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/bpf-cgroup.h>
78 #include <linux/uaccess.h>
79 #include <asm/ioctls.h>
80 #include <linux/memblock.h>
81 #include <linux/highmem.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 ____cacheline_aligned_in_smp;
127 EXPORT_SYMBOL(udp_memory_allocated);
128 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
129 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
131 #define MAX_UDP_PORTS 65536
132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
134 static int udp_lib_lport_inuse(struct net *net, __u16 num,
135 const struct udp_hslot *hslot,
136 unsigned long *bitmap,
137 struct sock *sk, unsigned int log)
140 kuid_t uid = sock_i_uid(sk);
142 sk_for_each(sk2, &hslot->head) {
143 if (net_eq(sock_net(sk2), net) &&
145 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
146 (!sk2->sk_reuse || !sk->sk_reuse) &&
147 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
148 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
149 inet_rcv_saddr_equal(sk, sk2, true)) {
150 if (sk2->sk_reuseport && sk->sk_reuseport &&
151 !rcu_access_pointer(sk->sk_reuseport_cb) &&
152 uid_eq(uid, sock_i_uid(sk2))) {
158 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
167 * Note: we still hold spinlock of primary hash chain, so no other writer
168 * can insert/delete a socket with local_port == num
170 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
171 struct udp_hslot *hslot2,
175 kuid_t uid = sock_i_uid(sk);
178 spin_lock(&hslot2->lock);
179 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
180 if (net_eq(sock_net(sk2), net) &&
182 (udp_sk(sk2)->udp_port_hash == num) &&
183 (!sk2->sk_reuse || !sk->sk_reuse) &&
184 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
185 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
186 inet_rcv_saddr_equal(sk, sk2, true)) {
187 if (sk2->sk_reuseport && sk->sk_reuseport &&
188 !rcu_access_pointer(sk->sk_reuseport_cb) &&
189 uid_eq(uid, sock_i_uid(sk2))) {
197 spin_unlock(&hslot2->lock);
201 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
203 struct net *net = sock_net(sk);
204 kuid_t uid = sock_i_uid(sk);
207 sk_for_each(sk2, &hslot->head) {
208 if (net_eq(sock_net(sk2), net) &&
210 sk2->sk_family == sk->sk_family &&
211 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
212 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
213 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
214 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
215 inet_rcv_saddr_equal(sk, sk2, false)) {
216 return reuseport_add_sock(sk, sk2,
217 inet_rcv_saddr_any(sk));
221 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
225 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
227 * @sk: socket struct in question
228 * @snum: port number to look up
229 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
232 int udp_lib_get_port(struct sock *sk, unsigned short snum,
233 unsigned int hash2_nulladdr)
235 struct udp_hslot *hslot, *hslot2;
236 struct udp_table *udptable = sk->sk_prot->h.udp_table;
238 struct net *net = sock_net(sk);
241 int low, high, remaining;
243 unsigned short first, last;
244 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
246 inet_get_local_port_range(net, &low, &high);
247 remaining = (high - low) + 1;
249 rand = get_random_u32();
250 first = reciprocal_scale(rand, remaining) + low;
252 * force rand to be an odd multiple of UDP_HTABLE_SIZE
254 rand = (rand | 1) * (udptable->mask + 1);
255 last = first + udptable->mask + 1;
257 hslot = udp_hashslot(udptable, net, first);
258 bitmap_zero(bitmap, PORTS_PER_CHAIN);
259 spin_lock_bh(&hslot->lock);
260 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
265 * Iterate on all possible values of snum for this hash.
266 * Using steps of an odd multiple of UDP_HTABLE_SIZE
267 * give us randomization and full range coverage.
270 if (low <= snum && snum <= high &&
271 !test_bit(snum >> udptable->log, bitmap) &&
272 !inet_is_local_reserved_port(net, snum))
275 } while (snum != first);
276 spin_unlock_bh(&hslot->lock);
278 } while (++first != last);
281 hslot = udp_hashslot(udptable, net, snum);
282 spin_lock_bh(&hslot->lock);
283 if (hslot->count > 10) {
285 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
287 slot2 &= udptable->mask;
288 hash2_nulladdr &= udptable->mask;
290 hslot2 = udp_hashslot2(udptable, slot2);
291 if (hslot->count < hslot2->count)
292 goto scan_primary_hash;
294 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
295 if (!exist && (hash2_nulladdr != slot2)) {
296 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
297 exist = udp_lib_lport_inuse2(net, snum, hslot2,
306 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
310 inet_sk(sk)->inet_num = snum;
311 udp_sk(sk)->udp_port_hash = snum;
312 udp_sk(sk)->udp_portaddr_hash ^= snum;
313 if (sk_unhashed(sk)) {
314 if (sk->sk_reuseport &&
315 udp_reuseport_add_sock(sk, hslot)) {
316 inet_sk(sk)->inet_num = 0;
317 udp_sk(sk)->udp_port_hash = 0;
318 udp_sk(sk)->udp_portaddr_hash ^= snum;
322 sk_add_node_rcu(sk, &hslot->head);
324 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
326 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
327 spin_lock(&hslot2->lock);
328 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
329 sk->sk_family == AF_INET6)
330 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
333 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
336 spin_unlock(&hslot2->lock);
338 sock_set_flag(sk, SOCK_RCU_FREE);
341 spin_unlock_bh(&hslot->lock);
345 EXPORT_SYMBOL(udp_lib_get_port);
347 int udp_v4_get_port(struct sock *sk, unsigned short snum)
349 unsigned int hash2_nulladdr =
350 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
351 unsigned int hash2_partial =
352 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
354 /* precompute partial secondary hash */
355 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
356 return udp_lib_get_port(sk, snum, hash2_nulladdr);
359 static int compute_score(struct sock *sk, struct net *net,
360 __be32 saddr, __be16 sport,
361 __be32 daddr, unsigned short hnum,
365 struct inet_sock *inet;
368 if (!net_eq(sock_net(sk), net) ||
369 udp_sk(sk)->udp_port_hash != hnum ||
373 if (sk->sk_rcv_saddr != daddr)
376 score = (sk->sk_family == PF_INET) ? 2 : 1;
379 if (inet->inet_daddr) {
380 if (inet->inet_daddr != saddr)
385 if (inet->inet_dport) {
386 if (inet->inet_dport != sport)
391 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
395 if (sk->sk_bound_dev_if)
398 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
403 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
404 const __u16 lport, const __be32 faddr,
407 static u32 udp_ehash_secret __read_mostly;
409 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
411 return __inet_ehashfn(laddr, lport, faddr, fport,
412 udp_ehash_secret + net_hash_mix(net));
415 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
417 __be32 saddr, __be16 sport,
418 __be32 daddr, unsigned short hnum)
420 struct sock *reuse_sk = NULL;
423 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
424 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
425 reuse_sk = reuseport_select_sock(sk, hash, skb,
426 sizeof(struct udphdr));
431 /* called with rcu_read_lock() */
432 static struct sock *udp4_lib_lookup2(struct net *net,
433 __be32 saddr, __be16 sport,
434 __be32 daddr, unsigned int hnum,
436 struct udp_hslot *hslot2,
439 struct sock *sk, *result;
444 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
445 score = compute_score(sk, net, saddr, sport,
446 daddr, hnum, dif, sdif);
447 if (score > badness) {
448 result = lookup_reuseport(net, sk, skb,
449 saddr, sport, daddr, hnum);
450 /* Fall back to scoring if group has connections */
451 if (result && !reuseport_has_conns(sk))
454 result = result ? : sk;
461 static struct sock *udp4_lookup_run_bpf(struct net *net,
462 struct udp_table *udptable,
464 __be32 saddr, __be16 sport,
465 __be32 daddr, u16 hnum, const int dif)
467 struct sock *sk, *reuse_sk;
470 if (udptable != &udp_table)
471 return NULL; /* only UDP is supported */
473 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport,
474 daddr, hnum, dif, &sk);
475 if (no_reuseport || IS_ERR_OR_NULL(sk))
478 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
484 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
485 * harder than this. -DaveM
487 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
488 __be16 sport, __be32 daddr, __be16 dport, int dif,
489 int sdif, struct udp_table *udptable, struct sk_buff *skb)
491 unsigned short hnum = ntohs(dport);
492 unsigned int hash2, slot2;
493 struct udp_hslot *hslot2;
494 struct sock *result, *sk;
496 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
497 slot2 = hash2 & udptable->mask;
498 hslot2 = &udptable->hash2[slot2];
500 /* Lookup connected or non-wildcard socket */
501 result = udp4_lib_lookup2(net, saddr, sport,
502 daddr, hnum, dif, sdif,
504 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
507 /* Lookup redirect from BPF */
508 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
509 sk = udp4_lookup_run_bpf(net, udptable, skb,
510 saddr, sport, daddr, hnum, dif);
517 /* Got non-wildcard socket or error on first lookup */
521 /* Lookup wildcard sockets */
522 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
523 slot2 = hash2 & udptable->mask;
524 hslot2 = &udptable->hash2[slot2];
526 result = udp4_lib_lookup2(net, saddr, sport,
527 htonl(INADDR_ANY), hnum, dif, sdif,
534 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
536 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
537 __be16 sport, __be16 dport,
538 struct udp_table *udptable)
540 const struct iphdr *iph = ip_hdr(skb);
542 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
543 iph->daddr, dport, inet_iif(skb),
544 inet_sdif(skb), udptable, skb);
547 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
548 __be16 sport, __be16 dport)
550 const struct iphdr *iph = ip_hdr(skb);
552 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
553 iph->daddr, dport, inet_iif(skb),
554 inet_sdif(skb), &udp_table, NULL);
557 /* Must be called under rcu_read_lock().
558 * Does increment socket refcount.
560 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
561 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
562 __be32 daddr, __be16 dport, int dif)
566 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
567 dif, 0, &udp_table, NULL);
568 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
572 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
575 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
576 __be16 loc_port, __be32 loc_addr,
577 __be16 rmt_port, __be32 rmt_addr,
578 int dif, int sdif, unsigned short hnum)
580 struct inet_sock *inet = inet_sk(sk);
582 if (!net_eq(sock_net(sk), net) ||
583 udp_sk(sk)->udp_port_hash != hnum ||
584 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
585 (inet->inet_dport != rmt_port && inet->inet_dport) ||
586 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
587 ipv6_only_sock(sk) ||
588 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
590 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
595 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
596 void udp_encap_enable(void)
598 static_branch_inc(&udp_encap_needed_key);
600 EXPORT_SYMBOL(udp_encap_enable);
602 void udp_encap_disable(void)
604 static_branch_dec(&udp_encap_needed_key);
606 EXPORT_SYMBOL(udp_encap_disable);
608 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
609 * through error handlers in encapsulations looking for a match.
611 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
615 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
616 int (*handler)(struct sk_buff *skb, u32 info);
617 const struct ip_tunnel_encap_ops *encap;
619 encap = rcu_dereference(iptun_encaps[i]);
622 handler = encap->err_handler;
623 if (handler && !handler(skb, info))
630 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
631 * reversing source and destination port: this will match tunnels that force the
632 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
633 * lwtunnels might actually break this assumption by being configured with
634 * different destination ports on endpoints, in this case we won't be able to
635 * trace ICMP messages back to them.
637 * If this doesn't match any socket, probe tunnels with arbitrary destination
638 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
639 * we've sent packets to won't necessarily match the local destination port.
641 * Then ask the tunnel implementation to match the error against a valid
644 * Return an error if we can't find a match, the socket if we need further
645 * processing, zero otherwise.
647 static struct sock *__udp4_lib_err_encap(struct net *net,
648 const struct iphdr *iph,
650 struct udp_table *udptable,
652 struct sk_buff *skb, u32 info)
654 int (*lookup)(struct sock *sk, struct sk_buff *skb);
655 int network_offset, transport_offset;
658 network_offset = skb_network_offset(skb);
659 transport_offset = skb_transport_offset(skb);
661 /* Network header needs to point to the outer IPv4 header inside ICMP */
662 skb_reset_network_header(skb);
664 /* Transport header needs to point to the UDP header */
665 skb_set_transport_header(skb, iph->ihl << 2);
670 lookup = READ_ONCE(up->encap_err_lookup);
671 if (lookup && lookup(sk, skb))
677 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
678 iph->saddr, uh->dest, skb->dev->ifindex, 0,
683 lookup = READ_ONCE(up->encap_err_lookup);
684 if (!lookup || lookup(sk, skb))
690 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
692 skb_set_transport_header(skb, transport_offset);
693 skb_set_network_header(skb, network_offset);
699 * This routine is called by the ICMP module when it gets some
700 * sort of error condition. If err < 0 then the socket should
701 * be closed and the error returned to the user. If err > 0
702 * it's just the icmp type << 8 | icmp code.
703 * Header points to the ip header of the error packet. We move
704 * on past this. Then (as it used to claim before adjustment)
705 * header points to the first 8 bytes of the udp header. We need
706 * to find the appropriate port.
709 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
711 struct inet_sock *inet;
712 const struct iphdr *iph = (const struct iphdr *)skb->data;
713 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
714 const int type = icmp_hdr(skb)->type;
715 const int code = icmp_hdr(skb)->code;
720 struct net *net = dev_net(skb->dev);
722 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
723 iph->saddr, uh->source, skb->dev->ifindex,
724 inet_sdif(skb), udptable, NULL);
726 if (!sk || udp_sk(sk)->encap_type) {
727 /* No socket for error: try tunnels before discarding */
728 if (static_branch_unlikely(&udp_encap_needed_key)) {
729 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
734 sk = ERR_PTR(-ENOENT);
737 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
750 case ICMP_TIME_EXCEEDED:
753 case ICMP_SOURCE_QUENCH:
755 case ICMP_PARAMETERPROB:
759 case ICMP_DEST_UNREACH:
760 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
761 ipv4_sk_update_pmtu(skb, sk, info);
762 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
770 if (code <= NR_ICMP_UNREACH) {
771 harderr = icmp_err_convert[code].fatal;
772 err = icmp_err_convert[code].errno;
776 ipv4_sk_redirect(skb, sk);
781 * RFC1122: OK. Passes ICMP errors back to application, as per
785 /* ...not for tunnels though: we don't have a sending socket */
786 if (udp_sk(sk)->encap_err_rcv)
787 udp_sk(sk)->encap_err_rcv(sk, skb, iph->ihl << 2);
790 if (!inet->recverr) {
791 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
794 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
802 int udp_err(struct sk_buff *skb, u32 info)
804 return __udp4_lib_err(skb, info, &udp_table);
808 * Throw away all pending data and cancel the corking. Socket is locked.
810 void udp_flush_pending_frames(struct sock *sk)
812 struct udp_sock *up = udp_sk(sk);
817 ip_flush_pending_frames(sk);
820 EXPORT_SYMBOL(udp_flush_pending_frames);
823 * udp4_hwcsum - handle outgoing HW checksumming
824 * @skb: sk_buff containing the filled-in UDP header
825 * (checksum field must be zeroed out)
826 * @src: source IP address
827 * @dst: destination IP address
829 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
831 struct udphdr *uh = udp_hdr(skb);
832 int offset = skb_transport_offset(skb);
833 int len = skb->len - offset;
837 if (!skb_has_frag_list(skb)) {
839 * Only one fragment on the socket.
841 skb->csum_start = skb_transport_header(skb) - skb->head;
842 skb->csum_offset = offsetof(struct udphdr, check);
843 uh->check = ~csum_tcpudp_magic(src, dst, len,
846 struct sk_buff *frags;
849 * HW-checksum won't work as there are two or more
850 * fragments on the socket so that all csums of sk_buffs
853 skb_walk_frags(skb, frags) {
854 csum = csum_add(csum, frags->csum);
858 csum = skb_checksum(skb, offset, hlen, csum);
859 skb->ip_summed = CHECKSUM_NONE;
861 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
863 uh->check = CSUM_MANGLED_0;
866 EXPORT_SYMBOL_GPL(udp4_hwcsum);
868 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
869 * for the simple case like when setting the checksum for a UDP tunnel.
871 void udp_set_csum(bool nocheck, struct sk_buff *skb,
872 __be32 saddr, __be32 daddr, int len)
874 struct udphdr *uh = udp_hdr(skb);
878 } else if (skb_is_gso(skb)) {
879 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
880 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
882 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
884 uh->check = CSUM_MANGLED_0;
886 skb->ip_summed = CHECKSUM_PARTIAL;
887 skb->csum_start = skb_transport_header(skb) - skb->head;
888 skb->csum_offset = offsetof(struct udphdr, check);
889 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
892 EXPORT_SYMBOL(udp_set_csum);
894 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
895 struct inet_cork *cork)
897 struct sock *sk = skb->sk;
898 struct inet_sock *inet = inet_sk(sk);
901 int is_udplite = IS_UDPLITE(sk);
902 int offset = skb_transport_offset(skb);
903 int len = skb->len - offset;
904 int datalen = len - sizeof(*uh);
908 * Create a UDP header
911 uh->source = inet->inet_sport;
912 uh->dest = fl4->fl4_dport;
913 uh->len = htons(len);
916 if (cork->gso_size) {
917 const int hlen = skb_network_header_len(skb) +
918 sizeof(struct udphdr);
920 if (hlen + cork->gso_size > cork->fragsize) {
924 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
928 if (sk->sk_no_check_tx) {
932 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
933 dst_xfrm(skb_dst(skb))) {
938 if (datalen > cork->gso_size) {
939 skb_shinfo(skb)->gso_size = cork->gso_size;
940 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
941 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
947 if (is_udplite) /* UDP-Lite */
948 csum = udplite_csum(skb);
950 else if (sk->sk_no_check_tx) { /* UDP csum off */
952 skb->ip_summed = CHECKSUM_NONE;
955 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
958 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
962 csum = udp_csum(skb);
964 /* add protocol-dependent pseudo-header */
965 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
966 sk->sk_protocol, csum);
968 uh->check = CSUM_MANGLED_0;
971 err = ip_send_skb(sock_net(sk), skb);
973 if (err == -ENOBUFS && !inet->recverr) {
974 UDP_INC_STATS(sock_net(sk),
975 UDP_MIB_SNDBUFERRORS, is_udplite);
979 UDP_INC_STATS(sock_net(sk),
980 UDP_MIB_OUTDATAGRAMS, is_udplite);
985 * Push out all pending data as one UDP datagram. Socket is locked.
987 int udp_push_pending_frames(struct sock *sk)
989 struct udp_sock *up = udp_sk(sk);
990 struct inet_sock *inet = inet_sk(sk);
991 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
995 skb = ip_finish_skb(sk, fl4);
999 err = udp_send_skb(skb, fl4, &inet->cork.base);
1006 EXPORT_SYMBOL(udp_push_pending_frames);
1008 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1010 switch (cmsg->cmsg_type) {
1012 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1014 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1021 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1023 struct cmsghdr *cmsg;
1024 bool need_ip = false;
1027 for_each_cmsghdr(cmsg, msg) {
1028 if (!CMSG_OK(msg, cmsg))
1031 if (cmsg->cmsg_level != SOL_UDP) {
1036 err = __udp_cmsg_send(cmsg, gso_size);
1043 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1045 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1047 struct inet_sock *inet = inet_sk(sk);
1048 struct udp_sock *up = udp_sk(sk);
1049 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1050 struct flowi4 fl4_stack;
1053 struct ipcm_cookie ipc;
1054 struct rtable *rt = NULL;
1057 __be32 daddr, faddr, saddr;
1060 int err, is_udplite = IS_UDPLITE(sk);
1061 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1062 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1063 struct sk_buff *skb;
1064 struct ip_options_data opt_copy;
1073 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1076 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1078 fl4 = &inet->cork.fl.u.ip4;
1081 * There are pending frames.
1082 * The socket lock must be held while it's corked.
1085 if (likely(up->pending)) {
1086 if (unlikely(up->pending != AF_INET)) {
1090 goto do_append_data;
1094 ulen += sizeof(struct udphdr);
1097 * Get and verify the address.
1100 if (msg->msg_namelen < sizeof(*usin))
1102 if (usin->sin_family != AF_INET) {
1103 if (usin->sin_family != AF_UNSPEC)
1104 return -EAFNOSUPPORT;
1107 daddr = usin->sin_addr.s_addr;
1108 dport = usin->sin_port;
1112 if (sk->sk_state != TCP_ESTABLISHED)
1113 return -EDESTADDRREQ;
1114 daddr = inet->inet_daddr;
1115 dport = inet->inet_dport;
1116 /* Open fast path for connected socket.
1117 Route will not be used, if at least one option is set.
1122 ipcm_init_sk(&ipc, inet);
1123 ipc.gso_size = READ_ONCE(up->gso_size);
1125 if (msg->msg_controllen) {
1126 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1128 err = ip_cmsg_send(sk, msg, &ipc,
1129 sk->sk_family == AF_INET6);
1130 if (unlikely(err < 0)) {
1139 struct ip_options_rcu *inet_opt;
1142 inet_opt = rcu_dereference(inet->inet_opt);
1144 memcpy(&opt_copy, inet_opt,
1145 sizeof(*inet_opt) + inet_opt->opt.optlen);
1146 ipc.opt = &opt_copy.opt;
1151 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1152 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1153 (struct sockaddr *)usin, &ipc.addr);
1157 if (usin->sin_port == 0) {
1158 /* BPF program set invalid port. Reject it. */
1162 daddr = usin->sin_addr.s_addr;
1163 dport = usin->sin_port;
1168 ipc.addr = faddr = daddr;
1170 if (ipc.opt && ipc.opt->opt.srr) {
1175 faddr = ipc.opt->opt.faddr;
1178 tos = get_rttos(&ipc, inet);
1179 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1180 (msg->msg_flags & MSG_DONTROUTE) ||
1181 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1186 if (ipv4_is_multicast(daddr)) {
1187 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1188 ipc.oif = inet->mc_index;
1190 saddr = inet->mc_addr;
1192 } else if (!ipc.oif) {
1193 ipc.oif = inet->uc_index;
1194 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1195 /* oif is set, packet is to local broadcast and
1196 * uc_index is set. oif is most likely set
1197 * by sk_bound_dev_if. If uc_index != oif check if the
1198 * oif is an L3 master and uc_index is an L3 slave.
1199 * If so, we want to allow the send using the uc_index.
1201 if (ipc.oif != inet->uc_index &&
1202 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1204 ipc.oif = inet->uc_index;
1209 rt = (struct rtable *)sk_dst_check(sk, 0);
1212 struct net *net = sock_net(sk);
1213 __u8 flow_flags = inet_sk_flowi_flags(sk);
1217 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1218 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1220 faddr, saddr, dport, inet->inet_sport,
1223 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1224 rt = ip_route_output_flow(net, fl4, sk);
1228 if (err == -ENETUNREACH)
1229 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1234 if ((rt->rt_flags & RTCF_BROADCAST) &&
1235 !sock_flag(sk, SOCK_BROADCAST))
1238 sk_dst_set(sk, dst_clone(&rt->dst));
1241 if (msg->msg_flags&MSG_CONFIRM)
1247 daddr = ipc.addr = fl4->daddr;
1249 /* Lockless fast path for the non-corking case. */
1251 struct inet_cork cork;
1253 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1254 sizeof(struct udphdr), &ipc, &rt,
1255 &cork, msg->msg_flags);
1257 if (!IS_ERR_OR_NULL(skb))
1258 err = udp_send_skb(skb, fl4, &cork);
1263 if (unlikely(up->pending)) {
1264 /* The socket is already corked while preparing it. */
1265 /* ... which is an evident application bug. --ANK */
1268 net_dbg_ratelimited("socket already corked\n");
1273 * Now cork the socket to pend data.
1275 fl4 = &inet->cork.fl.u.ip4;
1278 fl4->fl4_dport = dport;
1279 fl4->fl4_sport = inet->inet_sport;
1280 up->pending = AF_INET;
1284 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1285 sizeof(struct udphdr), &ipc, &rt,
1286 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1288 udp_flush_pending_frames(sk);
1290 err = udp_push_pending_frames(sk);
1291 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1303 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1304 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1305 * we don't have a good statistic (IpOutDiscards but it can be too many
1306 * things). We could add another new stat but at least for now that
1307 * seems like overkill.
1309 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1310 UDP_INC_STATS(sock_net(sk),
1311 UDP_MIB_SNDBUFERRORS, is_udplite);
1316 if (msg->msg_flags & MSG_PROBE)
1317 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1318 if (!(msg->msg_flags&MSG_PROBE) || len)
1319 goto back_from_confirm;
1323 EXPORT_SYMBOL(udp_sendmsg);
1325 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1326 size_t size, int flags)
1328 struct inet_sock *inet = inet_sk(sk);
1329 struct udp_sock *up = udp_sk(sk);
1332 if (flags & MSG_SENDPAGE_NOTLAST)
1336 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1338 /* Call udp_sendmsg to specify destination address which
1339 * sendpage interface can't pass.
1340 * This will succeed only when the socket is connected.
1342 ret = udp_sendmsg(sk, &msg, 0);
1349 if (unlikely(!up->pending)) {
1352 net_dbg_ratelimited("cork failed\n");
1356 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1357 page, offset, size, flags);
1358 if (ret == -EOPNOTSUPP) {
1360 return sock_no_sendpage(sk->sk_socket, page, offset,
1364 udp_flush_pending_frames(sk);
1369 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1370 ret = udp_push_pending_frames(sk);
1378 #define UDP_SKB_IS_STATELESS 0x80000000
1380 /* all head states (dst, sk, nf conntrack) except skb extensions are
1381 * cleared by udp_rcv().
1383 * We need to preserve secpath, if present, to eventually process
1384 * IP_CMSG_PASSSEC at recvmsg() time.
1386 * Other extensions can be cleared.
1388 static bool udp_try_make_stateless(struct sk_buff *skb)
1390 if (!skb_has_extensions(skb))
1393 if (!secpath_exists(skb)) {
1401 static void udp_set_dev_scratch(struct sk_buff *skb)
1403 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1405 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1406 scratch->_tsize_state = skb->truesize;
1407 #if BITS_PER_LONG == 64
1408 scratch->len = skb->len;
1409 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1410 scratch->is_linear = !skb_is_nonlinear(skb);
1412 if (udp_try_make_stateless(skb))
1413 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1416 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1418 /* We come here after udp_lib_checksum_complete() returned 0.
1419 * This means that __skb_checksum_complete() might have
1420 * set skb->csum_valid to 1.
1421 * On 64bit platforms, we can set csum_unnecessary
1422 * to true, but only if the skb is not shared.
1424 #if BITS_PER_LONG == 64
1425 if (!skb_shared(skb))
1426 udp_skb_scratch(skb)->csum_unnecessary = true;
1430 static int udp_skb_truesize(struct sk_buff *skb)
1432 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1435 static bool udp_skb_has_head_state(struct sk_buff *skb)
1437 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1440 /* fully reclaim rmem/fwd memory allocated for skb */
1441 static void udp_rmem_release(struct sock *sk, int size, int partial,
1442 bool rx_queue_lock_held)
1444 struct udp_sock *up = udp_sk(sk);
1445 struct sk_buff_head *sk_queue;
1448 if (likely(partial)) {
1449 up->forward_deficit += size;
1450 size = up->forward_deficit;
1451 if (size < READ_ONCE(up->forward_threshold) &&
1452 !skb_queue_empty(&up->reader_queue))
1455 size += up->forward_deficit;
1457 up->forward_deficit = 0;
1459 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1460 * if the called don't held it already
1462 sk_queue = &sk->sk_receive_queue;
1463 if (!rx_queue_lock_held)
1464 spin_lock(&sk_queue->lock);
1467 sk->sk_forward_alloc += size;
1468 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1469 sk->sk_forward_alloc -= amt;
1472 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1474 atomic_sub(size, &sk->sk_rmem_alloc);
1476 /* this can save us from acquiring the rx queue lock on next receive */
1477 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1479 if (!rx_queue_lock_held)
1480 spin_unlock(&sk_queue->lock);
1483 /* Note: called with reader_queue.lock held.
1484 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1485 * This avoids a cache line miss while receive_queue lock is held.
1486 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1488 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1490 prefetch(&skb->data);
1491 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1493 EXPORT_SYMBOL(udp_skb_destructor);
1495 /* as above, but the caller held the rx queue lock, too */
1496 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1498 prefetch(&skb->data);
1499 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1502 /* Idea of busylocks is to let producers grab an extra spinlock
1503 * to relieve pressure on the receive_queue spinlock shared by consumer.
1504 * Under flood, this means that only one producer can be in line
1505 * trying to acquire the receive_queue spinlock.
1506 * These busylock can be allocated on a per cpu manner, instead of a
1507 * per socket one (that would consume a cache line per socket)
1509 static int udp_busylocks_log __read_mostly;
1510 static spinlock_t *udp_busylocks __read_mostly;
1512 static spinlock_t *busylock_acquire(void *ptr)
1516 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1521 static void busylock_release(spinlock_t *busy)
1527 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1529 struct sk_buff_head *list = &sk->sk_receive_queue;
1530 int rmem, delta, amt, err = -ENOMEM;
1531 spinlock_t *busy = NULL;
1534 /* try to avoid the costly atomic add/sub pair when the receive
1535 * queue is full; always allow at least a packet
1537 rmem = atomic_read(&sk->sk_rmem_alloc);
1538 if (rmem > sk->sk_rcvbuf)
1541 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1542 * having linear skbs :
1543 * - Reduce memory overhead and thus increase receive queue capacity
1544 * - Less cache line misses at copyout() time
1545 * - Less work at consume_skb() (less alien page frag freeing)
1547 if (rmem > (sk->sk_rcvbuf >> 1)) {
1550 busy = busylock_acquire(sk);
1552 size = skb->truesize;
1553 udp_set_dev_scratch(skb);
1555 /* we drop only if the receive buf is full and the receive
1556 * queue contains some other skb
1558 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1559 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1562 spin_lock(&list->lock);
1563 if (size >= sk->sk_forward_alloc) {
1564 amt = sk_mem_pages(size);
1565 delta = amt << PAGE_SHIFT;
1566 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1568 spin_unlock(&list->lock);
1572 sk->sk_forward_alloc += delta;
1575 sk->sk_forward_alloc -= size;
1577 /* no need to setup a destructor, we will explicitly release the
1578 * forward allocated memory on dequeue
1580 sock_skb_set_dropcount(sk, skb);
1582 __skb_queue_tail(list, skb);
1583 spin_unlock(&list->lock);
1585 if (!sock_flag(sk, SOCK_DEAD))
1586 sk->sk_data_ready(sk);
1588 busylock_release(busy);
1592 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1595 atomic_inc(&sk->sk_drops);
1596 busylock_release(busy);
1599 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1601 void udp_destruct_common(struct sock *sk)
1603 /* reclaim completely the forward allocated memory */
1604 struct udp_sock *up = udp_sk(sk);
1605 unsigned int total = 0;
1606 struct sk_buff *skb;
1608 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1609 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1610 total += skb->truesize;
1613 udp_rmem_release(sk, total, 0, true);
1615 EXPORT_SYMBOL_GPL(udp_destruct_common);
1617 static void udp_destruct_sock(struct sock *sk)
1619 udp_destruct_common(sk);
1620 inet_sock_destruct(sk);
1623 int udp_init_sock(struct sock *sk)
1625 udp_lib_init_sock(sk);
1626 sk->sk_destruct = udp_destruct_sock;
1627 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1631 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1633 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1634 bool slow = lock_sock_fast(sk);
1636 sk_peek_offset_bwd(sk, len);
1637 unlock_sock_fast(sk, slow);
1640 if (!skb_unref(skb))
1643 /* In the more common cases we cleared the head states previously,
1644 * see __udp_queue_rcv_skb().
1646 if (unlikely(udp_skb_has_head_state(skb)))
1647 skb_release_head_state(skb);
1648 __consume_stateless_skb(skb);
1650 EXPORT_SYMBOL_GPL(skb_consume_udp);
1652 static struct sk_buff *__first_packet_length(struct sock *sk,
1653 struct sk_buff_head *rcvq,
1656 struct sk_buff *skb;
1658 while ((skb = skb_peek(rcvq)) != NULL) {
1659 if (udp_lib_checksum_complete(skb)) {
1660 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1662 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1664 atomic_inc(&sk->sk_drops);
1665 __skb_unlink(skb, rcvq);
1666 *total += skb->truesize;
1669 udp_skb_csum_unnecessary_set(skb);
1677 * first_packet_length - return length of first packet in receive queue
1680 * Drops all bad checksum frames, until a valid one is found.
1681 * Returns the length of found skb, or -1 if none is found.
1683 static int first_packet_length(struct sock *sk)
1685 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1686 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1687 struct sk_buff *skb;
1691 spin_lock_bh(&rcvq->lock);
1692 skb = __first_packet_length(sk, rcvq, &total);
1693 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1694 spin_lock(&sk_queue->lock);
1695 skb_queue_splice_tail_init(sk_queue, rcvq);
1696 spin_unlock(&sk_queue->lock);
1698 skb = __first_packet_length(sk, rcvq, &total);
1700 res = skb ? skb->len : -1;
1702 udp_rmem_release(sk, total, 1, false);
1703 spin_unlock_bh(&rcvq->lock);
1708 * IOCTL requests applicable to the UDP protocol
1711 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1716 int amount = sk_wmem_alloc_get(sk);
1718 return put_user(amount, (int __user *)arg);
1723 int amount = max_t(int, 0, first_packet_length(sk));
1725 return put_user(amount, (int __user *)arg);
1729 return -ENOIOCTLCMD;
1734 EXPORT_SYMBOL(udp_ioctl);
1736 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1739 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1740 struct sk_buff_head *queue;
1741 struct sk_buff *last;
1745 queue = &udp_sk(sk)->reader_queue;
1746 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1748 struct sk_buff *skb;
1750 error = sock_error(sk);
1756 spin_lock_bh(&queue->lock);
1757 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1760 if (!(flags & MSG_PEEK))
1761 udp_skb_destructor(sk, skb);
1762 spin_unlock_bh(&queue->lock);
1766 if (skb_queue_empty_lockless(sk_queue)) {
1767 spin_unlock_bh(&queue->lock);
1771 /* refill the reader queue and walk it again
1772 * keep both queues locked to avoid re-acquiring
1773 * the sk_receive_queue lock if fwd memory scheduling
1776 spin_lock(&sk_queue->lock);
1777 skb_queue_splice_tail_init(sk_queue, queue);
1779 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1781 if (skb && !(flags & MSG_PEEK))
1782 udp_skb_dtor_locked(sk, skb);
1783 spin_unlock(&sk_queue->lock);
1784 spin_unlock_bh(&queue->lock);
1789 if (!sk_can_busy_loop(sk))
1792 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1793 } while (!skb_queue_empty_lockless(sk_queue));
1795 /* sk_queue is empty, reader_queue may contain peeked packets */
1797 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1799 (struct sk_buff *)sk_queue));
1804 EXPORT_SYMBOL(__skb_recv_udp);
1806 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1808 struct sk_buff *skb;
1812 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1816 if (udp_lib_checksum_complete(skb)) {
1817 int is_udplite = IS_UDPLITE(sk);
1818 struct net *net = sock_net(sk);
1820 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1821 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1822 atomic_inc(&sk->sk_drops);
1827 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1828 copied = recv_actor(sk, skb);
1833 EXPORT_SYMBOL(udp_read_skb);
1836 * This should be easy, if there is something there we
1837 * return it, otherwise we block.
1840 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1843 struct inet_sock *inet = inet_sk(sk);
1844 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1845 struct sk_buff *skb;
1846 unsigned int ulen, copied;
1847 int off, err, peeking = flags & MSG_PEEK;
1848 int is_udplite = IS_UDPLITE(sk);
1849 bool checksum_valid = false;
1851 if (flags & MSG_ERRQUEUE)
1852 return ip_recv_error(sk, msg, len, addr_len);
1855 off = sk_peek_offset(sk, flags);
1856 skb = __skb_recv_udp(sk, flags, &off, &err);
1860 ulen = udp_skb_len(skb);
1862 if (copied > ulen - off)
1863 copied = ulen - off;
1864 else if (copied < ulen)
1865 msg->msg_flags |= MSG_TRUNC;
1868 * If checksum is needed at all, try to do it while copying the
1869 * data. If the data is truncated, or if we only want a partial
1870 * coverage checksum (UDP-Lite), do it before the copy.
1873 if (copied < ulen || peeking ||
1874 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1875 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1876 !__udp_lib_checksum_complete(skb);
1877 if (!checksum_valid)
1881 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1882 if (udp_skb_is_linear(skb))
1883 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1885 err = skb_copy_datagram_msg(skb, off, msg, copied);
1887 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1893 if (unlikely(err)) {
1895 atomic_inc(&sk->sk_drops);
1896 UDP_INC_STATS(sock_net(sk),
1897 UDP_MIB_INERRORS, is_udplite);
1904 UDP_INC_STATS(sock_net(sk),
1905 UDP_MIB_INDATAGRAMS, is_udplite);
1907 sock_recv_cmsgs(msg, sk, skb);
1909 /* Copy the address. */
1911 sin->sin_family = AF_INET;
1912 sin->sin_port = udp_hdr(skb)->source;
1913 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1914 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1915 *addr_len = sizeof(*sin);
1917 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1918 (struct sockaddr *)sin);
1921 if (udp_sk(sk)->gro_enabled)
1922 udp_cmsg_recv(msg, sk, skb);
1924 if (inet->cmsg_flags)
1925 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1928 if (flags & MSG_TRUNC)
1931 skb_consume_udp(sk, skb, peeking ? -err : err);
1935 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1936 udp_skb_destructor)) {
1937 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1938 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1942 /* starting over for a new packet, but check if we need to yield */
1944 msg->msg_flags &= ~MSG_TRUNC;
1948 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1950 /* This check is replicated from __ip4_datagram_connect() and
1951 * intended to prevent BPF program called below from accessing bytes
1952 * that are out of the bound specified by user in addr_len.
1954 if (addr_len < sizeof(struct sockaddr_in))
1957 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1959 EXPORT_SYMBOL(udp_pre_connect);
1961 int __udp_disconnect(struct sock *sk, int flags)
1963 struct inet_sock *inet = inet_sk(sk);
1965 * 1003.1g - break association.
1968 sk->sk_state = TCP_CLOSE;
1969 inet->inet_daddr = 0;
1970 inet->inet_dport = 0;
1971 sock_rps_reset_rxhash(sk);
1972 sk->sk_bound_dev_if = 0;
1973 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1974 inet_reset_saddr(sk);
1975 if (sk->sk_prot->rehash &&
1976 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1977 sk->sk_prot->rehash(sk);
1980 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1981 sk->sk_prot->unhash(sk);
1982 inet->inet_sport = 0;
1987 EXPORT_SYMBOL(__udp_disconnect);
1989 int udp_disconnect(struct sock *sk, int flags)
1992 __udp_disconnect(sk, flags);
1996 EXPORT_SYMBOL(udp_disconnect);
1998 void udp_lib_unhash(struct sock *sk)
2000 if (sk_hashed(sk)) {
2001 struct udp_table *udptable = sk->sk_prot->h.udp_table;
2002 struct udp_hslot *hslot, *hslot2;
2004 hslot = udp_hashslot(udptable, sock_net(sk),
2005 udp_sk(sk)->udp_port_hash);
2006 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2008 spin_lock_bh(&hslot->lock);
2009 if (rcu_access_pointer(sk->sk_reuseport_cb))
2010 reuseport_detach_sock(sk);
2011 if (sk_del_node_init_rcu(sk)) {
2013 inet_sk(sk)->inet_num = 0;
2014 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2016 spin_lock(&hslot2->lock);
2017 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2019 spin_unlock(&hslot2->lock);
2021 spin_unlock_bh(&hslot->lock);
2024 EXPORT_SYMBOL(udp_lib_unhash);
2027 * inet_rcv_saddr was changed, we must rehash secondary hash
2029 void udp_lib_rehash(struct sock *sk, u16 newhash)
2031 if (sk_hashed(sk)) {
2032 struct udp_table *udptable = sk->sk_prot->h.udp_table;
2033 struct udp_hslot *hslot, *hslot2, *nhslot2;
2035 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2036 nhslot2 = udp_hashslot2(udptable, newhash);
2037 udp_sk(sk)->udp_portaddr_hash = newhash;
2039 if (hslot2 != nhslot2 ||
2040 rcu_access_pointer(sk->sk_reuseport_cb)) {
2041 hslot = udp_hashslot(udptable, sock_net(sk),
2042 udp_sk(sk)->udp_port_hash);
2043 /* we must lock primary chain too */
2044 spin_lock_bh(&hslot->lock);
2045 if (rcu_access_pointer(sk->sk_reuseport_cb))
2046 reuseport_detach_sock(sk);
2048 if (hslot2 != nhslot2) {
2049 spin_lock(&hslot2->lock);
2050 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2052 spin_unlock(&hslot2->lock);
2054 spin_lock(&nhslot2->lock);
2055 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2058 spin_unlock(&nhslot2->lock);
2061 spin_unlock_bh(&hslot->lock);
2065 EXPORT_SYMBOL(udp_lib_rehash);
2067 void udp_v4_rehash(struct sock *sk)
2069 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2070 inet_sk(sk)->inet_rcv_saddr,
2071 inet_sk(sk)->inet_num);
2072 udp_lib_rehash(sk, new_hash);
2075 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2079 if (inet_sk(sk)->inet_daddr) {
2080 sock_rps_save_rxhash(sk, skb);
2081 sk_mark_napi_id(sk, skb);
2082 sk_incoming_cpu_update(sk);
2084 sk_mark_napi_id_once(sk, skb);
2087 rc = __udp_enqueue_schedule_skb(sk, skb);
2089 int is_udplite = IS_UDPLITE(sk);
2092 /* Note that an ENOMEM error is charged twice */
2093 if (rc == -ENOMEM) {
2094 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2096 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2098 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2100 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2102 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2103 kfree_skb_reason(skb, drop_reason);
2104 trace_udp_fail_queue_rcv_skb(rc, sk);
2114 * >0: "udp encap" protocol resubmission
2116 * Note that in the success and error cases, the skb is assumed to
2117 * have either been requeued or freed.
2119 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2121 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2122 struct udp_sock *up = udp_sk(sk);
2123 int is_udplite = IS_UDPLITE(sk);
2126 * Charge it to the socket, dropping if the queue is full.
2128 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2129 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2134 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2135 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2138 * This is an encapsulation socket so pass the skb to
2139 * the socket's udp_encap_rcv() hook. Otherwise, just
2140 * fall through and pass this up the UDP socket.
2141 * up->encap_rcv() returns the following value:
2142 * =0 if skb was successfully passed to the encap
2143 * handler or was discarded by it.
2144 * >0 if skb should be passed on to UDP.
2145 * <0 if skb should be resubmitted as proto -N
2148 /* if we're overly short, let UDP handle it */
2149 encap_rcv = READ_ONCE(up->encap_rcv);
2153 /* Verify checksum before giving to encap */
2154 if (udp_lib_checksum_complete(skb))
2157 ret = encap_rcv(sk, skb);
2159 __UDP_INC_STATS(sock_net(sk),
2160 UDP_MIB_INDATAGRAMS,
2166 /* FALLTHROUGH -- it's a UDP Packet */
2170 * UDP-Lite specific tests, ignored on UDP sockets
2172 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2175 * MIB statistics other than incrementing the error count are
2176 * disabled for the following two types of errors: these depend
2177 * on the application settings, not on the functioning of the
2178 * protocol stack as such.
2180 * RFC 3828 here recommends (sec 3.3): "There should also be a
2181 * way ... to ... at least let the receiving application block
2182 * delivery of packets with coverage values less than a value
2183 * provided by the application."
2185 if (up->pcrlen == 0) { /* full coverage was set */
2186 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2187 UDP_SKB_CB(skb)->cscov, skb->len);
2190 /* The next case involves violating the min. coverage requested
2191 * by the receiver. This is subtle: if receiver wants x and x is
2192 * greater than the buffersize/MTU then receiver will complain
2193 * that it wants x while sender emits packets of smaller size y.
2194 * Therefore the above ...()->partial_cov statement is essential.
2196 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2197 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2198 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2203 prefetch(&sk->sk_rmem_alloc);
2204 if (rcu_access_pointer(sk->sk_filter) &&
2205 udp_lib_checksum_complete(skb))
2208 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2209 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2213 udp_csum_pull_header(skb);
2215 ipv4_pktinfo_prepare(sk, skb);
2216 return __udp_queue_rcv_skb(sk, skb);
2219 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2220 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2222 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2223 atomic_inc(&sk->sk_drops);
2224 kfree_skb_reason(skb, drop_reason);
2228 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2230 struct sk_buff *next, *segs;
2233 if (likely(!udp_unexpected_gso(sk, skb)))
2234 return udp_queue_rcv_one_skb(sk, skb);
2236 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2237 __skb_push(skb, -skb_mac_offset(skb));
2238 segs = udp_rcv_segment(sk, skb, true);
2239 skb_list_walk_safe(segs, skb, next) {
2240 __skb_pull(skb, skb_transport_offset(skb));
2242 udp_post_segment_fix_csum(skb);
2243 ret = udp_queue_rcv_one_skb(sk, skb);
2245 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2250 /* For TCP sockets, sk_rx_dst is protected by socket lock
2251 * For UDP, we use xchg() to guard against concurrent changes.
2253 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2255 struct dst_entry *old;
2257 if (dst_hold_safe(dst)) {
2258 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2264 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2267 * Multicasts and broadcasts go to each listener.
2269 * Note: called only from the BH handler context.
2271 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2273 __be32 saddr, __be32 daddr,
2274 struct udp_table *udptable,
2277 struct sock *sk, *first = NULL;
2278 unsigned short hnum = ntohs(uh->dest);
2279 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2280 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2281 unsigned int offset = offsetof(typeof(*sk), sk_node);
2282 int dif = skb->dev->ifindex;
2283 int sdif = inet_sdif(skb);
2284 struct hlist_node *node;
2285 struct sk_buff *nskb;
2288 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2290 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2292 hslot = &udptable->hash2[hash2];
2293 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2296 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2297 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2298 uh->source, saddr, dif, sdif, hnum))
2305 nskb = skb_clone(skb, GFP_ATOMIC);
2307 if (unlikely(!nskb)) {
2308 atomic_inc(&sk->sk_drops);
2309 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2311 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2315 if (udp_queue_rcv_skb(sk, nskb) > 0)
2319 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2320 if (use_hash2 && hash2 != hash2_any) {
2326 if (udp_queue_rcv_skb(first, skb) > 0)
2330 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2331 proto == IPPROTO_UDPLITE);
2336 /* Initialize UDP checksum. If exited with zero value (success),
2337 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2338 * Otherwise, csum completion requires checksumming packet body,
2339 * including udp header and folding it to skb->csum.
2341 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2346 UDP_SKB_CB(skb)->partial_cov = 0;
2347 UDP_SKB_CB(skb)->cscov = skb->len;
2349 if (proto == IPPROTO_UDPLITE) {
2350 err = udplite_checksum_init(skb, uh);
2354 if (UDP_SKB_CB(skb)->partial_cov) {
2355 skb->csum = inet_compute_pseudo(skb, proto);
2360 /* Note, we are only interested in != 0 or == 0, thus the
2363 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2364 inet_compute_pseudo);
2368 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2369 /* If SW calculated the value, we know it's bad */
2370 if (skb->csum_complete_sw)
2373 /* HW says the value is bad. Let's validate that.
2374 * skb->csum is no longer the full packet checksum,
2375 * so don't treat it as such.
2377 skb_checksum_complete_unset(skb);
2383 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2384 * return code conversion for ip layer consumption
2386 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2391 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2392 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2394 ret = udp_queue_rcv_skb(sk, skb);
2396 /* a return value > 0 means to resubmit the input, but
2397 * it wants the return to be -protocol, or 0
2405 * All we need to do is get the socket, and then do a checksum.
2408 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2413 unsigned short ulen;
2414 struct rtable *rt = skb_rtable(skb);
2415 __be32 saddr, daddr;
2416 struct net *net = dev_net(skb->dev);
2420 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2423 * Validate the packet.
2425 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2426 goto drop; /* No space for header. */
2429 ulen = ntohs(uh->len);
2430 saddr = ip_hdr(skb)->saddr;
2431 daddr = ip_hdr(skb)->daddr;
2433 if (ulen > skb->len)
2436 if (proto == IPPROTO_UDP) {
2437 /* UDP validates ulen. */
2438 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2443 if (udp4_csum_init(skb, uh, proto))
2446 sk = skb_steal_sock(skb, &refcounted);
2448 struct dst_entry *dst = skb_dst(skb);
2451 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2452 udp_sk_rx_dst_set(sk, dst);
2454 ret = udp_unicast_rcv_skb(sk, skb, uh);
2460 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2461 return __udp4_lib_mcast_deliver(net, skb, uh,
2462 saddr, daddr, udptable, proto);
2464 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2466 return udp_unicast_rcv_skb(sk, skb, uh);
2468 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2472 /* No socket. Drop packet silently, if checksum is wrong */
2473 if (udp_lib_checksum_complete(skb))
2476 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2477 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2478 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2481 * Hmm. We got an UDP packet to a port to which we
2482 * don't wanna listen. Ignore it.
2484 kfree_skb_reason(skb, drop_reason);
2488 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2489 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2490 proto == IPPROTO_UDPLITE ? "Lite" : "",
2491 &saddr, ntohs(uh->source),
2493 &daddr, ntohs(uh->dest));
2498 * RFC1122: OK. Discards the bad packet silently (as far as
2499 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2501 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2502 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2503 proto == IPPROTO_UDPLITE ? "Lite" : "",
2504 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2506 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2508 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2509 kfree_skb_reason(skb, drop_reason);
2513 /* We can only early demux multicast if there is a single matching socket.
2514 * If more than one socket found returns NULL
2516 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2517 __be16 loc_port, __be32 loc_addr,
2518 __be16 rmt_port, __be32 rmt_addr,
2521 struct sock *sk, *result;
2522 unsigned short hnum = ntohs(loc_port);
2523 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2524 struct udp_hslot *hslot = &udp_table.hash[slot];
2526 /* Do not bother scanning a too big list */
2527 if (hslot->count > 10)
2531 sk_for_each_rcu(sk, &hslot->head) {
2532 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2533 rmt_port, rmt_addr, dif, sdif, hnum)) {
2543 /* For unicast we should only early demux connected sockets or we can
2544 * break forwarding setups. The chains here can be long so only check
2545 * if the first socket is an exact match and if not move on.
2547 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2548 __be16 loc_port, __be32 loc_addr,
2549 __be16 rmt_port, __be32 rmt_addr,
2552 unsigned short hnum = ntohs(loc_port);
2553 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2554 unsigned int slot2 = hash2 & udp_table.mask;
2555 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2556 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2557 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2560 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2561 if (inet_match(net, sk, acookie, ports, dif, sdif))
2563 /* Only check first socket in chain */
2569 int udp_v4_early_demux(struct sk_buff *skb)
2571 struct net *net = dev_net(skb->dev);
2572 struct in_device *in_dev = NULL;
2573 const struct iphdr *iph;
2574 const struct udphdr *uh;
2575 struct sock *sk = NULL;
2576 struct dst_entry *dst;
2577 int dif = skb->dev->ifindex;
2578 int sdif = inet_sdif(skb);
2581 /* validate the packet */
2582 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2588 if (skb->pkt_type == PACKET_MULTICAST) {
2589 in_dev = __in_dev_get_rcu(skb->dev);
2594 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2599 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2600 uh->source, iph->saddr,
2602 } else if (skb->pkt_type == PACKET_HOST) {
2603 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2604 uh->source, iph->saddr, dif, sdif);
2607 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2611 skb->destructor = sock_efree;
2612 dst = rcu_dereference(sk->sk_rx_dst);
2615 dst = dst_check(dst, 0);
2619 /* set noref for now.
2620 * any place which wants to hold dst has to call
2623 skb_dst_set_noref(skb, dst);
2625 /* for unconnected multicast sockets we need to validate
2626 * the source on each packet
2628 if (!inet_sk(sk)->inet_daddr && in_dev)
2629 return ip_mc_validate_source(skb, iph->daddr,
2631 iph->tos & IPTOS_RT_MASK,
2632 skb->dev, in_dev, &itag);
2637 int udp_rcv(struct sk_buff *skb)
2639 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2642 void udp_destroy_sock(struct sock *sk)
2644 struct udp_sock *up = udp_sk(sk);
2645 bool slow = lock_sock_fast(sk);
2647 /* protects from races with udp_abort() */
2648 sock_set_flag(sk, SOCK_DEAD);
2649 udp_flush_pending_frames(sk);
2650 unlock_sock_fast(sk, slow);
2651 if (static_branch_unlikely(&udp_encap_needed_key)) {
2652 if (up->encap_type) {
2653 void (*encap_destroy)(struct sock *sk);
2654 encap_destroy = READ_ONCE(up->encap_destroy);
2658 if (up->encap_enabled)
2659 static_branch_dec(&udp_encap_needed_key);
2664 * Socket option code for UDP
2666 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2667 sockptr_t optval, unsigned int optlen,
2668 int (*push_pending_frames)(struct sock *))
2670 struct udp_sock *up = udp_sk(sk);
2673 int is_udplite = IS_UDPLITE(sk);
2675 if (level == SOL_SOCKET) {
2676 err = sk_setsockopt(sk, level, optname, optval, optlen);
2678 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2679 sockopt_lock_sock(sk);
2680 /* paired with READ_ONCE in udp_rmem_release() */
2681 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2682 sockopt_release_sock(sk);
2687 if (optlen < sizeof(int))
2690 if (copy_from_sockptr(&val, optval, sizeof(val)))
2693 valbool = val ? 1 : 0;
2698 WRITE_ONCE(up->corkflag, 1);
2700 WRITE_ONCE(up->corkflag, 0);
2702 push_pending_frames(sk);
2711 case UDP_ENCAP_ESPINUDP:
2712 case UDP_ENCAP_ESPINUDP_NON_IKE:
2713 #if IS_ENABLED(CONFIG_IPV6)
2714 if (sk->sk_family == AF_INET6)
2715 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2718 up->encap_rcv = xfrm4_udp_encap_rcv;
2721 case UDP_ENCAP_L2TPINUDP:
2722 up->encap_type = val;
2724 udp_tunnel_encap_enable(sk->sk_socket);
2733 case UDP_NO_CHECK6_TX:
2734 up->no_check6_tx = valbool;
2737 case UDP_NO_CHECK6_RX:
2738 up->no_check6_rx = valbool;
2742 if (val < 0 || val > USHRT_MAX)
2744 WRITE_ONCE(up->gso_size, val);
2750 /* when enabling GRO, accept the related GSO packet type */
2752 udp_tunnel_encap_enable(sk->sk_socket);
2753 up->gro_enabled = valbool;
2754 up->accept_udp_l4 = valbool;
2759 * UDP-Lite's partial checksum coverage (RFC 3828).
2761 /* The sender sets actual checksum coverage length via this option.
2762 * The case coverage > packet length is handled by send module. */
2763 case UDPLITE_SEND_CSCOV:
2764 if (!is_udplite) /* Disable the option on UDP sockets */
2765 return -ENOPROTOOPT;
2766 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2768 else if (val > USHRT_MAX)
2771 up->pcflag |= UDPLITE_SEND_CC;
2774 /* The receiver specifies a minimum checksum coverage value. To make
2775 * sense, this should be set to at least 8 (as done below). If zero is
2776 * used, this again means full checksum coverage. */
2777 case UDPLITE_RECV_CSCOV:
2778 if (!is_udplite) /* Disable the option on UDP sockets */
2779 return -ENOPROTOOPT;
2780 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2782 else if (val > USHRT_MAX)
2785 up->pcflag |= UDPLITE_RECV_CC;
2795 EXPORT_SYMBOL(udp_lib_setsockopt);
2797 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2798 unsigned int optlen)
2800 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2801 return udp_lib_setsockopt(sk, level, optname,
2803 udp_push_pending_frames);
2804 return ip_setsockopt(sk, level, optname, optval, optlen);
2807 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2808 char __user *optval, int __user *optlen)
2810 struct udp_sock *up = udp_sk(sk);
2813 if (get_user(len, optlen))
2816 len = min_t(unsigned int, len, sizeof(int));
2823 val = READ_ONCE(up->corkflag);
2827 val = up->encap_type;
2830 case UDP_NO_CHECK6_TX:
2831 val = up->no_check6_tx;
2834 case UDP_NO_CHECK6_RX:
2835 val = up->no_check6_rx;
2839 val = READ_ONCE(up->gso_size);
2843 val = up->gro_enabled;
2846 /* The following two cannot be changed on UDP sockets, the return is
2847 * always 0 (which corresponds to the full checksum coverage of UDP). */
2848 case UDPLITE_SEND_CSCOV:
2852 case UDPLITE_RECV_CSCOV:
2857 return -ENOPROTOOPT;
2860 if (put_user(len, optlen))
2862 if (copy_to_user(optval, &val, len))
2866 EXPORT_SYMBOL(udp_lib_getsockopt);
2868 int udp_getsockopt(struct sock *sk, int level, int optname,
2869 char __user *optval, int __user *optlen)
2871 if (level == SOL_UDP || level == SOL_UDPLITE)
2872 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2873 return ip_getsockopt(sk, level, optname, optval, optlen);
2877 * udp_poll - wait for a UDP event.
2878 * @file: - file struct
2880 * @wait: - poll table
2882 * This is same as datagram poll, except for the special case of
2883 * blocking sockets. If application is using a blocking fd
2884 * and a packet with checksum error is in the queue;
2885 * then it could get return from select indicating data available
2886 * but then block when reading it. Add special case code
2887 * to work around these arguably broken applications.
2889 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2891 __poll_t mask = datagram_poll(file, sock, wait);
2892 struct sock *sk = sock->sk;
2894 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2895 mask |= EPOLLIN | EPOLLRDNORM;
2897 /* Check for false positives due to checksum errors */
2898 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2899 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2900 mask &= ~(EPOLLIN | EPOLLRDNORM);
2902 /* psock ingress_msg queue should not contain any bad checksum frames */
2903 if (sk_is_readable(sk))
2904 mask |= EPOLLIN | EPOLLRDNORM;
2908 EXPORT_SYMBOL(udp_poll);
2910 int udp_abort(struct sock *sk, int err)
2914 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2917 if (sock_flag(sk, SOCK_DEAD))
2921 sk_error_report(sk);
2922 __udp_disconnect(sk, 0);
2929 EXPORT_SYMBOL_GPL(udp_abort);
2931 struct proto udp_prot = {
2933 .owner = THIS_MODULE,
2934 .close = udp_lib_close,
2935 .pre_connect = udp_pre_connect,
2936 .connect = ip4_datagram_connect,
2937 .disconnect = udp_disconnect,
2939 .init = udp_init_sock,
2940 .destroy = udp_destroy_sock,
2941 .setsockopt = udp_setsockopt,
2942 .getsockopt = udp_getsockopt,
2943 .sendmsg = udp_sendmsg,
2944 .recvmsg = udp_recvmsg,
2945 .sendpage = udp_sendpage,
2946 .release_cb = ip4_datagram_release_cb,
2947 .hash = udp_lib_hash,
2948 .unhash = udp_lib_unhash,
2949 .rehash = udp_v4_rehash,
2950 .get_port = udp_v4_get_port,
2951 .put_port = udp_lib_unhash,
2952 #ifdef CONFIG_BPF_SYSCALL
2953 .psock_update_sk_prot = udp_bpf_update_proto,
2955 .memory_allocated = &udp_memory_allocated,
2956 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2958 .sysctl_mem = sysctl_udp_mem,
2959 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2960 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2961 .obj_size = sizeof(struct udp_sock),
2962 .h.udp_table = &udp_table,
2963 .diag_destroy = udp_abort,
2965 EXPORT_SYMBOL(udp_prot);
2967 /* ------------------------------------------------------------------------ */
2968 #ifdef CONFIG_PROC_FS
2970 static struct sock *udp_get_first(struct seq_file *seq, int start)
2973 struct udp_seq_afinfo *afinfo;
2974 struct udp_iter_state *state = seq->private;
2975 struct net *net = seq_file_net(seq);
2977 if (state->bpf_seq_afinfo)
2978 afinfo = state->bpf_seq_afinfo;
2980 afinfo = pde_data(file_inode(seq->file));
2982 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2984 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2986 if (hlist_empty(&hslot->head))
2989 spin_lock_bh(&hslot->lock);
2990 sk_for_each(sk, &hslot->head) {
2991 if (!net_eq(sock_net(sk), net))
2993 if (afinfo->family == AF_UNSPEC ||
2994 sk->sk_family == afinfo->family)
2997 spin_unlock_bh(&hslot->lock);
3004 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3006 struct udp_seq_afinfo *afinfo;
3007 struct udp_iter_state *state = seq->private;
3008 struct net *net = seq_file_net(seq);
3010 if (state->bpf_seq_afinfo)
3011 afinfo = state->bpf_seq_afinfo;
3013 afinfo = pde_data(file_inode(seq->file));
3017 } while (sk && (!net_eq(sock_net(sk), net) ||
3018 (afinfo->family != AF_UNSPEC &&
3019 sk->sk_family != afinfo->family)));
3022 if (state->bucket <= afinfo->udp_table->mask)
3023 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
3024 return udp_get_first(seq, state->bucket + 1);
3029 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3031 struct sock *sk = udp_get_first(seq, 0);
3034 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3036 return pos ? NULL : sk;
3039 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3041 struct udp_iter_state *state = seq->private;
3042 state->bucket = MAX_UDP_PORTS;
3044 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3046 EXPORT_SYMBOL(udp_seq_start);
3048 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3052 if (v == SEQ_START_TOKEN)
3053 sk = udp_get_idx(seq, 0);
3055 sk = udp_get_next(seq, v);
3060 EXPORT_SYMBOL(udp_seq_next);
3062 void udp_seq_stop(struct seq_file *seq, void *v)
3064 struct udp_seq_afinfo *afinfo;
3065 struct udp_iter_state *state = seq->private;
3067 if (state->bpf_seq_afinfo)
3068 afinfo = state->bpf_seq_afinfo;
3070 afinfo = pde_data(file_inode(seq->file));
3072 if (state->bucket <= afinfo->udp_table->mask)
3073 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
3075 EXPORT_SYMBOL(udp_seq_stop);
3077 /* ------------------------------------------------------------------------ */
3078 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3081 struct inet_sock *inet = inet_sk(sp);
3082 __be32 dest = inet->inet_daddr;
3083 __be32 src = inet->inet_rcv_saddr;
3084 __u16 destp = ntohs(inet->inet_dport);
3085 __u16 srcp = ntohs(inet->inet_sport);
3087 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3088 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3089 bucket, src, srcp, dest, destp, sp->sk_state,
3090 sk_wmem_alloc_get(sp),
3093 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3095 refcount_read(&sp->sk_refcnt), sp,
3096 atomic_read(&sp->sk_drops));
3099 int udp4_seq_show(struct seq_file *seq, void *v)
3101 seq_setwidth(seq, 127);
3102 if (v == SEQ_START_TOKEN)
3103 seq_puts(seq, " sl local_address rem_address st tx_queue "
3104 "rx_queue tr tm->when retrnsmt uid timeout "
3105 "inode ref pointer drops");
3107 struct udp_iter_state *state = seq->private;
3109 udp4_format_sock(v, seq, state->bucket);
3115 #ifdef CONFIG_BPF_SYSCALL
3116 struct bpf_iter__udp {
3117 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3118 __bpf_md_ptr(struct udp_sock *, udp_sk);
3119 uid_t uid __aligned(8);
3120 int bucket __aligned(8);
3123 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3124 struct udp_sock *udp_sk, uid_t uid, int bucket)
3126 struct bpf_iter__udp ctx;
3128 meta->seq_num--; /* skip SEQ_START_TOKEN */
3130 ctx.udp_sk = udp_sk;
3132 ctx.bucket = bucket;
3133 return bpf_iter_run_prog(prog, &ctx);
3136 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3138 struct udp_iter_state *state = seq->private;
3139 struct bpf_iter_meta meta;
3140 struct bpf_prog *prog;
3141 struct sock *sk = v;
3144 if (v == SEQ_START_TOKEN)
3147 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3149 prog = bpf_iter_get_info(&meta, false);
3150 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3153 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3155 struct bpf_iter_meta meta;
3156 struct bpf_prog *prog;
3160 prog = bpf_iter_get_info(&meta, true);
3162 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3165 udp_seq_stop(seq, v);
3168 static const struct seq_operations bpf_iter_udp_seq_ops = {
3169 .start = udp_seq_start,
3170 .next = udp_seq_next,
3171 .stop = bpf_iter_udp_seq_stop,
3172 .show = bpf_iter_udp_seq_show,
3176 const struct seq_operations udp_seq_ops = {
3177 .start = udp_seq_start,
3178 .next = udp_seq_next,
3179 .stop = udp_seq_stop,
3180 .show = udp4_seq_show,
3182 EXPORT_SYMBOL(udp_seq_ops);
3184 static struct udp_seq_afinfo udp4_seq_afinfo = {
3186 .udp_table = &udp_table,
3189 static int __net_init udp4_proc_init_net(struct net *net)
3191 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3192 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3197 static void __net_exit udp4_proc_exit_net(struct net *net)
3199 remove_proc_entry("udp", net->proc_net);
3202 static struct pernet_operations udp4_net_ops = {
3203 .init = udp4_proc_init_net,
3204 .exit = udp4_proc_exit_net,
3207 int __init udp4_proc_init(void)
3209 return register_pernet_subsys(&udp4_net_ops);
3212 void udp4_proc_exit(void)
3214 unregister_pernet_subsys(&udp4_net_ops);
3216 #endif /* CONFIG_PROC_FS */
3218 static __initdata unsigned long uhash_entries;
3219 static int __init set_uhash_entries(char *str)
3226 ret = kstrtoul(str, 0, &uhash_entries);
3230 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3231 uhash_entries = UDP_HTABLE_SIZE_MIN;
3234 __setup("uhash_entries=", set_uhash_entries);
3236 void __init udp_table_init(struct udp_table *table, const char *name)
3240 table->hash = alloc_large_system_hash(name,
3241 2 * sizeof(struct udp_hslot),
3243 21, /* one slot per 2 MB */
3247 UDP_HTABLE_SIZE_MIN,
3250 table->hash2 = table->hash + (table->mask + 1);
3251 for (i = 0; i <= table->mask; i++) {
3252 INIT_HLIST_HEAD(&table->hash[i].head);
3253 table->hash[i].count = 0;
3254 spin_lock_init(&table->hash[i].lock);
3256 for (i = 0; i <= table->mask; i++) {
3257 INIT_HLIST_HEAD(&table->hash2[i].head);
3258 table->hash2[i].count = 0;
3259 spin_lock_init(&table->hash2[i].lock);
3263 u32 udp_flow_hashrnd(void)
3265 static u32 hashrnd __read_mostly;
3267 net_get_random_once(&hashrnd, sizeof(hashrnd));
3271 EXPORT_SYMBOL(udp_flow_hashrnd);
3273 static int __net_init udp_sysctl_init(struct net *net)
3275 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3276 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3278 #ifdef CONFIG_NET_L3_MASTER_DEV
3279 net->ipv4.sysctl_udp_l3mdev_accept = 0;
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 /* 16 spinlocks per cpu */
3358 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3359 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3362 panic("UDP: failed to alloc udp_busylocks\n");
3363 for (i = 0; i < (1U << udp_busylocks_log); i++)
3364 spin_lock_init(udp_busylocks + i);
3366 if (register_pernet_subsys(&udp_sysctl_ops))
3367 panic("UDP: failed to init sysctl parameters.\n");
3369 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3370 bpf_iter_register();