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>
118 #if IS_ENABLED(CONFIG_IPV6)
119 #include <net/ipv6_stubs.h>
122 struct udp_table udp_table __read_mostly;
123 EXPORT_SYMBOL(udp_table);
125 long sysctl_udp_mem[3] __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_mem);
128 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
129 EXPORT_SYMBOL(udp_memory_allocated);
130 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
131 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
133 #define MAX_UDP_PORTS 65536
134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
136 static struct udp_table *udp_get_table_prot(struct sock *sk)
138 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
141 static int udp_lib_lport_inuse(struct net *net, __u16 num,
142 const struct udp_hslot *hslot,
143 unsigned long *bitmap,
144 struct sock *sk, unsigned int log)
147 kuid_t uid = sock_i_uid(sk);
149 sk_for_each(sk2, &hslot->head) {
150 if (net_eq(sock_net(sk2), net) &&
152 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
153 (!sk2->sk_reuse || !sk->sk_reuse) &&
154 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
155 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
156 inet_rcv_saddr_equal(sk, sk2, true)) {
157 if (sk2->sk_reuseport && sk->sk_reuseport &&
158 !rcu_access_pointer(sk->sk_reuseport_cb) &&
159 uid_eq(uid, sock_i_uid(sk2))) {
165 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
174 * Note: we still hold spinlock of primary hash chain, so no other writer
175 * can insert/delete a socket with local_port == num
177 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
178 struct udp_hslot *hslot2,
182 kuid_t uid = sock_i_uid(sk);
185 spin_lock(&hslot2->lock);
186 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
187 if (net_eq(sock_net(sk2), net) &&
189 (udp_sk(sk2)->udp_port_hash == num) &&
190 (!sk2->sk_reuse || !sk->sk_reuse) &&
191 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
192 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
193 inet_rcv_saddr_equal(sk, sk2, true)) {
194 if (sk2->sk_reuseport && sk->sk_reuseport &&
195 !rcu_access_pointer(sk->sk_reuseport_cb) &&
196 uid_eq(uid, sock_i_uid(sk2))) {
204 spin_unlock(&hslot2->lock);
208 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
210 struct net *net = sock_net(sk);
211 kuid_t uid = sock_i_uid(sk);
214 sk_for_each(sk2, &hslot->head) {
215 if (net_eq(sock_net(sk2), net) &&
217 sk2->sk_family == sk->sk_family &&
218 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
219 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
220 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
221 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
222 inet_rcv_saddr_equal(sk, sk2, false)) {
223 return reuseport_add_sock(sk, sk2,
224 inet_rcv_saddr_any(sk));
228 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
234 * @sk: socket struct in question
235 * @snum: port number to look up
236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
239 int udp_lib_get_port(struct sock *sk, unsigned short snum,
240 unsigned int hash2_nulladdr)
242 struct udp_table *udptable = udp_get_table_prot(sk);
243 struct udp_hslot *hslot, *hslot2;
244 struct net *net = sock_net(sk);
245 int error = -EADDRINUSE;
248 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
249 unsigned short first, last;
250 int low, high, remaining;
253 inet_sk_get_local_port_range(sk, &low, &high);
254 remaining = (high - low) + 1;
256 rand = get_random_u32();
257 first = reciprocal_scale(rand, remaining) + low;
259 * force rand to be an odd multiple of UDP_HTABLE_SIZE
261 rand = (rand | 1) * (udptable->mask + 1);
262 last = first + udptable->mask + 1;
264 hslot = udp_hashslot(udptable, net, first);
265 bitmap_zero(bitmap, PORTS_PER_CHAIN);
266 spin_lock_bh(&hslot->lock);
267 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
272 * Iterate on all possible values of snum for this hash.
273 * Using steps of an odd multiple of UDP_HTABLE_SIZE
274 * give us randomization and full range coverage.
277 if (low <= snum && snum <= high &&
278 !test_bit(snum >> udptable->log, bitmap) &&
279 !inet_is_local_reserved_port(net, snum))
282 } while (snum != first);
283 spin_unlock_bh(&hslot->lock);
285 } while (++first != last);
288 hslot = udp_hashslot(udptable, net, snum);
289 spin_lock_bh(&hslot->lock);
290 if (hslot->count > 10) {
292 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
294 slot2 &= udptable->mask;
295 hash2_nulladdr &= udptable->mask;
297 hslot2 = udp_hashslot2(udptable, slot2);
298 if (hslot->count < hslot2->count)
299 goto scan_primary_hash;
301 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
302 if (!exist && (hash2_nulladdr != slot2)) {
303 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
304 exist = udp_lib_lport_inuse2(net, snum, hslot2,
313 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
317 inet_sk(sk)->inet_num = snum;
318 udp_sk(sk)->udp_port_hash = snum;
319 udp_sk(sk)->udp_portaddr_hash ^= snum;
320 if (sk_unhashed(sk)) {
321 if (sk->sk_reuseport &&
322 udp_reuseport_add_sock(sk, hslot)) {
323 inet_sk(sk)->inet_num = 0;
324 udp_sk(sk)->udp_port_hash = 0;
325 udp_sk(sk)->udp_portaddr_hash ^= snum;
329 sk_add_node_rcu(sk, &hslot->head);
331 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
333 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
334 spin_lock(&hslot2->lock);
335 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
336 sk->sk_family == AF_INET6)
337 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
340 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
343 spin_unlock(&hslot2->lock);
345 sock_set_flag(sk, SOCK_RCU_FREE);
348 spin_unlock_bh(&hslot->lock);
352 EXPORT_SYMBOL(udp_lib_get_port);
354 int udp_v4_get_port(struct sock *sk, unsigned short snum)
356 unsigned int hash2_nulladdr =
357 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
358 unsigned int hash2_partial =
359 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
361 /* precompute partial secondary hash */
362 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
363 return udp_lib_get_port(sk, snum, hash2_nulladdr);
366 static int compute_score(struct sock *sk, struct net *net,
367 __be32 saddr, __be16 sport,
368 __be32 daddr, unsigned short hnum,
372 struct inet_sock *inet;
375 if (!net_eq(sock_net(sk), net) ||
376 udp_sk(sk)->udp_port_hash != hnum ||
380 if (sk->sk_rcv_saddr != daddr)
383 score = (sk->sk_family == PF_INET) ? 2 : 1;
386 if (inet->inet_daddr) {
387 if (inet->inet_daddr != saddr)
392 if (inet->inet_dport) {
393 if (inet->inet_dport != sport)
398 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
402 if (sk->sk_bound_dev_if)
405 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
410 INDIRECT_CALLABLE_SCOPE
411 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
412 const __be32 faddr, const __be16 fport)
414 static u32 udp_ehash_secret __read_mostly;
416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
418 return __inet_ehashfn(laddr, lport, faddr, fport,
419 udp_ehash_secret + net_hash_mix(net));
422 /* called with rcu_read_lock() */
423 static struct sock *udp4_lib_lookup2(struct net *net,
424 __be32 saddr, __be16 sport,
425 __be32 daddr, unsigned int hnum,
427 struct udp_hslot *hslot2,
430 struct sock *sk, *result;
435 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
436 score = compute_score(sk, net, saddr, sport,
437 daddr, hnum, dif, sdif);
438 if (score > badness) {
441 if (sk->sk_state == TCP_ESTABLISHED) {
446 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
447 saddr, sport, daddr, hnum, udp_ehashfn);
453 /* Fall back to scoring if group has connections */
454 if (!reuseport_has_conns(sk))
457 /* Reuseport logic returned an error, keep original score. */
461 badness = compute_score(result, net, saddr, sport,
462 daddr, hnum, dif, sdif);
469 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
470 * harder than this. -DaveM
472 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
473 __be16 sport, __be32 daddr, __be16 dport, int dif,
474 int sdif, struct udp_table *udptable, struct sk_buff *skb)
476 unsigned short hnum = ntohs(dport);
477 unsigned int hash2, slot2;
478 struct udp_hslot *hslot2;
479 struct sock *result, *sk;
481 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
482 slot2 = hash2 & udptable->mask;
483 hslot2 = &udptable->hash2[slot2];
485 /* Lookup connected or non-wildcard socket */
486 result = udp4_lib_lookup2(net, saddr, sport,
487 daddr, hnum, dif, sdif,
489 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
492 /* Lookup redirect from BPF */
493 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
494 udptable == net->ipv4.udp_table) {
495 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
496 saddr, sport, daddr, hnum, dif,
504 /* Got non-wildcard socket or error on first lookup */
508 /* Lookup wildcard sockets */
509 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
510 slot2 = hash2 & udptable->mask;
511 hslot2 = &udptable->hash2[slot2];
513 result = udp4_lib_lookup2(net, saddr, sport,
514 htonl(INADDR_ANY), hnum, dif, sdif,
521 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
523 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
524 __be16 sport, __be16 dport,
525 struct udp_table *udptable)
527 const struct iphdr *iph = ip_hdr(skb);
529 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
530 iph->daddr, dport, inet_iif(skb),
531 inet_sdif(skb), udptable, skb);
534 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
535 __be16 sport, __be16 dport)
537 const struct iphdr *iph = ip_hdr(skb);
538 struct net *net = dev_net(skb->dev);
541 inet_get_iif_sdif(skb, &iif, &sdif);
543 return __udp4_lib_lookup(net, iph->saddr, sport,
544 iph->daddr, dport, iif,
545 sdif, net->ipv4.udp_table, NULL);
548 /* Must be called under rcu_read_lock().
549 * Does increment socket refcount.
551 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
552 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
553 __be32 daddr, __be16 dport, int dif)
557 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
558 dif, 0, net->ipv4.udp_table, NULL);
559 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
563 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
566 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
567 __be16 loc_port, __be32 loc_addr,
568 __be16 rmt_port, __be32 rmt_addr,
569 int dif, int sdif, unsigned short hnum)
571 const struct inet_sock *inet = inet_sk(sk);
573 if (!net_eq(sock_net(sk), net) ||
574 udp_sk(sk)->udp_port_hash != hnum ||
575 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
576 (inet->inet_dport != rmt_port && inet->inet_dport) ||
577 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
578 ipv6_only_sock(sk) ||
579 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
581 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
586 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
587 EXPORT_SYMBOL(udp_encap_needed_key);
589 #if IS_ENABLED(CONFIG_IPV6)
590 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
591 EXPORT_SYMBOL(udpv6_encap_needed_key);
594 void udp_encap_enable(void)
596 static_branch_inc(&udp_encap_needed_key);
598 EXPORT_SYMBOL(udp_encap_enable);
600 void udp_encap_disable(void)
602 static_branch_dec(&udp_encap_needed_key);
604 EXPORT_SYMBOL(udp_encap_disable);
606 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
607 * through error handlers in encapsulations looking for a match.
609 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
613 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
614 int (*handler)(struct sk_buff *skb, u32 info);
615 const struct ip_tunnel_encap_ops *encap;
617 encap = rcu_dereference(iptun_encaps[i]);
620 handler = encap->err_handler;
621 if (handler && !handler(skb, info))
628 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
629 * reversing source and destination port: this will match tunnels that force the
630 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
631 * lwtunnels might actually break this assumption by being configured with
632 * different destination ports on endpoints, in this case we won't be able to
633 * trace ICMP messages back to them.
635 * If this doesn't match any socket, probe tunnels with arbitrary destination
636 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
637 * we've sent packets to won't necessarily match the local destination port.
639 * Then ask the tunnel implementation to match the error against a valid
642 * Return an error if we can't find a match, the socket if we need further
643 * processing, zero otherwise.
645 static struct sock *__udp4_lib_err_encap(struct net *net,
646 const struct iphdr *iph,
648 struct udp_table *udptable,
650 struct sk_buff *skb, u32 info)
652 int (*lookup)(struct sock *sk, struct sk_buff *skb);
653 int network_offset, transport_offset;
656 network_offset = skb_network_offset(skb);
657 transport_offset = skb_transport_offset(skb);
659 /* Network header needs to point to the outer IPv4 header inside ICMP */
660 skb_reset_network_header(skb);
662 /* Transport header needs to point to the UDP header */
663 skb_set_transport_header(skb, iph->ihl << 2);
668 lookup = READ_ONCE(up->encap_err_lookup);
669 if (lookup && lookup(sk, skb))
675 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
676 iph->saddr, uh->dest, skb->dev->ifindex, 0,
681 lookup = READ_ONCE(up->encap_err_lookup);
682 if (!lookup || lookup(sk, skb))
688 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
690 skb_set_transport_header(skb, transport_offset);
691 skb_set_network_header(skb, network_offset);
697 * This routine is called by the ICMP module when it gets some
698 * sort of error condition. If err < 0 then the socket should
699 * be closed and the error returned to the user. If err > 0
700 * it's just the icmp type << 8 | icmp code.
701 * Header points to the ip header of the error packet. We move
702 * on past this. Then (as it used to claim before adjustment)
703 * header points to the first 8 bytes of the udp header. We need
704 * to find the appropriate port.
707 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
709 struct inet_sock *inet;
710 const struct iphdr *iph = (const struct iphdr *)skb->data;
711 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
712 const int type = icmp_hdr(skb)->type;
713 const int code = icmp_hdr(skb)->code;
718 struct net *net = dev_net(skb->dev);
720 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
721 iph->saddr, uh->source, skb->dev->ifindex,
722 inet_sdif(skb), udptable, NULL);
724 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
725 /* No socket for error: try tunnels before discarding */
726 if (static_branch_unlikely(&udp_encap_needed_key)) {
727 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
732 sk = ERR_PTR(-ENOENT);
735 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
748 case ICMP_TIME_EXCEEDED:
751 case ICMP_SOURCE_QUENCH:
753 case ICMP_PARAMETERPROB:
757 case ICMP_DEST_UNREACH:
758 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
759 ipv4_sk_update_pmtu(skb, sk, info);
760 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
768 if (code <= NR_ICMP_UNREACH) {
769 harderr = icmp_err_convert[code].fatal;
770 err = icmp_err_convert[code].errno;
774 ipv4_sk_redirect(skb, sk);
779 * RFC1122: OK. Passes ICMP errors back to application, as per
783 /* ...not for tunnels though: we don't have a sending socket */
784 if (udp_sk(sk)->encap_err_rcv)
785 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
789 if (!inet_test_bit(RECVERR, sk)) {
790 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
793 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
801 int udp_err(struct sk_buff *skb, u32 info)
803 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
807 * Throw away all pending data and cancel the corking. Socket is locked.
809 void udp_flush_pending_frames(struct sock *sk)
811 struct udp_sock *up = udp_sk(sk);
815 WRITE_ONCE(up->pending, 0);
816 ip_flush_pending_frames(sk);
819 EXPORT_SYMBOL(udp_flush_pending_frames);
822 * udp4_hwcsum - handle outgoing HW checksumming
823 * @skb: sk_buff containing the filled-in UDP header
824 * (checksum field must be zeroed out)
825 * @src: source IP address
826 * @dst: destination IP address
828 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
830 struct udphdr *uh = udp_hdr(skb);
831 int offset = skb_transport_offset(skb);
832 int len = skb->len - offset;
836 if (!skb_has_frag_list(skb)) {
838 * Only one fragment on the socket.
840 skb->csum_start = skb_transport_header(skb) - skb->head;
841 skb->csum_offset = offsetof(struct udphdr, check);
842 uh->check = ~csum_tcpudp_magic(src, dst, len,
845 struct sk_buff *frags;
848 * HW-checksum won't work as there are two or more
849 * fragments on the socket so that all csums of sk_buffs
852 skb_walk_frags(skb, frags) {
853 csum = csum_add(csum, frags->csum);
857 csum = skb_checksum(skb, offset, hlen, csum);
858 skb->ip_summed = CHECKSUM_NONE;
860 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
862 uh->check = CSUM_MANGLED_0;
865 EXPORT_SYMBOL_GPL(udp4_hwcsum);
867 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
868 * for the simple case like when setting the checksum for a UDP tunnel.
870 void udp_set_csum(bool nocheck, struct sk_buff *skb,
871 __be32 saddr, __be32 daddr, int len)
873 struct udphdr *uh = udp_hdr(skb);
877 } else if (skb_is_gso(skb)) {
878 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
879 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
881 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
883 uh->check = CSUM_MANGLED_0;
885 skb->ip_summed = CHECKSUM_PARTIAL;
886 skb->csum_start = skb_transport_header(skb) - skb->head;
887 skb->csum_offset = offsetof(struct udphdr, check);
888 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
891 EXPORT_SYMBOL(udp_set_csum);
893 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
894 struct inet_cork *cork)
896 struct sock *sk = skb->sk;
897 struct inet_sock *inet = inet_sk(sk);
900 int is_udplite = IS_UDPLITE(sk);
901 int offset = skb_transport_offset(skb);
902 int len = skb->len - offset;
903 int datalen = len - sizeof(*uh);
907 * Create a UDP header
910 uh->source = inet->inet_sport;
911 uh->dest = fl4->fl4_dport;
912 uh->len = htons(len);
915 if (cork->gso_size) {
916 const int hlen = skb_network_header_len(skb) +
917 sizeof(struct udphdr);
919 if (hlen + cork->gso_size > cork->fragsize) {
923 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
927 if (sk->sk_no_check_tx) {
931 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
932 dst_xfrm(skb_dst(skb))) {
937 if (datalen > cork->gso_size) {
938 skb_shinfo(skb)->gso_size = cork->gso_size;
939 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
940 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
946 if (is_udplite) /* UDP-Lite */
947 csum = udplite_csum(skb);
949 else if (sk->sk_no_check_tx) { /* UDP csum off */
951 skb->ip_summed = CHECKSUM_NONE;
954 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
957 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
961 csum = udp_csum(skb);
963 /* add protocol-dependent pseudo-header */
964 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
965 sk->sk_protocol, csum);
967 uh->check = CSUM_MANGLED_0;
970 err = ip_send_skb(sock_net(sk), skb);
972 if (err == -ENOBUFS &&
973 !inet_test_bit(RECVERR, sk)) {
974 UDP_INC_STATS(sock_net(sk),
975 UDP_MIB_SNDBUFERRORS, is_udplite);
979 UDP_INC_STATS(sock_net(sk),
980 UDP_MIB_OUTDATAGRAMS, is_udplite);
985 * Push out all pending data as one UDP datagram. Socket is locked.
987 int udp_push_pending_frames(struct sock *sk)
989 struct udp_sock *up = udp_sk(sk);
990 struct inet_sock *inet = inet_sk(sk);
991 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
995 skb = ip_finish_skb(sk, fl4);
999 err = udp_send_skb(skb, fl4, &inet->cork.base);
1003 WRITE_ONCE(up->pending, 0);
1006 EXPORT_SYMBOL(udp_push_pending_frames);
1008 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1010 switch (cmsg->cmsg_type) {
1012 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1014 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1021 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1023 struct cmsghdr *cmsg;
1024 bool need_ip = false;
1027 for_each_cmsghdr(cmsg, msg) {
1028 if (!CMSG_OK(msg, cmsg))
1031 if (cmsg->cmsg_level != SOL_UDP) {
1036 err = __udp_cmsg_send(cmsg, gso_size);
1043 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1045 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1047 struct inet_sock *inet = inet_sk(sk);
1048 struct udp_sock *up = udp_sk(sk);
1049 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1050 struct flowi4 fl4_stack;
1053 struct ipcm_cookie ipc;
1054 struct rtable *rt = NULL;
1057 __be32 daddr, faddr, saddr;
1060 int err, is_udplite = IS_UDPLITE(sk);
1061 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1062 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1063 struct sk_buff *skb;
1064 struct ip_options_data opt_copy;
1074 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1077 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1079 fl4 = &inet->cork.fl.u.ip4;
1080 if (READ_ONCE(up->pending)) {
1082 * There are pending frames.
1083 * The socket lock must be held while it's corked.
1086 if (likely(up->pending)) {
1087 if (unlikely(up->pending != AF_INET)) {
1091 goto do_append_data;
1095 ulen += sizeof(struct udphdr);
1098 * Get and verify the address.
1101 if (msg->msg_namelen < sizeof(*usin))
1103 if (usin->sin_family != AF_INET) {
1104 if (usin->sin_family != AF_UNSPEC)
1105 return -EAFNOSUPPORT;
1108 daddr = usin->sin_addr.s_addr;
1109 dport = usin->sin_port;
1113 if (sk->sk_state != TCP_ESTABLISHED)
1114 return -EDESTADDRREQ;
1115 daddr = inet->inet_daddr;
1116 dport = inet->inet_dport;
1117 /* Open fast path for connected socket.
1118 Route will not be used, if at least one option is set.
1123 ipcm_init_sk(&ipc, inet);
1124 ipc.gso_size = READ_ONCE(up->gso_size);
1126 if (msg->msg_controllen) {
1127 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1129 err = ip_cmsg_send(sk, msg, &ipc,
1130 sk->sk_family == AF_INET6);
1133 if (unlikely(err < 0)) {
1141 struct ip_options_rcu *inet_opt;
1144 inet_opt = rcu_dereference(inet->inet_opt);
1146 memcpy(&opt_copy, inet_opt,
1147 sizeof(*inet_opt) + inet_opt->opt.optlen);
1148 ipc.opt = &opt_copy.opt;
1153 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1154 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1155 (struct sockaddr *)usin,
1161 if (usin->sin_port == 0) {
1162 /* BPF program set invalid port. Reject it. */
1166 daddr = usin->sin_addr.s_addr;
1167 dport = usin->sin_port;
1172 ipc.addr = faddr = daddr;
1174 if (ipc.opt && ipc.opt->opt.srr) {
1179 faddr = ipc.opt->opt.faddr;
1182 tos = get_rttos(&ipc, inet);
1183 scope = ip_sendmsg_scope(inet, &ipc, msg);
1184 if (scope == RT_SCOPE_LINK)
1187 uc_index = READ_ONCE(inet->uc_index);
1188 if (ipv4_is_multicast(daddr)) {
1189 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1190 ipc.oif = READ_ONCE(inet->mc_index);
1192 saddr = READ_ONCE(inet->mc_addr);
1194 } else if (!ipc.oif) {
1196 } else if (ipv4_is_lbcast(daddr) && uc_index) {
1197 /* oif is set, packet is to local broadcast and
1198 * uc_index is set. oif is most likely set
1199 * by sk_bound_dev_if. If uc_index != oif check if the
1200 * oif is an L3 master and uc_index is an L3 slave.
1201 * If so, we want to allow the send using the uc_index.
1203 if (ipc.oif != uc_index &&
1204 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1211 rt = (struct rtable *)sk_dst_check(sk, 0);
1214 struct net *net = sock_net(sk);
1215 __u8 flow_flags = inet_sk_flowi_flags(sk);
1219 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1220 sk->sk_protocol, flow_flags, faddr, saddr,
1221 dport, inet->inet_sport, sk->sk_uid);
1223 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1224 rt = ip_route_output_flow(net, fl4, sk);
1228 if (err == -ENETUNREACH)
1229 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1234 if ((rt->rt_flags & RTCF_BROADCAST) &&
1235 !sock_flag(sk, SOCK_BROADCAST))
1238 sk_dst_set(sk, dst_clone(&rt->dst));
1241 if (msg->msg_flags&MSG_CONFIRM)
1247 daddr = ipc.addr = fl4->daddr;
1249 /* Lockless fast path for the non-corking case. */
1251 struct inet_cork cork;
1253 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1254 sizeof(struct udphdr), &ipc, &rt,
1255 &cork, msg->msg_flags);
1257 if (!IS_ERR_OR_NULL(skb))
1258 err = udp_send_skb(skb, fl4, &cork);
1263 if (unlikely(up->pending)) {
1264 /* The socket is already corked while preparing it. */
1265 /* ... which is an evident application bug. --ANK */
1268 net_dbg_ratelimited("socket already corked\n");
1273 * Now cork the socket to pend data.
1275 fl4 = &inet->cork.fl.u.ip4;
1278 fl4->fl4_dport = dport;
1279 fl4->fl4_sport = inet->inet_sport;
1280 WRITE_ONCE(up->pending, AF_INET);
1284 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1285 sizeof(struct udphdr), &ipc, &rt,
1286 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1288 udp_flush_pending_frames(sk);
1290 err = udp_push_pending_frames(sk);
1291 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1292 WRITE_ONCE(up->pending, 0);
1303 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1304 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1305 * we don't have a good statistic (IpOutDiscards but it can be too many
1306 * things). We could add another new stat but at least for now that
1307 * seems like overkill.
1309 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1310 UDP_INC_STATS(sock_net(sk),
1311 UDP_MIB_SNDBUFERRORS, is_udplite);
1316 if (msg->msg_flags & MSG_PROBE)
1317 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1318 if (!(msg->msg_flags&MSG_PROBE) || len)
1319 goto back_from_confirm;
1323 EXPORT_SYMBOL(udp_sendmsg);
1325 void udp_splice_eof(struct socket *sock)
1327 struct sock *sk = sock->sk;
1328 struct udp_sock *up = udp_sk(sk);
1330 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1334 if (up->pending && !udp_test_bit(CORK, sk))
1335 udp_push_pending_frames(sk);
1338 EXPORT_SYMBOL_GPL(udp_splice_eof);
1340 #define UDP_SKB_IS_STATELESS 0x80000000
1342 /* all head states (dst, sk, nf conntrack) except skb extensions are
1343 * cleared by udp_rcv().
1345 * We need to preserve secpath, if present, to eventually process
1346 * IP_CMSG_PASSSEC at recvmsg() time.
1348 * Other extensions can be cleared.
1350 static bool udp_try_make_stateless(struct sk_buff *skb)
1352 if (!skb_has_extensions(skb))
1355 if (!secpath_exists(skb)) {
1363 static void udp_set_dev_scratch(struct sk_buff *skb)
1365 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1367 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1368 scratch->_tsize_state = skb->truesize;
1369 #if BITS_PER_LONG == 64
1370 scratch->len = skb->len;
1371 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1372 scratch->is_linear = !skb_is_nonlinear(skb);
1374 if (udp_try_make_stateless(skb))
1375 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1378 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1380 /* We come here after udp_lib_checksum_complete() returned 0.
1381 * This means that __skb_checksum_complete() might have
1382 * set skb->csum_valid to 1.
1383 * On 64bit platforms, we can set csum_unnecessary
1384 * to true, but only if the skb is not shared.
1386 #if BITS_PER_LONG == 64
1387 if (!skb_shared(skb))
1388 udp_skb_scratch(skb)->csum_unnecessary = true;
1392 static int udp_skb_truesize(struct sk_buff *skb)
1394 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1397 static bool udp_skb_has_head_state(struct sk_buff *skb)
1399 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1402 /* fully reclaim rmem/fwd memory allocated for skb */
1403 static void udp_rmem_release(struct sock *sk, int size, int partial,
1404 bool rx_queue_lock_held)
1406 struct udp_sock *up = udp_sk(sk);
1407 struct sk_buff_head *sk_queue;
1410 if (likely(partial)) {
1411 up->forward_deficit += size;
1412 size = up->forward_deficit;
1413 if (size < READ_ONCE(up->forward_threshold) &&
1414 !skb_queue_empty(&up->reader_queue))
1417 size += up->forward_deficit;
1419 up->forward_deficit = 0;
1421 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1422 * if the called don't held it already
1424 sk_queue = &sk->sk_receive_queue;
1425 if (!rx_queue_lock_held)
1426 spin_lock(&sk_queue->lock);
1429 sk_forward_alloc_add(sk, size);
1430 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1431 sk_forward_alloc_add(sk, -amt);
1434 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1436 atomic_sub(size, &sk->sk_rmem_alloc);
1438 /* this can save us from acquiring the rx queue lock on next receive */
1439 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1441 if (!rx_queue_lock_held)
1442 spin_unlock(&sk_queue->lock);
1445 /* Note: called with reader_queue.lock held.
1446 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1447 * This avoids a cache line miss while receive_queue lock is held.
1448 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1450 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1452 prefetch(&skb->data);
1453 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1455 EXPORT_SYMBOL(udp_skb_destructor);
1457 /* as above, but the caller held the rx queue lock, too */
1458 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1460 prefetch(&skb->data);
1461 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1464 /* Idea of busylocks is to let producers grab an extra spinlock
1465 * to relieve pressure on the receive_queue spinlock shared by consumer.
1466 * Under flood, this means that only one producer can be in line
1467 * trying to acquire the receive_queue spinlock.
1468 * These busylock can be allocated on a per cpu manner, instead of a
1469 * per socket one (that would consume a cache line per socket)
1471 static int udp_busylocks_log __read_mostly;
1472 static spinlock_t *udp_busylocks __read_mostly;
1474 static spinlock_t *busylock_acquire(void *ptr)
1478 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1483 static void busylock_release(spinlock_t *busy)
1489 static int udp_rmem_schedule(struct sock *sk, int size)
1493 delta = size - sk->sk_forward_alloc;
1494 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1500 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1502 struct sk_buff_head *list = &sk->sk_receive_queue;
1503 int rmem, err = -ENOMEM;
1504 spinlock_t *busy = NULL;
1507 /* try to avoid the costly atomic add/sub pair when the receive
1508 * queue is full; always allow at least a packet
1510 rmem = atomic_read(&sk->sk_rmem_alloc);
1511 if (rmem > sk->sk_rcvbuf)
1514 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1515 * having linear skbs :
1516 * - Reduce memory overhead and thus increase receive queue capacity
1517 * - Less cache line misses at copyout() time
1518 * - Less work at consume_skb() (less alien page frag freeing)
1520 if (rmem > (sk->sk_rcvbuf >> 1)) {
1523 busy = busylock_acquire(sk);
1525 size = skb->truesize;
1526 udp_set_dev_scratch(skb);
1528 /* we drop only if the receive buf is full and the receive
1529 * queue contains some other skb
1531 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1532 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1535 spin_lock(&list->lock);
1536 err = udp_rmem_schedule(sk, size);
1538 spin_unlock(&list->lock);
1542 sk_forward_alloc_add(sk, -size);
1544 /* no need to setup a destructor, we will explicitly release the
1545 * forward allocated memory on dequeue
1547 sock_skb_set_dropcount(sk, skb);
1549 __skb_queue_tail(list, skb);
1550 spin_unlock(&list->lock);
1552 if (!sock_flag(sk, SOCK_DEAD))
1553 INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
1555 busylock_release(busy);
1559 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1562 atomic_inc(&sk->sk_drops);
1563 busylock_release(busy);
1566 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1568 void udp_destruct_common(struct sock *sk)
1570 /* reclaim completely the forward allocated memory */
1571 struct udp_sock *up = udp_sk(sk);
1572 unsigned int total = 0;
1573 struct sk_buff *skb;
1575 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1576 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1577 total += skb->truesize;
1580 udp_rmem_release(sk, total, 0, true);
1582 EXPORT_SYMBOL_GPL(udp_destruct_common);
1584 static void udp_destruct_sock(struct sock *sk)
1586 udp_destruct_common(sk);
1587 inet_sock_destruct(sk);
1590 int udp_init_sock(struct sock *sk)
1592 udp_lib_init_sock(sk);
1593 sk->sk_destruct = udp_destruct_sock;
1594 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1598 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1600 sk_peek_offset_bwd(sk, len);
1602 if (!skb_unref(skb))
1605 /* In the more common cases we cleared the head states previously,
1606 * see __udp_queue_rcv_skb().
1608 if (unlikely(udp_skb_has_head_state(skb)))
1609 skb_release_head_state(skb);
1610 __consume_stateless_skb(skb);
1612 EXPORT_SYMBOL_GPL(skb_consume_udp);
1614 static struct sk_buff *__first_packet_length(struct sock *sk,
1615 struct sk_buff_head *rcvq,
1618 struct sk_buff *skb;
1620 while ((skb = skb_peek(rcvq)) != NULL) {
1621 if (udp_lib_checksum_complete(skb)) {
1622 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1624 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1626 atomic_inc(&sk->sk_drops);
1627 __skb_unlink(skb, rcvq);
1628 *total += skb->truesize;
1631 udp_skb_csum_unnecessary_set(skb);
1639 * first_packet_length - return length of first packet in receive queue
1642 * Drops all bad checksum frames, until a valid one is found.
1643 * Returns the length of found skb, or -1 if none is found.
1645 static int first_packet_length(struct sock *sk)
1647 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1648 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1649 struct sk_buff *skb;
1653 spin_lock_bh(&rcvq->lock);
1654 skb = __first_packet_length(sk, rcvq, &total);
1655 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1656 spin_lock(&sk_queue->lock);
1657 skb_queue_splice_tail_init(sk_queue, rcvq);
1658 spin_unlock(&sk_queue->lock);
1660 skb = __first_packet_length(sk, rcvq, &total);
1662 res = skb ? skb->len : -1;
1664 udp_rmem_release(sk, total, 1, false);
1665 spin_unlock_bh(&rcvq->lock);
1670 * IOCTL requests applicable to the UDP protocol
1673 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1678 *karg = sk_wmem_alloc_get(sk);
1684 *karg = max_t(int, 0, first_packet_length(sk));
1689 return -ENOIOCTLCMD;
1694 EXPORT_SYMBOL(udp_ioctl);
1696 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1699 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1700 struct sk_buff_head *queue;
1701 struct sk_buff *last;
1705 queue = &udp_sk(sk)->reader_queue;
1706 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1708 struct sk_buff *skb;
1710 error = sock_error(sk);
1716 spin_lock_bh(&queue->lock);
1717 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1720 if (!(flags & MSG_PEEK))
1721 udp_skb_destructor(sk, skb);
1722 spin_unlock_bh(&queue->lock);
1726 if (skb_queue_empty_lockless(sk_queue)) {
1727 spin_unlock_bh(&queue->lock);
1731 /* refill the reader queue and walk it again
1732 * keep both queues locked to avoid re-acquiring
1733 * the sk_receive_queue lock if fwd memory scheduling
1736 spin_lock(&sk_queue->lock);
1737 skb_queue_splice_tail_init(sk_queue, queue);
1739 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1741 if (skb && !(flags & MSG_PEEK))
1742 udp_skb_dtor_locked(sk, skb);
1743 spin_unlock(&sk_queue->lock);
1744 spin_unlock_bh(&queue->lock);
1749 if (!sk_can_busy_loop(sk))
1752 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1753 } while (!skb_queue_empty_lockless(sk_queue));
1755 /* sk_queue is empty, reader_queue may contain peeked packets */
1757 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1759 (struct sk_buff *)sk_queue));
1764 EXPORT_SYMBOL(__skb_recv_udp);
1766 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1768 struct sk_buff *skb;
1772 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1776 if (udp_lib_checksum_complete(skb)) {
1777 int is_udplite = IS_UDPLITE(sk);
1778 struct net *net = sock_net(sk);
1780 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1781 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1782 atomic_inc(&sk->sk_drops);
1787 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1788 return recv_actor(sk, skb);
1790 EXPORT_SYMBOL(udp_read_skb);
1793 * This should be easy, if there is something there we
1794 * return it, otherwise we block.
1797 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1800 struct inet_sock *inet = inet_sk(sk);
1801 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1802 struct sk_buff *skb;
1803 unsigned int ulen, copied;
1804 int off, err, peeking = flags & MSG_PEEK;
1805 int is_udplite = IS_UDPLITE(sk);
1806 bool checksum_valid = false;
1808 if (flags & MSG_ERRQUEUE)
1809 return ip_recv_error(sk, msg, len, addr_len);
1812 off = sk_peek_offset(sk, flags);
1813 skb = __skb_recv_udp(sk, flags, &off, &err);
1817 ulen = udp_skb_len(skb);
1819 if (copied > ulen - off)
1820 copied = ulen - off;
1821 else if (copied < ulen)
1822 msg->msg_flags |= MSG_TRUNC;
1825 * If checksum is needed at all, try to do it while copying the
1826 * data. If the data is truncated, or if we only want a partial
1827 * coverage checksum (UDP-Lite), do it before the copy.
1830 if (copied < ulen || peeking ||
1831 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1832 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1833 !__udp_lib_checksum_complete(skb);
1834 if (!checksum_valid)
1838 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1839 if (udp_skb_is_linear(skb))
1840 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1842 err = skb_copy_datagram_msg(skb, off, msg, copied);
1844 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1850 if (unlikely(err)) {
1852 atomic_inc(&sk->sk_drops);
1853 UDP_INC_STATS(sock_net(sk),
1854 UDP_MIB_INERRORS, is_udplite);
1861 UDP_INC_STATS(sock_net(sk),
1862 UDP_MIB_INDATAGRAMS, is_udplite);
1864 sock_recv_cmsgs(msg, sk, skb);
1866 /* Copy the address. */
1868 sin->sin_family = AF_INET;
1869 sin->sin_port = udp_hdr(skb)->source;
1870 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1871 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1872 *addr_len = sizeof(*sin);
1874 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1875 (struct sockaddr *)sin,
1879 if (udp_test_bit(GRO_ENABLED, sk))
1880 udp_cmsg_recv(msg, sk, skb);
1882 if (inet_cmsg_flags(inet))
1883 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1886 if (flags & MSG_TRUNC)
1889 skb_consume_udp(sk, skb, peeking ? -err : err);
1893 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1894 udp_skb_destructor)) {
1895 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1896 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1900 /* starting over for a new packet, but check if we need to yield */
1902 msg->msg_flags &= ~MSG_TRUNC;
1906 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1908 /* This check is replicated from __ip4_datagram_connect() and
1909 * intended to prevent BPF program called below from accessing bytes
1910 * that are out of the bound specified by user in addr_len.
1912 if (addr_len < sizeof(struct sockaddr_in))
1915 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
1917 EXPORT_SYMBOL(udp_pre_connect);
1919 int __udp_disconnect(struct sock *sk, int flags)
1921 struct inet_sock *inet = inet_sk(sk);
1923 * 1003.1g - break association.
1926 sk->sk_state = TCP_CLOSE;
1927 inet->inet_daddr = 0;
1928 inet->inet_dport = 0;
1929 sock_rps_reset_rxhash(sk);
1930 sk->sk_bound_dev_if = 0;
1931 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1932 inet_reset_saddr(sk);
1933 if (sk->sk_prot->rehash &&
1934 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1935 sk->sk_prot->rehash(sk);
1938 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1939 sk->sk_prot->unhash(sk);
1940 inet->inet_sport = 0;
1945 EXPORT_SYMBOL(__udp_disconnect);
1947 int udp_disconnect(struct sock *sk, int flags)
1950 __udp_disconnect(sk, flags);
1954 EXPORT_SYMBOL(udp_disconnect);
1956 void udp_lib_unhash(struct sock *sk)
1958 if (sk_hashed(sk)) {
1959 struct udp_table *udptable = udp_get_table_prot(sk);
1960 struct udp_hslot *hslot, *hslot2;
1962 hslot = udp_hashslot(udptable, sock_net(sk),
1963 udp_sk(sk)->udp_port_hash);
1964 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1966 spin_lock_bh(&hslot->lock);
1967 if (rcu_access_pointer(sk->sk_reuseport_cb))
1968 reuseport_detach_sock(sk);
1969 if (sk_del_node_init_rcu(sk)) {
1971 inet_sk(sk)->inet_num = 0;
1972 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1974 spin_lock(&hslot2->lock);
1975 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1977 spin_unlock(&hslot2->lock);
1979 spin_unlock_bh(&hslot->lock);
1982 EXPORT_SYMBOL(udp_lib_unhash);
1985 * inet_rcv_saddr was changed, we must rehash secondary hash
1987 void udp_lib_rehash(struct sock *sk, u16 newhash)
1989 if (sk_hashed(sk)) {
1990 struct udp_table *udptable = udp_get_table_prot(sk);
1991 struct udp_hslot *hslot, *hslot2, *nhslot2;
1993 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1994 nhslot2 = udp_hashslot2(udptable, newhash);
1995 udp_sk(sk)->udp_portaddr_hash = newhash;
1997 if (hslot2 != nhslot2 ||
1998 rcu_access_pointer(sk->sk_reuseport_cb)) {
1999 hslot = udp_hashslot(udptable, sock_net(sk),
2000 udp_sk(sk)->udp_port_hash);
2001 /* we must lock primary chain too */
2002 spin_lock_bh(&hslot->lock);
2003 if (rcu_access_pointer(sk->sk_reuseport_cb))
2004 reuseport_detach_sock(sk);
2006 if (hslot2 != nhslot2) {
2007 spin_lock(&hslot2->lock);
2008 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2010 spin_unlock(&hslot2->lock);
2012 spin_lock(&nhslot2->lock);
2013 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2016 spin_unlock(&nhslot2->lock);
2019 spin_unlock_bh(&hslot->lock);
2023 EXPORT_SYMBOL(udp_lib_rehash);
2025 void udp_v4_rehash(struct sock *sk)
2027 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2028 inet_sk(sk)->inet_rcv_saddr,
2029 inet_sk(sk)->inet_num);
2030 udp_lib_rehash(sk, new_hash);
2033 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2037 if (inet_sk(sk)->inet_daddr) {
2038 sock_rps_save_rxhash(sk, skb);
2039 sk_mark_napi_id(sk, skb);
2040 sk_incoming_cpu_update(sk);
2042 sk_mark_napi_id_once(sk, skb);
2045 rc = __udp_enqueue_schedule_skb(sk, skb);
2047 int is_udplite = IS_UDPLITE(sk);
2050 /* Note that an ENOMEM error is charged twice */
2051 if (rc == -ENOMEM) {
2052 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2054 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2056 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2058 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2060 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2061 kfree_skb_reason(skb, drop_reason);
2062 trace_udp_fail_queue_rcv_skb(rc, sk);
2072 * >0: "udp encap" protocol resubmission
2074 * Note that in the success and error cases, the skb is assumed to
2075 * have either been requeued or freed.
2077 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2079 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2080 struct udp_sock *up = udp_sk(sk);
2081 int is_udplite = IS_UDPLITE(sk);
2084 * Charge it to the socket, dropping if the queue is full.
2086 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2087 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2092 if (static_branch_unlikely(&udp_encap_needed_key) &&
2093 READ_ONCE(up->encap_type)) {
2094 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2097 * This is an encapsulation socket so pass the skb to
2098 * the socket's udp_encap_rcv() hook. Otherwise, just
2099 * fall through and pass this up the UDP socket.
2100 * up->encap_rcv() returns the following value:
2101 * =0 if skb was successfully passed to the encap
2102 * handler or was discarded by it.
2103 * >0 if skb should be passed on to UDP.
2104 * <0 if skb should be resubmitted as proto -N
2107 /* if we're overly short, let UDP handle it */
2108 encap_rcv = READ_ONCE(up->encap_rcv);
2112 /* Verify checksum before giving to encap */
2113 if (udp_lib_checksum_complete(skb))
2116 ret = encap_rcv(sk, skb);
2118 __UDP_INC_STATS(sock_net(sk),
2119 UDP_MIB_INDATAGRAMS,
2125 /* FALLTHROUGH -- it's a UDP Packet */
2129 * UDP-Lite specific tests, ignored on UDP sockets
2131 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2132 u16 pcrlen = READ_ONCE(up->pcrlen);
2135 * MIB statistics other than incrementing the error count are
2136 * disabled for the following two types of errors: these depend
2137 * on the application settings, not on the functioning of the
2138 * protocol stack as such.
2140 * RFC 3828 here recommends (sec 3.3): "There should also be a
2141 * way ... to ... at least let the receiving application block
2142 * delivery of packets with coverage values less than a value
2143 * provided by the application."
2145 if (pcrlen == 0) { /* full coverage was set */
2146 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2147 UDP_SKB_CB(skb)->cscov, skb->len);
2150 /* The next case involves violating the min. coverage requested
2151 * by the receiver. This is subtle: if receiver wants x and x is
2152 * greater than the buffersize/MTU then receiver will complain
2153 * that it wants x while sender emits packets of smaller size y.
2154 * Therefore the above ...()->partial_cov statement is essential.
2156 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2157 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2158 UDP_SKB_CB(skb)->cscov, pcrlen);
2163 prefetch(&sk->sk_rmem_alloc);
2164 if (rcu_access_pointer(sk->sk_filter) &&
2165 udp_lib_checksum_complete(skb))
2168 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2169 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2173 udp_csum_pull_header(skb);
2175 ipv4_pktinfo_prepare(sk, skb, true);
2176 return __udp_queue_rcv_skb(sk, skb);
2179 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2180 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2182 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2183 atomic_inc(&sk->sk_drops);
2184 kfree_skb_reason(skb, drop_reason);
2188 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2190 struct sk_buff *next, *segs;
2193 if (likely(!udp_unexpected_gso(sk, skb)))
2194 return udp_queue_rcv_one_skb(sk, skb);
2196 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2197 __skb_push(skb, -skb_mac_offset(skb));
2198 segs = udp_rcv_segment(sk, skb, true);
2199 skb_list_walk_safe(segs, skb, next) {
2200 __skb_pull(skb, skb_transport_offset(skb));
2202 udp_post_segment_fix_csum(skb);
2203 ret = udp_queue_rcv_one_skb(sk, skb);
2205 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2210 /* For TCP sockets, sk_rx_dst is protected by socket lock
2211 * For UDP, we use xchg() to guard against concurrent changes.
2213 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2215 struct dst_entry *old;
2217 if (dst_hold_safe(dst)) {
2218 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2224 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2227 * Multicasts and broadcasts go to each listener.
2229 * Note: called only from the BH handler context.
2231 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2233 __be32 saddr, __be32 daddr,
2234 struct udp_table *udptable,
2237 struct sock *sk, *first = NULL;
2238 unsigned short hnum = ntohs(uh->dest);
2239 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2240 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2241 unsigned int offset = offsetof(typeof(*sk), sk_node);
2242 int dif = skb->dev->ifindex;
2243 int sdif = inet_sdif(skb);
2244 struct hlist_node *node;
2245 struct sk_buff *nskb;
2248 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2250 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2252 hslot = &udptable->hash2[hash2];
2253 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2256 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2257 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2258 uh->source, saddr, dif, sdif, hnum))
2265 nskb = skb_clone(skb, GFP_ATOMIC);
2267 if (unlikely(!nskb)) {
2268 atomic_inc(&sk->sk_drops);
2269 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2271 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2275 if (udp_queue_rcv_skb(sk, nskb) > 0)
2279 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2280 if (use_hash2 && hash2 != hash2_any) {
2286 if (udp_queue_rcv_skb(first, skb) > 0)
2290 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2291 proto == IPPROTO_UDPLITE);
2296 /* Initialize UDP checksum. If exited with zero value (success),
2297 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2298 * Otherwise, csum completion requires checksumming packet body,
2299 * including udp header and folding it to skb->csum.
2301 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2306 UDP_SKB_CB(skb)->partial_cov = 0;
2307 UDP_SKB_CB(skb)->cscov = skb->len;
2309 if (proto == IPPROTO_UDPLITE) {
2310 err = udplite_checksum_init(skb, uh);
2314 if (UDP_SKB_CB(skb)->partial_cov) {
2315 skb->csum = inet_compute_pseudo(skb, proto);
2320 /* Note, we are only interested in != 0 or == 0, thus the
2323 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2324 inet_compute_pseudo);
2328 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2329 /* If SW calculated the value, we know it's bad */
2330 if (skb->csum_complete_sw)
2333 /* HW says the value is bad. Let's validate that.
2334 * skb->csum is no longer the full packet checksum,
2335 * so don't treat it as such.
2337 skb_checksum_complete_unset(skb);
2343 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2344 * return code conversion for ip layer consumption
2346 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2351 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2352 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2354 ret = udp_queue_rcv_skb(sk, skb);
2356 /* a return value > 0 means to resubmit the input, but
2357 * it wants the return to be -protocol, or 0
2365 * All we need to do is get the socket, and then do a checksum.
2368 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2373 unsigned short ulen;
2374 struct rtable *rt = skb_rtable(skb);
2375 __be32 saddr, daddr;
2376 struct net *net = dev_net(skb->dev);
2380 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2383 * Validate the packet.
2385 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2386 goto drop; /* No space for header. */
2389 ulen = ntohs(uh->len);
2390 saddr = ip_hdr(skb)->saddr;
2391 daddr = ip_hdr(skb)->daddr;
2393 if (ulen > skb->len)
2396 if (proto == IPPROTO_UDP) {
2397 /* UDP validates ulen. */
2398 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2403 if (udp4_csum_init(skb, uh, proto))
2406 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2407 &refcounted, udp_ehashfn);
2412 struct dst_entry *dst = skb_dst(skb);
2415 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2416 udp_sk_rx_dst_set(sk, dst);
2418 ret = udp_unicast_rcv_skb(sk, skb, uh);
2424 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2425 return __udp4_lib_mcast_deliver(net, skb, uh,
2426 saddr, daddr, udptable, proto);
2428 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2430 return udp_unicast_rcv_skb(sk, skb, uh);
2432 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2436 /* No socket. Drop packet silently, if checksum is wrong */
2437 if (udp_lib_checksum_complete(skb))
2440 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2441 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2442 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2445 * Hmm. We got an UDP packet to a port to which we
2446 * don't wanna listen. Ignore it.
2448 kfree_skb_reason(skb, drop_reason);
2452 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2453 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2454 proto == IPPROTO_UDPLITE ? "Lite" : "",
2455 &saddr, ntohs(uh->source),
2457 &daddr, ntohs(uh->dest));
2462 * RFC1122: OK. Discards the bad packet silently (as far as
2463 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2465 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2466 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2467 proto == IPPROTO_UDPLITE ? "Lite" : "",
2468 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2470 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2472 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2473 kfree_skb_reason(skb, drop_reason);
2477 /* We can only early demux multicast if there is a single matching socket.
2478 * If more than one socket found returns NULL
2480 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2481 __be16 loc_port, __be32 loc_addr,
2482 __be16 rmt_port, __be32 rmt_addr,
2485 struct udp_table *udptable = net->ipv4.udp_table;
2486 unsigned short hnum = ntohs(loc_port);
2487 struct sock *sk, *result;
2488 struct udp_hslot *hslot;
2491 slot = udp_hashfn(net, hnum, udptable->mask);
2492 hslot = &udptable->hash[slot];
2494 /* Do not bother scanning a too big list */
2495 if (hslot->count > 10)
2499 sk_for_each_rcu(sk, &hslot->head) {
2500 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2501 rmt_port, rmt_addr, dif, sdif, hnum)) {
2511 /* For unicast we should only early demux connected sockets or we can
2512 * break forwarding setups. The chains here can be long so only check
2513 * if the first socket is an exact match and if not move on.
2515 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2516 __be16 loc_port, __be32 loc_addr,
2517 __be16 rmt_port, __be32 rmt_addr,
2520 struct udp_table *udptable = net->ipv4.udp_table;
2521 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2522 unsigned short hnum = ntohs(loc_port);
2523 unsigned int hash2, slot2;
2524 struct udp_hslot *hslot2;
2528 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2529 slot2 = hash2 & udptable->mask;
2530 hslot2 = &udptable->hash2[slot2];
2531 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2533 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2534 if (inet_match(net, sk, acookie, ports, dif, sdif))
2536 /* Only check first socket in chain */
2542 int udp_v4_early_demux(struct sk_buff *skb)
2544 struct net *net = dev_net(skb->dev);
2545 struct in_device *in_dev = NULL;
2546 const struct iphdr *iph;
2547 const struct udphdr *uh;
2548 struct sock *sk = NULL;
2549 struct dst_entry *dst;
2550 int dif = skb->dev->ifindex;
2551 int sdif = inet_sdif(skb);
2554 /* validate the packet */
2555 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2561 if (skb->pkt_type == PACKET_MULTICAST) {
2562 in_dev = __in_dev_get_rcu(skb->dev);
2567 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2572 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2573 uh->source, iph->saddr,
2575 } else if (skb->pkt_type == PACKET_HOST) {
2576 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2577 uh->source, iph->saddr, dif, sdif);
2580 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2584 skb->destructor = sock_efree;
2585 dst = rcu_dereference(sk->sk_rx_dst);
2588 dst = dst_check(dst, 0);
2592 /* set noref for now.
2593 * any place which wants to hold dst has to call
2596 skb_dst_set_noref(skb, dst);
2598 /* for unconnected multicast sockets we need to validate
2599 * the source on each packet
2601 if (!inet_sk(sk)->inet_daddr && in_dev)
2602 return ip_mc_validate_source(skb, iph->daddr,
2604 iph->tos & IPTOS_RT_MASK,
2605 skb->dev, in_dev, &itag);
2610 int udp_rcv(struct sk_buff *skb)
2612 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2615 void udp_destroy_sock(struct sock *sk)
2617 struct udp_sock *up = udp_sk(sk);
2618 bool slow = lock_sock_fast(sk);
2620 /* protects from races with udp_abort() */
2621 sock_set_flag(sk, SOCK_DEAD);
2622 udp_flush_pending_frames(sk);
2623 unlock_sock_fast(sk, slow);
2624 if (static_branch_unlikely(&udp_encap_needed_key)) {
2625 if (up->encap_type) {
2626 void (*encap_destroy)(struct sock *sk);
2627 encap_destroy = READ_ONCE(up->encap_destroy);
2631 if (udp_test_bit(ENCAP_ENABLED, sk))
2632 static_branch_dec(&udp_encap_needed_key);
2636 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2640 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2641 if (family == AF_INET)
2642 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2643 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2644 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2650 * Socket option code for UDP
2652 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2653 sockptr_t optval, unsigned int optlen,
2654 int (*push_pending_frames)(struct sock *))
2656 struct udp_sock *up = udp_sk(sk);
2659 int is_udplite = IS_UDPLITE(sk);
2661 if (level == SOL_SOCKET) {
2662 err = sk_setsockopt(sk, level, optname, optval, optlen);
2664 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2665 sockopt_lock_sock(sk);
2666 /* paired with READ_ONCE in udp_rmem_release() */
2667 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2668 sockopt_release_sock(sk);
2673 if (optlen < sizeof(int))
2676 if (copy_from_sockptr(&val, optval, sizeof(val)))
2679 valbool = val ? 1 : 0;
2684 udp_set_bit(CORK, sk);
2686 udp_clear_bit(CORK, sk);
2688 push_pending_frames(sk);
2697 case UDP_ENCAP_ESPINUDP:
2698 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2700 case UDP_ENCAP_ESPINUDP_NON_IKE:
2701 #if IS_ENABLED(CONFIG_IPV6)
2702 if (sk->sk_family == AF_INET6)
2703 WRITE_ONCE(up->encap_rcv,
2704 ipv6_stub->xfrm6_udp_encap_rcv);
2707 WRITE_ONCE(up->encap_rcv,
2708 xfrm4_udp_encap_rcv);
2711 case UDP_ENCAP_L2TPINUDP:
2712 WRITE_ONCE(up->encap_type, val);
2713 udp_tunnel_encap_enable(sk);
2721 case UDP_NO_CHECK6_TX:
2722 udp_set_no_check6_tx(sk, valbool);
2725 case UDP_NO_CHECK6_RX:
2726 udp_set_no_check6_rx(sk, valbool);
2730 if (val < 0 || val > USHRT_MAX)
2732 WRITE_ONCE(up->gso_size, val);
2737 /* when enabling GRO, accept the related GSO packet type */
2739 udp_tunnel_encap_enable(sk);
2740 udp_assign_bit(GRO_ENABLED, sk, valbool);
2741 udp_assign_bit(ACCEPT_L4, sk, valbool);
2742 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
2746 * UDP-Lite's partial checksum coverage (RFC 3828).
2748 /* The sender sets actual checksum coverage length via this option.
2749 * The case coverage > packet length is handled by send module. */
2750 case UDPLITE_SEND_CSCOV:
2751 if (!is_udplite) /* Disable the option on UDP sockets */
2752 return -ENOPROTOOPT;
2753 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2755 else if (val > USHRT_MAX)
2757 WRITE_ONCE(up->pcslen, val);
2758 udp_set_bit(UDPLITE_SEND_CC, sk);
2761 /* The receiver specifies a minimum checksum coverage value. To make
2762 * sense, this should be set to at least 8 (as done below). If zero is
2763 * used, this again means full checksum coverage. */
2764 case UDPLITE_RECV_CSCOV:
2765 if (!is_udplite) /* Disable the option on UDP sockets */
2766 return -ENOPROTOOPT;
2767 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2769 else if (val > USHRT_MAX)
2771 WRITE_ONCE(up->pcrlen, val);
2772 udp_set_bit(UDPLITE_RECV_CC, sk);
2782 EXPORT_SYMBOL(udp_lib_setsockopt);
2784 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2785 unsigned int optlen)
2787 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2788 return udp_lib_setsockopt(sk, level, optname,
2790 udp_push_pending_frames);
2791 return ip_setsockopt(sk, level, optname, optval, optlen);
2794 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2795 char __user *optval, int __user *optlen)
2797 struct udp_sock *up = udp_sk(sk);
2800 if (get_user(len, optlen))
2806 len = min_t(unsigned int, len, sizeof(int));
2810 val = udp_test_bit(CORK, sk);
2814 val = READ_ONCE(up->encap_type);
2817 case UDP_NO_CHECK6_TX:
2818 val = udp_get_no_check6_tx(sk);
2821 case UDP_NO_CHECK6_RX:
2822 val = udp_get_no_check6_rx(sk);
2826 val = READ_ONCE(up->gso_size);
2830 val = udp_test_bit(GRO_ENABLED, sk);
2833 /* The following two cannot be changed on UDP sockets, the return is
2834 * always 0 (which corresponds to the full checksum coverage of UDP). */
2835 case UDPLITE_SEND_CSCOV:
2836 val = READ_ONCE(up->pcslen);
2839 case UDPLITE_RECV_CSCOV:
2840 val = READ_ONCE(up->pcrlen);
2844 return -ENOPROTOOPT;
2847 if (put_user(len, optlen))
2849 if (copy_to_user(optval, &val, len))
2853 EXPORT_SYMBOL(udp_lib_getsockopt);
2855 int udp_getsockopt(struct sock *sk, int level, int optname,
2856 char __user *optval, int __user *optlen)
2858 if (level == SOL_UDP || level == SOL_UDPLITE)
2859 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2860 return ip_getsockopt(sk, level, optname, optval, optlen);
2864 * udp_poll - wait for a UDP event.
2865 * @file: - file struct
2867 * @wait: - poll table
2869 * This is same as datagram poll, except for the special case of
2870 * blocking sockets. If application is using a blocking fd
2871 * and a packet with checksum error is in the queue;
2872 * then it could get return from select indicating data available
2873 * but then block when reading it. Add special case code
2874 * to work around these arguably broken applications.
2876 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2878 __poll_t mask = datagram_poll(file, sock, wait);
2879 struct sock *sk = sock->sk;
2881 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2882 mask |= EPOLLIN | EPOLLRDNORM;
2884 /* Check for false positives due to checksum errors */
2885 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2886 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2887 mask &= ~(EPOLLIN | EPOLLRDNORM);
2889 /* psock ingress_msg queue should not contain any bad checksum frames */
2890 if (sk_is_readable(sk))
2891 mask |= EPOLLIN | EPOLLRDNORM;
2895 EXPORT_SYMBOL(udp_poll);
2897 int udp_abort(struct sock *sk, int err)
2899 if (!has_current_bpf_ctx())
2902 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2905 if (sock_flag(sk, SOCK_DEAD))
2909 sk_error_report(sk);
2910 __udp_disconnect(sk, 0);
2913 if (!has_current_bpf_ctx())
2918 EXPORT_SYMBOL_GPL(udp_abort);
2920 struct proto udp_prot = {
2922 .owner = THIS_MODULE,
2923 .close = udp_lib_close,
2924 .pre_connect = udp_pre_connect,
2925 .connect = ip4_datagram_connect,
2926 .disconnect = udp_disconnect,
2928 .init = udp_init_sock,
2929 .destroy = udp_destroy_sock,
2930 .setsockopt = udp_setsockopt,
2931 .getsockopt = udp_getsockopt,
2932 .sendmsg = udp_sendmsg,
2933 .recvmsg = udp_recvmsg,
2934 .splice_eof = udp_splice_eof,
2935 .release_cb = ip4_datagram_release_cb,
2936 .hash = udp_lib_hash,
2937 .unhash = udp_lib_unhash,
2938 .rehash = udp_v4_rehash,
2939 .get_port = udp_v4_get_port,
2940 .put_port = udp_lib_unhash,
2941 #ifdef CONFIG_BPF_SYSCALL
2942 .psock_update_sk_prot = udp_bpf_update_proto,
2944 .memory_allocated = &udp_memory_allocated,
2945 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2947 .sysctl_mem = sysctl_udp_mem,
2948 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2949 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2950 .obj_size = sizeof(struct udp_sock),
2951 .h.udp_table = NULL,
2952 .diag_destroy = udp_abort,
2954 EXPORT_SYMBOL(udp_prot);
2956 /* ------------------------------------------------------------------------ */
2957 #ifdef CONFIG_PROC_FS
2959 static unsigned short seq_file_family(const struct seq_file *seq);
2960 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2962 unsigned short family = seq_file_family(seq);
2964 /* AF_UNSPEC is used as a match all */
2965 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2966 net_eq(sock_net(sk), seq_file_net(seq)));
2969 #ifdef CONFIG_BPF_SYSCALL
2970 static const struct seq_operations bpf_iter_udp_seq_ops;
2972 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2975 const struct udp_seq_afinfo *afinfo;
2977 #ifdef CONFIG_BPF_SYSCALL
2978 if (seq->op == &bpf_iter_udp_seq_ops)
2979 return net->ipv4.udp_table;
2982 afinfo = pde_data(file_inode(seq->file));
2983 return afinfo->udp_table ? : net->ipv4.udp_table;
2986 static struct sock *udp_get_first(struct seq_file *seq, int start)
2988 struct udp_iter_state *state = seq->private;
2989 struct net *net = seq_file_net(seq);
2990 struct udp_table *udptable;
2993 udptable = udp_get_table_seq(seq, net);
2995 for (state->bucket = start; state->bucket <= udptable->mask;
2997 struct udp_hslot *hslot = &udptable->hash[state->bucket];
2999 if (hlist_empty(&hslot->head))
3002 spin_lock_bh(&hslot->lock);
3003 sk_for_each(sk, &hslot->head) {
3004 if (seq_sk_match(seq, sk))
3007 spin_unlock_bh(&hslot->lock);
3014 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3016 struct udp_iter_state *state = seq->private;
3017 struct net *net = seq_file_net(seq);
3018 struct udp_table *udptable;
3022 } while (sk && !seq_sk_match(seq, sk));
3025 udptable = udp_get_table_seq(seq, net);
3027 if (state->bucket <= udptable->mask)
3028 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3030 return udp_get_first(seq, state->bucket + 1);
3035 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3037 struct sock *sk = udp_get_first(seq, 0);
3040 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3042 return pos ? NULL : sk;
3045 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3047 struct udp_iter_state *state = seq->private;
3048 state->bucket = MAX_UDP_PORTS;
3050 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3052 EXPORT_SYMBOL(udp_seq_start);
3054 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3058 if (v == SEQ_START_TOKEN)
3059 sk = udp_get_idx(seq, 0);
3061 sk = udp_get_next(seq, v);
3066 EXPORT_SYMBOL(udp_seq_next);
3068 void udp_seq_stop(struct seq_file *seq, void *v)
3070 struct udp_iter_state *state = seq->private;
3071 struct udp_table *udptable;
3073 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3075 if (state->bucket <= udptable->mask)
3076 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3078 EXPORT_SYMBOL(udp_seq_stop);
3080 /* ------------------------------------------------------------------------ */
3081 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3084 struct inet_sock *inet = inet_sk(sp);
3085 __be32 dest = inet->inet_daddr;
3086 __be32 src = inet->inet_rcv_saddr;
3087 __u16 destp = ntohs(inet->inet_dport);
3088 __u16 srcp = ntohs(inet->inet_sport);
3090 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3091 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3092 bucket, src, srcp, dest, destp, sp->sk_state,
3093 sk_wmem_alloc_get(sp),
3096 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3098 refcount_read(&sp->sk_refcnt), sp,
3099 atomic_read(&sp->sk_drops));
3102 int udp4_seq_show(struct seq_file *seq, void *v)
3104 seq_setwidth(seq, 127);
3105 if (v == SEQ_START_TOKEN)
3106 seq_puts(seq, " sl local_address rem_address st tx_queue "
3107 "rx_queue tr tm->when retrnsmt uid timeout "
3108 "inode ref pointer drops");
3110 struct udp_iter_state *state = seq->private;
3112 udp4_format_sock(v, seq, state->bucket);
3118 #ifdef CONFIG_BPF_SYSCALL
3119 struct bpf_iter__udp {
3120 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3121 __bpf_md_ptr(struct udp_sock *, udp_sk);
3122 uid_t uid __aligned(8);
3123 int bucket __aligned(8);
3126 struct bpf_udp_iter_state {
3127 struct udp_iter_state state;
3128 unsigned int cur_sk;
3129 unsigned int end_sk;
3130 unsigned int max_sk;
3132 struct sock **batch;
3133 bool st_bucket_done;
3136 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3137 unsigned int new_batch_sz);
3138 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3140 struct bpf_udp_iter_state *iter = seq->private;
3141 struct udp_iter_state *state = &iter->state;
3142 struct net *net = seq_file_net(seq);
3143 int resume_bucket, resume_offset;
3144 struct udp_table *udptable;
3145 unsigned int batch_sks = 0;
3146 bool resized = false;
3149 resume_bucket = state->bucket;
3150 resume_offset = iter->offset;
3152 /* The current batch is done, so advance the bucket. */
3153 if (iter->st_bucket_done)
3156 udptable = udp_get_table_seq(seq, net);
3159 /* New batch for the next bucket.
3160 * Iterate over the hash table to find a bucket with sockets matching
3161 * the iterator attributes, and return the first matching socket from
3162 * the bucket. The remaining matched sockets from the bucket are batched
3163 * before releasing the bucket lock. This allows BPF programs that are
3164 * called in seq_show to acquire the bucket lock if needed.
3168 iter->st_bucket_done = false;
3171 for (; state->bucket <= udptable->mask; state->bucket++) {
3172 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3174 if (hlist_empty(&hslot2->head))
3178 spin_lock_bh(&hslot2->lock);
3179 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3180 if (seq_sk_match(seq, sk)) {
3181 /* Resume from the last iterated socket at the
3182 * offset in the bucket before iterator was stopped.
3184 if (state->bucket == resume_bucket &&
3185 iter->offset < resume_offset) {
3189 if (iter->end_sk < iter->max_sk) {
3191 iter->batch[iter->end_sk++] = sk;
3196 spin_unlock_bh(&hslot2->lock);
3202 /* All done: no batch made. */
3206 if (iter->end_sk == batch_sks) {
3207 /* Batching is done for the current bucket; return the first
3208 * socket to be iterated from the batch.
3210 iter->st_bucket_done = true;
3213 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3215 /* After allocating a larger batch, retry one more time to grab
3221 return iter->batch[0];
3224 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3226 struct bpf_udp_iter_state *iter = seq->private;
3229 /* Whenever seq_next() is called, the iter->cur_sk is
3230 * done with seq_show(), so unref the iter->cur_sk.
3232 if (iter->cur_sk < iter->end_sk) {
3233 sock_put(iter->batch[iter->cur_sk++]);
3237 /* After updating iter->cur_sk, check if there are more sockets
3238 * available in the current bucket batch.
3240 if (iter->cur_sk < iter->end_sk)
3241 sk = iter->batch[iter->cur_sk];
3243 /* Prepare a new batch. */
3244 sk = bpf_iter_udp_batch(seq);
3250 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3252 /* bpf iter does not support lseek, so it always
3253 * continue from where it was stop()-ped.
3256 return bpf_iter_udp_batch(seq);
3258 return SEQ_START_TOKEN;
3261 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3262 struct udp_sock *udp_sk, uid_t uid, int bucket)
3264 struct bpf_iter__udp ctx;
3266 meta->seq_num--; /* skip SEQ_START_TOKEN */
3268 ctx.udp_sk = udp_sk;
3270 ctx.bucket = bucket;
3271 return bpf_iter_run_prog(prog, &ctx);
3274 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3276 struct udp_iter_state *state = seq->private;
3277 struct bpf_iter_meta meta;
3278 struct bpf_prog *prog;
3279 struct sock *sk = v;
3283 if (v == SEQ_START_TOKEN)
3288 if (unlikely(sk_unhashed(sk))) {
3293 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3295 prog = bpf_iter_get_info(&meta, false);
3296 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3303 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3305 while (iter->cur_sk < iter->end_sk)
3306 sock_put(iter->batch[iter->cur_sk++]);
3309 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3311 struct bpf_udp_iter_state *iter = seq->private;
3312 struct bpf_iter_meta meta;
3313 struct bpf_prog *prog;
3317 prog = bpf_iter_get_info(&meta, true);
3319 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3322 if (iter->cur_sk < iter->end_sk) {
3323 bpf_iter_udp_put_batch(iter);
3324 iter->st_bucket_done = false;
3328 static const struct seq_operations bpf_iter_udp_seq_ops = {
3329 .start = bpf_iter_udp_seq_start,
3330 .next = bpf_iter_udp_seq_next,
3331 .stop = bpf_iter_udp_seq_stop,
3332 .show = bpf_iter_udp_seq_show,
3336 static unsigned short seq_file_family(const struct seq_file *seq)
3338 const struct udp_seq_afinfo *afinfo;
3340 #ifdef CONFIG_BPF_SYSCALL
3341 /* BPF iterator: bpf programs to filter sockets. */
3342 if (seq->op == &bpf_iter_udp_seq_ops)
3346 /* Proc fs iterator */
3347 afinfo = pde_data(file_inode(seq->file));
3348 return afinfo->family;
3351 const struct seq_operations udp_seq_ops = {
3352 .start = udp_seq_start,
3353 .next = udp_seq_next,
3354 .stop = udp_seq_stop,
3355 .show = udp4_seq_show,
3357 EXPORT_SYMBOL(udp_seq_ops);
3359 static struct udp_seq_afinfo udp4_seq_afinfo = {
3364 static int __net_init udp4_proc_init_net(struct net *net)
3366 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3367 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3372 static void __net_exit udp4_proc_exit_net(struct net *net)
3374 remove_proc_entry("udp", net->proc_net);
3377 static struct pernet_operations udp4_net_ops = {
3378 .init = udp4_proc_init_net,
3379 .exit = udp4_proc_exit_net,
3382 int __init udp4_proc_init(void)
3384 return register_pernet_subsys(&udp4_net_ops);
3387 void udp4_proc_exit(void)
3389 unregister_pernet_subsys(&udp4_net_ops);
3391 #endif /* CONFIG_PROC_FS */
3393 static __initdata unsigned long uhash_entries;
3394 static int __init set_uhash_entries(char *str)
3401 ret = kstrtoul(str, 0, &uhash_entries);
3405 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3406 uhash_entries = UDP_HTABLE_SIZE_MIN;
3409 __setup("uhash_entries=", set_uhash_entries);
3411 void __init udp_table_init(struct udp_table *table, const char *name)
3415 table->hash = alloc_large_system_hash(name,
3416 2 * sizeof(struct udp_hslot),
3418 21, /* one slot per 2 MB */
3422 UDP_HTABLE_SIZE_MIN,
3423 UDP_HTABLE_SIZE_MAX);
3425 table->hash2 = table->hash + (table->mask + 1);
3426 for (i = 0; i <= table->mask; i++) {
3427 INIT_HLIST_HEAD(&table->hash[i].head);
3428 table->hash[i].count = 0;
3429 spin_lock_init(&table->hash[i].lock);
3431 for (i = 0; i <= table->mask; i++) {
3432 INIT_HLIST_HEAD(&table->hash2[i].head);
3433 table->hash2[i].count = 0;
3434 spin_lock_init(&table->hash2[i].lock);
3438 u32 udp_flow_hashrnd(void)
3440 static u32 hashrnd __read_mostly;
3442 net_get_random_once(&hashrnd, sizeof(hashrnd));
3446 EXPORT_SYMBOL(udp_flow_hashrnd);
3448 static void __net_init udp_sysctl_init(struct net *net)
3450 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3451 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3453 #ifdef CONFIG_NET_L3_MASTER_DEV
3454 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3458 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3460 struct udp_table *udptable;
3463 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3467 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3468 GFP_KERNEL_ACCOUNT);
3469 if (!udptable->hash)
3472 udptable->hash2 = udptable->hash + hash_entries;
3473 udptable->mask = hash_entries - 1;
3474 udptable->log = ilog2(hash_entries);
3476 for (i = 0; i < hash_entries; i++) {
3477 INIT_HLIST_HEAD(&udptable->hash[i].head);
3478 udptable->hash[i].count = 0;
3479 spin_lock_init(&udptable->hash[i].lock);
3481 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3482 udptable->hash2[i].count = 0;
3483 spin_lock_init(&udptable->hash2[i].lock);
3494 static void __net_exit udp_pernet_table_free(struct net *net)
3496 struct udp_table *udptable = net->ipv4.udp_table;
3498 if (udptable == &udp_table)
3501 kvfree(udptable->hash);
3505 static void __net_init udp_set_table(struct net *net)
3507 struct udp_table *udptable;
3508 unsigned int hash_entries;
3509 struct net *old_net;
3511 if (net_eq(net, &init_net))
3514 old_net = current->nsproxy->net_ns;
3515 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3519 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3520 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3521 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3523 hash_entries = roundup_pow_of_two(hash_entries);
3525 udptable = udp_pernet_table_alloc(hash_entries);
3527 net->ipv4.udp_table = udptable;
3529 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3530 "for a netns, fallback to the global one\n",
3533 net->ipv4.udp_table = &udp_table;
3537 static int __net_init udp_pernet_init(struct net *net)
3539 udp_sysctl_init(net);
3545 static void __net_exit udp_pernet_exit(struct net *net)
3547 udp_pernet_table_free(net);
3550 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3551 .init = udp_pernet_init,
3552 .exit = udp_pernet_exit,
3555 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3556 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3557 struct udp_sock *udp_sk, uid_t uid, int bucket)
3559 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3560 unsigned int new_batch_sz)
3562 struct sock **new_batch;
3564 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3565 GFP_USER | __GFP_NOWARN);
3569 bpf_iter_udp_put_batch(iter);
3570 kvfree(iter->batch);
3571 iter->batch = new_batch;
3572 iter->max_sk = new_batch_sz;
3577 #define INIT_BATCH_SZ 16
3579 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3581 struct bpf_udp_iter_state *iter = priv_data;
3584 ret = bpf_iter_init_seq_net(priv_data, aux);
3588 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3590 bpf_iter_fini_seq_net(priv_data);
3595 static void bpf_iter_fini_udp(void *priv_data)
3597 struct bpf_udp_iter_state *iter = priv_data;
3599 bpf_iter_fini_seq_net(priv_data);
3600 kvfree(iter->batch);
3603 static const struct bpf_iter_seq_info udp_seq_info = {
3604 .seq_ops = &bpf_iter_udp_seq_ops,
3605 .init_seq_private = bpf_iter_init_udp,
3606 .fini_seq_private = bpf_iter_fini_udp,
3607 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3610 static struct bpf_iter_reg udp_reg_info = {
3612 .ctx_arg_info_size = 1,
3614 { offsetof(struct bpf_iter__udp, udp_sk),
3615 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3617 .seq_info = &udp_seq_info,
3620 static void __init bpf_iter_register(void)
3622 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3623 if (bpf_iter_reg_target(&udp_reg_info))
3624 pr_warn("Warning: could not register bpf iterator udp\n");
3628 void __init udp_init(void)
3630 unsigned long limit;
3633 udp_table_init(&udp_table, "UDP");
3634 limit = nr_free_buffer_pages() / 8;
3635 limit = max(limit, 128UL);
3636 sysctl_udp_mem[0] = limit / 4 * 3;
3637 sysctl_udp_mem[1] = limit;
3638 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3640 /* 16 spinlocks per cpu */
3641 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3642 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3645 panic("UDP: failed to alloc udp_busylocks\n");
3646 for (i = 0; i < (1U << udp_busylocks_log); i++)
3647 spin_lock_init(udp_busylocks + i);
3649 if (register_pernet_subsys(&udp_sysctl_ops))
3650 panic("UDP: failed to init sysctl parameters.\n");
3652 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3653 bpf_iter_register();