1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * The User Datagram Protocol (UDP).
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
72 * James Chapman : Add L2TP encapsulation type.
75 #define pr_fmt(fmt) "UDP: " fmt
77 #include <linux/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <linux/memblock.h>
80 #include <linux/highmem.h>
81 #include <linux/swap.h>
82 #include <linux/types.h>
83 #include <linux/fcntl.h>
84 #include <linux/module.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/igmp.h>
88 #include <linux/inetdevice.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
93 #include <linux/inet.h>
94 #include <linux/netdevice.h>
95 #include <linux/slab.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/inet_hashtables.h>
103 #include <net/ip_tunnels.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <trace/events/udp.h>
108 #include <linux/static_key.h>
109 #include <linux/btf_ids.h>
110 #include <trace/events/skb.h>
111 #include <net/busy_poll.h>
112 #include "udp_impl.h"
113 #include <net/sock_reuseport.h>
114 #include <net/addrconf.h>
115 #include <net/udp_tunnel.h>
116 #if IS_ENABLED(CONFIG_IPV6)
117 #include <net/ipv6_stubs.h>
120 struct udp_table udp_table __read_mostly;
121 EXPORT_SYMBOL(udp_table);
123 long sysctl_udp_mem[3] __read_mostly;
124 EXPORT_SYMBOL(sysctl_udp_mem);
126 atomic_long_t udp_memory_allocated;
127 EXPORT_SYMBOL(udp_memory_allocated);
129 #define MAX_UDP_PORTS 65536
130 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
132 static int udp_lib_lport_inuse(struct net *net, __u16 num,
133 const struct udp_hslot *hslot,
134 unsigned long *bitmap,
135 struct sock *sk, unsigned int log)
138 kuid_t uid = sock_i_uid(sk);
140 sk_for_each(sk2, &hslot->head) {
141 if (net_eq(sock_net(sk2), net) &&
143 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
144 (!sk2->sk_reuse || !sk->sk_reuse) &&
145 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
146 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
147 inet_rcv_saddr_equal(sk, sk2, true)) {
148 if (sk2->sk_reuseport && sk->sk_reuseport &&
149 !rcu_access_pointer(sk->sk_reuseport_cb) &&
150 uid_eq(uid, sock_i_uid(sk2))) {
156 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
165 * Note: we still hold spinlock of primary hash chain, so no other writer
166 * can insert/delete a socket with local_port == num
168 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
169 struct udp_hslot *hslot2,
173 kuid_t uid = sock_i_uid(sk);
176 spin_lock(&hslot2->lock);
177 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
178 if (net_eq(sock_net(sk2), net) &&
180 (udp_sk(sk2)->udp_port_hash == num) &&
181 (!sk2->sk_reuse || !sk->sk_reuse) &&
182 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
183 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
184 inet_rcv_saddr_equal(sk, sk2, true)) {
185 if (sk2->sk_reuseport && sk->sk_reuseport &&
186 !rcu_access_pointer(sk->sk_reuseport_cb) &&
187 uid_eq(uid, sock_i_uid(sk2))) {
195 spin_unlock(&hslot2->lock);
199 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
201 struct net *net = sock_net(sk);
202 kuid_t uid = sock_i_uid(sk);
205 sk_for_each(sk2, &hslot->head) {
206 if (net_eq(sock_net(sk2), net) &&
208 sk2->sk_family == sk->sk_family &&
209 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
210 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
211 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
212 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
213 inet_rcv_saddr_equal(sk, sk2, false)) {
214 return reuseport_add_sock(sk, sk2,
215 inet_rcv_saddr_any(sk));
219 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
223 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
225 * @sk: socket struct in question
226 * @snum: port number to look up
227 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
230 int udp_lib_get_port(struct sock *sk, unsigned short snum,
231 unsigned int hash2_nulladdr)
233 struct udp_hslot *hslot, *hslot2;
234 struct udp_table *udptable = sk->sk_prot->h.udp_table;
236 struct net *net = sock_net(sk);
239 int low, high, remaining;
241 unsigned short first, last;
242 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
244 inet_get_local_port_range(net, &low, &high);
245 remaining = (high - low) + 1;
247 rand = prandom_u32();
248 first = reciprocal_scale(rand, remaining) + low;
250 * force rand to be an odd multiple of UDP_HTABLE_SIZE
252 rand = (rand | 1) * (udptable->mask + 1);
253 last = first + udptable->mask + 1;
255 hslot = udp_hashslot(udptable, net, first);
256 bitmap_zero(bitmap, PORTS_PER_CHAIN);
257 spin_lock_bh(&hslot->lock);
258 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
263 * Iterate on all possible values of snum for this hash.
264 * Using steps of an odd multiple of UDP_HTABLE_SIZE
265 * give us randomization and full range coverage.
268 if (low <= snum && snum <= high &&
269 !test_bit(snum >> udptable->log, bitmap) &&
270 !inet_is_local_reserved_port(net, snum))
273 } while (snum != first);
274 spin_unlock_bh(&hslot->lock);
276 } while (++first != last);
279 hslot = udp_hashslot(udptable, net, snum);
280 spin_lock_bh(&hslot->lock);
281 if (hslot->count > 10) {
283 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
285 slot2 &= udptable->mask;
286 hash2_nulladdr &= udptable->mask;
288 hslot2 = udp_hashslot2(udptable, slot2);
289 if (hslot->count < hslot2->count)
290 goto scan_primary_hash;
292 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
293 if (!exist && (hash2_nulladdr != slot2)) {
294 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
295 exist = udp_lib_lport_inuse2(net, snum, hslot2,
304 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
308 inet_sk(sk)->inet_num = snum;
309 udp_sk(sk)->udp_port_hash = snum;
310 udp_sk(sk)->udp_portaddr_hash ^= snum;
311 if (sk_unhashed(sk)) {
312 if (sk->sk_reuseport &&
313 udp_reuseport_add_sock(sk, hslot)) {
314 inet_sk(sk)->inet_num = 0;
315 udp_sk(sk)->udp_port_hash = 0;
316 udp_sk(sk)->udp_portaddr_hash ^= snum;
320 sk_add_node_rcu(sk, &hslot->head);
322 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
324 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
325 spin_lock(&hslot2->lock);
326 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
327 sk->sk_family == AF_INET6)
328 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
331 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
334 spin_unlock(&hslot2->lock);
336 sock_set_flag(sk, SOCK_RCU_FREE);
339 spin_unlock_bh(&hslot->lock);
343 EXPORT_SYMBOL(udp_lib_get_port);
345 int udp_v4_get_port(struct sock *sk, unsigned short snum)
347 unsigned int hash2_nulladdr =
348 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
349 unsigned int hash2_partial =
350 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
352 /* precompute partial secondary hash */
353 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
354 return udp_lib_get_port(sk, snum, hash2_nulladdr);
357 static int compute_score(struct sock *sk, struct net *net,
358 __be32 saddr, __be16 sport,
359 __be32 daddr, unsigned short hnum,
363 struct inet_sock *inet;
366 if (!net_eq(sock_net(sk), net) ||
367 udp_sk(sk)->udp_port_hash != hnum ||
371 if (sk->sk_rcv_saddr != daddr)
374 score = (sk->sk_family == PF_INET) ? 2 : 1;
377 if (inet->inet_daddr) {
378 if (inet->inet_daddr != saddr)
383 if (inet->inet_dport) {
384 if (inet->inet_dport != sport)
389 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
395 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
400 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
401 const __u16 lport, const __be32 faddr,
404 static u32 udp_ehash_secret __read_mostly;
406 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
408 return __inet_ehashfn(laddr, lport, faddr, fport,
409 udp_ehash_secret + net_hash_mix(net));
412 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
414 __be32 saddr, __be16 sport,
415 __be32 daddr, unsigned short hnum)
417 struct sock *reuse_sk = NULL;
420 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
421 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
422 reuse_sk = reuseport_select_sock(sk, hash, skb,
423 sizeof(struct udphdr));
428 /* called with rcu_read_lock() */
429 static struct sock *udp4_lib_lookup2(struct net *net,
430 __be32 saddr, __be16 sport,
431 __be32 daddr, unsigned int hnum,
433 struct udp_hslot *hslot2,
436 struct sock *sk, *result;
441 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
442 score = compute_score(sk, net, saddr, sport,
443 daddr, hnum, dif, sdif);
444 if (score > badness) {
445 result = lookup_reuseport(net, sk, skb,
446 saddr, sport, daddr, hnum);
447 /* Fall back to scoring if group has connections */
448 if (result && !reuseport_has_conns(sk, false))
451 result = result ? : sk;
458 static struct sock *udp4_lookup_run_bpf(struct net *net,
459 struct udp_table *udptable,
461 __be32 saddr, __be16 sport,
462 __be32 daddr, u16 hnum)
464 struct sock *sk, *reuse_sk;
467 if (udptable != &udp_table)
468 return NULL; /* only UDP is supported */
470 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP,
471 saddr, sport, daddr, hnum, &sk);
472 if (no_reuseport || IS_ERR_OR_NULL(sk))
475 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
481 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
482 * harder than this. -DaveM
484 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
485 __be16 sport, __be32 daddr, __be16 dport, int dif,
486 int sdif, struct udp_table *udptable, struct sk_buff *skb)
488 unsigned short hnum = ntohs(dport);
489 unsigned int hash2, slot2;
490 struct udp_hslot *hslot2;
491 struct sock *result, *sk;
493 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
494 slot2 = hash2 & udptable->mask;
495 hslot2 = &udptable->hash2[slot2];
497 /* Lookup connected or non-wildcard socket */
498 result = udp4_lib_lookup2(net, saddr, sport,
499 daddr, hnum, dif, sdif,
501 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
504 /* Lookup redirect from BPF */
505 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
506 sk = udp4_lookup_run_bpf(net, udptable, skb,
507 saddr, sport, daddr, hnum);
514 /* Got non-wildcard socket or error on first lookup */
518 /* Lookup wildcard sockets */
519 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
520 slot2 = hash2 & udptable->mask;
521 hslot2 = &udptable->hash2[slot2];
523 result = udp4_lib_lookup2(net, saddr, sport,
524 htonl(INADDR_ANY), hnum, dif, sdif,
531 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
533 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
534 __be16 sport, __be16 dport,
535 struct udp_table *udptable)
537 const struct iphdr *iph = ip_hdr(skb);
539 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
540 iph->daddr, dport, inet_iif(skb),
541 inet_sdif(skb), udptable, skb);
544 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
545 __be16 sport, __be16 dport)
547 const struct iphdr *iph = ip_hdr(skb);
549 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
550 iph->daddr, dport, inet_iif(skb),
551 inet_sdif(skb), &udp_table, NULL);
554 /* Must be called under rcu_read_lock().
555 * Does increment socket refcount.
557 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
558 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
559 __be32 daddr, __be16 dport, int dif)
563 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
564 dif, 0, &udp_table, NULL);
565 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
569 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
572 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
573 __be16 loc_port, __be32 loc_addr,
574 __be16 rmt_port, __be32 rmt_addr,
575 int dif, int sdif, unsigned short hnum)
577 struct inet_sock *inet = inet_sk(sk);
579 if (!net_eq(sock_net(sk), net) ||
580 udp_sk(sk)->udp_port_hash != hnum ||
581 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
582 (inet->inet_dport != rmt_port && inet->inet_dport) ||
583 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
584 ipv6_only_sock(sk) ||
585 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
587 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
592 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
593 void udp_encap_enable(void)
595 static_branch_inc(&udp_encap_needed_key);
597 EXPORT_SYMBOL(udp_encap_enable);
599 void udp_encap_disable(void)
601 static_branch_dec(&udp_encap_needed_key);
603 EXPORT_SYMBOL(udp_encap_disable);
605 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
606 * through error handlers in encapsulations looking for a match.
608 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
612 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
613 int (*handler)(struct sk_buff *skb, u32 info);
614 const struct ip_tunnel_encap_ops *encap;
616 encap = rcu_dereference(iptun_encaps[i]);
619 handler = encap->err_handler;
620 if (handler && !handler(skb, info))
627 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
628 * reversing source and destination port: this will match tunnels that force the
629 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
630 * lwtunnels might actually break this assumption by being configured with
631 * different destination ports on endpoints, in this case we won't be able to
632 * trace ICMP messages back to them.
634 * If this doesn't match any socket, probe tunnels with arbitrary destination
635 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
636 * we've sent packets to won't necessarily match the local destination port.
638 * Then ask the tunnel implementation to match the error against a valid
641 * Return an error if we can't find a match, the socket if we need further
642 * processing, zero otherwise.
644 static struct sock *__udp4_lib_err_encap(struct net *net,
645 const struct iphdr *iph,
647 struct udp_table *udptable,
648 struct sk_buff *skb, u32 info)
650 int network_offset, transport_offset;
653 network_offset = skb_network_offset(skb);
654 transport_offset = skb_transport_offset(skb);
656 /* Network header needs to point to the outer IPv4 header inside ICMP */
657 skb_reset_network_header(skb);
659 /* Transport header needs to point to the UDP header */
660 skb_set_transport_header(skb, iph->ihl << 2);
662 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
663 iph->saddr, uh->dest, skb->dev->ifindex, 0,
666 int (*lookup)(struct sock *sk, struct sk_buff *skb);
667 struct udp_sock *up = udp_sk(sk);
669 lookup = READ_ONCE(up->encap_err_lookup);
670 if (!lookup || lookup(sk, skb))
675 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
677 skb_set_transport_header(skb, transport_offset);
678 skb_set_network_header(skb, network_offset);
684 * This routine is called by the ICMP module when it gets some
685 * sort of error condition. If err < 0 then the socket should
686 * be closed and the error returned to the user. If err > 0
687 * it's just the icmp type << 8 | icmp code.
688 * Header points to the ip header of the error packet. We move
689 * on past this. Then (as it used to claim before adjustment)
690 * header points to the first 8 bytes of the udp header. We need
691 * to find the appropriate port.
694 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
696 struct inet_sock *inet;
697 const struct iphdr *iph = (const struct iphdr *)skb->data;
698 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
699 const int type = icmp_hdr(skb)->type;
700 const int code = icmp_hdr(skb)->code;
705 struct net *net = dev_net(skb->dev);
707 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
708 iph->saddr, uh->source, skb->dev->ifindex,
709 inet_sdif(skb), udptable, NULL);
710 if (!sk || udp_sk(sk)->encap_type) {
711 /* No socket for error: try tunnels before discarding */
712 sk = ERR_PTR(-ENOENT);
713 if (static_branch_unlikely(&udp_encap_needed_key)) {
714 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
721 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
734 case ICMP_TIME_EXCEEDED:
737 case ICMP_SOURCE_QUENCH:
739 case ICMP_PARAMETERPROB:
743 case ICMP_DEST_UNREACH:
744 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
745 ipv4_sk_update_pmtu(skb, sk, info);
746 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
754 if (code <= NR_ICMP_UNREACH) {
755 harderr = icmp_err_convert[code].fatal;
756 err = icmp_err_convert[code].errno;
760 ipv4_sk_redirect(skb, sk);
765 * RFC1122: OK. Passes ICMP errors back to application, as per
769 /* ...not for tunnels though: we don't have a sending socket */
772 if (!inet->recverr) {
773 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
776 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
779 sk->sk_error_report(sk);
784 int udp_err(struct sk_buff *skb, u32 info)
786 return __udp4_lib_err(skb, info, &udp_table);
790 * Throw away all pending data and cancel the corking. Socket is locked.
792 void udp_flush_pending_frames(struct sock *sk)
794 struct udp_sock *up = udp_sk(sk);
799 ip_flush_pending_frames(sk);
802 EXPORT_SYMBOL(udp_flush_pending_frames);
805 * udp4_hwcsum - handle outgoing HW checksumming
806 * @skb: sk_buff containing the filled-in UDP header
807 * (checksum field must be zeroed out)
808 * @src: source IP address
809 * @dst: destination IP address
811 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
813 struct udphdr *uh = udp_hdr(skb);
814 int offset = skb_transport_offset(skb);
815 int len = skb->len - offset;
819 if (!skb_has_frag_list(skb)) {
821 * Only one fragment on the socket.
823 skb->csum_start = skb_transport_header(skb) - skb->head;
824 skb->csum_offset = offsetof(struct udphdr, check);
825 uh->check = ~csum_tcpudp_magic(src, dst, len,
828 struct sk_buff *frags;
831 * HW-checksum won't work as there are two or more
832 * fragments on the socket so that all csums of sk_buffs
835 skb_walk_frags(skb, frags) {
836 csum = csum_add(csum, frags->csum);
840 csum = skb_checksum(skb, offset, hlen, csum);
841 skb->ip_summed = CHECKSUM_NONE;
843 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
845 uh->check = CSUM_MANGLED_0;
848 EXPORT_SYMBOL_GPL(udp4_hwcsum);
850 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
851 * for the simple case like when setting the checksum for a UDP tunnel.
853 void udp_set_csum(bool nocheck, struct sk_buff *skb,
854 __be32 saddr, __be32 daddr, int len)
856 struct udphdr *uh = udp_hdr(skb);
860 } else if (skb_is_gso(skb)) {
861 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
862 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
864 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
866 uh->check = CSUM_MANGLED_0;
868 skb->ip_summed = CHECKSUM_PARTIAL;
869 skb->csum_start = skb_transport_header(skb) - skb->head;
870 skb->csum_offset = offsetof(struct udphdr, check);
871 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
874 EXPORT_SYMBOL(udp_set_csum);
876 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
877 struct inet_cork *cork)
879 struct sock *sk = skb->sk;
880 struct inet_sock *inet = inet_sk(sk);
883 int is_udplite = IS_UDPLITE(sk);
884 int offset = skb_transport_offset(skb);
885 int len = skb->len - offset;
886 int datalen = len - sizeof(*uh);
890 * Create a UDP header
893 uh->source = inet->inet_sport;
894 uh->dest = fl4->fl4_dport;
895 uh->len = htons(len);
898 if (cork->gso_size) {
899 const int hlen = skb_network_header_len(skb) +
900 sizeof(struct udphdr);
902 if (hlen + cork->gso_size > cork->fragsize) {
906 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
910 if (sk->sk_no_check_tx) {
914 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
915 dst_xfrm(skb_dst(skb))) {
920 if (datalen > cork->gso_size) {
921 skb_shinfo(skb)->gso_size = cork->gso_size;
922 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
923 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
929 if (is_udplite) /* UDP-Lite */
930 csum = udplite_csum(skb);
932 else if (sk->sk_no_check_tx) { /* UDP csum off */
934 skb->ip_summed = CHECKSUM_NONE;
937 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
940 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
944 csum = udp_csum(skb);
946 /* add protocol-dependent pseudo-header */
947 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
948 sk->sk_protocol, csum);
950 uh->check = CSUM_MANGLED_0;
953 err = ip_send_skb(sock_net(sk), skb);
955 if (err == -ENOBUFS && !inet->recverr) {
956 UDP_INC_STATS(sock_net(sk),
957 UDP_MIB_SNDBUFERRORS, is_udplite);
961 UDP_INC_STATS(sock_net(sk),
962 UDP_MIB_OUTDATAGRAMS, is_udplite);
967 * Push out all pending data as one UDP datagram. Socket is locked.
969 int udp_push_pending_frames(struct sock *sk)
971 struct udp_sock *up = udp_sk(sk);
972 struct inet_sock *inet = inet_sk(sk);
973 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
977 skb = ip_finish_skb(sk, fl4);
981 err = udp_send_skb(skb, fl4, &inet->cork.base);
988 EXPORT_SYMBOL(udp_push_pending_frames);
990 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
992 switch (cmsg->cmsg_type) {
994 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
996 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1003 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1005 struct cmsghdr *cmsg;
1006 bool need_ip = false;
1009 for_each_cmsghdr(cmsg, msg) {
1010 if (!CMSG_OK(msg, cmsg))
1013 if (cmsg->cmsg_level != SOL_UDP) {
1018 err = __udp_cmsg_send(cmsg, gso_size);
1025 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1027 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1029 struct inet_sock *inet = inet_sk(sk);
1030 struct udp_sock *up = udp_sk(sk);
1031 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1032 struct flowi4 fl4_stack;
1035 struct ipcm_cookie ipc;
1036 struct rtable *rt = NULL;
1039 __be32 daddr, faddr, saddr;
1042 int err, is_udplite = IS_UDPLITE(sk);
1043 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1044 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1045 struct sk_buff *skb;
1046 struct ip_options_data opt_copy;
1055 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1058 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1060 fl4 = &inet->cork.fl.u.ip4;
1063 * There are pending frames.
1064 * The socket lock must be held while it's corked.
1067 if (likely(up->pending)) {
1068 if (unlikely(up->pending != AF_INET)) {
1072 goto do_append_data;
1076 ulen += sizeof(struct udphdr);
1079 * Get and verify the address.
1082 if (msg->msg_namelen < sizeof(*usin))
1084 if (usin->sin_family != AF_INET) {
1085 if (usin->sin_family != AF_UNSPEC)
1086 return -EAFNOSUPPORT;
1089 daddr = usin->sin_addr.s_addr;
1090 dport = usin->sin_port;
1094 if (sk->sk_state != TCP_ESTABLISHED)
1095 return -EDESTADDRREQ;
1096 daddr = inet->inet_daddr;
1097 dport = inet->inet_dport;
1098 /* Open fast path for connected socket.
1099 Route will not be used, if at least one option is set.
1104 ipcm_init_sk(&ipc, inet);
1105 ipc.gso_size = up->gso_size;
1107 if (msg->msg_controllen) {
1108 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1110 err = ip_cmsg_send(sk, msg, &ipc,
1111 sk->sk_family == AF_INET6);
1112 if (unlikely(err < 0)) {
1121 struct ip_options_rcu *inet_opt;
1124 inet_opt = rcu_dereference(inet->inet_opt);
1126 memcpy(&opt_copy, inet_opt,
1127 sizeof(*inet_opt) + inet_opt->opt.optlen);
1128 ipc.opt = &opt_copy.opt;
1133 if (cgroup_bpf_enabled(BPF_CGROUP_UDP4_SENDMSG) && !connected) {
1134 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1135 (struct sockaddr *)usin, &ipc.addr);
1139 if (usin->sin_port == 0) {
1140 /* BPF program set invalid port. Reject it. */
1144 daddr = usin->sin_addr.s_addr;
1145 dport = usin->sin_port;
1150 ipc.addr = faddr = daddr;
1152 if (ipc.opt && ipc.opt->opt.srr) {
1157 faddr = ipc.opt->opt.faddr;
1160 tos = get_rttos(&ipc, inet);
1161 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1162 (msg->msg_flags & MSG_DONTROUTE) ||
1163 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1168 if (ipv4_is_multicast(daddr)) {
1169 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1170 ipc.oif = inet->mc_index;
1172 saddr = inet->mc_addr;
1174 } else if (!ipc.oif) {
1175 ipc.oif = inet->uc_index;
1176 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1177 /* oif is set, packet is to local broadcast and
1178 * uc_index is set. oif is most likely set
1179 * by sk_bound_dev_if. If uc_index != oif check if the
1180 * oif is an L3 master and uc_index is an L3 slave.
1181 * If so, we want to allow the send using the uc_index.
1183 if (ipc.oif != inet->uc_index &&
1184 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1186 ipc.oif = inet->uc_index;
1191 rt = (struct rtable *)sk_dst_check(sk, 0);
1194 struct net *net = sock_net(sk);
1195 __u8 flow_flags = inet_sk_flowi_flags(sk);
1199 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1200 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1202 faddr, saddr, dport, inet->inet_sport,
1205 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1206 rt = ip_route_output_flow(net, fl4, sk);
1210 if (err == -ENETUNREACH)
1211 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1216 if ((rt->rt_flags & RTCF_BROADCAST) &&
1217 !sock_flag(sk, SOCK_BROADCAST))
1220 sk_dst_set(sk, dst_clone(&rt->dst));
1223 if (msg->msg_flags&MSG_CONFIRM)
1229 daddr = ipc.addr = fl4->daddr;
1231 /* Lockless fast path for the non-corking case. */
1233 struct inet_cork cork;
1235 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1236 sizeof(struct udphdr), &ipc, &rt,
1237 &cork, msg->msg_flags);
1239 if (!IS_ERR_OR_NULL(skb))
1240 err = udp_send_skb(skb, fl4, &cork);
1245 if (unlikely(up->pending)) {
1246 /* The socket is already corked while preparing it. */
1247 /* ... which is an evident application bug. --ANK */
1250 net_dbg_ratelimited("socket already corked\n");
1255 * Now cork the socket to pend data.
1257 fl4 = &inet->cork.fl.u.ip4;
1260 fl4->fl4_dport = dport;
1261 fl4->fl4_sport = inet->inet_sport;
1262 up->pending = AF_INET;
1266 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1267 sizeof(struct udphdr), &ipc, &rt,
1268 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1270 udp_flush_pending_frames(sk);
1272 err = udp_push_pending_frames(sk);
1273 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1285 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1286 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1287 * we don't have a good statistic (IpOutDiscards but it can be too many
1288 * things). We could add another new stat but at least for now that
1289 * seems like overkill.
1291 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1292 UDP_INC_STATS(sock_net(sk),
1293 UDP_MIB_SNDBUFERRORS, is_udplite);
1298 if (msg->msg_flags & MSG_PROBE)
1299 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1300 if (!(msg->msg_flags&MSG_PROBE) || len)
1301 goto back_from_confirm;
1305 EXPORT_SYMBOL(udp_sendmsg);
1307 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1308 size_t size, int flags)
1310 struct inet_sock *inet = inet_sk(sk);
1311 struct udp_sock *up = udp_sk(sk);
1314 if (flags & MSG_SENDPAGE_NOTLAST)
1318 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1320 /* Call udp_sendmsg to specify destination address which
1321 * sendpage interface can't pass.
1322 * This will succeed only when the socket is connected.
1324 ret = udp_sendmsg(sk, &msg, 0);
1331 if (unlikely(!up->pending)) {
1334 net_dbg_ratelimited("cork failed\n");
1338 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1339 page, offset, size, flags);
1340 if (ret == -EOPNOTSUPP) {
1342 return sock_no_sendpage(sk->sk_socket, page, offset,
1346 udp_flush_pending_frames(sk);
1351 if (!(up->corkflag || (flags&MSG_MORE)))
1352 ret = udp_push_pending_frames(sk);
1360 #define UDP_SKB_IS_STATELESS 0x80000000
1362 /* all head states (dst, sk, nf conntrack) except skb extensions are
1363 * cleared by udp_rcv().
1365 * We need to preserve secpath, if present, to eventually process
1366 * IP_CMSG_PASSSEC at recvmsg() time.
1368 * Other extensions can be cleared.
1370 static bool udp_try_make_stateless(struct sk_buff *skb)
1372 if (!skb_has_extensions(skb))
1375 if (!secpath_exists(skb)) {
1383 static void udp_set_dev_scratch(struct sk_buff *skb)
1385 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1387 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1388 scratch->_tsize_state = skb->truesize;
1389 #if BITS_PER_LONG == 64
1390 scratch->len = skb->len;
1391 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1392 scratch->is_linear = !skb_is_nonlinear(skb);
1394 if (udp_try_make_stateless(skb))
1395 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1398 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1400 /* We come here after udp_lib_checksum_complete() returned 0.
1401 * This means that __skb_checksum_complete() might have
1402 * set skb->csum_valid to 1.
1403 * On 64bit platforms, we can set csum_unnecessary
1404 * to true, but only if the skb is not shared.
1406 #if BITS_PER_LONG == 64
1407 if (!skb_shared(skb))
1408 udp_skb_scratch(skb)->csum_unnecessary = true;
1412 static int udp_skb_truesize(struct sk_buff *skb)
1414 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1417 static bool udp_skb_has_head_state(struct sk_buff *skb)
1419 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1422 /* fully reclaim rmem/fwd memory allocated for skb */
1423 static void udp_rmem_release(struct sock *sk, int size, int partial,
1424 bool rx_queue_lock_held)
1426 struct udp_sock *up = udp_sk(sk);
1427 struct sk_buff_head *sk_queue;
1430 if (likely(partial)) {
1431 up->forward_deficit += size;
1432 size = up->forward_deficit;
1433 if (size < (sk->sk_rcvbuf >> 2) &&
1434 !skb_queue_empty(&up->reader_queue))
1437 size += up->forward_deficit;
1439 up->forward_deficit = 0;
1441 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1442 * if the called don't held it already
1444 sk_queue = &sk->sk_receive_queue;
1445 if (!rx_queue_lock_held)
1446 spin_lock(&sk_queue->lock);
1449 sk->sk_forward_alloc += size;
1450 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1451 sk->sk_forward_alloc -= amt;
1454 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1456 atomic_sub(size, &sk->sk_rmem_alloc);
1458 /* this can save us from acquiring the rx queue lock on next receive */
1459 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1461 if (!rx_queue_lock_held)
1462 spin_unlock(&sk_queue->lock);
1465 /* Note: called with reader_queue.lock held.
1466 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1467 * This avoids a cache line miss while receive_queue lock is held.
1468 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1470 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1472 prefetch(&skb->data);
1473 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1475 EXPORT_SYMBOL(udp_skb_destructor);
1477 /* as above, but the caller held the rx queue lock, too */
1478 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1480 prefetch(&skb->data);
1481 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1484 /* Idea of busylocks is to let producers grab an extra spinlock
1485 * to relieve pressure on the receive_queue spinlock shared by consumer.
1486 * Under flood, this means that only one producer can be in line
1487 * trying to acquire the receive_queue spinlock.
1488 * These busylock can be allocated on a per cpu manner, instead of a
1489 * per socket one (that would consume a cache line per socket)
1491 static int udp_busylocks_log __read_mostly;
1492 static spinlock_t *udp_busylocks __read_mostly;
1494 static spinlock_t *busylock_acquire(void *ptr)
1498 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1503 static void busylock_release(spinlock_t *busy)
1509 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1511 struct sk_buff_head *list = &sk->sk_receive_queue;
1512 int rmem, delta, amt, err = -ENOMEM;
1513 spinlock_t *busy = NULL;
1516 /* try to avoid the costly atomic add/sub pair when the receive
1517 * queue is full; always allow at least a packet
1519 rmem = atomic_read(&sk->sk_rmem_alloc);
1520 if (rmem > sk->sk_rcvbuf)
1523 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1524 * having linear skbs :
1525 * - Reduce memory overhead and thus increase receive queue capacity
1526 * - Less cache line misses at copyout() time
1527 * - Less work at consume_skb() (less alien page frag freeing)
1529 if (rmem > (sk->sk_rcvbuf >> 1)) {
1532 busy = busylock_acquire(sk);
1534 size = skb->truesize;
1535 udp_set_dev_scratch(skb);
1537 /* we drop only if the receive buf is full and the receive
1538 * queue contains some other skb
1540 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1541 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1544 spin_lock(&list->lock);
1545 if (size >= sk->sk_forward_alloc) {
1546 amt = sk_mem_pages(size);
1547 delta = amt << SK_MEM_QUANTUM_SHIFT;
1548 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1550 spin_unlock(&list->lock);
1554 sk->sk_forward_alloc += delta;
1557 sk->sk_forward_alloc -= size;
1559 /* no need to setup a destructor, we will explicitly release the
1560 * forward allocated memory on dequeue
1562 sock_skb_set_dropcount(sk, skb);
1564 __skb_queue_tail(list, skb);
1565 spin_unlock(&list->lock);
1567 if (!sock_flag(sk, SOCK_DEAD))
1568 sk->sk_data_ready(sk);
1570 busylock_release(busy);
1574 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1577 atomic_inc(&sk->sk_drops);
1578 busylock_release(busy);
1581 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1583 void udp_destruct_sock(struct sock *sk)
1585 /* reclaim completely the forward allocated memory */
1586 struct udp_sock *up = udp_sk(sk);
1587 unsigned int total = 0;
1588 struct sk_buff *skb;
1590 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1591 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1592 total += skb->truesize;
1595 udp_rmem_release(sk, total, 0, true);
1597 inet_sock_destruct(sk);
1599 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1601 int udp_init_sock(struct sock *sk)
1603 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1604 sk->sk_destruct = udp_destruct_sock;
1607 EXPORT_SYMBOL_GPL(udp_init_sock);
1609 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1611 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1612 bool slow = lock_sock_fast(sk);
1614 sk_peek_offset_bwd(sk, len);
1615 unlock_sock_fast(sk, slow);
1618 if (!skb_unref(skb))
1621 /* In the more common cases we cleared the head states previously,
1622 * see __udp_queue_rcv_skb().
1624 if (unlikely(udp_skb_has_head_state(skb)))
1625 skb_release_head_state(skb);
1626 __consume_stateless_skb(skb);
1628 EXPORT_SYMBOL_GPL(skb_consume_udp);
1630 static struct sk_buff *__first_packet_length(struct sock *sk,
1631 struct sk_buff_head *rcvq,
1634 struct sk_buff *skb;
1636 while ((skb = skb_peek(rcvq)) != NULL) {
1637 if (udp_lib_checksum_complete(skb)) {
1638 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1640 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1642 atomic_inc(&sk->sk_drops);
1643 __skb_unlink(skb, rcvq);
1644 *total += skb->truesize;
1647 udp_skb_csum_unnecessary_set(skb);
1655 * first_packet_length - return length of first packet in receive queue
1658 * Drops all bad checksum frames, until a valid one is found.
1659 * Returns the length of found skb, or -1 if none is found.
1661 static int first_packet_length(struct sock *sk)
1663 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1664 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1665 struct sk_buff *skb;
1669 spin_lock_bh(&rcvq->lock);
1670 skb = __first_packet_length(sk, rcvq, &total);
1671 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1672 spin_lock(&sk_queue->lock);
1673 skb_queue_splice_tail_init(sk_queue, rcvq);
1674 spin_unlock(&sk_queue->lock);
1676 skb = __first_packet_length(sk, rcvq, &total);
1678 res = skb ? skb->len : -1;
1680 udp_rmem_release(sk, total, 1, false);
1681 spin_unlock_bh(&rcvq->lock);
1686 * IOCTL requests applicable to the UDP protocol
1689 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1694 int amount = sk_wmem_alloc_get(sk);
1696 return put_user(amount, (int __user *)arg);
1701 int amount = max_t(int, 0, first_packet_length(sk));
1703 return put_user(amount, (int __user *)arg);
1707 return -ENOIOCTLCMD;
1712 EXPORT_SYMBOL(udp_ioctl);
1714 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1715 int noblock, int *off, int *err)
1717 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1718 struct sk_buff_head *queue;
1719 struct sk_buff *last;
1723 queue = &udp_sk(sk)->reader_queue;
1724 flags |= noblock ? MSG_DONTWAIT : 0;
1725 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1727 struct sk_buff *skb;
1729 error = sock_error(sk);
1735 spin_lock_bh(&queue->lock);
1736 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1739 if (!(flags & MSG_PEEK))
1740 udp_skb_destructor(sk, skb);
1741 spin_unlock_bh(&queue->lock);
1745 if (skb_queue_empty_lockless(sk_queue)) {
1746 spin_unlock_bh(&queue->lock);
1750 /* refill the reader queue and walk it again
1751 * keep both queues locked to avoid re-acquiring
1752 * the sk_receive_queue lock if fwd memory scheduling
1755 spin_lock(&sk_queue->lock);
1756 skb_queue_splice_tail_init(sk_queue, queue);
1758 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1760 if (skb && !(flags & MSG_PEEK))
1761 udp_skb_dtor_locked(sk, skb);
1762 spin_unlock(&sk_queue->lock);
1763 spin_unlock_bh(&queue->lock);
1768 if (!sk_can_busy_loop(sk))
1771 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1772 } while (!skb_queue_empty_lockless(sk_queue));
1774 /* sk_queue is empty, reader_queue may contain peeked packets */
1776 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1778 (struct sk_buff *)sk_queue));
1783 EXPORT_SYMBOL(__skb_recv_udp);
1785 int udp_read_sock(struct sock *sk, read_descriptor_t *desc,
1786 sk_read_actor_t recv_actor)
1791 struct sk_buff *skb;
1794 skb = skb_recv_udp(sk, 0, 1, &err);
1797 used = recv_actor(desc, skb, 0, skb->len);
1802 } else if (used <= skb->len) {
1812 EXPORT_SYMBOL(udp_read_sock);
1815 * This should be easy, if there is something there we
1816 * return it, otherwise we block.
1819 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1820 int flags, int *addr_len)
1822 struct inet_sock *inet = inet_sk(sk);
1823 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1824 struct sk_buff *skb;
1825 unsigned int ulen, copied;
1826 int off, err, peeking = flags & MSG_PEEK;
1827 int is_udplite = IS_UDPLITE(sk);
1828 bool checksum_valid = false;
1830 if (flags & MSG_ERRQUEUE)
1831 return ip_recv_error(sk, msg, len, addr_len);
1834 off = sk_peek_offset(sk, flags);
1835 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1839 ulen = udp_skb_len(skb);
1841 if (copied > ulen - off)
1842 copied = ulen - off;
1843 else if (copied < ulen)
1844 msg->msg_flags |= MSG_TRUNC;
1847 * If checksum is needed at all, try to do it while copying the
1848 * data. If the data is truncated, or if we only want a partial
1849 * coverage checksum (UDP-Lite), do it before the copy.
1852 if (copied < ulen || peeking ||
1853 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1854 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1855 !__udp_lib_checksum_complete(skb);
1856 if (!checksum_valid)
1860 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1861 if (udp_skb_is_linear(skb))
1862 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1864 err = skb_copy_datagram_msg(skb, off, msg, copied);
1866 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1872 if (unlikely(err)) {
1874 atomic_inc(&sk->sk_drops);
1875 UDP_INC_STATS(sock_net(sk),
1876 UDP_MIB_INERRORS, is_udplite);
1883 UDP_INC_STATS(sock_net(sk),
1884 UDP_MIB_INDATAGRAMS, is_udplite);
1886 sock_recv_ts_and_drops(msg, sk, skb);
1888 /* Copy the address. */
1890 sin->sin_family = AF_INET;
1891 sin->sin_port = udp_hdr(skb)->source;
1892 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1893 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1894 *addr_len = sizeof(*sin);
1896 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1897 (struct sockaddr *)sin);
1900 if (udp_sk(sk)->gro_enabled)
1901 udp_cmsg_recv(msg, sk, skb);
1903 if (inet->cmsg_flags)
1904 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1907 if (flags & MSG_TRUNC)
1910 skb_consume_udp(sk, skb, peeking ? -err : err);
1914 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1915 udp_skb_destructor)) {
1916 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1917 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1921 /* starting over for a new packet, but check if we need to yield */
1923 msg->msg_flags &= ~MSG_TRUNC;
1927 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1929 /* This check is replicated from __ip4_datagram_connect() and
1930 * intended to prevent BPF program called below from accessing bytes
1931 * that are out of the bound specified by user in addr_len.
1933 if (addr_len < sizeof(struct sockaddr_in))
1936 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1938 EXPORT_SYMBOL(udp_pre_connect);
1940 int __udp_disconnect(struct sock *sk, int flags)
1942 struct inet_sock *inet = inet_sk(sk);
1944 * 1003.1g - break association.
1947 sk->sk_state = TCP_CLOSE;
1948 inet->inet_daddr = 0;
1949 inet->inet_dport = 0;
1950 sock_rps_reset_rxhash(sk);
1951 sk->sk_bound_dev_if = 0;
1952 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1953 inet_reset_saddr(sk);
1954 if (sk->sk_prot->rehash &&
1955 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1956 sk->sk_prot->rehash(sk);
1959 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1960 sk->sk_prot->unhash(sk);
1961 inet->inet_sport = 0;
1966 EXPORT_SYMBOL(__udp_disconnect);
1968 int udp_disconnect(struct sock *sk, int flags)
1971 __udp_disconnect(sk, flags);
1975 EXPORT_SYMBOL(udp_disconnect);
1977 void udp_lib_unhash(struct sock *sk)
1979 if (sk_hashed(sk)) {
1980 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1981 struct udp_hslot *hslot, *hslot2;
1983 hslot = udp_hashslot(udptable, sock_net(sk),
1984 udp_sk(sk)->udp_port_hash);
1985 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1987 spin_lock_bh(&hslot->lock);
1988 if (rcu_access_pointer(sk->sk_reuseport_cb))
1989 reuseport_detach_sock(sk);
1990 if (sk_del_node_init_rcu(sk)) {
1992 inet_sk(sk)->inet_num = 0;
1993 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1995 spin_lock(&hslot2->lock);
1996 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1998 spin_unlock(&hslot2->lock);
2000 spin_unlock_bh(&hslot->lock);
2003 EXPORT_SYMBOL(udp_lib_unhash);
2006 * inet_rcv_saddr was changed, we must rehash secondary hash
2008 void udp_lib_rehash(struct sock *sk, u16 newhash)
2010 if (sk_hashed(sk)) {
2011 struct udp_table *udptable = sk->sk_prot->h.udp_table;
2012 struct udp_hslot *hslot, *hslot2, *nhslot2;
2014 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2015 nhslot2 = udp_hashslot2(udptable, newhash);
2016 udp_sk(sk)->udp_portaddr_hash = newhash;
2018 if (hslot2 != nhslot2 ||
2019 rcu_access_pointer(sk->sk_reuseport_cb)) {
2020 hslot = udp_hashslot(udptable, sock_net(sk),
2021 udp_sk(sk)->udp_port_hash);
2022 /* we must lock primary chain too */
2023 spin_lock_bh(&hslot->lock);
2024 if (rcu_access_pointer(sk->sk_reuseport_cb))
2025 reuseport_detach_sock(sk);
2027 if (hslot2 != nhslot2) {
2028 spin_lock(&hslot2->lock);
2029 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2031 spin_unlock(&hslot2->lock);
2033 spin_lock(&nhslot2->lock);
2034 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2037 spin_unlock(&nhslot2->lock);
2040 spin_unlock_bh(&hslot->lock);
2044 EXPORT_SYMBOL(udp_lib_rehash);
2046 void udp_v4_rehash(struct sock *sk)
2048 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2049 inet_sk(sk)->inet_rcv_saddr,
2050 inet_sk(sk)->inet_num);
2051 udp_lib_rehash(sk, new_hash);
2054 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2058 if (inet_sk(sk)->inet_daddr) {
2059 sock_rps_save_rxhash(sk, skb);
2060 sk_mark_napi_id(sk, skb);
2061 sk_incoming_cpu_update(sk);
2063 sk_mark_napi_id_once(sk, skb);
2066 rc = __udp_enqueue_schedule_skb(sk, skb);
2068 int is_udplite = IS_UDPLITE(sk);
2070 /* Note that an ENOMEM error is charged twice */
2072 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2075 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2077 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2079 trace_udp_fail_queue_rcv_skb(rc, sk);
2089 * >0: "udp encap" protocol resubmission
2091 * Note that in the success and error cases, the skb is assumed to
2092 * have either been requeued or freed.
2094 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2096 struct udp_sock *up = udp_sk(sk);
2097 int is_udplite = IS_UDPLITE(sk);
2100 * Charge it to the socket, dropping if the queue is full.
2102 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2106 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2107 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2110 * This is an encapsulation socket so pass the skb to
2111 * the socket's udp_encap_rcv() hook. Otherwise, just
2112 * fall through and pass this up the UDP socket.
2113 * up->encap_rcv() returns the following value:
2114 * =0 if skb was successfully passed to the encap
2115 * handler or was discarded by it.
2116 * >0 if skb should be passed on to UDP.
2117 * <0 if skb should be resubmitted as proto -N
2120 /* if we're overly short, let UDP handle it */
2121 encap_rcv = READ_ONCE(up->encap_rcv);
2125 /* Verify checksum before giving to encap */
2126 if (udp_lib_checksum_complete(skb))
2129 ret = encap_rcv(sk, skb);
2131 __UDP_INC_STATS(sock_net(sk),
2132 UDP_MIB_INDATAGRAMS,
2138 /* FALLTHROUGH -- it's a UDP Packet */
2142 * UDP-Lite specific tests, ignored on UDP sockets
2144 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2147 * MIB statistics other than incrementing the error count are
2148 * disabled for the following two types of errors: these depend
2149 * on the application settings, not on the functioning of the
2150 * protocol stack as such.
2152 * RFC 3828 here recommends (sec 3.3): "There should also be a
2153 * way ... to ... at least let the receiving application block
2154 * delivery of packets with coverage values less than a value
2155 * provided by the application."
2157 if (up->pcrlen == 0) { /* full coverage was set */
2158 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2159 UDP_SKB_CB(skb)->cscov, skb->len);
2162 /* The next case involves violating the min. coverage requested
2163 * by the receiver. This is subtle: if receiver wants x and x is
2164 * greater than the buffersize/MTU then receiver will complain
2165 * that it wants x while sender emits packets of smaller size y.
2166 * Therefore the above ...()->partial_cov statement is essential.
2168 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2169 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2170 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2175 prefetch(&sk->sk_rmem_alloc);
2176 if (rcu_access_pointer(sk->sk_filter) &&
2177 udp_lib_checksum_complete(skb))
2180 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2183 udp_csum_pull_header(skb);
2185 ipv4_pktinfo_prepare(sk, skb);
2186 return __udp_queue_rcv_skb(sk, skb);
2189 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2191 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2192 atomic_inc(&sk->sk_drops);
2197 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2199 struct sk_buff *next, *segs;
2202 if (likely(!udp_unexpected_gso(sk, skb)))
2203 return udp_queue_rcv_one_skb(sk, skb);
2205 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2206 __skb_push(skb, -skb_mac_offset(skb));
2207 segs = udp_rcv_segment(sk, skb, true);
2208 skb_list_walk_safe(segs, skb, next) {
2209 __skb_pull(skb, skb_transport_offset(skb));
2211 udp_post_segment_fix_csum(skb);
2212 ret = udp_queue_rcv_one_skb(sk, skb);
2214 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2219 /* For TCP sockets, sk_rx_dst is protected by socket lock
2220 * For UDP, we use xchg() to guard against concurrent changes.
2222 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2224 struct dst_entry *old;
2226 if (dst_hold_safe(dst)) {
2227 old = xchg(&sk->sk_rx_dst, dst);
2233 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2236 * Multicasts and broadcasts go to each listener.
2238 * Note: called only from the BH handler context.
2240 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2242 __be32 saddr, __be32 daddr,
2243 struct udp_table *udptable,
2246 struct sock *sk, *first = NULL;
2247 unsigned short hnum = ntohs(uh->dest);
2248 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2249 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2250 unsigned int offset = offsetof(typeof(*sk), sk_node);
2251 int dif = skb->dev->ifindex;
2252 int sdif = inet_sdif(skb);
2253 struct hlist_node *node;
2254 struct sk_buff *nskb;
2257 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2259 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2261 hslot = &udptable->hash2[hash2];
2262 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2265 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2266 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2267 uh->source, saddr, dif, sdif, hnum))
2274 nskb = skb_clone(skb, GFP_ATOMIC);
2276 if (unlikely(!nskb)) {
2277 atomic_inc(&sk->sk_drops);
2278 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2280 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2284 if (udp_queue_rcv_skb(sk, nskb) > 0)
2288 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2289 if (use_hash2 && hash2 != hash2_any) {
2295 if (udp_queue_rcv_skb(first, skb) > 0)
2299 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2300 proto == IPPROTO_UDPLITE);
2305 /* Initialize UDP checksum. If exited with zero value (success),
2306 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2307 * Otherwise, csum completion requires checksumming packet body,
2308 * including udp header and folding it to skb->csum.
2310 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2315 UDP_SKB_CB(skb)->partial_cov = 0;
2316 UDP_SKB_CB(skb)->cscov = skb->len;
2318 if (proto == IPPROTO_UDPLITE) {
2319 err = udplite_checksum_init(skb, uh);
2323 if (UDP_SKB_CB(skb)->partial_cov) {
2324 skb->csum = inet_compute_pseudo(skb, proto);
2329 /* Note, we are only interested in != 0 or == 0, thus the
2332 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2333 inet_compute_pseudo);
2337 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2338 /* If SW calculated the value, we know it's bad */
2339 if (skb->csum_complete_sw)
2342 /* HW says the value is bad. Let's validate that.
2343 * skb->csum is no longer the full packet checksum,
2344 * so don't treat it as such.
2346 skb_checksum_complete_unset(skb);
2352 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2353 * return code conversion for ip layer consumption
2355 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2360 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2361 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2363 ret = udp_queue_rcv_skb(sk, skb);
2365 /* a return value > 0 means to resubmit the input, but
2366 * it wants the return to be -protocol, or 0
2374 * All we need to do is get the socket, and then do a checksum.
2377 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2382 unsigned short ulen;
2383 struct rtable *rt = skb_rtable(skb);
2384 __be32 saddr, daddr;
2385 struct net *net = dev_net(skb->dev);
2389 * Validate the packet.
2391 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2392 goto drop; /* No space for header. */
2395 ulen = ntohs(uh->len);
2396 saddr = ip_hdr(skb)->saddr;
2397 daddr = ip_hdr(skb)->daddr;
2399 if (ulen > skb->len)
2402 if (proto == IPPROTO_UDP) {
2403 /* UDP validates ulen. */
2404 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2409 if (udp4_csum_init(skb, uh, proto))
2412 sk = skb_steal_sock(skb, &refcounted);
2414 struct dst_entry *dst = skb_dst(skb);
2417 if (unlikely(sk->sk_rx_dst != dst))
2418 udp_sk_rx_dst_set(sk, dst);
2420 ret = udp_unicast_rcv_skb(sk, skb, uh);
2426 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2427 return __udp4_lib_mcast_deliver(net, skb, uh,
2428 saddr, daddr, udptable, proto);
2430 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2432 return udp_unicast_rcv_skb(sk, skb, uh);
2434 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2438 /* No socket. Drop packet silently, if checksum is wrong */
2439 if (udp_lib_checksum_complete(skb))
2442 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2443 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2446 * Hmm. We got an UDP packet to a port to which we
2447 * don't wanna listen. Ignore it.
2453 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2454 proto == IPPROTO_UDPLITE ? "Lite" : "",
2455 &saddr, ntohs(uh->source),
2457 &daddr, ntohs(uh->dest));
2462 * RFC1122: OK. Discards the bad packet silently (as far as
2463 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2465 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2466 proto == IPPROTO_UDPLITE ? "Lite" : "",
2467 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2469 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2471 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2476 /* We can only early demux multicast if there is a single matching socket.
2477 * If more than one socket found returns NULL
2479 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2480 __be16 loc_port, __be32 loc_addr,
2481 __be16 rmt_port, __be32 rmt_addr,
2484 struct sock *sk, *result;
2485 unsigned short hnum = ntohs(loc_port);
2486 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2487 struct udp_hslot *hslot = &udp_table.hash[slot];
2489 /* Do not bother scanning a too big list */
2490 if (hslot->count > 10)
2494 sk_for_each_rcu(sk, &hslot->head) {
2495 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2496 rmt_port, rmt_addr, dif, sdif, hnum)) {
2506 /* For unicast we should only early demux connected sockets or we can
2507 * break forwarding setups. The chains here can be long so only check
2508 * if the first socket is an exact match and if not move on.
2510 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2511 __be16 loc_port, __be32 loc_addr,
2512 __be16 rmt_port, __be32 rmt_addr,
2515 unsigned short hnum = ntohs(loc_port);
2516 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2517 unsigned int slot2 = hash2 & udp_table.mask;
2518 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2519 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2520 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2523 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2524 if (INET_MATCH(sk, net, acookie, rmt_addr,
2525 loc_addr, ports, dif, sdif))
2527 /* Only check first socket in chain */
2533 int udp_v4_early_demux(struct sk_buff *skb)
2535 struct net *net = dev_net(skb->dev);
2536 struct in_device *in_dev = NULL;
2537 const struct iphdr *iph;
2538 const struct udphdr *uh;
2539 struct sock *sk = NULL;
2540 struct dst_entry *dst;
2541 int dif = skb->dev->ifindex;
2542 int sdif = inet_sdif(skb);
2545 /* validate the packet */
2546 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2552 if (skb->pkt_type == PACKET_MULTICAST) {
2553 in_dev = __in_dev_get_rcu(skb->dev);
2558 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2563 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2564 uh->source, iph->saddr,
2566 } else if (skb->pkt_type == PACKET_HOST) {
2567 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2568 uh->source, iph->saddr, dif, sdif);
2571 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2575 skb->destructor = sock_efree;
2576 dst = READ_ONCE(sk->sk_rx_dst);
2579 dst = dst_check(dst, 0);
2583 /* set noref for now.
2584 * any place which wants to hold dst has to call
2587 skb_dst_set_noref(skb, dst);
2589 /* for unconnected multicast sockets we need to validate
2590 * the source on each packet
2592 if (!inet_sk(sk)->inet_daddr && in_dev)
2593 return ip_mc_validate_source(skb, iph->daddr,
2595 iph->tos & IPTOS_RT_MASK,
2596 skb->dev, in_dev, &itag);
2601 int udp_rcv(struct sk_buff *skb)
2603 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2606 void udp_destroy_sock(struct sock *sk)
2608 struct udp_sock *up = udp_sk(sk);
2609 bool slow = lock_sock_fast(sk);
2610 udp_flush_pending_frames(sk);
2611 unlock_sock_fast(sk, slow);
2612 if (static_branch_unlikely(&udp_encap_needed_key)) {
2613 if (up->encap_type) {
2614 void (*encap_destroy)(struct sock *sk);
2615 encap_destroy = READ_ONCE(up->encap_destroy);
2619 if (up->encap_enabled)
2620 static_branch_dec(&udp_encap_needed_key);
2625 * Socket option code for UDP
2627 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2628 sockptr_t optval, unsigned int optlen,
2629 int (*push_pending_frames)(struct sock *))
2631 struct udp_sock *up = udp_sk(sk);
2634 int is_udplite = IS_UDPLITE(sk);
2636 if (optlen < sizeof(int))
2639 if (copy_from_sockptr(&val, optval, sizeof(val)))
2642 valbool = val ? 1 : 0;
2651 push_pending_frames(sk);
2660 case UDP_ENCAP_ESPINUDP:
2661 case UDP_ENCAP_ESPINUDP_NON_IKE:
2662 #if IS_ENABLED(CONFIG_IPV6)
2663 if (sk->sk_family == AF_INET6)
2664 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2667 up->encap_rcv = xfrm4_udp_encap_rcv;
2670 case UDP_ENCAP_L2TPINUDP:
2671 up->encap_type = val;
2673 udp_tunnel_encap_enable(sk->sk_socket);
2682 case UDP_NO_CHECK6_TX:
2683 up->no_check6_tx = valbool;
2686 case UDP_NO_CHECK6_RX:
2687 up->no_check6_rx = valbool;
2691 if (val < 0 || val > USHRT_MAX)
2699 /* when enabling GRO, accept the related GSO packet type */
2701 udp_tunnel_encap_enable(sk->sk_socket);
2702 up->gro_enabled = valbool;
2703 up->accept_udp_l4 = valbool;
2708 * UDP-Lite's partial checksum coverage (RFC 3828).
2710 /* The sender sets actual checksum coverage length via this option.
2711 * The case coverage > packet length is handled by send module. */
2712 case UDPLITE_SEND_CSCOV:
2713 if (!is_udplite) /* Disable the option on UDP sockets */
2714 return -ENOPROTOOPT;
2715 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2717 else if (val > USHRT_MAX)
2720 up->pcflag |= UDPLITE_SEND_CC;
2723 /* The receiver specifies a minimum checksum coverage value. To make
2724 * sense, this should be set to at least 8 (as done below). If zero is
2725 * used, this again means full checksum coverage. */
2726 case UDPLITE_RECV_CSCOV:
2727 if (!is_udplite) /* Disable the option on UDP sockets */
2728 return -ENOPROTOOPT;
2729 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2731 else if (val > USHRT_MAX)
2734 up->pcflag |= UDPLITE_RECV_CC;
2744 EXPORT_SYMBOL(udp_lib_setsockopt);
2746 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2747 unsigned int optlen)
2749 if (level == SOL_UDP || level == SOL_UDPLITE)
2750 return udp_lib_setsockopt(sk, level, optname,
2752 udp_push_pending_frames);
2753 return ip_setsockopt(sk, level, optname, optval, optlen);
2756 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2757 char __user *optval, int __user *optlen)
2759 struct udp_sock *up = udp_sk(sk);
2762 if (get_user(len, optlen))
2765 len = min_t(unsigned int, len, sizeof(int));
2776 val = up->encap_type;
2779 case UDP_NO_CHECK6_TX:
2780 val = up->no_check6_tx;
2783 case UDP_NO_CHECK6_RX:
2784 val = up->no_check6_rx;
2792 val = up->gro_enabled;
2795 /* The following two cannot be changed on UDP sockets, the return is
2796 * always 0 (which corresponds to the full checksum coverage of UDP). */
2797 case UDPLITE_SEND_CSCOV:
2801 case UDPLITE_RECV_CSCOV:
2806 return -ENOPROTOOPT;
2809 if (put_user(len, optlen))
2811 if (copy_to_user(optval, &val, len))
2815 EXPORT_SYMBOL(udp_lib_getsockopt);
2817 int udp_getsockopt(struct sock *sk, int level, int optname,
2818 char __user *optval, int __user *optlen)
2820 if (level == SOL_UDP || level == SOL_UDPLITE)
2821 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2822 return ip_getsockopt(sk, level, optname, optval, optlen);
2826 * udp_poll - wait for a UDP event.
2827 * @file: - file struct
2829 * @wait: - poll table
2831 * This is same as datagram poll, except for the special case of
2832 * blocking sockets. If application is using a blocking fd
2833 * and a packet with checksum error is in the queue;
2834 * then it could get return from select indicating data available
2835 * but then block when reading it. Add special case code
2836 * to work around these arguably broken applications.
2838 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2840 __poll_t mask = datagram_poll(file, sock, wait);
2841 struct sock *sk = sock->sk;
2843 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2844 mask |= EPOLLIN | EPOLLRDNORM;
2846 /* Check for false positives due to checksum errors */
2847 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2848 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2849 mask &= ~(EPOLLIN | EPOLLRDNORM);
2854 EXPORT_SYMBOL(udp_poll);
2856 int udp_abort(struct sock *sk, int err)
2861 sk->sk_error_report(sk);
2862 __udp_disconnect(sk, 0);
2868 EXPORT_SYMBOL_GPL(udp_abort);
2870 struct proto udp_prot = {
2872 .owner = THIS_MODULE,
2873 .close = udp_lib_close,
2874 .pre_connect = udp_pre_connect,
2875 .connect = ip4_datagram_connect,
2876 .disconnect = udp_disconnect,
2878 .init = udp_init_sock,
2879 .destroy = udp_destroy_sock,
2880 .setsockopt = udp_setsockopt,
2881 .getsockopt = udp_getsockopt,
2882 .sendmsg = udp_sendmsg,
2883 .recvmsg = udp_recvmsg,
2884 .sendpage = udp_sendpage,
2885 .release_cb = ip4_datagram_release_cb,
2886 .hash = udp_lib_hash,
2887 .unhash = udp_lib_unhash,
2888 .rehash = udp_v4_rehash,
2889 .get_port = udp_v4_get_port,
2890 #ifdef CONFIG_BPF_SYSCALL
2891 .psock_update_sk_prot = udp_bpf_update_proto,
2893 .memory_allocated = &udp_memory_allocated,
2894 .sysctl_mem = sysctl_udp_mem,
2895 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2896 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2897 .obj_size = sizeof(struct udp_sock),
2898 .h.udp_table = &udp_table,
2899 .diag_destroy = udp_abort,
2901 EXPORT_SYMBOL(udp_prot);
2903 /* ------------------------------------------------------------------------ */
2904 #ifdef CONFIG_PROC_FS
2906 static struct sock *udp_get_first(struct seq_file *seq, int start)
2909 struct udp_seq_afinfo *afinfo;
2910 struct udp_iter_state *state = seq->private;
2911 struct net *net = seq_file_net(seq);
2913 if (state->bpf_seq_afinfo)
2914 afinfo = state->bpf_seq_afinfo;
2916 afinfo = PDE_DATA(file_inode(seq->file));
2918 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2920 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2922 if (hlist_empty(&hslot->head))
2925 spin_lock_bh(&hslot->lock);
2926 sk_for_each(sk, &hslot->head) {
2927 if (!net_eq(sock_net(sk), net))
2929 if (afinfo->family == AF_UNSPEC ||
2930 sk->sk_family == afinfo->family)
2933 spin_unlock_bh(&hslot->lock);
2940 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2942 struct udp_seq_afinfo *afinfo;
2943 struct udp_iter_state *state = seq->private;
2944 struct net *net = seq_file_net(seq);
2946 if (state->bpf_seq_afinfo)
2947 afinfo = state->bpf_seq_afinfo;
2949 afinfo = PDE_DATA(file_inode(seq->file));
2953 } while (sk && (!net_eq(sock_net(sk), net) ||
2954 (afinfo->family != AF_UNSPEC &&
2955 sk->sk_family != afinfo->family)));
2958 if (state->bucket <= afinfo->udp_table->mask)
2959 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2960 return udp_get_first(seq, state->bucket + 1);
2965 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2967 struct sock *sk = udp_get_first(seq, 0);
2970 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2972 return pos ? NULL : sk;
2975 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2977 struct udp_iter_state *state = seq->private;
2978 state->bucket = MAX_UDP_PORTS;
2980 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2982 EXPORT_SYMBOL(udp_seq_start);
2984 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2988 if (v == SEQ_START_TOKEN)
2989 sk = udp_get_idx(seq, 0);
2991 sk = udp_get_next(seq, v);
2996 EXPORT_SYMBOL(udp_seq_next);
2998 void udp_seq_stop(struct seq_file *seq, void *v)
3000 struct udp_seq_afinfo *afinfo;
3001 struct udp_iter_state *state = seq->private;
3003 if (state->bpf_seq_afinfo)
3004 afinfo = state->bpf_seq_afinfo;
3006 afinfo = PDE_DATA(file_inode(seq->file));
3008 if (state->bucket <= afinfo->udp_table->mask)
3009 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
3011 EXPORT_SYMBOL(udp_seq_stop);
3013 /* ------------------------------------------------------------------------ */
3014 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3017 struct inet_sock *inet = inet_sk(sp);
3018 __be32 dest = inet->inet_daddr;
3019 __be32 src = inet->inet_rcv_saddr;
3020 __u16 destp = ntohs(inet->inet_dport);
3021 __u16 srcp = ntohs(inet->inet_sport);
3023 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3024 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3025 bucket, src, srcp, dest, destp, sp->sk_state,
3026 sk_wmem_alloc_get(sp),
3029 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3031 refcount_read(&sp->sk_refcnt), sp,
3032 atomic_read(&sp->sk_drops));
3035 int udp4_seq_show(struct seq_file *seq, void *v)
3037 seq_setwidth(seq, 127);
3038 if (v == SEQ_START_TOKEN)
3039 seq_puts(seq, " sl local_address rem_address st tx_queue "
3040 "rx_queue tr tm->when retrnsmt uid timeout "
3041 "inode ref pointer drops");
3043 struct udp_iter_state *state = seq->private;
3045 udp4_format_sock(v, seq, state->bucket);
3051 #ifdef CONFIG_BPF_SYSCALL
3052 struct bpf_iter__udp {
3053 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3054 __bpf_md_ptr(struct udp_sock *, udp_sk);
3055 uid_t uid __aligned(8);
3056 int bucket __aligned(8);
3059 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3060 struct udp_sock *udp_sk, uid_t uid, int bucket)
3062 struct bpf_iter__udp ctx;
3064 meta->seq_num--; /* skip SEQ_START_TOKEN */
3066 ctx.udp_sk = udp_sk;
3068 ctx.bucket = bucket;
3069 return bpf_iter_run_prog(prog, &ctx);
3072 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3074 struct udp_iter_state *state = seq->private;
3075 struct bpf_iter_meta meta;
3076 struct bpf_prog *prog;
3077 struct sock *sk = v;
3080 if (v == SEQ_START_TOKEN)
3083 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3085 prog = bpf_iter_get_info(&meta, false);
3086 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3089 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3091 struct bpf_iter_meta meta;
3092 struct bpf_prog *prog;
3096 prog = bpf_iter_get_info(&meta, true);
3098 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3101 udp_seq_stop(seq, v);
3104 static const struct seq_operations bpf_iter_udp_seq_ops = {
3105 .start = udp_seq_start,
3106 .next = udp_seq_next,
3107 .stop = bpf_iter_udp_seq_stop,
3108 .show = bpf_iter_udp_seq_show,
3112 const struct seq_operations udp_seq_ops = {
3113 .start = udp_seq_start,
3114 .next = udp_seq_next,
3115 .stop = udp_seq_stop,
3116 .show = udp4_seq_show,
3118 EXPORT_SYMBOL(udp_seq_ops);
3120 static struct udp_seq_afinfo udp4_seq_afinfo = {
3122 .udp_table = &udp_table,
3125 static int __net_init udp4_proc_init_net(struct net *net)
3127 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3128 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3133 static void __net_exit udp4_proc_exit_net(struct net *net)
3135 remove_proc_entry("udp", net->proc_net);
3138 static struct pernet_operations udp4_net_ops = {
3139 .init = udp4_proc_init_net,
3140 .exit = udp4_proc_exit_net,
3143 int __init udp4_proc_init(void)
3145 return register_pernet_subsys(&udp4_net_ops);
3148 void udp4_proc_exit(void)
3150 unregister_pernet_subsys(&udp4_net_ops);
3152 #endif /* CONFIG_PROC_FS */
3154 static __initdata unsigned long uhash_entries;
3155 static int __init set_uhash_entries(char *str)
3162 ret = kstrtoul(str, 0, &uhash_entries);
3166 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3167 uhash_entries = UDP_HTABLE_SIZE_MIN;
3170 __setup("uhash_entries=", set_uhash_entries);
3172 void __init udp_table_init(struct udp_table *table, const char *name)
3176 table->hash = alloc_large_system_hash(name,
3177 2 * sizeof(struct udp_hslot),
3179 21, /* one slot per 2 MB */
3183 UDP_HTABLE_SIZE_MIN,
3186 table->hash2 = table->hash + (table->mask + 1);
3187 for (i = 0; i <= table->mask; i++) {
3188 INIT_HLIST_HEAD(&table->hash[i].head);
3189 table->hash[i].count = 0;
3190 spin_lock_init(&table->hash[i].lock);
3192 for (i = 0; i <= table->mask; i++) {
3193 INIT_HLIST_HEAD(&table->hash2[i].head);
3194 table->hash2[i].count = 0;
3195 spin_lock_init(&table->hash2[i].lock);
3199 u32 udp_flow_hashrnd(void)
3201 static u32 hashrnd __read_mostly;
3203 net_get_random_once(&hashrnd, sizeof(hashrnd));
3207 EXPORT_SYMBOL(udp_flow_hashrnd);
3209 static void __udp_sysctl_init(struct net *net)
3211 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3212 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3214 #ifdef CONFIG_NET_L3_MASTER_DEV
3215 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3219 static int __net_init udp_sysctl_init(struct net *net)
3221 __udp_sysctl_init(net);
3225 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3226 .init = udp_sysctl_init,
3229 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3230 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3231 struct udp_sock *udp_sk, uid_t uid, int bucket)
3233 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3235 struct udp_iter_state *st = priv_data;
3236 struct udp_seq_afinfo *afinfo;
3239 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3243 afinfo->family = AF_UNSPEC;
3244 afinfo->udp_table = &udp_table;
3245 st->bpf_seq_afinfo = afinfo;
3246 ret = bpf_iter_init_seq_net(priv_data, aux);
3252 static void bpf_iter_fini_udp(void *priv_data)
3254 struct udp_iter_state *st = priv_data;
3256 kfree(st->bpf_seq_afinfo);
3257 bpf_iter_fini_seq_net(priv_data);
3260 static const struct bpf_iter_seq_info udp_seq_info = {
3261 .seq_ops = &bpf_iter_udp_seq_ops,
3262 .init_seq_private = bpf_iter_init_udp,
3263 .fini_seq_private = bpf_iter_fini_udp,
3264 .seq_priv_size = sizeof(struct udp_iter_state),
3267 static struct bpf_iter_reg udp_reg_info = {
3269 .ctx_arg_info_size = 1,
3271 { offsetof(struct bpf_iter__udp, udp_sk),
3272 PTR_TO_BTF_ID_OR_NULL },
3274 .seq_info = &udp_seq_info,
3277 static void __init bpf_iter_register(void)
3279 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3280 if (bpf_iter_reg_target(&udp_reg_info))
3281 pr_warn("Warning: could not register bpf iterator udp\n");
3285 void __init udp_init(void)
3287 unsigned long limit;
3290 udp_table_init(&udp_table, "UDP");
3291 limit = nr_free_buffer_pages() / 8;
3292 limit = max(limit, 128UL);
3293 sysctl_udp_mem[0] = limit / 4 * 3;
3294 sysctl_udp_mem[1] = limit;
3295 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3297 __udp_sysctl_init(&init_net);
3299 /* 16 spinlocks per cpu */
3300 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3301 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3304 panic("UDP: failed to alloc udp_busylocks\n");
3305 for (i = 0; i < (1U << udp_busylocks_log); i++)
3306 spin_lock_init(udp_busylocks + i);
3308 if (register_pernet_subsys(&udp_sysctl_ops))
3309 panic("UDP: failed to init sysctl parameters.\n");
3311 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3312 bpf_iter_register();