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>
118 struct udp_table udp_table __read_mostly;
119 EXPORT_SYMBOL(udp_table);
121 long sysctl_udp_mem[3] __read_mostly;
122 EXPORT_SYMBOL(sysctl_udp_mem);
124 int sysctl_udp_rmem_min __read_mostly;
125 EXPORT_SYMBOL(sysctl_udp_rmem_min);
127 int sysctl_udp_wmem_min __read_mostly;
128 EXPORT_SYMBOL(sysctl_udp_wmem_min);
130 atomic_long_t udp_memory_allocated;
131 EXPORT_SYMBOL(udp_memory_allocated);
133 #define MAX_UDP_PORTS 65536
134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
136 static int udp_lib_lport_inuse(struct net *net, __u16 num,
137 const struct udp_hslot *hslot,
138 unsigned long *bitmap,
140 int (*saddr_comp)(const struct sock *sk1,
141 const struct sock *sk2,
142 bool match_wildcard),
146 kuid_t uid = sock_i_uid(sk);
148 sk_for_each(sk2, &hslot->head) {
149 if (net_eq(sock_net(sk2), net) &&
151 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
152 (!sk2->sk_reuse || !sk->sk_reuse) &&
153 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
154 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
155 (!sk2->sk_reuseport || !sk->sk_reuseport ||
156 rcu_access_pointer(sk->sk_reuseport_cb) ||
157 !uid_eq(uid, sock_i_uid(sk2))) &&
158 saddr_comp(sk, sk2, true)) {
161 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap);
168 * Note: we still hold spinlock of primary hash chain, so no other writer
169 * can insert/delete a socket with local_port == num
171 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
172 struct udp_hslot *hslot2,
174 int (*saddr_comp)(const struct sock *sk1,
175 const struct sock *sk2,
176 bool match_wildcard))
179 kuid_t uid = sock_i_uid(sk);
182 spin_lock(&hslot2->lock);
183 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
184 if (net_eq(sock_net(sk2), net) &&
186 (udp_sk(sk2)->udp_port_hash == num) &&
187 (!sk2->sk_reuse || !sk->sk_reuse) &&
188 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
189 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
190 (!sk2->sk_reuseport || !sk->sk_reuseport ||
191 rcu_access_pointer(sk->sk_reuseport_cb) ||
192 !uid_eq(uid, sock_i_uid(sk2))) &&
193 saddr_comp(sk, sk2, true)) {
198 spin_unlock(&hslot2->lock);
202 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot,
203 int (*saddr_same)(const struct sock *sk1,
204 const struct sock *sk2,
205 bool match_wildcard))
207 struct net *net = sock_net(sk);
208 kuid_t uid = sock_i_uid(sk);
211 sk_for_each(sk2, &hslot->head) {
212 if (net_eq(sock_net(sk2), net) &&
214 sk2->sk_family == sk->sk_family &&
215 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
216 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
217 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
218 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
219 (*saddr_same)(sk, sk2, false)) {
220 return reuseport_add_sock(sk, sk2);
224 /* Initial allocation may have already happened via setsockopt */
225 if (!rcu_access_pointer(sk->sk_reuseport_cb))
226 return reuseport_alloc(sk);
231 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
233 * @sk: socket struct in question
234 * @snum: port number to look up
235 * @saddr_comp: AF-dependent comparison of bound local IP addresses
236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
239 int udp_lib_get_port(struct sock *sk, unsigned short snum,
240 int (*saddr_comp)(const struct sock *sk1,
241 const struct sock *sk2,
242 bool match_wildcard),
243 unsigned int hash2_nulladdr)
245 struct udp_hslot *hslot, *hslot2;
246 struct udp_table *udptable = sk->sk_prot->h.udp_table;
248 struct net *net = sock_net(sk);
251 int low, high, remaining;
253 unsigned short first, last;
254 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
256 inet_get_local_port_range(net, &low, &high);
257 remaining = (high - low) + 1;
259 rand = prandom_u32();
260 first = reciprocal_scale(rand, remaining) + low;
262 * force rand to be an odd multiple of UDP_HTABLE_SIZE
264 rand = (rand | 1) * (udptable->mask + 1);
265 last = first + udptable->mask + 1;
267 hslot = udp_hashslot(udptable, net, first);
268 bitmap_zero(bitmap, PORTS_PER_CHAIN);
269 spin_lock_bh(&hslot->lock);
270 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
271 saddr_comp, udptable->log);
275 * Iterate on all possible values of snum for this hash.
276 * Using steps of an odd multiple of UDP_HTABLE_SIZE
277 * give us randomization and full range coverage.
280 if (low <= snum && snum <= high &&
281 !test_bit(snum >> udptable->log, bitmap) &&
282 !inet_is_local_reserved_port(net, snum))
285 } while (snum != first);
286 spin_unlock_bh(&hslot->lock);
287 } while (++first != last);
290 hslot = udp_hashslot(udptable, net, snum);
291 spin_lock_bh(&hslot->lock);
292 if (hslot->count > 10) {
294 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
296 slot2 &= udptable->mask;
297 hash2_nulladdr &= udptable->mask;
299 hslot2 = udp_hashslot2(udptable, slot2);
300 if (hslot->count < hslot2->count)
301 goto scan_primary_hash;
303 exist = udp_lib_lport_inuse2(net, snum, hslot2,
305 if (!exist && (hash2_nulladdr != slot2)) {
306 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
307 exist = udp_lib_lport_inuse2(net, snum, hslot2,
316 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
321 inet_sk(sk)->inet_num = snum;
322 udp_sk(sk)->udp_port_hash = snum;
323 udp_sk(sk)->udp_portaddr_hash ^= snum;
324 if (sk_unhashed(sk)) {
325 if (sk->sk_reuseport &&
326 udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
327 inet_sk(sk)->inet_num = 0;
328 udp_sk(sk)->udp_port_hash = 0;
329 udp_sk(sk)->udp_portaddr_hash ^= snum;
333 sk_add_node_rcu(sk, &hslot->head);
335 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
337 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
338 spin_lock(&hslot2->lock);
339 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
342 spin_unlock(&hslot2->lock);
344 sock_set_flag(sk, SOCK_RCU_FREE);
347 spin_unlock_bh(&hslot->lock);
351 EXPORT_SYMBOL(udp_lib_get_port);
353 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
354 * match_wildcard == false: addresses must be exactly the same, i.e.
355 * 0.0.0.0 only equals to 0.0.0.0
357 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
360 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
362 if (!ipv6_only_sock(sk2)) {
363 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
365 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
366 return match_wildcard;
371 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
374 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
377 int udp_v4_get_port(struct sock *sk, unsigned short snum)
379 unsigned int hash2_nulladdr =
380 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
381 unsigned int hash2_partial =
382 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
384 /* precompute partial secondary hash */
385 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
386 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
389 static inline int compute_score(struct sock *sk, struct net *net,
390 __be32 saddr, unsigned short hnum, __be16 sport,
391 __be32 daddr, __be16 dport, int dif)
394 struct inet_sock *inet;
396 if (!net_eq(sock_net(sk), net) ||
397 udp_sk(sk)->udp_port_hash != hnum ||
401 score = (sk->sk_family == PF_INET) ? 2 : 1;
404 if (inet->inet_rcv_saddr) {
405 if (inet->inet_rcv_saddr != daddr)
410 if (inet->inet_daddr) {
411 if (inet->inet_daddr != saddr)
416 if (inet->inet_dport) {
417 if (inet->inet_dport != sport)
422 if (sk->sk_bound_dev_if) {
423 if (sk->sk_bound_dev_if != dif)
427 if (sk->sk_incoming_cpu == raw_smp_processor_id())
433 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
435 static inline int compute_score2(struct sock *sk, struct net *net,
436 __be32 saddr, __be16 sport,
437 __be32 daddr, unsigned int hnum, int dif)
440 struct inet_sock *inet;
442 if (!net_eq(sock_net(sk), net) ||
448 if (inet->inet_rcv_saddr != daddr ||
449 inet->inet_num != hnum)
452 score = (sk->sk_family == PF_INET) ? 2 : 1;
454 if (inet->inet_daddr) {
455 if (inet->inet_daddr != saddr)
460 if (inet->inet_dport) {
461 if (inet->inet_dport != sport)
466 if (sk->sk_bound_dev_if) {
467 if (sk->sk_bound_dev_if != dif)
472 if (sk->sk_incoming_cpu == raw_smp_processor_id())
478 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
479 const __u16 lport, const __be32 faddr,
482 static u32 udp_ehash_secret __read_mostly;
484 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
486 return __inet_ehashfn(laddr, lport, faddr, fport,
487 udp_ehash_secret + net_hash_mix(net));
490 /* called with read_rcu_lock() */
491 static struct sock *udp4_lib_lookup2(struct net *net,
492 __be32 saddr, __be16 sport,
493 __be32 daddr, unsigned int hnum, int dif,
494 struct udp_hslot *hslot2, unsigned int slot2,
497 struct sock *sk, *result;
498 int score, badness, matches = 0, reuseport = 0;
503 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
504 score = compute_score2(sk, net, saddr, sport,
506 if (score > badness) {
507 reuseport = sk->sk_reuseport;
509 hash = udp_ehashfn(net, daddr, hnum,
511 result = reuseport_select_sock(sk, hash, skb,
512 sizeof(struct udphdr));
519 } else if (score == badness && reuseport) {
521 if (reciprocal_scale(hash, matches) == 0)
523 hash = next_pseudo_random32(hash);
529 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
530 * harder than this. -DaveM
532 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
533 __be16 sport, __be32 daddr, __be16 dport,
534 int dif, struct udp_table *udptable, struct sk_buff *skb)
536 struct sock *sk, *result;
537 unsigned short hnum = ntohs(dport);
538 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
539 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
540 int score, badness, matches = 0, reuseport = 0;
543 if (hslot->count > 10) {
544 hash2 = udp4_portaddr_hash(net, daddr, hnum);
545 slot2 = hash2 & udptable->mask;
546 hslot2 = &udptable->hash2[slot2];
547 if (hslot->count < hslot2->count)
550 result = udp4_lib_lookup2(net, saddr, sport,
554 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
555 slot2 = hash2 & udptable->mask;
556 hslot2 = &udptable->hash2[slot2];
557 if (hslot->count < hslot2->count)
560 result = udp4_lib_lookup2(net, saddr, sport,
561 htonl(INADDR_ANY), hnum, dif,
569 sk_for_each_rcu(sk, &hslot->head) {
570 score = compute_score(sk, net, saddr, hnum, sport,
572 if (score > badness) {
573 reuseport = sk->sk_reuseport;
575 hash = udp_ehashfn(net, daddr, hnum,
577 result = reuseport_select_sock(sk, hash, skb,
578 sizeof(struct udphdr));
585 } else if (score == badness && reuseport) {
587 if (reciprocal_scale(hash, matches) == 0)
589 hash = next_pseudo_random32(hash);
594 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
596 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
597 __be16 sport, __be16 dport,
598 struct udp_table *udptable)
600 const struct iphdr *iph = ip_hdr(skb);
602 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
603 iph->daddr, dport, inet_iif(skb),
607 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
608 __be16 sport, __be16 dport)
610 const struct iphdr *iph = ip_hdr(skb);
611 const struct net_device *dev =
612 skb_dst(skb) ? skb_dst(skb)->dev : skb->dev;
614 return __udp4_lib_lookup(dev_net(dev), iph->saddr, sport,
615 iph->daddr, dport, inet_iif(skb),
618 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
620 /* Must be called under rcu_read_lock().
621 * Does increment socket refcount.
623 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
624 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY)
625 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
626 __be32 daddr, __be16 dport, int dif)
630 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
631 dif, &udp_table, NULL);
632 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
636 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
639 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
640 __be16 loc_port, __be32 loc_addr,
641 __be16 rmt_port, __be32 rmt_addr,
642 int dif, unsigned short hnum)
644 struct inet_sock *inet = inet_sk(sk);
646 if (!net_eq(sock_net(sk), net) ||
647 udp_sk(sk)->udp_port_hash != hnum ||
648 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
649 (inet->inet_dport != rmt_port && inet->inet_dport) ||
650 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
651 ipv6_only_sock(sk) ||
652 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
654 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
660 * This routine is called by the ICMP module when it gets some
661 * sort of error condition. If err < 0 then the socket should
662 * be closed and the error returned to the user. If err > 0
663 * it's just the icmp type << 8 | icmp code.
664 * Header points to the ip header of the error packet. We move
665 * on past this. Then (as it used to claim before adjustment)
666 * header points to the first 8 bytes of the udp header. We need
667 * to find the appropriate port.
670 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
672 struct inet_sock *inet;
673 const struct iphdr *iph = (const struct iphdr *)skb->data;
674 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
675 const int type = icmp_hdr(skb)->type;
676 const int code = icmp_hdr(skb)->code;
680 struct net *net = dev_net(skb->dev);
682 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
683 iph->saddr, uh->source, skb->dev->ifindex, udptable,
686 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
687 return; /* No socket for error */
696 case ICMP_TIME_EXCEEDED:
699 case ICMP_SOURCE_QUENCH:
701 case ICMP_PARAMETERPROB:
705 case ICMP_DEST_UNREACH:
706 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
707 ipv4_sk_update_pmtu(skb, sk, info);
708 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
716 if (code <= NR_ICMP_UNREACH) {
717 harderr = icmp_err_convert[code].fatal;
718 err = icmp_err_convert[code].errno;
722 ipv4_sk_redirect(skb, sk);
727 * RFC1122: OK. Passes ICMP errors back to application, as per
730 if (!inet->recverr) {
731 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
734 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
737 sk->sk_error_report(sk);
742 void udp_err(struct sk_buff *skb, u32 info)
744 __udp4_lib_err(skb, info, &udp_table);
748 * Throw away all pending data and cancel the corking. Socket is locked.
750 void udp_flush_pending_frames(struct sock *sk)
752 struct udp_sock *up = udp_sk(sk);
757 ip_flush_pending_frames(sk);
760 EXPORT_SYMBOL(udp_flush_pending_frames);
763 * udp4_hwcsum - handle outgoing HW checksumming
764 * @skb: sk_buff containing the filled-in UDP header
765 * (checksum field must be zeroed out)
766 * @src: source IP address
767 * @dst: destination IP address
769 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
771 struct udphdr *uh = udp_hdr(skb);
772 int offset = skb_transport_offset(skb);
773 int len = skb->len - offset;
777 if (!skb_has_frag_list(skb)) {
779 * Only one fragment on the socket.
781 skb->csum_start = skb_transport_header(skb) - skb->head;
782 skb->csum_offset = offsetof(struct udphdr, check);
783 uh->check = ~csum_tcpudp_magic(src, dst, len,
786 struct sk_buff *frags;
789 * HW-checksum won't work as there are two or more
790 * fragments on the socket so that all csums of sk_buffs
793 skb_walk_frags(skb, frags) {
794 csum = csum_add(csum, frags->csum);
798 csum = skb_checksum(skb, offset, hlen, csum);
799 skb->ip_summed = CHECKSUM_NONE;
801 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
803 uh->check = CSUM_MANGLED_0;
806 EXPORT_SYMBOL_GPL(udp4_hwcsum);
808 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
809 * for the simple case like when setting the checksum for a UDP tunnel.
811 void udp_set_csum(bool nocheck, struct sk_buff *skb,
812 __be32 saddr, __be32 daddr, int len)
814 struct udphdr *uh = udp_hdr(skb);
818 } else if (skb_is_gso(skb)) {
819 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
820 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
822 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
824 uh->check = CSUM_MANGLED_0;
826 skb->ip_summed = CHECKSUM_PARTIAL;
827 skb->csum_start = skb_transport_header(skb) - skb->head;
828 skb->csum_offset = offsetof(struct udphdr, check);
829 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
832 EXPORT_SYMBOL(udp_set_csum);
834 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
836 struct sock *sk = skb->sk;
837 struct inet_sock *inet = inet_sk(sk);
840 int is_udplite = IS_UDPLITE(sk);
841 int offset = skb_transport_offset(skb);
842 int len = skb->len - offset;
846 * Create a UDP header
849 uh->source = inet->inet_sport;
850 uh->dest = fl4->fl4_dport;
851 uh->len = htons(len);
854 if (is_udplite) /* UDP-Lite */
855 csum = udplite_csum(skb);
857 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
859 skb->ip_summed = CHECKSUM_NONE;
862 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
864 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
868 csum = udp_csum(skb);
870 /* add protocol-dependent pseudo-header */
871 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
872 sk->sk_protocol, csum);
874 uh->check = CSUM_MANGLED_0;
877 err = ip_send_skb(sock_net(sk), skb);
879 if (err == -ENOBUFS && !inet->recverr) {
880 UDP_INC_STATS_USER(sock_net(sk),
881 UDP_MIB_SNDBUFERRORS, is_udplite);
885 UDP_INC_STATS_USER(sock_net(sk),
886 UDP_MIB_OUTDATAGRAMS, is_udplite);
891 * Push out all pending data as one UDP datagram. Socket is locked.
893 int udp_push_pending_frames(struct sock *sk)
895 struct udp_sock *up = udp_sk(sk);
896 struct inet_sock *inet = inet_sk(sk);
897 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
901 skb = ip_finish_skb(sk, fl4);
905 err = udp_send_skb(skb, fl4);
912 EXPORT_SYMBOL(udp_push_pending_frames);
914 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
916 struct inet_sock *inet = inet_sk(sk);
917 struct udp_sock *up = udp_sk(sk);
918 struct flowi4 fl4_stack;
921 struct ipcm_cookie ipc;
922 struct rtable *rt = NULL;
925 __be32 daddr, faddr, saddr;
928 int err, is_udplite = IS_UDPLITE(sk);
929 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
930 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
932 struct ip_options_data opt_copy;
941 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
949 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
951 fl4 = &inet->cork.fl.u.ip4;
954 * There are pending frames.
955 * The socket lock must be held while it's corked.
958 if (likely(up->pending)) {
959 if (unlikely(up->pending != AF_INET)) {
967 ulen += sizeof(struct udphdr);
970 * Get and verify the address.
973 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
974 if (msg->msg_namelen < sizeof(*usin))
976 if (usin->sin_family != AF_INET) {
977 if (usin->sin_family != AF_UNSPEC)
978 return -EAFNOSUPPORT;
981 daddr = usin->sin_addr.s_addr;
982 dport = usin->sin_port;
986 if (sk->sk_state != TCP_ESTABLISHED)
987 return -EDESTADDRREQ;
988 daddr = inet->inet_daddr;
989 dport = inet->inet_dport;
990 /* Open fast path for connected socket.
991 Route will not be used, if at least one option is set.
996 ipc.sockc.tsflags = sk->sk_tsflags;
997 ipc.addr = inet->inet_saddr;
998 ipc.oif = sk->sk_bound_dev_if;
1000 if (msg->msg_controllen) {
1001 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
1002 if (unlikely(err)) {
1011 struct ip_options_rcu *inet_opt;
1014 inet_opt = rcu_dereference(inet->inet_opt);
1016 memcpy(&opt_copy, inet_opt,
1017 sizeof(*inet_opt) + inet_opt->opt.optlen);
1018 ipc.opt = &opt_copy.opt;
1024 ipc.addr = faddr = daddr;
1026 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
1028 if (ipc.opt && ipc.opt->opt.srr) {
1031 faddr = ipc.opt->opt.faddr;
1034 tos = get_rttos(&ipc, inet);
1035 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1036 (msg->msg_flags & MSG_DONTROUTE) ||
1037 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1042 if (ipv4_is_multicast(daddr)) {
1044 ipc.oif = inet->mc_index;
1046 saddr = inet->mc_addr;
1048 } else if (!ipc.oif)
1049 ipc.oif = inet->uc_index;
1052 rt = (struct rtable *)sk_dst_check(sk, 0);
1055 struct net *net = sock_net(sk);
1056 __u8 flow_flags = inet_sk_flowi_flags(sk);
1060 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1061 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1063 faddr, saddr, dport, inet->inet_sport);
1065 if (!saddr && ipc.oif) {
1066 err = l3mdev_get_saddr(net, ipc.oif, fl4);
1071 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1072 rt = ip_route_output_flow(net, fl4, sk);
1076 if (err == -ENETUNREACH)
1077 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1082 if ((rt->rt_flags & RTCF_BROADCAST) &&
1083 !sock_flag(sk, SOCK_BROADCAST))
1086 sk_dst_set(sk, dst_clone(&rt->dst));
1089 if (msg->msg_flags&MSG_CONFIRM)
1095 daddr = ipc.addr = fl4->daddr;
1097 /* Lockless fast path for the non-corking case. */
1099 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1100 sizeof(struct udphdr), &ipc, &rt,
1103 if (!IS_ERR_OR_NULL(skb))
1104 err = udp_send_skb(skb, fl4);
1109 if (unlikely(up->pending)) {
1110 /* The socket is already corked while preparing it. */
1111 /* ... which is an evident application bug. --ANK */
1114 net_dbg_ratelimited("cork app bug 2\n");
1119 * Now cork the socket to pend data.
1121 fl4 = &inet->cork.fl.u.ip4;
1124 fl4->fl4_dport = dport;
1125 fl4->fl4_sport = inet->inet_sport;
1126 up->pending = AF_INET;
1130 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1131 sizeof(struct udphdr), &ipc, &rt,
1132 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1134 udp_flush_pending_frames(sk);
1136 err = udp_push_pending_frames(sk);
1137 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1148 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1149 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1150 * we don't have a good statistic (IpOutDiscards but it can be too many
1151 * things). We could add another new stat but at least for now that
1152 * seems like overkill.
1154 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1155 UDP_INC_STATS_USER(sock_net(sk),
1156 UDP_MIB_SNDBUFERRORS, is_udplite);
1161 dst_confirm(&rt->dst);
1162 if (!(msg->msg_flags&MSG_PROBE) || len)
1163 goto back_from_confirm;
1167 EXPORT_SYMBOL(udp_sendmsg);
1169 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1170 size_t size, int flags)
1172 struct inet_sock *inet = inet_sk(sk);
1173 struct udp_sock *up = udp_sk(sk);
1176 if (flags & MSG_SENDPAGE_NOTLAST)
1180 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1182 /* Call udp_sendmsg to specify destination address which
1183 * sendpage interface can't pass.
1184 * This will succeed only when the socket is connected.
1186 ret = udp_sendmsg(sk, &msg, 0);
1193 if (unlikely(!up->pending)) {
1196 net_dbg_ratelimited("udp cork app bug 3\n");
1200 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1201 page, offset, size, flags);
1202 if (ret == -EOPNOTSUPP) {
1204 return sock_no_sendpage(sk->sk_socket, page, offset,
1208 udp_flush_pending_frames(sk);
1213 if (!(up->corkflag || (flags&MSG_MORE)))
1214 ret = udp_push_pending_frames(sk);
1223 * first_packet_length - return length of first packet in receive queue
1226 * Drops all bad checksum frames, until a valid one is found.
1227 * Returns the length of found skb, or 0 if none is found.
1229 static unsigned int first_packet_length(struct sock *sk)
1231 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1232 struct sk_buff *skb;
1235 __skb_queue_head_init(&list_kill);
1237 spin_lock_bh(&rcvq->lock);
1238 while ((skb = skb_peek(rcvq)) != NULL &&
1239 udp_lib_checksum_complete(skb)) {
1240 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1242 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1244 atomic_inc(&sk->sk_drops);
1245 __skb_unlink(skb, rcvq);
1246 __skb_queue_tail(&list_kill, skb);
1248 res = skb ? skb->len : 0;
1249 spin_unlock_bh(&rcvq->lock);
1251 if (!skb_queue_empty(&list_kill)) {
1252 bool slow = lock_sock_fast(sk);
1254 __skb_queue_purge(&list_kill);
1255 sk_mem_reclaim_partial(sk);
1256 unlock_sock_fast(sk, slow);
1262 * IOCTL requests applicable to the UDP protocol
1265 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1270 int amount = sk_wmem_alloc_get(sk);
1272 return put_user(amount, (int __user *)arg);
1277 unsigned int amount = first_packet_length(sk);
1281 * We will only return the amount
1282 * of this packet since that is all
1283 * that will be read.
1285 return put_user(amount, (int __user *)arg);
1289 return -ENOIOCTLCMD;
1294 EXPORT_SYMBOL(udp_ioctl);
1297 * This should be easy, if there is something there we
1298 * return it, otherwise we block.
1301 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1302 int flags, int *addr_len)
1304 struct inet_sock *inet = inet_sk(sk);
1305 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1306 struct sk_buff *skb;
1307 unsigned int ulen, copied;
1308 int peeked, peeking, off;
1310 int is_udplite = IS_UDPLITE(sk);
1311 bool checksum_valid = false;
1314 if (flags & MSG_ERRQUEUE)
1315 return ip_recv_error(sk, msg, len, addr_len);
1318 peeking = off = sk_peek_offset(sk, flags);
1319 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1320 &peeked, &off, &err);
1326 if (copied > ulen - off)
1327 copied = ulen - off;
1328 else if (copied < ulen)
1329 msg->msg_flags |= MSG_TRUNC;
1332 * If checksum is needed at all, try to do it while copying the
1333 * data. If the data is truncated, or if we only want a partial
1334 * coverage checksum (UDP-Lite), do it before the copy.
1337 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov || peeking) {
1338 checksum_valid = !udp_lib_checksum_complete(skb);
1339 if (!checksum_valid)
1343 if (checksum_valid || skb_csum_unnecessary(skb))
1344 err = skb_copy_datagram_msg(skb, off, msg, copied);
1346 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1352 if (unlikely(err)) {
1353 trace_kfree_skb(skb, udp_recvmsg);
1355 atomic_inc(&sk->sk_drops);
1356 UDP_INC_STATS_USER(sock_net(sk),
1357 UDP_MIB_INERRORS, is_udplite);
1359 skb_free_datagram_locked(sk, skb);
1364 UDP_INC_STATS_USER(sock_net(sk),
1365 UDP_MIB_INDATAGRAMS, is_udplite);
1367 sock_recv_ts_and_drops(msg, sk, skb);
1369 /* Copy the address. */
1371 sin->sin_family = AF_INET;
1372 sin->sin_port = udp_hdr(skb)->source;
1373 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1374 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1375 *addr_len = sizeof(*sin);
1377 if (inet->cmsg_flags)
1378 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr) + off);
1381 if (flags & MSG_TRUNC)
1384 __skb_free_datagram_locked(sk, skb, peeking ? -err : err);
1388 slow = lock_sock_fast(sk);
1389 if (!skb_kill_datagram(sk, skb, flags)) {
1390 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1391 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1393 unlock_sock_fast(sk, slow);
1395 /* starting over for a new packet, but check if we need to yield */
1397 msg->msg_flags &= ~MSG_TRUNC;
1401 int udp_disconnect(struct sock *sk, int flags)
1403 struct inet_sock *inet = inet_sk(sk);
1405 * 1003.1g - break association.
1408 sk->sk_state = TCP_CLOSE;
1409 inet->inet_daddr = 0;
1410 inet->inet_dport = 0;
1411 sock_rps_reset_rxhash(sk);
1412 sk->sk_bound_dev_if = 0;
1413 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1414 inet_reset_saddr(sk);
1416 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1417 sk->sk_prot->unhash(sk);
1418 inet->inet_sport = 0;
1423 EXPORT_SYMBOL(udp_disconnect);
1425 void udp_lib_unhash(struct sock *sk)
1427 if (sk_hashed(sk)) {
1428 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1429 struct udp_hslot *hslot, *hslot2;
1431 hslot = udp_hashslot(udptable, sock_net(sk),
1432 udp_sk(sk)->udp_port_hash);
1433 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1435 spin_lock_bh(&hslot->lock);
1436 if (rcu_access_pointer(sk->sk_reuseport_cb))
1437 reuseport_detach_sock(sk);
1438 if (sk_del_node_init_rcu(sk)) {
1440 inet_sk(sk)->inet_num = 0;
1441 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1443 spin_lock(&hslot2->lock);
1444 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1446 spin_unlock(&hslot2->lock);
1448 spin_unlock_bh(&hslot->lock);
1451 EXPORT_SYMBOL(udp_lib_unhash);
1454 * inet_rcv_saddr was changed, we must rehash secondary hash
1456 void udp_lib_rehash(struct sock *sk, u16 newhash)
1458 if (sk_hashed(sk)) {
1459 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1460 struct udp_hslot *hslot, *hslot2, *nhslot2;
1462 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1463 nhslot2 = udp_hashslot2(udptable, newhash);
1464 udp_sk(sk)->udp_portaddr_hash = newhash;
1466 if (hslot2 != nhslot2 ||
1467 rcu_access_pointer(sk->sk_reuseport_cb)) {
1468 hslot = udp_hashslot(udptable, sock_net(sk),
1469 udp_sk(sk)->udp_port_hash);
1470 /* we must lock primary chain too */
1471 spin_lock_bh(&hslot->lock);
1472 if (rcu_access_pointer(sk->sk_reuseport_cb))
1473 reuseport_detach_sock(sk);
1475 if (hslot2 != nhslot2) {
1476 spin_lock(&hslot2->lock);
1477 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1479 spin_unlock(&hslot2->lock);
1481 spin_lock(&nhslot2->lock);
1482 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1485 spin_unlock(&nhslot2->lock);
1488 spin_unlock_bh(&hslot->lock);
1492 EXPORT_SYMBOL(udp_lib_rehash);
1494 static void udp_v4_rehash(struct sock *sk)
1496 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1497 inet_sk(sk)->inet_rcv_saddr,
1498 inet_sk(sk)->inet_num);
1499 udp_lib_rehash(sk, new_hash);
1502 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1506 if (inet_sk(sk)->inet_daddr) {
1507 sock_rps_save_rxhash(sk, skb);
1508 sk_mark_napi_id(sk, skb);
1509 sk_incoming_cpu_update(sk);
1512 rc = __sock_queue_rcv_skb(sk, skb);
1514 int is_udplite = IS_UDPLITE(sk);
1516 /* Note that an ENOMEM error is charged twice */
1518 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1520 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1522 trace_udp_fail_queue_rcv_skb(rc, sk);
1530 static struct static_key udp_encap_needed __read_mostly;
1531 void udp_encap_enable(void)
1533 if (!static_key_enabled(&udp_encap_needed))
1534 static_key_slow_inc(&udp_encap_needed);
1536 EXPORT_SYMBOL(udp_encap_enable);
1541 * >0: "udp encap" protocol resubmission
1543 * Note that in the success and error cases, the skb is assumed to
1544 * have either been requeued or freed.
1546 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1548 struct udp_sock *up = udp_sk(sk);
1550 int is_udplite = IS_UDPLITE(sk);
1553 * Charge it to the socket, dropping if the queue is full.
1555 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1559 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1560 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1563 * This is an encapsulation socket so pass the skb to
1564 * the socket's udp_encap_rcv() hook. Otherwise, just
1565 * fall through and pass this up the UDP socket.
1566 * up->encap_rcv() returns the following value:
1567 * =0 if skb was successfully passed to the encap
1568 * handler or was discarded by it.
1569 * >0 if skb should be passed on to UDP.
1570 * <0 if skb should be resubmitted as proto -N
1573 /* if we're overly short, let UDP handle it */
1574 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1575 if (skb->len > sizeof(struct udphdr) && encap_rcv) {
1578 /* Verify checksum before giving to encap */
1579 if (udp_lib_checksum_complete(skb))
1582 ret = encap_rcv(sk, skb);
1584 UDP_INC_STATS_BH(sock_net(sk),
1585 UDP_MIB_INDATAGRAMS,
1591 /* FALLTHROUGH -- it's a UDP Packet */
1595 * UDP-Lite specific tests, ignored on UDP sockets
1597 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1600 * MIB statistics other than incrementing the error count are
1601 * disabled for the following two types of errors: these depend
1602 * on the application settings, not on the functioning of the
1603 * protocol stack as such.
1605 * RFC 3828 here recommends (sec 3.3): "There should also be a
1606 * way ... to ... at least let the receiving application block
1607 * delivery of packets with coverage values less than a value
1608 * provided by the application."
1610 if (up->pcrlen == 0) { /* full coverage was set */
1611 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1612 UDP_SKB_CB(skb)->cscov, skb->len);
1615 /* The next case involves violating the min. coverage requested
1616 * by the receiver. This is subtle: if receiver wants x and x is
1617 * greater than the buffersize/MTU then receiver will complain
1618 * that it wants x while sender emits packets of smaller size y.
1619 * Therefore the above ...()->partial_cov statement is essential.
1621 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1622 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1623 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1628 if (rcu_access_pointer(sk->sk_filter)) {
1629 if (udp_lib_checksum_complete(skb))
1631 if (sk_filter(sk, skb))
1635 udp_csum_pull_header(skb);
1636 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1637 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1644 ipv4_pktinfo_prepare(sk, skb);
1646 if (!sock_owned_by_user(sk))
1647 rc = __udp_queue_rcv_skb(sk, skb);
1648 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1657 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1659 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1660 atomic_inc(&sk->sk_drops);
1665 /* For TCP sockets, sk_rx_dst is protected by socket lock
1666 * For UDP, we use xchg() to guard against concurrent changes.
1668 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1670 struct dst_entry *old;
1673 old = xchg(&sk->sk_rx_dst, dst);
1678 * Multicasts and broadcasts go to each listener.
1680 * Note: called only from the BH handler context.
1682 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1684 __be32 saddr, __be32 daddr,
1685 struct udp_table *udptable,
1688 struct sock *sk, *first = NULL;
1689 unsigned short hnum = ntohs(uh->dest);
1690 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1691 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1692 unsigned int offset = offsetof(typeof(*sk), sk_node);
1693 int dif = skb->dev->ifindex;
1694 struct hlist_node *node;
1695 struct sk_buff *nskb;
1698 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1700 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1702 hslot = &udp_table.hash2[hash2];
1703 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1706 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1707 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1708 uh->source, saddr, dif, hnum))
1715 nskb = skb_clone(skb, GFP_ATOMIC);
1717 if (unlikely(!nskb)) {
1718 atomic_inc(&sk->sk_drops);
1719 UDP_INC_STATS_BH(net, UDP_MIB_RCVBUFERRORS,
1721 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS,
1725 if (udp_queue_rcv_skb(sk, nskb) > 0)
1729 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1730 if (use_hash2 && hash2 != hash2_any) {
1736 if (udp_queue_rcv_skb(first, skb) > 0)
1740 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
1741 proto == IPPROTO_UDPLITE);
1746 /* Initialize UDP checksum. If exited with zero value (success),
1747 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1748 * Otherwise, csum completion requires chacksumming packet body,
1749 * including udp header and folding it to skb->csum.
1751 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1756 UDP_SKB_CB(skb)->partial_cov = 0;
1757 UDP_SKB_CB(skb)->cscov = skb->len;
1759 if (proto == IPPROTO_UDPLITE) {
1760 err = udplite_checksum_init(skb, uh);
1765 return skb_checksum_init_zero_check(skb, proto, uh->check,
1766 inet_compute_pseudo);
1770 * All we need to do is get the socket, and then do a checksum.
1773 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1778 unsigned short ulen;
1779 struct rtable *rt = skb_rtable(skb);
1780 __be32 saddr, daddr;
1781 struct net *net = dev_net(skb->dev);
1784 * Validate the packet.
1786 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1787 goto drop; /* No space for header. */
1790 ulen = ntohs(uh->len);
1791 saddr = ip_hdr(skb)->saddr;
1792 daddr = ip_hdr(skb)->daddr;
1794 if (ulen > skb->len)
1797 if (proto == IPPROTO_UDP) {
1798 /* UDP validates ulen. */
1799 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1804 if (udp4_csum_init(skb, uh, proto))
1807 sk = skb_steal_sock(skb);
1809 struct dst_entry *dst = skb_dst(skb);
1812 if (unlikely(sk->sk_rx_dst != dst))
1813 udp_sk_rx_dst_set(sk, dst);
1815 ret = udp_queue_rcv_skb(sk, skb);
1817 /* a return value > 0 means to resubmit the input, but
1818 * it wants the return to be -protocol, or 0
1825 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1826 return __udp4_lib_mcast_deliver(net, skb, uh,
1827 saddr, daddr, udptable, proto);
1829 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1833 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1834 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1835 inet_compute_pseudo);
1837 ret = udp_queue_rcv_skb(sk, skb);
1839 /* a return value > 0 means to resubmit the input, but
1840 * it wants the return to be -protocol, or 0
1847 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1851 /* No socket. Drop packet silently, if checksum is wrong */
1852 if (udp_lib_checksum_complete(skb))
1855 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1856 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1859 * Hmm. We got an UDP packet to a port to which we
1860 * don't wanna listen. Ignore it.
1866 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1867 proto == IPPROTO_UDPLITE ? "Lite" : "",
1868 &saddr, ntohs(uh->source),
1870 &daddr, ntohs(uh->dest));
1875 * RFC1122: OK. Discards the bad packet silently (as far as
1876 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1878 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1879 proto == IPPROTO_UDPLITE ? "Lite" : "",
1880 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1882 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1884 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1889 /* We can only early demux multicast if there is a single matching socket.
1890 * If more than one socket found returns NULL
1892 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1893 __be16 loc_port, __be32 loc_addr,
1894 __be16 rmt_port, __be32 rmt_addr,
1897 struct sock *sk, *result;
1898 unsigned short hnum = ntohs(loc_port);
1899 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
1900 struct udp_hslot *hslot = &udp_table.hash[slot];
1902 /* Do not bother scanning a too big list */
1903 if (hslot->count > 10)
1907 sk_for_each_rcu(sk, &hslot->head) {
1908 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
1909 rmt_port, rmt_addr, dif, hnum)) {
1919 /* For unicast we should only early demux connected sockets or we can
1920 * break forwarding setups. The chains here can be long so only check
1921 * if the first socket is an exact match and if not move on.
1923 static struct sock *__udp4_lib_demux_lookup(struct net *net,
1924 __be16 loc_port, __be32 loc_addr,
1925 __be16 rmt_port, __be32 rmt_addr,
1928 unsigned short hnum = ntohs(loc_port);
1929 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1930 unsigned int slot2 = hash2 & udp_table.mask;
1931 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1932 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
1933 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1936 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1937 if (INET_MATCH(sk, net, acookie, rmt_addr,
1938 loc_addr, ports, dif))
1940 /* Only check first socket in chain */
1946 void udp_v4_early_demux(struct sk_buff *skb)
1948 struct net *net = dev_net(skb->dev);
1949 const struct iphdr *iph;
1950 const struct udphdr *uh;
1951 struct sock *sk = NULL;
1952 struct dst_entry *dst;
1953 int dif = skb->dev->ifindex;
1956 /* validate the packet */
1957 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1963 if (skb->pkt_type == PACKET_BROADCAST ||
1964 skb->pkt_type == PACKET_MULTICAST) {
1965 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1970 /* we are supposed to accept bcast packets */
1971 if (skb->pkt_type == PACKET_MULTICAST) {
1972 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
1978 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1979 uh->source, iph->saddr, dif);
1980 } else if (skb->pkt_type == PACKET_HOST) {
1981 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1982 uh->source, iph->saddr, dif);
1985 if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2))
1989 skb->destructor = sock_efree;
1990 dst = READ_ONCE(sk->sk_rx_dst);
1993 dst = dst_check(dst, 0);
1995 /* DST_NOCACHE can not be used without taking a reference */
1996 if (dst->flags & DST_NOCACHE) {
1997 if (likely(atomic_inc_not_zero(&dst->__refcnt)))
1998 skb_dst_set(skb, dst);
2000 skb_dst_set_noref(skb, dst);
2005 int udp_rcv(struct sk_buff *skb)
2007 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2010 void udp_destroy_sock(struct sock *sk)
2012 struct udp_sock *up = udp_sk(sk);
2013 bool slow = lock_sock_fast(sk);
2014 udp_flush_pending_frames(sk);
2015 unlock_sock_fast(sk, slow);
2016 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2017 void (*encap_destroy)(struct sock *sk);
2018 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2025 * Socket option code for UDP
2027 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2028 char __user *optval, unsigned int optlen,
2029 int (*push_pending_frames)(struct sock *))
2031 struct udp_sock *up = udp_sk(sk);
2034 int is_udplite = IS_UDPLITE(sk);
2036 if (optlen < sizeof(int))
2039 if (get_user(val, (int __user *)optval))
2042 valbool = val ? 1 : 0;
2051 push_pending_frames(sk);
2059 case UDP_ENCAP_ESPINUDP:
2060 case UDP_ENCAP_ESPINUDP_NON_IKE:
2061 up->encap_rcv = xfrm4_udp_encap_rcv;
2063 case UDP_ENCAP_L2TPINUDP:
2064 up->encap_type = val;
2073 case UDP_NO_CHECK6_TX:
2074 up->no_check6_tx = valbool;
2077 case UDP_NO_CHECK6_RX:
2078 up->no_check6_rx = valbool;
2082 * UDP-Lite's partial checksum coverage (RFC 3828).
2084 /* The sender sets actual checksum coverage length via this option.
2085 * The case coverage > packet length is handled by send module. */
2086 case UDPLITE_SEND_CSCOV:
2087 if (!is_udplite) /* Disable the option on UDP sockets */
2088 return -ENOPROTOOPT;
2089 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2091 else if (val > USHRT_MAX)
2094 up->pcflag |= UDPLITE_SEND_CC;
2097 /* The receiver specifies a minimum checksum coverage value. To make
2098 * sense, this should be set to at least 8 (as done below). If zero is
2099 * used, this again means full checksum coverage. */
2100 case UDPLITE_RECV_CSCOV:
2101 if (!is_udplite) /* Disable the option on UDP sockets */
2102 return -ENOPROTOOPT;
2103 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2105 else if (val > USHRT_MAX)
2108 up->pcflag |= UDPLITE_RECV_CC;
2118 EXPORT_SYMBOL(udp_lib_setsockopt);
2120 int 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 ip_setsockopt(sk, level, optname, optval, optlen);
2129 #ifdef CONFIG_COMPAT
2130 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2131 char __user *optval, unsigned int optlen)
2133 if (level == SOL_UDP || level == SOL_UDPLITE)
2134 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2135 udp_push_pending_frames);
2136 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2140 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2141 char __user *optval, int __user *optlen)
2143 struct udp_sock *up = udp_sk(sk);
2146 if (get_user(len, optlen))
2149 len = min_t(unsigned int, len, sizeof(int));
2160 val = up->encap_type;
2163 case UDP_NO_CHECK6_TX:
2164 val = up->no_check6_tx;
2167 case UDP_NO_CHECK6_RX:
2168 val = up->no_check6_rx;
2171 /* The following two cannot be changed on UDP sockets, the return is
2172 * always 0 (which corresponds to the full checksum coverage of UDP). */
2173 case UDPLITE_SEND_CSCOV:
2177 case UDPLITE_RECV_CSCOV:
2182 return -ENOPROTOOPT;
2185 if (put_user(len, optlen))
2187 if (copy_to_user(optval, &val, len))
2191 EXPORT_SYMBOL(udp_lib_getsockopt);
2193 int udp_getsockopt(struct sock *sk, int level, int optname,
2194 char __user *optval, int __user *optlen)
2196 if (level == SOL_UDP || level == SOL_UDPLITE)
2197 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2198 return ip_getsockopt(sk, level, optname, optval, optlen);
2201 #ifdef CONFIG_COMPAT
2202 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2203 char __user *optval, int __user *optlen)
2205 if (level == SOL_UDP || level == SOL_UDPLITE)
2206 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2207 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2211 * udp_poll - wait for a UDP event.
2212 * @file - file struct
2214 * @wait - poll table
2216 * This is same as datagram poll, except for the special case of
2217 * blocking sockets. If application is using a blocking fd
2218 * and a packet with checksum error is in the queue;
2219 * then it could get return from select indicating data available
2220 * but then block when reading it. Add special case code
2221 * to work around these arguably broken applications.
2223 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2225 unsigned int mask = datagram_poll(file, sock, wait);
2226 struct sock *sk = sock->sk;
2228 sock_rps_record_flow(sk);
2230 /* Check for false positives due to checksum errors */
2231 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2232 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2233 mask &= ~(POLLIN | POLLRDNORM);
2238 EXPORT_SYMBOL(udp_poll);
2240 struct proto udp_prot = {
2242 .owner = THIS_MODULE,
2243 .close = udp_lib_close,
2244 .connect = ip4_datagram_connect,
2245 .disconnect = udp_disconnect,
2247 .destroy = udp_destroy_sock,
2248 .setsockopt = udp_setsockopt,
2249 .getsockopt = udp_getsockopt,
2250 .sendmsg = udp_sendmsg,
2251 .recvmsg = udp_recvmsg,
2252 .sendpage = udp_sendpage,
2253 .backlog_rcv = __udp_queue_rcv_skb,
2254 .release_cb = ip4_datagram_release_cb,
2255 .hash = udp_lib_hash,
2256 .unhash = udp_lib_unhash,
2257 .rehash = udp_v4_rehash,
2258 .get_port = udp_v4_get_port,
2259 .memory_allocated = &udp_memory_allocated,
2260 .sysctl_mem = sysctl_udp_mem,
2261 .sysctl_wmem = &sysctl_udp_wmem_min,
2262 .sysctl_rmem = &sysctl_udp_rmem_min,
2263 .obj_size = sizeof(struct udp_sock),
2264 .slab_flags = SLAB_DESTROY_BY_RCU,
2265 .h.udp_table = &udp_table,
2266 #ifdef CONFIG_COMPAT
2267 .compat_setsockopt = compat_udp_setsockopt,
2268 .compat_getsockopt = compat_udp_getsockopt,
2270 .clear_sk = sk_prot_clear_portaddr_nulls,
2272 EXPORT_SYMBOL(udp_prot);
2274 /* ------------------------------------------------------------------------ */
2275 #ifdef CONFIG_PROC_FS
2277 static struct sock *udp_get_first(struct seq_file *seq, int start)
2280 struct udp_iter_state *state = seq->private;
2281 struct net *net = seq_file_net(seq);
2283 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2285 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2287 if (hlist_empty(&hslot->head))
2290 spin_lock_bh(&hslot->lock);
2291 sk_for_each(sk, &hslot->head) {
2292 if (!net_eq(sock_net(sk), net))
2294 if (sk->sk_family == state->family)
2297 spin_unlock_bh(&hslot->lock);
2304 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2306 struct udp_iter_state *state = seq->private;
2307 struct net *net = seq_file_net(seq);
2311 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2314 if (state->bucket <= state->udp_table->mask)
2315 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2316 return udp_get_first(seq, state->bucket + 1);
2321 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2323 struct sock *sk = udp_get_first(seq, 0);
2326 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2328 return pos ? NULL : sk;
2331 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2333 struct udp_iter_state *state = seq->private;
2334 state->bucket = MAX_UDP_PORTS;
2336 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2339 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2343 if (v == SEQ_START_TOKEN)
2344 sk = udp_get_idx(seq, 0);
2346 sk = udp_get_next(seq, v);
2352 static void udp_seq_stop(struct seq_file *seq, void *v)
2354 struct udp_iter_state *state = seq->private;
2356 if (state->bucket <= state->udp_table->mask)
2357 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2360 int udp_seq_open(struct inode *inode, struct file *file)
2362 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2363 struct udp_iter_state *s;
2366 err = seq_open_net(inode, file, &afinfo->seq_ops,
2367 sizeof(struct udp_iter_state));
2371 s = ((struct seq_file *)file->private_data)->private;
2372 s->family = afinfo->family;
2373 s->udp_table = afinfo->udp_table;
2376 EXPORT_SYMBOL(udp_seq_open);
2378 /* ------------------------------------------------------------------------ */
2379 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2381 struct proc_dir_entry *p;
2384 afinfo->seq_ops.start = udp_seq_start;
2385 afinfo->seq_ops.next = udp_seq_next;
2386 afinfo->seq_ops.stop = udp_seq_stop;
2388 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2389 afinfo->seq_fops, afinfo);
2394 EXPORT_SYMBOL(udp_proc_register);
2396 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2398 remove_proc_entry(afinfo->name, net->proc_net);
2400 EXPORT_SYMBOL(udp_proc_unregister);
2402 /* ------------------------------------------------------------------------ */
2403 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2406 struct inet_sock *inet = inet_sk(sp);
2407 __be32 dest = inet->inet_daddr;
2408 __be32 src = inet->inet_rcv_saddr;
2409 __u16 destp = ntohs(inet->inet_dport);
2410 __u16 srcp = ntohs(inet->inet_sport);
2412 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2413 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2414 bucket, src, srcp, dest, destp, sp->sk_state,
2415 sk_wmem_alloc_get(sp),
2416 sk_rmem_alloc_get(sp),
2418 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2420 atomic_read(&sp->sk_refcnt), sp,
2421 atomic_read(&sp->sk_drops));
2424 int udp4_seq_show(struct seq_file *seq, void *v)
2426 seq_setwidth(seq, 127);
2427 if (v == SEQ_START_TOKEN)
2428 seq_puts(seq, " sl local_address rem_address st tx_queue "
2429 "rx_queue tr tm->when retrnsmt uid timeout "
2430 "inode ref pointer drops");
2432 struct udp_iter_state *state = seq->private;
2434 udp4_format_sock(v, seq, state->bucket);
2440 static const struct file_operations udp_afinfo_seq_fops = {
2441 .owner = THIS_MODULE,
2442 .open = udp_seq_open,
2444 .llseek = seq_lseek,
2445 .release = seq_release_net
2448 /* ------------------------------------------------------------------------ */
2449 static struct udp_seq_afinfo udp4_seq_afinfo = {
2452 .udp_table = &udp_table,
2453 .seq_fops = &udp_afinfo_seq_fops,
2455 .show = udp4_seq_show,
2459 static int __net_init udp4_proc_init_net(struct net *net)
2461 return udp_proc_register(net, &udp4_seq_afinfo);
2464 static void __net_exit udp4_proc_exit_net(struct net *net)
2466 udp_proc_unregister(net, &udp4_seq_afinfo);
2469 static struct pernet_operations udp4_net_ops = {
2470 .init = udp4_proc_init_net,
2471 .exit = udp4_proc_exit_net,
2474 int __init udp4_proc_init(void)
2476 return register_pernet_subsys(&udp4_net_ops);
2479 void udp4_proc_exit(void)
2481 unregister_pernet_subsys(&udp4_net_ops);
2483 #endif /* CONFIG_PROC_FS */
2485 static __initdata unsigned long uhash_entries;
2486 static int __init set_uhash_entries(char *str)
2493 ret = kstrtoul(str, 0, &uhash_entries);
2497 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2498 uhash_entries = UDP_HTABLE_SIZE_MIN;
2501 __setup("uhash_entries=", set_uhash_entries);
2503 void __init udp_table_init(struct udp_table *table, const char *name)
2507 table->hash = alloc_large_system_hash(name,
2508 2 * sizeof(struct udp_hslot),
2510 21, /* one slot per 2 MB */
2514 UDP_HTABLE_SIZE_MIN,
2517 table->hash2 = table->hash + (table->mask + 1);
2518 for (i = 0; i <= table->mask; i++) {
2519 INIT_HLIST_HEAD(&table->hash[i].head);
2520 table->hash[i].count = 0;
2521 spin_lock_init(&table->hash[i].lock);
2523 for (i = 0; i <= table->mask; i++) {
2524 INIT_HLIST_HEAD(&table->hash2[i].head);
2525 table->hash2[i].count = 0;
2526 spin_lock_init(&table->hash2[i].lock);
2530 u32 udp_flow_hashrnd(void)
2532 static u32 hashrnd __read_mostly;
2534 net_get_random_once(&hashrnd, sizeof(hashrnd));
2538 EXPORT_SYMBOL(udp_flow_hashrnd);
2540 void __init udp_init(void)
2542 unsigned long limit;
2544 udp_table_init(&udp_table, "UDP");
2545 limit = nr_free_buffer_pages() / 8;
2546 limit = max(limit, 128UL);
2547 sysctl_udp_mem[0] = limit / 4 * 3;
2548 sysctl_udp_mem[1] = limit;
2549 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2551 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2552 sysctl_udp_wmem_min = SK_MEM_QUANTUM;