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 struct udp_table *udp_get_table_prot(struct sock *sk)
136 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
139 static int udp_lib_lport_inuse(struct net *net, __u16 num,
140 const struct udp_hslot *hslot,
141 unsigned long *bitmap,
142 struct sock *sk, unsigned int log)
145 kuid_t uid = sock_i_uid(sk);
147 sk_for_each(sk2, &hslot->head) {
148 if (net_eq(sock_net(sk2), net) &&
150 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
151 (!sk2->sk_reuse || !sk->sk_reuse) &&
152 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
153 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
154 inet_rcv_saddr_equal(sk, sk2, true)) {
155 if (sk2->sk_reuseport && sk->sk_reuseport &&
156 !rcu_access_pointer(sk->sk_reuseport_cb) &&
157 uid_eq(uid, sock_i_uid(sk2))) {
163 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
172 * Note: we still hold spinlock of primary hash chain, so no other writer
173 * can insert/delete a socket with local_port == num
175 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
176 struct udp_hslot *hslot2,
180 kuid_t uid = sock_i_uid(sk);
183 spin_lock(&hslot2->lock);
184 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
185 if (net_eq(sock_net(sk2), net) &&
187 (udp_sk(sk2)->udp_port_hash == num) &&
188 (!sk2->sk_reuse || !sk->sk_reuse) &&
189 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
190 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
191 inet_rcv_saddr_equal(sk, sk2, true)) {
192 if (sk2->sk_reuseport && sk->sk_reuseport &&
193 !rcu_access_pointer(sk->sk_reuseport_cb) &&
194 uid_eq(uid, sock_i_uid(sk2))) {
202 spin_unlock(&hslot2->lock);
206 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
208 struct net *net = sock_net(sk);
209 kuid_t uid = sock_i_uid(sk);
212 sk_for_each(sk2, &hslot->head) {
213 if (net_eq(sock_net(sk2), net) &&
215 sk2->sk_family == sk->sk_family &&
216 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
217 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
218 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
219 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
220 inet_rcv_saddr_equal(sk, sk2, false)) {
221 return reuseport_add_sock(sk, sk2,
222 inet_rcv_saddr_any(sk));
226 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
230 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
232 * @sk: socket struct in question
233 * @snum: port number to look up
234 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237 int udp_lib_get_port(struct sock *sk, unsigned short snum,
238 unsigned int hash2_nulladdr)
240 struct udp_table *udptable = udp_get_table_prot(sk);
241 struct udp_hslot *hslot, *hslot2;
242 struct net *net = sock_net(sk);
246 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
247 unsigned short first, last;
248 int low, high, remaining;
251 inet_get_local_port_range(net, &low, &high);
252 remaining = (high - low) + 1;
254 rand = get_random_u32();
255 first = reciprocal_scale(rand, remaining) + low;
257 * force rand to be an odd multiple of UDP_HTABLE_SIZE
259 rand = (rand | 1) * (udptable->mask + 1);
260 last = first + udptable->mask + 1;
262 hslot = udp_hashslot(udptable, net, first);
263 bitmap_zero(bitmap, PORTS_PER_CHAIN);
264 spin_lock_bh(&hslot->lock);
265 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
270 * Iterate on all possible values of snum for this hash.
271 * Using steps of an odd multiple of UDP_HTABLE_SIZE
272 * give us randomization and full range coverage.
275 if (low <= snum && snum <= high &&
276 !test_bit(snum >> udptable->log, bitmap) &&
277 !inet_is_local_reserved_port(net, snum))
280 } while (snum != first);
281 spin_unlock_bh(&hslot->lock);
283 } while (++first != last);
286 hslot = udp_hashslot(udptable, net, snum);
287 spin_lock_bh(&hslot->lock);
288 if (hslot->count > 10) {
290 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
292 slot2 &= udptable->mask;
293 hash2_nulladdr &= udptable->mask;
295 hslot2 = udp_hashslot2(udptable, slot2);
296 if (hslot->count < hslot2->count)
297 goto scan_primary_hash;
299 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
300 if (!exist && (hash2_nulladdr != slot2)) {
301 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
302 exist = udp_lib_lport_inuse2(net, snum, hslot2,
311 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
315 inet_sk(sk)->inet_num = snum;
316 udp_sk(sk)->udp_port_hash = snum;
317 udp_sk(sk)->udp_portaddr_hash ^= snum;
318 if (sk_unhashed(sk)) {
319 if (sk->sk_reuseport &&
320 udp_reuseport_add_sock(sk, hslot)) {
321 inet_sk(sk)->inet_num = 0;
322 udp_sk(sk)->udp_port_hash = 0;
323 udp_sk(sk)->udp_portaddr_hash ^= snum;
327 sk_add_node_rcu(sk, &hslot->head);
329 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
331 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
332 spin_lock(&hslot2->lock);
333 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
334 sk->sk_family == AF_INET6)
335 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
338 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
341 spin_unlock(&hslot2->lock);
343 sock_set_flag(sk, SOCK_RCU_FREE);
346 spin_unlock_bh(&hslot->lock);
350 EXPORT_SYMBOL(udp_lib_get_port);
352 int udp_v4_get_port(struct sock *sk, unsigned short snum)
354 unsigned int hash2_nulladdr =
355 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
356 unsigned int hash2_partial =
357 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
359 /* precompute partial secondary hash */
360 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
361 return udp_lib_get_port(sk, snum, hash2_nulladdr);
364 static int compute_score(struct sock *sk, struct net *net,
365 __be32 saddr, __be16 sport,
366 __be32 daddr, unsigned short hnum,
370 struct inet_sock *inet;
373 if (!net_eq(sock_net(sk), net) ||
374 udp_sk(sk)->udp_port_hash != hnum ||
378 if (sk->sk_rcv_saddr != daddr)
381 score = (sk->sk_family == PF_INET) ? 2 : 1;
384 if (inet->inet_daddr) {
385 if (inet->inet_daddr != saddr)
390 if (inet->inet_dport) {
391 if (inet->inet_dport != sport)
396 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
400 if (sk->sk_bound_dev_if)
403 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
408 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
409 const __u16 lport, const __be32 faddr,
412 static u32 udp_ehash_secret __read_mostly;
414 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
416 return __inet_ehashfn(laddr, lport, faddr, fport,
417 udp_ehash_secret + net_hash_mix(net));
420 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
422 __be32 saddr, __be16 sport,
423 __be32 daddr, unsigned short hnum)
425 struct sock *reuse_sk = NULL;
428 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
429 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
430 reuse_sk = reuseport_select_sock(sk, hash, skb,
431 sizeof(struct udphdr));
436 /* called with rcu_read_lock() */
437 static struct sock *udp4_lib_lookup2(struct net *net,
438 __be32 saddr, __be16 sport,
439 __be32 daddr, unsigned int hnum,
441 struct udp_hslot *hslot2,
444 struct sock *sk, *result;
449 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
450 score = compute_score(sk, net, saddr, sport,
451 daddr, hnum, dif, sdif);
452 if (score > badness) {
453 result = lookup_reuseport(net, sk, skb,
454 saddr, sport, daddr, hnum);
455 /* Fall back to scoring if group has connections */
456 if (result && !reuseport_has_conns(sk))
459 result = result ? : sk;
466 static struct sock *udp4_lookup_run_bpf(struct net *net,
467 struct udp_table *udptable,
469 __be32 saddr, __be16 sport,
470 __be32 daddr, u16 hnum, const int dif)
472 struct sock *sk, *reuse_sk;
475 if (udptable != &udp_table)
476 return NULL; /* only UDP is supported */
478 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport,
479 daddr, hnum, dif, &sk);
480 if (no_reuseport || IS_ERR_OR_NULL(sk))
483 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
489 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
490 * harder than this. -DaveM
492 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
493 __be16 sport, __be32 daddr, __be16 dport, int dif,
494 int sdif, struct udp_table *udptable, struct sk_buff *skb)
496 unsigned short hnum = ntohs(dport);
497 unsigned int hash2, slot2;
498 struct udp_hslot *hslot2;
499 struct sock *result, *sk;
501 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
502 slot2 = hash2 & udptable->mask;
503 hslot2 = &udptable->hash2[slot2];
505 /* Lookup connected or non-wildcard socket */
506 result = udp4_lib_lookup2(net, saddr, sport,
507 daddr, hnum, dif, sdif,
509 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
512 /* Lookup redirect from BPF */
513 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
514 sk = udp4_lookup_run_bpf(net, udptable, skb,
515 saddr, sport, daddr, hnum, dif);
522 /* Got non-wildcard socket or error on first lookup */
526 /* Lookup wildcard sockets */
527 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
528 slot2 = hash2 & udptable->mask;
529 hslot2 = &udptable->hash2[slot2];
531 result = udp4_lib_lookup2(net, saddr, sport,
532 htonl(INADDR_ANY), hnum, dif, sdif,
539 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
541 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
542 __be16 sport, __be16 dport,
543 struct udp_table *udptable)
545 const struct iphdr *iph = ip_hdr(skb);
547 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
548 iph->daddr, dport, inet_iif(skb),
549 inet_sdif(skb), udptable, skb);
552 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
553 __be16 sport, __be16 dport)
555 const struct iphdr *iph = ip_hdr(skb);
557 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
558 iph->daddr, dport, inet_iif(skb),
559 inet_sdif(skb), &udp_table, NULL);
562 /* Must be called under rcu_read_lock().
563 * Does increment socket refcount.
565 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
566 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
567 __be32 daddr, __be16 dport, int dif)
571 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
572 dif, 0, &udp_table, NULL);
573 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
577 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
580 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
581 __be16 loc_port, __be32 loc_addr,
582 __be16 rmt_port, __be32 rmt_addr,
583 int dif, int sdif, unsigned short hnum)
585 struct inet_sock *inet = inet_sk(sk);
587 if (!net_eq(sock_net(sk), net) ||
588 udp_sk(sk)->udp_port_hash != hnum ||
589 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
590 (inet->inet_dport != rmt_port && inet->inet_dport) ||
591 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
592 ipv6_only_sock(sk) ||
593 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
595 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
600 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
601 void udp_encap_enable(void)
603 static_branch_inc(&udp_encap_needed_key);
605 EXPORT_SYMBOL(udp_encap_enable);
607 void udp_encap_disable(void)
609 static_branch_dec(&udp_encap_needed_key);
611 EXPORT_SYMBOL(udp_encap_disable);
613 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
614 * through error handlers in encapsulations looking for a match.
616 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
620 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
621 int (*handler)(struct sk_buff *skb, u32 info);
622 const struct ip_tunnel_encap_ops *encap;
624 encap = rcu_dereference(iptun_encaps[i]);
627 handler = encap->err_handler;
628 if (handler && !handler(skb, info))
635 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
636 * reversing source and destination port: this will match tunnels that force the
637 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
638 * lwtunnels might actually break this assumption by being configured with
639 * different destination ports on endpoints, in this case we won't be able to
640 * trace ICMP messages back to them.
642 * If this doesn't match any socket, probe tunnels with arbitrary destination
643 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
644 * we've sent packets to won't necessarily match the local destination port.
646 * Then ask the tunnel implementation to match the error against a valid
649 * Return an error if we can't find a match, the socket if we need further
650 * processing, zero otherwise.
652 static struct sock *__udp4_lib_err_encap(struct net *net,
653 const struct iphdr *iph,
655 struct udp_table *udptable,
657 struct sk_buff *skb, u32 info)
659 int (*lookup)(struct sock *sk, struct sk_buff *skb);
660 int network_offset, transport_offset;
663 network_offset = skb_network_offset(skb);
664 transport_offset = skb_transport_offset(skb);
666 /* Network header needs to point to the outer IPv4 header inside ICMP */
667 skb_reset_network_header(skb);
669 /* Transport header needs to point to the UDP header */
670 skb_set_transport_header(skb, iph->ihl << 2);
675 lookup = READ_ONCE(up->encap_err_lookup);
676 if (lookup && lookup(sk, skb))
682 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
683 iph->saddr, uh->dest, skb->dev->ifindex, 0,
688 lookup = READ_ONCE(up->encap_err_lookup);
689 if (!lookup || lookup(sk, skb))
695 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
697 skb_set_transport_header(skb, transport_offset);
698 skb_set_network_header(skb, network_offset);
704 * This routine is called by the ICMP module when it gets some
705 * sort of error condition. If err < 0 then the socket should
706 * be closed and the error returned to the user. If err > 0
707 * it's just the icmp type << 8 | icmp code.
708 * Header points to the ip header of the error packet. We move
709 * on past this. Then (as it used to claim before adjustment)
710 * header points to the first 8 bytes of the udp header. We need
711 * to find the appropriate port.
714 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
716 struct inet_sock *inet;
717 const struct iphdr *iph = (const struct iphdr *)skb->data;
718 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
719 const int type = icmp_hdr(skb)->type;
720 const int code = icmp_hdr(skb)->code;
725 struct net *net = dev_net(skb->dev);
727 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
728 iph->saddr, uh->source, skb->dev->ifindex,
729 inet_sdif(skb), udptable, NULL);
731 if (!sk || udp_sk(sk)->encap_type) {
732 /* No socket for error: try tunnels before discarding */
733 if (static_branch_unlikely(&udp_encap_needed_key)) {
734 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
739 sk = ERR_PTR(-ENOENT);
742 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
755 case ICMP_TIME_EXCEEDED:
758 case ICMP_SOURCE_QUENCH:
760 case ICMP_PARAMETERPROB:
764 case ICMP_DEST_UNREACH:
765 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
766 ipv4_sk_update_pmtu(skb, sk, info);
767 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
775 if (code <= NR_ICMP_UNREACH) {
776 harderr = icmp_err_convert[code].fatal;
777 err = icmp_err_convert[code].errno;
781 ipv4_sk_redirect(skb, sk);
786 * RFC1122: OK. Passes ICMP errors back to application, as per
790 /* ...not for tunnels though: we don't have a sending socket */
791 if (udp_sk(sk)->encap_err_rcv)
792 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
796 if (!inet->recverr) {
797 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
800 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
808 int udp_err(struct sk_buff *skb, u32 info)
810 return __udp4_lib_err(skb, info, &udp_table);
814 * Throw away all pending data and cancel the corking. Socket is locked.
816 void udp_flush_pending_frames(struct sock *sk)
818 struct udp_sock *up = udp_sk(sk);
823 ip_flush_pending_frames(sk);
826 EXPORT_SYMBOL(udp_flush_pending_frames);
829 * udp4_hwcsum - handle outgoing HW checksumming
830 * @skb: sk_buff containing the filled-in UDP header
831 * (checksum field must be zeroed out)
832 * @src: source IP address
833 * @dst: destination IP address
835 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
837 struct udphdr *uh = udp_hdr(skb);
838 int offset = skb_transport_offset(skb);
839 int len = skb->len - offset;
843 if (!skb_has_frag_list(skb)) {
845 * Only one fragment on the socket.
847 skb->csum_start = skb_transport_header(skb) - skb->head;
848 skb->csum_offset = offsetof(struct udphdr, check);
849 uh->check = ~csum_tcpudp_magic(src, dst, len,
852 struct sk_buff *frags;
855 * HW-checksum won't work as there are two or more
856 * fragments on the socket so that all csums of sk_buffs
859 skb_walk_frags(skb, frags) {
860 csum = csum_add(csum, frags->csum);
864 csum = skb_checksum(skb, offset, hlen, csum);
865 skb->ip_summed = CHECKSUM_NONE;
867 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
869 uh->check = CSUM_MANGLED_0;
872 EXPORT_SYMBOL_GPL(udp4_hwcsum);
874 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
875 * for the simple case like when setting the checksum for a UDP tunnel.
877 void udp_set_csum(bool nocheck, struct sk_buff *skb,
878 __be32 saddr, __be32 daddr, int len)
880 struct udphdr *uh = udp_hdr(skb);
884 } else if (skb_is_gso(skb)) {
885 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
886 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
888 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
890 uh->check = CSUM_MANGLED_0;
892 skb->ip_summed = CHECKSUM_PARTIAL;
893 skb->csum_start = skb_transport_header(skb) - skb->head;
894 skb->csum_offset = offsetof(struct udphdr, check);
895 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
898 EXPORT_SYMBOL(udp_set_csum);
900 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
901 struct inet_cork *cork)
903 struct sock *sk = skb->sk;
904 struct inet_sock *inet = inet_sk(sk);
907 int is_udplite = IS_UDPLITE(sk);
908 int offset = skb_transport_offset(skb);
909 int len = skb->len - offset;
910 int datalen = len - sizeof(*uh);
914 * Create a UDP header
917 uh->source = inet->inet_sport;
918 uh->dest = fl4->fl4_dport;
919 uh->len = htons(len);
922 if (cork->gso_size) {
923 const int hlen = skb_network_header_len(skb) +
924 sizeof(struct udphdr);
926 if (hlen + cork->gso_size > cork->fragsize) {
930 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
934 if (sk->sk_no_check_tx) {
938 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
939 dst_xfrm(skb_dst(skb))) {
944 if (datalen > cork->gso_size) {
945 skb_shinfo(skb)->gso_size = cork->gso_size;
946 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
947 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
953 if (is_udplite) /* UDP-Lite */
954 csum = udplite_csum(skb);
956 else if (sk->sk_no_check_tx) { /* UDP csum off */
958 skb->ip_summed = CHECKSUM_NONE;
961 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
964 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
968 csum = udp_csum(skb);
970 /* add protocol-dependent pseudo-header */
971 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
972 sk->sk_protocol, csum);
974 uh->check = CSUM_MANGLED_0;
977 err = ip_send_skb(sock_net(sk), skb);
979 if (err == -ENOBUFS && !inet->recverr) {
980 UDP_INC_STATS(sock_net(sk),
981 UDP_MIB_SNDBUFERRORS, is_udplite);
985 UDP_INC_STATS(sock_net(sk),
986 UDP_MIB_OUTDATAGRAMS, is_udplite);
991 * Push out all pending data as one UDP datagram. Socket is locked.
993 int udp_push_pending_frames(struct sock *sk)
995 struct udp_sock *up = udp_sk(sk);
996 struct inet_sock *inet = inet_sk(sk);
997 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1001 skb = ip_finish_skb(sk, fl4);
1005 err = udp_send_skb(skb, fl4, &inet->cork.base);
1012 EXPORT_SYMBOL(udp_push_pending_frames);
1014 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1016 switch (cmsg->cmsg_type) {
1018 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1020 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1027 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1029 struct cmsghdr *cmsg;
1030 bool need_ip = false;
1033 for_each_cmsghdr(cmsg, msg) {
1034 if (!CMSG_OK(msg, cmsg))
1037 if (cmsg->cmsg_level != SOL_UDP) {
1042 err = __udp_cmsg_send(cmsg, gso_size);
1049 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1051 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1053 struct inet_sock *inet = inet_sk(sk);
1054 struct udp_sock *up = udp_sk(sk);
1055 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1056 struct flowi4 fl4_stack;
1059 struct ipcm_cookie ipc;
1060 struct rtable *rt = NULL;
1063 __be32 daddr, faddr, saddr;
1066 int err, is_udplite = IS_UDPLITE(sk);
1067 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1068 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1069 struct sk_buff *skb;
1070 struct ip_options_data opt_copy;
1079 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1082 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1084 fl4 = &inet->cork.fl.u.ip4;
1087 * There are pending frames.
1088 * The socket lock must be held while it's corked.
1091 if (likely(up->pending)) {
1092 if (unlikely(up->pending != AF_INET)) {
1096 goto do_append_data;
1100 ulen += sizeof(struct udphdr);
1103 * Get and verify the address.
1106 if (msg->msg_namelen < sizeof(*usin))
1108 if (usin->sin_family != AF_INET) {
1109 if (usin->sin_family != AF_UNSPEC)
1110 return -EAFNOSUPPORT;
1113 daddr = usin->sin_addr.s_addr;
1114 dport = usin->sin_port;
1118 if (sk->sk_state != TCP_ESTABLISHED)
1119 return -EDESTADDRREQ;
1120 daddr = inet->inet_daddr;
1121 dport = inet->inet_dport;
1122 /* Open fast path for connected socket.
1123 Route will not be used, if at least one option is set.
1128 ipcm_init_sk(&ipc, inet);
1129 ipc.gso_size = READ_ONCE(up->gso_size);
1131 if (msg->msg_controllen) {
1132 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1134 err = ip_cmsg_send(sk, msg, &ipc,
1135 sk->sk_family == AF_INET6);
1136 if (unlikely(err < 0)) {
1145 struct ip_options_rcu *inet_opt;
1148 inet_opt = rcu_dereference(inet->inet_opt);
1150 memcpy(&opt_copy, inet_opt,
1151 sizeof(*inet_opt) + inet_opt->opt.optlen);
1152 ipc.opt = &opt_copy.opt;
1157 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1158 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1159 (struct sockaddr *)usin, &ipc.addr);
1163 if (usin->sin_port == 0) {
1164 /* BPF program set invalid port. Reject it. */
1168 daddr = usin->sin_addr.s_addr;
1169 dport = usin->sin_port;
1174 ipc.addr = faddr = daddr;
1176 if (ipc.opt && ipc.opt->opt.srr) {
1181 faddr = ipc.opt->opt.faddr;
1184 tos = get_rttos(&ipc, inet);
1185 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1186 (msg->msg_flags & MSG_DONTROUTE) ||
1187 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1192 if (ipv4_is_multicast(daddr)) {
1193 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1194 ipc.oif = inet->mc_index;
1196 saddr = inet->mc_addr;
1198 } else if (!ipc.oif) {
1199 ipc.oif = inet->uc_index;
1200 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1201 /* oif is set, packet is to local broadcast and
1202 * uc_index is set. oif is most likely set
1203 * by sk_bound_dev_if. If uc_index != oif check if the
1204 * oif is an L3 master and uc_index is an L3 slave.
1205 * If so, we want to allow the send using the uc_index.
1207 if (ipc.oif != inet->uc_index &&
1208 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1210 ipc.oif = inet->uc_index;
1215 rt = (struct rtable *)sk_dst_check(sk, 0);
1218 struct net *net = sock_net(sk);
1219 __u8 flow_flags = inet_sk_flowi_flags(sk);
1223 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1224 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1226 faddr, saddr, dport, inet->inet_sport,
1229 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1230 rt = ip_route_output_flow(net, fl4, sk);
1234 if (err == -ENETUNREACH)
1235 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1240 if ((rt->rt_flags & RTCF_BROADCAST) &&
1241 !sock_flag(sk, SOCK_BROADCAST))
1244 sk_dst_set(sk, dst_clone(&rt->dst));
1247 if (msg->msg_flags&MSG_CONFIRM)
1253 daddr = ipc.addr = fl4->daddr;
1255 /* Lockless fast path for the non-corking case. */
1257 struct inet_cork cork;
1259 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1260 sizeof(struct udphdr), &ipc, &rt,
1261 &cork, msg->msg_flags);
1263 if (!IS_ERR_OR_NULL(skb))
1264 err = udp_send_skb(skb, fl4, &cork);
1269 if (unlikely(up->pending)) {
1270 /* The socket is already corked while preparing it. */
1271 /* ... which is an evident application bug. --ANK */
1274 net_dbg_ratelimited("socket already corked\n");
1279 * Now cork the socket to pend data.
1281 fl4 = &inet->cork.fl.u.ip4;
1284 fl4->fl4_dport = dport;
1285 fl4->fl4_sport = inet->inet_sport;
1286 up->pending = AF_INET;
1290 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1291 sizeof(struct udphdr), &ipc, &rt,
1292 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1294 udp_flush_pending_frames(sk);
1296 err = udp_push_pending_frames(sk);
1297 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1309 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1310 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1311 * we don't have a good statistic (IpOutDiscards but it can be too many
1312 * things). We could add another new stat but at least for now that
1313 * seems like overkill.
1315 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1316 UDP_INC_STATS(sock_net(sk),
1317 UDP_MIB_SNDBUFERRORS, is_udplite);
1322 if (msg->msg_flags & MSG_PROBE)
1323 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1324 if (!(msg->msg_flags&MSG_PROBE) || len)
1325 goto back_from_confirm;
1329 EXPORT_SYMBOL(udp_sendmsg);
1331 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1332 size_t size, int flags)
1334 struct inet_sock *inet = inet_sk(sk);
1335 struct udp_sock *up = udp_sk(sk);
1338 if (flags & MSG_SENDPAGE_NOTLAST)
1342 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1344 /* Call udp_sendmsg to specify destination address which
1345 * sendpage interface can't pass.
1346 * This will succeed only when the socket is connected.
1348 ret = udp_sendmsg(sk, &msg, 0);
1355 if (unlikely(!up->pending)) {
1358 net_dbg_ratelimited("cork failed\n");
1362 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1363 page, offset, size, flags);
1364 if (ret == -EOPNOTSUPP) {
1366 return sock_no_sendpage(sk->sk_socket, page, offset,
1370 udp_flush_pending_frames(sk);
1375 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1376 ret = udp_push_pending_frames(sk);
1384 #define UDP_SKB_IS_STATELESS 0x80000000
1386 /* all head states (dst, sk, nf conntrack) except skb extensions are
1387 * cleared by udp_rcv().
1389 * We need to preserve secpath, if present, to eventually process
1390 * IP_CMSG_PASSSEC at recvmsg() time.
1392 * Other extensions can be cleared.
1394 static bool udp_try_make_stateless(struct sk_buff *skb)
1396 if (!skb_has_extensions(skb))
1399 if (!secpath_exists(skb)) {
1407 static void udp_set_dev_scratch(struct sk_buff *skb)
1409 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1411 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1412 scratch->_tsize_state = skb->truesize;
1413 #if BITS_PER_LONG == 64
1414 scratch->len = skb->len;
1415 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1416 scratch->is_linear = !skb_is_nonlinear(skb);
1418 if (udp_try_make_stateless(skb))
1419 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1422 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1424 /* We come here after udp_lib_checksum_complete() returned 0.
1425 * This means that __skb_checksum_complete() might have
1426 * set skb->csum_valid to 1.
1427 * On 64bit platforms, we can set csum_unnecessary
1428 * to true, but only if the skb is not shared.
1430 #if BITS_PER_LONG == 64
1431 if (!skb_shared(skb))
1432 udp_skb_scratch(skb)->csum_unnecessary = true;
1436 static int udp_skb_truesize(struct sk_buff *skb)
1438 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1441 static bool udp_skb_has_head_state(struct sk_buff *skb)
1443 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1446 /* fully reclaim rmem/fwd memory allocated for skb */
1447 static void udp_rmem_release(struct sock *sk, int size, int partial,
1448 bool rx_queue_lock_held)
1450 struct udp_sock *up = udp_sk(sk);
1451 struct sk_buff_head *sk_queue;
1454 if (likely(partial)) {
1455 up->forward_deficit += size;
1456 size = up->forward_deficit;
1457 if (size < READ_ONCE(up->forward_threshold) &&
1458 !skb_queue_empty(&up->reader_queue))
1461 size += up->forward_deficit;
1463 up->forward_deficit = 0;
1465 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1466 * if the called don't held it already
1468 sk_queue = &sk->sk_receive_queue;
1469 if (!rx_queue_lock_held)
1470 spin_lock(&sk_queue->lock);
1473 sk->sk_forward_alloc += size;
1474 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1475 sk->sk_forward_alloc -= amt;
1478 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1480 atomic_sub(size, &sk->sk_rmem_alloc);
1482 /* this can save us from acquiring the rx queue lock on next receive */
1483 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1485 if (!rx_queue_lock_held)
1486 spin_unlock(&sk_queue->lock);
1489 /* Note: called with reader_queue.lock held.
1490 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1491 * This avoids a cache line miss while receive_queue lock is held.
1492 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1494 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1496 prefetch(&skb->data);
1497 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1499 EXPORT_SYMBOL(udp_skb_destructor);
1501 /* as above, but the caller held the rx queue lock, too */
1502 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1504 prefetch(&skb->data);
1505 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1508 /* Idea of busylocks is to let producers grab an extra spinlock
1509 * to relieve pressure on the receive_queue spinlock shared by consumer.
1510 * Under flood, this means that only one producer can be in line
1511 * trying to acquire the receive_queue spinlock.
1512 * These busylock can be allocated on a per cpu manner, instead of a
1513 * per socket one (that would consume a cache line per socket)
1515 static int udp_busylocks_log __read_mostly;
1516 static spinlock_t *udp_busylocks __read_mostly;
1518 static spinlock_t *busylock_acquire(void *ptr)
1522 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1527 static void busylock_release(spinlock_t *busy)
1533 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1535 struct sk_buff_head *list = &sk->sk_receive_queue;
1536 int rmem, delta, amt, err = -ENOMEM;
1537 spinlock_t *busy = NULL;
1540 /* try to avoid the costly atomic add/sub pair when the receive
1541 * queue is full; always allow at least a packet
1543 rmem = atomic_read(&sk->sk_rmem_alloc);
1544 if (rmem > sk->sk_rcvbuf)
1547 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1548 * having linear skbs :
1549 * - Reduce memory overhead and thus increase receive queue capacity
1550 * - Less cache line misses at copyout() time
1551 * - Less work at consume_skb() (less alien page frag freeing)
1553 if (rmem > (sk->sk_rcvbuf >> 1)) {
1556 busy = busylock_acquire(sk);
1558 size = skb->truesize;
1559 udp_set_dev_scratch(skb);
1561 /* we drop only if the receive buf is full and the receive
1562 * queue contains some other skb
1564 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1565 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1568 spin_lock(&list->lock);
1569 if (size >= sk->sk_forward_alloc) {
1570 amt = sk_mem_pages(size);
1571 delta = amt << PAGE_SHIFT;
1572 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1574 spin_unlock(&list->lock);
1578 sk->sk_forward_alloc += delta;
1581 sk->sk_forward_alloc -= size;
1583 /* no need to setup a destructor, we will explicitly release the
1584 * forward allocated memory on dequeue
1586 sock_skb_set_dropcount(sk, skb);
1588 __skb_queue_tail(list, skb);
1589 spin_unlock(&list->lock);
1591 if (!sock_flag(sk, SOCK_DEAD))
1592 sk->sk_data_ready(sk);
1594 busylock_release(busy);
1598 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1601 atomic_inc(&sk->sk_drops);
1602 busylock_release(busy);
1605 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1607 void udp_destruct_common(struct sock *sk)
1609 /* reclaim completely the forward allocated memory */
1610 struct udp_sock *up = udp_sk(sk);
1611 unsigned int total = 0;
1612 struct sk_buff *skb;
1614 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1615 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1616 total += skb->truesize;
1619 udp_rmem_release(sk, total, 0, true);
1621 EXPORT_SYMBOL_GPL(udp_destruct_common);
1623 static void udp_destruct_sock(struct sock *sk)
1625 udp_destruct_common(sk);
1626 inet_sock_destruct(sk);
1629 int udp_init_sock(struct sock *sk)
1631 udp_lib_init_sock(sk);
1632 sk->sk_destruct = udp_destruct_sock;
1633 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1637 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1639 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1640 bool slow = lock_sock_fast(sk);
1642 sk_peek_offset_bwd(sk, len);
1643 unlock_sock_fast(sk, slow);
1646 if (!skb_unref(skb))
1649 /* In the more common cases we cleared the head states previously,
1650 * see __udp_queue_rcv_skb().
1652 if (unlikely(udp_skb_has_head_state(skb)))
1653 skb_release_head_state(skb);
1654 __consume_stateless_skb(skb);
1656 EXPORT_SYMBOL_GPL(skb_consume_udp);
1658 static struct sk_buff *__first_packet_length(struct sock *sk,
1659 struct sk_buff_head *rcvq,
1662 struct sk_buff *skb;
1664 while ((skb = skb_peek(rcvq)) != NULL) {
1665 if (udp_lib_checksum_complete(skb)) {
1666 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1668 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1670 atomic_inc(&sk->sk_drops);
1671 __skb_unlink(skb, rcvq);
1672 *total += skb->truesize;
1675 udp_skb_csum_unnecessary_set(skb);
1683 * first_packet_length - return length of first packet in receive queue
1686 * Drops all bad checksum frames, until a valid one is found.
1687 * Returns the length of found skb, or -1 if none is found.
1689 static int first_packet_length(struct sock *sk)
1691 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1692 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1693 struct sk_buff *skb;
1697 spin_lock_bh(&rcvq->lock);
1698 skb = __first_packet_length(sk, rcvq, &total);
1699 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1700 spin_lock(&sk_queue->lock);
1701 skb_queue_splice_tail_init(sk_queue, rcvq);
1702 spin_unlock(&sk_queue->lock);
1704 skb = __first_packet_length(sk, rcvq, &total);
1706 res = skb ? skb->len : -1;
1708 udp_rmem_release(sk, total, 1, false);
1709 spin_unlock_bh(&rcvq->lock);
1714 * IOCTL requests applicable to the UDP protocol
1717 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1722 int amount = sk_wmem_alloc_get(sk);
1724 return put_user(amount, (int __user *)arg);
1729 int amount = max_t(int, 0, first_packet_length(sk));
1731 return put_user(amount, (int __user *)arg);
1735 return -ENOIOCTLCMD;
1740 EXPORT_SYMBOL(udp_ioctl);
1742 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1745 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1746 struct sk_buff_head *queue;
1747 struct sk_buff *last;
1751 queue = &udp_sk(sk)->reader_queue;
1752 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1754 struct sk_buff *skb;
1756 error = sock_error(sk);
1762 spin_lock_bh(&queue->lock);
1763 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1766 if (!(flags & MSG_PEEK))
1767 udp_skb_destructor(sk, skb);
1768 spin_unlock_bh(&queue->lock);
1772 if (skb_queue_empty_lockless(sk_queue)) {
1773 spin_unlock_bh(&queue->lock);
1777 /* refill the reader queue and walk it again
1778 * keep both queues locked to avoid re-acquiring
1779 * the sk_receive_queue lock if fwd memory scheduling
1782 spin_lock(&sk_queue->lock);
1783 skb_queue_splice_tail_init(sk_queue, queue);
1785 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1787 if (skb && !(flags & MSG_PEEK))
1788 udp_skb_dtor_locked(sk, skb);
1789 spin_unlock(&sk_queue->lock);
1790 spin_unlock_bh(&queue->lock);
1795 if (!sk_can_busy_loop(sk))
1798 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1799 } while (!skb_queue_empty_lockless(sk_queue));
1801 /* sk_queue is empty, reader_queue may contain peeked packets */
1803 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1805 (struct sk_buff *)sk_queue));
1810 EXPORT_SYMBOL(__skb_recv_udp);
1812 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1814 struct sk_buff *skb;
1818 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1822 if (udp_lib_checksum_complete(skb)) {
1823 int is_udplite = IS_UDPLITE(sk);
1824 struct net *net = sock_net(sk);
1826 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1827 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1828 atomic_inc(&sk->sk_drops);
1833 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1834 copied = recv_actor(sk, skb);
1839 EXPORT_SYMBOL(udp_read_skb);
1842 * This should be easy, if there is something there we
1843 * return it, otherwise we block.
1846 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1849 struct inet_sock *inet = inet_sk(sk);
1850 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1851 struct sk_buff *skb;
1852 unsigned int ulen, copied;
1853 int off, err, peeking = flags & MSG_PEEK;
1854 int is_udplite = IS_UDPLITE(sk);
1855 bool checksum_valid = false;
1857 if (flags & MSG_ERRQUEUE)
1858 return ip_recv_error(sk, msg, len, addr_len);
1861 off = sk_peek_offset(sk, flags);
1862 skb = __skb_recv_udp(sk, flags, &off, &err);
1866 ulen = udp_skb_len(skb);
1868 if (copied > ulen - off)
1869 copied = ulen - off;
1870 else if (copied < ulen)
1871 msg->msg_flags |= MSG_TRUNC;
1874 * If checksum is needed at all, try to do it while copying the
1875 * data. If the data is truncated, or if we only want a partial
1876 * coverage checksum (UDP-Lite), do it before the copy.
1879 if (copied < ulen || peeking ||
1880 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1881 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1882 !__udp_lib_checksum_complete(skb);
1883 if (!checksum_valid)
1887 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1888 if (udp_skb_is_linear(skb))
1889 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1891 err = skb_copy_datagram_msg(skb, off, msg, copied);
1893 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1899 if (unlikely(err)) {
1901 atomic_inc(&sk->sk_drops);
1902 UDP_INC_STATS(sock_net(sk),
1903 UDP_MIB_INERRORS, is_udplite);
1910 UDP_INC_STATS(sock_net(sk),
1911 UDP_MIB_INDATAGRAMS, is_udplite);
1913 sock_recv_cmsgs(msg, sk, skb);
1915 /* Copy the address. */
1917 sin->sin_family = AF_INET;
1918 sin->sin_port = udp_hdr(skb)->source;
1919 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1920 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1921 *addr_len = sizeof(*sin);
1923 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1924 (struct sockaddr *)sin);
1927 if (udp_sk(sk)->gro_enabled)
1928 udp_cmsg_recv(msg, sk, skb);
1930 if (inet->cmsg_flags)
1931 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1934 if (flags & MSG_TRUNC)
1937 skb_consume_udp(sk, skb, peeking ? -err : err);
1941 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1942 udp_skb_destructor)) {
1943 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1944 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1948 /* starting over for a new packet, but check if we need to yield */
1950 msg->msg_flags &= ~MSG_TRUNC;
1954 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1956 /* This check is replicated from __ip4_datagram_connect() and
1957 * intended to prevent BPF program called below from accessing bytes
1958 * that are out of the bound specified by user in addr_len.
1960 if (addr_len < sizeof(struct sockaddr_in))
1963 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1965 EXPORT_SYMBOL(udp_pre_connect);
1967 int __udp_disconnect(struct sock *sk, int flags)
1969 struct inet_sock *inet = inet_sk(sk);
1971 * 1003.1g - break association.
1974 sk->sk_state = TCP_CLOSE;
1975 inet->inet_daddr = 0;
1976 inet->inet_dport = 0;
1977 sock_rps_reset_rxhash(sk);
1978 sk->sk_bound_dev_if = 0;
1979 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1980 inet_reset_saddr(sk);
1981 if (sk->sk_prot->rehash &&
1982 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1983 sk->sk_prot->rehash(sk);
1986 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1987 sk->sk_prot->unhash(sk);
1988 inet->inet_sport = 0;
1993 EXPORT_SYMBOL(__udp_disconnect);
1995 int udp_disconnect(struct sock *sk, int flags)
1998 __udp_disconnect(sk, flags);
2002 EXPORT_SYMBOL(udp_disconnect);
2004 void udp_lib_unhash(struct sock *sk)
2006 if (sk_hashed(sk)) {
2007 struct udp_table *udptable = udp_get_table_prot(sk);
2008 struct udp_hslot *hslot, *hslot2;
2010 hslot = udp_hashslot(udptable, sock_net(sk),
2011 udp_sk(sk)->udp_port_hash);
2012 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2014 spin_lock_bh(&hslot->lock);
2015 if (rcu_access_pointer(sk->sk_reuseport_cb))
2016 reuseport_detach_sock(sk);
2017 if (sk_del_node_init_rcu(sk)) {
2019 inet_sk(sk)->inet_num = 0;
2020 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2022 spin_lock(&hslot2->lock);
2023 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2025 spin_unlock(&hslot2->lock);
2027 spin_unlock_bh(&hslot->lock);
2030 EXPORT_SYMBOL(udp_lib_unhash);
2033 * inet_rcv_saddr was changed, we must rehash secondary hash
2035 void udp_lib_rehash(struct sock *sk, u16 newhash)
2037 if (sk_hashed(sk)) {
2038 struct udp_table *udptable = udp_get_table_prot(sk);
2039 struct udp_hslot *hslot, *hslot2, *nhslot2;
2041 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2042 nhslot2 = udp_hashslot2(udptable, newhash);
2043 udp_sk(sk)->udp_portaddr_hash = newhash;
2045 if (hslot2 != nhslot2 ||
2046 rcu_access_pointer(sk->sk_reuseport_cb)) {
2047 hslot = udp_hashslot(udptable, sock_net(sk),
2048 udp_sk(sk)->udp_port_hash);
2049 /* we must lock primary chain too */
2050 spin_lock_bh(&hslot->lock);
2051 if (rcu_access_pointer(sk->sk_reuseport_cb))
2052 reuseport_detach_sock(sk);
2054 if (hslot2 != nhslot2) {
2055 spin_lock(&hslot2->lock);
2056 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2058 spin_unlock(&hslot2->lock);
2060 spin_lock(&nhslot2->lock);
2061 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2064 spin_unlock(&nhslot2->lock);
2067 spin_unlock_bh(&hslot->lock);
2071 EXPORT_SYMBOL(udp_lib_rehash);
2073 void udp_v4_rehash(struct sock *sk)
2075 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2076 inet_sk(sk)->inet_rcv_saddr,
2077 inet_sk(sk)->inet_num);
2078 udp_lib_rehash(sk, new_hash);
2081 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2085 if (inet_sk(sk)->inet_daddr) {
2086 sock_rps_save_rxhash(sk, skb);
2087 sk_mark_napi_id(sk, skb);
2088 sk_incoming_cpu_update(sk);
2090 sk_mark_napi_id_once(sk, skb);
2093 rc = __udp_enqueue_schedule_skb(sk, skb);
2095 int is_udplite = IS_UDPLITE(sk);
2098 /* Note that an ENOMEM error is charged twice */
2099 if (rc == -ENOMEM) {
2100 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2102 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2104 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2106 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2108 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2109 kfree_skb_reason(skb, drop_reason);
2110 trace_udp_fail_queue_rcv_skb(rc, sk);
2120 * >0: "udp encap" protocol resubmission
2122 * Note that in the success and error cases, the skb is assumed to
2123 * have either been requeued or freed.
2125 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2127 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2128 struct udp_sock *up = udp_sk(sk);
2129 int is_udplite = IS_UDPLITE(sk);
2132 * Charge it to the socket, dropping if the queue is full.
2134 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2135 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2140 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2141 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2144 * This is an encapsulation socket so pass the skb to
2145 * the socket's udp_encap_rcv() hook. Otherwise, just
2146 * fall through and pass this up the UDP socket.
2147 * up->encap_rcv() returns the following value:
2148 * =0 if skb was successfully passed to the encap
2149 * handler or was discarded by it.
2150 * >0 if skb should be passed on to UDP.
2151 * <0 if skb should be resubmitted as proto -N
2154 /* if we're overly short, let UDP handle it */
2155 encap_rcv = READ_ONCE(up->encap_rcv);
2159 /* Verify checksum before giving to encap */
2160 if (udp_lib_checksum_complete(skb))
2163 ret = encap_rcv(sk, skb);
2165 __UDP_INC_STATS(sock_net(sk),
2166 UDP_MIB_INDATAGRAMS,
2172 /* FALLTHROUGH -- it's a UDP Packet */
2176 * UDP-Lite specific tests, ignored on UDP sockets
2178 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2181 * MIB statistics other than incrementing the error count are
2182 * disabled for the following two types of errors: these depend
2183 * on the application settings, not on the functioning of the
2184 * protocol stack as such.
2186 * RFC 3828 here recommends (sec 3.3): "There should also be a
2187 * way ... to ... at least let the receiving application block
2188 * delivery of packets with coverage values less than a value
2189 * provided by the application."
2191 if (up->pcrlen == 0) { /* full coverage was set */
2192 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2193 UDP_SKB_CB(skb)->cscov, skb->len);
2196 /* The next case involves violating the min. coverage requested
2197 * by the receiver. This is subtle: if receiver wants x and x is
2198 * greater than the buffersize/MTU then receiver will complain
2199 * that it wants x while sender emits packets of smaller size y.
2200 * Therefore the above ...()->partial_cov statement is essential.
2202 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2203 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2204 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2209 prefetch(&sk->sk_rmem_alloc);
2210 if (rcu_access_pointer(sk->sk_filter) &&
2211 udp_lib_checksum_complete(skb))
2214 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2215 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2219 udp_csum_pull_header(skb);
2221 ipv4_pktinfo_prepare(sk, skb);
2222 return __udp_queue_rcv_skb(sk, skb);
2225 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2226 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2228 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2229 atomic_inc(&sk->sk_drops);
2230 kfree_skb_reason(skb, drop_reason);
2234 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2236 struct sk_buff *next, *segs;
2239 if (likely(!udp_unexpected_gso(sk, skb)))
2240 return udp_queue_rcv_one_skb(sk, skb);
2242 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2243 __skb_push(skb, -skb_mac_offset(skb));
2244 segs = udp_rcv_segment(sk, skb, true);
2245 skb_list_walk_safe(segs, skb, next) {
2246 __skb_pull(skb, skb_transport_offset(skb));
2248 udp_post_segment_fix_csum(skb);
2249 ret = udp_queue_rcv_one_skb(sk, skb);
2251 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2256 /* For TCP sockets, sk_rx_dst is protected by socket lock
2257 * For UDP, we use xchg() to guard against concurrent changes.
2259 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2261 struct dst_entry *old;
2263 if (dst_hold_safe(dst)) {
2264 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2270 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2273 * Multicasts and broadcasts go to each listener.
2275 * Note: called only from the BH handler context.
2277 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2279 __be32 saddr, __be32 daddr,
2280 struct udp_table *udptable,
2283 struct sock *sk, *first = NULL;
2284 unsigned short hnum = ntohs(uh->dest);
2285 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2286 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2287 unsigned int offset = offsetof(typeof(*sk), sk_node);
2288 int dif = skb->dev->ifindex;
2289 int sdif = inet_sdif(skb);
2290 struct hlist_node *node;
2291 struct sk_buff *nskb;
2294 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2296 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2298 hslot = &udptable->hash2[hash2];
2299 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2302 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2303 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2304 uh->source, saddr, dif, sdif, hnum))
2311 nskb = skb_clone(skb, GFP_ATOMIC);
2313 if (unlikely(!nskb)) {
2314 atomic_inc(&sk->sk_drops);
2315 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2317 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2321 if (udp_queue_rcv_skb(sk, nskb) > 0)
2325 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2326 if (use_hash2 && hash2 != hash2_any) {
2332 if (udp_queue_rcv_skb(first, skb) > 0)
2336 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2337 proto == IPPROTO_UDPLITE);
2342 /* Initialize UDP checksum. If exited with zero value (success),
2343 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2344 * Otherwise, csum completion requires checksumming packet body,
2345 * including udp header and folding it to skb->csum.
2347 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2352 UDP_SKB_CB(skb)->partial_cov = 0;
2353 UDP_SKB_CB(skb)->cscov = skb->len;
2355 if (proto == IPPROTO_UDPLITE) {
2356 err = udplite_checksum_init(skb, uh);
2360 if (UDP_SKB_CB(skb)->partial_cov) {
2361 skb->csum = inet_compute_pseudo(skb, proto);
2366 /* Note, we are only interested in != 0 or == 0, thus the
2369 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2370 inet_compute_pseudo);
2374 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2375 /* If SW calculated the value, we know it's bad */
2376 if (skb->csum_complete_sw)
2379 /* HW says the value is bad. Let's validate that.
2380 * skb->csum is no longer the full packet checksum,
2381 * so don't treat it as such.
2383 skb_checksum_complete_unset(skb);
2389 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2390 * return code conversion for ip layer consumption
2392 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2397 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2398 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2400 ret = udp_queue_rcv_skb(sk, skb);
2402 /* a return value > 0 means to resubmit the input, but
2403 * it wants the return to be -protocol, or 0
2411 * All we need to do is get the socket, and then do a checksum.
2414 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2419 unsigned short ulen;
2420 struct rtable *rt = skb_rtable(skb);
2421 __be32 saddr, daddr;
2422 struct net *net = dev_net(skb->dev);
2426 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2429 * Validate the packet.
2431 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2432 goto drop; /* No space for header. */
2435 ulen = ntohs(uh->len);
2436 saddr = ip_hdr(skb)->saddr;
2437 daddr = ip_hdr(skb)->daddr;
2439 if (ulen > skb->len)
2442 if (proto == IPPROTO_UDP) {
2443 /* UDP validates ulen. */
2444 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2449 if (udp4_csum_init(skb, uh, proto))
2452 sk = skb_steal_sock(skb, &refcounted);
2454 struct dst_entry *dst = skb_dst(skb);
2457 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2458 udp_sk_rx_dst_set(sk, dst);
2460 ret = udp_unicast_rcv_skb(sk, skb, uh);
2466 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2467 return __udp4_lib_mcast_deliver(net, skb, uh,
2468 saddr, daddr, udptable, proto);
2470 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2472 return udp_unicast_rcv_skb(sk, skb, uh);
2474 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2478 /* No socket. Drop packet silently, if checksum is wrong */
2479 if (udp_lib_checksum_complete(skb))
2482 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2483 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2484 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2487 * Hmm. We got an UDP packet to a port to which we
2488 * don't wanna listen. Ignore it.
2490 kfree_skb_reason(skb, drop_reason);
2494 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2495 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2496 proto == IPPROTO_UDPLITE ? "Lite" : "",
2497 &saddr, ntohs(uh->source),
2499 &daddr, ntohs(uh->dest));
2504 * RFC1122: OK. Discards the bad packet silently (as far as
2505 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2507 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2508 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2509 proto == IPPROTO_UDPLITE ? "Lite" : "",
2510 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2512 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2514 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2515 kfree_skb_reason(skb, drop_reason);
2519 /* We can only early demux multicast if there is a single matching socket.
2520 * If more than one socket found returns NULL
2522 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2523 __be16 loc_port, __be32 loc_addr,
2524 __be16 rmt_port, __be32 rmt_addr,
2527 unsigned short hnum = ntohs(loc_port);
2528 struct sock *sk, *result;
2529 struct udp_hslot *hslot;
2532 slot = udp_hashfn(net, hnum, udp_table.mask);
2533 hslot = &udp_table.hash[slot];
2535 /* Do not bother scanning a too big list */
2536 if (hslot->count > 10)
2540 sk_for_each_rcu(sk, &hslot->head) {
2541 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2542 rmt_port, rmt_addr, dif, sdif, hnum)) {
2552 /* For unicast we should only early demux connected sockets or we can
2553 * break forwarding setups. The chains here can be long so only check
2554 * if the first socket is an exact match and if not move on.
2556 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2557 __be16 loc_port, __be32 loc_addr,
2558 __be16 rmt_port, __be32 rmt_addr,
2561 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2562 unsigned short hnum = ntohs(loc_port);
2563 unsigned int hash2, slot2;
2564 struct udp_hslot *hslot2;
2568 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2569 slot2 = hash2 & udp_table.mask;
2570 hslot2 = &udp_table.hash2[slot2];
2571 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2573 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2574 if (inet_match(net, sk, acookie, ports, dif, sdif))
2576 /* Only check first socket in chain */
2582 int udp_v4_early_demux(struct sk_buff *skb)
2584 struct net *net = dev_net(skb->dev);
2585 struct in_device *in_dev = NULL;
2586 const struct iphdr *iph;
2587 const struct udphdr *uh;
2588 struct sock *sk = NULL;
2589 struct dst_entry *dst;
2590 int dif = skb->dev->ifindex;
2591 int sdif = inet_sdif(skb);
2594 /* validate the packet */
2595 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2601 if (skb->pkt_type == PACKET_MULTICAST) {
2602 in_dev = __in_dev_get_rcu(skb->dev);
2607 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2612 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2613 uh->source, iph->saddr,
2615 } else if (skb->pkt_type == PACKET_HOST) {
2616 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2617 uh->source, iph->saddr, dif, sdif);
2620 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2624 skb->destructor = sock_efree;
2625 dst = rcu_dereference(sk->sk_rx_dst);
2628 dst = dst_check(dst, 0);
2632 /* set noref for now.
2633 * any place which wants to hold dst has to call
2636 skb_dst_set_noref(skb, dst);
2638 /* for unconnected multicast sockets we need to validate
2639 * the source on each packet
2641 if (!inet_sk(sk)->inet_daddr && in_dev)
2642 return ip_mc_validate_source(skb, iph->daddr,
2644 iph->tos & IPTOS_RT_MASK,
2645 skb->dev, in_dev, &itag);
2650 int udp_rcv(struct sk_buff *skb)
2652 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2655 void udp_destroy_sock(struct sock *sk)
2657 struct udp_sock *up = udp_sk(sk);
2658 bool slow = lock_sock_fast(sk);
2660 /* protects from races with udp_abort() */
2661 sock_set_flag(sk, SOCK_DEAD);
2662 udp_flush_pending_frames(sk);
2663 unlock_sock_fast(sk, slow);
2664 if (static_branch_unlikely(&udp_encap_needed_key)) {
2665 if (up->encap_type) {
2666 void (*encap_destroy)(struct sock *sk);
2667 encap_destroy = READ_ONCE(up->encap_destroy);
2671 if (up->encap_enabled)
2672 static_branch_dec(&udp_encap_needed_key);
2677 * Socket option code for UDP
2679 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2680 sockptr_t optval, unsigned int optlen,
2681 int (*push_pending_frames)(struct sock *))
2683 struct udp_sock *up = udp_sk(sk);
2686 int is_udplite = IS_UDPLITE(sk);
2688 if (level == SOL_SOCKET) {
2689 err = sk_setsockopt(sk, level, optname, optval, optlen);
2691 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2692 sockopt_lock_sock(sk);
2693 /* paired with READ_ONCE in udp_rmem_release() */
2694 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2695 sockopt_release_sock(sk);
2700 if (optlen < sizeof(int))
2703 if (copy_from_sockptr(&val, optval, sizeof(val)))
2706 valbool = val ? 1 : 0;
2711 WRITE_ONCE(up->corkflag, 1);
2713 WRITE_ONCE(up->corkflag, 0);
2715 push_pending_frames(sk);
2724 case UDP_ENCAP_ESPINUDP:
2725 case UDP_ENCAP_ESPINUDP_NON_IKE:
2726 #if IS_ENABLED(CONFIG_IPV6)
2727 if (sk->sk_family == AF_INET6)
2728 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2731 up->encap_rcv = xfrm4_udp_encap_rcv;
2734 case UDP_ENCAP_L2TPINUDP:
2735 up->encap_type = val;
2737 udp_tunnel_encap_enable(sk->sk_socket);
2746 case UDP_NO_CHECK6_TX:
2747 up->no_check6_tx = valbool;
2750 case UDP_NO_CHECK6_RX:
2751 up->no_check6_rx = valbool;
2755 if (val < 0 || val > USHRT_MAX)
2757 WRITE_ONCE(up->gso_size, val);
2763 /* when enabling GRO, accept the related GSO packet type */
2765 udp_tunnel_encap_enable(sk->sk_socket);
2766 up->gro_enabled = valbool;
2767 up->accept_udp_l4 = valbool;
2772 * UDP-Lite's partial checksum coverage (RFC 3828).
2774 /* The sender sets actual checksum coverage length via this option.
2775 * The case coverage > packet length is handled by send module. */
2776 case UDPLITE_SEND_CSCOV:
2777 if (!is_udplite) /* Disable the option on UDP sockets */
2778 return -ENOPROTOOPT;
2779 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2781 else if (val > USHRT_MAX)
2784 up->pcflag |= UDPLITE_SEND_CC;
2787 /* The receiver specifies a minimum checksum coverage value. To make
2788 * sense, this should be set to at least 8 (as done below). If zero is
2789 * used, this again means full checksum coverage. */
2790 case UDPLITE_RECV_CSCOV:
2791 if (!is_udplite) /* Disable the option on UDP sockets */
2792 return -ENOPROTOOPT;
2793 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2795 else if (val > USHRT_MAX)
2798 up->pcflag |= UDPLITE_RECV_CC;
2808 EXPORT_SYMBOL(udp_lib_setsockopt);
2810 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2811 unsigned int optlen)
2813 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2814 return udp_lib_setsockopt(sk, level, optname,
2816 udp_push_pending_frames);
2817 return ip_setsockopt(sk, level, optname, optval, optlen);
2820 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2821 char __user *optval, int __user *optlen)
2823 struct udp_sock *up = udp_sk(sk);
2826 if (get_user(len, optlen))
2829 len = min_t(unsigned int, len, sizeof(int));
2836 val = READ_ONCE(up->corkflag);
2840 val = up->encap_type;
2843 case UDP_NO_CHECK6_TX:
2844 val = up->no_check6_tx;
2847 case UDP_NO_CHECK6_RX:
2848 val = up->no_check6_rx;
2852 val = READ_ONCE(up->gso_size);
2856 val = up->gro_enabled;
2859 /* The following two cannot be changed on UDP sockets, the return is
2860 * always 0 (which corresponds to the full checksum coverage of UDP). */
2861 case UDPLITE_SEND_CSCOV:
2865 case UDPLITE_RECV_CSCOV:
2870 return -ENOPROTOOPT;
2873 if (put_user(len, optlen))
2875 if (copy_to_user(optval, &val, len))
2879 EXPORT_SYMBOL(udp_lib_getsockopt);
2881 int udp_getsockopt(struct sock *sk, int level, int optname,
2882 char __user *optval, int __user *optlen)
2884 if (level == SOL_UDP || level == SOL_UDPLITE)
2885 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2886 return ip_getsockopt(sk, level, optname, optval, optlen);
2890 * udp_poll - wait for a UDP event.
2891 * @file: - file struct
2893 * @wait: - poll table
2895 * This is same as datagram poll, except for the special case of
2896 * blocking sockets. If application is using a blocking fd
2897 * and a packet with checksum error is in the queue;
2898 * then it could get return from select indicating data available
2899 * but then block when reading it. Add special case code
2900 * to work around these arguably broken applications.
2902 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2904 __poll_t mask = datagram_poll(file, sock, wait);
2905 struct sock *sk = sock->sk;
2907 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2908 mask |= EPOLLIN | EPOLLRDNORM;
2910 /* Check for false positives due to checksum errors */
2911 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2912 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2913 mask &= ~(EPOLLIN | EPOLLRDNORM);
2915 /* psock ingress_msg queue should not contain any bad checksum frames */
2916 if (sk_is_readable(sk))
2917 mask |= EPOLLIN | EPOLLRDNORM;
2921 EXPORT_SYMBOL(udp_poll);
2923 int udp_abort(struct sock *sk, int err)
2927 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2930 if (sock_flag(sk, SOCK_DEAD))
2934 sk_error_report(sk);
2935 __udp_disconnect(sk, 0);
2942 EXPORT_SYMBOL_GPL(udp_abort);
2944 struct proto udp_prot = {
2946 .owner = THIS_MODULE,
2947 .close = udp_lib_close,
2948 .pre_connect = udp_pre_connect,
2949 .connect = ip4_datagram_connect,
2950 .disconnect = udp_disconnect,
2952 .init = udp_init_sock,
2953 .destroy = udp_destroy_sock,
2954 .setsockopt = udp_setsockopt,
2955 .getsockopt = udp_getsockopt,
2956 .sendmsg = udp_sendmsg,
2957 .recvmsg = udp_recvmsg,
2958 .sendpage = udp_sendpage,
2959 .release_cb = ip4_datagram_release_cb,
2960 .hash = udp_lib_hash,
2961 .unhash = udp_lib_unhash,
2962 .rehash = udp_v4_rehash,
2963 .get_port = udp_v4_get_port,
2964 .put_port = udp_lib_unhash,
2965 #ifdef CONFIG_BPF_SYSCALL
2966 .psock_update_sk_prot = udp_bpf_update_proto,
2968 .memory_allocated = &udp_memory_allocated,
2969 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2971 .sysctl_mem = sysctl_udp_mem,
2972 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2973 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2974 .obj_size = sizeof(struct udp_sock),
2975 .h.udp_table = NULL,
2976 .diag_destroy = udp_abort,
2978 EXPORT_SYMBOL(udp_prot);
2980 /* ------------------------------------------------------------------------ */
2981 #ifdef CONFIG_PROC_FS
2983 static struct udp_table *udp_get_table_afinfo(struct udp_seq_afinfo *afinfo,
2986 return afinfo->udp_table ? : net->ipv4.udp_table;
2989 static struct sock *udp_get_first(struct seq_file *seq, int start)
2991 struct udp_iter_state *state = seq->private;
2992 struct net *net = seq_file_net(seq);
2993 struct udp_seq_afinfo *afinfo;
2994 struct udp_table *udptable;
2997 if (state->bpf_seq_afinfo)
2998 afinfo = state->bpf_seq_afinfo;
3000 afinfo = pde_data(file_inode(seq->file));
3002 udptable = udp_get_table_afinfo(afinfo, net);
3004 for (state->bucket = start; state->bucket <= udptable->mask;
3006 struct udp_hslot *hslot = &udptable->hash[state->bucket];
3008 if (hlist_empty(&hslot->head))
3011 spin_lock_bh(&hslot->lock);
3012 sk_for_each(sk, &hslot->head) {
3013 if (!net_eq(sock_net(sk), net))
3015 if (afinfo->family == AF_UNSPEC ||
3016 sk->sk_family == afinfo->family)
3019 spin_unlock_bh(&hslot->lock);
3026 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3028 struct udp_iter_state *state = seq->private;
3029 struct net *net = seq_file_net(seq);
3030 struct udp_seq_afinfo *afinfo;
3031 struct udp_table *udptable;
3033 if (state->bpf_seq_afinfo)
3034 afinfo = state->bpf_seq_afinfo;
3036 afinfo = pde_data(file_inode(seq->file));
3040 } while (sk && (!net_eq(sock_net(sk), net) ||
3041 (afinfo->family != AF_UNSPEC &&
3042 sk->sk_family != afinfo->family)));
3045 udptable = udp_get_table_afinfo(afinfo, net);
3047 if (state->bucket <= udptable->mask)
3048 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3050 return udp_get_first(seq, state->bucket + 1);
3055 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3057 struct sock *sk = udp_get_first(seq, 0);
3060 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3062 return pos ? NULL : sk;
3065 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3067 struct udp_iter_state *state = seq->private;
3068 state->bucket = MAX_UDP_PORTS;
3070 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3072 EXPORT_SYMBOL(udp_seq_start);
3074 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3078 if (v == SEQ_START_TOKEN)
3079 sk = udp_get_idx(seq, 0);
3081 sk = udp_get_next(seq, v);
3086 EXPORT_SYMBOL(udp_seq_next);
3088 void udp_seq_stop(struct seq_file *seq, void *v)
3090 struct udp_iter_state *state = seq->private;
3091 struct udp_seq_afinfo *afinfo;
3092 struct udp_table *udptable;
3094 if (state->bpf_seq_afinfo)
3095 afinfo = state->bpf_seq_afinfo;
3097 afinfo = pde_data(file_inode(seq->file));
3099 udptable = udp_get_table_afinfo(afinfo, seq_file_net(seq));
3101 if (state->bucket <= udptable->mask)
3102 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3104 EXPORT_SYMBOL(udp_seq_stop);
3106 /* ------------------------------------------------------------------------ */
3107 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3110 struct inet_sock *inet = inet_sk(sp);
3111 __be32 dest = inet->inet_daddr;
3112 __be32 src = inet->inet_rcv_saddr;
3113 __u16 destp = ntohs(inet->inet_dport);
3114 __u16 srcp = ntohs(inet->inet_sport);
3116 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3117 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3118 bucket, src, srcp, dest, destp, sp->sk_state,
3119 sk_wmem_alloc_get(sp),
3122 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3124 refcount_read(&sp->sk_refcnt), sp,
3125 atomic_read(&sp->sk_drops));
3128 int udp4_seq_show(struct seq_file *seq, void *v)
3130 seq_setwidth(seq, 127);
3131 if (v == SEQ_START_TOKEN)
3132 seq_puts(seq, " sl local_address rem_address st tx_queue "
3133 "rx_queue tr tm->when retrnsmt uid timeout "
3134 "inode ref pointer drops");
3136 struct udp_iter_state *state = seq->private;
3138 udp4_format_sock(v, seq, state->bucket);
3144 #ifdef CONFIG_BPF_SYSCALL
3145 struct bpf_iter__udp {
3146 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3147 __bpf_md_ptr(struct udp_sock *, udp_sk);
3148 uid_t uid __aligned(8);
3149 int bucket __aligned(8);
3152 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3153 struct udp_sock *udp_sk, uid_t uid, int bucket)
3155 struct bpf_iter__udp ctx;
3157 meta->seq_num--; /* skip SEQ_START_TOKEN */
3159 ctx.udp_sk = udp_sk;
3161 ctx.bucket = bucket;
3162 return bpf_iter_run_prog(prog, &ctx);
3165 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3167 struct udp_iter_state *state = seq->private;
3168 struct bpf_iter_meta meta;
3169 struct bpf_prog *prog;
3170 struct sock *sk = v;
3173 if (v == SEQ_START_TOKEN)
3176 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3178 prog = bpf_iter_get_info(&meta, false);
3179 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3182 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3184 struct bpf_iter_meta meta;
3185 struct bpf_prog *prog;
3189 prog = bpf_iter_get_info(&meta, true);
3191 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3194 udp_seq_stop(seq, v);
3197 static const struct seq_operations bpf_iter_udp_seq_ops = {
3198 .start = udp_seq_start,
3199 .next = udp_seq_next,
3200 .stop = bpf_iter_udp_seq_stop,
3201 .show = bpf_iter_udp_seq_show,
3205 const struct seq_operations udp_seq_ops = {
3206 .start = udp_seq_start,
3207 .next = udp_seq_next,
3208 .stop = udp_seq_stop,
3209 .show = udp4_seq_show,
3211 EXPORT_SYMBOL(udp_seq_ops);
3213 static struct udp_seq_afinfo udp4_seq_afinfo = {
3218 static int __net_init udp4_proc_init_net(struct net *net)
3220 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3221 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3226 static void __net_exit udp4_proc_exit_net(struct net *net)
3228 remove_proc_entry("udp", net->proc_net);
3231 static struct pernet_operations udp4_net_ops = {
3232 .init = udp4_proc_init_net,
3233 .exit = udp4_proc_exit_net,
3236 int __init udp4_proc_init(void)
3238 return register_pernet_subsys(&udp4_net_ops);
3241 void udp4_proc_exit(void)
3243 unregister_pernet_subsys(&udp4_net_ops);
3245 #endif /* CONFIG_PROC_FS */
3247 static __initdata unsigned long uhash_entries;
3248 static int __init set_uhash_entries(char *str)
3255 ret = kstrtoul(str, 0, &uhash_entries);
3259 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3260 uhash_entries = UDP_HTABLE_SIZE_MIN;
3263 __setup("uhash_entries=", set_uhash_entries);
3265 void __init udp_table_init(struct udp_table *table, const char *name)
3269 table->hash = alloc_large_system_hash(name,
3270 2 * sizeof(struct udp_hslot),
3272 21, /* one slot per 2 MB */
3276 UDP_HTABLE_SIZE_MIN,
3279 table->hash2 = table->hash + (table->mask + 1);
3280 for (i = 0; i <= table->mask; i++) {
3281 INIT_HLIST_HEAD(&table->hash[i].head);
3282 table->hash[i].count = 0;
3283 spin_lock_init(&table->hash[i].lock);
3285 for (i = 0; i <= table->mask; i++) {
3286 INIT_HLIST_HEAD(&table->hash2[i].head);
3287 table->hash2[i].count = 0;
3288 spin_lock_init(&table->hash2[i].lock);
3292 u32 udp_flow_hashrnd(void)
3294 static u32 hashrnd __read_mostly;
3296 net_get_random_once(&hashrnd, sizeof(hashrnd));
3300 EXPORT_SYMBOL(udp_flow_hashrnd);
3302 static int __net_init udp_sysctl_init(struct net *net)
3304 net->ipv4.udp_table = &udp_table;
3306 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3307 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3309 #ifdef CONFIG_NET_L3_MASTER_DEV
3310 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3316 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3317 .init = udp_sysctl_init,
3320 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3321 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3322 struct udp_sock *udp_sk, uid_t uid, int bucket)
3324 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3326 struct udp_iter_state *st = priv_data;
3327 struct udp_seq_afinfo *afinfo;
3330 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3334 afinfo->family = AF_UNSPEC;
3335 afinfo->udp_table = NULL;
3336 st->bpf_seq_afinfo = afinfo;
3337 ret = bpf_iter_init_seq_net(priv_data, aux);
3343 static void bpf_iter_fini_udp(void *priv_data)
3345 struct udp_iter_state *st = priv_data;
3347 kfree(st->bpf_seq_afinfo);
3348 bpf_iter_fini_seq_net(priv_data);
3351 static const struct bpf_iter_seq_info udp_seq_info = {
3352 .seq_ops = &bpf_iter_udp_seq_ops,
3353 .init_seq_private = bpf_iter_init_udp,
3354 .fini_seq_private = bpf_iter_fini_udp,
3355 .seq_priv_size = sizeof(struct udp_iter_state),
3358 static struct bpf_iter_reg udp_reg_info = {
3360 .ctx_arg_info_size = 1,
3362 { offsetof(struct bpf_iter__udp, udp_sk),
3363 PTR_TO_BTF_ID_OR_NULL },
3365 .seq_info = &udp_seq_info,
3368 static void __init bpf_iter_register(void)
3370 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3371 if (bpf_iter_reg_target(&udp_reg_info))
3372 pr_warn("Warning: could not register bpf iterator udp\n");
3376 void __init udp_init(void)
3378 unsigned long limit;
3381 udp_table_init(&udp_table, "UDP");
3382 limit = nr_free_buffer_pages() / 8;
3383 limit = max(limit, 128UL);
3384 sysctl_udp_mem[0] = limit / 4 * 3;
3385 sysctl_udp_mem[1] = limit;
3386 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3388 /* 16 spinlocks per cpu */
3389 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3390 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3393 panic("UDP: failed to alloc udp_busylocks\n");
3394 for (i = 0; i < (1U << udp_busylocks_log); i++)
3395 spin_lock_init(udp_busylocks + i);
3397 if (register_pernet_subsys(&udp_sysctl_ops))
3398 panic("UDP: failed to init sysctl parameters.\n");
3400 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3401 bpf_iter_register();