1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * The User Datagram Protocol (UDP).
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
72 * James Chapman : Add L2TP encapsulation type.
75 #define pr_fmt(fmt) "UDP: " fmt
77 #include <linux/bpf-cgroup.h>
78 #include <linux/uaccess.h>
79 #include <asm/ioctls.h>
80 #include <linux/memblock.h>
81 #include <linux/highmem.h>
82 #include <linux/types.h>
83 #include <linux/fcntl.h>
84 #include <linux/module.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/igmp.h>
88 #include <linux/inetdevice.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
93 #include <linux/inet.h>
94 #include <linux/netdevice.h>
95 #include <linux/slab.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/inet_hashtables.h>
103 #include <net/ip_tunnels.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <trace/events/udp.h>
109 #include <linux/static_key.h>
110 #include <linux/btf_ids.h>
111 #include <trace/events/skb.h>
112 #include <net/busy_poll.h>
113 #include "udp_impl.h"
114 #include <net/sock_reuseport.h>
115 #include <net/addrconf.h>
116 #include <net/udp_tunnel.h>
118 #if IS_ENABLED(CONFIG_IPV6)
119 #include <net/ipv6_stubs.h>
122 struct udp_table udp_table __read_mostly;
123 EXPORT_SYMBOL(udp_table);
125 long sysctl_udp_mem[3] __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_mem);
128 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
129 EXPORT_SYMBOL(udp_memory_allocated);
130 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
131 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
133 #define MAX_UDP_PORTS 65536
134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
136 static struct udp_table *udp_get_table_prot(struct sock *sk)
138 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
141 static int udp_lib_lport_inuse(struct net *net, __u16 num,
142 const struct udp_hslot *hslot,
143 unsigned long *bitmap,
144 struct sock *sk, unsigned int log)
147 kuid_t uid = sock_i_uid(sk);
149 sk_for_each(sk2, &hslot->head) {
150 if (net_eq(sock_net(sk2), net) &&
152 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
153 (!sk2->sk_reuse || !sk->sk_reuse) &&
154 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
155 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
156 inet_rcv_saddr_equal(sk, sk2, true)) {
157 if (sk2->sk_reuseport && sk->sk_reuseport &&
158 !rcu_access_pointer(sk->sk_reuseport_cb) &&
159 uid_eq(uid, sock_i_uid(sk2))) {
165 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
174 * Note: we still hold spinlock of primary hash chain, so no other writer
175 * can insert/delete a socket with local_port == num
177 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
178 struct udp_hslot *hslot2,
182 kuid_t uid = sock_i_uid(sk);
185 spin_lock(&hslot2->lock);
186 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
187 if (net_eq(sock_net(sk2), net) &&
189 (udp_sk(sk2)->udp_port_hash == num) &&
190 (!sk2->sk_reuse || !sk->sk_reuse) &&
191 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
192 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
193 inet_rcv_saddr_equal(sk, sk2, true)) {
194 if (sk2->sk_reuseport && sk->sk_reuseport &&
195 !rcu_access_pointer(sk->sk_reuseport_cb) &&
196 uid_eq(uid, sock_i_uid(sk2))) {
204 spin_unlock(&hslot2->lock);
208 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
210 struct net *net = sock_net(sk);
211 kuid_t uid = sock_i_uid(sk);
214 sk_for_each(sk2, &hslot->head) {
215 if (net_eq(sock_net(sk2), net) &&
217 sk2->sk_family == sk->sk_family &&
218 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
219 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
220 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
221 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
222 inet_rcv_saddr_equal(sk, sk2, false)) {
223 return reuseport_add_sock(sk, sk2,
224 inet_rcv_saddr_any(sk));
228 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
234 * @sk: socket struct in question
235 * @snum: port number to look up
236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
239 int udp_lib_get_port(struct sock *sk, unsigned short snum,
240 unsigned int hash2_nulladdr)
242 struct udp_table *udptable = udp_get_table_prot(sk);
243 struct udp_hslot *hslot, *hslot2;
244 struct net *net = sock_net(sk);
245 int error = -EADDRINUSE;
248 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
249 unsigned short first, last;
250 int low, high, remaining;
253 inet_sk_get_local_port_range(sk, &low, &high);
254 remaining = (high - low) + 1;
256 rand = get_random_u32();
257 first = reciprocal_scale(rand, remaining) + low;
259 * force rand to be an odd multiple of UDP_HTABLE_SIZE
261 rand = (rand | 1) * (udptable->mask + 1);
262 last = first + udptable->mask + 1;
264 hslot = udp_hashslot(udptable, net, first);
265 bitmap_zero(bitmap, PORTS_PER_CHAIN);
266 spin_lock_bh(&hslot->lock);
267 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
272 * Iterate on all possible values of snum for this hash.
273 * Using steps of an odd multiple of UDP_HTABLE_SIZE
274 * give us randomization and full range coverage.
277 if (low <= snum && snum <= high &&
278 !test_bit(snum >> udptable->log, bitmap) &&
279 !inet_is_local_reserved_port(net, snum))
282 } while (snum != first);
283 spin_unlock_bh(&hslot->lock);
285 } while (++first != last);
288 hslot = udp_hashslot(udptable, net, snum);
289 spin_lock_bh(&hslot->lock);
290 if (hslot->count > 10) {
292 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
294 slot2 &= udptable->mask;
295 hash2_nulladdr &= udptable->mask;
297 hslot2 = udp_hashslot2(udptable, slot2);
298 if (hslot->count < hslot2->count)
299 goto scan_primary_hash;
301 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
302 if (!exist && (hash2_nulladdr != slot2)) {
303 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
304 exist = udp_lib_lport_inuse2(net, snum, hslot2,
313 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
317 inet_sk(sk)->inet_num = snum;
318 udp_sk(sk)->udp_port_hash = snum;
319 udp_sk(sk)->udp_portaddr_hash ^= snum;
320 if (sk_unhashed(sk)) {
321 if (sk->sk_reuseport &&
322 udp_reuseport_add_sock(sk, hslot)) {
323 inet_sk(sk)->inet_num = 0;
324 udp_sk(sk)->udp_port_hash = 0;
325 udp_sk(sk)->udp_portaddr_hash ^= snum;
329 sk_add_node_rcu(sk, &hslot->head);
331 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
333 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
334 spin_lock(&hslot2->lock);
335 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
336 sk->sk_family == AF_INET6)
337 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
340 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
343 spin_unlock(&hslot2->lock);
345 sock_set_flag(sk, SOCK_RCU_FREE);
348 spin_unlock_bh(&hslot->lock);
352 EXPORT_SYMBOL(udp_lib_get_port);
354 int udp_v4_get_port(struct sock *sk, unsigned short snum)
356 unsigned int hash2_nulladdr =
357 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
358 unsigned int hash2_partial =
359 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
361 /* precompute partial secondary hash */
362 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
363 return udp_lib_get_port(sk, snum, hash2_nulladdr);
366 static int compute_score(struct sock *sk, struct net *net,
367 __be32 saddr, __be16 sport,
368 __be32 daddr, unsigned short hnum,
372 struct inet_sock *inet;
375 if (!net_eq(sock_net(sk), net) ||
376 udp_sk(sk)->udp_port_hash != hnum ||
380 if (sk->sk_rcv_saddr != daddr)
383 score = (sk->sk_family == PF_INET) ? 2 : 1;
386 if (inet->inet_daddr) {
387 if (inet->inet_daddr != saddr)
392 if (inet->inet_dport) {
393 if (inet->inet_dport != sport)
398 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
402 if (sk->sk_bound_dev_if)
405 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
410 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
411 const __u16 lport, const __be32 faddr,
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 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
424 __be32 saddr, __be16 sport,
425 __be32 daddr, unsigned short hnum)
427 struct sock *reuse_sk = NULL;
430 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
431 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
432 reuse_sk = reuseport_select_sock(sk, hash, skb,
433 sizeof(struct udphdr));
438 /* called with rcu_read_lock() */
439 static struct sock *udp4_lib_lookup2(struct net *net,
440 __be32 saddr, __be16 sport,
441 __be32 daddr, unsigned int hnum,
443 struct udp_hslot *hslot2,
446 struct sock *sk, *result;
451 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
452 score = compute_score(sk, net, saddr, sport,
453 daddr, hnum, dif, sdif);
454 if (score > badness) {
455 result = lookup_reuseport(net, sk, skb,
456 saddr, sport, daddr, hnum);
457 /* Fall back to scoring if group has connections */
458 if (result && !reuseport_has_conns(sk))
461 result = result ? : sk;
468 static struct sock *udp4_lookup_run_bpf(struct net *net,
469 struct udp_table *udptable,
471 __be32 saddr, __be16 sport,
472 __be32 daddr, u16 hnum, const int dif)
474 struct sock *sk, *reuse_sk;
477 if (udptable != net->ipv4.udp_table)
478 return NULL; /* only UDP is supported */
480 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport,
481 daddr, hnum, dif, &sk);
482 if (no_reuseport || IS_ERR_OR_NULL(sk))
485 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
491 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
492 * harder than this. -DaveM
494 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
495 __be16 sport, __be32 daddr, __be16 dport, int dif,
496 int sdif, struct udp_table *udptable, struct sk_buff *skb)
498 unsigned short hnum = ntohs(dport);
499 unsigned int hash2, slot2;
500 struct udp_hslot *hslot2;
501 struct sock *result, *sk;
503 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
504 slot2 = hash2 & udptable->mask;
505 hslot2 = &udptable->hash2[slot2];
507 /* Lookup connected or non-wildcard socket */
508 result = udp4_lib_lookup2(net, saddr, sport,
509 daddr, hnum, dif, sdif,
511 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
514 /* Lookup redirect from BPF */
515 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
516 sk = udp4_lookup_run_bpf(net, udptable, skb,
517 saddr, sport, daddr, hnum, dif);
524 /* Got non-wildcard socket or error on first lookup */
528 /* Lookup wildcard sockets */
529 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
530 slot2 = hash2 & udptable->mask;
531 hslot2 = &udptable->hash2[slot2];
533 result = udp4_lib_lookup2(net, saddr, sport,
534 htonl(INADDR_ANY), hnum, dif, sdif,
541 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
543 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
544 __be16 sport, __be16 dport,
545 struct udp_table *udptable)
547 const struct iphdr *iph = ip_hdr(skb);
549 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
550 iph->daddr, dport, inet_iif(skb),
551 inet_sdif(skb), udptable, skb);
554 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
555 __be16 sport, __be16 dport)
557 const struct iphdr *iph = ip_hdr(skb);
558 struct net *net = dev_net(skb->dev);
561 inet_get_iif_sdif(skb, &iif, &sdif);
563 return __udp4_lib_lookup(net, iph->saddr, sport,
564 iph->daddr, dport, iif,
565 sdif, net->ipv4.udp_table, NULL);
568 /* Must be called under rcu_read_lock().
569 * Does increment socket refcount.
571 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
572 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
573 __be32 daddr, __be16 dport, int dif)
577 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
578 dif, 0, net->ipv4.udp_table, NULL);
579 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
583 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
586 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
587 __be16 loc_port, __be32 loc_addr,
588 __be16 rmt_port, __be32 rmt_addr,
589 int dif, int sdif, unsigned short hnum)
591 const struct inet_sock *inet = inet_sk(sk);
593 if (!net_eq(sock_net(sk), net) ||
594 udp_sk(sk)->udp_port_hash != hnum ||
595 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
596 (inet->inet_dport != rmt_port && inet->inet_dport) ||
597 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
598 ipv6_only_sock(sk) ||
599 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
601 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
606 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
607 void udp_encap_enable(void)
609 static_branch_inc(&udp_encap_needed_key);
611 EXPORT_SYMBOL(udp_encap_enable);
613 void udp_encap_disable(void)
615 static_branch_dec(&udp_encap_needed_key);
617 EXPORT_SYMBOL(udp_encap_disable);
619 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
620 * through error handlers in encapsulations looking for a match.
622 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
626 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
627 int (*handler)(struct sk_buff *skb, u32 info);
628 const struct ip_tunnel_encap_ops *encap;
630 encap = rcu_dereference(iptun_encaps[i]);
633 handler = encap->err_handler;
634 if (handler && !handler(skb, info))
641 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
642 * reversing source and destination port: this will match tunnels that force the
643 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
644 * lwtunnels might actually break this assumption by being configured with
645 * different destination ports on endpoints, in this case we won't be able to
646 * trace ICMP messages back to them.
648 * If this doesn't match any socket, probe tunnels with arbitrary destination
649 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
650 * we've sent packets to won't necessarily match the local destination port.
652 * Then ask the tunnel implementation to match the error against a valid
655 * Return an error if we can't find a match, the socket if we need further
656 * processing, zero otherwise.
658 static struct sock *__udp4_lib_err_encap(struct net *net,
659 const struct iphdr *iph,
661 struct udp_table *udptable,
663 struct sk_buff *skb, u32 info)
665 int (*lookup)(struct sock *sk, struct sk_buff *skb);
666 int network_offset, transport_offset;
669 network_offset = skb_network_offset(skb);
670 transport_offset = skb_transport_offset(skb);
672 /* Network header needs to point to the outer IPv4 header inside ICMP */
673 skb_reset_network_header(skb);
675 /* Transport header needs to point to the UDP header */
676 skb_set_transport_header(skb, iph->ihl << 2);
681 lookup = READ_ONCE(up->encap_err_lookup);
682 if (lookup && lookup(sk, skb))
688 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
689 iph->saddr, uh->dest, skb->dev->ifindex, 0,
694 lookup = READ_ONCE(up->encap_err_lookup);
695 if (!lookup || lookup(sk, skb))
701 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
703 skb_set_transport_header(skb, transport_offset);
704 skb_set_network_header(skb, network_offset);
710 * This routine is called by the ICMP module when it gets some
711 * sort of error condition. If err < 0 then the socket should
712 * be closed and the error returned to the user. If err > 0
713 * it's just the icmp type << 8 | icmp code.
714 * Header points to the ip header of the error packet. We move
715 * on past this. Then (as it used to claim before adjustment)
716 * header points to the first 8 bytes of the udp header. We need
717 * to find the appropriate port.
720 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
722 struct inet_sock *inet;
723 const struct iphdr *iph = (const struct iphdr *)skb->data;
724 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
725 const int type = icmp_hdr(skb)->type;
726 const int code = icmp_hdr(skb)->code;
731 struct net *net = dev_net(skb->dev);
733 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
734 iph->saddr, uh->source, skb->dev->ifindex,
735 inet_sdif(skb), udptable, NULL);
737 if (!sk || udp_sk(sk)->encap_type) {
738 /* No socket for error: try tunnels before discarding */
739 if (static_branch_unlikely(&udp_encap_needed_key)) {
740 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
745 sk = ERR_PTR(-ENOENT);
748 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
761 case ICMP_TIME_EXCEEDED:
764 case ICMP_SOURCE_QUENCH:
766 case ICMP_PARAMETERPROB:
770 case ICMP_DEST_UNREACH:
771 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
772 ipv4_sk_update_pmtu(skb, sk, info);
773 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
781 if (code <= NR_ICMP_UNREACH) {
782 harderr = icmp_err_convert[code].fatal;
783 err = icmp_err_convert[code].errno;
787 ipv4_sk_redirect(skb, sk);
792 * RFC1122: OK. Passes ICMP errors back to application, as per
796 /* ...not for tunnels though: we don't have a sending socket */
797 if (udp_sk(sk)->encap_err_rcv)
798 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
802 if (!inet->recverr) {
803 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
806 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
814 int udp_err(struct sk_buff *skb, u32 info)
816 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
820 * Throw away all pending data and cancel the corking. Socket is locked.
822 void udp_flush_pending_frames(struct sock *sk)
824 struct udp_sock *up = udp_sk(sk);
829 ip_flush_pending_frames(sk);
832 EXPORT_SYMBOL(udp_flush_pending_frames);
835 * udp4_hwcsum - handle outgoing HW checksumming
836 * @skb: sk_buff containing the filled-in UDP header
837 * (checksum field must be zeroed out)
838 * @src: source IP address
839 * @dst: destination IP address
841 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
843 struct udphdr *uh = udp_hdr(skb);
844 int offset = skb_transport_offset(skb);
845 int len = skb->len - offset;
849 if (!skb_has_frag_list(skb)) {
851 * Only one fragment on the socket.
853 skb->csum_start = skb_transport_header(skb) - skb->head;
854 skb->csum_offset = offsetof(struct udphdr, check);
855 uh->check = ~csum_tcpudp_magic(src, dst, len,
858 struct sk_buff *frags;
861 * HW-checksum won't work as there are two or more
862 * fragments on the socket so that all csums of sk_buffs
865 skb_walk_frags(skb, frags) {
866 csum = csum_add(csum, frags->csum);
870 csum = skb_checksum(skb, offset, hlen, csum);
871 skb->ip_summed = CHECKSUM_NONE;
873 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
875 uh->check = CSUM_MANGLED_0;
878 EXPORT_SYMBOL_GPL(udp4_hwcsum);
880 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
881 * for the simple case like when setting the checksum for a UDP tunnel.
883 void udp_set_csum(bool nocheck, struct sk_buff *skb,
884 __be32 saddr, __be32 daddr, int len)
886 struct udphdr *uh = udp_hdr(skb);
890 } else if (skb_is_gso(skb)) {
891 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
892 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
894 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
896 uh->check = CSUM_MANGLED_0;
898 skb->ip_summed = CHECKSUM_PARTIAL;
899 skb->csum_start = skb_transport_header(skb) - skb->head;
900 skb->csum_offset = offsetof(struct udphdr, check);
901 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
904 EXPORT_SYMBOL(udp_set_csum);
906 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
907 struct inet_cork *cork)
909 struct sock *sk = skb->sk;
910 struct inet_sock *inet = inet_sk(sk);
913 int is_udplite = IS_UDPLITE(sk);
914 int offset = skb_transport_offset(skb);
915 int len = skb->len - offset;
916 int datalen = len - sizeof(*uh);
920 * Create a UDP header
923 uh->source = inet->inet_sport;
924 uh->dest = fl4->fl4_dport;
925 uh->len = htons(len);
928 if (cork->gso_size) {
929 const int hlen = skb_network_header_len(skb) +
930 sizeof(struct udphdr);
932 if (hlen + cork->gso_size > cork->fragsize) {
936 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
940 if (sk->sk_no_check_tx) {
944 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
945 dst_xfrm(skb_dst(skb))) {
950 if (datalen > cork->gso_size) {
951 skb_shinfo(skb)->gso_size = cork->gso_size;
952 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
953 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
959 if (is_udplite) /* UDP-Lite */
960 csum = udplite_csum(skb);
962 else if (sk->sk_no_check_tx) { /* UDP csum off */
964 skb->ip_summed = CHECKSUM_NONE;
967 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
970 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
974 csum = udp_csum(skb);
976 /* add protocol-dependent pseudo-header */
977 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
978 sk->sk_protocol, csum);
980 uh->check = CSUM_MANGLED_0;
983 err = ip_send_skb(sock_net(sk), skb);
985 if (err == -ENOBUFS && !inet->recverr) {
986 UDP_INC_STATS(sock_net(sk),
987 UDP_MIB_SNDBUFERRORS, is_udplite);
991 UDP_INC_STATS(sock_net(sk),
992 UDP_MIB_OUTDATAGRAMS, is_udplite);
997 * Push out all pending data as one UDP datagram. Socket is locked.
999 int udp_push_pending_frames(struct sock *sk)
1001 struct udp_sock *up = udp_sk(sk);
1002 struct inet_sock *inet = inet_sk(sk);
1003 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1004 struct sk_buff *skb;
1007 skb = ip_finish_skb(sk, fl4);
1011 err = udp_send_skb(skb, fl4, &inet->cork.base);
1018 EXPORT_SYMBOL(udp_push_pending_frames);
1020 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1022 switch (cmsg->cmsg_type) {
1024 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1026 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1033 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1035 struct cmsghdr *cmsg;
1036 bool need_ip = false;
1039 for_each_cmsghdr(cmsg, msg) {
1040 if (!CMSG_OK(msg, cmsg))
1043 if (cmsg->cmsg_level != SOL_UDP) {
1048 err = __udp_cmsg_send(cmsg, gso_size);
1055 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1057 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1059 struct inet_sock *inet = inet_sk(sk);
1060 struct udp_sock *up = udp_sk(sk);
1061 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1062 struct flowi4 fl4_stack;
1065 struct ipcm_cookie ipc;
1066 struct rtable *rt = NULL;
1069 __be32 daddr, faddr, saddr;
1072 int err, is_udplite = IS_UDPLITE(sk);
1073 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1074 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1075 struct sk_buff *skb;
1076 struct ip_options_data opt_copy;
1085 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1088 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1090 fl4 = &inet->cork.fl.u.ip4;
1093 * There are pending frames.
1094 * The socket lock must be held while it's corked.
1097 if (likely(up->pending)) {
1098 if (unlikely(up->pending != AF_INET)) {
1102 goto do_append_data;
1106 ulen += sizeof(struct udphdr);
1109 * Get and verify the address.
1112 if (msg->msg_namelen < sizeof(*usin))
1114 if (usin->sin_family != AF_INET) {
1115 if (usin->sin_family != AF_UNSPEC)
1116 return -EAFNOSUPPORT;
1119 daddr = usin->sin_addr.s_addr;
1120 dport = usin->sin_port;
1124 if (sk->sk_state != TCP_ESTABLISHED)
1125 return -EDESTADDRREQ;
1126 daddr = inet->inet_daddr;
1127 dport = inet->inet_dport;
1128 /* Open fast path for connected socket.
1129 Route will not be used, if at least one option is set.
1134 ipcm_init_sk(&ipc, inet);
1135 ipc.gso_size = READ_ONCE(up->gso_size);
1137 if (msg->msg_controllen) {
1138 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1140 err = ip_cmsg_send(sk, msg, &ipc,
1141 sk->sk_family == AF_INET6);
1142 if (unlikely(err < 0)) {
1151 struct ip_options_rcu *inet_opt;
1154 inet_opt = rcu_dereference(inet->inet_opt);
1156 memcpy(&opt_copy, inet_opt,
1157 sizeof(*inet_opt) + inet_opt->opt.optlen);
1158 ipc.opt = &opt_copy.opt;
1163 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1164 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1165 (struct sockaddr *)usin, &ipc.addr);
1169 if (usin->sin_port == 0) {
1170 /* BPF program set invalid port. Reject it. */
1174 daddr = usin->sin_addr.s_addr;
1175 dport = usin->sin_port;
1180 ipc.addr = faddr = daddr;
1182 if (ipc.opt && ipc.opt->opt.srr) {
1187 faddr = ipc.opt->opt.faddr;
1190 tos = get_rttos(&ipc, inet);
1191 scope = ip_sendmsg_scope(inet, &ipc, msg);
1192 if (scope == RT_SCOPE_LINK)
1195 if (ipv4_is_multicast(daddr)) {
1196 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1197 ipc.oif = inet->mc_index;
1199 saddr = inet->mc_addr;
1201 } else if (!ipc.oif) {
1202 ipc.oif = inet->uc_index;
1203 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1204 /* oif is set, packet is to local broadcast and
1205 * uc_index is set. oif is most likely set
1206 * by sk_bound_dev_if. If uc_index != oif check if the
1207 * oif is an L3 master and uc_index is an L3 slave.
1208 * If so, we want to allow the send using the uc_index.
1210 if (ipc.oif != inet->uc_index &&
1211 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1213 ipc.oif = inet->uc_index;
1218 rt = (struct rtable *)sk_dst_check(sk, 0);
1221 struct net *net = sock_net(sk);
1222 __u8 flow_flags = inet_sk_flowi_flags(sk);
1226 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1227 sk->sk_protocol, flow_flags, faddr, saddr,
1228 dport, inet->inet_sport, sk->sk_uid);
1230 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1231 rt = ip_route_output_flow(net, fl4, sk);
1235 if (err == -ENETUNREACH)
1236 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1241 if ((rt->rt_flags & RTCF_BROADCAST) &&
1242 !sock_flag(sk, SOCK_BROADCAST))
1245 sk_dst_set(sk, dst_clone(&rt->dst));
1248 if (msg->msg_flags&MSG_CONFIRM)
1254 daddr = ipc.addr = fl4->daddr;
1256 /* Lockless fast path for the non-corking case. */
1258 struct inet_cork cork;
1260 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1261 sizeof(struct udphdr), &ipc, &rt,
1262 &cork, msg->msg_flags);
1264 if (!IS_ERR_OR_NULL(skb))
1265 err = udp_send_skb(skb, fl4, &cork);
1270 if (unlikely(up->pending)) {
1271 /* The socket is already corked while preparing it. */
1272 /* ... which is an evident application bug. --ANK */
1275 net_dbg_ratelimited("socket already corked\n");
1280 * Now cork the socket to pend data.
1282 fl4 = &inet->cork.fl.u.ip4;
1285 fl4->fl4_dport = dport;
1286 fl4->fl4_sport = inet->inet_sport;
1287 up->pending = AF_INET;
1291 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1292 sizeof(struct udphdr), &ipc, &rt,
1293 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1295 udp_flush_pending_frames(sk);
1297 err = udp_push_pending_frames(sk);
1298 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1310 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1311 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1312 * we don't have a good statistic (IpOutDiscards but it can be too many
1313 * things). We could add another new stat but at least for now that
1314 * seems like overkill.
1316 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1317 UDP_INC_STATS(sock_net(sk),
1318 UDP_MIB_SNDBUFERRORS, is_udplite);
1323 if (msg->msg_flags & MSG_PROBE)
1324 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1325 if (!(msg->msg_flags&MSG_PROBE) || len)
1326 goto back_from_confirm;
1330 EXPORT_SYMBOL(udp_sendmsg);
1332 void udp_splice_eof(struct socket *sock)
1334 struct sock *sk = sock->sk;
1335 struct udp_sock *up = udp_sk(sk);
1337 if (!up->pending || READ_ONCE(up->corkflag))
1341 if (up->pending && !READ_ONCE(up->corkflag))
1342 udp_push_pending_frames(sk);
1345 EXPORT_SYMBOL_GPL(udp_splice_eof);
1347 #define UDP_SKB_IS_STATELESS 0x80000000
1349 /* all head states (dst, sk, nf conntrack) except skb extensions are
1350 * cleared by udp_rcv().
1352 * We need to preserve secpath, if present, to eventually process
1353 * IP_CMSG_PASSSEC at recvmsg() time.
1355 * Other extensions can be cleared.
1357 static bool udp_try_make_stateless(struct sk_buff *skb)
1359 if (!skb_has_extensions(skb))
1362 if (!secpath_exists(skb)) {
1370 static void udp_set_dev_scratch(struct sk_buff *skb)
1372 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1374 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1375 scratch->_tsize_state = skb->truesize;
1376 #if BITS_PER_LONG == 64
1377 scratch->len = skb->len;
1378 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1379 scratch->is_linear = !skb_is_nonlinear(skb);
1381 if (udp_try_make_stateless(skb))
1382 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1385 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1387 /* We come here after udp_lib_checksum_complete() returned 0.
1388 * This means that __skb_checksum_complete() might have
1389 * set skb->csum_valid to 1.
1390 * On 64bit platforms, we can set csum_unnecessary
1391 * to true, but only if the skb is not shared.
1393 #if BITS_PER_LONG == 64
1394 if (!skb_shared(skb))
1395 udp_skb_scratch(skb)->csum_unnecessary = true;
1399 static int udp_skb_truesize(struct sk_buff *skb)
1401 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1404 static bool udp_skb_has_head_state(struct sk_buff *skb)
1406 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1409 /* fully reclaim rmem/fwd memory allocated for skb */
1410 static void udp_rmem_release(struct sock *sk, int size, int partial,
1411 bool rx_queue_lock_held)
1413 struct udp_sock *up = udp_sk(sk);
1414 struct sk_buff_head *sk_queue;
1417 if (likely(partial)) {
1418 up->forward_deficit += size;
1419 size = up->forward_deficit;
1420 if (size < READ_ONCE(up->forward_threshold) &&
1421 !skb_queue_empty(&up->reader_queue))
1424 size += up->forward_deficit;
1426 up->forward_deficit = 0;
1428 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1429 * if the called don't held it already
1431 sk_queue = &sk->sk_receive_queue;
1432 if (!rx_queue_lock_held)
1433 spin_lock(&sk_queue->lock);
1436 sk->sk_forward_alloc += size;
1437 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1438 sk->sk_forward_alloc -= amt;
1441 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1443 atomic_sub(size, &sk->sk_rmem_alloc);
1445 /* this can save us from acquiring the rx queue lock on next receive */
1446 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1448 if (!rx_queue_lock_held)
1449 spin_unlock(&sk_queue->lock);
1452 /* Note: called with reader_queue.lock held.
1453 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1454 * This avoids a cache line miss while receive_queue lock is held.
1455 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1457 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1459 prefetch(&skb->data);
1460 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1462 EXPORT_SYMBOL(udp_skb_destructor);
1464 /* as above, but the caller held the rx queue lock, too */
1465 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1467 prefetch(&skb->data);
1468 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1471 /* Idea of busylocks is to let producers grab an extra spinlock
1472 * to relieve pressure on the receive_queue spinlock shared by consumer.
1473 * Under flood, this means that only one producer can be in line
1474 * trying to acquire the receive_queue spinlock.
1475 * These busylock can be allocated on a per cpu manner, instead of a
1476 * per socket one (that would consume a cache line per socket)
1478 static int udp_busylocks_log __read_mostly;
1479 static spinlock_t *udp_busylocks __read_mostly;
1481 static spinlock_t *busylock_acquire(void *ptr)
1485 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1490 static void busylock_release(spinlock_t *busy)
1496 static int udp_rmem_schedule(struct sock *sk, int size)
1500 delta = size - sk->sk_forward_alloc;
1501 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1507 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1509 struct sk_buff_head *list = &sk->sk_receive_queue;
1510 int rmem, err = -ENOMEM;
1511 spinlock_t *busy = NULL;
1514 /* try to avoid the costly atomic add/sub pair when the receive
1515 * queue is full; always allow at least a packet
1517 rmem = atomic_read(&sk->sk_rmem_alloc);
1518 if (rmem > sk->sk_rcvbuf)
1521 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1522 * having linear skbs :
1523 * - Reduce memory overhead and thus increase receive queue capacity
1524 * - Less cache line misses at copyout() time
1525 * - Less work at consume_skb() (less alien page frag freeing)
1527 if (rmem > (sk->sk_rcvbuf >> 1)) {
1530 busy = busylock_acquire(sk);
1532 size = skb->truesize;
1533 udp_set_dev_scratch(skb);
1535 /* we drop only if the receive buf is full and the receive
1536 * queue contains some other skb
1538 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1539 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1542 spin_lock(&list->lock);
1543 err = udp_rmem_schedule(sk, size);
1545 spin_unlock(&list->lock);
1549 sk->sk_forward_alloc -= size;
1551 /* no need to setup a destructor, we will explicitly release the
1552 * forward allocated memory on dequeue
1554 sock_skb_set_dropcount(sk, skb);
1556 __skb_queue_tail(list, skb);
1557 spin_unlock(&list->lock);
1559 if (!sock_flag(sk, SOCK_DEAD))
1560 sk->sk_data_ready(sk);
1562 busylock_release(busy);
1566 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1569 atomic_inc(&sk->sk_drops);
1570 busylock_release(busy);
1573 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1575 void udp_destruct_common(struct sock *sk)
1577 /* reclaim completely the forward allocated memory */
1578 struct udp_sock *up = udp_sk(sk);
1579 unsigned int total = 0;
1580 struct sk_buff *skb;
1582 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1583 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1584 total += skb->truesize;
1587 udp_rmem_release(sk, total, 0, true);
1589 EXPORT_SYMBOL_GPL(udp_destruct_common);
1591 static void udp_destruct_sock(struct sock *sk)
1593 udp_destruct_common(sk);
1594 inet_sock_destruct(sk);
1597 int udp_init_sock(struct sock *sk)
1599 udp_lib_init_sock(sk);
1600 sk->sk_destruct = udp_destruct_sock;
1601 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1605 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1607 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1608 bool slow = lock_sock_fast(sk);
1610 sk_peek_offset_bwd(sk, len);
1611 unlock_sock_fast(sk, slow);
1614 if (!skb_unref(skb))
1617 /* In the more common cases we cleared the head states previously,
1618 * see __udp_queue_rcv_skb().
1620 if (unlikely(udp_skb_has_head_state(skb)))
1621 skb_release_head_state(skb);
1622 __consume_stateless_skb(skb);
1624 EXPORT_SYMBOL_GPL(skb_consume_udp);
1626 static struct sk_buff *__first_packet_length(struct sock *sk,
1627 struct sk_buff_head *rcvq,
1630 struct sk_buff *skb;
1632 while ((skb = skb_peek(rcvq)) != NULL) {
1633 if (udp_lib_checksum_complete(skb)) {
1634 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1636 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1638 atomic_inc(&sk->sk_drops);
1639 __skb_unlink(skb, rcvq);
1640 *total += skb->truesize;
1643 udp_skb_csum_unnecessary_set(skb);
1651 * first_packet_length - return length of first packet in receive queue
1654 * Drops all bad checksum frames, until a valid one is found.
1655 * Returns the length of found skb, or -1 if none is found.
1657 static int first_packet_length(struct sock *sk)
1659 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1660 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1661 struct sk_buff *skb;
1665 spin_lock_bh(&rcvq->lock);
1666 skb = __first_packet_length(sk, rcvq, &total);
1667 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1668 spin_lock(&sk_queue->lock);
1669 skb_queue_splice_tail_init(sk_queue, rcvq);
1670 spin_unlock(&sk_queue->lock);
1672 skb = __first_packet_length(sk, rcvq, &total);
1674 res = skb ? skb->len : -1;
1676 udp_rmem_release(sk, total, 1, false);
1677 spin_unlock_bh(&rcvq->lock);
1682 * IOCTL requests applicable to the UDP protocol
1685 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1690 *karg = sk_wmem_alloc_get(sk);
1696 *karg = max_t(int, 0, first_packet_length(sk));
1701 return -ENOIOCTLCMD;
1706 EXPORT_SYMBOL(udp_ioctl);
1708 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1711 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1712 struct sk_buff_head *queue;
1713 struct sk_buff *last;
1717 queue = &udp_sk(sk)->reader_queue;
1718 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1720 struct sk_buff *skb;
1722 error = sock_error(sk);
1728 spin_lock_bh(&queue->lock);
1729 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1732 if (!(flags & MSG_PEEK))
1733 udp_skb_destructor(sk, skb);
1734 spin_unlock_bh(&queue->lock);
1738 if (skb_queue_empty_lockless(sk_queue)) {
1739 spin_unlock_bh(&queue->lock);
1743 /* refill the reader queue and walk it again
1744 * keep both queues locked to avoid re-acquiring
1745 * the sk_receive_queue lock if fwd memory scheduling
1748 spin_lock(&sk_queue->lock);
1749 skb_queue_splice_tail_init(sk_queue, queue);
1751 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1753 if (skb && !(flags & MSG_PEEK))
1754 udp_skb_dtor_locked(sk, skb);
1755 spin_unlock(&sk_queue->lock);
1756 spin_unlock_bh(&queue->lock);
1761 if (!sk_can_busy_loop(sk))
1764 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1765 } while (!skb_queue_empty_lockless(sk_queue));
1767 /* sk_queue is empty, reader_queue may contain peeked packets */
1769 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1771 (struct sk_buff *)sk_queue));
1776 EXPORT_SYMBOL(__skb_recv_udp);
1778 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1780 struct sk_buff *skb;
1784 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1788 if (udp_lib_checksum_complete(skb)) {
1789 int is_udplite = IS_UDPLITE(sk);
1790 struct net *net = sock_net(sk);
1792 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1793 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1794 atomic_inc(&sk->sk_drops);
1799 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1800 return recv_actor(sk, skb);
1802 EXPORT_SYMBOL(udp_read_skb);
1805 * This should be easy, if there is something there we
1806 * return it, otherwise we block.
1809 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1812 struct inet_sock *inet = inet_sk(sk);
1813 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1814 struct sk_buff *skb;
1815 unsigned int ulen, copied;
1816 int off, err, peeking = flags & MSG_PEEK;
1817 int is_udplite = IS_UDPLITE(sk);
1818 bool checksum_valid = false;
1820 if (flags & MSG_ERRQUEUE)
1821 return ip_recv_error(sk, msg, len, addr_len);
1824 off = sk_peek_offset(sk, flags);
1825 skb = __skb_recv_udp(sk, flags, &off, &err);
1829 ulen = udp_skb_len(skb);
1831 if (copied > ulen - off)
1832 copied = ulen - off;
1833 else if (copied < ulen)
1834 msg->msg_flags |= MSG_TRUNC;
1837 * If checksum is needed at all, try to do it while copying the
1838 * data. If the data is truncated, or if we only want a partial
1839 * coverage checksum (UDP-Lite), do it before the copy.
1842 if (copied < ulen || peeking ||
1843 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1844 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1845 !__udp_lib_checksum_complete(skb);
1846 if (!checksum_valid)
1850 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1851 if (udp_skb_is_linear(skb))
1852 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1854 err = skb_copy_datagram_msg(skb, off, msg, copied);
1856 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1862 if (unlikely(err)) {
1864 atomic_inc(&sk->sk_drops);
1865 UDP_INC_STATS(sock_net(sk),
1866 UDP_MIB_INERRORS, is_udplite);
1873 UDP_INC_STATS(sock_net(sk),
1874 UDP_MIB_INDATAGRAMS, is_udplite);
1876 sock_recv_cmsgs(msg, sk, skb);
1878 /* Copy the address. */
1880 sin->sin_family = AF_INET;
1881 sin->sin_port = udp_hdr(skb)->source;
1882 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1883 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1884 *addr_len = sizeof(*sin);
1886 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1887 (struct sockaddr *)sin);
1890 if (udp_sk(sk)->gro_enabled)
1891 udp_cmsg_recv(msg, sk, skb);
1893 if (inet->cmsg_flags)
1894 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1897 if (flags & MSG_TRUNC)
1900 skb_consume_udp(sk, skb, peeking ? -err : err);
1904 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1905 udp_skb_destructor)) {
1906 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1907 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1911 /* starting over for a new packet, but check if we need to yield */
1913 msg->msg_flags &= ~MSG_TRUNC;
1917 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1919 /* This check is replicated from __ip4_datagram_connect() and
1920 * intended to prevent BPF program called below from accessing bytes
1921 * that are out of the bound specified by user in addr_len.
1923 if (addr_len < sizeof(struct sockaddr_in))
1926 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1928 EXPORT_SYMBOL(udp_pre_connect);
1930 int __udp_disconnect(struct sock *sk, int flags)
1932 struct inet_sock *inet = inet_sk(sk);
1934 * 1003.1g - break association.
1937 sk->sk_state = TCP_CLOSE;
1938 inet->inet_daddr = 0;
1939 inet->inet_dport = 0;
1940 sock_rps_reset_rxhash(sk);
1941 sk->sk_bound_dev_if = 0;
1942 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1943 inet_reset_saddr(sk);
1944 if (sk->sk_prot->rehash &&
1945 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1946 sk->sk_prot->rehash(sk);
1949 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1950 sk->sk_prot->unhash(sk);
1951 inet->inet_sport = 0;
1956 EXPORT_SYMBOL(__udp_disconnect);
1958 int udp_disconnect(struct sock *sk, int flags)
1961 __udp_disconnect(sk, flags);
1965 EXPORT_SYMBOL(udp_disconnect);
1967 void udp_lib_unhash(struct sock *sk)
1969 if (sk_hashed(sk)) {
1970 struct udp_table *udptable = udp_get_table_prot(sk);
1971 struct udp_hslot *hslot, *hslot2;
1973 hslot = udp_hashslot(udptable, sock_net(sk),
1974 udp_sk(sk)->udp_port_hash);
1975 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1977 spin_lock_bh(&hslot->lock);
1978 if (rcu_access_pointer(sk->sk_reuseport_cb))
1979 reuseport_detach_sock(sk);
1980 if (sk_del_node_init_rcu(sk)) {
1982 inet_sk(sk)->inet_num = 0;
1983 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1985 spin_lock(&hslot2->lock);
1986 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1988 spin_unlock(&hslot2->lock);
1990 spin_unlock_bh(&hslot->lock);
1993 EXPORT_SYMBOL(udp_lib_unhash);
1996 * inet_rcv_saddr was changed, we must rehash secondary hash
1998 void udp_lib_rehash(struct sock *sk, u16 newhash)
2000 if (sk_hashed(sk)) {
2001 struct udp_table *udptable = udp_get_table_prot(sk);
2002 struct udp_hslot *hslot, *hslot2, *nhslot2;
2004 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2005 nhslot2 = udp_hashslot2(udptable, newhash);
2006 udp_sk(sk)->udp_portaddr_hash = newhash;
2008 if (hslot2 != nhslot2 ||
2009 rcu_access_pointer(sk->sk_reuseport_cb)) {
2010 hslot = udp_hashslot(udptable, sock_net(sk),
2011 udp_sk(sk)->udp_port_hash);
2012 /* we must lock primary chain too */
2013 spin_lock_bh(&hslot->lock);
2014 if (rcu_access_pointer(sk->sk_reuseport_cb))
2015 reuseport_detach_sock(sk);
2017 if (hslot2 != nhslot2) {
2018 spin_lock(&hslot2->lock);
2019 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2021 spin_unlock(&hslot2->lock);
2023 spin_lock(&nhslot2->lock);
2024 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2027 spin_unlock(&nhslot2->lock);
2030 spin_unlock_bh(&hslot->lock);
2034 EXPORT_SYMBOL(udp_lib_rehash);
2036 void udp_v4_rehash(struct sock *sk)
2038 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2039 inet_sk(sk)->inet_rcv_saddr,
2040 inet_sk(sk)->inet_num);
2041 udp_lib_rehash(sk, new_hash);
2044 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2048 if (inet_sk(sk)->inet_daddr) {
2049 sock_rps_save_rxhash(sk, skb);
2050 sk_mark_napi_id(sk, skb);
2051 sk_incoming_cpu_update(sk);
2053 sk_mark_napi_id_once(sk, skb);
2056 rc = __udp_enqueue_schedule_skb(sk, skb);
2058 int is_udplite = IS_UDPLITE(sk);
2061 /* Note that an ENOMEM error is charged twice */
2062 if (rc == -ENOMEM) {
2063 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2065 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2067 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2069 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2071 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2072 kfree_skb_reason(skb, drop_reason);
2073 trace_udp_fail_queue_rcv_skb(rc, sk);
2083 * >0: "udp encap" protocol resubmission
2085 * Note that in the success and error cases, the skb is assumed to
2086 * have either been requeued or freed.
2088 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2090 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2091 struct udp_sock *up = udp_sk(sk);
2092 int is_udplite = IS_UDPLITE(sk);
2095 * Charge it to the socket, dropping if the queue is full.
2097 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2098 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2103 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2104 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2107 * This is an encapsulation socket so pass the skb to
2108 * the socket's udp_encap_rcv() hook. Otherwise, just
2109 * fall through and pass this up the UDP socket.
2110 * up->encap_rcv() returns the following value:
2111 * =0 if skb was successfully passed to the encap
2112 * handler or was discarded by it.
2113 * >0 if skb should be passed on to UDP.
2114 * <0 if skb should be resubmitted as proto -N
2117 /* if we're overly short, let UDP handle it */
2118 encap_rcv = READ_ONCE(up->encap_rcv);
2122 /* Verify checksum before giving to encap */
2123 if (udp_lib_checksum_complete(skb))
2126 ret = encap_rcv(sk, skb);
2128 __UDP_INC_STATS(sock_net(sk),
2129 UDP_MIB_INDATAGRAMS,
2135 /* FALLTHROUGH -- it's a UDP Packet */
2139 * UDP-Lite specific tests, ignored on UDP sockets
2141 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2144 * MIB statistics other than incrementing the error count are
2145 * disabled for the following two types of errors: these depend
2146 * on the application settings, not on the functioning of the
2147 * protocol stack as such.
2149 * RFC 3828 here recommends (sec 3.3): "There should also be a
2150 * way ... to ... at least let the receiving application block
2151 * delivery of packets with coverage values less than a value
2152 * provided by the application."
2154 if (up->pcrlen == 0) { /* full coverage was set */
2155 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2156 UDP_SKB_CB(skb)->cscov, skb->len);
2159 /* The next case involves violating the min. coverage requested
2160 * by the receiver. This is subtle: if receiver wants x and x is
2161 * greater than the buffersize/MTU then receiver will complain
2162 * that it wants x while sender emits packets of smaller size y.
2163 * Therefore the above ...()->partial_cov statement is essential.
2165 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2166 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2167 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2172 prefetch(&sk->sk_rmem_alloc);
2173 if (rcu_access_pointer(sk->sk_filter) &&
2174 udp_lib_checksum_complete(skb))
2177 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2178 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2182 udp_csum_pull_header(skb);
2184 ipv4_pktinfo_prepare(sk, skb);
2185 return __udp_queue_rcv_skb(sk, skb);
2188 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2189 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2191 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2192 atomic_inc(&sk->sk_drops);
2193 kfree_skb_reason(skb, drop_reason);
2197 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2199 struct sk_buff *next, *segs;
2202 if (likely(!udp_unexpected_gso(sk, skb)))
2203 return udp_queue_rcv_one_skb(sk, skb);
2205 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2206 __skb_push(skb, -skb_mac_offset(skb));
2207 segs = udp_rcv_segment(sk, skb, true);
2208 skb_list_walk_safe(segs, skb, next) {
2209 __skb_pull(skb, skb_transport_offset(skb));
2211 udp_post_segment_fix_csum(skb);
2212 ret = udp_queue_rcv_one_skb(sk, skb);
2214 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2219 /* For TCP sockets, sk_rx_dst is protected by socket lock
2220 * For UDP, we use xchg() to guard against concurrent changes.
2222 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2224 struct dst_entry *old;
2226 if (dst_hold_safe(dst)) {
2227 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2233 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2236 * Multicasts and broadcasts go to each listener.
2238 * Note: called only from the BH handler context.
2240 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2242 __be32 saddr, __be32 daddr,
2243 struct udp_table *udptable,
2246 struct sock *sk, *first = NULL;
2247 unsigned short hnum = ntohs(uh->dest);
2248 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2249 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2250 unsigned int offset = offsetof(typeof(*sk), sk_node);
2251 int dif = skb->dev->ifindex;
2252 int sdif = inet_sdif(skb);
2253 struct hlist_node *node;
2254 struct sk_buff *nskb;
2257 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2259 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2261 hslot = &udptable->hash2[hash2];
2262 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2265 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2266 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2267 uh->source, saddr, dif, sdif, hnum))
2274 nskb = skb_clone(skb, GFP_ATOMIC);
2276 if (unlikely(!nskb)) {
2277 atomic_inc(&sk->sk_drops);
2278 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2280 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2284 if (udp_queue_rcv_skb(sk, nskb) > 0)
2288 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2289 if (use_hash2 && hash2 != hash2_any) {
2295 if (udp_queue_rcv_skb(first, skb) > 0)
2299 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2300 proto == IPPROTO_UDPLITE);
2305 /* Initialize UDP checksum. If exited with zero value (success),
2306 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2307 * Otherwise, csum completion requires checksumming packet body,
2308 * including udp header and folding it to skb->csum.
2310 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2315 UDP_SKB_CB(skb)->partial_cov = 0;
2316 UDP_SKB_CB(skb)->cscov = skb->len;
2318 if (proto == IPPROTO_UDPLITE) {
2319 err = udplite_checksum_init(skb, uh);
2323 if (UDP_SKB_CB(skb)->partial_cov) {
2324 skb->csum = inet_compute_pseudo(skb, proto);
2329 /* Note, we are only interested in != 0 or == 0, thus the
2332 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2333 inet_compute_pseudo);
2337 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2338 /* If SW calculated the value, we know it's bad */
2339 if (skb->csum_complete_sw)
2342 /* HW says the value is bad. Let's validate that.
2343 * skb->csum is no longer the full packet checksum,
2344 * so don't treat it as such.
2346 skb_checksum_complete_unset(skb);
2352 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2353 * return code conversion for ip layer consumption
2355 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2360 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2361 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2363 ret = udp_queue_rcv_skb(sk, skb);
2365 /* a return value > 0 means to resubmit the input, but
2366 * it wants the return to be -protocol, or 0
2374 * All we need to do is get the socket, and then do a checksum.
2377 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2382 unsigned short ulen;
2383 struct rtable *rt = skb_rtable(skb);
2384 __be32 saddr, daddr;
2385 struct net *net = dev_net(skb->dev);
2389 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2392 * Validate the packet.
2394 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2395 goto drop; /* No space for header. */
2398 ulen = ntohs(uh->len);
2399 saddr = ip_hdr(skb)->saddr;
2400 daddr = ip_hdr(skb)->daddr;
2402 if (ulen > skb->len)
2405 if (proto == IPPROTO_UDP) {
2406 /* UDP validates ulen. */
2407 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2412 if (udp4_csum_init(skb, uh, proto))
2415 sk = skb_steal_sock(skb, &refcounted);
2417 struct dst_entry *dst = skb_dst(skb);
2420 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2421 udp_sk_rx_dst_set(sk, dst);
2423 ret = udp_unicast_rcv_skb(sk, skb, uh);
2429 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2430 return __udp4_lib_mcast_deliver(net, skb, uh,
2431 saddr, daddr, udptable, proto);
2433 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2435 return udp_unicast_rcv_skb(sk, skb, uh);
2437 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2441 /* No socket. Drop packet silently, if checksum is wrong */
2442 if (udp_lib_checksum_complete(skb))
2445 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2446 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2447 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2450 * Hmm. We got an UDP packet to a port to which we
2451 * don't wanna listen. Ignore it.
2453 kfree_skb_reason(skb, drop_reason);
2457 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2458 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2459 proto == IPPROTO_UDPLITE ? "Lite" : "",
2460 &saddr, ntohs(uh->source),
2462 &daddr, ntohs(uh->dest));
2467 * RFC1122: OK. Discards the bad packet silently (as far as
2468 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2470 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2471 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2472 proto == IPPROTO_UDPLITE ? "Lite" : "",
2473 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2475 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2477 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2478 kfree_skb_reason(skb, drop_reason);
2482 /* We can only early demux multicast if there is a single matching socket.
2483 * If more than one socket found returns NULL
2485 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2486 __be16 loc_port, __be32 loc_addr,
2487 __be16 rmt_port, __be32 rmt_addr,
2490 struct udp_table *udptable = net->ipv4.udp_table;
2491 unsigned short hnum = ntohs(loc_port);
2492 struct sock *sk, *result;
2493 struct udp_hslot *hslot;
2496 slot = udp_hashfn(net, hnum, udptable->mask);
2497 hslot = &udptable->hash[slot];
2499 /* Do not bother scanning a too big list */
2500 if (hslot->count > 10)
2504 sk_for_each_rcu(sk, &hslot->head) {
2505 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2506 rmt_port, rmt_addr, dif, sdif, hnum)) {
2516 /* For unicast we should only early demux connected sockets or we can
2517 * break forwarding setups. The chains here can be long so only check
2518 * if the first socket is an exact match and if not move on.
2520 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2521 __be16 loc_port, __be32 loc_addr,
2522 __be16 rmt_port, __be32 rmt_addr,
2525 struct udp_table *udptable = net->ipv4.udp_table;
2526 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2527 unsigned short hnum = ntohs(loc_port);
2528 unsigned int hash2, slot2;
2529 struct udp_hslot *hslot2;
2533 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2534 slot2 = hash2 & udptable->mask;
2535 hslot2 = &udptable->hash2[slot2];
2536 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2538 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2539 if (inet_match(net, sk, acookie, ports, dif, sdif))
2541 /* Only check first socket in chain */
2547 int udp_v4_early_demux(struct sk_buff *skb)
2549 struct net *net = dev_net(skb->dev);
2550 struct in_device *in_dev = NULL;
2551 const struct iphdr *iph;
2552 const struct udphdr *uh;
2553 struct sock *sk = NULL;
2554 struct dst_entry *dst;
2555 int dif = skb->dev->ifindex;
2556 int sdif = inet_sdif(skb);
2559 /* validate the packet */
2560 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2566 if (skb->pkt_type == PACKET_MULTICAST) {
2567 in_dev = __in_dev_get_rcu(skb->dev);
2572 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2577 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2578 uh->source, iph->saddr,
2580 } else if (skb->pkt_type == PACKET_HOST) {
2581 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2582 uh->source, iph->saddr, dif, sdif);
2585 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2589 skb->destructor = sock_efree;
2590 dst = rcu_dereference(sk->sk_rx_dst);
2593 dst = dst_check(dst, 0);
2597 /* set noref for now.
2598 * any place which wants to hold dst has to call
2601 skb_dst_set_noref(skb, dst);
2603 /* for unconnected multicast sockets we need to validate
2604 * the source on each packet
2606 if (!inet_sk(sk)->inet_daddr && in_dev)
2607 return ip_mc_validate_source(skb, iph->daddr,
2609 iph->tos & IPTOS_RT_MASK,
2610 skb->dev, in_dev, &itag);
2615 int udp_rcv(struct sk_buff *skb)
2617 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2620 void udp_destroy_sock(struct sock *sk)
2622 struct udp_sock *up = udp_sk(sk);
2623 bool slow = lock_sock_fast(sk);
2625 /* protects from races with udp_abort() */
2626 sock_set_flag(sk, SOCK_DEAD);
2627 udp_flush_pending_frames(sk);
2628 unlock_sock_fast(sk, slow);
2629 if (static_branch_unlikely(&udp_encap_needed_key)) {
2630 if (up->encap_type) {
2631 void (*encap_destroy)(struct sock *sk);
2632 encap_destroy = READ_ONCE(up->encap_destroy);
2636 if (up->encap_enabled)
2637 static_branch_dec(&udp_encap_needed_key);
2642 * Socket option code for UDP
2644 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2645 sockptr_t optval, unsigned int optlen,
2646 int (*push_pending_frames)(struct sock *))
2648 struct udp_sock *up = udp_sk(sk);
2651 int is_udplite = IS_UDPLITE(sk);
2653 if (level == SOL_SOCKET) {
2654 err = sk_setsockopt(sk, level, optname, optval, optlen);
2656 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2657 sockopt_lock_sock(sk);
2658 /* paired with READ_ONCE in udp_rmem_release() */
2659 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2660 sockopt_release_sock(sk);
2665 if (optlen < sizeof(int))
2668 if (copy_from_sockptr(&val, optval, sizeof(val)))
2671 valbool = val ? 1 : 0;
2676 WRITE_ONCE(up->corkflag, 1);
2678 WRITE_ONCE(up->corkflag, 0);
2680 push_pending_frames(sk);
2689 case UDP_ENCAP_ESPINUDP:
2690 case UDP_ENCAP_ESPINUDP_NON_IKE:
2691 #if IS_ENABLED(CONFIG_IPV6)
2692 if (sk->sk_family == AF_INET6)
2693 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2696 up->encap_rcv = xfrm4_udp_encap_rcv;
2699 case UDP_ENCAP_L2TPINUDP:
2700 up->encap_type = val;
2702 udp_tunnel_encap_enable(sk->sk_socket);
2711 case UDP_NO_CHECK6_TX:
2712 up->no_check6_tx = valbool;
2715 case UDP_NO_CHECK6_RX:
2716 up->no_check6_rx = valbool;
2720 if (val < 0 || val > USHRT_MAX)
2722 WRITE_ONCE(up->gso_size, val);
2728 /* when enabling GRO, accept the related GSO packet type */
2730 udp_tunnel_encap_enable(sk->sk_socket);
2731 up->gro_enabled = valbool;
2732 up->accept_udp_l4 = valbool;
2737 * UDP-Lite's partial checksum coverage (RFC 3828).
2739 /* The sender sets actual checksum coverage length via this option.
2740 * The case coverage > packet length is handled by send module. */
2741 case UDPLITE_SEND_CSCOV:
2742 if (!is_udplite) /* Disable the option on UDP sockets */
2743 return -ENOPROTOOPT;
2744 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2746 else if (val > USHRT_MAX)
2749 up->pcflag |= UDPLITE_SEND_CC;
2752 /* The receiver specifies a minimum checksum coverage value. To make
2753 * sense, this should be set to at least 8 (as done below). If zero is
2754 * used, this again means full checksum coverage. */
2755 case UDPLITE_RECV_CSCOV:
2756 if (!is_udplite) /* Disable the option on UDP sockets */
2757 return -ENOPROTOOPT;
2758 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2760 else if (val > USHRT_MAX)
2763 up->pcflag |= UDPLITE_RECV_CC;
2773 EXPORT_SYMBOL(udp_lib_setsockopt);
2775 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2776 unsigned int optlen)
2778 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2779 return udp_lib_setsockopt(sk, level, optname,
2781 udp_push_pending_frames);
2782 return ip_setsockopt(sk, level, optname, optval, optlen);
2785 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2786 char __user *optval, int __user *optlen)
2788 struct udp_sock *up = udp_sk(sk);
2791 if (get_user(len, optlen))
2794 len = min_t(unsigned int, len, sizeof(int));
2801 val = READ_ONCE(up->corkflag);
2805 val = up->encap_type;
2808 case UDP_NO_CHECK6_TX:
2809 val = up->no_check6_tx;
2812 case UDP_NO_CHECK6_RX:
2813 val = up->no_check6_rx;
2817 val = READ_ONCE(up->gso_size);
2821 val = up->gro_enabled;
2824 /* The following two cannot be changed on UDP sockets, the return is
2825 * always 0 (which corresponds to the full checksum coverage of UDP). */
2826 case UDPLITE_SEND_CSCOV:
2830 case UDPLITE_RECV_CSCOV:
2835 return -ENOPROTOOPT;
2838 if (put_user(len, optlen))
2840 if (copy_to_user(optval, &val, len))
2844 EXPORT_SYMBOL(udp_lib_getsockopt);
2846 int udp_getsockopt(struct sock *sk, int level, int optname,
2847 char __user *optval, int __user *optlen)
2849 if (level == SOL_UDP || level == SOL_UDPLITE)
2850 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2851 return ip_getsockopt(sk, level, optname, optval, optlen);
2855 * udp_poll - wait for a UDP event.
2856 * @file: - file struct
2858 * @wait: - poll table
2860 * This is same as datagram poll, except for the special case of
2861 * blocking sockets. If application is using a blocking fd
2862 * and a packet with checksum error is in the queue;
2863 * then it could get return from select indicating data available
2864 * but then block when reading it. Add special case code
2865 * to work around these arguably broken applications.
2867 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2869 __poll_t mask = datagram_poll(file, sock, wait);
2870 struct sock *sk = sock->sk;
2872 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2873 mask |= EPOLLIN | EPOLLRDNORM;
2875 /* Check for false positives due to checksum errors */
2876 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2877 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2878 mask &= ~(EPOLLIN | EPOLLRDNORM);
2880 /* psock ingress_msg queue should not contain any bad checksum frames */
2881 if (sk_is_readable(sk))
2882 mask |= EPOLLIN | EPOLLRDNORM;
2886 EXPORT_SYMBOL(udp_poll);
2888 int udp_abort(struct sock *sk, int err)
2890 if (!has_current_bpf_ctx())
2893 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2896 if (sock_flag(sk, SOCK_DEAD))
2900 sk_error_report(sk);
2901 __udp_disconnect(sk, 0);
2904 if (!has_current_bpf_ctx())
2909 EXPORT_SYMBOL_GPL(udp_abort);
2911 struct proto udp_prot = {
2913 .owner = THIS_MODULE,
2914 .close = udp_lib_close,
2915 .pre_connect = udp_pre_connect,
2916 .connect = ip4_datagram_connect,
2917 .disconnect = udp_disconnect,
2919 .init = udp_init_sock,
2920 .destroy = udp_destroy_sock,
2921 .setsockopt = udp_setsockopt,
2922 .getsockopt = udp_getsockopt,
2923 .sendmsg = udp_sendmsg,
2924 .recvmsg = udp_recvmsg,
2925 .splice_eof = udp_splice_eof,
2926 .release_cb = ip4_datagram_release_cb,
2927 .hash = udp_lib_hash,
2928 .unhash = udp_lib_unhash,
2929 .rehash = udp_v4_rehash,
2930 .get_port = udp_v4_get_port,
2931 .put_port = udp_lib_unhash,
2932 #ifdef CONFIG_BPF_SYSCALL
2933 .psock_update_sk_prot = udp_bpf_update_proto,
2935 .memory_allocated = &udp_memory_allocated,
2936 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2938 .sysctl_mem = sysctl_udp_mem,
2939 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2940 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2941 .obj_size = sizeof(struct udp_sock),
2942 .h.udp_table = NULL,
2943 .diag_destroy = udp_abort,
2945 EXPORT_SYMBOL(udp_prot);
2947 /* ------------------------------------------------------------------------ */
2948 #ifdef CONFIG_PROC_FS
2950 static unsigned short seq_file_family(const struct seq_file *seq);
2951 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2953 unsigned short family = seq_file_family(seq);
2955 /* AF_UNSPEC is used as a match all */
2956 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2957 net_eq(sock_net(sk), seq_file_net(seq)));
2960 #ifdef CONFIG_BPF_SYSCALL
2961 static const struct seq_operations bpf_iter_udp_seq_ops;
2963 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2966 const struct udp_seq_afinfo *afinfo;
2968 #ifdef CONFIG_BPF_SYSCALL
2969 if (seq->op == &bpf_iter_udp_seq_ops)
2970 return net->ipv4.udp_table;
2973 afinfo = pde_data(file_inode(seq->file));
2974 return afinfo->udp_table ? : net->ipv4.udp_table;
2977 static struct sock *udp_get_first(struct seq_file *seq, int start)
2979 struct udp_iter_state *state = seq->private;
2980 struct net *net = seq_file_net(seq);
2981 struct udp_table *udptable;
2984 udptable = udp_get_table_seq(seq, net);
2986 for (state->bucket = start; state->bucket <= udptable->mask;
2988 struct udp_hslot *hslot = &udptable->hash[state->bucket];
2990 if (hlist_empty(&hslot->head))
2993 spin_lock_bh(&hslot->lock);
2994 sk_for_each(sk, &hslot->head) {
2995 if (seq_sk_match(seq, sk))
2998 spin_unlock_bh(&hslot->lock);
3005 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3007 struct udp_iter_state *state = seq->private;
3008 struct net *net = seq_file_net(seq);
3009 struct udp_table *udptable;
3013 } while (sk && !seq_sk_match(seq, sk));
3016 udptable = udp_get_table_seq(seq, net);
3018 if (state->bucket <= udptable->mask)
3019 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3021 return udp_get_first(seq, state->bucket + 1);
3026 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3028 struct sock *sk = udp_get_first(seq, 0);
3031 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3033 return pos ? NULL : sk;
3036 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3038 struct udp_iter_state *state = seq->private;
3039 state->bucket = MAX_UDP_PORTS;
3041 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3043 EXPORT_SYMBOL(udp_seq_start);
3045 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3049 if (v == SEQ_START_TOKEN)
3050 sk = udp_get_idx(seq, 0);
3052 sk = udp_get_next(seq, v);
3057 EXPORT_SYMBOL(udp_seq_next);
3059 void udp_seq_stop(struct seq_file *seq, void *v)
3061 struct udp_iter_state *state = seq->private;
3062 struct udp_table *udptable;
3064 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3066 if (state->bucket <= udptable->mask)
3067 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3069 EXPORT_SYMBOL(udp_seq_stop);
3071 /* ------------------------------------------------------------------------ */
3072 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3075 struct inet_sock *inet = inet_sk(sp);
3076 __be32 dest = inet->inet_daddr;
3077 __be32 src = inet->inet_rcv_saddr;
3078 __u16 destp = ntohs(inet->inet_dport);
3079 __u16 srcp = ntohs(inet->inet_sport);
3081 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3082 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3083 bucket, src, srcp, dest, destp, sp->sk_state,
3084 sk_wmem_alloc_get(sp),
3087 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3089 refcount_read(&sp->sk_refcnt), sp,
3090 atomic_read(&sp->sk_drops));
3093 int udp4_seq_show(struct seq_file *seq, void *v)
3095 seq_setwidth(seq, 127);
3096 if (v == SEQ_START_TOKEN)
3097 seq_puts(seq, " sl local_address rem_address st tx_queue "
3098 "rx_queue tr tm->when retrnsmt uid timeout "
3099 "inode ref pointer drops");
3101 struct udp_iter_state *state = seq->private;
3103 udp4_format_sock(v, seq, state->bucket);
3109 #ifdef CONFIG_BPF_SYSCALL
3110 struct bpf_iter__udp {
3111 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3112 __bpf_md_ptr(struct udp_sock *, udp_sk);
3113 uid_t uid __aligned(8);
3114 int bucket __aligned(8);
3117 struct bpf_udp_iter_state {
3118 struct udp_iter_state state;
3119 unsigned int cur_sk;
3120 unsigned int end_sk;
3121 unsigned int max_sk;
3123 struct sock **batch;
3124 bool st_bucket_done;
3127 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3128 unsigned int new_batch_sz);
3129 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3131 struct bpf_udp_iter_state *iter = seq->private;
3132 struct udp_iter_state *state = &iter->state;
3133 struct net *net = seq_file_net(seq);
3134 struct udp_table *udptable;
3135 unsigned int batch_sks = 0;
3136 bool resized = false;
3139 /* The current batch is done, so advance the bucket. */
3140 if (iter->st_bucket_done) {
3145 udptable = udp_get_table_seq(seq, net);
3148 /* New batch for the next bucket.
3149 * Iterate over the hash table to find a bucket with sockets matching
3150 * the iterator attributes, and return the first matching socket from
3151 * the bucket. The remaining matched sockets from the bucket are batched
3152 * before releasing the bucket lock. This allows BPF programs that are
3153 * called in seq_show to acquire the bucket lock if needed.
3157 iter->st_bucket_done = false;
3160 for (; state->bucket <= udptable->mask; state->bucket++) {
3161 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3163 if (hlist_empty(&hslot2->head)) {
3168 spin_lock_bh(&hslot2->lock);
3169 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3170 if (seq_sk_match(seq, sk)) {
3171 /* Resume from the last iterated socket at the
3172 * offset in the bucket before iterator was stopped.
3178 if (iter->end_sk < iter->max_sk) {
3180 iter->batch[iter->end_sk++] = sk;
3185 spin_unlock_bh(&hslot2->lock);
3190 /* Reset the current bucket's offset before moving to the next bucket. */
3194 /* All done: no batch made. */
3198 if (iter->end_sk == batch_sks) {
3199 /* Batching is done for the current bucket; return the first
3200 * socket to be iterated from the batch.
3202 iter->st_bucket_done = true;
3205 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3207 /* After allocating a larger batch, retry one more time to grab
3214 return iter->batch[0];
3217 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3219 struct bpf_udp_iter_state *iter = seq->private;
3222 /* Whenever seq_next() is called, the iter->cur_sk is
3223 * done with seq_show(), so unref the iter->cur_sk.
3225 if (iter->cur_sk < iter->end_sk) {
3226 sock_put(iter->batch[iter->cur_sk++]);
3230 /* After updating iter->cur_sk, check if there are more sockets
3231 * available in the current bucket batch.
3233 if (iter->cur_sk < iter->end_sk)
3234 sk = iter->batch[iter->cur_sk];
3236 /* Prepare a new batch. */
3237 sk = bpf_iter_udp_batch(seq);
3243 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3245 /* bpf iter does not support lseek, so it always
3246 * continue from where it was stop()-ped.
3249 return bpf_iter_udp_batch(seq);
3251 return SEQ_START_TOKEN;
3254 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3255 struct udp_sock *udp_sk, uid_t uid, int bucket)
3257 struct bpf_iter__udp ctx;
3259 meta->seq_num--; /* skip SEQ_START_TOKEN */
3261 ctx.udp_sk = udp_sk;
3263 ctx.bucket = bucket;
3264 return bpf_iter_run_prog(prog, &ctx);
3267 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3269 struct udp_iter_state *state = seq->private;
3270 struct bpf_iter_meta meta;
3271 struct bpf_prog *prog;
3272 struct sock *sk = v;
3276 if (v == SEQ_START_TOKEN)
3281 if (unlikely(sk_unhashed(sk))) {
3286 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3288 prog = bpf_iter_get_info(&meta, false);
3289 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3296 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3298 while (iter->cur_sk < iter->end_sk)
3299 sock_put(iter->batch[iter->cur_sk++]);
3302 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3304 struct bpf_udp_iter_state *iter = seq->private;
3305 struct bpf_iter_meta meta;
3306 struct bpf_prog *prog;
3310 prog = bpf_iter_get_info(&meta, true);
3312 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3315 if (iter->cur_sk < iter->end_sk) {
3316 bpf_iter_udp_put_batch(iter);
3317 iter->st_bucket_done = false;
3321 static const struct seq_operations bpf_iter_udp_seq_ops = {
3322 .start = bpf_iter_udp_seq_start,
3323 .next = bpf_iter_udp_seq_next,
3324 .stop = bpf_iter_udp_seq_stop,
3325 .show = bpf_iter_udp_seq_show,
3329 static unsigned short seq_file_family(const struct seq_file *seq)
3331 const struct udp_seq_afinfo *afinfo;
3333 #ifdef CONFIG_BPF_SYSCALL
3334 /* BPF iterator: bpf programs to filter sockets. */
3335 if (seq->op == &bpf_iter_udp_seq_ops)
3339 /* Proc fs iterator */
3340 afinfo = pde_data(file_inode(seq->file));
3341 return afinfo->family;
3344 const struct seq_operations udp_seq_ops = {
3345 .start = udp_seq_start,
3346 .next = udp_seq_next,
3347 .stop = udp_seq_stop,
3348 .show = udp4_seq_show,
3350 EXPORT_SYMBOL(udp_seq_ops);
3352 static struct udp_seq_afinfo udp4_seq_afinfo = {
3357 static int __net_init udp4_proc_init_net(struct net *net)
3359 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3360 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3365 static void __net_exit udp4_proc_exit_net(struct net *net)
3367 remove_proc_entry("udp", net->proc_net);
3370 static struct pernet_operations udp4_net_ops = {
3371 .init = udp4_proc_init_net,
3372 .exit = udp4_proc_exit_net,
3375 int __init udp4_proc_init(void)
3377 return register_pernet_subsys(&udp4_net_ops);
3380 void udp4_proc_exit(void)
3382 unregister_pernet_subsys(&udp4_net_ops);
3384 #endif /* CONFIG_PROC_FS */
3386 static __initdata unsigned long uhash_entries;
3387 static int __init set_uhash_entries(char *str)
3394 ret = kstrtoul(str, 0, &uhash_entries);
3398 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3399 uhash_entries = UDP_HTABLE_SIZE_MIN;
3402 __setup("uhash_entries=", set_uhash_entries);
3404 void __init udp_table_init(struct udp_table *table, const char *name)
3408 table->hash = alloc_large_system_hash(name,
3409 2 * sizeof(struct udp_hslot),
3411 21, /* one slot per 2 MB */
3415 UDP_HTABLE_SIZE_MIN,
3416 UDP_HTABLE_SIZE_MAX);
3418 table->hash2 = table->hash + (table->mask + 1);
3419 for (i = 0; i <= table->mask; i++) {
3420 INIT_HLIST_HEAD(&table->hash[i].head);
3421 table->hash[i].count = 0;
3422 spin_lock_init(&table->hash[i].lock);
3424 for (i = 0; i <= table->mask; i++) {
3425 INIT_HLIST_HEAD(&table->hash2[i].head);
3426 table->hash2[i].count = 0;
3427 spin_lock_init(&table->hash2[i].lock);
3431 u32 udp_flow_hashrnd(void)
3433 static u32 hashrnd __read_mostly;
3435 net_get_random_once(&hashrnd, sizeof(hashrnd));
3439 EXPORT_SYMBOL(udp_flow_hashrnd);
3441 static void __net_init udp_sysctl_init(struct net *net)
3443 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3444 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3446 #ifdef CONFIG_NET_L3_MASTER_DEV
3447 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3451 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3453 struct udp_table *udptable;
3456 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3460 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3461 GFP_KERNEL_ACCOUNT);
3462 if (!udptable->hash)
3465 udptable->hash2 = udptable->hash + hash_entries;
3466 udptable->mask = hash_entries - 1;
3467 udptable->log = ilog2(hash_entries);
3469 for (i = 0; i < hash_entries; i++) {
3470 INIT_HLIST_HEAD(&udptable->hash[i].head);
3471 udptable->hash[i].count = 0;
3472 spin_lock_init(&udptable->hash[i].lock);
3474 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3475 udptable->hash2[i].count = 0;
3476 spin_lock_init(&udptable->hash2[i].lock);
3487 static void __net_exit udp_pernet_table_free(struct net *net)
3489 struct udp_table *udptable = net->ipv4.udp_table;
3491 if (udptable == &udp_table)
3494 kvfree(udptable->hash);
3498 static void __net_init udp_set_table(struct net *net)
3500 struct udp_table *udptable;
3501 unsigned int hash_entries;
3502 struct net *old_net;
3504 if (net_eq(net, &init_net))
3507 old_net = current->nsproxy->net_ns;
3508 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3512 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3513 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3514 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3516 hash_entries = roundup_pow_of_two(hash_entries);
3518 udptable = udp_pernet_table_alloc(hash_entries);
3520 net->ipv4.udp_table = udptable;
3522 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3523 "for a netns, fallback to the global one\n",
3526 net->ipv4.udp_table = &udp_table;
3530 static int __net_init udp_pernet_init(struct net *net)
3532 udp_sysctl_init(net);
3538 static void __net_exit udp_pernet_exit(struct net *net)
3540 udp_pernet_table_free(net);
3543 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3544 .init = udp_pernet_init,
3545 .exit = udp_pernet_exit,
3548 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3549 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3550 struct udp_sock *udp_sk, uid_t uid, int bucket)
3552 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3553 unsigned int new_batch_sz)
3555 struct sock **new_batch;
3557 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3558 GFP_USER | __GFP_NOWARN);
3562 bpf_iter_udp_put_batch(iter);
3563 kvfree(iter->batch);
3564 iter->batch = new_batch;
3565 iter->max_sk = new_batch_sz;
3570 #define INIT_BATCH_SZ 16
3572 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3574 struct bpf_udp_iter_state *iter = priv_data;
3577 ret = bpf_iter_init_seq_net(priv_data, aux);
3581 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3583 bpf_iter_fini_seq_net(priv_data);
3588 static void bpf_iter_fini_udp(void *priv_data)
3590 struct bpf_udp_iter_state *iter = priv_data;
3592 bpf_iter_fini_seq_net(priv_data);
3593 kvfree(iter->batch);
3596 static const struct bpf_iter_seq_info udp_seq_info = {
3597 .seq_ops = &bpf_iter_udp_seq_ops,
3598 .init_seq_private = bpf_iter_init_udp,
3599 .fini_seq_private = bpf_iter_fini_udp,
3600 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3603 static struct bpf_iter_reg udp_reg_info = {
3605 .ctx_arg_info_size = 1,
3607 { offsetof(struct bpf_iter__udp, udp_sk),
3608 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3610 .seq_info = &udp_seq_info,
3613 static void __init bpf_iter_register(void)
3615 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3616 if (bpf_iter_reg_target(&udp_reg_info))
3617 pr_warn("Warning: could not register bpf iterator udp\n");
3621 void __init udp_init(void)
3623 unsigned long limit;
3626 udp_table_init(&udp_table, "UDP");
3627 limit = nr_free_buffer_pages() / 8;
3628 limit = max(limit, 128UL);
3629 sysctl_udp_mem[0] = limit / 4 * 3;
3630 sysctl_udp_mem[1] = limit;
3631 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3633 /* 16 spinlocks per cpu */
3634 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3635 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3638 panic("UDP: failed to alloc udp_busylocks\n");
3639 for (i = 0; i < (1U << udp_busylocks_log); i++)
3640 spin_lock_init(udp_busylocks + i);
3642 if (register_pernet_subsys(&udp_sysctl_ops))
3643 panic("UDP: failed to init sysctl parameters.\n");
3645 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3646 bpf_iter_register();