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 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,
389 if (sk->sk_bound_dev_if)
392 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
397 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
398 const __u16 lport, const __be32 faddr,
401 static u32 udp_ehash_secret __read_mostly;
403 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
405 return __inet_ehashfn(laddr, lport, faddr, fport,
406 udp_ehash_secret + net_hash_mix(net));
409 /* called with rcu_read_lock() */
410 static struct sock *udp4_lib_lookup2(struct net *net,
411 __be32 saddr, __be16 sport,
412 __be32 daddr, unsigned int hnum,
414 struct udp_hslot *hslot2,
417 struct sock *sk, *result, *reuseport_result;
423 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
424 score = compute_score(sk, net, saddr, sport,
425 daddr, hnum, dif, sdif);
426 if (score > badness) {
427 reuseport_result = NULL;
429 if (sk->sk_reuseport &&
430 sk->sk_state != TCP_ESTABLISHED) {
431 hash = udp_ehashfn(net, daddr, hnum,
433 reuseport_result = reuseport_select_sock(sk, hash, skb,
434 sizeof(struct udphdr));
435 if (reuseport_result && !reuseport_has_conns(sk, false))
436 return reuseport_result;
439 result = reuseport_result ? : sk;
446 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
447 * harder than this. -DaveM
449 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
450 __be16 sport, __be32 daddr, __be16 dport, int dif,
451 int sdif, struct udp_table *udptable, struct sk_buff *skb)
454 unsigned short hnum = ntohs(dport);
455 unsigned int hash2, slot2;
456 struct udp_hslot *hslot2;
458 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
459 slot2 = hash2 & udptable->mask;
460 hslot2 = &udptable->hash2[slot2];
462 result = udp4_lib_lookup2(net, saddr, sport,
463 daddr, hnum, dif, sdif,
466 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
467 slot2 = hash2 & udptable->mask;
468 hslot2 = &udptable->hash2[slot2];
470 result = udp4_lib_lookup2(net, saddr, sport,
471 htonl(INADDR_ANY), hnum, dif, sdif,
478 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
480 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
481 __be16 sport, __be16 dport,
482 struct udp_table *udptable)
484 const struct iphdr *iph = ip_hdr(skb);
486 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
487 iph->daddr, dport, inet_iif(skb),
488 inet_sdif(skb), udptable, skb);
491 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
492 __be16 sport, __be16 dport)
494 const struct iphdr *iph = ip_hdr(skb);
496 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
497 iph->daddr, dport, inet_iif(skb),
498 inet_sdif(skb), &udp_table, NULL);
500 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
502 /* Must be called under rcu_read_lock().
503 * Does increment socket refcount.
505 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
506 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
507 __be32 daddr, __be16 dport, int dif)
511 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
512 dif, 0, &udp_table, NULL);
513 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
517 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
520 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
521 __be16 loc_port, __be32 loc_addr,
522 __be16 rmt_port, __be32 rmt_addr,
523 int dif, int sdif, unsigned short hnum)
525 struct inet_sock *inet = inet_sk(sk);
527 if (!net_eq(sock_net(sk), net) ||
528 udp_sk(sk)->udp_port_hash != hnum ||
529 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
530 (inet->inet_dport != rmt_port && inet->inet_dport) ||
531 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
532 ipv6_only_sock(sk) ||
533 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
535 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
540 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
541 void udp_encap_enable(void)
543 static_branch_inc(&udp_encap_needed_key);
545 EXPORT_SYMBOL(udp_encap_enable);
547 void udp_encap_disable(void)
549 static_branch_dec(&udp_encap_needed_key);
551 EXPORT_SYMBOL(udp_encap_disable);
553 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
554 * through error handlers in encapsulations looking for a match.
556 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
560 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
561 int (*handler)(struct sk_buff *skb, u32 info);
562 const struct ip_tunnel_encap_ops *encap;
564 encap = rcu_dereference(iptun_encaps[i]);
567 handler = encap->err_handler;
568 if (handler && !handler(skb, info))
575 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
576 * reversing source and destination port: this will match tunnels that force the
577 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
578 * lwtunnels might actually break this assumption by being configured with
579 * different destination ports on endpoints, in this case we won't be able to
580 * trace ICMP messages back to them.
582 * If this doesn't match any socket, probe tunnels with arbitrary destination
583 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
584 * we've sent packets to won't necessarily match the local destination port.
586 * Then ask the tunnel implementation to match the error against a valid
589 * Return an error if we can't find a match, the socket if we need further
590 * processing, zero otherwise.
592 static struct sock *__udp4_lib_err_encap(struct net *net,
593 const struct iphdr *iph,
595 struct udp_table *udptable,
596 struct sk_buff *skb, u32 info)
598 int network_offset, transport_offset;
601 network_offset = skb_network_offset(skb);
602 transport_offset = skb_transport_offset(skb);
604 /* Network header needs to point to the outer IPv4 header inside ICMP */
605 skb_reset_network_header(skb);
607 /* Transport header needs to point to the UDP header */
608 skb_set_transport_header(skb, iph->ihl << 2);
610 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
611 iph->saddr, uh->dest, skb->dev->ifindex, 0,
614 int (*lookup)(struct sock *sk, struct sk_buff *skb);
615 struct udp_sock *up = udp_sk(sk);
617 lookup = READ_ONCE(up->encap_err_lookup);
618 if (!lookup || lookup(sk, skb))
623 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
625 skb_set_transport_header(skb, transport_offset);
626 skb_set_network_header(skb, network_offset);
632 * This routine is called by the ICMP module when it gets some
633 * sort of error condition. If err < 0 then the socket should
634 * be closed and the error returned to the user. If err > 0
635 * it's just the icmp type << 8 | icmp code.
636 * Header points to the ip header of the error packet. We move
637 * on past this. Then (as it used to claim before adjustment)
638 * header points to the first 8 bytes of the udp header. We need
639 * to find the appropriate port.
642 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
644 struct inet_sock *inet;
645 const struct iphdr *iph = (const struct iphdr *)skb->data;
646 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
647 const int type = icmp_hdr(skb)->type;
648 const int code = icmp_hdr(skb)->code;
653 struct net *net = dev_net(skb->dev);
655 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
656 iph->saddr, uh->source, skb->dev->ifindex,
657 inet_sdif(skb), udptable, NULL);
659 /* No socket for error: try tunnels before discarding */
660 sk = ERR_PTR(-ENOENT);
661 if (static_branch_unlikely(&udp_encap_needed_key)) {
662 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
669 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
682 case ICMP_TIME_EXCEEDED:
685 case ICMP_SOURCE_QUENCH:
687 case ICMP_PARAMETERPROB:
691 case ICMP_DEST_UNREACH:
692 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
693 ipv4_sk_update_pmtu(skb, sk, info);
694 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
702 if (code <= NR_ICMP_UNREACH) {
703 harderr = icmp_err_convert[code].fatal;
704 err = icmp_err_convert[code].errno;
708 ipv4_sk_redirect(skb, sk);
713 * RFC1122: OK. Passes ICMP errors back to application, as per
717 /* ...not for tunnels though: we don't have a sending socket */
720 if (!inet->recverr) {
721 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
724 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
727 sk->sk_error_report(sk);
732 int udp_err(struct sk_buff *skb, u32 info)
734 return __udp4_lib_err(skb, info, &udp_table);
738 * Throw away all pending data and cancel the corking. Socket is locked.
740 void udp_flush_pending_frames(struct sock *sk)
742 struct udp_sock *up = udp_sk(sk);
747 ip_flush_pending_frames(sk);
750 EXPORT_SYMBOL(udp_flush_pending_frames);
753 * udp4_hwcsum - handle outgoing HW checksumming
754 * @skb: sk_buff containing the filled-in UDP header
755 * (checksum field must be zeroed out)
756 * @src: source IP address
757 * @dst: destination IP address
759 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
761 struct udphdr *uh = udp_hdr(skb);
762 int offset = skb_transport_offset(skb);
763 int len = skb->len - offset;
767 if (!skb_has_frag_list(skb)) {
769 * Only one fragment on the socket.
771 skb->csum_start = skb_transport_header(skb) - skb->head;
772 skb->csum_offset = offsetof(struct udphdr, check);
773 uh->check = ~csum_tcpudp_magic(src, dst, len,
776 struct sk_buff *frags;
779 * HW-checksum won't work as there are two or more
780 * fragments on the socket so that all csums of sk_buffs
783 skb_walk_frags(skb, frags) {
784 csum = csum_add(csum, frags->csum);
788 csum = skb_checksum(skb, offset, hlen, csum);
789 skb->ip_summed = CHECKSUM_NONE;
791 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
793 uh->check = CSUM_MANGLED_0;
796 EXPORT_SYMBOL_GPL(udp4_hwcsum);
798 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
799 * for the simple case like when setting the checksum for a UDP tunnel.
801 void udp_set_csum(bool nocheck, struct sk_buff *skb,
802 __be32 saddr, __be32 daddr, int len)
804 struct udphdr *uh = udp_hdr(skb);
808 } else if (skb_is_gso(skb)) {
809 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
810 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
812 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
814 uh->check = CSUM_MANGLED_0;
816 skb->ip_summed = CHECKSUM_PARTIAL;
817 skb->csum_start = skb_transport_header(skb) - skb->head;
818 skb->csum_offset = offsetof(struct udphdr, check);
819 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
822 EXPORT_SYMBOL(udp_set_csum);
824 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
825 struct inet_cork *cork)
827 struct sock *sk = skb->sk;
828 struct inet_sock *inet = inet_sk(sk);
831 int is_udplite = IS_UDPLITE(sk);
832 int offset = skb_transport_offset(skb);
833 int len = skb->len - offset;
834 int datalen = len - sizeof(*uh);
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 (datalen > 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 if (datalen > cork->gso_size) {
869 skb_shinfo(skb)->gso_size = cork->gso_size;
870 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
871 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
877 if (is_udplite) /* UDP-Lite */
878 csum = udplite_csum(skb);
880 else if (sk->sk_no_check_tx) { /* UDP csum off */
882 skb->ip_summed = CHECKSUM_NONE;
885 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
888 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
892 csum = udp_csum(skb);
894 /* add protocol-dependent pseudo-header */
895 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
896 sk->sk_protocol, csum);
898 uh->check = CSUM_MANGLED_0;
901 err = ip_send_skb(sock_net(sk), skb);
903 if (err == -ENOBUFS && !inet->recverr) {
904 UDP_INC_STATS(sock_net(sk),
905 UDP_MIB_SNDBUFERRORS, is_udplite);
909 UDP_INC_STATS(sock_net(sk),
910 UDP_MIB_OUTDATAGRAMS, is_udplite);
915 * Push out all pending data as one UDP datagram. Socket is locked.
917 int udp_push_pending_frames(struct sock *sk)
919 struct udp_sock *up = udp_sk(sk);
920 struct inet_sock *inet = inet_sk(sk);
921 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
925 skb = ip_finish_skb(sk, fl4);
929 err = udp_send_skb(skb, fl4, &inet->cork.base);
936 EXPORT_SYMBOL(udp_push_pending_frames);
938 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
940 switch (cmsg->cmsg_type) {
942 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
944 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
951 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
953 struct cmsghdr *cmsg;
954 bool need_ip = false;
957 for_each_cmsghdr(cmsg, msg) {
958 if (!CMSG_OK(msg, cmsg))
961 if (cmsg->cmsg_level != SOL_UDP) {
966 err = __udp_cmsg_send(cmsg, gso_size);
973 EXPORT_SYMBOL_GPL(udp_cmsg_send);
975 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
977 struct inet_sock *inet = inet_sk(sk);
978 struct udp_sock *up = udp_sk(sk);
979 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
980 struct flowi4 fl4_stack;
983 struct ipcm_cookie ipc;
984 struct rtable *rt = NULL;
987 __be32 daddr, faddr, saddr;
990 int err, is_udplite = IS_UDPLITE(sk);
991 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
992 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
994 struct ip_options_data opt_copy;
1003 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1006 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1008 fl4 = &inet->cork.fl.u.ip4;
1011 * There are pending frames.
1012 * The socket lock must be held while it's corked.
1015 if (likely(up->pending)) {
1016 if (unlikely(up->pending != AF_INET)) {
1020 goto do_append_data;
1024 ulen += sizeof(struct udphdr);
1027 * Get and verify the address.
1030 if (msg->msg_namelen < sizeof(*usin))
1032 if (usin->sin_family != AF_INET) {
1033 if (usin->sin_family != AF_UNSPEC)
1034 return -EAFNOSUPPORT;
1037 daddr = usin->sin_addr.s_addr;
1038 dport = usin->sin_port;
1042 if (sk->sk_state != TCP_ESTABLISHED)
1043 return -EDESTADDRREQ;
1044 daddr = inet->inet_daddr;
1045 dport = inet->inet_dport;
1046 /* Open fast path for connected socket.
1047 Route will not be used, if at least one option is set.
1052 ipcm_init_sk(&ipc, inet);
1053 ipc.gso_size = READ_ONCE(up->gso_size);
1055 if (msg->msg_controllen) {
1056 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1058 err = ip_cmsg_send(sk, msg, &ipc,
1059 sk->sk_family == AF_INET6);
1060 if (unlikely(err < 0)) {
1069 struct ip_options_rcu *inet_opt;
1072 inet_opt = rcu_dereference(inet->inet_opt);
1074 memcpy(&opt_copy, inet_opt,
1075 sizeof(*inet_opt) + inet_opt->opt.optlen);
1076 ipc.opt = &opt_copy.opt;
1081 if (cgroup_bpf_enabled && !connected) {
1082 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1083 (struct sockaddr *)usin, &ipc.addr);
1087 if (usin->sin_port == 0) {
1088 /* BPF program set invalid port. Reject it. */
1092 daddr = usin->sin_addr.s_addr;
1093 dport = usin->sin_port;
1098 ipc.addr = faddr = daddr;
1100 if (ipc.opt && ipc.opt->opt.srr) {
1105 faddr = ipc.opt->opt.faddr;
1108 tos = get_rttos(&ipc, inet);
1109 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1110 (msg->msg_flags & MSG_DONTROUTE) ||
1111 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1116 if (ipv4_is_multicast(daddr)) {
1117 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1118 ipc.oif = inet->mc_index;
1120 saddr = inet->mc_addr;
1122 } else if (!ipc.oif) {
1123 ipc.oif = inet->uc_index;
1124 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1125 /* oif is set, packet is to local broadcast and
1126 * and uc_index is set. oif is most likely set
1127 * by sk_bound_dev_if. If uc_index != oif check if the
1128 * oif is an L3 master and uc_index is an L3 slave.
1129 * If so, we want to allow the send using the uc_index.
1131 if (ipc.oif != inet->uc_index &&
1132 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1134 ipc.oif = inet->uc_index;
1139 rt = (struct rtable *)sk_dst_check(sk, 0);
1142 struct net *net = sock_net(sk);
1143 __u8 flow_flags = inet_sk_flowi_flags(sk);
1147 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1148 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1150 faddr, saddr, dport, inet->inet_sport,
1153 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1154 rt = ip_route_output_flow(net, fl4, sk);
1158 if (err == -ENETUNREACH)
1159 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1164 if ((rt->rt_flags & RTCF_BROADCAST) &&
1165 !sock_flag(sk, SOCK_BROADCAST))
1168 sk_dst_set(sk, dst_clone(&rt->dst));
1171 if (msg->msg_flags&MSG_CONFIRM)
1177 daddr = ipc.addr = fl4->daddr;
1179 /* Lockless fast path for the non-corking case. */
1181 struct inet_cork cork;
1183 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1184 sizeof(struct udphdr), &ipc, &rt,
1185 &cork, msg->msg_flags);
1187 if (!IS_ERR_OR_NULL(skb))
1188 err = udp_send_skb(skb, fl4, &cork);
1193 if (unlikely(up->pending)) {
1194 /* The socket is already corked while preparing it. */
1195 /* ... which is an evident application bug. --ANK */
1198 net_dbg_ratelimited("socket already corked\n");
1203 * Now cork the socket to pend data.
1205 fl4 = &inet->cork.fl.u.ip4;
1208 fl4->fl4_dport = dport;
1209 fl4->fl4_sport = inet->inet_sport;
1210 up->pending = AF_INET;
1214 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1215 sizeof(struct udphdr), &ipc, &rt,
1216 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1218 udp_flush_pending_frames(sk);
1220 err = udp_push_pending_frames(sk);
1221 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1233 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1234 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1235 * we don't have a good statistic (IpOutDiscards but it can be too many
1236 * things). We could add another new stat but at least for now that
1237 * seems like overkill.
1239 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1240 UDP_INC_STATS(sock_net(sk),
1241 UDP_MIB_SNDBUFERRORS, is_udplite);
1246 if (msg->msg_flags & MSG_PROBE)
1247 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1248 if (!(msg->msg_flags&MSG_PROBE) || len)
1249 goto back_from_confirm;
1253 EXPORT_SYMBOL(udp_sendmsg);
1255 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1256 size_t size, int flags)
1258 struct inet_sock *inet = inet_sk(sk);
1259 struct udp_sock *up = udp_sk(sk);
1262 if (flags & MSG_SENDPAGE_NOTLAST)
1266 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1268 /* Call udp_sendmsg to specify destination address which
1269 * sendpage interface can't pass.
1270 * This will succeed only when the socket is connected.
1272 ret = udp_sendmsg(sk, &msg, 0);
1279 if (unlikely(!up->pending)) {
1282 net_dbg_ratelimited("cork failed\n");
1286 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1287 page, offset, size, flags);
1288 if (ret == -EOPNOTSUPP) {
1290 return sock_no_sendpage(sk->sk_socket, page, offset,
1294 udp_flush_pending_frames(sk);
1299 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1300 ret = udp_push_pending_frames(sk);
1308 #define UDP_SKB_IS_STATELESS 0x80000000
1310 /* all head states (dst, sk, nf conntrack) except skb extensions are
1311 * cleared by udp_rcv().
1313 * We need to preserve secpath, if present, to eventually process
1314 * IP_CMSG_PASSSEC at recvmsg() time.
1316 * Other extensions can be cleared.
1318 static bool udp_try_make_stateless(struct sk_buff *skb)
1320 if (!skb_has_extensions(skb))
1323 if (!secpath_exists(skb)) {
1331 static void udp_set_dev_scratch(struct sk_buff *skb)
1333 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1335 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1336 scratch->_tsize_state = skb->truesize;
1337 #if BITS_PER_LONG == 64
1338 scratch->len = skb->len;
1339 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1340 scratch->is_linear = !skb_is_nonlinear(skb);
1342 if (udp_try_make_stateless(skb))
1343 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1346 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1348 /* We come here after udp_lib_checksum_complete() returned 0.
1349 * This means that __skb_checksum_complete() might have
1350 * set skb->csum_valid to 1.
1351 * On 64bit platforms, we can set csum_unnecessary
1352 * to true, but only if the skb is not shared.
1354 #if BITS_PER_LONG == 64
1355 if (!skb_shared(skb))
1356 udp_skb_scratch(skb)->csum_unnecessary = true;
1360 static int udp_skb_truesize(struct sk_buff *skb)
1362 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1365 static bool udp_skb_has_head_state(struct sk_buff *skb)
1367 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1370 /* fully reclaim rmem/fwd memory allocated for skb */
1371 static void udp_rmem_release(struct sock *sk, int size, int partial,
1372 bool rx_queue_lock_held)
1374 struct udp_sock *up = udp_sk(sk);
1375 struct sk_buff_head *sk_queue;
1378 if (likely(partial)) {
1379 up->forward_deficit += size;
1380 size = up->forward_deficit;
1381 if (size < (sk->sk_rcvbuf >> 2) &&
1382 !skb_queue_empty(&up->reader_queue))
1385 size += up->forward_deficit;
1387 up->forward_deficit = 0;
1389 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1390 * if the called don't held it already
1392 sk_queue = &sk->sk_receive_queue;
1393 if (!rx_queue_lock_held)
1394 spin_lock(&sk_queue->lock);
1397 sk->sk_forward_alloc += size;
1398 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1399 sk->sk_forward_alloc -= amt;
1402 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1404 atomic_sub(size, &sk->sk_rmem_alloc);
1406 /* this can save us from acquiring the rx queue lock on next receive */
1407 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1409 if (!rx_queue_lock_held)
1410 spin_unlock(&sk_queue->lock);
1413 /* Note: called with reader_queue.lock held.
1414 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1415 * This avoids a cache line miss while receive_queue lock is held.
1416 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1418 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1420 prefetch(&skb->data);
1421 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1423 EXPORT_SYMBOL(udp_skb_destructor);
1425 /* as above, but the caller held the rx queue lock, too */
1426 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1428 prefetch(&skb->data);
1429 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1432 /* Idea of busylocks is to let producers grab an extra spinlock
1433 * to relieve pressure on the receive_queue spinlock shared by consumer.
1434 * Under flood, this means that only one producer can be in line
1435 * trying to acquire the receive_queue spinlock.
1436 * These busylock can be allocated on a per cpu manner, instead of a
1437 * per socket one (that would consume a cache line per socket)
1439 static int udp_busylocks_log __read_mostly;
1440 static spinlock_t *udp_busylocks __read_mostly;
1442 static spinlock_t *busylock_acquire(void *ptr)
1446 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1451 static void busylock_release(spinlock_t *busy)
1457 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1459 struct sk_buff_head *list = &sk->sk_receive_queue;
1460 int rmem, delta, amt, err = -ENOMEM;
1461 spinlock_t *busy = NULL;
1464 /* try to avoid the costly atomic add/sub pair when the receive
1465 * queue is full; always allow at least a packet
1467 rmem = atomic_read(&sk->sk_rmem_alloc);
1468 if (rmem > sk->sk_rcvbuf)
1471 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1472 * having linear skbs :
1473 * - Reduce memory overhead and thus increase receive queue capacity
1474 * - Less cache line misses at copyout() time
1475 * - Less work at consume_skb() (less alien page frag freeing)
1477 if (rmem > (sk->sk_rcvbuf >> 1)) {
1480 busy = busylock_acquire(sk);
1482 size = skb->truesize;
1483 udp_set_dev_scratch(skb);
1485 /* we drop only if the receive buf is full and the receive
1486 * queue contains some other skb
1488 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1489 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1492 spin_lock(&list->lock);
1493 if (size >= sk->sk_forward_alloc) {
1494 amt = sk_mem_pages(size);
1495 delta = amt << SK_MEM_QUANTUM_SHIFT;
1496 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1498 spin_unlock(&list->lock);
1502 sk->sk_forward_alloc += delta;
1505 sk->sk_forward_alloc -= size;
1507 /* no need to setup a destructor, we will explicitly release the
1508 * forward allocated memory on dequeue
1510 sock_skb_set_dropcount(sk, skb);
1512 __skb_queue_tail(list, skb);
1513 spin_unlock(&list->lock);
1515 if (!sock_flag(sk, SOCK_DEAD))
1516 sk->sk_data_ready(sk);
1518 busylock_release(busy);
1522 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1525 atomic_inc(&sk->sk_drops);
1526 busylock_release(busy);
1529 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1531 void udp_destruct_common(struct sock *sk)
1533 /* reclaim completely the forward allocated memory */
1534 struct udp_sock *up = udp_sk(sk);
1535 unsigned int total = 0;
1536 struct sk_buff *skb;
1538 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1539 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1540 total += skb->truesize;
1543 udp_rmem_release(sk, total, 0, true);
1545 EXPORT_SYMBOL_GPL(udp_destruct_common);
1547 static void udp_destruct_sock(struct sock *sk)
1549 udp_destruct_common(sk);
1550 inet_sock_destruct(sk);
1553 int udp_init_sock(struct sock *sk)
1555 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1556 sk->sk_destruct = udp_destruct_sock;
1560 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1562 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1563 bool slow = lock_sock_fast(sk);
1565 sk_peek_offset_bwd(sk, len);
1566 unlock_sock_fast(sk, slow);
1569 if (!skb_unref(skb))
1572 /* In the more common cases we cleared the head states previously,
1573 * see __udp_queue_rcv_skb().
1575 if (unlikely(udp_skb_has_head_state(skb)))
1576 skb_release_head_state(skb);
1577 __consume_stateless_skb(skb);
1579 EXPORT_SYMBOL_GPL(skb_consume_udp);
1581 static struct sk_buff *__first_packet_length(struct sock *sk,
1582 struct sk_buff_head *rcvq,
1585 struct sk_buff *skb;
1587 while ((skb = skb_peek(rcvq)) != NULL) {
1588 if (udp_lib_checksum_complete(skb)) {
1589 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1591 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1593 atomic_inc(&sk->sk_drops);
1594 __skb_unlink(skb, rcvq);
1595 *total += skb->truesize;
1598 udp_skb_csum_unnecessary_set(skb);
1606 * first_packet_length - return length of first packet in receive queue
1609 * Drops all bad checksum frames, until a valid one is found.
1610 * Returns the length of found skb, or -1 if none is found.
1612 static int first_packet_length(struct sock *sk)
1614 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1615 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1616 struct sk_buff *skb;
1620 spin_lock_bh(&rcvq->lock);
1621 skb = __first_packet_length(sk, rcvq, &total);
1622 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1623 spin_lock(&sk_queue->lock);
1624 skb_queue_splice_tail_init(sk_queue, rcvq);
1625 spin_unlock(&sk_queue->lock);
1627 skb = __first_packet_length(sk, rcvq, &total);
1629 res = skb ? skb->len : -1;
1631 udp_rmem_release(sk, total, 1, false);
1632 spin_unlock_bh(&rcvq->lock);
1637 * IOCTL requests applicable to the UDP protocol
1640 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1645 int amount = sk_wmem_alloc_get(sk);
1647 return put_user(amount, (int __user *)arg);
1652 int amount = max_t(int, 0, first_packet_length(sk));
1654 return put_user(amount, (int __user *)arg);
1658 return -ENOIOCTLCMD;
1663 EXPORT_SYMBOL(udp_ioctl);
1665 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1666 int noblock, int *off, int *err)
1668 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1669 struct sk_buff_head *queue;
1670 struct sk_buff *last;
1674 queue = &udp_sk(sk)->reader_queue;
1675 flags |= noblock ? MSG_DONTWAIT : 0;
1676 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1678 struct sk_buff *skb;
1680 error = sock_error(sk);
1686 spin_lock_bh(&queue->lock);
1687 skb = __skb_try_recv_from_queue(sk, queue, flags,
1691 spin_unlock_bh(&queue->lock);
1695 if (skb_queue_empty_lockless(sk_queue)) {
1696 spin_unlock_bh(&queue->lock);
1700 /* refill the reader queue and walk it again
1701 * keep both queues locked to avoid re-acquiring
1702 * the sk_receive_queue lock if fwd memory scheduling
1705 spin_lock(&sk_queue->lock);
1706 skb_queue_splice_tail_init(sk_queue, queue);
1708 skb = __skb_try_recv_from_queue(sk, queue, flags,
1709 udp_skb_dtor_locked,
1711 spin_unlock(&sk_queue->lock);
1712 spin_unlock_bh(&queue->lock);
1717 if (!sk_can_busy_loop(sk))
1720 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1721 } while (!skb_queue_empty_lockless(sk_queue));
1723 /* sk_queue is empty, reader_queue may contain peeked packets */
1725 !__skb_wait_for_more_packets(sk, &error, &timeo,
1726 (struct sk_buff *)sk_queue));
1731 EXPORT_SYMBOL(__skb_recv_udp);
1734 * This should be easy, if there is something there we
1735 * return it, otherwise we block.
1738 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1739 int flags, int *addr_len)
1741 struct inet_sock *inet = inet_sk(sk);
1742 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1743 struct sk_buff *skb;
1744 unsigned int ulen, copied;
1745 int off, err, peeking = flags & MSG_PEEK;
1746 int is_udplite = IS_UDPLITE(sk);
1747 bool checksum_valid = false;
1749 if (flags & MSG_ERRQUEUE)
1750 return ip_recv_error(sk, msg, len, addr_len);
1753 off = sk_peek_offset(sk, flags);
1754 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1758 ulen = udp_skb_len(skb);
1760 if (copied > ulen - off)
1761 copied = ulen - off;
1762 else if (copied < ulen)
1763 msg->msg_flags |= MSG_TRUNC;
1766 * If checksum is needed at all, try to do it while copying the
1767 * data. If the data is truncated, or if we only want a partial
1768 * coverage checksum (UDP-Lite), do it before the copy.
1771 if (copied < ulen || peeking ||
1772 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1773 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1774 !__udp_lib_checksum_complete(skb);
1775 if (!checksum_valid)
1779 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1780 if (udp_skb_is_linear(skb))
1781 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1783 err = skb_copy_datagram_msg(skb, off, msg, copied);
1785 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1791 if (unlikely(err)) {
1793 atomic_inc(&sk->sk_drops);
1794 UDP_INC_STATS(sock_net(sk),
1795 UDP_MIB_INERRORS, is_udplite);
1802 UDP_INC_STATS(sock_net(sk),
1803 UDP_MIB_INDATAGRAMS, is_udplite);
1805 sock_recv_ts_and_drops(msg, sk, skb);
1807 /* Copy the address. */
1809 sin->sin_family = AF_INET;
1810 sin->sin_port = udp_hdr(skb)->source;
1811 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1812 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1813 *addr_len = sizeof(*sin);
1815 if (cgroup_bpf_enabled)
1816 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1817 (struct sockaddr *)sin);
1820 if (udp_sk(sk)->gro_enabled)
1821 udp_cmsg_recv(msg, sk, skb);
1823 if (inet->cmsg_flags)
1824 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1827 if (flags & MSG_TRUNC)
1830 skb_consume_udp(sk, skb, peeking ? -err : err);
1834 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1835 udp_skb_destructor)) {
1836 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1837 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1841 /* starting over for a new packet, but check if we need to yield */
1843 msg->msg_flags &= ~MSG_TRUNC;
1847 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1849 /* This check is replicated from __ip4_datagram_connect() and
1850 * intended to prevent BPF program called below from accessing bytes
1851 * that are out of the bound specified by user in addr_len.
1853 if (addr_len < sizeof(struct sockaddr_in))
1856 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1858 EXPORT_SYMBOL(udp_pre_connect);
1860 int __udp_disconnect(struct sock *sk, int flags)
1862 struct inet_sock *inet = inet_sk(sk);
1864 * 1003.1g - break association.
1867 sk->sk_state = TCP_CLOSE;
1868 inet->inet_daddr = 0;
1869 inet->inet_dport = 0;
1870 sock_rps_reset_rxhash(sk);
1871 sk->sk_bound_dev_if = 0;
1872 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1873 inet_reset_saddr(sk);
1874 if (sk->sk_prot->rehash &&
1875 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1876 sk->sk_prot->rehash(sk);
1879 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1880 sk->sk_prot->unhash(sk);
1881 inet->inet_sport = 0;
1886 EXPORT_SYMBOL(__udp_disconnect);
1888 int udp_disconnect(struct sock *sk, int flags)
1891 __udp_disconnect(sk, flags);
1895 EXPORT_SYMBOL(udp_disconnect);
1897 void udp_lib_unhash(struct sock *sk)
1899 if (sk_hashed(sk)) {
1900 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1901 struct udp_hslot *hslot, *hslot2;
1903 hslot = udp_hashslot(udptable, sock_net(sk),
1904 udp_sk(sk)->udp_port_hash);
1905 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1907 spin_lock_bh(&hslot->lock);
1908 if (rcu_access_pointer(sk->sk_reuseport_cb))
1909 reuseport_detach_sock(sk);
1910 if (sk_del_node_init_rcu(sk)) {
1912 inet_sk(sk)->inet_num = 0;
1913 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1915 spin_lock(&hslot2->lock);
1916 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1918 spin_unlock(&hslot2->lock);
1920 spin_unlock_bh(&hslot->lock);
1923 EXPORT_SYMBOL(udp_lib_unhash);
1926 * inet_rcv_saddr was changed, we must rehash secondary hash
1928 void udp_lib_rehash(struct sock *sk, u16 newhash)
1930 if (sk_hashed(sk)) {
1931 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1932 struct udp_hslot *hslot, *hslot2, *nhslot2;
1934 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1935 nhslot2 = udp_hashslot2(udptable, newhash);
1936 udp_sk(sk)->udp_portaddr_hash = newhash;
1938 if (hslot2 != nhslot2 ||
1939 rcu_access_pointer(sk->sk_reuseport_cb)) {
1940 hslot = udp_hashslot(udptable, sock_net(sk),
1941 udp_sk(sk)->udp_port_hash);
1942 /* we must lock primary chain too */
1943 spin_lock_bh(&hslot->lock);
1944 if (rcu_access_pointer(sk->sk_reuseport_cb))
1945 reuseport_detach_sock(sk);
1947 if (hslot2 != nhslot2) {
1948 spin_lock(&hslot2->lock);
1949 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1951 spin_unlock(&hslot2->lock);
1953 spin_lock(&nhslot2->lock);
1954 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1957 spin_unlock(&nhslot2->lock);
1960 spin_unlock_bh(&hslot->lock);
1964 EXPORT_SYMBOL(udp_lib_rehash);
1966 void udp_v4_rehash(struct sock *sk)
1968 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1969 inet_sk(sk)->inet_rcv_saddr,
1970 inet_sk(sk)->inet_num);
1971 udp_lib_rehash(sk, new_hash);
1974 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1978 if (inet_sk(sk)->inet_daddr) {
1979 sock_rps_save_rxhash(sk, skb);
1980 sk_mark_napi_id(sk, skb);
1981 sk_incoming_cpu_update(sk);
1983 sk_mark_napi_id_once(sk, skb);
1986 rc = __udp_enqueue_schedule_skb(sk, skb);
1988 int is_udplite = IS_UDPLITE(sk);
1990 /* Note that an ENOMEM error is charged twice */
1992 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1994 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1996 trace_udp_fail_queue_rcv_skb(rc, sk);
2006 * >0: "udp encap" protocol resubmission
2008 * Note that in the success and error cases, the skb is assumed to
2009 * have either been requeued or freed.
2011 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2013 struct udp_sock *up = udp_sk(sk);
2014 int is_udplite = IS_UDPLITE(sk);
2017 * Charge it to the socket, dropping if the queue is full.
2019 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2023 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2024 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2027 * This is an encapsulation socket so pass the skb to
2028 * the socket's udp_encap_rcv() hook. Otherwise, just
2029 * fall through and pass this up the UDP socket.
2030 * up->encap_rcv() returns the following value:
2031 * =0 if skb was successfully passed to the encap
2032 * handler or was discarded by it.
2033 * >0 if skb should be passed on to UDP.
2034 * <0 if skb should be resubmitted as proto -N
2037 /* if we're overly short, let UDP handle it */
2038 encap_rcv = READ_ONCE(up->encap_rcv);
2042 /* Verify checksum before giving to encap */
2043 if (udp_lib_checksum_complete(skb))
2046 ret = encap_rcv(sk, skb);
2048 __UDP_INC_STATS(sock_net(sk),
2049 UDP_MIB_INDATAGRAMS,
2055 /* FALLTHROUGH -- it's a UDP Packet */
2059 * UDP-Lite specific tests, ignored on UDP sockets
2061 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2064 * MIB statistics other than incrementing the error count are
2065 * disabled for the following two types of errors: these depend
2066 * on the application settings, not on the functioning of the
2067 * protocol stack as such.
2069 * RFC 3828 here recommends (sec 3.3): "There should also be a
2070 * way ... to ... at least let the receiving application block
2071 * delivery of packets with coverage values less than a value
2072 * provided by the application."
2074 if (up->pcrlen == 0) { /* full coverage was set */
2075 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2076 UDP_SKB_CB(skb)->cscov, skb->len);
2079 /* The next case involves violating the min. coverage requested
2080 * by the receiver. This is subtle: if receiver wants x and x is
2081 * greater than the buffersize/MTU then receiver will complain
2082 * that it wants x while sender emits packets of smaller size y.
2083 * Therefore the above ...()->partial_cov statement is essential.
2085 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2086 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2087 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2092 prefetch(&sk->sk_rmem_alloc);
2093 if (rcu_access_pointer(sk->sk_filter) &&
2094 udp_lib_checksum_complete(skb))
2097 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2100 udp_csum_pull_header(skb);
2102 ipv4_pktinfo_prepare(sk, skb);
2103 return __udp_queue_rcv_skb(sk, skb);
2106 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2108 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2109 atomic_inc(&sk->sk_drops);
2114 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2116 struct sk_buff *next, *segs;
2119 if (likely(!udp_unexpected_gso(sk, skb)))
2120 return udp_queue_rcv_one_skb(sk, skb);
2122 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET);
2123 __skb_push(skb, -skb_mac_offset(skb));
2124 segs = udp_rcv_segment(sk, skb, true);
2125 for (skb = segs; skb; skb = next) {
2127 __skb_pull(skb, skb_transport_offset(skb));
2128 ret = udp_queue_rcv_one_skb(sk, skb);
2130 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2135 /* For TCP sockets, sk_rx_dst is protected by socket lock
2136 * For UDP, we use xchg() to guard against concurrent changes.
2138 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2140 struct dst_entry *old;
2142 if (dst_hold_safe(dst)) {
2143 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2149 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2152 * Multicasts and broadcasts go to each listener.
2154 * Note: called only from the BH handler context.
2156 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2158 __be32 saddr, __be32 daddr,
2159 struct udp_table *udptable,
2162 struct sock *sk, *first = NULL;
2163 unsigned short hnum = ntohs(uh->dest);
2164 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2165 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2166 unsigned int offset = offsetof(typeof(*sk), sk_node);
2167 int dif = skb->dev->ifindex;
2168 int sdif = inet_sdif(skb);
2169 struct hlist_node *node;
2170 struct sk_buff *nskb;
2173 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2175 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2177 hslot = &udptable->hash2[hash2];
2178 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2181 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2182 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2183 uh->source, saddr, dif, sdif, hnum))
2190 nskb = skb_clone(skb, GFP_ATOMIC);
2192 if (unlikely(!nskb)) {
2193 atomic_inc(&sk->sk_drops);
2194 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2196 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2200 if (udp_queue_rcv_skb(sk, nskb) > 0)
2204 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2205 if (use_hash2 && hash2 != hash2_any) {
2211 if (udp_queue_rcv_skb(first, skb) > 0)
2215 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2216 proto == IPPROTO_UDPLITE);
2221 /* Initialize UDP checksum. If exited with zero value (success),
2222 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2223 * Otherwise, csum completion requires checksumming packet body,
2224 * including udp header and folding it to skb->csum.
2226 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2231 UDP_SKB_CB(skb)->partial_cov = 0;
2232 UDP_SKB_CB(skb)->cscov = skb->len;
2234 if (proto == IPPROTO_UDPLITE) {
2235 err = udplite_checksum_init(skb, uh);
2239 if (UDP_SKB_CB(skb)->partial_cov) {
2240 skb->csum = inet_compute_pseudo(skb, proto);
2245 /* Note, we are only interested in != 0 or == 0, thus the
2248 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2249 inet_compute_pseudo);
2253 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2254 /* If SW calculated the value, we know it's bad */
2255 if (skb->csum_complete_sw)
2258 /* HW says the value is bad. Let's validate that.
2259 * skb->csum is no longer the full packet checksum,
2260 * so don't treat it as such.
2262 skb_checksum_complete_unset(skb);
2268 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2269 * return code conversion for ip layer consumption
2271 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2276 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2277 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2279 ret = udp_queue_rcv_skb(sk, skb);
2281 /* a return value > 0 means to resubmit the input, but
2282 * it wants the return to be -protocol, or 0
2290 * All we need to do is get the socket, and then do a checksum.
2293 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2298 unsigned short ulen;
2299 struct rtable *rt = skb_rtable(skb);
2300 __be32 saddr, daddr;
2301 struct net *net = dev_net(skb->dev);
2304 * Validate the packet.
2306 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2307 goto drop; /* No space for header. */
2310 ulen = ntohs(uh->len);
2311 saddr = ip_hdr(skb)->saddr;
2312 daddr = ip_hdr(skb)->daddr;
2314 if (ulen > skb->len)
2317 if (proto == IPPROTO_UDP) {
2318 /* UDP validates ulen. */
2319 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2324 if (udp4_csum_init(skb, uh, proto))
2327 sk = skb_steal_sock(skb);
2329 struct dst_entry *dst = skb_dst(skb);
2332 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2333 udp_sk_rx_dst_set(sk, dst);
2335 ret = udp_unicast_rcv_skb(sk, skb, uh);
2340 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2341 return __udp4_lib_mcast_deliver(net, skb, uh,
2342 saddr, daddr, udptable, proto);
2344 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2346 return udp_unicast_rcv_skb(sk, skb, uh);
2348 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2352 /* No socket. Drop packet silently, if checksum is wrong */
2353 if (udp_lib_checksum_complete(skb))
2356 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2357 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2360 * Hmm. We got an UDP packet to a port to which we
2361 * don't wanna listen. Ignore it.
2367 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2368 proto == IPPROTO_UDPLITE ? "Lite" : "",
2369 &saddr, ntohs(uh->source),
2371 &daddr, ntohs(uh->dest));
2376 * RFC1122: OK. Discards the bad packet silently (as far as
2377 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2379 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2380 proto == IPPROTO_UDPLITE ? "Lite" : "",
2381 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2383 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2385 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2390 /* We can only early demux multicast if there is a single matching socket.
2391 * If more than one socket found returns NULL
2393 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2394 __be16 loc_port, __be32 loc_addr,
2395 __be16 rmt_port, __be32 rmt_addr,
2398 struct sock *sk, *result;
2399 unsigned short hnum = ntohs(loc_port);
2400 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2401 struct udp_hslot *hslot = &udp_table.hash[slot];
2403 /* Do not bother scanning a too big list */
2404 if (hslot->count > 10)
2408 sk_for_each_rcu(sk, &hslot->head) {
2409 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2410 rmt_port, rmt_addr, dif, sdif, hnum)) {
2420 /* For unicast we should only early demux connected sockets or we can
2421 * break forwarding setups. The chains here can be long so only check
2422 * if the first socket is an exact match and if not move on.
2424 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2425 __be16 loc_port, __be32 loc_addr,
2426 __be16 rmt_port, __be32 rmt_addr,
2429 unsigned short hnum = ntohs(loc_port);
2430 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2431 unsigned int slot2 = hash2 & udp_table.mask;
2432 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2433 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2434 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2437 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2438 if (INET_MATCH(sk, net, acookie, rmt_addr,
2439 loc_addr, ports, dif, sdif))
2441 /* Only check first socket in chain */
2447 int udp_v4_early_demux(struct sk_buff *skb)
2449 struct net *net = dev_net(skb->dev);
2450 struct in_device *in_dev = NULL;
2451 const struct iphdr *iph;
2452 const struct udphdr *uh;
2453 struct sock *sk = NULL;
2454 struct dst_entry *dst;
2455 int dif = skb->dev->ifindex;
2456 int sdif = inet_sdif(skb);
2459 /* validate the packet */
2460 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2466 if (skb->pkt_type == PACKET_MULTICAST) {
2467 in_dev = __in_dev_get_rcu(skb->dev);
2472 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2477 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2478 uh->source, iph->saddr,
2480 } else if (skb->pkt_type == PACKET_HOST) {
2481 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2482 uh->source, iph->saddr, dif, sdif);
2485 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2489 skb->destructor = sock_efree;
2490 dst = rcu_dereference(sk->sk_rx_dst);
2493 dst = dst_check(dst, 0);
2497 /* set noref for now.
2498 * any place which wants to hold dst has to call
2501 skb_dst_set_noref(skb, dst);
2503 /* for unconnected multicast sockets we need to validate
2504 * the source on each packet
2506 if (!inet_sk(sk)->inet_daddr && in_dev)
2507 return ip_mc_validate_source(skb, iph->daddr,
2509 iph->tos & IPTOS_RT_MASK,
2510 skb->dev, in_dev, &itag);
2515 int udp_rcv(struct sk_buff *skb)
2517 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2520 void udp_destroy_sock(struct sock *sk)
2522 struct udp_sock *up = udp_sk(sk);
2523 bool slow = lock_sock_fast(sk);
2525 /* protects from races with udp_abort() */
2526 sock_set_flag(sk, SOCK_DEAD);
2527 udp_flush_pending_frames(sk);
2528 unlock_sock_fast(sk, slow);
2529 if (static_branch_unlikely(&udp_encap_needed_key)) {
2530 if (up->encap_type) {
2531 void (*encap_destroy)(struct sock *sk);
2532 encap_destroy = READ_ONCE(up->encap_destroy);
2536 if (up->encap_enabled)
2537 static_branch_dec(&udp_encap_needed_key);
2542 * Socket option code for UDP
2544 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2545 char __user *optval, unsigned int optlen,
2546 int (*push_pending_frames)(struct sock *))
2548 struct udp_sock *up = udp_sk(sk);
2551 int is_udplite = IS_UDPLITE(sk);
2553 if (optlen < sizeof(int))
2556 if (get_user(val, (int __user *)optval))
2559 valbool = val ? 1 : 0;
2564 WRITE_ONCE(up->corkflag, 1);
2566 WRITE_ONCE(up->corkflag, 0);
2568 push_pending_frames(sk);
2576 case UDP_ENCAP_ESPINUDP:
2577 case UDP_ENCAP_ESPINUDP_NON_IKE:
2578 up->encap_rcv = xfrm4_udp_encap_rcv;
2580 case UDP_ENCAP_L2TPINUDP:
2581 up->encap_type = val;
2583 udp_tunnel_encap_enable(sk->sk_socket);
2592 case UDP_NO_CHECK6_TX:
2593 up->no_check6_tx = valbool;
2596 case UDP_NO_CHECK6_RX:
2597 up->no_check6_rx = valbool;
2601 if (val < 0 || val > USHRT_MAX)
2603 WRITE_ONCE(up->gso_size, val);
2609 udp_tunnel_encap_enable(sk->sk_socket);
2610 up->gro_enabled = valbool;
2615 * UDP-Lite's partial checksum coverage (RFC 3828).
2617 /* The sender sets actual checksum coverage length via this option.
2618 * The case coverage > packet length is handled by send module. */
2619 case UDPLITE_SEND_CSCOV:
2620 if (!is_udplite) /* Disable the option on UDP sockets */
2621 return -ENOPROTOOPT;
2622 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2624 else if (val > USHRT_MAX)
2627 up->pcflag |= UDPLITE_SEND_CC;
2630 /* The receiver specifies a minimum checksum coverage value. To make
2631 * sense, this should be set to at least 8 (as done below). If zero is
2632 * used, this again means full checksum coverage. */
2633 case UDPLITE_RECV_CSCOV:
2634 if (!is_udplite) /* Disable the option on UDP sockets */
2635 return -ENOPROTOOPT;
2636 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2638 else if (val > USHRT_MAX)
2641 up->pcflag |= UDPLITE_RECV_CC;
2651 EXPORT_SYMBOL(udp_lib_setsockopt);
2653 int udp_setsockopt(struct sock *sk, int level, int optname,
2654 char __user *optval, unsigned int optlen)
2656 if (level == SOL_UDP || level == SOL_UDPLITE)
2657 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2658 udp_push_pending_frames);
2659 return ip_setsockopt(sk, level, optname, optval, optlen);
2662 #ifdef CONFIG_COMPAT
2663 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2664 char __user *optval, unsigned int optlen)
2666 if (level == SOL_UDP || level == SOL_UDPLITE)
2667 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2668 udp_push_pending_frames);
2669 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2673 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2674 char __user *optval, int __user *optlen)
2676 struct udp_sock *up = udp_sk(sk);
2679 if (get_user(len, optlen))
2682 len = min_t(unsigned int, len, sizeof(int));
2689 val = READ_ONCE(up->corkflag);
2693 val = up->encap_type;
2696 case UDP_NO_CHECK6_TX:
2697 val = up->no_check6_tx;
2700 case UDP_NO_CHECK6_RX:
2701 val = up->no_check6_rx;
2705 val = READ_ONCE(up->gso_size);
2709 val = up->gro_enabled;
2712 /* The following two cannot be changed on UDP sockets, the return is
2713 * always 0 (which corresponds to the full checksum coverage of UDP). */
2714 case UDPLITE_SEND_CSCOV:
2718 case UDPLITE_RECV_CSCOV:
2723 return -ENOPROTOOPT;
2726 if (put_user(len, optlen))
2728 if (copy_to_user(optval, &val, len))
2732 EXPORT_SYMBOL(udp_lib_getsockopt);
2734 int udp_getsockopt(struct sock *sk, int level, int optname,
2735 char __user *optval, int __user *optlen)
2737 if (level == SOL_UDP || level == SOL_UDPLITE)
2738 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2739 return ip_getsockopt(sk, level, optname, optval, optlen);
2742 #ifdef CONFIG_COMPAT
2743 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2744 char __user *optval, int __user *optlen)
2746 if (level == SOL_UDP || level == SOL_UDPLITE)
2747 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2748 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2752 * udp_poll - wait for a UDP event.
2753 * @file - file struct
2755 * @wait - poll table
2757 * This is same as datagram poll, except for the special case of
2758 * blocking sockets. If application is using a blocking fd
2759 * and a packet with checksum error is in the queue;
2760 * then it could get return from select indicating data available
2761 * but then block when reading it. Add special case code
2762 * to work around these arguably broken applications.
2764 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2766 __poll_t mask = datagram_poll(file, sock, wait);
2767 struct sock *sk = sock->sk;
2769 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2770 mask |= EPOLLIN | EPOLLRDNORM;
2772 /* Check for false positives due to checksum errors */
2773 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2774 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2775 mask &= ~(EPOLLIN | EPOLLRDNORM);
2780 EXPORT_SYMBOL(udp_poll);
2782 int udp_abort(struct sock *sk, int err)
2786 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2789 if (sock_flag(sk, SOCK_DEAD))
2793 sk->sk_error_report(sk);
2794 __udp_disconnect(sk, 0);
2801 EXPORT_SYMBOL_GPL(udp_abort);
2803 struct proto udp_prot = {
2805 .owner = THIS_MODULE,
2806 .close = udp_lib_close,
2807 .pre_connect = udp_pre_connect,
2808 .connect = ip4_datagram_connect,
2809 .disconnect = udp_disconnect,
2811 .init = udp_init_sock,
2812 .destroy = udp_destroy_sock,
2813 .setsockopt = udp_setsockopt,
2814 .getsockopt = udp_getsockopt,
2815 .sendmsg = udp_sendmsg,
2816 .recvmsg = udp_recvmsg,
2817 .sendpage = udp_sendpage,
2818 .release_cb = ip4_datagram_release_cb,
2819 .hash = udp_lib_hash,
2820 .unhash = udp_lib_unhash,
2821 .rehash = udp_v4_rehash,
2822 .get_port = udp_v4_get_port,
2823 .memory_allocated = &udp_memory_allocated,
2824 .sysctl_mem = sysctl_udp_mem,
2825 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2826 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2827 .obj_size = sizeof(struct udp_sock),
2828 .h.udp_table = &udp_table,
2829 #ifdef CONFIG_COMPAT
2830 .compat_setsockopt = compat_udp_setsockopt,
2831 .compat_getsockopt = compat_udp_getsockopt,
2833 .diag_destroy = udp_abort,
2835 EXPORT_SYMBOL(udp_prot);
2837 /* ------------------------------------------------------------------------ */
2838 #ifdef CONFIG_PROC_FS
2840 static struct sock *udp_get_first(struct seq_file *seq, int start)
2843 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2844 struct udp_iter_state *state = seq->private;
2845 struct net *net = seq_file_net(seq);
2847 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2849 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2851 if (hlist_empty(&hslot->head))
2854 spin_lock_bh(&hslot->lock);
2855 sk_for_each(sk, &hslot->head) {
2856 if (!net_eq(sock_net(sk), net))
2858 if (sk->sk_family == afinfo->family)
2861 spin_unlock_bh(&hslot->lock);
2868 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2870 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2871 struct udp_iter_state *state = seq->private;
2872 struct net *net = seq_file_net(seq);
2876 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2879 if (state->bucket <= afinfo->udp_table->mask)
2880 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2881 return udp_get_first(seq, state->bucket + 1);
2886 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2888 struct sock *sk = udp_get_first(seq, 0);
2891 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2893 return pos ? NULL : sk;
2896 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2898 struct udp_iter_state *state = seq->private;
2899 state->bucket = MAX_UDP_PORTS;
2901 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2903 EXPORT_SYMBOL(udp_seq_start);
2905 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2909 if (v == SEQ_START_TOKEN)
2910 sk = udp_get_idx(seq, 0);
2912 sk = udp_get_next(seq, v);
2917 EXPORT_SYMBOL(udp_seq_next);
2919 void udp_seq_stop(struct seq_file *seq, void *v)
2921 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2922 struct udp_iter_state *state = seq->private;
2924 if (state->bucket <= afinfo->udp_table->mask)
2925 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2927 EXPORT_SYMBOL(udp_seq_stop);
2929 /* ------------------------------------------------------------------------ */
2930 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2933 struct inet_sock *inet = inet_sk(sp);
2934 __be32 dest = inet->inet_daddr;
2935 __be32 src = inet->inet_rcv_saddr;
2936 __u16 destp = ntohs(inet->inet_dport);
2937 __u16 srcp = ntohs(inet->inet_sport);
2939 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2940 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2941 bucket, src, srcp, dest, destp, sp->sk_state,
2942 sk_wmem_alloc_get(sp),
2945 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2947 refcount_read(&sp->sk_refcnt), sp,
2948 atomic_read(&sp->sk_drops));
2951 int udp4_seq_show(struct seq_file *seq, void *v)
2953 seq_setwidth(seq, 127);
2954 if (v == SEQ_START_TOKEN)
2955 seq_puts(seq, " sl local_address rem_address st tx_queue "
2956 "rx_queue tr tm->when retrnsmt uid timeout "
2957 "inode ref pointer drops");
2959 struct udp_iter_state *state = seq->private;
2961 udp4_format_sock(v, seq, state->bucket);
2967 const struct seq_operations udp_seq_ops = {
2968 .start = udp_seq_start,
2969 .next = udp_seq_next,
2970 .stop = udp_seq_stop,
2971 .show = udp4_seq_show,
2973 EXPORT_SYMBOL(udp_seq_ops);
2975 static struct udp_seq_afinfo udp4_seq_afinfo = {
2977 .udp_table = &udp_table,
2980 static int __net_init udp4_proc_init_net(struct net *net)
2982 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2983 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2988 static void __net_exit udp4_proc_exit_net(struct net *net)
2990 remove_proc_entry("udp", net->proc_net);
2993 static struct pernet_operations udp4_net_ops = {
2994 .init = udp4_proc_init_net,
2995 .exit = udp4_proc_exit_net,
2998 int __init udp4_proc_init(void)
3000 return register_pernet_subsys(&udp4_net_ops);
3003 void udp4_proc_exit(void)
3005 unregister_pernet_subsys(&udp4_net_ops);
3007 #endif /* CONFIG_PROC_FS */
3009 static __initdata unsigned long uhash_entries;
3010 static int __init set_uhash_entries(char *str)
3017 ret = kstrtoul(str, 0, &uhash_entries);
3021 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3022 uhash_entries = UDP_HTABLE_SIZE_MIN;
3025 __setup("uhash_entries=", set_uhash_entries);
3027 void __init udp_table_init(struct udp_table *table, const char *name)
3031 table->hash = alloc_large_system_hash(name,
3032 2 * sizeof(struct udp_hslot),
3034 21, /* one slot per 2 MB */
3038 UDP_HTABLE_SIZE_MIN,
3041 table->hash2 = table->hash + (table->mask + 1);
3042 for (i = 0; i <= table->mask; i++) {
3043 INIT_HLIST_HEAD(&table->hash[i].head);
3044 table->hash[i].count = 0;
3045 spin_lock_init(&table->hash[i].lock);
3047 for (i = 0; i <= table->mask; i++) {
3048 INIT_HLIST_HEAD(&table->hash2[i].head);
3049 table->hash2[i].count = 0;
3050 spin_lock_init(&table->hash2[i].lock);
3054 u32 udp_flow_hashrnd(void)
3056 static u32 hashrnd __read_mostly;
3058 net_get_random_once(&hashrnd, sizeof(hashrnd));
3062 EXPORT_SYMBOL(udp_flow_hashrnd);
3064 static void __udp_sysctl_init(struct net *net)
3066 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3067 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3069 #ifdef CONFIG_NET_L3_MASTER_DEV
3070 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3074 static int __net_init udp_sysctl_init(struct net *net)
3076 __udp_sysctl_init(net);
3080 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3081 .init = udp_sysctl_init,
3084 void __init udp_init(void)
3086 unsigned long limit;
3089 udp_table_init(&udp_table, "UDP");
3090 limit = nr_free_buffer_pages() / 8;
3091 limit = max(limit, 128UL);
3092 sysctl_udp_mem[0] = limit / 4 * 3;
3093 sysctl_udp_mem[1] = limit;
3094 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3096 __udp_sysctl_init(&init_net);
3098 /* 16 spinlocks per cpu */
3099 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3100 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3103 panic("UDP: failed to alloc udp_busylocks\n");
3104 for (i = 0; i < (1U << udp_busylocks_log); i++)
3105 spin_lock_init(udp_busylocks + i);
3107 if (register_pernet_subsys(&udp_sysctl_ops))
3108 panic("UDP: failed to init sysctl parameters.\n");