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>
107 #include <net/xfrm.h>
108 #include <trace/events/udp.h>
109 #include <linux/static_key.h>
110 #include <linux/btf_ids.h>
111 #include <trace/events/skb.h>
112 #include <net/busy_poll.h>
113 #include "udp_impl.h"
114 #include <net/sock_reuseport.h>
115 #include <net/addrconf.h>
116 #include <net/udp_tunnel.h>
118 #if IS_ENABLED(CONFIG_IPV6)
119 #include <net/ipv6_stubs.h>
122 struct udp_table udp_table __read_mostly;
123 EXPORT_SYMBOL(udp_table);
125 long sysctl_udp_mem[3] __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_mem);
128 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
129 EXPORT_SYMBOL(udp_memory_allocated);
130 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
131 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
133 #define MAX_UDP_PORTS 65536
134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
136 static struct udp_table *udp_get_table_prot(struct sock *sk)
138 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
141 static int udp_lib_lport_inuse(struct net *net, __u16 num,
142 const struct udp_hslot *hslot,
143 unsigned long *bitmap,
144 struct sock *sk, unsigned int log)
147 kuid_t uid = sock_i_uid(sk);
149 sk_for_each(sk2, &hslot->head) {
150 if (net_eq(sock_net(sk2), net) &&
152 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
153 (!sk2->sk_reuse || !sk->sk_reuse) &&
154 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
155 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
156 inet_rcv_saddr_equal(sk, sk2, true)) {
157 if (sk2->sk_reuseport && sk->sk_reuseport &&
158 !rcu_access_pointer(sk->sk_reuseport_cb) &&
159 uid_eq(uid, sock_i_uid(sk2))) {
165 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
174 * Note: we still hold spinlock of primary hash chain, so no other writer
175 * can insert/delete a socket with local_port == num
177 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
178 struct udp_hslot *hslot2,
182 kuid_t uid = sock_i_uid(sk);
185 spin_lock(&hslot2->lock);
186 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
187 if (net_eq(sock_net(sk2), net) &&
189 (udp_sk(sk2)->udp_port_hash == num) &&
190 (!sk2->sk_reuse || !sk->sk_reuse) &&
191 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
192 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
193 inet_rcv_saddr_equal(sk, sk2, true)) {
194 if (sk2->sk_reuseport && sk->sk_reuseport &&
195 !rcu_access_pointer(sk->sk_reuseport_cb) &&
196 uid_eq(uid, sock_i_uid(sk2))) {
204 spin_unlock(&hslot2->lock);
208 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
210 struct net *net = sock_net(sk);
211 kuid_t uid = sock_i_uid(sk);
214 sk_for_each(sk2, &hslot->head) {
215 if (net_eq(sock_net(sk2), net) &&
217 sk2->sk_family == sk->sk_family &&
218 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
219 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
220 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
221 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
222 inet_rcv_saddr_equal(sk, sk2, false)) {
223 return reuseport_add_sock(sk, sk2,
224 inet_rcv_saddr_any(sk));
228 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
234 * @sk: socket struct in question
235 * @snum: port number to look up
236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
239 int udp_lib_get_port(struct sock *sk, unsigned short snum,
240 unsigned int hash2_nulladdr)
242 struct udp_table *udptable = udp_get_table_prot(sk);
243 struct udp_hslot *hslot, *hslot2;
244 struct net *net = sock_net(sk);
245 int error = -EADDRINUSE;
248 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
249 unsigned short first, last;
250 int low, high, remaining;
253 inet_sk_get_local_port_range(sk, &low, &high);
254 remaining = (high - low) + 1;
256 rand = get_random_u32();
257 first = reciprocal_scale(rand, remaining) + low;
259 * force rand to be an odd multiple of UDP_HTABLE_SIZE
261 rand = (rand | 1) * (udptable->mask + 1);
262 last = first + udptable->mask + 1;
264 hslot = udp_hashslot(udptable, net, first);
265 bitmap_zero(bitmap, PORTS_PER_CHAIN);
266 spin_lock_bh(&hslot->lock);
267 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
272 * Iterate on all possible values of snum for this hash.
273 * Using steps of an odd multiple of UDP_HTABLE_SIZE
274 * give us randomization and full range coverage.
277 if (low <= snum && snum <= high &&
278 !test_bit(snum >> udptable->log, bitmap) &&
279 !inet_is_local_reserved_port(net, snum))
282 } while (snum != first);
283 spin_unlock_bh(&hslot->lock);
285 } while (++first != last);
288 hslot = udp_hashslot(udptable, net, snum);
289 spin_lock_bh(&hslot->lock);
290 if (hslot->count > 10) {
292 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
294 slot2 &= udptable->mask;
295 hash2_nulladdr &= udptable->mask;
297 hslot2 = udp_hashslot2(udptable, slot2);
298 if (hslot->count < hslot2->count)
299 goto scan_primary_hash;
301 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
302 if (!exist && (hash2_nulladdr != slot2)) {
303 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
304 exist = udp_lib_lport_inuse2(net, snum, hslot2,
313 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
317 inet_sk(sk)->inet_num = snum;
318 udp_sk(sk)->udp_port_hash = snum;
319 udp_sk(sk)->udp_portaddr_hash ^= snum;
320 if (sk_unhashed(sk)) {
321 if (sk->sk_reuseport &&
322 udp_reuseport_add_sock(sk, hslot)) {
323 inet_sk(sk)->inet_num = 0;
324 udp_sk(sk)->udp_port_hash = 0;
325 udp_sk(sk)->udp_portaddr_hash ^= snum;
329 sk_add_node_rcu(sk, &hslot->head);
331 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
333 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
334 spin_lock(&hslot2->lock);
335 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
336 sk->sk_family == AF_INET6)
337 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
340 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
343 spin_unlock(&hslot2->lock);
345 sock_set_flag(sk, SOCK_RCU_FREE);
348 spin_unlock_bh(&hslot->lock);
352 EXPORT_SYMBOL(udp_lib_get_port);
354 int udp_v4_get_port(struct sock *sk, unsigned short snum)
356 unsigned int hash2_nulladdr =
357 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
358 unsigned int hash2_partial =
359 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
361 /* precompute partial secondary hash */
362 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
363 return udp_lib_get_port(sk, snum, hash2_nulladdr);
366 static int compute_score(struct sock *sk, struct net *net,
367 __be32 saddr, __be16 sport,
368 __be32 daddr, unsigned short hnum,
372 struct inet_sock *inet;
375 if (!net_eq(sock_net(sk), net) ||
376 udp_sk(sk)->udp_port_hash != hnum ||
380 if (sk->sk_rcv_saddr != daddr)
383 score = (sk->sk_family == PF_INET) ? 2 : 1;
386 if (inet->inet_daddr) {
387 if (inet->inet_daddr != saddr)
392 if (inet->inet_dport) {
393 if (inet->inet_dport != sport)
398 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
402 if (sk->sk_bound_dev_if)
405 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
410 INDIRECT_CALLABLE_SCOPE
411 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
412 const __be32 faddr, const __be16 fport)
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 /* called with rcu_read_lock() */
421 static struct sock *udp4_lib_lookup2(struct net *net,
422 __be32 saddr, __be16 sport,
423 __be32 daddr, unsigned int hnum,
425 struct udp_hslot *hslot2,
428 struct sock *sk, *result;
434 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
435 need_rescore = false;
437 score = compute_score(need_rescore ? result : sk, net, saddr,
438 sport, daddr, hnum, dif, sdif);
439 if (score > badness) {
445 if (sk->sk_state == TCP_ESTABLISHED) {
450 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
451 saddr, sport, daddr, hnum, udp_ehashfn);
457 /* Fall back to scoring if group has connections */
458 if (!reuseport_has_conns(sk))
461 /* Reuseport logic returned an error, keep original score. */
465 /* compute_score is too long of a function to be
466 * inlined, and calling it again here yields
467 * measureable overhead for some
468 * workloads. Work around it by jumping
469 * backwards to rescore 'result'.
478 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
479 * harder than this. -DaveM
481 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
482 __be16 sport, __be32 daddr, __be16 dport, int dif,
483 int sdif, struct udp_table *udptable, struct sk_buff *skb)
485 unsigned short hnum = ntohs(dport);
486 unsigned int hash2, slot2;
487 struct udp_hslot *hslot2;
488 struct sock *result, *sk;
490 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
491 slot2 = hash2 & udptable->mask;
492 hslot2 = &udptable->hash2[slot2];
494 /* Lookup connected or non-wildcard socket */
495 result = udp4_lib_lookup2(net, saddr, sport,
496 daddr, hnum, dif, sdif,
498 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
501 /* Lookup redirect from BPF */
502 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
503 udptable == net->ipv4.udp_table) {
504 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
505 saddr, sport, daddr, hnum, dif,
513 /* Got non-wildcard socket or error on first lookup */
517 /* Lookup wildcard sockets */
518 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
519 slot2 = hash2 & udptable->mask;
520 hslot2 = &udptable->hash2[slot2];
522 result = udp4_lib_lookup2(net, saddr, sport,
523 htonl(INADDR_ANY), hnum, dif, sdif,
530 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
532 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
533 __be16 sport, __be16 dport,
534 struct udp_table *udptable)
536 const struct iphdr *iph = ip_hdr(skb);
538 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
539 iph->daddr, dport, inet_iif(skb),
540 inet_sdif(skb), udptable, skb);
543 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
544 __be16 sport, __be16 dport)
546 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
547 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
548 struct net *net = dev_net(skb->dev);
551 inet_get_iif_sdif(skb, &iif, &sdif);
553 return __udp4_lib_lookup(net, iph->saddr, sport,
554 iph->daddr, dport, iif,
555 sdif, net->ipv4.udp_table, NULL);
558 /* Must be called under rcu_read_lock().
559 * Does increment socket refcount.
561 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
562 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
563 __be32 daddr, __be16 dport, int dif)
567 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
568 dif, 0, net->ipv4.udp_table, NULL);
569 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
573 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
576 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
577 __be16 loc_port, __be32 loc_addr,
578 __be16 rmt_port, __be32 rmt_addr,
579 int dif, int sdif, unsigned short hnum)
581 const struct inet_sock *inet = inet_sk(sk);
583 if (!net_eq(sock_net(sk), net) ||
584 udp_sk(sk)->udp_port_hash != hnum ||
585 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
586 (inet->inet_dport != rmt_port && inet->inet_dport) ||
587 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
588 ipv6_only_sock(sk) ||
589 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
591 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
596 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
597 EXPORT_SYMBOL(udp_encap_needed_key);
599 #if IS_ENABLED(CONFIG_IPV6)
600 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
601 EXPORT_SYMBOL(udpv6_encap_needed_key);
604 void udp_encap_enable(void)
606 static_branch_inc(&udp_encap_needed_key);
608 EXPORT_SYMBOL(udp_encap_enable);
610 void udp_encap_disable(void)
612 static_branch_dec(&udp_encap_needed_key);
614 EXPORT_SYMBOL(udp_encap_disable);
616 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
617 * through error handlers in encapsulations looking for a match.
619 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
623 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
624 int (*handler)(struct sk_buff *skb, u32 info);
625 const struct ip_tunnel_encap_ops *encap;
627 encap = rcu_dereference(iptun_encaps[i]);
630 handler = encap->err_handler;
631 if (handler && !handler(skb, info))
638 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
639 * reversing source and destination port: this will match tunnels that force the
640 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
641 * lwtunnels might actually break this assumption by being configured with
642 * different destination ports on endpoints, in this case we won't be able to
643 * trace ICMP messages back to them.
645 * If this doesn't match any socket, probe tunnels with arbitrary destination
646 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
647 * we've sent packets to won't necessarily match the local destination port.
649 * Then ask the tunnel implementation to match the error against a valid
652 * Return an error if we can't find a match, the socket if we need further
653 * processing, zero otherwise.
655 static struct sock *__udp4_lib_err_encap(struct net *net,
656 const struct iphdr *iph,
658 struct udp_table *udptable,
660 struct sk_buff *skb, u32 info)
662 int (*lookup)(struct sock *sk, struct sk_buff *skb);
663 int network_offset, transport_offset;
666 network_offset = skb_network_offset(skb);
667 transport_offset = skb_transport_offset(skb);
669 /* Network header needs to point to the outer IPv4 header inside ICMP */
670 skb_reset_network_header(skb);
672 /* Transport header needs to point to the UDP header */
673 skb_set_transport_header(skb, iph->ihl << 2);
678 lookup = READ_ONCE(up->encap_err_lookup);
679 if (lookup && lookup(sk, skb))
685 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
686 iph->saddr, uh->dest, skb->dev->ifindex, 0,
691 lookup = READ_ONCE(up->encap_err_lookup);
692 if (!lookup || lookup(sk, skb))
698 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
700 skb_set_transport_header(skb, transport_offset);
701 skb_set_network_header(skb, network_offset);
707 * This routine is called by the ICMP module when it gets some
708 * sort of error condition. If err < 0 then the socket should
709 * be closed and the error returned to the user. If err > 0
710 * it's just the icmp type << 8 | icmp code.
711 * Header points to the ip header of the error packet. We move
712 * on past this. Then (as it used to claim before adjustment)
713 * header points to the first 8 bytes of the udp header. We need
714 * to find the appropriate port.
717 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
719 struct inet_sock *inet;
720 const struct iphdr *iph = (const struct iphdr *)skb->data;
721 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
722 const int type = icmp_hdr(skb)->type;
723 const int code = icmp_hdr(skb)->code;
728 struct net *net = dev_net(skb->dev);
730 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
731 iph->saddr, uh->source, skb->dev->ifindex,
732 inet_sdif(skb), udptable, NULL);
734 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
735 /* No socket for error: try tunnels before discarding */
736 if (static_branch_unlikely(&udp_encap_needed_key)) {
737 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
742 sk = ERR_PTR(-ENOENT);
745 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
758 case ICMP_TIME_EXCEEDED:
761 case ICMP_SOURCE_QUENCH:
763 case ICMP_PARAMETERPROB:
767 case ICMP_DEST_UNREACH:
768 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
769 ipv4_sk_update_pmtu(skb, sk, info);
770 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
778 if (code <= NR_ICMP_UNREACH) {
779 harderr = icmp_err_convert[code].fatal;
780 err = icmp_err_convert[code].errno;
784 ipv4_sk_redirect(skb, sk);
789 * RFC1122: OK. Passes ICMP errors back to application, as per
793 /* ...not for tunnels though: we don't have a sending socket */
794 if (udp_sk(sk)->encap_err_rcv)
795 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
799 if (!inet_test_bit(RECVERR, sk)) {
800 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
803 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
811 int udp_err(struct sk_buff *skb, u32 info)
813 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
817 * Throw away all pending data and cancel the corking. Socket is locked.
819 void udp_flush_pending_frames(struct sock *sk)
821 struct udp_sock *up = udp_sk(sk);
825 WRITE_ONCE(up->pending, 0);
826 ip_flush_pending_frames(sk);
829 EXPORT_SYMBOL(udp_flush_pending_frames);
832 * udp4_hwcsum - handle outgoing HW checksumming
833 * @skb: sk_buff containing the filled-in UDP header
834 * (checksum field must be zeroed out)
835 * @src: source IP address
836 * @dst: destination IP address
838 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
840 struct udphdr *uh = udp_hdr(skb);
841 int offset = skb_transport_offset(skb);
842 int len = skb->len - offset;
846 if (!skb_has_frag_list(skb)) {
848 * Only one fragment on the socket.
850 skb->csum_start = skb_transport_header(skb) - skb->head;
851 skb->csum_offset = offsetof(struct udphdr, check);
852 uh->check = ~csum_tcpudp_magic(src, dst, len,
855 struct sk_buff *frags;
858 * HW-checksum won't work as there are two or more
859 * fragments on the socket so that all csums of sk_buffs
862 skb_walk_frags(skb, frags) {
863 csum = csum_add(csum, frags->csum);
867 csum = skb_checksum(skb, offset, hlen, csum);
868 skb->ip_summed = CHECKSUM_NONE;
870 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
872 uh->check = CSUM_MANGLED_0;
875 EXPORT_SYMBOL_GPL(udp4_hwcsum);
877 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
878 * for the simple case like when setting the checksum for a UDP tunnel.
880 void udp_set_csum(bool nocheck, struct sk_buff *skb,
881 __be32 saddr, __be32 daddr, int len)
883 struct udphdr *uh = udp_hdr(skb);
887 } else if (skb_is_gso(skb)) {
888 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
889 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
891 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
893 uh->check = CSUM_MANGLED_0;
895 skb->ip_summed = CHECKSUM_PARTIAL;
896 skb->csum_start = skb_transport_header(skb) - skb->head;
897 skb->csum_offset = offsetof(struct udphdr, check);
898 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
901 EXPORT_SYMBOL(udp_set_csum);
903 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
904 struct inet_cork *cork)
906 struct sock *sk = skb->sk;
907 struct inet_sock *inet = inet_sk(sk);
910 int is_udplite = IS_UDPLITE(sk);
911 int offset = skb_transport_offset(skb);
912 int len = skb->len - offset;
913 int datalen = len - sizeof(*uh);
917 * Create a UDP header
920 uh->source = inet->inet_sport;
921 uh->dest = fl4->fl4_dport;
922 uh->len = htons(len);
925 if (cork->gso_size) {
926 const int hlen = skb_network_header_len(skb) +
927 sizeof(struct udphdr);
929 if (hlen + cork->gso_size > cork->fragsize) {
933 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
937 if (sk->sk_no_check_tx) {
941 if (is_udplite || dst_xfrm(skb_dst(skb))) {
946 if (datalen > cork->gso_size) {
947 skb_shinfo(skb)->gso_size = cork->gso_size;
948 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
949 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
955 if (is_udplite) /* UDP-Lite */
956 csum = udplite_csum(skb);
958 else if (sk->sk_no_check_tx) { /* UDP csum off */
960 skb->ip_summed = CHECKSUM_NONE;
963 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
966 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
970 csum = udp_csum(skb);
972 /* add protocol-dependent pseudo-header */
973 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
974 sk->sk_protocol, csum);
976 uh->check = CSUM_MANGLED_0;
979 err = ip_send_skb(sock_net(sk), skb);
981 if (err == -ENOBUFS &&
982 !inet_test_bit(RECVERR, sk)) {
983 UDP_INC_STATS(sock_net(sk),
984 UDP_MIB_SNDBUFERRORS, is_udplite);
988 UDP_INC_STATS(sock_net(sk),
989 UDP_MIB_OUTDATAGRAMS, is_udplite);
994 * Push out all pending data as one UDP datagram. Socket is locked.
996 int udp_push_pending_frames(struct sock *sk)
998 struct udp_sock *up = udp_sk(sk);
999 struct inet_sock *inet = inet_sk(sk);
1000 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1001 struct sk_buff *skb;
1004 skb = ip_finish_skb(sk, fl4);
1008 err = udp_send_skb(skb, fl4, &inet->cork.base);
1012 WRITE_ONCE(up->pending, 0);
1015 EXPORT_SYMBOL(udp_push_pending_frames);
1017 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1019 switch (cmsg->cmsg_type) {
1021 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1023 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1030 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1032 struct cmsghdr *cmsg;
1033 bool need_ip = false;
1036 for_each_cmsghdr(cmsg, msg) {
1037 if (!CMSG_OK(msg, cmsg))
1040 if (cmsg->cmsg_level != SOL_UDP) {
1045 err = __udp_cmsg_send(cmsg, gso_size);
1052 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1054 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1056 struct inet_sock *inet = inet_sk(sk);
1057 struct udp_sock *up = udp_sk(sk);
1058 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1059 struct flowi4 fl4_stack;
1062 struct ipcm_cookie ipc;
1063 struct rtable *rt = NULL;
1066 __be32 daddr, faddr, saddr;
1069 int err, is_udplite = IS_UDPLITE(sk);
1070 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1071 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1072 struct sk_buff *skb;
1073 struct ip_options_data opt_copy;
1083 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1086 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1088 fl4 = &inet->cork.fl.u.ip4;
1089 if (READ_ONCE(up->pending)) {
1091 * There are pending frames.
1092 * The socket lock must be held while it's corked.
1095 if (likely(up->pending)) {
1096 if (unlikely(up->pending != AF_INET)) {
1100 goto do_append_data;
1104 ulen += sizeof(struct udphdr);
1107 * Get and verify the address.
1110 if (msg->msg_namelen < sizeof(*usin))
1112 if (usin->sin_family != AF_INET) {
1113 if (usin->sin_family != AF_UNSPEC)
1114 return -EAFNOSUPPORT;
1117 daddr = usin->sin_addr.s_addr;
1118 dport = usin->sin_port;
1122 if (sk->sk_state != TCP_ESTABLISHED)
1123 return -EDESTADDRREQ;
1124 daddr = inet->inet_daddr;
1125 dport = inet->inet_dport;
1126 /* Open fast path for connected socket.
1127 Route will not be used, if at least one option is set.
1132 ipcm_init_sk(&ipc, inet);
1133 ipc.gso_size = READ_ONCE(up->gso_size);
1135 if (msg->msg_controllen) {
1136 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1138 err = ip_cmsg_send(sk, msg, &ipc,
1139 sk->sk_family == AF_INET6);
1142 if (unlikely(err < 0)) {
1150 struct ip_options_rcu *inet_opt;
1153 inet_opt = rcu_dereference(inet->inet_opt);
1155 memcpy(&opt_copy, inet_opt,
1156 sizeof(*inet_opt) + inet_opt->opt.optlen);
1157 ipc.opt = &opt_copy.opt;
1162 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1163 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1164 (struct sockaddr *)usin,
1170 if (usin->sin_port == 0) {
1171 /* BPF program set invalid port. Reject it. */
1175 daddr = usin->sin_addr.s_addr;
1176 dport = usin->sin_port;
1181 ipc.addr = faddr = daddr;
1183 if (ipc.opt && ipc.opt->opt.srr) {
1188 faddr = ipc.opt->opt.faddr;
1191 tos = get_rttos(&ipc, inet);
1192 scope = ip_sendmsg_scope(inet, &ipc, msg);
1193 if (scope == RT_SCOPE_LINK)
1196 uc_index = READ_ONCE(inet->uc_index);
1197 if (ipv4_is_multicast(daddr)) {
1198 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1199 ipc.oif = READ_ONCE(inet->mc_index);
1201 saddr = READ_ONCE(inet->mc_addr);
1203 } else if (!ipc.oif) {
1205 } else if (ipv4_is_lbcast(daddr) && uc_index) {
1206 /* oif is set, packet is to local broadcast and
1207 * uc_index is set. oif is most likely set
1208 * by sk_bound_dev_if. If uc_index != oif check if the
1209 * oif is an L3 master and uc_index is an L3 slave.
1210 * If so, we want to allow the send using the uc_index.
1212 if (ipc.oif != uc_index &&
1213 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1220 rt = dst_rtable(sk_dst_check(sk, 0));
1223 struct net *net = sock_net(sk);
1224 __u8 flow_flags = inet_sk_flowi_flags(sk);
1228 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1229 sk->sk_protocol, flow_flags, faddr, saddr,
1230 dport, inet->inet_sport, sk->sk_uid);
1232 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1233 rt = ip_route_output_flow(net, fl4, sk);
1237 if (err == -ENETUNREACH)
1238 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1243 if ((rt->rt_flags & RTCF_BROADCAST) &&
1244 !sock_flag(sk, SOCK_BROADCAST))
1247 sk_dst_set(sk, dst_clone(&rt->dst));
1250 if (msg->msg_flags&MSG_CONFIRM)
1256 daddr = ipc.addr = fl4->daddr;
1258 /* Lockless fast path for the non-corking case. */
1260 struct inet_cork cork;
1262 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1263 sizeof(struct udphdr), &ipc, &rt,
1264 &cork, msg->msg_flags);
1266 if (!IS_ERR_OR_NULL(skb))
1267 err = udp_send_skb(skb, fl4, &cork);
1272 if (unlikely(up->pending)) {
1273 /* The socket is already corked while preparing it. */
1274 /* ... which is an evident application bug. --ANK */
1277 net_dbg_ratelimited("socket already corked\n");
1282 * Now cork the socket to pend data.
1284 fl4 = &inet->cork.fl.u.ip4;
1287 fl4->fl4_dport = dport;
1288 fl4->fl4_sport = inet->inet_sport;
1289 WRITE_ONCE(up->pending, AF_INET);
1293 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1294 sizeof(struct udphdr), &ipc, &rt,
1295 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1297 udp_flush_pending_frames(sk);
1299 err = udp_push_pending_frames(sk);
1300 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1301 WRITE_ONCE(up->pending, 0);
1312 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1313 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1314 * we don't have a good statistic (IpOutDiscards but it can be too many
1315 * things). We could add another new stat but at least for now that
1316 * seems like overkill.
1318 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1319 UDP_INC_STATS(sock_net(sk),
1320 UDP_MIB_SNDBUFERRORS, is_udplite);
1325 if (msg->msg_flags & MSG_PROBE)
1326 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1327 if (!(msg->msg_flags&MSG_PROBE) || len)
1328 goto back_from_confirm;
1332 EXPORT_SYMBOL(udp_sendmsg);
1334 void udp_splice_eof(struct socket *sock)
1336 struct sock *sk = sock->sk;
1337 struct udp_sock *up = udp_sk(sk);
1339 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1343 if (up->pending && !udp_test_bit(CORK, sk))
1344 udp_push_pending_frames(sk);
1347 EXPORT_SYMBOL_GPL(udp_splice_eof);
1349 #define UDP_SKB_IS_STATELESS 0x80000000
1351 /* all head states (dst, sk, nf conntrack) except skb extensions are
1352 * cleared by udp_rcv().
1354 * We need to preserve secpath, if present, to eventually process
1355 * IP_CMSG_PASSSEC at recvmsg() time.
1357 * Other extensions can be cleared.
1359 static bool udp_try_make_stateless(struct sk_buff *skb)
1361 if (!skb_has_extensions(skb))
1364 if (!secpath_exists(skb)) {
1372 static void udp_set_dev_scratch(struct sk_buff *skb)
1374 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1376 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1377 scratch->_tsize_state = skb->truesize;
1378 #if BITS_PER_LONG == 64
1379 scratch->len = skb->len;
1380 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1381 scratch->is_linear = !skb_is_nonlinear(skb);
1383 if (udp_try_make_stateless(skb))
1384 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1387 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1389 /* We come here after udp_lib_checksum_complete() returned 0.
1390 * This means that __skb_checksum_complete() might have
1391 * set skb->csum_valid to 1.
1392 * On 64bit platforms, we can set csum_unnecessary
1393 * to true, but only if the skb is not shared.
1395 #if BITS_PER_LONG == 64
1396 if (!skb_shared(skb))
1397 udp_skb_scratch(skb)->csum_unnecessary = true;
1401 static int udp_skb_truesize(struct sk_buff *skb)
1403 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1406 static bool udp_skb_has_head_state(struct sk_buff *skb)
1408 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1411 /* fully reclaim rmem/fwd memory allocated for skb */
1412 static void udp_rmem_release(struct sock *sk, int size, int partial,
1413 bool rx_queue_lock_held)
1415 struct udp_sock *up = udp_sk(sk);
1416 struct sk_buff_head *sk_queue;
1419 if (likely(partial)) {
1420 up->forward_deficit += size;
1421 size = up->forward_deficit;
1422 if (size < READ_ONCE(up->forward_threshold) &&
1423 !skb_queue_empty(&up->reader_queue))
1426 size += up->forward_deficit;
1428 up->forward_deficit = 0;
1430 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1431 * if the called don't held it already
1433 sk_queue = &sk->sk_receive_queue;
1434 if (!rx_queue_lock_held)
1435 spin_lock(&sk_queue->lock);
1438 sk_forward_alloc_add(sk, size);
1439 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1440 sk_forward_alloc_add(sk, -amt);
1443 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1445 atomic_sub(size, &sk->sk_rmem_alloc);
1447 /* this can save us from acquiring the rx queue lock on next receive */
1448 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1450 if (!rx_queue_lock_held)
1451 spin_unlock(&sk_queue->lock);
1454 /* Note: called with reader_queue.lock held.
1455 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1456 * This avoids a cache line miss while receive_queue lock is held.
1457 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1459 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1461 prefetch(&skb->data);
1462 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1464 EXPORT_SYMBOL(udp_skb_destructor);
1466 /* as above, but the caller held the rx queue lock, too */
1467 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1469 prefetch(&skb->data);
1470 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1473 /* Idea of busylocks is to let producers grab an extra spinlock
1474 * to relieve pressure on the receive_queue spinlock shared by consumer.
1475 * Under flood, this means that only one producer can be in line
1476 * trying to acquire the receive_queue spinlock.
1477 * These busylock can be allocated on a per cpu manner, instead of a
1478 * per socket one (that would consume a cache line per socket)
1480 static int udp_busylocks_log __read_mostly;
1481 static spinlock_t *udp_busylocks __read_mostly;
1483 static spinlock_t *busylock_acquire(void *ptr)
1487 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1492 static void busylock_release(spinlock_t *busy)
1498 static int udp_rmem_schedule(struct sock *sk, int size)
1502 delta = size - sk->sk_forward_alloc;
1503 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1509 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1511 struct sk_buff_head *list = &sk->sk_receive_queue;
1512 int rmem, err = -ENOMEM;
1513 spinlock_t *busy = NULL;
1514 bool becomes_readable;
1517 /* Immediately drop when the receive queue is full.
1518 * Always allow at least one packet.
1520 rmem = atomic_read(&sk->sk_rmem_alloc);
1521 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1525 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1526 * having linear skbs :
1527 * - Reduce memory overhead and thus increase receive queue capacity
1528 * - Less cache line misses at copyout() time
1529 * - Less work at consume_skb() (less alien page frag freeing)
1531 if (rmem > (rcvbuf >> 1)) {
1534 busy = busylock_acquire(sk);
1536 size = skb->truesize;
1537 udp_set_dev_scratch(skb);
1539 atomic_add(size, &sk->sk_rmem_alloc);
1541 spin_lock(&list->lock);
1542 err = udp_rmem_schedule(sk, size);
1544 spin_unlock(&list->lock);
1548 sk_forward_alloc_add(sk, -size);
1550 /* no need to setup a destructor, we will explicitly release the
1551 * forward allocated memory on dequeue
1553 sock_skb_set_dropcount(sk, skb);
1555 becomes_readable = skb_queue_empty(list);
1556 __skb_queue_tail(list, skb);
1557 spin_unlock(&list->lock);
1559 if (!sock_flag(sk, SOCK_DEAD)) {
1560 if (becomes_readable ||
1561 sk->sk_data_ready != sock_def_readable ||
1562 READ_ONCE(sk->sk_peek_off) >= 0)
1563 INDIRECT_CALL_1(sk->sk_data_ready,
1564 sock_def_readable, sk);
1566 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
1568 busylock_release(busy);
1572 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1575 atomic_inc(&sk->sk_drops);
1576 busylock_release(busy);
1579 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1581 void udp_destruct_common(struct sock *sk)
1583 /* reclaim completely the forward allocated memory */
1584 struct udp_sock *up = udp_sk(sk);
1585 unsigned int total = 0;
1586 struct sk_buff *skb;
1588 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1589 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1590 total += skb->truesize;
1593 udp_rmem_release(sk, total, 0, true);
1595 EXPORT_SYMBOL_GPL(udp_destruct_common);
1597 static void udp_destruct_sock(struct sock *sk)
1599 udp_destruct_common(sk);
1600 inet_sock_destruct(sk);
1603 int udp_init_sock(struct sock *sk)
1605 udp_lib_init_sock(sk);
1606 sk->sk_destruct = udp_destruct_sock;
1607 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1611 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1613 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
1614 sk_peek_offset_bwd(sk, len);
1616 if (!skb_unref(skb))
1619 /* In the more common cases we cleared the head states previously,
1620 * see __udp_queue_rcv_skb().
1622 if (unlikely(udp_skb_has_head_state(skb)))
1623 skb_release_head_state(skb);
1624 __consume_stateless_skb(skb);
1626 EXPORT_SYMBOL_GPL(skb_consume_udp);
1628 static struct sk_buff *__first_packet_length(struct sock *sk,
1629 struct sk_buff_head *rcvq,
1632 struct sk_buff *skb;
1634 while ((skb = skb_peek(rcvq)) != NULL) {
1635 if (udp_lib_checksum_complete(skb)) {
1636 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1638 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1640 atomic_inc(&sk->sk_drops);
1641 __skb_unlink(skb, rcvq);
1642 *total += skb->truesize;
1645 udp_skb_csum_unnecessary_set(skb);
1653 * first_packet_length - return length of first packet in receive queue
1656 * Drops all bad checksum frames, until a valid one is found.
1657 * Returns the length of found skb, or -1 if none is found.
1659 static int first_packet_length(struct sock *sk)
1661 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1662 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1663 struct sk_buff *skb;
1667 spin_lock_bh(&rcvq->lock);
1668 skb = __first_packet_length(sk, rcvq, &total);
1669 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1670 spin_lock(&sk_queue->lock);
1671 skb_queue_splice_tail_init(sk_queue, rcvq);
1672 spin_unlock(&sk_queue->lock);
1674 skb = __first_packet_length(sk, rcvq, &total);
1676 res = skb ? skb->len : -1;
1678 udp_rmem_release(sk, total, 1, false);
1679 spin_unlock_bh(&rcvq->lock);
1684 * IOCTL requests applicable to the UDP protocol
1687 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1692 *karg = sk_wmem_alloc_get(sk);
1698 *karg = max_t(int, 0, first_packet_length(sk));
1703 return -ENOIOCTLCMD;
1708 EXPORT_SYMBOL(udp_ioctl);
1710 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1713 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1714 struct sk_buff_head *queue;
1715 struct sk_buff *last;
1719 queue = &udp_sk(sk)->reader_queue;
1720 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1722 struct sk_buff *skb;
1724 error = sock_error(sk);
1730 spin_lock_bh(&queue->lock);
1731 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1734 if (!(flags & MSG_PEEK))
1735 udp_skb_destructor(sk, skb);
1736 spin_unlock_bh(&queue->lock);
1740 if (skb_queue_empty_lockless(sk_queue)) {
1741 spin_unlock_bh(&queue->lock);
1745 /* refill the reader queue and walk it again
1746 * keep both queues locked to avoid re-acquiring
1747 * the sk_receive_queue lock if fwd memory scheduling
1750 spin_lock(&sk_queue->lock);
1751 skb_queue_splice_tail_init(sk_queue, queue);
1753 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1755 if (skb && !(flags & MSG_PEEK))
1756 udp_skb_dtor_locked(sk, skb);
1757 spin_unlock(&sk_queue->lock);
1758 spin_unlock_bh(&queue->lock);
1763 if (!sk_can_busy_loop(sk))
1766 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1767 } while (!skb_queue_empty_lockless(sk_queue));
1769 /* sk_queue is empty, reader_queue may contain peeked packets */
1771 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1773 (struct sk_buff *)sk_queue));
1778 EXPORT_SYMBOL(__skb_recv_udp);
1780 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1782 struct sk_buff *skb;
1786 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1790 if (udp_lib_checksum_complete(skb)) {
1791 int is_udplite = IS_UDPLITE(sk);
1792 struct net *net = sock_net(sk);
1794 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1795 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1796 atomic_inc(&sk->sk_drops);
1801 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1802 return recv_actor(sk, skb);
1804 EXPORT_SYMBOL(udp_read_skb);
1807 * This should be easy, if there is something there we
1808 * return it, otherwise we block.
1811 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1814 struct inet_sock *inet = inet_sk(sk);
1815 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1816 struct sk_buff *skb;
1817 unsigned int ulen, copied;
1818 int off, err, peeking = flags & MSG_PEEK;
1819 int is_udplite = IS_UDPLITE(sk);
1820 bool checksum_valid = false;
1822 if (flags & MSG_ERRQUEUE)
1823 return ip_recv_error(sk, msg, len, addr_len);
1826 off = sk_peek_offset(sk, flags);
1827 skb = __skb_recv_udp(sk, flags, &off, &err);
1831 ulen = udp_skb_len(skb);
1833 if (copied > ulen - off)
1834 copied = ulen - off;
1835 else if (copied < ulen)
1836 msg->msg_flags |= MSG_TRUNC;
1839 * If checksum is needed at all, try to do it while copying the
1840 * data. If the data is truncated, or if we only want a partial
1841 * coverage checksum (UDP-Lite), do it before the copy.
1844 if (copied < ulen || peeking ||
1845 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1846 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1847 !__udp_lib_checksum_complete(skb);
1848 if (!checksum_valid)
1852 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1853 if (udp_skb_is_linear(skb))
1854 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1856 err = skb_copy_datagram_msg(skb, off, msg, copied);
1858 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1864 if (unlikely(err)) {
1866 atomic_inc(&sk->sk_drops);
1867 UDP_INC_STATS(sock_net(sk),
1868 UDP_MIB_INERRORS, is_udplite);
1875 UDP_INC_STATS(sock_net(sk),
1876 UDP_MIB_INDATAGRAMS, is_udplite);
1878 sock_recv_cmsgs(msg, sk, skb);
1880 /* Copy the address. */
1882 sin->sin_family = AF_INET;
1883 sin->sin_port = udp_hdr(skb)->source;
1884 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1885 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1886 *addr_len = sizeof(*sin);
1888 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1889 (struct sockaddr *)sin,
1893 if (udp_test_bit(GRO_ENABLED, sk))
1894 udp_cmsg_recv(msg, sk, skb);
1896 if (inet_cmsg_flags(inet))
1897 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1900 if (flags & MSG_TRUNC)
1903 skb_consume_udp(sk, skb, peeking ? -err : err);
1907 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1908 udp_skb_destructor)) {
1909 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1910 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1914 /* starting over for a new packet, but check if we need to yield */
1916 msg->msg_flags &= ~MSG_TRUNC;
1920 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1922 /* This check is replicated from __ip4_datagram_connect() and
1923 * intended to prevent BPF program called below from accessing bytes
1924 * that are out of the bound specified by user in addr_len.
1926 if (addr_len < sizeof(struct sockaddr_in))
1929 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
1931 EXPORT_SYMBOL(udp_pre_connect);
1933 int __udp_disconnect(struct sock *sk, int flags)
1935 struct inet_sock *inet = inet_sk(sk);
1937 * 1003.1g - break association.
1940 sk->sk_state = TCP_CLOSE;
1941 inet->inet_daddr = 0;
1942 inet->inet_dport = 0;
1943 sock_rps_reset_rxhash(sk);
1944 sk->sk_bound_dev_if = 0;
1945 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1946 inet_reset_saddr(sk);
1947 if (sk->sk_prot->rehash &&
1948 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1949 sk->sk_prot->rehash(sk);
1952 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1953 sk->sk_prot->unhash(sk);
1954 inet->inet_sport = 0;
1959 EXPORT_SYMBOL(__udp_disconnect);
1961 int udp_disconnect(struct sock *sk, int flags)
1964 __udp_disconnect(sk, flags);
1968 EXPORT_SYMBOL(udp_disconnect);
1970 void udp_lib_unhash(struct sock *sk)
1972 if (sk_hashed(sk)) {
1973 struct udp_table *udptable = udp_get_table_prot(sk);
1974 struct udp_hslot *hslot, *hslot2;
1976 hslot = udp_hashslot(udptable, sock_net(sk),
1977 udp_sk(sk)->udp_port_hash);
1978 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1980 spin_lock_bh(&hslot->lock);
1981 if (rcu_access_pointer(sk->sk_reuseport_cb))
1982 reuseport_detach_sock(sk);
1983 if (sk_del_node_init_rcu(sk)) {
1985 inet_sk(sk)->inet_num = 0;
1986 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1988 spin_lock(&hslot2->lock);
1989 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1991 spin_unlock(&hslot2->lock);
1993 spin_unlock_bh(&hslot->lock);
1996 EXPORT_SYMBOL(udp_lib_unhash);
1999 * inet_rcv_saddr was changed, we must rehash secondary hash
2001 void udp_lib_rehash(struct sock *sk, u16 newhash)
2003 if (sk_hashed(sk)) {
2004 struct udp_table *udptable = udp_get_table_prot(sk);
2005 struct udp_hslot *hslot, *hslot2, *nhslot2;
2007 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2008 nhslot2 = udp_hashslot2(udptable, newhash);
2009 udp_sk(sk)->udp_portaddr_hash = newhash;
2011 if (hslot2 != nhslot2 ||
2012 rcu_access_pointer(sk->sk_reuseport_cb)) {
2013 hslot = udp_hashslot(udptable, sock_net(sk),
2014 udp_sk(sk)->udp_port_hash);
2015 /* we must lock primary chain too */
2016 spin_lock_bh(&hslot->lock);
2017 if (rcu_access_pointer(sk->sk_reuseport_cb))
2018 reuseport_detach_sock(sk);
2020 if (hslot2 != nhslot2) {
2021 spin_lock(&hslot2->lock);
2022 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2024 spin_unlock(&hslot2->lock);
2026 spin_lock(&nhslot2->lock);
2027 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2030 spin_unlock(&nhslot2->lock);
2033 spin_unlock_bh(&hslot->lock);
2037 EXPORT_SYMBOL(udp_lib_rehash);
2039 void udp_v4_rehash(struct sock *sk)
2041 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2042 inet_sk(sk)->inet_rcv_saddr,
2043 inet_sk(sk)->inet_num);
2044 udp_lib_rehash(sk, new_hash);
2047 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2051 if (inet_sk(sk)->inet_daddr) {
2052 sock_rps_save_rxhash(sk, skb);
2053 sk_mark_napi_id(sk, skb);
2054 sk_incoming_cpu_update(sk);
2056 sk_mark_napi_id_once(sk, skb);
2059 rc = __udp_enqueue_schedule_skb(sk, skb);
2061 int is_udplite = IS_UDPLITE(sk);
2064 /* Note that an ENOMEM error is charged twice */
2065 if (rc == -ENOMEM) {
2066 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2068 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2070 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2072 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2074 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2075 trace_udp_fail_queue_rcv_skb(rc, sk, skb);
2076 sk_skb_reason_drop(sk, skb, drop_reason);
2086 * >0: "udp encap" protocol resubmission
2088 * Note that in the success and error cases, the skb is assumed to
2089 * have either been requeued or freed.
2091 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2093 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2094 struct udp_sock *up = udp_sk(sk);
2095 int is_udplite = IS_UDPLITE(sk);
2098 * Charge it to the socket, dropping if the queue is full.
2100 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2101 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2106 if (static_branch_unlikely(&udp_encap_needed_key) &&
2107 READ_ONCE(up->encap_type)) {
2108 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2111 * This is an encapsulation socket so pass the skb to
2112 * the socket's udp_encap_rcv() hook. Otherwise, just
2113 * fall through and pass this up the UDP socket.
2114 * up->encap_rcv() returns the following value:
2115 * =0 if skb was successfully passed to the encap
2116 * handler or was discarded by it.
2117 * >0 if skb should be passed on to UDP.
2118 * <0 if skb should be resubmitted as proto -N
2121 /* if we're overly short, let UDP handle it */
2122 encap_rcv = READ_ONCE(up->encap_rcv);
2126 /* Verify checksum before giving to encap */
2127 if (udp_lib_checksum_complete(skb))
2130 ret = encap_rcv(sk, skb);
2132 __UDP_INC_STATS(sock_net(sk),
2133 UDP_MIB_INDATAGRAMS,
2139 /* FALLTHROUGH -- it's a UDP Packet */
2143 * UDP-Lite specific tests, ignored on UDP sockets
2145 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2146 u16 pcrlen = READ_ONCE(up->pcrlen);
2149 * MIB statistics other than incrementing the error count are
2150 * disabled for the following two types of errors: these depend
2151 * on the application settings, not on the functioning of the
2152 * protocol stack as such.
2154 * RFC 3828 here recommends (sec 3.3): "There should also be a
2155 * way ... to ... at least let the receiving application block
2156 * delivery of packets with coverage values less than a value
2157 * provided by the application."
2159 if (pcrlen == 0) { /* full coverage was set */
2160 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2161 UDP_SKB_CB(skb)->cscov, skb->len);
2164 /* The next case involves violating the min. coverage requested
2165 * by the receiver. This is subtle: if receiver wants x and x is
2166 * greater than the buffersize/MTU then receiver will complain
2167 * that it wants x while sender emits packets of smaller size y.
2168 * Therefore the above ...()->partial_cov statement is essential.
2170 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2171 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2172 UDP_SKB_CB(skb)->cscov, pcrlen);
2177 prefetch(&sk->sk_rmem_alloc);
2178 if (rcu_access_pointer(sk->sk_filter) &&
2179 udp_lib_checksum_complete(skb))
2182 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2183 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2187 udp_csum_pull_header(skb);
2189 ipv4_pktinfo_prepare(sk, skb, true);
2190 return __udp_queue_rcv_skb(sk, skb);
2193 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2194 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2196 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2197 atomic_inc(&sk->sk_drops);
2198 sk_skb_reason_drop(sk, skb, drop_reason);
2202 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2204 struct sk_buff *next, *segs;
2207 if (likely(!udp_unexpected_gso(sk, skb)))
2208 return udp_queue_rcv_one_skb(sk, skb);
2210 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2211 __skb_push(skb, -skb_mac_offset(skb));
2212 segs = udp_rcv_segment(sk, skb, true);
2213 skb_list_walk_safe(segs, skb, next) {
2214 __skb_pull(skb, skb_transport_offset(skb));
2216 udp_post_segment_fix_csum(skb);
2217 ret = udp_queue_rcv_one_skb(sk, skb);
2219 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2224 /* For TCP sockets, sk_rx_dst is protected by socket lock
2225 * For UDP, we use xchg() to guard against concurrent changes.
2227 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2229 struct dst_entry *old;
2231 if (dst_hold_safe(dst)) {
2232 old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst)));
2238 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2241 * Multicasts and broadcasts go to each listener.
2243 * Note: called only from the BH handler context.
2245 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2247 __be32 saddr, __be32 daddr,
2248 struct udp_table *udptable,
2251 struct sock *sk, *first = NULL;
2252 unsigned short hnum = ntohs(uh->dest);
2253 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2254 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2255 unsigned int offset = offsetof(typeof(*sk), sk_node);
2256 int dif = skb->dev->ifindex;
2257 int sdif = inet_sdif(skb);
2258 struct hlist_node *node;
2259 struct sk_buff *nskb;
2262 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2264 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2266 hslot = &udptable->hash2[hash2];
2267 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2270 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2271 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2272 uh->source, saddr, dif, sdif, hnum))
2279 nskb = skb_clone(skb, GFP_ATOMIC);
2281 if (unlikely(!nskb)) {
2282 atomic_inc(&sk->sk_drops);
2283 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2285 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2289 if (udp_queue_rcv_skb(sk, nskb) > 0)
2293 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2294 if (use_hash2 && hash2 != hash2_any) {
2300 if (udp_queue_rcv_skb(first, skb) > 0)
2304 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2305 proto == IPPROTO_UDPLITE);
2310 /* Initialize UDP checksum. If exited with zero value (success),
2311 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2312 * Otherwise, csum completion requires checksumming packet body,
2313 * including udp header and folding it to skb->csum.
2315 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2320 UDP_SKB_CB(skb)->partial_cov = 0;
2321 UDP_SKB_CB(skb)->cscov = skb->len;
2323 if (proto == IPPROTO_UDPLITE) {
2324 err = udplite_checksum_init(skb, uh);
2328 if (UDP_SKB_CB(skb)->partial_cov) {
2329 skb->csum = inet_compute_pseudo(skb, proto);
2334 /* Note, we are only interested in != 0 or == 0, thus the
2337 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2338 inet_compute_pseudo);
2342 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2343 /* If SW calculated the value, we know it's bad */
2344 if (skb->csum_complete_sw)
2347 /* HW says the value is bad. Let's validate that.
2348 * skb->csum is no longer the full packet checksum,
2349 * so don't treat it as such.
2351 skb_checksum_complete_unset(skb);
2357 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2358 * return code conversion for ip layer consumption
2360 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2365 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2366 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2368 ret = udp_queue_rcv_skb(sk, skb);
2370 /* a return value > 0 means to resubmit the input, but
2371 * it wants the return to be -protocol, or 0
2379 * All we need to do is get the socket, and then do a checksum.
2382 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2385 struct sock *sk = NULL;
2387 unsigned short ulen;
2388 struct rtable *rt = skb_rtable(skb);
2389 __be32 saddr, daddr;
2390 struct net *net = dev_net(skb->dev);
2394 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2397 * Validate the packet.
2399 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2400 goto drop; /* No space for header. */
2403 ulen = ntohs(uh->len);
2404 saddr = ip_hdr(skb)->saddr;
2405 daddr = ip_hdr(skb)->daddr;
2407 if (ulen > skb->len)
2410 if (proto == IPPROTO_UDP) {
2411 /* UDP validates ulen. */
2412 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2417 if (udp4_csum_init(skb, uh, proto))
2420 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2421 &refcounted, udp_ehashfn);
2426 struct dst_entry *dst = skb_dst(skb);
2429 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2430 udp_sk_rx_dst_set(sk, dst);
2432 ret = udp_unicast_rcv_skb(sk, skb, uh);
2438 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2439 return __udp4_lib_mcast_deliver(net, skb, uh,
2440 saddr, daddr, udptable, proto);
2442 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2444 return udp_unicast_rcv_skb(sk, skb, uh);
2446 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2450 /* No socket. Drop packet silently, if checksum is wrong */
2451 if (udp_lib_checksum_complete(skb))
2454 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2455 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2456 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2459 * Hmm. We got an UDP packet to a port to which we
2460 * don't wanna listen. Ignore it.
2462 sk_skb_reason_drop(sk, skb, drop_reason);
2466 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2467 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2468 proto == IPPROTO_UDPLITE ? "Lite" : "",
2469 &saddr, ntohs(uh->source),
2471 &daddr, ntohs(uh->dest));
2476 * RFC1122: OK. Discards the bad packet silently (as far as
2477 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2479 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2480 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2481 proto == IPPROTO_UDPLITE ? "Lite" : "",
2482 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2484 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2486 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2487 sk_skb_reason_drop(sk, skb, drop_reason);
2491 /* We can only early demux multicast if there is a single matching socket.
2492 * If more than one socket found returns NULL
2494 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2495 __be16 loc_port, __be32 loc_addr,
2496 __be16 rmt_port, __be32 rmt_addr,
2499 struct udp_table *udptable = net->ipv4.udp_table;
2500 unsigned short hnum = ntohs(loc_port);
2501 struct sock *sk, *result;
2502 struct udp_hslot *hslot;
2505 slot = udp_hashfn(net, hnum, udptable->mask);
2506 hslot = &udptable->hash[slot];
2508 /* Do not bother scanning a too big list */
2509 if (hslot->count > 10)
2513 sk_for_each_rcu(sk, &hslot->head) {
2514 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2515 rmt_port, rmt_addr, dif, sdif, hnum)) {
2525 /* For unicast we should only early demux connected sockets or we can
2526 * break forwarding setups. The chains here can be long so only check
2527 * if the first socket is an exact match and if not move on.
2529 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2530 __be16 loc_port, __be32 loc_addr,
2531 __be16 rmt_port, __be32 rmt_addr,
2534 struct udp_table *udptable = net->ipv4.udp_table;
2535 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2536 unsigned short hnum = ntohs(loc_port);
2537 unsigned int hash2, slot2;
2538 struct udp_hslot *hslot2;
2542 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2543 slot2 = hash2 & udptable->mask;
2544 hslot2 = &udptable->hash2[slot2];
2545 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2547 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2548 if (inet_match(net, sk, acookie, ports, dif, sdif))
2550 /* Only check first socket in chain */
2556 int udp_v4_early_demux(struct sk_buff *skb)
2558 struct net *net = dev_net(skb->dev);
2559 struct in_device *in_dev = NULL;
2560 const struct iphdr *iph;
2561 const struct udphdr *uh;
2562 struct sock *sk = NULL;
2563 struct dst_entry *dst;
2564 int dif = skb->dev->ifindex;
2565 int sdif = inet_sdif(skb);
2568 /* validate the packet */
2569 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2575 if (skb->pkt_type == PACKET_MULTICAST) {
2576 in_dev = __in_dev_get_rcu(skb->dev);
2581 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2586 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2587 uh->source, iph->saddr,
2589 } else if (skb->pkt_type == PACKET_HOST) {
2590 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2591 uh->source, iph->saddr, dif, sdif);
2598 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2599 skb->destructor = sock_pfree;
2600 dst = rcu_dereference(sk->sk_rx_dst);
2603 dst = dst_check(dst, 0);
2607 /* set noref for now.
2608 * any place which wants to hold dst has to call
2611 skb_dst_set_noref(skb, dst);
2613 /* for unconnected multicast sockets we need to validate
2614 * the source on each packet
2616 if (!inet_sk(sk)->inet_daddr && in_dev)
2617 return ip_mc_validate_source(skb, iph->daddr,
2619 iph->tos & IPTOS_RT_MASK,
2620 skb->dev, in_dev, &itag);
2625 int udp_rcv(struct sk_buff *skb)
2627 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2630 void udp_destroy_sock(struct sock *sk)
2632 struct udp_sock *up = udp_sk(sk);
2633 bool slow = lock_sock_fast(sk);
2635 /* protects from races with udp_abort() */
2636 sock_set_flag(sk, SOCK_DEAD);
2637 udp_flush_pending_frames(sk);
2638 unlock_sock_fast(sk, slow);
2639 if (static_branch_unlikely(&udp_encap_needed_key)) {
2640 if (up->encap_type) {
2641 void (*encap_destroy)(struct sock *sk);
2642 encap_destroy = READ_ONCE(up->encap_destroy);
2646 if (udp_test_bit(ENCAP_ENABLED, sk))
2647 static_branch_dec(&udp_encap_needed_key);
2651 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2655 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2656 if (family == AF_INET)
2657 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2658 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2659 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2665 * Socket option code for UDP
2667 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2668 sockptr_t optval, unsigned int optlen,
2669 int (*push_pending_frames)(struct sock *))
2671 struct udp_sock *up = udp_sk(sk);
2674 int is_udplite = IS_UDPLITE(sk);
2676 if (level == SOL_SOCKET) {
2677 err = sk_setsockopt(sk, level, optname, optval, optlen);
2679 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2680 sockopt_lock_sock(sk);
2681 /* paired with READ_ONCE in udp_rmem_release() */
2682 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2683 sockopt_release_sock(sk);
2688 if (optlen < sizeof(int))
2691 if (copy_from_sockptr(&val, optval, sizeof(val)))
2694 valbool = val ? 1 : 0;
2699 udp_set_bit(CORK, sk);
2701 udp_clear_bit(CORK, sk);
2703 push_pending_frames(sk);
2712 case UDP_ENCAP_ESPINUDP:
2713 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2714 #if IS_ENABLED(CONFIG_IPV6)
2715 if (sk->sk_family == AF_INET6)
2716 WRITE_ONCE(up->encap_rcv,
2717 ipv6_stub->xfrm6_udp_encap_rcv);
2720 WRITE_ONCE(up->encap_rcv,
2721 xfrm4_udp_encap_rcv);
2724 case UDP_ENCAP_L2TPINUDP:
2725 WRITE_ONCE(up->encap_type, val);
2726 udp_tunnel_encap_enable(sk);
2734 case UDP_NO_CHECK6_TX:
2735 udp_set_no_check6_tx(sk, valbool);
2738 case UDP_NO_CHECK6_RX:
2739 udp_set_no_check6_rx(sk, valbool);
2743 if (val < 0 || val > USHRT_MAX)
2745 WRITE_ONCE(up->gso_size, val);
2750 /* when enabling GRO, accept the related GSO packet type */
2752 udp_tunnel_encap_enable(sk);
2753 udp_assign_bit(GRO_ENABLED, sk, valbool);
2754 udp_assign_bit(ACCEPT_L4, sk, valbool);
2755 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
2759 * UDP-Lite's partial checksum coverage (RFC 3828).
2761 /* The sender sets actual checksum coverage length via this option.
2762 * The case coverage > packet length is handled by send module. */
2763 case UDPLITE_SEND_CSCOV:
2764 if (!is_udplite) /* Disable the option on UDP sockets */
2765 return -ENOPROTOOPT;
2766 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2768 else if (val > USHRT_MAX)
2770 WRITE_ONCE(up->pcslen, val);
2771 udp_set_bit(UDPLITE_SEND_CC, sk);
2774 /* The receiver specifies a minimum checksum coverage value. To make
2775 * sense, this should be set to at least 8 (as done below). If zero is
2776 * used, this again means full checksum coverage. */
2777 case UDPLITE_RECV_CSCOV:
2778 if (!is_udplite) /* Disable the option on UDP sockets */
2779 return -ENOPROTOOPT;
2780 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2782 else if (val > USHRT_MAX)
2784 WRITE_ONCE(up->pcrlen, val);
2785 udp_set_bit(UDPLITE_RECV_CC, sk);
2795 EXPORT_SYMBOL(udp_lib_setsockopt);
2797 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2798 unsigned int optlen)
2800 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2801 return udp_lib_setsockopt(sk, level, optname,
2803 udp_push_pending_frames);
2804 return ip_setsockopt(sk, level, optname, optval, optlen);
2807 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2808 char __user *optval, int __user *optlen)
2810 struct udp_sock *up = udp_sk(sk);
2813 if (get_user(len, optlen))
2819 len = min_t(unsigned int, len, sizeof(int));
2823 val = udp_test_bit(CORK, sk);
2827 val = READ_ONCE(up->encap_type);
2830 case UDP_NO_CHECK6_TX:
2831 val = udp_get_no_check6_tx(sk);
2834 case UDP_NO_CHECK6_RX:
2835 val = udp_get_no_check6_rx(sk);
2839 val = READ_ONCE(up->gso_size);
2843 val = udp_test_bit(GRO_ENABLED, sk);
2846 /* The following two cannot be changed on UDP sockets, the return is
2847 * always 0 (which corresponds to the full checksum coverage of UDP). */
2848 case UDPLITE_SEND_CSCOV:
2849 val = READ_ONCE(up->pcslen);
2852 case UDPLITE_RECV_CSCOV:
2853 val = READ_ONCE(up->pcrlen);
2857 return -ENOPROTOOPT;
2860 if (put_user(len, optlen))
2862 if (copy_to_user(optval, &val, len))
2866 EXPORT_SYMBOL(udp_lib_getsockopt);
2868 int udp_getsockopt(struct sock *sk, int level, int optname,
2869 char __user *optval, int __user *optlen)
2871 if (level == SOL_UDP || level == SOL_UDPLITE)
2872 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2873 return ip_getsockopt(sk, level, optname, optval, optlen);
2877 * udp_poll - wait for a UDP event.
2878 * @file: - file struct
2880 * @wait: - poll table
2882 * This is same as datagram poll, except for the special case of
2883 * blocking sockets. If application is using a blocking fd
2884 * and a packet with checksum error is in the queue;
2885 * then it could get return from select indicating data available
2886 * but then block when reading it. Add special case code
2887 * to work around these arguably broken applications.
2889 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2891 __poll_t mask = datagram_poll(file, sock, wait);
2892 struct sock *sk = sock->sk;
2894 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2895 mask |= EPOLLIN | EPOLLRDNORM;
2897 /* Check for false positives due to checksum errors */
2898 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2899 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2900 mask &= ~(EPOLLIN | EPOLLRDNORM);
2902 /* psock ingress_msg queue should not contain any bad checksum frames */
2903 if (sk_is_readable(sk))
2904 mask |= EPOLLIN | EPOLLRDNORM;
2908 EXPORT_SYMBOL(udp_poll);
2910 int udp_abort(struct sock *sk, int err)
2912 if (!has_current_bpf_ctx())
2915 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2918 if (sock_flag(sk, SOCK_DEAD))
2922 sk_error_report(sk);
2923 __udp_disconnect(sk, 0);
2926 if (!has_current_bpf_ctx())
2931 EXPORT_SYMBOL_GPL(udp_abort);
2933 struct proto udp_prot = {
2935 .owner = THIS_MODULE,
2936 .close = udp_lib_close,
2937 .pre_connect = udp_pre_connect,
2938 .connect = ip4_datagram_connect,
2939 .disconnect = udp_disconnect,
2941 .init = udp_init_sock,
2942 .destroy = udp_destroy_sock,
2943 .setsockopt = udp_setsockopt,
2944 .getsockopt = udp_getsockopt,
2945 .sendmsg = udp_sendmsg,
2946 .recvmsg = udp_recvmsg,
2947 .splice_eof = udp_splice_eof,
2948 .release_cb = ip4_datagram_release_cb,
2949 .hash = udp_lib_hash,
2950 .unhash = udp_lib_unhash,
2951 .rehash = udp_v4_rehash,
2952 .get_port = udp_v4_get_port,
2953 .put_port = udp_lib_unhash,
2954 #ifdef CONFIG_BPF_SYSCALL
2955 .psock_update_sk_prot = udp_bpf_update_proto,
2957 .memory_allocated = &udp_memory_allocated,
2958 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2960 .sysctl_mem = sysctl_udp_mem,
2961 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2962 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2963 .obj_size = sizeof(struct udp_sock),
2964 .h.udp_table = NULL,
2965 .diag_destroy = udp_abort,
2967 EXPORT_SYMBOL(udp_prot);
2969 /* ------------------------------------------------------------------------ */
2970 #ifdef CONFIG_PROC_FS
2972 static unsigned short seq_file_family(const struct seq_file *seq);
2973 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2975 unsigned short family = seq_file_family(seq);
2977 /* AF_UNSPEC is used as a match all */
2978 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2979 net_eq(sock_net(sk), seq_file_net(seq)));
2982 #ifdef CONFIG_BPF_SYSCALL
2983 static const struct seq_operations bpf_iter_udp_seq_ops;
2985 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2988 const struct udp_seq_afinfo *afinfo;
2990 #ifdef CONFIG_BPF_SYSCALL
2991 if (seq->op == &bpf_iter_udp_seq_ops)
2992 return net->ipv4.udp_table;
2995 afinfo = pde_data(file_inode(seq->file));
2996 return afinfo->udp_table ? : net->ipv4.udp_table;
2999 static struct sock *udp_get_first(struct seq_file *seq, int start)
3001 struct udp_iter_state *state = seq->private;
3002 struct net *net = seq_file_net(seq);
3003 struct udp_table *udptable;
3006 udptable = udp_get_table_seq(seq, net);
3008 for (state->bucket = start; state->bucket <= udptable->mask;
3010 struct udp_hslot *hslot = &udptable->hash[state->bucket];
3012 if (hlist_empty(&hslot->head))
3015 spin_lock_bh(&hslot->lock);
3016 sk_for_each(sk, &hslot->head) {
3017 if (seq_sk_match(seq, sk))
3020 spin_unlock_bh(&hslot->lock);
3027 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3029 struct udp_iter_state *state = seq->private;
3030 struct net *net = seq_file_net(seq);
3031 struct udp_table *udptable;
3035 } while (sk && !seq_sk_match(seq, sk));
3038 udptable = udp_get_table_seq(seq, net);
3040 if (state->bucket <= udptable->mask)
3041 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3043 return udp_get_first(seq, state->bucket + 1);
3048 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3050 struct sock *sk = udp_get_first(seq, 0);
3053 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3055 return pos ? NULL : sk;
3058 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3060 struct udp_iter_state *state = seq->private;
3061 state->bucket = MAX_UDP_PORTS;
3063 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3065 EXPORT_SYMBOL(udp_seq_start);
3067 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3071 if (v == SEQ_START_TOKEN)
3072 sk = udp_get_idx(seq, 0);
3074 sk = udp_get_next(seq, v);
3079 EXPORT_SYMBOL(udp_seq_next);
3081 void udp_seq_stop(struct seq_file *seq, void *v)
3083 struct udp_iter_state *state = seq->private;
3084 struct udp_table *udptable;
3086 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3088 if (state->bucket <= udptable->mask)
3089 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3091 EXPORT_SYMBOL(udp_seq_stop);
3093 /* ------------------------------------------------------------------------ */
3094 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3097 struct inet_sock *inet = inet_sk(sp);
3098 __be32 dest = inet->inet_daddr;
3099 __be32 src = inet->inet_rcv_saddr;
3100 __u16 destp = ntohs(inet->inet_dport);
3101 __u16 srcp = ntohs(inet->inet_sport);
3103 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3104 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3105 bucket, src, srcp, dest, destp, sp->sk_state,
3106 sk_wmem_alloc_get(sp),
3109 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3111 refcount_read(&sp->sk_refcnt), sp,
3112 atomic_read(&sp->sk_drops));
3115 int udp4_seq_show(struct seq_file *seq, void *v)
3117 seq_setwidth(seq, 127);
3118 if (v == SEQ_START_TOKEN)
3119 seq_puts(seq, " sl local_address rem_address st tx_queue "
3120 "rx_queue tr tm->when retrnsmt uid timeout "
3121 "inode ref pointer drops");
3123 struct udp_iter_state *state = seq->private;
3125 udp4_format_sock(v, seq, state->bucket);
3131 #ifdef CONFIG_BPF_SYSCALL
3132 struct bpf_iter__udp {
3133 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3134 __bpf_md_ptr(struct udp_sock *, udp_sk);
3135 uid_t uid __aligned(8);
3136 int bucket __aligned(8);
3139 struct bpf_udp_iter_state {
3140 struct udp_iter_state state;
3141 unsigned int cur_sk;
3142 unsigned int end_sk;
3143 unsigned int max_sk;
3145 struct sock **batch;
3146 bool st_bucket_done;
3149 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3150 unsigned int new_batch_sz);
3151 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3153 struct bpf_udp_iter_state *iter = seq->private;
3154 struct udp_iter_state *state = &iter->state;
3155 struct net *net = seq_file_net(seq);
3156 int resume_bucket, resume_offset;
3157 struct udp_table *udptable;
3158 unsigned int batch_sks = 0;
3159 bool resized = false;
3162 resume_bucket = state->bucket;
3163 resume_offset = iter->offset;
3165 /* The current batch is done, so advance the bucket. */
3166 if (iter->st_bucket_done)
3169 udptable = udp_get_table_seq(seq, net);
3172 /* New batch for the next bucket.
3173 * Iterate over the hash table to find a bucket with sockets matching
3174 * the iterator attributes, and return the first matching socket from
3175 * the bucket. The remaining matched sockets from the bucket are batched
3176 * before releasing the bucket lock. This allows BPF programs that are
3177 * called in seq_show to acquire the bucket lock if needed.
3181 iter->st_bucket_done = false;
3184 for (; state->bucket <= udptable->mask; state->bucket++) {
3185 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3187 if (hlist_empty(&hslot2->head))
3191 spin_lock_bh(&hslot2->lock);
3192 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3193 if (seq_sk_match(seq, sk)) {
3194 /* Resume from the last iterated socket at the
3195 * offset in the bucket before iterator was stopped.
3197 if (state->bucket == resume_bucket &&
3198 iter->offset < resume_offset) {
3202 if (iter->end_sk < iter->max_sk) {
3204 iter->batch[iter->end_sk++] = sk;
3209 spin_unlock_bh(&hslot2->lock);
3215 /* All done: no batch made. */
3219 if (iter->end_sk == batch_sks) {
3220 /* Batching is done for the current bucket; return the first
3221 * socket to be iterated from the batch.
3223 iter->st_bucket_done = true;
3226 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3228 /* After allocating a larger batch, retry one more time to grab
3234 return iter->batch[0];
3237 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3239 struct bpf_udp_iter_state *iter = seq->private;
3242 /* Whenever seq_next() is called, the iter->cur_sk is
3243 * done with seq_show(), so unref the iter->cur_sk.
3245 if (iter->cur_sk < iter->end_sk) {
3246 sock_put(iter->batch[iter->cur_sk++]);
3250 /* After updating iter->cur_sk, check if there are more sockets
3251 * available in the current bucket batch.
3253 if (iter->cur_sk < iter->end_sk)
3254 sk = iter->batch[iter->cur_sk];
3256 /* Prepare a new batch. */
3257 sk = bpf_iter_udp_batch(seq);
3263 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3265 /* bpf iter does not support lseek, so it always
3266 * continue from where it was stop()-ped.
3269 return bpf_iter_udp_batch(seq);
3271 return SEQ_START_TOKEN;
3274 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3275 struct udp_sock *udp_sk, uid_t uid, int bucket)
3277 struct bpf_iter__udp ctx;
3279 meta->seq_num--; /* skip SEQ_START_TOKEN */
3281 ctx.udp_sk = udp_sk;
3283 ctx.bucket = bucket;
3284 return bpf_iter_run_prog(prog, &ctx);
3287 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3289 struct udp_iter_state *state = seq->private;
3290 struct bpf_iter_meta meta;
3291 struct bpf_prog *prog;
3292 struct sock *sk = v;
3296 if (v == SEQ_START_TOKEN)
3301 if (unlikely(sk_unhashed(sk))) {
3306 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3308 prog = bpf_iter_get_info(&meta, false);
3309 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3316 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3318 while (iter->cur_sk < iter->end_sk)
3319 sock_put(iter->batch[iter->cur_sk++]);
3322 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3324 struct bpf_udp_iter_state *iter = seq->private;
3325 struct bpf_iter_meta meta;
3326 struct bpf_prog *prog;
3330 prog = bpf_iter_get_info(&meta, true);
3332 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3335 if (iter->cur_sk < iter->end_sk) {
3336 bpf_iter_udp_put_batch(iter);
3337 iter->st_bucket_done = false;
3341 static const struct seq_operations bpf_iter_udp_seq_ops = {
3342 .start = bpf_iter_udp_seq_start,
3343 .next = bpf_iter_udp_seq_next,
3344 .stop = bpf_iter_udp_seq_stop,
3345 .show = bpf_iter_udp_seq_show,
3349 static unsigned short seq_file_family(const struct seq_file *seq)
3351 const struct udp_seq_afinfo *afinfo;
3353 #ifdef CONFIG_BPF_SYSCALL
3354 /* BPF iterator: bpf programs to filter sockets. */
3355 if (seq->op == &bpf_iter_udp_seq_ops)
3359 /* Proc fs iterator */
3360 afinfo = pde_data(file_inode(seq->file));
3361 return afinfo->family;
3364 const struct seq_operations udp_seq_ops = {
3365 .start = udp_seq_start,
3366 .next = udp_seq_next,
3367 .stop = udp_seq_stop,
3368 .show = udp4_seq_show,
3370 EXPORT_SYMBOL(udp_seq_ops);
3372 static struct udp_seq_afinfo udp4_seq_afinfo = {
3377 static int __net_init udp4_proc_init_net(struct net *net)
3379 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3380 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3385 static void __net_exit udp4_proc_exit_net(struct net *net)
3387 remove_proc_entry("udp", net->proc_net);
3390 static struct pernet_operations udp4_net_ops = {
3391 .init = udp4_proc_init_net,
3392 .exit = udp4_proc_exit_net,
3395 int __init udp4_proc_init(void)
3397 return register_pernet_subsys(&udp4_net_ops);
3400 void udp4_proc_exit(void)
3402 unregister_pernet_subsys(&udp4_net_ops);
3404 #endif /* CONFIG_PROC_FS */
3406 static __initdata unsigned long uhash_entries;
3407 static int __init set_uhash_entries(char *str)
3414 ret = kstrtoul(str, 0, &uhash_entries);
3418 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3419 uhash_entries = UDP_HTABLE_SIZE_MIN;
3422 __setup("uhash_entries=", set_uhash_entries);
3424 void __init udp_table_init(struct udp_table *table, const char *name)
3428 table->hash = alloc_large_system_hash(name,
3429 2 * sizeof(struct udp_hslot),
3431 21, /* one slot per 2 MB */
3435 UDP_HTABLE_SIZE_MIN,
3436 UDP_HTABLE_SIZE_MAX);
3438 table->hash2 = table->hash + (table->mask + 1);
3439 for (i = 0; i <= table->mask; i++) {
3440 INIT_HLIST_HEAD(&table->hash[i].head);
3441 table->hash[i].count = 0;
3442 spin_lock_init(&table->hash[i].lock);
3444 for (i = 0; i <= table->mask; i++) {
3445 INIT_HLIST_HEAD(&table->hash2[i].head);
3446 table->hash2[i].count = 0;
3447 spin_lock_init(&table->hash2[i].lock);
3451 u32 udp_flow_hashrnd(void)
3453 static u32 hashrnd __read_mostly;
3455 net_get_random_once(&hashrnd, sizeof(hashrnd));
3459 EXPORT_SYMBOL(udp_flow_hashrnd);
3461 static void __net_init udp_sysctl_init(struct net *net)
3463 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3464 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3466 #ifdef CONFIG_NET_L3_MASTER_DEV
3467 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3471 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3473 struct udp_table *udptable;
3476 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3480 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3481 GFP_KERNEL_ACCOUNT);
3482 if (!udptable->hash)
3485 udptable->hash2 = udptable->hash + hash_entries;
3486 udptable->mask = hash_entries - 1;
3487 udptable->log = ilog2(hash_entries);
3489 for (i = 0; i < hash_entries; i++) {
3490 INIT_HLIST_HEAD(&udptable->hash[i].head);
3491 udptable->hash[i].count = 0;
3492 spin_lock_init(&udptable->hash[i].lock);
3494 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3495 udptable->hash2[i].count = 0;
3496 spin_lock_init(&udptable->hash2[i].lock);
3507 static void __net_exit udp_pernet_table_free(struct net *net)
3509 struct udp_table *udptable = net->ipv4.udp_table;
3511 if (udptable == &udp_table)
3514 kvfree(udptable->hash);
3518 static void __net_init udp_set_table(struct net *net)
3520 struct udp_table *udptable;
3521 unsigned int hash_entries;
3522 struct net *old_net;
3524 if (net_eq(net, &init_net))
3527 old_net = current->nsproxy->net_ns;
3528 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3532 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3533 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3534 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3536 hash_entries = roundup_pow_of_two(hash_entries);
3538 udptable = udp_pernet_table_alloc(hash_entries);
3540 net->ipv4.udp_table = udptable;
3542 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3543 "for a netns, fallback to the global one\n",
3546 net->ipv4.udp_table = &udp_table;
3550 static int __net_init udp_pernet_init(struct net *net)
3552 udp_sysctl_init(net);
3558 static void __net_exit udp_pernet_exit(struct net *net)
3560 udp_pernet_table_free(net);
3563 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3564 .init = udp_pernet_init,
3565 .exit = udp_pernet_exit,
3568 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3569 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3570 struct udp_sock *udp_sk, uid_t uid, int bucket)
3572 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3573 unsigned int new_batch_sz)
3575 struct sock **new_batch;
3577 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3578 GFP_USER | __GFP_NOWARN);
3582 bpf_iter_udp_put_batch(iter);
3583 kvfree(iter->batch);
3584 iter->batch = new_batch;
3585 iter->max_sk = new_batch_sz;
3590 #define INIT_BATCH_SZ 16
3592 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3594 struct bpf_udp_iter_state *iter = priv_data;
3597 ret = bpf_iter_init_seq_net(priv_data, aux);
3601 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3603 bpf_iter_fini_seq_net(priv_data);
3608 static void bpf_iter_fini_udp(void *priv_data)
3610 struct bpf_udp_iter_state *iter = priv_data;
3612 bpf_iter_fini_seq_net(priv_data);
3613 kvfree(iter->batch);
3616 static const struct bpf_iter_seq_info udp_seq_info = {
3617 .seq_ops = &bpf_iter_udp_seq_ops,
3618 .init_seq_private = bpf_iter_init_udp,
3619 .fini_seq_private = bpf_iter_fini_udp,
3620 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3623 static struct bpf_iter_reg udp_reg_info = {
3625 .ctx_arg_info_size = 1,
3627 { offsetof(struct bpf_iter__udp, udp_sk),
3628 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3630 .seq_info = &udp_seq_info,
3633 static void __init bpf_iter_register(void)
3635 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3636 if (bpf_iter_reg_target(&udp_reg_info))
3637 pr_warn("Warning: could not register bpf iterator udp\n");
3641 void __init udp_init(void)
3643 unsigned long limit;
3646 udp_table_init(&udp_table, "UDP");
3647 limit = nr_free_buffer_pages() / 8;
3648 limit = max(limit, 128UL);
3649 sysctl_udp_mem[0] = limit / 4 * 3;
3650 sysctl_udp_mem[1] = limit;
3651 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3653 /* 16 spinlocks per cpu */
3654 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3655 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3658 panic("UDP: failed to alloc udp_busylocks\n");
3659 for (i = 0; i < (1U << udp_busylocks_log); i++)
3660 spin_lock_init(udp_busylocks + i);
3662 if (register_pernet_subsys(&udp_sysctl_ops))
3663 panic("UDP: failed to init sysctl parameters.\n");
3665 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3666 bpf_iter_register();