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).
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
13 * Hirokazu Takahashi, <taka@valinux.co.jp>
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.
71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
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,
393 if (sk->sk_bound_dev_if)
396 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
401 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
402 const __u16 lport, const __be32 faddr,
405 static u32 udp_ehash_secret __read_mostly;
407 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
409 return __inet_ehashfn(laddr, lport, faddr, fport,
410 udp_ehash_secret + net_hash_mix(net));
413 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
415 __be32 saddr, __be16 sport,
416 __be32 daddr, unsigned short hnum)
418 struct sock *reuse_sk = NULL;
421 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
422 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
423 reuse_sk = reuseport_select_sock(sk, hash, skb,
424 sizeof(struct udphdr));
429 /* called with rcu_read_lock() */
430 static struct sock *udp4_lib_lookup2(struct net *net,
431 __be32 saddr, __be16 sport,
432 __be32 daddr, unsigned int hnum,
434 struct udp_hslot *hslot2,
437 struct sock *sk, *result;
442 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
443 score = compute_score(sk, net, saddr, sport,
444 daddr, hnum, dif, sdif);
445 if (score > badness) {
447 result = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
453 /* Fall back to scoring if group has connections */
454 if (!reuseport_has_conns(sk))
457 /* Reuseport logic returned an error, keep original score. */
461 badness = compute_score(result, net, saddr, sport,
462 daddr, hnum, dif, sdif);
469 static struct sock *udp4_lookup_run_bpf(struct net *net,
470 struct udp_table *udptable,
472 __be32 saddr, __be16 sport,
473 __be32 daddr, u16 hnum)
475 struct sock *sk, *reuse_sk;
478 if (udptable != &udp_table)
479 return NULL; /* only UDP is supported */
481 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP,
482 saddr, sport, daddr, hnum, &sk);
483 if (no_reuseport || IS_ERR_OR_NULL(sk))
486 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
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, int dif,
497 int sdif, struct udp_table *udptable, struct sk_buff *skb)
499 unsigned short hnum = ntohs(dport);
500 unsigned int hash2, slot2;
501 struct udp_hslot *hslot2;
502 struct sock *result, *sk;
504 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
505 slot2 = hash2 & udptable->mask;
506 hslot2 = &udptable->hash2[slot2];
508 /* Lookup connected or non-wildcard socket */
509 result = udp4_lib_lookup2(net, saddr, sport,
510 daddr, hnum, dif, sdif,
512 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
515 /* Lookup redirect from BPF */
516 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
517 sk = udp4_lookup_run_bpf(net, udptable, skb,
518 saddr, sport, daddr, hnum);
525 /* Got non-wildcard socket or error on first lookup */
529 /* Lookup wildcard sockets */
530 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
531 slot2 = hash2 & udptable->mask;
532 hslot2 = &udptable->hash2[slot2];
534 result = udp4_lib_lookup2(net, saddr, sport,
535 htonl(INADDR_ANY), hnum, dif, sdif,
542 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
544 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
545 __be16 sport, __be16 dport,
546 struct udp_table *udptable)
548 const struct iphdr *iph = ip_hdr(skb);
550 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
551 iph->daddr, dport, inet_iif(skb),
552 inet_sdif(skb), udptable, skb);
555 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
556 __be16 sport, __be16 dport)
558 const struct iphdr *iph = ip_hdr(skb);
560 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
561 iph->daddr, dport, inet_iif(skb),
562 inet_sdif(skb), &udp_table, NULL);
565 /* Must be called under rcu_read_lock().
566 * Does increment socket refcount.
568 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
569 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
570 __be32 daddr, __be16 dport, int dif)
574 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
575 dif, 0, &udp_table, NULL);
576 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
580 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
583 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
584 __be16 loc_port, __be32 loc_addr,
585 __be16 rmt_port, __be32 rmt_addr,
586 int dif, int sdif, unsigned short hnum)
588 struct inet_sock *inet = inet_sk(sk);
590 if (!net_eq(sock_net(sk), net) ||
591 udp_sk(sk)->udp_port_hash != hnum ||
592 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
593 (inet->inet_dport != rmt_port && inet->inet_dport) ||
594 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
595 ipv6_only_sock(sk) ||
596 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
598 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
603 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
604 void udp_encap_enable(void)
606 static_branch_inc(&udp_encap_needed_key);
608 EXPORT_SYMBOL(udp_encap_enable);
610 void udp_encap_disable(void)
612 static_branch_dec(&udp_encap_needed_key);
614 EXPORT_SYMBOL(udp_encap_disable);
616 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
617 * through error handlers in encapsulations looking for a match.
619 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
623 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
624 int (*handler)(struct sk_buff *skb, u32 info);
625 const struct ip_tunnel_encap_ops *encap;
627 encap = rcu_dereference(iptun_encaps[i]);
630 handler = encap->err_handler;
631 if (handler && !handler(skb, info))
638 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
639 * reversing source and destination port: this will match tunnels that force the
640 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
641 * lwtunnels might actually break this assumption by being configured with
642 * different destination ports on endpoints, in this case we won't be able to
643 * trace ICMP messages back to them.
645 * If this doesn't match any socket, probe tunnels with arbitrary destination
646 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
647 * we've sent packets to won't necessarily match the local destination port.
649 * Then ask the tunnel implementation to match the error against a valid
652 * Return an error if we can't find a match, the socket if we need further
653 * processing, zero otherwise.
655 static struct sock *__udp4_lib_err_encap(struct net *net,
656 const struct iphdr *iph,
658 struct udp_table *udptable,
660 struct sk_buff *skb, u32 info)
662 int (*lookup)(struct sock *sk, struct sk_buff *skb);
663 int network_offset, transport_offset;
666 network_offset = skb_network_offset(skb);
667 transport_offset = skb_transport_offset(skb);
669 /* Network header needs to point to the outer IPv4 header inside ICMP */
670 skb_reset_network_header(skb);
672 /* Transport header needs to point to the UDP header */
673 skb_set_transport_header(skb, iph->ihl << 2);
678 lookup = READ_ONCE(up->encap_err_lookup);
679 if (lookup && lookup(sk, skb))
685 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
686 iph->saddr, uh->dest, skb->dev->ifindex, 0,
691 lookup = READ_ONCE(up->encap_err_lookup);
692 if (!lookup || lookup(sk, skb))
698 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
700 skb_set_transport_header(skb, transport_offset);
701 skb_set_network_header(skb, network_offset);
707 * This routine is called by the ICMP module when it gets some
708 * sort of error condition. If err < 0 then the socket should
709 * be closed and the error returned to the user. If err > 0
710 * it's just the icmp type << 8 | icmp code.
711 * Header points to the ip header of the error packet. We move
712 * on past this. Then (as it used to claim before adjustment)
713 * header points to the first 8 bytes of the udp header. We need
714 * to find the appropriate port.
717 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
719 struct inet_sock *inet;
720 const struct iphdr *iph = (const struct iphdr *)skb->data;
721 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
722 const int type = icmp_hdr(skb)->type;
723 const int code = icmp_hdr(skb)->code;
728 struct net *net = dev_net(skb->dev);
730 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
731 iph->saddr, uh->source, skb->dev->ifindex,
732 inet_sdif(skb), udptable, NULL);
734 if (!sk || udp_sk(sk)->encap_type) {
735 /* No socket for error: try tunnels before discarding */
736 if (static_branch_unlikely(&udp_encap_needed_key)) {
737 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
742 sk = ERR_PTR(-ENOENT);
745 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
758 case ICMP_TIME_EXCEEDED:
761 case ICMP_SOURCE_QUENCH:
763 case ICMP_PARAMETERPROB:
767 case ICMP_DEST_UNREACH:
768 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
769 ipv4_sk_update_pmtu(skb, sk, info);
770 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
778 if (code <= NR_ICMP_UNREACH) {
779 harderr = icmp_err_convert[code].fatal;
780 err = icmp_err_convert[code].errno;
784 ipv4_sk_redirect(skb, sk);
789 * RFC1122: OK. Passes ICMP errors back to application, as per
793 /* ...not for tunnels though: we don't have a sending socket */
794 if (udp_sk(sk)->encap_err_rcv)
795 udp_sk(sk)->encap_err_rcv(sk, skb, iph->ihl << 2);
798 if (!inet->recverr) {
799 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
802 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
810 int udp_err(struct sk_buff *skb, u32 info)
812 return __udp4_lib_err(skb, info, &udp_table);
816 * Throw away all pending data and cancel the corking. Socket is locked.
818 void udp_flush_pending_frames(struct sock *sk)
820 struct udp_sock *up = udp_sk(sk);
825 ip_flush_pending_frames(sk);
828 EXPORT_SYMBOL(udp_flush_pending_frames);
831 * udp4_hwcsum - handle outgoing HW checksumming
832 * @skb: sk_buff containing the filled-in UDP header
833 * (checksum field must be zeroed out)
834 * @src: source IP address
835 * @dst: destination IP address
837 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
839 struct udphdr *uh = udp_hdr(skb);
840 int offset = skb_transport_offset(skb);
841 int len = skb->len - offset;
845 if (!skb_has_frag_list(skb)) {
847 * Only one fragment on the socket.
849 skb->csum_start = skb_transport_header(skb) - skb->head;
850 skb->csum_offset = offsetof(struct udphdr, check);
851 uh->check = ~csum_tcpudp_magic(src, dst, len,
854 struct sk_buff *frags;
857 * HW-checksum won't work as there are two or more
858 * fragments on the socket so that all csums of sk_buffs
861 skb_walk_frags(skb, frags) {
862 csum = csum_add(csum, frags->csum);
866 csum = skb_checksum(skb, offset, hlen, csum);
867 skb->ip_summed = CHECKSUM_NONE;
869 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
871 uh->check = CSUM_MANGLED_0;
874 EXPORT_SYMBOL_GPL(udp4_hwcsum);
876 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
877 * for the simple case like when setting the checksum for a UDP tunnel.
879 void udp_set_csum(bool nocheck, struct sk_buff *skb,
880 __be32 saddr, __be32 daddr, int len)
882 struct udphdr *uh = udp_hdr(skb);
886 } else if (skb_is_gso(skb)) {
887 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
888 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
890 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
892 uh->check = CSUM_MANGLED_0;
894 skb->ip_summed = CHECKSUM_PARTIAL;
895 skb->csum_start = skb_transport_header(skb) - skb->head;
896 skb->csum_offset = offsetof(struct udphdr, check);
897 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
900 EXPORT_SYMBOL(udp_set_csum);
902 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
903 struct inet_cork *cork)
905 struct sock *sk = skb->sk;
906 struct inet_sock *inet = inet_sk(sk);
909 int is_udplite = IS_UDPLITE(sk);
910 int offset = skb_transport_offset(skb);
911 int len = skb->len - offset;
912 int datalen = len - sizeof(*uh);
916 * Create a UDP header
919 uh->source = inet->inet_sport;
920 uh->dest = fl4->fl4_dport;
921 uh->len = htons(len);
924 if (cork->gso_size) {
925 const int hlen = skb_network_header_len(skb) +
926 sizeof(struct udphdr);
928 if (hlen + cork->gso_size > cork->fragsize) {
932 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
936 if (sk->sk_no_check_tx) {
940 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
941 dst_xfrm(skb_dst(skb))) {
946 if (datalen > cork->gso_size) {
947 skb_shinfo(skb)->gso_size = cork->gso_size;
948 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
949 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
955 if (is_udplite) /* UDP-Lite */
956 csum = udplite_csum(skb);
958 else if (sk->sk_no_check_tx) { /* UDP csum off */
960 skb->ip_summed = CHECKSUM_NONE;
963 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
966 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
970 csum = udp_csum(skb);
972 /* add protocol-dependent pseudo-header */
973 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
974 sk->sk_protocol, csum);
976 uh->check = CSUM_MANGLED_0;
979 err = ip_send_skb(sock_net(sk), skb);
981 if (err == -ENOBUFS && !inet->recverr) {
982 UDP_INC_STATS(sock_net(sk),
983 UDP_MIB_SNDBUFERRORS, is_udplite);
987 UDP_INC_STATS(sock_net(sk),
988 UDP_MIB_OUTDATAGRAMS, is_udplite);
993 * Push out all pending data as one UDP datagram. Socket is locked.
995 int udp_push_pending_frames(struct sock *sk)
997 struct udp_sock *up = udp_sk(sk);
998 struct inet_sock *inet = inet_sk(sk);
999 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1000 struct sk_buff *skb;
1003 skb = ip_finish_skb(sk, fl4);
1007 err = udp_send_skb(skb, fl4, &inet->cork.base);
1014 EXPORT_SYMBOL(udp_push_pending_frames);
1016 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1018 switch (cmsg->cmsg_type) {
1020 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1022 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1029 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1031 struct cmsghdr *cmsg;
1032 bool need_ip = false;
1035 for_each_cmsghdr(cmsg, msg) {
1036 if (!CMSG_OK(msg, cmsg))
1039 if (cmsg->cmsg_level != SOL_UDP) {
1044 err = __udp_cmsg_send(cmsg, gso_size);
1051 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1053 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1055 struct inet_sock *inet = inet_sk(sk);
1056 struct udp_sock *up = udp_sk(sk);
1057 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1058 struct flowi4 fl4_stack;
1061 struct ipcm_cookie ipc;
1062 struct rtable *rt = NULL;
1065 __be32 daddr, faddr, saddr;
1068 int err, is_udplite = IS_UDPLITE(sk);
1069 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1070 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1071 struct sk_buff *skb;
1072 struct ip_options_data opt_copy;
1081 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1084 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1086 fl4 = &inet->cork.fl.u.ip4;
1089 * There are pending frames.
1090 * The socket lock must be held while it's corked.
1093 if (likely(up->pending)) {
1094 if (unlikely(up->pending != AF_INET)) {
1098 goto do_append_data;
1102 ulen += sizeof(struct udphdr);
1105 * Get and verify the address.
1108 if (msg->msg_namelen < sizeof(*usin))
1110 if (usin->sin_family != AF_INET) {
1111 if (usin->sin_family != AF_UNSPEC)
1112 return -EAFNOSUPPORT;
1115 daddr = usin->sin_addr.s_addr;
1116 dport = usin->sin_port;
1120 if (sk->sk_state != TCP_ESTABLISHED)
1121 return -EDESTADDRREQ;
1122 daddr = inet->inet_daddr;
1123 dport = inet->inet_dport;
1124 /* Open fast path for connected socket.
1125 Route will not be used, if at least one option is set.
1130 ipcm_init_sk(&ipc, inet);
1131 ipc.gso_size = READ_ONCE(up->gso_size);
1133 if (msg->msg_controllen) {
1134 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1136 err = ip_cmsg_send(sk, msg, &ipc,
1137 sk->sk_family == AF_INET6);
1138 if (unlikely(err < 0)) {
1147 struct ip_options_rcu *inet_opt;
1150 inet_opt = rcu_dereference(inet->inet_opt);
1152 memcpy(&opt_copy, inet_opt,
1153 sizeof(*inet_opt) + inet_opt->opt.optlen);
1154 ipc.opt = &opt_copy.opt;
1159 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1160 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1161 (struct sockaddr *)usin, &ipc.addr);
1165 if (usin->sin_port == 0) {
1166 /* BPF program set invalid port. Reject it. */
1170 daddr = usin->sin_addr.s_addr;
1171 dport = usin->sin_port;
1176 ipc.addr = faddr = daddr;
1178 if (ipc.opt && ipc.opt->opt.srr) {
1183 faddr = ipc.opt->opt.faddr;
1186 tos = get_rttos(&ipc, inet);
1187 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1188 (msg->msg_flags & MSG_DONTROUTE) ||
1189 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1194 if (ipv4_is_multicast(daddr)) {
1195 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1196 ipc.oif = inet->mc_index;
1198 saddr = inet->mc_addr;
1200 } else if (!ipc.oif) {
1201 ipc.oif = inet->uc_index;
1202 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1203 /* oif is set, packet is to local broadcast and
1204 * uc_index is set. oif is most likely set
1205 * by sk_bound_dev_if. If uc_index != oif check if the
1206 * oif is an L3 master and uc_index is an L3 slave.
1207 * If so, we want to allow the send using the uc_index.
1209 if (ipc.oif != inet->uc_index &&
1210 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1212 ipc.oif = inet->uc_index;
1217 rt = (struct rtable *)sk_dst_check(sk, 0);
1220 struct net *net = sock_net(sk);
1221 __u8 flow_flags = inet_sk_flowi_flags(sk);
1225 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1226 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1228 faddr, saddr, dport, inet->inet_sport,
1231 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1232 rt = ip_route_output_flow(net, fl4, sk);
1236 if (err == -ENETUNREACH)
1237 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1242 if ((rt->rt_flags & RTCF_BROADCAST) &&
1243 !sock_flag(sk, SOCK_BROADCAST))
1246 sk_dst_set(sk, dst_clone(&rt->dst));
1249 if (msg->msg_flags&MSG_CONFIRM)
1255 daddr = ipc.addr = fl4->daddr;
1257 /* Lockless fast path for the non-corking case. */
1259 struct inet_cork cork;
1261 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1262 sizeof(struct udphdr), &ipc, &rt,
1263 &cork, msg->msg_flags);
1265 if (!IS_ERR_OR_NULL(skb))
1266 err = udp_send_skb(skb, fl4, &cork);
1271 if (unlikely(up->pending)) {
1272 /* The socket is already corked while preparing it. */
1273 /* ... which is an evident application bug. --ANK */
1276 net_dbg_ratelimited("socket already corked\n");
1281 * Now cork the socket to pend data.
1283 fl4 = &inet->cork.fl.u.ip4;
1286 fl4->fl4_dport = dport;
1287 fl4->fl4_sport = inet->inet_sport;
1288 up->pending = AF_INET;
1292 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1293 sizeof(struct udphdr), &ipc, &rt,
1294 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1296 udp_flush_pending_frames(sk);
1298 err = udp_push_pending_frames(sk);
1299 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1311 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1312 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1313 * we don't have a good statistic (IpOutDiscards but it can be too many
1314 * things). We could add another new stat but at least for now that
1315 * seems like overkill.
1317 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1318 UDP_INC_STATS(sock_net(sk),
1319 UDP_MIB_SNDBUFERRORS, is_udplite);
1324 if (msg->msg_flags & MSG_PROBE)
1325 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1326 if (!(msg->msg_flags&MSG_PROBE) || len)
1327 goto back_from_confirm;
1331 EXPORT_SYMBOL(udp_sendmsg);
1333 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1334 size_t size, int flags)
1336 struct inet_sock *inet = inet_sk(sk);
1337 struct udp_sock *up = udp_sk(sk);
1340 if (flags & MSG_SENDPAGE_NOTLAST)
1344 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1346 /* Call udp_sendmsg to specify destination address which
1347 * sendpage interface can't pass.
1348 * This will succeed only when the socket is connected.
1350 ret = udp_sendmsg(sk, &msg, 0);
1357 if (unlikely(!up->pending)) {
1360 net_dbg_ratelimited("cork failed\n");
1364 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1365 page, offset, size, flags);
1366 if (ret == -EOPNOTSUPP) {
1368 return sock_no_sendpage(sk->sk_socket, page, offset,
1372 udp_flush_pending_frames(sk);
1377 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1378 ret = udp_push_pending_frames(sk);
1386 #define UDP_SKB_IS_STATELESS 0x80000000
1388 /* all head states (dst, sk, nf conntrack) except skb extensions are
1389 * cleared by udp_rcv().
1391 * We need to preserve secpath, if present, to eventually process
1392 * IP_CMSG_PASSSEC at recvmsg() time.
1394 * Other extensions can be cleared.
1396 static bool udp_try_make_stateless(struct sk_buff *skb)
1398 if (!skb_has_extensions(skb))
1401 if (!secpath_exists(skb)) {
1409 static void udp_set_dev_scratch(struct sk_buff *skb)
1411 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1413 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1414 scratch->_tsize_state = skb->truesize;
1415 #if BITS_PER_LONG == 64
1416 scratch->len = skb->len;
1417 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1418 scratch->is_linear = !skb_is_nonlinear(skb);
1420 if (udp_try_make_stateless(skb))
1421 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1424 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1426 /* We come here after udp_lib_checksum_complete() returned 0.
1427 * This means that __skb_checksum_complete() might have
1428 * set skb->csum_valid to 1.
1429 * On 64bit platforms, we can set csum_unnecessary
1430 * to true, but only if the skb is not shared.
1432 #if BITS_PER_LONG == 64
1433 if (!skb_shared(skb))
1434 udp_skb_scratch(skb)->csum_unnecessary = true;
1438 static int udp_skb_truesize(struct sk_buff *skb)
1440 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1443 static bool udp_skb_has_head_state(struct sk_buff *skb)
1445 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1448 /* fully reclaim rmem/fwd memory allocated for skb */
1449 static void udp_rmem_release(struct sock *sk, int size, int partial,
1450 bool rx_queue_lock_held)
1452 struct udp_sock *up = udp_sk(sk);
1453 struct sk_buff_head *sk_queue;
1456 if (likely(partial)) {
1457 up->forward_deficit += size;
1458 size = up->forward_deficit;
1459 if (size < (sk->sk_rcvbuf >> 2) &&
1460 !skb_queue_empty(&up->reader_queue))
1463 size += up->forward_deficit;
1465 up->forward_deficit = 0;
1467 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1468 * if the called don't held it already
1470 sk_queue = &sk->sk_receive_queue;
1471 if (!rx_queue_lock_held)
1472 spin_lock(&sk_queue->lock);
1475 sk->sk_forward_alloc += size;
1476 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1477 sk->sk_forward_alloc -= amt;
1480 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1482 atomic_sub(size, &sk->sk_rmem_alloc);
1484 /* this can save us from acquiring the rx queue lock on next receive */
1485 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1487 if (!rx_queue_lock_held)
1488 spin_unlock(&sk_queue->lock);
1491 /* Note: called with reader_queue.lock held.
1492 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1493 * This avoids a cache line miss while receive_queue lock is held.
1494 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1496 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1498 prefetch(&skb->data);
1499 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1501 EXPORT_SYMBOL(udp_skb_destructor);
1503 /* as above, but the caller held the rx queue lock, too */
1504 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1506 prefetch(&skb->data);
1507 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1510 /* Idea of busylocks is to let producers grab an extra spinlock
1511 * to relieve pressure on the receive_queue spinlock shared by consumer.
1512 * Under flood, this means that only one producer can be in line
1513 * trying to acquire the receive_queue spinlock.
1514 * These busylock can be allocated on a per cpu manner, instead of a
1515 * per socket one (that would consume a cache line per socket)
1517 static int udp_busylocks_log __read_mostly;
1518 static spinlock_t *udp_busylocks __read_mostly;
1520 static spinlock_t *busylock_acquire(void *ptr)
1524 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1529 static void busylock_release(spinlock_t *busy)
1535 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1537 struct sk_buff_head *list = &sk->sk_receive_queue;
1538 int rmem, delta, amt, err = -ENOMEM;
1539 spinlock_t *busy = NULL;
1542 /* try to avoid the costly atomic add/sub pair when the receive
1543 * queue is full; always allow at least a packet
1545 rmem = atomic_read(&sk->sk_rmem_alloc);
1546 if (rmem > sk->sk_rcvbuf)
1549 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1550 * having linear skbs :
1551 * - Reduce memory overhead and thus increase receive queue capacity
1552 * - Less cache line misses at copyout() time
1553 * - Less work at consume_skb() (less alien page frag freeing)
1555 if (rmem > (sk->sk_rcvbuf >> 1)) {
1558 busy = busylock_acquire(sk);
1560 size = skb->truesize;
1561 udp_set_dev_scratch(skb);
1563 /* we drop only if the receive buf is full and the receive
1564 * queue contains some other skb
1566 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1567 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1570 spin_lock(&list->lock);
1571 if (size >= sk->sk_forward_alloc) {
1572 amt = sk_mem_pages(size);
1573 delta = amt << SK_MEM_QUANTUM_SHIFT;
1574 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1576 spin_unlock(&list->lock);
1580 sk->sk_forward_alloc += delta;
1583 sk->sk_forward_alloc -= size;
1585 /* no need to setup a destructor, we will explicitly release the
1586 * forward allocated memory on dequeue
1588 sock_skb_set_dropcount(sk, skb);
1590 __skb_queue_tail(list, skb);
1591 spin_unlock(&list->lock);
1593 if (!sock_flag(sk, SOCK_DEAD))
1594 sk->sk_data_ready(sk);
1596 busylock_release(busy);
1600 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1603 atomic_inc(&sk->sk_drops);
1604 busylock_release(busy);
1607 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1609 void udp_destruct_common(struct sock *sk)
1611 /* reclaim completely the forward allocated memory */
1612 struct udp_sock *up = udp_sk(sk);
1613 unsigned int total = 0;
1614 struct sk_buff *skb;
1616 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1617 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1618 total += skb->truesize;
1621 udp_rmem_release(sk, total, 0, true);
1623 EXPORT_SYMBOL_GPL(udp_destruct_common);
1625 static void udp_destruct_sock(struct sock *sk)
1627 udp_destruct_common(sk);
1628 inet_sock_destruct(sk);
1631 int udp_init_sock(struct sock *sk)
1633 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1634 sk->sk_destruct = udp_destruct_sock;
1638 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1640 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1641 bool slow = lock_sock_fast(sk);
1643 sk_peek_offset_bwd(sk, len);
1644 unlock_sock_fast(sk, slow);
1647 if (!skb_unref(skb))
1650 /* In the more common cases we cleared the head states previously,
1651 * see __udp_queue_rcv_skb().
1653 if (unlikely(udp_skb_has_head_state(skb)))
1654 skb_release_head_state(skb);
1655 __consume_stateless_skb(skb);
1657 EXPORT_SYMBOL_GPL(skb_consume_udp);
1659 static struct sk_buff *__first_packet_length(struct sock *sk,
1660 struct sk_buff_head *rcvq,
1663 struct sk_buff *skb;
1665 while ((skb = skb_peek(rcvq)) != NULL) {
1666 if (udp_lib_checksum_complete(skb)) {
1667 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1669 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1671 atomic_inc(&sk->sk_drops);
1672 __skb_unlink(skb, rcvq);
1673 *total += skb->truesize;
1676 udp_skb_csum_unnecessary_set(skb);
1684 * first_packet_length - return length of first packet in receive queue
1687 * Drops all bad checksum frames, until a valid one is found.
1688 * Returns the length of found skb, or -1 if none is found.
1690 static int first_packet_length(struct sock *sk)
1692 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1693 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1694 struct sk_buff *skb;
1698 spin_lock_bh(&rcvq->lock);
1699 skb = __first_packet_length(sk, rcvq, &total);
1700 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1701 spin_lock(&sk_queue->lock);
1702 skb_queue_splice_tail_init(sk_queue, rcvq);
1703 spin_unlock(&sk_queue->lock);
1705 skb = __first_packet_length(sk, rcvq, &total);
1707 res = skb ? skb->len : -1;
1709 udp_rmem_release(sk, total, 1, false);
1710 spin_unlock_bh(&rcvq->lock);
1715 * IOCTL requests applicable to the UDP protocol
1718 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1723 int amount = sk_wmem_alloc_get(sk);
1725 return put_user(amount, (int __user *)arg);
1730 int amount = max_t(int, 0, first_packet_length(sk));
1732 return put_user(amount, (int __user *)arg);
1736 return -ENOIOCTLCMD;
1741 EXPORT_SYMBOL(udp_ioctl);
1743 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1744 int noblock, int *off, int *err)
1746 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1747 struct sk_buff_head *queue;
1748 struct sk_buff *last;
1752 queue = &udp_sk(sk)->reader_queue;
1753 flags |= noblock ? MSG_DONTWAIT : 0;
1754 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1756 struct sk_buff *skb;
1758 error = sock_error(sk);
1764 spin_lock_bh(&queue->lock);
1765 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1768 if (!(flags & MSG_PEEK))
1769 udp_skb_destructor(sk, skb);
1770 spin_unlock_bh(&queue->lock);
1774 if (skb_queue_empty_lockless(sk_queue)) {
1775 spin_unlock_bh(&queue->lock);
1779 /* refill the reader queue and walk it again
1780 * keep both queues locked to avoid re-acquiring
1781 * the sk_receive_queue lock if fwd memory scheduling
1784 spin_lock(&sk_queue->lock);
1785 skb_queue_splice_tail_init(sk_queue, queue);
1787 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1789 if (skb && !(flags & MSG_PEEK))
1790 udp_skb_dtor_locked(sk, skb);
1791 spin_unlock(&sk_queue->lock);
1792 spin_unlock_bh(&queue->lock);
1797 if (!sk_can_busy_loop(sk))
1800 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1801 } while (!skb_queue_empty_lockless(sk_queue));
1803 /* sk_queue is empty, reader_queue may contain peeked packets */
1805 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1807 (struct sk_buff *)sk_queue));
1812 EXPORT_SYMBOL(__skb_recv_udp);
1814 int udp_read_sock(struct sock *sk, read_descriptor_t *desc,
1815 sk_read_actor_t recv_actor)
1820 struct sk_buff *skb;
1823 skb = skb_recv_udp(sk, 0, 1, &err);
1827 if (udp_lib_checksum_complete(skb)) {
1828 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1830 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1832 atomic_inc(&sk->sk_drops);
1837 used = recv_actor(desc, skb, 0, skb->len);
1843 } else if (used <= skb->len) {
1854 EXPORT_SYMBOL(udp_read_sock);
1857 * This should be easy, if there is something there we
1858 * return it, otherwise we block.
1861 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1862 int flags, int *addr_len)
1864 struct inet_sock *inet = inet_sk(sk);
1865 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1866 struct sk_buff *skb;
1867 unsigned int ulen, copied;
1868 int off, err, peeking = flags & MSG_PEEK;
1869 int is_udplite = IS_UDPLITE(sk);
1870 bool checksum_valid = false;
1872 if (flags & MSG_ERRQUEUE)
1873 return ip_recv_error(sk, msg, len, addr_len);
1876 off = sk_peek_offset(sk, flags);
1877 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1881 ulen = udp_skb_len(skb);
1883 if (copied > ulen - off)
1884 copied = ulen - off;
1885 else if (copied < ulen)
1886 msg->msg_flags |= MSG_TRUNC;
1889 * If checksum is needed at all, try to do it while copying the
1890 * data. If the data is truncated, or if we only want a partial
1891 * coverage checksum (UDP-Lite), do it before the copy.
1894 if (copied < ulen || peeking ||
1895 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1896 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1897 !__udp_lib_checksum_complete(skb);
1898 if (!checksum_valid)
1902 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1903 if (udp_skb_is_linear(skb))
1904 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1906 err = skb_copy_datagram_msg(skb, off, msg, copied);
1908 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1914 if (unlikely(err)) {
1916 atomic_inc(&sk->sk_drops);
1917 UDP_INC_STATS(sock_net(sk),
1918 UDP_MIB_INERRORS, is_udplite);
1925 UDP_INC_STATS(sock_net(sk),
1926 UDP_MIB_INDATAGRAMS, is_udplite);
1928 sock_recv_ts_and_drops(msg, sk, skb);
1930 /* Copy the address. */
1932 sin->sin_family = AF_INET;
1933 sin->sin_port = udp_hdr(skb)->source;
1934 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1935 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1936 *addr_len = sizeof(*sin);
1938 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1939 (struct sockaddr *)sin);
1942 if (udp_sk(sk)->gro_enabled)
1943 udp_cmsg_recv(msg, sk, skb);
1945 if (inet->cmsg_flags)
1946 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1949 if (flags & MSG_TRUNC)
1952 skb_consume_udp(sk, skb, peeking ? -err : err);
1956 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1957 udp_skb_destructor)) {
1958 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1959 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1963 /* starting over for a new packet, but check if we need to yield */
1965 msg->msg_flags &= ~MSG_TRUNC;
1969 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1971 /* This check is replicated from __ip4_datagram_connect() and
1972 * intended to prevent BPF program called below from accessing bytes
1973 * that are out of the bound specified by user in addr_len.
1975 if (addr_len < sizeof(struct sockaddr_in))
1978 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1980 EXPORT_SYMBOL(udp_pre_connect);
1982 int __udp_disconnect(struct sock *sk, int flags)
1984 struct inet_sock *inet = inet_sk(sk);
1986 * 1003.1g - break association.
1989 sk->sk_state = TCP_CLOSE;
1990 inet->inet_daddr = 0;
1991 inet->inet_dport = 0;
1992 sock_rps_reset_rxhash(sk);
1993 sk->sk_bound_dev_if = 0;
1994 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1995 inet_reset_saddr(sk);
1996 if (sk->sk_prot->rehash &&
1997 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1998 sk->sk_prot->rehash(sk);
2001 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
2002 sk->sk_prot->unhash(sk);
2003 inet->inet_sport = 0;
2008 EXPORT_SYMBOL(__udp_disconnect);
2010 int udp_disconnect(struct sock *sk, int flags)
2013 __udp_disconnect(sk, flags);
2017 EXPORT_SYMBOL(udp_disconnect);
2019 void udp_lib_unhash(struct sock *sk)
2021 if (sk_hashed(sk)) {
2022 struct udp_table *udptable = sk->sk_prot->h.udp_table;
2023 struct udp_hslot *hslot, *hslot2;
2025 hslot = udp_hashslot(udptable, sock_net(sk),
2026 udp_sk(sk)->udp_port_hash);
2027 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2029 spin_lock_bh(&hslot->lock);
2030 if (rcu_access_pointer(sk->sk_reuseport_cb))
2031 reuseport_detach_sock(sk);
2032 if (sk_del_node_init_rcu(sk)) {
2034 inet_sk(sk)->inet_num = 0;
2035 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2037 spin_lock(&hslot2->lock);
2038 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2040 spin_unlock(&hslot2->lock);
2042 spin_unlock_bh(&hslot->lock);
2045 EXPORT_SYMBOL(udp_lib_unhash);
2048 * inet_rcv_saddr was changed, we must rehash secondary hash
2050 void udp_lib_rehash(struct sock *sk, u16 newhash)
2052 if (sk_hashed(sk)) {
2053 struct udp_table *udptable = sk->sk_prot->h.udp_table;
2054 struct udp_hslot *hslot, *hslot2, *nhslot2;
2056 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2057 nhslot2 = udp_hashslot2(udptable, newhash);
2058 udp_sk(sk)->udp_portaddr_hash = newhash;
2060 if (hslot2 != nhslot2 ||
2061 rcu_access_pointer(sk->sk_reuseport_cb)) {
2062 hslot = udp_hashslot(udptable, sock_net(sk),
2063 udp_sk(sk)->udp_port_hash);
2064 /* we must lock primary chain too */
2065 spin_lock_bh(&hslot->lock);
2066 if (rcu_access_pointer(sk->sk_reuseport_cb))
2067 reuseport_detach_sock(sk);
2069 if (hslot2 != nhslot2) {
2070 spin_lock(&hslot2->lock);
2071 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2073 spin_unlock(&hslot2->lock);
2075 spin_lock(&nhslot2->lock);
2076 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2079 spin_unlock(&nhslot2->lock);
2082 spin_unlock_bh(&hslot->lock);
2086 EXPORT_SYMBOL(udp_lib_rehash);
2088 void udp_v4_rehash(struct sock *sk)
2090 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2091 inet_sk(sk)->inet_rcv_saddr,
2092 inet_sk(sk)->inet_num);
2093 udp_lib_rehash(sk, new_hash);
2096 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2100 if (inet_sk(sk)->inet_daddr) {
2101 sock_rps_save_rxhash(sk, skb);
2102 sk_mark_napi_id(sk, skb);
2103 sk_incoming_cpu_update(sk);
2105 sk_mark_napi_id_once(sk, skb);
2108 rc = __udp_enqueue_schedule_skb(sk, skb);
2110 int is_udplite = IS_UDPLITE(sk);
2112 /* Note that an ENOMEM error is charged twice */
2114 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2117 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2119 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2121 trace_udp_fail_queue_rcv_skb(rc, sk);
2131 * >0: "udp encap" protocol resubmission
2133 * Note that in the success and error cases, the skb is assumed to
2134 * have either been requeued or freed.
2136 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2138 struct udp_sock *up = udp_sk(sk);
2139 int is_udplite = IS_UDPLITE(sk);
2142 * Charge it to the socket, dropping if the queue is full.
2144 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2148 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2149 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2152 * This is an encapsulation socket so pass the skb to
2153 * the socket's udp_encap_rcv() hook. Otherwise, just
2154 * fall through and pass this up the UDP socket.
2155 * up->encap_rcv() returns the following value:
2156 * =0 if skb was successfully passed to the encap
2157 * handler or was discarded by it.
2158 * >0 if skb should be passed on to UDP.
2159 * <0 if skb should be resubmitted as proto -N
2162 /* if we're overly short, let UDP handle it */
2163 encap_rcv = READ_ONCE(up->encap_rcv);
2167 /* Verify checksum before giving to encap */
2168 if (udp_lib_checksum_complete(skb))
2171 ret = encap_rcv(sk, skb);
2173 __UDP_INC_STATS(sock_net(sk),
2174 UDP_MIB_INDATAGRAMS,
2180 /* FALLTHROUGH -- it's a UDP Packet */
2184 * UDP-Lite specific tests, ignored on UDP sockets
2186 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2189 * MIB statistics other than incrementing the error count are
2190 * disabled for the following two types of errors: these depend
2191 * on the application settings, not on the functioning of the
2192 * protocol stack as such.
2194 * RFC 3828 here recommends (sec 3.3): "There should also be a
2195 * way ... to ... at least let the receiving application block
2196 * delivery of packets with coverage values less than a value
2197 * provided by the application."
2199 if (up->pcrlen == 0) { /* full coverage was set */
2200 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2201 UDP_SKB_CB(skb)->cscov, skb->len);
2204 /* The next case involves violating the min. coverage requested
2205 * by the receiver. This is subtle: if receiver wants x and x is
2206 * greater than the buffersize/MTU then receiver will complain
2207 * that it wants x while sender emits packets of smaller size y.
2208 * Therefore the above ...()->partial_cov statement is essential.
2210 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2211 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2212 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2217 prefetch(&sk->sk_rmem_alloc);
2218 if (rcu_access_pointer(sk->sk_filter) &&
2219 udp_lib_checksum_complete(skb))
2222 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2225 udp_csum_pull_header(skb);
2227 ipv4_pktinfo_prepare(sk, skb);
2228 return __udp_queue_rcv_skb(sk, skb);
2231 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2233 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2234 atomic_inc(&sk->sk_drops);
2239 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2241 struct sk_buff *next, *segs;
2244 if (likely(!udp_unexpected_gso(sk, skb)))
2245 return udp_queue_rcv_one_skb(sk, skb);
2247 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2248 __skb_push(skb, -skb_mac_offset(skb));
2249 segs = udp_rcv_segment(sk, skb, true);
2250 skb_list_walk_safe(segs, skb, next) {
2251 __skb_pull(skb, skb_transport_offset(skb));
2253 udp_post_segment_fix_csum(skb);
2254 ret = udp_queue_rcv_one_skb(sk, skb);
2256 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2261 /* For TCP sockets, sk_rx_dst is protected by socket lock
2262 * For UDP, we use xchg() to guard against concurrent changes.
2264 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2266 struct dst_entry *old;
2268 if (dst_hold_safe(dst)) {
2269 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2275 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2278 * Multicasts and broadcasts go to each listener.
2280 * Note: called only from the BH handler context.
2282 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2284 __be32 saddr, __be32 daddr,
2285 struct udp_table *udptable,
2288 struct sock *sk, *first = NULL;
2289 unsigned short hnum = ntohs(uh->dest);
2290 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2291 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2292 unsigned int offset = offsetof(typeof(*sk), sk_node);
2293 int dif = skb->dev->ifindex;
2294 int sdif = inet_sdif(skb);
2295 struct hlist_node *node;
2296 struct sk_buff *nskb;
2299 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2301 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2303 hslot = &udptable->hash2[hash2];
2304 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2307 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2308 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2309 uh->source, saddr, dif, sdif, hnum))
2316 nskb = skb_clone(skb, GFP_ATOMIC);
2318 if (unlikely(!nskb)) {
2319 atomic_inc(&sk->sk_drops);
2320 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2322 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2326 if (udp_queue_rcv_skb(sk, nskb) > 0)
2330 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2331 if (use_hash2 && hash2 != hash2_any) {
2337 if (udp_queue_rcv_skb(first, skb) > 0)
2341 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2342 proto == IPPROTO_UDPLITE);
2347 /* Initialize UDP checksum. If exited with zero value (success),
2348 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2349 * Otherwise, csum completion requires checksumming packet body,
2350 * including udp header and folding it to skb->csum.
2352 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2357 UDP_SKB_CB(skb)->partial_cov = 0;
2358 UDP_SKB_CB(skb)->cscov = skb->len;
2360 if (proto == IPPROTO_UDPLITE) {
2361 err = udplite_checksum_init(skb, uh);
2365 if (UDP_SKB_CB(skb)->partial_cov) {
2366 skb->csum = inet_compute_pseudo(skb, proto);
2371 /* Note, we are only interested in != 0 or == 0, thus the
2374 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2375 inet_compute_pseudo);
2379 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2380 /* If SW calculated the value, we know it's bad */
2381 if (skb->csum_complete_sw)
2384 /* HW says the value is bad. Let's validate that.
2385 * skb->csum is no longer the full packet checksum,
2386 * so don't treat it as such.
2388 skb_checksum_complete_unset(skb);
2394 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2395 * return code conversion for ip layer consumption
2397 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2402 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2403 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2405 ret = udp_queue_rcv_skb(sk, skb);
2407 /* a return value > 0 means to resubmit the input, but
2408 * it wants the return to be -protocol, or 0
2416 * All we need to do is get the socket, and then do a checksum.
2419 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2424 unsigned short ulen;
2425 struct rtable *rt = skb_rtable(skb);
2426 __be32 saddr, daddr;
2427 struct net *net = dev_net(skb->dev);
2431 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2434 * Validate the packet.
2436 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2437 goto drop; /* No space for header. */
2440 ulen = ntohs(uh->len);
2441 saddr = ip_hdr(skb)->saddr;
2442 daddr = ip_hdr(skb)->daddr;
2444 if (ulen > skb->len)
2447 if (proto == IPPROTO_UDP) {
2448 /* UDP validates ulen. */
2449 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2454 if (udp4_csum_init(skb, uh, proto))
2457 sk = skb_steal_sock(skb, &refcounted);
2459 struct dst_entry *dst = skb_dst(skb);
2462 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2463 udp_sk_rx_dst_set(sk, dst);
2465 ret = udp_unicast_rcv_skb(sk, skb, uh);
2471 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2472 return __udp4_lib_mcast_deliver(net, skb, uh,
2473 saddr, daddr, udptable, proto);
2475 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2477 return udp_unicast_rcv_skb(sk, skb, uh);
2479 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2483 /* No socket. Drop packet silently, if checksum is wrong */
2484 if (udp_lib_checksum_complete(skb))
2487 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2488 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2489 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2492 * Hmm. We got an UDP packet to a port to which we
2493 * don't wanna listen. Ignore it.
2495 kfree_skb_reason(skb, drop_reason);
2499 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2500 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2501 proto == IPPROTO_UDPLITE ? "Lite" : "",
2502 &saddr, ntohs(uh->source),
2504 &daddr, ntohs(uh->dest));
2509 * RFC1122: OK. Discards the bad packet silently (as far as
2510 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2512 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2513 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2514 proto == IPPROTO_UDPLITE ? "Lite" : "",
2515 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2517 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2519 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2520 kfree_skb_reason(skb, drop_reason);
2524 /* We can only early demux multicast if there is a single matching socket.
2525 * If more than one socket found returns NULL
2527 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2528 __be16 loc_port, __be32 loc_addr,
2529 __be16 rmt_port, __be32 rmt_addr,
2532 struct sock *sk, *result;
2533 unsigned short hnum = ntohs(loc_port);
2534 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2535 struct udp_hslot *hslot = &udp_table.hash[slot];
2537 /* Do not bother scanning a too big list */
2538 if (hslot->count > 10)
2542 sk_for_each_rcu(sk, &hslot->head) {
2543 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2544 rmt_port, rmt_addr, dif, sdif, hnum)) {
2554 /* For unicast we should only early demux connected sockets or we can
2555 * break forwarding setups. The chains here can be long so only check
2556 * if the first socket is an exact match and if not move on.
2558 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2559 __be16 loc_port, __be32 loc_addr,
2560 __be16 rmt_port, __be32 rmt_addr,
2563 unsigned short hnum = ntohs(loc_port);
2564 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2565 unsigned int slot2 = hash2 & udp_table.mask;
2566 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2567 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2568 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2571 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2572 if (INET_MATCH(net, sk, acookie, ports, dif, sdif))
2574 /* Only check first socket in chain */
2580 int udp_v4_early_demux(struct sk_buff *skb)
2582 struct net *net = dev_net(skb->dev);
2583 struct in_device *in_dev = NULL;
2584 const struct iphdr *iph;
2585 const struct udphdr *uh;
2586 struct sock *sk = NULL;
2587 struct dst_entry *dst;
2588 int dif = skb->dev->ifindex;
2589 int sdif = inet_sdif(skb);
2592 /* validate the packet */
2593 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2599 if (skb->pkt_type == PACKET_MULTICAST) {
2600 in_dev = __in_dev_get_rcu(skb->dev);
2605 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2610 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2611 uh->source, iph->saddr,
2613 } else if (skb->pkt_type == PACKET_HOST) {
2614 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2615 uh->source, iph->saddr, dif, sdif);
2618 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2622 skb->destructor = sock_efree;
2623 dst = rcu_dereference(sk->sk_rx_dst);
2626 dst = dst_check(dst, 0);
2630 /* set noref for now.
2631 * any place which wants to hold dst has to call
2634 skb_dst_set_noref(skb, dst);
2636 /* for unconnected multicast sockets we need to validate
2637 * the source on each packet
2639 if (!inet_sk(sk)->inet_daddr && in_dev)
2640 return ip_mc_validate_source(skb, iph->daddr,
2642 iph->tos & IPTOS_RT_MASK,
2643 skb->dev, in_dev, &itag);
2648 int udp_rcv(struct sk_buff *skb)
2650 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2653 void udp_destroy_sock(struct sock *sk)
2655 struct udp_sock *up = udp_sk(sk);
2656 bool slow = lock_sock_fast(sk);
2658 /* protects from races with udp_abort() */
2659 sock_set_flag(sk, SOCK_DEAD);
2660 udp_flush_pending_frames(sk);
2661 unlock_sock_fast(sk, slow);
2662 if (static_branch_unlikely(&udp_encap_needed_key)) {
2663 if (up->encap_type) {
2664 void (*encap_destroy)(struct sock *sk);
2665 encap_destroy = READ_ONCE(up->encap_destroy);
2669 if (up->encap_enabled)
2670 static_branch_dec(&udp_encap_needed_key);
2675 * Socket option code for UDP
2677 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2678 sockptr_t optval, unsigned int optlen,
2679 int (*push_pending_frames)(struct sock *))
2681 struct udp_sock *up = udp_sk(sk);
2684 int is_udplite = IS_UDPLITE(sk);
2686 if (optlen < sizeof(int))
2689 if (copy_from_sockptr(&val, optval, sizeof(val)))
2692 valbool = val ? 1 : 0;
2697 WRITE_ONCE(up->corkflag, 1);
2699 WRITE_ONCE(up->corkflag, 0);
2701 push_pending_frames(sk);
2710 case UDP_ENCAP_ESPINUDP:
2711 case UDP_ENCAP_ESPINUDP_NON_IKE:
2712 #if IS_ENABLED(CONFIG_IPV6)
2713 if (sk->sk_family == AF_INET6)
2714 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2717 up->encap_rcv = xfrm4_udp_encap_rcv;
2720 case UDP_ENCAP_L2TPINUDP:
2721 up->encap_type = val;
2723 udp_tunnel_encap_enable(sk->sk_socket);
2732 case UDP_NO_CHECK6_TX:
2733 up->no_check6_tx = valbool;
2736 case UDP_NO_CHECK6_RX:
2737 up->no_check6_rx = valbool;
2741 if (val < 0 || val > USHRT_MAX)
2743 WRITE_ONCE(up->gso_size, val);
2749 /* when enabling GRO, accept the related GSO packet type */
2751 udp_tunnel_encap_enable(sk->sk_socket);
2752 up->gro_enabled = valbool;
2753 up->accept_udp_l4 = valbool;
2758 * UDP-Lite's partial checksum coverage (RFC 3828).
2760 /* The sender sets actual checksum coverage length via this option.
2761 * The case coverage > packet length is handled by send module. */
2762 case UDPLITE_SEND_CSCOV:
2763 if (!is_udplite) /* Disable the option on UDP sockets */
2764 return -ENOPROTOOPT;
2765 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2767 else if (val > USHRT_MAX)
2770 up->pcflag |= UDPLITE_SEND_CC;
2773 /* The receiver specifies a minimum checksum coverage value. To make
2774 * sense, this should be set to at least 8 (as done below). If zero is
2775 * used, this again means full checksum coverage. */
2776 case UDPLITE_RECV_CSCOV:
2777 if (!is_udplite) /* Disable the option on UDP sockets */
2778 return -ENOPROTOOPT;
2779 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2781 else if (val > USHRT_MAX)
2784 up->pcflag |= UDPLITE_RECV_CC;
2794 EXPORT_SYMBOL(udp_lib_setsockopt);
2796 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2797 unsigned int optlen)
2799 if (level == SOL_UDP || level == SOL_UDPLITE)
2800 return udp_lib_setsockopt(sk, level, optname,
2802 udp_push_pending_frames);
2803 return ip_setsockopt(sk, level, optname, optval, optlen);
2806 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2807 char __user *optval, int __user *optlen)
2809 struct udp_sock *up = udp_sk(sk);
2812 if (get_user(len, optlen))
2815 len = min_t(unsigned int, len, sizeof(int));
2822 val = READ_ONCE(up->corkflag);
2826 val = up->encap_type;
2829 case UDP_NO_CHECK6_TX:
2830 val = up->no_check6_tx;
2833 case UDP_NO_CHECK6_RX:
2834 val = up->no_check6_rx;
2838 val = READ_ONCE(up->gso_size);
2842 val = up->gro_enabled;
2845 /* The following two cannot be changed on UDP sockets, the return is
2846 * always 0 (which corresponds to the full checksum coverage of UDP). */
2847 case UDPLITE_SEND_CSCOV:
2851 case UDPLITE_RECV_CSCOV:
2856 return -ENOPROTOOPT;
2859 if (put_user(len, optlen))
2861 if (copy_to_user(optval, &val, len))
2865 EXPORT_SYMBOL(udp_lib_getsockopt);
2867 int udp_getsockopt(struct sock *sk, int level, int optname,
2868 char __user *optval, int __user *optlen)
2870 if (level == SOL_UDP || level == SOL_UDPLITE)
2871 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2872 return ip_getsockopt(sk, level, optname, optval, optlen);
2876 * udp_poll - wait for a UDP event.
2877 * @file: - file struct
2879 * @wait: - poll table
2881 * This is same as datagram poll, except for the special case of
2882 * blocking sockets. If application is using a blocking fd
2883 * and a packet with checksum error is in the queue;
2884 * then it could get return from select indicating data available
2885 * but then block when reading it. Add special case code
2886 * to work around these arguably broken applications.
2888 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2890 __poll_t mask = datagram_poll(file, sock, wait);
2891 struct sock *sk = sock->sk;
2893 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2894 mask |= EPOLLIN | EPOLLRDNORM;
2896 /* Check for false positives due to checksum errors */
2897 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2898 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2899 mask &= ~(EPOLLIN | EPOLLRDNORM);
2901 /* psock ingress_msg queue should not contain any bad checksum frames */
2902 if (sk_is_readable(sk))
2903 mask |= EPOLLIN | EPOLLRDNORM;
2907 EXPORT_SYMBOL(udp_poll);
2909 int udp_abort(struct sock *sk, int err)
2913 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2916 if (sock_flag(sk, SOCK_DEAD))
2920 sk_error_report(sk);
2921 __udp_disconnect(sk, 0);
2928 EXPORT_SYMBOL_GPL(udp_abort);
2930 struct proto udp_prot = {
2932 .owner = THIS_MODULE,
2933 .close = udp_lib_close,
2934 .pre_connect = udp_pre_connect,
2935 .connect = ip4_datagram_connect,
2936 .disconnect = udp_disconnect,
2938 .init = udp_init_sock,
2939 .destroy = udp_destroy_sock,
2940 .setsockopt = udp_setsockopt,
2941 .getsockopt = udp_getsockopt,
2942 .sendmsg = udp_sendmsg,
2943 .recvmsg = udp_recvmsg,
2944 .sendpage = udp_sendpage,
2945 .release_cb = ip4_datagram_release_cb,
2946 .hash = udp_lib_hash,
2947 .unhash = udp_lib_unhash,
2948 .rehash = udp_v4_rehash,
2949 .get_port = udp_v4_get_port,
2950 #ifdef CONFIG_BPF_SYSCALL
2951 .psock_update_sk_prot = udp_bpf_update_proto,
2953 .memory_allocated = &udp_memory_allocated,
2954 .sysctl_mem = sysctl_udp_mem,
2955 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2956 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2957 .obj_size = sizeof(struct udp_sock),
2958 .h.udp_table = &udp_table,
2959 .diag_destroy = udp_abort,
2961 EXPORT_SYMBOL(udp_prot);
2963 /* ------------------------------------------------------------------------ */
2964 #ifdef CONFIG_PROC_FS
2966 static struct sock *udp_get_first(struct seq_file *seq, int start)
2969 struct udp_seq_afinfo *afinfo;
2970 struct udp_iter_state *state = seq->private;
2971 struct net *net = seq_file_net(seq);
2973 if (state->bpf_seq_afinfo)
2974 afinfo = state->bpf_seq_afinfo;
2976 afinfo = PDE_DATA(file_inode(seq->file));
2978 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2980 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2982 if (hlist_empty(&hslot->head))
2985 spin_lock_bh(&hslot->lock);
2986 sk_for_each(sk, &hslot->head) {
2987 if (!net_eq(sock_net(sk), net))
2989 if (afinfo->family == AF_UNSPEC ||
2990 sk->sk_family == afinfo->family)
2993 spin_unlock_bh(&hslot->lock);
3000 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3002 struct udp_seq_afinfo *afinfo;
3003 struct udp_iter_state *state = seq->private;
3004 struct net *net = seq_file_net(seq);
3006 if (state->bpf_seq_afinfo)
3007 afinfo = state->bpf_seq_afinfo;
3009 afinfo = PDE_DATA(file_inode(seq->file));
3013 } while (sk && (!net_eq(sock_net(sk), net) ||
3014 (afinfo->family != AF_UNSPEC &&
3015 sk->sk_family != afinfo->family)));
3018 if (state->bucket <= afinfo->udp_table->mask)
3019 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
3020 return udp_get_first(seq, state->bucket + 1);
3025 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3027 struct sock *sk = udp_get_first(seq, 0);
3030 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3032 return pos ? NULL : sk;
3035 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3037 struct udp_iter_state *state = seq->private;
3038 state->bucket = MAX_UDP_PORTS;
3040 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3042 EXPORT_SYMBOL(udp_seq_start);
3044 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3048 if (v == SEQ_START_TOKEN)
3049 sk = udp_get_idx(seq, 0);
3051 sk = udp_get_next(seq, v);
3056 EXPORT_SYMBOL(udp_seq_next);
3058 void udp_seq_stop(struct seq_file *seq, void *v)
3060 struct udp_seq_afinfo *afinfo;
3061 struct udp_iter_state *state = seq->private;
3063 if (state->bpf_seq_afinfo)
3064 afinfo = state->bpf_seq_afinfo;
3066 afinfo = PDE_DATA(file_inode(seq->file));
3068 if (state->bucket <= afinfo->udp_table->mask)
3069 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
3071 EXPORT_SYMBOL(udp_seq_stop);
3073 /* ------------------------------------------------------------------------ */
3074 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3077 struct inet_sock *inet = inet_sk(sp);
3078 __be32 dest = inet->inet_daddr;
3079 __be32 src = inet->inet_rcv_saddr;
3080 __u16 destp = ntohs(inet->inet_dport);
3081 __u16 srcp = ntohs(inet->inet_sport);
3083 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3084 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3085 bucket, src, srcp, dest, destp, sp->sk_state,
3086 sk_wmem_alloc_get(sp),
3089 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3091 refcount_read(&sp->sk_refcnt), sp,
3092 atomic_read(&sp->sk_drops));
3095 int udp4_seq_show(struct seq_file *seq, void *v)
3097 seq_setwidth(seq, 127);
3098 if (v == SEQ_START_TOKEN)
3099 seq_puts(seq, " sl local_address rem_address st tx_queue "
3100 "rx_queue tr tm->when retrnsmt uid timeout "
3101 "inode ref pointer drops");
3103 struct udp_iter_state *state = seq->private;
3105 udp4_format_sock(v, seq, state->bucket);
3111 #ifdef CONFIG_BPF_SYSCALL
3112 struct bpf_iter__udp {
3113 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3114 __bpf_md_ptr(struct udp_sock *, udp_sk);
3115 uid_t uid __aligned(8);
3116 int bucket __aligned(8);
3119 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3120 struct udp_sock *udp_sk, uid_t uid, int bucket)
3122 struct bpf_iter__udp ctx;
3124 meta->seq_num--; /* skip SEQ_START_TOKEN */
3126 ctx.udp_sk = udp_sk;
3128 ctx.bucket = bucket;
3129 return bpf_iter_run_prog(prog, &ctx);
3132 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3134 struct udp_iter_state *state = seq->private;
3135 struct bpf_iter_meta meta;
3136 struct bpf_prog *prog;
3137 struct sock *sk = v;
3140 if (v == SEQ_START_TOKEN)
3143 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3145 prog = bpf_iter_get_info(&meta, false);
3146 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3149 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3151 struct bpf_iter_meta meta;
3152 struct bpf_prog *prog;
3156 prog = bpf_iter_get_info(&meta, true);
3158 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3161 udp_seq_stop(seq, v);
3164 static const struct seq_operations bpf_iter_udp_seq_ops = {
3165 .start = udp_seq_start,
3166 .next = udp_seq_next,
3167 .stop = bpf_iter_udp_seq_stop,
3168 .show = bpf_iter_udp_seq_show,
3172 const struct seq_operations udp_seq_ops = {
3173 .start = udp_seq_start,
3174 .next = udp_seq_next,
3175 .stop = udp_seq_stop,
3176 .show = udp4_seq_show,
3178 EXPORT_SYMBOL(udp_seq_ops);
3180 static struct udp_seq_afinfo udp4_seq_afinfo = {
3182 .udp_table = &udp_table,
3185 static int __net_init udp4_proc_init_net(struct net *net)
3187 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3188 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3193 static void __net_exit udp4_proc_exit_net(struct net *net)
3195 remove_proc_entry("udp", net->proc_net);
3198 static struct pernet_operations udp4_net_ops = {
3199 .init = udp4_proc_init_net,
3200 .exit = udp4_proc_exit_net,
3203 int __init udp4_proc_init(void)
3205 return register_pernet_subsys(&udp4_net_ops);
3208 void udp4_proc_exit(void)
3210 unregister_pernet_subsys(&udp4_net_ops);
3212 #endif /* CONFIG_PROC_FS */
3214 static __initdata unsigned long uhash_entries;
3215 static int __init set_uhash_entries(char *str)
3222 ret = kstrtoul(str, 0, &uhash_entries);
3226 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3227 uhash_entries = UDP_HTABLE_SIZE_MIN;
3230 __setup("uhash_entries=", set_uhash_entries);
3232 void __init udp_table_init(struct udp_table *table, const char *name)
3236 table->hash = alloc_large_system_hash(name,
3237 2 * sizeof(struct udp_hslot),
3239 21, /* one slot per 2 MB */
3243 UDP_HTABLE_SIZE_MIN,
3246 table->hash2 = table->hash + (table->mask + 1);
3247 for (i = 0; i <= table->mask; i++) {
3248 INIT_HLIST_HEAD(&table->hash[i].head);
3249 table->hash[i].count = 0;
3250 spin_lock_init(&table->hash[i].lock);
3252 for (i = 0; i <= table->mask; i++) {
3253 INIT_HLIST_HEAD(&table->hash2[i].head);
3254 table->hash2[i].count = 0;
3255 spin_lock_init(&table->hash2[i].lock);
3259 u32 udp_flow_hashrnd(void)
3261 static u32 hashrnd __read_mostly;
3263 net_get_random_once(&hashrnd, sizeof(hashrnd));
3267 EXPORT_SYMBOL(udp_flow_hashrnd);
3269 static void __udp_sysctl_init(struct net *net)
3271 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3272 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3274 #ifdef CONFIG_NET_L3_MASTER_DEV
3275 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3279 static int __net_init udp_sysctl_init(struct net *net)
3281 __udp_sysctl_init(net);
3285 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3286 .init = udp_sysctl_init,
3289 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3290 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3291 struct udp_sock *udp_sk, uid_t uid, int bucket)
3293 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3295 struct udp_iter_state *st = priv_data;
3296 struct udp_seq_afinfo *afinfo;
3299 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3303 afinfo->family = AF_UNSPEC;
3304 afinfo->udp_table = &udp_table;
3305 st->bpf_seq_afinfo = afinfo;
3306 ret = bpf_iter_init_seq_net(priv_data, aux);
3312 static void bpf_iter_fini_udp(void *priv_data)
3314 struct udp_iter_state *st = priv_data;
3316 kfree(st->bpf_seq_afinfo);
3317 bpf_iter_fini_seq_net(priv_data);
3320 static const struct bpf_iter_seq_info udp_seq_info = {
3321 .seq_ops = &bpf_iter_udp_seq_ops,
3322 .init_seq_private = bpf_iter_init_udp,
3323 .fini_seq_private = bpf_iter_fini_udp,
3324 .seq_priv_size = sizeof(struct udp_iter_state),
3327 static struct bpf_iter_reg udp_reg_info = {
3329 .ctx_arg_info_size = 1,
3331 { offsetof(struct bpf_iter__udp, udp_sk),
3332 PTR_TO_BTF_ID_OR_NULL },
3334 .seq_info = &udp_seq_info,
3337 static void __init bpf_iter_register(void)
3339 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3340 if (bpf_iter_reg_target(&udp_reg_info))
3341 pr_warn("Warning: could not register bpf iterator udp\n");
3345 void __init udp_init(void)
3347 unsigned long limit;
3350 udp_table_init(&udp_table, "UDP");
3351 limit = nr_free_buffer_pages() / 8;
3352 limit = max(limit, 128UL);
3353 sysctl_udp_mem[0] = limit / 4 * 3;
3354 sysctl_udp_mem[1] = limit;
3355 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3357 __udp_sysctl_init(&init_net);
3359 /* 16 spinlocks per cpu */
3360 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3361 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3364 panic("UDP: failed to alloc udp_busylocks\n");
3365 for (i = 0; i < (1U << udp_busylocks_log); i++)
3366 spin_lock_init(udp_busylocks + i);
3368 if (register_pernet_subsys(&udp_sysctl_ops))
3369 panic("UDP: failed to init sysctl parameters.\n");
3371 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3372 bpf_iter_register();