2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The User Datagram Protocol (UDP).
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
71 * James Chapman : Add L2TP encapsulation type.
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
80 #define pr_fmt(fmt) "UDP: " fmt
82 #include <asm/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/bootmem.h>
85 #include <linux/highmem.h>
86 #include <linux/swap.h>
87 #include <linux/types.h>
88 #include <linux/fcntl.h>
89 #include <linux/module.h>
90 #include <linux/socket.h>
91 #include <linux/sockios.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
95 #include <linux/errno.h>
96 #include <linux/timer.h>
98 #include <linux/inet.h>
99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/route.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <trace/events/udp.h>
112 #include <linux/static_key.h>
113 #include <trace/events/skb.h>
114 #include <net/busy_poll.h>
115 #include "udp_impl.h"
116 #include <net/sock_reuseport.h>
117 #include <net/addrconf.h>
119 struct udp_table udp_table __read_mostly;
120 EXPORT_SYMBOL(udp_table);
122 long sysctl_udp_mem[3] __read_mostly;
123 EXPORT_SYMBOL(sysctl_udp_mem);
125 int sysctl_udp_rmem_min __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_rmem_min);
128 int sysctl_udp_wmem_min __read_mostly;
129 EXPORT_SYMBOL(sysctl_udp_wmem_min);
131 atomic_long_t udp_memory_allocated;
132 EXPORT_SYMBOL(udp_memory_allocated);
134 #define MAX_UDP_PORTS 65536
135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
137 static int udp_lib_lport_inuse(struct net *net, __u16 num,
138 const struct udp_hslot *hslot,
139 unsigned long *bitmap,
141 int (*saddr_comp)(const struct sock *sk1,
142 const struct sock *sk2,
143 bool match_wildcard),
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 (!sk2->sk_reuseport || !sk->sk_reuseport ||
157 rcu_access_pointer(sk->sk_reuseport_cb) ||
158 !uid_eq(uid, sock_i_uid(sk2))) &&
159 saddr_comp(sk, sk2, true)) {
162 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap);
169 * Note: we still hold spinlock of primary hash chain, so no other writer
170 * can insert/delete a socket with local_port == num
172 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
173 struct udp_hslot *hslot2,
175 int (*saddr_comp)(const struct sock *sk1,
176 const struct sock *sk2,
177 bool match_wildcard))
180 kuid_t uid = sock_i_uid(sk);
183 spin_lock(&hslot2->lock);
184 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
185 if (net_eq(sock_net(sk2), net) &&
187 (udp_sk(sk2)->udp_port_hash == num) &&
188 (!sk2->sk_reuse || !sk->sk_reuse) &&
189 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
190 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
191 (!sk2->sk_reuseport || !sk->sk_reuseport ||
192 rcu_access_pointer(sk->sk_reuseport_cb) ||
193 !uid_eq(uid, sock_i_uid(sk2))) &&
194 saddr_comp(sk, sk2, true)) {
199 spin_unlock(&hslot2->lock);
203 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot,
204 int (*saddr_same)(const struct sock *sk1,
205 const struct sock *sk2,
206 bool match_wildcard))
208 struct net *net = sock_net(sk);
209 kuid_t uid = sock_i_uid(sk);
212 sk_for_each(sk2, &hslot->head) {
213 if (net_eq(sock_net(sk2), net) &&
215 sk2->sk_family == sk->sk_family &&
216 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
217 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
218 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
219 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
220 (*saddr_same)(sk, sk2, false)) {
221 return reuseport_add_sock(sk, sk2);
225 /* Initial allocation may have already happened via setsockopt */
226 if (!rcu_access_pointer(sk->sk_reuseport_cb))
227 return reuseport_alloc(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 * @saddr_comp: AF-dependent comparison of bound local IP addresses
237 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
240 int udp_lib_get_port(struct sock *sk, unsigned short snum,
241 int (*saddr_comp)(const struct sock *sk1,
242 const struct sock *sk2,
243 bool match_wildcard),
244 unsigned int hash2_nulladdr)
246 struct udp_hslot *hslot, *hslot2;
247 struct udp_table *udptable = sk->sk_prot->h.udp_table;
249 struct net *net = sock_net(sk);
252 int low, high, remaining;
254 unsigned short first, last;
255 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
257 inet_get_local_port_range(net, &low, &high);
258 remaining = (high - low) + 1;
260 rand = prandom_u32();
261 first = reciprocal_scale(rand, remaining) + low;
263 * force rand to be an odd multiple of UDP_HTABLE_SIZE
265 rand = (rand | 1) * (udptable->mask + 1);
266 last = first + udptable->mask + 1;
268 hslot = udp_hashslot(udptable, net, first);
269 bitmap_zero(bitmap, PORTS_PER_CHAIN);
270 spin_lock_bh(&hslot->lock);
271 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
272 saddr_comp, udptable->log);
276 * Iterate on all possible values of snum for this hash.
277 * Using steps of an odd multiple of UDP_HTABLE_SIZE
278 * give us randomization and full range coverage.
281 if (low <= snum && snum <= high &&
282 !test_bit(snum >> udptable->log, bitmap) &&
283 !inet_is_local_reserved_port(net, snum))
286 } while (snum != first);
287 spin_unlock_bh(&hslot->lock);
288 } while (++first != last);
291 hslot = udp_hashslot(udptable, net, snum);
292 spin_lock_bh(&hslot->lock);
293 if (hslot->count > 10) {
295 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
297 slot2 &= udptable->mask;
298 hash2_nulladdr &= udptable->mask;
300 hslot2 = udp_hashslot2(udptable, slot2);
301 if (hslot->count < hslot2->count)
302 goto scan_primary_hash;
304 exist = udp_lib_lport_inuse2(net, snum, hslot2,
306 if (!exist && (hash2_nulladdr != slot2)) {
307 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
308 exist = udp_lib_lport_inuse2(net, snum, hslot2,
317 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
322 inet_sk(sk)->inet_num = snum;
323 udp_sk(sk)->udp_port_hash = snum;
324 udp_sk(sk)->udp_portaddr_hash ^= snum;
325 if (sk_unhashed(sk)) {
326 if (sk->sk_reuseport &&
327 udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
328 inet_sk(sk)->inet_num = 0;
329 udp_sk(sk)->udp_port_hash = 0;
330 udp_sk(sk)->udp_portaddr_hash ^= snum;
334 sk_add_node_rcu(sk, &hslot->head);
336 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
338 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
339 spin_lock(&hslot2->lock);
340 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
341 sk->sk_family == AF_INET6)
342 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
345 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
348 spin_unlock(&hslot2->lock);
350 sock_set_flag(sk, SOCK_RCU_FREE);
353 spin_unlock_bh(&hslot->lock);
357 EXPORT_SYMBOL(udp_lib_get_port);
359 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
360 * match_wildcard == false: addresses must be exactly the same, i.e.
361 * 0.0.0.0 only equals to 0.0.0.0
363 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
366 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
368 if (!ipv6_only_sock(sk2)) {
369 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
371 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
372 return match_wildcard;
377 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
380 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
383 int udp_v4_get_port(struct sock *sk, unsigned short snum)
385 unsigned int hash2_nulladdr =
386 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
387 unsigned int hash2_partial =
388 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
390 /* precompute partial secondary hash */
391 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
392 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
395 static int compute_score(struct sock *sk, struct net *net,
396 __be32 saddr, __be16 sport,
397 __be32 daddr, unsigned short hnum, int dif)
400 struct inet_sock *inet;
402 if (!net_eq(sock_net(sk), net) ||
403 udp_sk(sk)->udp_port_hash != hnum ||
407 score = (sk->sk_family == PF_INET) ? 2 : 1;
410 if (inet->inet_rcv_saddr) {
411 if (inet->inet_rcv_saddr != daddr)
416 if (inet->inet_daddr) {
417 if (inet->inet_daddr != saddr)
422 if (inet->inet_dport) {
423 if (inet->inet_dport != sport)
428 if (sk->sk_bound_dev_if) {
429 if (sk->sk_bound_dev_if != dif)
433 if (sk->sk_incoming_cpu == raw_smp_processor_id())
438 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
439 const __u16 lport, const __be32 faddr,
442 static u32 udp_ehash_secret __read_mostly;
444 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
446 return __inet_ehashfn(laddr, lport, faddr, fport,
447 udp_ehash_secret + net_hash_mix(net));
450 /* called with rcu_read_lock() */
451 static struct sock *udp4_lib_lookup2(struct net *net,
452 __be32 saddr, __be16 sport,
453 __be32 daddr, unsigned int hnum, int dif,
454 struct udp_hslot *hslot2,
457 struct sock *sk, *result;
458 int score, badness, matches = 0, reuseport = 0;
463 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
464 score = compute_score(sk, net, saddr, sport,
466 if (score > badness) {
467 reuseport = sk->sk_reuseport;
469 hash = udp_ehashfn(net, daddr, hnum,
471 result = reuseport_select_sock(sk, hash, skb,
472 sizeof(struct udphdr));
479 } else if (score == badness && reuseport) {
481 if (reciprocal_scale(hash, matches) == 0)
483 hash = next_pseudo_random32(hash);
489 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
490 * harder than this. -DaveM
492 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
493 __be16 sport, __be32 daddr, __be16 dport,
494 int dif, struct udp_table *udptable, struct sk_buff *skb)
496 struct sock *sk, *result;
497 unsigned short hnum = ntohs(dport);
498 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
499 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
500 int score, badness, matches = 0, reuseport = 0;
503 if (hslot->count > 10) {
504 hash2 = udp4_portaddr_hash(net, daddr, hnum);
505 slot2 = hash2 & udptable->mask;
506 hslot2 = &udptable->hash2[slot2];
507 if (hslot->count < hslot2->count)
510 result = udp4_lib_lookup2(net, saddr, sport,
514 unsigned int old_slot2 = slot2;
515 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
516 slot2 = hash2 & udptable->mask;
517 /* avoid searching the same slot again. */
518 if (unlikely(slot2 == old_slot2))
521 hslot2 = &udptable->hash2[slot2];
522 if (hslot->count < hslot2->count)
525 result = udp4_lib_lookup2(net, saddr, sport,
534 sk_for_each_rcu(sk, &hslot->head) {
535 score = compute_score(sk, net, saddr, sport,
537 if (score > badness) {
538 reuseport = sk->sk_reuseport;
540 hash = udp_ehashfn(net, daddr, hnum,
542 result = reuseport_select_sock(sk, hash, skb,
543 sizeof(struct udphdr));
550 } else if (score == badness && reuseport) {
552 if (reciprocal_scale(hash, matches) == 0)
554 hash = next_pseudo_random32(hash);
559 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
561 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
562 __be16 sport, __be16 dport,
563 struct udp_table *udptable)
565 const struct iphdr *iph = ip_hdr(skb);
567 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
568 iph->daddr, dport, inet_iif(skb),
572 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
573 __be16 sport, __be16 dport)
575 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
577 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
579 /* Must be called under rcu_read_lock().
580 * Does increment socket refcount.
582 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
583 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY)
584 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
585 __be32 daddr, __be16 dport, int dif)
589 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
590 dif, &udp_table, NULL);
591 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
595 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
598 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
599 __be16 loc_port, __be32 loc_addr,
600 __be16 rmt_port, __be32 rmt_addr,
601 int dif, unsigned short hnum)
603 struct inet_sock *inet = inet_sk(sk);
605 if (!net_eq(sock_net(sk), net) ||
606 udp_sk(sk)->udp_port_hash != hnum ||
607 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
608 (inet->inet_dport != rmt_port && inet->inet_dport) ||
609 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
610 ipv6_only_sock(sk) ||
611 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
613 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
619 * This routine is called by the ICMP module when it gets some
620 * sort of error condition. If err < 0 then the socket should
621 * be closed and the error returned to the user. If err > 0
622 * it's just the icmp type << 8 | icmp code.
623 * Header points to the ip header of the error packet. We move
624 * on past this. Then (as it used to claim before adjustment)
625 * header points to the first 8 bytes of the udp header. We need
626 * to find the appropriate port.
629 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
631 struct inet_sock *inet;
632 const struct iphdr *iph = (const struct iphdr *)skb->data;
633 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
634 const int type = icmp_hdr(skb)->type;
635 const int code = icmp_hdr(skb)->code;
639 struct net *net = dev_net(skb->dev);
641 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
642 iph->saddr, uh->source, skb->dev->ifindex, udptable,
645 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
646 return; /* No socket for error */
655 case ICMP_TIME_EXCEEDED:
658 case ICMP_SOURCE_QUENCH:
660 case ICMP_PARAMETERPROB:
664 case ICMP_DEST_UNREACH:
665 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
666 ipv4_sk_update_pmtu(skb, sk, info);
667 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
675 if (code <= NR_ICMP_UNREACH) {
676 harderr = icmp_err_convert[code].fatal;
677 err = icmp_err_convert[code].errno;
681 ipv4_sk_redirect(skb, sk);
686 * RFC1122: OK. Passes ICMP errors back to application, as per
689 if (!inet->recverr) {
690 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
693 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
696 sk->sk_error_report(sk);
701 void udp_err(struct sk_buff *skb, u32 info)
703 __udp4_lib_err(skb, info, &udp_table);
707 * Throw away all pending data and cancel the corking. Socket is locked.
709 void udp_flush_pending_frames(struct sock *sk)
711 struct udp_sock *up = udp_sk(sk);
716 ip_flush_pending_frames(sk);
719 EXPORT_SYMBOL(udp_flush_pending_frames);
722 * udp4_hwcsum - handle outgoing HW checksumming
723 * @skb: sk_buff containing the filled-in UDP header
724 * (checksum field must be zeroed out)
725 * @src: source IP address
726 * @dst: destination IP address
728 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
730 struct udphdr *uh = udp_hdr(skb);
731 int offset = skb_transport_offset(skb);
732 int len = skb->len - offset;
736 if (!skb_has_frag_list(skb)) {
738 * Only one fragment on the socket.
740 skb->csum_start = skb_transport_header(skb) - skb->head;
741 skb->csum_offset = offsetof(struct udphdr, check);
742 uh->check = ~csum_tcpudp_magic(src, dst, len,
745 struct sk_buff *frags;
748 * HW-checksum won't work as there are two or more
749 * fragments on the socket so that all csums of sk_buffs
752 skb_walk_frags(skb, frags) {
753 csum = csum_add(csum, frags->csum);
757 csum = skb_checksum(skb, offset, hlen, csum);
758 skb->ip_summed = CHECKSUM_NONE;
760 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
762 uh->check = CSUM_MANGLED_0;
765 EXPORT_SYMBOL_GPL(udp4_hwcsum);
767 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
768 * for the simple case like when setting the checksum for a UDP tunnel.
770 void udp_set_csum(bool nocheck, struct sk_buff *skb,
771 __be32 saddr, __be32 daddr, int len)
773 struct udphdr *uh = udp_hdr(skb);
777 } else if (skb_is_gso(skb)) {
778 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
779 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
781 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
783 uh->check = CSUM_MANGLED_0;
785 skb->ip_summed = CHECKSUM_PARTIAL;
786 skb->csum_start = skb_transport_header(skb) - skb->head;
787 skb->csum_offset = offsetof(struct udphdr, check);
788 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
791 EXPORT_SYMBOL(udp_set_csum);
793 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
795 struct sock *sk = skb->sk;
796 struct inet_sock *inet = inet_sk(sk);
799 int is_udplite = IS_UDPLITE(sk);
800 int offset = skb_transport_offset(skb);
801 int len = skb->len - offset;
805 * Create a UDP header
808 uh->source = inet->inet_sport;
809 uh->dest = fl4->fl4_dport;
810 uh->len = htons(len);
813 if (is_udplite) /* UDP-Lite */
814 csum = udplite_csum(skb);
816 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
818 skb->ip_summed = CHECKSUM_NONE;
821 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
823 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
827 csum = udp_csum(skb);
829 /* add protocol-dependent pseudo-header */
830 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
831 sk->sk_protocol, csum);
833 uh->check = CSUM_MANGLED_0;
836 err = ip_send_skb(sock_net(sk), skb);
838 if (err == -ENOBUFS && !inet->recverr) {
839 UDP_INC_STATS(sock_net(sk),
840 UDP_MIB_SNDBUFERRORS, is_udplite);
844 UDP_INC_STATS(sock_net(sk),
845 UDP_MIB_OUTDATAGRAMS, is_udplite);
850 * Push out all pending data as one UDP datagram. Socket is locked.
852 int udp_push_pending_frames(struct sock *sk)
854 struct udp_sock *up = udp_sk(sk);
855 struct inet_sock *inet = inet_sk(sk);
856 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
860 skb = ip_finish_skb(sk, fl4);
864 err = udp_send_skb(skb, fl4);
871 EXPORT_SYMBOL(udp_push_pending_frames);
873 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
875 struct inet_sock *inet = inet_sk(sk);
876 struct udp_sock *up = udp_sk(sk);
877 struct flowi4 fl4_stack;
880 struct ipcm_cookie ipc;
881 struct rtable *rt = NULL;
884 __be32 daddr, faddr, saddr;
887 int err, is_udplite = IS_UDPLITE(sk);
888 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
889 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
891 struct ip_options_data opt_copy;
900 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
908 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
910 fl4 = &inet->cork.fl.u.ip4;
913 * There are pending frames.
914 * The socket lock must be held while it's corked.
917 if (likely(up->pending)) {
918 if (unlikely(up->pending != AF_INET)) {
926 ulen += sizeof(struct udphdr);
929 * Get and verify the address.
932 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
933 if (msg->msg_namelen < sizeof(*usin))
935 if (usin->sin_family != AF_INET) {
936 if (usin->sin_family != AF_UNSPEC)
937 return -EAFNOSUPPORT;
940 daddr = usin->sin_addr.s_addr;
941 dport = usin->sin_port;
945 if (sk->sk_state != TCP_ESTABLISHED)
946 return -EDESTADDRREQ;
947 daddr = inet->inet_daddr;
948 dport = inet->inet_dport;
949 /* Open fast path for connected socket.
950 Route will not be used, if at least one option is set.
955 ipc.sockc.tsflags = sk->sk_tsflags;
956 ipc.addr = inet->inet_saddr;
957 ipc.oif = sk->sk_bound_dev_if;
959 if (msg->msg_controllen) {
960 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
970 struct ip_options_rcu *inet_opt;
973 inet_opt = rcu_dereference(inet->inet_opt);
975 memcpy(&opt_copy, inet_opt,
976 sizeof(*inet_opt) + inet_opt->opt.optlen);
977 ipc.opt = &opt_copy.opt;
983 ipc.addr = faddr = daddr;
985 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
987 if (ipc.opt && ipc.opt->opt.srr) {
990 faddr = ipc.opt->opt.faddr;
993 tos = get_rttos(&ipc, inet);
994 if (sock_flag(sk, SOCK_LOCALROUTE) ||
995 (msg->msg_flags & MSG_DONTROUTE) ||
996 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1001 if (ipv4_is_multicast(daddr)) {
1003 ipc.oif = inet->mc_index;
1005 saddr = inet->mc_addr;
1007 } else if (!ipc.oif)
1008 ipc.oif = inet->uc_index;
1011 rt = (struct rtable *)sk_dst_check(sk, 0);
1014 struct net *net = sock_net(sk);
1015 __u8 flow_flags = inet_sk_flowi_flags(sk);
1019 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1020 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1022 faddr, saddr, dport, inet->inet_sport,
1025 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1026 rt = ip_route_output_flow(net, fl4, sk);
1030 if (err == -ENETUNREACH)
1031 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1036 if ((rt->rt_flags & RTCF_BROADCAST) &&
1037 !sock_flag(sk, SOCK_BROADCAST))
1040 sk_dst_set(sk, dst_clone(&rt->dst));
1043 if (msg->msg_flags&MSG_CONFIRM)
1049 daddr = ipc.addr = fl4->daddr;
1051 /* Lockless fast path for the non-corking case. */
1053 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1054 sizeof(struct udphdr), &ipc, &rt,
1057 if (!IS_ERR_OR_NULL(skb))
1058 err = udp_send_skb(skb, fl4);
1063 if (unlikely(up->pending)) {
1064 /* The socket is already corked while preparing it. */
1065 /* ... which is an evident application bug. --ANK */
1068 net_dbg_ratelimited("cork app bug 2\n");
1073 * Now cork the socket to pend data.
1075 fl4 = &inet->cork.fl.u.ip4;
1078 fl4->fl4_dport = dport;
1079 fl4->fl4_sport = inet->inet_sport;
1080 up->pending = AF_INET;
1084 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1085 sizeof(struct udphdr), &ipc, &rt,
1086 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1088 udp_flush_pending_frames(sk);
1090 err = udp_push_pending_frames(sk);
1091 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1102 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1103 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1104 * we don't have a good statistic (IpOutDiscards but it can be too many
1105 * things). We could add another new stat but at least for now that
1106 * seems like overkill.
1108 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1109 UDP_INC_STATS(sock_net(sk),
1110 UDP_MIB_SNDBUFERRORS, is_udplite);
1115 dst_confirm(&rt->dst);
1116 if (!(msg->msg_flags&MSG_PROBE) || len)
1117 goto back_from_confirm;
1121 EXPORT_SYMBOL(udp_sendmsg);
1123 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1124 size_t size, int flags)
1126 struct inet_sock *inet = inet_sk(sk);
1127 struct udp_sock *up = udp_sk(sk);
1130 if (flags & MSG_SENDPAGE_NOTLAST)
1134 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1136 /* Call udp_sendmsg to specify destination address which
1137 * sendpage interface can't pass.
1138 * This will succeed only when the socket is connected.
1140 ret = udp_sendmsg(sk, &msg, 0);
1147 if (unlikely(!up->pending)) {
1150 net_dbg_ratelimited("udp cork app bug 3\n");
1154 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1155 page, offset, size, flags);
1156 if (ret == -EOPNOTSUPP) {
1158 return sock_no_sendpage(sk->sk_socket, page, offset,
1162 udp_flush_pending_frames(sk);
1167 if (!(up->corkflag || (flags&MSG_MORE)))
1168 ret = udp_push_pending_frames(sk);
1176 static void udp_rmem_release(struct sock *sk, int size, int partial)
1180 atomic_sub(size, &sk->sk_rmem_alloc);
1182 spin_lock_bh(&sk->sk_receive_queue.lock);
1183 sk->sk_forward_alloc += size;
1184 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1185 sk->sk_forward_alloc -= amt;
1186 spin_unlock_bh(&sk->sk_receive_queue.lock);
1189 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1192 static void udp_rmem_free(struct sk_buff *skb)
1194 udp_rmem_release(skb->sk, skb->truesize, 1);
1197 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1199 struct sk_buff_head *list = &sk->sk_receive_queue;
1200 int rmem, delta, amt, err = -ENOMEM;
1201 int size = skb->truesize;
1203 /* try to avoid the costly atomic add/sub pair when the receive
1204 * queue is full; always allow at least a packet
1206 rmem = atomic_read(&sk->sk_rmem_alloc);
1207 if (rmem && (rmem + size > sk->sk_rcvbuf))
1210 /* we drop only if the receive buf is full and the receive
1211 * queue contains some other skb
1213 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1214 if ((rmem > sk->sk_rcvbuf) && (rmem > size))
1217 spin_lock(&list->lock);
1218 if (size >= sk->sk_forward_alloc) {
1219 amt = sk_mem_pages(size);
1220 delta = amt << SK_MEM_QUANTUM_SHIFT;
1221 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1223 spin_unlock(&list->lock);
1227 sk->sk_forward_alloc += delta;
1230 sk->sk_forward_alloc -= size;
1232 /* the skb owner in now the udp socket */
1234 skb->destructor = udp_rmem_free;
1236 sock_skb_set_dropcount(sk, skb);
1238 __skb_queue_tail(list, skb);
1239 spin_unlock(&list->lock);
1241 if (!sock_flag(sk, SOCK_DEAD))
1242 sk->sk_data_ready(sk);
1247 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1250 atomic_inc(&sk->sk_drops);
1253 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1255 static void udp_destruct_sock(struct sock *sk)
1257 /* reclaim completely the forward allocated memory */
1258 __skb_queue_purge(&sk->sk_receive_queue);
1259 udp_rmem_release(sk, 0, 0);
1260 inet_sock_destruct(sk);
1263 int udp_init_sock(struct sock *sk)
1265 sk->sk_destruct = udp_destruct_sock;
1268 EXPORT_SYMBOL_GPL(udp_init_sock);
1270 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1272 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1273 bool slow = lock_sock_fast(sk);
1275 sk_peek_offset_bwd(sk, len);
1276 unlock_sock_fast(sk, slow);
1280 EXPORT_SYMBOL_GPL(skb_consume_udp);
1283 * first_packet_length - return length of first packet in receive queue
1286 * Drops all bad checksum frames, until a valid one is found.
1287 * Returns the length of found skb, or -1 if none is found.
1289 static int first_packet_length(struct sock *sk)
1291 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1292 struct sk_buff *skb;
1295 __skb_queue_head_init(&list_kill);
1297 spin_lock_bh(&rcvq->lock);
1298 while ((skb = skb_peek(rcvq)) != NULL &&
1299 udp_lib_checksum_complete(skb)) {
1300 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1302 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1304 atomic_inc(&sk->sk_drops);
1305 __skb_unlink(skb, rcvq);
1306 __skb_queue_tail(&list_kill, skb);
1308 res = skb ? skb->len : -1;
1309 spin_unlock_bh(&rcvq->lock);
1311 __skb_queue_purge(&list_kill);
1316 * IOCTL requests applicable to the UDP protocol
1319 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1324 int amount = sk_wmem_alloc_get(sk);
1326 return put_user(amount, (int __user *)arg);
1331 int amount = max_t(int, 0, first_packet_length(sk));
1333 return put_user(amount, (int __user *)arg);
1337 return -ENOIOCTLCMD;
1342 EXPORT_SYMBOL(udp_ioctl);
1345 * This should be easy, if there is something there we
1346 * return it, otherwise we block.
1349 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1350 int flags, int *addr_len)
1352 struct inet_sock *inet = inet_sk(sk);
1353 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1354 struct sk_buff *skb;
1355 unsigned int ulen, copied;
1356 int peeked, peeking, off;
1358 int is_udplite = IS_UDPLITE(sk);
1359 bool checksum_valid = false;
1361 if (flags & MSG_ERRQUEUE)
1362 return ip_recv_error(sk, msg, len, addr_len);
1365 peeking = off = sk_peek_offset(sk, flags);
1366 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1367 &peeked, &off, &err);
1373 if (copied > ulen - off)
1374 copied = ulen - off;
1375 else if (copied < ulen)
1376 msg->msg_flags |= MSG_TRUNC;
1379 * If checksum is needed at all, try to do it while copying the
1380 * data. If the data is truncated, or if we only want a partial
1381 * coverage checksum (UDP-Lite), do it before the copy.
1384 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov || peeking) {
1385 checksum_valid = !udp_lib_checksum_complete(skb);
1386 if (!checksum_valid)
1390 if (checksum_valid || skb_csum_unnecessary(skb))
1391 err = skb_copy_datagram_msg(skb, off, msg, copied);
1393 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1399 if (unlikely(err)) {
1401 atomic_inc(&sk->sk_drops);
1402 UDP_INC_STATS(sock_net(sk),
1403 UDP_MIB_INERRORS, is_udplite);
1410 UDP_INC_STATS(sock_net(sk),
1411 UDP_MIB_INDATAGRAMS, is_udplite);
1413 sock_recv_ts_and_drops(msg, sk, skb);
1415 /* Copy the address. */
1417 sin->sin_family = AF_INET;
1418 sin->sin_port = udp_hdr(skb)->source;
1419 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1420 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1421 *addr_len = sizeof(*sin);
1423 if (inet->cmsg_flags)
1424 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr), off);
1427 if (flags & MSG_TRUNC)
1430 skb_consume_udp(sk, skb, peeking ? -err : err);
1434 if (!__sk_queue_drop_skb(sk, skb, flags)) {
1435 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1436 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1440 /* starting over for a new packet, but check if we need to yield */
1442 msg->msg_flags &= ~MSG_TRUNC;
1446 int __udp_disconnect(struct sock *sk, int flags)
1448 struct inet_sock *inet = inet_sk(sk);
1450 * 1003.1g - break association.
1453 sk->sk_state = TCP_CLOSE;
1454 inet->inet_daddr = 0;
1455 inet->inet_dport = 0;
1456 sock_rps_reset_rxhash(sk);
1457 sk->sk_bound_dev_if = 0;
1458 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1459 inet_reset_saddr(sk);
1461 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1462 sk->sk_prot->unhash(sk);
1463 inet->inet_sport = 0;
1468 EXPORT_SYMBOL(__udp_disconnect);
1470 int udp_disconnect(struct sock *sk, int flags)
1473 __udp_disconnect(sk, flags);
1477 EXPORT_SYMBOL(udp_disconnect);
1479 void udp_lib_unhash(struct sock *sk)
1481 if (sk_hashed(sk)) {
1482 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1483 struct udp_hslot *hslot, *hslot2;
1485 hslot = udp_hashslot(udptable, sock_net(sk),
1486 udp_sk(sk)->udp_port_hash);
1487 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1489 spin_lock_bh(&hslot->lock);
1490 if (rcu_access_pointer(sk->sk_reuseport_cb))
1491 reuseport_detach_sock(sk);
1492 if (sk_del_node_init_rcu(sk)) {
1494 inet_sk(sk)->inet_num = 0;
1495 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1497 spin_lock(&hslot2->lock);
1498 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1500 spin_unlock(&hslot2->lock);
1502 spin_unlock_bh(&hslot->lock);
1505 EXPORT_SYMBOL(udp_lib_unhash);
1508 * inet_rcv_saddr was changed, we must rehash secondary hash
1510 void udp_lib_rehash(struct sock *sk, u16 newhash)
1512 if (sk_hashed(sk)) {
1513 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1514 struct udp_hslot *hslot, *hslot2, *nhslot2;
1516 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1517 nhslot2 = udp_hashslot2(udptable, newhash);
1518 udp_sk(sk)->udp_portaddr_hash = newhash;
1520 if (hslot2 != nhslot2 ||
1521 rcu_access_pointer(sk->sk_reuseport_cb)) {
1522 hslot = udp_hashslot(udptable, sock_net(sk),
1523 udp_sk(sk)->udp_port_hash);
1524 /* we must lock primary chain too */
1525 spin_lock_bh(&hslot->lock);
1526 if (rcu_access_pointer(sk->sk_reuseport_cb))
1527 reuseport_detach_sock(sk);
1529 if (hslot2 != nhslot2) {
1530 spin_lock(&hslot2->lock);
1531 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1533 spin_unlock(&hslot2->lock);
1535 spin_lock(&nhslot2->lock);
1536 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1539 spin_unlock(&nhslot2->lock);
1542 spin_unlock_bh(&hslot->lock);
1546 EXPORT_SYMBOL(udp_lib_rehash);
1548 static void udp_v4_rehash(struct sock *sk)
1550 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1551 inet_sk(sk)->inet_rcv_saddr,
1552 inet_sk(sk)->inet_num);
1553 udp_lib_rehash(sk, new_hash);
1556 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1560 if (inet_sk(sk)->inet_daddr) {
1561 sock_rps_save_rxhash(sk, skb);
1562 sk_mark_napi_id(sk, skb);
1563 sk_incoming_cpu_update(sk);
1566 rc = __udp_enqueue_schedule_skb(sk, skb);
1568 int is_udplite = IS_UDPLITE(sk);
1570 /* Note that an ENOMEM error is charged twice */
1572 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1574 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1576 trace_udp_fail_queue_rcv_skb(rc, sk);
1583 static struct static_key udp_encap_needed __read_mostly;
1584 void udp_encap_enable(void)
1586 if (!static_key_enabled(&udp_encap_needed))
1587 static_key_slow_inc(&udp_encap_needed);
1589 EXPORT_SYMBOL(udp_encap_enable);
1594 * >0: "udp encap" protocol resubmission
1596 * Note that in the success and error cases, the skb is assumed to
1597 * have either been requeued or freed.
1599 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1601 struct udp_sock *up = udp_sk(sk);
1602 int is_udplite = IS_UDPLITE(sk);
1605 * Charge it to the socket, dropping if the queue is full.
1607 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1611 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1612 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1615 * This is an encapsulation socket so pass the skb to
1616 * the socket's udp_encap_rcv() hook. Otherwise, just
1617 * fall through and pass this up the UDP socket.
1618 * up->encap_rcv() returns the following value:
1619 * =0 if skb was successfully passed to the encap
1620 * handler or was discarded by it.
1621 * >0 if skb should be passed on to UDP.
1622 * <0 if skb should be resubmitted as proto -N
1625 /* if we're overly short, let UDP handle it */
1626 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1630 /* Verify checksum before giving to encap */
1631 if (udp_lib_checksum_complete(skb))
1634 ret = encap_rcv(sk, skb);
1636 __UDP_INC_STATS(sock_net(sk),
1637 UDP_MIB_INDATAGRAMS,
1643 /* FALLTHROUGH -- it's a UDP Packet */
1647 * UDP-Lite specific tests, ignored on UDP sockets
1649 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1652 * MIB statistics other than incrementing the error count are
1653 * disabled for the following two types of errors: these depend
1654 * on the application settings, not on the functioning of the
1655 * protocol stack as such.
1657 * RFC 3828 here recommends (sec 3.3): "There should also be a
1658 * way ... to ... at least let the receiving application block
1659 * delivery of packets with coverage values less than a value
1660 * provided by the application."
1662 if (up->pcrlen == 0) { /* full coverage was set */
1663 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1664 UDP_SKB_CB(skb)->cscov, skb->len);
1667 /* The next case involves violating the min. coverage requested
1668 * by the receiver. This is subtle: if receiver wants x and x is
1669 * greater than the buffersize/MTU then receiver will complain
1670 * that it wants x while sender emits packets of smaller size y.
1671 * Therefore the above ...()->partial_cov statement is essential.
1673 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1674 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1675 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1680 if (rcu_access_pointer(sk->sk_filter) &&
1681 udp_lib_checksum_complete(skb))
1684 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1687 udp_csum_pull_header(skb);
1689 ipv4_pktinfo_prepare(sk, skb);
1690 return __udp_queue_rcv_skb(sk, skb);
1693 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1695 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1696 atomic_inc(&sk->sk_drops);
1701 /* For TCP sockets, sk_rx_dst is protected by socket lock
1702 * For UDP, we use xchg() to guard against concurrent changes.
1704 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1706 struct dst_entry *old;
1709 old = xchg(&sk->sk_rx_dst, dst);
1714 * Multicasts and broadcasts go to each listener.
1716 * Note: called only from the BH handler context.
1718 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1720 __be32 saddr, __be32 daddr,
1721 struct udp_table *udptable,
1724 struct sock *sk, *first = NULL;
1725 unsigned short hnum = ntohs(uh->dest);
1726 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1727 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1728 unsigned int offset = offsetof(typeof(*sk), sk_node);
1729 int dif = skb->dev->ifindex;
1730 struct hlist_node *node;
1731 struct sk_buff *nskb;
1734 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1736 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1738 hslot = &udp_table.hash2[hash2];
1739 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1742 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1743 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1744 uh->source, saddr, dif, hnum))
1751 nskb = skb_clone(skb, GFP_ATOMIC);
1753 if (unlikely(!nskb)) {
1754 atomic_inc(&sk->sk_drops);
1755 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1757 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1761 if (udp_queue_rcv_skb(sk, nskb) > 0)
1765 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1766 if (use_hash2 && hash2 != hash2_any) {
1772 if (udp_queue_rcv_skb(first, skb) > 0)
1776 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
1777 proto == IPPROTO_UDPLITE);
1782 /* Initialize UDP checksum. If exited with zero value (success),
1783 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1784 * Otherwise, csum completion requires chacksumming packet body,
1785 * including udp header and folding it to skb->csum.
1787 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1792 UDP_SKB_CB(skb)->partial_cov = 0;
1793 UDP_SKB_CB(skb)->cscov = skb->len;
1795 if (proto == IPPROTO_UDPLITE) {
1796 err = udplite_checksum_init(skb, uh);
1801 /* Note, we are only interested in != 0 or == 0, thus the
1804 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
1805 inet_compute_pseudo);
1809 * All we need to do is get the socket, and then do a checksum.
1812 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1817 unsigned short ulen;
1818 struct rtable *rt = skb_rtable(skb);
1819 __be32 saddr, daddr;
1820 struct net *net = dev_net(skb->dev);
1823 * Validate the packet.
1825 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1826 goto drop; /* No space for header. */
1829 ulen = ntohs(uh->len);
1830 saddr = ip_hdr(skb)->saddr;
1831 daddr = ip_hdr(skb)->daddr;
1833 if (ulen > skb->len)
1836 if (proto == IPPROTO_UDP) {
1837 /* UDP validates ulen. */
1838 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1843 if (udp4_csum_init(skb, uh, proto))
1846 sk = skb_steal_sock(skb);
1848 struct dst_entry *dst = skb_dst(skb);
1851 if (unlikely(sk->sk_rx_dst != dst))
1852 udp_sk_rx_dst_set(sk, dst);
1854 ret = udp_queue_rcv_skb(sk, skb);
1856 /* a return value > 0 means to resubmit the input, but
1857 * it wants the return to be -protocol, or 0
1864 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1865 return __udp4_lib_mcast_deliver(net, skb, uh,
1866 saddr, daddr, udptable, proto);
1868 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1872 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1873 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1874 inet_compute_pseudo);
1876 ret = udp_queue_rcv_skb(sk, skb);
1878 /* a return value > 0 means to resubmit the input, but
1879 * it wants the return to be -protocol, or 0
1886 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1890 /* No socket. Drop packet silently, if checksum is wrong */
1891 if (udp_lib_checksum_complete(skb))
1894 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1895 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1898 * Hmm. We got an UDP packet to a port to which we
1899 * don't wanna listen. Ignore it.
1905 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1906 proto == IPPROTO_UDPLITE ? "Lite" : "",
1907 &saddr, ntohs(uh->source),
1909 &daddr, ntohs(uh->dest));
1914 * RFC1122: OK. Discards the bad packet silently (as far as
1915 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1917 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1918 proto == IPPROTO_UDPLITE ? "Lite" : "",
1919 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1921 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1923 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1928 /* We can only early demux multicast if there is a single matching socket.
1929 * If more than one socket found returns NULL
1931 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1932 __be16 loc_port, __be32 loc_addr,
1933 __be16 rmt_port, __be32 rmt_addr,
1936 struct sock *sk, *result;
1937 unsigned short hnum = ntohs(loc_port);
1938 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
1939 struct udp_hslot *hslot = &udp_table.hash[slot];
1941 /* Do not bother scanning a too big list */
1942 if (hslot->count > 10)
1946 sk_for_each_rcu(sk, &hslot->head) {
1947 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
1948 rmt_port, rmt_addr, dif, hnum)) {
1958 /* For unicast we should only early demux connected sockets or we can
1959 * break forwarding setups. The chains here can be long so only check
1960 * if the first socket is an exact match and if not move on.
1962 static struct sock *__udp4_lib_demux_lookup(struct net *net,
1963 __be16 loc_port, __be32 loc_addr,
1964 __be16 rmt_port, __be32 rmt_addr,
1967 unsigned short hnum = ntohs(loc_port);
1968 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1969 unsigned int slot2 = hash2 & udp_table.mask;
1970 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1971 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
1972 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1975 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1976 if (INET_MATCH(sk, net, acookie, rmt_addr,
1977 loc_addr, ports, dif))
1979 /* Only check first socket in chain */
1985 void udp_v4_early_demux(struct sk_buff *skb)
1987 struct net *net = dev_net(skb->dev);
1988 const struct iphdr *iph;
1989 const struct udphdr *uh;
1990 struct sock *sk = NULL;
1991 struct dst_entry *dst;
1992 int dif = skb->dev->ifindex;
1995 /* validate the packet */
1996 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2002 if (skb->pkt_type == PACKET_BROADCAST ||
2003 skb->pkt_type == PACKET_MULTICAST) {
2004 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2009 /* we are supposed to accept bcast packets */
2010 if (skb->pkt_type == PACKET_MULTICAST) {
2011 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2017 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2018 uh->source, iph->saddr, dif);
2019 } else if (skb->pkt_type == PACKET_HOST) {
2020 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2021 uh->source, iph->saddr, dif);
2024 if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2))
2028 skb->destructor = sock_efree;
2029 dst = READ_ONCE(sk->sk_rx_dst);
2032 dst = dst_check(dst, 0);
2034 /* DST_NOCACHE can not be used without taking a reference */
2035 if (dst->flags & DST_NOCACHE) {
2036 if (likely(atomic_inc_not_zero(&dst->__refcnt)))
2037 skb_dst_set(skb, dst);
2039 skb_dst_set_noref(skb, dst);
2044 int udp_rcv(struct sk_buff *skb)
2046 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2049 void udp_destroy_sock(struct sock *sk)
2051 struct udp_sock *up = udp_sk(sk);
2052 bool slow = lock_sock_fast(sk);
2053 udp_flush_pending_frames(sk);
2054 unlock_sock_fast(sk, slow);
2055 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2056 void (*encap_destroy)(struct sock *sk);
2057 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2064 * Socket option code for UDP
2066 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2067 char __user *optval, unsigned int optlen,
2068 int (*push_pending_frames)(struct sock *))
2070 struct udp_sock *up = udp_sk(sk);
2073 int is_udplite = IS_UDPLITE(sk);
2075 if (optlen < sizeof(int))
2078 if (get_user(val, (int __user *)optval))
2081 valbool = val ? 1 : 0;
2090 push_pending_frames(sk);
2098 case UDP_ENCAP_ESPINUDP:
2099 case UDP_ENCAP_ESPINUDP_NON_IKE:
2100 up->encap_rcv = xfrm4_udp_encap_rcv;
2102 case UDP_ENCAP_L2TPINUDP:
2103 up->encap_type = val;
2112 case UDP_NO_CHECK6_TX:
2113 up->no_check6_tx = valbool;
2116 case UDP_NO_CHECK6_RX:
2117 up->no_check6_rx = valbool;
2121 * UDP-Lite's partial checksum coverage (RFC 3828).
2123 /* The sender sets actual checksum coverage length via this option.
2124 * The case coverage > packet length is handled by send module. */
2125 case UDPLITE_SEND_CSCOV:
2126 if (!is_udplite) /* Disable the option on UDP sockets */
2127 return -ENOPROTOOPT;
2128 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2130 else if (val > USHRT_MAX)
2133 up->pcflag |= UDPLITE_SEND_CC;
2136 /* The receiver specifies a minimum checksum coverage value. To make
2137 * sense, this should be set to at least 8 (as done below). If zero is
2138 * used, this again means full checksum coverage. */
2139 case UDPLITE_RECV_CSCOV:
2140 if (!is_udplite) /* Disable the option on UDP sockets */
2141 return -ENOPROTOOPT;
2142 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2144 else if (val > USHRT_MAX)
2147 up->pcflag |= UDPLITE_RECV_CC;
2157 EXPORT_SYMBOL(udp_lib_setsockopt);
2159 int udp_setsockopt(struct sock *sk, int level, int optname,
2160 char __user *optval, unsigned int optlen)
2162 if (level == SOL_UDP || level == SOL_UDPLITE)
2163 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2164 udp_push_pending_frames);
2165 return ip_setsockopt(sk, level, optname, optval, optlen);
2168 #ifdef CONFIG_COMPAT
2169 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2170 char __user *optval, unsigned int optlen)
2172 if (level == SOL_UDP || level == SOL_UDPLITE)
2173 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2174 udp_push_pending_frames);
2175 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2179 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2180 char __user *optval, int __user *optlen)
2182 struct udp_sock *up = udp_sk(sk);
2185 if (get_user(len, optlen))
2188 len = min_t(unsigned int, len, sizeof(int));
2199 val = up->encap_type;
2202 case UDP_NO_CHECK6_TX:
2203 val = up->no_check6_tx;
2206 case UDP_NO_CHECK6_RX:
2207 val = up->no_check6_rx;
2210 /* The following two cannot be changed on UDP sockets, the return is
2211 * always 0 (which corresponds to the full checksum coverage of UDP). */
2212 case UDPLITE_SEND_CSCOV:
2216 case UDPLITE_RECV_CSCOV:
2221 return -ENOPROTOOPT;
2224 if (put_user(len, optlen))
2226 if (copy_to_user(optval, &val, len))
2230 EXPORT_SYMBOL(udp_lib_getsockopt);
2232 int udp_getsockopt(struct sock *sk, int level, int optname,
2233 char __user *optval, int __user *optlen)
2235 if (level == SOL_UDP || level == SOL_UDPLITE)
2236 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2237 return ip_getsockopt(sk, level, optname, optval, optlen);
2240 #ifdef CONFIG_COMPAT
2241 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2242 char __user *optval, int __user *optlen)
2244 if (level == SOL_UDP || level == SOL_UDPLITE)
2245 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2246 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2250 * udp_poll - wait for a UDP event.
2251 * @file - file struct
2253 * @wait - poll table
2255 * This is same as datagram poll, except for the special case of
2256 * blocking sockets. If application is using a blocking fd
2257 * and a packet with checksum error is in the queue;
2258 * then it could get return from select indicating data available
2259 * but then block when reading it. Add special case code
2260 * to work around these arguably broken applications.
2262 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2264 unsigned int mask = datagram_poll(file, sock, wait);
2265 struct sock *sk = sock->sk;
2267 sock_rps_record_flow(sk);
2269 /* Check for false positives due to checksum errors */
2270 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2271 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2272 mask &= ~(POLLIN | POLLRDNORM);
2277 EXPORT_SYMBOL(udp_poll);
2279 int udp_abort(struct sock *sk, int err)
2284 sk->sk_error_report(sk);
2285 __udp_disconnect(sk, 0);
2291 EXPORT_SYMBOL_GPL(udp_abort);
2293 struct proto udp_prot = {
2295 .owner = THIS_MODULE,
2296 .close = udp_lib_close,
2297 .connect = ip4_datagram_connect,
2298 .disconnect = udp_disconnect,
2300 .init = udp_init_sock,
2301 .destroy = udp_destroy_sock,
2302 .setsockopt = udp_setsockopt,
2303 .getsockopt = udp_getsockopt,
2304 .sendmsg = udp_sendmsg,
2305 .recvmsg = udp_recvmsg,
2306 .sendpage = udp_sendpage,
2307 .release_cb = ip4_datagram_release_cb,
2308 .hash = udp_lib_hash,
2309 .unhash = udp_lib_unhash,
2310 .rehash = udp_v4_rehash,
2311 .get_port = udp_v4_get_port,
2312 .memory_allocated = &udp_memory_allocated,
2313 .sysctl_mem = sysctl_udp_mem,
2314 .sysctl_wmem = &sysctl_udp_wmem_min,
2315 .sysctl_rmem = &sysctl_udp_rmem_min,
2316 .obj_size = sizeof(struct udp_sock),
2317 .h.udp_table = &udp_table,
2318 #ifdef CONFIG_COMPAT
2319 .compat_setsockopt = compat_udp_setsockopt,
2320 .compat_getsockopt = compat_udp_getsockopt,
2322 .diag_destroy = udp_abort,
2324 EXPORT_SYMBOL(udp_prot);
2326 /* ------------------------------------------------------------------------ */
2327 #ifdef CONFIG_PROC_FS
2329 static struct sock *udp_get_first(struct seq_file *seq, int start)
2332 struct udp_iter_state *state = seq->private;
2333 struct net *net = seq_file_net(seq);
2335 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2337 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2339 if (hlist_empty(&hslot->head))
2342 spin_lock_bh(&hslot->lock);
2343 sk_for_each(sk, &hslot->head) {
2344 if (!net_eq(sock_net(sk), net))
2346 if (sk->sk_family == state->family)
2349 spin_unlock_bh(&hslot->lock);
2356 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2358 struct udp_iter_state *state = seq->private;
2359 struct net *net = seq_file_net(seq);
2363 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2366 if (state->bucket <= state->udp_table->mask)
2367 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2368 return udp_get_first(seq, state->bucket + 1);
2373 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2375 struct sock *sk = udp_get_first(seq, 0);
2378 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2380 return pos ? NULL : sk;
2383 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2385 struct udp_iter_state *state = seq->private;
2386 state->bucket = MAX_UDP_PORTS;
2388 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2391 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2395 if (v == SEQ_START_TOKEN)
2396 sk = udp_get_idx(seq, 0);
2398 sk = udp_get_next(seq, v);
2404 static void udp_seq_stop(struct seq_file *seq, void *v)
2406 struct udp_iter_state *state = seq->private;
2408 if (state->bucket <= state->udp_table->mask)
2409 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2412 int udp_seq_open(struct inode *inode, struct file *file)
2414 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2415 struct udp_iter_state *s;
2418 err = seq_open_net(inode, file, &afinfo->seq_ops,
2419 sizeof(struct udp_iter_state));
2423 s = ((struct seq_file *)file->private_data)->private;
2424 s->family = afinfo->family;
2425 s->udp_table = afinfo->udp_table;
2428 EXPORT_SYMBOL(udp_seq_open);
2430 /* ------------------------------------------------------------------------ */
2431 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2433 struct proc_dir_entry *p;
2436 afinfo->seq_ops.start = udp_seq_start;
2437 afinfo->seq_ops.next = udp_seq_next;
2438 afinfo->seq_ops.stop = udp_seq_stop;
2440 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2441 afinfo->seq_fops, afinfo);
2446 EXPORT_SYMBOL(udp_proc_register);
2448 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2450 remove_proc_entry(afinfo->name, net->proc_net);
2452 EXPORT_SYMBOL(udp_proc_unregister);
2454 /* ------------------------------------------------------------------------ */
2455 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2458 struct inet_sock *inet = inet_sk(sp);
2459 __be32 dest = inet->inet_daddr;
2460 __be32 src = inet->inet_rcv_saddr;
2461 __u16 destp = ntohs(inet->inet_dport);
2462 __u16 srcp = ntohs(inet->inet_sport);
2464 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2465 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2466 bucket, src, srcp, dest, destp, sp->sk_state,
2467 sk_wmem_alloc_get(sp),
2468 sk_rmem_alloc_get(sp),
2470 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2472 atomic_read(&sp->sk_refcnt), sp,
2473 atomic_read(&sp->sk_drops));
2476 int udp4_seq_show(struct seq_file *seq, void *v)
2478 seq_setwidth(seq, 127);
2479 if (v == SEQ_START_TOKEN)
2480 seq_puts(seq, " sl local_address rem_address st tx_queue "
2481 "rx_queue tr tm->when retrnsmt uid timeout "
2482 "inode ref pointer drops");
2484 struct udp_iter_state *state = seq->private;
2486 udp4_format_sock(v, seq, state->bucket);
2492 static const struct file_operations udp_afinfo_seq_fops = {
2493 .owner = THIS_MODULE,
2494 .open = udp_seq_open,
2496 .llseek = seq_lseek,
2497 .release = seq_release_net
2500 /* ------------------------------------------------------------------------ */
2501 static struct udp_seq_afinfo udp4_seq_afinfo = {
2504 .udp_table = &udp_table,
2505 .seq_fops = &udp_afinfo_seq_fops,
2507 .show = udp4_seq_show,
2511 static int __net_init udp4_proc_init_net(struct net *net)
2513 return udp_proc_register(net, &udp4_seq_afinfo);
2516 static void __net_exit udp4_proc_exit_net(struct net *net)
2518 udp_proc_unregister(net, &udp4_seq_afinfo);
2521 static struct pernet_operations udp4_net_ops = {
2522 .init = udp4_proc_init_net,
2523 .exit = udp4_proc_exit_net,
2526 int __init udp4_proc_init(void)
2528 return register_pernet_subsys(&udp4_net_ops);
2531 void udp4_proc_exit(void)
2533 unregister_pernet_subsys(&udp4_net_ops);
2535 #endif /* CONFIG_PROC_FS */
2537 static __initdata unsigned long uhash_entries;
2538 static int __init set_uhash_entries(char *str)
2545 ret = kstrtoul(str, 0, &uhash_entries);
2549 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2550 uhash_entries = UDP_HTABLE_SIZE_MIN;
2553 __setup("uhash_entries=", set_uhash_entries);
2555 void __init udp_table_init(struct udp_table *table, const char *name)
2559 table->hash = alloc_large_system_hash(name,
2560 2 * sizeof(struct udp_hslot),
2562 21, /* one slot per 2 MB */
2566 UDP_HTABLE_SIZE_MIN,
2569 table->hash2 = table->hash + (table->mask + 1);
2570 for (i = 0; i <= table->mask; i++) {
2571 INIT_HLIST_HEAD(&table->hash[i].head);
2572 table->hash[i].count = 0;
2573 spin_lock_init(&table->hash[i].lock);
2575 for (i = 0; i <= table->mask; i++) {
2576 INIT_HLIST_HEAD(&table->hash2[i].head);
2577 table->hash2[i].count = 0;
2578 spin_lock_init(&table->hash2[i].lock);
2582 u32 udp_flow_hashrnd(void)
2584 static u32 hashrnd __read_mostly;
2586 net_get_random_once(&hashrnd, sizeof(hashrnd));
2590 EXPORT_SYMBOL(udp_flow_hashrnd);
2592 void __init udp_init(void)
2594 unsigned long limit;
2596 udp_table_init(&udp_table, "UDP");
2597 limit = nr_free_buffer_pages() / 8;
2598 limit = max(limit, 128UL);
2599 sysctl_udp_mem[0] = limit / 4 * 3;
2600 sysctl_udp_mem[1] = limit;
2601 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2603 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2604 sysctl_udp_wmem_min = SK_MEM_QUANTUM;