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).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
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.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
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 return reuseport_alloc(sk);
229 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
231 * @sk: socket struct in question
232 * @snum: port number to look up
233 * @saddr_comp: AF-dependent comparison of bound local IP addresses
234 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237 int udp_lib_get_port(struct sock *sk, unsigned short snum,
238 int (*saddr_comp)(const struct sock *sk1,
239 const struct sock *sk2,
240 bool match_wildcard),
241 unsigned int hash2_nulladdr)
243 struct udp_hslot *hslot, *hslot2;
244 struct udp_table *udptable = sk->sk_prot->h.udp_table;
246 struct net *net = sock_net(sk);
249 int low, high, remaining;
251 unsigned short first, last;
252 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
254 inet_get_local_port_range(net, &low, &high);
255 remaining = (high - low) + 1;
257 rand = prandom_u32();
258 first = reciprocal_scale(rand, remaining) + low;
260 * force rand to be an odd multiple of UDP_HTABLE_SIZE
262 rand = (rand | 1) * (udptable->mask + 1);
263 last = first + udptable->mask + 1;
265 hslot = udp_hashslot(udptable, net, first);
266 bitmap_zero(bitmap, PORTS_PER_CHAIN);
267 spin_lock_bh(&hslot->lock);
268 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
269 saddr_comp, udptable->log);
273 * Iterate on all possible values of snum for this hash.
274 * Using steps of an odd multiple of UDP_HTABLE_SIZE
275 * give us randomization and full range coverage.
278 if (low <= snum && snum <= high &&
279 !test_bit(snum >> udptable->log, bitmap) &&
280 !inet_is_local_reserved_port(net, snum))
283 } while (snum != first);
284 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,
303 if (!exist && (hash2_nulladdr != slot2)) {
304 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
305 exist = udp_lib_lport_inuse2(net, snum, hslot2,
314 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
319 inet_sk(sk)->inet_num = snum;
320 udp_sk(sk)->udp_port_hash = snum;
321 udp_sk(sk)->udp_portaddr_hash ^= snum;
322 if (sk_unhashed(sk)) {
323 if (sk->sk_reuseport &&
324 udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
325 inet_sk(sk)->inet_num = 0;
326 udp_sk(sk)->udp_port_hash = 0;
327 udp_sk(sk)->udp_portaddr_hash ^= snum;
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);
347 sock_set_flag(sk, SOCK_RCU_FREE);
350 spin_unlock_bh(&hslot->lock);
354 EXPORT_SYMBOL(udp_lib_get_port);
356 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
357 * match_wildcard == false: addresses must be exactly the same, i.e.
358 * 0.0.0.0 only equals to 0.0.0.0
360 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
363 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
365 if (!ipv6_only_sock(sk2)) {
366 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
368 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
369 return match_wildcard;
374 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
377 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
380 int udp_v4_get_port(struct sock *sk, unsigned short snum)
382 unsigned int hash2_nulladdr =
383 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
384 unsigned int hash2_partial =
385 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
387 /* precompute partial secondary hash */
388 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
389 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
392 static int compute_score(struct sock *sk, struct net *net,
393 __be32 saddr, __be16 sport,
394 __be32 daddr, unsigned short hnum, int dif)
397 struct inet_sock *inet;
399 if (!net_eq(sock_net(sk), net) ||
400 udp_sk(sk)->udp_port_hash != hnum ||
404 score = (sk->sk_family == PF_INET) ? 2 : 1;
407 if (inet->inet_rcv_saddr) {
408 if (inet->inet_rcv_saddr != daddr)
413 if (inet->inet_daddr) {
414 if (inet->inet_daddr != saddr)
419 if (inet->inet_dport) {
420 if (inet->inet_dport != sport)
425 if (sk->sk_bound_dev_if) {
426 if (sk->sk_bound_dev_if != dif)
430 if (sk->sk_incoming_cpu == raw_smp_processor_id())
435 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
436 const __u16 lport, const __be32 faddr,
439 static u32 udp_ehash_secret __read_mostly;
441 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
443 return __inet_ehashfn(laddr, lport, faddr, fport,
444 udp_ehash_secret + net_hash_mix(net));
447 /* called with rcu_read_lock() */
448 static struct sock *udp4_lib_lookup2(struct net *net,
449 __be32 saddr, __be16 sport,
450 __be32 daddr, unsigned int hnum, int dif,
451 struct udp_hslot *hslot2,
454 struct sock *sk, *result;
455 int score, badness, matches = 0, reuseport = 0;
460 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
461 score = compute_score(sk, net, saddr, sport,
463 if (score > badness) {
464 reuseport = sk->sk_reuseport;
466 hash = udp_ehashfn(net, daddr, hnum,
468 result = reuseport_select_sock(sk, hash, skb,
469 sizeof(struct udphdr));
476 } else if (score == badness && reuseport) {
478 if (reciprocal_scale(hash, matches) == 0)
480 hash = next_pseudo_random32(hash);
486 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
487 * harder than this. -DaveM
489 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
490 __be16 sport, __be32 daddr, __be16 dport,
491 int dif, struct udp_table *udptable, struct sk_buff *skb)
493 struct sock *sk, *result;
494 unsigned short hnum = ntohs(dport);
495 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
496 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
497 int score, badness, matches = 0, reuseport = 0;
500 if (hslot->count > 10) {
501 hash2 = udp4_portaddr_hash(net, daddr, hnum);
502 slot2 = hash2 & udptable->mask;
503 hslot2 = &udptable->hash2[slot2];
504 if (hslot->count < hslot2->count)
507 result = udp4_lib_lookup2(net, saddr, sport,
511 unsigned int old_slot2 = slot2;
512 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
513 slot2 = hash2 & udptable->mask;
514 /* avoid searching the same slot again. */
515 if (unlikely(slot2 == old_slot2))
518 hslot2 = &udptable->hash2[slot2];
519 if (hslot->count < hslot2->count)
522 result = udp4_lib_lookup2(net, saddr, sport,
531 sk_for_each_rcu(sk, &hslot->head) {
532 score = compute_score(sk, net, saddr, sport,
534 if (score > badness) {
535 reuseport = sk->sk_reuseport;
537 hash = udp_ehashfn(net, daddr, hnum,
539 result = reuseport_select_sock(sk, hash, skb,
540 sizeof(struct udphdr));
547 } else if (score == badness && reuseport) {
549 if (reciprocal_scale(hash, matches) == 0)
551 hash = next_pseudo_random32(hash);
556 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
558 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
559 __be16 sport, __be16 dport,
560 struct udp_table *udptable)
562 const struct iphdr *iph = ip_hdr(skb);
564 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
565 iph->daddr, dport, inet_iif(skb),
569 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
570 __be16 sport, __be16 dport)
572 const struct iphdr *iph = ip_hdr(skb);
574 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
575 iph->daddr, dport, inet_iif(skb),
578 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
580 /* Must be called under rcu_read_lock().
581 * Does increment socket refcount.
583 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
584 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY)
585 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
586 __be32 daddr, __be16 dport, int dif)
590 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
591 dif, &udp_table, NULL);
592 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
596 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
599 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
600 __be16 loc_port, __be32 loc_addr,
601 __be16 rmt_port, __be32 rmt_addr,
602 int dif, unsigned short hnum)
604 struct inet_sock *inet = inet_sk(sk);
606 if (!net_eq(sock_net(sk), net) ||
607 udp_sk(sk)->udp_port_hash != hnum ||
608 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
609 (inet->inet_dport != rmt_port && inet->inet_dport) ||
610 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
611 ipv6_only_sock(sk) ||
612 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
614 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
620 * This routine is called by the ICMP module when it gets some
621 * sort of error condition. If err < 0 then the socket should
622 * be closed and the error returned to the user. If err > 0
623 * it's just the icmp type << 8 | icmp code.
624 * Header points to the ip header of the error packet. We move
625 * on past this. Then (as it used to claim before adjustment)
626 * header points to the first 8 bytes of the udp header. We need
627 * to find the appropriate port.
630 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
632 struct inet_sock *inet;
633 const struct iphdr *iph = (const struct iphdr *)skb->data;
634 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
635 const int type = icmp_hdr(skb)->type;
636 const int code = icmp_hdr(skb)->code;
640 struct net *net = dev_net(skb->dev);
642 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
643 iph->saddr, uh->source, skb->dev->ifindex, udptable,
646 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
647 return; /* No socket for error */
656 case ICMP_TIME_EXCEEDED:
659 case ICMP_SOURCE_QUENCH:
661 case ICMP_PARAMETERPROB:
665 case ICMP_DEST_UNREACH:
666 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
667 ipv4_sk_update_pmtu(skb, sk, info);
668 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
676 if (code <= NR_ICMP_UNREACH) {
677 harderr = icmp_err_convert[code].fatal;
678 err = icmp_err_convert[code].errno;
682 ipv4_sk_redirect(skb, sk);
687 * RFC1122: OK. Passes ICMP errors back to application, as per
690 if (!inet->recverr) {
691 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
694 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
697 sk->sk_error_report(sk);
702 void udp_err(struct sk_buff *skb, u32 info)
704 __udp4_lib_err(skb, info, &udp_table);
708 * Throw away all pending data and cancel the corking. Socket is locked.
710 void udp_flush_pending_frames(struct sock *sk)
712 struct udp_sock *up = udp_sk(sk);
717 ip_flush_pending_frames(sk);
720 EXPORT_SYMBOL(udp_flush_pending_frames);
723 * udp4_hwcsum - handle outgoing HW checksumming
724 * @skb: sk_buff containing the filled-in UDP header
725 * (checksum field must be zeroed out)
726 * @src: source IP address
727 * @dst: destination IP address
729 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
731 struct udphdr *uh = udp_hdr(skb);
732 int offset = skb_transport_offset(skb);
733 int len = skb->len - offset;
737 if (!skb_has_frag_list(skb)) {
739 * Only one fragment on the socket.
741 skb->csum_start = skb_transport_header(skb) - skb->head;
742 skb->csum_offset = offsetof(struct udphdr, check);
743 uh->check = ~csum_tcpudp_magic(src, dst, len,
746 struct sk_buff *frags;
749 * HW-checksum won't work as there are two or more
750 * fragments on the socket so that all csums of sk_buffs
753 skb_walk_frags(skb, frags) {
754 csum = csum_add(csum, frags->csum);
758 csum = skb_checksum(skb, offset, hlen, csum);
759 skb->ip_summed = CHECKSUM_NONE;
761 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
763 uh->check = CSUM_MANGLED_0;
766 EXPORT_SYMBOL_GPL(udp4_hwcsum);
768 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
769 * for the simple case like when setting the checksum for a UDP tunnel.
771 void udp_set_csum(bool nocheck, struct sk_buff *skb,
772 __be32 saddr, __be32 daddr, int len)
774 struct udphdr *uh = udp_hdr(skb);
778 } else if (skb_is_gso(skb)) {
779 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
780 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
782 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
784 uh->check = CSUM_MANGLED_0;
786 skb->ip_summed = CHECKSUM_PARTIAL;
787 skb->csum_start = skb_transport_header(skb) - skb->head;
788 skb->csum_offset = offsetof(struct udphdr, check);
789 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
792 EXPORT_SYMBOL(udp_set_csum);
794 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
796 struct sock *sk = skb->sk;
797 struct inet_sock *inet = inet_sk(sk);
800 int is_udplite = IS_UDPLITE(sk);
801 int offset = skb_transport_offset(skb);
802 int len = skb->len - offset;
806 * Create a UDP header
809 uh->source = inet->inet_sport;
810 uh->dest = fl4->fl4_dport;
811 uh->len = htons(len);
814 if (is_udplite) /* UDP-Lite */
815 csum = udplite_csum(skb);
817 else if (sk->sk_no_check_tx && !skb_is_gso(skb)) { /* UDP csum off */
819 skb->ip_summed = CHECKSUM_NONE;
822 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
824 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
828 csum = udp_csum(skb);
830 /* add protocol-dependent pseudo-header */
831 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
832 sk->sk_protocol, csum);
834 uh->check = CSUM_MANGLED_0;
837 err = ip_send_skb(sock_net(sk), skb);
839 if (err == -ENOBUFS && !inet->recverr) {
840 UDP_INC_STATS(sock_net(sk),
841 UDP_MIB_SNDBUFERRORS, is_udplite);
845 UDP_INC_STATS(sock_net(sk),
846 UDP_MIB_OUTDATAGRAMS, is_udplite);
851 * Push out all pending data as one UDP datagram. Socket is locked.
853 int udp_push_pending_frames(struct sock *sk)
855 struct udp_sock *up = udp_sk(sk);
856 struct inet_sock *inet = inet_sk(sk);
857 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
861 skb = ip_finish_skb(sk, fl4);
865 err = udp_send_skb(skb, fl4);
872 EXPORT_SYMBOL(udp_push_pending_frames);
874 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
876 struct inet_sock *inet = inet_sk(sk);
877 struct udp_sock *up = udp_sk(sk);
878 struct flowi4 fl4_stack;
881 struct ipcm_cookie ipc;
882 struct rtable *rt = NULL;
885 __be32 daddr, faddr, saddr;
888 int err, is_udplite = IS_UDPLITE(sk);
889 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
890 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
892 struct ip_options_data opt_copy;
901 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
909 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
911 fl4 = &inet->cork.fl.u.ip4;
914 * There are pending frames.
915 * The socket lock must be held while it's corked.
918 if (likely(up->pending)) {
919 if (unlikely(up->pending != AF_INET)) {
927 ulen += sizeof(struct udphdr);
930 * Get and verify the address.
933 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
934 if (msg->msg_namelen < sizeof(*usin))
936 if (usin->sin_family != AF_INET) {
937 if (usin->sin_family != AF_UNSPEC)
938 return -EAFNOSUPPORT;
941 daddr = usin->sin_addr.s_addr;
942 dport = usin->sin_port;
946 if (sk->sk_state != TCP_ESTABLISHED)
947 return -EDESTADDRREQ;
948 daddr = inet->inet_daddr;
949 dport = inet->inet_dport;
950 /* Open fast path for connected socket.
951 Route will not be used, if at least one option is set.
956 ipc.sockc.tsflags = sk->sk_tsflags;
957 ipc.addr = inet->inet_saddr;
958 ipc.oif = sk->sk_bound_dev_if;
960 if (msg->msg_controllen) {
961 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
971 struct ip_options_rcu *inet_opt;
974 inet_opt = rcu_dereference(inet->inet_opt);
976 memcpy(&opt_copy, inet_opt,
977 sizeof(*inet_opt) + inet_opt->opt.optlen);
978 ipc.opt = &opt_copy.opt;
984 ipc.addr = faddr = daddr;
986 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
988 if (ipc.opt && ipc.opt->opt.srr) {
993 faddr = ipc.opt->opt.faddr;
996 tos = get_rttos(&ipc, inet);
997 if (sock_flag(sk, SOCK_LOCALROUTE) ||
998 (msg->msg_flags & MSG_DONTROUTE) ||
999 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1004 if (ipv4_is_multicast(daddr)) {
1006 ipc.oif = inet->mc_index;
1008 saddr = inet->mc_addr;
1010 } else if (!ipc.oif)
1011 ipc.oif = inet->uc_index;
1014 rt = (struct rtable *)sk_dst_check(sk, 0);
1017 struct net *net = sock_net(sk);
1018 __u8 flow_flags = inet_sk_flowi_flags(sk);
1022 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1023 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1025 faddr, saddr, dport, inet->inet_sport);
1027 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1028 rt = ip_route_output_flow(net, fl4, sk);
1032 if (err == -ENETUNREACH)
1033 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1038 if ((rt->rt_flags & RTCF_BROADCAST) &&
1039 !sock_flag(sk, SOCK_BROADCAST))
1042 sk_dst_set(sk, dst_clone(&rt->dst));
1045 if (msg->msg_flags&MSG_CONFIRM)
1051 daddr = ipc.addr = fl4->daddr;
1053 /* Lockless fast path for the non-corking case. */
1055 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1056 sizeof(struct udphdr), &ipc, &rt,
1059 if (!IS_ERR_OR_NULL(skb))
1060 err = udp_send_skb(skb, fl4);
1065 if (unlikely(up->pending)) {
1066 /* The socket is already corked while preparing it. */
1067 /* ... which is an evident application bug. --ANK */
1070 net_dbg_ratelimited("cork app bug 2\n");
1075 * Now cork the socket to pend data.
1077 fl4 = &inet->cork.fl.u.ip4;
1080 fl4->fl4_dport = dport;
1081 fl4->fl4_sport = inet->inet_sport;
1082 up->pending = AF_INET;
1086 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1087 sizeof(struct udphdr), &ipc, &rt,
1088 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1090 udp_flush_pending_frames(sk);
1092 err = udp_push_pending_frames(sk);
1093 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1105 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1106 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1107 * we don't have a good statistic (IpOutDiscards but it can be too many
1108 * things). We could add another new stat but at least for now that
1109 * seems like overkill.
1111 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1112 UDP_INC_STATS(sock_net(sk),
1113 UDP_MIB_SNDBUFERRORS, is_udplite);
1118 dst_confirm(&rt->dst);
1119 if (!(msg->msg_flags&MSG_PROBE) || len)
1120 goto back_from_confirm;
1124 EXPORT_SYMBOL(udp_sendmsg);
1126 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1127 size_t size, int flags)
1129 struct inet_sock *inet = inet_sk(sk);
1130 struct udp_sock *up = udp_sk(sk);
1133 if (flags & MSG_SENDPAGE_NOTLAST)
1137 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1139 /* Call udp_sendmsg to specify destination address which
1140 * sendpage interface can't pass.
1141 * This will succeed only when the socket is connected.
1143 ret = udp_sendmsg(sk, &msg, 0);
1150 if (unlikely(!up->pending)) {
1153 net_dbg_ratelimited("udp cork app bug 3\n");
1157 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1158 page, offset, size, flags);
1159 if (ret == -EOPNOTSUPP) {
1161 return sock_no_sendpage(sk->sk_socket, page, offset,
1165 udp_flush_pending_frames(sk);
1170 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1171 ret = udp_push_pending_frames(sk);
1180 * first_packet_length - return length of first packet in receive queue
1183 * Drops all bad checksum frames, until a valid one is found.
1184 * Returns the length of found skb, or -1 if none is found.
1186 static int first_packet_length(struct sock *sk)
1188 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1189 struct sk_buff *skb;
1192 __skb_queue_head_init(&list_kill);
1194 spin_lock_bh(&rcvq->lock);
1195 while ((skb = skb_peek(rcvq)) != NULL &&
1196 udp_lib_checksum_complete(skb)) {
1197 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1199 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1201 atomic_inc(&sk->sk_drops);
1202 __skb_unlink(skb, rcvq);
1203 __skb_queue_tail(&list_kill, skb);
1205 res = skb ? skb->len : -1;
1206 spin_unlock_bh(&rcvq->lock);
1208 if (!skb_queue_empty(&list_kill)) {
1209 bool slow = lock_sock_fast(sk);
1211 __skb_queue_purge(&list_kill);
1212 sk_mem_reclaim_partial(sk);
1213 unlock_sock_fast(sk, slow);
1219 * IOCTL requests applicable to the UDP protocol
1222 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1227 int amount = sk_wmem_alloc_get(sk);
1229 return put_user(amount, (int __user *)arg);
1234 int amount = max_t(int, 0, first_packet_length(sk));
1236 return put_user(amount, (int __user *)arg);
1240 return -ENOIOCTLCMD;
1245 EXPORT_SYMBOL(udp_ioctl);
1248 * This should be easy, if there is something there we
1249 * return it, otherwise we block.
1252 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1253 int flags, int *addr_len)
1255 struct inet_sock *inet = inet_sk(sk);
1256 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1257 struct sk_buff *skb;
1258 unsigned int ulen, copied;
1259 int peeked, peeking, off;
1261 int is_udplite = IS_UDPLITE(sk);
1262 bool checksum_valid = false;
1265 if (flags & MSG_ERRQUEUE)
1266 return ip_recv_error(sk, msg, len, addr_len);
1269 peeking = off = sk_peek_offset(sk, flags);
1270 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1271 &peeked, &off, &err);
1277 if (copied > ulen - off)
1278 copied = ulen - off;
1279 else if (copied < ulen)
1280 msg->msg_flags |= MSG_TRUNC;
1283 * If checksum is needed at all, try to do it while copying the
1284 * data. If the data is truncated, or if we only want a partial
1285 * coverage checksum (UDP-Lite), do it before the copy.
1288 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov || peeking) {
1289 checksum_valid = !udp_lib_checksum_complete(skb);
1290 if (!checksum_valid)
1294 if (checksum_valid || skb_csum_unnecessary(skb))
1295 err = skb_copy_datagram_msg(skb, off, msg, copied);
1297 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1303 if (unlikely(err)) {
1304 trace_kfree_skb(skb, udp_recvmsg);
1306 atomic_inc(&sk->sk_drops);
1307 UDP_INC_STATS(sock_net(sk),
1308 UDP_MIB_INERRORS, is_udplite);
1310 skb_free_datagram_locked(sk, skb);
1315 UDP_INC_STATS(sock_net(sk),
1316 UDP_MIB_INDATAGRAMS, is_udplite);
1318 sock_recv_ts_and_drops(msg, sk, skb);
1320 /* Copy the address. */
1322 sin->sin_family = AF_INET;
1323 sin->sin_port = udp_hdr(skb)->source;
1324 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1325 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1326 *addr_len = sizeof(*sin);
1328 if (inet->cmsg_flags)
1329 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr), off);
1332 if (flags & MSG_TRUNC)
1335 __skb_free_datagram_locked(sk, skb, peeking ? -err : err);
1339 slow = lock_sock_fast(sk);
1340 if (!skb_kill_datagram(sk, skb, flags)) {
1341 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1342 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1344 unlock_sock_fast(sk, slow);
1346 /* starting over for a new packet, but check if we need to yield */
1348 msg->msg_flags &= ~MSG_TRUNC;
1352 int __udp_disconnect(struct sock *sk, int flags)
1354 struct inet_sock *inet = inet_sk(sk);
1356 * 1003.1g - break association.
1359 sk->sk_state = TCP_CLOSE;
1360 inet->inet_daddr = 0;
1361 inet->inet_dport = 0;
1362 sock_rps_reset_rxhash(sk);
1363 sk->sk_bound_dev_if = 0;
1364 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1365 inet_reset_saddr(sk);
1367 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1368 sk->sk_prot->unhash(sk);
1369 inet->inet_sport = 0;
1374 EXPORT_SYMBOL(__udp_disconnect);
1376 int udp_disconnect(struct sock *sk, int flags)
1379 __udp_disconnect(sk, flags);
1383 EXPORT_SYMBOL(udp_disconnect);
1385 void udp_lib_unhash(struct sock *sk)
1387 if (sk_hashed(sk)) {
1388 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1389 struct udp_hslot *hslot, *hslot2;
1391 hslot = udp_hashslot(udptable, sock_net(sk),
1392 udp_sk(sk)->udp_port_hash);
1393 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1395 spin_lock_bh(&hslot->lock);
1396 if (rcu_access_pointer(sk->sk_reuseport_cb))
1397 reuseport_detach_sock(sk);
1398 if (sk_del_node_init_rcu(sk)) {
1400 inet_sk(sk)->inet_num = 0;
1401 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1403 spin_lock(&hslot2->lock);
1404 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1406 spin_unlock(&hslot2->lock);
1408 spin_unlock_bh(&hslot->lock);
1411 EXPORT_SYMBOL(udp_lib_unhash);
1414 * inet_rcv_saddr was changed, we must rehash secondary hash
1416 void udp_lib_rehash(struct sock *sk, u16 newhash)
1418 if (sk_hashed(sk)) {
1419 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1420 struct udp_hslot *hslot, *hslot2, *nhslot2;
1422 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1423 nhslot2 = udp_hashslot2(udptable, newhash);
1424 udp_sk(sk)->udp_portaddr_hash = newhash;
1426 if (hslot2 != nhslot2 ||
1427 rcu_access_pointer(sk->sk_reuseport_cb)) {
1428 hslot = udp_hashslot(udptable, sock_net(sk),
1429 udp_sk(sk)->udp_port_hash);
1430 /* we must lock primary chain too */
1431 spin_lock_bh(&hslot->lock);
1432 if (rcu_access_pointer(sk->sk_reuseport_cb))
1433 reuseport_detach_sock(sk);
1435 if (hslot2 != nhslot2) {
1436 spin_lock(&hslot2->lock);
1437 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1439 spin_unlock(&hslot2->lock);
1441 spin_lock(&nhslot2->lock);
1442 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1445 spin_unlock(&nhslot2->lock);
1448 spin_unlock_bh(&hslot->lock);
1452 EXPORT_SYMBOL(udp_lib_rehash);
1454 static void udp_v4_rehash(struct sock *sk)
1456 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1457 inet_sk(sk)->inet_rcv_saddr,
1458 inet_sk(sk)->inet_num);
1459 udp_lib_rehash(sk, new_hash);
1462 int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1466 if (inet_sk(sk)->inet_daddr) {
1467 sock_rps_save_rxhash(sk, skb);
1468 sk_mark_napi_id(sk, skb);
1469 sk_incoming_cpu_update(sk);
1472 rc = __sock_queue_rcv_skb(sk, skb);
1474 int is_udplite = IS_UDPLITE(sk);
1476 /* Note that an ENOMEM error is charged twice */
1478 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1480 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1482 trace_udp_fail_queue_rcv_skb(rc, sk);
1490 static struct static_key udp_encap_needed __read_mostly;
1491 void udp_encap_enable(void)
1493 if (!static_key_enabled(&udp_encap_needed))
1494 static_key_slow_inc(&udp_encap_needed);
1496 EXPORT_SYMBOL(udp_encap_enable);
1501 * >0: "udp encap" protocol resubmission
1503 * Note that in the success and error cases, the skb is assumed to
1504 * have either been requeued or freed.
1506 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1508 struct udp_sock *up = udp_sk(sk);
1510 int is_udplite = IS_UDPLITE(sk);
1513 * Charge it to the socket, dropping if the queue is full.
1515 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1519 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1520 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1523 * This is an encapsulation socket so pass the skb to
1524 * the socket's udp_encap_rcv() hook. Otherwise, just
1525 * fall through and pass this up the UDP socket.
1526 * up->encap_rcv() returns the following value:
1527 * =0 if skb was successfully passed to the encap
1528 * handler or was discarded by it.
1529 * >0 if skb should be passed on to UDP.
1530 * <0 if skb should be resubmitted as proto -N
1533 /* if we're overly short, let UDP handle it */
1534 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1538 /* Verify checksum before giving to encap */
1539 if (udp_lib_checksum_complete(skb))
1542 ret = encap_rcv(sk, skb);
1544 __UDP_INC_STATS(sock_net(sk),
1545 UDP_MIB_INDATAGRAMS,
1551 /* FALLTHROUGH -- it's a UDP Packet */
1555 * UDP-Lite specific tests, ignored on UDP sockets
1557 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1560 * MIB statistics other than incrementing the error count are
1561 * disabled for the following two types of errors: these depend
1562 * on the application settings, not on the functioning of the
1563 * protocol stack as such.
1565 * RFC 3828 here recommends (sec 3.3): "There should also be a
1566 * way ... to ... at least let the receiving application block
1567 * delivery of packets with coverage values less than a value
1568 * provided by the application."
1570 if (up->pcrlen == 0) { /* full coverage was set */
1571 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1572 UDP_SKB_CB(skb)->cscov, skb->len);
1575 /* The next case involves violating the min. coverage requested
1576 * by the receiver. This is subtle: if receiver wants x and x is
1577 * greater than the buffersize/MTU then receiver will complain
1578 * that it wants x while sender emits packets of smaller size y.
1579 * Therefore the above ...()->partial_cov statement is essential.
1581 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1582 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1583 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1588 if (rcu_access_pointer(sk->sk_filter) &&
1589 udp_lib_checksum_complete(skb))
1592 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1595 udp_csum_pull_header(skb);
1596 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1597 __UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1604 ipv4_pktinfo_prepare(sk, skb);
1606 if (!sock_owned_by_user(sk))
1607 rc = __udp_queue_rcv_skb(sk, skb);
1608 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1617 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1619 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1620 atomic_inc(&sk->sk_drops);
1625 /* For TCP sockets, sk_rx_dst is protected by socket lock
1626 * For UDP, we use xchg() to guard against concurrent changes.
1628 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1630 struct dst_entry *old;
1633 old = xchg(&sk->sk_rx_dst, dst);
1638 * Multicasts and broadcasts go to each listener.
1640 * Note: called only from the BH handler context.
1642 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1644 __be32 saddr, __be32 daddr,
1645 struct udp_table *udptable,
1648 struct sock *sk, *first = NULL;
1649 unsigned short hnum = ntohs(uh->dest);
1650 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1651 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1652 unsigned int offset = offsetof(typeof(*sk), sk_node);
1653 int dif = skb->dev->ifindex;
1654 struct hlist_node *node;
1655 struct sk_buff *nskb;
1658 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1660 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1662 hslot = &udptable->hash2[hash2];
1663 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1666 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1667 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1668 uh->source, saddr, dif, hnum))
1675 nskb = skb_clone(skb, GFP_ATOMIC);
1677 if (unlikely(!nskb)) {
1678 atomic_inc(&sk->sk_drops);
1679 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1681 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1685 if (udp_queue_rcv_skb(sk, nskb) > 0)
1689 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1690 if (use_hash2 && hash2 != hash2_any) {
1696 if (udp_queue_rcv_skb(first, skb) > 0)
1700 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
1701 proto == IPPROTO_UDPLITE);
1706 /* Initialize UDP checksum. If exited with zero value (success),
1707 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1708 * Otherwise, csum completion requires chacksumming packet body,
1709 * including udp header and folding it to skb->csum.
1711 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1716 UDP_SKB_CB(skb)->partial_cov = 0;
1717 UDP_SKB_CB(skb)->cscov = skb->len;
1719 if (proto == IPPROTO_UDPLITE) {
1720 err = udplite_checksum_init(skb, uh);
1724 if (UDP_SKB_CB(skb)->partial_cov) {
1725 skb->csum = inet_compute_pseudo(skb, proto);
1730 /* Note, we are only interested in != 0 or == 0, thus the
1733 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
1734 inet_compute_pseudo);
1738 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
1739 /* If SW calculated the value, we know it's bad */
1740 if (skb->csum_complete_sw)
1743 /* HW says the value is bad. Let's validate that.
1744 * skb->csum is no longer the full packet checksum,
1745 * so don't treat it as such.
1747 skb_checksum_complete_unset(skb);
1753 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
1754 * return code conversion for ip layer consumption
1756 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
1761 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1762 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1763 inet_compute_pseudo);
1765 ret = udp_queue_rcv_skb(sk, skb);
1767 /* a return value > 0 means to resubmit the input, but
1768 * it wants the return to be -protocol, or 0
1776 * All we need to do is get the socket, and then do a checksum.
1779 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1784 unsigned short ulen;
1785 struct rtable *rt = skb_rtable(skb);
1786 __be32 saddr, daddr;
1787 struct net *net = dev_net(skb->dev);
1790 * Validate the packet.
1792 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1793 goto drop; /* No space for header. */
1796 ulen = ntohs(uh->len);
1797 saddr = ip_hdr(skb)->saddr;
1798 daddr = ip_hdr(skb)->daddr;
1800 if (ulen > skb->len)
1803 if (proto == IPPROTO_UDP) {
1804 /* UDP validates ulen. */
1805 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1810 if (udp4_csum_init(skb, uh, proto))
1813 sk = skb_steal_sock(skb);
1815 struct dst_entry *dst = skb_dst(skb);
1818 if (unlikely(sk->sk_rx_dst != dst))
1819 udp_sk_rx_dst_set(sk, dst);
1821 ret = udp_unicast_rcv_skb(sk, skb, uh);
1826 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1827 return __udp4_lib_mcast_deliver(net, skb, uh,
1828 saddr, daddr, udptable, proto);
1830 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1832 return udp_unicast_rcv_skb(sk, skb, uh);
1834 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1838 /* No socket. Drop packet silently, if checksum is wrong */
1839 if (udp_lib_checksum_complete(skb))
1842 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1843 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1846 * Hmm. We got an UDP packet to a port to which we
1847 * don't wanna listen. Ignore it.
1853 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1854 proto == IPPROTO_UDPLITE ? "Lite" : "",
1855 &saddr, ntohs(uh->source),
1857 &daddr, ntohs(uh->dest));
1862 * RFC1122: OK. Discards the bad packet silently (as far as
1863 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1865 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1866 proto == IPPROTO_UDPLITE ? "Lite" : "",
1867 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1869 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1871 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1876 /* We can only early demux multicast if there is a single matching socket.
1877 * If more than one socket found returns NULL
1879 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1880 __be16 loc_port, __be32 loc_addr,
1881 __be16 rmt_port, __be32 rmt_addr,
1884 struct sock *sk, *result;
1885 unsigned short hnum = ntohs(loc_port);
1886 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
1887 struct udp_hslot *hslot = &udp_table.hash[slot];
1889 /* Do not bother scanning a too big list */
1890 if (hslot->count > 10)
1894 sk_for_each_rcu(sk, &hslot->head) {
1895 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
1896 rmt_port, rmt_addr, dif, hnum)) {
1906 /* For unicast we should only early demux connected sockets or we can
1907 * break forwarding setups. The chains here can be long so only check
1908 * if the first socket is an exact match and if not move on.
1910 static struct sock *__udp4_lib_demux_lookup(struct net *net,
1911 __be16 loc_port, __be32 loc_addr,
1912 __be16 rmt_port, __be32 rmt_addr,
1915 unsigned short hnum = ntohs(loc_port);
1916 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1917 unsigned int slot2 = hash2 & udp_table.mask;
1918 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1919 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
1920 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1923 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1924 if (INET_MATCH(sk, net, acookie, rmt_addr,
1925 loc_addr, ports, dif))
1927 /* Only check first socket in chain */
1933 void udp_v4_early_demux(struct sk_buff *skb)
1935 struct net *net = dev_net(skb->dev);
1936 const struct iphdr *iph;
1937 const struct udphdr *uh;
1938 struct sock *sk = NULL;
1939 struct dst_entry *dst;
1940 int dif = skb->dev->ifindex;
1943 /* validate the packet */
1944 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1950 if (skb->pkt_type == PACKET_BROADCAST ||
1951 skb->pkt_type == PACKET_MULTICAST) {
1952 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1957 /* we are supposed to accept bcast packets */
1958 if (skb->pkt_type == PACKET_MULTICAST) {
1959 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
1965 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1966 uh->source, iph->saddr, dif);
1967 } else if (skb->pkt_type == PACKET_HOST) {
1968 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1969 uh->source, iph->saddr, dif);
1972 if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2))
1976 skb->destructor = sock_efree;
1977 dst = READ_ONCE(sk->sk_rx_dst);
1980 dst = dst_check(dst, 0);
1982 /* DST_NOCACHE can not be used without taking a reference */
1983 if (dst->flags & DST_NOCACHE) {
1984 if (likely(atomic_inc_not_zero(&dst->__refcnt)))
1985 skb_dst_set(skb, dst);
1987 skb_dst_set_noref(skb, dst);
1992 int udp_rcv(struct sk_buff *skb)
1994 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1997 void udp_destroy_sock(struct sock *sk)
1999 struct udp_sock *up = udp_sk(sk);
2000 bool slow = lock_sock_fast(sk);
2002 /* protects from races with udp_abort() */
2003 sock_set_flag(sk, SOCK_DEAD);
2004 udp_flush_pending_frames(sk);
2005 unlock_sock_fast(sk, slow);
2006 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2007 void (*encap_destroy)(struct sock *sk);
2008 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2015 * Socket option code for UDP
2017 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2018 char __user *optval, unsigned int optlen,
2019 int (*push_pending_frames)(struct sock *))
2021 struct udp_sock *up = udp_sk(sk);
2024 int is_udplite = IS_UDPLITE(sk);
2026 if (optlen < sizeof(int))
2029 if (get_user(val, (int __user *)optval))
2032 valbool = val ? 1 : 0;
2037 WRITE_ONCE(up->corkflag, 1);
2039 WRITE_ONCE(up->corkflag, 0);
2041 push_pending_frames(sk);
2049 case UDP_ENCAP_ESPINUDP:
2050 case UDP_ENCAP_ESPINUDP_NON_IKE:
2051 up->encap_rcv = xfrm4_udp_encap_rcv;
2053 case UDP_ENCAP_L2TPINUDP:
2054 up->encap_type = val;
2063 case UDP_NO_CHECK6_TX:
2064 up->no_check6_tx = valbool;
2067 case UDP_NO_CHECK6_RX:
2068 up->no_check6_rx = valbool;
2072 * UDP-Lite's partial checksum coverage (RFC 3828).
2074 /* The sender sets actual checksum coverage length via this option.
2075 * The case coverage > packet length is handled by send module. */
2076 case UDPLITE_SEND_CSCOV:
2077 if (!is_udplite) /* Disable the option on UDP sockets */
2078 return -ENOPROTOOPT;
2079 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2081 else if (val > USHRT_MAX)
2084 up->pcflag |= UDPLITE_SEND_CC;
2087 /* The receiver specifies a minimum checksum coverage value. To make
2088 * sense, this should be set to at least 8 (as done below). If zero is
2089 * used, this again means full checksum coverage. */
2090 case UDPLITE_RECV_CSCOV:
2091 if (!is_udplite) /* Disable the option on UDP sockets */
2092 return -ENOPROTOOPT;
2093 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2095 else if (val > USHRT_MAX)
2098 up->pcflag |= UDPLITE_RECV_CC;
2108 EXPORT_SYMBOL(udp_lib_setsockopt);
2110 int udp_setsockopt(struct sock *sk, int level, int optname,
2111 char __user *optval, unsigned int optlen)
2113 if (level == SOL_UDP || level == SOL_UDPLITE)
2114 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2115 udp_push_pending_frames);
2116 return ip_setsockopt(sk, level, optname, optval, optlen);
2119 #ifdef CONFIG_COMPAT
2120 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2121 char __user *optval, unsigned int optlen)
2123 if (level == SOL_UDP || level == SOL_UDPLITE)
2124 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2125 udp_push_pending_frames);
2126 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2130 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2131 char __user *optval, int __user *optlen)
2133 struct udp_sock *up = udp_sk(sk);
2136 if (get_user(len, optlen))
2139 len = min_t(unsigned int, len, sizeof(int));
2146 val = READ_ONCE(up->corkflag);
2150 val = up->encap_type;
2153 case UDP_NO_CHECK6_TX:
2154 val = up->no_check6_tx;
2157 case UDP_NO_CHECK6_RX:
2158 val = up->no_check6_rx;
2161 /* The following two cannot be changed on UDP sockets, the return is
2162 * always 0 (which corresponds to the full checksum coverage of UDP). */
2163 case UDPLITE_SEND_CSCOV:
2167 case UDPLITE_RECV_CSCOV:
2172 return -ENOPROTOOPT;
2175 if (put_user(len, optlen))
2177 if (copy_to_user(optval, &val, len))
2181 EXPORT_SYMBOL(udp_lib_getsockopt);
2183 int udp_getsockopt(struct sock *sk, int level, int optname,
2184 char __user *optval, int __user *optlen)
2186 if (level == SOL_UDP || level == SOL_UDPLITE)
2187 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2188 return ip_getsockopt(sk, level, optname, optval, optlen);
2191 #ifdef CONFIG_COMPAT
2192 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2193 char __user *optval, int __user *optlen)
2195 if (level == SOL_UDP || level == SOL_UDPLITE)
2196 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2197 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2201 * udp_poll - wait for a UDP event.
2202 * @file - file struct
2204 * @wait - poll table
2206 * This is same as datagram poll, except for the special case of
2207 * blocking sockets. If application is using a blocking fd
2208 * and a packet with checksum error is in the queue;
2209 * then it could get return from select indicating data available
2210 * but then block when reading it. Add special case code
2211 * to work around these arguably broken applications.
2213 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2215 unsigned int mask = datagram_poll(file, sock, wait);
2216 struct sock *sk = sock->sk;
2218 sock_rps_record_flow(sk);
2220 /* Check for false positives due to checksum errors */
2221 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2222 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2223 mask &= ~(POLLIN | POLLRDNORM);
2228 EXPORT_SYMBOL(udp_poll);
2230 int udp_abort(struct sock *sk, int err)
2234 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2237 if (sock_flag(sk, SOCK_DEAD))
2241 sk->sk_error_report(sk);
2242 __udp_disconnect(sk, 0);
2249 EXPORT_SYMBOL_GPL(udp_abort);
2251 struct proto udp_prot = {
2253 .owner = THIS_MODULE,
2254 .close = udp_lib_close,
2255 .connect = ip4_datagram_connect,
2256 .disconnect = udp_disconnect,
2258 .destroy = udp_destroy_sock,
2259 .setsockopt = udp_setsockopt,
2260 .getsockopt = udp_getsockopt,
2261 .sendmsg = udp_sendmsg,
2262 .recvmsg = udp_recvmsg,
2263 .sendpage = udp_sendpage,
2264 .backlog_rcv = __udp_queue_rcv_skb,
2265 .release_cb = ip4_datagram_release_cb,
2266 .hash = udp_lib_hash,
2267 .unhash = udp_lib_unhash,
2268 .rehash = udp_v4_rehash,
2269 .get_port = udp_v4_get_port,
2270 .memory_allocated = &udp_memory_allocated,
2271 .sysctl_mem = sysctl_udp_mem,
2272 .sysctl_wmem = &sysctl_udp_wmem_min,
2273 .sysctl_rmem = &sysctl_udp_rmem_min,
2274 .obj_size = sizeof(struct udp_sock),
2275 .h.udp_table = &udp_table,
2276 #ifdef CONFIG_COMPAT
2277 .compat_setsockopt = compat_udp_setsockopt,
2278 .compat_getsockopt = compat_udp_getsockopt,
2280 .diag_destroy = udp_abort,
2282 EXPORT_SYMBOL(udp_prot);
2284 /* ------------------------------------------------------------------------ */
2285 #ifdef CONFIG_PROC_FS
2287 static struct sock *udp_get_first(struct seq_file *seq, int start)
2290 struct udp_iter_state *state = seq->private;
2291 struct net *net = seq_file_net(seq);
2293 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2295 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2297 if (hlist_empty(&hslot->head))
2300 spin_lock_bh(&hslot->lock);
2301 sk_for_each(sk, &hslot->head) {
2302 if (!net_eq(sock_net(sk), net))
2304 if (sk->sk_family == state->family)
2307 spin_unlock_bh(&hslot->lock);
2314 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2316 struct udp_iter_state *state = seq->private;
2317 struct net *net = seq_file_net(seq);
2321 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2324 if (state->bucket <= state->udp_table->mask)
2325 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2326 return udp_get_first(seq, state->bucket + 1);
2331 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2333 struct sock *sk = udp_get_first(seq, 0);
2336 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2338 return pos ? NULL : sk;
2341 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2343 struct udp_iter_state *state = seq->private;
2344 state->bucket = MAX_UDP_PORTS;
2346 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2349 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2353 if (v == SEQ_START_TOKEN)
2354 sk = udp_get_idx(seq, 0);
2356 sk = udp_get_next(seq, v);
2362 static void udp_seq_stop(struct seq_file *seq, void *v)
2364 struct udp_iter_state *state = seq->private;
2366 if (state->bucket <= state->udp_table->mask)
2367 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2370 int udp_seq_open(struct inode *inode, struct file *file)
2372 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2373 struct udp_iter_state *s;
2376 err = seq_open_net(inode, file, &afinfo->seq_ops,
2377 sizeof(struct udp_iter_state));
2381 s = ((struct seq_file *)file->private_data)->private;
2382 s->family = afinfo->family;
2383 s->udp_table = afinfo->udp_table;
2386 EXPORT_SYMBOL(udp_seq_open);
2388 /* ------------------------------------------------------------------------ */
2389 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2391 struct proc_dir_entry *p;
2394 afinfo->seq_ops.start = udp_seq_start;
2395 afinfo->seq_ops.next = udp_seq_next;
2396 afinfo->seq_ops.stop = udp_seq_stop;
2398 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2399 afinfo->seq_fops, afinfo);
2404 EXPORT_SYMBOL(udp_proc_register);
2406 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2408 remove_proc_entry(afinfo->name, net->proc_net);
2410 EXPORT_SYMBOL(udp_proc_unregister);
2412 /* ------------------------------------------------------------------------ */
2413 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2416 struct inet_sock *inet = inet_sk(sp);
2417 __be32 dest = inet->inet_daddr;
2418 __be32 src = inet->inet_rcv_saddr;
2419 __u16 destp = ntohs(inet->inet_dport);
2420 __u16 srcp = ntohs(inet->inet_sport);
2422 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2423 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2424 bucket, src, srcp, dest, destp, sp->sk_state,
2425 sk_wmem_alloc_get(sp),
2426 sk_rmem_alloc_get(sp),
2428 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2430 atomic_read(&sp->sk_refcnt), sp,
2431 atomic_read(&sp->sk_drops));
2434 int udp4_seq_show(struct seq_file *seq, void *v)
2436 seq_setwidth(seq, 127);
2437 if (v == SEQ_START_TOKEN)
2438 seq_puts(seq, " sl local_address rem_address st tx_queue "
2439 "rx_queue tr tm->when retrnsmt uid timeout "
2440 "inode ref pointer drops");
2442 struct udp_iter_state *state = seq->private;
2444 udp4_format_sock(v, seq, state->bucket);
2450 static const struct file_operations udp_afinfo_seq_fops = {
2451 .owner = THIS_MODULE,
2452 .open = udp_seq_open,
2454 .llseek = seq_lseek,
2455 .release = seq_release_net
2458 /* ------------------------------------------------------------------------ */
2459 static struct udp_seq_afinfo udp4_seq_afinfo = {
2462 .udp_table = &udp_table,
2463 .seq_fops = &udp_afinfo_seq_fops,
2465 .show = udp4_seq_show,
2469 static int __net_init udp4_proc_init_net(struct net *net)
2471 return udp_proc_register(net, &udp4_seq_afinfo);
2474 static void __net_exit udp4_proc_exit_net(struct net *net)
2476 udp_proc_unregister(net, &udp4_seq_afinfo);
2479 static struct pernet_operations udp4_net_ops = {
2480 .init = udp4_proc_init_net,
2481 .exit = udp4_proc_exit_net,
2484 int __init udp4_proc_init(void)
2486 return register_pernet_subsys(&udp4_net_ops);
2489 void udp4_proc_exit(void)
2491 unregister_pernet_subsys(&udp4_net_ops);
2493 #endif /* CONFIG_PROC_FS */
2495 static __initdata unsigned long uhash_entries;
2496 static int __init set_uhash_entries(char *str)
2503 ret = kstrtoul(str, 0, &uhash_entries);
2507 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2508 uhash_entries = UDP_HTABLE_SIZE_MIN;
2511 __setup("uhash_entries=", set_uhash_entries);
2513 void __init udp_table_init(struct udp_table *table, const char *name)
2517 table->hash = alloc_large_system_hash(name,
2518 2 * sizeof(struct udp_hslot),
2520 21, /* one slot per 2 MB */
2524 UDP_HTABLE_SIZE_MIN,
2527 table->hash2 = table->hash + (table->mask + 1);
2528 for (i = 0; i <= table->mask; i++) {
2529 INIT_HLIST_HEAD(&table->hash[i].head);
2530 table->hash[i].count = 0;
2531 spin_lock_init(&table->hash[i].lock);
2533 for (i = 0; i <= table->mask; i++) {
2534 INIT_HLIST_HEAD(&table->hash2[i].head);
2535 table->hash2[i].count = 0;
2536 spin_lock_init(&table->hash2[i].lock);
2540 u32 udp_flow_hashrnd(void)
2542 static u32 hashrnd __read_mostly;
2544 net_get_random_once(&hashrnd, sizeof(hashrnd));
2548 EXPORT_SYMBOL(udp_flow_hashrnd);
2550 void __init udp_init(void)
2552 unsigned long limit;
2554 udp_table_init(&udp_table, "UDP");
2555 limit = nr_free_buffer_pages() / 8;
2556 limit = max(limit, 128UL);
2557 sysctl_udp_mem[0] = limit / 4 * 3;
2558 sysctl_udp_mem[1] = limit;
2559 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2561 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2562 sysctl_udp_wmem_min = SK_MEM_QUANTUM;