1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * The User Datagram Protocol (UDP).
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
13 * Hirokazu Takahashi, <taka@valinux.co.jp>
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
72 * James Chapman : Add L2TP encapsulation type.
75 #define pr_fmt(fmt) "UDP: " fmt
77 #include <linux/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>
117 #if IS_ENABLED(CONFIG_IPV6)
118 #include <net/ipv6_stubs.h>
121 struct udp_table udp_table __read_mostly;
122 EXPORT_SYMBOL(udp_table);
124 long sysctl_udp_mem[3] __read_mostly;
125 EXPORT_SYMBOL(sysctl_udp_mem);
127 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
128 EXPORT_SYMBOL(udp_memory_allocated);
129 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
130 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
132 #define MAX_UDP_PORTS 65536
133 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
135 static struct udp_table *udp_get_table_prot(struct sock *sk)
137 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
140 static int udp_lib_lport_inuse(struct net *net, __u16 num,
141 const struct udp_hslot *hslot,
142 unsigned long *bitmap,
143 struct sock *sk, unsigned int log)
146 kuid_t uid = sock_i_uid(sk);
148 sk_for_each(sk2, &hslot->head) {
149 if (net_eq(sock_net(sk2), net) &&
151 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
152 (!sk2->sk_reuse || !sk->sk_reuse) &&
153 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
154 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
155 inet_rcv_saddr_equal(sk, sk2, true)) {
156 if (sk2->sk_reuseport && sk->sk_reuseport &&
157 !rcu_access_pointer(sk->sk_reuseport_cb) &&
158 uid_eq(uid, sock_i_uid(sk2))) {
164 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
173 * Note: we still hold spinlock of primary hash chain, so no other writer
174 * can insert/delete a socket with local_port == num
176 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
177 struct udp_hslot *hslot2,
181 kuid_t uid = sock_i_uid(sk);
184 spin_lock(&hslot2->lock);
185 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
186 if (net_eq(sock_net(sk2), net) &&
188 (udp_sk(sk2)->udp_port_hash == num) &&
189 (!sk2->sk_reuse || !sk->sk_reuse) &&
190 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
191 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
192 inet_rcv_saddr_equal(sk, sk2, true)) {
193 if (sk2->sk_reuseport && sk->sk_reuseport &&
194 !rcu_access_pointer(sk->sk_reuseport_cb) &&
195 uid_eq(uid, sock_i_uid(sk2))) {
203 spin_unlock(&hslot2->lock);
207 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
209 struct net *net = sock_net(sk);
210 kuid_t uid = sock_i_uid(sk);
213 sk_for_each(sk2, &hslot->head) {
214 if (net_eq(sock_net(sk2), net) &&
216 sk2->sk_family == sk->sk_family &&
217 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
218 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
219 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
220 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
221 inet_rcv_saddr_equal(sk, sk2, false)) {
222 return reuseport_add_sock(sk, sk2,
223 inet_rcv_saddr_any(sk));
227 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
231 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
233 * @sk: socket struct in question
234 * @snum: port number to look up
235 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
238 int udp_lib_get_port(struct sock *sk, unsigned short snum,
239 unsigned int hash2_nulladdr)
241 struct udp_table *udptable = udp_get_table_prot(sk);
242 struct udp_hslot *hslot, *hslot2;
243 struct net *net = sock_net(sk);
244 int error = -EADDRINUSE;
247 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
248 unsigned short first, last;
249 int low, high, remaining;
252 inet_sk_get_local_port_range(sk, &low, &high);
253 remaining = (high - low) + 1;
255 rand = get_random_u32();
256 first = reciprocal_scale(rand, remaining) + low;
258 * force rand to be an odd multiple of UDP_HTABLE_SIZE
260 rand = (rand | 1) * (udptable->mask + 1);
261 last = first + udptable->mask + 1;
263 hslot = udp_hashslot(udptable, net, first);
264 bitmap_zero(bitmap, PORTS_PER_CHAIN);
265 spin_lock_bh(&hslot->lock);
266 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
271 * Iterate on all possible values of snum for this hash.
272 * Using steps of an odd multiple of UDP_HTABLE_SIZE
273 * give us randomization and full range coverage.
276 if (low <= snum && snum <= high &&
277 !test_bit(snum >> udptable->log, bitmap) &&
278 !inet_is_local_reserved_port(net, snum))
281 } while (snum != first);
282 spin_unlock_bh(&hslot->lock);
284 } while (++first != last);
287 hslot = udp_hashslot(udptable, net, snum);
288 spin_lock_bh(&hslot->lock);
289 if (hslot->count > 10) {
291 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
293 slot2 &= udptable->mask;
294 hash2_nulladdr &= udptable->mask;
296 hslot2 = udp_hashslot2(udptable, slot2);
297 if (hslot->count < hslot2->count)
298 goto scan_primary_hash;
300 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
301 if (!exist && (hash2_nulladdr != slot2)) {
302 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
303 exist = udp_lib_lport_inuse2(net, snum, hslot2,
312 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
316 inet_sk(sk)->inet_num = snum;
317 udp_sk(sk)->udp_port_hash = snum;
318 udp_sk(sk)->udp_portaddr_hash ^= snum;
319 if (sk_unhashed(sk)) {
320 if (sk->sk_reuseport &&
321 udp_reuseport_add_sock(sk, hslot)) {
322 inet_sk(sk)->inet_num = 0;
323 udp_sk(sk)->udp_port_hash = 0;
324 udp_sk(sk)->udp_portaddr_hash ^= snum;
328 sk_add_node_rcu(sk, &hslot->head);
330 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
332 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
333 spin_lock(&hslot2->lock);
334 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
335 sk->sk_family == AF_INET6)
336 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
339 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
342 spin_unlock(&hslot2->lock);
344 sock_set_flag(sk, SOCK_RCU_FREE);
347 spin_unlock_bh(&hslot->lock);
351 EXPORT_SYMBOL(udp_lib_get_port);
353 int udp_v4_get_port(struct sock *sk, unsigned short snum)
355 unsigned int hash2_nulladdr =
356 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
357 unsigned int hash2_partial =
358 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
360 /* precompute partial secondary hash */
361 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
362 return udp_lib_get_port(sk, snum, hash2_nulladdr);
365 static int compute_score(struct sock *sk, struct net *net,
366 __be32 saddr, __be16 sport,
367 __be32 daddr, unsigned short hnum,
371 struct inet_sock *inet;
374 if (!net_eq(sock_net(sk), net) ||
375 udp_sk(sk)->udp_port_hash != hnum ||
379 if (sk->sk_rcv_saddr != daddr)
382 score = (sk->sk_family == PF_INET) ? 2 : 1;
385 if (inet->inet_daddr) {
386 if (inet->inet_daddr != saddr)
391 if (inet->inet_dport) {
392 if (inet->inet_dport != sport)
397 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
401 if (sk->sk_bound_dev_if)
404 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
409 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
410 const __u16 lport, const __be32 faddr,
413 static u32 udp_ehash_secret __read_mostly;
415 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
417 return __inet_ehashfn(laddr, lport, faddr, fport,
418 udp_ehash_secret + net_hash_mix(net));
421 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
423 __be32 saddr, __be16 sport,
424 __be32 daddr, unsigned short hnum)
426 struct sock *reuse_sk = NULL;
429 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
430 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
431 reuse_sk = reuseport_select_sock(sk, hash, skb,
432 sizeof(struct udphdr));
437 /* called with rcu_read_lock() */
438 static struct sock *udp4_lib_lookup2(struct net *net,
439 __be32 saddr, __be16 sport,
440 __be32 daddr, unsigned int hnum,
442 struct udp_hslot *hslot2,
445 struct sock *sk, *result;
450 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
451 score = compute_score(sk, net, saddr, sport,
452 daddr, hnum, dif, sdif);
453 if (score > badness) {
455 result = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
461 /* Fall back to scoring if group has connections */
462 if (!reuseport_has_conns(sk))
465 /* Reuseport logic returned an error, keep original score. */
469 badness = compute_score(result, net, saddr, sport,
470 daddr, hnum, dif, sdif);
477 static struct sock *udp4_lookup_run_bpf(struct net *net,
478 struct udp_table *udptable,
480 __be32 saddr, __be16 sport,
481 __be32 daddr, u16 hnum, const int dif)
483 struct sock *sk, *reuse_sk;
486 if (udptable != net->ipv4.udp_table)
487 return NULL; /* only UDP is supported */
489 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport,
490 daddr, hnum, dif, &sk);
491 if (no_reuseport || IS_ERR_OR_NULL(sk))
494 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
500 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
501 * harder than this. -DaveM
503 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
504 __be16 sport, __be32 daddr, __be16 dport, int dif,
505 int sdif, struct udp_table *udptable, struct sk_buff *skb)
507 unsigned short hnum = ntohs(dport);
508 unsigned int hash2, slot2;
509 struct udp_hslot *hslot2;
510 struct sock *result, *sk;
512 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
513 slot2 = hash2 & udptable->mask;
514 hslot2 = &udptable->hash2[slot2];
516 /* Lookup connected or non-wildcard socket */
517 result = udp4_lib_lookup2(net, saddr, sport,
518 daddr, hnum, dif, sdif,
520 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
523 /* Lookup redirect from BPF */
524 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
525 sk = udp4_lookup_run_bpf(net, udptable, skb,
526 saddr, sport, daddr, hnum, dif);
533 /* Got non-wildcard socket or error on first lookup */
537 /* Lookup wildcard sockets */
538 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
539 slot2 = hash2 & udptable->mask;
540 hslot2 = &udptable->hash2[slot2];
542 result = udp4_lib_lookup2(net, saddr, sport,
543 htonl(INADDR_ANY), hnum, dif, sdif,
550 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
552 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
553 __be16 sport, __be16 dport,
554 struct udp_table *udptable)
556 const struct iphdr *iph = ip_hdr(skb);
558 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
559 iph->daddr, dport, inet_iif(skb),
560 inet_sdif(skb), udptable, skb);
563 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
564 __be16 sport, __be16 dport)
566 const struct iphdr *iph = ip_hdr(skb);
567 struct net *net = dev_net(skb->dev);
569 return __udp4_lib_lookup(net, iph->saddr, sport,
570 iph->daddr, dport, inet_iif(skb),
571 inet_sdif(skb), net->ipv4.udp_table, NULL);
574 /* Must be called under rcu_read_lock().
575 * Does increment socket refcount.
577 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
578 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
579 __be32 daddr, __be16 dport, int dif)
583 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
584 dif, 0, net->ipv4.udp_table, NULL);
585 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
589 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
592 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
593 __be16 loc_port, __be32 loc_addr,
594 __be16 rmt_port, __be32 rmt_addr,
595 int dif, int sdif, unsigned short hnum)
597 const struct inet_sock *inet = inet_sk(sk);
599 if (!net_eq(sock_net(sk), net) ||
600 udp_sk(sk)->udp_port_hash != hnum ||
601 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
602 (inet->inet_dport != rmt_port && inet->inet_dport) ||
603 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
604 ipv6_only_sock(sk) ||
605 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
607 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
612 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
613 void udp_encap_enable(void)
615 static_branch_inc(&udp_encap_needed_key);
617 EXPORT_SYMBOL(udp_encap_enable);
619 void udp_encap_disable(void)
621 static_branch_dec(&udp_encap_needed_key);
623 EXPORT_SYMBOL(udp_encap_disable);
625 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
626 * through error handlers in encapsulations looking for a match.
628 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
632 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
633 int (*handler)(struct sk_buff *skb, u32 info);
634 const struct ip_tunnel_encap_ops *encap;
636 encap = rcu_dereference(iptun_encaps[i]);
639 handler = encap->err_handler;
640 if (handler && !handler(skb, info))
647 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
648 * reversing source and destination port: this will match tunnels that force the
649 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
650 * lwtunnels might actually break this assumption by being configured with
651 * different destination ports on endpoints, in this case we won't be able to
652 * trace ICMP messages back to them.
654 * If this doesn't match any socket, probe tunnels with arbitrary destination
655 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
656 * we've sent packets to won't necessarily match the local destination port.
658 * Then ask the tunnel implementation to match the error against a valid
661 * Return an error if we can't find a match, the socket if we need further
662 * processing, zero otherwise.
664 static struct sock *__udp4_lib_err_encap(struct net *net,
665 const struct iphdr *iph,
667 struct udp_table *udptable,
669 struct sk_buff *skb, u32 info)
671 int (*lookup)(struct sock *sk, struct sk_buff *skb);
672 int network_offset, transport_offset;
675 network_offset = skb_network_offset(skb);
676 transport_offset = skb_transport_offset(skb);
678 /* Network header needs to point to the outer IPv4 header inside ICMP */
679 skb_reset_network_header(skb);
681 /* Transport header needs to point to the UDP header */
682 skb_set_transport_header(skb, iph->ihl << 2);
687 lookup = READ_ONCE(up->encap_err_lookup);
688 if (lookup && lookup(sk, skb))
694 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
695 iph->saddr, uh->dest, skb->dev->ifindex, 0,
700 lookup = READ_ONCE(up->encap_err_lookup);
701 if (!lookup || lookup(sk, skb))
707 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
709 skb_set_transport_header(skb, transport_offset);
710 skb_set_network_header(skb, network_offset);
716 * This routine is called by the ICMP module when it gets some
717 * sort of error condition. If err < 0 then the socket should
718 * be closed and the error returned to the user. If err > 0
719 * it's just the icmp type << 8 | icmp code.
720 * Header points to the ip header of the error packet. We move
721 * on past this. Then (as it used to claim before adjustment)
722 * header points to the first 8 bytes of the udp header. We need
723 * to find the appropriate port.
726 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
728 struct inet_sock *inet;
729 const struct iphdr *iph = (const struct iphdr *)skb->data;
730 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
731 const int type = icmp_hdr(skb)->type;
732 const int code = icmp_hdr(skb)->code;
737 struct net *net = dev_net(skb->dev);
739 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
740 iph->saddr, uh->source, skb->dev->ifindex,
741 inet_sdif(skb), udptable, NULL);
743 if (!sk || udp_sk(sk)->encap_type) {
744 /* No socket for error: try tunnels before discarding */
745 if (static_branch_unlikely(&udp_encap_needed_key)) {
746 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
751 sk = ERR_PTR(-ENOENT);
754 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
767 case ICMP_TIME_EXCEEDED:
770 case ICMP_SOURCE_QUENCH:
772 case ICMP_PARAMETERPROB:
776 case ICMP_DEST_UNREACH:
777 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
778 ipv4_sk_update_pmtu(skb, sk, info);
779 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
787 if (code <= NR_ICMP_UNREACH) {
788 harderr = icmp_err_convert[code].fatal;
789 err = icmp_err_convert[code].errno;
793 ipv4_sk_redirect(skb, sk);
798 * RFC1122: OK. Passes ICMP errors back to application, as per
802 /* ...not for tunnels though: we don't have a sending socket */
803 if (udp_sk(sk)->encap_err_rcv)
804 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
808 if (!inet->recverr) {
809 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
812 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
820 int udp_err(struct sk_buff *skb, u32 info)
822 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
826 * Throw away all pending data and cancel the corking. Socket is locked.
828 void udp_flush_pending_frames(struct sock *sk)
830 struct udp_sock *up = udp_sk(sk);
835 ip_flush_pending_frames(sk);
838 EXPORT_SYMBOL(udp_flush_pending_frames);
841 * udp4_hwcsum - handle outgoing HW checksumming
842 * @skb: sk_buff containing the filled-in UDP header
843 * (checksum field must be zeroed out)
844 * @src: source IP address
845 * @dst: destination IP address
847 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
849 struct udphdr *uh = udp_hdr(skb);
850 int offset = skb_transport_offset(skb);
851 int len = skb->len - offset;
855 if (!skb_has_frag_list(skb)) {
857 * Only one fragment on the socket.
859 skb->csum_start = skb_transport_header(skb) - skb->head;
860 skb->csum_offset = offsetof(struct udphdr, check);
861 uh->check = ~csum_tcpudp_magic(src, dst, len,
864 struct sk_buff *frags;
867 * HW-checksum won't work as there are two or more
868 * fragments on the socket so that all csums of sk_buffs
871 skb_walk_frags(skb, frags) {
872 csum = csum_add(csum, frags->csum);
876 csum = skb_checksum(skb, offset, hlen, csum);
877 skb->ip_summed = CHECKSUM_NONE;
879 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
881 uh->check = CSUM_MANGLED_0;
884 EXPORT_SYMBOL_GPL(udp4_hwcsum);
886 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
887 * for the simple case like when setting the checksum for a UDP tunnel.
889 void udp_set_csum(bool nocheck, struct sk_buff *skb,
890 __be32 saddr, __be32 daddr, int len)
892 struct udphdr *uh = udp_hdr(skb);
896 } else if (skb_is_gso(skb)) {
897 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
898 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
900 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
902 uh->check = CSUM_MANGLED_0;
904 skb->ip_summed = CHECKSUM_PARTIAL;
905 skb->csum_start = skb_transport_header(skb) - skb->head;
906 skb->csum_offset = offsetof(struct udphdr, check);
907 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
910 EXPORT_SYMBOL(udp_set_csum);
912 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
913 struct inet_cork *cork)
915 struct sock *sk = skb->sk;
916 struct inet_sock *inet = inet_sk(sk);
919 int is_udplite = IS_UDPLITE(sk);
920 int offset = skb_transport_offset(skb);
921 int len = skb->len - offset;
922 int datalen = len - sizeof(*uh);
926 * Create a UDP header
929 uh->source = inet->inet_sport;
930 uh->dest = fl4->fl4_dport;
931 uh->len = htons(len);
934 if (cork->gso_size) {
935 const int hlen = skb_network_header_len(skb) +
936 sizeof(struct udphdr);
938 if (hlen + cork->gso_size > cork->fragsize) {
942 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
946 if (sk->sk_no_check_tx) {
950 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
951 dst_xfrm(skb_dst(skb))) {
956 if (datalen > cork->gso_size) {
957 skb_shinfo(skb)->gso_size = cork->gso_size;
958 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
959 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
965 if (is_udplite) /* UDP-Lite */
966 csum = udplite_csum(skb);
968 else if (sk->sk_no_check_tx) { /* UDP csum off */
970 skb->ip_summed = CHECKSUM_NONE;
973 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
976 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
980 csum = udp_csum(skb);
982 /* add protocol-dependent pseudo-header */
983 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
984 sk->sk_protocol, csum);
986 uh->check = CSUM_MANGLED_0;
989 err = ip_send_skb(sock_net(sk), skb);
991 if (err == -ENOBUFS && !inet->recverr) {
992 UDP_INC_STATS(sock_net(sk),
993 UDP_MIB_SNDBUFERRORS, is_udplite);
997 UDP_INC_STATS(sock_net(sk),
998 UDP_MIB_OUTDATAGRAMS, is_udplite);
1003 * Push out all pending data as one UDP datagram. Socket is locked.
1005 int udp_push_pending_frames(struct sock *sk)
1007 struct udp_sock *up = udp_sk(sk);
1008 struct inet_sock *inet = inet_sk(sk);
1009 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1010 struct sk_buff *skb;
1013 skb = ip_finish_skb(sk, fl4);
1017 err = udp_send_skb(skb, fl4, &inet->cork.base);
1024 EXPORT_SYMBOL(udp_push_pending_frames);
1026 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1028 switch (cmsg->cmsg_type) {
1030 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1032 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1039 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1041 struct cmsghdr *cmsg;
1042 bool need_ip = false;
1045 for_each_cmsghdr(cmsg, msg) {
1046 if (!CMSG_OK(msg, cmsg))
1049 if (cmsg->cmsg_level != SOL_UDP) {
1054 err = __udp_cmsg_send(cmsg, gso_size);
1061 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1063 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1065 struct inet_sock *inet = inet_sk(sk);
1066 struct udp_sock *up = udp_sk(sk);
1067 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1068 struct flowi4 fl4_stack;
1071 struct ipcm_cookie ipc;
1072 struct rtable *rt = NULL;
1075 __be32 daddr, faddr, saddr;
1078 int err, is_udplite = IS_UDPLITE(sk);
1079 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1080 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1081 struct sk_buff *skb;
1082 struct ip_options_data opt_copy;
1091 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1094 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1096 fl4 = &inet->cork.fl.u.ip4;
1099 * There are pending frames.
1100 * The socket lock must be held while it's corked.
1103 if (likely(up->pending)) {
1104 if (unlikely(up->pending != AF_INET)) {
1108 goto do_append_data;
1112 ulen += sizeof(struct udphdr);
1115 * Get and verify the address.
1118 if (msg->msg_namelen < sizeof(*usin))
1120 if (usin->sin_family != AF_INET) {
1121 if (usin->sin_family != AF_UNSPEC)
1122 return -EAFNOSUPPORT;
1125 daddr = usin->sin_addr.s_addr;
1126 dport = usin->sin_port;
1130 if (sk->sk_state != TCP_ESTABLISHED)
1131 return -EDESTADDRREQ;
1132 daddr = inet->inet_daddr;
1133 dport = inet->inet_dport;
1134 /* Open fast path for connected socket.
1135 Route will not be used, if at least one option is set.
1140 ipcm_init_sk(&ipc, inet);
1141 ipc.gso_size = READ_ONCE(up->gso_size);
1143 if (msg->msg_controllen) {
1144 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1146 err = ip_cmsg_send(sk, msg, &ipc,
1147 sk->sk_family == AF_INET6);
1148 if (unlikely(err < 0)) {
1157 struct ip_options_rcu *inet_opt;
1160 inet_opt = rcu_dereference(inet->inet_opt);
1162 memcpy(&opt_copy, inet_opt,
1163 sizeof(*inet_opt) + inet_opt->opt.optlen);
1164 ipc.opt = &opt_copy.opt;
1169 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1170 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1171 (struct sockaddr *)usin, &ipc.addr);
1175 if (usin->sin_port == 0) {
1176 /* BPF program set invalid port. Reject it. */
1180 daddr = usin->sin_addr.s_addr;
1181 dport = usin->sin_port;
1186 ipc.addr = faddr = daddr;
1188 if (ipc.opt && ipc.opt->opt.srr) {
1193 faddr = ipc.opt->opt.faddr;
1196 tos = get_rttos(&ipc, inet);
1197 scope = ip_sendmsg_scope(inet, &ipc, msg);
1198 if (scope == RT_SCOPE_LINK)
1201 if (ipv4_is_multicast(daddr)) {
1202 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1203 ipc.oif = inet->mc_index;
1205 saddr = inet->mc_addr;
1207 } else if (!ipc.oif) {
1208 ipc.oif = inet->uc_index;
1209 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1210 /* oif is set, packet is to local broadcast and
1211 * uc_index is set. oif is most likely set
1212 * by sk_bound_dev_if. If uc_index != oif check if the
1213 * oif is an L3 master and uc_index is an L3 slave.
1214 * If so, we want to allow the send using the uc_index.
1216 if (ipc.oif != inet->uc_index &&
1217 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1219 ipc.oif = inet->uc_index;
1224 rt = (struct rtable *)sk_dst_check(sk, 0);
1227 struct net *net = sock_net(sk);
1228 __u8 flow_flags = inet_sk_flowi_flags(sk);
1232 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1233 sk->sk_protocol, flow_flags, faddr, saddr,
1234 dport, inet->inet_sport, sk->sk_uid);
1236 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1237 rt = ip_route_output_flow(net, fl4, sk);
1241 if (err == -ENETUNREACH)
1242 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1247 if ((rt->rt_flags & RTCF_BROADCAST) &&
1248 !sock_flag(sk, SOCK_BROADCAST))
1251 sk_dst_set(sk, dst_clone(&rt->dst));
1254 if (msg->msg_flags&MSG_CONFIRM)
1260 daddr = ipc.addr = fl4->daddr;
1262 /* Lockless fast path for the non-corking case. */
1264 struct inet_cork cork;
1266 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1267 sizeof(struct udphdr), &ipc, &rt,
1268 &cork, msg->msg_flags);
1270 if (!IS_ERR_OR_NULL(skb))
1271 err = udp_send_skb(skb, fl4, &cork);
1276 if (unlikely(up->pending)) {
1277 /* The socket is already corked while preparing it. */
1278 /* ... which is an evident application bug. --ANK */
1281 net_dbg_ratelimited("socket already corked\n");
1286 * Now cork the socket to pend data.
1288 fl4 = &inet->cork.fl.u.ip4;
1291 fl4->fl4_dport = dport;
1292 fl4->fl4_sport = inet->inet_sport;
1293 up->pending = AF_INET;
1297 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1298 sizeof(struct udphdr), &ipc, &rt,
1299 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1301 udp_flush_pending_frames(sk);
1303 err = udp_push_pending_frames(sk);
1304 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1316 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1317 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1318 * we don't have a good statistic (IpOutDiscards but it can be too many
1319 * things). We could add another new stat but at least for now that
1320 * seems like overkill.
1322 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1323 UDP_INC_STATS(sock_net(sk),
1324 UDP_MIB_SNDBUFERRORS, is_udplite);
1329 if (msg->msg_flags & MSG_PROBE)
1330 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1331 if (!(msg->msg_flags&MSG_PROBE) || len)
1332 goto back_from_confirm;
1336 EXPORT_SYMBOL(udp_sendmsg);
1338 void udp_splice_eof(struct socket *sock)
1340 struct sock *sk = sock->sk;
1341 struct udp_sock *up = udp_sk(sk);
1343 if (!up->pending || READ_ONCE(up->corkflag))
1347 if (up->pending && !READ_ONCE(up->corkflag))
1348 udp_push_pending_frames(sk);
1351 EXPORT_SYMBOL_GPL(udp_splice_eof);
1353 #define UDP_SKB_IS_STATELESS 0x80000000
1355 /* all head states (dst, sk, nf conntrack) except skb extensions are
1356 * cleared by udp_rcv().
1358 * We need to preserve secpath, if present, to eventually process
1359 * IP_CMSG_PASSSEC at recvmsg() time.
1361 * Other extensions can be cleared.
1363 static bool udp_try_make_stateless(struct sk_buff *skb)
1365 if (!skb_has_extensions(skb))
1368 if (!secpath_exists(skb)) {
1376 static void udp_set_dev_scratch(struct sk_buff *skb)
1378 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1380 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1381 scratch->_tsize_state = skb->truesize;
1382 #if BITS_PER_LONG == 64
1383 scratch->len = skb->len;
1384 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1385 scratch->is_linear = !skb_is_nonlinear(skb);
1387 if (udp_try_make_stateless(skb))
1388 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1391 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1393 /* We come here after udp_lib_checksum_complete() returned 0.
1394 * This means that __skb_checksum_complete() might have
1395 * set skb->csum_valid to 1.
1396 * On 64bit platforms, we can set csum_unnecessary
1397 * to true, but only if the skb is not shared.
1399 #if BITS_PER_LONG == 64
1400 if (!skb_shared(skb))
1401 udp_skb_scratch(skb)->csum_unnecessary = true;
1405 static int udp_skb_truesize(struct sk_buff *skb)
1407 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1410 static bool udp_skb_has_head_state(struct sk_buff *skb)
1412 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1415 /* fully reclaim rmem/fwd memory allocated for skb */
1416 static void udp_rmem_release(struct sock *sk, int size, int partial,
1417 bool rx_queue_lock_held)
1419 struct udp_sock *up = udp_sk(sk);
1420 struct sk_buff_head *sk_queue;
1423 if (likely(partial)) {
1424 up->forward_deficit += size;
1425 size = up->forward_deficit;
1426 if (size < READ_ONCE(up->forward_threshold) &&
1427 !skb_queue_empty(&up->reader_queue))
1430 size += up->forward_deficit;
1432 up->forward_deficit = 0;
1434 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1435 * if the called don't held it already
1437 sk_queue = &sk->sk_receive_queue;
1438 if (!rx_queue_lock_held)
1439 spin_lock(&sk_queue->lock);
1442 sk->sk_forward_alloc += size;
1443 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1444 sk->sk_forward_alloc -= amt;
1447 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1449 atomic_sub(size, &sk->sk_rmem_alloc);
1451 /* this can save us from acquiring the rx queue lock on next receive */
1452 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1454 if (!rx_queue_lock_held)
1455 spin_unlock(&sk_queue->lock);
1458 /* Note: called with reader_queue.lock held.
1459 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1460 * This avoids a cache line miss while receive_queue lock is held.
1461 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1463 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1465 prefetch(&skb->data);
1466 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1468 EXPORT_SYMBOL(udp_skb_destructor);
1470 /* as above, but the caller held the rx queue lock, too */
1471 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1473 prefetch(&skb->data);
1474 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1477 /* Idea of busylocks is to let producers grab an extra spinlock
1478 * to relieve pressure on the receive_queue spinlock shared by consumer.
1479 * Under flood, this means that only one producer can be in line
1480 * trying to acquire the receive_queue spinlock.
1481 * These busylock can be allocated on a per cpu manner, instead of a
1482 * per socket one (that would consume a cache line per socket)
1484 static int udp_busylocks_log __read_mostly;
1485 static spinlock_t *udp_busylocks __read_mostly;
1487 static spinlock_t *busylock_acquire(void *ptr)
1491 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1496 static void busylock_release(spinlock_t *busy)
1502 static int udp_rmem_schedule(struct sock *sk, int size)
1506 delta = size - sk->sk_forward_alloc;
1507 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1513 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1515 struct sk_buff_head *list = &sk->sk_receive_queue;
1516 int rmem, err = -ENOMEM;
1517 spinlock_t *busy = NULL;
1520 /* try to avoid the costly atomic add/sub pair when the receive
1521 * queue is full; always allow at least a packet
1523 rmem = atomic_read(&sk->sk_rmem_alloc);
1524 if (rmem > sk->sk_rcvbuf)
1527 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1528 * having linear skbs :
1529 * - Reduce memory overhead and thus increase receive queue capacity
1530 * - Less cache line misses at copyout() time
1531 * - Less work at consume_skb() (less alien page frag freeing)
1533 if (rmem > (sk->sk_rcvbuf >> 1)) {
1536 busy = busylock_acquire(sk);
1538 size = skb->truesize;
1539 udp_set_dev_scratch(skb);
1541 /* we drop only if the receive buf is full and the receive
1542 * queue contains some other skb
1544 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1545 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1548 spin_lock(&list->lock);
1549 err = udp_rmem_schedule(sk, size);
1551 spin_unlock(&list->lock);
1555 sk->sk_forward_alloc -= size;
1557 /* no need to setup a destructor, we will explicitly release the
1558 * forward allocated memory on dequeue
1560 sock_skb_set_dropcount(sk, skb);
1562 __skb_queue_tail(list, skb);
1563 spin_unlock(&list->lock);
1565 if (!sock_flag(sk, SOCK_DEAD))
1566 INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
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(sk->sk_peek_off) >= 0)) {
1614 bool slow = lock_sock_fast(sk);
1616 sk_peek_offset_bwd(sk, len);
1617 unlock_sock_fast(sk, slow);
1620 if (!skb_unref(skb))
1623 /* In the more common cases we cleared the head states previously,
1624 * see __udp_queue_rcv_skb().
1626 if (unlikely(udp_skb_has_head_state(skb)))
1627 skb_release_head_state(skb);
1628 __consume_stateless_skb(skb);
1630 EXPORT_SYMBOL_GPL(skb_consume_udp);
1632 static struct sk_buff *__first_packet_length(struct sock *sk,
1633 struct sk_buff_head *rcvq,
1636 struct sk_buff *skb;
1638 while ((skb = skb_peek(rcvq)) != NULL) {
1639 if (udp_lib_checksum_complete(skb)) {
1640 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1642 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1644 atomic_inc(&sk->sk_drops);
1645 __skb_unlink(skb, rcvq);
1646 *total += skb->truesize;
1649 udp_skb_csum_unnecessary_set(skb);
1657 * first_packet_length - return length of first packet in receive queue
1660 * Drops all bad checksum frames, until a valid one is found.
1661 * Returns the length of found skb, or -1 if none is found.
1663 static int first_packet_length(struct sock *sk)
1665 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1666 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1667 struct sk_buff *skb;
1671 spin_lock_bh(&rcvq->lock);
1672 skb = __first_packet_length(sk, rcvq, &total);
1673 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1674 spin_lock(&sk_queue->lock);
1675 skb_queue_splice_tail_init(sk_queue, rcvq);
1676 spin_unlock(&sk_queue->lock);
1678 skb = __first_packet_length(sk, rcvq, &total);
1680 res = skb ? skb->len : -1;
1682 udp_rmem_release(sk, total, 1, false);
1683 spin_unlock_bh(&rcvq->lock);
1688 * IOCTL requests applicable to the UDP protocol
1691 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1696 *karg = sk_wmem_alloc_get(sk);
1702 *karg = max_t(int, 0, first_packet_length(sk));
1707 return -ENOIOCTLCMD;
1712 EXPORT_SYMBOL(udp_ioctl);
1714 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1717 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1718 struct sk_buff_head *queue;
1719 struct sk_buff *last;
1723 queue = &udp_sk(sk)->reader_queue;
1724 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1726 struct sk_buff *skb;
1728 error = sock_error(sk);
1734 spin_lock_bh(&queue->lock);
1735 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1738 if (!(flags & MSG_PEEK))
1739 udp_skb_destructor(sk, skb);
1740 spin_unlock_bh(&queue->lock);
1744 if (skb_queue_empty_lockless(sk_queue)) {
1745 spin_unlock_bh(&queue->lock);
1749 /* refill the reader queue and walk it again
1750 * keep both queues locked to avoid re-acquiring
1751 * the sk_receive_queue lock if fwd memory scheduling
1754 spin_lock(&sk_queue->lock);
1755 skb_queue_splice_tail_init(sk_queue, queue);
1757 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1759 if (skb && !(flags & MSG_PEEK))
1760 udp_skb_dtor_locked(sk, skb);
1761 spin_unlock(&sk_queue->lock);
1762 spin_unlock_bh(&queue->lock);
1767 if (!sk_can_busy_loop(sk))
1770 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1771 } while (!skb_queue_empty_lockless(sk_queue));
1773 /* sk_queue is empty, reader_queue may contain peeked packets */
1775 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1777 (struct sk_buff *)sk_queue));
1782 EXPORT_SYMBOL(__skb_recv_udp);
1784 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1786 struct sk_buff *skb;
1790 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1794 if (udp_lib_checksum_complete(skb)) {
1795 int is_udplite = IS_UDPLITE(sk);
1796 struct net *net = sock_net(sk);
1798 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1799 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1800 atomic_inc(&sk->sk_drops);
1805 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1806 return recv_actor(sk, skb);
1808 EXPORT_SYMBOL(udp_read_skb);
1811 * This should be easy, if there is something there we
1812 * return it, otherwise we block.
1815 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1818 struct inet_sock *inet = inet_sk(sk);
1819 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1820 struct sk_buff *skb;
1821 unsigned int ulen, copied;
1822 int off, err, peeking = flags & MSG_PEEK;
1823 int is_udplite = IS_UDPLITE(sk);
1824 bool checksum_valid = false;
1826 if (flags & MSG_ERRQUEUE)
1827 return ip_recv_error(sk, msg, len, addr_len);
1830 off = sk_peek_offset(sk, flags);
1831 skb = __skb_recv_udp(sk, flags, &off, &err);
1835 ulen = udp_skb_len(skb);
1837 if (copied > ulen - off)
1838 copied = ulen - off;
1839 else if (copied < ulen)
1840 msg->msg_flags |= MSG_TRUNC;
1843 * If checksum is needed at all, try to do it while copying the
1844 * data. If the data is truncated, or if we only want a partial
1845 * coverage checksum (UDP-Lite), do it before the copy.
1848 if (copied < ulen || peeking ||
1849 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1850 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1851 !__udp_lib_checksum_complete(skb);
1852 if (!checksum_valid)
1856 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1857 if (udp_skb_is_linear(skb))
1858 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1860 err = skb_copy_datagram_msg(skb, off, msg, copied);
1862 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1868 if (unlikely(err)) {
1870 atomic_inc(&sk->sk_drops);
1871 UDP_INC_STATS(sock_net(sk),
1872 UDP_MIB_INERRORS, is_udplite);
1879 UDP_INC_STATS(sock_net(sk),
1880 UDP_MIB_INDATAGRAMS, is_udplite);
1882 sock_recv_cmsgs(msg, sk, skb);
1884 /* Copy the address. */
1886 sin->sin_family = AF_INET;
1887 sin->sin_port = udp_hdr(skb)->source;
1888 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1889 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1890 *addr_len = sizeof(*sin);
1892 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1893 (struct sockaddr *)sin);
1896 if (udp_sk(sk)->gro_enabled)
1897 udp_cmsg_recv(msg, sk, skb);
1899 if (inet->cmsg_flags)
1900 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1903 if (flags & MSG_TRUNC)
1906 skb_consume_udp(sk, skb, peeking ? -err : err);
1910 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1911 udp_skb_destructor)) {
1912 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1913 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1917 /* starting over for a new packet, but check if we need to yield */
1919 msg->msg_flags &= ~MSG_TRUNC;
1923 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1925 /* This check is replicated from __ip4_datagram_connect() and
1926 * intended to prevent BPF program called below from accessing bytes
1927 * that are out of the bound specified by user in addr_len.
1929 if (addr_len < sizeof(struct sockaddr_in))
1932 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1934 EXPORT_SYMBOL(udp_pre_connect);
1936 int __udp_disconnect(struct sock *sk, int flags)
1938 struct inet_sock *inet = inet_sk(sk);
1940 * 1003.1g - break association.
1943 sk->sk_state = TCP_CLOSE;
1944 inet->inet_daddr = 0;
1945 inet->inet_dport = 0;
1946 sock_rps_reset_rxhash(sk);
1947 sk->sk_bound_dev_if = 0;
1948 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1949 inet_reset_saddr(sk);
1950 if (sk->sk_prot->rehash &&
1951 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1952 sk->sk_prot->rehash(sk);
1955 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1956 sk->sk_prot->unhash(sk);
1957 inet->inet_sport = 0;
1962 EXPORT_SYMBOL(__udp_disconnect);
1964 int udp_disconnect(struct sock *sk, int flags)
1967 __udp_disconnect(sk, flags);
1971 EXPORT_SYMBOL(udp_disconnect);
1973 void udp_lib_unhash(struct sock *sk)
1975 if (sk_hashed(sk)) {
1976 struct udp_table *udptable = udp_get_table_prot(sk);
1977 struct udp_hslot *hslot, *hslot2;
1979 hslot = udp_hashslot(udptable, sock_net(sk),
1980 udp_sk(sk)->udp_port_hash);
1981 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1983 spin_lock_bh(&hslot->lock);
1984 if (rcu_access_pointer(sk->sk_reuseport_cb))
1985 reuseport_detach_sock(sk);
1986 if (sk_del_node_init_rcu(sk)) {
1988 inet_sk(sk)->inet_num = 0;
1989 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1991 spin_lock(&hslot2->lock);
1992 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1994 spin_unlock(&hslot2->lock);
1996 spin_unlock_bh(&hslot->lock);
1999 EXPORT_SYMBOL(udp_lib_unhash);
2002 * inet_rcv_saddr was changed, we must rehash secondary hash
2004 void udp_lib_rehash(struct sock *sk, u16 newhash)
2006 if (sk_hashed(sk)) {
2007 struct udp_table *udptable = udp_get_table_prot(sk);
2008 struct udp_hslot *hslot, *hslot2, *nhslot2;
2010 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2011 nhslot2 = udp_hashslot2(udptable, newhash);
2012 udp_sk(sk)->udp_portaddr_hash = newhash;
2014 if (hslot2 != nhslot2 ||
2015 rcu_access_pointer(sk->sk_reuseport_cb)) {
2016 hslot = udp_hashslot(udptable, sock_net(sk),
2017 udp_sk(sk)->udp_port_hash);
2018 /* we must lock primary chain too */
2019 spin_lock_bh(&hslot->lock);
2020 if (rcu_access_pointer(sk->sk_reuseport_cb))
2021 reuseport_detach_sock(sk);
2023 if (hslot2 != nhslot2) {
2024 spin_lock(&hslot2->lock);
2025 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2027 spin_unlock(&hslot2->lock);
2029 spin_lock(&nhslot2->lock);
2030 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2033 spin_unlock(&nhslot2->lock);
2036 spin_unlock_bh(&hslot->lock);
2040 EXPORT_SYMBOL(udp_lib_rehash);
2042 void udp_v4_rehash(struct sock *sk)
2044 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2045 inet_sk(sk)->inet_rcv_saddr,
2046 inet_sk(sk)->inet_num);
2047 udp_lib_rehash(sk, new_hash);
2050 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2054 if (inet_sk(sk)->inet_daddr) {
2055 sock_rps_save_rxhash(sk, skb);
2056 sk_mark_napi_id(sk, skb);
2057 sk_incoming_cpu_update(sk);
2059 sk_mark_napi_id_once(sk, skb);
2062 rc = __udp_enqueue_schedule_skb(sk, skb);
2064 int is_udplite = IS_UDPLITE(sk);
2067 /* Note that an ENOMEM error is charged twice */
2068 if (rc == -ENOMEM) {
2069 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2071 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2073 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2075 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2077 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2078 kfree_skb_reason(skb, drop_reason);
2079 trace_udp_fail_queue_rcv_skb(rc, sk);
2089 * >0: "udp encap" protocol resubmission
2091 * Note that in the success and error cases, the skb is assumed to
2092 * have either been requeued or freed.
2094 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2096 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2097 struct udp_sock *up = udp_sk(sk);
2098 int is_udplite = IS_UDPLITE(sk);
2101 * Charge it to the socket, dropping if the queue is full.
2103 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2104 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2109 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2110 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2113 * This is an encapsulation socket so pass the skb to
2114 * the socket's udp_encap_rcv() hook. Otherwise, just
2115 * fall through and pass this up the UDP socket.
2116 * up->encap_rcv() returns the following value:
2117 * =0 if skb was successfully passed to the encap
2118 * handler or was discarded by it.
2119 * >0 if skb should be passed on to UDP.
2120 * <0 if skb should be resubmitted as proto -N
2123 /* if we're overly short, let UDP handle it */
2124 encap_rcv = READ_ONCE(up->encap_rcv);
2128 /* Verify checksum before giving to encap */
2129 if (udp_lib_checksum_complete(skb))
2132 ret = encap_rcv(sk, skb);
2134 __UDP_INC_STATS(sock_net(sk),
2135 UDP_MIB_INDATAGRAMS,
2141 /* FALLTHROUGH -- it's a UDP Packet */
2145 * UDP-Lite specific tests, ignored on UDP sockets
2147 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2150 * MIB statistics other than incrementing the error count are
2151 * disabled for the following two types of errors: these depend
2152 * on the application settings, not on the functioning of the
2153 * protocol stack as such.
2155 * RFC 3828 here recommends (sec 3.3): "There should also be a
2156 * way ... to ... at least let the receiving application block
2157 * delivery of packets with coverage values less than a value
2158 * provided by the application."
2160 if (up->pcrlen == 0) { /* full coverage was set */
2161 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2162 UDP_SKB_CB(skb)->cscov, skb->len);
2165 /* The next case involves violating the min. coverage requested
2166 * by the receiver. This is subtle: if receiver wants x and x is
2167 * greater than the buffersize/MTU then receiver will complain
2168 * that it wants x while sender emits packets of smaller size y.
2169 * Therefore the above ...()->partial_cov statement is essential.
2171 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2172 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2173 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2178 prefetch(&sk->sk_rmem_alloc);
2179 if (rcu_access_pointer(sk->sk_filter) &&
2180 udp_lib_checksum_complete(skb))
2183 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2184 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2188 udp_csum_pull_header(skb);
2190 ipv4_pktinfo_prepare(sk, skb);
2191 return __udp_queue_rcv_skb(sk, skb);
2194 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2195 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2197 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2198 atomic_inc(&sk->sk_drops);
2199 kfree_skb_reason(skb, drop_reason);
2203 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2205 struct sk_buff *next, *segs;
2208 if (likely(!udp_unexpected_gso(sk, skb)))
2209 return udp_queue_rcv_one_skb(sk, skb);
2211 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2212 __skb_push(skb, -skb_mac_offset(skb));
2213 segs = udp_rcv_segment(sk, skb, true);
2214 skb_list_walk_safe(segs, skb, next) {
2215 __skb_pull(skb, skb_transport_offset(skb));
2217 udp_post_segment_fix_csum(skb);
2218 ret = udp_queue_rcv_one_skb(sk, skb);
2220 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2225 /* For TCP sockets, sk_rx_dst is protected by socket lock
2226 * For UDP, we use xchg() to guard against concurrent changes.
2228 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2230 struct dst_entry *old;
2232 if (dst_hold_safe(dst)) {
2233 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2239 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2242 * Multicasts and broadcasts go to each listener.
2244 * Note: called only from the BH handler context.
2246 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2248 __be32 saddr, __be32 daddr,
2249 struct udp_table *udptable,
2252 struct sock *sk, *first = NULL;
2253 unsigned short hnum = ntohs(uh->dest);
2254 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2255 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2256 unsigned int offset = offsetof(typeof(*sk), sk_node);
2257 int dif = skb->dev->ifindex;
2258 int sdif = inet_sdif(skb);
2259 struct hlist_node *node;
2260 struct sk_buff *nskb;
2263 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2265 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2267 hslot = &udptable->hash2[hash2];
2268 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2271 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2272 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2273 uh->source, saddr, dif, sdif, hnum))
2280 nskb = skb_clone(skb, GFP_ATOMIC);
2282 if (unlikely(!nskb)) {
2283 atomic_inc(&sk->sk_drops);
2284 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2286 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2290 if (udp_queue_rcv_skb(sk, nskb) > 0)
2294 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2295 if (use_hash2 && hash2 != hash2_any) {
2301 if (udp_queue_rcv_skb(first, skb) > 0)
2305 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2306 proto == IPPROTO_UDPLITE);
2311 /* Initialize UDP checksum. If exited with zero value (success),
2312 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2313 * Otherwise, csum completion requires checksumming packet body,
2314 * including udp header and folding it to skb->csum.
2316 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2321 UDP_SKB_CB(skb)->partial_cov = 0;
2322 UDP_SKB_CB(skb)->cscov = skb->len;
2324 if (proto == IPPROTO_UDPLITE) {
2325 err = udplite_checksum_init(skb, uh);
2329 if (UDP_SKB_CB(skb)->partial_cov) {
2330 skb->csum = inet_compute_pseudo(skb, proto);
2335 /* Note, we are only interested in != 0 or == 0, thus the
2338 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2339 inet_compute_pseudo);
2343 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2344 /* If SW calculated the value, we know it's bad */
2345 if (skb->csum_complete_sw)
2348 /* HW says the value is bad. Let's validate that.
2349 * skb->csum is no longer the full packet checksum,
2350 * so don't treat it as such.
2352 skb_checksum_complete_unset(skb);
2358 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2359 * return code conversion for ip layer consumption
2361 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2366 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2367 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2369 ret = udp_queue_rcv_skb(sk, skb);
2371 /* a return value > 0 means to resubmit the input, but
2372 * it wants the return to be -protocol, or 0
2380 * All we need to do is get the socket, and then do a checksum.
2383 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2388 unsigned short ulen;
2389 struct rtable *rt = skb_rtable(skb);
2390 __be32 saddr, daddr;
2391 struct net *net = dev_net(skb->dev);
2395 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2398 * Validate the packet.
2400 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2401 goto drop; /* No space for header. */
2404 ulen = ntohs(uh->len);
2405 saddr = ip_hdr(skb)->saddr;
2406 daddr = ip_hdr(skb)->daddr;
2408 if (ulen > skb->len)
2411 if (proto == IPPROTO_UDP) {
2412 /* UDP validates ulen. */
2413 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2418 if (udp4_csum_init(skb, uh, proto))
2421 sk = skb_steal_sock(skb, &refcounted);
2423 struct dst_entry *dst = skb_dst(skb);
2426 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2427 udp_sk_rx_dst_set(sk, dst);
2429 ret = udp_unicast_rcv_skb(sk, skb, uh);
2435 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2436 return __udp4_lib_mcast_deliver(net, skb, uh,
2437 saddr, daddr, udptable, proto);
2439 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2441 return udp_unicast_rcv_skb(sk, skb, uh);
2443 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2447 /* No socket. Drop packet silently, if checksum is wrong */
2448 if (udp_lib_checksum_complete(skb))
2451 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2452 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2453 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2456 * Hmm. We got an UDP packet to a port to which we
2457 * don't wanna listen. Ignore it.
2459 kfree_skb_reason(skb, drop_reason);
2463 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2464 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2465 proto == IPPROTO_UDPLITE ? "Lite" : "",
2466 &saddr, ntohs(uh->source),
2468 &daddr, ntohs(uh->dest));
2473 * RFC1122: OK. Discards the bad packet silently (as far as
2474 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2476 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2477 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2478 proto == IPPROTO_UDPLITE ? "Lite" : "",
2479 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2481 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2483 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2484 kfree_skb_reason(skb, drop_reason);
2488 /* We can only early demux multicast if there is a single matching socket.
2489 * If more than one socket found returns NULL
2491 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2492 __be16 loc_port, __be32 loc_addr,
2493 __be16 rmt_port, __be32 rmt_addr,
2496 struct udp_table *udptable = net->ipv4.udp_table;
2497 unsigned short hnum = ntohs(loc_port);
2498 struct sock *sk, *result;
2499 struct udp_hslot *hslot;
2502 slot = udp_hashfn(net, hnum, udptable->mask);
2503 hslot = &udptable->hash[slot];
2505 /* Do not bother scanning a too big list */
2506 if (hslot->count > 10)
2510 sk_for_each_rcu(sk, &hslot->head) {
2511 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2512 rmt_port, rmt_addr, dif, sdif, hnum)) {
2522 /* For unicast we should only early demux connected sockets or we can
2523 * break forwarding setups. The chains here can be long so only check
2524 * if the first socket is an exact match and if not move on.
2526 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2527 __be16 loc_port, __be32 loc_addr,
2528 __be16 rmt_port, __be32 rmt_addr,
2531 struct udp_table *udptable = net->ipv4.udp_table;
2532 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2533 unsigned short hnum = ntohs(loc_port);
2534 unsigned int hash2, slot2;
2535 struct udp_hslot *hslot2;
2539 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2540 slot2 = hash2 & udptable->mask;
2541 hslot2 = &udptable->hash2[slot2];
2542 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2544 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2545 if (inet_match(net, sk, acookie, ports, dif, sdif))
2547 /* Only check first socket in chain */
2553 int udp_v4_early_demux(struct sk_buff *skb)
2555 struct net *net = dev_net(skb->dev);
2556 struct in_device *in_dev = NULL;
2557 const struct iphdr *iph;
2558 const struct udphdr *uh;
2559 struct sock *sk = NULL;
2560 struct dst_entry *dst;
2561 int dif = skb->dev->ifindex;
2562 int sdif = inet_sdif(skb);
2565 /* validate the packet */
2566 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2572 if (skb->pkt_type == PACKET_MULTICAST) {
2573 in_dev = __in_dev_get_rcu(skb->dev);
2578 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2583 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2584 uh->source, iph->saddr,
2586 } else if (skb->pkt_type == PACKET_HOST) {
2587 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2588 uh->source, iph->saddr, dif, sdif);
2591 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2595 skb->destructor = sock_efree;
2596 dst = rcu_dereference(sk->sk_rx_dst);
2599 dst = dst_check(dst, 0);
2603 /* set noref for now.
2604 * any place which wants to hold dst has to call
2607 skb_dst_set_noref(skb, dst);
2609 /* for unconnected multicast sockets we need to validate
2610 * the source on each packet
2612 if (!inet_sk(sk)->inet_daddr && in_dev)
2613 return ip_mc_validate_source(skb, iph->daddr,
2615 iph->tos & IPTOS_RT_MASK,
2616 skb->dev, in_dev, &itag);
2621 int udp_rcv(struct sk_buff *skb)
2623 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2626 void udp_destroy_sock(struct sock *sk)
2628 struct udp_sock *up = udp_sk(sk);
2629 bool slow = lock_sock_fast(sk);
2631 /* protects from races with udp_abort() */
2632 sock_set_flag(sk, SOCK_DEAD);
2633 udp_flush_pending_frames(sk);
2634 unlock_sock_fast(sk, slow);
2635 if (static_branch_unlikely(&udp_encap_needed_key)) {
2636 if (up->encap_type) {
2637 void (*encap_destroy)(struct sock *sk);
2638 encap_destroy = READ_ONCE(up->encap_destroy);
2642 if (up->encap_enabled)
2643 static_branch_dec(&udp_encap_needed_key);
2648 * Socket option code for UDP
2650 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2651 sockptr_t optval, unsigned int optlen,
2652 int (*push_pending_frames)(struct sock *))
2654 struct udp_sock *up = udp_sk(sk);
2657 int is_udplite = IS_UDPLITE(sk);
2659 if (level == SOL_SOCKET) {
2660 err = sk_setsockopt(sk, level, optname, optval, optlen);
2662 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2663 sockopt_lock_sock(sk);
2664 /* paired with READ_ONCE in udp_rmem_release() */
2665 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2666 sockopt_release_sock(sk);
2671 if (optlen < sizeof(int))
2674 if (copy_from_sockptr(&val, optval, sizeof(val)))
2677 valbool = val ? 1 : 0;
2682 WRITE_ONCE(up->corkflag, 1);
2684 WRITE_ONCE(up->corkflag, 0);
2686 push_pending_frames(sk);
2695 case UDP_ENCAP_ESPINUDP:
2696 case UDP_ENCAP_ESPINUDP_NON_IKE:
2697 #if IS_ENABLED(CONFIG_IPV6)
2698 if (sk->sk_family == AF_INET6)
2699 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2702 up->encap_rcv = xfrm4_udp_encap_rcv;
2705 case UDP_ENCAP_L2TPINUDP:
2706 up->encap_type = val;
2708 udp_tunnel_encap_enable(sk->sk_socket);
2717 case UDP_NO_CHECK6_TX:
2718 up->no_check6_tx = valbool;
2721 case UDP_NO_CHECK6_RX:
2722 up->no_check6_rx = valbool;
2726 if (val < 0 || val > USHRT_MAX)
2728 WRITE_ONCE(up->gso_size, val);
2734 /* when enabling GRO, accept the related GSO packet type */
2736 udp_tunnel_encap_enable(sk->sk_socket);
2737 up->gro_enabled = valbool;
2738 up->accept_udp_l4 = valbool;
2743 * UDP-Lite's partial checksum coverage (RFC 3828).
2745 /* The sender sets actual checksum coverage length via this option.
2746 * The case coverage > packet length is handled by send module. */
2747 case UDPLITE_SEND_CSCOV:
2748 if (!is_udplite) /* Disable the option on UDP sockets */
2749 return -ENOPROTOOPT;
2750 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2752 else if (val > USHRT_MAX)
2755 up->pcflag |= UDPLITE_SEND_CC;
2758 /* The receiver specifies a minimum checksum coverage value. To make
2759 * sense, this should be set to at least 8 (as done below). If zero is
2760 * used, this again means full checksum coverage. */
2761 case UDPLITE_RECV_CSCOV:
2762 if (!is_udplite) /* Disable the option on UDP sockets */
2763 return -ENOPROTOOPT;
2764 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2766 else if (val > USHRT_MAX)
2769 up->pcflag |= UDPLITE_RECV_CC;
2779 EXPORT_SYMBOL(udp_lib_setsockopt);
2781 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2782 unsigned int optlen)
2784 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2785 return udp_lib_setsockopt(sk, level, optname,
2787 udp_push_pending_frames);
2788 return ip_setsockopt(sk, level, optname, optval, optlen);
2791 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2792 char __user *optval, int __user *optlen)
2794 struct udp_sock *up = udp_sk(sk);
2797 if (get_user(len, optlen))
2800 len = min_t(unsigned int, len, sizeof(int));
2807 val = READ_ONCE(up->corkflag);
2811 val = up->encap_type;
2814 case UDP_NO_CHECK6_TX:
2815 val = up->no_check6_tx;
2818 case UDP_NO_CHECK6_RX:
2819 val = up->no_check6_rx;
2823 val = READ_ONCE(up->gso_size);
2827 val = up->gro_enabled;
2830 /* The following two cannot be changed on UDP sockets, the return is
2831 * always 0 (which corresponds to the full checksum coverage of UDP). */
2832 case UDPLITE_SEND_CSCOV:
2836 case UDPLITE_RECV_CSCOV:
2841 return -ENOPROTOOPT;
2844 if (put_user(len, optlen))
2846 if (copy_to_user(optval, &val, len))
2850 EXPORT_SYMBOL(udp_lib_getsockopt);
2852 int udp_getsockopt(struct sock *sk, int level, int optname,
2853 char __user *optval, int __user *optlen)
2855 if (level == SOL_UDP || level == SOL_UDPLITE)
2856 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2857 return ip_getsockopt(sk, level, optname, optval, optlen);
2861 * udp_poll - wait for a UDP event.
2862 * @file: - file struct
2864 * @wait: - poll table
2866 * This is same as datagram poll, except for the special case of
2867 * blocking sockets. If application is using a blocking fd
2868 * and a packet with checksum error is in the queue;
2869 * then it could get return from select indicating data available
2870 * but then block when reading it. Add special case code
2871 * to work around these arguably broken applications.
2873 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2875 __poll_t mask = datagram_poll(file, sock, wait);
2876 struct sock *sk = sock->sk;
2878 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2879 mask |= EPOLLIN | EPOLLRDNORM;
2881 /* Check for false positives due to checksum errors */
2882 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2883 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2884 mask &= ~(EPOLLIN | EPOLLRDNORM);
2886 /* psock ingress_msg queue should not contain any bad checksum frames */
2887 if (sk_is_readable(sk))
2888 mask |= EPOLLIN | EPOLLRDNORM;
2892 EXPORT_SYMBOL(udp_poll);
2894 int udp_abort(struct sock *sk, int err)
2896 if (!has_current_bpf_ctx())
2899 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2902 if (sock_flag(sk, SOCK_DEAD))
2906 sk_error_report(sk);
2907 __udp_disconnect(sk, 0);
2910 if (!has_current_bpf_ctx())
2915 EXPORT_SYMBOL_GPL(udp_abort);
2917 struct proto udp_prot = {
2919 .owner = THIS_MODULE,
2920 .close = udp_lib_close,
2921 .pre_connect = udp_pre_connect,
2922 .connect = ip4_datagram_connect,
2923 .disconnect = udp_disconnect,
2925 .init = udp_init_sock,
2926 .destroy = udp_destroy_sock,
2927 .setsockopt = udp_setsockopt,
2928 .getsockopt = udp_getsockopt,
2929 .sendmsg = udp_sendmsg,
2930 .recvmsg = udp_recvmsg,
2931 .splice_eof = udp_splice_eof,
2932 .release_cb = ip4_datagram_release_cb,
2933 .hash = udp_lib_hash,
2934 .unhash = udp_lib_unhash,
2935 .rehash = udp_v4_rehash,
2936 .get_port = udp_v4_get_port,
2937 .put_port = udp_lib_unhash,
2938 #ifdef CONFIG_BPF_SYSCALL
2939 .psock_update_sk_prot = udp_bpf_update_proto,
2941 .memory_allocated = &udp_memory_allocated,
2942 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2944 .sysctl_mem = sysctl_udp_mem,
2945 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2946 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2947 .obj_size = sizeof(struct udp_sock),
2948 .h.udp_table = NULL,
2949 .diag_destroy = udp_abort,
2951 EXPORT_SYMBOL(udp_prot);
2953 /* ------------------------------------------------------------------------ */
2954 #ifdef CONFIG_PROC_FS
2956 static unsigned short seq_file_family(const struct seq_file *seq);
2957 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2959 unsigned short family = seq_file_family(seq);
2961 /* AF_UNSPEC is used as a match all */
2962 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2963 net_eq(sock_net(sk), seq_file_net(seq)));
2966 #ifdef CONFIG_BPF_SYSCALL
2967 static const struct seq_operations bpf_iter_udp_seq_ops;
2969 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2972 const struct udp_seq_afinfo *afinfo;
2974 #ifdef CONFIG_BPF_SYSCALL
2975 if (seq->op == &bpf_iter_udp_seq_ops)
2976 return net->ipv4.udp_table;
2979 afinfo = pde_data(file_inode(seq->file));
2980 return afinfo->udp_table ? : net->ipv4.udp_table;
2983 static struct sock *udp_get_first(struct seq_file *seq, int start)
2985 struct udp_iter_state *state = seq->private;
2986 struct net *net = seq_file_net(seq);
2987 struct udp_table *udptable;
2990 udptable = udp_get_table_seq(seq, net);
2992 for (state->bucket = start; state->bucket <= udptable->mask;
2994 struct udp_hslot *hslot = &udptable->hash[state->bucket];
2996 if (hlist_empty(&hslot->head))
2999 spin_lock_bh(&hslot->lock);
3000 sk_for_each(sk, &hslot->head) {
3001 if (seq_sk_match(seq, sk))
3004 spin_unlock_bh(&hslot->lock);
3011 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3013 struct udp_iter_state *state = seq->private;
3014 struct net *net = seq_file_net(seq);
3015 struct udp_table *udptable;
3019 } while (sk && !seq_sk_match(seq, sk));
3022 udptable = udp_get_table_seq(seq, net);
3024 if (state->bucket <= udptable->mask)
3025 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3027 return udp_get_first(seq, state->bucket + 1);
3032 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3034 struct sock *sk = udp_get_first(seq, 0);
3037 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3039 return pos ? NULL : sk;
3042 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3044 struct udp_iter_state *state = seq->private;
3045 state->bucket = MAX_UDP_PORTS;
3047 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3049 EXPORT_SYMBOL(udp_seq_start);
3051 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3055 if (v == SEQ_START_TOKEN)
3056 sk = udp_get_idx(seq, 0);
3058 sk = udp_get_next(seq, v);
3063 EXPORT_SYMBOL(udp_seq_next);
3065 void udp_seq_stop(struct seq_file *seq, void *v)
3067 struct udp_iter_state *state = seq->private;
3068 struct udp_table *udptable;
3070 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3072 if (state->bucket <= udptable->mask)
3073 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3075 EXPORT_SYMBOL(udp_seq_stop);
3077 /* ------------------------------------------------------------------------ */
3078 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3081 struct inet_sock *inet = inet_sk(sp);
3082 __be32 dest = inet->inet_daddr;
3083 __be32 src = inet->inet_rcv_saddr;
3084 __u16 destp = ntohs(inet->inet_dport);
3085 __u16 srcp = ntohs(inet->inet_sport);
3087 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3088 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3089 bucket, src, srcp, dest, destp, sp->sk_state,
3090 sk_wmem_alloc_get(sp),
3093 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3095 refcount_read(&sp->sk_refcnt), sp,
3096 atomic_read(&sp->sk_drops));
3099 int udp4_seq_show(struct seq_file *seq, void *v)
3101 seq_setwidth(seq, 127);
3102 if (v == SEQ_START_TOKEN)
3103 seq_puts(seq, " sl local_address rem_address st tx_queue "
3104 "rx_queue tr tm->when retrnsmt uid timeout "
3105 "inode ref pointer drops");
3107 struct udp_iter_state *state = seq->private;
3109 udp4_format_sock(v, seq, state->bucket);
3115 #ifdef CONFIG_BPF_SYSCALL
3116 struct bpf_iter__udp {
3117 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3118 __bpf_md_ptr(struct udp_sock *, udp_sk);
3119 uid_t uid __aligned(8);
3120 int bucket __aligned(8);
3123 struct bpf_udp_iter_state {
3124 struct udp_iter_state state;
3125 unsigned int cur_sk;
3126 unsigned int end_sk;
3127 unsigned int max_sk;
3129 struct sock **batch;
3130 bool st_bucket_done;
3133 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3134 unsigned int new_batch_sz);
3135 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3137 struct bpf_udp_iter_state *iter = seq->private;
3138 struct udp_iter_state *state = &iter->state;
3139 struct net *net = seq_file_net(seq);
3140 struct udp_table *udptable;
3141 unsigned int batch_sks = 0;
3142 bool resized = false;
3145 /* The current batch is done, so advance the bucket. */
3146 if (iter->st_bucket_done) {
3151 udptable = udp_get_table_seq(seq, net);
3154 /* New batch for the next bucket.
3155 * Iterate over the hash table to find a bucket with sockets matching
3156 * the iterator attributes, and return the first matching socket from
3157 * the bucket. The remaining matched sockets from the bucket are batched
3158 * before releasing the bucket lock. This allows BPF programs that are
3159 * called in seq_show to acquire the bucket lock if needed.
3163 iter->st_bucket_done = false;
3166 for (; state->bucket <= udptable->mask; state->bucket++) {
3167 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3169 if (hlist_empty(&hslot2->head)) {
3174 spin_lock_bh(&hslot2->lock);
3175 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3176 if (seq_sk_match(seq, sk)) {
3177 /* Resume from the last iterated socket at the
3178 * offset in the bucket before iterator was stopped.
3184 if (iter->end_sk < iter->max_sk) {
3186 iter->batch[iter->end_sk++] = sk;
3191 spin_unlock_bh(&hslot2->lock);
3196 /* Reset the current bucket's offset before moving to the next bucket. */
3200 /* All done: no batch made. */
3204 if (iter->end_sk == batch_sks) {
3205 /* Batching is done for the current bucket; return the first
3206 * socket to be iterated from the batch.
3208 iter->st_bucket_done = true;
3211 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3213 /* After allocating a larger batch, retry one more time to grab
3220 return iter->batch[0];
3223 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3225 struct bpf_udp_iter_state *iter = seq->private;
3228 /* Whenever seq_next() is called, the iter->cur_sk is
3229 * done with seq_show(), so unref the iter->cur_sk.
3231 if (iter->cur_sk < iter->end_sk) {
3232 sock_put(iter->batch[iter->cur_sk++]);
3236 /* After updating iter->cur_sk, check if there are more sockets
3237 * available in the current bucket batch.
3239 if (iter->cur_sk < iter->end_sk)
3240 sk = iter->batch[iter->cur_sk];
3242 /* Prepare a new batch. */
3243 sk = bpf_iter_udp_batch(seq);
3249 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3251 /* bpf iter does not support lseek, so it always
3252 * continue from where it was stop()-ped.
3255 return bpf_iter_udp_batch(seq);
3257 return SEQ_START_TOKEN;
3260 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3261 struct udp_sock *udp_sk, uid_t uid, int bucket)
3263 struct bpf_iter__udp ctx;
3265 meta->seq_num--; /* skip SEQ_START_TOKEN */
3267 ctx.udp_sk = udp_sk;
3269 ctx.bucket = bucket;
3270 return bpf_iter_run_prog(prog, &ctx);
3273 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3275 struct udp_iter_state *state = seq->private;
3276 struct bpf_iter_meta meta;
3277 struct bpf_prog *prog;
3278 struct sock *sk = v;
3282 if (v == SEQ_START_TOKEN)
3287 if (unlikely(sk_unhashed(sk))) {
3292 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3294 prog = bpf_iter_get_info(&meta, false);
3295 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3302 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3304 while (iter->cur_sk < iter->end_sk)
3305 sock_put(iter->batch[iter->cur_sk++]);
3308 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3310 struct bpf_udp_iter_state *iter = seq->private;
3311 struct bpf_iter_meta meta;
3312 struct bpf_prog *prog;
3316 prog = bpf_iter_get_info(&meta, true);
3318 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3321 if (iter->cur_sk < iter->end_sk) {
3322 bpf_iter_udp_put_batch(iter);
3323 iter->st_bucket_done = false;
3327 static const struct seq_operations bpf_iter_udp_seq_ops = {
3328 .start = bpf_iter_udp_seq_start,
3329 .next = bpf_iter_udp_seq_next,
3330 .stop = bpf_iter_udp_seq_stop,
3331 .show = bpf_iter_udp_seq_show,
3335 static unsigned short seq_file_family(const struct seq_file *seq)
3337 const struct udp_seq_afinfo *afinfo;
3339 #ifdef CONFIG_BPF_SYSCALL
3340 /* BPF iterator: bpf programs to filter sockets. */
3341 if (seq->op == &bpf_iter_udp_seq_ops)
3345 /* Proc fs iterator */
3346 afinfo = pde_data(file_inode(seq->file));
3347 return afinfo->family;
3350 const struct seq_operations udp_seq_ops = {
3351 .start = udp_seq_start,
3352 .next = udp_seq_next,
3353 .stop = udp_seq_stop,
3354 .show = udp4_seq_show,
3356 EXPORT_SYMBOL(udp_seq_ops);
3358 static struct udp_seq_afinfo udp4_seq_afinfo = {
3363 static int __net_init udp4_proc_init_net(struct net *net)
3365 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3366 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3371 static void __net_exit udp4_proc_exit_net(struct net *net)
3373 remove_proc_entry("udp", net->proc_net);
3376 static struct pernet_operations udp4_net_ops = {
3377 .init = udp4_proc_init_net,
3378 .exit = udp4_proc_exit_net,
3381 int __init udp4_proc_init(void)
3383 return register_pernet_subsys(&udp4_net_ops);
3386 void udp4_proc_exit(void)
3388 unregister_pernet_subsys(&udp4_net_ops);
3390 #endif /* CONFIG_PROC_FS */
3392 static __initdata unsigned long uhash_entries;
3393 static int __init set_uhash_entries(char *str)
3400 ret = kstrtoul(str, 0, &uhash_entries);
3404 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3405 uhash_entries = UDP_HTABLE_SIZE_MIN;
3408 __setup("uhash_entries=", set_uhash_entries);
3410 void __init udp_table_init(struct udp_table *table, const char *name)
3414 table->hash = alloc_large_system_hash(name,
3415 2 * sizeof(struct udp_hslot),
3417 21, /* one slot per 2 MB */
3421 UDP_HTABLE_SIZE_MIN,
3422 UDP_HTABLE_SIZE_MAX);
3424 table->hash2 = table->hash + (table->mask + 1);
3425 for (i = 0; i <= table->mask; i++) {
3426 INIT_HLIST_HEAD(&table->hash[i].head);
3427 table->hash[i].count = 0;
3428 spin_lock_init(&table->hash[i].lock);
3430 for (i = 0; i <= table->mask; i++) {
3431 INIT_HLIST_HEAD(&table->hash2[i].head);
3432 table->hash2[i].count = 0;
3433 spin_lock_init(&table->hash2[i].lock);
3437 u32 udp_flow_hashrnd(void)
3439 static u32 hashrnd __read_mostly;
3441 net_get_random_once(&hashrnd, sizeof(hashrnd));
3445 EXPORT_SYMBOL(udp_flow_hashrnd);
3447 static void __net_init udp_sysctl_init(struct net *net)
3449 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3450 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3452 #ifdef CONFIG_NET_L3_MASTER_DEV
3453 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3457 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3459 struct udp_table *udptable;
3462 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3466 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3467 GFP_KERNEL_ACCOUNT);
3468 if (!udptable->hash)
3471 udptable->hash2 = udptable->hash + hash_entries;
3472 udptable->mask = hash_entries - 1;
3473 udptable->log = ilog2(hash_entries);
3475 for (i = 0; i < hash_entries; i++) {
3476 INIT_HLIST_HEAD(&udptable->hash[i].head);
3477 udptable->hash[i].count = 0;
3478 spin_lock_init(&udptable->hash[i].lock);
3480 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3481 udptable->hash2[i].count = 0;
3482 spin_lock_init(&udptable->hash2[i].lock);
3493 static void __net_exit udp_pernet_table_free(struct net *net)
3495 struct udp_table *udptable = net->ipv4.udp_table;
3497 if (udptable == &udp_table)
3500 kvfree(udptable->hash);
3504 static void __net_init udp_set_table(struct net *net)
3506 struct udp_table *udptable;
3507 unsigned int hash_entries;
3508 struct net *old_net;
3510 if (net_eq(net, &init_net))
3513 old_net = current->nsproxy->net_ns;
3514 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3518 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3519 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3520 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3522 hash_entries = roundup_pow_of_two(hash_entries);
3524 udptable = udp_pernet_table_alloc(hash_entries);
3526 net->ipv4.udp_table = udptable;
3528 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3529 "for a netns, fallback to the global one\n",
3532 net->ipv4.udp_table = &udp_table;
3536 static int __net_init udp_pernet_init(struct net *net)
3538 udp_sysctl_init(net);
3544 static void __net_exit udp_pernet_exit(struct net *net)
3546 udp_pernet_table_free(net);
3549 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3550 .init = udp_pernet_init,
3551 .exit = udp_pernet_exit,
3554 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3555 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3556 struct udp_sock *udp_sk, uid_t uid, int bucket)
3558 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3559 unsigned int new_batch_sz)
3561 struct sock **new_batch;
3563 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3564 GFP_USER | __GFP_NOWARN);
3568 bpf_iter_udp_put_batch(iter);
3569 kvfree(iter->batch);
3570 iter->batch = new_batch;
3571 iter->max_sk = new_batch_sz;
3576 #define INIT_BATCH_SZ 16
3578 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3580 struct bpf_udp_iter_state *iter = priv_data;
3583 ret = bpf_iter_init_seq_net(priv_data, aux);
3587 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3589 bpf_iter_fini_seq_net(priv_data);
3594 static void bpf_iter_fini_udp(void *priv_data)
3596 struct bpf_udp_iter_state *iter = priv_data;
3598 bpf_iter_fini_seq_net(priv_data);
3599 kvfree(iter->batch);
3602 static const struct bpf_iter_seq_info udp_seq_info = {
3603 .seq_ops = &bpf_iter_udp_seq_ops,
3604 .init_seq_private = bpf_iter_init_udp,
3605 .fini_seq_private = bpf_iter_fini_udp,
3606 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3609 static struct bpf_iter_reg udp_reg_info = {
3611 .ctx_arg_info_size = 1,
3613 { offsetof(struct bpf_iter__udp, udp_sk),
3614 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3616 .seq_info = &udp_seq_info,
3619 static void __init bpf_iter_register(void)
3621 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3622 if (bpf_iter_reg_target(&udp_reg_info))
3623 pr_warn("Warning: could not register bpf iterator udp\n");
3627 void __init udp_init(void)
3629 unsigned long limit;
3632 udp_table_init(&udp_table, "UDP");
3633 limit = nr_free_buffer_pages() / 8;
3634 limit = max(limit, 128UL);
3635 sysctl_udp_mem[0] = limit / 4 * 3;
3636 sysctl_udp_mem[1] = limit;
3637 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3639 /* 16 spinlocks per cpu */
3640 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3641 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3644 panic("UDP: failed to alloc udp_busylocks\n");
3645 for (i = 0; i < (1U << udp_busylocks_log); i++)
3646 spin_lock_init(udp_busylocks + i);
3648 if (register_pernet_subsys(&udp_sysctl_ops))
3649 panic("UDP: failed to init sysctl parameters.\n");
3651 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3652 bpf_iter_register();