1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * The User Datagram Protocol (UDP).
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
13 * Hirokazu Takahashi, <taka@valinux.co.jp>
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
72 * James Chapman : Add L2TP encapsulation type.
75 #define pr_fmt(fmt) "UDP: " fmt
77 #include <linux/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <linux/memblock.h>
80 #include <linux/highmem.h>
81 #include <linux/swap.h>
82 #include <linux/types.h>
83 #include <linux/fcntl.h>
84 #include <linux/module.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/igmp.h>
88 #include <linux/inetdevice.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
93 #include <linux/inet.h>
94 #include <linux/netdevice.h>
95 #include <linux/slab.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/inet_hashtables.h>
103 #include <net/ip_tunnels.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <trace/events/udp.h>
108 #include <linux/static_key.h>
109 #include <trace/events/skb.h>
110 #include <net/busy_poll.h>
111 #include "udp_impl.h"
112 #include <net/sock_reuseport.h>
113 #include <net/addrconf.h>
114 #include <net/udp_tunnel.h>
116 struct udp_table udp_table __read_mostly;
117 EXPORT_SYMBOL(udp_table);
119 long sysctl_udp_mem[3] __read_mostly;
120 EXPORT_SYMBOL(sysctl_udp_mem);
122 atomic_long_t udp_memory_allocated;
123 EXPORT_SYMBOL(udp_memory_allocated);
125 #define MAX_UDP_PORTS 65536
126 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
128 /* IPCB reference means this can not be used from early demux */
129 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
131 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
132 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
133 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
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_hslot *hslot, *hslot2;
241 struct udp_table *udptable = sk->sk_prot->h.udp_table;
243 struct net *net = sock_net(sk);
246 int low, high, remaining;
248 unsigned short first, last;
249 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
251 inet_get_local_port_range(net, &low, &high);
252 remaining = (high - low) + 1;
254 rand = prandom_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,
367 int dif, int sdif, bool exact_dif)
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,
402 if (sk->sk_incoming_cpu == raw_smp_processor_id())
407 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
408 const __u16 lport, const __be32 faddr,
411 static u32 udp_ehash_secret __read_mostly;
413 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
415 return __inet_ehashfn(laddr, lport, faddr, fport,
416 udp_ehash_secret + net_hash_mix(net));
419 /* called with rcu_read_lock() */
420 static struct sock *udp4_lib_lookup2(struct net *net,
421 __be32 saddr, __be16 sport,
422 __be32 daddr, unsigned int hnum,
423 int dif, int sdif, bool exact_dif,
424 struct udp_hslot *hslot2,
427 struct sock *sk, *result;
433 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
434 score = compute_score(sk, net, saddr, sport,
435 daddr, hnum, dif, sdif, exact_dif);
436 if (score > badness) {
437 if (sk->sk_reuseport) {
438 hash = udp_ehashfn(net, daddr, hnum,
440 result = reuseport_select_sock(sk, hash, skb,
441 sizeof(struct udphdr));
452 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
453 * harder than this. -DaveM
455 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
456 __be16 sport, __be32 daddr, __be16 dport, int dif,
457 int sdif, struct udp_table *udptable, struct sk_buff *skb)
460 unsigned short hnum = ntohs(dport);
461 unsigned int hash2, slot2;
462 struct udp_hslot *hslot2;
463 bool exact_dif = udp_lib_exact_dif_match(net, skb);
465 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
466 slot2 = hash2 & udptable->mask;
467 hslot2 = &udptable->hash2[slot2];
469 result = udp4_lib_lookup2(net, saddr, sport,
470 daddr, hnum, dif, sdif,
471 exact_dif, hslot2, skb);
473 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
474 slot2 = hash2 & udptable->mask;
475 hslot2 = &udptable->hash2[slot2];
477 result = udp4_lib_lookup2(net, saddr, sport,
478 htonl(INADDR_ANY), hnum, dif, sdif,
479 exact_dif, hslot2, skb);
481 if (unlikely(IS_ERR(result)))
485 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
487 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
488 __be16 sport, __be16 dport,
489 struct udp_table *udptable)
491 const struct iphdr *iph = ip_hdr(skb);
493 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
494 iph->daddr, dport, inet_iif(skb),
495 inet_sdif(skb), udptable, skb);
498 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
499 __be16 sport, __be16 dport)
501 const struct iphdr *iph = ip_hdr(skb);
503 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
504 iph->daddr, dport, inet_iif(skb),
505 inet_sdif(skb), &udp_table, NULL);
507 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
509 /* Must be called under rcu_read_lock().
510 * Does increment socket refcount.
512 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
513 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
514 __be32 daddr, __be16 dport, int dif)
518 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
519 dif, 0, &udp_table, NULL);
520 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
524 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
527 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
528 __be16 loc_port, __be32 loc_addr,
529 __be16 rmt_port, __be32 rmt_addr,
530 int dif, int sdif, unsigned short hnum)
532 struct inet_sock *inet = inet_sk(sk);
534 if (!net_eq(sock_net(sk), net) ||
535 udp_sk(sk)->udp_port_hash != hnum ||
536 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
537 (inet->inet_dport != rmt_port && inet->inet_dport) ||
538 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
539 ipv6_only_sock(sk) ||
540 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
542 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
547 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
548 void udp_encap_enable(void)
550 static_branch_inc(&udp_encap_needed_key);
552 EXPORT_SYMBOL(udp_encap_enable);
554 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
555 * through error handlers in encapsulations looking for a match.
557 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
561 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
562 int (*handler)(struct sk_buff *skb, u32 info);
563 const struct ip_tunnel_encap_ops *encap;
565 encap = rcu_dereference(iptun_encaps[i]);
568 handler = encap->err_handler;
569 if (handler && !handler(skb, info))
576 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
577 * reversing source and destination port: this will match tunnels that force the
578 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
579 * lwtunnels might actually break this assumption by being configured with
580 * different destination ports on endpoints, in this case we won't be able to
581 * trace ICMP messages back to them.
583 * If this doesn't match any socket, probe tunnels with arbitrary destination
584 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
585 * we've sent packets to won't necessarily match the local destination port.
587 * Then ask the tunnel implementation to match the error against a valid
590 * Return an error if we can't find a match, the socket if we need further
591 * processing, zero otherwise.
593 static struct sock *__udp4_lib_err_encap(struct net *net,
594 const struct iphdr *iph,
596 struct udp_table *udptable,
597 struct sk_buff *skb, u32 info)
599 int network_offset, transport_offset;
602 network_offset = skb_network_offset(skb);
603 transport_offset = skb_transport_offset(skb);
605 /* Network header needs to point to the outer IPv4 header inside ICMP */
606 skb_reset_network_header(skb);
608 /* Transport header needs to point to the UDP header */
609 skb_set_transport_header(skb, iph->ihl << 2);
611 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
612 iph->saddr, uh->dest, skb->dev->ifindex, 0,
615 int (*lookup)(struct sock *sk, struct sk_buff *skb);
616 struct udp_sock *up = udp_sk(sk);
618 lookup = READ_ONCE(up->encap_err_lookup);
619 if (!lookup || lookup(sk, skb))
624 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
626 skb_set_transport_header(skb, transport_offset);
627 skb_set_network_header(skb, network_offset);
633 * This routine is called by the ICMP module when it gets some
634 * sort of error condition. If err < 0 then the socket should
635 * be closed and the error returned to the user. If err > 0
636 * it's just the icmp type << 8 | icmp code.
637 * Header points to the ip header of the error packet. We move
638 * on past this. Then (as it used to claim before adjustment)
639 * header points to the first 8 bytes of the udp header. We need
640 * to find the appropriate port.
643 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
645 struct inet_sock *inet;
646 const struct iphdr *iph = (const struct iphdr *)skb->data;
647 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
648 const int type = icmp_hdr(skb)->type;
649 const int code = icmp_hdr(skb)->code;
654 struct net *net = dev_net(skb->dev);
656 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
657 iph->saddr, uh->source, skb->dev->ifindex,
658 inet_sdif(skb), udptable, NULL);
660 /* No socket for error: try tunnels before discarding */
661 sk = ERR_PTR(-ENOENT);
662 if (static_branch_unlikely(&udp_encap_needed_key)) {
663 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
670 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
683 case ICMP_TIME_EXCEEDED:
686 case ICMP_SOURCE_QUENCH:
688 case ICMP_PARAMETERPROB:
692 case ICMP_DEST_UNREACH:
693 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
694 ipv4_sk_update_pmtu(skb, sk, info);
695 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
703 if (code <= NR_ICMP_UNREACH) {
704 harderr = icmp_err_convert[code].fatal;
705 err = icmp_err_convert[code].errno;
709 ipv4_sk_redirect(skb, sk);
714 * RFC1122: OK. Passes ICMP errors back to application, as per
718 /* ...not for tunnels though: we don't have a sending socket */
721 if (!inet->recverr) {
722 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
725 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
728 sk->sk_error_report(sk);
733 int udp_err(struct sk_buff *skb, u32 info)
735 return __udp4_lib_err(skb, info, &udp_table);
739 * Throw away all pending data and cancel the corking. Socket is locked.
741 void udp_flush_pending_frames(struct sock *sk)
743 struct udp_sock *up = udp_sk(sk);
748 ip_flush_pending_frames(sk);
751 EXPORT_SYMBOL(udp_flush_pending_frames);
754 * udp4_hwcsum - handle outgoing HW checksumming
755 * @skb: sk_buff containing the filled-in UDP header
756 * (checksum field must be zeroed out)
757 * @src: source IP address
758 * @dst: destination IP address
760 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
762 struct udphdr *uh = udp_hdr(skb);
763 int offset = skb_transport_offset(skb);
764 int len = skb->len - offset;
768 if (!skb_has_frag_list(skb)) {
770 * Only one fragment on the socket.
772 skb->csum_start = skb_transport_header(skb) - skb->head;
773 skb->csum_offset = offsetof(struct udphdr, check);
774 uh->check = ~csum_tcpudp_magic(src, dst, len,
777 struct sk_buff *frags;
780 * HW-checksum won't work as there are two or more
781 * fragments on the socket so that all csums of sk_buffs
784 skb_walk_frags(skb, frags) {
785 csum = csum_add(csum, frags->csum);
789 csum = skb_checksum(skb, offset, hlen, csum);
790 skb->ip_summed = CHECKSUM_NONE;
792 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
794 uh->check = CSUM_MANGLED_0;
797 EXPORT_SYMBOL_GPL(udp4_hwcsum);
799 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
800 * for the simple case like when setting the checksum for a UDP tunnel.
802 void udp_set_csum(bool nocheck, struct sk_buff *skb,
803 __be32 saddr, __be32 daddr, int len)
805 struct udphdr *uh = udp_hdr(skb);
809 } else if (skb_is_gso(skb)) {
810 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
811 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
813 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
815 uh->check = CSUM_MANGLED_0;
817 skb->ip_summed = CHECKSUM_PARTIAL;
818 skb->csum_start = skb_transport_header(skb) - skb->head;
819 skb->csum_offset = offsetof(struct udphdr, check);
820 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
823 EXPORT_SYMBOL(udp_set_csum);
825 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
826 struct inet_cork *cork)
828 struct sock *sk = skb->sk;
829 struct inet_sock *inet = inet_sk(sk);
832 int is_udplite = IS_UDPLITE(sk);
833 int offset = skb_transport_offset(skb);
834 int len = skb->len - offset;
838 * Create a UDP header
841 uh->source = inet->inet_sport;
842 uh->dest = fl4->fl4_dport;
843 uh->len = htons(len);
846 if (cork->gso_size) {
847 const int hlen = skb_network_header_len(skb) +
848 sizeof(struct udphdr);
850 if (hlen + cork->gso_size > cork->fragsize) {
854 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
858 if (sk->sk_no_check_tx) {
862 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
863 dst_xfrm(skb_dst(skb))) {
868 skb_shinfo(skb)->gso_size = cork->gso_size;
869 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
870 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh),
875 if (is_udplite) /* UDP-Lite */
876 csum = udplite_csum(skb);
878 else if (sk->sk_no_check_tx) { /* UDP csum off */
880 skb->ip_summed = CHECKSUM_NONE;
883 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
886 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
890 csum = udp_csum(skb);
892 /* add protocol-dependent pseudo-header */
893 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
894 sk->sk_protocol, csum);
896 uh->check = CSUM_MANGLED_0;
899 err = ip_send_skb(sock_net(sk), skb);
901 if (err == -ENOBUFS && !inet->recverr) {
902 UDP_INC_STATS(sock_net(sk),
903 UDP_MIB_SNDBUFERRORS, is_udplite);
907 UDP_INC_STATS(sock_net(sk),
908 UDP_MIB_OUTDATAGRAMS, is_udplite);
913 * Push out all pending data as one UDP datagram. Socket is locked.
915 int udp_push_pending_frames(struct sock *sk)
917 struct udp_sock *up = udp_sk(sk);
918 struct inet_sock *inet = inet_sk(sk);
919 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
923 skb = ip_finish_skb(sk, fl4);
927 err = udp_send_skb(skb, fl4, &inet->cork.base);
934 EXPORT_SYMBOL(udp_push_pending_frames);
936 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
938 switch (cmsg->cmsg_type) {
940 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
942 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
949 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
951 struct cmsghdr *cmsg;
952 bool need_ip = false;
955 for_each_cmsghdr(cmsg, msg) {
956 if (!CMSG_OK(msg, cmsg))
959 if (cmsg->cmsg_level != SOL_UDP) {
964 err = __udp_cmsg_send(cmsg, gso_size);
971 EXPORT_SYMBOL_GPL(udp_cmsg_send);
973 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
975 struct inet_sock *inet = inet_sk(sk);
976 struct udp_sock *up = udp_sk(sk);
977 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
978 struct flowi4 fl4_stack;
981 struct ipcm_cookie ipc;
982 struct rtable *rt = NULL;
985 __be32 daddr, faddr, saddr;
988 int err, is_udplite = IS_UDPLITE(sk);
989 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
990 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
992 struct ip_options_data opt_copy;
1001 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1004 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1006 fl4 = &inet->cork.fl.u.ip4;
1009 * There are pending frames.
1010 * The socket lock must be held while it's corked.
1013 if (likely(up->pending)) {
1014 if (unlikely(up->pending != AF_INET)) {
1018 goto do_append_data;
1022 ulen += sizeof(struct udphdr);
1025 * Get and verify the address.
1028 if (msg->msg_namelen < sizeof(*usin))
1030 if (usin->sin_family != AF_INET) {
1031 if (usin->sin_family != AF_UNSPEC)
1032 return -EAFNOSUPPORT;
1035 daddr = usin->sin_addr.s_addr;
1036 dport = usin->sin_port;
1040 if (sk->sk_state != TCP_ESTABLISHED)
1041 return -EDESTADDRREQ;
1042 daddr = inet->inet_daddr;
1043 dport = inet->inet_dport;
1044 /* Open fast path for connected socket.
1045 Route will not be used, if at least one option is set.
1050 ipcm_init_sk(&ipc, inet);
1051 ipc.gso_size = up->gso_size;
1053 if (msg->msg_controllen) {
1054 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1056 err = ip_cmsg_send(sk, msg, &ipc,
1057 sk->sk_family == AF_INET6);
1058 if (unlikely(err < 0)) {
1067 struct ip_options_rcu *inet_opt;
1070 inet_opt = rcu_dereference(inet->inet_opt);
1072 memcpy(&opt_copy, inet_opt,
1073 sizeof(*inet_opt) + inet_opt->opt.optlen);
1074 ipc.opt = &opt_copy.opt;
1079 if (cgroup_bpf_enabled && !connected) {
1080 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1081 (struct sockaddr *)usin, &ipc.addr);
1085 if (usin->sin_port == 0) {
1086 /* BPF program set invalid port. Reject it. */
1090 daddr = usin->sin_addr.s_addr;
1091 dport = usin->sin_port;
1096 ipc.addr = faddr = daddr;
1098 if (ipc.opt && ipc.opt->opt.srr) {
1103 faddr = ipc.opt->opt.faddr;
1106 tos = get_rttos(&ipc, inet);
1107 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1108 (msg->msg_flags & MSG_DONTROUTE) ||
1109 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1114 if (ipv4_is_multicast(daddr)) {
1115 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1116 ipc.oif = inet->mc_index;
1118 saddr = inet->mc_addr;
1120 } else if (!ipc.oif) {
1121 ipc.oif = inet->uc_index;
1122 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1123 /* oif is set, packet is to local broadcast and
1124 * and uc_index is set. oif is most likely set
1125 * by sk_bound_dev_if. If uc_index != oif check if the
1126 * oif is an L3 master and uc_index is an L3 slave.
1127 * If so, we want to allow the send using the uc_index.
1129 if (ipc.oif != inet->uc_index &&
1130 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1132 ipc.oif = inet->uc_index;
1137 rt = (struct rtable *)sk_dst_check(sk, 0);
1140 struct net *net = sock_net(sk);
1141 __u8 flow_flags = inet_sk_flowi_flags(sk);
1145 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1146 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1148 faddr, saddr, dport, inet->inet_sport,
1151 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1152 rt = ip_route_output_flow(net, fl4, sk);
1156 if (err == -ENETUNREACH)
1157 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1162 if ((rt->rt_flags & RTCF_BROADCAST) &&
1163 !sock_flag(sk, SOCK_BROADCAST))
1166 sk_dst_set(sk, dst_clone(&rt->dst));
1169 if (msg->msg_flags&MSG_CONFIRM)
1175 daddr = ipc.addr = fl4->daddr;
1177 /* Lockless fast path for the non-corking case. */
1179 struct inet_cork cork;
1181 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1182 sizeof(struct udphdr), &ipc, &rt,
1183 &cork, msg->msg_flags);
1185 if (!IS_ERR_OR_NULL(skb))
1186 err = udp_send_skb(skb, fl4, &cork);
1191 if (unlikely(up->pending)) {
1192 /* The socket is already corked while preparing it. */
1193 /* ... which is an evident application bug. --ANK */
1196 net_dbg_ratelimited("socket already corked\n");
1201 * Now cork the socket to pend data.
1203 fl4 = &inet->cork.fl.u.ip4;
1206 fl4->fl4_dport = dport;
1207 fl4->fl4_sport = inet->inet_sport;
1208 up->pending = AF_INET;
1212 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1213 sizeof(struct udphdr), &ipc, &rt,
1214 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1216 udp_flush_pending_frames(sk);
1218 err = udp_push_pending_frames(sk);
1219 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1231 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1232 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1233 * we don't have a good statistic (IpOutDiscards but it can be too many
1234 * things). We could add another new stat but at least for now that
1235 * seems like overkill.
1237 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1238 UDP_INC_STATS(sock_net(sk),
1239 UDP_MIB_SNDBUFERRORS, is_udplite);
1244 if (msg->msg_flags & MSG_PROBE)
1245 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1246 if (!(msg->msg_flags&MSG_PROBE) || len)
1247 goto back_from_confirm;
1251 EXPORT_SYMBOL(udp_sendmsg);
1253 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1254 size_t size, int flags)
1256 struct inet_sock *inet = inet_sk(sk);
1257 struct udp_sock *up = udp_sk(sk);
1260 if (flags & MSG_SENDPAGE_NOTLAST)
1264 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1266 /* Call udp_sendmsg to specify destination address which
1267 * sendpage interface can't pass.
1268 * This will succeed only when the socket is connected.
1270 ret = udp_sendmsg(sk, &msg, 0);
1277 if (unlikely(!up->pending)) {
1280 net_dbg_ratelimited("cork failed\n");
1284 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1285 page, offset, size, flags);
1286 if (ret == -EOPNOTSUPP) {
1288 return sock_no_sendpage(sk->sk_socket, page, offset,
1292 udp_flush_pending_frames(sk);
1297 if (!(up->corkflag || (flags&MSG_MORE)))
1298 ret = udp_push_pending_frames(sk);
1306 #define UDP_SKB_IS_STATELESS 0x80000000
1308 static void udp_set_dev_scratch(struct sk_buff *skb)
1310 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1312 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1313 scratch->_tsize_state = skb->truesize;
1314 #if BITS_PER_LONG == 64
1315 scratch->len = skb->len;
1316 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1317 scratch->is_linear = !skb_is_nonlinear(skb);
1319 /* all head states execept sp (dst, sk, nf) are always cleared by
1320 * udp_rcv() and we need to preserve secpath, if present, to eventually
1321 * process IP_CMSG_PASSSEC at recvmsg() time
1323 if (likely(!skb_sec_path(skb)))
1324 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1327 static int udp_skb_truesize(struct sk_buff *skb)
1329 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1332 static bool udp_skb_has_head_state(struct sk_buff *skb)
1334 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1337 /* fully reclaim rmem/fwd memory allocated for skb */
1338 static void udp_rmem_release(struct sock *sk, int size, int partial,
1339 bool rx_queue_lock_held)
1341 struct udp_sock *up = udp_sk(sk);
1342 struct sk_buff_head *sk_queue;
1345 if (likely(partial)) {
1346 up->forward_deficit += size;
1347 size = up->forward_deficit;
1348 if (size < (sk->sk_rcvbuf >> 2))
1351 size += up->forward_deficit;
1353 up->forward_deficit = 0;
1355 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1356 * if the called don't held it already
1358 sk_queue = &sk->sk_receive_queue;
1359 if (!rx_queue_lock_held)
1360 spin_lock(&sk_queue->lock);
1363 sk->sk_forward_alloc += size;
1364 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1365 sk->sk_forward_alloc -= amt;
1368 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1370 atomic_sub(size, &sk->sk_rmem_alloc);
1372 /* this can save us from acquiring the rx queue lock on next receive */
1373 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1375 if (!rx_queue_lock_held)
1376 spin_unlock(&sk_queue->lock);
1379 /* Note: called with reader_queue.lock held.
1380 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1381 * This avoids a cache line miss while receive_queue lock is held.
1382 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1384 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1386 prefetch(&skb->data);
1387 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1389 EXPORT_SYMBOL(udp_skb_destructor);
1391 /* as above, but the caller held the rx queue lock, too */
1392 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1394 prefetch(&skb->data);
1395 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1398 /* Idea of busylocks is to let producers grab an extra spinlock
1399 * to relieve pressure on the receive_queue spinlock shared by consumer.
1400 * Under flood, this means that only one producer can be in line
1401 * trying to acquire the receive_queue spinlock.
1402 * These busylock can be allocated on a per cpu manner, instead of a
1403 * per socket one (that would consume a cache line per socket)
1405 static int udp_busylocks_log __read_mostly;
1406 static spinlock_t *udp_busylocks __read_mostly;
1408 static spinlock_t *busylock_acquire(void *ptr)
1412 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1417 static void busylock_release(spinlock_t *busy)
1423 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1425 struct sk_buff_head *list = &sk->sk_receive_queue;
1426 int rmem, delta, amt, err = -ENOMEM;
1427 spinlock_t *busy = NULL;
1430 /* try to avoid the costly atomic add/sub pair when the receive
1431 * queue is full; always allow at least a packet
1433 rmem = atomic_read(&sk->sk_rmem_alloc);
1434 if (rmem > sk->sk_rcvbuf)
1437 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1438 * having linear skbs :
1439 * - Reduce memory overhead and thus increase receive queue capacity
1440 * - Less cache line misses at copyout() time
1441 * - Less work at consume_skb() (less alien page frag freeing)
1443 if (rmem > (sk->sk_rcvbuf >> 1)) {
1446 busy = busylock_acquire(sk);
1448 size = skb->truesize;
1449 udp_set_dev_scratch(skb);
1451 /* we drop only if the receive buf is full and the receive
1452 * queue contains some other skb
1454 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1455 if (rmem > (size + sk->sk_rcvbuf))
1458 spin_lock(&list->lock);
1459 if (size >= sk->sk_forward_alloc) {
1460 amt = sk_mem_pages(size);
1461 delta = amt << SK_MEM_QUANTUM_SHIFT;
1462 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1464 spin_unlock(&list->lock);
1468 sk->sk_forward_alloc += delta;
1471 sk->sk_forward_alloc -= size;
1473 /* no need to setup a destructor, we will explicitly release the
1474 * forward allocated memory on dequeue
1476 sock_skb_set_dropcount(sk, skb);
1478 __skb_queue_tail(list, skb);
1479 spin_unlock(&list->lock);
1481 if (!sock_flag(sk, SOCK_DEAD))
1482 sk->sk_data_ready(sk);
1484 busylock_release(busy);
1488 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1491 atomic_inc(&sk->sk_drops);
1492 busylock_release(busy);
1495 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1497 void udp_destruct_sock(struct sock *sk)
1499 /* reclaim completely the forward allocated memory */
1500 struct udp_sock *up = udp_sk(sk);
1501 unsigned int total = 0;
1502 struct sk_buff *skb;
1504 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1505 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1506 total += skb->truesize;
1509 udp_rmem_release(sk, total, 0, true);
1511 inet_sock_destruct(sk);
1513 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1515 int udp_init_sock(struct sock *sk)
1517 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1518 sk->sk_destruct = udp_destruct_sock;
1521 EXPORT_SYMBOL_GPL(udp_init_sock);
1523 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1525 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1526 bool slow = lock_sock_fast(sk);
1528 sk_peek_offset_bwd(sk, len);
1529 unlock_sock_fast(sk, slow);
1532 if (!skb_unref(skb))
1535 /* In the more common cases we cleared the head states previously,
1536 * see __udp_queue_rcv_skb().
1538 if (unlikely(udp_skb_has_head_state(skb)))
1539 skb_release_head_state(skb);
1540 __consume_stateless_skb(skb);
1542 EXPORT_SYMBOL_GPL(skb_consume_udp);
1544 static struct sk_buff *__first_packet_length(struct sock *sk,
1545 struct sk_buff_head *rcvq,
1548 struct sk_buff *skb;
1550 while ((skb = skb_peek(rcvq)) != NULL) {
1551 if (udp_lib_checksum_complete(skb)) {
1552 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1554 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1556 atomic_inc(&sk->sk_drops);
1557 __skb_unlink(skb, rcvq);
1558 *total += skb->truesize;
1561 /* the csum related bits could be changed, refresh
1564 udp_set_dev_scratch(skb);
1572 * first_packet_length - return length of first packet in receive queue
1575 * Drops all bad checksum frames, until a valid one is found.
1576 * Returns the length of found skb, or -1 if none is found.
1578 static int first_packet_length(struct sock *sk)
1580 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1581 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1582 struct sk_buff *skb;
1586 spin_lock_bh(&rcvq->lock);
1587 skb = __first_packet_length(sk, rcvq, &total);
1588 if (!skb && !skb_queue_empty(sk_queue)) {
1589 spin_lock(&sk_queue->lock);
1590 skb_queue_splice_tail_init(sk_queue, rcvq);
1591 spin_unlock(&sk_queue->lock);
1593 skb = __first_packet_length(sk, rcvq, &total);
1595 res = skb ? skb->len : -1;
1597 udp_rmem_release(sk, total, 1, false);
1598 spin_unlock_bh(&rcvq->lock);
1603 * IOCTL requests applicable to the UDP protocol
1606 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1611 int amount = sk_wmem_alloc_get(sk);
1613 return put_user(amount, (int __user *)arg);
1618 int amount = max_t(int, 0, first_packet_length(sk));
1620 return put_user(amount, (int __user *)arg);
1624 return -ENOIOCTLCMD;
1629 EXPORT_SYMBOL(udp_ioctl);
1631 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1632 int noblock, int *off, int *err)
1634 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1635 struct sk_buff_head *queue;
1636 struct sk_buff *last;
1640 queue = &udp_sk(sk)->reader_queue;
1641 flags |= noblock ? MSG_DONTWAIT : 0;
1642 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1644 struct sk_buff *skb;
1646 error = sock_error(sk);
1652 spin_lock_bh(&queue->lock);
1653 skb = __skb_try_recv_from_queue(sk, queue, flags,
1657 spin_unlock_bh(&queue->lock);
1661 if (skb_queue_empty(sk_queue)) {
1662 spin_unlock_bh(&queue->lock);
1666 /* refill the reader queue and walk it again
1667 * keep both queues locked to avoid re-acquiring
1668 * the sk_receive_queue lock if fwd memory scheduling
1671 spin_lock(&sk_queue->lock);
1672 skb_queue_splice_tail_init(sk_queue, queue);
1674 skb = __skb_try_recv_from_queue(sk, queue, flags,
1675 udp_skb_dtor_locked,
1677 spin_unlock(&sk_queue->lock);
1678 spin_unlock_bh(&queue->lock);
1683 if (!sk_can_busy_loop(sk))
1686 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1687 } while (!skb_queue_empty(sk_queue));
1689 /* sk_queue is empty, reader_queue may contain peeked packets */
1691 !__skb_wait_for_more_packets(sk, &error, &timeo,
1692 (struct sk_buff *)sk_queue));
1697 EXPORT_SYMBOL(__skb_recv_udp);
1700 * This should be easy, if there is something there we
1701 * return it, otherwise we block.
1704 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1705 int flags, int *addr_len)
1707 struct inet_sock *inet = inet_sk(sk);
1708 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1709 struct sk_buff *skb;
1710 unsigned int ulen, copied;
1711 int off, err, peeking = flags & MSG_PEEK;
1712 int is_udplite = IS_UDPLITE(sk);
1713 bool checksum_valid = false;
1715 if (flags & MSG_ERRQUEUE)
1716 return ip_recv_error(sk, msg, len, addr_len);
1719 off = sk_peek_offset(sk, flags);
1720 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1724 ulen = udp_skb_len(skb);
1726 if (copied > ulen - off)
1727 copied = ulen - off;
1728 else if (copied < ulen)
1729 msg->msg_flags |= MSG_TRUNC;
1732 * If checksum is needed at all, try to do it while copying the
1733 * data. If the data is truncated, or if we only want a partial
1734 * coverage checksum (UDP-Lite), do it before the copy.
1737 if (copied < ulen || peeking ||
1738 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1739 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1740 !__udp_lib_checksum_complete(skb);
1741 if (!checksum_valid)
1745 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1746 if (udp_skb_is_linear(skb))
1747 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1749 err = skb_copy_datagram_msg(skb, off, msg, copied);
1751 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1757 if (unlikely(err)) {
1759 atomic_inc(&sk->sk_drops);
1760 UDP_INC_STATS(sock_net(sk),
1761 UDP_MIB_INERRORS, is_udplite);
1768 UDP_INC_STATS(sock_net(sk),
1769 UDP_MIB_INDATAGRAMS, is_udplite);
1771 sock_recv_ts_and_drops(msg, sk, skb);
1773 /* Copy the address. */
1775 sin->sin_family = AF_INET;
1776 sin->sin_port = udp_hdr(skb)->source;
1777 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1778 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1779 *addr_len = sizeof(*sin);
1781 if (cgroup_bpf_enabled)
1782 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1783 (struct sockaddr *)sin);
1786 if (udp_sk(sk)->gro_enabled)
1787 udp_cmsg_recv(msg, sk, skb);
1789 if (inet->cmsg_flags)
1790 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1793 if (flags & MSG_TRUNC)
1796 skb_consume_udp(sk, skb, peeking ? -err : err);
1800 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1801 udp_skb_destructor)) {
1802 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1803 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1807 /* starting over for a new packet, but check if we need to yield */
1809 msg->msg_flags &= ~MSG_TRUNC;
1813 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1815 /* This check is replicated from __ip4_datagram_connect() and
1816 * intended to prevent BPF program called below from accessing bytes
1817 * that are out of the bound specified by user in addr_len.
1819 if (addr_len < sizeof(struct sockaddr_in))
1822 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1824 EXPORT_SYMBOL(udp_pre_connect);
1826 int __udp_disconnect(struct sock *sk, int flags)
1828 struct inet_sock *inet = inet_sk(sk);
1830 * 1003.1g - break association.
1833 sk->sk_state = TCP_CLOSE;
1834 inet->inet_daddr = 0;
1835 inet->inet_dport = 0;
1836 sock_rps_reset_rxhash(sk);
1837 sk->sk_bound_dev_if = 0;
1838 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1839 inet_reset_saddr(sk);
1841 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1842 sk->sk_prot->unhash(sk);
1843 inet->inet_sport = 0;
1848 EXPORT_SYMBOL(__udp_disconnect);
1850 int udp_disconnect(struct sock *sk, int flags)
1853 __udp_disconnect(sk, flags);
1857 EXPORT_SYMBOL(udp_disconnect);
1859 void udp_lib_unhash(struct sock *sk)
1861 if (sk_hashed(sk)) {
1862 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1863 struct udp_hslot *hslot, *hslot2;
1865 hslot = udp_hashslot(udptable, sock_net(sk),
1866 udp_sk(sk)->udp_port_hash);
1867 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1869 spin_lock_bh(&hslot->lock);
1870 if (rcu_access_pointer(sk->sk_reuseport_cb))
1871 reuseport_detach_sock(sk);
1872 if (sk_del_node_init_rcu(sk)) {
1874 inet_sk(sk)->inet_num = 0;
1875 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1877 spin_lock(&hslot2->lock);
1878 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1880 spin_unlock(&hslot2->lock);
1882 spin_unlock_bh(&hslot->lock);
1885 EXPORT_SYMBOL(udp_lib_unhash);
1888 * inet_rcv_saddr was changed, we must rehash secondary hash
1890 void udp_lib_rehash(struct sock *sk, u16 newhash)
1892 if (sk_hashed(sk)) {
1893 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1894 struct udp_hslot *hslot, *hslot2, *nhslot2;
1896 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1897 nhslot2 = udp_hashslot2(udptable, newhash);
1898 udp_sk(sk)->udp_portaddr_hash = newhash;
1900 if (hslot2 != nhslot2 ||
1901 rcu_access_pointer(sk->sk_reuseport_cb)) {
1902 hslot = udp_hashslot(udptable, sock_net(sk),
1903 udp_sk(sk)->udp_port_hash);
1904 /* we must lock primary chain too */
1905 spin_lock_bh(&hslot->lock);
1906 if (rcu_access_pointer(sk->sk_reuseport_cb))
1907 reuseport_detach_sock(sk);
1909 if (hslot2 != nhslot2) {
1910 spin_lock(&hslot2->lock);
1911 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1913 spin_unlock(&hslot2->lock);
1915 spin_lock(&nhslot2->lock);
1916 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1919 spin_unlock(&nhslot2->lock);
1922 spin_unlock_bh(&hslot->lock);
1926 EXPORT_SYMBOL(udp_lib_rehash);
1928 void udp_v4_rehash(struct sock *sk)
1930 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1931 inet_sk(sk)->inet_rcv_saddr,
1932 inet_sk(sk)->inet_num);
1933 udp_lib_rehash(sk, new_hash);
1936 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1940 if (inet_sk(sk)->inet_daddr) {
1941 sock_rps_save_rxhash(sk, skb);
1942 sk_mark_napi_id(sk, skb);
1943 sk_incoming_cpu_update(sk);
1945 sk_mark_napi_id_once(sk, skb);
1948 rc = __udp_enqueue_schedule_skb(sk, skb);
1950 int is_udplite = IS_UDPLITE(sk);
1952 /* Note that an ENOMEM error is charged twice */
1954 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1956 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1958 trace_udp_fail_queue_rcv_skb(rc, sk);
1968 * >0: "udp encap" protocol resubmission
1970 * Note that in the success and error cases, the skb is assumed to
1971 * have either been requeued or freed.
1973 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
1975 struct udp_sock *up = udp_sk(sk);
1976 int is_udplite = IS_UDPLITE(sk);
1979 * Charge it to the socket, dropping if the queue is full.
1981 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1985 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
1986 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1989 * This is an encapsulation socket so pass the skb to
1990 * the socket's udp_encap_rcv() hook. Otherwise, just
1991 * fall through and pass this up the UDP socket.
1992 * up->encap_rcv() returns the following value:
1993 * =0 if skb was successfully passed to the encap
1994 * handler or was discarded by it.
1995 * >0 if skb should be passed on to UDP.
1996 * <0 if skb should be resubmitted as proto -N
1999 /* if we're overly short, let UDP handle it */
2000 encap_rcv = READ_ONCE(up->encap_rcv);
2004 /* Verify checksum before giving to encap */
2005 if (udp_lib_checksum_complete(skb))
2008 ret = encap_rcv(sk, skb);
2010 __UDP_INC_STATS(sock_net(sk),
2011 UDP_MIB_INDATAGRAMS,
2017 /* FALLTHROUGH -- it's a UDP Packet */
2021 * UDP-Lite specific tests, ignored on UDP sockets
2023 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2026 * MIB statistics other than incrementing the error count are
2027 * disabled for the following two types of errors: these depend
2028 * on the application settings, not on the functioning of the
2029 * protocol stack as such.
2031 * RFC 3828 here recommends (sec 3.3): "There should also be a
2032 * way ... to ... at least let the receiving application block
2033 * delivery of packets with coverage values less than a value
2034 * provided by the application."
2036 if (up->pcrlen == 0) { /* full coverage was set */
2037 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2038 UDP_SKB_CB(skb)->cscov, skb->len);
2041 /* The next case involves violating the min. coverage requested
2042 * by the receiver. This is subtle: if receiver wants x and x is
2043 * greater than the buffersize/MTU then receiver will complain
2044 * that it wants x while sender emits packets of smaller size y.
2045 * Therefore the above ...()->partial_cov statement is essential.
2047 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2048 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2049 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2054 prefetch(&sk->sk_rmem_alloc);
2055 if (rcu_access_pointer(sk->sk_filter) &&
2056 udp_lib_checksum_complete(skb))
2059 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2062 udp_csum_pull_header(skb);
2064 ipv4_pktinfo_prepare(sk, skb);
2065 return __udp_queue_rcv_skb(sk, skb);
2068 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2070 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2071 atomic_inc(&sk->sk_drops);
2076 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2078 struct sk_buff *next, *segs;
2081 if (likely(!udp_unexpected_gso(sk, skb)))
2082 return udp_queue_rcv_one_skb(sk, skb);
2084 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET);
2085 __skb_push(skb, -skb_mac_offset(skb));
2086 segs = udp_rcv_segment(sk, skb, true);
2087 for (skb = segs; skb; skb = next) {
2089 __skb_pull(skb, skb_transport_offset(skb));
2090 ret = udp_queue_rcv_one_skb(sk, skb);
2092 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2097 /* For TCP sockets, sk_rx_dst is protected by socket lock
2098 * For UDP, we use xchg() to guard against concurrent changes.
2100 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2102 struct dst_entry *old;
2104 if (dst_hold_safe(dst)) {
2105 old = xchg(&sk->sk_rx_dst, dst);
2111 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2114 * Multicasts and broadcasts go to each listener.
2116 * Note: called only from the BH handler context.
2118 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2120 __be32 saddr, __be32 daddr,
2121 struct udp_table *udptable,
2124 struct sock *sk, *first = NULL;
2125 unsigned short hnum = ntohs(uh->dest);
2126 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2127 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2128 unsigned int offset = offsetof(typeof(*sk), sk_node);
2129 int dif = skb->dev->ifindex;
2130 int sdif = inet_sdif(skb);
2131 struct hlist_node *node;
2132 struct sk_buff *nskb;
2135 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2137 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2139 hslot = &udptable->hash2[hash2];
2140 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2143 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2144 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2145 uh->source, saddr, dif, sdif, hnum))
2152 nskb = skb_clone(skb, GFP_ATOMIC);
2154 if (unlikely(!nskb)) {
2155 atomic_inc(&sk->sk_drops);
2156 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2158 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2162 if (udp_queue_rcv_skb(sk, nskb) > 0)
2166 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2167 if (use_hash2 && hash2 != hash2_any) {
2173 if (udp_queue_rcv_skb(first, skb) > 0)
2177 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2178 proto == IPPROTO_UDPLITE);
2183 /* Initialize UDP checksum. If exited with zero value (success),
2184 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2185 * Otherwise, csum completion requires chacksumming packet body,
2186 * including udp header and folding it to skb->csum.
2188 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2193 UDP_SKB_CB(skb)->partial_cov = 0;
2194 UDP_SKB_CB(skb)->cscov = skb->len;
2196 if (proto == IPPROTO_UDPLITE) {
2197 err = udplite_checksum_init(skb, uh);
2201 if (UDP_SKB_CB(skb)->partial_cov) {
2202 skb->csum = inet_compute_pseudo(skb, proto);
2207 /* Note, we are only interested in != 0 or == 0, thus the
2210 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2211 inet_compute_pseudo);
2215 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2216 /* If SW calculated the value, we know it's bad */
2217 if (skb->csum_complete_sw)
2220 /* HW says the value is bad. Let's validate that.
2221 * skb->csum is no longer the full packet checksum,
2222 * so don't treat it as such.
2224 skb_checksum_complete_unset(skb);
2230 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2231 * return code conversion for ip layer consumption
2233 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2238 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2239 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2240 inet_compute_pseudo);
2242 ret = udp_queue_rcv_skb(sk, skb);
2244 /* a return value > 0 means to resubmit the input, but
2245 * it wants the return to be -protocol, or 0
2253 * All we need to do is get the socket, and then do a checksum.
2256 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2261 unsigned short ulen;
2262 struct rtable *rt = skb_rtable(skb);
2263 __be32 saddr, daddr;
2264 struct net *net = dev_net(skb->dev);
2267 * Validate the packet.
2269 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2270 goto drop; /* No space for header. */
2273 ulen = ntohs(uh->len);
2274 saddr = ip_hdr(skb)->saddr;
2275 daddr = ip_hdr(skb)->daddr;
2277 if (ulen > skb->len)
2280 if (proto == IPPROTO_UDP) {
2281 /* UDP validates ulen. */
2282 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2287 if (udp4_csum_init(skb, uh, proto))
2290 sk = skb_steal_sock(skb);
2292 struct dst_entry *dst = skb_dst(skb);
2295 if (unlikely(sk->sk_rx_dst != dst))
2296 udp_sk_rx_dst_set(sk, dst);
2298 ret = udp_unicast_rcv_skb(sk, skb, uh);
2303 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2304 return __udp4_lib_mcast_deliver(net, skb, uh,
2305 saddr, daddr, udptable, proto);
2307 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2309 return udp_unicast_rcv_skb(sk, skb, uh);
2311 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2315 /* No socket. Drop packet silently, if checksum is wrong */
2316 if (udp_lib_checksum_complete(skb))
2319 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2320 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2323 * Hmm. We got an UDP packet to a port to which we
2324 * don't wanna listen. Ignore it.
2330 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2331 proto == IPPROTO_UDPLITE ? "Lite" : "",
2332 &saddr, ntohs(uh->source),
2334 &daddr, ntohs(uh->dest));
2339 * RFC1122: OK. Discards the bad packet silently (as far as
2340 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2342 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2343 proto == IPPROTO_UDPLITE ? "Lite" : "",
2344 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2346 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2348 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2353 /* We can only early demux multicast if there is a single matching socket.
2354 * If more than one socket found returns NULL
2356 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2357 __be16 loc_port, __be32 loc_addr,
2358 __be16 rmt_port, __be32 rmt_addr,
2361 struct sock *sk, *result;
2362 unsigned short hnum = ntohs(loc_port);
2363 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2364 struct udp_hslot *hslot = &udp_table.hash[slot];
2366 /* Do not bother scanning a too big list */
2367 if (hslot->count > 10)
2371 sk_for_each_rcu(sk, &hslot->head) {
2372 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2373 rmt_port, rmt_addr, dif, sdif, hnum)) {
2383 /* For unicast we should only early demux connected sockets or we can
2384 * break forwarding setups. The chains here can be long so only check
2385 * if the first socket is an exact match and if not move on.
2387 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2388 __be16 loc_port, __be32 loc_addr,
2389 __be16 rmt_port, __be32 rmt_addr,
2392 unsigned short hnum = ntohs(loc_port);
2393 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2394 unsigned int slot2 = hash2 & udp_table.mask;
2395 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2396 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2397 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2400 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2401 if (INET_MATCH(sk, net, acookie, rmt_addr,
2402 loc_addr, ports, dif, sdif))
2404 /* Only check first socket in chain */
2410 int udp_v4_early_demux(struct sk_buff *skb)
2412 struct net *net = dev_net(skb->dev);
2413 struct in_device *in_dev = NULL;
2414 const struct iphdr *iph;
2415 const struct udphdr *uh;
2416 struct sock *sk = NULL;
2417 struct dst_entry *dst;
2418 int dif = skb->dev->ifindex;
2419 int sdif = inet_sdif(skb);
2422 /* validate the packet */
2423 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2429 if (skb->pkt_type == PACKET_MULTICAST) {
2430 in_dev = __in_dev_get_rcu(skb->dev);
2435 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2440 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2441 uh->source, iph->saddr,
2443 } else if (skb->pkt_type == PACKET_HOST) {
2444 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2445 uh->source, iph->saddr, dif, sdif);
2448 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2452 skb->destructor = sock_efree;
2453 dst = READ_ONCE(sk->sk_rx_dst);
2456 dst = dst_check(dst, 0);
2460 /* set noref for now.
2461 * any place which wants to hold dst has to call
2464 skb_dst_set_noref(skb, dst);
2466 /* for unconnected multicast sockets we need to validate
2467 * the source on each packet
2469 if (!inet_sk(sk)->inet_daddr && in_dev)
2470 return ip_mc_validate_source(skb, iph->daddr,
2471 iph->saddr, iph->tos,
2472 skb->dev, in_dev, &itag);
2477 int udp_rcv(struct sk_buff *skb)
2479 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2482 void udp_destroy_sock(struct sock *sk)
2484 struct udp_sock *up = udp_sk(sk);
2485 bool slow = lock_sock_fast(sk);
2486 udp_flush_pending_frames(sk);
2487 unlock_sock_fast(sk, slow);
2488 if (static_branch_unlikely(&udp_encap_needed_key)) {
2489 if (up->encap_type) {
2490 void (*encap_destroy)(struct sock *sk);
2491 encap_destroy = READ_ONCE(up->encap_destroy);
2495 if (up->encap_enabled)
2496 static_branch_dec(&udp_encap_needed_key);
2501 * Socket option code for UDP
2503 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2504 char __user *optval, unsigned int optlen,
2505 int (*push_pending_frames)(struct sock *))
2507 struct udp_sock *up = udp_sk(sk);
2510 int is_udplite = IS_UDPLITE(sk);
2512 if (optlen < sizeof(int))
2515 if (get_user(val, (int __user *)optval))
2518 valbool = val ? 1 : 0;
2527 push_pending_frames(sk);
2535 case UDP_ENCAP_ESPINUDP:
2536 case UDP_ENCAP_ESPINUDP_NON_IKE:
2537 up->encap_rcv = xfrm4_udp_encap_rcv;
2539 case UDP_ENCAP_L2TPINUDP:
2540 up->encap_type = val;
2542 udp_tunnel_encap_enable(sk->sk_socket);
2551 case UDP_NO_CHECK6_TX:
2552 up->no_check6_tx = valbool;
2555 case UDP_NO_CHECK6_RX:
2556 up->no_check6_rx = valbool;
2560 if (val < 0 || val > USHRT_MAX)
2568 udp_tunnel_encap_enable(sk->sk_socket);
2569 up->gro_enabled = valbool;
2574 * UDP-Lite's partial checksum coverage (RFC 3828).
2576 /* The sender sets actual checksum coverage length via this option.
2577 * The case coverage > packet length is handled by send module. */
2578 case UDPLITE_SEND_CSCOV:
2579 if (!is_udplite) /* Disable the option on UDP sockets */
2580 return -ENOPROTOOPT;
2581 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2583 else if (val > USHRT_MAX)
2586 up->pcflag |= UDPLITE_SEND_CC;
2589 /* The receiver specifies a minimum checksum coverage value. To make
2590 * sense, this should be set to at least 8 (as done below). If zero is
2591 * used, this again means full checksum coverage. */
2592 case UDPLITE_RECV_CSCOV:
2593 if (!is_udplite) /* Disable the option on UDP sockets */
2594 return -ENOPROTOOPT;
2595 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2597 else if (val > USHRT_MAX)
2600 up->pcflag |= UDPLITE_RECV_CC;
2610 EXPORT_SYMBOL(udp_lib_setsockopt);
2612 int udp_setsockopt(struct sock *sk, int level, int optname,
2613 char __user *optval, unsigned int optlen)
2615 if (level == SOL_UDP || level == SOL_UDPLITE)
2616 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2617 udp_push_pending_frames);
2618 return ip_setsockopt(sk, level, optname, optval, optlen);
2621 #ifdef CONFIG_COMPAT
2622 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2623 char __user *optval, unsigned int optlen)
2625 if (level == SOL_UDP || level == SOL_UDPLITE)
2626 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2627 udp_push_pending_frames);
2628 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2632 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2633 char __user *optval, int __user *optlen)
2635 struct udp_sock *up = udp_sk(sk);
2638 if (get_user(len, optlen))
2641 len = min_t(unsigned int, len, sizeof(int));
2652 val = up->encap_type;
2655 case UDP_NO_CHECK6_TX:
2656 val = up->no_check6_tx;
2659 case UDP_NO_CHECK6_RX:
2660 val = up->no_check6_rx;
2667 /* The following two cannot be changed on UDP sockets, the return is
2668 * always 0 (which corresponds to the full checksum coverage of UDP). */
2669 case UDPLITE_SEND_CSCOV:
2673 case UDPLITE_RECV_CSCOV:
2678 return -ENOPROTOOPT;
2681 if (put_user(len, optlen))
2683 if (copy_to_user(optval, &val, len))
2687 EXPORT_SYMBOL(udp_lib_getsockopt);
2689 int udp_getsockopt(struct sock *sk, int level, int optname,
2690 char __user *optval, int __user *optlen)
2692 if (level == SOL_UDP || level == SOL_UDPLITE)
2693 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2694 return ip_getsockopt(sk, level, optname, optval, optlen);
2697 #ifdef CONFIG_COMPAT
2698 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2699 char __user *optval, int __user *optlen)
2701 if (level == SOL_UDP || level == SOL_UDPLITE)
2702 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2703 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2707 * udp_poll - wait for a UDP event.
2708 * @file - file struct
2710 * @wait - poll table
2712 * This is same as datagram poll, except for the special case of
2713 * blocking sockets. If application is using a blocking fd
2714 * and a packet with checksum error is in the queue;
2715 * then it could get return from select indicating data available
2716 * but then block when reading it. Add special case code
2717 * to work around these arguably broken applications.
2719 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2721 __poll_t mask = datagram_poll(file, sock, wait);
2722 struct sock *sk = sock->sk;
2724 if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2725 mask |= EPOLLIN | EPOLLRDNORM;
2727 /* Check for false positives due to checksum errors */
2728 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2729 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2730 mask &= ~(EPOLLIN | EPOLLRDNORM);
2735 EXPORT_SYMBOL(udp_poll);
2737 int udp_abort(struct sock *sk, int err)
2742 sk->sk_error_report(sk);
2743 __udp_disconnect(sk, 0);
2749 EXPORT_SYMBOL_GPL(udp_abort);
2751 struct proto udp_prot = {
2753 .owner = THIS_MODULE,
2754 .close = udp_lib_close,
2755 .pre_connect = udp_pre_connect,
2756 .connect = ip4_datagram_connect,
2757 .disconnect = udp_disconnect,
2759 .init = udp_init_sock,
2760 .destroy = udp_destroy_sock,
2761 .setsockopt = udp_setsockopt,
2762 .getsockopt = udp_getsockopt,
2763 .sendmsg = udp_sendmsg,
2764 .recvmsg = udp_recvmsg,
2765 .sendpage = udp_sendpage,
2766 .release_cb = ip4_datagram_release_cb,
2767 .hash = udp_lib_hash,
2768 .unhash = udp_lib_unhash,
2769 .rehash = udp_v4_rehash,
2770 .get_port = udp_v4_get_port,
2771 .memory_allocated = &udp_memory_allocated,
2772 .sysctl_mem = sysctl_udp_mem,
2773 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2774 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2775 .obj_size = sizeof(struct udp_sock),
2776 .h.udp_table = &udp_table,
2777 #ifdef CONFIG_COMPAT
2778 .compat_setsockopt = compat_udp_setsockopt,
2779 .compat_getsockopt = compat_udp_getsockopt,
2781 .diag_destroy = udp_abort,
2783 EXPORT_SYMBOL(udp_prot);
2785 /* ------------------------------------------------------------------------ */
2786 #ifdef CONFIG_PROC_FS
2788 static struct sock *udp_get_first(struct seq_file *seq, int start)
2791 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2792 struct udp_iter_state *state = seq->private;
2793 struct net *net = seq_file_net(seq);
2795 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2797 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2799 if (hlist_empty(&hslot->head))
2802 spin_lock_bh(&hslot->lock);
2803 sk_for_each(sk, &hslot->head) {
2804 if (!net_eq(sock_net(sk), net))
2806 if (sk->sk_family == afinfo->family)
2809 spin_unlock_bh(&hslot->lock);
2816 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2818 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2819 struct udp_iter_state *state = seq->private;
2820 struct net *net = seq_file_net(seq);
2824 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2827 if (state->bucket <= afinfo->udp_table->mask)
2828 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2829 return udp_get_first(seq, state->bucket + 1);
2834 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2836 struct sock *sk = udp_get_first(seq, 0);
2839 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2841 return pos ? NULL : sk;
2844 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2846 struct udp_iter_state *state = seq->private;
2847 state->bucket = MAX_UDP_PORTS;
2849 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2851 EXPORT_SYMBOL(udp_seq_start);
2853 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2857 if (v == SEQ_START_TOKEN)
2858 sk = udp_get_idx(seq, 0);
2860 sk = udp_get_next(seq, v);
2865 EXPORT_SYMBOL(udp_seq_next);
2867 void udp_seq_stop(struct seq_file *seq, void *v)
2869 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2870 struct udp_iter_state *state = seq->private;
2872 if (state->bucket <= afinfo->udp_table->mask)
2873 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2875 EXPORT_SYMBOL(udp_seq_stop);
2877 /* ------------------------------------------------------------------------ */
2878 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2881 struct inet_sock *inet = inet_sk(sp);
2882 __be32 dest = inet->inet_daddr;
2883 __be32 src = inet->inet_rcv_saddr;
2884 __u16 destp = ntohs(inet->inet_dport);
2885 __u16 srcp = ntohs(inet->inet_sport);
2887 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2888 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2889 bucket, src, srcp, dest, destp, sp->sk_state,
2890 sk_wmem_alloc_get(sp),
2893 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2895 refcount_read(&sp->sk_refcnt), sp,
2896 atomic_read(&sp->sk_drops));
2899 int udp4_seq_show(struct seq_file *seq, void *v)
2901 seq_setwidth(seq, 127);
2902 if (v == SEQ_START_TOKEN)
2903 seq_puts(seq, " sl local_address rem_address st tx_queue "
2904 "rx_queue tr tm->when retrnsmt uid timeout "
2905 "inode ref pointer drops");
2907 struct udp_iter_state *state = seq->private;
2909 udp4_format_sock(v, seq, state->bucket);
2915 const struct seq_operations udp_seq_ops = {
2916 .start = udp_seq_start,
2917 .next = udp_seq_next,
2918 .stop = udp_seq_stop,
2919 .show = udp4_seq_show,
2921 EXPORT_SYMBOL(udp_seq_ops);
2923 static struct udp_seq_afinfo udp4_seq_afinfo = {
2925 .udp_table = &udp_table,
2928 static int __net_init udp4_proc_init_net(struct net *net)
2930 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2931 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2936 static void __net_exit udp4_proc_exit_net(struct net *net)
2938 remove_proc_entry("udp", net->proc_net);
2941 static struct pernet_operations udp4_net_ops = {
2942 .init = udp4_proc_init_net,
2943 .exit = udp4_proc_exit_net,
2946 int __init udp4_proc_init(void)
2948 return register_pernet_subsys(&udp4_net_ops);
2951 void udp4_proc_exit(void)
2953 unregister_pernet_subsys(&udp4_net_ops);
2955 #endif /* CONFIG_PROC_FS */
2957 static __initdata unsigned long uhash_entries;
2958 static int __init set_uhash_entries(char *str)
2965 ret = kstrtoul(str, 0, &uhash_entries);
2969 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2970 uhash_entries = UDP_HTABLE_SIZE_MIN;
2973 __setup("uhash_entries=", set_uhash_entries);
2975 void __init udp_table_init(struct udp_table *table, const char *name)
2979 table->hash = alloc_large_system_hash(name,
2980 2 * sizeof(struct udp_hslot),
2982 21, /* one slot per 2 MB */
2986 UDP_HTABLE_SIZE_MIN,
2989 table->hash2 = table->hash + (table->mask + 1);
2990 for (i = 0; i <= table->mask; i++) {
2991 INIT_HLIST_HEAD(&table->hash[i].head);
2992 table->hash[i].count = 0;
2993 spin_lock_init(&table->hash[i].lock);
2995 for (i = 0; i <= table->mask; i++) {
2996 INIT_HLIST_HEAD(&table->hash2[i].head);
2997 table->hash2[i].count = 0;
2998 spin_lock_init(&table->hash2[i].lock);
3002 u32 udp_flow_hashrnd(void)
3004 static u32 hashrnd __read_mostly;
3006 net_get_random_once(&hashrnd, sizeof(hashrnd));
3010 EXPORT_SYMBOL(udp_flow_hashrnd);
3012 static void __udp_sysctl_init(struct net *net)
3014 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3015 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3017 #ifdef CONFIG_NET_L3_MASTER_DEV
3018 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3022 static int __net_init udp_sysctl_init(struct net *net)
3024 __udp_sysctl_init(net);
3028 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3029 .init = udp_sysctl_init,
3032 void __init udp_init(void)
3034 unsigned long limit;
3037 udp_table_init(&udp_table, "UDP");
3038 limit = nr_free_buffer_pages() / 8;
3039 limit = max(limit, 128UL);
3040 sysctl_udp_mem[0] = limit / 4 * 3;
3041 sysctl_udp_mem[1] = limit;
3042 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3044 __udp_sysctl_init(&init_net);
3046 /* 16 spinlocks per cpu */
3047 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3048 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3051 panic("UDP: failed to alloc udp_busylocks\n");
3052 for (i = 0; i < (1U << udp_busylocks_log); i++)
3053 spin_lock_init(udp_busylocks + i);
3055 if (register_pernet_subsys(&udp_sysctl_ops))
3056 panic("UDP: failed to init sysctl parameters.\n");