1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * The User Datagram Protocol (UDP).
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
72 * James Chapman : Add L2TP encapsulation type.
75 #define pr_fmt(fmt) "UDP: " fmt
77 #include <linux/bpf-cgroup.h>
78 #include <linux/uaccess.h>
79 #include <asm/ioctls.h>
80 #include <linux/memblock.h>
81 #include <linux/highmem.h>
82 #include <linux/types.h>
83 #include <linux/fcntl.h>
84 #include <linux/module.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/igmp.h>
88 #include <linux/inetdevice.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
93 #include <linux/inet.h>
94 #include <linux/netdevice.h>
95 #include <linux/slab.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/inet_hashtables.h>
103 #include <net/ip_tunnels.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <trace/events/udp.h>
108 #include <linux/static_key.h>
109 #include <linux/btf_ids.h>
110 #include <trace/events/skb.h>
111 #include <net/busy_poll.h>
112 #include "udp_impl.h"
113 #include <net/sock_reuseport.h>
114 #include <net/addrconf.h>
115 #include <net/udp_tunnel.h>
116 #if IS_ENABLED(CONFIG_IPV6)
117 #include <net/ipv6_stubs.h>
120 struct udp_table udp_table __read_mostly;
121 EXPORT_SYMBOL(udp_table);
123 long sysctl_udp_mem[3] __read_mostly;
124 EXPORT_SYMBOL(sysctl_udp_mem);
126 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
127 EXPORT_SYMBOL(udp_memory_allocated);
128 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
129 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
131 #define MAX_UDP_PORTS 65536
132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
134 static struct udp_table *udp_get_table_prot(struct sock *sk)
136 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
139 static int udp_lib_lport_inuse(struct net *net, __u16 num,
140 const struct udp_hslot *hslot,
141 unsigned long *bitmap,
142 struct sock *sk, unsigned int log)
145 kuid_t uid = sock_i_uid(sk);
147 sk_for_each(sk2, &hslot->head) {
148 if (net_eq(sock_net(sk2), net) &&
150 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
151 (!sk2->sk_reuse || !sk->sk_reuse) &&
152 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
153 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
154 inet_rcv_saddr_equal(sk, sk2, true)) {
155 if (sk2->sk_reuseport && sk->sk_reuseport &&
156 !rcu_access_pointer(sk->sk_reuseport_cb) &&
157 uid_eq(uid, sock_i_uid(sk2))) {
163 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
172 * Note: we still hold spinlock of primary hash chain, so no other writer
173 * can insert/delete a socket with local_port == num
175 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
176 struct udp_hslot *hslot2,
180 kuid_t uid = sock_i_uid(sk);
183 spin_lock(&hslot2->lock);
184 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
185 if (net_eq(sock_net(sk2), net) &&
187 (udp_sk(sk2)->udp_port_hash == num) &&
188 (!sk2->sk_reuse || !sk->sk_reuse) &&
189 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
190 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
191 inet_rcv_saddr_equal(sk, sk2, true)) {
192 if (sk2->sk_reuseport && sk->sk_reuseport &&
193 !rcu_access_pointer(sk->sk_reuseport_cb) &&
194 uid_eq(uid, sock_i_uid(sk2))) {
202 spin_unlock(&hslot2->lock);
206 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
208 struct net *net = sock_net(sk);
209 kuid_t uid = sock_i_uid(sk);
212 sk_for_each(sk2, &hslot->head) {
213 if (net_eq(sock_net(sk2), net) &&
215 sk2->sk_family == sk->sk_family &&
216 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
217 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
218 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
219 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
220 inet_rcv_saddr_equal(sk, sk2, false)) {
221 return reuseport_add_sock(sk, sk2,
222 inet_rcv_saddr_any(sk));
226 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
230 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
232 * @sk: socket struct in question
233 * @snum: port number to look up
234 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237 int udp_lib_get_port(struct sock *sk, unsigned short snum,
238 unsigned int hash2_nulladdr)
240 struct udp_table *udptable = udp_get_table_prot(sk);
241 struct udp_hslot *hslot, *hslot2;
242 struct net *net = sock_net(sk);
243 int error = -EADDRINUSE;
246 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
247 unsigned short first, last;
248 int low, high, remaining;
251 inet_get_local_port_range(net, &low, &high);
252 remaining = (high - low) + 1;
254 rand = get_random_u32();
255 first = reciprocal_scale(rand, remaining) + low;
257 * force rand to be an odd multiple of UDP_HTABLE_SIZE
259 rand = (rand | 1) * (udptable->mask + 1);
260 last = first + udptable->mask + 1;
262 hslot = udp_hashslot(udptable, net, first);
263 bitmap_zero(bitmap, PORTS_PER_CHAIN);
264 spin_lock_bh(&hslot->lock);
265 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
270 * Iterate on all possible values of snum for this hash.
271 * Using steps of an odd multiple of UDP_HTABLE_SIZE
272 * give us randomization and full range coverage.
275 if (low <= snum && snum <= high &&
276 !test_bit(snum >> udptable->log, bitmap) &&
277 !inet_is_local_reserved_port(net, snum))
280 } while (snum != first);
281 spin_unlock_bh(&hslot->lock);
283 } while (++first != last);
286 hslot = udp_hashslot(udptable, net, snum);
287 spin_lock_bh(&hslot->lock);
288 if (hslot->count > 10) {
290 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
292 slot2 &= udptable->mask;
293 hash2_nulladdr &= udptable->mask;
295 hslot2 = udp_hashslot2(udptable, slot2);
296 if (hslot->count < hslot2->count)
297 goto scan_primary_hash;
299 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
300 if (!exist && (hash2_nulladdr != slot2)) {
301 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
302 exist = udp_lib_lport_inuse2(net, snum, hslot2,
311 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
315 inet_sk(sk)->inet_num = snum;
316 udp_sk(sk)->udp_port_hash = snum;
317 udp_sk(sk)->udp_portaddr_hash ^= snum;
318 if (sk_unhashed(sk)) {
319 if (sk->sk_reuseport &&
320 udp_reuseport_add_sock(sk, hslot)) {
321 inet_sk(sk)->inet_num = 0;
322 udp_sk(sk)->udp_port_hash = 0;
323 udp_sk(sk)->udp_portaddr_hash ^= snum;
327 sk_add_node_rcu(sk, &hslot->head);
329 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
331 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
332 spin_lock(&hslot2->lock);
333 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
334 sk->sk_family == AF_INET6)
335 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
338 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
341 spin_unlock(&hslot2->lock);
343 sock_set_flag(sk, SOCK_RCU_FREE);
346 spin_unlock_bh(&hslot->lock);
350 EXPORT_SYMBOL(udp_lib_get_port);
352 int udp_v4_get_port(struct sock *sk, unsigned short snum)
354 unsigned int hash2_nulladdr =
355 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
356 unsigned int hash2_partial =
357 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
359 /* precompute partial secondary hash */
360 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
361 return udp_lib_get_port(sk, snum, hash2_nulladdr);
364 static int compute_score(struct sock *sk, struct net *net,
365 __be32 saddr, __be16 sport,
366 __be32 daddr, unsigned short hnum,
370 struct inet_sock *inet;
373 if (!net_eq(sock_net(sk), net) ||
374 udp_sk(sk)->udp_port_hash != hnum ||
378 if (sk->sk_rcv_saddr != daddr)
381 score = (sk->sk_family == PF_INET) ? 2 : 1;
384 if (inet->inet_daddr) {
385 if (inet->inet_daddr != saddr)
390 if (inet->inet_dport) {
391 if (inet->inet_dport != sport)
396 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
400 if (sk->sk_bound_dev_if)
403 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
408 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
409 const __u16 lport, const __be32 faddr,
412 static u32 udp_ehash_secret __read_mostly;
414 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
416 return __inet_ehashfn(laddr, lport, faddr, fport,
417 udp_ehash_secret + net_hash_mix(net));
420 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
422 __be32 saddr, __be16 sport,
423 __be32 daddr, unsigned short hnum)
425 struct sock *reuse_sk = NULL;
428 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
429 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
430 reuse_sk = reuseport_select_sock(sk, hash, skb,
431 sizeof(struct udphdr));
436 /* called with rcu_read_lock() */
437 static struct sock *udp4_lib_lookup2(struct net *net,
438 __be32 saddr, __be16 sport,
439 __be32 daddr, unsigned int hnum,
441 struct udp_hslot *hslot2,
444 struct sock *sk, *result;
449 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
450 score = compute_score(sk, net, saddr, sport,
451 daddr, hnum, dif, sdif);
452 if (score > badness) {
453 result = lookup_reuseport(net, sk, skb,
454 saddr, sport, daddr, hnum);
455 /* Fall back to scoring if group has connections */
456 if (result && !reuseport_has_conns(sk))
459 result = result ? : sk;
466 static struct sock *udp4_lookup_run_bpf(struct net *net,
467 struct udp_table *udptable,
469 __be32 saddr, __be16 sport,
470 __be32 daddr, u16 hnum, const int dif)
472 struct sock *sk, *reuse_sk;
475 if (udptable != net->ipv4.udp_table)
476 return NULL; /* only UDP is supported */
478 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport,
479 daddr, hnum, dif, &sk);
480 if (no_reuseport || IS_ERR_OR_NULL(sk))
483 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
489 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
490 * harder than this. -DaveM
492 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
493 __be16 sport, __be32 daddr, __be16 dport, int dif,
494 int sdif, struct udp_table *udptable, struct sk_buff *skb)
496 unsigned short hnum = ntohs(dport);
497 unsigned int hash2, slot2;
498 struct udp_hslot *hslot2;
499 struct sock *result, *sk;
501 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
502 slot2 = hash2 & udptable->mask;
503 hslot2 = &udptable->hash2[slot2];
505 /* Lookup connected or non-wildcard socket */
506 result = udp4_lib_lookup2(net, saddr, sport,
507 daddr, hnum, dif, sdif,
509 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
512 /* Lookup redirect from BPF */
513 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
514 sk = udp4_lookup_run_bpf(net, udptable, skb,
515 saddr, sport, daddr, hnum, dif);
522 /* Got non-wildcard socket or error on first lookup */
526 /* Lookup wildcard sockets */
527 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
528 slot2 = hash2 & udptable->mask;
529 hslot2 = &udptable->hash2[slot2];
531 result = udp4_lib_lookup2(net, saddr, sport,
532 htonl(INADDR_ANY), hnum, dif, sdif,
539 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
541 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
542 __be16 sport, __be16 dport,
543 struct udp_table *udptable)
545 const struct iphdr *iph = ip_hdr(skb);
547 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
548 iph->daddr, dport, inet_iif(skb),
549 inet_sdif(skb), udptable, skb);
552 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
553 __be16 sport, __be16 dport)
555 const struct iphdr *iph = ip_hdr(skb);
556 struct net *net = dev_net(skb->dev);
558 return __udp4_lib_lookup(net, iph->saddr, sport,
559 iph->daddr, dport, inet_iif(skb),
560 inet_sdif(skb), net->ipv4.udp_table, NULL);
563 /* Must be called under rcu_read_lock().
564 * Does increment socket refcount.
566 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
567 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
568 __be32 daddr, __be16 dport, int dif)
572 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
573 dif, 0, net->ipv4.udp_table, NULL);
574 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
578 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
581 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
582 __be16 loc_port, __be32 loc_addr,
583 __be16 rmt_port, __be32 rmt_addr,
584 int dif, int sdif, unsigned short hnum)
586 struct inet_sock *inet = inet_sk(sk);
588 if (!net_eq(sock_net(sk), net) ||
589 udp_sk(sk)->udp_port_hash != hnum ||
590 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
591 (inet->inet_dport != rmt_port && inet->inet_dport) ||
592 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
593 ipv6_only_sock(sk) ||
594 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
596 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
601 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
602 void udp_encap_enable(void)
604 static_branch_inc(&udp_encap_needed_key);
606 EXPORT_SYMBOL(udp_encap_enable);
608 void udp_encap_disable(void)
610 static_branch_dec(&udp_encap_needed_key);
612 EXPORT_SYMBOL(udp_encap_disable);
614 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
615 * through error handlers in encapsulations looking for a match.
617 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
621 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
622 int (*handler)(struct sk_buff *skb, u32 info);
623 const struct ip_tunnel_encap_ops *encap;
625 encap = rcu_dereference(iptun_encaps[i]);
628 handler = encap->err_handler;
629 if (handler && !handler(skb, info))
636 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
637 * reversing source and destination port: this will match tunnels that force the
638 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
639 * lwtunnels might actually break this assumption by being configured with
640 * different destination ports on endpoints, in this case we won't be able to
641 * trace ICMP messages back to them.
643 * If this doesn't match any socket, probe tunnels with arbitrary destination
644 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
645 * we've sent packets to won't necessarily match the local destination port.
647 * Then ask the tunnel implementation to match the error against a valid
650 * Return an error if we can't find a match, the socket if we need further
651 * processing, zero otherwise.
653 static struct sock *__udp4_lib_err_encap(struct net *net,
654 const struct iphdr *iph,
656 struct udp_table *udptable,
658 struct sk_buff *skb, u32 info)
660 int (*lookup)(struct sock *sk, struct sk_buff *skb);
661 int network_offset, transport_offset;
664 network_offset = skb_network_offset(skb);
665 transport_offset = skb_transport_offset(skb);
667 /* Network header needs to point to the outer IPv4 header inside ICMP */
668 skb_reset_network_header(skb);
670 /* Transport header needs to point to the UDP header */
671 skb_set_transport_header(skb, iph->ihl << 2);
676 lookup = READ_ONCE(up->encap_err_lookup);
677 if (lookup && lookup(sk, skb))
683 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
684 iph->saddr, uh->dest, skb->dev->ifindex, 0,
689 lookup = READ_ONCE(up->encap_err_lookup);
690 if (!lookup || lookup(sk, skb))
696 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
698 skb_set_transport_header(skb, transport_offset);
699 skb_set_network_header(skb, network_offset);
705 * This routine is called by the ICMP module when it gets some
706 * sort of error condition. If err < 0 then the socket should
707 * be closed and the error returned to the user. If err > 0
708 * it's just the icmp type << 8 | icmp code.
709 * Header points to the ip header of the error packet. We move
710 * on past this. Then (as it used to claim before adjustment)
711 * header points to the first 8 bytes of the udp header. We need
712 * to find the appropriate port.
715 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
717 struct inet_sock *inet;
718 const struct iphdr *iph = (const struct iphdr *)skb->data;
719 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
720 const int type = icmp_hdr(skb)->type;
721 const int code = icmp_hdr(skb)->code;
726 struct net *net = dev_net(skb->dev);
728 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
729 iph->saddr, uh->source, skb->dev->ifindex,
730 inet_sdif(skb), udptable, NULL);
732 if (!sk || udp_sk(sk)->encap_type) {
733 /* No socket for error: try tunnels before discarding */
734 if (static_branch_unlikely(&udp_encap_needed_key)) {
735 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
740 sk = ERR_PTR(-ENOENT);
743 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
756 case ICMP_TIME_EXCEEDED:
759 case ICMP_SOURCE_QUENCH:
761 case ICMP_PARAMETERPROB:
765 case ICMP_DEST_UNREACH:
766 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
767 ipv4_sk_update_pmtu(skb, sk, info);
768 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
776 if (code <= NR_ICMP_UNREACH) {
777 harderr = icmp_err_convert[code].fatal;
778 err = icmp_err_convert[code].errno;
782 ipv4_sk_redirect(skb, sk);
787 * RFC1122: OK. Passes ICMP errors back to application, as per
791 /* ...not for tunnels though: we don't have a sending socket */
792 if (udp_sk(sk)->encap_err_rcv)
793 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
797 if (!inet->recverr) {
798 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
801 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
809 int udp_err(struct sk_buff *skb, u32 info)
811 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
815 * Throw away all pending data and cancel the corking. Socket is locked.
817 void udp_flush_pending_frames(struct sock *sk)
819 struct udp_sock *up = udp_sk(sk);
824 ip_flush_pending_frames(sk);
827 EXPORT_SYMBOL(udp_flush_pending_frames);
830 * udp4_hwcsum - handle outgoing HW checksumming
831 * @skb: sk_buff containing the filled-in UDP header
832 * (checksum field must be zeroed out)
833 * @src: source IP address
834 * @dst: destination IP address
836 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
838 struct udphdr *uh = udp_hdr(skb);
839 int offset = skb_transport_offset(skb);
840 int len = skb->len - offset;
844 if (!skb_has_frag_list(skb)) {
846 * Only one fragment on the socket.
848 skb->csum_start = skb_transport_header(skb) - skb->head;
849 skb->csum_offset = offsetof(struct udphdr, check);
850 uh->check = ~csum_tcpudp_magic(src, dst, len,
853 struct sk_buff *frags;
856 * HW-checksum won't work as there are two or more
857 * fragments on the socket so that all csums of sk_buffs
860 skb_walk_frags(skb, frags) {
861 csum = csum_add(csum, frags->csum);
865 csum = skb_checksum(skb, offset, hlen, csum);
866 skb->ip_summed = CHECKSUM_NONE;
868 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
870 uh->check = CSUM_MANGLED_0;
873 EXPORT_SYMBOL_GPL(udp4_hwcsum);
875 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
876 * for the simple case like when setting the checksum for a UDP tunnel.
878 void udp_set_csum(bool nocheck, struct sk_buff *skb,
879 __be32 saddr, __be32 daddr, int len)
881 struct udphdr *uh = udp_hdr(skb);
885 } else if (skb_is_gso(skb)) {
886 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
887 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
889 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
891 uh->check = CSUM_MANGLED_0;
893 skb->ip_summed = CHECKSUM_PARTIAL;
894 skb->csum_start = skb_transport_header(skb) - skb->head;
895 skb->csum_offset = offsetof(struct udphdr, check);
896 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
899 EXPORT_SYMBOL(udp_set_csum);
901 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
902 struct inet_cork *cork)
904 struct sock *sk = skb->sk;
905 struct inet_sock *inet = inet_sk(sk);
908 int is_udplite = IS_UDPLITE(sk);
909 int offset = skb_transport_offset(skb);
910 int len = skb->len - offset;
911 int datalen = len - sizeof(*uh);
915 * Create a UDP header
918 uh->source = inet->inet_sport;
919 uh->dest = fl4->fl4_dport;
920 uh->len = htons(len);
923 if (cork->gso_size) {
924 const int hlen = skb_network_header_len(skb) +
925 sizeof(struct udphdr);
927 if (hlen + cork->gso_size > cork->fragsize) {
931 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
935 if (sk->sk_no_check_tx) {
939 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
940 dst_xfrm(skb_dst(skb))) {
945 if (datalen > cork->gso_size) {
946 skb_shinfo(skb)->gso_size = cork->gso_size;
947 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
948 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
954 if (is_udplite) /* UDP-Lite */
955 csum = udplite_csum(skb);
957 else if (sk->sk_no_check_tx) { /* UDP csum off */
959 skb->ip_summed = CHECKSUM_NONE;
962 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
965 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
969 csum = udp_csum(skb);
971 /* add protocol-dependent pseudo-header */
972 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
973 sk->sk_protocol, csum);
975 uh->check = CSUM_MANGLED_0;
978 err = ip_send_skb(sock_net(sk), skb);
980 if (err == -ENOBUFS && !inet->recverr) {
981 UDP_INC_STATS(sock_net(sk),
982 UDP_MIB_SNDBUFERRORS, is_udplite);
986 UDP_INC_STATS(sock_net(sk),
987 UDP_MIB_OUTDATAGRAMS, is_udplite);
992 * Push out all pending data as one UDP datagram. Socket is locked.
994 int udp_push_pending_frames(struct sock *sk)
996 struct udp_sock *up = udp_sk(sk);
997 struct inet_sock *inet = inet_sk(sk);
998 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1002 skb = ip_finish_skb(sk, fl4);
1006 err = udp_send_skb(skb, fl4, &inet->cork.base);
1013 EXPORT_SYMBOL(udp_push_pending_frames);
1015 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1017 switch (cmsg->cmsg_type) {
1019 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1021 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1028 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1030 struct cmsghdr *cmsg;
1031 bool need_ip = false;
1034 for_each_cmsghdr(cmsg, msg) {
1035 if (!CMSG_OK(msg, cmsg))
1038 if (cmsg->cmsg_level != SOL_UDP) {
1043 err = __udp_cmsg_send(cmsg, gso_size);
1050 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1052 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1054 struct inet_sock *inet = inet_sk(sk);
1055 struct udp_sock *up = udp_sk(sk);
1056 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1057 struct flowi4 fl4_stack;
1060 struct ipcm_cookie ipc;
1061 struct rtable *rt = NULL;
1064 __be32 daddr, faddr, saddr;
1067 int err, is_udplite = IS_UDPLITE(sk);
1068 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1069 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1070 struct sk_buff *skb;
1071 struct ip_options_data opt_copy;
1080 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1083 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1085 fl4 = &inet->cork.fl.u.ip4;
1088 * There are pending frames.
1089 * The socket lock must be held while it's corked.
1092 if (likely(up->pending)) {
1093 if (unlikely(up->pending != AF_INET)) {
1097 goto do_append_data;
1101 ulen += sizeof(struct udphdr);
1104 * Get and verify the address.
1107 if (msg->msg_namelen < sizeof(*usin))
1109 if (usin->sin_family != AF_INET) {
1110 if (usin->sin_family != AF_UNSPEC)
1111 return -EAFNOSUPPORT;
1114 daddr = usin->sin_addr.s_addr;
1115 dport = usin->sin_port;
1119 if (sk->sk_state != TCP_ESTABLISHED)
1120 return -EDESTADDRREQ;
1121 daddr = inet->inet_daddr;
1122 dport = inet->inet_dport;
1123 /* Open fast path for connected socket.
1124 Route will not be used, if at least one option is set.
1129 ipcm_init_sk(&ipc, inet);
1130 ipc.gso_size = READ_ONCE(up->gso_size);
1132 if (msg->msg_controllen) {
1133 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1135 err = ip_cmsg_send(sk, msg, &ipc,
1136 sk->sk_family == AF_INET6);
1137 if (unlikely(err < 0)) {
1146 struct ip_options_rcu *inet_opt;
1149 inet_opt = rcu_dereference(inet->inet_opt);
1151 memcpy(&opt_copy, inet_opt,
1152 sizeof(*inet_opt) + inet_opt->opt.optlen);
1153 ipc.opt = &opt_copy.opt;
1158 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1159 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1160 (struct sockaddr *)usin, &ipc.addr);
1164 if (usin->sin_port == 0) {
1165 /* BPF program set invalid port. Reject it. */
1169 daddr = usin->sin_addr.s_addr;
1170 dport = usin->sin_port;
1175 ipc.addr = faddr = daddr;
1177 if (ipc.opt && ipc.opt->opt.srr) {
1182 faddr = ipc.opt->opt.faddr;
1185 tos = get_rttos(&ipc, inet);
1186 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1187 (msg->msg_flags & MSG_DONTROUTE) ||
1188 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1193 if (ipv4_is_multicast(daddr)) {
1194 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1195 ipc.oif = inet->mc_index;
1197 saddr = inet->mc_addr;
1199 } else if (!ipc.oif) {
1200 ipc.oif = inet->uc_index;
1201 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1202 /* oif is set, packet is to local broadcast and
1203 * uc_index is set. oif is most likely set
1204 * by sk_bound_dev_if. If uc_index != oif check if the
1205 * oif is an L3 master and uc_index is an L3 slave.
1206 * If so, we want to allow the send using the uc_index.
1208 if (ipc.oif != inet->uc_index &&
1209 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1211 ipc.oif = inet->uc_index;
1216 rt = (struct rtable *)sk_dst_check(sk, 0);
1219 struct net *net = sock_net(sk);
1220 __u8 flow_flags = inet_sk_flowi_flags(sk);
1224 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1225 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1227 faddr, saddr, dport, inet->inet_sport,
1230 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1231 rt = ip_route_output_flow(net, fl4, sk);
1235 if (err == -ENETUNREACH)
1236 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1241 if ((rt->rt_flags & RTCF_BROADCAST) &&
1242 !sock_flag(sk, SOCK_BROADCAST))
1245 sk_dst_set(sk, dst_clone(&rt->dst));
1248 if (msg->msg_flags&MSG_CONFIRM)
1254 daddr = ipc.addr = fl4->daddr;
1256 /* Lockless fast path for the non-corking case. */
1258 struct inet_cork cork;
1260 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1261 sizeof(struct udphdr), &ipc, &rt,
1262 &cork, msg->msg_flags);
1264 if (!IS_ERR_OR_NULL(skb))
1265 err = udp_send_skb(skb, fl4, &cork);
1270 if (unlikely(up->pending)) {
1271 /* The socket is already corked while preparing it. */
1272 /* ... which is an evident application bug. --ANK */
1275 net_dbg_ratelimited("socket already corked\n");
1280 * Now cork the socket to pend data.
1282 fl4 = &inet->cork.fl.u.ip4;
1285 fl4->fl4_dport = dport;
1286 fl4->fl4_sport = inet->inet_sport;
1287 up->pending = AF_INET;
1291 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1292 sizeof(struct udphdr), &ipc, &rt,
1293 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1295 udp_flush_pending_frames(sk);
1297 err = udp_push_pending_frames(sk);
1298 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1310 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1311 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1312 * we don't have a good statistic (IpOutDiscards but it can be too many
1313 * things). We could add another new stat but at least for now that
1314 * seems like overkill.
1316 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1317 UDP_INC_STATS(sock_net(sk),
1318 UDP_MIB_SNDBUFERRORS, is_udplite);
1323 if (msg->msg_flags & MSG_PROBE)
1324 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1325 if (!(msg->msg_flags&MSG_PROBE) || len)
1326 goto back_from_confirm;
1330 EXPORT_SYMBOL(udp_sendmsg);
1332 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1333 size_t size, int flags)
1335 struct inet_sock *inet = inet_sk(sk);
1336 struct udp_sock *up = udp_sk(sk);
1339 if (flags & MSG_SENDPAGE_NOTLAST)
1343 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1345 /* Call udp_sendmsg to specify destination address which
1346 * sendpage interface can't pass.
1347 * This will succeed only when the socket is connected.
1349 ret = udp_sendmsg(sk, &msg, 0);
1356 if (unlikely(!up->pending)) {
1359 net_dbg_ratelimited("cork failed\n");
1363 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1364 page, offset, size, flags);
1365 if (ret == -EOPNOTSUPP) {
1367 return sock_no_sendpage(sk->sk_socket, page, offset,
1371 udp_flush_pending_frames(sk);
1376 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1377 ret = udp_push_pending_frames(sk);
1385 #define UDP_SKB_IS_STATELESS 0x80000000
1387 /* all head states (dst, sk, nf conntrack) except skb extensions are
1388 * cleared by udp_rcv().
1390 * We need to preserve secpath, if present, to eventually process
1391 * IP_CMSG_PASSSEC at recvmsg() time.
1393 * Other extensions can be cleared.
1395 static bool udp_try_make_stateless(struct sk_buff *skb)
1397 if (!skb_has_extensions(skb))
1400 if (!secpath_exists(skb)) {
1408 static void udp_set_dev_scratch(struct sk_buff *skb)
1410 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1412 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1413 scratch->_tsize_state = skb->truesize;
1414 #if BITS_PER_LONG == 64
1415 scratch->len = skb->len;
1416 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1417 scratch->is_linear = !skb_is_nonlinear(skb);
1419 if (udp_try_make_stateless(skb))
1420 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1423 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1425 /* We come here after udp_lib_checksum_complete() returned 0.
1426 * This means that __skb_checksum_complete() might have
1427 * set skb->csum_valid to 1.
1428 * On 64bit platforms, we can set csum_unnecessary
1429 * to true, but only if the skb is not shared.
1431 #if BITS_PER_LONG == 64
1432 if (!skb_shared(skb))
1433 udp_skb_scratch(skb)->csum_unnecessary = true;
1437 static int udp_skb_truesize(struct sk_buff *skb)
1439 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1442 static bool udp_skb_has_head_state(struct sk_buff *skb)
1444 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1447 /* fully reclaim rmem/fwd memory allocated for skb */
1448 static void udp_rmem_release(struct sock *sk, int size, int partial,
1449 bool rx_queue_lock_held)
1451 struct udp_sock *up = udp_sk(sk);
1452 struct sk_buff_head *sk_queue;
1455 if (likely(partial)) {
1456 up->forward_deficit += size;
1457 size = up->forward_deficit;
1458 if (size < READ_ONCE(up->forward_threshold) &&
1459 !skb_queue_empty(&up->reader_queue))
1462 size += up->forward_deficit;
1464 up->forward_deficit = 0;
1466 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1467 * if the called don't held it already
1469 sk_queue = &sk->sk_receive_queue;
1470 if (!rx_queue_lock_held)
1471 spin_lock(&sk_queue->lock);
1474 sk->sk_forward_alloc += size;
1475 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1476 sk->sk_forward_alloc -= amt;
1479 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1481 atomic_sub(size, &sk->sk_rmem_alloc);
1483 /* this can save us from acquiring the rx queue lock on next receive */
1484 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1486 if (!rx_queue_lock_held)
1487 spin_unlock(&sk_queue->lock);
1490 /* Note: called with reader_queue.lock held.
1491 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1492 * This avoids a cache line miss while receive_queue lock is held.
1493 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1495 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1497 prefetch(&skb->data);
1498 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1500 EXPORT_SYMBOL(udp_skb_destructor);
1502 /* as above, but the caller held the rx queue lock, too */
1503 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1505 prefetch(&skb->data);
1506 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1509 /* Idea of busylocks is to let producers grab an extra spinlock
1510 * to relieve pressure on the receive_queue spinlock shared by consumer.
1511 * Under flood, this means that only one producer can be in line
1512 * trying to acquire the receive_queue spinlock.
1513 * These busylock can be allocated on a per cpu manner, instead of a
1514 * per socket one (that would consume a cache line per socket)
1516 static int udp_busylocks_log __read_mostly;
1517 static spinlock_t *udp_busylocks __read_mostly;
1519 static spinlock_t *busylock_acquire(void *ptr)
1523 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1528 static void busylock_release(spinlock_t *busy)
1534 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1536 struct sk_buff_head *list = &sk->sk_receive_queue;
1537 int rmem, delta, amt, err = -ENOMEM;
1538 spinlock_t *busy = NULL;
1541 /* try to avoid the costly atomic add/sub pair when the receive
1542 * queue is full; always allow at least a packet
1544 rmem = atomic_read(&sk->sk_rmem_alloc);
1545 if (rmem > sk->sk_rcvbuf)
1548 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1549 * having linear skbs :
1550 * - Reduce memory overhead and thus increase receive queue capacity
1551 * - Less cache line misses at copyout() time
1552 * - Less work at consume_skb() (less alien page frag freeing)
1554 if (rmem > (sk->sk_rcvbuf >> 1)) {
1557 busy = busylock_acquire(sk);
1559 size = skb->truesize;
1560 udp_set_dev_scratch(skb);
1562 /* we drop only if the receive buf is full and the receive
1563 * queue contains some other skb
1565 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1566 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1569 spin_lock(&list->lock);
1570 if (size >= sk->sk_forward_alloc) {
1571 amt = sk_mem_pages(size);
1572 delta = amt << PAGE_SHIFT;
1573 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1575 spin_unlock(&list->lock);
1579 sk->sk_forward_alloc += delta;
1582 sk->sk_forward_alloc -= size;
1584 /* no need to setup a destructor, we will explicitly release the
1585 * forward allocated memory on dequeue
1587 sock_skb_set_dropcount(sk, skb);
1589 __skb_queue_tail(list, skb);
1590 spin_unlock(&list->lock);
1592 if (!sock_flag(sk, SOCK_DEAD))
1593 sk->sk_data_ready(sk);
1595 busylock_release(busy);
1599 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1602 atomic_inc(&sk->sk_drops);
1603 busylock_release(busy);
1606 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1608 void udp_destruct_common(struct sock *sk)
1610 /* reclaim completely the forward allocated memory */
1611 struct udp_sock *up = udp_sk(sk);
1612 unsigned int total = 0;
1613 struct sk_buff *skb;
1615 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1616 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1617 total += skb->truesize;
1620 udp_rmem_release(sk, total, 0, true);
1622 EXPORT_SYMBOL_GPL(udp_destruct_common);
1624 static void udp_destruct_sock(struct sock *sk)
1626 udp_destruct_common(sk);
1627 inet_sock_destruct(sk);
1630 int udp_init_sock(struct sock *sk)
1632 udp_lib_init_sock(sk);
1633 sk->sk_destruct = udp_destruct_sock;
1634 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1638 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1640 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1641 bool slow = lock_sock_fast(sk);
1643 sk_peek_offset_bwd(sk, len);
1644 unlock_sock_fast(sk, slow);
1647 if (!skb_unref(skb))
1650 /* In the more common cases we cleared the head states previously,
1651 * see __udp_queue_rcv_skb().
1653 if (unlikely(udp_skb_has_head_state(skb)))
1654 skb_release_head_state(skb);
1655 __consume_stateless_skb(skb);
1657 EXPORT_SYMBOL_GPL(skb_consume_udp);
1659 static struct sk_buff *__first_packet_length(struct sock *sk,
1660 struct sk_buff_head *rcvq,
1663 struct sk_buff *skb;
1665 while ((skb = skb_peek(rcvq)) != NULL) {
1666 if (udp_lib_checksum_complete(skb)) {
1667 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1669 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1671 atomic_inc(&sk->sk_drops);
1672 __skb_unlink(skb, rcvq);
1673 *total += skb->truesize;
1676 udp_skb_csum_unnecessary_set(skb);
1684 * first_packet_length - return length of first packet in receive queue
1687 * Drops all bad checksum frames, until a valid one is found.
1688 * Returns the length of found skb, or -1 if none is found.
1690 static int first_packet_length(struct sock *sk)
1692 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1693 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1694 struct sk_buff *skb;
1698 spin_lock_bh(&rcvq->lock);
1699 skb = __first_packet_length(sk, rcvq, &total);
1700 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1701 spin_lock(&sk_queue->lock);
1702 skb_queue_splice_tail_init(sk_queue, rcvq);
1703 spin_unlock(&sk_queue->lock);
1705 skb = __first_packet_length(sk, rcvq, &total);
1707 res = skb ? skb->len : -1;
1709 udp_rmem_release(sk, total, 1, false);
1710 spin_unlock_bh(&rcvq->lock);
1715 * IOCTL requests applicable to the UDP protocol
1718 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1723 int amount = sk_wmem_alloc_get(sk);
1725 return put_user(amount, (int __user *)arg);
1730 int amount = max_t(int, 0, first_packet_length(sk));
1732 return put_user(amount, (int __user *)arg);
1736 return -ENOIOCTLCMD;
1741 EXPORT_SYMBOL(udp_ioctl);
1743 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1746 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1747 struct sk_buff_head *queue;
1748 struct sk_buff *last;
1752 queue = &udp_sk(sk)->reader_queue;
1753 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1755 struct sk_buff *skb;
1757 error = sock_error(sk);
1763 spin_lock_bh(&queue->lock);
1764 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1767 if (!(flags & MSG_PEEK))
1768 udp_skb_destructor(sk, skb);
1769 spin_unlock_bh(&queue->lock);
1773 if (skb_queue_empty_lockless(sk_queue)) {
1774 spin_unlock_bh(&queue->lock);
1778 /* refill the reader queue and walk it again
1779 * keep both queues locked to avoid re-acquiring
1780 * the sk_receive_queue lock if fwd memory scheduling
1783 spin_lock(&sk_queue->lock);
1784 skb_queue_splice_tail_init(sk_queue, queue);
1786 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1788 if (skb && !(flags & MSG_PEEK))
1789 udp_skb_dtor_locked(sk, skb);
1790 spin_unlock(&sk_queue->lock);
1791 spin_unlock_bh(&queue->lock);
1796 if (!sk_can_busy_loop(sk))
1799 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1800 } while (!skb_queue_empty_lockless(sk_queue));
1802 /* sk_queue is empty, reader_queue may contain peeked packets */
1804 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1806 (struct sk_buff *)sk_queue));
1811 EXPORT_SYMBOL(__skb_recv_udp);
1813 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1815 struct sk_buff *skb;
1819 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1823 if (udp_lib_checksum_complete(skb)) {
1824 int is_udplite = IS_UDPLITE(sk);
1825 struct net *net = sock_net(sk);
1827 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1828 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1829 atomic_inc(&sk->sk_drops);
1834 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1835 copied = recv_actor(sk, skb);
1840 EXPORT_SYMBOL(udp_read_skb);
1843 * This should be easy, if there is something there we
1844 * return it, otherwise we block.
1847 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1850 struct inet_sock *inet = inet_sk(sk);
1851 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1852 struct sk_buff *skb;
1853 unsigned int ulen, copied;
1854 int off, err, peeking = flags & MSG_PEEK;
1855 int is_udplite = IS_UDPLITE(sk);
1856 bool checksum_valid = false;
1858 if (flags & MSG_ERRQUEUE)
1859 return ip_recv_error(sk, msg, len, addr_len);
1862 off = sk_peek_offset(sk, flags);
1863 skb = __skb_recv_udp(sk, flags, &off, &err);
1867 ulen = udp_skb_len(skb);
1869 if (copied > ulen - off)
1870 copied = ulen - off;
1871 else if (copied < ulen)
1872 msg->msg_flags |= MSG_TRUNC;
1875 * If checksum is needed at all, try to do it while copying the
1876 * data. If the data is truncated, or if we only want a partial
1877 * coverage checksum (UDP-Lite), do it before the copy.
1880 if (copied < ulen || peeking ||
1881 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1882 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1883 !__udp_lib_checksum_complete(skb);
1884 if (!checksum_valid)
1888 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1889 if (udp_skb_is_linear(skb))
1890 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1892 err = skb_copy_datagram_msg(skb, off, msg, copied);
1894 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1900 if (unlikely(err)) {
1902 atomic_inc(&sk->sk_drops);
1903 UDP_INC_STATS(sock_net(sk),
1904 UDP_MIB_INERRORS, is_udplite);
1911 UDP_INC_STATS(sock_net(sk),
1912 UDP_MIB_INDATAGRAMS, is_udplite);
1914 sock_recv_cmsgs(msg, sk, skb);
1916 /* Copy the address. */
1918 sin->sin_family = AF_INET;
1919 sin->sin_port = udp_hdr(skb)->source;
1920 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1921 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1922 *addr_len = sizeof(*sin);
1924 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1925 (struct sockaddr *)sin);
1928 if (udp_sk(sk)->gro_enabled)
1929 udp_cmsg_recv(msg, sk, skb);
1931 if (inet->cmsg_flags)
1932 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1935 if (flags & MSG_TRUNC)
1938 skb_consume_udp(sk, skb, peeking ? -err : err);
1942 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1943 udp_skb_destructor)) {
1944 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1945 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1949 /* starting over for a new packet, but check if we need to yield */
1951 msg->msg_flags &= ~MSG_TRUNC;
1955 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1957 /* This check is replicated from __ip4_datagram_connect() and
1958 * intended to prevent BPF program called below from accessing bytes
1959 * that are out of the bound specified by user in addr_len.
1961 if (addr_len < sizeof(struct sockaddr_in))
1964 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1966 EXPORT_SYMBOL(udp_pre_connect);
1968 int __udp_disconnect(struct sock *sk, int flags)
1970 struct inet_sock *inet = inet_sk(sk);
1972 * 1003.1g - break association.
1975 sk->sk_state = TCP_CLOSE;
1976 inet->inet_daddr = 0;
1977 inet->inet_dport = 0;
1978 sock_rps_reset_rxhash(sk);
1979 sk->sk_bound_dev_if = 0;
1980 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1981 inet_reset_saddr(sk);
1982 if (sk->sk_prot->rehash &&
1983 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1984 sk->sk_prot->rehash(sk);
1987 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1988 sk->sk_prot->unhash(sk);
1989 inet->inet_sport = 0;
1994 EXPORT_SYMBOL(__udp_disconnect);
1996 int udp_disconnect(struct sock *sk, int flags)
1999 __udp_disconnect(sk, flags);
2003 EXPORT_SYMBOL(udp_disconnect);
2005 void udp_lib_unhash(struct sock *sk)
2007 if (sk_hashed(sk)) {
2008 struct udp_table *udptable = udp_get_table_prot(sk);
2009 struct udp_hslot *hslot, *hslot2;
2011 hslot = udp_hashslot(udptable, sock_net(sk),
2012 udp_sk(sk)->udp_port_hash);
2013 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2015 spin_lock_bh(&hslot->lock);
2016 if (rcu_access_pointer(sk->sk_reuseport_cb))
2017 reuseport_detach_sock(sk);
2018 if (sk_del_node_init_rcu(sk)) {
2020 inet_sk(sk)->inet_num = 0;
2021 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2023 spin_lock(&hslot2->lock);
2024 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2026 spin_unlock(&hslot2->lock);
2028 spin_unlock_bh(&hslot->lock);
2031 EXPORT_SYMBOL(udp_lib_unhash);
2034 * inet_rcv_saddr was changed, we must rehash secondary hash
2036 void udp_lib_rehash(struct sock *sk, u16 newhash)
2038 if (sk_hashed(sk)) {
2039 struct udp_table *udptable = udp_get_table_prot(sk);
2040 struct udp_hslot *hslot, *hslot2, *nhslot2;
2042 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2043 nhslot2 = udp_hashslot2(udptable, newhash);
2044 udp_sk(sk)->udp_portaddr_hash = newhash;
2046 if (hslot2 != nhslot2 ||
2047 rcu_access_pointer(sk->sk_reuseport_cb)) {
2048 hslot = udp_hashslot(udptable, sock_net(sk),
2049 udp_sk(sk)->udp_port_hash);
2050 /* we must lock primary chain too */
2051 spin_lock_bh(&hslot->lock);
2052 if (rcu_access_pointer(sk->sk_reuseport_cb))
2053 reuseport_detach_sock(sk);
2055 if (hslot2 != nhslot2) {
2056 spin_lock(&hslot2->lock);
2057 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2059 spin_unlock(&hslot2->lock);
2061 spin_lock(&nhslot2->lock);
2062 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2065 spin_unlock(&nhslot2->lock);
2068 spin_unlock_bh(&hslot->lock);
2072 EXPORT_SYMBOL(udp_lib_rehash);
2074 void udp_v4_rehash(struct sock *sk)
2076 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2077 inet_sk(sk)->inet_rcv_saddr,
2078 inet_sk(sk)->inet_num);
2079 udp_lib_rehash(sk, new_hash);
2082 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2086 if (inet_sk(sk)->inet_daddr) {
2087 sock_rps_save_rxhash(sk, skb);
2088 sk_mark_napi_id(sk, skb);
2089 sk_incoming_cpu_update(sk);
2091 sk_mark_napi_id_once(sk, skb);
2094 rc = __udp_enqueue_schedule_skb(sk, skb);
2096 int is_udplite = IS_UDPLITE(sk);
2099 /* Note that an ENOMEM error is charged twice */
2100 if (rc == -ENOMEM) {
2101 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2103 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2105 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2107 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2109 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2110 kfree_skb_reason(skb, drop_reason);
2111 trace_udp_fail_queue_rcv_skb(rc, sk);
2121 * >0: "udp encap" protocol resubmission
2123 * Note that in the success and error cases, the skb is assumed to
2124 * have either been requeued or freed.
2126 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2128 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2129 struct udp_sock *up = udp_sk(sk);
2130 int is_udplite = IS_UDPLITE(sk);
2133 * Charge it to the socket, dropping if the queue is full.
2135 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2136 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2141 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2142 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2145 * This is an encapsulation socket so pass the skb to
2146 * the socket's udp_encap_rcv() hook. Otherwise, just
2147 * fall through and pass this up the UDP socket.
2148 * up->encap_rcv() returns the following value:
2149 * =0 if skb was successfully passed to the encap
2150 * handler or was discarded by it.
2151 * >0 if skb should be passed on to UDP.
2152 * <0 if skb should be resubmitted as proto -N
2155 /* if we're overly short, let UDP handle it */
2156 encap_rcv = READ_ONCE(up->encap_rcv);
2160 /* Verify checksum before giving to encap */
2161 if (udp_lib_checksum_complete(skb))
2164 ret = encap_rcv(sk, skb);
2166 __UDP_INC_STATS(sock_net(sk),
2167 UDP_MIB_INDATAGRAMS,
2173 /* FALLTHROUGH -- it's a UDP Packet */
2177 * UDP-Lite specific tests, ignored on UDP sockets
2179 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2182 * MIB statistics other than incrementing the error count are
2183 * disabled for the following two types of errors: these depend
2184 * on the application settings, not on the functioning of the
2185 * protocol stack as such.
2187 * RFC 3828 here recommends (sec 3.3): "There should also be a
2188 * way ... to ... at least let the receiving application block
2189 * delivery of packets with coverage values less than a value
2190 * provided by the application."
2192 if (up->pcrlen == 0) { /* full coverage was set */
2193 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2194 UDP_SKB_CB(skb)->cscov, skb->len);
2197 /* The next case involves violating the min. coverage requested
2198 * by the receiver. This is subtle: if receiver wants x and x is
2199 * greater than the buffersize/MTU then receiver will complain
2200 * that it wants x while sender emits packets of smaller size y.
2201 * Therefore the above ...()->partial_cov statement is essential.
2203 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2204 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2205 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2210 prefetch(&sk->sk_rmem_alloc);
2211 if (rcu_access_pointer(sk->sk_filter) &&
2212 udp_lib_checksum_complete(skb))
2215 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2216 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2220 udp_csum_pull_header(skb);
2222 ipv4_pktinfo_prepare(sk, skb);
2223 return __udp_queue_rcv_skb(sk, skb);
2226 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2227 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2229 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2230 atomic_inc(&sk->sk_drops);
2231 kfree_skb_reason(skb, drop_reason);
2235 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2237 struct sk_buff *next, *segs;
2240 if (likely(!udp_unexpected_gso(sk, skb)))
2241 return udp_queue_rcv_one_skb(sk, skb);
2243 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2244 __skb_push(skb, -skb_mac_offset(skb));
2245 segs = udp_rcv_segment(sk, skb, true);
2246 skb_list_walk_safe(segs, skb, next) {
2247 __skb_pull(skb, skb_transport_offset(skb));
2249 udp_post_segment_fix_csum(skb);
2250 ret = udp_queue_rcv_one_skb(sk, skb);
2252 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2257 /* For TCP sockets, sk_rx_dst is protected by socket lock
2258 * For UDP, we use xchg() to guard against concurrent changes.
2260 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2262 struct dst_entry *old;
2264 if (dst_hold_safe(dst)) {
2265 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2271 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2274 * Multicasts and broadcasts go to each listener.
2276 * Note: called only from the BH handler context.
2278 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2280 __be32 saddr, __be32 daddr,
2281 struct udp_table *udptable,
2284 struct sock *sk, *first = NULL;
2285 unsigned short hnum = ntohs(uh->dest);
2286 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2287 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2288 unsigned int offset = offsetof(typeof(*sk), sk_node);
2289 int dif = skb->dev->ifindex;
2290 int sdif = inet_sdif(skb);
2291 struct hlist_node *node;
2292 struct sk_buff *nskb;
2295 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2297 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2299 hslot = &udptable->hash2[hash2];
2300 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2303 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2304 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2305 uh->source, saddr, dif, sdif, hnum))
2312 nskb = skb_clone(skb, GFP_ATOMIC);
2314 if (unlikely(!nskb)) {
2315 atomic_inc(&sk->sk_drops);
2316 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2318 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2322 if (udp_queue_rcv_skb(sk, nskb) > 0)
2326 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2327 if (use_hash2 && hash2 != hash2_any) {
2333 if (udp_queue_rcv_skb(first, skb) > 0)
2337 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2338 proto == IPPROTO_UDPLITE);
2343 /* Initialize UDP checksum. If exited with zero value (success),
2344 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2345 * Otherwise, csum completion requires checksumming packet body,
2346 * including udp header and folding it to skb->csum.
2348 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2353 UDP_SKB_CB(skb)->partial_cov = 0;
2354 UDP_SKB_CB(skb)->cscov = skb->len;
2356 if (proto == IPPROTO_UDPLITE) {
2357 err = udplite_checksum_init(skb, uh);
2361 if (UDP_SKB_CB(skb)->partial_cov) {
2362 skb->csum = inet_compute_pseudo(skb, proto);
2367 /* Note, we are only interested in != 0 or == 0, thus the
2370 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2371 inet_compute_pseudo);
2375 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2376 /* If SW calculated the value, we know it's bad */
2377 if (skb->csum_complete_sw)
2380 /* HW says the value is bad. Let's validate that.
2381 * skb->csum is no longer the full packet checksum,
2382 * so don't treat it as such.
2384 skb_checksum_complete_unset(skb);
2390 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2391 * return code conversion for ip layer consumption
2393 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2398 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2399 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2401 ret = udp_queue_rcv_skb(sk, skb);
2403 /* a return value > 0 means to resubmit the input, but
2404 * it wants the return to be -protocol, or 0
2412 * All we need to do is get the socket, and then do a checksum.
2415 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2420 unsigned short ulen;
2421 struct rtable *rt = skb_rtable(skb);
2422 __be32 saddr, daddr;
2423 struct net *net = dev_net(skb->dev);
2427 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2430 * Validate the packet.
2432 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2433 goto drop; /* No space for header. */
2436 ulen = ntohs(uh->len);
2437 saddr = ip_hdr(skb)->saddr;
2438 daddr = ip_hdr(skb)->daddr;
2440 if (ulen > skb->len)
2443 if (proto == IPPROTO_UDP) {
2444 /* UDP validates ulen. */
2445 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2450 if (udp4_csum_init(skb, uh, proto))
2453 sk = skb_steal_sock(skb, &refcounted);
2455 struct dst_entry *dst = skb_dst(skb);
2458 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2459 udp_sk_rx_dst_set(sk, dst);
2461 ret = udp_unicast_rcv_skb(sk, skb, uh);
2467 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2468 return __udp4_lib_mcast_deliver(net, skb, uh,
2469 saddr, daddr, udptable, proto);
2471 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2473 return udp_unicast_rcv_skb(sk, skb, uh);
2475 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2479 /* No socket. Drop packet silently, if checksum is wrong */
2480 if (udp_lib_checksum_complete(skb))
2483 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2484 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2485 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2488 * Hmm. We got an UDP packet to a port to which we
2489 * don't wanna listen. Ignore it.
2491 kfree_skb_reason(skb, drop_reason);
2495 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2496 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2497 proto == IPPROTO_UDPLITE ? "Lite" : "",
2498 &saddr, ntohs(uh->source),
2500 &daddr, ntohs(uh->dest));
2505 * RFC1122: OK. Discards the bad packet silently (as far as
2506 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2508 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2509 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2510 proto == IPPROTO_UDPLITE ? "Lite" : "",
2511 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2513 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2515 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2516 kfree_skb_reason(skb, drop_reason);
2520 /* We can only early demux multicast if there is a single matching socket.
2521 * If more than one socket found returns NULL
2523 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2524 __be16 loc_port, __be32 loc_addr,
2525 __be16 rmt_port, __be32 rmt_addr,
2528 struct udp_table *udptable = net->ipv4.udp_table;
2529 unsigned short hnum = ntohs(loc_port);
2530 struct sock *sk, *result;
2531 struct udp_hslot *hslot;
2534 slot = udp_hashfn(net, hnum, udptable->mask);
2535 hslot = &udptable->hash[slot];
2537 /* Do not bother scanning a too big list */
2538 if (hslot->count > 10)
2542 sk_for_each_rcu(sk, &hslot->head) {
2543 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2544 rmt_port, rmt_addr, dif, sdif, hnum)) {
2554 /* For unicast we should only early demux connected sockets or we can
2555 * break forwarding setups. The chains here can be long so only check
2556 * if the first socket is an exact match and if not move on.
2558 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2559 __be16 loc_port, __be32 loc_addr,
2560 __be16 rmt_port, __be32 rmt_addr,
2563 struct udp_table *udptable = net->ipv4.udp_table;
2564 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2565 unsigned short hnum = ntohs(loc_port);
2566 unsigned int hash2, slot2;
2567 struct udp_hslot *hslot2;
2571 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2572 slot2 = hash2 & udptable->mask;
2573 hslot2 = &udptable->hash2[slot2];
2574 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2576 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2577 if (inet_match(net, sk, acookie, ports, dif, sdif))
2579 /* Only check first socket in chain */
2585 int udp_v4_early_demux(struct sk_buff *skb)
2587 struct net *net = dev_net(skb->dev);
2588 struct in_device *in_dev = NULL;
2589 const struct iphdr *iph;
2590 const struct udphdr *uh;
2591 struct sock *sk = NULL;
2592 struct dst_entry *dst;
2593 int dif = skb->dev->ifindex;
2594 int sdif = inet_sdif(skb);
2597 /* validate the packet */
2598 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2604 if (skb->pkt_type == PACKET_MULTICAST) {
2605 in_dev = __in_dev_get_rcu(skb->dev);
2610 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2615 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2616 uh->source, iph->saddr,
2618 } else if (skb->pkt_type == PACKET_HOST) {
2619 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2620 uh->source, iph->saddr, dif, sdif);
2623 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2627 skb->destructor = sock_efree;
2628 dst = rcu_dereference(sk->sk_rx_dst);
2631 dst = dst_check(dst, 0);
2635 /* set noref for now.
2636 * any place which wants to hold dst has to call
2639 skb_dst_set_noref(skb, dst);
2641 /* for unconnected multicast sockets we need to validate
2642 * the source on each packet
2644 if (!inet_sk(sk)->inet_daddr && in_dev)
2645 return ip_mc_validate_source(skb, iph->daddr,
2647 iph->tos & IPTOS_RT_MASK,
2648 skb->dev, in_dev, &itag);
2653 int udp_rcv(struct sk_buff *skb)
2655 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2658 void udp_destroy_sock(struct sock *sk)
2660 struct udp_sock *up = udp_sk(sk);
2661 bool slow = lock_sock_fast(sk);
2663 /* protects from races with udp_abort() */
2664 sock_set_flag(sk, SOCK_DEAD);
2665 udp_flush_pending_frames(sk);
2666 unlock_sock_fast(sk, slow);
2667 if (static_branch_unlikely(&udp_encap_needed_key)) {
2668 if (up->encap_type) {
2669 void (*encap_destroy)(struct sock *sk);
2670 encap_destroy = READ_ONCE(up->encap_destroy);
2674 if (up->encap_enabled)
2675 static_branch_dec(&udp_encap_needed_key);
2680 * Socket option code for UDP
2682 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2683 sockptr_t optval, unsigned int optlen,
2684 int (*push_pending_frames)(struct sock *))
2686 struct udp_sock *up = udp_sk(sk);
2689 int is_udplite = IS_UDPLITE(sk);
2691 if (level == SOL_SOCKET) {
2692 err = sk_setsockopt(sk, level, optname, optval, optlen);
2694 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2695 sockopt_lock_sock(sk);
2696 /* paired with READ_ONCE in udp_rmem_release() */
2697 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2698 sockopt_release_sock(sk);
2703 if (optlen < sizeof(int))
2706 if (copy_from_sockptr(&val, optval, sizeof(val)))
2709 valbool = val ? 1 : 0;
2714 WRITE_ONCE(up->corkflag, 1);
2716 WRITE_ONCE(up->corkflag, 0);
2718 push_pending_frames(sk);
2727 case UDP_ENCAP_ESPINUDP:
2728 case UDP_ENCAP_ESPINUDP_NON_IKE:
2729 #if IS_ENABLED(CONFIG_IPV6)
2730 if (sk->sk_family == AF_INET6)
2731 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2734 up->encap_rcv = xfrm4_udp_encap_rcv;
2737 case UDP_ENCAP_L2TPINUDP:
2738 up->encap_type = val;
2740 udp_tunnel_encap_enable(sk->sk_socket);
2749 case UDP_NO_CHECK6_TX:
2750 up->no_check6_tx = valbool;
2753 case UDP_NO_CHECK6_RX:
2754 up->no_check6_rx = valbool;
2758 if (val < 0 || val > USHRT_MAX)
2760 WRITE_ONCE(up->gso_size, val);
2766 /* when enabling GRO, accept the related GSO packet type */
2768 udp_tunnel_encap_enable(sk->sk_socket);
2769 up->gro_enabled = valbool;
2770 up->accept_udp_l4 = valbool;
2775 * UDP-Lite's partial checksum coverage (RFC 3828).
2777 /* The sender sets actual checksum coverage length via this option.
2778 * The case coverage > packet length is handled by send module. */
2779 case UDPLITE_SEND_CSCOV:
2780 if (!is_udplite) /* Disable the option on UDP sockets */
2781 return -ENOPROTOOPT;
2782 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2784 else if (val > USHRT_MAX)
2787 up->pcflag |= UDPLITE_SEND_CC;
2790 /* The receiver specifies a minimum checksum coverage value. To make
2791 * sense, this should be set to at least 8 (as done below). If zero is
2792 * used, this again means full checksum coverage. */
2793 case UDPLITE_RECV_CSCOV:
2794 if (!is_udplite) /* Disable the option on UDP sockets */
2795 return -ENOPROTOOPT;
2796 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2798 else if (val > USHRT_MAX)
2801 up->pcflag |= UDPLITE_RECV_CC;
2811 EXPORT_SYMBOL(udp_lib_setsockopt);
2813 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2814 unsigned int optlen)
2816 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2817 return udp_lib_setsockopt(sk, level, optname,
2819 udp_push_pending_frames);
2820 return ip_setsockopt(sk, level, optname, optval, optlen);
2823 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2824 char __user *optval, int __user *optlen)
2826 struct udp_sock *up = udp_sk(sk);
2829 if (get_user(len, optlen))
2832 len = min_t(unsigned int, len, sizeof(int));
2839 val = READ_ONCE(up->corkflag);
2843 val = up->encap_type;
2846 case UDP_NO_CHECK6_TX:
2847 val = up->no_check6_tx;
2850 case UDP_NO_CHECK6_RX:
2851 val = up->no_check6_rx;
2855 val = READ_ONCE(up->gso_size);
2859 val = up->gro_enabled;
2862 /* The following two cannot be changed on UDP sockets, the return is
2863 * always 0 (which corresponds to the full checksum coverage of UDP). */
2864 case UDPLITE_SEND_CSCOV:
2868 case UDPLITE_RECV_CSCOV:
2873 return -ENOPROTOOPT;
2876 if (put_user(len, optlen))
2878 if (copy_to_user(optval, &val, len))
2882 EXPORT_SYMBOL(udp_lib_getsockopt);
2884 int udp_getsockopt(struct sock *sk, int level, int optname,
2885 char __user *optval, int __user *optlen)
2887 if (level == SOL_UDP || level == SOL_UDPLITE)
2888 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2889 return ip_getsockopt(sk, level, optname, optval, optlen);
2893 * udp_poll - wait for a UDP event.
2894 * @file: - file struct
2896 * @wait: - poll table
2898 * This is same as datagram poll, except for the special case of
2899 * blocking sockets. If application is using a blocking fd
2900 * and a packet with checksum error is in the queue;
2901 * then it could get return from select indicating data available
2902 * but then block when reading it. Add special case code
2903 * to work around these arguably broken applications.
2905 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2907 __poll_t mask = datagram_poll(file, sock, wait);
2908 struct sock *sk = sock->sk;
2910 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2911 mask |= EPOLLIN | EPOLLRDNORM;
2913 /* Check for false positives due to checksum errors */
2914 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2915 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2916 mask &= ~(EPOLLIN | EPOLLRDNORM);
2918 /* psock ingress_msg queue should not contain any bad checksum frames */
2919 if (sk_is_readable(sk))
2920 mask |= EPOLLIN | EPOLLRDNORM;
2924 EXPORT_SYMBOL(udp_poll);
2926 int udp_abort(struct sock *sk, int err)
2930 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2933 if (sock_flag(sk, SOCK_DEAD))
2937 sk_error_report(sk);
2938 __udp_disconnect(sk, 0);
2945 EXPORT_SYMBOL_GPL(udp_abort);
2947 struct proto udp_prot = {
2949 .owner = THIS_MODULE,
2950 .close = udp_lib_close,
2951 .pre_connect = udp_pre_connect,
2952 .connect = ip4_datagram_connect,
2953 .disconnect = udp_disconnect,
2955 .init = udp_init_sock,
2956 .destroy = udp_destroy_sock,
2957 .setsockopt = udp_setsockopt,
2958 .getsockopt = udp_getsockopt,
2959 .sendmsg = udp_sendmsg,
2960 .recvmsg = udp_recvmsg,
2961 .sendpage = udp_sendpage,
2962 .release_cb = ip4_datagram_release_cb,
2963 .hash = udp_lib_hash,
2964 .unhash = udp_lib_unhash,
2965 .rehash = udp_v4_rehash,
2966 .get_port = udp_v4_get_port,
2967 .put_port = udp_lib_unhash,
2968 #ifdef CONFIG_BPF_SYSCALL
2969 .psock_update_sk_prot = udp_bpf_update_proto,
2971 .memory_allocated = &udp_memory_allocated,
2972 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2974 .sysctl_mem = sysctl_udp_mem,
2975 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2976 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2977 .obj_size = sizeof(struct udp_sock),
2978 .h.udp_table = NULL,
2979 .diag_destroy = udp_abort,
2981 EXPORT_SYMBOL(udp_prot);
2983 /* ------------------------------------------------------------------------ */
2984 #ifdef CONFIG_PROC_FS
2986 static struct udp_table *udp_get_table_afinfo(struct udp_seq_afinfo *afinfo,
2989 return afinfo->udp_table ? : net->ipv4.udp_table;
2992 static struct sock *udp_get_first(struct seq_file *seq, int start)
2994 struct udp_iter_state *state = seq->private;
2995 struct net *net = seq_file_net(seq);
2996 struct udp_seq_afinfo *afinfo;
2997 struct udp_table *udptable;
3000 if (state->bpf_seq_afinfo)
3001 afinfo = state->bpf_seq_afinfo;
3003 afinfo = pde_data(file_inode(seq->file));
3005 udptable = udp_get_table_afinfo(afinfo, net);
3007 for (state->bucket = start; state->bucket <= udptable->mask;
3009 struct udp_hslot *hslot = &udptable->hash[state->bucket];
3011 if (hlist_empty(&hslot->head))
3014 spin_lock_bh(&hslot->lock);
3015 sk_for_each(sk, &hslot->head) {
3016 if (!net_eq(sock_net(sk), net))
3018 if (afinfo->family == AF_UNSPEC ||
3019 sk->sk_family == afinfo->family)
3022 spin_unlock_bh(&hslot->lock);
3029 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3031 struct udp_iter_state *state = seq->private;
3032 struct net *net = seq_file_net(seq);
3033 struct udp_seq_afinfo *afinfo;
3034 struct udp_table *udptable;
3036 if (state->bpf_seq_afinfo)
3037 afinfo = state->bpf_seq_afinfo;
3039 afinfo = pde_data(file_inode(seq->file));
3043 } while (sk && (!net_eq(sock_net(sk), net) ||
3044 (afinfo->family != AF_UNSPEC &&
3045 sk->sk_family != afinfo->family)));
3048 udptable = udp_get_table_afinfo(afinfo, net);
3050 if (state->bucket <= udptable->mask)
3051 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3053 return udp_get_first(seq, state->bucket + 1);
3058 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3060 struct sock *sk = udp_get_first(seq, 0);
3063 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3065 return pos ? NULL : sk;
3068 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3070 struct udp_iter_state *state = seq->private;
3071 state->bucket = MAX_UDP_PORTS;
3073 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3075 EXPORT_SYMBOL(udp_seq_start);
3077 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3081 if (v == SEQ_START_TOKEN)
3082 sk = udp_get_idx(seq, 0);
3084 sk = udp_get_next(seq, v);
3089 EXPORT_SYMBOL(udp_seq_next);
3091 void udp_seq_stop(struct seq_file *seq, void *v)
3093 struct udp_iter_state *state = seq->private;
3094 struct udp_seq_afinfo *afinfo;
3095 struct udp_table *udptable;
3097 if (state->bpf_seq_afinfo)
3098 afinfo = state->bpf_seq_afinfo;
3100 afinfo = pde_data(file_inode(seq->file));
3102 udptable = udp_get_table_afinfo(afinfo, seq_file_net(seq));
3104 if (state->bucket <= udptable->mask)
3105 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3107 EXPORT_SYMBOL(udp_seq_stop);
3109 /* ------------------------------------------------------------------------ */
3110 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3113 struct inet_sock *inet = inet_sk(sp);
3114 __be32 dest = inet->inet_daddr;
3115 __be32 src = inet->inet_rcv_saddr;
3116 __u16 destp = ntohs(inet->inet_dport);
3117 __u16 srcp = ntohs(inet->inet_sport);
3119 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3120 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3121 bucket, src, srcp, dest, destp, sp->sk_state,
3122 sk_wmem_alloc_get(sp),
3125 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3127 refcount_read(&sp->sk_refcnt), sp,
3128 atomic_read(&sp->sk_drops));
3131 int udp4_seq_show(struct seq_file *seq, void *v)
3133 seq_setwidth(seq, 127);
3134 if (v == SEQ_START_TOKEN)
3135 seq_puts(seq, " sl local_address rem_address st tx_queue "
3136 "rx_queue tr tm->when retrnsmt uid timeout "
3137 "inode ref pointer drops");
3139 struct udp_iter_state *state = seq->private;
3141 udp4_format_sock(v, seq, state->bucket);
3147 #ifdef CONFIG_BPF_SYSCALL
3148 struct bpf_iter__udp {
3149 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3150 __bpf_md_ptr(struct udp_sock *, udp_sk);
3151 uid_t uid __aligned(8);
3152 int bucket __aligned(8);
3155 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3156 struct udp_sock *udp_sk, uid_t uid, int bucket)
3158 struct bpf_iter__udp ctx;
3160 meta->seq_num--; /* skip SEQ_START_TOKEN */
3162 ctx.udp_sk = udp_sk;
3164 ctx.bucket = bucket;
3165 return bpf_iter_run_prog(prog, &ctx);
3168 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3170 struct udp_iter_state *state = seq->private;
3171 struct bpf_iter_meta meta;
3172 struct bpf_prog *prog;
3173 struct sock *sk = v;
3176 if (v == SEQ_START_TOKEN)
3179 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3181 prog = bpf_iter_get_info(&meta, false);
3182 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3185 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3187 struct bpf_iter_meta meta;
3188 struct bpf_prog *prog;
3192 prog = bpf_iter_get_info(&meta, true);
3194 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3197 udp_seq_stop(seq, v);
3200 static const struct seq_operations bpf_iter_udp_seq_ops = {
3201 .start = udp_seq_start,
3202 .next = udp_seq_next,
3203 .stop = bpf_iter_udp_seq_stop,
3204 .show = bpf_iter_udp_seq_show,
3208 const struct seq_operations udp_seq_ops = {
3209 .start = udp_seq_start,
3210 .next = udp_seq_next,
3211 .stop = udp_seq_stop,
3212 .show = udp4_seq_show,
3214 EXPORT_SYMBOL(udp_seq_ops);
3216 static struct udp_seq_afinfo udp4_seq_afinfo = {
3221 static int __net_init udp4_proc_init_net(struct net *net)
3223 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3224 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3229 static void __net_exit udp4_proc_exit_net(struct net *net)
3231 remove_proc_entry("udp", net->proc_net);
3234 static struct pernet_operations udp4_net_ops = {
3235 .init = udp4_proc_init_net,
3236 .exit = udp4_proc_exit_net,
3239 int __init udp4_proc_init(void)
3241 return register_pernet_subsys(&udp4_net_ops);
3244 void udp4_proc_exit(void)
3246 unregister_pernet_subsys(&udp4_net_ops);
3248 #endif /* CONFIG_PROC_FS */
3250 static __initdata unsigned long uhash_entries;
3251 static int __init set_uhash_entries(char *str)
3258 ret = kstrtoul(str, 0, &uhash_entries);
3262 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3263 uhash_entries = UDP_HTABLE_SIZE_MIN;
3266 __setup("uhash_entries=", set_uhash_entries);
3268 void __init udp_table_init(struct udp_table *table, const char *name)
3272 table->hash = alloc_large_system_hash(name,
3273 2 * sizeof(struct udp_hslot),
3275 21, /* one slot per 2 MB */
3279 UDP_HTABLE_SIZE_MIN,
3280 UDP_HTABLE_SIZE_MAX);
3282 table->hash2 = table->hash + (table->mask + 1);
3283 for (i = 0; i <= table->mask; i++) {
3284 INIT_HLIST_HEAD(&table->hash[i].head);
3285 table->hash[i].count = 0;
3286 spin_lock_init(&table->hash[i].lock);
3288 for (i = 0; i <= table->mask; i++) {
3289 INIT_HLIST_HEAD(&table->hash2[i].head);
3290 table->hash2[i].count = 0;
3291 spin_lock_init(&table->hash2[i].lock);
3295 u32 udp_flow_hashrnd(void)
3297 static u32 hashrnd __read_mostly;
3299 net_get_random_once(&hashrnd, sizeof(hashrnd));
3303 EXPORT_SYMBOL(udp_flow_hashrnd);
3305 static void __net_init udp_sysctl_init(struct net *net)
3307 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3308 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3310 #ifdef CONFIG_NET_L3_MASTER_DEV
3311 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3315 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3317 struct udp_table *udptable;
3320 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3324 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3325 GFP_KERNEL_ACCOUNT);
3326 if (!udptable->hash)
3329 udptable->hash2 = udptable->hash + hash_entries;
3330 udptable->mask = hash_entries - 1;
3331 udptable->log = ilog2(hash_entries);
3333 for (i = 0; i < hash_entries; i++) {
3334 INIT_HLIST_HEAD(&udptable->hash[i].head);
3335 udptable->hash[i].count = 0;
3336 spin_lock_init(&udptable->hash[i].lock);
3338 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3339 udptable->hash2[i].count = 0;
3340 spin_lock_init(&udptable->hash2[i].lock);
3351 static void __net_exit udp_pernet_table_free(struct net *net)
3353 struct udp_table *udptable = net->ipv4.udp_table;
3355 if (udptable == &udp_table)
3358 kvfree(udptable->hash);
3362 static void __net_init udp_set_table(struct net *net)
3364 struct udp_table *udptable;
3365 unsigned int hash_entries;
3366 struct net *old_net;
3368 if (net_eq(net, &init_net))
3371 old_net = current->nsproxy->net_ns;
3372 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3376 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3377 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3378 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3380 hash_entries = roundup_pow_of_two(hash_entries);
3382 udptable = udp_pernet_table_alloc(hash_entries);
3384 net->ipv4.udp_table = udptable;
3386 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3387 "for a netns, fallback to the global one\n",
3390 net->ipv4.udp_table = &udp_table;
3394 static int __net_init udp_pernet_init(struct net *net)
3396 udp_sysctl_init(net);
3402 static void __net_exit udp_pernet_exit(struct net *net)
3404 udp_pernet_table_free(net);
3407 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3408 .init = udp_pernet_init,
3409 .exit = udp_pernet_exit,
3412 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3413 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3414 struct udp_sock *udp_sk, uid_t uid, int bucket)
3416 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3418 struct udp_iter_state *st = priv_data;
3419 struct udp_seq_afinfo *afinfo;
3422 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3426 afinfo->family = AF_UNSPEC;
3427 afinfo->udp_table = NULL;
3428 st->bpf_seq_afinfo = afinfo;
3429 ret = bpf_iter_init_seq_net(priv_data, aux);
3435 static void bpf_iter_fini_udp(void *priv_data)
3437 struct udp_iter_state *st = priv_data;
3439 kfree(st->bpf_seq_afinfo);
3440 bpf_iter_fini_seq_net(priv_data);
3443 static const struct bpf_iter_seq_info udp_seq_info = {
3444 .seq_ops = &bpf_iter_udp_seq_ops,
3445 .init_seq_private = bpf_iter_init_udp,
3446 .fini_seq_private = bpf_iter_fini_udp,
3447 .seq_priv_size = sizeof(struct udp_iter_state),
3450 static struct bpf_iter_reg udp_reg_info = {
3452 .ctx_arg_info_size = 1,
3454 { offsetof(struct bpf_iter__udp, udp_sk),
3455 PTR_TO_BTF_ID_OR_NULL },
3457 .seq_info = &udp_seq_info,
3460 static void __init bpf_iter_register(void)
3462 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3463 if (bpf_iter_reg_target(&udp_reg_info))
3464 pr_warn("Warning: could not register bpf iterator udp\n");
3468 void __init udp_init(void)
3470 unsigned long limit;
3473 udp_table_init(&udp_table, "UDP");
3474 limit = nr_free_buffer_pages() / 8;
3475 limit = max(limit, 128UL);
3476 sysctl_udp_mem[0] = limit / 4 * 3;
3477 sysctl_udp_mem[1] = limit;
3478 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3480 /* 16 spinlocks per cpu */
3481 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3482 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3485 panic("UDP: failed to alloc udp_busylocks\n");
3486 for (i = 0; i < (1U << udp_busylocks_log); i++)
3487 spin_lock_init(udp_busylocks + i);
3489 if (register_pernet_subsys(&udp_sysctl_ops))
3490 panic("UDP: failed to init sysctl parameters.\n");
3492 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3493 bpf_iter_register();