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 sock_set_flag(sk, SOCK_RCU_FREE);
331 sk_add_node_rcu(sk, &hslot->head);
333 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
335 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
336 spin_lock(&hslot2->lock);
337 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
338 sk->sk_family == AF_INET6)
339 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
342 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
345 spin_unlock(&hslot2->lock);
350 spin_unlock_bh(&hslot->lock);
354 EXPORT_SYMBOL(udp_lib_get_port);
356 int udp_v4_get_port(struct sock *sk, unsigned short snum)
358 unsigned int hash2_nulladdr =
359 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
360 unsigned int hash2_partial =
361 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
363 /* precompute partial secondary hash */
364 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
365 return udp_lib_get_port(sk, snum, hash2_nulladdr);
368 static int compute_score(struct sock *sk, struct net *net,
369 __be32 saddr, __be16 sport,
370 __be32 daddr, unsigned short hnum,
374 struct inet_sock *inet;
377 if (!net_eq(sock_net(sk), net) ||
378 udp_sk(sk)->udp_port_hash != hnum ||
382 if (sk->sk_rcv_saddr != daddr)
385 score = (sk->sk_family == PF_INET) ? 2 : 1;
388 if (inet->inet_daddr) {
389 if (inet->inet_daddr != saddr)
394 if (inet->inet_dport) {
395 if (inet->inet_dport != sport)
400 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
404 if (sk->sk_bound_dev_if)
407 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
412 INDIRECT_CALLABLE_SCOPE
413 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
414 const __be32 faddr, const __be16 fport)
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;
436 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
437 need_rescore = false;
439 score = compute_score(need_rescore ? result : sk, net, saddr,
440 sport, daddr, hnum, dif, sdif);
441 if (score > badness) {
447 if (sk->sk_state == TCP_ESTABLISHED) {
452 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
453 saddr, sport, daddr, hnum, udp_ehashfn);
459 /* Fall back to scoring if group has connections */
460 if (!reuseport_has_conns(sk))
463 /* Reuseport logic returned an error, keep original score. */
467 /* compute_score is too long of a function to be
468 * inlined, and calling it again here yields
469 * measureable overhead for some
470 * workloads. Work around it by jumping
471 * backwards to rescore 'result'.
480 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
481 * harder than this. -DaveM
483 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
484 __be16 sport, __be32 daddr, __be16 dport, int dif,
485 int sdif, struct udp_table *udptable, struct sk_buff *skb)
487 unsigned short hnum = ntohs(dport);
488 unsigned int hash2, slot2;
489 struct udp_hslot *hslot2;
490 struct sock *result, *sk;
492 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
493 slot2 = hash2 & udptable->mask;
494 hslot2 = &udptable->hash2[slot2];
496 /* Lookup connected or non-wildcard socket */
497 result = udp4_lib_lookup2(net, saddr, sport,
498 daddr, hnum, dif, sdif,
500 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
503 /* Lookup redirect from BPF */
504 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
505 udptable == net->ipv4.udp_table) {
506 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
507 saddr, sport, daddr, hnum, dif,
515 /* Got non-wildcard socket or error on first lookup */
519 /* Lookup wildcard sockets */
520 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
521 slot2 = hash2 & udptable->mask;
522 hslot2 = &udptable->hash2[slot2];
524 result = udp4_lib_lookup2(net, saddr, sport,
525 htonl(INADDR_ANY), hnum, dif, sdif,
532 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
534 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
535 __be16 sport, __be16 dport,
536 struct udp_table *udptable)
538 const struct iphdr *iph = ip_hdr(skb);
540 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
541 iph->daddr, dport, inet_iif(skb),
542 inet_sdif(skb), udptable, skb);
545 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
546 __be16 sport, __be16 dport)
548 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
549 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
550 struct net *net = dev_net(skb->dev);
553 inet_get_iif_sdif(skb, &iif, &sdif);
555 return __udp4_lib_lookup(net, iph->saddr, sport,
556 iph->daddr, dport, iif,
557 sdif, net->ipv4.udp_table, NULL);
560 /* Must be called under rcu_read_lock().
561 * Does increment socket refcount.
563 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
564 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
565 __be32 daddr, __be16 dport, int dif)
569 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
570 dif, 0, net->ipv4.udp_table, NULL);
571 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
575 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
578 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
579 __be16 loc_port, __be32 loc_addr,
580 __be16 rmt_port, __be32 rmt_addr,
581 int dif, int sdif, unsigned short hnum)
583 const struct inet_sock *inet = inet_sk(sk);
585 if (!net_eq(sock_net(sk), net) ||
586 udp_sk(sk)->udp_port_hash != hnum ||
587 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
588 (inet->inet_dport != rmt_port && inet->inet_dport) ||
589 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
590 ipv6_only_sock(sk) ||
591 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
593 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
598 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
599 EXPORT_SYMBOL(udp_encap_needed_key);
601 #if IS_ENABLED(CONFIG_IPV6)
602 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
603 EXPORT_SYMBOL(udpv6_encap_needed_key);
606 void udp_encap_enable(void)
608 static_branch_inc(&udp_encap_needed_key);
610 EXPORT_SYMBOL(udp_encap_enable);
612 void udp_encap_disable(void)
614 static_branch_dec(&udp_encap_needed_key);
616 EXPORT_SYMBOL(udp_encap_disable);
618 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
619 * through error handlers in encapsulations looking for a match.
621 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
625 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
626 int (*handler)(struct sk_buff *skb, u32 info);
627 const struct ip_tunnel_encap_ops *encap;
629 encap = rcu_dereference(iptun_encaps[i]);
632 handler = encap->err_handler;
633 if (handler && !handler(skb, info))
640 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
641 * reversing source and destination port: this will match tunnels that force the
642 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
643 * lwtunnels might actually break this assumption by being configured with
644 * different destination ports on endpoints, in this case we won't be able to
645 * trace ICMP messages back to them.
647 * If this doesn't match any socket, probe tunnels with arbitrary destination
648 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
649 * we've sent packets to won't necessarily match the local destination port.
651 * Then ask the tunnel implementation to match the error against a valid
654 * Return an error if we can't find a match, the socket if we need further
655 * processing, zero otherwise.
657 static struct sock *__udp4_lib_err_encap(struct net *net,
658 const struct iphdr *iph,
660 struct udp_table *udptable,
662 struct sk_buff *skb, u32 info)
664 int (*lookup)(struct sock *sk, struct sk_buff *skb);
665 int network_offset, transport_offset;
668 network_offset = skb_network_offset(skb);
669 transport_offset = skb_transport_offset(skb);
671 /* Network header needs to point to the outer IPv4 header inside ICMP */
672 skb_reset_network_header(skb);
674 /* Transport header needs to point to the UDP header */
675 skb_set_transport_header(skb, iph->ihl << 2);
680 lookup = READ_ONCE(up->encap_err_lookup);
681 if (lookup && lookup(sk, skb))
687 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
688 iph->saddr, uh->dest, skb->dev->ifindex, 0,
693 lookup = READ_ONCE(up->encap_err_lookup);
694 if (!lookup || lookup(sk, skb))
700 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
702 skb_set_transport_header(skb, transport_offset);
703 skb_set_network_header(skb, network_offset);
709 * This routine is called by the ICMP module when it gets some
710 * sort of error condition. If err < 0 then the socket should
711 * be closed and the error returned to the user. If err > 0
712 * it's just the icmp type << 8 | icmp code.
713 * Header points to the ip header of the error packet. We move
714 * on past this. Then (as it used to claim before adjustment)
715 * header points to the first 8 bytes of the udp header. We need
716 * to find the appropriate port.
719 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
721 struct inet_sock *inet;
722 const struct iphdr *iph = (const struct iphdr *)skb->data;
723 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
724 const int type = icmp_hdr(skb)->type;
725 const int code = icmp_hdr(skb)->code;
730 struct net *net = dev_net(skb->dev);
732 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
733 iph->saddr, uh->source, skb->dev->ifindex,
734 inet_sdif(skb), udptable, NULL);
736 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
737 /* No socket for error: try tunnels before discarding */
738 if (static_branch_unlikely(&udp_encap_needed_key)) {
739 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
744 sk = ERR_PTR(-ENOENT);
747 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
760 case ICMP_TIME_EXCEEDED:
763 case ICMP_SOURCE_QUENCH:
765 case ICMP_PARAMETERPROB:
769 case ICMP_DEST_UNREACH:
770 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
771 ipv4_sk_update_pmtu(skb, sk, info);
772 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
780 if (code <= NR_ICMP_UNREACH) {
781 harderr = icmp_err_convert[code].fatal;
782 err = icmp_err_convert[code].errno;
786 ipv4_sk_redirect(skb, sk);
791 * RFC1122: OK. Passes ICMP errors back to application, as per
795 /* ...not for tunnels though: we don't have a sending socket */
796 if (udp_sk(sk)->encap_err_rcv)
797 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
801 if (!inet_test_bit(RECVERR, sk)) {
802 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
805 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
813 int udp_err(struct sk_buff *skb, u32 info)
815 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
819 * Throw away all pending data and cancel the corking. Socket is locked.
821 void udp_flush_pending_frames(struct sock *sk)
823 struct udp_sock *up = udp_sk(sk);
827 WRITE_ONCE(up->pending, 0);
828 ip_flush_pending_frames(sk);
831 EXPORT_SYMBOL(udp_flush_pending_frames);
834 * udp4_hwcsum - handle outgoing HW checksumming
835 * @skb: sk_buff containing the filled-in UDP header
836 * (checksum field must be zeroed out)
837 * @src: source IP address
838 * @dst: destination IP address
840 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
842 struct udphdr *uh = udp_hdr(skb);
843 int offset = skb_transport_offset(skb);
844 int len = skb->len - offset;
848 if (!skb_has_frag_list(skb)) {
850 * Only one fragment on the socket.
852 skb->csum_start = skb_transport_header(skb) - skb->head;
853 skb->csum_offset = offsetof(struct udphdr, check);
854 uh->check = ~csum_tcpudp_magic(src, dst, len,
857 struct sk_buff *frags;
860 * HW-checksum won't work as there are two or more
861 * fragments on the socket so that all csums of sk_buffs
864 skb_walk_frags(skb, frags) {
865 csum = csum_add(csum, frags->csum);
869 csum = skb_checksum(skb, offset, hlen, csum);
870 skb->ip_summed = CHECKSUM_NONE;
872 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
874 uh->check = CSUM_MANGLED_0;
877 EXPORT_SYMBOL_GPL(udp4_hwcsum);
879 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
880 * for the simple case like when setting the checksum for a UDP tunnel.
882 void udp_set_csum(bool nocheck, struct sk_buff *skb,
883 __be32 saddr, __be32 daddr, int len)
885 struct udphdr *uh = udp_hdr(skb);
889 } else if (skb_is_gso(skb)) {
890 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
891 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
893 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
895 uh->check = CSUM_MANGLED_0;
897 skb->ip_summed = CHECKSUM_PARTIAL;
898 skb->csum_start = skb_transport_header(skb) - skb->head;
899 skb->csum_offset = offsetof(struct udphdr, check);
900 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
903 EXPORT_SYMBOL(udp_set_csum);
905 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
906 struct inet_cork *cork)
908 struct sock *sk = skb->sk;
909 struct inet_sock *inet = inet_sk(sk);
912 int is_udplite = IS_UDPLITE(sk);
913 int offset = skb_transport_offset(skb);
914 int len = skb->len - offset;
915 int datalen = len - sizeof(*uh);
919 * Create a UDP header
922 uh->source = inet->inet_sport;
923 uh->dest = fl4->fl4_dport;
924 uh->len = htons(len);
927 if (cork->gso_size) {
928 const int hlen = skb_network_header_len(skb) +
929 sizeof(struct udphdr);
931 if (hlen + cork->gso_size > cork->fragsize) {
935 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
939 if (sk->sk_no_check_tx) {
943 if (is_udplite || dst_xfrm(skb_dst(skb))) {
948 if (datalen > cork->gso_size) {
949 skb_shinfo(skb)->gso_size = cork->gso_size;
950 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
951 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
957 if (is_udplite) /* UDP-Lite */
958 csum = udplite_csum(skb);
960 else if (sk->sk_no_check_tx) { /* UDP csum off */
962 skb->ip_summed = CHECKSUM_NONE;
965 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
968 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
972 csum = udp_csum(skb);
974 /* add protocol-dependent pseudo-header */
975 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
976 sk->sk_protocol, csum);
978 uh->check = CSUM_MANGLED_0;
981 err = ip_send_skb(sock_net(sk), skb);
983 if (err == -ENOBUFS &&
984 !inet_test_bit(RECVERR, sk)) {
985 UDP_INC_STATS(sock_net(sk),
986 UDP_MIB_SNDBUFERRORS, is_udplite);
990 UDP_INC_STATS(sock_net(sk),
991 UDP_MIB_OUTDATAGRAMS, is_udplite);
996 * Push out all pending data as one UDP datagram. Socket is locked.
998 int udp_push_pending_frames(struct sock *sk)
1000 struct udp_sock *up = udp_sk(sk);
1001 struct inet_sock *inet = inet_sk(sk);
1002 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1003 struct sk_buff *skb;
1006 skb = ip_finish_skb(sk, fl4);
1010 err = udp_send_skb(skb, fl4, &inet->cork.base);
1014 WRITE_ONCE(up->pending, 0);
1017 EXPORT_SYMBOL(udp_push_pending_frames);
1019 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1021 switch (cmsg->cmsg_type) {
1023 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1025 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1032 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1034 struct cmsghdr *cmsg;
1035 bool need_ip = false;
1038 for_each_cmsghdr(cmsg, msg) {
1039 if (!CMSG_OK(msg, cmsg))
1042 if (cmsg->cmsg_level != SOL_UDP) {
1047 err = __udp_cmsg_send(cmsg, gso_size);
1054 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1056 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1058 struct inet_sock *inet = inet_sk(sk);
1059 struct udp_sock *up = udp_sk(sk);
1060 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1061 struct flowi4 fl4_stack;
1064 struct ipcm_cookie ipc;
1065 struct rtable *rt = NULL;
1068 __be32 daddr, faddr, saddr;
1071 int err, is_udplite = IS_UDPLITE(sk);
1072 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1073 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1074 struct sk_buff *skb;
1075 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;
1091 if (READ_ONCE(up->pending)) {
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);
1144 if (unlikely(err < 0)) {
1152 struct ip_options_rcu *inet_opt;
1155 inet_opt = rcu_dereference(inet->inet_opt);
1157 memcpy(&opt_copy, inet_opt,
1158 sizeof(*inet_opt) + inet_opt->opt.optlen);
1159 ipc.opt = &opt_copy.opt;
1164 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1165 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1166 (struct sockaddr *)usin,
1172 if (usin->sin_port == 0) {
1173 /* BPF program set invalid port. Reject it. */
1177 daddr = usin->sin_addr.s_addr;
1178 dport = usin->sin_port;
1183 ipc.addr = faddr = daddr;
1185 if (ipc.opt && ipc.opt->opt.srr) {
1190 faddr = ipc.opt->opt.faddr;
1193 tos = get_rttos(&ipc, inet);
1194 scope = ip_sendmsg_scope(inet, &ipc, msg);
1195 if (scope == RT_SCOPE_LINK)
1198 uc_index = READ_ONCE(inet->uc_index);
1199 if (ipv4_is_multicast(daddr)) {
1200 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1201 ipc.oif = READ_ONCE(inet->mc_index);
1203 saddr = READ_ONCE(inet->mc_addr);
1205 } else if (!ipc.oif) {
1207 } else if (ipv4_is_lbcast(daddr) && uc_index) {
1208 /* oif is set, packet is to local broadcast and
1209 * uc_index is set. oif is most likely set
1210 * by sk_bound_dev_if. If uc_index != oif check if the
1211 * oif is an L3 master and uc_index is an L3 slave.
1212 * If so, we want to allow the send using the uc_index.
1214 if (ipc.oif != uc_index &&
1215 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1222 rt = dst_rtable(sk_dst_check(sk, 0));
1225 struct net *net = sock_net(sk);
1226 __u8 flow_flags = inet_sk_flowi_flags(sk);
1230 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1231 sk->sk_protocol, flow_flags, faddr, saddr,
1232 dport, inet->inet_sport, sk->sk_uid);
1234 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1235 rt = ip_route_output_flow(net, fl4, sk);
1239 if (err == -ENETUNREACH)
1240 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1245 if ((rt->rt_flags & RTCF_BROADCAST) &&
1246 !sock_flag(sk, SOCK_BROADCAST))
1249 sk_dst_set(sk, dst_clone(&rt->dst));
1252 if (msg->msg_flags&MSG_CONFIRM)
1258 daddr = ipc.addr = fl4->daddr;
1260 /* Lockless fast path for the non-corking case. */
1262 struct inet_cork cork;
1264 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1265 sizeof(struct udphdr), &ipc, &rt,
1266 &cork, msg->msg_flags);
1268 if (!IS_ERR_OR_NULL(skb))
1269 err = udp_send_skb(skb, fl4, &cork);
1274 if (unlikely(up->pending)) {
1275 /* The socket is already corked while preparing it. */
1276 /* ... which is an evident application bug. --ANK */
1279 net_dbg_ratelimited("socket already corked\n");
1284 * Now cork the socket to pend data.
1286 fl4 = &inet->cork.fl.u.ip4;
1289 fl4->fl4_dport = dport;
1290 fl4->fl4_sport = inet->inet_sport;
1291 WRITE_ONCE(up->pending, AF_INET);
1295 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1296 sizeof(struct udphdr), &ipc, &rt,
1297 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1299 udp_flush_pending_frames(sk);
1301 err = udp_push_pending_frames(sk);
1302 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1303 WRITE_ONCE(up->pending, 0);
1314 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1315 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1316 * we don't have a good statistic (IpOutDiscards but it can be too many
1317 * things). We could add another new stat but at least for now that
1318 * seems like overkill.
1320 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1321 UDP_INC_STATS(sock_net(sk),
1322 UDP_MIB_SNDBUFERRORS, is_udplite);
1327 if (msg->msg_flags & MSG_PROBE)
1328 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1329 if (!(msg->msg_flags&MSG_PROBE) || len)
1330 goto back_from_confirm;
1334 EXPORT_SYMBOL(udp_sendmsg);
1336 void udp_splice_eof(struct socket *sock)
1338 struct sock *sk = sock->sk;
1339 struct udp_sock *up = udp_sk(sk);
1341 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1345 if (up->pending && !udp_test_bit(CORK, sk))
1346 udp_push_pending_frames(sk);
1349 EXPORT_SYMBOL_GPL(udp_splice_eof);
1351 #define UDP_SKB_IS_STATELESS 0x80000000
1353 /* all head states (dst, sk, nf conntrack) except skb extensions are
1354 * cleared by udp_rcv().
1356 * We need to preserve secpath, if present, to eventually process
1357 * IP_CMSG_PASSSEC at recvmsg() time.
1359 * Other extensions can be cleared.
1361 static bool udp_try_make_stateless(struct sk_buff *skb)
1363 if (!skb_has_extensions(skb))
1366 if (!secpath_exists(skb)) {
1374 static void udp_set_dev_scratch(struct sk_buff *skb)
1376 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1378 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1379 scratch->_tsize_state = skb->truesize;
1380 #if BITS_PER_LONG == 64
1381 scratch->len = skb->len;
1382 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1383 scratch->is_linear = !skb_is_nonlinear(skb);
1385 if (udp_try_make_stateless(skb))
1386 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1389 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1391 /* We come here after udp_lib_checksum_complete() returned 0.
1392 * This means that __skb_checksum_complete() might have
1393 * set skb->csum_valid to 1.
1394 * On 64bit platforms, we can set csum_unnecessary
1395 * to true, but only if the skb is not shared.
1397 #if BITS_PER_LONG == 64
1398 if (!skb_shared(skb))
1399 udp_skb_scratch(skb)->csum_unnecessary = true;
1403 static int udp_skb_truesize(struct sk_buff *skb)
1405 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1408 static bool udp_skb_has_head_state(struct sk_buff *skb)
1410 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1413 /* fully reclaim rmem/fwd memory allocated for skb */
1414 static void udp_rmem_release(struct sock *sk, int size, int partial,
1415 bool rx_queue_lock_held)
1417 struct udp_sock *up = udp_sk(sk);
1418 struct sk_buff_head *sk_queue;
1421 if (likely(partial)) {
1422 up->forward_deficit += size;
1423 size = up->forward_deficit;
1424 if (size < READ_ONCE(up->forward_threshold) &&
1425 !skb_queue_empty(&up->reader_queue))
1428 size += up->forward_deficit;
1430 up->forward_deficit = 0;
1432 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1433 * if the called don't held it already
1435 sk_queue = &sk->sk_receive_queue;
1436 if (!rx_queue_lock_held)
1437 spin_lock(&sk_queue->lock);
1440 sk_forward_alloc_add(sk, size);
1441 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1442 sk_forward_alloc_add(sk, -amt);
1445 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1447 atomic_sub(size, &sk->sk_rmem_alloc);
1449 /* this can save us from acquiring the rx queue lock on next receive */
1450 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1452 if (!rx_queue_lock_held)
1453 spin_unlock(&sk_queue->lock);
1456 /* Note: called with reader_queue.lock held.
1457 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1458 * This avoids a cache line miss while receive_queue lock is held.
1459 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1461 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1463 prefetch(&skb->data);
1464 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1466 EXPORT_SYMBOL(udp_skb_destructor);
1468 /* as above, but the caller held the rx queue lock, too */
1469 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1471 prefetch(&skb->data);
1472 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1475 /* Idea of busylocks is to let producers grab an extra spinlock
1476 * to relieve pressure on the receive_queue spinlock shared by consumer.
1477 * Under flood, this means that only one producer can be in line
1478 * trying to acquire the receive_queue spinlock.
1479 * These busylock can be allocated on a per cpu manner, instead of a
1480 * per socket one (that would consume a cache line per socket)
1482 static int udp_busylocks_log __read_mostly;
1483 static spinlock_t *udp_busylocks __read_mostly;
1485 static spinlock_t *busylock_acquire(void *ptr)
1489 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1494 static void busylock_release(spinlock_t *busy)
1500 static int udp_rmem_schedule(struct sock *sk, int size)
1504 delta = size - sk->sk_forward_alloc;
1505 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1511 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1513 struct sk_buff_head *list = &sk->sk_receive_queue;
1514 int rmem, err = -ENOMEM;
1515 spinlock_t *busy = NULL;
1516 bool becomes_readable;
1519 /* Immediately drop when the receive queue is full.
1520 * Always allow at least one packet.
1522 rmem = atomic_read(&sk->sk_rmem_alloc);
1523 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1527 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1528 * having linear skbs :
1529 * - Reduce memory overhead and thus increase receive queue capacity
1530 * - Less cache line misses at copyout() time
1531 * - Less work at consume_skb() (less alien page frag freeing)
1533 if (rmem > (rcvbuf >> 1)) {
1536 busy = busylock_acquire(sk);
1538 size = skb->truesize;
1539 udp_set_dev_scratch(skb);
1541 atomic_add(size, &sk->sk_rmem_alloc);
1543 spin_lock(&list->lock);
1544 err = udp_rmem_schedule(sk, size);
1546 spin_unlock(&list->lock);
1550 sk_forward_alloc_add(sk, -size);
1552 /* no need to setup a destructor, we will explicitly release the
1553 * forward allocated memory on dequeue
1555 sock_skb_set_dropcount(sk, skb);
1557 becomes_readable = skb_queue_empty(list);
1558 __skb_queue_tail(list, skb);
1559 spin_unlock(&list->lock);
1561 if (!sock_flag(sk, SOCK_DEAD)) {
1562 if (becomes_readable ||
1563 sk->sk_data_ready != sock_def_readable ||
1564 READ_ONCE(sk->sk_peek_off) >= 0)
1565 INDIRECT_CALL_1(sk->sk_data_ready,
1566 sock_def_readable, sk);
1568 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
1570 busylock_release(busy);
1574 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1577 atomic_inc(&sk->sk_drops);
1578 busylock_release(busy);
1581 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1583 void udp_destruct_common(struct sock *sk)
1585 /* reclaim completely the forward allocated memory */
1586 struct udp_sock *up = udp_sk(sk);
1587 unsigned int total = 0;
1588 struct sk_buff *skb;
1590 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1591 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1592 total += skb->truesize;
1595 udp_rmem_release(sk, total, 0, true);
1597 EXPORT_SYMBOL_GPL(udp_destruct_common);
1599 static void udp_destruct_sock(struct sock *sk)
1601 udp_destruct_common(sk);
1602 inet_sock_destruct(sk);
1605 int udp_init_sock(struct sock *sk)
1607 udp_lib_init_sock(sk);
1608 sk->sk_destruct = udp_destruct_sock;
1609 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1613 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1615 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
1616 sk_peek_offset_bwd(sk, len);
1618 if (!skb_unref(skb))
1621 /* In the more common cases we cleared the head states previously,
1622 * see __udp_queue_rcv_skb().
1624 if (unlikely(udp_skb_has_head_state(skb)))
1625 skb_release_head_state(skb);
1626 __consume_stateless_skb(skb);
1628 EXPORT_SYMBOL_GPL(skb_consume_udp);
1630 static struct sk_buff *__first_packet_length(struct sock *sk,
1631 struct sk_buff_head *rcvq,
1634 struct sk_buff *skb;
1636 while ((skb = skb_peek(rcvq)) != NULL) {
1637 if (udp_lib_checksum_complete(skb)) {
1638 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1640 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1642 atomic_inc(&sk->sk_drops);
1643 __skb_unlink(skb, rcvq);
1644 *total += skb->truesize;
1647 udp_skb_csum_unnecessary_set(skb);
1655 * first_packet_length - return length of first packet in receive queue
1658 * Drops all bad checksum frames, until a valid one is found.
1659 * Returns the length of found skb, or -1 if none is found.
1661 static int first_packet_length(struct sock *sk)
1663 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1664 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1665 struct sk_buff *skb;
1669 spin_lock_bh(&rcvq->lock);
1670 skb = __first_packet_length(sk, rcvq, &total);
1671 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1672 spin_lock(&sk_queue->lock);
1673 skb_queue_splice_tail_init(sk_queue, rcvq);
1674 spin_unlock(&sk_queue->lock);
1676 skb = __first_packet_length(sk, rcvq, &total);
1678 res = skb ? skb->len : -1;
1680 udp_rmem_release(sk, total, 1, false);
1681 spin_unlock_bh(&rcvq->lock);
1686 * IOCTL requests applicable to the UDP protocol
1689 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1694 *karg = sk_wmem_alloc_get(sk);
1700 *karg = max_t(int, 0, first_packet_length(sk));
1705 return -ENOIOCTLCMD;
1710 EXPORT_SYMBOL(udp_ioctl);
1712 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1715 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1716 struct sk_buff_head *queue;
1717 struct sk_buff *last;
1721 queue = &udp_sk(sk)->reader_queue;
1722 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1724 struct sk_buff *skb;
1726 error = sock_error(sk);
1732 spin_lock_bh(&queue->lock);
1733 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1736 if (!(flags & MSG_PEEK))
1737 udp_skb_destructor(sk, skb);
1738 spin_unlock_bh(&queue->lock);
1742 if (skb_queue_empty_lockless(sk_queue)) {
1743 spin_unlock_bh(&queue->lock);
1747 /* refill the reader queue and walk it again
1748 * keep both queues locked to avoid re-acquiring
1749 * the sk_receive_queue lock if fwd memory scheduling
1752 spin_lock(&sk_queue->lock);
1753 skb_queue_splice_tail_init(sk_queue, queue);
1755 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1757 if (skb && !(flags & MSG_PEEK))
1758 udp_skb_dtor_locked(sk, skb);
1759 spin_unlock(&sk_queue->lock);
1760 spin_unlock_bh(&queue->lock);
1765 if (!sk_can_busy_loop(sk))
1768 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1769 } while (!skb_queue_empty_lockless(sk_queue));
1771 /* sk_queue is empty, reader_queue may contain peeked packets */
1773 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1775 (struct sk_buff *)sk_queue));
1780 EXPORT_SYMBOL(__skb_recv_udp);
1782 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1784 struct sk_buff *skb;
1788 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1792 if (udp_lib_checksum_complete(skb)) {
1793 int is_udplite = IS_UDPLITE(sk);
1794 struct net *net = sock_net(sk);
1796 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1797 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1798 atomic_inc(&sk->sk_drops);
1803 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1804 return recv_actor(sk, skb);
1806 EXPORT_SYMBOL(udp_read_skb);
1809 * This should be easy, if there is something there we
1810 * return it, otherwise we block.
1813 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1816 struct inet_sock *inet = inet_sk(sk);
1817 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1818 struct sk_buff *skb;
1819 unsigned int ulen, copied;
1820 int off, err, peeking = flags & MSG_PEEK;
1821 int is_udplite = IS_UDPLITE(sk);
1822 bool checksum_valid = false;
1824 if (flags & MSG_ERRQUEUE)
1825 return ip_recv_error(sk, msg, len, addr_len);
1828 off = sk_peek_offset(sk, flags);
1829 skb = __skb_recv_udp(sk, flags, &off, &err);
1833 ulen = udp_skb_len(skb);
1835 if (copied > ulen - off)
1836 copied = ulen - off;
1837 else if (copied < ulen)
1838 msg->msg_flags |= MSG_TRUNC;
1841 * If checksum is needed at all, try to do it while copying the
1842 * data. If the data is truncated, or if we only want a partial
1843 * coverage checksum (UDP-Lite), do it before the copy.
1846 if (copied < ulen || peeking ||
1847 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1848 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1849 !__udp_lib_checksum_complete(skb);
1850 if (!checksum_valid)
1854 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1855 if (udp_skb_is_linear(skb))
1856 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1858 err = skb_copy_datagram_msg(skb, off, msg, copied);
1860 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1866 if (unlikely(err)) {
1868 atomic_inc(&sk->sk_drops);
1869 UDP_INC_STATS(sock_net(sk),
1870 UDP_MIB_INERRORS, is_udplite);
1877 UDP_INC_STATS(sock_net(sk),
1878 UDP_MIB_INDATAGRAMS, is_udplite);
1880 sock_recv_cmsgs(msg, sk, skb);
1882 /* Copy the address. */
1884 sin->sin_family = AF_INET;
1885 sin->sin_port = udp_hdr(skb)->source;
1886 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1887 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1888 *addr_len = sizeof(*sin);
1890 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1891 (struct sockaddr *)sin,
1895 if (udp_test_bit(GRO_ENABLED, sk))
1896 udp_cmsg_recv(msg, sk, skb);
1898 if (inet_cmsg_flags(inet))
1899 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1902 if (flags & MSG_TRUNC)
1905 skb_consume_udp(sk, skb, peeking ? -err : err);
1909 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1910 udp_skb_destructor)) {
1911 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1912 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1916 /* starting over for a new packet, but check if we need to yield */
1918 msg->msg_flags &= ~MSG_TRUNC;
1922 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1924 /* This check is replicated from __ip4_datagram_connect() and
1925 * intended to prevent BPF program called below from accessing bytes
1926 * that are out of the bound specified by user in addr_len.
1928 if (addr_len < sizeof(struct sockaddr_in))
1931 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
1933 EXPORT_SYMBOL(udp_pre_connect);
1935 int __udp_disconnect(struct sock *sk, int flags)
1937 struct inet_sock *inet = inet_sk(sk);
1939 * 1003.1g - break association.
1942 sk->sk_state = TCP_CLOSE;
1943 inet->inet_daddr = 0;
1944 inet->inet_dport = 0;
1945 sock_rps_reset_rxhash(sk);
1946 sk->sk_bound_dev_if = 0;
1947 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1948 inet_reset_saddr(sk);
1949 if (sk->sk_prot->rehash &&
1950 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1951 sk->sk_prot->rehash(sk);
1954 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1955 sk->sk_prot->unhash(sk);
1956 inet->inet_sport = 0;
1961 EXPORT_SYMBOL(__udp_disconnect);
1963 int udp_disconnect(struct sock *sk, int flags)
1966 __udp_disconnect(sk, flags);
1970 EXPORT_SYMBOL(udp_disconnect);
1972 void udp_lib_unhash(struct sock *sk)
1974 if (sk_hashed(sk)) {
1975 struct udp_table *udptable = udp_get_table_prot(sk);
1976 struct udp_hslot *hslot, *hslot2;
1978 hslot = udp_hashslot(udptable, sock_net(sk),
1979 udp_sk(sk)->udp_port_hash);
1980 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1982 spin_lock_bh(&hslot->lock);
1983 if (rcu_access_pointer(sk->sk_reuseport_cb))
1984 reuseport_detach_sock(sk);
1985 if (sk_del_node_init_rcu(sk)) {
1987 inet_sk(sk)->inet_num = 0;
1988 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1990 spin_lock(&hslot2->lock);
1991 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1993 spin_unlock(&hslot2->lock);
1995 spin_unlock_bh(&hslot->lock);
1998 EXPORT_SYMBOL(udp_lib_unhash);
2001 * inet_rcv_saddr was changed, we must rehash secondary hash
2003 void udp_lib_rehash(struct sock *sk, u16 newhash)
2005 if (sk_hashed(sk)) {
2006 struct udp_table *udptable = udp_get_table_prot(sk);
2007 struct udp_hslot *hslot, *hslot2, *nhslot2;
2009 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2010 nhslot2 = udp_hashslot2(udptable, newhash);
2011 udp_sk(sk)->udp_portaddr_hash = newhash;
2013 if (hslot2 != nhslot2 ||
2014 rcu_access_pointer(sk->sk_reuseport_cb)) {
2015 hslot = udp_hashslot(udptable, sock_net(sk),
2016 udp_sk(sk)->udp_port_hash);
2017 /* we must lock primary chain too */
2018 spin_lock_bh(&hslot->lock);
2019 if (rcu_access_pointer(sk->sk_reuseport_cb))
2020 reuseport_detach_sock(sk);
2022 if (hslot2 != nhslot2) {
2023 spin_lock(&hslot2->lock);
2024 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2026 spin_unlock(&hslot2->lock);
2028 spin_lock(&nhslot2->lock);
2029 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2032 spin_unlock(&nhslot2->lock);
2035 spin_unlock_bh(&hslot->lock);
2039 EXPORT_SYMBOL(udp_lib_rehash);
2041 void udp_v4_rehash(struct sock *sk)
2043 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2044 inet_sk(sk)->inet_rcv_saddr,
2045 inet_sk(sk)->inet_num);
2046 udp_lib_rehash(sk, new_hash);
2049 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2053 if (inet_sk(sk)->inet_daddr) {
2054 sock_rps_save_rxhash(sk, skb);
2055 sk_mark_napi_id(sk, skb);
2056 sk_incoming_cpu_update(sk);
2058 sk_mark_napi_id_once(sk, skb);
2061 rc = __udp_enqueue_schedule_skb(sk, skb);
2063 int is_udplite = IS_UDPLITE(sk);
2066 /* Note that an ENOMEM error is charged twice */
2067 if (rc == -ENOMEM) {
2068 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2070 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2072 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2074 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2076 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2077 trace_udp_fail_queue_rcv_skb(rc, sk, skb);
2078 sk_skb_reason_drop(sk, skb, drop_reason);
2088 * >0: "udp encap" protocol resubmission
2090 * Note that in the success and error cases, the skb is assumed to
2091 * have either been requeued or freed.
2093 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2095 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2096 struct udp_sock *up = udp_sk(sk);
2097 int is_udplite = IS_UDPLITE(sk);
2100 * Charge it to the socket, dropping if the queue is full.
2102 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2103 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2108 if (static_branch_unlikely(&udp_encap_needed_key) &&
2109 READ_ONCE(up->encap_type)) {
2110 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2113 * This is an encapsulation socket so pass the skb to
2114 * the socket's udp_encap_rcv() hook. Otherwise, just
2115 * fall through and pass this up the UDP socket.
2116 * up->encap_rcv() returns the following value:
2117 * =0 if skb was successfully passed to the encap
2118 * handler or was discarded by it.
2119 * >0 if skb should be passed on to UDP.
2120 * <0 if skb should be resubmitted as proto -N
2123 /* if we're overly short, let UDP handle it */
2124 encap_rcv = READ_ONCE(up->encap_rcv);
2128 /* Verify checksum before giving to encap */
2129 if (udp_lib_checksum_complete(skb))
2132 ret = encap_rcv(sk, skb);
2134 __UDP_INC_STATS(sock_net(sk),
2135 UDP_MIB_INDATAGRAMS,
2141 /* FALLTHROUGH -- it's a UDP Packet */
2145 * UDP-Lite specific tests, ignored on UDP sockets
2147 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2148 u16 pcrlen = READ_ONCE(up->pcrlen);
2151 * MIB statistics other than incrementing the error count are
2152 * disabled for the following two types of errors: these depend
2153 * on the application settings, not on the functioning of the
2154 * protocol stack as such.
2156 * RFC 3828 here recommends (sec 3.3): "There should also be a
2157 * way ... to ... at least let the receiving application block
2158 * delivery of packets with coverage values less than a value
2159 * provided by the application."
2161 if (pcrlen == 0) { /* full coverage was set */
2162 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2163 UDP_SKB_CB(skb)->cscov, skb->len);
2166 /* The next case involves violating the min. coverage requested
2167 * by the receiver. This is subtle: if receiver wants x and x is
2168 * greater than the buffersize/MTU then receiver will complain
2169 * that it wants x while sender emits packets of smaller size y.
2170 * Therefore the above ...()->partial_cov statement is essential.
2172 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2173 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2174 UDP_SKB_CB(skb)->cscov, pcrlen);
2179 prefetch(&sk->sk_rmem_alloc);
2180 if (rcu_access_pointer(sk->sk_filter) &&
2181 udp_lib_checksum_complete(skb))
2184 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2185 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2189 udp_csum_pull_header(skb);
2191 ipv4_pktinfo_prepare(sk, skb, true);
2192 return __udp_queue_rcv_skb(sk, skb);
2195 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2196 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2198 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2199 atomic_inc(&sk->sk_drops);
2200 sk_skb_reason_drop(sk, skb, drop_reason);
2204 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2206 struct sk_buff *next, *segs;
2209 if (likely(!udp_unexpected_gso(sk, skb)))
2210 return udp_queue_rcv_one_skb(sk, skb);
2212 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2213 __skb_push(skb, -skb_mac_offset(skb));
2214 segs = udp_rcv_segment(sk, skb, true);
2215 skb_list_walk_safe(segs, skb, next) {
2216 __skb_pull(skb, skb_transport_offset(skb));
2218 udp_post_segment_fix_csum(skb);
2219 ret = udp_queue_rcv_one_skb(sk, skb);
2221 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2226 /* For TCP sockets, sk_rx_dst is protected by socket lock
2227 * For UDP, we use xchg() to guard against concurrent changes.
2229 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2231 struct dst_entry *old;
2233 if (dst_hold_safe(dst)) {
2234 old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst)));
2240 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2243 * Multicasts and broadcasts go to each listener.
2245 * Note: called only from the BH handler context.
2247 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2249 __be32 saddr, __be32 daddr,
2250 struct udp_table *udptable,
2253 struct sock *sk, *first = NULL;
2254 unsigned short hnum = ntohs(uh->dest);
2255 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2256 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2257 unsigned int offset = offsetof(typeof(*sk), sk_node);
2258 int dif = skb->dev->ifindex;
2259 int sdif = inet_sdif(skb);
2260 struct hlist_node *node;
2261 struct sk_buff *nskb;
2264 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2266 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2268 hslot = &udptable->hash2[hash2];
2269 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2272 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2273 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2274 uh->source, saddr, dif, sdif, hnum))
2281 nskb = skb_clone(skb, GFP_ATOMIC);
2283 if (unlikely(!nskb)) {
2284 atomic_inc(&sk->sk_drops);
2285 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2287 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2291 if (udp_queue_rcv_skb(sk, nskb) > 0)
2295 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2296 if (use_hash2 && hash2 != hash2_any) {
2302 if (udp_queue_rcv_skb(first, skb) > 0)
2306 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2307 proto == IPPROTO_UDPLITE);
2312 /* Initialize UDP checksum. If exited with zero value (success),
2313 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2314 * Otherwise, csum completion requires checksumming packet body,
2315 * including udp header and folding it to skb->csum.
2317 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2322 UDP_SKB_CB(skb)->partial_cov = 0;
2323 UDP_SKB_CB(skb)->cscov = skb->len;
2325 if (proto == IPPROTO_UDPLITE) {
2326 err = udplite_checksum_init(skb, uh);
2330 if (UDP_SKB_CB(skb)->partial_cov) {
2331 skb->csum = inet_compute_pseudo(skb, proto);
2336 /* Note, we are only interested in != 0 or == 0, thus the
2339 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2340 inet_compute_pseudo);
2344 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2345 /* If SW calculated the value, we know it's bad */
2346 if (skb->csum_complete_sw)
2349 /* HW says the value is bad. Let's validate that.
2350 * skb->csum is no longer the full packet checksum,
2351 * so don't treat it as such.
2353 skb_checksum_complete_unset(skb);
2359 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2360 * return code conversion for ip layer consumption
2362 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2367 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2368 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2370 ret = udp_queue_rcv_skb(sk, skb);
2372 /* a return value > 0 means to resubmit the input, but
2373 * it wants the return to be -protocol, or 0
2381 * All we need to do is get the socket, and then do a checksum.
2384 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2387 struct sock *sk = NULL;
2389 unsigned short ulen;
2390 struct rtable *rt = skb_rtable(skb);
2391 __be32 saddr, daddr;
2392 struct net *net = dev_net(skb->dev);
2396 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2399 * Validate the packet.
2401 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2402 goto drop; /* No space for header. */
2405 ulen = ntohs(uh->len);
2406 saddr = ip_hdr(skb)->saddr;
2407 daddr = ip_hdr(skb)->daddr;
2409 if (ulen > skb->len)
2412 if (proto == IPPROTO_UDP) {
2413 /* UDP validates ulen. */
2414 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2419 if (udp4_csum_init(skb, uh, proto))
2422 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2423 &refcounted, udp_ehashfn);
2428 struct dst_entry *dst = skb_dst(skb);
2431 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2432 udp_sk_rx_dst_set(sk, dst);
2434 ret = udp_unicast_rcv_skb(sk, skb, uh);
2440 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2441 return __udp4_lib_mcast_deliver(net, skb, uh,
2442 saddr, daddr, udptable, proto);
2444 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2446 return udp_unicast_rcv_skb(sk, skb, uh);
2448 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2452 /* No socket. Drop packet silently, if checksum is wrong */
2453 if (udp_lib_checksum_complete(skb))
2456 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2457 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2458 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2461 * Hmm. We got an UDP packet to a port to which we
2462 * don't wanna listen. Ignore it.
2464 sk_skb_reason_drop(sk, skb, drop_reason);
2468 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2469 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2470 proto == IPPROTO_UDPLITE ? "Lite" : "",
2471 &saddr, ntohs(uh->source),
2473 &daddr, ntohs(uh->dest));
2478 * RFC1122: OK. Discards the bad packet silently (as far as
2479 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2481 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2482 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2483 proto == IPPROTO_UDPLITE ? "Lite" : "",
2484 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2486 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2488 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2489 sk_skb_reason_drop(sk, skb, drop_reason);
2493 /* We can only early demux multicast if there is a single matching socket.
2494 * If more than one socket found returns NULL
2496 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2497 __be16 loc_port, __be32 loc_addr,
2498 __be16 rmt_port, __be32 rmt_addr,
2501 struct udp_table *udptable = net->ipv4.udp_table;
2502 unsigned short hnum = ntohs(loc_port);
2503 struct sock *sk, *result;
2504 struct udp_hslot *hslot;
2507 slot = udp_hashfn(net, hnum, udptable->mask);
2508 hslot = &udptable->hash[slot];
2510 /* Do not bother scanning a too big list */
2511 if (hslot->count > 10)
2515 sk_for_each_rcu(sk, &hslot->head) {
2516 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2517 rmt_port, rmt_addr, dif, sdif, hnum)) {
2527 /* For unicast we should only early demux connected sockets or we can
2528 * break forwarding setups. The chains here can be long so only check
2529 * if the first socket is an exact match and if not move on.
2531 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2532 __be16 loc_port, __be32 loc_addr,
2533 __be16 rmt_port, __be32 rmt_addr,
2536 struct udp_table *udptable = net->ipv4.udp_table;
2537 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2538 unsigned short hnum = ntohs(loc_port);
2539 unsigned int hash2, slot2;
2540 struct udp_hslot *hslot2;
2544 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2545 slot2 = hash2 & udptable->mask;
2546 hslot2 = &udptable->hash2[slot2];
2547 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2549 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2550 if (inet_match(net, sk, acookie, ports, dif, sdif))
2552 /* Only check first socket in chain */
2558 int udp_v4_early_demux(struct sk_buff *skb)
2560 struct net *net = dev_net(skb->dev);
2561 struct in_device *in_dev = NULL;
2562 const struct iphdr *iph;
2563 const struct udphdr *uh;
2564 struct sock *sk = NULL;
2565 struct dst_entry *dst;
2566 int dif = skb->dev->ifindex;
2567 int sdif = inet_sdif(skb);
2570 /* validate the packet */
2571 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2577 if (skb->pkt_type == PACKET_MULTICAST) {
2578 in_dev = __in_dev_get_rcu(skb->dev);
2583 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2588 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2589 uh->source, iph->saddr,
2591 } else if (skb->pkt_type == PACKET_HOST) {
2592 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2593 uh->source, iph->saddr, dif, sdif);
2600 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2601 skb->destructor = sock_pfree;
2602 dst = rcu_dereference(sk->sk_rx_dst);
2605 dst = dst_check(dst, 0);
2609 /* set noref for now.
2610 * any place which wants to hold dst has to call
2613 skb_dst_set_noref(skb, dst);
2615 /* for unconnected multicast sockets we need to validate
2616 * the source on each packet
2618 if (!inet_sk(sk)->inet_daddr && in_dev)
2619 return ip_mc_validate_source(skb, iph->daddr,
2621 iph->tos & IPTOS_RT_MASK,
2622 skb->dev, in_dev, &itag);
2627 int udp_rcv(struct sk_buff *skb)
2629 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2632 void udp_destroy_sock(struct sock *sk)
2634 struct udp_sock *up = udp_sk(sk);
2635 bool slow = lock_sock_fast(sk);
2637 /* protects from races with udp_abort() */
2638 sock_set_flag(sk, SOCK_DEAD);
2639 udp_flush_pending_frames(sk);
2640 unlock_sock_fast(sk, slow);
2641 if (static_branch_unlikely(&udp_encap_needed_key)) {
2642 if (up->encap_type) {
2643 void (*encap_destroy)(struct sock *sk);
2644 encap_destroy = READ_ONCE(up->encap_destroy);
2648 if (udp_test_bit(ENCAP_ENABLED, sk))
2649 static_branch_dec(&udp_encap_needed_key);
2653 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2657 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2658 if (family == AF_INET)
2659 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2660 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2661 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2667 * Socket option code for UDP
2669 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2670 sockptr_t optval, unsigned int optlen,
2671 int (*push_pending_frames)(struct sock *))
2673 struct udp_sock *up = udp_sk(sk);
2676 int is_udplite = IS_UDPLITE(sk);
2678 if (level == SOL_SOCKET) {
2679 err = sk_setsockopt(sk, level, optname, optval, optlen);
2681 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2682 sockopt_lock_sock(sk);
2683 /* paired with READ_ONCE in udp_rmem_release() */
2684 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2685 sockopt_release_sock(sk);
2690 if (optlen < sizeof(int))
2693 if (copy_from_sockptr(&val, optval, sizeof(val)))
2696 valbool = val ? 1 : 0;
2701 udp_set_bit(CORK, sk);
2703 udp_clear_bit(CORK, sk);
2705 push_pending_frames(sk);
2714 case UDP_ENCAP_ESPINUDP:
2715 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2716 #if IS_ENABLED(CONFIG_IPV6)
2717 if (sk->sk_family == AF_INET6)
2718 WRITE_ONCE(up->encap_rcv,
2719 ipv6_stub->xfrm6_udp_encap_rcv);
2722 WRITE_ONCE(up->encap_rcv,
2723 xfrm4_udp_encap_rcv);
2726 case UDP_ENCAP_L2TPINUDP:
2727 WRITE_ONCE(up->encap_type, val);
2728 udp_tunnel_encap_enable(sk);
2736 case UDP_NO_CHECK6_TX:
2737 udp_set_no_check6_tx(sk, valbool);
2740 case UDP_NO_CHECK6_RX:
2741 udp_set_no_check6_rx(sk, valbool);
2745 if (val < 0 || val > USHRT_MAX)
2747 WRITE_ONCE(up->gso_size, val);
2752 /* when enabling GRO, accept the related GSO packet type */
2754 udp_tunnel_encap_enable(sk);
2755 udp_assign_bit(GRO_ENABLED, sk, valbool);
2756 udp_assign_bit(ACCEPT_L4, sk, valbool);
2757 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
2761 * UDP-Lite's partial checksum coverage (RFC 3828).
2763 /* The sender sets actual checksum coverage length via this option.
2764 * The case coverage > packet length is handled by send module. */
2765 case UDPLITE_SEND_CSCOV:
2766 if (!is_udplite) /* Disable the option on UDP sockets */
2767 return -ENOPROTOOPT;
2768 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2770 else if (val > USHRT_MAX)
2772 WRITE_ONCE(up->pcslen, val);
2773 udp_set_bit(UDPLITE_SEND_CC, sk);
2776 /* The receiver specifies a minimum checksum coverage value. To make
2777 * sense, this should be set to at least 8 (as done below). If zero is
2778 * used, this again means full checksum coverage. */
2779 case UDPLITE_RECV_CSCOV:
2780 if (!is_udplite) /* Disable the option on UDP sockets */
2781 return -ENOPROTOOPT;
2782 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2784 else if (val > USHRT_MAX)
2786 WRITE_ONCE(up->pcrlen, val);
2787 udp_set_bit(UDPLITE_RECV_CC, sk);
2797 EXPORT_SYMBOL(udp_lib_setsockopt);
2799 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2800 unsigned int optlen)
2802 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2803 return udp_lib_setsockopt(sk, level, optname,
2805 udp_push_pending_frames);
2806 return ip_setsockopt(sk, level, optname, optval, optlen);
2809 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2810 char __user *optval, int __user *optlen)
2812 struct udp_sock *up = udp_sk(sk);
2815 if (get_user(len, optlen))
2821 len = min_t(unsigned int, len, sizeof(int));
2825 val = udp_test_bit(CORK, sk);
2829 val = READ_ONCE(up->encap_type);
2832 case UDP_NO_CHECK6_TX:
2833 val = udp_get_no_check6_tx(sk);
2836 case UDP_NO_CHECK6_RX:
2837 val = udp_get_no_check6_rx(sk);
2841 val = READ_ONCE(up->gso_size);
2845 val = udp_test_bit(GRO_ENABLED, sk);
2848 /* The following two cannot be changed on UDP sockets, the return is
2849 * always 0 (which corresponds to the full checksum coverage of UDP). */
2850 case UDPLITE_SEND_CSCOV:
2851 val = READ_ONCE(up->pcslen);
2854 case UDPLITE_RECV_CSCOV:
2855 val = READ_ONCE(up->pcrlen);
2859 return -ENOPROTOOPT;
2862 if (put_user(len, optlen))
2864 if (copy_to_user(optval, &val, len))
2868 EXPORT_SYMBOL(udp_lib_getsockopt);
2870 int udp_getsockopt(struct sock *sk, int level, int optname,
2871 char __user *optval, int __user *optlen)
2873 if (level == SOL_UDP || level == SOL_UDPLITE)
2874 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2875 return ip_getsockopt(sk, level, optname, optval, optlen);
2879 * udp_poll - wait for a UDP event.
2880 * @file: - file struct
2882 * @wait: - poll table
2884 * This is same as datagram poll, except for the special case of
2885 * blocking sockets. If application is using a blocking fd
2886 * and a packet with checksum error is in the queue;
2887 * then it could get return from select indicating data available
2888 * but then block when reading it. Add special case code
2889 * to work around these arguably broken applications.
2891 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2893 __poll_t mask = datagram_poll(file, sock, wait);
2894 struct sock *sk = sock->sk;
2896 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2897 mask |= EPOLLIN | EPOLLRDNORM;
2899 /* Check for false positives due to checksum errors */
2900 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2901 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2902 mask &= ~(EPOLLIN | EPOLLRDNORM);
2904 /* psock ingress_msg queue should not contain any bad checksum frames */
2905 if (sk_is_readable(sk))
2906 mask |= EPOLLIN | EPOLLRDNORM;
2910 EXPORT_SYMBOL(udp_poll);
2912 int udp_abort(struct sock *sk, int err)
2914 if (!has_current_bpf_ctx())
2917 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2920 if (sock_flag(sk, SOCK_DEAD))
2924 sk_error_report(sk);
2925 __udp_disconnect(sk, 0);
2928 if (!has_current_bpf_ctx())
2933 EXPORT_SYMBOL_GPL(udp_abort);
2935 struct proto udp_prot = {
2937 .owner = THIS_MODULE,
2938 .close = udp_lib_close,
2939 .pre_connect = udp_pre_connect,
2940 .connect = ip4_datagram_connect,
2941 .disconnect = udp_disconnect,
2943 .init = udp_init_sock,
2944 .destroy = udp_destroy_sock,
2945 .setsockopt = udp_setsockopt,
2946 .getsockopt = udp_getsockopt,
2947 .sendmsg = udp_sendmsg,
2948 .recvmsg = udp_recvmsg,
2949 .splice_eof = udp_splice_eof,
2950 .release_cb = ip4_datagram_release_cb,
2951 .hash = udp_lib_hash,
2952 .unhash = udp_lib_unhash,
2953 .rehash = udp_v4_rehash,
2954 .get_port = udp_v4_get_port,
2955 .put_port = udp_lib_unhash,
2956 #ifdef CONFIG_BPF_SYSCALL
2957 .psock_update_sk_prot = udp_bpf_update_proto,
2959 .memory_allocated = &udp_memory_allocated,
2960 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2962 .sysctl_mem = sysctl_udp_mem,
2963 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2964 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2965 .obj_size = sizeof(struct udp_sock),
2966 .h.udp_table = NULL,
2967 .diag_destroy = udp_abort,
2969 EXPORT_SYMBOL(udp_prot);
2971 /* ------------------------------------------------------------------------ */
2972 #ifdef CONFIG_PROC_FS
2974 static unsigned short seq_file_family(const struct seq_file *seq);
2975 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2977 unsigned short family = seq_file_family(seq);
2979 /* AF_UNSPEC is used as a match all */
2980 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2981 net_eq(sock_net(sk), seq_file_net(seq)));
2984 #ifdef CONFIG_BPF_SYSCALL
2985 static const struct seq_operations bpf_iter_udp_seq_ops;
2987 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2990 const struct udp_seq_afinfo *afinfo;
2992 #ifdef CONFIG_BPF_SYSCALL
2993 if (seq->op == &bpf_iter_udp_seq_ops)
2994 return net->ipv4.udp_table;
2997 afinfo = pde_data(file_inode(seq->file));
2998 return afinfo->udp_table ? : net->ipv4.udp_table;
3001 static struct sock *udp_get_first(struct seq_file *seq, int start)
3003 struct udp_iter_state *state = seq->private;
3004 struct net *net = seq_file_net(seq);
3005 struct udp_table *udptable;
3008 udptable = udp_get_table_seq(seq, net);
3010 for (state->bucket = start; state->bucket <= udptable->mask;
3012 struct udp_hslot *hslot = &udptable->hash[state->bucket];
3014 if (hlist_empty(&hslot->head))
3017 spin_lock_bh(&hslot->lock);
3018 sk_for_each(sk, &hslot->head) {
3019 if (seq_sk_match(seq, sk))
3022 spin_unlock_bh(&hslot->lock);
3029 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3031 struct udp_iter_state *state = seq->private;
3032 struct net *net = seq_file_net(seq);
3033 struct udp_table *udptable;
3037 } while (sk && !seq_sk_match(seq, sk));
3040 udptable = udp_get_table_seq(seq, net);
3042 if (state->bucket <= udptable->mask)
3043 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3045 return udp_get_first(seq, state->bucket + 1);
3050 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3052 struct sock *sk = udp_get_first(seq, 0);
3055 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3057 return pos ? NULL : sk;
3060 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3062 struct udp_iter_state *state = seq->private;
3063 state->bucket = MAX_UDP_PORTS;
3065 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3067 EXPORT_SYMBOL(udp_seq_start);
3069 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3073 if (v == SEQ_START_TOKEN)
3074 sk = udp_get_idx(seq, 0);
3076 sk = udp_get_next(seq, v);
3081 EXPORT_SYMBOL(udp_seq_next);
3083 void udp_seq_stop(struct seq_file *seq, void *v)
3085 struct udp_iter_state *state = seq->private;
3086 struct udp_table *udptable;
3088 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3090 if (state->bucket <= udptable->mask)
3091 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3093 EXPORT_SYMBOL(udp_seq_stop);
3095 /* ------------------------------------------------------------------------ */
3096 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3099 struct inet_sock *inet = inet_sk(sp);
3100 __be32 dest = inet->inet_daddr;
3101 __be32 src = inet->inet_rcv_saddr;
3102 __u16 destp = ntohs(inet->inet_dport);
3103 __u16 srcp = ntohs(inet->inet_sport);
3105 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3106 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3107 bucket, src, srcp, dest, destp, sp->sk_state,
3108 sk_wmem_alloc_get(sp),
3111 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3113 refcount_read(&sp->sk_refcnt), sp,
3114 atomic_read(&sp->sk_drops));
3117 int udp4_seq_show(struct seq_file *seq, void *v)
3119 seq_setwidth(seq, 127);
3120 if (v == SEQ_START_TOKEN)
3121 seq_puts(seq, " sl local_address rem_address st tx_queue "
3122 "rx_queue tr tm->when retrnsmt uid timeout "
3123 "inode ref pointer drops");
3125 struct udp_iter_state *state = seq->private;
3127 udp4_format_sock(v, seq, state->bucket);
3133 #ifdef CONFIG_BPF_SYSCALL
3134 struct bpf_iter__udp {
3135 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3136 __bpf_md_ptr(struct udp_sock *, udp_sk);
3137 uid_t uid __aligned(8);
3138 int bucket __aligned(8);
3141 struct bpf_udp_iter_state {
3142 struct udp_iter_state state;
3143 unsigned int cur_sk;
3144 unsigned int end_sk;
3145 unsigned int max_sk;
3147 struct sock **batch;
3148 bool st_bucket_done;
3151 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3152 unsigned int new_batch_sz);
3153 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3155 struct bpf_udp_iter_state *iter = seq->private;
3156 struct udp_iter_state *state = &iter->state;
3157 struct net *net = seq_file_net(seq);
3158 int resume_bucket, resume_offset;
3159 struct udp_table *udptable;
3160 unsigned int batch_sks = 0;
3161 bool resized = false;
3164 resume_bucket = state->bucket;
3165 resume_offset = iter->offset;
3167 /* The current batch is done, so advance the bucket. */
3168 if (iter->st_bucket_done)
3171 udptable = udp_get_table_seq(seq, net);
3174 /* New batch for the next bucket.
3175 * Iterate over the hash table to find a bucket with sockets matching
3176 * the iterator attributes, and return the first matching socket from
3177 * the bucket. The remaining matched sockets from the bucket are batched
3178 * before releasing the bucket lock. This allows BPF programs that are
3179 * called in seq_show to acquire the bucket lock if needed.
3183 iter->st_bucket_done = false;
3186 for (; state->bucket <= udptable->mask; state->bucket++) {
3187 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3189 if (hlist_empty(&hslot2->head))
3193 spin_lock_bh(&hslot2->lock);
3194 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3195 if (seq_sk_match(seq, sk)) {
3196 /* Resume from the last iterated socket at the
3197 * offset in the bucket before iterator was stopped.
3199 if (state->bucket == resume_bucket &&
3200 iter->offset < resume_offset) {
3204 if (iter->end_sk < iter->max_sk) {
3206 iter->batch[iter->end_sk++] = sk;
3211 spin_unlock_bh(&hslot2->lock);
3217 /* All done: no batch made. */
3221 if (iter->end_sk == batch_sks) {
3222 /* Batching is done for the current bucket; return the first
3223 * socket to be iterated from the batch.
3225 iter->st_bucket_done = true;
3228 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3230 /* After allocating a larger batch, retry one more time to grab
3236 return iter->batch[0];
3239 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3241 struct bpf_udp_iter_state *iter = seq->private;
3244 /* Whenever seq_next() is called, the iter->cur_sk is
3245 * done with seq_show(), so unref the iter->cur_sk.
3247 if (iter->cur_sk < iter->end_sk) {
3248 sock_put(iter->batch[iter->cur_sk++]);
3252 /* After updating iter->cur_sk, check if there are more sockets
3253 * available in the current bucket batch.
3255 if (iter->cur_sk < iter->end_sk)
3256 sk = iter->batch[iter->cur_sk];
3258 /* Prepare a new batch. */
3259 sk = bpf_iter_udp_batch(seq);
3265 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3267 /* bpf iter does not support lseek, so it always
3268 * continue from where it was stop()-ped.
3271 return bpf_iter_udp_batch(seq);
3273 return SEQ_START_TOKEN;
3276 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3277 struct udp_sock *udp_sk, uid_t uid, int bucket)
3279 struct bpf_iter__udp ctx;
3281 meta->seq_num--; /* skip SEQ_START_TOKEN */
3283 ctx.udp_sk = udp_sk;
3285 ctx.bucket = bucket;
3286 return bpf_iter_run_prog(prog, &ctx);
3289 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3291 struct udp_iter_state *state = seq->private;
3292 struct bpf_iter_meta meta;
3293 struct bpf_prog *prog;
3294 struct sock *sk = v;
3298 if (v == SEQ_START_TOKEN)
3303 if (unlikely(sk_unhashed(sk))) {
3308 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3310 prog = bpf_iter_get_info(&meta, false);
3311 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3318 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3320 while (iter->cur_sk < iter->end_sk)
3321 sock_put(iter->batch[iter->cur_sk++]);
3324 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3326 struct bpf_udp_iter_state *iter = seq->private;
3327 struct bpf_iter_meta meta;
3328 struct bpf_prog *prog;
3332 prog = bpf_iter_get_info(&meta, true);
3334 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3337 if (iter->cur_sk < iter->end_sk) {
3338 bpf_iter_udp_put_batch(iter);
3339 iter->st_bucket_done = false;
3343 static const struct seq_operations bpf_iter_udp_seq_ops = {
3344 .start = bpf_iter_udp_seq_start,
3345 .next = bpf_iter_udp_seq_next,
3346 .stop = bpf_iter_udp_seq_stop,
3347 .show = bpf_iter_udp_seq_show,
3351 static unsigned short seq_file_family(const struct seq_file *seq)
3353 const struct udp_seq_afinfo *afinfo;
3355 #ifdef CONFIG_BPF_SYSCALL
3356 /* BPF iterator: bpf programs to filter sockets. */
3357 if (seq->op == &bpf_iter_udp_seq_ops)
3361 /* Proc fs iterator */
3362 afinfo = pde_data(file_inode(seq->file));
3363 return afinfo->family;
3366 const struct seq_operations udp_seq_ops = {
3367 .start = udp_seq_start,
3368 .next = udp_seq_next,
3369 .stop = udp_seq_stop,
3370 .show = udp4_seq_show,
3372 EXPORT_SYMBOL(udp_seq_ops);
3374 static struct udp_seq_afinfo udp4_seq_afinfo = {
3379 static int __net_init udp4_proc_init_net(struct net *net)
3381 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3382 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3387 static void __net_exit udp4_proc_exit_net(struct net *net)
3389 remove_proc_entry("udp", net->proc_net);
3392 static struct pernet_operations udp4_net_ops = {
3393 .init = udp4_proc_init_net,
3394 .exit = udp4_proc_exit_net,
3397 int __init udp4_proc_init(void)
3399 return register_pernet_subsys(&udp4_net_ops);
3402 void udp4_proc_exit(void)
3404 unregister_pernet_subsys(&udp4_net_ops);
3406 #endif /* CONFIG_PROC_FS */
3408 static __initdata unsigned long uhash_entries;
3409 static int __init set_uhash_entries(char *str)
3416 ret = kstrtoul(str, 0, &uhash_entries);
3420 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3421 uhash_entries = UDP_HTABLE_SIZE_MIN;
3424 __setup("uhash_entries=", set_uhash_entries);
3426 void __init udp_table_init(struct udp_table *table, const char *name)
3430 table->hash = alloc_large_system_hash(name,
3431 2 * sizeof(struct udp_hslot),
3433 21, /* one slot per 2 MB */
3437 UDP_HTABLE_SIZE_MIN,
3438 UDP_HTABLE_SIZE_MAX);
3440 table->hash2 = table->hash + (table->mask + 1);
3441 for (i = 0; i <= table->mask; i++) {
3442 INIT_HLIST_HEAD(&table->hash[i].head);
3443 table->hash[i].count = 0;
3444 spin_lock_init(&table->hash[i].lock);
3446 for (i = 0; i <= table->mask; i++) {
3447 INIT_HLIST_HEAD(&table->hash2[i].head);
3448 table->hash2[i].count = 0;
3449 spin_lock_init(&table->hash2[i].lock);
3453 u32 udp_flow_hashrnd(void)
3455 static u32 hashrnd __read_mostly;
3457 net_get_random_once(&hashrnd, sizeof(hashrnd));
3461 EXPORT_SYMBOL(udp_flow_hashrnd);
3463 static void __net_init udp_sysctl_init(struct net *net)
3465 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3466 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3468 #ifdef CONFIG_NET_L3_MASTER_DEV
3469 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3473 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3475 struct udp_table *udptable;
3478 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3482 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3483 GFP_KERNEL_ACCOUNT);
3484 if (!udptable->hash)
3487 udptable->hash2 = udptable->hash + hash_entries;
3488 udptable->mask = hash_entries - 1;
3489 udptable->log = ilog2(hash_entries);
3491 for (i = 0; i < hash_entries; i++) {
3492 INIT_HLIST_HEAD(&udptable->hash[i].head);
3493 udptable->hash[i].count = 0;
3494 spin_lock_init(&udptable->hash[i].lock);
3496 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3497 udptable->hash2[i].count = 0;
3498 spin_lock_init(&udptable->hash2[i].lock);
3509 static void __net_exit udp_pernet_table_free(struct net *net)
3511 struct udp_table *udptable = net->ipv4.udp_table;
3513 if (udptable == &udp_table)
3516 kvfree(udptable->hash);
3520 static void __net_init udp_set_table(struct net *net)
3522 struct udp_table *udptable;
3523 unsigned int hash_entries;
3524 struct net *old_net;
3526 if (net_eq(net, &init_net))
3529 old_net = current->nsproxy->net_ns;
3530 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3534 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3535 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3536 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3538 hash_entries = roundup_pow_of_two(hash_entries);
3540 udptable = udp_pernet_table_alloc(hash_entries);
3542 net->ipv4.udp_table = udptable;
3544 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3545 "for a netns, fallback to the global one\n",
3548 net->ipv4.udp_table = &udp_table;
3552 static int __net_init udp_pernet_init(struct net *net)
3554 udp_sysctl_init(net);
3560 static void __net_exit udp_pernet_exit(struct net *net)
3562 udp_pernet_table_free(net);
3565 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3566 .init = udp_pernet_init,
3567 .exit = udp_pernet_exit,
3570 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3571 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3572 struct udp_sock *udp_sk, uid_t uid, int bucket)
3574 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3575 unsigned int new_batch_sz)
3577 struct sock **new_batch;
3579 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3580 GFP_USER | __GFP_NOWARN);
3584 bpf_iter_udp_put_batch(iter);
3585 kvfree(iter->batch);
3586 iter->batch = new_batch;
3587 iter->max_sk = new_batch_sz;
3592 #define INIT_BATCH_SZ 16
3594 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3596 struct bpf_udp_iter_state *iter = priv_data;
3599 ret = bpf_iter_init_seq_net(priv_data, aux);
3603 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3605 bpf_iter_fini_seq_net(priv_data);
3610 static void bpf_iter_fini_udp(void *priv_data)
3612 struct bpf_udp_iter_state *iter = priv_data;
3614 bpf_iter_fini_seq_net(priv_data);
3615 kvfree(iter->batch);
3618 static const struct bpf_iter_seq_info udp_seq_info = {
3619 .seq_ops = &bpf_iter_udp_seq_ops,
3620 .init_seq_private = bpf_iter_init_udp,
3621 .fini_seq_private = bpf_iter_fini_udp,
3622 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3625 static struct bpf_iter_reg udp_reg_info = {
3627 .ctx_arg_info_size = 1,
3629 { offsetof(struct bpf_iter__udp, udp_sk),
3630 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3632 .seq_info = &udp_seq_info,
3635 static void __init bpf_iter_register(void)
3637 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3638 if (bpf_iter_reg_target(&udp_reg_info))
3639 pr_warn("Warning: could not register bpf iterator udp\n");
3643 void __init udp_init(void)
3645 unsigned long limit;
3648 udp_table_init(&udp_table, "UDP");
3649 limit = nr_free_buffer_pages() / 8;
3650 limit = max(limit, 128UL);
3651 sysctl_udp_mem[0] = limit / 4 * 3;
3652 sysctl_udp_mem[1] = limit;
3653 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3655 /* 16 spinlocks per cpu */
3656 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3657 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3660 panic("UDP: failed to alloc udp_busylocks\n");
3661 for (i = 0; i < (1U << udp_busylocks_log); i++)
3662 spin_lock_init(udp_busylocks + i);
3664 if (register_pernet_subsys(&udp_sysctl_ops))
3665 panic("UDP: failed to init sysctl parameters.\n");
3667 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3668 bpf_iter_register();