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 void udp_encap_enable(void)
589 static_branch_inc(&udp_encap_needed_key);
591 EXPORT_SYMBOL(udp_encap_enable);
593 void udp_encap_disable(void)
595 static_branch_dec(&udp_encap_needed_key);
597 EXPORT_SYMBOL(udp_encap_disable);
599 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
600 * through error handlers in encapsulations looking for a match.
602 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
606 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
607 int (*handler)(struct sk_buff *skb, u32 info);
608 const struct ip_tunnel_encap_ops *encap;
610 encap = rcu_dereference(iptun_encaps[i]);
613 handler = encap->err_handler;
614 if (handler && !handler(skb, info))
621 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
622 * reversing source and destination port: this will match tunnels that force the
623 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
624 * lwtunnels might actually break this assumption by being configured with
625 * different destination ports on endpoints, in this case we won't be able to
626 * trace ICMP messages back to them.
628 * If this doesn't match any socket, probe tunnels with arbitrary destination
629 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
630 * we've sent packets to won't necessarily match the local destination port.
632 * Then ask the tunnel implementation to match the error against a valid
635 * Return an error if we can't find a match, the socket if we need further
636 * processing, zero otherwise.
638 static struct sock *__udp4_lib_err_encap(struct net *net,
639 const struct iphdr *iph,
641 struct udp_table *udptable,
643 struct sk_buff *skb, u32 info)
645 int (*lookup)(struct sock *sk, struct sk_buff *skb);
646 int network_offset, transport_offset;
649 network_offset = skb_network_offset(skb);
650 transport_offset = skb_transport_offset(skb);
652 /* Network header needs to point to the outer IPv4 header inside ICMP */
653 skb_reset_network_header(skb);
655 /* Transport header needs to point to the UDP header */
656 skb_set_transport_header(skb, iph->ihl << 2);
661 lookup = READ_ONCE(up->encap_err_lookup);
662 if (lookup && lookup(sk, skb))
668 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
669 iph->saddr, uh->dest, skb->dev->ifindex, 0,
674 lookup = READ_ONCE(up->encap_err_lookup);
675 if (!lookup || lookup(sk, skb))
681 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
683 skb_set_transport_header(skb, transport_offset);
684 skb_set_network_header(skb, network_offset);
690 * This routine is called by the ICMP module when it gets some
691 * sort of error condition. If err < 0 then the socket should
692 * be closed and the error returned to the user. If err > 0
693 * it's just the icmp type << 8 | icmp code.
694 * Header points to the ip header of the error packet. We move
695 * on past this. Then (as it used to claim before adjustment)
696 * header points to the first 8 bytes of the udp header. We need
697 * to find the appropriate port.
700 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
702 struct inet_sock *inet;
703 const struct iphdr *iph = (const struct iphdr *)skb->data;
704 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
705 const int type = icmp_hdr(skb)->type;
706 const int code = icmp_hdr(skb)->code;
711 struct net *net = dev_net(skb->dev);
713 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
714 iph->saddr, uh->source, skb->dev->ifindex,
715 inet_sdif(skb), udptable, NULL);
717 if (!sk || udp_sk(sk)->encap_type) {
718 /* No socket for error: try tunnels before discarding */
719 if (static_branch_unlikely(&udp_encap_needed_key)) {
720 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
725 sk = ERR_PTR(-ENOENT);
728 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
741 case ICMP_TIME_EXCEEDED:
744 case ICMP_SOURCE_QUENCH:
746 case ICMP_PARAMETERPROB:
750 case ICMP_DEST_UNREACH:
751 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
752 ipv4_sk_update_pmtu(skb, sk, info);
753 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
761 if (code <= NR_ICMP_UNREACH) {
762 harderr = icmp_err_convert[code].fatal;
763 err = icmp_err_convert[code].errno;
767 ipv4_sk_redirect(skb, sk);
772 * RFC1122: OK. Passes ICMP errors back to application, as per
776 /* ...not for tunnels though: we don't have a sending socket */
777 if (udp_sk(sk)->encap_err_rcv)
778 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
782 if (!inet_test_bit(RECVERR, sk)) {
783 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
786 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
794 int udp_err(struct sk_buff *skb, u32 info)
796 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
800 * Throw away all pending data and cancel the corking. Socket is locked.
802 void udp_flush_pending_frames(struct sock *sk)
804 struct udp_sock *up = udp_sk(sk);
809 ip_flush_pending_frames(sk);
812 EXPORT_SYMBOL(udp_flush_pending_frames);
815 * udp4_hwcsum - handle outgoing HW checksumming
816 * @skb: sk_buff containing the filled-in UDP header
817 * (checksum field must be zeroed out)
818 * @src: source IP address
819 * @dst: destination IP address
821 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
823 struct udphdr *uh = udp_hdr(skb);
824 int offset = skb_transport_offset(skb);
825 int len = skb->len - offset;
829 if (!skb_has_frag_list(skb)) {
831 * Only one fragment on the socket.
833 skb->csum_start = skb_transport_header(skb) - skb->head;
834 skb->csum_offset = offsetof(struct udphdr, check);
835 uh->check = ~csum_tcpudp_magic(src, dst, len,
838 struct sk_buff *frags;
841 * HW-checksum won't work as there are two or more
842 * fragments on the socket so that all csums of sk_buffs
845 skb_walk_frags(skb, frags) {
846 csum = csum_add(csum, frags->csum);
850 csum = skb_checksum(skb, offset, hlen, csum);
851 skb->ip_summed = CHECKSUM_NONE;
853 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
855 uh->check = CSUM_MANGLED_0;
858 EXPORT_SYMBOL_GPL(udp4_hwcsum);
860 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
861 * for the simple case like when setting the checksum for a UDP tunnel.
863 void udp_set_csum(bool nocheck, struct sk_buff *skb,
864 __be32 saddr, __be32 daddr, int len)
866 struct udphdr *uh = udp_hdr(skb);
870 } else if (skb_is_gso(skb)) {
871 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
872 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
874 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
876 uh->check = CSUM_MANGLED_0;
878 skb->ip_summed = CHECKSUM_PARTIAL;
879 skb->csum_start = skb_transport_header(skb) - skb->head;
880 skb->csum_offset = offsetof(struct udphdr, check);
881 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
884 EXPORT_SYMBOL(udp_set_csum);
886 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
887 struct inet_cork *cork)
889 struct sock *sk = skb->sk;
890 struct inet_sock *inet = inet_sk(sk);
893 int is_udplite = IS_UDPLITE(sk);
894 int offset = skb_transport_offset(skb);
895 int len = skb->len - offset;
896 int datalen = len - sizeof(*uh);
900 * Create a UDP header
903 uh->source = inet->inet_sport;
904 uh->dest = fl4->fl4_dport;
905 uh->len = htons(len);
908 if (cork->gso_size) {
909 const int hlen = skb_network_header_len(skb) +
910 sizeof(struct udphdr);
912 if (hlen + cork->gso_size > cork->fragsize) {
916 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
920 if (sk->sk_no_check_tx) {
924 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
925 dst_xfrm(skb_dst(skb))) {
930 if (datalen > cork->gso_size) {
931 skb_shinfo(skb)->gso_size = cork->gso_size;
932 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
933 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
939 if (is_udplite) /* UDP-Lite */
940 csum = udplite_csum(skb);
942 else if (sk->sk_no_check_tx) { /* UDP csum off */
944 skb->ip_summed = CHECKSUM_NONE;
947 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
950 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
954 csum = udp_csum(skb);
956 /* add protocol-dependent pseudo-header */
957 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
958 sk->sk_protocol, csum);
960 uh->check = CSUM_MANGLED_0;
963 err = ip_send_skb(sock_net(sk), skb);
965 if (err == -ENOBUFS &&
966 !inet_test_bit(RECVERR, sk)) {
967 UDP_INC_STATS(sock_net(sk),
968 UDP_MIB_SNDBUFERRORS, is_udplite);
972 UDP_INC_STATS(sock_net(sk),
973 UDP_MIB_OUTDATAGRAMS, is_udplite);
978 * Push out all pending data as one UDP datagram. Socket is locked.
980 int udp_push_pending_frames(struct sock *sk)
982 struct udp_sock *up = udp_sk(sk);
983 struct inet_sock *inet = inet_sk(sk);
984 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
988 skb = ip_finish_skb(sk, fl4);
992 err = udp_send_skb(skb, fl4, &inet->cork.base);
999 EXPORT_SYMBOL(udp_push_pending_frames);
1001 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1003 switch (cmsg->cmsg_type) {
1005 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1007 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1014 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1016 struct cmsghdr *cmsg;
1017 bool need_ip = false;
1020 for_each_cmsghdr(cmsg, msg) {
1021 if (!CMSG_OK(msg, cmsg))
1024 if (cmsg->cmsg_level != SOL_UDP) {
1029 err = __udp_cmsg_send(cmsg, gso_size);
1036 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1038 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1040 struct inet_sock *inet = inet_sk(sk);
1041 struct udp_sock *up = udp_sk(sk);
1042 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1043 struct flowi4 fl4_stack;
1046 struct ipcm_cookie ipc;
1047 struct rtable *rt = NULL;
1050 __be32 daddr, faddr, saddr;
1053 int err, is_udplite = IS_UDPLITE(sk);
1054 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1055 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1056 struct sk_buff *skb;
1057 struct ip_options_data opt_copy;
1066 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1069 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1071 fl4 = &inet->cork.fl.u.ip4;
1074 * There are pending frames.
1075 * The socket lock must be held while it's corked.
1078 if (likely(up->pending)) {
1079 if (unlikely(up->pending != AF_INET)) {
1083 goto do_append_data;
1087 ulen += sizeof(struct udphdr);
1090 * Get and verify the address.
1093 if (msg->msg_namelen < sizeof(*usin))
1095 if (usin->sin_family != AF_INET) {
1096 if (usin->sin_family != AF_UNSPEC)
1097 return -EAFNOSUPPORT;
1100 daddr = usin->sin_addr.s_addr;
1101 dport = usin->sin_port;
1105 if (sk->sk_state != TCP_ESTABLISHED)
1106 return -EDESTADDRREQ;
1107 daddr = inet->inet_daddr;
1108 dport = inet->inet_dport;
1109 /* Open fast path for connected socket.
1110 Route will not be used, if at least one option is set.
1115 ipcm_init_sk(&ipc, inet);
1116 ipc.gso_size = READ_ONCE(up->gso_size);
1118 if (msg->msg_controllen) {
1119 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1121 err = ip_cmsg_send(sk, msg, &ipc,
1122 sk->sk_family == AF_INET6);
1123 if (unlikely(err < 0)) {
1132 struct ip_options_rcu *inet_opt;
1135 inet_opt = rcu_dereference(inet->inet_opt);
1137 memcpy(&opt_copy, inet_opt,
1138 sizeof(*inet_opt) + inet_opt->opt.optlen);
1139 ipc.opt = &opt_copy.opt;
1144 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1145 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1146 (struct sockaddr *)usin, &ipc.addr);
1150 if (usin->sin_port == 0) {
1151 /* BPF program set invalid port. Reject it. */
1155 daddr = usin->sin_addr.s_addr;
1156 dport = usin->sin_port;
1161 ipc.addr = faddr = daddr;
1163 if (ipc.opt && ipc.opt->opt.srr) {
1168 faddr = ipc.opt->opt.faddr;
1171 tos = get_rttos(&ipc, inet);
1172 scope = ip_sendmsg_scope(inet, &ipc, msg);
1173 if (scope == RT_SCOPE_LINK)
1176 if (ipv4_is_multicast(daddr)) {
1177 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1178 ipc.oif = inet->mc_index;
1180 saddr = inet->mc_addr;
1182 } else if (!ipc.oif) {
1183 ipc.oif = inet->uc_index;
1184 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1185 /* oif is set, packet is to local broadcast and
1186 * uc_index is set. oif is most likely set
1187 * by sk_bound_dev_if. If uc_index != oif check if the
1188 * oif is an L3 master and uc_index is an L3 slave.
1189 * If so, we want to allow the send using the uc_index.
1191 if (ipc.oif != inet->uc_index &&
1192 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1194 ipc.oif = inet->uc_index;
1199 rt = (struct rtable *)sk_dst_check(sk, 0);
1202 struct net *net = sock_net(sk);
1203 __u8 flow_flags = inet_sk_flowi_flags(sk);
1207 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1208 sk->sk_protocol, flow_flags, faddr, saddr,
1209 dport, inet->inet_sport, sk->sk_uid);
1211 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1212 rt = ip_route_output_flow(net, fl4, sk);
1216 if (err == -ENETUNREACH)
1217 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1222 if ((rt->rt_flags & RTCF_BROADCAST) &&
1223 !sock_flag(sk, SOCK_BROADCAST))
1226 sk_dst_set(sk, dst_clone(&rt->dst));
1229 if (msg->msg_flags&MSG_CONFIRM)
1235 daddr = ipc.addr = fl4->daddr;
1237 /* Lockless fast path for the non-corking case. */
1239 struct inet_cork cork;
1241 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1242 sizeof(struct udphdr), &ipc, &rt,
1243 &cork, msg->msg_flags);
1245 if (!IS_ERR_OR_NULL(skb))
1246 err = udp_send_skb(skb, fl4, &cork);
1251 if (unlikely(up->pending)) {
1252 /* The socket is already corked while preparing it. */
1253 /* ... which is an evident application bug. --ANK */
1256 net_dbg_ratelimited("socket already corked\n");
1261 * Now cork the socket to pend data.
1263 fl4 = &inet->cork.fl.u.ip4;
1266 fl4->fl4_dport = dport;
1267 fl4->fl4_sport = inet->inet_sport;
1268 up->pending = AF_INET;
1272 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1273 sizeof(struct udphdr), &ipc, &rt,
1274 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1276 udp_flush_pending_frames(sk);
1278 err = udp_push_pending_frames(sk);
1279 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1291 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1292 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1293 * we don't have a good statistic (IpOutDiscards but it can be too many
1294 * things). We could add another new stat but at least for now that
1295 * seems like overkill.
1297 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1298 UDP_INC_STATS(sock_net(sk),
1299 UDP_MIB_SNDBUFERRORS, is_udplite);
1304 if (msg->msg_flags & MSG_PROBE)
1305 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1306 if (!(msg->msg_flags&MSG_PROBE) || len)
1307 goto back_from_confirm;
1311 EXPORT_SYMBOL(udp_sendmsg);
1313 void udp_splice_eof(struct socket *sock)
1315 struct sock *sk = sock->sk;
1316 struct udp_sock *up = udp_sk(sk);
1318 if (!up->pending || READ_ONCE(up->corkflag))
1322 if (up->pending && !READ_ONCE(up->corkflag))
1323 udp_push_pending_frames(sk);
1326 EXPORT_SYMBOL_GPL(udp_splice_eof);
1328 #define UDP_SKB_IS_STATELESS 0x80000000
1330 /* all head states (dst, sk, nf conntrack) except skb extensions are
1331 * cleared by udp_rcv().
1333 * We need to preserve secpath, if present, to eventually process
1334 * IP_CMSG_PASSSEC at recvmsg() time.
1336 * Other extensions can be cleared.
1338 static bool udp_try_make_stateless(struct sk_buff *skb)
1340 if (!skb_has_extensions(skb))
1343 if (!secpath_exists(skb)) {
1351 static void udp_set_dev_scratch(struct sk_buff *skb)
1353 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1355 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1356 scratch->_tsize_state = skb->truesize;
1357 #if BITS_PER_LONG == 64
1358 scratch->len = skb->len;
1359 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1360 scratch->is_linear = !skb_is_nonlinear(skb);
1362 if (udp_try_make_stateless(skb))
1363 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1366 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1368 /* We come here after udp_lib_checksum_complete() returned 0.
1369 * This means that __skb_checksum_complete() might have
1370 * set skb->csum_valid to 1.
1371 * On 64bit platforms, we can set csum_unnecessary
1372 * to true, but only if the skb is not shared.
1374 #if BITS_PER_LONG == 64
1375 if (!skb_shared(skb))
1376 udp_skb_scratch(skb)->csum_unnecessary = true;
1380 static int udp_skb_truesize(struct sk_buff *skb)
1382 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1385 static bool udp_skb_has_head_state(struct sk_buff *skb)
1387 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1390 /* fully reclaim rmem/fwd memory allocated for skb */
1391 static void udp_rmem_release(struct sock *sk, int size, int partial,
1392 bool rx_queue_lock_held)
1394 struct udp_sock *up = udp_sk(sk);
1395 struct sk_buff_head *sk_queue;
1398 if (likely(partial)) {
1399 up->forward_deficit += size;
1400 size = up->forward_deficit;
1401 if (size < READ_ONCE(up->forward_threshold) &&
1402 !skb_queue_empty(&up->reader_queue))
1405 size += up->forward_deficit;
1407 up->forward_deficit = 0;
1409 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1410 * if the called don't held it already
1412 sk_queue = &sk->sk_receive_queue;
1413 if (!rx_queue_lock_held)
1414 spin_lock(&sk_queue->lock);
1417 sk->sk_forward_alloc += size;
1418 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1419 sk->sk_forward_alloc -= amt;
1422 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1424 atomic_sub(size, &sk->sk_rmem_alloc);
1426 /* this can save us from acquiring the rx queue lock on next receive */
1427 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1429 if (!rx_queue_lock_held)
1430 spin_unlock(&sk_queue->lock);
1433 /* Note: called with reader_queue.lock held.
1434 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1435 * This avoids a cache line miss while receive_queue lock is held.
1436 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1438 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1440 prefetch(&skb->data);
1441 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1443 EXPORT_SYMBOL(udp_skb_destructor);
1445 /* as above, but the caller held the rx queue lock, too */
1446 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1448 prefetch(&skb->data);
1449 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1452 /* Idea of busylocks is to let producers grab an extra spinlock
1453 * to relieve pressure on the receive_queue spinlock shared by consumer.
1454 * Under flood, this means that only one producer can be in line
1455 * trying to acquire the receive_queue spinlock.
1456 * These busylock can be allocated on a per cpu manner, instead of a
1457 * per socket one (that would consume a cache line per socket)
1459 static int udp_busylocks_log __read_mostly;
1460 static spinlock_t *udp_busylocks __read_mostly;
1462 static spinlock_t *busylock_acquire(void *ptr)
1466 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1471 static void busylock_release(spinlock_t *busy)
1477 static int udp_rmem_schedule(struct sock *sk, int size)
1481 delta = size - sk->sk_forward_alloc;
1482 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1488 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1490 struct sk_buff_head *list = &sk->sk_receive_queue;
1491 int rmem, err = -ENOMEM;
1492 spinlock_t *busy = NULL;
1495 /* try to avoid the costly atomic add/sub pair when the receive
1496 * queue is full; always allow at least a packet
1498 rmem = atomic_read(&sk->sk_rmem_alloc);
1499 if (rmem > sk->sk_rcvbuf)
1502 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1503 * having linear skbs :
1504 * - Reduce memory overhead and thus increase receive queue capacity
1505 * - Less cache line misses at copyout() time
1506 * - Less work at consume_skb() (less alien page frag freeing)
1508 if (rmem > (sk->sk_rcvbuf >> 1)) {
1511 busy = busylock_acquire(sk);
1513 size = skb->truesize;
1514 udp_set_dev_scratch(skb);
1516 /* we drop only if the receive buf is full and the receive
1517 * queue contains some other skb
1519 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1520 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1523 spin_lock(&list->lock);
1524 err = udp_rmem_schedule(sk, size);
1526 spin_unlock(&list->lock);
1530 sk->sk_forward_alloc -= size;
1532 /* no need to setup a destructor, we will explicitly release the
1533 * forward allocated memory on dequeue
1535 sock_skb_set_dropcount(sk, skb);
1537 __skb_queue_tail(list, skb);
1538 spin_unlock(&list->lock);
1540 if (!sock_flag(sk, SOCK_DEAD))
1541 INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
1543 busylock_release(busy);
1547 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1550 atomic_inc(&sk->sk_drops);
1551 busylock_release(busy);
1554 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1556 void udp_destruct_common(struct sock *sk)
1558 /* reclaim completely the forward allocated memory */
1559 struct udp_sock *up = udp_sk(sk);
1560 unsigned int total = 0;
1561 struct sk_buff *skb;
1563 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1564 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1565 total += skb->truesize;
1568 udp_rmem_release(sk, total, 0, true);
1570 EXPORT_SYMBOL_GPL(udp_destruct_common);
1572 static void udp_destruct_sock(struct sock *sk)
1574 udp_destruct_common(sk);
1575 inet_sock_destruct(sk);
1578 int udp_init_sock(struct sock *sk)
1580 udp_lib_init_sock(sk);
1581 sk->sk_destruct = udp_destruct_sock;
1582 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1586 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1588 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1589 bool slow = lock_sock_fast(sk);
1591 sk_peek_offset_bwd(sk, len);
1592 unlock_sock_fast(sk, slow);
1595 if (!skb_unref(skb))
1598 /* In the more common cases we cleared the head states previously,
1599 * see __udp_queue_rcv_skb().
1601 if (unlikely(udp_skb_has_head_state(skb)))
1602 skb_release_head_state(skb);
1603 __consume_stateless_skb(skb);
1605 EXPORT_SYMBOL_GPL(skb_consume_udp);
1607 static struct sk_buff *__first_packet_length(struct sock *sk,
1608 struct sk_buff_head *rcvq,
1611 struct sk_buff *skb;
1613 while ((skb = skb_peek(rcvq)) != NULL) {
1614 if (udp_lib_checksum_complete(skb)) {
1615 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1617 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1619 atomic_inc(&sk->sk_drops);
1620 __skb_unlink(skb, rcvq);
1621 *total += skb->truesize;
1624 udp_skb_csum_unnecessary_set(skb);
1632 * first_packet_length - return length of first packet in receive queue
1635 * Drops all bad checksum frames, until a valid one is found.
1636 * Returns the length of found skb, or -1 if none is found.
1638 static int first_packet_length(struct sock *sk)
1640 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1641 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1642 struct sk_buff *skb;
1646 spin_lock_bh(&rcvq->lock);
1647 skb = __first_packet_length(sk, rcvq, &total);
1648 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1649 spin_lock(&sk_queue->lock);
1650 skb_queue_splice_tail_init(sk_queue, rcvq);
1651 spin_unlock(&sk_queue->lock);
1653 skb = __first_packet_length(sk, rcvq, &total);
1655 res = skb ? skb->len : -1;
1657 udp_rmem_release(sk, total, 1, false);
1658 spin_unlock_bh(&rcvq->lock);
1663 * IOCTL requests applicable to the UDP protocol
1666 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1671 *karg = sk_wmem_alloc_get(sk);
1677 *karg = max_t(int, 0, first_packet_length(sk));
1682 return -ENOIOCTLCMD;
1687 EXPORT_SYMBOL(udp_ioctl);
1689 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1692 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1693 struct sk_buff_head *queue;
1694 struct sk_buff *last;
1698 queue = &udp_sk(sk)->reader_queue;
1699 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1701 struct sk_buff *skb;
1703 error = sock_error(sk);
1709 spin_lock_bh(&queue->lock);
1710 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1713 if (!(flags & MSG_PEEK))
1714 udp_skb_destructor(sk, skb);
1715 spin_unlock_bh(&queue->lock);
1719 if (skb_queue_empty_lockless(sk_queue)) {
1720 spin_unlock_bh(&queue->lock);
1724 /* refill the reader queue and walk it again
1725 * keep both queues locked to avoid re-acquiring
1726 * the sk_receive_queue lock if fwd memory scheduling
1729 spin_lock(&sk_queue->lock);
1730 skb_queue_splice_tail_init(sk_queue, queue);
1732 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1734 if (skb && !(flags & MSG_PEEK))
1735 udp_skb_dtor_locked(sk, skb);
1736 spin_unlock(&sk_queue->lock);
1737 spin_unlock_bh(&queue->lock);
1742 if (!sk_can_busy_loop(sk))
1745 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1746 } while (!skb_queue_empty_lockless(sk_queue));
1748 /* sk_queue is empty, reader_queue may contain peeked packets */
1750 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1752 (struct sk_buff *)sk_queue));
1757 EXPORT_SYMBOL(__skb_recv_udp);
1759 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1761 struct sk_buff *skb;
1765 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1769 if (udp_lib_checksum_complete(skb)) {
1770 int is_udplite = IS_UDPLITE(sk);
1771 struct net *net = sock_net(sk);
1773 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1774 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1775 atomic_inc(&sk->sk_drops);
1780 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1781 return recv_actor(sk, skb);
1783 EXPORT_SYMBOL(udp_read_skb);
1786 * This should be easy, if there is something there we
1787 * return it, otherwise we block.
1790 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1793 struct inet_sock *inet = inet_sk(sk);
1794 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1795 struct sk_buff *skb;
1796 unsigned int ulen, copied;
1797 int off, err, peeking = flags & MSG_PEEK;
1798 int is_udplite = IS_UDPLITE(sk);
1799 bool checksum_valid = false;
1801 if (flags & MSG_ERRQUEUE)
1802 return ip_recv_error(sk, msg, len, addr_len);
1805 off = sk_peek_offset(sk, flags);
1806 skb = __skb_recv_udp(sk, flags, &off, &err);
1810 ulen = udp_skb_len(skb);
1812 if (copied > ulen - off)
1813 copied = ulen - off;
1814 else if (copied < ulen)
1815 msg->msg_flags |= MSG_TRUNC;
1818 * If checksum is needed at all, try to do it while copying the
1819 * data. If the data is truncated, or if we only want a partial
1820 * coverage checksum (UDP-Lite), do it before the copy.
1823 if (copied < ulen || peeking ||
1824 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1825 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1826 !__udp_lib_checksum_complete(skb);
1827 if (!checksum_valid)
1831 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1832 if (udp_skb_is_linear(skb))
1833 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1835 err = skb_copy_datagram_msg(skb, off, msg, copied);
1837 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1843 if (unlikely(err)) {
1845 atomic_inc(&sk->sk_drops);
1846 UDP_INC_STATS(sock_net(sk),
1847 UDP_MIB_INERRORS, is_udplite);
1854 UDP_INC_STATS(sock_net(sk),
1855 UDP_MIB_INDATAGRAMS, is_udplite);
1857 sock_recv_cmsgs(msg, sk, skb);
1859 /* Copy the address. */
1861 sin->sin_family = AF_INET;
1862 sin->sin_port = udp_hdr(skb)->source;
1863 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1864 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1865 *addr_len = sizeof(*sin);
1867 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1868 (struct sockaddr *)sin);
1871 if (udp_sk(sk)->gro_enabled)
1872 udp_cmsg_recv(msg, sk, skb);
1874 if (inet_cmsg_flags(inet))
1875 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1878 if (flags & MSG_TRUNC)
1881 skb_consume_udp(sk, skb, peeking ? -err : err);
1885 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1886 udp_skb_destructor)) {
1887 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1888 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1892 /* starting over for a new packet, but check if we need to yield */
1894 msg->msg_flags &= ~MSG_TRUNC;
1898 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1900 /* This check is replicated from __ip4_datagram_connect() and
1901 * intended to prevent BPF program called below from accessing bytes
1902 * that are out of the bound specified by user in addr_len.
1904 if (addr_len < sizeof(struct sockaddr_in))
1907 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1909 EXPORT_SYMBOL(udp_pre_connect);
1911 int __udp_disconnect(struct sock *sk, int flags)
1913 struct inet_sock *inet = inet_sk(sk);
1915 * 1003.1g - break association.
1918 sk->sk_state = TCP_CLOSE;
1919 inet->inet_daddr = 0;
1920 inet->inet_dport = 0;
1921 sock_rps_reset_rxhash(sk);
1922 sk->sk_bound_dev_if = 0;
1923 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1924 inet_reset_saddr(sk);
1925 if (sk->sk_prot->rehash &&
1926 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1927 sk->sk_prot->rehash(sk);
1930 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1931 sk->sk_prot->unhash(sk);
1932 inet->inet_sport = 0;
1937 EXPORT_SYMBOL(__udp_disconnect);
1939 int udp_disconnect(struct sock *sk, int flags)
1942 __udp_disconnect(sk, flags);
1946 EXPORT_SYMBOL(udp_disconnect);
1948 void udp_lib_unhash(struct sock *sk)
1950 if (sk_hashed(sk)) {
1951 struct udp_table *udptable = udp_get_table_prot(sk);
1952 struct udp_hslot *hslot, *hslot2;
1954 hslot = udp_hashslot(udptable, sock_net(sk),
1955 udp_sk(sk)->udp_port_hash);
1956 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1958 spin_lock_bh(&hslot->lock);
1959 if (rcu_access_pointer(sk->sk_reuseport_cb))
1960 reuseport_detach_sock(sk);
1961 if (sk_del_node_init_rcu(sk)) {
1963 inet_sk(sk)->inet_num = 0;
1964 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1966 spin_lock(&hslot2->lock);
1967 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1969 spin_unlock(&hslot2->lock);
1971 spin_unlock_bh(&hslot->lock);
1974 EXPORT_SYMBOL(udp_lib_unhash);
1977 * inet_rcv_saddr was changed, we must rehash secondary hash
1979 void udp_lib_rehash(struct sock *sk, u16 newhash)
1981 if (sk_hashed(sk)) {
1982 struct udp_table *udptable = udp_get_table_prot(sk);
1983 struct udp_hslot *hslot, *hslot2, *nhslot2;
1985 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1986 nhslot2 = udp_hashslot2(udptable, newhash);
1987 udp_sk(sk)->udp_portaddr_hash = newhash;
1989 if (hslot2 != nhslot2 ||
1990 rcu_access_pointer(sk->sk_reuseport_cb)) {
1991 hslot = udp_hashslot(udptable, sock_net(sk),
1992 udp_sk(sk)->udp_port_hash);
1993 /* we must lock primary chain too */
1994 spin_lock_bh(&hslot->lock);
1995 if (rcu_access_pointer(sk->sk_reuseport_cb))
1996 reuseport_detach_sock(sk);
1998 if (hslot2 != nhslot2) {
1999 spin_lock(&hslot2->lock);
2000 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2002 spin_unlock(&hslot2->lock);
2004 spin_lock(&nhslot2->lock);
2005 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2008 spin_unlock(&nhslot2->lock);
2011 spin_unlock_bh(&hslot->lock);
2015 EXPORT_SYMBOL(udp_lib_rehash);
2017 void udp_v4_rehash(struct sock *sk)
2019 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2020 inet_sk(sk)->inet_rcv_saddr,
2021 inet_sk(sk)->inet_num);
2022 udp_lib_rehash(sk, new_hash);
2025 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2029 if (inet_sk(sk)->inet_daddr) {
2030 sock_rps_save_rxhash(sk, skb);
2031 sk_mark_napi_id(sk, skb);
2032 sk_incoming_cpu_update(sk);
2034 sk_mark_napi_id_once(sk, skb);
2037 rc = __udp_enqueue_schedule_skb(sk, skb);
2039 int is_udplite = IS_UDPLITE(sk);
2042 /* Note that an ENOMEM error is charged twice */
2043 if (rc == -ENOMEM) {
2044 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2046 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2048 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2050 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2052 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2053 kfree_skb_reason(skb, drop_reason);
2054 trace_udp_fail_queue_rcv_skb(rc, sk);
2064 * >0: "udp encap" protocol resubmission
2066 * Note that in the success and error cases, the skb is assumed to
2067 * have either been requeued or freed.
2069 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2071 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2072 struct udp_sock *up = udp_sk(sk);
2073 int is_udplite = IS_UDPLITE(sk);
2076 * Charge it to the socket, dropping if the queue is full.
2078 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2079 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2084 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2085 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2088 * This is an encapsulation socket so pass the skb to
2089 * the socket's udp_encap_rcv() hook. Otherwise, just
2090 * fall through and pass this up the UDP socket.
2091 * up->encap_rcv() returns the following value:
2092 * =0 if skb was successfully passed to the encap
2093 * handler or was discarded by it.
2094 * >0 if skb should be passed on to UDP.
2095 * <0 if skb should be resubmitted as proto -N
2098 /* if we're overly short, let UDP handle it */
2099 encap_rcv = READ_ONCE(up->encap_rcv);
2103 /* Verify checksum before giving to encap */
2104 if (udp_lib_checksum_complete(skb))
2107 ret = encap_rcv(sk, skb);
2109 __UDP_INC_STATS(sock_net(sk),
2110 UDP_MIB_INDATAGRAMS,
2116 /* FALLTHROUGH -- it's a UDP Packet */
2120 * UDP-Lite specific tests, ignored on UDP sockets
2122 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2125 * MIB statistics other than incrementing the error count are
2126 * disabled for the following two types of errors: these depend
2127 * on the application settings, not on the functioning of the
2128 * protocol stack as such.
2130 * RFC 3828 here recommends (sec 3.3): "There should also be a
2131 * way ... to ... at least let the receiving application block
2132 * delivery of packets with coverage values less than a value
2133 * provided by the application."
2135 if (up->pcrlen == 0) { /* full coverage was set */
2136 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2137 UDP_SKB_CB(skb)->cscov, skb->len);
2140 /* The next case involves violating the min. coverage requested
2141 * by the receiver. This is subtle: if receiver wants x and x is
2142 * greater than the buffersize/MTU then receiver will complain
2143 * that it wants x while sender emits packets of smaller size y.
2144 * Therefore the above ...()->partial_cov statement is essential.
2146 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2147 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2148 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2153 prefetch(&sk->sk_rmem_alloc);
2154 if (rcu_access_pointer(sk->sk_filter) &&
2155 udp_lib_checksum_complete(skb))
2158 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2159 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2163 udp_csum_pull_header(skb);
2165 ipv4_pktinfo_prepare(sk, skb);
2166 return __udp_queue_rcv_skb(sk, skb);
2169 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2170 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2172 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2173 atomic_inc(&sk->sk_drops);
2174 kfree_skb_reason(skb, drop_reason);
2178 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2180 struct sk_buff *next, *segs;
2183 if (likely(!udp_unexpected_gso(sk, skb)))
2184 return udp_queue_rcv_one_skb(sk, skb);
2186 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2187 __skb_push(skb, -skb_mac_offset(skb));
2188 segs = udp_rcv_segment(sk, skb, true);
2189 skb_list_walk_safe(segs, skb, next) {
2190 __skb_pull(skb, skb_transport_offset(skb));
2192 udp_post_segment_fix_csum(skb);
2193 ret = udp_queue_rcv_one_skb(sk, skb);
2195 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2200 /* For TCP sockets, sk_rx_dst is protected by socket lock
2201 * For UDP, we use xchg() to guard against concurrent changes.
2203 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2205 struct dst_entry *old;
2207 if (dst_hold_safe(dst)) {
2208 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2214 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2217 * Multicasts and broadcasts go to each listener.
2219 * Note: called only from the BH handler context.
2221 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2223 __be32 saddr, __be32 daddr,
2224 struct udp_table *udptable,
2227 struct sock *sk, *first = NULL;
2228 unsigned short hnum = ntohs(uh->dest);
2229 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2230 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2231 unsigned int offset = offsetof(typeof(*sk), sk_node);
2232 int dif = skb->dev->ifindex;
2233 int sdif = inet_sdif(skb);
2234 struct hlist_node *node;
2235 struct sk_buff *nskb;
2238 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2240 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2242 hslot = &udptable->hash2[hash2];
2243 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2246 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2247 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2248 uh->source, saddr, dif, sdif, hnum))
2255 nskb = skb_clone(skb, GFP_ATOMIC);
2257 if (unlikely(!nskb)) {
2258 atomic_inc(&sk->sk_drops);
2259 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2261 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2265 if (udp_queue_rcv_skb(sk, nskb) > 0)
2269 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2270 if (use_hash2 && hash2 != hash2_any) {
2276 if (udp_queue_rcv_skb(first, skb) > 0)
2280 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2281 proto == IPPROTO_UDPLITE);
2286 /* Initialize UDP checksum. If exited with zero value (success),
2287 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2288 * Otherwise, csum completion requires checksumming packet body,
2289 * including udp header and folding it to skb->csum.
2291 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2296 UDP_SKB_CB(skb)->partial_cov = 0;
2297 UDP_SKB_CB(skb)->cscov = skb->len;
2299 if (proto == IPPROTO_UDPLITE) {
2300 err = udplite_checksum_init(skb, uh);
2304 if (UDP_SKB_CB(skb)->partial_cov) {
2305 skb->csum = inet_compute_pseudo(skb, proto);
2310 /* Note, we are only interested in != 0 or == 0, thus the
2313 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2314 inet_compute_pseudo);
2318 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2319 /* If SW calculated the value, we know it's bad */
2320 if (skb->csum_complete_sw)
2323 /* HW says the value is bad. Let's validate that.
2324 * skb->csum is no longer the full packet checksum,
2325 * so don't treat it as such.
2327 skb_checksum_complete_unset(skb);
2333 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2334 * return code conversion for ip layer consumption
2336 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2341 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2342 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2344 ret = udp_queue_rcv_skb(sk, skb);
2346 /* a return value > 0 means to resubmit the input, but
2347 * it wants the return to be -protocol, or 0
2355 * All we need to do is get the socket, and then do a checksum.
2358 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2363 unsigned short ulen;
2364 struct rtable *rt = skb_rtable(skb);
2365 __be32 saddr, daddr;
2366 struct net *net = dev_net(skb->dev);
2370 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2373 * Validate the packet.
2375 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2376 goto drop; /* No space for header. */
2379 ulen = ntohs(uh->len);
2380 saddr = ip_hdr(skb)->saddr;
2381 daddr = ip_hdr(skb)->daddr;
2383 if (ulen > skb->len)
2386 if (proto == IPPROTO_UDP) {
2387 /* UDP validates ulen. */
2388 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2393 if (udp4_csum_init(skb, uh, proto))
2396 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2397 &refcounted, udp_ehashfn);
2402 struct dst_entry *dst = skb_dst(skb);
2405 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2406 udp_sk_rx_dst_set(sk, dst);
2408 ret = udp_unicast_rcv_skb(sk, skb, uh);
2414 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2415 return __udp4_lib_mcast_deliver(net, skb, uh,
2416 saddr, daddr, udptable, proto);
2418 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2420 return udp_unicast_rcv_skb(sk, skb, uh);
2422 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2426 /* No socket. Drop packet silently, if checksum is wrong */
2427 if (udp_lib_checksum_complete(skb))
2430 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2431 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2432 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2435 * Hmm. We got an UDP packet to a port to which we
2436 * don't wanna listen. Ignore it.
2438 kfree_skb_reason(skb, drop_reason);
2442 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2443 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2444 proto == IPPROTO_UDPLITE ? "Lite" : "",
2445 &saddr, ntohs(uh->source),
2447 &daddr, ntohs(uh->dest));
2452 * RFC1122: OK. Discards the bad packet silently (as far as
2453 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2455 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2456 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2457 proto == IPPROTO_UDPLITE ? "Lite" : "",
2458 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2460 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2462 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2463 kfree_skb_reason(skb, drop_reason);
2467 /* We can only early demux multicast if there is a single matching socket.
2468 * If more than one socket found returns NULL
2470 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2471 __be16 loc_port, __be32 loc_addr,
2472 __be16 rmt_port, __be32 rmt_addr,
2475 struct udp_table *udptable = net->ipv4.udp_table;
2476 unsigned short hnum = ntohs(loc_port);
2477 struct sock *sk, *result;
2478 struct udp_hslot *hslot;
2481 slot = udp_hashfn(net, hnum, udptable->mask);
2482 hslot = &udptable->hash[slot];
2484 /* Do not bother scanning a too big list */
2485 if (hslot->count > 10)
2489 sk_for_each_rcu(sk, &hslot->head) {
2490 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2491 rmt_port, rmt_addr, dif, sdif, hnum)) {
2501 /* For unicast we should only early demux connected sockets or we can
2502 * break forwarding setups. The chains here can be long so only check
2503 * if the first socket is an exact match and if not move on.
2505 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2506 __be16 loc_port, __be32 loc_addr,
2507 __be16 rmt_port, __be32 rmt_addr,
2510 struct udp_table *udptable = net->ipv4.udp_table;
2511 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2512 unsigned short hnum = ntohs(loc_port);
2513 unsigned int hash2, slot2;
2514 struct udp_hslot *hslot2;
2518 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2519 slot2 = hash2 & udptable->mask;
2520 hslot2 = &udptable->hash2[slot2];
2521 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2523 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2524 if (inet_match(net, sk, acookie, ports, dif, sdif))
2526 /* Only check first socket in chain */
2532 int udp_v4_early_demux(struct sk_buff *skb)
2534 struct net *net = dev_net(skb->dev);
2535 struct in_device *in_dev = NULL;
2536 const struct iphdr *iph;
2537 const struct udphdr *uh;
2538 struct sock *sk = NULL;
2539 struct dst_entry *dst;
2540 int dif = skb->dev->ifindex;
2541 int sdif = inet_sdif(skb);
2544 /* validate the packet */
2545 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2551 if (skb->pkt_type == PACKET_MULTICAST) {
2552 in_dev = __in_dev_get_rcu(skb->dev);
2557 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2562 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2563 uh->source, iph->saddr,
2565 } else if (skb->pkt_type == PACKET_HOST) {
2566 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2567 uh->source, iph->saddr, dif, sdif);
2570 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2574 skb->destructor = sock_efree;
2575 dst = rcu_dereference(sk->sk_rx_dst);
2578 dst = dst_check(dst, 0);
2582 /* set noref for now.
2583 * any place which wants to hold dst has to call
2586 skb_dst_set_noref(skb, dst);
2588 /* for unconnected multicast sockets we need to validate
2589 * the source on each packet
2591 if (!inet_sk(sk)->inet_daddr && in_dev)
2592 return ip_mc_validate_source(skb, iph->daddr,
2594 iph->tos & IPTOS_RT_MASK,
2595 skb->dev, in_dev, &itag);
2600 int udp_rcv(struct sk_buff *skb)
2602 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2605 void udp_destroy_sock(struct sock *sk)
2607 struct udp_sock *up = udp_sk(sk);
2608 bool slow = lock_sock_fast(sk);
2610 /* protects from races with udp_abort() */
2611 sock_set_flag(sk, SOCK_DEAD);
2612 udp_flush_pending_frames(sk);
2613 unlock_sock_fast(sk, slow);
2614 if (static_branch_unlikely(&udp_encap_needed_key)) {
2615 if (up->encap_type) {
2616 void (*encap_destroy)(struct sock *sk);
2617 encap_destroy = READ_ONCE(up->encap_destroy);
2621 if (up->encap_enabled)
2622 static_branch_dec(&udp_encap_needed_key);
2627 * Socket option code for UDP
2629 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2630 sockptr_t optval, unsigned int optlen,
2631 int (*push_pending_frames)(struct sock *))
2633 struct udp_sock *up = udp_sk(sk);
2636 int is_udplite = IS_UDPLITE(sk);
2638 if (level == SOL_SOCKET) {
2639 err = sk_setsockopt(sk, level, optname, optval, optlen);
2641 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2642 sockopt_lock_sock(sk);
2643 /* paired with READ_ONCE in udp_rmem_release() */
2644 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2645 sockopt_release_sock(sk);
2650 if (optlen < sizeof(int))
2653 if (copy_from_sockptr(&val, optval, sizeof(val)))
2656 valbool = val ? 1 : 0;
2661 WRITE_ONCE(up->corkflag, 1);
2663 WRITE_ONCE(up->corkflag, 0);
2665 push_pending_frames(sk);
2674 case UDP_ENCAP_ESPINUDP:
2675 case UDP_ENCAP_ESPINUDP_NON_IKE:
2676 #if IS_ENABLED(CONFIG_IPV6)
2677 if (sk->sk_family == AF_INET6)
2678 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2681 up->encap_rcv = xfrm4_udp_encap_rcv;
2684 case UDP_ENCAP_L2TPINUDP:
2685 up->encap_type = val;
2687 udp_tunnel_encap_enable(sk->sk_socket);
2696 case UDP_NO_CHECK6_TX:
2697 up->no_check6_tx = valbool;
2700 case UDP_NO_CHECK6_RX:
2701 up->no_check6_rx = valbool;
2705 if (val < 0 || val > USHRT_MAX)
2707 WRITE_ONCE(up->gso_size, val);
2713 /* when enabling GRO, accept the related GSO packet type */
2715 udp_tunnel_encap_enable(sk->sk_socket);
2716 up->gro_enabled = valbool;
2717 up->accept_udp_l4 = valbool;
2722 * UDP-Lite's partial checksum coverage (RFC 3828).
2724 /* The sender sets actual checksum coverage length via this option.
2725 * The case coverage > packet length is handled by send module. */
2726 case UDPLITE_SEND_CSCOV:
2727 if (!is_udplite) /* Disable the option on UDP sockets */
2728 return -ENOPROTOOPT;
2729 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2731 else if (val > USHRT_MAX)
2734 up->pcflag |= UDPLITE_SEND_CC;
2737 /* The receiver specifies a minimum checksum coverage value. To make
2738 * sense, this should be set to at least 8 (as done below). If zero is
2739 * used, this again means full checksum coverage. */
2740 case UDPLITE_RECV_CSCOV:
2741 if (!is_udplite) /* Disable the option on UDP sockets */
2742 return -ENOPROTOOPT;
2743 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2745 else if (val > USHRT_MAX)
2748 up->pcflag |= UDPLITE_RECV_CC;
2758 EXPORT_SYMBOL(udp_lib_setsockopt);
2760 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2761 unsigned int optlen)
2763 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2764 return udp_lib_setsockopt(sk, level, optname,
2766 udp_push_pending_frames);
2767 return ip_setsockopt(sk, level, optname, optval, optlen);
2770 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2771 char __user *optval, int __user *optlen)
2773 struct udp_sock *up = udp_sk(sk);
2776 if (get_user(len, optlen))
2779 len = min_t(unsigned int, len, sizeof(int));
2786 val = READ_ONCE(up->corkflag);
2790 val = up->encap_type;
2793 case UDP_NO_CHECK6_TX:
2794 val = up->no_check6_tx;
2797 case UDP_NO_CHECK6_RX:
2798 val = up->no_check6_rx;
2802 val = READ_ONCE(up->gso_size);
2806 val = up->gro_enabled;
2809 /* The following two cannot be changed on UDP sockets, the return is
2810 * always 0 (which corresponds to the full checksum coverage of UDP). */
2811 case UDPLITE_SEND_CSCOV:
2815 case UDPLITE_RECV_CSCOV:
2820 return -ENOPROTOOPT;
2823 if (put_user(len, optlen))
2825 if (copy_to_user(optval, &val, len))
2829 EXPORT_SYMBOL(udp_lib_getsockopt);
2831 int udp_getsockopt(struct sock *sk, int level, int optname,
2832 char __user *optval, int __user *optlen)
2834 if (level == SOL_UDP || level == SOL_UDPLITE)
2835 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2836 return ip_getsockopt(sk, level, optname, optval, optlen);
2840 * udp_poll - wait for a UDP event.
2841 * @file: - file struct
2843 * @wait: - poll table
2845 * This is same as datagram poll, except for the special case of
2846 * blocking sockets. If application is using a blocking fd
2847 * and a packet with checksum error is in the queue;
2848 * then it could get return from select indicating data available
2849 * but then block when reading it. Add special case code
2850 * to work around these arguably broken applications.
2852 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2854 __poll_t mask = datagram_poll(file, sock, wait);
2855 struct sock *sk = sock->sk;
2857 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2858 mask |= EPOLLIN | EPOLLRDNORM;
2860 /* Check for false positives due to checksum errors */
2861 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2862 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2863 mask &= ~(EPOLLIN | EPOLLRDNORM);
2865 /* psock ingress_msg queue should not contain any bad checksum frames */
2866 if (sk_is_readable(sk))
2867 mask |= EPOLLIN | EPOLLRDNORM;
2871 EXPORT_SYMBOL(udp_poll);
2873 int udp_abort(struct sock *sk, int err)
2875 if (!has_current_bpf_ctx())
2878 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2881 if (sock_flag(sk, SOCK_DEAD))
2885 sk_error_report(sk);
2886 __udp_disconnect(sk, 0);
2889 if (!has_current_bpf_ctx())
2894 EXPORT_SYMBOL_GPL(udp_abort);
2896 struct proto udp_prot = {
2898 .owner = THIS_MODULE,
2899 .close = udp_lib_close,
2900 .pre_connect = udp_pre_connect,
2901 .connect = ip4_datagram_connect,
2902 .disconnect = udp_disconnect,
2904 .init = udp_init_sock,
2905 .destroy = udp_destroy_sock,
2906 .setsockopt = udp_setsockopt,
2907 .getsockopt = udp_getsockopt,
2908 .sendmsg = udp_sendmsg,
2909 .recvmsg = udp_recvmsg,
2910 .splice_eof = udp_splice_eof,
2911 .release_cb = ip4_datagram_release_cb,
2912 .hash = udp_lib_hash,
2913 .unhash = udp_lib_unhash,
2914 .rehash = udp_v4_rehash,
2915 .get_port = udp_v4_get_port,
2916 .put_port = udp_lib_unhash,
2917 #ifdef CONFIG_BPF_SYSCALL
2918 .psock_update_sk_prot = udp_bpf_update_proto,
2920 .memory_allocated = &udp_memory_allocated,
2921 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2923 .sysctl_mem = sysctl_udp_mem,
2924 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2925 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2926 .obj_size = sizeof(struct udp_sock),
2927 .h.udp_table = NULL,
2928 .diag_destroy = udp_abort,
2930 EXPORT_SYMBOL(udp_prot);
2932 /* ------------------------------------------------------------------------ */
2933 #ifdef CONFIG_PROC_FS
2935 static unsigned short seq_file_family(const struct seq_file *seq);
2936 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2938 unsigned short family = seq_file_family(seq);
2940 /* AF_UNSPEC is used as a match all */
2941 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2942 net_eq(sock_net(sk), seq_file_net(seq)));
2945 #ifdef CONFIG_BPF_SYSCALL
2946 static const struct seq_operations bpf_iter_udp_seq_ops;
2948 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2951 const struct udp_seq_afinfo *afinfo;
2953 #ifdef CONFIG_BPF_SYSCALL
2954 if (seq->op == &bpf_iter_udp_seq_ops)
2955 return net->ipv4.udp_table;
2958 afinfo = pde_data(file_inode(seq->file));
2959 return afinfo->udp_table ? : net->ipv4.udp_table;
2962 static struct sock *udp_get_first(struct seq_file *seq, int start)
2964 struct udp_iter_state *state = seq->private;
2965 struct net *net = seq_file_net(seq);
2966 struct udp_table *udptable;
2969 udptable = udp_get_table_seq(seq, net);
2971 for (state->bucket = start; state->bucket <= udptable->mask;
2973 struct udp_hslot *hslot = &udptable->hash[state->bucket];
2975 if (hlist_empty(&hslot->head))
2978 spin_lock_bh(&hslot->lock);
2979 sk_for_each(sk, &hslot->head) {
2980 if (seq_sk_match(seq, sk))
2983 spin_unlock_bh(&hslot->lock);
2990 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2992 struct udp_iter_state *state = seq->private;
2993 struct net *net = seq_file_net(seq);
2994 struct udp_table *udptable;
2998 } while (sk && !seq_sk_match(seq, sk));
3001 udptable = udp_get_table_seq(seq, net);
3003 if (state->bucket <= udptable->mask)
3004 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3006 return udp_get_first(seq, state->bucket + 1);
3011 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3013 struct sock *sk = udp_get_first(seq, 0);
3016 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3018 return pos ? NULL : sk;
3021 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3023 struct udp_iter_state *state = seq->private;
3024 state->bucket = MAX_UDP_PORTS;
3026 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3028 EXPORT_SYMBOL(udp_seq_start);
3030 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3034 if (v == SEQ_START_TOKEN)
3035 sk = udp_get_idx(seq, 0);
3037 sk = udp_get_next(seq, v);
3042 EXPORT_SYMBOL(udp_seq_next);
3044 void udp_seq_stop(struct seq_file *seq, void *v)
3046 struct udp_iter_state *state = seq->private;
3047 struct udp_table *udptable;
3049 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3051 if (state->bucket <= udptable->mask)
3052 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3054 EXPORT_SYMBOL(udp_seq_stop);
3056 /* ------------------------------------------------------------------------ */
3057 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3060 struct inet_sock *inet = inet_sk(sp);
3061 __be32 dest = inet->inet_daddr;
3062 __be32 src = inet->inet_rcv_saddr;
3063 __u16 destp = ntohs(inet->inet_dport);
3064 __u16 srcp = ntohs(inet->inet_sport);
3066 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3067 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3068 bucket, src, srcp, dest, destp, sp->sk_state,
3069 sk_wmem_alloc_get(sp),
3072 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3074 refcount_read(&sp->sk_refcnt), sp,
3075 atomic_read(&sp->sk_drops));
3078 int udp4_seq_show(struct seq_file *seq, void *v)
3080 seq_setwidth(seq, 127);
3081 if (v == SEQ_START_TOKEN)
3082 seq_puts(seq, " sl local_address rem_address st tx_queue "
3083 "rx_queue tr tm->when retrnsmt uid timeout "
3084 "inode ref pointer drops");
3086 struct udp_iter_state *state = seq->private;
3088 udp4_format_sock(v, seq, state->bucket);
3094 #ifdef CONFIG_BPF_SYSCALL
3095 struct bpf_iter__udp {
3096 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3097 __bpf_md_ptr(struct udp_sock *, udp_sk);
3098 uid_t uid __aligned(8);
3099 int bucket __aligned(8);
3102 struct bpf_udp_iter_state {
3103 struct udp_iter_state state;
3104 unsigned int cur_sk;
3105 unsigned int end_sk;
3106 unsigned int max_sk;
3108 struct sock **batch;
3109 bool st_bucket_done;
3112 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3113 unsigned int new_batch_sz);
3114 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3116 struct bpf_udp_iter_state *iter = seq->private;
3117 struct udp_iter_state *state = &iter->state;
3118 struct net *net = seq_file_net(seq);
3119 struct udp_table *udptable;
3120 unsigned int batch_sks = 0;
3121 bool resized = false;
3124 /* The current batch is done, so advance the bucket. */
3125 if (iter->st_bucket_done) {
3130 udptable = udp_get_table_seq(seq, net);
3133 /* New batch for the next bucket.
3134 * Iterate over the hash table to find a bucket with sockets matching
3135 * the iterator attributes, and return the first matching socket from
3136 * the bucket. The remaining matched sockets from the bucket are batched
3137 * before releasing the bucket lock. This allows BPF programs that are
3138 * called in seq_show to acquire the bucket lock if needed.
3142 iter->st_bucket_done = false;
3145 for (; state->bucket <= udptable->mask; state->bucket++) {
3146 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3148 if (hlist_empty(&hslot2->head)) {
3153 spin_lock_bh(&hslot2->lock);
3154 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3155 if (seq_sk_match(seq, sk)) {
3156 /* Resume from the last iterated socket at the
3157 * offset in the bucket before iterator was stopped.
3163 if (iter->end_sk < iter->max_sk) {
3165 iter->batch[iter->end_sk++] = sk;
3170 spin_unlock_bh(&hslot2->lock);
3175 /* Reset the current bucket's offset before moving to the next bucket. */
3179 /* All done: no batch made. */
3183 if (iter->end_sk == batch_sks) {
3184 /* Batching is done for the current bucket; return the first
3185 * socket to be iterated from the batch.
3187 iter->st_bucket_done = true;
3190 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3192 /* After allocating a larger batch, retry one more time to grab
3199 return iter->batch[0];
3202 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3204 struct bpf_udp_iter_state *iter = seq->private;
3207 /* Whenever seq_next() is called, the iter->cur_sk is
3208 * done with seq_show(), so unref the iter->cur_sk.
3210 if (iter->cur_sk < iter->end_sk) {
3211 sock_put(iter->batch[iter->cur_sk++]);
3215 /* After updating iter->cur_sk, check if there are more sockets
3216 * available in the current bucket batch.
3218 if (iter->cur_sk < iter->end_sk)
3219 sk = iter->batch[iter->cur_sk];
3221 /* Prepare a new batch. */
3222 sk = bpf_iter_udp_batch(seq);
3228 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3230 /* bpf iter does not support lseek, so it always
3231 * continue from where it was stop()-ped.
3234 return bpf_iter_udp_batch(seq);
3236 return SEQ_START_TOKEN;
3239 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3240 struct udp_sock *udp_sk, uid_t uid, int bucket)
3242 struct bpf_iter__udp ctx;
3244 meta->seq_num--; /* skip SEQ_START_TOKEN */
3246 ctx.udp_sk = udp_sk;
3248 ctx.bucket = bucket;
3249 return bpf_iter_run_prog(prog, &ctx);
3252 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3254 struct udp_iter_state *state = seq->private;
3255 struct bpf_iter_meta meta;
3256 struct bpf_prog *prog;
3257 struct sock *sk = v;
3261 if (v == SEQ_START_TOKEN)
3266 if (unlikely(sk_unhashed(sk))) {
3271 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3273 prog = bpf_iter_get_info(&meta, false);
3274 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3281 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3283 while (iter->cur_sk < iter->end_sk)
3284 sock_put(iter->batch[iter->cur_sk++]);
3287 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3289 struct bpf_udp_iter_state *iter = seq->private;
3290 struct bpf_iter_meta meta;
3291 struct bpf_prog *prog;
3295 prog = bpf_iter_get_info(&meta, true);
3297 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3300 if (iter->cur_sk < iter->end_sk) {
3301 bpf_iter_udp_put_batch(iter);
3302 iter->st_bucket_done = false;
3306 static const struct seq_operations bpf_iter_udp_seq_ops = {
3307 .start = bpf_iter_udp_seq_start,
3308 .next = bpf_iter_udp_seq_next,
3309 .stop = bpf_iter_udp_seq_stop,
3310 .show = bpf_iter_udp_seq_show,
3314 static unsigned short seq_file_family(const struct seq_file *seq)
3316 const struct udp_seq_afinfo *afinfo;
3318 #ifdef CONFIG_BPF_SYSCALL
3319 /* BPF iterator: bpf programs to filter sockets. */
3320 if (seq->op == &bpf_iter_udp_seq_ops)
3324 /* Proc fs iterator */
3325 afinfo = pde_data(file_inode(seq->file));
3326 return afinfo->family;
3329 const struct seq_operations udp_seq_ops = {
3330 .start = udp_seq_start,
3331 .next = udp_seq_next,
3332 .stop = udp_seq_stop,
3333 .show = udp4_seq_show,
3335 EXPORT_SYMBOL(udp_seq_ops);
3337 static struct udp_seq_afinfo udp4_seq_afinfo = {
3342 static int __net_init udp4_proc_init_net(struct net *net)
3344 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3345 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3350 static void __net_exit udp4_proc_exit_net(struct net *net)
3352 remove_proc_entry("udp", net->proc_net);
3355 static struct pernet_operations udp4_net_ops = {
3356 .init = udp4_proc_init_net,
3357 .exit = udp4_proc_exit_net,
3360 int __init udp4_proc_init(void)
3362 return register_pernet_subsys(&udp4_net_ops);
3365 void udp4_proc_exit(void)
3367 unregister_pernet_subsys(&udp4_net_ops);
3369 #endif /* CONFIG_PROC_FS */
3371 static __initdata unsigned long uhash_entries;
3372 static int __init set_uhash_entries(char *str)
3379 ret = kstrtoul(str, 0, &uhash_entries);
3383 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3384 uhash_entries = UDP_HTABLE_SIZE_MIN;
3387 __setup("uhash_entries=", set_uhash_entries);
3389 void __init udp_table_init(struct udp_table *table, const char *name)
3393 table->hash = alloc_large_system_hash(name,
3394 2 * sizeof(struct udp_hslot),
3396 21, /* one slot per 2 MB */
3400 UDP_HTABLE_SIZE_MIN,
3401 UDP_HTABLE_SIZE_MAX);
3403 table->hash2 = table->hash + (table->mask + 1);
3404 for (i = 0; i <= table->mask; i++) {
3405 INIT_HLIST_HEAD(&table->hash[i].head);
3406 table->hash[i].count = 0;
3407 spin_lock_init(&table->hash[i].lock);
3409 for (i = 0; i <= table->mask; i++) {
3410 INIT_HLIST_HEAD(&table->hash2[i].head);
3411 table->hash2[i].count = 0;
3412 spin_lock_init(&table->hash2[i].lock);
3416 u32 udp_flow_hashrnd(void)
3418 static u32 hashrnd __read_mostly;
3420 net_get_random_once(&hashrnd, sizeof(hashrnd));
3424 EXPORT_SYMBOL(udp_flow_hashrnd);
3426 static void __net_init udp_sysctl_init(struct net *net)
3428 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3429 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3431 #ifdef CONFIG_NET_L3_MASTER_DEV
3432 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3436 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3438 struct udp_table *udptable;
3441 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3445 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3446 GFP_KERNEL_ACCOUNT);
3447 if (!udptable->hash)
3450 udptable->hash2 = udptable->hash + hash_entries;
3451 udptable->mask = hash_entries - 1;
3452 udptable->log = ilog2(hash_entries);
3454 for (i = 0; i < hash_entries; i++) {
3455 INIT_HLIST_HEAD(&udptable->hash[i].head);
3456 udptable->hash[i].count = 0;
3457 spin_lock_init(&udptable->hash[i].lock);
3459 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3460 udptable->hash2[i].count = 0;
3461 spin_lock_init(&udptable->hash2[i].lock);
3472 static void __net_exit udp_pernet_table_free(struct net *net)
3474 struct udp_table *udptable = net->ipv4.udp_table;
3476 if (udptable == &udp_table)
3479 kvfree(udptable->hash);
3483 static void __net_init udp_set_table(struct net *net)
3485 struct udp_table *udptable;
3486 unsigned int hash_entries;
3487 struct net *old_net;
3489 if (net_eq(net, &init_net))
3492 old_net = current->nsproxy->net_ns;
3493 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3497 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3498 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3499 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3501 hash_entries = roundup_pow_of_two(hash_entries);
3503 udptable = udp_pernet_table_alloc(hash_entries);
3505 net->ipv4.udp_table = udptable;
3507 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3508 "for a netns, fallback to the global one\n",
3511 net->ipv4.udp_table = &udp_table;
3515 static int __net_init udp_pernet_init(struct net *net)
3517 udp_sysctl_init(net);
3523 static void __net_exit udp_pernet_exit(struct net *net)
3525 udp_pernet_table_free(net);
3528 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3529 .init = udp_pernet_init,
3530 .exit = udp_pernet_exit,
3533 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3534 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3535 struct udp_sock *udp_sk, uid_t uid, int bucket)
3537 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3538 unsigned int new_batch_sz)
3540 struct sock **new_batch;
3542 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3543 GFP_USER | __GFP_NOWARN);
3547 bpf_iter_udp_put_batch(iter);
3548 kvfree(iter->batch);
3549 iter->batch = new_batch;
3550 iter->max_sk = new_batch_sz;
3555 #define INIT_BATCH_SZ 16
3557 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3559 struct bpf_udp_iter_state *iter = priv_data;
3562 ret = bpf_iter_init_seq_net(priv_data, aux);
3566 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3568 bpf_iter_fini_seq_net(priv_data);
3573 static void bpf_iter_fini_udp(void *priv_data)
3575 struct bpf_udp_iter_state *iter = priv_data;
3577 bpf_iter_fini_seq_net(priv_data);
3578 kvfree(iter->batch);
3581 static const struct bpf_iter_seq_info udp_seq_info = {
3582 .seq_ops = &bpf_iter_udp_seq_ops,
3583 .init_seq_private = bpf_iter_init_udp,
3584 .fini_seq_private = bpf_iter_fini_udp,
3585 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3588 static struct bpf_iter_reg udp_reg_info = {
3590 .ctx_arg_info_size = 1,
3592 { offsetof(struct bpf_iter__udp, udp_sk),
3593 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3595 .seq_info = &udp_seq_info,
3598 static void __init bpf_iter_register(void)
3600 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3601 if (bpf_iter_reg_target(&udp_reg_info))
3602 pr_warn("Warning: could not register bpf iterator udp\n");
3606 void __init udp_init(void)
3608 unsigned long limit;
3611 udp_table_init(&udp_table, "UDP");
3612 limit = nr_free_buffer_pages() / 8;
3613 limit = max(limit, 128UL);
3614 sysctl_udp_mem[0] = limit / 4 * 3;
3615 sysctl_udp_mem[1] = limit;
3616 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3618 /* 16 spinlocks per cpu */
3619 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3620 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3623 panic("UDP: failed to alloc udp_busylocks\n");
3624 for (i = 0; i < (1U << udp_busylocks_log); i++)
3625 spin_lock_init(udp_busylocks + i);
3627 if (register_pernet_subsys(&udp_sysctl_ops))
3628 panic("UDP: failed to init sysctl parameters.\n");
3630 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3631 bpf_iter_register();