1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * The User Datagram Protocol (UDP).
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
72 * James Chapman : Add L2TP encapsulation type.
75 #define pr_fmt(fmt) "UDP: " fmt
77 #include <linux/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <linux/memblock.h>
80 #include <linux/highmem.h>
81 #include <linux/swap.h>
82 #include <linux/types.h>
83 #include <linux/fcntl.h>
84 #include <linux/module.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/igmp.h>
88 #include <linux/inetdevice.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
93 #include <linux/inet.h>
94 #include <linux/netdevice.h>
95 #include <linux/slab.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/inet_hashtables.h>
103 #include <net/ip_tunnels.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <trace/events/udp.h>
108 #include <linux/static_key.h>
109 #include <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 static int udp_lib_lport_inuse(struct net *net, __u16 num,
129 const struct udp_hslot *hslot,
130 unsigned long *bitmap,
131 struct sock *sk, unsigned int log)
134 kuid_t uid = sock_i_uid(sk);
136 sk_for_each(sk2, &hslot->head) {
137 if (net_eq(sock_net(sk2), net) &&
139 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
140 (!sk2->sk_reuse || !sk->sk_reuse) &&
141 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
142 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
143 inet_rcv_saddr_equal(sk, sk2, true)) {
144 if (sk2->sk_reuseport && sk->sk_reuseport &&
145 !rcu_access_pointer(sk->sk_reuseport_cb) &&
146 uid_eq(uid, sock_i_uid(sk2))) {
152 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
161 * Note: we still hold spinlock of primary hash chain, so no other writer
162 * can insert/delete a socket with local_port == num
164 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
165 struct udp_hslot *hslot2,
169 kuid_t uid = sock_i_uid(sk);
172 spin_lock(&hslot2->lock);
173 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
174 if (net_eq(sock_net(sk2), net) &&
176 (udp_sk(sk2)->udp_port_hash == num) &&
177 (!sk2->sk_reuse || !sk->sk_reuse) &&
178 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
179 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
180 inet_rcv_saddr_equal(sk, sk2, true)) {
181 if (sk2->sk_reuseport && sk->sk_reuseport &&
182 !rcu_access_pointer(sk->sk_reuseport_cb) &&
183 uid_eq(uid, sock_i_uid(sk2))) {
191 spin_unlock(&hslot2->lock);
195 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
197 struct net *net = sock_net(sk);
198 kuid_t uid = sock_i_uid(sk);
201 sk_for_each(sk2, &hslot->head) {
202 if (net_eq(sock_net(sk2), net) &&
204 sk2->sk_family == sk->sk_family &&
205 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
206 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
207 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
208 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
209 inet_rcv_saddr_equal(sk, sk2, false)) {
210 return reuseport_add_sock(sk, sk2,
211 inet_rcv_saddr_any(sk));
215 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
219 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
221 * @sk: socket struct in question
222 * @snum: port number to look up
223 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
226 int udp_lib_get_port(struct sock *sk, unsigned short snum,
227 unsigned int hash2_nulladdr)
229 struct udp_hslot *hslot, *hslot2;
230 struct udp_table *udptable = sk->sk_prot->h.udp_table;
232 struct net *net = sock_net(sk);
235 int low, high, remaining;
237 unsigned short first, last;
238 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
240 inet_get_local_port_range(net, &low, &high);
241 remaining = (high - low) + 1;
243 rand = prandom_u32();
244 first = reciprocal_scale(rand, remaining) + low;
246 * force rand to be an odd multiple of UDP_HTABLE_SIZE
248 rand = (rand | 1) * (udptable->mask + 1);
249 last = first + udptable->mask + 1;
251 hslot = udp_hashslot(udptable, net, first);
252 bitmap_zero(bitmap, PORTS_PER_CHAIN);
253 spin_lock_bh(&hslot->lock);
254 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
259 * Iterate on all possible values of snum for this hash.
260 * Using steps of an odd multiple of UDP_HTABLE_SIZE
261 * give us randomization and full range coverage.
264 if (low <= snum && snum <= high &&
265 !test_bit(snum >> udptable->log, bitmap) &&
266 !inet_is_local_reserved_port(net, snum))
269 } while (snum != first);
270 spin_unlock_bh(&hslot->lock);
272 } while (++first != last);
275 hslot = udp_hashslot(udptable, net, snum);
276 spin_lock_bh(&hslot->lock);
277 if (hslot->count > 10) {
279 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
281 slot2 &= udptable->mask;
282 hash2_nulladdr &= udptable->mask;
284 hslot2 = udp_hashslot2(udptable, slot2);
285 if (hslot->count < hslot2->count)
286 goto scan_primary_hash;
288 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
289 if (!exist && (hash2_nulladdr != slot2)) {
290 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
291 exist = udp_lib_lport_inuse2(net, snum, hslot2,
300 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
304 inet_sk(sk)->inet_num = snum;
305 udp_sk(sk)->udp_port_hash = snum;
306 udp_sk(sk)->udp_portaddr_hash ^= snum;
307 if (sk_unhashed(sk)) {
308 if (sk->sk_reuseport &&
309 udp_reuseport_add_sock(sk, hslot)) {
310 inet_sk(sk)->inet_num = 0;
311 udp_sk(sk)->udp_port_hash = 0;
312 udp_sk(sk)->udp_portaddr_hash ^= snum;
316 sk_add_node_rcu(sk, &hslot->head);
318 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
320 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
321 spin_lock(&hslot2->lock);
322 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
323 sk->sk_family == AF_INET6)
324 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
327 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
330 spin_unlock(&hslot2->lock);
332 sock_set_flag(sk, SOCK_RCU_FREE);
335 spin_unlock_bh(&hslot->lock);
339 EXPORT_SYMBOL(udp_lib_get_port);
341 int udp_v4_get_port(struct sock *sk, unsigned short snum)
343 unsigned int hash2_nulladdr =
344 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
345 unsigned int hash2_partial =
346 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
348 /* precompute partial secondary hash */
349 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
350 return udp_lib_get_port(sk, snum, hash2_nulladdr);
353 static int compute_score(struct sock *sk, struct net *net,
354 __be32 saddr, __be16 sport,
355 __be32 daddr, unsigned short hnum,
359 struct inet_sock *inet;
362 if (!net_eq(sock_net(sk), net) ||
363 udp_sk(sk)->udp_port_hash != hnum ||
367 if (sk->sk_rcv_saddr != daddr)
370 score = (sk->sk_family == PF_INET) ? 2 : 1;
373 if (inet->inet_daddr) {
374 if (inet->inet_daddr != saddr)
379 if (inet->inet_dport) {
380 if (inet->inet_dport != sport)
385 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
391 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
396 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
397 const __u16 lport, const __be32 faddr,
400 static u32 udp_ehash_secret __read_mostly;
402 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
404 return __inet_ehashfn(laddr, lport, faddr, fport,
405 udp_ehash_secret + net_hash_mix(net));
408 /* called with rcu_read_lock() */
409 static struct sock *udp4_lib_lookup2(struct net *net,
410 __be32 saddr, __be16 sport,
411 __be32 daddr, unsigned int hnum,
413 struct udp_hslot *hslot2,
416 struct sock *sk, *result;
422 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
423 score = compute_score(sk, net, saddr, sport,
424 daddr, hnum, dif, sdif);
425 if (score > badness) {
426 if (sk->sk_reuseport &&
427 sk->sk_state != TCP_ESTABLISHED) {
428 hash = udp_ehashfn(net, daddr, hnum,
430 result = reuseport_select_sock(sk, hash, skb,
431 sizeof(struct udphdr));
432 if (result && !reuseport_has_conns(sk, false))
442 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
443 * harder than this. -DaveM
445 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
446 __be16 sport, __be32 daddr, __be16 dport, int dif,
447 int sdif, struct udp_table *udptable, struct sk_buff *skb)
450 unsigned short hnum = ntohs(dport);
451 unsigned int hash2, slot2;
452 struct udp_hslot *hslot2;
454 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
455 slot2 = hash2 & udptable->mask;
456 hslot2 = &udptable->hash2[slot2];
458 result = udp4_lib_lookup2(net, saddr, sport,
459 daddr, hnum, dif, sdif,
462 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
463 slot2 = hash2 & udptable->mask;
464 hslot2 = &udptable->hash2[slot2];
466 result = udp4_lib_lookup2(net, saddr, sport,
467 htonl(INADDR_ANY), hnum, dif, sdif,
474 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
476 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
477 __be16 sport, __be16 dport,
478 struct udp_table *udptable)
480 const struct iphdr *iph = ip_hdr(skb);
482 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
483 iph->daddr, dport, inet_iif(skb),
484 inet_sdif(skb), udptable, skb);
487 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
488 __be16 sport, __be16 dport)
490 const struct iphdr *iph = ip_hdr(skb);
492 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
493 iph->daddr, dport, inet_iif(skb),
494 inet_sdif(skb), &udp_table, NULL);
496 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
498 /* Must be called under rcu_read_lock().
499 * Does increment socket refcount.
501 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
502 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
503 __be32 daddr, __be16 dport, int dif)
507 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
508 dif, 0, &udp_table, NULL);
509 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
513 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
516 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
517 __be16 loc_port, __be32 loc_addr,
518 __be16 rmt_port, __be32 rmt_addr,
519 int dif, int sdif, unsigned short hnum)
521 struct inet_sock *inet = inet_sk(sk);
523 if (!net_eq(sock_net(sk), net) ||
524 udp_sk(sk)->udp_port_hash != hnum ||
525 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
526 (inet->inet_dport != rmt_port && inet->inet_dport) ||
527 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
528 ipv6_only_sock(sk) ||
529 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
531 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
536 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
537 void udp_encap_enable(void)
539 static_branch_inc(&udp_encap_needed_key);
541 EXPORT_SYMBOL(udp_encap_enable);
543 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
544 * through error handlers in encapsulations looking for a match.
546 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
550 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
551 int (*handler)(struct sk_buff *skb, u32 info);
552 const struct ip_tunnel_encap_ops *encap;
554 encap = rcu_dereference(iptun_encaps[i]);
557 handler = encap->err_handler;
558 if (handler && !handler(skb, info))
565 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
566 * reversing source and destination port: this will match tunnels that force the
567 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
568 * lwtunnels might actually break this assumption by being configured with
569 * different destination ports on endpoints, in this case we won't be able to
570 * trace ICMP messages back to them.
572 * If this doesn't match any socket, probe tunnels with arbitrary destination
573 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
574 * we've sent packets to won't necessarily match the local destination port.
576 * Then ask the tunnel implementation to match the error against a valid
579 * Return an error if we can't find a match, the socket if we need further
580 * processing, zero otherwise.
582 static struct sock *__udp4_lib_err_encap(struct net *net,
583 const struct iphdr *iph,
585 struct udp_table *udptable,
586 struct sk_buff *skb, u32 info)
588 int network_offset, transport_offset;
591 network_offset = skb_network_offset(skb);
592 transport_offset = skb_transport_offset(skb);
594 /* Network header needs to point to the outer IPv4 header inside ICMP */
595 skb_reset_network_header(skb);
597 /* Transport header needs to point to the UDP header */
598 skb_set_transport_header(skb, iph->ihl << 2);
600 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
601 iph->saddr, uh->dest, skb->dev->ifindex, 0,
604 int (*lookup)(struct sock *sk, struct sk_buff *skb);
605 struct udp_sock *up = udp_sk(sk);
607 lookup = READ_ONCE(up->encap_err_lookup);
608 if (!lookup || lookup(sk, skb))
613 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
615 skb_set_transport_header(skb, transport_offset);
616 skb_set_network_header(skb, network_offset);
622 * This routine is called by the ICMP module when it gets some
623 * sort of error condition. If err < 0 then the socket should
624 * be closed and the error returned to the user. If err > 0
625 * it's just the icmp type << 8 | icmp code.
626 * Header points to the ip header of the error packet. We move
627 * on past this. Then (as it used to claim before adjustment)
628 * header points to the first 8 bytes of the udp header. We need
629 * to find the appropriate port.
632 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
634 struct inet_sock *inet;
635 const struct iphdr *iph = (const struct iphdr *)skb->data;
636 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
637 const int type = icmp_hdr(skb)->type;
638 const int code = icmp_hdr(skb)->code;
643 struct net *net = dev_net(skb->dev);
645 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
646 iph->saddr, uh->source, skb->dev->ifindex,
647 inet_sdif(skb), udptable, NULL);
649 /* No socket for error: try tunnels before discarding */
650 sk = ERR_PTR(-ENOENT);
651 if (static_branch_unlikely(&udp_encap_needed_key)) {
652 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
659 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
672 case ICMP_TIME_EXCEEDED:
675 case ICMP_SOURCE_QUENCH:
677 case ICMP_PARAMETERPROB:
681 case ICMP_DEST_UNREACH:
682 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
683 ipv4_sk_update_pmtu(skb, sk, info);
684 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
692 if (code <= NR_ICMP_UNREACH) {
693 harderr = icmp_err_convert[code].fatal;
694 err = icmp_err_convert[code].errno;
698 ipv4_sk_redirect(skb, sk);
703 * RFC1122: OK. Passes ICMP errors back to application, as per
707 /* ...not for tunnels though: we don't have a sending socket */
710 if (!inet->recverr) {
711 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
714 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
717 sk->sk_error_report(sk);
722 int udp_err(struct sk_buff *skb, u32 info)
724 return __udp4_lib_err(skb, info, &udp_table);
728 * Throw away all pending data and cancel the corking. Socket is locked.
730 void udp_flush_pending_frames(struct sock *sk)
732 struct udp_sock *up = udp_sk(sk);
737 ip_flush_pending_frames(sk);
740 EXPORT_SYMBOL(udp_flush_pending_frames);
743 * udp4_hwcsum - handle outgoing HW checksumming
744 * @skb: sk_buff containing the filled-in UDP header
745 * (checksum field must be zeroed out)
746 * @src: source IP address
747 * @dst: destination IP address
749 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
751 struct udphdr *uh = udp_hdr(skb);
752 int offset = skb_transport_offset(skb);
753 int len = skb->len - offset;
757 if (!skb_has_frag_list(skb)) {
759 * Only one fragment on the socket.
761 skb->csum_start = skb_transport_header(skb) - skb->head;
762 skb->csum_offset = offsetof(struct udphdr, check);
763 uh->check = ~csum_tcpudp_magic(src, dst, len,
766 struct sk_buff *frags;
769 * HW-checksum won't work as there are two or more
770 * fragments on the socket so that all csums of sk_buffs
773 skb_walk_frags(skb, frags) {
774 csum = csum_add(csum, frags->csum);
778 csum = skb_checksum(skb, offset, hlen, csum);
779 skb->ip_summed = CHECKSUM_NONE;
781 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
783 uh->check = CSUM_MANGLED_0;
786 EXPORT_SYMBOL_GPL(udp4_hwcsum);
788 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
789 * for the simple case like when setting the checksum for a UDP tunnel.
791 void udp_set_csum(bool nocheck, struct sk_buff *skb,
792 __be32 saddr, __be32 daddr, int len)
794 struct udphdr *uh = udp_hdr(skb);
798 } else if (skb_is_gso(skb)) {
799 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
800 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
802 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
804 uh->check = CSUM_MANGLED_0;
806 skb->ip_summed = CHECKSUM_PARTIAL;
807 skb->csum_start = skb_transport_header(skb) - skb->head;
808 skb->csum_offset = offsetof(struct udphdr, check);
809 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
812 EXPORT_SYMBOL(udp_set_csum);
814 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
815 struct inet_cork *cork)
817 struct sock *sk = skb->sk;
818 struct inet_sock *inet = inet_sk(sk);
821 int is_udplite = IS_UDPLITE(sk);
822 int offset = skb_transport_offset(skb);
823 int len = skb->len - offset;
824 int datalen = len - sizeof(*uh);
828 * Create a UDP header
831 uh->source = inet->inet_sport;
832 uh->dest = fl4->fl4_dport;
833 uh->len = htons(len);
836 if (cork->gso_size) {
837 const int hlen = skb_network_header_len(skb) +
838 sizeof(struct udphdr);
840 if (hlen + cork->gso_size > cork->fragsize) {
844 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
848 if (sk->sk_no_check_tx) {
852 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
853 dst_xfrm(skb_dst(skb))) {
858 if (datalen > cork->gso_size) {
859 skb_shinfo(skb)->gso_size = cork->gso_size;
860 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
861 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
867 if (is_udplite) /* UDP-Lite */
868 csum = udplite_csum(skb);
870 else if (sk->sk_no_check_tx) { /* UDP csum off */
872 skb->ip_summed = CHECKSUM_NONE;
875 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
878 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
882 csum = udp_csum(skb);
884 /* add protocol-dependent pseudo-header */
885 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
886 sk->sk_protocol, csum);
888 uh->check = CSUM_MANGLED_0;
891 err = ip_send_skb(sock_net(sk), skb);
893 if (err == -ENOBUFS && !inet->recverr) {
894 UDP_INC_STATS(sock_net(sk),
895 UDP_MIB_SNDBUFERRORS, is_udplite);
899 UDP_INC_STATS(sock_net(sk),
900 UDP_MIB_OUTDATAGRAMS, is_udplite);
905 * Push out all pending data as one UDP datagram. Socket is locked.
907 int udp_push_pending_frames(struct sock *sk)
909 struct udp_sock *up = udp_sk(sk);
910 struct inet_sock *inet = inet_sk(sk);
911 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
915 skb = ip_finish_skb(sk, fl4);
919 err = udp_send_skb(skb, fl4, &inet->cork.base);
926 EXPORT_SYMBOL(udp_push_pending_frames);
928 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
930 switch (cmsg->cmsg_type) {
932 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
934 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
941 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
943 struct cmsghdr *cmsg;
944 bool need_ip = false;
947 for_each_cmsghdr(cmsg, msg) {
948 if (!CMSG_OK(msg, cmsg))
951 if (cmsg->cmsg_level != SOL_UDP) {
956 err = __udp_cmsg_send(cmsg, gso_size);
963 EXPORT_SYMBOL_GPL(udp_cmsg_send);
965 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
967 struct inet_sock *inet = inet_sk(sk);
968 struct udp_sock *up = udp_sk(sk);
969 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
970 struct flowi4 fl4_stack;
973 struct ipcm_cookie ipc;
974 struct rtable *rt = NULL;
977 __be32 daddr, faddr, saddr;
980 int err, is_udplite = IS_UDPLITE(sk);
981 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
982 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
984 struct ip_options_data opt_copy;
993 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
996 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
998 fl4 = &inet->cork.fl.u.ip4;
1001 * There are pending frames.
1002 * The socket lock must be held while it's corked.
1005 if (likely(up->pending)) {
1006 if (unlikely(up->pending != AF_INET)) {
1010 goto do_append_data;
1014 ulen += sizeof(struct udphdr);
1017 * Get and verify the address.
1020 if (msg->msg_namelen < sizeof(*usin))
1022 if (usin->sin_family != AF_INET) {
1023 if (usin->sin_family != AF_UNSPEC)
1024 return -EAFNOSUPPORT;
1027 daddr = usin->sin_addr.s_addr;
1028 dport = usin->sin_port;
1032 if (sk->sk_state != TCP_ESTABLISHED)
1033 return -EDESTADDRREQ;
1034 daddr = inet->inet_daddr;
1035 dport = inet->inet_dport;
1036 /* Open fast path for connected socket.
1037 Route will not be used, if at least one option is set.
1042 ipcm_init_sk(&ipc, inet);
1043 ipc.gso_size = up->gso_size;
1045 if (msg->msg_controllen) {
1046 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1048 err = ip_cmsg_send(sk, msg, &ipc,
1049 sk->sk_family == AF_INET6);
1050 if (unlikely(err < 0)) {
1059 struct ip_options_rcu *inet_opt;
1062 inet_opt = rcu_dereference(inet->inet_opt);
1064 memcpy(&opt_copy, inet_opt,
1065 sizeof(*inet_opt) + inet_opt->opt.optlen);
1066 ipc.opt = &opt_copy.opt;
1071 if (cgroup_bpf_enabled && !connected) {
1072 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1073 (struct sockaddr *)usin, &ipc.addr);
1077 if (usin->sin_port == 0) {
1078 /* BPF program set invalid port. Reject it. */
1082 daddr = usin->sin_addr.s_addr;
1083 dport = usin->sin_port;
1088 ipc.addr = faddr = daddr;
1090 if (ipc.opt && ipc.opt->opt.srr) {
1095 faddr = ipc.opt->opt.faddr;
1098 tos = get_rttos(&ipc, inet);
1099 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1100 (msg->msg_flags & MSG_DONTROUTE) ||
1101 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1106 if (ipv4_is_multicast(daddr)) {
1107 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1108 ipc.oif = inet->mc_index;
1110 saddr = inet->mc_addr;
1112 } else if (!ipc.oif) {
1113 ipc.oif = inet->uc_index;
1114 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1115 /* oif is set, packet is to local broadcast and
1116 * and uc_index is set. oif is most likely set
1117 * by sk_bound_dev_if. If uc_index != oif check if the
1118 * oif is an L3 master and uc_index is an L3 slave.
1119 * If so, we want to allow the send using the uc_index.
1121 if (ipc.oif != inet->uc_index &&
1122 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1124 ipc.oif = inet->uc_index;
1129 rt = (struct rtable *)sk_dst_check(sk, 0);
1132 struct net *net = sock_net(sk);
1133 __u8 flow_flags = inet_sk_flowi_flags(sk);
1137 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1138 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1140 faddr, saddr, dport, inet->inet_sport,
1143 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1144 rt = ip_route_output_flow(net, fl4, sk);
1148 if (err == -ENETUNREACH)
1149 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1154 if ((rt->rt_flags & RTCF_BROADCAST) &&
1155 !sock_flag(sk, SOCK_BROADCAST))
1158 sk_dst_set(sk, dst_clone(&rt->dst));
1161 if (msg->msg_flags&MSG_CONFIRM)
1167 daddr = ipc.addr = fl4->daddr;
1169 /* Lockless fast path for the non-corking case. */
1171 struct inet_cork cork;
1173 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1174 sizeof(struct udphdr), &ipc, &rt,
1175 &cork, msg->msg_flags);
1177 if (!IS_ERR_OR_NULL(skb))
1178 err = udp_send_skb(skb, fl4, &cork);
1183 if (unlikely(up->pending)) {
1184 /* The socket is already corked while preparing it. */
1185 /* ... which is an evident application bug. --ANK */
1188 net_dbg_ratelimited("socket already corked\n");
1193 * Now cork the socket to pend data.
1195 fl4 = &inet->cork.fl.u.ip4;
1198 fl4->fl4_dport = dport;
1199 fl4->fl4_sport = inet->inet_sport;
1200 up->pending = AF_INET;
1204 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1205 sizeof(struct udphdr), &ipc, &rt,
1206 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1208 udp_flush_pending_frames(sk);
1210 err = udp_push_pending_frames(sk);
1211 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1223 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1224 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1225 * we don't have a good statistic (IpOutDiscards but it can be too many
1226 * things). We could add another new stat but at least for now that
1227 * seems like overkill.
1229 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1230 UDP_INC_STATS(sock_net(sk),
1231 UDP_MIB_SNDBUFERRORS, is_udplite);
1236 if (msg->msg_flags & MSG_PROBE)
1237 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1238 if (!(msg->msg_flags&MSG_PROBE) || len)
1239 goto back_from_confirm;
1243 EXPORT_SYMBOL(udp_sendmsg);
1245 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1246 size_t size, int flags)
1248 struct inet_sock *inet = inet_sk(sk);
1249 struct udp_sock *up = udp_sk(sk);
1252 if (flags & MSG_SENDPAGE_NOTLAST)
1256 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1258 /* Call udp_sendmsg to specify destination address which
1259 * sendpage interface can't pass.
1260 * This will succeed only when the socket is connected.
1262 ret = udp_sendmsg(sk, &msg, 0);
1269 if (unlikely(!up->pending)) {
1272 net_dbg_ratelimited("cork failed\n");
1276 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1277 page, offset, size, flags);
1278 if (ret == -EOPNOTSUPP) {
1280 return sock_no_sendpage(sk->sk_socket, page, offset,
1284 udp_flush_pending_frames(sk);
1289 if (!(up->corkflag || (flags&MSG_MORE)))
1290 ret = udp_push_pending_frames(sk);
1298 #define UDP_SKB_IS_STATELESS 0x80000000
1300 /* all head states (dst, sk, nf conntrack) except skb extensions are
1301 * cleared by udp_rcv().
1303 * We need to preserve secpath, if present, to eventually process
1304 * IP_CMSG_PASSSEC at recvmsg() time.
1306 * Other extensions can be cleared.
1308 static bool udp_try_make_stateless(struct sk_buff *skb)
1310 if (!skb_has_extensions(skb))
1313 if (!secpath_exists(skb)) {
1321 static void udp_set_dev_scratch(struct sk_buff *skb)
1323 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1325 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1326 scratch->_tsize_state = skb->truesize;
1327 #if BITS_PER_LONG == 64
1328 scratch->len = skb->len;
1329 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1330 scratch->is_linear = !skb_is_nonlinear(skb);
1332 if (udp_try_make_stateless(skb))
1333 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1336 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1338 /* We come here after udp_lib_checksum_complete() returned 0.
1339 * This means that __skb_checksum_complete() might have
1340 * set skb->csum_valid to 1.
1341 * On 64bit platforms, we can set csum_unnecessary
1342 * to true, but only if the skb is not shared.
1344 #if BITS_PER_LONG == 64
1345 if (!skb_shared(skb))
1346 udp_skb_scratch(skb)->csum_unnecessary = true;
1350 static int udp_skb_truesize(struct sk_buff *skb)
1352 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1355 static bool udp_skb_has_head_state(struct sk_buff *skb)
1357 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1360 /* fully reclaim rmem/fwd memory allocated for skb */
1361 static void udp_rmem_release(struct sock *sk, int size, int partial,
1362 bool rx_queue_lock_held)
1364 struct udp_sock *up = udp_sk(sk);
1365 struct sk_buff_head *sk_queue;
1368 if (likely(partial)) {
1369 up->forward_deficit += size;
1370 size = up->forward_deficit;
1371 if (size < (sk->sk_rcvbuf >> 2) &&
1372 !skb_queue_empty(&up->reader_queue))
1375 size += up->forward_deficit;
1377 up->forward_deficit = 0;
1379 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1380 * if the called don't held it already
1382 sk_queue = &sk->sk_receive_queue;
1383 if (!rx_queue_lock_held)
1384 spin_lock(&sk_queue->lock);
1387 sk->sk_forward_alloc += size;
1388 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1389 sk->sk_forward_alloc -= amt;
1392 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1394 atomic_sub(size, &sk->sk_rmem_alloc);
1396 /* this can save us from acquiring the rx queue lock on next receive */
1397 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1399 if (!rx_queue_lock_held)
1400 spin_unlock(&sk_queue->lock);
1403 /* Note: called with reader_queue.lock held.
1404 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1405 * This avoids a cache line miss while receive_queue lock is held.
1406 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1408 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1410 prefetch(&skb->data);
1411 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1413 EXPORT_SYMBOL(udp_skb_destructor);
1415 /* as above, but the caller held the rx queue lock, too */
1416 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1418 prefetch(&skb->data);
1419 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1422 /* Idea of busylocks is to let producers grab an extra spinlock
1423 * to relieve pressure on the receive_queue spinlock shared by consumer.
1424 * Under flood, this means that only one producer can be in line
1425 * trying to acquire the receive_queue spinlock.
1426 * These busylock can be allocated on a per cpu manner, instead of a
1427 * per socket one (that would consume a cache line per socket)
1429 static int udp_busylocks_log __read_mostly;
1430 static spinlock_t *udp_busylocks __read_mostly;
1432 static spinlock_t *busylock_acquire(void *ptr)
1436 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1441 static void busylock_release(spinlock_t *busy)
1447 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1449 struct sk_buff_head *list = &sk->sk_receive_queue;
1450 int rmem, delta, amt, err = -ENOMEM;
1451 spinlock_t *busy = NULL;
1454 /* try to avoid the costly atomic add/sub pair when the receive
1455 * queue is full; always allow at least a packet
1457 rmem = atomic_read(&sk->sk_rmem_alloc);
1458 if (rmem > sk->sk_rcvbuf)
1461 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1462 * having linear skbs :
1463 * - Reduce memory overhead and thus increase receive queue capacity
1464 * - Less cache line misses at copyout() time
1465 * - Less work at consume_skb() (less alien page frag freeing)
1467 if (rmem > (sk->sk_rcvbuf >> 1)) {
1470 busy = busylock_acquire(sk);
1472 size = skb->truesize;
1473 udp_set_dev_scratch(skb);
1475 /* we drop only if the receive buf is full and the receive
1476 * queue contains some other skb
1478 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1479 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1482 spin_lock(&list->lock);
1483 if (size >= sk->sk_forward_alloc) {
1484 amt = sk_mem_pages(size);
1485 delta = amt << SK_MEM_QUANTUM_SHIFT;
1486 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1488 spin_unlock(&list->lock);
1492 sk->sk_forward_alloc += delta;
1495 sk->sk_forward_alloc -= size;
1497 /* no need to setup a destructor, we will explicitly release the
1498 * forward allocated memory on dequeue
1500 sock_skb_set_dropcount(sk, skb);
1502 __skb_queue_tail(list, skb);
1503 spin_unlock(&list->lock);
1505 if (!sock_flag(sk, SOCK_DEAD))
1506 sk->sk_data_ready(sk);
1508 busylock_release(busy);
1512 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1515 atomic_inc(&sk->sk_drops);
1516 busylock_release(busy);
1519 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1521 void udp_destruct_sock(struct sock *sk)
1523 /* reclaim completely the forward allocated memory */
1524 struct udp_sock *up = udp_sk(sk);
1525 unsigned int total = 0;
1526 struct sk_buff *skb;
1528 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1529 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1530 total += skb->truesize;
1533 udp_rmem_release(sk, total, 0, true);
1535 inet_sock_destruct(sk);
1537 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1539 int udp_init_sock(struct sock *sk)
1541 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1542 sk->sk_destruct = udp_destruct_sock;
1545 EXPORT_SYMBOL_GPL(udp_init_sock);
1547 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1549 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1550 bool slow = lock_sock_fast(sk);
1552 sk_peek_offset_bwd(sk, len);
1553 unlock_sock_fast(sk, slow);
1556 if (!skb_unref(skb))
1559 /* In the more common cases we cleared the head states previously,
1560 * see __udp_queue_rcv_skb().
1562 if (unlikely(udp_skb_has_head_state(skb)))
1563 skb_release_head_state(skb);
1564 __consume_stateless_skb(skb);
1566 EXPORT_SYMBOL_GPL(skb_consume_udp);
1568 static struct sk_buff *__first_packet_length(struct sock *sk,
1569 struct sk_buff_head *rcvq,
1572 struct sk_buff *skb;
1574 while ((skb = skb_peek(rcvq)) != NULL) {
1575 if (udp_lib_checksum_complete(skb)) {
1576 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1578 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1580 atomic_inc(&sk->sk_drops);
1581 __skb_unlink(skb, rcvq);
1582 *total += skb->truesize;
1585 udp_skb_csum_unnecessary_set(skb);
1593 * first_packet_length - return length of first packet in receive queue
1596 * Drops all bad checksum frames, until a valid one is found.
1597 * Returns the length of found skb, or -1 if none is found.
1599 static int first_packet_length(struct sock *sk)
1601 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1602 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1603 struct sk_buff *skb;
1607 spin_lock_bh(&rcvq->lock);
1608 skb = __first_packet_length(sk, rcvq, &total);
1609 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1610 spin_lock(&sk_queue->lock);
1611 skb_queue_splice_tail_init(sk_queue, rcvq);
1612 spin_unlock(&sk_queue->lock);
1614 skb = __first_packet_length(sk, rcvq, &total);
1616 res = skb ? skb->len : -1;
1618 udp_rmem_release(sk, total, 1, false);
1619 spin_unlock_bh(&rcvq->lock);
1624 * IOCTL requests applicable to the UDP protocol
1627 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1632 int amount = sk_wmem_alloc_get(sk);
1634 return put_user(amount, (int __user *)arg);
1639 int amount = max_t(int, 0, first_packet_length(sk));
1641 return put_user(amount, (int __user *)arg);
1645 return -ENOIOCTLCMD;
1650 EXPORT_SYMBOL(udp_ioctl);
1652 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1653 int noblock, int *off, int *err)
1655 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1656 struct sk_buff_head *queue;
1657 struct sk_buff *last;
1661 queue = &udp_sk(sk)->reader_queue;
1662 flags |= noblock ? MSG_DONTWAIT : 0;
1663 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1665 struct sk_buff *skb;
1667 error = sock_error(sk);
1673 spin_lock_bh(&queue->lock);
1674 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1677 if (!(flags & MSG_PEEK))
1678 udp_skb_destructor(sk, skb);
1679 spin_unlock_bh(&queue->lock);
1683 if (skb_queue_empty_lockless(sk_queue)) {
1684 spin_unlock_bh(&queue->lock);
1688 /* refill the reader queue and walk it again
1689 * keep both queues locked to avoid re-acquiring
1690 * the sk_receive_queue lock if fwd memory scheduling
1693 spin_lock(&sk_queue->lock);
1694 skb_queue_splice_tail_init(sk_queue, queue);
1696 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1698 if (skb && !(flags & MSG_PEEK))
1699 udp_skb_dtor_locked(sk, skb);
1700 spin_unlock(&sk_queue->lock);
1701 spin_unlock_bh(&queue->lock);
1706 if (!sk_can_busy_loop(sk))
1709 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1710 } while (!skb_queue_empty_lockless(sk_queue));
1712 /* sk_queue is empty, reader_queue may contain peeked packets */
1714 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1716 (struct sk_buff *)sk_queue));
1721 EXPORT_SYMBOL(__skb_recv_udp);
1724 * This should be easy, if there is something there we
1725 * return it, otherwise we block.
1728 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1729 int flags, int *addr_len)
1731 struct inet_sock *inet = inet_sk(sk);
1732 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1733 struct sk_buff *skb;
1734 unsigned int ulen, copied;
1735 int off, err, peeking = flags & MSG_PEEK;
1736 int is_udplite = IS_UDPLITE(sk);
1737 bool checksum_valid = false;
1739 if (flags & MSG_ERRQUEUE)
1740 return ip_recv_error(sk, msg, len, addr_len);
1743 off = sk_peek_offset(sk, flags);
1744 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1748 ulen = udp_skb_len(skb);
1750 if (copied > ulen - off)
1751 copied = ulen - off;
1752 else if (copied < ulen)
1753 msg->msg_flags |= MSG_TRUNC;
1756 * If checksum is needed at all, try to do it while copying the
1757 * data. If the data is truncated, or if we only want a partial
1758 * coverage checksum (UDP-Lite), do it before the copy.
1761 if (copied < ulen || peeking ||
1762 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1763 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1764 !__udp_lib_checksum_complete(skb);
1765 if (!checksum_valid)
1769 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1770 if (udp_skb_is_linear(skb))
1771 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1773 err = skb_copy_datagram_msg(skb, off, msg, copied);
1775 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1781 if (unlikely(err)) {
1783 atomic_inc(&sk->sk_drops);
1784 UDP_INC_STATS(sock_net(sk),
1785 UDP_MIB_INERRORS, is_udplite);
1792 UDP_INC_STATS(sock_net(sk),
1793 UDP_MIB_INDATAGRAMS, is_udplite);
1795 sock_recv_ts_and_drops(msg, sk, skb);
1797 /* Copy the address. */
1799 sin->sin_family = AF_INET;
1800 sin->sin_port = udp_hdr(skb)->source;
1801 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1802 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1803 *addr_len = sizeof(*sin);
1805 if (cgroup_bpf_enabled)
1806 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1807 (struct sockaddr *)sin);
1810 if (udp_sk(sk)->gro_enabled)
1811 udp_cmsg_recv(msg, sk, skb);
1813 if (inet->cmsg_flags)
1814 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1817 if (flags & MSG_TRUNC)
1820 skb_consume_udp(sk, skb, peeking ? -err : err);
1824 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1825 udp_skb_destructor)) {
1826 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1827 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1831 /* starting over for a new packet, but check if we need to yield */
1833 msg->msg_flags &= ~MSG_TRUNC;
1837 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1839 /* This check is replicated from __ip4_datagram_connect() and
1840 * intended to prevent BPF program called below from accessing bytes
1841 * that are out of the bound specified by user in addr_len.
1843 if (addr_len < sizeof(struct sockaddr_in))
1846 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1848 EXPORT_SYMBOL(udp_pre_connect);
1850 int __udp_disconnect(struct sock *sk, int flags)
1852 struct inet_sock *inet = inet_sk(sk);
1854 * 1003.1g - break association.
1857 sk->sk_state = TCP_CLOSE;
1858 inet->inet_daddr = 0;
1859 inet->inet_dport = 0;
1860 sock_rps_reset_rxhash(sk);
1861 sk->sk_bound_dev_if = 0;
1862 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1863 inet_reset_saddr(sk);
1864 if (sk->sk_prot->rehash &&
1865 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1866 sk->sk_prot->rehash(sk);
1869 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1870 sk->sk_prot->unhash(sk);
1871 inet->inet_sport = 0;
1876 EXPORT_SYMBOL(__udp_disconnect);
1878 int udp_disconnect(struct sock *sk, int flags)
1881 __udp_disconnect(sk, flags);
1885 EXPORT_SYMBOL(udp_disconnect);
1887 void udp_lib_unhash(struct sock *sk)
1889 if (sk_hashed(sk)) {
1890 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1891 struct udp_hslot *hslot, *hslot2;
1893 hslot = udp_hashslot(udptable, sock_net(sk),
1894 udp_sk(sk)->udp_port_hash);
1895 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1897 spin_lock_bh(&hslot->lock);
1898 if (rcu_access_pointer(sk->sk_reuseport_cb))
1899 reuseport_detach_sock(sk);
1900 if (sk_del_node_init_rcu(sk)) {
1902 inet_sk(sk)->inet_num = 0;
1903 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1905 spin_lock(&hslot2->lock);
1906 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1908 spin_unlock(&hslot2->lock);
1910 spin_unlock_bh(&hslot->lock);
1913 EXPORT_SYMBOL(udp_lib_unhash);
1916 * inet_rcv_saddr was changed, we must rehash secondary hash
1918 void udp_lib_rehash(struct sock *sk, u16 newhash)
1920 if (sk_hashed(sk)) {
1921 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1922 struct udp_hslot *hslot, *hslot2, *nhslot2;
1924 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1925 nhslot2 = udp_hashslot2(udptable, newhash);
1926 udp_sk(sk)->udp_portaddr_hash = newhash;
1928 if (hslot2 != nhslot2 ||
1929 rcu_access_pointer(sk->sk_reuseport_cb)) {
1930 hslot = udp_hashslot(udptable, sock_net(sk),
1931 udp_sk(sk)->udp_port_hash);
1932 /* we must lock primary chain too */
1933 spin_lock_bh(&hslot->lock);
1934 if (rcu_access_pointer(sk->sk_reuseport_cb))
1935 reuseport_detach_sock(sk);
1937 if (hslot2 != nhslot2) {
1938 spin_lock(&hslot2->lock);
1939 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1941 spin_unlock(&hslot2->lock);
1943 spin_lock(&nhslot2->lock);
1944 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1947 spin_unlock(&nhslot2->lock);
1950 spin_unlock_bh(&hslot->lock);
1954 EXPORT_SYMBOL(udp_lib_rehash);
1956 void udp_v4_rehash(struct sock *sk)
1958 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1959 inet_sk(sk)->inet_rcv_saddr,
1960 inet_sk(sk)->inet_num);
1961 udp_lib_rehash(sk, new_hash);
1964 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1968 if (inet_sk(sk)->inet_daddr) {
1969 sock_rps_save_rxhash(sk, skb);
1970 sk_mark_napi_id(sk, skb);
1971 sk_incoming_cpu_update(sk);
1973 sk_mark_napi_id_once(sk, skb);
1976 rc = __udp_enqueue_schedule_skb(sk, skb);
1978 int is_udplite = IS_UDPLITE(sk);
1980 /* Note that an ENOMEM error is charged twice */
1982 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1984 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1986 trace_udp_fail_queue_rcv_skb(rc, sk);
1996 * >0: "udp encap" protocol resubmission
1998 * Note that in the success and error cases, the skb is assumed to
1999 * have either been requeued or freed.
2001 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2003 struct udp_sock *up = udp_sk(sk);
2004 int is_udplite = IS_UDPLITE(sk);
2007 * Charge it to the socket, dropping if the queue is full.
2009 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2013 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2014 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2017 * This is an encapsulation socket so pass the skb to
2018 * the socket's udp_encap_rcv() hook. Otherwise, just
2019 * fall through and pass this up the UDP socket.
2020 * up->encap_rcv() returns the following value:
2021 * =0 if skb was successfully passed to the encap
2022 * handler or was discarded by it.
2023 * >0 if skb should be passed on to UDP.
2024 * <0 if skb should be resubmitted as proto -N
2027 /* if we're overly short, let UDP handle it */
2028 encap_rcv = READ_ONCE(up->encap_rcv);
2032 /* Verify checksum before giving to encap */
2033 if (udp_lib_checksum_complete(skb))
2036 ret = encap_rcv(sk, skb);
2038 __UDP_INC_STATS(sock_net(sk),
2039 UDP_MIB_INDATAGRAMS,
2045 /* FALLTHROUGH -- it's a UDP Packet */
2049 * UDP-Lite specific tests, ignored on UDP sockets
2051 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2054 * MIB statistics other than incrementing the error count are
2055 * disabled for the following two types of errors: these depend
2056 * on the application settings, not on the functioning of the
2057 * protocol stack as such.
2059 * RFC 3828 here recommends (sec 3.3): "There should also be a
2060 * way ... to ... at least let the receiving application block
2061 * delivery of packets with coverage values less than a value
2062 * provided by the application."
2064 if (up->pcrlen == 0) { /* full coverage was set */
2065 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2066 UDP_SKB_CB(skb)->cscov, skb->len);
2069 /* The next case involves violating the min. coverage requested
2070 * by the receiver. This is subtle: if receiver wants x and x is
2071 * greater than the buffersize/MTU then receiver will complain
2072 * that it wants x while sender emits packets of smaller size y.
2073 * Therefore the above ...()->partial_cov statement is essential.
2075 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2076 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2077 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2082 prefetch(&sk->sk_rmem_alloc);
2083 if (rcu_access_pointer(sk->sk_filter) &&
2084 udp_lib_checksum_complete(skb))
2087 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2090 udp_csum_pull_header(skb);
2092 ipv4_pktinfo_prepare(sk, skb);
2093 return __udp_queue_rcv_skb(sk, skb);
2096 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2098 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2099 atomic_inc(&sk->sk_drops);
2104 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2106 struct sk_buff *next, *segs;
2109 if (likely(!udp_unexpected_gso(sk, skb)))
2110 return udp_queue_rcv_one_skb(sk, skb);
2112 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2113 __skb_push(skb, -skb_mac_offset(skb));
2114 segs = udp_rcv_segment(sk, skb, true);
2115 skb_list_walk_safe(segs, skb, next) {
2116 __skb_pull(skb, skb_transport_offset(skb));
2117 ret = udp_queue_rcv_one_skb(sk, skb);
2119 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2124 /* For TCP sockets, sk_rx_dst is protected by socket lock
2125 * For UDP, we use xchg() to guard against concurrent changes.
2127 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2129 struct dst_entry *old;
2131 if (dst_hold_safe(dst)) {
2132 old = xchg(&sk->sk_rx_dst, dst);
2138 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2141 * Multicasts and broadcasts go to each listener.
2143 * Note: called only from the BH handler context.
2145 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2147 __be32 saddr, __be32 daddr,
2148 struct udp_table *udptable,
2151 struct sock *sk, *first = NULL;
2152 unsigned short hnum = ntohs(uh->dest);
2153 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2154 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2155 unsigned int offset = offsetof(typeof(*sk), sk_node);
2156 int dif = skb->dev->ifindex;
2157 int sdif = inet_sdif(skb);
2158 struct hlist_node *node;
2159 struct sk_buff *nskb;
2162 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2164 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2166 hslot = &udptable->hash2[hash2];
2167 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2170 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2171 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2172 uh->source, saddr, dif, sdif, hnum))
2179 nskb = skb_clone(skb, GFP_ATOMIC);
2181 if (unlikely(!nskb)) {
2182 atomic_inc(&sk->sk_drops);
2183 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2185 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2189 if (udp_queue_rcv_skb(sk, nskb) > 0)
2193 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2194 if (use_hash2 && hash2 != hash2_any) {
2200 if (udp_queue_rcv_skb(first, skb) > 0)
2204 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2205 proto == IPPROTO_UDPLITE);
2210 /* Initialize UDP checksum. If exited with zero value (success),
2211 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2212 * Otherwise, csum completion requires checksumming packet body,
2213 * including udp header and folding it to skb->csum.
2215 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2220 UDP_SKB_CB(skb)->partial_cov = 0;
2221 UDP_SKB_CB(skb)->cscov = skb->len;
2223 if (proto == IPPROTO_UDPLITE) {
2224 err = udplite_checksum_init(skb, uh);
2228 if (UDP_SKB_CB(skb)->partial_cov) {
2229 skb->csum = inet_compute_pseudo(skb, proto);
2234 /* Note, we are only interested in != 0 or == 0, thus the
2237 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2238 inet_compute_pseudo);
2242 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2243 /* If SW calculated the value, we know it's bad */
2244 if (skb->csum_complete_sw)
2247 /* HW says the value is bad. Let's validate that.
2248 * skb->csum is no longer the full packet checksum,
2249 * so don't treat it as such.
2251 skb_checksum_complete_unset(skb);
2257 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2258 * return code conversion for ip layer consumption
2260 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2265 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2266 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2268 ret = udp_queue_rcv_skb(sk, skb);
2270 /* a return value > 0 means to resubmit the input, but
2271 * it wants the return to be -protocol, or 0
2279 * All we need to do is get the socket, and then do a checksum.
2282 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2287 unsigned short ulen;
2288 struct rtable *rt = skb_rtable(skb);
2289 __be32 saddr, daddr;
2290 struct net *net = dev_net(skb->dev);
2294 * Validate the packet.
2296 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2297 goto drop; /* No space for header. */
2300 ulen = ntohs(uh->len);
2301 saddr = ip_hdr(skb)->saddr;
2302 daddr = ip_hdr(skb)->daddr;
2304 if (ulen > skb->len)
2307 if (proto == IPPROTO_UDP) {
2308 /* UDP validates ulen. */
2309 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2314 if (udp4_csum_init(skb, uh, proto))
2317 sk = skb_steal_sock(skb, &refcounted);
2319 struct dst_entry *dst = skb_dst(skb);
2322 if (unlikely(sk->sk_rx_dst != dst))
2323 udp_sk_rx_dst_set(sk, dst);
2325 ret = udp_unicast_rcv_skb(sk, skb, uh);
2331 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2332 return __udp4_lib_mcast_deliver(net, skb, uh,
2333 saddr, daddr, udptable, proto);
2335 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2337 return udp_unicast_rcv_skb(sk, skb, uh);
2339 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2343 /* No socket. Drop packet silently, if checksum is wrong */
2344 if (udp_lib_checksum_complete(skb))
2347 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2348 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2351 * Hmm. We got an UDP packet to a port to which we
2352 * don't wanna listen. Ignore it.
2358 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2359 proto == IPPROTO_UDPLITE ? "Lite" : "",
2360 &saddr, ntohs(uh->source),
2362 &daddr, ntohs(uh->dest));
2367 * RFC1122: OK. Discards the bad packet silently (as far as
2368 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2370 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2371 proto == IPPROTO_UDPLITE ? "Lite" : "",
2372 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2374 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2376 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2381 /* We can only early demux multicast if there is a single matching socket.
2382 * If more than one socket found returns NULL
2384 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2385 __be16 loc_port, __be32 loc_addr,
2386 __be16 rmt_port, __be32 rmt_addr,
2389 struct sock *sk, *result;
2390 unsigned short hnum = ntohs(loc_port);
2391 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2392 struct udp_hslot *hslot = &udp_table.hash[slot];
2394 /* Do not bother scanning a too big list */
2395 if (hslot->count > 10)
2399 sk_for_each_rcu(sk, &hslot->head) {
2400 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2401 rmt_port, rmt_addr, dif, sdif, hnum)) {
2411 /* For unicast we should only early demux connected sockets or we can
2412 * break forwarding setups. The chains here can be long so only check
2413 * if the first socket is an exact match and if not move on.
2415 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2416 __be16 loc_port, __be32 loc_addr,
2417 __be16 rmt_port, __be32 rmt_addr,
2420 unsigned short hnum = ntohs(loc_port);
2421 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2422 unsigned int slot2 = hash2 & udp_table.mask;
2423 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2424 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2425 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2428 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2429 if (INET_MATCH(sk, net, acookie, rmt_addr,
2430 loc_addr, ports, dif, sdif))
2432 /* Only check first socket in chain */
2438 int udp_v4_early_demux(struct sk_buff *skb)
2440 struct net *net = dev_net(skb->dev);
2441 struct in_device *in_dev = NULL;
2442 const struct iphdr *iph;
2443 const struct udphdr *uh;
2444 struct sock *sk = NULL;
2445 struct dst_entry *dst;
2446 int dif = skb->dev->ifindex;
2447 int sdif = inet_sdif(skb);
2450 /* validate the packet */
2451 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2457 if (skb->pkt_type == PACKET_MULTICAST) {
2458 in_dev = __in_dev_get_rcu(skb->dev);
2463 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2468 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2469 uh->source, iph->saddr,
2471 } else if (skb->pkt_type == PACKET_HOST) {
2472 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2473 uh->source, iph->saddr, dif, sdif);
2476 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2480 skb->destructor = sock_efree;
2481 dst = READ_ONCE(sk->sk_rx_dst);
2484 dst = dst_check(dst, 0);
2488 /* set noref for now.
2489 * any place which wants to hold dst has to call
2492 skb_dst_set_noref(skb, dst);
2494 /* for unconnected multicast sockets we need to validate
2495 * the source on each packet
2497 if (!inet_sk(sk)->inet_daddr && in_dev)
2498 return ip_mc_validate_source(skb, iph->daddr,
2499 iph->saddr, iph->tos,
2500 skb->dev, in_dev, &itag);
2505 int udp_rcv(struct sk_buff *skb)
2507 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2510 void udp_destroy_sock(struct sock *sk)
2512 struct udp_sock *up = udp_sk(sk);
2513 bool slow = lock_sock_fast(sk);
2514 udp_flush_pending_frames(sk);
2515 unlock_sock_fast(sk, slow);
2516 if (static_branch_unlikely(&udp_encap_needed_key)) {
2517 if (up->encap_type) {
2518 void (*encap_destroy)(struct sock *sk);
2519 encap_destroy = READ_ONCE(up->encap_destroy);
2523 if (up->encap_enabled)
2524 static_branch_dec(&udp_encap_needed_key);
2529 * Socket option code for UDP
2531 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2532 char __user *optval, unsigned int optlen,
2533 int (*push_pending_frames)(struct sock *))
2535 struct udp_sock *up = udp_sk(sk);
2538 int is_udplite = IS_UDPLITE(sk);
2540 if (optlen < sizeof(int))
2543 if (get_user(val, (int __user *)optval))
2546 valbool = val ? 1 : 0;
2555 push_pending_frames(sk);
2564 case UDP_ENCAP_ESPINUDP:
2565 case UDP_ENCAP_ESPINUDP_NON_IKE:
2566 up->encap_rcv = xfrm4_udp_encap_rcv;
2569 case UDP_ENCAP_L2TPINUDP:
2570 up->encap_type = val;
2572 udp_tunnel_encap_enable(sk->sk_socket);
2581 case UDP_NO_CHECK6_TX:
2582 up->no_check6_tx = valbool;
2585 case UDP_NO_CHECK6_RX:
2586 up->no_check6_rx = valbool;
2590 if (val < 0 || val > USHRT_MAX)
2598 udp_tunnel_encap_enable(sk->sk_socket);
2599 up->gro_enabled = valbool;
2604 * UDP-Lite's partial checksum coverage (RFC 3828).
2606 /* The sender sets actual checksum coverage length via this option.
2607 * The case coverage > packet length is handled by send module. */
2608 case UDPLITE_SEND_CSCOV:
2609 if (!is_udplite) /* Disable the option on UDP sockets */
2610 return -ENOPROTOOPT;
2611 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2613 else if (val > USHRT_MAX)
2616 up->pcflag |= UDPLITE_SEND_CC;
2619 /* The receiver specifies a minimum checksum coverage value. To make
2620 * sense, this should be set to at least 8 (as done below). If zero is
2621 * used, this again means full checksum coverage. */
2622 case UDPLITE_RECV_CSCOV:
2623 if (!is_udplite) /* Disable the option on UDP sockets */
2624 return -ENOPROTOOPT;
2625 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2627 else if (val > USHRT_MAX)
2630 up->pcflag |= UDPLITE_RECV_CC;
2640 EXPORT_SYMBOL(udp_lib_setsockopt);
2642 int udp_setsockopt(struct sock *sk, int level, int optname,
2643 char __user *optval, unsigned int optlen)
2645 if (level == SOL_UDP || level == SOL_UDPLITE)
2646 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2647 udp_push_pending_frames);
2648 return ip_setsockopt(sk, level, optname, optval, optlen);
2651 #ifdef CONFIG_COMPAT
2652 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2653 char __user *optval, unsigned int optlen)
2655 if (level == SOL_UDP || level == SOL_UDPLITE)
2656 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2657 udp_push_pending_frames);
2658 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2662 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2663 char __user *optval, int __user *optlen)
2665 struct udp_sock *up = udp_sk(sk);
2668 if (get_user(len, optlen))
2671 len = min_t(unsigned int, len, sizeof(int));
2682 val = up->encap_type;
2685 case UDP_NO_CHECK6_TX:
2686 val = up->no_check6_tx;
2689 case UDP_NO_CHECK6_RX:
2690 val = up->no_check6_rx;
2697 /* The following two cannot be changed on UDP sockets, the return is
2698 * always 0 (which corresponds to the full checksum coverage of UDP). */
2699 case UDPLITE_SEND_CSCOV:
2703 case UDPLITE_RECV_CSCOV:
2708 return -ENOPROTOOPT;
2711 if (put_user(len, optlen))
2713 if (copy_to_user(optval, &val, len))
2717 EXPORT_SYMBOL(udp_lib_getsockopt);
2719 int udp_getsockopt(struct sock *sk, int level, int optname,
2720 char __user *optval, int __user *optlen)
2722 if (level == SOL_UDP || level == SOL_UDPLITE)
2723 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2724 return ip_getsockopt(sk, level, optname, optval, optlen);
2727 #ifdef CONFIG_COMPAT
2728 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2729 char __user *optval, int __user *optlen)
2731 if (level == SOL_UDP || level == SOL_UDPLITE)
2732 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2733 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2737 * udp_poll - wait for a UDP event.
2738 * @file - file struct
2740 * @wait - poll table
2742 * This is same as datagram poll, except for the special case of
2743 * blocking sockets. If application is using a blocking fd
2744 * and a packet with checksum error is in the queue;
2745 * then it could get return from select indicating data available
2746 * but then block when reading it. Add special case code
2747 * to work around these arguably broken applications.
2749 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2751 __poll_t mask = datagram_poll(file, sock, wait);
2752 struct sock *sk = sock->sk;
2754 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2755 mask |= EPOLLIN | EPOLLRDNORM;
2757 /* Check for false positives due to checksum errors */
2758 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2759 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2760 mask &= ~(EPOLLIN | EPOLLRDNORM);
2765 EXPORT_SYMBOL(udp_poll);
2767 int udp_abort(struct sock *sk, int err)
2772 sk->sk_error_report(sk);
2773 __udp_disconnect(sk, 0);
2779 EXPORT_SYMBOL_GPL(udp_abort);
2781 struct proto udp_prot = {
2783 .owner = THIS_MODULE,
2784 .close = udp_lib_close,
2785 .pre_connect = udp_pre_connect,
2786 .connect = ip4_datagram_connect,
2787 .disconnect = udp_disconnect,
2789 .init = udp_init_sock,
2790 .destroy = udp_destroy_sock,
2791 .setsockopt = udp_setsockopt,
2792 .getsockopt = udp_getsockopt,
2793 .sendmsg = udp_sendmsg,
2794 .recvmsg = udp_recvmsg,
2795 .sendpage = udp_sendpage,
2796 .release_cb = ip4_datagram_release_cb,
2797 .hash = udp_lib_hash,
2798 .unhash = udp_lib_unhash,
2799 .rehash = udp_v4_rehash,
2800 .get_port = udp_v4_get_port,
2801 .memory_allocated = &udp_memory_allocated,
2802 .sysctl_mem = sysctl_udp_mem,
2803 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2804 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2805 .obj_size = sizeof(struct udp_sock),
2806 .h.udp_table = &udp_table,
2807 #ifdef CONFIG_COMPAT
2808 .compat_setsockopt = compat_udp_setsockopt,
2809 .compat_getsockopt = compat_udp_getsockopt,
2811 .diag_destroy = udp_abort,
2813 EXPORT_SYMBOL(udp_prot);
2815 /* ------------------------------------------------------------------------ */
2816 #ifdef CONFIG_PROC_FS
2818 static struct sock *udp_get_first(struct seq_file *seq, int start)
2821 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2822 struct udp_iter_state *state = seq->private;
2823 struct net *net = seq_file_net(seq);
2825 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2827 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2829 if (hlist_empty(&hslot->head))
2832 spin_lock_bh(&hslot->lock);
2833 sk_for_each(sk, &hslot->head) {
2834 if (!net_eq(sock_net(sk), net))
2836 if (sk->sk_family == afinfo->family)
2839 spin_unlock_bh(&hslot->lock);
2846 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2848 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2849 struct udp_iter_state *state = seq->private;
2850 struct net *net = seq_file_net(seq);
2854 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2857 if (state->bucket <= afinfo->udp_table->mask)
2858 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2859 return udp_get_first(seq, state->bucket + 1);
2864 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2866 struct sock *sk = udp_get_first(seq, 0);
2869 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2871 return pos ? NULL : sk;
2874 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2876 struct udp_iter_state *state = seq->private;
2877 state->bucket = MAX_UDP_PORTS;
2879 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2881 EXPORT_SYMBOL(udp_seq_start);
2883 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2887 if (v == SEQ_START_TOKEN)
2888 sk = udp_get_idx(seq, 0);
2890 sk = udp_get_next(seq, v);
2895 EXPORT_SYMBOL(udp_seq_next);
2897 void udp_seq_stop(struct seq_file *seq, void *v)
2899 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2900 struct udp_iter_state *state = seq->private;
2902 if (state->bucket <= afinfo->udp_table->mask)
2903 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2905 EXPORT_SYMBOL(udp_seq_stop);
2907 /* ------------------------------------------------------------------------ */
2908 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2911 struct inet_sock *inet = inet_sk(sp);
2912 __be32 dest = inet->inet_daddr;
2913 __be32 src = inet->inet_rcv_saddr;
2914 __u16 destp = ntohs(inet->inet_dport);
2915 __u16 srcp = ntohs(inet->inet_sport);
2917 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2918 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2919 bucket, src, srcp, dest, destp, sp->sk_state,
2920 sk_wmem_alloc_get(sp),
2923 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2925 refcount_read(&sp->sk_refcnt), sp,
2926 atomic_read(&sp->sk_drops));
2929 int udp4_seq_show(struct seq_file *seq, void *v)
2931 seq_setwidth(seq, 127);
2932 if (v == SEQ_START_TOKEN)
2933 seq_puts(seq, " sl local_address rem_address st tx_queue "
2934 "rx_queue tr tm->when retrnsmt uid timeout "
2935 "inode ref pointer drops");
2937 struct udp_iter_state *state = seq->private;
2939 udp4_format_sock(v, seq, state->bucket);
2945 const struct seq_operations udp_seq_ops = {
2946 .start = udp_seq_start,
2947 .next = udp_seq_next,
2948 .stop = udp_seq_stop,
2949 .show = udp4_seq_show,
2951 EXPORT_SYMBOL(udp_seq_ops);
2953 static struct udp_seq_afinfo udp4_seq_afinfo = {
2955 .udp_table = &udp_table,
2958 static int __net_init udp4_proc_init_net(struct net *net)
2960 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2961 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2966 static void __net_exit udp4_proc_exit_net(struct net *net)
2968 remove_proc_entry("udp", net->proc_net);
2971 static struct pernet_operations udp4_net_ops = {
2972 .init = udp4_proc_init_net,
2973 .exit = udp4_proc_exit_net,
2976 int __init udp4_proc_init(void)
2978 return register_pernet_subsys(&udp4_net_ops);
2981 void udp4_proc_exit(void)
2983 unregister_pernet_subsys(&udp4_net_ops);
2985 #endif /* CONFIG_PROC_FS */
2987 static __initdata unsigned long uhash_entries;
2988 static int __init set_uhash_entries(char *str)
2995 ret = kstrtoul(str, 0, &uhash_entries);
2999 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3000 uhash_entries = UDP_HTABLE_SIZE_MIN;
3003 __setup("uhash_entries=", set_uhash_entries);
3005 void __init udp_table_init(struct udp_table *table, const char *name)
3009 table->hash = alloc_large_system_hash(name,
3010 2 * sizeof(struct udp_hslot),
3012 21, /* one slot per 2 MB */
3016 UDP_HTABLE_SIZE_MIN,
3019 table->hash2 = table->hash + (table->mask + 1);
3020 for (i = 0; i <= table->mask; i++) {
3021 INIT_HLIST_HEAD(&table->hash[i].head);
3022 table->hash[i].count = 0;
3023 spin_lock_init(&table->hash[i].lock);
3025 for (i = 0; i <= table->mask; i++) {
3026 INIT_HLIST_HEAD(&table->hash2[i].head);
3027 table->hash2[i].count = 0;
3028 spin_lock_init(&table->hash2[i].lock);
3032 u32 udp_flow_hashrnd(void)
3034 static u32 hashrnd __read_mostly;
3036 net_get_random_once(&hashrnd, sizeof(hashrnd));
3040 EXPORT_SYMBOL(udp_flow_hashrnd);
3042 static void __udp_sysctl_init(struct net *net)
3044 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3045 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3047 #ifdef CONFIG_NET_L3_MASTER_DEV
3048 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3052 static int __net_init udp_sysctl_init(struct net *net)
3054 __udp_sysctl_init(net);
3058 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3059 .init = udp_sysctl_init,
3062 void __init udp_init(void)
3064 unsigned long limit;
3067 udp_table_init(&udp_table, "UDP");
3068 limit = nr_free_buffer_pages() / 8;
3069 limit = max(limit, 128UL);
3070 sysctl_udp_mem[0] = limit / 4 * 3;
3071 sysctl_udp_mem[1] = limit;
3072 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3074 __udp_sysctl_init(&init_net);
3076 /* 16 spinlocks per cpu */
3077 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3078 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3081 panic("UDP: failed to alloc udp_busylocks\n");
3082 for (i = 0; i < (1U << udp_busylocks_log); i++)
3083 spin_lock_init(udp_busylocks + i);
3085 if (register_pernet_subsys(&udp_sysctl_ops))
3086 panic("UDP: failed to init sysctl parameters.\n");