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 INDIRECT_CALLABLE_SCOPE
410 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
411 const __be32 faddr, const __be16 fport)
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 /* called with rcu_read_lock() */
422 static struct sock *udp4_lib_lookup2(struct net *net,
423 __be32 saddr, __be16 sport,
424 __be32 daddr, unsigned int hnum,
426 struct udp_hslot *hslot2,
429 struct sock *sk, *result;
434 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
435 score = compute_score(sk, net, saddr, sport,
436 daddr, hnum, dif, sdif);
437 if (score > badness) {
440 if (sk->sk_state == TCP_ESTABLISHED) {
445 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
446 saddr, sport, daddr, hnum, udp_ehashfn);
452 /* Fall back to scoring if group has connections */
453 if (!reuseport_has_conns(sk))
456 /* Reuseport logic returned an error, keep original score. */
460 badness = compute_score(result, net, saddr, sport,
461 daddr, hnum, dif, sdif);
468 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
469 * harder than this. -DaveM
471 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
472 __be16 sport, __be32 daddr, __be16 dport, int dif,
473 int sdif, struct udp_table *udptable, struct sk_buff *skb)
475 unsigned short hnum = ntohs(dport);
476 unsigned int hash2, slot2;
477 struct udp_hslot *hslot2;
478 struct sock *result, *sk;
480 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
481 slot2 = hash2 & udptable->mask;
482 hslot2 = &udptable->hash2[slot2];
484 /* Lookup connected or non-wildcard socket */
485 result = udp4_lib_lookup2(net, saddr, sport,
486 daddr, hnum, dif, sdif,
488 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
491 /* Lookup redirect from BPF */
492 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
493 udptable == net->ipv4.udp_table) {
494 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
495 saddr, sport, daddr, hnum, dif,
503 /* Got non-wildcard socket or error on first lookup */
507 /* Lookup wildcard sockets */
508 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
509 slot2 = hash2 & udptable->mask;
510 hslot2 = &udptable->hash2[slot2];
512 result = udp4_lib_lookup2(net, saddr, sport,
513 htonl(INADDR_ANY), hnum, dif, sdif,
520 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
522 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
523 __be16 sport, __be16 dport,
524 struct udp_table *udptable)
526 const struct iphdr *iph = ip_hdr(skb);
528 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
529 iph->daddr, dport, inet_iif(skb),
530 inet_sdif(skb), udptable, skb);
533 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
534 __be16 sport, __be16 dport)
536 const struct iphdr *iph = ip_hdr(skb);
537 struct net *net = dev_net(skb->dev);
539 return __udp4_lib_lookup(net, iph->saddr, sport,
540 iph->daddr, dport, inet_iif(skb),
541 inet_sdif(skb), net->ipv4.udp_table, NULL);
544 /* Must be called under rcu_read_lock().
545 * Does increment socket refcount.
547 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
548 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
549 __be32 daddr, __be16 dport, int dif)
553 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
554 dif, 0, net->ipv4.udp_table, NULL);
555 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
559 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
562 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
563 __be16 loc_port, __be32 loc_addr,
564 __be16 rmt_port, __be32 rmt_addr,
565 int dif, int sdif, unsigned short hnum)
567 const struct inet_sock *inet = inet_sk(sk);
569 if (!net_eq(sock_net(sk), net) ||
570 udp_sk(sk)->udp_port_hash != hnum ||
571 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
572 (inet->inet_dport != rmt_port && inet->inet_dport) ||
573 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
574 ipv6_only_sock(sk) ||
575 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
577 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
582 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
583 void udp_encap_enable(void)
585 static_branch_inc(&udp_encap_needed_key);
587 EXPORT_SYMBOL(udp_encap_enable);
589 void udp_encap_disable(void)
591 static_branch_dec(&udp_encap_needed_key);
593 EXPORT_SYMBOL(udp_encap_disable);
595 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
596 * through error handlers in encapsulations looking for a match.
598 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
602 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
603 int (*handler)(struct sk_buff *skb, u32 info);
604 const struct ip_tunnel_encap_ops *encap;
606 encap = rcu_dereference(iptun_encaps[i]);
609 handler = encap->err_handler;
610 if (handler && !handler(skb, info))
617 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
618 * reversing source and destination port: this will match tunnels that force the
619 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
620 * lwtunnels might actually break this assumption by being configured with
621 * different destination ports on endpoints, in this case we won't be able to
622 * trace ICMP messages back to them.
624 * If this doesn't match any socket, probe tunnels with arbitrary destination
625 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
626 * we've sent packets to won't necessarily match the local destination port.
628 * Then ask the tunnel implementation to match the error against a valid
631 * Return an error if we can't find a match, the socket if we need further
632 * processing, zero otherwise.
634 static struct sock *__udp4_lib_err_encap(struct net *net,
635 const struct iphdr *iph,
637 struct udp_table *udptable,
639 struct sk_buff *skb, u32 info)
641 int (*lookup)(struct sock *sk, struct sk_buff *skb);
642 int network_offset, transport_offset;
645 network_offset = skb_network_offset(skb);
646 transport_offset = skb_transport_offset(skb);
648 /* Network header needs to point to the outer IPv4 header inside ICMP */
649 skb_reset_network_header(skb);
651 /* Transport header needs to point to the UDP header */
652 skb_set_transport_header(skb, iph->ihl << 2);
657 lookup = READ_ONCE(up->encap_err_lookup);
658 if (lookup && lookup(sk, skb))
664 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
665 iph->saddr, uh->dest, skb->dev->ifindex, 0,
670 lookup = READ_ONCE(up->encap_err_lookup);
671 if (!lookup || lookup(sk, skb))
677 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
679 skb_set_transport_header(skb, transport_offset);
680 skb_set_network_header(skb, network_offset);
686 * This routine is called by the ICMP module when it gets some
687 * sort of error condition. If err < 0 then the socket should
688 * be closed and the error returned to the user. If err > 0
689 * it's just the icmp type << 8 | icmp code.
690 * Header points to the ip header of the error packet. We move
691 * on past this. Then (as it used to claim before adjustment)
692 * header points to the first 8 bytes of the udp header. We need
693 * to find the appropriate port.
696 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
698 struct inet_sock *inet;
699 const struct iphdr *iph = (const struct iphdr *)skb->data;
700 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
701 const int type = icmp_hdr(skb)->type;
702 const int code = icmp_hdr(skb)->code;
707 struct net *net = dev_net(skb->dev);
709 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
710 iph->saddr, uh->source, skb->dev->ifindex,
711 inet_sdif(skb), udptable, NULL);
713 if (!sk || udp_sk(sk)->encap_type) {
714 /* No socket for error: try tunnels before discarding */
715 if (static_branch_unlikely(&udp_encap_needed_key)) {
716 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
721 sk = ERR_PTR(-ENOENT);
724 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
737 case ICMP_TIME_EXCEEDED:
740 case ICMP_SOURCE_QUENCH:
742 case ICMP_PARAMETERPROB:
746 case ICMP_DEST_UNREACH:
747 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
748 ipv4_sk_update_pmtu(skb, sk, info);
749 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
757 if (code <= NR_ICMP_UNREACH) {
758 harderr = icmp_err_convert[code].fatal;
759 err = icmp_err_convert[code].errno;
763 ipv4_sk_redirect(skb, sk);
768 * RFC1122: OK. Passes ICMP errors back to application, as per
772 /* ...not for tunnels though: we don't have a sending socket */
773 if (udp_sk(sk)->encap_err_rcv)
774 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
778 if (!inet->recverr) {
779 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
782 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
790 int udp_err(struct sk_buff *skb, u32 info)
792 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
796 * Throw away all pending data and cancel the corking. Socket is locked.
798 void udp_flush_pending_frames(struct sock *sk)
800 struct udp_sock *up = udp_sk(sk);
805 ip_flush_pending_frames(sk);
808 EXPORT_SYMBOL(udp_flush_pending_frames);
811 * udp4_hwcsum - handle outgoing HW checksumming
812 * @skb: sk_buff containing the filled-in UDP header
813 * (checksum field must be zeroed out)
814 * @src: source IP address
815 * @dst: destination IP address
817 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
819 struct udphdr *uh = udp_hdr(skb);
820 int offset = skb_transport_offset(skb);
821 int len = skb->len - offset;
825 if (!skb_has_frag_list(skb)) {
827 * Only one fragment on the socket.
829 skb->csum_start = skb_transport_header(skb) - skb->head;
830 skb->csum_offset = offsetof(struct udphdr, check);
831 uh->check = ~csum_tcpudp_magic(src, dst, len,
834 struct sk_buff *frags;
837 * HW-checksum won't work as there are two or more
838 * fragments on the socket so that all csums of sk_buffs
841 skb_walk_frags(skb, frags) {
842 csum = csum_add(csum, frags->csum);
846 csum = skb_checksum(skb, offset, hlen, csum);
847 skb->ip_summed = CHECKSUM_NONE;
849 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
851 uh->check = CSUM_MANGLED_0;
854 EXPORT_SYMBOL_GPL(udp4_hwcsum);
856 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
857 * for the simple case like when setting the checksum for a UDP tunnel.
859 void udp_set_csum(bool nocheck, struct sk_buff *skb,
860 __be32 saddr, __be32 daddr, int len)
862 struct udphdr *uh = udp_hdr(skb);
866 } else if (skb_is_gso(skb)) {
867 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
868 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
870 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
872 uh->check = CSUM_MANGLED_0;
874 skb->ip_summed = CHECKSUM_PARTIAL;
875 skb->csum_start = skb_transport_header(skb) - skb->head;
876 skb->csum_offset = offsetof(struct udphdr, check);
877 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
880 EXPORT_SYMBOL(udp_set_csum);
882 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
883 struct inet_cork *cork)
885 struct sock *sk = skb->sk;
886 struct inet_sock *inet = inet_sk(sk);
889 int is_udplite = IS_UDPLITE(sk);
890 int offset = skb_transport_offset(skb);
891 int len = skb->len - offset;
892 int datalen = len - sizeof(*uh);
896 * Create a UDP header
899 uh->source = inet->inet_sport;
900 uh->dest = fl4->fl4_dport;
901 uh->len = htons(len);
904 if (cork->gso_size) {
905 const int hlen = skb_network_header_len(skb) +
906 sizeof(struct udphdr);
908 if (hlen + cork->gso_size > cork->fragsize) {
912 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
916 if (sk->sk_no_check_tx) {
920 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
921 dst_xfrm(skb_dst(skb))) {
926 if (datalen > cork->gso_size) {
927 skb_shinfo(skb)->gso_size = cork->gso_size;
928 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
929 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
935 if (is_udplite) /* UDP-Lite */
936 csum = udplite_csum(skb);
938 else if (sk->sk_no_check_tx) { /* UDP csum off */
940 skb->ip_summed = CHECKSUM_NONE;
943 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
946 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
950 csum = udp_csum(skb);
952 /* add protocol-dependent pseudo-header */
953 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
954 sk->sk_protocol, csum);
956 uh->check = CSUM_MANGLED_0;
959 err = ip_send_skb(sock_net(sk), skb);
961 if (err == -ENOBUFS && !inet->recverr) {
962 UDP_INC_STATS(sock_net(sk),
963 UDP_MIB_SNDBUFERRORS, is_udplite);
967 UDP_INC_STATS(sock_net(sk),
968 UDP_MIB_OUTDATAGRAMS, is_udplite);
973 * Push out all pending data as one UDP datagram. Socket is locked.
975 int udp_push_pending_frames(struct sock *sk)
977 struct udp_sock *up = udp_sk(sk);
978 struct inet_sock *inet = inet_sk(sk);
979 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
983 skb = ip_finish_skb(sk, fl4);
987 err = udp_send_skb(skb, fl4, &inet->cork.base);
994 EXPORT_SYMBOL(udp_push_pending_frames);
996 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
998 switch (cmsg->cmsg_type) {
1000 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1002 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1009 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1011 struct cmsghdr *cmsg;
1012 bool need_ip = false;
1015 for_each_cmsghdr(cmsg, msg) {
1016 if (!CMSG_OK(msg, cmsg))
1019 if (cmsg->cmsg_level != SOL_UDP) {
1024 err = __udp_cmsg_send(cmsg, gso_size);
1031 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1033 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1035 struct inet_sock *inet = inet_sk(sk);
1036 struct udp_sock *up = udp_sk(sk);
1037 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1038 struct flowi4 fl4_stack;
1041 struct ipcm_cookie ipc;
1042 struct rtable *rt = NULL;
1045 __be32 daddr, faddr, saddr;
1048 int err, is_udplite = IS_UDPLITE(sk);
1049 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1050 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1051 struct sk_buff *skb;
1052 struct ip_options_data opt_copy;
1061 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1064 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1066 fl4 = &inet->cork.fl.u.ip4;
1069 * There are pending frames.
1070 * The socket lock must be held while it's corked.
1073 if (likely(up->pending)) {
1074 if (unlikely(up->pending != AF_INET)) {
1078 goto do_append_data;
1082 ulen += sizeof(struct udphdr);
1085 * Get and verify the address.
1088 if (msg->msg_namelen < sizeof(*usin))
1090 if (usin->sin_family != AF_INET) {
1091 if (usin->sin_family != AF_UNSPEC)
1092 return -EAFNOSUPPORT;
1095 daddr = usin->sin_addr.s_addr;
1096 dport = usin->sin_port;
1100 if (sk->sk_state != TCP_ESTABLISHED)
1101 return -EDESTADDRREQ;
1102 daddr = inet->inet_daddr;
1103 dport = inet->inet_dport;
1104 /* Open fast path for connected socket.
1105 Route will not be used, if at least one option is set.
1110 ipcm_init_sk(&ipc, inet);
1111 ipc.gso_size = READ_ONCE(up->gso_size);
1113 if (msg->msg_controllen) {
1114 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1116 err = ip_cmsg_send(sk, msg, &ipc,
1117 sk->sk_family == AF_INET6);
1118 if (unlikely(err < 0)) {
1127 struct ip_options_rcu *inet_opt;
1130 inet_opt = rcu_dereference(inet->inet_opt);
1132 memcpy(&opt_copy, inet_opt,
1133 sizeof(*inet_opt) + inet_opt->opt.optlen);
1134 ipc.opt = &opt_copy.opt;
1139 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1140 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1141 (struct sockaddr *)usin, &ipc.addr);
1145 if (usin->sin_port == 0) {
1146 /* BPF program set invalid port. Reject it. */
1150 daddr = usin->sin_addr.s_addr;
1151 dport = usin->sin_port;
1156 ipc.addr = faddr = daddr;
1158 if (ipc.opt && ipc.opt->opt.srr) {
1163 faddr = ipc.opt->opt.faddr;
1166 tos = get_rttos(&ipc, inet);
1167 scope = ip_sendmsg_scope(inet, &ipc, msg);
1168 if (scope == RT_SCOPE_LINK)
1171 if (ipv4_is_multicast(daddr)) {
1172 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1173 ipc.oif = inet->mc_index;
1175 saddr = inet->mc_addr;
1177 } else if (!ipc.oif) {
1178 ipc.oif = inet->uc_index;
1179 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1180 /* oif is set, packet is to local broadcast and
1181 * uc_index is set. oif is most likely set
1182 * by sk_bound_dev_if. If uc_index != oif check if the
1183 * oif is an L3 master and uc_index is an L3 slave.
1184 * If so, we want to allow the send using the uc_index.
1186 if (ipc.oif != inet->uc_index &&
1187 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1189 ipc.oif = inet->uc_index;
1194 rt = (struct rtable *)sk_dst_check(sk, 0);
1197 struct net *net = sock_net(sk);
1198 __u8 flow_flags = inet_sk_flowi_flags(sk);
1202 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1203 sk->sk_protocol, flow_flags, faddr, saddr,
1204 dport, inet->inet_sport, sk->sk_uid);
1206 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1207 rt = ip_route_output_flow(net, fl4, sk);
1211 if (err == -ENETUNREACH)
1212 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1217 if ((rt->rt_flags & RTCF_BROADCAST) &&
1218 !sock_flag(sk, SOCK_BROADCAST))
1221 sk_dst_set(sk, dst_clone(&rt->dst));
1224 if (msg->msg_flags&MSG_CONFIRM)
1230 daddr = ipc.addr = fl4->daddr;
1232 /* Lockless fast path for the non-corking case. */
1234 struct inet_cork cork;
1236 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1237 sizeof(struct udphdr), &ipc, &rt,
1238 &cork, msg->msg_flags);
1240 if (!IS_ERR_OR_NULL(skb))
1241 err = udp_send_skb(skb, fl4, &cork);
1246 if (unlikely(up->pending)) {
1247 /* The socket is already corked while preparing it. */
1248 /* ... which is an evident application bug. --ANK */
1251 net_dbg_ratelimited("socket already corked\n");
1256 * Now cork the socket to pend data.
1258 fl4 = &inet->cork.fl.u.ip4;
1261 fl4->fl4_dport = dport;
1262 fl4->fl4_sport = inet->inet_sport;
1263 up->pending = AF_INET;
1267 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1268 sizeof(struct udphdr), &ipc, &rt,
1269 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1271 udp_flush_pending_frames(sk);
1273 err = udp_push_pending_frames(sk);
1274 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1286 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1287 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1288 * we don't have a good statistic (IpOutDiscards but it can be too many
1289 * things). We could add another new stat but at least for now that
1290 * seems like overkill.
1292 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1293 UDP_INC_STATS(sock_net(sk),
1294 UDP_MIB_SNDBUFERRORS, is_udplite);
1299 if (msg->msg_flags & MSG_PROBE)
1300 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1301 if (!(msg->msg_flags&MSG_PROBE) || len)
1302 goto back_from_confirm;
1306 EXPORT_SYMBOL(udp_sendmsg);
1308 void udp_splice_eof(struct socket *sock)
1310 struct sock *sk = sock->sk;
1311 struct udp_sock *up = udp_sk(sk);
1313 if (!up->pending || READ_ONCE(up->corkflag))
1317 if (up->pending && !READ_ONCE(up->corkflag))
1318 udp_push_pending_frames(sk);
1321 EXPORT_SYMBOL_GPL(udp_splice_eof);
1323 #define UDP_SKB_IS_STATELESS 0x80000000
1325 /* all head states (dst, sk, nf conntrack) except skb extensions are
1326 * cleared by udp_rcv().
1328 * We need to preserve secpath, if present, to eventually process
1329 * IP_CMSG_PASSSEC at recvmsg() time.
1331 * Other extensions can be cleared.
1333 static bool udp_try_make_stateless(struct sk_buff *skb)
1335 if (!skb_has_extensions(skb))
1338 if (!secpath_exists(skb)) {
1346 static void udp_set_dev_scratch(struct sk_buff *skb)
1348 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1350 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1351 scratch->_tsize_state = skb->truesize;
1352 #if BITS_PER_LONG == 64
1353 scratch->len = skb->len;
1354 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1355 scratch->is_linear = !skb_is_nonlinear(skb);
1357 if (udp_try_make_stateless(skb))
1358 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1361 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1363 /* We come here after udp_lib_checksum_complete() returned 0.
1364 * This means that __skb_checksum_complete() might have
1365 * set skb->csum_valid to 1.
1366 * On 64bit platforms, we can set csum_unnecessary
1367 * to true, but only if the skb is not shared.
1369 #if BITS_PER_LONG == 64
1370 if (!skb_shared(skb))
1371 udp_skb_scratch(skb)->csum_unnecessary = true;
1375 static int udp_skb_truesize(struct sk_buff *skb)
1377 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1380 static bool udp_skb_has_head_state(struct sk_buff *skb)
1382 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1385 /* fully reclaim rmem/fwd memory allocated for skb */
1386 static void udp_rmem_release(struct sock *sk, int size, int partial,
1387 bool rx_queue_lock_held)
1389 struct udp_sock *up = udp_sk(sk);
1390 struct sk_buff_head *sk_queue;
1393 if (likely(partial)) {
1394 up->forward_deficit += size;
1395 size = up->forward_deficit;
1396 if (size < READ_ONCE(up->forward_threshold) &&
1397 !skb_queue_empty(&up->reader_queue))
1400 size += up->forward_deficit;
1402 up->forward_deficit = 0;
1404 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1405 * if the called don't held it already
1407 sk_queue = &sk->sk_receive_queue;
1408 if (!rx_queue_lock_held)
1409 spin_lock(&sk_queue->lock);
1412 sk->sk_forward_alloc += size;
1413 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1414 sk->sk_forward_alloc -= amt;
1417 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1419 atomic_sub(size, &sk->sk_rmem_alloc);
1421 /* this can save us from acquiring the rx queue lock on next receive */
1422 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1424 if (!rx_queue_lock_held)
1425 spin_unlock(&sk_queue->lock);
1428 /* Note: called with reader_queue.lock held.
1429 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1430 * This avoids a cache line miss while receive_queue lock is held.
1431 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1433 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1435 prefetch(&skb->data);
1436 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1438 EXPORT_SYMBOL(udp_skb_destructor);
1440 /* as above, but the caller held the rx queue lock, too */
1441 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1443 prefetch(&skb->data);
1444 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1447 /* Idea of busylocks is to let producers grab an extra spinlock
1448 * to relieve pressure on the receive_queue spinlock shared by consumer.
1449 * Under flood, this means that only one producer can be in line
1450 * trying to acquire the receive_queue spinlock.
1451 * These busylock can be allocated on a per cpu manner, instead of a
1452 * per socket one (that would consume a cache line per socket)
1454 static int udp_busylocks_log __read_mostly;
1455 static spinlock_t *udp_busylocks __read_mostly;
1457 static spinlock_t *busylock_acquire(void *ptr)
1461 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1466 static void busylock_release(spinlock_t *busy)
1472 static int udp_rmem_schedule(struct sock *sk, int size)
1476 delta = size - sk->sk_forward_alloc;
1477 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1483 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1485 struct sk_buff_head *list = &sk->sk_receive_queue;
1486 int rmem, err = -ENOMEM;
1487 spinlock_t *busy = NULL;
1490 /* try to avoid the costly atomic add/sub pair when the receive
1491 * queue is full; always allow at least a packet
1493 rmem = atomic_read(&sk->sk_rmem_alloc);
1494 if (rmem > sk->sk_rcvbuf)
1497 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1498 * having linear skbs :
1499 * - Reduce memory overhead and thus increase receive queue capacity
1500 * - Less cache line misses at copyout() time
1501 * - Less work at consume_skb() (less alien page frag freeing)
1503 if (rmem > (sk->sk_rcvbuf >> 1)) {
1506 busy = busylock_acquire(sk);
1508 size = skb->truesize;
1509 udp_set_dev_scratch(skb);
1511 /* we drop only if the receive buf is full and the receive
1512 * queue contains some other skb
1514 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1515 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1518 spin_lock(&list->lock);
1519 err = udp_rmem_schedule(sk, size);
1521 spin_unlock(&list->lock);
1525 sk->sk_forward_alloc -= size;
1527 /* no need to setup a destructor, we will explicitly release the
1528 * forward allocated memory on dequeue
1530 sock_skb_set_dropcount(sk, skb);
1532 __skb_queue_tail(list, skb);
1533 spin_unlock(&list->lock);
1535 if (!sock_flag(sk, SOCK_DEAD))
1536 INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
1538 busylock_release(busy);
1542 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1545 atomic_inc(&sk->sk_drops);
1546 busylock_release(busy);
1549 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1551 void udp_destruct_common(struct sock *sk)
1553 /* reclaim completely the forward allocated memory */
1554 struct udp_sock *up = udp_sk(sk);
1555 unsigned int total = 0;
1556 struct sk_buff *skb;
1558 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1559 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1560 total += skb->truesize;
1563 udp_rmem_release(sk, total, 0, true);
1565 EXPORT_SYMBOL_GPL(udp_destruct_common);
1567 static void udp_destruct_sock(struct sock *sk)
1569 udp_destruct_common(sk);
1570 inet_sock_destruct(sk);
1573 int udp_init_sock(struct sock *sk)
1575 udp_lib_init_sock(sk);
1576 sk->sk_destruct = udp_destruct_sock;
1577 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1581 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1583 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1584 bool slow = lock_sock_fast(sk);
1586 sk_peek_offset_bwd(sk, len);
1587 unlock_sock_fast(sk, slow);
1590 if (!skb_unref(skb))
1593 /* In the more common cases we cleared the head states previously,
1594 * see __udp_queue_rcv_skb().
1596 if (unlikely(udp_skb_has_head_state(skb)))
1597 skb_release_head_state(skb);
1598 __consume_stateless_skb(skb);
1600 EXPORT_SYMBOL_GPL(skb_consume_udp);
1602 static struct sk_buff *__first_packet_length(struct sock *sk,
1603 struct sk_buff_head *rcvq,
1606 struct sk_buff *skb;
1608 while ((skb = skb_peek(rcvq)) != NULL) {
1609 if (udp_lib_checksum_complete(skb)) {
1610 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1612 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1614 atomic_inc(&sk->sk_drops);
1615 __skb_unlink(skb, rcvq);
1616 *total += skb->truesize;
1619 udp_skb_csum_unnecessary_set(skb);
1627 * first_packet_length - return length of first packet in receive queue
1630 * Drops all bad checksum frames, until a valid one is found.
1631 * Returns the length of found skb, or -1 if none is found.
1633 static int first_packet_length(struct sock *sk)
1635 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1636 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1637 struct sk_buff *skb;
1641 spin_lock_bh(&rcvq->lock);
1642 skb = __first_packet_length(sk, rcvq, &total);
1643 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1644 spin_lock(&sk_queue->lock);
1645 skb_queue_splice_tail_init(sk_queue, rcvq);
1646 spin_unlock(&sk_queue->lock);
1648 skb = __first_packet_length(sk, rcvq, &total);
1650 res = skb ? skb->len : -1;
1652 udp_rmem_release(sk, total, 1, false);
1653 spin_unlock_bh(&rcvq->lock);
1658 * IOCTL requests applicable to the UDP protocol
1661 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1666 *karg = sk_wmem_alloc_get(sk);
1672 *karg = max_t(int, 0, first_packet_length(sk));
1677 return -ENOIOCTLCMD;
1682 EXPORT_SYMBOL(udp_ioctl);
1684 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1687 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1688 struct sk_buff_head *queue;
1689 struct sk_buff *last;
1693 queue = &udp_sk(sk)->reader_queue;
1694 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1696 struct sk_buff *skb;
1698 error = sock_error(sk);
1704 spin_lock_bh(&queue->lock);
1705 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1708 if (!(flags & MSG_PEEK))
1709 udp_skb_destructor(sk, skb);
1710 spin_unlock_bh(&queue->lock);
1714 if (skb_queue_empty_lockless(sk_queue)) {
1715 spin_unlock_bh(&queue->lock);
1719 /* refill the reader queue and walk it again
1720 * keep both queues locked to avoid re-acquiring
1721 * the sk_receive_queue lock if fwd memory scheduling
1724 spin_lock(&sk_queue->lock);
1725 skb_queue_splice_tail_init(sk_queue, queue);
1727 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1729 if (skb && !(flags & MSG_PEEK))
1730 udp_skb_dtor_locked(sk, skb);
1731 spin_unlock(&sk_queue->lock);
1732 spin_unlock_bh(&queue->lock);
1737 if (!sk_can_busy_loop(sk))
1740 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1741 } while (!skb_queue_empty_lockless(sk_queue));
1743 /* sk_queue is empty, reader_queue may contain peeked packets */
1745 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1747 (struct sk_buff *)sk_queue));
1752 EXPORT_SYMBOL(__skb_recv_udp);
1754 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1756 struct sk_buff *skb;
1760 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1764 if (udp_lib_checksum_complete(skb)) {
1765 int is_udplite = IS_UDPLITE(sk);
1766 struct net *net = sock_net(sk);
1768 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1769 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1770 atomic_inc(&sk->sk_drops);
1775 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1776 return recv_actor(sk, skb);
1778 EXPORT_SYMBOL(udp_read_skb);
1781 * This should be easy, if there is something there we
1782 * return it, otherwise we block.
1785 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1788 struct inet_sock *inet = inet_sk(sk);
1789 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1790 struct sk_buff *skb;
1791 unsigned int ulen, copied;
1792 int off, err, peeking = flags & MSG_PEEK;
1793 int is_udplite = IS_UDPLITE(sk);
1794 bool checksum_valid = false;
1796 if (flags & MSG_ERRQUEUE)
1797 return ip_recv_error(sk, msg, len, addr_len);
1800 off = sk_peek_offset(sk, flags);
1801 skb = __skb_recv_udp(sk, flags, &off, &err);
1805 ulen = udp_skb_len(skb);
1807 if (copied > ulen - off)
1808 copied = ulen - off;
1809 else if (copied < ulen)
1810 msg->msg_flags |= MSG_TRUNC;
1813 * If checksum is needed at all, try to do it while copying the
1814 * data. If the data is truncated, or if we only want a partial
1815 * coverage checksum (UDP-Lite), do it before the copy.
1818 if (copied < ulen || peeking ||
1819 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1820 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1821 !__udp_lib_checksum_complete(skb);
1822 if (!checksum_valid)
1826 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1827 if (udp_skb_is_linear(skb))
1828 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1830 err = skb_copy_datagram_msg(skb, off, msg, copied);
1832 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1838 if (unlikely(err)) {
1840 atomic_inc(&sk->sk_drops);
1841 UDP_INC_STATS(sock_net(sk),
1842 UDP_MIB_INERRORS, is_udplite);
1849 UDP_INC_STATS(sock_net(sk),
1850 UDP_MIB_INDATAGRAMS, is_udplite);
1852 sock_recv_cmsgs(msg, sk, skb);
1854 /* Copy the address. */
1856 sin->sin_family = AF_INET;
1857 sin->sin_port = udp_hdr(skb)->source;
1858 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1859 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1860 *addr_len = sizeof(*sin);
1862 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1863 (struct sockaddr *)sin);
1866 if (udp_sk(sk)->gro_enabled)
1867 udp_cmsg_recv(msg, sk, skb);
1869 if (inet->cmsg_flags)
1870 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1873 if (flags & MSG_TRUNC)
1876 skb_consume_udp(sk, skb, peeking ? -err : err);
1880 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1881 udp_skb_destructor)) {
1882 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1883 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1887 /* starting over for a new packet, but check if we need to yield */
1889 msg->msg_flags &= ~MSG_TRUNC;
1893 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1895 /* This check is replicated from __ip4_datagram_connect() and
1896 * intended to prevent BPF program called below from accessing bytes
1897 * that are out of the bound specified by user in addr_len.
1899 if (addr_len < sizeof(struct sockaddr_in))
1902 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1904 EXPORT_SYMBOL(udp_pre_connect);
1906 int __udp_disconnect(struct sock *sk, int flags)
1908 struct inet_sock *inet = inet_sk(sk);
1910 * 1003.1g - break association.
1913 sk->sk_state = TCP_CLOSE;
1914 inet->inet_daddr = 0;
1915 inet->inet_dport = 0;
1916 sock_rps_reset_rxhash(sk);
1917 sk->sk_bound_dev_if = 0;
1918 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1919 inet_reset_saddr(sk);
1920 if (sk->sk_prot->rehash &&
1921 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1922 sk->sk_prot->rehash(sk);
1925 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1926 sk->sk_prot->unhash(sk);
1927 inet->inet_sport = 0;
1932 EXPORT_SYMBOL(__udp_disconnect);
1934 int udp_disconnect(struct sock *sk, int flags)
1937 __udp_disconnect(sk, flags);
1941 EXPORT_SYMBOL(udp_disconnect);
1943 void udp_lib_unhash(struct sock *sk)
1945 if (sk_hashed(sk)) {
1946 struct udp_table *udptable = udp_get_table_prot(sk);
1947 struct udp_hslot *hslot, *hslot2;
1949 hslot = udp_hashslot(udptable, sock_net(sk),
1950 udp_sk(sk)->udp_port_hash);
1951 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1953 spin_lock_bh(&hslot->lock);
1954 if (rcu_access_pointer(sk->sk_reuseport_cb))
1955 reuseport_detach_sock(sk);
1956 if (sk_del_node_init_rcu(sk)) {
1958 inet_sk(sk)->inet_num = 0;
1959 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1961 spin_lock(&hslot2->lock);
1962 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1964 spin_unlock(&hslot2->lock);
1966 spin_unlock_bh(&hslot->lock);
1969 EXPORT_SYMBOL(udp_lib_unhash);
1972 * inet_rcv_saddr was changed, we must rehash secondary hash
1974 void udp_lib_rehash(struct sock *sk, u16 newhash)
1976 if (sk_hashed(sk)) {
1977 struct udp_table *udptable = udp_get_table_prot(sk);
1978 struct udp_hslot *hslot, *hslot2, *nhslot2;
1980 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1981 nhslot2 = udp_hashslot2(udptable, newhash);
1982 udp_sk(sk)->udp_portaddr_hash = newhash;
1984 if (hslot2 != nhslot2 ||
1985 rcu_access_pointer(sk->sk_reuseport_cb)) {
1986 hslot = udp_hashslot(udptable, sock_net(sk),
1987 udp_sk(sk)->udp_port_hash);
1988 /* we must lock primary chain too */
1989 spin_lock_bh(&hslot->lock);
1990 if (rcu_access_pointer(sk->sk_reuseport_cb))
1991 reuseport_detach_sock(sk);
1993 if (hslot2 != nhslot2) {
1994 spin_lock(&hslot2->lock);
1995 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1997 spin_unlock(&hslot2->lock);
1999 spin_lock(&nhslot2->lock);
2000 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2003 spin_unlock(&nhslot2->lock);
2006 spin_unlock_bh(&hslot->lock);
2010 EXPORT_SYMBOL(udp_lib_rehash);
2012 void udp_v4_rehash(struct sock *sk)
2014 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2015 inet_sk(sk)->inet_rcv_saddr,
2016 inet_sk(sk)->inet_num);
2017 udp_lib_rehash(sk, new_hash);
2020 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2024 if (inet_sk(sk)->inet_daddr) {
2025 sock_rps_save_rxhash(sk, skb);
2026 sk_mark_napi_id(sk, skb);
2027 sk_incoming_cpu_update(sk);
2029 sk_mark_napi_id_once(sk, skb);
2032 rc = __udp_enqueue_schedule_skb(sk, skb);
2034 int is_udplite = IS_UDPLITE(sk);
2037 /* Note that an ENOMEM error is charged twice */
2038 if (rc == -ENOMEM) {
2039 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2041 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2043 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2045 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2047 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2048 kfree_skb_reason(skb, drop_reason);
2049 trace_udp_fail_queue_rcv_skb(rc, sk);
2059 * >0: "udp encap" protocol resubmission
2061 * Note that in the success and error cases, the skb is assumed to
2062 * have either been requeued or freed.
2064 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2066 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2067 struct udp_sock *up = udp_sk(sk);
2068 int is_udplite = IS_UDPLITE(sk);
2071 * Charge it to the socket, dropping if the queue is full.
2073 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2074 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2079 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2080 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2083 * This is an encapsulation socket so pass the skb to
2084 * the socket's udp_encap_rcv() hook. Otherwise, just
2085 * fall through and pass this up the UDP socket.
2086 * up->encap_rcv() returns the following value:
2087 * =0 if skb was successfully passed to the encap
2088 * handler or was discarded by it.
2089 * >0 if skb should be passed on to UDP.
2090 * <0 if skb should be resubmitted as proto -N
2093 /* if we're overly short, let UDP handle it */
2094 encap_rcv = READ_ONCE(up->encap_rcv);
2098 /* Verify checksum before giving to encap */
2099 if (udp_lib_checksum_complete(skb))
2102 ret = encap_rcv(sk, skb);
2104 __UDP_INC_STATS(sock_net(sk),
2105 UDP_MIB_INDATAGRAMS,
2111 /* FALLTHROUGH -- it's a UDP Packet */
2115 * UDP-Lite specific tests, ignored on UDP sockets
2117 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2120 * MIB statistics other than incrementing the error count are
2121 * disabled for the following two types of errors: these depend
2122 * on the application settings, not on the functioning of the
2123 * protocol stack as such.
2125 * RFC 3828 here recommends (sec 3.3): "There should also be a
2126 * way ... to ... at least let the receiving application block
2127 * delivery of packets with coverage values less than a value
2128 * provided by the application."
2130 if (up->pcrlen == 0) { /* full coverage was set */
2131 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2132 UDP_SKB_CB(skb)->cscov, skb->len);
2135 /* The next case involves violating the min. coverage requested
2136 * by the receiver. This is subtle: if receiver wants x and x is
2137 * greater than the buffersize/MTU then receiver will complain
2138 * that it wants x while sender emits packets of smaller size y.
2139 * Therefore the above ...()->partial_cov statement is essential.
2141 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2142 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2143 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2148 prefetch(&sk->sk_rmem_alloc);
2149 if (rcu_access_pointer(sk->sk_filter) &&
2150 udp_lib_checksum_complete(skb))
2153 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2154 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2158 udp_csum_pull_header(skb);
2160 ipv4_pktinfo_prepare(sk, skb);
2161 return __udp_queue_rcv_skb(sk, skb);
2164 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2165 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2167 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2168 atomic_inc(&sk->sk_drops);
2169 kfree_skb_reason(skb, drop_reason);
2173 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2175 struct sk_buff *next, *segs;
2178 if (likely(!udp_unexpected_gso(sk, skb)))
2179 return udp_queue_rcv_one_skb(sk, skb);
2181 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2182 __skb_push(skb, -skb_mac_offset(skb));
2183 segs = udp_rcv_segment(sk, skb, true);
2184 skb_list_walk_safe(segs, skb, next) {
2185 __skb_pull(skb, skb_transport_offset(skb));
2187 udp_post_segment_fix_csum(skb);
2188 ret = udp_queue_rcv_one_skb(sk, skb);
2190 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2195 /* For TCP sockets, sk_rx_dst is protected by socket lock
2196 * For UDP, we use xchg() to guard against concurrent changes.
2198 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2200 struct dst_entry *old;
2202 if (dst_hold_safe(dst)) {
2203 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2209 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2212 * Multicasts and broadcasts go to each listener.
2214 * Note: called only from the BH handler context.
2216 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2218 __be32 saddr, __be32 daddr,
2219 struct udp_table *udptable,
2222 struct sock *sk, *first = NULL;
2223 unsigned short hnum = ntohs(uh->dest);
2224 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2225 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2226 unsigned int offset = offsetof(typeof(*sk), sk_node);
2227 int dif = skb->dev->ifindex;
2228 int sdif = inet_sdif(skb);
2229 struct hlist_node *node;
2230 struct sk_buff *nskb;
2233 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2235 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2237 hslot = &udptable->hash2[hash2];
2238 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2241 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2242 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2243 uh->source, saddr, dif, sdif, hnum))
2250 nskb = skb_clone(skb, GFP_ATOMIC);
2252 if (unlikely(!nskb)) {
2253 atomic_inc(&sk->sk_drops);
2254 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2256 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2260 if (udp_queue_rcv_skb(sk, nskb) > 0)
2264 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2265 if (use_hash2 && hash2 != hash2_any) {
2271 if (udp_queue_rcv_skb(first, skb) > 0)
2275 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2276 proto == IPPROTO_UDPLITE);
2281 /* Initialize UDP checksum. If exited with zero value (success),
2282 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2283 * Otherwise, csum completion requires checksumming packet body,
2284 * including udp header and folding it to skb->csum.
2286 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2291 UDP_SKB_CB(skb)->partial_cov = 0;
2292 UDP_SKB_CB(skb)->cscov = skb->len;
2294 if (proto == IPPROTO_UDPLITE) {
2295 err = udplite_checksum_init(skb, uh);
2299 if (UDP_SKB_CB(skb)->partial_cov) {
2300 skb->csum = inet_compute_pseudo(skb, proto);
2305 /* Note, we are only interested in != 0 or == 0, thus the
2308 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2309 inet_compute_pseudo);
2313 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2314 /* If SW calculated the value, we know it's bad */
2315 if (skb->csum_complete_sw)
2318 /* HW says the value is bad. Let's validate that.
2319 * skb->csum is no longer the full packet checksum,
2320 * so don't treat it as such.
2322 skb_checksum_complete_unset(skb);
2328 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2329 * return code conversion for ip layer consumption
2331 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2336 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2337 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2339 ret = udp_queue_rcv_skb(sk, skb);
2341 /* a return value > 0 means to resubmit the input, but
2342 * it wants the return to be -protocol, or 0
2350 * All we need to do is get the socket, and then do a checksum.
2353 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2358 unsigned short ulen;
2359 struct rtable *rt = skb_rtable(skb);
2360 __be32 saddr, daddr;
2361 struct net *net = dev_net(skb->dev);
2365 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2368 * Validate the packet.
2370 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2371 goto drop; /* No space for header. */
2374 ulen = ntohs(uh->len);
2375 saddr = ip_hdr(skb)->saddr;
2376 daddr = ip_hdr(skb)->daddr;
2378 if (ulen > skb->len)
2381 if (proto == IPPROTO_UDP) {
2382 /* UDP validates ulen. */
2383 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2388 if (udp4_csum_init(skb, uh, proto))
2391 sk = skb_steal_sock(skb, &refcounted);
2393 struct dst_entry *dst = skb_dst(skb);
2396 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2397 udp_sk_rx_dst_set(sk, dst);
2399 ret = udp_unicast_rcv_skb(sk, skb, uh);
2405 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2406 return __udp4_lib_mcast_deliver(net, skb, uh,
2407 saddr, daddr, udptable, proto);
2409 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2411 return udp_unicast_rcv_skb(sk, skb, uh);
2413 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2417 /* No socket. Drop packet silently, if checksum is wrong */
2418 if (udp_lib_checksum_complete(skb))
2421 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2422 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2423 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2426 * Hmm. We got an UDP packet to a port to which we
2427 * don't wanna listen. Ignore it.
2429 kfree_skb_reason(skb, drop_reason);
2433 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2434 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2435 proto == IPPROTO_UDPLITE ? "Lite" : "",
2436 &saddr, ntohs(uh->source),
2438 &daddr, ntohs(uh->dest));
2443 * RFC1122: OK. Discards the bad packet silently (as far as
2444 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2446 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2447 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2448 proto == IPPROTO_UDPLITE ? "Lite" : "",
2449 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2451 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2453 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2454 kfree_skb_reason(skb, drop_reason);
2458 /* We can only early demux multicast if there is a single matching socket.
2459 * If more than one socket found returns NULL
2461 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2462 __be16 loc_port, __be32 loc_addr,
2463 __be16 rmt_port, __be32 rmt_addr,
2466 struct udp_table *udptable = net->ipv4.udp_table;
2467 unsigned short hnum = ntohs(loc_port);
2468 struct sock *sk, *result;
2469 struct udp_hslot *hslot;
2472 slot = udp_hashfn(net, hnum, udptable->mask);
2473 hslot = &udptable->hash[slot];
2475 /* Do not bother scanning a too big list */
2476 if (hslot->count > 10)
2480 sk_for_each_rcu(sk, &hslot->head) {
2481 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2482 rmt_port, rmt_addr, dif, sdif, hnum)) {
2492 /* For unicast we should only early demux connected sockets or we can
2493 * break forwarding setups. The chains here can be long so only check
2494 * if the first socket is an exact match and if not move on.
2496 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2497 __be16 loc_port, __be32 loc_addr,
2498 __be16 rmt_port, __be32 rmt_addr,
2501 struct udp_table *udptable = net->ipv4.udp_table;
2502 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2503 unsigned short hnum = ntohs(loc_port);
2504 unsigned int hash2, slot2;
2505 struct udp_hslot *hslot2;
2509 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2510 slot2 = hash2 & udptable->mask;
2511 hslot2 = &udptable->hash2[slot2];
2512 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2514 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2515 if (inet_match(net, sk, acookie, ports, dif, sdif))
2517 /* Only check first socket in chain */
2523 int udp_v4_early_demux(struct sk_buff *skb)
2525 struct net *net = dev_net(skb->dev);
2526 struct in_device *in_dev = NULL;
2527 const struct iphdr *iph;
2528 const struct udphdr *uh;
2529 struct sock *sk = NULL;
2530 struct dst_entry *dst;
2531 int dif = skb->dev->ifindex;
2532 int sdif = inet_sdif(skb);
2535 /* validate the packet */
2536 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2542 if (skb->pkt_type == PACKET_MULTICAST) {
2543 in_dev = __in_dev_get_rcu(skb->dev);
2548 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2553 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2554 uh->source, iph->saddr,
2556 } else if (skb->pkt_type == PACKET_HOST) {
2557 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2558 uh->source, iph->saddr, dif, sdif);
2561 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2565 skb->destructor = sock_efree;
2566 dst = rcu_dereference(sk->sk_rx_dst);
2569 dst = dst_check(dst, 0);
2573 /* set noref for now.
2574 * any place which wants to hold dst has to call
2577 skb_dst_set_noref(skb, dst);
2579 /* for unconnected multicast sockets we need to validate
2580 * the source on each packet
2582 if (!inet_sk(sk)->inet_daddr && in_dev)
2583 return ip_mc_validate_source(skb, iph->daddr,
2585 iph->tos & IPTOS_RT_MASK,
2586 skb->dev, in_dev, &itag);
2591 int udp_rcv(struct sk_buff *skb)
2593 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2596 void udp_destroy_sock(struct sock *sk)
2598 struct udp_sock *up = udp_sk(sk);
2599 bool slow = lock_sock_fast(sk);
2601 /* protects from races with udp_abort() */
2602 sock_set_flag(sk, SOCK_DEAD);
2603 udp_flush_pending_frames(sk);
2604 unlock_sock_fast(sk, slow);
2605 if (static_branch_unlikely(&udp_encap_needed_key)) {
2606 if (up->encap_type) {
2607 void (*encap_destroy)(struct sock *sk);
2608 encap_destroy = READ_ONCE(up->encap_destroy);
2612 if (up->encap_enabled)
2613 static_branch_dec(&udp_encap_needed_key);
2618 * Socket option code for UDP
2620 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2621 sockptr_t optval, unsigned int optlen,
2622 int (*push_pending_frames)(struct sock *))
2624 struct udp_sock *up = udp_sk(sk);
2627 int is_udplite = IS_UDPLITE(sk);
2629 if (level == SOL_SOCKET) {
2630 err = sk_setsockopt(sk, level, optname, optval, optlen);
2632 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2633 sockopt_lock_sock(sk);
2634 /* paired with READ_ONCE in udp_rmem_release() */
2635 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2636 sockopt_release_sock(sk);
2641 if (optlen < sizeof(int))
2644 if (copy_from_sockptr(&val, optval, sizeof(val)))
2647 valbool = val ? 1 : 0;
2652 WRITE_ONCE(up->corkflag, 1);
2654 WRITE_ONCE(up->corkflag, 0);
2656 push_pending_frames(sk);
2665 case UDP_ENCAP_ESPINUDP:
2666 case UDP_ENCAP_ESPINUDP_NON_IKE:
2667 #if IS_ENABLED(CONFIG_IPV6)
2668 if (sk->sk_family == AF_INET6)
2669 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2672 up->encap_rcv = xfrm4_udp_encap_rcv;
2675 case UDP_ENCAP_L2TPINUDP:
2676 up->encap_type = val;
2678 udp_tunnel_encap_enable(sk->sk_socket);
2687 case UDP_NO_CHECK6_TX:
2688 up->no_check6_tx = valbool;
2691 case UDP_NO_CHECK6_RX:
2692 up->no_check6_rx = valbool;
2696 if (val < 0 || val > USHRT_MAX)
2698 WRITE_ONCE(up->gso_size, val);
2704 /* when enabling GRO, accept the related GSO packet type */
2706 udp_tunnel_encap_enable(sk->sk_socket);
2707 up->gro_enabled = valbool;
2708 up->accept_udp_l4 = valbool;
2713 * UDP-Lite's partial checksum coverage (RFC 3828).
2715 /* The sender sets actual checksum coverage length via this option.
2716 * The case coverage > packet length is handled by send module. */
2717 case UDPLITE_SEND_CSCOV:
2718 if (!is_udplite) /* Disable the option on UDP sockets */
2719 return -ENOPROTOOPT;
2720 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2722 else if (val > USHRT_MAX)
2725 up->pcflag |= UDPLITE_SEND_CC;
2728 /* The receiver specifies a minimum checksum coverage value. To make
2729 * sense, this should be set to at least 8 (as done below). If zero is
2730 * used, this again means full checksum coverage. */
2731 case UDPLITE_RECV_CSCOV:
2732 if (!is_udplite) /* Disable the option on UDP sockets */
2733 return -ENOPROTOOPT;
2734 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2736 else if (val > USHRT_MAX)
2739 up->pcflag |= UDPLITE_RECV_CC;
2749 EXPORT_SYMBOL(udp_lib_setsockopt);
2751 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2752 unsigned int optlen)
2754 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2755 return udp_lib_setsockopt(sk, level, optname,
2757 udp_push_pending_frames);
2758 return ip_setsockopt(sk, level, optname, optval, optlen);
2761 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2762 char __user *optval, int __user *optlen)
2764 struct udp_sock *up = udp_sk(sk);
2767 if (get_user(len, optlen))
2770 len = min_t(unsigned int, len, sizeof(int));
2777 val = READ_ONCE(up->corkflag);
2781 val = up->encap_type;
2784 case UDP_NO_CHECK6_TX:
2785 val = up->no_check6_tx;
2788 case UDP_NO_CHECK6_RX:
2789 val = up->no_check6_rx;
2793 val = READ_ONCE(up->gso_size);
2797 val = up->gro_enabled;
2800 /* The following two cannot be changed on UDP sockets, the return is
2801 * always 0 (which corresponds to the full checksum coverage of UDP). */
2802 case UDPLITE_SEND_CSCOV:
2806 case UDPLITE_RECV_CSCOV:
2811 return -ENOPROTOOPT;
2814 if (put_user(len, optlen))
2816 if (copy_to_user(optval, &val, len))
2820 EXPORT_SYMBOL(udp_lib_getsockopt);
2822 int udp_getsockopt(struct sock *sk, int level, int optname,
2823 char __user *optval, int __user *optlen)
2825 if (level == SOL_UDP || level == SOL_UDPLITE)
2826 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2827 return ip_getsockopt(sk, level, optname, optval, optlen);
2831 * udp_poll - wait for a UDP event.
2832 * @file: - file struct
2834 * @wait: - poll table
2836 * This is same as datagram poll, except for the special case of
2837 * blocking sockets. If application is using a blocking fd
2838 * and a packet with checksum error is in the queue;
2839 * then it could get return from select indicating data available
2840 * but then block when reading it. Add special case code
2841 * to work around these arguably broken applications.
2843 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2845 __poll_t mask = datagram_poll(file, sock, wait);
2846 struct sock *sk = sock->sk;
2848 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2849 mask |= EPOLLIN | EPOLLRDNORM;
2851 /* Check for false positives due to checksum errors */
2852 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2853 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2854 mask &= ~(EPOLLIN | EPOLLRDNORM);
2856 /* psock ingress_msg queue should not contain any bad checksum frames */
2857 if (sk_is_readable(sk))
2858 mask |= EPOLLIN | EPOLLRDNORM;
2862 EXPORT_SYMBOL(udp_poll);
2864 int udp_abort(struct sock *sk, int err)
2866 if (!has_current_bpf_ctx())
2869 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2872 if (sock_flag(sk, SOCK_DEAD))
2876 sk_error_report(sk);
2877 __udp_disconnect(sk, 0);
2880 if (!has_current_bpf_ctx())
2885 EXPORT_SYMBOL_GPL(udp_abort);
2887 struct proto udp_prot = {
2889 .owner = THIS_MODULE,
2890 .close = udp_lib_close,
2891 .pre_connect = udp_pre_connect,
2892 .connect = ip4_datagram_connect,
2893 .disconnect = udp_disconnect,
2895 .init = udp_init_sock,
2896 .destroy = udp_destroy_sock,
2897 .setsockopt = udp_setsockopt,
2898 .getsockopt = udp_getsockopt,
2899 .sendmsg = udp_sendmsg,
2900 .recvmsg = udp_recvmsg,
2901 .splice_eof = udp_splice_eof,
2902 .release_cb = ip4_datagram_release_cb,
2903 .hash = udp_lib_hash,
2904 .unhash = udp_lib_unhash,
2905 .rehash = udp_v4_rehash,
2906 .get_port = udp_v4_get_port,
2907 .put_port = udp_lib_unhash,
2908 #ifdef CONFIG_BPF_SYSCALL
2909 .psock_update_sk_prot = udp_bpf_update_proto,
2911 .memory_allocated = &udp_memory_allocated,
2912 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2914 .sysctl_mem = sysctl_udp_mem,
2915 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2916 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2917 .obj_size = sizeof(struct udp_sock),
2918 .h.udp_table = NULL,
2919 .diag_destroy = udp_abort,
2921 EXPORT_SYMBOL(udp_prot);
2923 /* ------------------------------------------------------------------------ */
2924 #ifdef CONFIG_PROC_FS
2926 static unsigned short seq_file_family(const struct seq_file *seq);
2927 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2929 unsigned short family = seq_file_family(seq);
2931 /* AF_UNSPEC is used as a match all */
2932 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2933 net_eq(sock_net(sk), seq_file_net(seq)));
2936 #ifdef CONFIG_BPF_SYSCALL
2937 static const struct seq_operations bpf_iter_udp_seq_ops;
2939 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2942 const struct udp_seq_afinfo *afinfo;
2944 #ifdef CONFIG_BPF_SYSCALL
2945 if (seq->op == &bpf_iter_udp_seq_ops)
2946 return net->ipv4.udp_table;
2949 afinfo = pde_data(file_inode(seq->file));
2950 return afinfo->udp_table ? : net->ipv4.udp_table;
2953 static struct sock *udp_get_first(struct seq_file *seq, int start)
2955 struct udp_iter_state *state = seq->private;
2956 struct net *net = seq_file_net(seq);
2957 struct udp_table *udptable;
2960 udptable = udp_get_table_seq(seq, net);
2962 for (state->bucket = start; state->bucket <= udptable->mask;
2964 struct udp_hslot *hslot = &udptable->hash[state->bucket];
2966 if (hlist_empty(&hslot->head))
2969 spin_lock_bh(&hslot->lock);
2970 sk_for_each(sk, &hslot->head) {
2971 if (seq_sk_match(seq, sk))
2974 spin_unlock_bh(&hslot->lock);
2981 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2983 struct udp_iter_state *state = seq->private;
2984 struct net *net = seq_file_net(seq);
2985 struct udp_table *udptable;
2989 } while (sk && !seq_sk_match(seq, sk));
2992 udptable = udp_get_table_seq(seq, net);
2994 if (state->bucket <= udptable->mask)
2995 spin_unlock_bh(&udptable->hash[state->bucket].lock);
2997 return udp_get_first(seq, state->bucket + 1);
3002 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3004 struct sock *sk = udp_get_first(seq, 0);
3007 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3009 return pos ? NULL : sk;
3012 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3014 struct udp_iter_state *state = seq->private;
3015 state->bucket = MAX_UDP_PORTS;
3017 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3019 EXPORT_SYMBOL(udp_seq_start);
3021 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3025 if (v == SEQ_START_TOKEN)
3026 sk = udp_get_idx(seq, 0);
3028 sk = udp_get_next(seq, v);
3033 EXPORT_SYMBOL(udp_seq_next);
3035 void udp_seq_stop(struct seq_file *seq, void *v)
3037 struct udp_iter_state *state = seq->private;
3038 struct udp_table *udptable;
3040 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3042 if (state->bucket <= udptable->mask)
3043 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3045 EXPORT_SYMBOL(udp_seq_stop);
3047 /* ------------------------------------------------------------------------ */
3048 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3051 struct inet_sock *inet = inet_sk(sp);
3052 __be32 dest = inet->inet_daddr;
3053 __be32 src = inet->inet_rcv_saddr;
3054 __u16 destp = ntohs(inet->inet_dport);
3055 __u16 srcp = ntohs(inet->inet_sport);
3057 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3058 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3059 bucket, src, srcp, dest, destp, sp->sk_state,
3060 sk_wmem_alloc_get(sp),
3063 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3065 refcount_read(&sp->sk_refcnt), sp,
3066 atomic_read(&sp->sk_drops));
3069 int udp4_seq_show(struct seq_file *seq, void *v)
3071 seq_setwidth(seq, 127);
3072 if (v == SEQ_START_TOKEN)
3073 seq_puts(seq, " sl local_address rem_address st tx_queue "
3074 "rx_queue tr tm->when retrnsmt uid timeout "
3075 "inode ref pointer drops");
3077 struct udp_iter_state *state = seq->private;
3079 udp4_format_sock(v, seq, state->bucket);
3085 #ifdef CONFIG_BPF_SYSCALL
3086 struct bpf_iter__udp {
3087 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3088 __bpf_md_ptr(struct udp_sock *, udp_sk);
3089 uid_t uid __aligned(8);
3090 int bucket __aligned(8);
3093 struct bpf_udp_iter_state {
3094 struct udp_iter_state state;
3095 unsigned int cur_sk;
3096 unsigned int end_sk;
3097 unsigned int max_sk;
3099 struct sock **batch;
3100 bool st_bucket_done;
3103 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3104 unsigned int new_batch_sz);
3105 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3107 struct bpf_udp_iter_state *iter = seq->private;
3108 struct udp_iter_state *state = &iter->state;
3109 struct net *net = seq_file_net(seq);
3110 struct udp_table *udptable;
3111 unsigned int batch_sks = 0;
3112 bool resized = false;
3115 /* The current batch is done, so advance the bucket. */
3116 if (iter->st_bucket_done) {
3121 udptable = udp_get_table_seq(seq, net);
3124 /* New batch for the next bucket.
3125 * Iterate over the hash table to find a bucket with sockets matching
3126 * the iterator attributes, and return the first matching socket from
3127 * the bucket. The remaining matched sockets from the bucket are batched
3128 * before releasing the bucket lock. This allows BPF programs that are
3129 * called in seq_show to acquire the bucket lock if needed.
3133 iter->st_bucket_done = false;
3136 for (; state->bucket <= udptable->mask; state->bucket++) {
3137 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3139 if (hlist_empty(&hslot2->head)) {
3144 spin_lock_bh(&hslot2->lock);
3145 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3146 if (seq_sk_match(seq, sk)) {
3147 /* Resume from the last iterated socket at the
3148 * offset in the bucket before iterator was stopped.
3154 if (iter->end_sk < iter->max_sk) {
3156 iter->batch[iter->end_sk++] = sk;
3161 spin_unlock_bh(&hslot2->lock);
3166 /* Reset the current bucket's offset before moving to the next bucket. */
3170 /* All done: no batch made. */
3174 if (iter->end_sk == batch_sks) {
3175 /* Batching is done for the current bucket; return the first
3176 * socket to be iterated from the batch.
3178 iter->st_bucket_done = true;
3181 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3183 /* After allocating a larger batch, retry one more time to grab
3190 return iter->batch[0];
3193 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3195 struct bpf_udp_iter_state *iter = seq->private;
3198 /* Whenever seq_next() is called, the iter->cur_sk is
3199 * done with seq_show(), so unref the iter->cur_sk.
3201 if (iter->cur_sk < iter->end_sk) {
3202 sock_put(iter->batch[iter->cur_sk++]);
3206 /* After updating iter->cur_sk, check if there are more sockets
3207 * available in the current bucket batch.
3209 if (iter->cur_sk < iter->end_sk)
3210 sk = iter->batch[iter->cur_sk];
3212 /* Prepare a new batch. */
3213 sk = bpf_iter_udp_batch(seq);
3219 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3221 /* bpf iter does not support lseek, so it always
3222 * continue from where it was stop()-ped.
3225 return bpf_iter_udp_batch(seq);
3227 return SEQ_START_TOKEN;
3230 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3231 struct udp_sock *udp_sk, uid_t uid, int bucket)
3233 struct bpf_iter__udp ctx;
3235 meta->seq_num--; /* skip SEQ_START_TOKEN */
3237 ctx.udp_sk = udp_sk;
3239 ctx.bucket = bucket;
3240 return bpf_iter_run_prog(prog, &ctx);
3243 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3245 struct udp_iter_state *state = seq->private;
3246 struct bpf_iter_meta meta;
3247 struct bpf_prog *prog;
3248 struct sock *sk = v;
3252 if (v == SEQ_START_TOKEN)
3257 if (unlikely(sk_unhashed(sk))) {
3262 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3264 prog = bpf_iter_get_info(&meta, false);
3265 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3272 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3274 while (iter->cur_sk < iter->end_sk)
3275 sock_put(iter->batch[iter->cur_sk++]);
3278 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3280 struct bpf_udp_iter_state *iter = seq->private;
3281 struct bpf_iter_meta meta;
3282 struct bpf_prog *prog;
3286 prog = bpf_iter_get_info(&meta, true);
3288 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3291 if (iter->cur_sk < iter->end_sk) {
3292 bpf_iter_udp_put_batch(iter);
3293 iter->st_bucket_done = false;
3297 static const struct seq_operations bpf_iter_udp_seq_ops = {
3298 .start = bpf_iter_udp_seq_start,
3299 .next = bpf_iter_udp_seq_next,
3300 .stop = bpf_iter_udp_seq_stop,
3301 .show = bpf_iter_udp_seq_show,
3305 static unsigned short seq_file_family(const struct seq_file *seq)
3307 const struct udp_seq_afinfo *afinfo;
3309 #ifdef CONFIG_BPF_SYSCALL
3310 /* BPF iterator: bpf programs to filter sockets. */
3311 if (seq->op == &bpf_iter_udp_seq_ops)
3315 /* Proc fs iterator */
3316 afinfo = pde_data(file_inode(seq->file));
3317 return afinfo->family;
3320 const struct seq_operations udp_seq_ops = {
3321 .start = udp_seq_start,
3322 .next = udp_seq_next,
3323 .stop = udp_seq_stop,
3324 .show = udp4_seq_show,
3326 EXPORT_SYMBOL(udp_seq_ops);
3328 static struct udp_seq_afinfo udp4_seq_afinfo = {
3333 static int __net_init udp4_proc_init_net(struct net *net)
3335 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3336 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3341 static void __net_exit udp4_proc_exit_net(struct net *net)
3343 remove_proc_entry("udp", net->proc_net);
3346 static struct pernet_operations udp4_net_ops = {
3347 .init = udp4_proc_init_net,
3348 .exit = udp4_proc_exit_net,
3351 int __init udp4_proc_init(void)
3353 return register_pernet_subsys(&udp4_net_ops);
3356 void udp4_proc_exit(void)
3358 unregister_pernet_subsys(&udp4_net_ops);
3360 #endif /* CONFIG_PROC_FS */
3362 static __initdata unsigned long uhash_entries;
3363 static int __init set_uhash_entries(char *str)
3370 ret = kstrtoul(str, 0, &uhash_entries);
3374 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3375 uhash_entries = UDP_HTABLE_SIZE_MIN;
3378 __setup("uhash_entries=", set_uhash_entries);
3380 void __init udp_table_init(struct udp_table *table, const char *name)
3384 table->hash = alloc_large_system_hash(name,
3385 2 * sizeof(struct udp_hslot),
3387 21, /* one slot per 2 MB */
3391 UDP_HTABLE_SIZE_MIN,
3392 UDP_HTABLE_SIZE_MAX);
3394 table->hash2 = table->hash + (table->mask + 1);
3395 for (i = 0; i <= table->mask; i++) {
3396 INIT_HLIST_HEAD(&table->hash[i].head);
3397 table->hash[i].count = 0;
3398 spin_lock_init(&table->hash[i].lock);
3400 for (i = 0; i <= table->mask; i++) {
3401 INIT_HLIST_HEAD(&table->hash2[i].head);
3402 table->hash2[i].count = 0;
3403 spin_lock_init(&table->hash2[i].lock);
3407 u32 udp_flow_hashrnd(void)
3409 static u32 hashrnd __read_mostly;
3411 net_get_random_once(&hashrnd, sizeof(hashrnd));
3415 EXPORT_SYMBOL(udp_flow_hashrnd);
3417 static void __net_init udp_sysctl_init(struct net *net)
3419 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3420 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3422 #ifdef CONFIG_NET_L3_MASTER_DEV
3423 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3427 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3429 struct udp_table *udptable;
3432 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3436 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3437 GFP_KERNEL_ACCOUNT);
3438 if (!udptable->hash)
3441 udptable->hash2 = udptable->hash + hash_entries;
3442 udptable->mask = hash_entries - 1;
3443 udptable->log = ilog2(hash_entries);
3445 for (i = 0; i < hash_entries; i++) {
3446 INIT_HLIST_HEAD(&udptable->hash[i].head);
3447 udptable->hash[i].count = 0;
3448 spin_lock_init(&udptable->hash[i].lock);
3450 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3451 udptable->hash2[i].count = 0;
3452 spin_lock_init(&udptable->hash2[i].lock);
3463 static void __net_exit udp_pernet_table_free(struct net *net)
3465 struct udp_table *udptable = net->ipv4.udp_table;
3467 if (udptable == &udp_table)
3470 kvfree(udptable->hash);
3474 static void __net_init udp_set_table(struct net *net)
3476 struct udp_table *udptable;
3477 unsigned int hash_entries;
3478 struct net *old_net;
3480 if (net_eq(net, &init_net))
3483 old_net = current->nsproxy->net_ns;
3484 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3488 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3489 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3490 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3492 hash_entries = roundup_pow_of_two(hash_entries);
3494 udptable = udp_pernet_table_alloc(hash_entries);
3496 net->ipv4.udp_table = udptable;
3498 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3499 "for a netns, fallback to the global one\n",
3502 net->ipv4.udp_table = &udp_table;
3506 static int __net_init udp_pernet_init(struct net *net)
3508 udp_sysctl_init(net);
3514 static void __net_exit udp_pernet_exit(struct net *net)
3516 udp_pernet_table_free(net);
3519 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3520 .init = udp_pernet_init,
3521 .exit = udp_pernet_exit,
3524 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3525 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3526 struct udp_sock *udp_sk, uid_t uid, int bucket)
3528 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3529 unsigned int new_batch_sz)
3531 struct sock **new_batch;
3533 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3534 GFP_USER | __GFP_NOWARN);
3538 bpf_iter_udp_put_batch(iter);
3539 kvfree(iter->batch);
3540 iter->batch = new_batch;
3541 iter->max_sk = new_batch_sz;
3546 #define INIT_BATCH_SZ 16
3548 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3550 struct bpf_udp_iter_state *iter = priv_data;
3553 ret = bpf_iter_init_seq_net(priv_data, aux);
3557 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3559 bpf_iter_fini_seq_net(priv_data);
3564 static void bpf_iter_fini_udp(void *priv_data)
3566 struct bpf_udp_iter_state *iter = priv_data;
3568 bpf_iter_fini_seq_net(priv_data);
3569 kvfree(iter->batch);
3572 static const struct bpf_iter_seq_info udp_seq_info = {
3573 .seq_ops = &bpf_iter_udp_seq_ops,
3574 .init_seq_private = bpf_iter_init_udp,
3575 .fini_seq_private = bpf_iter_fini_udp,
3576 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3579 static struct bpf_iter_reg udp_reg_info = {
3581 .ctx_arg_info_size = 1,
3583 { offsetof(struct bpf_iter__udp, udp_sk),
3584 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3586 .seq_info = &udp_seq_info,
3589 static void __init bpf_iter_register(void)
3591 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3592 if (bpf_iter_reg_target(&udp_reg_info))
3593 pr_warn("Warning: could not register bpf iterator udp\n");
3597 void __init udp_init(void)
3599 unsigned long limit;
3602 udp_table_init(&udp_table, "UDP");
3603 limit = nr_free_buffer_pages() / 8;
3604 limit = max(limit, 128UL);
3605 sysctl_udp_mem[0] = limit / 4 * 3;
3606 sysctl_udp_mem[1] = limit;
3607 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3609 /* 16 spinlocks per cpu */
3610 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3611 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3614 panic("UDP: failed to alloc udp_busylocks\n");
3615 for (i = 0; i < (1U << udp_busylocks_log); i++)
3616 spin_lock_init(udp_busylocks + i);
3618 if (register_pernet_subsys(&udp_sysctl_ops))
3619 panic("UDP: failed to init sysctl parameters.\n");
3621 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3622 bpf_iter_register();