2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The User Datagram Protocol (UDP).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
80 #define pr_fmt(fmt) "UDP: " fmt
82 #include <linux/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/memblock.h>
85 #include <linux/highmem.h>
86 #include <linux/swap.h>
87 #include <linux/types.h>
88 #include <linux/fcntl.h>
89 #include <linux/module.h>
90 #include <linux/socket.h>
91 #include <linux/sockios.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
95 #include <linux/errno.h>
96 #include <linux/timer.h>
98 #include <linux/inet.h>
99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/ip_tunnels.h>
109 #include <net/route.h>
110 #include <net/checksum.h>
111 #include <net/xfrm.h>
112 #include <trace/events/udp.h>
113 #include <linux/static_key.h>
114 #include <trace/events/skb.h>
115 #include <net/busy_poll.h>
116 #include "udp_impl.h"
117 #include <net/sock_reuseport.h>
118 #include <net/addrconf.h>
119 #include <net/udp_tunnel.h>
121 struct udp_table udp_table __read_mostly;
122 EXPORT_SYMBOL(udp_table);
124 long sysctl_udp_mem[3] __read_mostly;
125 EXPORT_SYMBOL(sysctl_udp_mem);
127 atomic_long_t udp_memory_allocated;
128 EXPORT_SYMBOL(udp_memory_allocated);
130 #define MAX_UDP_PORTS 65536
131 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
133 /* IPCB reference means this can not be used from early demux */
134 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
136 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
137 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
138 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
144 static int udp_lib_lport_inuse(struct net *net, __u16 num,
145 const struct udp_hslot *hslot,
146 unsigned long *bitmap,
147 struct sock *sk, unsigned int log)
150 kuid_t uid = sock_i_uid(sk);
152 sk_for_each(sk2, &hslot->head) {
153 if (net_eq(sock_net(sk2), net) &&
155 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
156 (!sk2->sk_reuse || !sk->sk_reuse) &&
157 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
158 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
159 inet_rcv_saddr_equal(sk, sk2, true)) {
160 if (sk2->sk_reuseport && sk->sk_reuseport &&
161 !rcu_access_pointer(sk->sk_reuseport_cb) &&
162 uid_eq(uid, sock_i_uid(sk2))) {
168 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
177 * Note: we still hold spinlock of primary hash chain, so no other writer
178 * can insert/delete a socket with local_port == num
180 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
181 struct udp_hslot *hslot2,
185 kuid_t uid = sock_i_uid(sk);
188 spin_lock(&hslot2->lock);
189 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
190 if (net_eq(sock_net(sk2), net) &&
192 (udp_sk(sk2)->udp_port_hash == num) &&
193 (!sk2->sk_reuse || !sk->sk_reuse) &&
194 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
195 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
196 inet_rcv_saddr_equal(sk, sk2, true)) {
197 if (sk2->sk_reuseport && sk->sk_reuseport &&
198 !rcu_access_pointer(sk->sk_reuseport_cb) &&
199 uid_eq(uid, sock_i_uid(sk2))) {
207 spin_unlock(&hslot2->lock);
211 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
213 struct net *net = sock_net(sk);
214 kuid_t uid = sock_i_uid(sk);
217 sk_for_each(sk2, &hslot->head) {
218 if (net_eq(sock_net(sk2), net) &&
220 sk2->sk_family == sk->sk_family &&
221 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
222 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
223 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
224 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
225 inet_rcv_saddr_equal(sk, sk2, false)) {
226 return reuseport_add_sock(sk, sk2,
227 inet_rcv_saddr_any(sk));
231 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
235 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
237 * @sk: socket struct in question
238 * @snum: port number to look up
239 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
242 int udp_lib_get_port(struct sock *sk, unsigned short snum,
243 unsigned int hash2_nulladdr)
245 struct udp_hslot *hslot, *hslot2;
246 struct udp_table *udptable = sk->sk_prot->h.udp_table;
248 struct net *net = sock_net(sk);
251 int low, high, remaining;
253 unsigned short first, last;
254 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
256 inet_get_local_port_range(net, &low, &high);
257 remaining = (high - low) + 1;
259 rand = prandom_u32();
260 first = reciprocal_scale(rand, remaining) + low;
262 * force rand to be an odd multiple of UDP_HTABLE_SIZE
264 rand = (rand | 1) * (udptable->mask + 1);
265 last = first + udptable->mask + 1;
267 hslot = udp_hashslot(udptable, net, first);
268 bitmap_zero(bitmap, PORTS_PER_CHAIN);
269 spin_lock_bh(&hslot->lock);
270 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
275 * Iterate on all possible values of snum for this hash.
276 * Using steps of an odd multiple of UDP_HTABLE_SIZE
277 * give us randomization and full range coverage.
280 if (low <= snum && snum <= high &&
281 !test_bit(snum >> udptable->log, bitmap) &&
282 !inet_is_local_reserved_port(net, snum))
285 } while (snum != first);
286 spin_unlock_bh(&hslot->lock);
288 } while (++first != last);
291 hslot = udp_hashslot(udptable, net, snum);
292 spin_lock_bh(&hslot->lock);
293 if (hslot->count > 10) {
295 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
297 slot2 &= udptable->mask;
298 hash2_nulladdr &= udptable->mask;
300 hslot2 = udp_hashslot2(udptable, slot2);
301 if (hslot->count < hslot2->count)
302 goto scan_primary_hash;
304 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
305 if (!exist && (hash2_nulladdr != slot2)) {
306 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
307 exist = udp_lib_lport_inuse2(net, snum, hslot2,
316 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
320 inet_sk(sk)->inet_num = snum;
321 udp_sk(sk)->udp_port_hash = snum;
322 udp_sk(sk)->udp_portaddr_hash ^= snum;
323 if (sk_unhashed(sk)) {
324 if (sk->sk_reuseport &&
325 udp_reuseport_add_sock(sk, hslot)) {
326 inet_sk(sk)->inet_num = 0;
327 udp_sk(sk)->udp_port_hash = 0;
328 udp_sk(sk)->udp_portaddr_hash ^= snum;
332 sk_add_node_rcu(sk, &hslot->head);
334 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
336 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
337 spin_lock(&hslot2->lock);
338 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
339 sk->sk_family == AF_INET6)
340 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
343 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
346 spin_unlock(&hslot2->lock);
348 sock_set_flag(sk, SOCK_RCU_FREE);
351 spin_unlock_bh(&hslot->lock);
355 EXPORT_SYMBOL(udp_lib_get_port);
357 int udp_v4_get_port(struct sock *sk, unsigned short snum)
359 unsigned int hash2_nulladdr =
360 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
361 unsigned int hash2_partial =
362 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
364 /* precompute partial secondary hash */
365 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
366 return udp_lib_get_port(sk, snum, hash2_nulladdr);
369 static int compute_score(struct sock *sk, struct net *net,
370 __be32 saddr, __be16 sport,
371 __be32 daddr, unsigned short hnum,
372 int dif, int sdif, bool exact_dif)
375 struct inet_sock *inet;
378 if (!net_eq(sock_net(sk), net) ||
379 udp_sk(sk)->udp_port_hash != hnum ||
383 if (sk->sk_rcv_saddr != daddr)
386 score = (sk->sk_family == PF_INET) ? 2 : 1;
389 if (inet->inet_daddr) {
390 if (inet->inet_daddr != saddr)
395 if (inet->inet_dport) {
396 if (inet->inet_dport != sport)
401 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
407 if (sk->sk_incoming_cpu == raw_smp_processor_id())
412 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
413 const __u16 lport, const __be32 faddr,
416 static u32 udp_ehash_secret __read_mostly;
418 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
420 return __inet_ehashfn(laddr, lport, faddr, fport,
421 udp_ehash_secret + net_hash_mix(net));
424 /* called with rcu_read_lock() */
425 static struct sock *udp4_lib_lookup2(struct net *net,
426 __be32 saddr, __be16 sport,
427 __be32 daddr, unsigned int hnum,
428 int dif, int sdif, bool exact_dif,
429 struct udp_hslot *hslot2,
432 struct sock *sk, *result;
438 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
439 score = compute_score(sk, net, saddr, sport,
440 daddr, hnum, dif, sdif, exact_dif);
441 if (score > badness) {
442 if (sk->sk_reuseport) {
443 hash = udp_ehashfn(net, daddr, hnum,
445 result = reuseport_select_sock(sk, hash, skb,
446 sizeof(struct udphdr));
457 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
458 * harder than this. -DaveM
460 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
461 __be16 sport, __be32 daddr, __be16 dport, int dif,
462 int sdif, struct udp_table *udptable, struct sk_buff *skb)
465 unsigned short hnum = ntohs(dport);
466 unsigned int hash2, slot2;
467 struct udp_hslot *hslot2;
468 bool exact_dif = udp_lib_exact_dif_match(net, skb);
470 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
471 slot2 = hash2 & udptable->mask;
472 hslot2 = &udptable->hash2[slot2];
474 result = udp4_lib_lookup2(net, saddr, sport,
475 daddr, hnum, dif, sdif,
476 exact_dif, hslot2, skb);
478 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
479 slot2 = hash2 & udptable->mask;
480 hslot2 = &udptable->hash2[slot2];
482 result = udp4_lib_lookup2(net, saddr, sport,
483 htonl(INADDR_ANY), hnum, dif, sdif,
484 exact_dif, hslot2, skb);
486 if (unlikely(IS_ERR(result)))
490 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
492 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
493 __be16 sport, __be16 dport,
494 struct udp_table *udptable)
496 const struct iphdr *iph = ip_hdr(skb);
498 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
499 iph->daddr, dport, inet_iif(skb),
500 inet_sdif(skb), udptable, skb);
503 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
504 __be16 sport, __be16 dport)
506 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
508 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
510 /* Must be called under rcu_read_lock().
511 * Does increment socket refcount.
513 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
514 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
515 __be32 daddr, __be16 dport, int dif)
519 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
520 dif, 0, &udp_table, NULL);
521 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
525 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
528 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
529 __be16 loc_port, __be32 loc_addr,
530 __be16 rmt_port, __be32 rmt_addr,
531 int dif, int sdif, unsigned short hnum)
533 struct inet_sock *inet = inet_sk(sk);
535 if (!net_eq(sock_net(sk), net) ||
536 udp_sk(sk)->udp_port_hash != hnum ||
537 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
538 (inet->inet_dport != rmt_port && inet->inet_dport) ||
539 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
540 ipv6_only_sock(sk) ||
541 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
542 sk->sk_bound_dev_if != sdif))
544 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
549 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
550 void udp_encap_enable(void)
552 static_branch_inc(&udp_encap_needed_key);
554 EXPORT_SYMBOL(udp_encap_enable);
556 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
557 * through error handlers in encapsulations looking for a match.
559 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
563 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
564 int (*handler)(struct sk_buff *skb, u32 info);
566 if (!iptun_encaps[i])
568 handler = rcu_dereference(iptun_encaps[i]->err_handler);
569 if (handler && !handler(skb, info))
576 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
577 * reversing source and destination port: this will match tunnels that force the
578 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
579 * lwtunnels might actually break this assumption by being configured with
580 * different destination ports on endpoints, in this case we won't be able to
581 * trace ICMP messages back to them.
583 * If this doesn't match any socket, probe tunnels with arbitrary destination
584 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
585 * we've sent packets to won't necessarily match the local destination port.
587 * Then ask the tunnel implementation to match the error against a valid
590 * Return an error if we can't find a match, the socket if we need further
591 * processing, zero otherwise.
593 static struct sock *__udp4_lib_err_encap(struct net *net,
594 const struct iphdr *iph,
596 struct udp_table *udptable,
597 struct sk_buff *skb, u32 info)
599 int network_offset, transport_offset;
602 network_offset = skb_network_offset(skb);
603 transport_offset = skb_transport_offset(skb);
605 /* Network header needs to point to the outer IPv4 header inside ICMP */
606 skb_reset_network_header(skb);
608 /* Transport header needs to point to the UDP header */
609 skb_set_transport_header(skb, iph->ihl << 2);
611 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
612 iph->saddr, uh->dest, skb->dev->ifindex, 0,
615 int (*lookup)(struct sock *sk, struct sk_buff *skb);
616 struct udp_sock *up = udp_sk(sk);
618 lookup = READ_ONCE(up->encap_err_lookup);
619 if (!lookup || lookup(sk, skb))
624 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
626 skb_set_transport_header(skb, transport_offset);
627 skb_set_network_header(skb, network_offset);
633 * This routine is called by the ICMP module when it gets some
634 * sort of error condition. If err < 0 then the socket should
635 * be closed and the error returned to the user. If err > 0
636 * it's just the icmp type << 8 | icmp code.
637 * Header points to the ip header of the error packet. We move
638 * on past this. Then (as it used to claim before adjustment)
639 * header points to the first 8 bytes of the udp header. We need
640 * to find the appropriate port.
643 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
645 struct inet_sock *inet;
646 const struct iphdr *iph = (const struct iphdr *)skb->data;
647 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
648 const int type = icmp_hdr(skb)->type;
649 const int code = icmp_hdr(skb)->code;
654 struct net *net = dev_net(skb->dev);
656 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
657 iph->saddr, uh->source, skb->dev->ifindex,
658 inet_sdif(skb), udptable, NULL);
660 /* No socket for error: try tunnels before discarding */
661 sk = ERR_PTR(-ENOENT);
662 if (static_branch_unlikely(&udp_encap_needed_key)) {
663 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
670 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
683 case ICMP_TIME_EXCEEDED:
686 case ICMP_SOURCE_QUENCH:
688 case ICMP_PARAMETERPROB:
692 case ICMP_DEST_UNREACH:
693 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
694 ipv4_sk_update_pmtu(skb, sk, info);
695 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
703 if (code <= NR_ICMP_UNREACH) {
704 harderr = icmp_err_convert[code].fatal;
705 err = icmp_err_convert[code].errno;
709 ipv4_sk_redirect(skb, sk);
714 * RFC1122: OK. Passes ICMP errors back to application, as per
718 /* ...not for tunnels though: we don't have a sending socket */
721 if (!inet->recverr) {
722 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
725 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
728 sk->sk_error_report(sk);
733 int udp_err(struct sk_buff *skb, u32 info)
735 return __udp4_lib_err(skb, info, &udp_table);
739 * Throw away all pending data and cancel the corking. Socket is locked.
741 void udp_flush_pending_frames(struct sock *sk)
743 struct udp_sock *up = udp_sk(sk);
748 ip_flush_pending_frames(sk);
751 EXPORT_SYMBOL(udp_flush_pending_frames);
754 * udp4_hwcsum - handle outgoing HW checksumming
755 * @skb: sk_buff containing the filled-in UDP header
756 * (checksum field must be zeroed out)
757 * @src: source IP address
758 * @dst: destination IP address
760 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
762 struct udphdr *uh = udp_hdr(skb);
763 int offset = skb_transport_offset(skb);
764 int len = skb->len - offset;
768 if (!skb_has_frag_list(skb)) {
770 * Only one fragment on the socket.
772 skb->csum_start = skb_transport_header(skb) - skb->head;
773 skb->csum_offset = offsetof(struct udphdr, check);
774 uh->check = ~csum_tcpudp_magic(src, dst, len,
777 struct sk_buff *frags;
780 * HW-checksum won't work as there are two or more
781 * fragments on the socket so that all csums of sk_buffs
784 skb_walk_frags(skb, frags) {
785 csum = csum_add(csum, frags->csum);
789 csum = skb_checksum(skb, offset, hlen, csum);
790 skb->ip_summed = CHECKSUM_NONE;
792 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
794 uh->check = CSUM_MANGLED_0;
797 EXPORT_SYMBOL_GPL(udp4_hwcsum);
799 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
800 * for the simple case like when setting the checksum for a UDP tunnel.
802 void udp_set_csum(bool nocheck, struct sk_buff *skb,
803 __be32 saddr, __be32 daddr, int len)
805 struct udphdr *uh = udp_hdr(skb);
809 } else if (skb_is_gso(skb)) {
810 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
811 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
813 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
815 uh->check = CSUM_MANGLED_0;
817 skb->ip_summed = CHECKSUM_PARTIAL;
818 skb->csum_start = skb_transport_header(skb) - skb->head;
819 skb->csum_offset = offsetof(struct udphdr, check);
820 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
823 EXPORT_SYMBOL(udp_set_csum);
825 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
826 struct inet_cork *cork)
828 struct sock *sk = skb->sk;
829 struct inet_sock *inet = inet_sk(sk);
832 int is_udplite = IS_UDPLITE(sk);
833 int offset = skb_transport_offset(skb);
834 int len = skb->len - offset;
838 * Create a UDP header
841 uh->source = inet->inet_sport;
842 uh->dest = fl4->fl4_dport;
843 uh->len = htons(len);
846 if (cork->gso_size) {
847 const int hlen = skb_network_header_len(skb) +
848 sizeof(struct udphdr);
850 if (hlen + cork->gso_size > cork->fragsize)
852 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS)
854 if (sk->sk_no_check_tx)
856 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
857 dst_xfrm(skb_dst(skb)))
860 skb_shinfo(skb)->gso_size = cork->gso_size;
861 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
862 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh),
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, sk->sk_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 static void udp_set_dev_scratch(struct sk_buff *skb)
1302 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1304 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1305 scratch->_tsize_state = skb->truesize;
1306 #if BITS_PER_LONG == 64
1307 scratch->len = skb->len;
1308 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1309 scratch->is_linear = !skb_is_nonlinear(skb);
1311 /* all head states execept sp (dst, sk, nf) are always cleared by
1312 * udp_rcv() and we need to preserve secpath, if present, to eventually
1313 * process IP_CMSG_PASSSEC at recvmsg() time
1315 if (likely(!skb_sec_path(skb)))
1316 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1319 static int udp_skb_truesize(struct sk_buff *skb)
1321 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1324 static bool udp_skb_has_head_state(struct sk_buff *skb)
1326 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1329 /* fully reclaim rmem/fwd memory allocated for skb */
1330 static void udp_rmem_release(struct sock *sk, int size, int partial,
1331 bool rx_queue_lock_held)
1333 struct udp_sock *up = udp_sk(sk);
1334 struct sk_buff_head *sk_queue;
1337 if (likely(partial)) {
1338 up->forward_deficit += size;
1339 size = up->forward_deficit;
1340 if (size < (sk->sk_rcvbuf >> 2))
1343 size += up->forward_deficit;
1345 up->forward_deficit = 0;
1347 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1348 * if the called don't held it already
1350 sk_queue = &sk->sk_receive_queue;
1351 if (!rx_queue_lock_held)
1352 spin_lock(&sk_queue->lock);
1355 sk->sk_forward_alloc += size;
1356 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1357 sk->sk_forward_alloc -= amt;
1360 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1362 atomic_sub(size, &sk->sk_rmem_alloc);
1364 /* this can save us from acquiring the rx queue lock on next receive */
1365 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1367 if (!rx_queue_lock_held)
1368 spin_unlock(&sk_queue->lock);
1371 /* Note: called with reader_queue.lock held.
1372 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1373 * This avoids a cache line miss while receive_queue lock is held.
1374 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1376 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1378 prefetch(&skb->data);
1379 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1381 EXPORT_SYMBOL(udp_skb_destructor);
1383 /* as above, but the caller held the rx queue lock, too */
1384 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1386 prefetch(&skb->data);
1387 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1390 /* Idea of busylocks is to let producers grab an extra spinlock
1391 * to relieve pressure on the receive_queue spinlock shared by consumer.
1392 * Under flood, this means that only one producer can be in line
1393 * trying to acquire the receive_queue spinlock.
1394 * These busylock can be allocated on a per cpu manner, instead of a
1395 * per socket one (that would consume a cache line per socket)
1397 static int udp_busylocks_log __read_mostly;
1398 static spinlock_t *udp_busylocks __read_mostly;
1400 static spinlock_t *busylock_acquire(void *ptr)
1404 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1409 static void busylock_release(spinlock_t *busy)
1415 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1417 struct sk_buff_head *list = &sk->sk_receive_queue;
1418 int rmem, delta, amt, err = -ENOMEM;
1419 spinlock_t *busy = NULL;
1422 /* try to avoid the costly atomic add/sub pair when the receive
1423 * queue is full; always allow at least a packet
1425 rmem = atomic_read(&sk->sk_rmem_alloc);
1426 if (rmem > sk->sk_rcvbuf)
1429 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1430 * having linear skbs :
1431 * - Reduce memory overhead and thus increase receive queue capacity
1432 * - Less cache line misses at copyout() time
1433 * - Less work at consume_skb() (less alien page frag freeing)
1435 if (rmem > (sk->sk_rcvbuf >> 1)) {
1438 busy = busylock_acquire(sk);
1440 size = skb->truesize;
1441 udp_set_dev_scratch(skb);
1443 /* we drop only if the receive buf is full and the receive
1444 * queue contains some other skb
1446 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1447 if (rmem > (size + sk->sk_rcvbuf))
1450 spin_lock(&list->lock);
1451 if (size >= sk->sk_forward_alloc) {
1452 amt = sk_mem_pages(size);
1453 delta = amt << SK_MEM_QUANTUM_SHIFT;
1454 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1456 spin_unlock(&list->lock);
1460 sk->sk_forward_alloc += delta;
1463 sk->sk_forward_alloc -= size;
1465 /* no need to setup a destructor, we will explicitly release the
1466 * forward allocated memory on dequeue
1468 sock_skb_set_dropcount(sk, skb);
1470 __skb_queue_tail(list, skb);
1471 spin_unlock(&list->lock);
1473 if (!sock_flag(sk, SOCK_DEAD))
1474 sk->sk_data_ready(sk);
1476 busylock_release(busy);
1480 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1483 atomic_inc(&sk->sk_drops);
1484 busylock_release(busy);
1487 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1489 void udp_destruct_sock(struct sock *sk)
1491 /* reclaim completely the forward allocated memory */
1492 struct udp_sock *up = udp_sk(sk);
1493 unsigned int total = 0;
1494 struct sk_buff *skb;
1496 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1497 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1498 total += skb->truesize;
1501 udp_rmem_release(sk, total, 0, true);
1503 inet_sock_destruct(sk);
1505 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1507 int udp_init_sock(struct sock *sk)
1509 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1510 sk->sk_destruct = udp_destruct_sock;
1513 EXPORT_SYMBOL_GPL(udp_init_sock);
1515 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1517 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1518 bool slow = lock_sock_fast(sk);
1520 sk_peek_offset_bwd(sk, len);
1521 unlock_sock_fast(sk, slow);
1524 if (!skb_unref(skb))
1527 /* In the more common cases we cleared the head states previously,
1528 * see __udp_queue_rcv_skb().
1530 if (unlikely(udp_skb_has_head_state(skb)))
1531 skb_release_head_state(skb);
1532 __consume_stateless_skb(skb);
1534 EXPORT_SYMBOL_GPL(skb_consume_udp);
1536 static struct sk_buff *__first_packet_length(struct sock *sk,
1537 struct sk_buff_head *rcvq,
1540 struct sk_buff *skb;
1542 while ((skb = skb_peek(rcvq)) != NULL) {
1543 if (udp_lib_checksum_complete(skb)) {
1544 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1546 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1548 atomic_inc(&sk->sk_drops);
1549 __skb_unlink(skb, rcvq);
1550 *total += skb->truesize;
1553 /* the csum related bits could be changed, refresh
1556 udp_set_dev_scratch(skb);
1564 * first_packet_length - return length of first packet in receive queue
1567 * Drops all bad checksum frames, until a valid one is found.
1568 * Returns the length of found skb, or -1 if none is found.
1570 static int first_packet_length(struct sock *sk)
1572 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1573 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1574 struct sk_buff *skb;
1578 spin_lock_bh(&rcvq->lock);
1579 skb = __first_packet_length(sk, rcvq, &total);
1580 if (!skb && !skb_queue_empty(sk_queue)) {
1581 spin_lock(&sk_queue->lock);
1582 skb_queue_splice_tail_init(sk_queue, rcvq);
1583 spin_unlock(&sk_queue->lock);
1585 skb = __first_packet_length(sk, rcvq, &total);
1587 res = skb ? skb->len : -1;
1589 udp_rmem_release(sk, total, 1, false);
1590 spin_unlock_bh(&rcvq->lock);
1595 * IOCTL requests applicable to the UDP protocol
1598 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1603 int amount = sk_wmem_alloc_get(sk);
1605 return put_user(amount, (int __user *)arg);
1610 int amount = max_t(int, 0, first_packet_length(sk));
1612 return put_user(amount, (int __user *)arg);
1616 return -ENOIOCTLCMD;
1621 EXPORT_SYMBOL(udp_ioctl);
1623 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1624 int noblock, int *peeked, int *off, int *err)
1626 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1627 struct sk_buff_head *queue;
1628 struct sk_buff *last;
1632 queue = &udp_sk(sk)->reader_queue;
1633 flags |= noblock ? MSG_DONTWAIT : 0;
1634 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1636 struct sk_buff *skb;
1638 error = sock_error(sk);
1645 spin_lock_bh(&queue->lock);
1646 skb = __skb_try_recv_from_queue(sk, queue, flags,
1651 spin_unlock_bh(&queue->lock);
1655 if (skb_queue_empty(sk_queue)) {
1656 spin_unlock_bh(&queue->lock);
1660 /* refill the reader queue and walk it again
1661 * keep both queues locked to avoid re-acquiring
1662 * the sk_receive_queue lock if fwd memory scheduling
1665 spin_lock(&sk_queue->lock);
1666 skb_queue_splice_tail_init(sk_queue, queue);
1668 skb = __skb_try_recv_from_queue(sk, queue, flags,
1669 udp_skb_dtor_locked,
1672 spin_unlock(&sk_queue->lock);
1673 spin_unlock_bh(&queue->lock);
1678 if (!sk_can_busy_loop(sk))
1681 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1682 } while (!skb_queue_empty(sk_queue));
1684 /* sk_queue is empty, reader_queue may contain peeked packets */
1686 !__skb_wait_for_more_packets(sk, &error, &timeo,
1687 (struct sk_buff *)sk_queue));
1692 EXPORT_SYMBOL(__skb_recv_udp);
1695 * This should be easy, if there is something there we
1696 * return it, otherwise we block.
1699 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1700 int flags, int *addr_len)
1702 struct inet_sock *inet = inet_sk(sk);
1703 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1704 struct sk_buff *skb;
1705 unsigned int ulen, copied;
1706 int peeked, peeking, off;
1708 int is_udplite = IS_UDPLITE(sk);
1709 bool checksum_valid = false;
1711 if (flags & MSG_ERRQUEUE)
1712 return ip_recv_error(sk, msg, len, addr_len);
1715 peeking = flags & MSG_PEEK;
1716 off = sk_peek_offset(sk, flags);
1717 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1721 ulen = udp_skb_len(skb);
1723 if (copied > ulen - off)
1724 copied = ulen - off;
1725 else if (copied < ulen)
1726 msg->msg_flags |= MSG_TRUNC;
1729 * If checksum is needed at all, try to do it while copying the
1730 * data. If the data is truncated, or if we only want a partial
1731 * coverage checksum (UDP-Lite), do it before the copy.
1734 if (copied < ulen || peeking ||
1735 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1736 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1737 !__udp_lib_checksum_complete(skb);
1738 if (!checksum_valid)
1742 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1743 if (udp_skb_is_linear(skb))
1744 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1746 err = skb_copy_datagram_msg(skb, off, msg, copied);
1748 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1754 if (unlikely(err)) {
1756 atomic_inc(&sk->sk_drops);
1757 UDP_INC_STATS(sock_net(sk),
1758 UDP_MIB_INERRORS, is_udplite);
1765 UDP_INC_STATS(sock_net(sk),
1766 UDP_MIB_INDATAGRAMS, is_udplite);
1768 sock_recv_ts_and_drops(msg, sk, skb);
1770 /* Copy the address. */
1772 sin->sin_family = AF_INET;
1773 sin->sin_port = udp_hdr(skb)->source;
1774 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1775 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1776 *addr_len = sizeof(*sin);
1779 if (udp_sk(sk)->gro_enabled)
1780 udp_cmsg_recv(msg, sk, skb);
1782 if (inet->cmsg_flags)
1783 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1786 if (flags & MSG_TRUNC)
1789 skb_consume_udp(sk, skb, peeking ? -err : err);
1793 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1794 udp_skb_destructor)) {
1795 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1796 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1800 /* starting over for a new packet, but check if we need to yield */
1802 msg->msg_flags &= ~MSG_TRUNC;
1806 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1808 /* This check is replicated from __ip4_datagram_connect() and
1809 * intended to prevent BPF program called below from accessing bytes
1810 * that are out of the bound specified by user in addr_len.
1812 if (addr_len < sizeof(struct sockaddr_in))
1815 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1817 EXPORT_SYMBOL(udp_pre_connect);
1819 int __udp_disconnect(struct sock *sk, int flags)
1821 struct inet_sock *inet = inet_sk(sk);
1823 * 1003.1g - break association.
1826 sk->sk_state = TCP_CLOSE;
1827 inet->inet_daddr = 0;
1828 inet->inet_dport = 0;
1829 sock_rps_reset_rxhash(sk);
1830 sk->sk_bound_dev_if = 0;
1831 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1832 inet_reset_saddr(sk);
1834 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1835 sk->sk_prot->unhash(sk);
1836 inet->inet_sport = 0;
1841 EXPORT_SYMBOL(__udp_disconnect);
1843 int udp_disconnect(struct sock *sk, int flags)
1846 __udp_disconnect(sk, flags);
1850 EXPORT_SYMBOL(udp_disconnect);
1852 void udp_lib_unhash(struct sock *sk)
1854 if (sk_hashed(sk)) {
1855 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1856 struct udp_hslot *hslot, *hslot2;
1858 hslot = udp_hashslot(udptable, sock_net(sk),
1859 udp_sk(sk)->udp_port_hash);
1860 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1862 spin_lock_bh(&hslot->lock);
1863 if (rcu_access_pointer(sk->sk_reuseport_cb))
1864 reuseport_detach_sock(sk);
1865 if (sk_del_node_init_rcu(sk)) {
1867 inet_sk(sk)->inet_num = 0;
1868 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1870 spin_lock(&hslot2->lock);
1871 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1873 spin_unlock(&hslot2->lock);
1875 spin_unlock_bh(&hslot->lock);
1878 EXPORT_SYMBOL(udp_lib_unhash);
1881 * inet_rcv_saddr was changed, we must rehash secondary hash
1883 void udp_lib_rehash(struct sock *sk, u16 newhash)
1885 if (sk_hashed(sk)) {
1886 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1887 struct udp_hslot *hslot, *hslot2, *nhslot2;
1889 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1890 nhslot2 = udp_hashslot2(udptable, newhash);
1891 udp_sk(sk)->udp_portaddr_hash = newhash;
1893 if (hslot2 != nhslot2 ||
1894 rcu_access_pointer(sk->sk_reuseport_cb)) {
1895 hslot = udp_hashslot(udptable, sock_net(sk),
1896 udp_sk(sk)->udp_port_hash);
1897 /* we must lock primary chain too */
1898 spin_lock_bh(&hslot->lock);
1899 if (rcu_access_pointer(sk->sk_reuseport_cb))
1900 reuseport_detach_sock(sk);
1902 if (hslot2 != nhslot2) {
1903 spin_lock(&hslot2->lock);
1904 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1906 spin_unlock(&hslot2->lock);
1908 spin_lock(&nhslot2->lock);
1909 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1912 spin_unlock(&nhslot2->lock);
1915 spin_unlock_bh(&hslot->lock);
1919 EXPORT_SYMBOL(udp_lib_rehash);
1921 static void udp_v4_rehash(struct sock *sk)
1923 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1924 inet_sk(sk)->inet_rcv_saddr,
1925 inet_sk(sk)->inet_num);
1926 udp_lib_rehash(sk, new_hash);
1929 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1933 if (inet_sk(sk)->inet_daddr) {
1934 sock_rps_save_rxhash(sk, skb);
1935 sk_mark_napi_id(sk, skb);
1936 sk_incoming_cpu_update(sk);
1938 sk_mark_napi_id_once(sk, skb);
1941 rc = __udp_enqueue_schedule_skb(sk, skb);
1943 int is_udplite = IS_UDPLITE(sk);
1945 /* Note that an ENOMEM error is charged twice */
1947 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1949 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1951 trace_udp_fail_queue_rcv_skb(rc, sk);
1961 * >0: "udp encap" protocol resubmission
1963 * Note that in the success and error cases, the skb is assumed to
1964 * have either been requeued or freed.
1966 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
1968 struct udp_sock *up = udp_sk(sk);
1969 int is_udplite = IS_UDPLITE(sk);
1972 * Charge it to the socket, dropping if the queue is full.
1974 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1978 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
1979 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1982 * This is an encapsulation socket so pass the skb to
1983 * the socket's udp_encap_rcv() hook. Otherwise, just
1984 * fall through and pass this up the UDP socket.
1985 * up->encap_rcv() returns the following value:
1986 * =0 if skb was successfully passed to the encap
1987 * handler or was discarded by it.
1988 * >0 if skb should be passed on to UDP.
1989 * <0 if skb should be resubmitted as proto -N
1992 /* if we're overly short, let UDP handle it */
1993 encap_rcv = READ_ONCE(up->encap_rcv);
1997 /* Verify checksum before giving to encap */
1998 if (udp_lib_checksum_complete(skb))
2001 ret = encap_rcv(sk, skb);
2003 __UDP_INC_STATS(sock_net(sk),
2004 UDP_MIB_INDATAGRAMS,
2010 /* FALLTHROUGH -- it's a UDP Packet */
2014 * UDP-Lite specific tests, ignored on UDP sockets
2016 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2019 * MIB statistics other than incrementing the error count are
2020 * disabled for the following two types of errors: these depend
2021 * on the application settings, not on the functioning of the
2022 * protocol stack as such.
2024 * RFC 3828 here recommends (sec 3.3): "There should also be a
2025 * way ... to ... at least let the receiving application block
2026 * delivery of packets with coverage values less than a value
2027 * provided by the application."
2029 if (up->pcrlen == 0) { /* full coverage was set */
2030 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2031 UDP_SKB_CB(skb)->cscov, skb->len);
2034 /* The next case involves violating the min. coverage requested
2035 * by the receiver. This is subtle: if receiver wants x and x is
2036 * greater than the buffersize/MTU then receiver will complain
2037 * that it wants x while sender emits packets of smaller size y.
2038 * Therefore the above ...()->partial_cov statement is essential.
2040 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2041 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2042 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2047 prefetch(&sk->sk_rmem_alloc);
2048 if (rcu_access_pointer(sk->sk_filter) &&
2049 udp_lib_checksum_complete(skb))
2052 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2055 udp_csum_pull_header(skb);
2057 ipv4_pktinfo_prepare(sk, skb);
2058 return __udp_queue_rcv_skb(sk, skb);
2061 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2063 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2064 atomic_inc(&sk->sk_drops);
2069 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2071 struct sk_buff *next, *segs;
2074 if (likely(!udp_unexpected_gso(sk, skb)))
2075 return udp_queue_rcv_one_skb(sk, skb);
2077 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET);
2078 __skb_push(skb, -skb_mac_offset(skb));
2079 segs = udp_rcv_segment(sk, skb, true);
2080 for (skb = segs; skb; skb = next) {
2082 __skb_pull(skb, skb_transport_offset(skb));
2083 ret = udp_queue_rcv_one_skb(sk, skb);
2085 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2090 /* For TCP sockets, sk_rx_dst is protected by socket lock
2091 * For UDP, we use xchg() to guard against concurrent changes.
2093 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2095 struct dst_entry *old;
2097 if (dst_hold_safe(dst)) {
2098 old = xchg(&sk->sk_rx_dst, dst);
2104 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2107 * Multicasts and broadcasts go to each listener.
2109 * Note: called only from the BH handler context.
2111 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2113 __be32 saddr, __be32 daddr,
2114 struct udp_table *udptable,
2117 struct sock *sk, *first = NULL;
2118 unsigned short hnum = ntohs(uh->dest);
2119 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2120 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2121 unsigned int offset = offsetof(typeof(*sk), sk_node);
2122 int dif = skb->dev->ifindex;
2123 int sdif = inet_sdif(skb);
2124 struct hlist_node *node;
2125 struct sk_buff *nskb;
2128 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2130 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2132 hslot = &udptable->hash2[hash2];
2133 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2136 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2137 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2138 uh->source, saddr, dif, sdif, hnum))
2145 nskb = skb_clone(skb, GFP_ATOMIC);
2147 if (unlikely(!nskb)) {
2148 atomic_inc(&sk->sk_drops);
2149 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2151 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2155 if (udp_queue_rcv_skb(sk, nskb) > 0)
2159 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2160 if (use_hash2 && hash2 != hash2_any) {
2166 if (udp_queue_rcv_skb(first, skb) > 0)
2170 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2171 proto == IPPROTO_UDPLITE);
2176 /* Initialize UDP checksum. If exited with zero value (success),
2177 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2178 * Otherwise, csum completion requires chacksumming packet body,
2179 * including udp header and folding it to skb->csum.
2181 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2186 UDP_SKB_CB(skb)->partial_cov = 0;
2187 UDP_SKB_CB(skb)->cscov = skb->len;
2189 if (proto == IPPROTO_UDPLITE) {
2190 err = udplite_checksum_init(skb, uh);
2194 if (UDP_SKB_CB(skb)->partial_cov) {
2195 skb->csum = inet_compute_pseudo(skb, proto);
2200 /* Note, we are only interested in != 0 or == 0, thus the
2203 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2204 inet_compute_pseudo);
2208 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2209 /* If SW calculated the value, we know it's bad */
2210 if (skb->csum_complete_sw)
2213 /* HW says the value is bad. Let's validate that.
2214 * skb->csum is no longer the full packet checksum,
2215 * so don't treat it as such.
2217 skb_checksum_complete_unset(skb);
2223 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2224 * return code conversion for ip layer consumption
2226 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2231 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2232 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2233 inet_compute_pseudo);
2235 ret = udp_queue_rcv_skb(sk, skb);
2237 /* a return value > 0 means to resubmit the input, but
2238 * it wants the return to be -protocol, or 0
2246 * All we need to do is get the socket, and then do a checksum.
2249 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2254 unsigned short ulen;
2255 struct rtable *rt = skb_rtable(skb);
2256 __be32 saddr, daddr;
2257 struct net *net = dev_net(skb->dev);
2260 * Validate the packet.
2262 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2263 goto drop; /* No space for header. */
2266 ulen = ntohs(uh->len);
2267 saddr = ip_hdr(skb)->saddr;
2268 daddr = ip_hdr(skb)->daddr;
2270 if (ulen > skb->len)
2273 if (proto == IPPROTO_UDP) {
2274 /* UDP validates ulen. */
2275 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2280 if (udp4_csum_init(skb, uh, proto))
2283 sk = skb_steal_sock(skb);
2285 struct dst_entry *dst = skb_dst(skb);
2288 if (unlikely(sk->sk_rx_dst != dst))
2289 udp_sk_rx_dst_set(sk, dst);
2291 ret = udp_unicast_rcv_skb(sk, skb, uh);
2296 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2297 return __udp4_lib_mcast_deliver(net, skb, uh,
2298 saddr, daddr, udptable, proto);
2300 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2302 return udp_unicast_rcv_skb(sk, skb, uh);
2304 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2308 /* No socket. Drop packet silently, if checksum is wrong */
2309 if (udp_lib_checksum_complete(skb))
2312 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2313 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2316 * Hmm. We got an UDP packet to a port to which we
2317 * don't wanna listen. Ignore it.
2323 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2324 proto == IPPROTO_UDPLITE ? "Lite" : "",
2325 &saddr, ntohs(uh->source),
2327 &daddr, ntohs(uh->dest));
2332 * RFC1122: OK. Discards the bad packet silently (as far as
2333 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2335 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2336 proto == IPPROTO_UDPLITE ? "Lite" : "",
2337 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2339 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2341 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2346 /* We can only early demux multicast if there is a single matching socket.
2347 * If more than one socket found returns NULL
2349 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2350 __be16 loc_port, __be32 loc_addr,
2351 __be16 rmt_port, __be32 rmt_addr,
2354 struct sock *sk, *result;
2355 unsigned short hnum = ntohs(loc_port);
2356 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2357 struct udp_hslot *hslot = &udp_table.hash[slot];
2359 /* Do not bother scanning a too big list */
2360 if (hslot->count > 10)
2364 sk_for_each_rcu(sk, &hslot->head) {
2365 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2366 rmt_port, rmt_addr, dif, sdif, hnum)) {
2376 /* For unicast we should only early demux connected sockets or we can
2377 * break forwarding setups. The chains here can be long so only check
2378 * if the first socket is an exact match and if not move on.
2380 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2381 __be16 loc_port, __be32 loc_addr,
2382 __be16 rmt_port, __be32 rmt_addr,
2385 unsigned short hnum = ntohs(loc_port);
2386 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2387 unsigned int slot2 = hash2 & udp_table.mask;
2388 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2389 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2390 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2393 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2394 if (INET_MATCH(sk, net, acookie, rmt_addr,
2395 loc_addr, ports, dif, sdif))
2397 /* Only check first socket in chain */
2403 int udp_v4_early_demux(struct sk_buff *skb)
2405 struct net *net = dev_net(skb->dev);
2406 struct in_device *in_dev = NULL;
2407 const struct iphdr *iph;
2408 const struct udphdr *uh;
2409 struct sock *sk = NULL;
2410 struct dst_entry *dst;
2411 int dif = skb->dev->ifindex;
2412 int sdif = inet_sdif(skb);
2415 /* validate the packet */
2416 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2422 if (skb->pkt_type == PACKET_MULTICAST) {
2423 in_dev = __in_dev_get_rcu(skb->dev);
2428 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2433 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2434 uh->source, iph->saddr,
2436 } else if (skb->pkt_type == PACKET_HOST) {
2437 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2438 uh->source, iph->saddr, dif, sdif);
2441 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2445 skb->destructor = sock_efree;
2446 dst = READ_ONCE(sk->sk_rx_dst);
2449 dst = dst_check(dst, 0);
2453 /* set noref for now.
2454 * any place which wants to hold dst has to call
2457 skb_dst_set_noref(skb, dst);
2459 /* for unconnected multicast sockets we need to validate
2460 * the source on each packet
2462 if (!inet_sk(sk)->inet_daddr && in_dev)
2463 return ip_mc_validate_source(skb, iph->daddr,
2464 iph->saddr, iph->tos,
2465 skb->dev, in_dev, &itag);
2470 int udp_rcv(struct sk_buff *skb)
2472 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2475 void udp_destroy_sock(struct sock *sk)
2477 struct udp_sock *up = udp_sk(sk);
2478 bool slow = lock_sock_fast(sk);
2479 udp_flush_pending_frames(sk);
2480 unlock_sock_fast(sk, slow);
2481 if (static_branch_unlikely(&udp_encap_needed_key)) {
2482 if (up->encap_type) {
2483 void (*encap_destroy)(struct sock *sk);
2484 encap_destroy = READ_ONCE(up->encap_destroy);
2488 if (up->encap_enabled)
2489 static_branch_dec(&udp_encap_needed_key);
2494 * Socket option code for UDP
2496 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2497 char __user *optval, unsigned int optlen,
2498 int (*push_pending_frames)(struct sock *))
2500 struct udp_sock *up = udp_sk(sk);
2503 int is_udplite = IS_UDPLITE(sk);
2505 if (optlen < sizeof(int))
2508 if (get_user(val, (int __user *)optval))
2511 valbool = val ? 1 : 0;
2520 push_pending_frames(sk);
2528 case UDP_ENCAP_ESPINUDP:
2529 case UDP_ENCAP_ESPINUDP_NON_IKE:
2530 up->encap_rcv = xfrm4_udp_encap_rcv;
2532 case UDP_ENCAP_L2TPINUDP:
2533 up->encap_type = val;
2535 udp_tunnel_encap_enable(sk->sk_socket);
2544 case UDP_NO_CHECK6_TX:
2545 up->no_check6_tx = valbool;
2548 case UDP_NO_CHECK6_RX:
2549 up->no_check6_rx = valbool;
2553 if (val < 0 || val > USHRT_MAX)
2561 udp_tunnel_encap_enable(sk->sk_socket);
2562 up->gro_enabled = valbool;
2567 * UDP-Lite's partial checksum coverage (RFC 3828).
2569 /* The sender sets actual checksum coverage length via this option.
2570 * The case coverage > packet length is handled by send module. */
2571 case UDPLITE_SEND_CSCOV:
2572 if (!is_udplite) /* Disable the option on UDP sockets */
2573 return -ENOPROTOOPT;
2574 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2576 else if (val > USHRT_MAX)
2579 up->pcflag |= UDPLITE_SEND_CC;
2582 /* The receiver specifies a minimum checksum coverage value. To make
2583 * sense, this should be set to at least 8 (as done below). If zero is
2584 * used, this again means full checksum coverage. */
2585 case UDPLITE_RECV_CSCOV:
2586 if (!is_udplite) /* Disable the option on UDP sockets */
2587 return -ENOPROTOOPT;
2588 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2590 else if (val > USHRT_MAX)
2593 up->pcflag |= UDPLITE_RECV_CC;
2603 EXPORT_SYMBOL(udp_lib_setsockopt);
2605 int udp_setsockopt(struct sock *sk, int level, int optname,
2606 char __user *optval, unsigned int optlen)
2608 if (level == SOL_UDP || level == SOL_UDPLITE)
2609 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2610 udp_push_pending_frames);
2611 return ip_setsockopt(sk, level, optname, optval, optlen);
2614 #ifdef CONFIG_COMPAT
2615 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2616 char __user *optval, unsigned int optlen)
2618 if (level == SOL_UDP || level == SOL_UDPLITE)
2619 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2620 udp_push_pending_frames);
2621 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2625 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2626 char __user *optval, int __user *optlen)
2628 struct udp_sock *up = udp_sk(sk);
2631 if (get_user(len, optlen))
2634 len = min_t(unsigned int, len, sizeof(int));
2645 val = up->encap_type;
2648 case UDP_NO_CHECK6_TX:
2649 val = up->no_check6_tx;
2652 case UDP_NO_CHECK6_RX:
2653 val = up->no_check6_rx;
2660 /* The following two cannot be changed on UDP sockets, the return is
2661 * always 0 (which corresponds to the full checksum coverage of UDP). */
2662 case UDPLITE_SEND_CSCOV:
2666 case UDPLITE_RECV_CSCOV:
2671 return -ENOPROTOOPT;
2674 if (put_user(len, optlen))
2676 if (copy_to_user(optval, &val, len))
2680 EXPORT_SYMBOL(udp_lib_getsockopt);
2682 int udp_getsockopt(struct sock *sk, int level, int optname,
2683 char __user *optval, int __user *optlen)
2685 if (level == SOL_UDP || level == SOL_UDPLITE)
2686 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2687 return ip_getsockopt(sk, level, optname, optval, optlen);
2690 #ifdef CONFIG_COMPAT
2691 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2692 char __user *optval, int __user *optlen)
2694 if (level == SOL_UDP || level == SOL_UDPLITE)
2695 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2696 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2700 * udp_poll - wait for a UDP event.
2701 * @file - file struct
2703 * @wait - poll table
2705 * This is same as datagram poll, except for the special case of
2706 * blocking sockets. If application is using a blocking fd
2707 * and a packet with checksum error is in the queue;
2708 * then it could get return from select indicating data available
2709 * but then block when reading it. Add special case code
2710 * to work around these arguably broken applications.
2712 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2714 __poll_t mask = datagram_poll(file, sock, wait);
2715 struct sock *sk = sock->sk;
2717 if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2718 mask |= EPOLLIN | EPOLLRDNORM;
2720 /* Check for false positives due to checksum errors */
2721 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2722 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2723 mask &= ~(EPOLLIN | EPOLLRDNORM);
2728 EXPORT_SYMBOL(udp_poll);
2730 int udp_abort(struct sock *sk, int err)
2735 sk->sk_error_report(sk);
2736 __udp_disconnect(sk, 0);
2742 EXPORT_SYMBOL_GPL(udp_abort);
2744 struct proto udp_prot = {
2746 .owner = THIS_MODULE,
2747 .close = udp_lib_close,
2748 .pre_connect = udp_pre_connect,
2749 .connect = ip4_datagram_connect,
2750 .disconnect = udp_disconnect,
2752 .init = udp_init_sock,
2753 .destroy = udp_destroy_sock,
2754 .setsockopt = udp_setsockopt,
2755 .getsockopt = udp_getsockopt,
2756 .sendmsg = udp_sendmsg,
2757 .recvmsg = udp_recvmsg,
2758 .sendpage = udp_sendpage,
2759 .release_cb = ip4_datagram_release_cb,
2760 .hash = udp_lib_hash,
2761 .unhash = udp_lib_unhash,
2762 .rehash = udp_v4_rehash,
2763 .get_port = udp_v4_get_port,
2764 .memory_allocated = &udp_memory_allocated,
2765 .sysctl_mem = sysctl_udp_mem,
2766 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2767 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2768 .obj_size = sizeof(struct udp_sock),
2769 .h.udp_table = &udp_table,
2770 #ifdef CONFIG_COMPAT
2771 .compat_setsockopt = compat_udp_setsockopt,
2772 .compat_getsockopt = compat_udp_getsockopt,
2774 .diag_destroy = udp_abort,
2776 EXPORT_SYMBOL(udp_prot);
2778 /* ------------------------------------------------------------------------ */
2779 #ifdef CONFIG_PROC_FS
2781 static struct sock *udp_get_first(struct seq_file *seq, int start)
2784 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2785 struct udp_iter_state *state = seq->private;
2786 struct net *net = seq_file_net(seq);
2788 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2790 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2792 if (hlist_empty(&hslot->head))
2795 spin_lock_bh(&hslot->lock);
2796 sk_for_each(sk, &hslot->head) {
2797 if (!net_eq(sock_net(sk), net))
2799 if (sk->sk_family == afinfo->family)
2802 spin_unlock_bh(&hslot->lock);
2809 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2811 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2812 struct udp_iter_state *state = seq->private;
2813 struct net *net = seq_file_net(seq);
2817 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2820 if (state->bucket <= afinfo->udp_table->mask)
2821 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2822 return udp_get_first(seq, state->bucket + 1);
2827 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2829 struct sock *sk = udp_get_first(seq, 0);
2832 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2834 return pos ? NULL : sk;
2837 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2839 struct udp_iter_state *state = seq->private;
2840 state->bucket = MAX_UDP_PORTS;
2842 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2844 EXPORT_SYMBOL(udp_seq_start);
2846 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2850 if (v == SEQ_START_TOKEN)
2851 sk = udp_get_idx(seq, 0);
2853 sk = udp_get_next(seq, v);
2858 EXPORT_SYMBOL(udp_seq_next);
2860 void udp_seq_stop(struct seq_file *seq, void *v)
2862 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2863 struct udp_iter_state *state = seq->private;
2865 if (state->bucket <= afinfo->udp_table->mask)
2866 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2868 EXPORT_SYMBOL(udp_seq_stop);
2870 /* ------------------------------------------------------------------------ */
2871 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2874 struct inet_sock *inet = inet_sk(sp);
2875 __be32 dest = inet->inet_daddr;
2876 __be32 src = inet->inet_rcv_saddr;
2877 __u16 destp = ntohs(inet->inet_dport);
2878 __u16 srcp = ntohs(inet->inet_sport);
2880 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2881 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2882 bucket, src, srcp, dest, destp, sp->sk_state,
2883 sk_wmem_alloc_get(sp),
2886 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2888 refcount_read(&sp->sk_refcnt), sp,
2889 atomic_read(&sp->sk_drops));
2892 int udp4_seq_show(struct seq_file *seq, void *v)
2894 seq_setwidth(seq, 127);
2895 if (v == SEQ_START_TOKEN)
2896 seq_puts(seq, " sl local_address rem_address st tx_queue "
2897 "rx_queue tr tm->when retrnsmt uid timeout "
2898 "inode ref pointer drops");
2900 struct udp_iter_state *state = seq->private;
2902 udp4_format_sock(v, seq, state->bucket);
2908 const struct seq_operations udp_seq_ops = {
2909 .start = udp_seq_start,
2910 .next = udp_seq_next,
2911 .stop = udp_seq_stop,
2912 .show = udp4_seq_show,
2914 EXPORT_SYMBOL(udp_seq_ops);
2916 static struct udp_seq_afinfo udp4_seq_afinfo = {
2918 .udp_table = &udp_table,
2921 static int __net_init udp4_proc_init_net(struct net *net)
2923 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2924 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2929 static void __net_exit udp4_proc_exit_net(struct net *net)
2931 remove_proc_entry("udp", net->proc_net);
2934 static struct pernet_operations udp4_net_ops = {
2935 .init = udp4_proc_init_net,
2936 .exit = udp4_proc_exit_net,
2939 int __init udp4_proc_init(void)
2941 return register_pernet_subsys(&udp4_net_ops);
2944 void udp4_proc_exit(void)
2946 unregister_pernet_subsys(&udp4_net_ops);
2948 #endif /* CONFIG_PROC_FS */
2950 static __initdata unsigned long uhash_entries;
2951 static int __init set_uhash_entries(char *str)
2958 ret = kstrtoul(str, 0, &uhash_entries);
2962 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2963 uhash_entries = UDP_HTABLE_SIZE_MIN;
2966 __setup("uhash_entries=", set_uhash_entries);
2968 void __init udp_table_init(struct udp_table *table, const char *name)
2972 table->hash = alloc_large_system_hash(name,
2973 2 * sizeof(struct udp_hslot),
2975 21, /* one slot per 2 MB */
2979 UDP_HTABLE_SIZE_MIN,
2982 table->hash2 = table->hash + (table->mask + 1);
2983 for (i = 0; i <= table->mask; i++) {
2984 INIT_HLIST_HEAD(&table->hash[i].head);
2985 table->hash[i].count = 0;
2986 spin_lock_init(&table->hash[i].lock);
2988 for (i = 0; i <= table->mask; i++) {
2989 INIT_HLIST_HEAD(&table->hash2[i].head);
2990 table->hash2[i].count = 0;
2991 spin_lock_init(&table->hash2[i].lock);
2995 u32 udp_flow_hashrnd(void)
2997 static u32 hashrnd __read_mostly;
2999 net_get_random_once(&hashrnd, sizeof(hashrnd));
3003 EXPORT_SYMBOL(udp_flow_hashrnd);
3005 static void __udp_sysctl_init(struct net *net)
3007 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3008 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3010 #ifdef CONFIG_NET_L3_MASTER_DEV
3011 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3015 static int __net_init udp_sysctl_init(struct net *net)
3017 __udp_sysctl_init(net);
3021 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3022 .init = udp_sysctl_init,
3025 void __init udp_init(void)
3027 unsigned long limit;
3030 udp_table_init(&udp_table, "UDP");
3031 limit = nr_free_buffer_pages() / 8;
3032 limit = max(limit, 128UL);
3033 sysctl_udp_mem[0] = limit / 4 * 3;
3034 sysctl_udp_mem[1] = limit;
3035 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3037 __udp_sysctl_init(&init_net);
3039 /* 16 spinlocks per cpu */
3040 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3041 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3044 panic("UDP: failed to alloc udp_busylocks\n");
3045 for (i = 0; i < (1U << udp_busylocks_log); i++)
3046 spin_lock_init(udp_busylocks + i);
3048 if (register_pernet_subsys(&udp_sysctl_ops))
3049 panic("UDP: failed to init sysctl parameters.\n");