2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The User Datagram Protocol (UDP).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
80 #define pr_fmt(fmt) "UDP: " fmt
82 #include <asm/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"
117 struct udp_table udp_table __read_mostly;
118 EXPORT_SYMBOL(udp_table);
120 long sysctl_udp_mem[3] __read_mostly;
121 EXPORT_SYMBOL(sysctl_udp_mem);
123 int sysctl_udp_rmem_min __read_mostly;
124 EXPORT_SYMBOL(sysctl_udp_rmem_min);
126 int sysctl_udp_wmem_min __read_mostly;
127 EXPORT_SYMBOL(sysctl_udp_wmem_min);
129 atomic_long_t udp_memory_allocated;
130 EXPORT_SYMBOL(udp_memory_allocated);
132 #define MAX_UDP_PORTS 65536
133 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
135 static int udp_lib_lport_inuse(struct net *net, __u16 num,
136 const struct udp_hslot *hslot,
137 unsigned long *bitmap,
139 int (*saddr_comp)(const struct sock *sk1,
140 const struct sock *sk2),
144 struct hlist_nulls_node *node;
145 kuid_t uid = sock_i_uid(sk);
147 sk_nulls_for_each(sk2, node, &hslot->head) {
148 if (net_eq(sock_net(sk2), net) &&
150 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
151 (!sk2->sk_reuse || !sk->sk_reuse) &&
152 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
153 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
154 (!sk2->sk_reuseport || !sk->sk_reuseport ||
155 !uid_eq(uid, sock_i_uid(sk2))) &&
156 saddr_comp(sk, sk2)) {
159 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap);
166 * Note: we still hold spinlock of primary hash chain, so no other writer
167 * can insert/delete a socket with local_port == num
169 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
170 struct udp_hslot *hslot2,
172 int (*saddr_comp)(const struct sock *sk1,
173 const struct sock *sk2))
176 struct hlist_nulls_node *node;
177 kuid_t uid = sock_i_uid(sk);
180 spin_lock(&hslot2->lock);
181 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) {
182 if (net_eq(sock_net(sk2), net) &&
184 (udp_sk(sk2)->udp_port_hash == num) &&
185 (!sk2->sk_reuse || !sk->sk_reuse) &&
186 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
187 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
188 (!sk2->sk_reuseport || !sk->sk_reuseport ||
189 !uid_eq(uid, sock_i_uid(sk2))) &&
190 saddr_comp(sk, sk2)) {
195 spin_unlock(&hslot2->lock);
200 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
202 * @sk: socket struct in question
203 * @snum: port number to look up
204 * @saddr_comp: AF-dependent comparison of bound local IP addresses
205 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
208 int udp_lib_get_port(struct sock *sk, unsigned short snum,
209 int (*saddr_comp)(const struct sock *sk1,
210 const struct sock *sk2),
211 unsigned int hash2_nulladdr)
213 struct udp_hslot *hslot, *hslot2;
214 struct udp_table *udptable = sk->sk_prot->h.udp_table;
216 struct net *net = sock_net(sk);
219 int low, high, remaining;
221 unsigned short first, last;
222 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
224 inet_get_local_port_range(net, &low, &high);
225 remaining = (high - low) + 1;
227 rand = prandom_u32();
228 first = reciprocal_scale(rand, remaining) + low;
230 * force rand to be an odd multiple of UDP_HTABLE_SIZE
232 rand = (rand | 1) * (udptable->mask + 1);
233 last = first + udptable->mask + 1;
235 hslot = udp_hashslot(udptable, net, first);
236 bitmap_zero(bitmap, PORTS_PER_CHAIN);
237 spin_lock_bh(&hslot->lock);
238 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
239 saddr_comp, udptable->log);
243 * Iterate on all possible values of snum for this hash.
244 * Using steps of an odd multiple of UDP_HTABLE_SIZE
245 * give us randomization and full range coverage.
248 if (low <= snum && snum <= high &&
249 !test_bit(snum >> udptable->log, bitmap) &&
250 !inet_is_local_reserved_port(net, snum))
253 } while (snum != first);
254 spin_unlock_bh(&hslot->lock);
255 } while (++first != last);
258 hslot = udp_hashslot(udptable, net, snum);
259 spin_lock_bh(&hslot->lock);
260 if (hslot->count > 10) {
262 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
264 slot2 &= udptable->mask;
265 hash2_nulladdr &= udptable->mask;
267 hslot2 = udp_hashslot2(udptable, slot2);
268 if (hslot->count < hslot2->count)
269 goto scan_primary_hash;
271 exist = udp_lib_lport_inuse2(net, snum, hslot2,
273 if (!exist && (hash2_nulladdr != slot2)) {
274 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
275 exist = udp_lib_lport_inuse2(net, snum, hslot2,
284 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
289 inet_sk(sk)->inet_num = snum;
290 udp_sk(sk)->udp_port_hash = snum;
291 udp_sk(sk)->udp_portaddr_hash ^= snum;
292 if (sk_unhashed(sk)) {
293 sk_nulls_add_node_rcu(sk, &hslot->head);
295 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
297 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
298 spin_lock(&hslot2->lock);
299 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
302 spin_unlock(&hslot2->lock);
306 spin_unlock_bh(&hslot->lock);
310 EXPORT_SYMBOL(udp_lib_get_port);
312 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
314 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
316 return (!ipv6_only_sock(sk2) &&
317 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
318 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
321 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
324 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
327 int udp_v4_get_port(struct sock *sk, unsigned short snum)
329 unsigned int hash2_nulladdr =
330 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
331 unsigned int hash2_partial =
332 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
334 /* precompute partial secondary hash */
335 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
336 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
339 static inline int compute_score(struct sock *sk, struct net *net,
340 __be32 saddr, unsigned short hnum, __be16 sport,
341 __be32 daddr, __be16 dport, int dif)
344 struct inet_sock *inet;
346 if (!net_eq(sock_net(sk), net) ||
347 udp_sk(sk)->udp_port_hash != hnum ||
351 score = (sk->sk_family == PF_INET) ? 2 : 1;
354 if (inet->inet_rcv_saddr) {
355 if (inet->inet_rcv_saddr != daddr)
360 if (inet->inet_daddr) {
361 if (inet->inet_daddr != saddr)
366 if (inet->inet_dport) {
367 if (inet->inet_dport != sport)
372 if (sk->sk_bound_dev_if) {
373 if (sk->sk_bound_dev_if != dif)
377 if (sk->sk_incoming_cpu == raw_smp_processor_id())
383 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
385 static inline int compute_score2(struct sock *sk, struct net *net,
386 __be32 saddr, __be16 sport,
387 __be32 daddr, unsigned int hnum, int dif)
390 struct inet_sock *inet;
392 if (!net_eq(sock_net(sk), net) ||
398 if (inet->inet_rcv_saddr != daddr ||
399 inet->inet_num != hnum)
402 score = (sk->sk_family == PF_INET) ? 2 : 1;
404 if (inet->inet_daddr) {
405 if (inet->inet_daddr != saddr)
410 if (inet->inet_dport) {
411 if (inet->inet_dport != sport)
416 if (sk->sk_bound_dev_if) {
417 if (sk->sk_bound_dev_if != dif)
422 if (sk->sk_incoming_cpu == raw_smp_processor_id())
428 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
429 const __u16 lport, const __be32 faddr,
432 static u32 udp_ehash_secret __read_mostly;
434 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
436 return __inet_ehashfn(laddr, lport, faddr, fport,
437 udp_ehash_secret + net_hash_mix(net));
440 /* called with read_rcu_lock() */
441 static struct sock *udp4_lib_lookup2(struct net *net,
442 __be32 saddr, __be16 sport,
443 __be32 daddr, unsigned int hnum, int dif,
444 struct udp_hslot *hslot2, unsigned int slot2)
446 struct sock *sk, *result;
447 struct hlist_nulls_node *node;
448 int score, badness, matches = 0, reuseport = 0;
454 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
455 score = compute_score2(sk, net, saddr, sport,
457 if (score > badness) {
460 reuseport = sk->sk_reuseport;
462 hash = udp_ehashfn(net, daddr, hnum,
466 } else if (score == badness && reuseport) {
468 if (reciprocal_scale(hash, matches) == 0)
470 hash = next_pseudo_random32(hash);
474 * if the nulls value we got at the end of this lookup is
475 * not the expected one, we must restart lookup.
476 * We probably met an item that was moved to another chain.
478 if (get_nulls_value(node) != slot2)
481 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
483 else if (unlikely(compute_score2(result, net, saddr, sport,
484 daddr, hnum, dif) < badness)) {
492 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
493 * harder than this. -DaveM
495 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
496 __be16 sport, __be32 daddr, __be16 dport,
497 int dif, struct udp_table *udptable)
499 struct sock *sk, *result;
500 struct hlist_nulls_node *node;
501 unsigned short hnum = ntohs(dport);
502 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
503 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
504 int score, badness, matches = 0, reuseport = 0;
508 if (hslot->count > 10) {
509 hash2 = udp4_portaddr_hash(net, daddr, hnum);
510 slot2 = hash2 & udptable->mask;
511 hslot2 = &udptable->hash2[slot2];
512 if (hslot->count < hslot2->count)
515 result = udp4_lib_lookup2(net, saddr, sport,
519 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
520 slot2 = hash2 & udptable->mask;
521 hslot2 = &udptable->hash2[slot2];
522 if (hslot->count < hslot2->count)
525 result = udp4_lib_lookup2(net, saddr, sport,
526 htonl(INADDR_ANY), hnum, dif,
535 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
536 score = compute_score(sk, net, saddr, hnum, sport,
538 if (score > badness) {
541 reuseport = sk->sk_reuseport;
543 hash = udp_ehashfn(net, daddr, hnum,
547 } else if (score == badness && reuseport) {
549 if (reciprocal_scale(hash, matches) == 0)
551 hash = next_pseudo_random32(hash);
555 * if the nulls value we got at the end of this lookup is
556 * not the expected one, we must restart lookup.
557 * We probably met an item that was moved to another chain.
559 if (get_nulls_value(node) != slot)
563 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
565 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
566 daddr, dport, dif) < badness)) {
574 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
576 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
577 __be16 sport, __be16 dport,
578 struct udp_table *udptable)
580 const struct iphdr *iph = ip_hdr(skb);
582 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
583 iph->daddr, dport, inet_iif(skb),
587 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
588 __be32 daddr, __be16 dport, int dif)
590 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
592 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
594 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
595 __be16 loc_port, __be32 loc_addr,
596 __be16 rmt_port, __be32 rmt_addr,
597 int dif, unsigned short hnum)
599 struct inet_sock *inet = inet_sk(sk);
601 if (!net_eq(sock_net(sk), net) ||
602 udp_sk(sk)->udp_port_hash != hnum ||
603 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
604 (inet->inet_dport != rmt_port && inet->inet_dport) ||
605 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
606 ipv6_only_sock(sk) ||
607 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
609 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
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, udptable);
640 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
641 return; /* No socket for error */
650 case ICMP_TIME_EXCEEDED:
653 case ICMP_SOURCE_QUENCH:
655 case ICMP_PARAMETERPROB:
659 case ICMP_DEST_UNREACH:
660 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
661 ipv4_sk_update_pmtu(skb, sk, info);
662 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
670 if (code <= NR_ICMP_UNREACH) {
671 harderr = icmp_err_convert[code].fatal;
672 err = icmp_err_convert[code].errno;
676 ipv4_sk_redirect(skb, sk);
681 * RFC1122: OK. Passes ICMP errors back to application, as per
684 if (!inet->recverr) {
685 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
688 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
691 sk->sk_error_report(sk);
696 void udp_err(struct sk_buff *skb, u32 info)
698 __udp4_lib_err(skb, info, &udp_table);
702 * Throw away all pending data and cancel the corking. Socket is locked.
704 void udp_flush_pending_frames(struct sock *sk)
706 struct udp_sock *up = udp_sk(sk);
711 ip_flush_pending_frames(sk);
714 EXPORT_SYMBOL(udp_flush_pending_frames);
717 * udp4_hwcsum - handle outgoing HW checksumming
718 * @skb: sk_buff containing the filled-in UDP header
719 * (checksum field must be zeroed out)
720 * @src: source IP address
721 * @dst: destination IP address
723 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
725 struct udphdr *uh = udp_hdr(skb);
726 int offset = skb_transport_offset(skb);
727 int len = skb->len - offset;
731 if (!skb_has_frag_list(skb)) {
733 * Only one fragment on the socket.
735 skb->csum_start = skb_transport_header(skb) - skb->head;
736 skb->csum_offset = offsetof(struct udphdr, check);
737 uh->check = ~csum_tcpudp_magic(src, dst, len,
740 struct sk_buff *frags;
743 * HW-checksum won't work as there are two or more
744 * fragments on the socket so that all csums of sk_buffs
747 skb_walk_frags(skb, frags) {
748 csum = csum_add(csum, frags->csum);
752 csum = skb_checksum(skb, offset, hlen, csum);
753 skb->ip_summed = CHECKSUM_NONE;
755 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
757 uh->check = CSUM_MANGLED_0;
760 EXPORT_SYMBOL_GPL(udp4_hwcsum);
762 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
763 * for the simple case like when setting the checksum for a UDP tunnel.
765 void udp_set_csum(bool nocheck, struct sk_buff *skb,
766 __be32 saddr, __be32 daddr, int len)
768 struct udphdr *uh = udp_hdr(skb);
772 else if (skb_is_gso(skb))
773 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
774 else if (skb_dst(skb) && skb_dst(skb)->dev &&
775 (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) {
777 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
779 skb->ip_summed = CHECKSUM_PARTIAL;
780 skb->csum_start = skb_transport_header(skb) - skb->head;
781 skb->csum_offset = offsetof(struct udphdr, check);
782 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
786 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
789 csum = skb_checksum(skb, 0, len, 0);
790 uh->check = udp_v4_check(len, saddr, daddr, csum);
792 uh->check = CSUM_MANGLED_0;
794 skb->ip_summed = CHECKSUM_UNNECESSARY;
797 EXPORT_SYMBOL(udp_set_csum);
799 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
801 struct sock *sk = skb->sk;
802 struct inet_sock *inet = inet_sk(sk);
805 int is_udplite = IS_UDPLITE(sk);
806 int offset = skb_transport_offset(skb);
807 int len = skb->len - offset;
811 * Create a UDP header
814 uh->source = inet->inet_sport;
815 uh->dest = fl4->fl4_dport;
816 uh->len = htons(len);
819 if (is_udplite) /* UDP-Lite */
820 csum = udplite_csum(skb);
822 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
824 skb->ip_summed = CHECKSUM_NONE;
827 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
829 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
833 csum = udp_csum(skb);
835 /* add protocol-dependent pseudo-header */
836 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
837 sk->sk_protocol, csum);
839 uh->check = CSUM_MANGLED_0;
842 err = ip_send_skb(sock_net(sk), skb);
844 if (err == -ENOBUFS && !inet->recverr) {
845 UDP_INC_STATS_USER(sock_net(sk),
846 UDP_MIB_SNDBUFERRORS, is_udplite);
850 UDP_INC_STATS_USER(sock_net(sk),
851 UDP_MIB_OUTDATAGRAMS, is_udplite);
856 * Push out all pending data as one UDP datagram. Socket is locked.
858 int udp_push_pending_frames(struct sock *sk)
860 struct udp_sock *up = udp_sk(sk);
861 struct inet_sock *inet = inet_sk(sk);
862 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
866 skb = ip_finish_skb(sk, fl4);
870 err = udp_send_skb(skb, fl4);
877 EXPORT_SYMBOL(udp_push_pending_frames);
879 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
881 struct inet_sock *inet = inet_sk(sk);
882 struct udp_sock *up = udp_sk(sk);
883 struct flowi4 fl4_stack;
886 struct ipcm_cookie ipc;
887 struct rtable *rt = NULL;
890 __be32 daddr, faddr, saddr;
893 int err, is_udplite = IS_UDPLITE(sk);
894 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
895 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
897 struct ip_options_data opt_copy;
906 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
914 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
916 fl4 = &inet->cork.fl.u.ip4;
919 * There are pending frames.
920 * The socket lock must be held while it's corked.
923 if (likely(up->pending)) {
924 if (unlikely(up->pending != AF_INET)) {
932 ulen += sizeof(struct udphdr);
935 * Get and verify the address.
938 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
939 if (msg->msg_namelen < sizeof(*usin))
941 if (usin->sin_family != AF_INET) {
942 if (usin->sin_family != AF_UNSPEC)
943 return -EAFNOSUPPORT;
946 daddr = usin->sin_addr.s_addr;
947 dport = usin->sin_port;
951 if (sk->sk_state != TCP_ESTABLISHED)
952 return -EDESTADDRREQ;
953 daddr = inet->inet_daddr;
954 dport = inet->inet_dport;
955 /* Open fast path for connected socket.
956 Route will not be used, if at least one option is set.
960 ipc.addr = inet->inet_saddr;
962 ipc.oif = sk->sk_bound_dev_if;
964 sock_tx_timestamp(sk, &ipc.tx_flags);
966 if (msg->msg_controllen) {
967 err = ip_cmsg_send(sock_net(sk), msg, &ipc,
968 sk->sk_family == AF_INET6);
976 struct ip_options_rcu *inet_opt;
979 inet_opt = rcu_dereference(inet->inet_opt);
981 memcpy(&opt_copy, inet_opt,
982 sizeof(*inet_opt) + inet_opt->opt.optlen);
983 ipc.opt = &opt_copy.opt;
989 ipc.addr = faddr = daddr;
991 if (ipc.opt && ipc.opt->opt.srr) {
994 faddr = ipc.opt->opt.faddr;
997 tos = get_rttos(&ipc, inet);
998 if (sock_flag(sk, SOCK_LOCALROUTE) ||
999 (msg->msg_flags & MSG_DONTROUTE) ||
1000 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1005 if (ipv4_is_multicast(daddr)) {
1007 ipc.oif = inet->mc_index;
1009 saddr = inet->mc_addr;
1011 } else if (!ipc.oif)
1012 ipc.oif = inet->uc_index;
1015 rt = (struct rtable *)sk_dst_check(sk, 0);
1018 struct net *net = sock_net(sk);
1019 __u8 flow_flags = inet_sk_flowi_flags(sk);
1023 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1024 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1026 faddr, saddr, dport, inet->inet_sport);
1028 if (!saddr && ipc.oif)
1029 l3mdev_get_saddr(net, ipc.oif, fl4);
1031 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1032 rt = ip_route_output_flow(net, fl4, sk);
1036 if (err == -ENETUNREACH)
1037 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1042 if ((rt->rt_flags & RTCF_BROADCAST) &&
1043 !sock_flag(sk, SOCK_BROADCAST))
1046 sk_dst_set(sk, dst_clone(&rt->dst));
1049 if (msg->msg_flags&MSG_CONFIRM)
1055 daddr = ipc.addr = fl4->daddr;
1057 /* Lockless fast path for the non-corking case. */
1059 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1060 sizeof(struct udphdr), &ipc, &rt,
1063 if (!IS_ERR_OR_NULL(skb))
1064 err = udp_send_skb(skb, fl4);
1069 if (unlikely(up->pending)) {
1070 /* The socket is already corked while preparing it. */
1071 /* ... which is an evident application bug. --ANK */
1074 net_dbg_ratelimited("cork app bug 2\n");
1079 * Now cork the socket to pend data.
1081 fl4 = &inet->cork.fl.u.ip4;
1084 fl4->fl4_dport = dport;
1085 fl4->fl4_sport = inet->inet_sport;
1086 up->pending = AF_INET;
1090 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1091 sizeof(struct udphdr), &ipc, &rt,
1092 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1094 udp_flush_pending_frames(sk);
1096 err = udp_push_pending_frames(sk);
1097 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1108 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1109 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1110 * we don't have a good statistic (IpOutDiscards but it can be too many
1111 * things). We could add another new stat but at least for now that
1112 * seems like overkill.
1114 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1115 UDP_INC_STATS_USER(sock_net(sk),
1116 UDP_MIB_SNDBUFERRORS, is_udplite);
1121 dst_confirm(&rt->dst);
1122 if (!(msg->msg_flags&MSG_PROBE) || len)
1123 goto back_from_confirm;
1127 EXPORT_SYMBOL(udp_sendmsg);
1129 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1130 size_t size, int flags)
1132 struct inet_sock *inet = inet_sk(sk);
1133 struct udp_sock *up = udp_sk(sk);
1136 if (flags & MSG_SENDPAGE_NOTLAST)
1140 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1142 /* Call udp_sendmsg to specify destination address which
1143 * sendpage interface can't pass.
1144 * This will succeed only when the socket is connected.
1146 ret = udp_sendmsg(sk, &msg, 0);
1153 if (unlikely(!up->pending)) {
1156 net_dbg_ratelimited("udp cork app bug 3\n");
1160 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1161 page, offset, size, flags);
1162 if (ret == -EOPNOTSUPP) {
1164 return sock_no_sendpage(sk->sk_socket, page, offset,
1168 udp_flush_pending_frames(sk);
1173 if (!(up->corkflag || (flags&MSG_MORE)))
1174 ret = udp_push_pending_frames(sk);
1183 * first_packet_length - return length of first packet in receive queue
1186 * Drops all bad checksum frames, until a valid one is found.
1187 * Returns the length of found skb, or 0 if none is found.
1189 static unsigned int first_packet_length(struct sock *sk)
1191 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1192 struct sk_buff *skb;
1195 __skb_queue_head_init(&list_kill);
1197 spin_lock_bh(&rcvq->lock);
1198 while ((skb = skb_peek(rcvq)) != NULL &&
1199 udp_lib_checksum_complete(skb)) {
1200 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1202 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1204 atomic_inc(&sk->sk_drops);
1205 __skb_unlink(skb, rcvq);
1206 __skb_queue_tail(&list_kill, skb);
1208 res = skb ? skb->len : 0;
1209 spin_unlock_bh(&rcvq->lock);
1211 if (!skb_queue_empty(&list_kill)) {
1212 bool slow = lock_sock_fast(sk);
1214 __skb_queue_purge(&list_kill);
1215 sk_mem_reclaim_partial(sk);
1216 unlock_sock_fast(sk, slow);
1222 * IOCTL requests applicable to the UDP protocol
1225 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1230 int amount = sk_wmem_alloc_get(sk);
1232 return put_user(amount, (int __user *)arg);
1237 unsigned int amount = first_packet_length(sk);
1241 * We will only return the amount
1242 * of this packet since that is all
1243 * that will be read.
1245 amount -= sizeof(struct udphdr);
1247 return put_user(amount, (int __user *)arg);
1251 return -ENOIOCTLCMD;
1256 EXPORT_SYMBOL(udp_ioctl);
1259 * This should be easy, if there is something there we
1260 * return it, otherwise we block.
1263 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1264 int flags, int *addr_len)
1266 struct inet_sock *inet = inet_sk(sk);
1267 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1268 struct sk_buff *skb;
1269 unsigned int ulen, copied;
1270 int peeked, off = 0;
1272 int is_udplite = IS_UDPLITE(sk);
1275 if (flags & MSG_ERRQUEUE)
1276 return ip_recv_error(sk, msg, len, addr_len);
1279 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1280 &peeked, &off, &err);
1284 ulen = skb->len - sizeof(struct udphdr);
1288 else if (copied < ulen)
1289 msg->msg_flags |= MSG_TRUNC;
1292 * If checksum is needed at all, try to do it while copying the
1293 * data. If the data is truncated, or if we only want a partial
1294 * coverage checksum (UDP-Lite), do it before the copy.
1297 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1298 if (udp_lib_checksum_complete(skb))
1302 if (skb_csum_unnecessary(skb))
1303 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
1306 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr),
1313 if (unlikely(err)) {
1314 trace_kfree_skb(skb, udp_recvmsg);
1316 atomic_inc(&sk->sk_drops);
1317 UDP_INC_STATS_USER(sock_net(sk),
1318 UDP_MIB_INERRORS, is_udplite);
1324 UDP_INC_STATS_USER(sock_net(sk),
1325 UDP_MIB_INDATAGRAMS, is_udplite);
1327 sock_recv_ts_and_drops(msg, sk, skb);
1329 /* Copy the address. */
1331 sin->sin_family = AF_INET;
1332 sin->sin_port = udp_hdr(skb)->source;
1333 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1334 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1335 *addr_len = sizeof(*sin);
1337 if (inet->cmsg_flags)
1338 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr));
1341 if (flags & MSG_TRUNC)
1345 skb_free_datagram_locked(sk, skb);
1350 slow = lock_sock_fast(sk);
1351 if (!skb_kill_datagram(sk, skb, flags)) {
1352 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1353 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1355 unlock_sock_fast(sk, slow);
1357 /* starting over for a new packet, but check if we need to yield */
1359 msg->msg_flags &= ~MSG_TRUNC;
1363 int udp_disconnect(struct sock *sk, int flags)
1365 struct inet_sock *inet = inet_sk(sk);
1367 * 1003.1g - break association.
1370 sk->sk_state = TCP_CLOSE;
1371 inet->inet_daddr = 0;
1372 inet->inet_dport = 0;
1373 sock_rps_reset_rxhash(sk);
1374 sk->sk_bound_dev_if = 0;
1375 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1376 inet_reset_saddr(sk);
1378 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1379 sk->sk_prot->unhash(sk);
1380 inet->inet_sport = 0;
1385 EXPORT_SYMBOL(udp_disconnect);
1387 void udp_lib_unhash(struct sock *sk)
1389 if (sk_hashed(sk)) {
1390 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1391 struct udp_hslot *hslot, *hslot2;
1393 hslot = udp_hashslot(udptable, sock_net(sk),
1394 udp_sk(sk)->udp_port_hash);
1395 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1397 spin_lock_bh(&hslot->lock);
1398 if (sk_nulls_del_node_init_rcu(sk)) {
1400 inet_sk(sk)->inet_num = 0;
1401 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1403 spin_lock(&hslot2->lock);
1404 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1406 spin_unlock(&hslot2->lock);
1408 spin_unlock_bh(&hslot->lock);
1411 EXPORT_SYMBOL(udp_lib_unhash);
1414 * inet_rcv_saddr was changed, we must rehash secondary hash
1416 void udp_lib_rehash(struct sock *sk, u16 newhash)
1418 if (sk_hashed(sk)) {
1419 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1420 struct udp_hslot *hslot, *hslot2, *nhslot2;
1422 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1423 nhslot2 = udp_hashslot2(udptable, newhash);
1424 udp_sk(sk)->udp_portaddr_hash = newhash;
1425 if (hslot2 != nhslot2) {
1426 hslot = udp_hashslot(udptable, sock_net(sk),
1427 udp_sk(sk)->udp_port_hash);
1428 /* we must lock primary chain too */
1429 spin_lock_bh(&hslot->lock);
1431 spin_lock(&hslot2->lock);
1432 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1434 spin_unlock(&hslot2->lock);
1436 spin_lock(&nhslot2->lock);
1437 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1440 spin_unlock(&nhslot2->lock);
1442 spin_unlock_bh(&hslot->lock);
1446 EXPORT_SYMBOL(udp_lib_rehash);
1448 static void udp_v4_rehash(struct sock *sk)
1450 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1451 inet_sk(sk)->inet_rcv_saddr,
1452 inet_sk(sk)->inet_num);
1453 udp_lib_rehash(sk, new_hash);
1456 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1460 if (inet_sk(sk)->inet_daddr) {
1461 sock_rps_save_rxhash(sk, skb);
1462 sk_mark_napi_id(sk, skb);
1463 sk_incoming_cpu_update(sk);
1466 rc = sock_queue_rcv_skb(sk, skb);
1468 int is_udplite = IS_UDPLITE(sk);
1470 /* Note that an ENOMEM error is charged twice */
1472 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1474 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1476 trace_udp_fail_queue_rcv_skb(rc, sk);
1484 static struct static_key udp_encap_needed __read_mostly;
1485 void udp_encap_enable(void)
1487 if (!static_key_enabled(&udp_encap_needed))
1488 static_key_slow_inc(&udp_encap_needed);
1490 EXPORT_SYMBOL(udp_encap_enable);
1495 * >0: "udp encap" protocol resubmission
1497 * Note that in the success and error cases, the skb is assumed to
1498 * have either been requeued or freed.
1500 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1502 struct udp_sock *up = udp_sk(sk);
1504 int is_udplite = IS_UDPLITE(sk);
1507 * Charge it to the socket, dropping if the queue is full.
1509 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1513 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1514 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1517 * This is an encapsulation socket so pass the skb to
1518 * the socket's udp_encap_rcv() hook. Otherwise, just
1519 * fall through and pass this up the UDP socket.
1520 * up->encap_rcv() returns the following value:
1521 * =0 if skb was successfully passed to the encap
1522 * handler or was discarded by it.
1523 * >0 if skb should be passed on to UDP.
1524 * <0 if skb should be resubmitted as proto -N
1527 /* if we're overly short, let UDP handle it */
1528 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1529 if (skb->len > sizeof(struct udphdr) && encap_rcv) {
1532 /* Verify checksum before giving to encap */
1533 if (udp_lib_checksum_complete(skb))
1536 ret = encap_rcv(sk, skb);
1538 UDP_INC_STATS_BH(sock_net(sk),
1539 UDP_MIB_INDATAGRAMS,
1545 /* FALLTHROUGH -- it's a UDP Packet */
1549 * UDP-Lite specific tests, ignored on UDP sockets
1551 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1554 * MIB statistics other than incrementing the error count are
1555 * disabled for the following two types of errors: these depend
1556 * on the application settings, not on the functioning of the
1557 * protocol stack as such.
1559 * RFC 3828 here recommends (sec 3.3): "There should also be a
1560 * way ... to ... at least let the receiving application block
1561 * delivery of packets with coverage values less than a value
1562 * provided by the application."
1564 if (up->pcrlen == 0) { /* full coverage was set */
1565 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1566 UDP_SKB_CB(skb)->cscov, skb->len);
1569 /* The next case involves violating the min. coverage requested
1570 * by the receiver. This is subtle: if receiver wants x and x is
1571 * greater than the buffersize/MTU then receiver will complain
1572 * that it wants x while sender emits packets of smaller size y.
1573 * Therefore the above ...()->partial_cov statement is essential.
1575 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1576 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1577 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1582 if (rcu_access_pointer(sk->sk_filter) &&
1583 udp_lib_checksum_complete(skb))
1586 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1587 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1594 ipv4_pktinfo_prepare(sk, skb);
1596 if (!sock_owned_by_user(sk))
1597 rc = __udp_queue_rcv_skb(sk, skb);
1598 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1607 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1609 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1610 atomic_inc(&sk->sk_drops);
1615 static void flush_stack(struct sock **stack, unsigned int count,
1616 struct sk_buff *skb, unsigned int final)
1619 struct sk_buff *skb1 = NULL;
1622 for (i = 0; i < count; i++) {
1625 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1628 atomic_inc(&sk->sk_drops);
1629 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1631 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1635 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1644 /* For TCP sockets, sk_rx_dst is protected by socket lock
1645 * For UDP, we use xchg() to guard against concurrent changes.
1647 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1649 struct dst_entry *old;
1652 old = xchg(&sk->sk_rx_dst, dst);
1657 * Multicasts and broadcasts go to each listener.
1659 * Note: called only from the BH handler context.
1661 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1663 __be32 saddr, __be32 daddr,
1664 struct udp_table *udptable,
1667 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1668 struct hlist_nulls_node *node;
1669 unsigned short hnum = ntohs(uh->dest);
1670 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1671 int dif = skb->dev->ifindex;
1672 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
1673 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1674 bool inner_flushed = false;
1677 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1679 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1681 hslot = &udp_table.hash2[hash2];
1682 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1685 spin_lock(&hslot->lock);
1686 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
1687 if (__udp_is_mcast_sock(net, sk,
1691 if (unlikely(count == ARRAY_SIZE(stack))) {
1692 flush_stack(stack, count, skb, ~0);
1693 inner_flushed = true;
1696 stack[count++] = sk;
1701 spin_unlock(&hslot->lock);
1703 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1704 if (use_hash2 && hash2 != hash2_any) {
1710 * do the slow work with no lock held
1713 flush_stack(stack, count, skb, count - 1);
1716 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
1717 proto == IPPROTO_UDPLITE);
1723 /* Initialize UDP checksum. If exited with zero value (success),
1724 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1725 * Otherwise, csum completion requires chacksumming packet body,
1726 * including udp header and folding it to skb->csum.
1728 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1733 UDP_SKB_CB(skb)->partial_cov = 0;
1734 UDP_SKB_CB(skb)->cscov = skb->len;
1736 if (proto == IPPROTO_UDPLITE) {
1737 err = udplite_checksum_init(skb, uh);
1742 return skb_checksum_init_zero_check(skb, proto, uh->check,
1743 inet_compute_pseudo);
1747 * All we need to do is get the socket, and then do a checksum.
1750 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1755 unsigned short ulen;
1756 struct rtable *rt = skb_rtable(skb);
1757 __be32 saddr, daddr;
1758 struct net *net = dev_net(skb->dev);
1761 * Validate the packet.
1763 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1764 goto drop; /* No space for header. */
1767 ulen = ntohs(uh->len);
1768 saddr = ip_hdr(skb)->saddr;
1769 daddr = ip_hdr(skb)->daddr;
1771 if (ulen > skb->len)
1774 if (proto == IPPROTO_UDP) {
1775 /* UDP validates ulen. */
1776 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1781 if (udp4_csum_init(skb, uh, proto))
1784 sk = skb_steal_sock(skb);
1786 struct dst_entry *dst = skb_dst(skb);
1789 if (unlikely(sk->sk_rx_dst != dst))
1790 udp_sk_rx_dst_set(sk, dst);
1792 ret = udp_queue_rcv_skb(sk, skb);
1794 /* a return value > 0 means to resubmit the input, but
1795 * it wants the return to be -protocol, or 0
1802 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1803 return __udp4_lib_mcast_deliver(net, skb, uh,
1804 saddr, daddr, udptable, proto);
1806 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1810 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1811 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1812 inet_compute_pseudo);
1814 ret = udp_queue_rcv_skb(sk, skb);
1817 /* a return value > 0 means to resubmit the input, but
1818 * it wants the return to be -protocol, or 0
1825 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1829 /* No socket. Drop packet silently, if checksum is wrong */
1830 if (udp_lib_checksum_complete(skb))
1833 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1834 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1837 * Hmm. We got an UDP packet to a port to which we
1838 * don't wanna listen. Ignore it.
1844 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1845 proto == IPPROTO_UDPLITE ? "Lite" : "",
1846 &saddr, ntohs(uh->source),
1848 &daddr, ntohs(uh->dest));
1853 * RFC1122: OK. Discards the bad packet silently (as far as
1854 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1856 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1857 proto == IPPROTO_UDPLITE ? "Lite" : "",
1858 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1860 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1862 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1867 /* We can only early demux multicast if there is a single matching socket.
1868 * If more than one socket found returns NULL
1870 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1871 __be16 loc_port, __be32 loc_addr,
1872 __be16 rmt_port, __be32 rmt_addr,
1875 struct sock *sk, *result;
1876 struct hlist_nulls_node *node;
1877 unsigned short hnum = ntohs(loc_port);
1878 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1879 struct udp_hslot *hslot = &udp_table.hash[slot];
1881 /* Do not bother scanning a too big list */
1882 if (hslot->count > 10)
1889 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1890 if (__udp_is_mcast_sock(net, sk,
1899 * if the nulls value we got at the end of this lookup is
1900 * not the expected one, we must restart lookup.
1901 * We probably met an item that was moved to another chain.
1903 if (get_nulls_value(node) != slot)
1908 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1910 else if (unlikely(!__udp_is_mcast_sock(net, result,
1922 /* For unicast we should only early demux connected sockets or we can
1923 * break forwarding setups. The chains here can be long so only check
1924 * if the first socket is an exact match and if not move on.
1926 static struct sock *__udp4_lib_demux_lookup(struct net *net,
1927 __be16 loc_port, __be32 loc_addr,
1928 __be16 rmt_port, __be32 rmt_addr,
1931 struct sock *sk, *result;
1932 struct hlist_nulls_node *node;
1933 unsigned short hnum = ntohs(loc_port);
1934 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1935 unsigned int slot2 = hash2 & udp_table.mask;
1936 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1937 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
1938 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1942 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
1943 if (INET_MATCH(sk, net, acookie,
1944 rmt_addr, loc_addr, ports, dif))
1946 /* Only check first socket in chain */
1951 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1953 else if (unlikely(!INET_MATCH(sk, net, acookie,
1964 void udp_v4_early_demux(struct sk_buff *skb)
1966 struct net *net = dev_net(skb->dev);
1967 const struct iphdr *iph;
1968 const struct udphdr *uh;
1970 struct dst_entry *dst;
1971 int dif = skb->dev->ifindex;
1974 /* validate the packet */
1975 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1981 if (skb->pkt_type == PACKET_BROADCAST ||
1982 skb->pkt_type == PACKET_MULTICAST) {
1983 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1988 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
1992 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1993 uh->source, iph->saddr, dif);
1994 } else if (skb->pkt_type == PACKET_HOST) {
1995 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1996 uh->source, iph->saddr, dif);
2005 skb->destructor = sock_efree;
2006 dst = READ_ONCE(sk->sk_rx_dst);
2009 dst = dst_check(dst, 0);
2011 /* DST_NOCACHE can not be used without taking a reference */
2012 if (dst->flags & DST_NOCACHE) {
2013 if (likely(atomic_inc_not_zero(&dst->__refcnt)))
2014 skb_dst_set(skb, dst);
2016 skb_dst_set_noref(skb, dst);
2021 int udp_rcv(struct sk_buff *skb)
2023 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2026 void udp_destroy_sock(struct sock *sk)
2028 struct udp_sock *up = udp_sk(sk);
2029 bool slow = lock_sock_fast(sk);
2030 udp_flush_pending_frames(sk);
2031 unlock_sock_fast(sk, slow);
2032 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2033 void (*encap_destroy)(struct sock *sk);
2034 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2041 * Socket option code for UDP
2043 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2044 char __user *optval, unsigned int optlen,
2045 int (*push_pending_frames)(struct sock *))
2047 struct udp_sock *up = udp_sk(sk);
2050 int is_udplite = IS_UDPLITE(sk);
2052 if (optlen < sizeof(int))
2055 if (get_user(val, (int __user *)optval))
2058 valbool = val ? 1 : 0;
2067 push_pending_frames(sk);
2075 case UDP_ENCAP_ESPINUDP:
2076 case UDP_ENCAP_ESPINUDP_NON_IKE:
2077 up->encap_rcv = xfrm4_udp_encap_rcv;
2079 case UDP_ENCAP_L2TPINUDP:
2080 up->encap_type = val;
2089 case UDP_NO_CHECK6_TX:
2090 up->no_check6_tx = valbool;
2093 case UDP_NO_CHECK6_RX:
2094 up->no_check6_rx = valbool;
2098 * UDP-Lite's partial checksum coverage (RFC 3828).
2100 /* The sender sets actual checksum coverage length via this option.
2101 * The case coverage > packet length is handled by send module. */
2102 case UDPLITE_SEND_CSCOV:
2103 if (!is_udplite) /* Disable the option on UDP sockets */
2104 return -ENOPROTOOPT;
2105 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2107 else if (val > USHRT_MAX)
2110 up->pcflag |= UDPLITE_SEND_CC;
2113 /* The receiver specifies a minimum checksum coverage value. To make
2114 * sense, this should be set to at least 8 (as done below). If zero is
2115 * used, this again means full checksum coverage. */
2116 case UDPLITE_RECV_CSCOV:
2117 if (!is_udplite) /* Disable the option on UDP sockets */
2118 return -ENOPROTOOPT;
2119 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2121 else if (val > USHRT_MAX)
2124 up->pcflag |= UDPLITE_RECV_CC;
2134 EXPORT_SYMBOL(udp_lib_setsockopt);
2136 int udp_setsockopt(struct sock *sk, int level, int optname,
2137 char __user *optval, unsigned int optlen)
2139 if (level == SOL_UDP || level == SOL_UDPLITE)
2140 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2141 udp_push_pending_frames);
2142 return ip_setsockopt(sk, level, optname, optval, optlen);
2145 #ifdef CONFIG_COMPAT
2146 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2147 char __user *optval, unsigned int optlen)
2149 if (level == SOL_UDP || level == SOL_UDPLITE)
2150 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2151 udp_push_pending_frames);
2152 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2156 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2157 char __user *optval, int __user *optlen)
2159 struct udp_sock *up = udp_sk(sk);
2162 if (get_user(len, optlen))
2165 len = min_t(unsigned int, len, sizeof(int));
2176 val = up->encap_type;
2179 case UDP_NO_CHECK6_TX:
2180 val = up->no_check6_tx;
2183 case UDP_NO_CHECK6_RX:
2184 val = up->no_check6_rx;
2187 /* The following two cannot be changed on UDP sockets, the return is
2188 * always 0 (which corresponds to the full checksum coverage of UDP). */
2189 case UDPLITE_SEND_CSCOV:
2193 case UDPLITE_RECV_CSCOV:
2198 return -ENOPROTOOPT;
2201 if (put_user(len, optlen))
2203 if (copy_to_user(optval, &val, len))
2207 EXPORT_SYMBOL(udp_lib_getsockopt);
2209 int udp_getsockopt(struct sock *sk, int level, int optname,
2210 char __user *optval, int __user *optlen)
2212 if (level == SOL_UDP || level == SOL_UDPLITE)
2213 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2214 return ip_getsockopt(sk, level, optname, optval, optlen);
2217 #ifdef CONFIG_COMPAT
2218 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2219 char __user *optval, int __user *optlen)
2221 if (level == SOL_UDP || level == SOL_UDPLITE)
2222 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2223 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2227 * udp_poll - wait for a UDP event.
2228 * @file - file struct
2230 * @wait - poll table
2232 * This is same as datagram poll, except for the special case of
2233 * blocking sockets. If application is using a blocking fd
2234 * and a packet with checksum error is in the queue;
2235 * then it could get return from select indicating data available
2236 * but then block when reading it. Add special case code
2237 * to work around these arguably broken applications.
2239 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2241 unsigned int mask = datagram_poll(file, sock, wait);
2242 struct sock *sk = sock->sk;
2244 sock_rps_record_flow(sk);
2246 /* Check for false positives due to checksum errors */
2247 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2248 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2249 mask &= ~(POLLIN | POLLRDNORM);
2254 EXPORT_SYMBOL(udp_poll);
2256 struct proto udp_prot = {
2258 .owner = THIS_MODULE,
2259 .close = udp_lib_close,
2260 .connect = ip4_datagram_connect,
2261 .disconnect = udp_disconnect,
2263 .destroy = udp_destroy_sock,
2264 .setsockopt = udp_setsockopt,
2265 .getsockopt = udp_getsockopt,
2266 .sendmsg = udp_sendmsg,
2267 .recvmsg = udp_recvmsg,
2268 .sendpage = udp_sendpage,
2269 .backlog_rcv = __udp_queue_rcv_skb,
2270 .release_cb = ip4_datagram_release_cb,
2271 .hash = udp_lib_hash,
2272 .unhash = udp_lib_unhash,
2273 .rehash = udp_v4_rehash,
2274 .get_port = udp_v4_get_port,
2275 .memory_allocated = &udp_memory_allocated,
2276 .sysctl_mem = sysctl_udp_mem,
2277 .sysctl_wmem = &sysctl_udp_wmem_min,
2278 .sysctl_rmem = &sysctl_udp_rmem_min,
2279 .obj_size = sizeof(struct udp_sock),
2280 .slab_flags = SLAB_DESTROY_BY_RCU,
2281 .h.udp_table = &udp_table,
2282 #ifdef CONFIG_COMPAT
2283 .compat_setsockopt = compat_udp_setsockopt,
2284 .compat_getsockopt = compat_udp_getsockopt,
2286 .clear_sk = sk_prot_clear_portaddr_nulls,
2288 EXPORT_SYMBOL(udp_prot);
2290 /* ------------------------------------------------------------------------ */
2291 #ifdef CONFIG_PROC_FS
2293 static struct sock *udp_get_first(struct seq_file *seq, int start)
2296 struct udp_iter_state *state = seq->private;
2297 struct net *net = seq_file_net(seq);
2299 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2301 struct hlist_nulls_node *node;
2302 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2304 if (hlist_nulls_empty(&hslot->head))
2307 spin_lock_bh(&hslot->lock);
2308 sk_nulls_for_each(sk, node, &hslot->head) {
2309 if (!net_eq(sock_net(sk), net))
2311 if (sk->sk_family == state->family)
2314 spin_unlock_bh(&hslot->lock);
2321 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2323 struct udp_iter_state *state = seq->private;
2324 struct net *net = seq_file_net(seq);
2327 sk = sk_nulls_next(sk);
2328 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2331 if (state->bucket <= state->udp_table->mask)
2332 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2333 return udp_get_first(seq, state->bucket + 1);
2338 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2340 struct sock *sk = udp_get_first(seq, 0);
2343 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2345 return pos ? NULL : sk;
2348 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2350 struct udp_iter_state *state = seq->private;
2351 state->bucket = MAX_UDP_PORTS;
2353 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2356 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2360 if (v == SEQ_START_TOKEN)
2361 sk = udp_get_idx(seq, 0);
2363 sk = udp_get_next(seq, v);
2369 static void udp_seq_stop(struct seq_file *seq, void *v)
2371 struct udp_iter_state *state = seq->private;
2373 if (state->bucket <= state->udp_table->mask)
2374 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2377 int udp_seq_open(struct inode *inode, struct file *file)
2379 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2380 struct udp_iter_state *s;
2383 err = seq_open_net(inode, file, &afinfo->seq_ops,
2384 sizeof(struct udp_iter_state));
2388 s = ((struct seq_file *)file->private_data)->private;
2389 s->family = afinfo->family;
2390 s->udp_table = afinfo->udp_table;
2393 EXPORT_SYMBOL(udp_seq_open);
2395 /* ------------------------------------------------------------------------ */
2396 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2398 struct proc_dir_entry *p;
2401 afinfo->seq_ops.start = udp_seq_start;
2402 afinfo->seq_ops.next = udp_seq_next;
2403 afinfo->seq_ops.stop = udp_seq_stop;
2405 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2406 afinfo->seq_fops, afinfo);
2411 EXPORT_SYMBOL(udp_proc_register);
2413 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2415 remove_proc_entry(afinfo->name, net->proc_net);
2417 EXPORT_SYMBOL(udp_proc_unregister);
2419 /* ------------------------------------------------------------------------ */
2420 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2423 struct inet_sock *inet = inet_sk(sp);
2424 __be32 dest = inet->inet_daddr;
2425 __be32 src = inet->inet_rcv_saddr;
2426 __u16 destp = ntohs(inet->inet_dport);
2427 __u16 srcp = ntohs(inet->inet_sport);
2429 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2430 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2431 bucket, src, srcp, dest, destp, sp->sk_state,
2432 sk_wmem_alloc_get(sp),
2433 sk_rmem_alloc_get(sp),
2435 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2437 atomic_read(&sp->sk_refcnt), sp,
2438 atomic_read(&sp->sk_drops));
2441 int udp4_seq_show(struct seq_file *seq, void *v)
2443 seq_setwidth(seq, 127);
2444 if (v == SEQ_START_TOKEN)
2445 seq_puts(seq, " sl local_address rem_address st tx_queue "
2446 "rx_queue tr tm->when retrnsmt uid timeout "
2447 "inode ref pointer drops");
2449 struct udp_iter_state *state = seq->private;
2451 udp4_format_sock(v, seq, state->bucket);
2457 static const struct file_operations udp_afinfo_seq_fops = {
2458 .owner = THIS_MODULE,
2459 .open = udp_seq_open,
2461 .llseek = seq_lseek,
2462 .release = seq_release_net
2465 /* ------------------------------------------------------------------------ */
2466 static struct udp_seq_afinfo udp4_seq_afinfo = {
2469 .udp_table = &udp_table,
2470 .seq_fops = &udp_afinfo_seq_fops,
2472 .show = udp4_seq_show,
2476 static int __net_init udp4_proc_init_net(struct net *net)
2478 return udp_proc_register(net, &udp4_seq_afinfo);
2481 static void __net_exit udp4_proc_exit_net(struct net *net)
2483 udp_proc_unregister(net, &udp4_seq_afinfo);
2486 static struct pernet_operations udp4_net_ops = {
2487 .init = udp4_proc_init_net,
2488 .exit = udp4_proc_exit_net,
2491 int __init udp4_proc_init(void)
2493 return register_pernet_subsys(&udp4_net_ops);
2496 void udp4_proc_exit(void)
2498 unregister_pernet_subsys(&udp4_net_ops);
2500 #endif /* CONFIG_PROC_FS */
2502 static __initdata unsigned long uhash_entries;
2503 static int __init set_uhash_entries(char *str)
2510 ret = kstrtoul(str, 0, &uhash_entries);
2514 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2515 uhash_entries = UDP_HTABLE_SIZE_MIN;
2518 __setup("uhash_entries=", set_uhash_entries);
2520 void __init udp_table_init(struct udp_table *table, const char *name)
2524 table->hash = alloc_large_system_hash(name,
2525 2 * sizeof(struct udp_hslot),
2527 21, /* one slot per 2 MB */
2531 UDP_HTABLE_SIZE_MIN,
2534 table->hash2 = table->hash + (table->mask + 1);
2535 for (i = 0; i <= table->mask; i++) {
2536 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2537 table->hash[i].count = 0;
2538 spin_lock_init(&table->hash[i].lock);
2540 for (i = 0; i <= table->mask; i++) {
2541 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2542 table->hash2[i].count = 0;
2543 spin_lock_init(&table->hash2[i].lock);
2547 u32 udp_flow_hashrnd(void)
2549 static u32 hashrnd __read_mostly;
2551 net_get_random_once(&hashrnd, sizeof(hashrnd));
2555 EXPORT_SYMBOL(udp_flow_hashrnd);
2557 void __init udp_init(void)
2559 unsigned long limit;
2561 udp_table_init(&udp_table, "UDP");
2562 limit = nr_free_buffer_pages() / 8;
2563 limit = max(limit, 128UL);
2564 sysctl_udp_mem[0] = limit / 4 * 3;
2565 sysctl_udp_mem[1] = limit;
2566 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2568 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2569 sysctl_udp_wmem_min = SK_MEM_QUANTUM;