if (level == SOL_UDP || level == SOL_UDPLITE)
2479 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2480 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2484 * udp_poll - wait for a UDP event.
2485 * @file - file struct
2487 * @wait - poll table
2489 * This is same as datagram poll, except for the special case of
2490 * blocking sockets. If application is using a blocking fd
2491 * and a packet with checksum error is in the queue;
2492 * then it could get return from select indicating data available
2493 * but then block when reading it. Add special case code
2494 * to work around these arguably broken applications.
2496 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2498 unsigned int mask = datagram_poll(file, sock, wait);
2499 struct sock *sk = sock->sk;
2501 if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2502 mask |= POLLIN | POLLRDNORM;
2504 sock_rps_record_flow(sk);
2506 /* Check for false positives due to checksum errors */
2507 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2508 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2509 mask &= ~(POLLIN | POLLRDNORM);
2514 EXPORT_SYMBOL(udp_poll);
2516 int udp_abort(struct sock *sk, int err)
2521 sk->sk_error_report(sk);
2522 __udp_disconnect(sk, 0);
2528 EXPORT_SYMBOL_GPL(udp_abort);
2530 struct proto udp_prot = {
2532 .owner = THIS_MODULE,
2533 .close = udp_lib_close,
2534 .connect = ip4_datagram_connect,
2535 .disconnect = udp_disconnect,
2537 .init = udp_init_sock,
2538 .destroy = udp_destroy_sock,
2539 .setsockopt = udp_setsockopt,
2540 .getsockopt = udp_getsockopt,
2541 .sendmsg = udp_sendmsg,
2542 .recvmsg = udp_recvmsg,
2543 .sendpage = udp_sendpage,
2544 .release_cb = ip4_datagram_release_cb,
2545 .hash = udp_lib_hash,
2546 .unhash = udp_lib_unhash,
2547 .rehash = udp_v4_rehash,
2548 .get_port = udp_v4_get_port,
2549 .memory_allocated = &udp_memory_allocated,
2550 .sysctl_mem = sysctl_udp_mem,
2551 .sysctl_wmem = &sysctl_udp_wmem_min,
2552 .sysctl_rmem = &sysctl_udp_rmem_min,
2553 .obj_size = sizeof(struct udp_sock),
2554 .h.udp_table = &udp_table,
2555 #ifdef CONFIG_COMPAT
2556 .compat_setsockopt = compat_udp_setsockopt,
2557 .compat_getsockopt = compat_udp_getsockopt,
2559 .diag_destroy = udp_abort,
2561 EXPORT_SYMBOL(udp_prot);
2563 /* ------------------------------------------------------------------------ */
2564 #ifdef CONFIG_PROC_FS
2566 static struct sock *udp_get_first(struct seq_file *seq, int start)
2569 struct udp_iter_state *state = seq->private;
2570 struct net *net = seq_file_net(seq);
2572 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2574 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2576 if (hlist_empty(&hslot->head))
2579 spin_lock_bh(&hslot->lock);
2580 sk_for_each(sk, &hslot->head) {
2581 if (!net_eq(sock_net(sk), net))
2583 if (sk->sk_family == state->family)
2586 spin_unlock_bh(&hslot->lock);
2593 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2595 struct udp_iter_state *state = seq->private;
2596 struct net *net = seq_file_net(seq);
2600 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2603 if (state->bucket <= state->udp_table->mask)
2604 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2605 return udp_get_first(seq, state->bucket + 1);
2610 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2612 struct sock *sk = udp_get_first(seq, 0);
2615 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2617 return pos ? NULL : sk;
2620 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2622 struct udp_iter_state *state = seq->private;
2623 state->bucket = MAX_UDP_PORTS;
2625 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2628 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2632 if (v == SEQ_START_TOKEN)
2633 sk = udp_get_idx(seq, 0);
2635 sk = udp_get_next(seq, v);
2641 static void udp_seq_stop(struct seq_file *seq, void *v)
2643 struct udp_iter_state *state = seq->private;
2645 if (state->bucket <= state->udp_table->mask)
2646 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2649 int udp_seq_open(struct inode *inode, struct file *file)
2651 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2652 struct udp_iter_state *s;
2655 err = seq_open_net(inode, file, &afinfo->seq_ops,
2656 sizeof(struct udp_iter_state));
2660 s = ((struct seq_file *)file->private_data)->private;
2661 s->family = afinfo->family;
2662 s->udp_table = afinfo->udp_table;
2665 EXPORT_SYMBOL(udp_seq_open);
2667 /* ------------------------------------------------------------------------ */
2668 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2670 struct proc_dir_entry *p;
2673 afinfo->seq_ops.start = udp_seq_start;
2674 afinfo->seq_ops.next = udp_seq_next;
2675 afinfo->seq_ops.stop = udp_seq_stop;
2677 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2678 afinfo->seq_fops, afinfo);
2683 EXPORT_SYMBOL(udp_proc_register);
2685 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2687 remove_proc_entry(afinfo->name, net->proc_net);
2689 EXPORT_SYMBOL(udp_proc_unregister);
2691 /* ------------------------------------------------------------------------ */
2692 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2695 struct inet_sock *inet = inet_sk(sp);
2696 __be32 dest = inet->inet_daddr;
2697 __be32 src = inet->inet_rcv_saddr;
2698 __u16 destp = ntohs(inet->inet_dport);
2699 __u16 srcp = ntohs(inet->inet_sport);
2701 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2702 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2703 bucket, src, srcp, dest, destp, sp->sk_state,
2704 sk_wmem_alloc_get(sp),
2705 sk_rmem_alloc_get(sp),
2707 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2709 refcount_read(&sp->sk_refcnt), sp,
2710 atomic_read(&sp->sk_drops));
2713 int udp4_seq_show(struct seq_file *seq, void *v)
2715 seq_setwidth(seq, 127);
2716 if (v == SEQ_START_TOKEN)
2717 seq_puts(seq, " sl local_address rem_address st tx_queue "
2718 "rx_queue tr tm->when retrnsmt uid timeout "
2719 "inode ref pointer drops");
2721 struct udp_iter_state *state = seq->private;
2723 udp4_format_sock(v, seq, state->bucket);
2729 static const struct file_operations udp_afinfo_seq_fops = {
2730 .owner = THIS_MODULE,
2731 .open = udp_seq_open,
2733 .llseek = seq_lseek,
2734 .release = seq_release_net
2737 /* ------------------------------------------------------------------------ */
2738 static struct udp_seq_afinfo udp4_seq_afinfo = {
2741 .udp_table = &udp_table,
2742 .seq_fops = &udp_afinfo_seq_fops,
2744 .show = udp4_seq_show,
2748 static int __net_init udp4_proc_init_net(struct net *net)
2750 return udp_proc_register(net, &udp4_seq_afinfo);
2753 static void __net_exit udp4_proc_exit_net(struct net *net)
2755 udp_proc_unregister(net, &udp4_seq_afinfo);
2758 static struct pernet_operations udp4_net_ops = {
2759 .init = udp4_proc_init_net,
2760 .exit = udp4_proc_exit_net,
2763 int __init udp4_proc_init(void)
2765 return register_pernet_subsys(&udp4_net_ops);
2768 void udp4_proc_exit(void)
2770 unregister_pernet_subsys(&udp4_net_ops);
2772 #endif /* CONFIG_PROC_FS */
2774 static __initdata unsigned long uhash_entries;
2775 static int __init set_uhash_entries(char *str)
2782 ret = kstrtoul(str, 0, &uhash_entries);
2786 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2787 uhash_entries = UDP_HTABLE_SIZE_MIN;
2790 __setup("uhash_entries=", set_uhash_entries);
2792 void __init udp_table_init(struct udp_table *table, const char *name)
2796 table->hash = alloc_large_system_hash(name,
2797 2 * sizeof(struct udp_hslot),
2799 21, /* one slot per 2 MB */
2803 UDP_HTABLE_SIZE_MIN,
2806 table->hash2 = table->hash + (table->mask + 1);
2807 for (i = 0; i <= table->mask; i++) {
2808 INIT_HLIST_HEAD(&table->hash[i].head);
2809 table->hash[i].count = 0;
2810 spin_lock_init(&table->hash[i].lock);
2812 for (i = 0; i <= table->mask; i++) {
2813 INIT_HLIST_HEAD(&table->hash2[i].head);
2814 table->hash2[i].count = 0;
2815 spin_lock_init(&table->hash2[i].lock);
2819 u32 udp_flow_hashrnd(void)
2821 static u32 hashrnd __read_mostly;
2823 net_get_random_once(&hashrnd, sizeof(hashrnd));
2827 EXPORT_SYMBOL(udp_flow_hashrnd);
2829 void __init udp_init(void)
2831 unsigned long limit;
2834 udp_table_init(&udp_table, "UDP");
2835 limit = nr_free_buffer_pages() / 8;
2836 limit = max(limit, 128UL);
2837 sysctl_udp_mem[0] = limit / 4 * 3;
2838 sysctl_udp_mem[1] = limit;
2839 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2841 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2842 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2844 /* 16 spinlocks per cpu */
2845 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2846 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2849 panic("UDP: failed to alloc udp_busylocks\n");
2850 for (i = 0; i < (1U << udp_busylocks_log); i++)
2851 spin_lock_init(udp_busylocks + i);