Merge branch 'x86/rdrand' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
[wandboard.git] / net / ipv4 / udp.c
blob5d075b5f70fcd61dabe0c38ac6864771b4022093
1 /*
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).
8 * Authors: Ross Biro
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>
14 * Fixes:
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
28 * does NOT close.
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
57 * for connect.
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
64 * datagrams.
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 #include <asm/system.h>
81 #include <asm/uaccess.h>
82 #include <asm/ioctls.h>
83 #include <linux/bootmem.h>
84 #include <linux/highmem.h>
85 #include <linux/swap.h>
86 #include <linux/types.h>
87 #include <linux/fcntl.h>
88 #include <linux/module.h>
89 #include <linux/socket.h>
90 #include <linux/sockios.h>
91 #include <linux/igmp.h>
92 #include <linux/in.h>
93 #include <linux/errno.h>
94 #include <linux/timer.h>
95 #include <linux/mm.h>
96 #include <linux/inet.h>
97 #include <linux/netdevice.h>
98 #include <linux/slab.h>
99 #include <net/tcp_states.h>
100 #include <linux/skbuff.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <net/net_namespace.h>
104 #include <net/icmp.h>
105 #include <net/route.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <trace/events/udp.h>
109 #include "udp_impl.h"
111 struct udp_table udp_table __read_mostly;
112 EXPORT_SYMBOL(udp_table);
114 long sysctl_udp_mem[3] __read_mostly;
115 EXPORT_SYMBOL(sysctl_udp_mem);
117 int sysctl_udp_rmem_min __read_mostly;
118 EXPORT_SYMBOL(sysctl_udp_rmem_min);
120 int sysctl_udp_wmem_min __read_mostly;
121 EXPORT_SYMBOL(sysctl_udp_wmem_min);
123 atomic_long_t udp_memory_allocated;
124 EXPORT_SYMBOL(udp_memory_allocated);
126 #define MAX_UDP_PORTS 65536
127 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
129 static int udp_lib_lport_inuse(struct net *net, __u16 num,
130 const struct udp_hslot *hslot,
131 unsigned long *bitmap,
132 struct sock *sk,
133 int (*saddr_comp)(const struct sock *sk1,
134 const struct sock *sk2),
135 unsigned int log)
137 struct sock *sk2;
138 struct hlist_nulls_node *node;
140 sk_nulls_for_each(sk2, node, &hslot->head)
141 if (net_eq(sock_net(sk2), net) &&
142 sk2 != sk &&
143 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
144 (!sk2->sk_reuse || !sk->sk_reuse) &&
145 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
146 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
147 (*saddr_comp)(sk, sk2)) {
148 if (bitmap)
149 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
150 bitmap);
151 else
152 return 1;
154 return 0;
158 * Note: we still hold spinlock of primary hash chain, so no other writer
159 * can insert/delete a socket with local_port == num
161 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
162 struct udp_hslot *hslot2,
163 struct sock *sk,
164 int (*saddr_comp)(const struct sock *sk1,
165 const struct sock *sk2))
167 struct sock *sk2;
168 struct hlist_nulls_node *node;
169 int res = 0;
171 spin_lock(&hslot2->lock);
172 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
173 if (net_eq(sock_net(sk2), net) &&
174 sk2 != sk &&
175 (udp_sk(sk2)->udp_port_hash == num) &&
176 (!sk2->sk_reuse || !sk->sk_reuse) &&
177 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
178 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
179 (*saddr_comp)(sk, sk2)) {
180 res = 1;
181 break;
183 spin_unlock(&hslot2->lock);
184 return res;
188 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
190 * @sk: socket struct in question
191 * @snum: port number to look up
192 * @saddr_comp: AF-dependent comparison of bound local IP addresses
193 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
194 * with NULL address
196 int udp_lib_get_port(struct sock *sk, unsigned short snum,
197 int (*saddr_comp)(const struct sock *sk1,
198 const struct sock *sk2),
199 unsigned int hash2_nulladdr)
201 struct udp_hslot *hslot, *hslot2;
202 struct udp_table *udptable = sk->sk_prot->h.udp_table;
203 int error = 1;
204 struct net *net = sock_net(sk);
206 if (!snum) {
207 int low, high, remaining;
208 unsigned rand;
209 unsigned short first, last;
210 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
212 inet_get_local_port_range(&low, &high);
213 remaining = (high - low) + 1;
215 rand = net_random();
216 first = (((u64)rand * remaining) >> 32) + low;
218 * force rand to be an odd multiple of UDP_HTABLE_SIZE
220 rand = (rand | 1) * (udptable->mask + 1);
221 last = first + udptable->mask + 1;
222 do {
223 hslot = udp_hashslot(udptable, net, first);
224 bitmap_zero(bitmap, PORTS_PER_CHAIN);
225 spin_lock_bh(&hslot->lock);
226 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
227 saddr_comp, udptable->log);
229 snum = first;
231 * Iterate on all possible values of snum for this hash.
232 * Using steps of an odd multiple of UDP_HTABLE_SIZE
233 * give us randomization and full range coverage.
235 do {
236 if (low <= snum && snum <= high &&
237 !test_bit(snum >> udptable->log, bitmap) &&
238 !inet_is_reserved_local_port(snum))
239 goto found;
240 snum += rand;
241 } while (snum != first);
242 spin_unlock_bh(&hslot->lock);
243 } while (++first != last);
244 goto fail;
245 } else {
246 hslot = udp_hashslot(udptable, net, snum);
247 spin_lock_bh(&hslot->lock);
248 if (hslot->count > 10) {
249 int exist;
250 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
252 slot2 &= udptable->mask;
253 hash2_nulladdr &= udptable->mask;
255 hslot2 = udp_hashslot2(udptable, slot2);
256 if (hslot->count < hslot2->count)
257 goto scan_primary_hash;
259 exist = udp_lib_lport_inuse2(net, snum, hslot2,
260 sk, saddr_comp);
261 if (!exist && (hash2_nulladdr != slot2)) {
262 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
263 exist = udp_lib_lport_inuse2(net, snum, hslot2,
264 sk, saddr_comp);
266 if (exist)
267 goto fail_unlock;
268 else
269 goto found;
271 scan_primary_hash:
272 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
273 saddr_comp, 0))
274 goto fail_unlock;
276 found:
277 inet_sk(sk)->inet_num = snum;
278 udp_sk(sk)->udp_port_hash = snum;
279 udp_sk(sk)->udp_portaddr_hash ^= snum;
280 if (sk_unhashed(sk)) {
281 sk_nulls_add_node_rcu(sk, &hslot->head);
282 hslot->count++;
283 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
285 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
286 spin_lock(&hslot2->lock);
287 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
288 &hslot2->head);
289 hslot2->count++;
290 spin_unlock(&hslot2->lock);
292 error = 0;
293 fail_unlock:
294 spin_unlock_bh(&hslot->lock);
295 fail:
296 return error;
298 EXPORT_SYMBOL(udp_lib_get_port);
300 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
302 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
304 return (!ipv6_only_sock(sk2) &&
305 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
306 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
309 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
310 unsigned int port)
312 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
315 int udp_v4_get_port(struct sock *sk, unsigned short snum)
317 unsigned int hash2_nulladdr =
318 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
319 unsigned int hash2_partial =
320 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
322 /* precompute partial secondary hash */
323 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
324 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
327 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
328 unsigned short hnum,
329 __be16 sport, __be32 daddr, __be16 dport, int dif)
331 int score = -1;
333 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
334 !ipv6_only_sock(sk)) {
335 struct inet_sock *inet = inet_sk(sk);
337 score = (sk->sk_family == PF_INET ? 1 : 0);
338 if (inet->inet_rcv_saddr) {
339 if (inet->inet_rcv_saddr != daddr)
340 return -1;
341 score += 2;
343 if (inet->inet_daddr) {
344 if (inet->inet_daddr != saddr)
345 return -1;
346 score += 2;
348 if (inet->inet_dport) {
349 if (inet->inet_dport != sport)
350 return -1;
351 score += 2;
353 if (sk->sk_bound_dev_if) {
354 if (sk->sk_bound_dev_if != dif)
355 return -1;
356 score += 2;
359 return score;
363 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
365 #define SCORE2_MAX (1 + 2 + 2 + 2)
366 static inline int compute_score2(struct sock *sk, struct net *net,
367 __be32 saddr, __be16 sport,
368 __be32 daddr, unsigned int hnum, int dif)
370 int score = -1;
372 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
373 struct inet_sock *inet = inet_sk(sk);
375 if (inet->inet_rcv_saddr != daddr)
376 return -1;
377 if (inet->inet_num != hnum)
378 return -1;
380 score = (sk->sk_family == PF_INET ? 1 : 0);
381 if (inet->inet_daddr) {
382 if (inet->inet_daddr != saddr)
383 return -1;
384 score += 2;
386 if (inet->inet_dport) {
387 if (inet->inet_dport != sport)
388 return -1;
389 score += 2;
391 if (sk->sk_bound_dev_if) {
392 if (sk->sk_bound_dev_if != dif)
393 return -1;
394 score += 2;
397 return score;
401 /* called with read_rcu_lock() */
402 static struct sock *udp4_lib_lookup2(struct net *net,
403 __be32 saddr, __be16 sport,
404 __be32 daddr, unsigned int hnum, int dif,
405 struct udp_hslot *hslot2, unsigned int slot2)
407 struct sock *sk, *result;
408 struct hlist_nulls_node *node;
409 int score, badness;
411 begin:
412 result = NULL;
413 badness = -1;
414 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
415 score = compute_score2(sk, net, saddr, sport,
416 daddr, hnum, dif);
417 if (score > badness) {
418 result = sk;
419 badness = score;
420 if (score == SCORE2_MAX)
421 goto exact_match;
425 * if the nulls value we got at the end of this lookup is
426 * not the expected one, we must restart lookup.
427 * We probably met an item that was moved to another chain.
429 if (get_nulls_value(node) != slot2)
430 goto begin;
432 if (result) {
433 exact_match:
434 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
435 result = NULL;
436 else if (unlikely(compute_score2(result, net, saddr, sport,
437 daddr, hnum, dif) < badness)) {
438 sock_put(result);
439 goto begin;
442 return result;
445 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
446 * harder than this. -DaveM
448 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
449 __be16 sport, __be32 daddr, __be16 dport,
450 int dif, struct udp_table *udptable)
452 struct sock *sk, *result;
453 struct hlist_nulls_node *node;
454 unsigned short hnum = ntohs(dport);
455 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
456 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
457 int score, badness;
459 rcu_read_lock();
460 if (hslot->count > 10) {
461 hash2 = udp4_portaddr_hash(net, daddr, hnum);
462 slot2 = hash2 & udptable->mask;
463 hslot2 = &udptable->hash2[slot2];
464 if (hslot->count < hslot2->count)
465 goto begin;
467 result = udp4_lib_lookup2(net, saddr, sport,
468 daddr, hnum, dif,
469 hslot2, slot2);
470 if (!result) {
471 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
472 slot2 = hash2 & udptable->mask;
473 hslot2 = &udptable->hash2[slot2];
474 if (hslot->count < hslot2->count)
475 goto begin;
477 result = udp4_lib_lookup2(net, saddr, sport,
478 htonl(INADDR_ANY), hnum, dif,
479 hslot2, slot2);
481 rcu_read_unlock();
482 return result;
484 begin:
485 result = NULL;
486 badness = -1;
487 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
488 score = compute_score(sk, net, saddr, hnum, sport,
489 daddr, dport, dif);
490 if (score > badness) {
491 result = sk;
492 badness = score;
496 * if the nulls value we got at the end of this lookup is
497 * not the expected one, we must restart lookup.
498 * We probably met an item that was moved to another chain.
500 if (get_nulls_value(node) != slot)
501 goto begin;
503 if (result) {
504 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
505 result = NULL;
506 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
507 daddr, dport, dif) < badness)) {
508 sock_put(result);
509 goto begin;
512 rcu_read_unlock();
513 return result;
515 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
517 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
518 __be16 sport, __be16 dport,
519 struct udp_table *udptable)
521 struct sock *sk;
522 const struct iphdr *iph = ip_hdr(skb);
524 if (unlikely(sk = skb_steal_sock(skb)))
525 return sk;
526 else
527 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
528 iph->daddr, dport, inet_iif(skb),
529 udptable);
532 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
533 __be32 daddr, __be16 dport, int dif)
535 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
537 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
539 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
540 __be16 loc_port, __be32 loc_addr,
541 __be16 rmt_port, __be32 rmt_addr,
542 int dif)
544 struct hlist_nulls_node *node;
545 struct sock *s = sk;
546 unsigned short hnum = ntohs(loc_port);
548 sk_nulls_for_each_from(s, node) {
549 struct inet_sock *inet = inet_sk(s);
551 if (!net_eq(sock_net(s), net) ||
552 udp_sk(s)->udp_port_hash != hnum ||
553 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
554 (inet->inet_dport != rmt_port && inet->inet_dport) ||
555 (inet->inet_rcv_saddr &&
556 inet->inet_rcv_saddr != loc_addr) ||
557 ipv6_only_sock(s) ||
558 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
559 continue;
560 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
561 continue;
562 goto found;
564 s = NULL;
565 found:
566 return s;
570 * This routine is called by the ICMP module when it gets some
571 * sort of error condition. If err < 0 then the socket should
572 * be closed and the error returned to the user. If err > 0
573 * it's just the icmp type << 8 | icmp code.
574 * Header points to the ip header of the error packet. We move
575 * on past this. Then (as it used to claim before adjustment)
576 * header points to the first 8 bytes of the udp header. We need
577 * to find the appropriate port.
580 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
582 struct inet_sock *inet;
583 const struct iphdr *iph = (const struct iphdr *)skb->data;
584 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
585 const int type = icmp_hdr(skb)->type;
586 const int code = icmp_hdr(skb)->code;
587 struct sock *sk;
588 int harderr;
589 int err;
590 struct net *net = dev_net(skb->dev);
592 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
593 iph->saddr, uh->source, skb->dev->ifindex, udptable);
594 if (sk == NULL) {
595 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
596 return; /* No socket for error */
599 err = 0;
600 harderr = 0;
601 inet = inet_sk(sk);
603 switch (type) {
604 default:
605 case ICMP_TIME_EXCEEDED:
606 err = EHOSTUNREACH;
607 break;
608 case ICMP_SOURCE_QUENCH:
609 goto out;
610 case ICMP_PARAMETERPROB:
611 err = EPROTO;
612 harderr = 1;
613 break;
614 case ICMP_DEST_UNREACH:
615 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
616 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
617 err = EMSGSIZE;
618 harderr = 1;
619 break;
621 goto out;
623 err = EHOSTUNREACH;
624 if (code <= NR_ICMP_UNREACH) {
625 harderr = icmp_err_convert[code].fatal;
626 err = icmp_err_convert[code].errno;
628 break;
632 * RFC1122: OK. Passes ICMP errors back to application, as per
633 * 4.1.3.3.
635 if (!inet->recverr) {
636 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
637 goto out;
638 } else
639 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
641 sk->sk_err = err;
642 sk->sk_error_report(sk);
643 out:
644 sock_put(sk);
647 void udp_err(struct sk_buff *skb, u32 info)
649 __udp4_lib_err(skb, info, &udp_table);
653 * Throw away all pending data and cancel the corking. Socket is locked.
655 void udp_flush_pending_frames(struct sock *sk)
657 struct udp_sock *up = udp_sk(sk);
659 if (up->pending) {
660 up->len = 0;
661 up->pending = 0;
662 ip_flush_pending_frames(sk);
665 EXPORT_SYMBOL(udp_flush_pending_frames);
668 * udp4_hwcsum - handle outgoing HW checksumming
669 * @skb: sk_buff containing the filled-in UDP header
670 * (checksum field must be zeroed out)
671 * @src: source IP address
672 * @dst: destination IP address
674 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
676 struct udphdr *uh = udp_hdr(skb);
677 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
678 int offset = skb_transport_offset(skb);
679 int len = skb->len - offset;
680 int hlen = len;
681 __wsum csum = 0;
683 if (!frags) {
685 * Only one fragment on the socket.
687 skb->csum_start = skb_transport_header(skb) - skb->head;
688 skb->csum_offset = offsetof(struct udphdr, check);
689 uh->check = ~csum_tcpudp_magic(src, dst, len,
690 IPPROTO_UDP, 0);
691 } else {
693 * HW-checksum won't work as there are two or more
694 * fragments on the socket so that all csums of sk_buffs
695 * should be together
697 do {
698 csum = csum_add(csum, frags->csum);
699 hlen -= frags->len;
700 } while ((frags = frags->next));
702 csum = skb_checksum(skb, offset, hlen, csum);
703 skb->ip_summed = CHECKSUM_NONE;
705 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
706 if (uh->check == 0)
707 uh->check = CSUM_MANGLED_0;
711 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
713 struct sock *sk = skb->sk;
714 struct inet_sock *inet = inet_sk(sk);
715 struct udphdr *uh;
716 int err = 0;
717 int is_udplite = IS_UDPLITE(sk);
718 int offset = skb_transport_offset(skb);
719 int len = skb->len - offset;
720 __wsum csum = 0;
723 * Create a UDP header
725 uh = udp_hdr(skb);
726 uh->source = inet->inet_sport;
727 uh->dest = fl4->fl4_dport;
728 uh->len = htons(len);
729 uh->check = 0;
731 if (is_udplite) /* UDP-Lite */
732 csum = udplite_csum(skb);
734 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
736 skb->ip_summed = CHECKSUM_NONE;
737 goto send;
739 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
741 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
742 goto send;
744 } else
745 csum = udp_csum(skb);
747 /* add protocol-dependent pseudo-header */
748 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
749 sk->sk_protocol, csum);
750 if (uh->check == 0)
751 uh->check = CSUM_MANGLED_0;
753 send:
754 err = ip_send_skb(skb);
755 if (err) {
756 if (err == -ENOBUFS && !inet->recverr) {
757 UDP_INC_STATS_USER(sock_net(sk),
758 UDP_MIB_SNDBUFERRORS, is_udplite);
759 err = 0;
761 } else
762 UDP_INC_STATS_USER(sock_net(sk),
763 UDP_MIB_OUTDATAGRAMS, is_udplite);
764 return err;
768 * Push out all pending data as one UDP datagram. Socket is locked.
770 static int udp_push_pending_frames(struct sock *sk)
772 struct udp_sock *up = udp_sk(sk);
773 struct inet_sock *inet = inet_sk(sk);
774 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
775 struct sk_buff *skb;
776 int err = 0;
778 skb = ip_finish_skb(sk, fl4);
779 if (!skb)
780 goto out;
782 err = udp_send_skb(skb, fl4);
784 out:
785 up->len = 0;
786 up->pending = 0;
787 return err;
790 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
791 size_t len)
793 struct inet_sock *inet = inet_sk(sk);
794 struct udp_sock *up = udp_sk(sk);
795 struct flowi4 fl4_stack;
796 struct flowi4 *fl4;
797 int ulen = len;
798 struct ipcm_cookie ipc;
799 struct rtable *rt = NULL;
800 int free = 0;
801 int connected = 0;
802 __be32 daddr, faddr, saddr;
803 __be16 dport;
804 u8 tos;
805 int err, is_udplite = IS_UDPLITE(sk);
806 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
807 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
808 struct sk_buff *skb;
809 struct ip_options_data opt_copy;
811 if (len > 0xFFFF)
812 return -EMSGSIZE;
815 * Check the flags.
818 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
819 return -EOPNOTSUPP;
821 ipc.opt = NULL;
822 ipc.tx_flags = 0;
824 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
826 fl4 = &inet->cork.fl.u.ip4;
827 if (up->pending) {
829 * There are pending frames.
830 * The socket lock must be held while it's corked.
832 lock_sock(sk);
833 if (likely(up->pending)) {
834 if (unlikely(up->pending != AF_INET)) {
835 release_sock(sk);
836 return -EINVAL;
838 goto do_append_data;
840 release_sock(sk);
842 ulen += sizeof(struct udphdr);
845 * Get and verify the address.
847 if (msg->msg_name) {
848 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
849 if (msg->msg_namelen < sizeof(*usin))
850 return -EINVAL;
851 if (usin->sin_family != AF_INET) {
852 if (usin->sin_family != AF_UNSPEC)
853 return -EAFNOSUPPORT;
856 daddr = usin->sin_addr.s_addr;
857 dport = usin->sin_port;
858 if (dport == 0)
859 return -EINVAL;
860 } else {
861 if (sk->sk_state != TCP_ESTABLISHED)
862 return -EDESTADDRREQ;
863 daddr = inet->inet_daddr;
864 dport = inet->inet_dport;
865 /* Open fast path for connected socket.
866 Route will not be used, if at least one option is set.
868 connected = 1;
870 ipc.addr = inet->inet_saddr;
872 ipc.oif = sk->sk_bound_dev_if;
873 err = sock_tx_timestamp(sk, &ipc.tx_flags);
874 if (err)
875 return err;
876 if (msg->msg_controllen) {
877 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
878 if (err)
879 return err;
880 if (ipc.opt)
881 free = 1;
882 connected = 0;
884 if (!ipc.opt) {
885 struct ip_options_rcu *inet_opt;
887 rcu_read_lock();
888 inet_opt = rcu_dereference(inet->inet_opt);
889 if (inet_opt) {
890 memcpy(&opt_copy, inet_opt,
891 sizeof(*inet_opt) + inet_opt->opt.optlen);
892 ipc.opt = &opt_copy.opt;
894 rcu_read_unlock();
897 saddr = ipc.addr;
898 ipc.addr = faddr = daddr;
900 if (ipc.opt && ipc.opt->opt.srr) {
901 if (!daddr)
902 return -EINVAL;
903 faddr = ipc.opt->opt.faddr;
904 connected = 0;
906 tos = RT_TOS(inet->tos);
907 if (sock_flag(sk, SOCK_LOCALROUTE) ||
908 (msg->msg_flags & MSG_DONTROUTE) ||
909 (ipc.opt && ipc.opt->opt.is_strictroute)) {
910 tos |= RTO_ONLINK;
911 connected = 0;
914 if (ipv4_is_multicast(daddr)) {
915 if (!ipc.oif)
916 ipc.oif = inet->mc_index;
917 if (!saddr)
918 saddr = inet->mc_addr;
919 connected = 0;
922 if (connected)
923 rt = (struct rtable *)sk_dst_check(sk, 0);
925 if (rt == NULL) {
926 struct net *net = sock_net(sk);
928 fl4 = &fl4_stack;
929 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
930 RT_SCOPE_UNIVERSE, sk->sk_protocol,
931 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
932 faddr, saddr, dport, inet->inet_sport);
934 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
935 rt = ip_route_output_flow(net, fl4, sk);
936 if (IS_ERR(rt)) {
937 err = PTR_ERR(rt);
938 rt = NULL;
939 if (err == -ENETUNREACH)
940 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
941 goto out;
944 err = -EACCES;
945 if ((rt->rt_flags & RTCF_BROADCAST) &&
946 !sock_flag(sk, SOCK_BROADCAST))
947 goto out;
948 if (connected)
949 sk_dst_set(sk, dst_clone(&rt->dst));
952 if (msg->msg_flags&MSG_CONFIRM)
953 goto do_confirm;
954 back_from_confirm:
956 saddr = fl4->saddr;
957 if (!ipc.addr)
958 daddr = ipc.addr = fl4->daddr;
960 /* Lockless fast path for the non-corking case. */
961 if (!corkreq) {
962 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
963 sizeof(struct udphdr), &ipc, &rt,
964 msg->msg_flags);
965 err = PTR_ERR(skb);
966 if (skb && !IS_ERR(skb))
967 err = udp_send_skb(skb, fl4);
968 goto out;
971 lock_sock(sk);
972 if (unlikely(up->pending)) {
973 /* The socket is already corked while preparing it. */
974 /* ... which is an evident application bug. --ANK */
975 release_sock(sk);
977 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
978 err = -EINVAL;
979 goto out;
982 * Now cork the socket to pend data.
984 fl4 = &inet->cork.fl.u.ip4;
985 fl4->daddr = daddr;
986 fl4->saddr = saddr;
987 fl4->fl4_dport = dport;
988 fl4->fl4_sport = inet->inet_sport;
989 up->pending = AF_INET;
991 do_append_data:
992 up->len += ulen;
993 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
994 sizeof(struct udphdr), &ipc, &rt,
995 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
996 if (err)
997 udp_flush_pending_frames(sk);
998 else if (!corkreq)
999 err = udp_push_pending_frames(sk);
1000 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1001 up->pending = 0;
1002 release_sock(sk);
1004 out:
1005 ip_rt_put(rt);
1006 if (free)
1007 kfree(ipc.opt);
1008 if (!err)
1009 return len;
1011 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1012 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1013 * we don't have a good statistic (IpOutDiscards but it can be too many
1014 * things). We could add another new stat but at least for now that
1015 * seems like overkill.
1017 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1018 UDP_INC_STATS_USER(sock_net(sk),
1019 UDP_MIB_SNDBUFERRORS, is_udplite);
1021 return err;
1023 do_confirm:
1024 dst_confirm(&rt->dst);
1025 if (!(msg->msg_flags&MSG_PROBE) || len)
1026 goto back_from_confirm;
1027 err = 0;
1028 goto out;
1030 EXPORT_SYMBOL(udp_sendmsg);
1032 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1033 size_t size, int flags)
1035 struct inet_sock *inet = inet_sk(sk);
1036 struct udp_sock *up = udp_sk(sk);
1037 int ret;
1039 if (!up->pending) {
1040 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1042 /* Call udp_sendmsg to specify destination address which
1043 * sendpage interface can't pass.
1044 * This will succeed only when the socket is connected.
1046 ret = udp_sendmsg(NULL, sk, &msg, 0);
1047 if (ret < 0)
1048 return ret;
1051 lock_sock(sk);
1053 if (unlikely(!up->pending)) {
1054 release_sock(sk);
1056 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
1057 return -EINVAL;
1060 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1061 page, offset, size, flags);
1062 if (ret == -EOPNOTSUPP) {
1063 release_sock(sk);
1064 return sock_no_sendpage(sk->sk_socket, page, offset,
1065 size, flags);
1067 if (ret < 0) {
1068 udp_flush_pending_frames(sk);
1069 goto out;
1072 up->len += size;
1073 if (!(up->corkflag || (flags&MSG_MORE)))
1074 ret = udp_push_pending_frames(sk);
1075 if (!ret)
1076 ret = size;
1077 out:
1078 release_sock(sk);
1079 return ret;
1084 * first_packet_length - return length of first packet in receive queue
1085 * @sk: socket
1087 * Drops all bad checksum frames, until a valid one is found.
1088 * Returns the length of found skb, or 0 if none is found.
1090 static unsigned int first_packet_length(struct sock *sk)
1092 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1093 struct sk_buff *skb;
1094 unsigned int res;
1096 __skb_queue_head_init(&list_kill);
1098 spin_lock_bh(&rcvq->lock);
1099 while ((skb = skb_peek(rcvq)) != NULL &&
1100 udp_lib_checksum_complete(skb)) {
1101 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1102 IS_UDPLITE(sk));
1103 atomic_inc(&sk->sk_drops);
1104 __skb_unlink(skb, rcvq);
1105 __skb_queue_tail(&list_kill, skb);
1107 res = skb ? skb->len : 0;
1108 spin_unlock_bh(&rcvq->lock);
1110 if (!skb_queue_empty(&list_kill)) {
1111 bool slow = lock_sock_fast(sk);
1113 __skb_queue_purge(&list_kill);
1114 sk_mem_reclaim_partial(sk);
1115 unlock_sock_fast(sk, slow);
1117 return res;
1121 * IOCTL requests applicable to the UDP protocol
1124 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1126 switch (cmd) {
1127 case SIOCOUTQ:
1129 int amount = sk_wmem_alloc_get(sk);
1131 return put_user(amount, (int __user *)arg);
1134 case SIOCINQ:
1136 unsigned int amount = first_packet_length(sk);
1138 if (amount)
1140 * We will only return the amount
1141 * of this packet since that is all
1142 * that will be read.
1144 amount -= sizeof(struct udphdr);
1146 return put_user(amount, (int __user *)arg);
1149 default:
1150 return -ENOIOCTLCMD;
1153 return 0;
1155 EXPORT_SYMBOL(udp_ioctl);
1158 * This should be easy, if there is something there we
1159 * return it, otherwise we block.
1162 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1163 size_t len, int noblock, int flags, int *addr_len)
1165 struct inet_sock *inet = inet_sk(sk);
1166 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1167 struct sk_buff *skb;
1168 unsigned int ulen, copied;
1169 int peeked;
1170 int err;
1171 int is_udplite = IS_UDPLITE(sk);
1172 bool slow;
1175 * Check any passed addresses
1177 if (addr_len)
1178 *addr_len = sizeof(*sin);
1180 if (flags & MSG_ERRQUEUE)
1181 return ip_recv_error(sk, msg, len);
1183 try_again:
1184 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1185 &peeked, &err);
1186 if (!skb)
1187 goto out;
1189 ulen = skb->len - sizeof(struct udphdr);
1190 copied = len;
1191 if (copied > ulen)
1192 copied = ulen;
1193 else if (copied < ulen)
1194 msg->msg_flags |= MSG_TRUNC;
1197 * If checksum is needed at all, try to do it while copying the
1198 * data. If the data is truncated, or if we only want a partial
1199 * coverage checksum (UDP-Lite), do it before the copy.
1202 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1203 if (udp_lib_checksum_complete(skb))
1204 goto csum_copy_err;
1207 if (skb_csum_unnecessary(skb))
1208 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1209 msg->msg_iov, copied);
1210 else {
1211 err = skb_copy_and_csum_datagram_iovec(skb,
1212 sizeof(struct udphdr),
1213 msg->msg_iov);
1215 if (err == -EINVAL)
1216 goto csum_copy_err;
1219 if (err)
1220 goto out_free;
1222 if (!peeked)
1223 UDP_INC_STATS_USER(sock_net(sk),
1224 UDP_MIB_INDATAGRAMS, is_udplite);
1226 sock_recv_ts_and_drops(msg, sk, skb);
1228 /* Copy the address. */
1229 if (sin) {
1230 sin->sin_family = AF_INET;
1231 sin->sin_port = udp_hdr(skb)->source;
1232 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1233 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1235 if (inet->cmsg_flags)
1236 ip_cmsg_recv(msg, skb);
1238 err = copied;
1239 if (flags & MSG_TRUNC)
1240 err = ulen;
1242 out_free:
1243 skb_free_datagram_locked(sk, skb);
1244 out:
1245 return err;
1247 csum_copy_err:
1248 slow = lock_sock_fast(sk);
1249 if (!skb_kill_datagram(sk, skb, flags))
1250 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1251 unlock_sock_fast(sk, slow);
1253 if (noblock)
1254 return -EAGAIN;
1256 /* starting over for a new packet */
1257 msg->msg_flags &= ~MSG_TRUNC;
1258 goto try_again;
1262 int udp_disconnect(struct sock *sk, int flags)
1264 struct inet_sock *inet = inet_sk(sk);
1266 * 1003.1g - break association.
1269 sk->sk_state = TCP_CLOSE;
1270 inet->inet_daddr = 0;
1271 inet->inet_dport = 0;
1272 sock_rps_reset_rxhash(sk);
1273 sk->sk_bound_dev_if = 0;
1274 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1275 inet_reset_saddr(sk);
1277 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1278 sk->sk_prot->unhash(sk);
1279 inet->inet_sport = 0;
1281 sk_dst_reset(sk);
1282 return 0;
1284 EXPORT_SYMBOL(udp_disconnect);
1286 void udp_lib_unhash(struct sock *sk)
1288 if (sk_hashed(sk)) {
1289 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1290 struct udp_hslot *hslot, *hslot2;
1292 hslot = udp_hashslot(udptable, sock_net(sk),
1293 udp_sk(sk)->udp_port_hash);
1294 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1296 spin_lock_bh(&hslot->lock);
1297 if (sk_nulls_del_node_init_rcu(sk)) {
1298 hslot->count--;
1299 inet_sk(sk)->inet_num = 0;
1300 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1302 spin_lock(&hslot2->lock);
1303 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1304 hslot2->count--;
1305 spin_unlock(&hslot2->lock);
1307 spin_unlock_bh(&hslot->lock);
1310 EXPORT_SYMBOL(udp_lib_unhash);
1313 * inet_rcv_saddr was changed, we must rehash secondary hash
1315 void udp_lib_rehash(struct sock *sk, u16 newhash)
1317 if (sk_hashed(sk)) {
1318 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1319 struct udp_hslot *hslot, *hslot2, *nhslot2;
1321 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1322 nhslot2 = udp_hashslot2(udptable, newhash);
1323 udp_sk(sk)->udp_portaddr_hash = newhash;
1324 if (hslot2 != nhslot2) {
1325 hslot = udp_hashslot(udptable, sock_net(sk),
1326 udp_sk(sk)->udp_port_hash);
1327 /* we must lock primary chain too */
1328 spin_lock_bh(&hslot->lock);
1330 spin_lock(&hslot2->lock);
1331 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1332 hslot2->count--;
1333 spin_unlock(&hslot2->lock);
1335 spin_lock(&nhslot2->lock);
1336 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1337 &nhslot2->head);
1338 nhslot2->count++;
1339 spin_unlock(&nhslot2->lock);
1341 spin_unlock_bh(&hslot->lock);
1345 EXPORT_SYMBOL(udp_lib_rehash);
1347 static void udp_v4_rehash(struct sock *sk)
1349 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1350 inet_sk(sk)->inet_rcv_saddr,
1351 inet_sk(sk)->inet_num);
1352 udp_lib_rehash(sk, new_hash);
1355 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1357 int rc;
1359 if (inet_sk(sk)->inet_daddr)
1360 sock_rps_save_rxhash(sk, skb);
1362 rc = sock_queue_rcv_skb(sk, skb);
1363 if (rc < 0) {
1364 int is_udplite = IS_UDPLITE(sk);
1366 /* Note that an ENOMEM error is charged twice */
1367 if (rc == -ENOMEM)
1368 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1369 is_udplite);
1370 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1371 kfree_skb(skb);
1372 trace_udp_fail_queue_rcv_skb(rc, sk);
1373 return -1;
1376 return 0;
1380 /* returns:
1381 * -1: error
1382 * 0: success
1383 * >0: "udp encap" protocol resubmission
1385 * Note that in the success and error cases, the skb is assumed to
1386 * have either been requeued or freed.
1388 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1390 struct udp_sock *up = udp_sk(sk);
1391 int rc;
1392 int is_udplite = IS_UDPLITE(sk);
1395 * Charge it to the socket, dropping if the queue is full.
1397 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1398 goto drop;
1399 nf_reset(skb);
1401 if (up->encap_type) {
1402 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1405 * This is an encapsulation socket so pass the skb to
1406 * the socket's udp_encap_rcv() hook. Otherwise, just
1407 * fall through and pass this up the UDP socket.
1408 * up->encap_rcv() returns the following value:
1409 * =0 if skb was successfully passed to the encap
1410 * handler or was discarded by it.
1411 * >0 if skb should be passed on to UDP.
1412 * <0 if skb should be resubmitted as proto -N
1415 /* if we're overly short, let UDP handle it */
1416 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1417 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1418 int ret;
1420 ret = encap_rcv(sk, skb);
1421 if (ret <= 0) {
1422 UDP_INC_STATS_BH(sock_net(sk),
1423 UDP_MIB_INDATAGRAMS,
1424 is_udplite);
1425 return -ret;
1429 /* FALLTHROUGH -- it's a UDP Packet */
1433 * UDP-Lite specific tests, ignored on UDP sockets
1435 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1438 * MIB statistics other than incrementing the error count are
1439 * disabled for the following two types of errors: these depend
1440 * on the application settings, not on the functioning of the
1441 * protocol stack as such.
1443 * RFC 3828 here recommends (sec 3.3): "There should also be a
1444 * way ... to ... at least let the receiving application block
1445 * delivery of packets with coverage values less than a value
1446 * provided by the application."
1448 if (up->pcrlen == 0) { /* full coverage was set */
1449 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1450 "%d while full coverage %d requested\n",
1451 UDP_SKB_CB(skb)->cscov, skb->len);
1452 goto drop;
1454 /* The next case involves violating the min. coverage requested
1455 * by the receiver. This is subtle: if receiver wants x and x is
1456 * greater than the buffersize/MTU then receiver will complain
1457 * that it wants x while sender emits packets of smaller size y.
1458 * Therefore the above ...()->partial_cov statement is essential.
1460 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1461 LIMIT_NETDEBUG(KERN_WARNING
1462 "UDPLITE: coverage %d too small, need min %d\n",
1463 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1464 goto drop;
1468 if (rcu_access_pointer(sk->sk_filter) &&
1469 udp_lib_checksum_complete(skb))
1470 goto drop;
1473 if (sk_rcvqueues_full(sk, skb))
1474 goto drop;
1476 rc = 0;
1478 ipv4_pktinfo_prepare(skb);
1479 bh_lock_sock(sk);
1480 if (!sock_owned_by_user(sk))
1481 rc = __udp_queue_rcv_skb(sk, skb);
1482 else if (sk_add_backlog(sk, skb)) {
1483 bh_unlock_sock(sk);
1484 goto drop;
1486 bh_unlock_sock(sk);
1488 return rc;
1490 drop:
1491 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1492 atomic_inc(&sk->sk_drops);
1493 kfree_skb(skb);
1494 return -1;
1498 static void flush_stack(struct sock **stack, unsigned int count,
1499 struct sk_buff *skb, unsigned int final)
1501 unsigned int i;
1502 struct sk_buff *skb1 = NULL;
1503 struct sock *sk;
1505 for (i = 0; i < count; i++) {
1506 sk = stack[i];
1507 if (likely(skb1 == NULL))
1508 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1510 if (!skb1) {
1511 atomic_inc(&sk->sk_drops);
1512 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1513 IS_UDPLITE(sk));
1514 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1515 IS_UDPLITE(sk));
1518 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1519 skb1 = NULL;
1521 if (unlikely(skb1))
1522 kfree_skb(skb1);
1526 * Multicasts and broadcasts go to each listener.
1528 * Note: called only from the BH handler context.
1530 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1531 struct udphdr *uh,
1532 __be32 saddr, __be32 daddr,
1533 struct udp_table *udptable)
1535 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1536 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1537 int dif;
1538 unsigned int i, count = 0;
1540 spin_lock(&hslot->lock);
1541 sk = sk_nulls_head(&hslot->head);
1542 dif = skb->dev->ifindex;
1543 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1544 while (sk) {
1545 stack[count++] = sk;
1546 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1547 daddr, uh->source, saddr, dif);
1548 if (unlikely(count == ARRAY_SIZE(stack))) {
1549 if (!sk)
1550 break;
1551 flush_stack(stack, count, skb, ~0);
1552 count = 0;
1556 * before releasing chain lock, we must take a reference on sockets
1558 for (i = 0; i < count; i++)
1559 sock_hold(stack[i]);
1561 spin_unlock(&hslot->lock);
1564 * do the slow work with no lock held
1566 if (count) {
1567 flush_stack(stack, count, skb, count - 1);
1569 for (i = 0; i < count; i++)
1570 sock_put(stack[i]);
1571 } else {
1572 kfree_skb(skb);
1574 return 0;
1577 /* Initialize UDP checksum. If exited with zero value (success),
1578 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1579 * Otherwise, csum completion requires chacksumming packet body,
1580 * including udp header and folding it to skb->csum.
1582 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1583 int proto)
1585 const struct iphdr *iph;
1586 int err;
1588 UDP_SKB_CB(skb)->partial_cov = 0;
1589 UDP_SKB_CB(skb)->cscov = skb->len;
1591 if (proto == IPPROTO_UDPLITE) {
1592 err = udplite_checksum_init(skb, uh);
1593 if (err)
1594 return err;
1597 iph = ip_hdr(skb);
1598 if (uh->check == 0) {
1599 skb->ip_summed = CHECKSUM_UNNECESSARY;
1600 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1601 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1602 proto, skb->csum))
1603 skb->ip_summed = CHECKSUM_UNNECESSARY;
1605 if (!skb_csum_unnecessary(skb))
1606 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1607 skb->len, proto, 0);
1608 /* Probably, we should checksum udp header (it should be in cache
1609 * in any case) and data in tiny packets (< rx copybreak).
1612 return 0;
1616 * All we need to do is get the socket, and then do a checksum.
1619 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1620 int proto)
1622 struct sock *sk;
1623 struct udphdr *uh;
1624 unsigned short ulen;
1625 struct rtable *rt = skb_rtable(skb);
1626 __be32 saddr, daddr;
1627 struct net *net = dev_net(skb->dev);
1630 * Validate the packet.
1632 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1633 goto drop; /* No space for header. */
1635 uh = udp_hdr(skb);
1636 ulen = ntohs(uh->len);
1637 saddr = ip_hdr(skb)->saddr;
1638 daddr = ip_hdr(skb)->daddr;
1640 if (ulen > skb->len)
1641 goto short_packet;
1643 if (proto == IPPROTO_UDP) {
1644 /* UDP validates ulen. */
1645 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1646 goto short_packet;
1647 uh = udp_hdr(skb);
1650 if (udp4_csum_init(skb, uh, proto))
1651 goto csum_error;
1653 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1654 return __udp4_lib_mcast_deliver(net, skb, uh,
1655 saddr, daddr, udptable);
1657 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1659 if (sk != NULL) {
1660 int ret = udp_queue_rcv_skb(sk, skb);
1661 sock_put(sk);
1663 /* a return value > 0 means to resubmit the input, but
1664 * it wants the return to be -protocol, or 0
1666 if (ret > 0)
1667 return -ret;
1668 return 0;
1671 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1672 goto drop;
1673 nf_reset(skb);
1675 /* No socket. Drop packet silently, if checksum is wrong */
1676 if (udp_lib_checksum_complete(skb))
1677 goto csum_error;
1679 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1680 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1683 * Hmm. We got an UDP packet to a port to which we
1684 * don't wanna listen. Ignore it.
1686 kfree_skb(skb);
1687 return 0;
1689 short_packet:
1690 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1691 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1692 &saddr,
1693 ntohs(uh->source),
1694 ulen,
1695 skb->len,
1696 &daddr,
1697 ntohs(uh->dest));
1698 goto drop;
1700 csum_error:
1702 * RFC1122: OK. Discards the bad packet silently (as far as
1703 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1705 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1706 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1707 &saddr,
1708 ntohs(uh->source),
1709 &daddr,
1710 ntohs(uh->dest),
1711 ulen);
1712 drop:
1713 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1714 kfree_skb(skb);
1715 return 0;
1718 int udp_rcv(struct sk_buff *skb)
1720 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1723 void udp_destroy_sock(struct sock *sk)
1725 bool slow = lock_sock_fast(sk);
1726 udp_flush_pending_frames(sk);
1727 unlock_sock_fast(sk, slow);
1731 * Socket option code for UDP
1733 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1734 char __user *optval, unsigned int optlen,
1735 int (*push_pending_frames)(struct sock *))
1737 struct udp_sock *up = udp_sk(sk);
1738 int val;
1739 int err = 0;
1740 int is_udplite = IS_UDPLITE(sk);
1742 if (optlen < sizeof(int))
1743 return -EINVAL;
1745 if (get_user(val, (int __user *)optval))
1746 return -EFAULT;
1748 switch (optname) {
1749 case UDP_CORK:
1750 if (val != 0) {
1751 up->corkflag = 1;
1752 } else {
1753 up->corkflag = 0;
1754 lock_sock(sk);
1755 (*push_pending_frames)(sk);
1756 release_sock(sk);
1758 break;
1760 case UDP_ENCAP:
1761 switch (val) {
1762 case 0:
1763 case UDP_ENCAP_ESPINUDP:
1764 case UDP_ENCAP_ESPINUDP_NON_IKE:
1765 up->encap_rcv = xfrm4_udp_encap_rcv;
1766 /* FALLTHROUGH */
1767 case UDP_ENCAP_L2TPINUDP:
1768 up->encap_type = val;
1769 break;
1770 default:
1771 err = -ENOPROTOOPT;
1772 break;
1774 break;
1777 * UDP-Lite's partial checksum coverage (RFC 3828).
1779 /* The sender sets actual checksum coverage length via this option.
1780 * The case coverage > packet length is handled by send module. */
1781 case UDPLITE_SEND_CSCOV:
1782 if (!is_udplite) /* Disable the option on UDP sockets */
1783 return -ENOPROTOOPT;
1784 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1785 val = 8;
1786 else if (val > USHRT_MAX)
1787 val = USHRT_MAX;
1788 up->pcslen = val;
1789 up->pcflag |= UDPLITE_SEND_CC;
1790 break;
1792 /* The receiver specifies a minimum checksum coverage value. To make
1793 * sense, this should be set to at least 8 (as done below). If zero is
1794 * used, this again means full checksum coverage. */
1795 case UDPLITE_RECV_CSCOV:
1796 if (!is_udplite) /* Disable the option on UDP sockets */
1797 return -ENOPROTOOPT;
1798 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1799 val = 8;
1800 else if (val > USHRT_MAX)
1801 val = USHRT_MAX;
1802 up->pcrlen = val;
1803 up->pcflag |= UDPLITE_RECV_CC;
1804 break;
1806 default:
1807 err = -ENOPROTOOPT;
1808 break;
1811 return err;
1813 EXPORT_SYMBOL(udp_lib_setsockopt);
1815 int udp_setsockopt(struct sock *sk, int level, int optname,
1816 char __user *optval, unsigned int optlen)
1818 if (level == SOL_UDP || level == SOL_UDPLITE)
1819 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1820 udp_push_pending_frames);
1821 return ip_setsockopt(sk, level, optname, optval, optlen);
1824 #ifdef CONFIG_COMPAT
1825 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1826 char __user *optval, unsigned int optlen)
1828 if (level == SOL_UDP || level == SOL_UDPLITE)
1829 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1830 udp_push_pending_frames);
1831 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1833 #endif
1835 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1836 char __user *optval, int __user *optlen)
1838 struct udp_sock *up = udp_sk(sk);
1839 int val, len;
1841 if (get_user(len, optlen))
1842 return -EFAULT;
1844 len = min_t(unsigned int, len, sizeof(int));
1846 if (len < 0)
1847 return -EINVAL;
1849 switch (optname) {
1850 case UDP_CORK:
1851 val = up->corkflag;
1852 break;
1854 case UDP_ENCAP:
1855 val = up->encap_type;
1856 break;
1858 /* The following two cannot be changed on UDP sockets, the return is
1859 * always 0 (which corresponds to the full checksum coverage of UDP). */
1860 case UDPLITE_SEND_CSCOV:
1861 val = up->pcslen;
1862 break;
1864 case UDPLITE_RECV_CSCOV:
1865 val = up->pcrlen;
1866 break;
1868 default:
1869 return -ENOPROTOOPT;
1872 if (put_user(len, optlen))
1873 return -EFAULT;
1874 if (copy_to_user(optval, &val, len))
1875 return -EFAULT;
1876 return 0;
1878 EXPORT_SYMBOL(udp_lib_getsockopt);
1880 int udp_getsockopt(struct sock *sk, int level, int optname,
1881 char __user *optval, int __user *optlen)
1883 if (level == SOL_UDP || level == SOL_UDPLITE)
1884 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1885 return ip_getsockopt(sk, level, optname, optval, optlen);
1888 #ifdef CONFIG_COMPAT
1889 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1890 char __user *optval, int __user *optlen)
1892 if (level == SOL_UDP || level == SOL_UDPLITE)
1893 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1894 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1896 #endif
1898 * udp_poll - wait for a UDP event.
1899 * @file - file struct
1900 * @sock - socket
1901 * @wait - poll table
1903 * This is same as datagram poll, except for the special case of
1904 * blocking sockets. If application is using a blocking fd
1905 * and a packet with checksum error is in the queue;
1906 * then it could get return from select indicating data available
1907 * but then block when reading it. Add special case code
1908 * to work around these arguably broken applications.
1910 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1912 unsigned int mask = datagram_poll(file, sock, wait);
1913 struct sock *sk = sock->sk;
1915 /* Check for false positives due to checksum errors */
1916 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1917 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1918 mask &= ~(POLLIN | POLLRDNORM);
1920 return mask;
1923 EXPORT_SYMBOL(udp_poll);
1925 struct proto udp_prot = {
1926 .name = "UDP",
1927 .owner = THIS_MODULE,
1928 .close = udp_lib_close,
1929 .connect = ip4_datagram_connect,
1930 .disconnect = udp_disconnect,
1931 .ioctl = udp_ioctl,
1932 .destroy = udp_destroy_sock,
1933 .setsockopt = udp_setsockopt,
1934 .getsockopt = udp_getsockopt,
1935 .sendmsg = udp_sendmsg,
1936 .recvmsg = udp_recvmsg,
1937 .sendpage = udp_sendpage,
1938 .backlog_rcv = __udp_queue_rcv_skb,
1939 .hash = udp_lib_hash,
1940 .unhash = udp_lib_unhash,
1941 .rehash = udp_v4_rehash,
1942 .get_port = udp_v4_get_port,
1943 .memory_allocated = &udp_memory_allocated,
1944 .sysctl_mem = sysctl_udp_mem,
1945 .sysctl_wmem = &sysctl_udp_wmem_min,
1946 .sysctl_rmem = &sysctl_udp_rmem_min,
1947 .obj_size = sizeof(struct udp_sock),
1948 .slab_flags = SLAB_DESTROY_BY_RCU,
1949 .h.udp_table = &udp_table,
1950 #ifdef CONFIG_COMPAT
1951 .compat_setsockopt = compat_udp_setsockopt,
1952 .compat_getsockopt = compat_udp_getsockopt,
1953 #endif
1954 .clear_sk = sk_prot_clear_portaddr_nulls,
1956 EXPORT_SYMBOL(udp_prot);
1958 /* ------------------------------------------------------------------------ */
1959 #ifdef CONFIG_PROC_FS
1961 static struct sock *udp_get_first(struct seq_file *seq, int start)
1963 struct sock *sk;
1964 struct udp_iter_state *state = seq->private;
1965 struct net *net = seq_file_net(seq);
1967 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1968 ++state->bucket) {
1969 struct hlist_nulls_node *node;
1970 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1972 if (hlist_nulls_empty(&hslot->head))
1973 continue;
1975 spin_lock_bh(&hslot->lock);
1976 sk_nulls_for_each(sk, node, &hslot->head) {
1977 if (!net_eq(sock_net(sk), net))
1978 continue;
1979 if (sk->sk_family == state->family)
1980 goto found;
1982 spin_unlock_bh(&hslot->lock);
1984 sk = NULL;
1985 found:
1986 return sk;
1989 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1991 struct udp_iter_state *state = seq->private;
1992 struct net *net = seq_file_net(seq);
1994 do {
1995 sk = sk_nulls_next(sk);
1996 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1998 if (!sk) {
1999 if (state->bucket <= state->udp_table->mask)
2000 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2001 return udp_get_first(seq, state->bucket + 1);
2003 return sk;
2006 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2008 struct sock *sk = udp_get_first(seq, 0);
2010 if (sk)
2011 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2012 --pos;
2013 return pos ? NULL : sk;
2016 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2018 struct udp_iter_state *state = seq->private;
2019 state->bucket = MAX_UDP_PORTS;
2021 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2024 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2026 struct sock *sk;
2028 if (v == SEQ_START_TOKEN)
2029 sk = udp_get_idx(seq, 0);
2030 else
2031 sk = udp_get_next(seq, v);
2033 ++*pos;
2034 return sk;
2037 static void udp_seq_stop(struct seq_file *seq, void *v)
2039 struct udp_iter_state *state = seq->private;
2041 if (state->bucket <= state->udp_table->mask)
2042 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2045 int udp_seq_open(struct inode *inode, struct file *file)
2047 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
2048 struct udp_iter_state *s;
2049 int err;
2051 err = seq_open_net(inode, file, &afinfo->seq_ops,
2052 sizeof(struct udp_iter_state));
2053 if (err < 0)
2054 return err;
2056 s = ((struct seq_file *)file->private_data)->private;
2057 s->family = afinfo->family;
2058 s->udp_table = afinfo->udp_table;
2059 return err;
2061 EXPORT_SYMBOL(udp_seq_open);
2063 /* ------------------------------------------------------------------------ */
2064 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2066 struct proc_dir_entry *p;
2067 int rc = 0;
2069 afinfo->seq_ops.start = udp_seq_start;
2070 afinfo->seq_ops.next = udp_seq_next;
2071 afinfo->seq_ops.stop = udp_seq_stop;
2073 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2074 afinfo->seq_fops, afinfo);
2075 if (!p)
2076 rc = -ENOMEM;
2077 return rc;
2079 EXPORT_SYMBOL(udp_proc_register);
2081 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2083 proc_net_remove(net, afinfo->name);
2085 EXPORT_SYMBOL(udp_proc_unregister);
2087 /* ------------------------------------------------------------------------ */
2088 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2089 int bucket, int *len)
2091 struct inet_sock *inet = inet_sk(sp);
2092 __be32 dest = inet->inet_daddr;
2093 __be32 src = inet->inet_rcv_saddr;
2094 __u16 destp = ntohs(inet->inet_dport);
2095 __u16 srcp = ntohs(inet->inet_sport);
2097 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2098 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2099 bucket, src, srcp, dest, destp, sp->sk_state,
2100 sk_wmem_alloc_get(sp),
2101 sk_rmem_alloc_get(sp),
2102 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
2103 atomic_read(&sp->sk_refcnt), sp,
2104 atomic_read(&sp->sk_drops), len);
2107 int udp4_seq_show(struct seq_file *seq, void *v)
2109 if (v == SEQ_START_TOKEN)
2110 seq_printf(seq, "%-127s\n",
2111 " sl local_address rem_address st tx_queue "
2112 "rx_queue tr tm->when retrnsmt uid timeout "
2113 "inode ref pointer drops");
2114 else {
2115 struct udp_iter_state *state = seq->private;
2116 int len;
2118 udp4_format_sock(v, seq, state->bucket, &len);
2119 seq_printf(seq, "%*s\n", 127 - len, "");
2121 return 0;
2124 static const struct file_operations udp_afinfo_seq_fops = {
2125 .owner = THIS_MODULE,
2126 .open = udp_seq_open,
2127 .read = seq_read,
2128 .llseek = seq_lseek,
2129 .release = seq_release_net
2132 /* ------------------------------------------------------------------------ */
2133 static struct udp_seq_afinfo udp4_seq_afinfo = {
2134 .name = "udp",
2135 .family = AF_INET,
2136 .udp_table = &udp_table,
2137 .seq_fops = &udp_afinfo_seq_fops,
2138 .seq_ops = {
2139 .show = udp4_seq_show,
2143 static int __net_init udp4_proc_init_net(struct net *net)
2145 return udp_proc_register(net, &udp4_seq_afinfo);
2148 static void __net_exit udp4_proc_exit_net(struct net *net)
2150 udp_proc_unregister(net, &udp4_seq_afinfo);
2153 static struct pernet_operations udp4_net_ops = {
2154 .init = udp4_proc_init_net,
2155 .exit = udp4_proc_exit_net,
2158 int __init udp4_proc_init(void)
2160 return register_pernet_subsys(&udp4_net_ops);
2163 void udp4_proc_exit(void)
2165 unregister_pernet_subsys(&udp4_net_ops);
2167 #endif /* CONFIG_PROC_FS */
2169 static __initdata unsigned long uhash_entries;
2170 static int __init set_uhash_entries(char *str)
2172 if (!str)
2173 return 0;
2174 uhash_entries = simple_strtoul(str, &str, 0);
2175 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2176 uhash_entries = UDP_HTABLE_SIZE_MIN;
2177 return 1;
2179 __setup("uhash_entries=", set_uhash_entries);
2181 void __init udp_table_init(struct udp_table *table, const char *name)
2183 unsigned int i;
2185 if (!CONFIG_BASE_SMALL)
2186 table->hash = alloc_large_system_hash(name,
2187 2 * sizeof(struct udp_hslot),
2188 uhash_entries,
2189 21, /* one slot per 2 MB */
2191 &table->log,
2192 &table->mask,
2193 64 * 1024);
2195 * Make sure hash table has the minimum size
2197 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2198 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
2199 2 * sizeof(struct udp_hslot), GFP_KERNEL);
2200 if (!table->hash)
2201 panic(name);
2202 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2203 table->mask = UDP_HTABLE_SIZE_MIN - 1;
2205 table->hash2 = table->hash + (table->mask + 1);
2206 for (i = 0; i <= table->mask; i++) {
2207 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2208 table->hash[i].count = 0;
2209 spin_lock_init(&table->hash[i].lock);
2211 for (i = 0; i <= table->mask; i++) {
2212 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2213 table->hash2[i].count = 0;
2214 spin_lock_init(&table->hash2[i].lock);
2218 void __init udp_init(void)
2220 unsigned long limit;
2222 udp_table_init(&udp_table, "UDP");
2223 limit = nr_free_buffer_pages() / 8;
2224 limit = max(limit, 128UL);
2225 sysctl_udp_mem[0] = limit / 4 * 3;
2226 sysctl_udp_mem[1] = limit;
2227 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2229 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2230 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2233 int udp4_ufo_send_check(struct sk_buff *skb)
2235 const struct iphdr *iph;
2236 struct udphdr *uh;
2238 if (!pskb_may_pull(skb, sizeof(*uh)))
2239 return -EINVAL;
2241 iph = ip_hdr(skb);
2242 uh = udp_hdr(skb);
2244 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2245 IPPROTO_UDP, 0);
2246 skb->csum_start = skb_transport_header(skb) - skb->head;
2247 skb->csum_offset = offsetof(struct udphdr, check);
2248 skb->ip_summed = CHECKSUM_PARTIAL;
2249 return 0;
2252 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
2253 netdev_features_t features)
2255 struct sk_buff *segs = ERR_PTR(-EINVAL);
2256 unsigned int mss;
2257 int offset;
2258 __wsum csum;
2260 mss = skb_shinfo(skb)->gso_size;
2261 if (unlikely(skb->len <= mss))
2262 goto out;
2264 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2265 /* Packet is from an untrusted source, reset gso_segs. */
2266 int type = skb_shinfo(skb)->gso_type;
2268 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2269 !(type & (SKB_GSO_UDP))))
2270 goto out;
2272 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2274 segs = NULL;
2275 goto out;
2278 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2279 * do checksum of UDP packets sent as multiple IP fragments.
2281 offset = skb_checksum_start_offset(skb);
2282 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2283 offset += skb->csum_offset;
2284 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2285 skb->ip_summed = CHECKSUM_NONE;
2287 /* Fragment the skb. IP headers of the fragments are updated in
2288 * inet_gso_segment()
2290 segs = skb_segment(skb, features);
2291 out:
2292 return segs;