Loading...
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.
5 *
6 * The User Datagram Protocol (UDP).
7 *
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>
13 *
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.
72 *
73 *
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.
78 */
79
80#define pr_fmt(fmt) "UDP: " fmt
81
82#include <linux/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>
94#include <linux/in.h>
95#include <linux/errno.h>
96#include <linux/timer.h>
97#include <linux/mm.h>
98#include <linux/inet.h>
99#include <linux/netdevice.h>
100#include <linux/slab.h>
101#include <net/tcp_states.h>
102#include <linux/skbuff.h>
103#include <linux/proc_fs.h>
104#include <linux/seq_file.h>
105#include <net/net_namespace.h>
106#include <net/icmp.h>
107#include <net/inet_hashtables.h>
108#include <net/route.h>
109#include <net/checksum.h>
110#include <net/xfrm.h>
111#include <trace/events/udp.h>
112#include <linux/static_key.h>
113#include <trace/events/skb.h>
114#include <net/busy_poll.h>
115#include "udp_impl.h"
116#include <net/sock_reuseport.h>
117#include <net/addrconf.h>
118
119struct udp_table udp_table __read_mostly;
120EXPORT_SYMBOL(udp_table);
121
122long sysctl_udp_mem[3] __read_mostly;
123EXPORT_SYMBOL(sysctl_udp_mem);
124
125atomic_long_t udp_memory_allocated;
126EXPORT_SYMBOL(udp_memory_allocated);
127
128#define MAX_UDP_PORTS 65536
129#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
130
131/* IPCB reference means this can not be used from early demux */
132static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
133{
134#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
135 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
136 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
137 return true;
138#endif
139 return false;
140}
141
142static int udp_lib_lport_inuse(struct net *net, __u16 num,
143 const struct udp_hslot *hslot,
144 unsigned long *bitmap,
145 struct sock *sk, unsigned int log)
146{
147 struct sock *sk2;
148 kuid_t uid = sock_i_uid(sk);
149
150 sk_for_each(sk2, &hslot->head) {
151 if (net_eq(sock_net(sk2), net) &&
152 sk2 != sk &&
153 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
154 (!sk2->sk_reuse || !sk->sk_reuse) &&
155 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
156 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
157 inet_rcv_saddr_equal(sk, sk2, true)) {
158 if (sk2->sk_reuseport && sk->sk_reuseport &&
159 !rcu_access_pointer(sk->sk_reuseport_cb) &&
160 uid_eq(uid, sock_i_uid(sk2))) {
161 if (!bitmap)
162 return 0;
163 } else {
164 if (!bitmap)
165 return 1;
166 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
167 bitmap);
168 }
169 }
170 }
171 return 0;
172}
173
174/*
175 * Note: we still hold spinlock of primary hash chain, so no other writer
176 * can insert/delete a socket with local_port == num
177 */
178static int udp_lib_lport_inuse2(struct net *net, __u16 num,
179 struct udp_hslot *hslot2,
180 struct sock *sk)
181{
182 struct sock *sk2;
183 kuid_t uid = sock_i_uid(sk);
184 int res = 0;
185
186 spin_lock(&hslot2->lock);
187 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
188 if (net_eq(sock_net(sk2), net) &&
189 sk2 != sk &&
190 (udp_sk(sk2)->udp_port_hash == num) &&
191 (!sk2->sk_reuse || !sk->sk_reuse) &&
192 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
193 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
194 inet_rcv_saddr_equal(sk, sk2, true)) {
195 if (sk2->sk_reuseport && sk->sk_reuseport &&
196 !rcu_access_pointer(sk->sk_reuseport_cb) &&
197 uid_eq(uid, sock_i_uid(sk2))) {
198 res = 0;
199 } else {
200 res = 1;
201 }
202 break;
203 }
204 }
205 spin_unlock(&hslot2->lock);
206 return res;
207}
208
209static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
210{
211 struct net *net = sock_net(sk);
212 kuid_t uid = sock_i_uid(sk);
213 struct sock *sk2;
214
215 sk_for_each(sk2, &hslot->head) {
216 if (net_eq(sock_net(sk2), net) &&
217 sk2 != sk &&
218 sk2->sk_family == sk->sk_family &&
219 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
220 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
221 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
222 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
223 inet_rcv_saddr_equal(sk, sk2, false)) {
224 return reuseport_add_sock(sk, sk2);
225 }
226 }
227
228 return reuseport_alloc(sk);
229}
230
231/**
232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
233 *
234 * @sk: socket struct in question
235 * @snum: port number to look up
236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237 * with NULL address
238 */
239int udp_lib_get_port(struct sock *sk, unsigned short snum,
240 unsigned int hash2_nulladdr)
241{
242 struct udp_hslot *hslot, *hslot2;
243 struct udp_table *udptable = sk->sk_prot->h.udp_table;
244 int error = 1;
245 struct net *net = sock_net(sk);
246
247 if (!snum) {
248 int low, high, remaining;
249 unsigned int rand;
250 unsigned short first, last;
251 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
252
253 inet_get_local_port_range(net, &low, &high);
254 remaining = (high - low) + 1;
255
256 rand = prandom_u32();
257 first = reciprocal_scale(rand, remaining) + low;
258 /*
259 * force rand to be an odd multiple of UDP_HTABLE_SIZE
260 */
261 rand = (rand | 1) * (udptable->mask + 1);
262 last = first + udptable->mask + 1;
263 do {
264 hslot = udp_hashslot(udptable, net, first);
265 bitmap_zero(bitmap, PORTS_PER_CHAIN);
266 spin_lock_bh(&hslot->lock);
267 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
268 udptable->log);
269
270 snum = first;
271 /*
272 * Iterate on all possible values of snum for this hash.
273 * Using steps of an odd multiple of UDP_HTABLE_SIZE
274 * give us randomization and full range coverage.
275 */
276 do {
277 if (low <= snum && snum <= high &&
278 !test_bit(snum >> udptable->log, bitmap) &&
279 !inet_is_local_reserved_port(net, snum))
280 goto found;
281 snum += rand;
282 } while (snum != first);
283 spin_unlock_bh(&hslot->lock);
284 cond_resched();
285 } while (++first != last);
286 goto fail;
287 } else {
288 hslot = udp_hashslot(udptable, net, snum);
289 spin_lock_bh(&hslot->lock);
290 if (hslot->count > 10) {
291 int exist;
292 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
293
294 slot2 &= udptable->mask;
295 hash2_nulladdr &= udptable->mask;
296
297 hslot2 = udp_hashslot2(udptable, slot2);
298 if (hslot->count < hslot2->count)
299 goto scan_primary_hash;
300
301 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
302 if (!exist && (hash2_nulladdr != slot2)) {
303 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
304 exist = udp_lib_lport_inuse2(net, snum, hslot2,
305 sk);
306 }
307 if (exist)
308 goto fail_unlock;
309 else
310 goto found;
311 }
312scan_primary_hash:
313 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
314 goto fail_unlock;
315 }
316found:
317 inet_sk(sk)->inet_num = snum;
318 udp_sk(sk)->udp_port_hash = snum;
319 udp_sk(sk)->udp_portaddr_hash ^= snum;
320 if (sk_unhashed(sk)) {
321 if (sk->sk_reuseport &&
322 udp_reuseport_add_sock(sk, hslot)) {
323 inet_sk(sk)->inet_num = 0;
324 udp_sk(sk)->udp_port_hash = 0;
325 udp_sk(sk)->udp_portaddr_hash ^= snum;
326 goto fail_unlock;
327 }
328
329 sk_add_node_rcu(sk, &hslot->head);
330 hslot->count++;
331 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
332
333 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
334 spin_lock(&hslot2->lock);
335 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
336 sk->sk_family == AF_INET6)
337 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
338 &hslot2->head);
339 else
340 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
341 &hslot2->head);
342 hslot2->count++;
343 spin_unlock(&hslot2->lock);
344 }
345 sock_set_flag(sk, SOCK_RCU_FREE);
346 error = 0;
347fail_unlock:
348 spin_unlock_bh(&hslot->lock);
349fail:
350 return error;
351}
352EXPORT_SYMBOL(udp_lib_get_port);
353
354int udp_v4_get_port(struct sock *sk, unsigned short snum)
355{
356 unsigned int hash2_nulladdr =
357 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
358 unsigned int hash2_partial =
359 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
360
361 /* precompute partial secondary hash */
362 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
363 return udp_lib_get_port(sk, snum, hash2_nulladdr);
364}
365
366static int compute_score(struct sock *sk, struct net *net,
367 __be32 saddr, __be16 sport,
368 __be32 daddr, unsigned short hnum,
369 int dif, int sdif, bool exact_dif)
370{
371 int score;
372 struct inet_sock *inet;
373
374 if (!net_eq(sock_net(sk), net) ||
375 udp_sk(sk)->udp_port_hash != hnum ||
376 ipv6_only_sock(sk))
377 return -1;
378
379 score = (sk->sk_family == PF_INET) ? 2 : 1;
380 inet = inet_sk(sk);
381
382 if (inet->inet_rcv_saddr) {
383 if (inet->inet_rcv_saddr != daddr)
384 return -1;
385 score += 4;
386 }
387
388 if (inet->inet_daddr) {
389 if (inet->inet_daddr != saddr)
390 return -1;
391 score += 4;
392 }
393
394 if (inet->inet_dport) {
395 if (inet->inet_dport != sport)
396 return -1;
397 score += 4;
398 }
399
400 if (sk->sk_bound_dev_if || exact_dif) {
401 bool dev_match = (sk->sk_bound_dev_if == dif ||
402 sk->sk_bound_dev_if == sdif);
403
404 if (!dev_match)
405 return -1;
406 if (sk->sk_bound_dev_if)
407 score += 4;
408 }
409
410 if (sk->sk_incoming_cpu == raw_smp_processor_id())
411 score++;
412 return score;
413}
414
415static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
416 const __u16 lport, const __be32 faddr,
417 const __be16 fport)
418{
419 static u32 udp_ehash_secret __read_mostly;
420
421 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
422
423 return __inet_ehashfn(laddr, lport, faddr, fport,
424 udp_ehash_secret + net_hash_mix(net));
425}
426
427/* called with rcu_read_lock() */
428static struct sock *udp4_lib_lookup2(struct net *net,
429 __be32 saddr, __be16 sport,
430 __be32 daddr, unsigned int hnum,
431 int dif, int sdif, bool exact_dif,
432 struct udp_hslot *hslot2,
433 struct sk_buff *skb)
434{
435 struct sock *sk, *result;
436 int score, badness;
437 u32 hash = 0;
438
439 result = NULL;
440 badness = 0;
441 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
442 score = compute_score(sk, net, saddr, sport,
443 daddr, hnum, dif, sdif, exact_dif);
444 if (score > badness) {
445 if (sk->sk_reuseport) {
446 hash = udp_ehashfn(net, daddr, hnum,
447 saddr, sport);
448 result = reuseport_select_sock(sk, hash, skb,
449 sizeof(struct udphdr));
450 if (result)
451 return result;
452 }
453 badness = score;
454 result = sk;
455 }
456 }
457 return result;
458}
459
460/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
461 * harder than this. -DaveM
462 */
463struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
464 __be16 sport, __be32 daddr, __be16 dport, int dif,
465 int sdif, struct udp_table *udptable, struct sk_buff *skb)
466{
467 struct sock *sk, *result;
468 unsigned short hnum = ntohs(dport);
469 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
470 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
471 bool exact_dif = udp_lib_exact_dif_match(net, skb);
472 int score, badness;
473 u32 hash = 0;
474
475 if (hslot->count > 10) {
476 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
477 slot2 = hash2 & udptable->mask;
478 hslot2 = &udptable->hash2[slot2];
479 if (hslot->count < hslot2->count)
480 goto begin;
481
482 result = udp4_lib_lookup2(net, saddr, sport,
483 daddr, hnum, dif, sdif,
484 exact_dif, hslot2, skb);
485 if (!result) {
486 unsigned int old_slot2 = slot2;
487 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
488 slot2 = hash2 & udptable->mask;
489 /* avoid searching the same slot again. */
490 if (unlikely(slot2 == old_slot2))
491 return result;
492
493 hslot2 = &udptable->hash2[slot2];
494 if (hslot->count < hslot2->count)
495 goto begin;
496
497 result = udp4_lib_lookup2(net, saddr, sport,
498 daddr, hnum, dif, sdif,
499 exact_dif, hslot2, skb);
500 }
501 return result;
502 }
503begin:
504 result = NULL;
505 badness = 0;
506 sk_for_each_rcu(sk, &hslot->head) {
507 score = compute_score(sk, net, saddr, sport,
508 daddr, hnum, dif, sdif, exact_dif);
509 if (score > badness) {
510 if (sk->sk_reuseport) {
511 hash = udp_ehashfn(net, daddr, hnum,
512 saddr, sport);
513 result = reuseport_select_sock(sk, hash, skb,
514 sizeof(struct udphdr));
515 if (result)
516 return result;
517 }
518 result = sk;
519 badness = score;
520 }
521 }
522 return result;
523}
524EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
525
526static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
527 __be16 sport, __be16 dport,
528 struct udp_table *udptable)
529{
530 const struct iphdr *iph = ip_hdr(skb);
531
532 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
533 iph->daddr, dport, inet_iif(skb),
534 inet_sdif(skb), udptable, skb);
535}
536
537struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
538 __be16 sport, __be16 dport)
539{
540 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
541}
542EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
543
544/* Must be called under rcu_read_lock().
545 * Does increment socket refcount.
546 */
547#if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
548 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
549 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
550struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
551 __be32 daddr, __be16 dport, int dif)
552{
553 struct sock *sk;
554
555 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
556 dif, 0, &udp_table, NULL);
557 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
558 sk = NULL;
559 return sk;
560}
561EXPORT_SYMBOL_GPL(udp4_lib_lookup);
562#endif
563
564static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
565 __be16 loc_port, __be32 loc_addr,
566 __be16 rmt_port, __be32 rmt_addr,
567 int dif, int sdif, unsigned short hnum)
568{
569 struct inet_sock *inet = inet_sk(sk);
570
571 if (!net_eq(sock_net(sk), net) ||
572 udp_sk(sk)->udp_port_hash != hnum ||
573 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
574 (inet->inet_dport != rmt_port && inet->inet_dport) ||
575 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
576 ipv6_only_sock(sk) ||
577 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
578 sk->sk_bound_dev_if != sdif))
579 return false;
580 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
581 return false;
582 return true;
583}
584
585/*
586 * This routine is called by the ICMP module when it gets some
587 * sort of error condition. If err < 0 then the socket should
588 * be closed and the error returned to the user. If err > 0
589 * it's just the icmp type << 8 | icmp code.
590 * Header points to the ip header of the error packet. We move
591 * on past this. Then (as it used to claim before adjustment)
592 * header points to the first 8 bytes of the udp header. We need
593 * to find the appropriate port.
594 */
595
596void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
597{
598 struct inet_sock *inet;
599 const struct iphdr *iph = (const struct iphdr *)skb->data;
600 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
601 const int type = icmp_hdr(skb)->type;
602 const int code = icmp_hdr(skb)->code;
603 struct sock *sk;
604 int harderr;
605 int err;
606 struct net *net = dev_net(skb->dev);
607
608 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
609 iph->saddr, uh->source, skb->dev->ifindex, 0,
610 udptable, NULL);
611 if (!sk) {
612 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
613 return; /* No socket for error */
614 }
615
616 err = 0;
617 harderr = 0;
618 inet = inet_sk(sk);
619
620 switch (type) {
621 default:
622 case ICMP_TIME_EXCEEDED:
623 err = EHOSTUNREACH;
624 break;
625 case ICMP_SOURCE_QUENCH:
626 goto out;
627 case ICMP_PARAMETERPROB:
628 err = EPROTO;
629 harderr = 1;
630 break;
631 case ICMP_DEST_UNREACH:
632 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
633 ipv4_sk_update_pmtu(skb, sk, info);
634 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
635 err = EMSGSIZE;
636 harderr = 1;
637 break;
638 }
639 goto out;
640 }
641 err = EHOSTUNREACH;
642 if (code <= NR_ICMP_UNREACH) {
643 harderr = icmp_err_convert[code].fatal;
644 err = icmp_err_convert[code].errno;
645 }
646 break;
647 case ICMP_REDIRECT:
648 ipv4_sk_redirect(skb, sk);
649 goto out;
650 }
651
652 /*
653 * RFC1122: OK. Passes ICMP errors back to application, as per
654 * 4.1.3.3.
655 */
656 if (!inet->recverr) {
657 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
658 goto out;
659 } else
660 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
661
662 sk->sk_err = err;
663 sk->sk_error_report(sk);
664out:
665 return;
666}
667
668void udp_err(struct sk_buff *skb, u32 info)
669{
670 __udp4_lib_err(skb, info, &udp_table);
671}
672
673/*
674 * Throw away all pending data and cancel the corking. Socket is locked.
675 */
676void udp_flush_pending_frames(struct sock *sk)
677{
678 struct udp_sock *up = udp_sk(sk);
679
680 if (up->pending) {
681 up->len = 0;
682 up->pending = 0;
683 ip_flush_pending_frames(sk);
684 }
685}
686EXPORT_SYMBOL(udp_flush_pending_frames);
687
688/**
689 * udp4_hwcsum - handle outgoing HW checksumming
690 * @skb: sk_buff containing the filled-in UDP header
691 * (checksum field must be zeroed out)
692 * @src: source IP address
693 * @dst: destination IP address
694 */
695void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
696{
697 struct udphdr *uh = udp_hdr(skb);
698 int offset = skb_transport_offset(skb);
699 int len = skb->len - offset;
700 int hlen = len;
701 __wsum csum = 0;
702
703 if (!skb_has_frag_list(skb)) {
704 /*
705 * Only one fragment on the socket.
706 */
707 skb->csum_start = skb_transport_header(skb) - skb->head;
708 skb->csum_offset = offsetof(struct udphdr, check);
709 uh->check = ~csum_tcpudp_magic(src, dst, len,
710 IPPROTO_UDP, 0);
711 } else {
712 struct sk_buff *frags;
713
714 /*
715 * HW-checksum won't work as there are two or more
716 * fragments on the socket so that all csums of sk_buffs
717 * should be together
718 */
719 skb_walk_frags(skb, frags) {
720 csum = csum_add(csum, frags->csum);
721 hlen -= frags->len;
722 }
723
724 csum = skb_checksum(skb, offset, hlen, csum);
725 skb->ip_summed = CHECKSUM_NONE;
726
727 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
728 if (uh->check == 0)
729 uh->check = CSUM_MANGLED_0;
730 }
731}
732EXPORT_SYMBOL_GPL(udp4_hwcsum);
733
734/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
735 * for the simple case like when setting the checksum for a UDP tunnel.
736 */
737void udp_set_csum(bool nocheck, struct sk_buff *skb,
738 __be32 saddr, __be32 daddr, int len)
739{
740 struct udphdr *uh = udp_hdr(skb);
741
742 if (nocheck) {
743 uh->check = 0;
744 } else if (skb_is_gso(skb)) {
745 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
746 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
747 uh->check = 0;
748 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
749 if (uh->check == 0)
750 uh->check = CSUM_MANGLED_0;
751 } else {
752 skb->ip_summed = CHECKSUM_PARTIAL;
753 skb->csum_start = skb_transport_header(skb) - skb->head;
754 skb->csum_offset = offsetof(struct udphdr, check);
755 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
756 }
757}
758EXPORT_SYMBOL(udp_set_csum);
759
760static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
761{
762 struct sock *sk = skb->sk;
763 struct inet_sock *inet = inet_sk(sk);
764 struct udphdr *uh;
765 int err = 0;
766 int is_udplite = IS_UDPLITE(sk);
767 int offset = skb_transport_offset(skb);
768 int len = skb->len - offset;
769 __wsum csum = 0;
770
771 /*
772 * Create a UDP header
773 */
774 uh = udp_hdr(skb);
775 uh->source = inet->inet_sport;
776 uh->dest = fl4->fl4_dport;
777 uh->len = htons(len);
778 uh->check = 0;
779
780 if (is_udplite) /* UDP-Lite */
781 csum = udplite_csum(skb);
782
783 else if (sk->sk_no_check_tx) { /* UDP csum off */
784
785 skb->ip_summed = CHECKSUM_NONE;
786 goto send;
787
788 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
789
790 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
791 goto send;
792
793 } else
794 csum = udp_csum(skb);
795
796 /* add protocol-dependent pseudo-header */
797 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
798 sk->sk_protocol, csum);
799 if (uh->check == 0)
800 uh->check = CSUM_MANGLED_0;
801
802send:
803 err = ip_send_skb(sock_net(sk), skb);
804 if (err) {
805 if (err == -ENOBUFS && !inet->recverr) {
806 UDP_INC_STATS(sock_net(sk),
807 UDP_MIB_SNDBUFERRORS, is_udplite);
808 err = 0;
809 }
810 } else
811 UDP_INC_STATS(sock_net(sk),
812 UDP_MIB_OUTDATAGRAMS, is_udplite);
813 return err;
814}
815
816/*
817 * Push out all pending data as one UDP datagram. Socket is locked.
818 */
819int udp_push_pending_frames(struct sock *sk)
820{
821 struct udp_sock *up = udp_sk(sk);
822 struct inet_sock *inet = inet_sk(sk);
823 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
824 struct sk_buff *skb;
825 int err = 0;
826
827 skb = ip_finish_skb(sk, fl4);
828 if (!skb)
829 goto out;
830
831 err = udp_send_skb(skb, fl4);
832
833out:
834 up->len = 0;
835 up->pending = 0;
836 return err;
837}
838EXPORT_SYMBOL(udp_push_pending_frames);
839
840int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
841{
842 struct inet_sock *inet = inet_sk(sk);
843 struct udp_sock *up = udp_sk(sk);
844 struct flowi4 fl4_stack;
845 struct flowi4 *fl4;
846 int ulen = len;
847 struct ipcm_cookie ipc;
848 struct rtable *rt = NULL;
849 int free = 0;
850 int connected = 0;
851 __be32 daddr, faddr, saddr;
852 __be16 dport;
853 u8 tos;
854 int err, is_udplite = IS_UDPLITE(sk);
855 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
856 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
857 struct sk_buff *skb;
858 struct ip_options_data opt_copy;
859
860 if (len > 0xFFFF)
861 return -EMSGSIZE;
862
863 /*
864 * Check the flags.
865 */
866
867 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
868 return -EOPNOTSUPP;
869
870 ipc.opt = NULL;
871 ipc.tx_flags = 0;
872 ipc.ttl = 0;
873 ipc.tos = -1;
874
875 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
876
877 fl4 = &inet->cork.fl.u.ip4;
878 if (up->pending) {
879 /*
880 * There are pending frames.
881 * The socket lock must be held while it's corked.
882 */
883 lock_sock(sk);
884 if (likely(up->pending)) {
885 if (unlikely(up->pending != AF_INET)) {
886 release_sock(sk);
887 return -EINVAL;
888 }
889 goto do_append_data;
890 }
891 release_sock(sk);
892 }
893 ulen += sizeof(struct udphdr);
894
895 /*
896 * Get and verify the address.
897 */
898 if (msg->msg_name) {
899 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
900 if (msg->msg_namelen < sizeof(*usin))
901 return -EINVAL;
902 if (usin->sin_family != AF_INET) {
903 if (usin->sin_family != AF_UNSPEC)
904 return -EAFNOSUPPORT;
905 }
906
907 daddr = usin->sin_addr.s_addr;
908 dport = usin->sin_port;
909 if (dport == 0)
910 return -EINVAL;
911 } else {
912 if (sk->sk_state != TCP_ESTABLISHED)
913 return -EDESTADDRREQ;
914 daddr = inet->inet_daddr;
915 dport = inet->inet_dport;
916 /* Open fast path for connected socket.
917 Route will not be used, if at least one option is set.
918 */
919 connected = 1;
920 }
921
922 ipc.sockc.tsflags = sk->sk_tsflags;
923 ipc.addr = inet->inet_saddr;
924 ipc.oif = sk->sk_bound_dev_if;
925
926 if (msg->msg_controllen) {
927 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
928 if (unlikely(err)) {
929 kfree(ipc.opt);
930 return err;
931 }
932 if (ipc.opt)
933 free = 1;
934 connected = 0;
935 }
936 if (!ipc.opt) {
937 struct ip_options_rcu *inet_opt;
938
939 rcu_read_lock();
940 inet_opt = rcu_dereference(inet->inet_opt);
941 if (inet_opt) {
942 memcpy(&opt_copy, inet_opt,
943 sizeof(*inet_opt) + inet_opt->opt.optlen);
944 ipc.opt = &opt_copy.opt;
945 }
946 rcu_read_unlock();
947 }
948
949 saddr = ipc.addr;
950 ipc.addr = faddr = daddr;
951
952 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
953
954 if (ipc.opt && ipc.opt->opt.srr) {
955 if (!daddr) {
956 err = -EINVAL;
957 goto out_free;
958 }
959 faddr = ipc.opt->opt.faddr;
960 connected = 0;
961 }
962 tos = get_rttos(&ipc, inet);
963 if (sock_flag(sk, SOCK_LOCALROUTE) ||
964 (msg->msg_flags & MSG_DONTROUTE) ||
965 (ipc.opt && ipc.opt->opt.is_strictroute)) {
966 tos |= RTO_ONLINK;
967 connected = 0;
968 }
969
970 if (ipv4_is_multicast(daddr)) {
971 if (!ipc.oif)
972 ipc.oif = inet->mc_index;
973 if (!saddr)
974 saddr = inet->mc_addr;
975 connected = 0;
976 } else if (!ipc.oif) {
977 ipc.oif = inet->uc_index;
978 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
979 /* oif is set, packet is to local broadcast and
980 * and uc_index is set. oif is most likely set
981 * by sk_bound_dev_if. If uc_index != oif check if the
982 * oif is an L3 master and uc_index is an L3 slave.
983 * If so, we want to allow the send using the uc_index.
984 */
985 if (ipc.oif != inet->uc_index &&
986 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
987 inet->uc_index)) {
988 ipc.oif = inet->uc_index;
989 }
990 }
991
992 if (connected)
993 rt = (struct rtable *)sk_dst_check(sk, 0);
994
995 if (!rt) {
996 struct net *net = sock_net(sk);
997 __u8 flow_flags = inet_sk_flowi_flags(sk);
998
999 fl4 = &fl4_stack;
1000
1001 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1002 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1003 flow_flags,
1004 faddr, saddr, dport, inet->inet_sport,
1005 sk->sk_uid);
1006
1007 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1008 rt = ip_route_output_flow(net, fl4, sk);
1009 if (IS_ERR(rt)) {
1010 err = PTR_ERR(rt);
1011 rt = NULL;
1012 if (err == -ENETUNREACH)
1013 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1014 goto out;
1015 }
1016
1017 err = -EACCES;
1018 if ((rt->rt_flags & RTCF_BROADCAST) &&
1019 !sock_flag(sk, SOCK_BROADCAST))
1020 goto out;
1021 if (connected)
1022 sk_dst_set(sk, dst_clone(&rt->dst));
1023 }
1024
1025 if (msg->msg_flags&MSG_CONFIRM)
1026 goto do_confirm;
1027back_from_confirm:
1028
1029 saddr = fl4->saddr;
1030 if (!ipc.addr)
1031 daddr = ipc.addr = fl4->daddr;
1032
1033 /* Lockless fast path for the non-corking case. */
1034 if (!corkreq) {
1035 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1036 sizeof(struct udphdr), &ipc, &rt,
1037 msg->msg_flags);
1038 err = PTR_ERR(skb);
1039 if (!IS_ERR_OR_NULL(skb))
1040 err = udp_send_skb(skb, fl4);
1041 goto out;
1042 }
1043
1044 lock_sock(sk);
1045 if (unlikely(up->pending)) {
1046 /* The socket is already corked while preparing it. */
1047 /* ... which is an evident application bug. --ANK */
1048 release_sock(sk);
1049
1050 net_dbg_ratelimited("socket already corked\n");
1051 err = -EINVAL;
1052 goto out;
1053 }
1054 /*
1055 * Now cork the socket to pend data.
1056 */
1057 fl4 = &inet->cork.fl.u.ip4;
1058 fl4->daddr = daddr;
1059 fl4->saddr = saddr;
1060 fl4->fl4_dport = dport;
1061 fl4->fl4_sport = inet->inet_sport;
1062 up->pending = AF_INET;
1063
1064do_append_data:
1065 up->len += ulen;
1066 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1067 sizeof(struct udphdr), &ipc, &rt,
1068 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1069 if (err)
1070 udp_flush_pending_frames(sk);
1071 else if (!corkreq)
1072 err = udp_push_pending_frames(sk);
1073 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1074 up->pending = 0;
1075 release_sock(sk);
1076
1077out:
1078 ip_rt_put(rt);
1079out_free:
1080 if (free)
1081 kfree(ipc.opt);
1082 if (!err)
1083 return len;
1084 /*
1085 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1086 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1087 * we don't have a good statistic (IpOutDiscards but it can be too many
1088 * things). We could add another new stat but at least for now that
1089 * seems like overkill.
1090 */
1091 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1092 UDP_INC_STATS(sock_net(sk),
1093 UDP_MIB_SNDBUFERRORS, is_udplite);
1094 }
1095 return err;
1096
1097do_confirm:
1098 if (msg->msg_flags & MSG_PROBE)
1099 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1100 if (!(msg->msg_flags&MSG_PROBE) || len)
1101 goto back_from_confirm;
1102 err = 0;
1103 goto out;
1104}
1105EXPORT_SYMBOL(udp_sendmsg);
1106
1107int udp_sendpage(struct sock *sk, struct page *page, int offset,
1108 size_t size, int flags)
1109{
1110 struct inet_sock *inet = inet_sk(sk);
1111 struct udp_sock *up = udp_sk(sk);
1112 int ret;
1113
1114 if (flags & MSG_SENDPAGE_NOTLAST)
1115 flags |= MSG_MORE;
1116
1117 if (!up->pending) {
1118 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1119
1120 /* Call udp_sendmsg to specify destination address which
1121 * sendpage interface can't pass.
1122 * This will succeed only when the socket is connected.
1123 */
1124 ret = udp_sendmsg(sk, &msg, 0);
1125 if (ret < 0)
1126 return ret;
1127 }
1128
1129 lock_sock(sk);
1130
1131 if (unlikely(!up->pending)) {
1132 release_sock(sk);
1133
1134 net_dbg_ratelimited("cork failed\n");
1135 return -EINVAL;
1136 }
1137
1138 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1139 page, offset, size, flags);
1140 if (ret == -EOPNOTSUPP) {
1141 release_sock(sk);
1142 return sock_no_sendpage(sk->sk_socket, page, offset,
1143 size, flags);
1144 }
1145 if (ret < 0) {
1146 udp_flush_pending_frames(sk);
1147 goto out;
1148 }
1149
1150 up->len += size;
1151 if (!(up->corkflag || (flags&MSG_MORE)))
1152 ret = udp_push_pending_frames(sk);
1153 if (!ret)
1154 ret = size;
1155out:
1156 release_sock(sk);
1157 return ret;
1158}
1159
1160#define UDP_SKB_IS_STATELESS 0x80000000
1161
1162static void udp_set_dev_scratch(struct sk_buff *skb)
1163{
1164 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1165
1166 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1167 scratch->_tsize_state = skb->truesize;
1168#if BITS_PER_LONG == 64
1169 scratch->len = skb->len;
1170 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1171 scratch->is_linear = !skb_is_nonlinear(skb);
1172#endif
1173 /* all head states execept sp (dst, sk, nf) are always cleared by
1174 * udp_rcv() and we need to preserve secpath, if present, to eventually
1175 * process IP_CMSG_PASSSEC at recvmsg() time
1176 */
1177 if (likely(!skb_sec_path(skb)))
1178 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1179}
1180
1181static int udp_skb_truesize(struct sk_buff *skb)
1182{
1183 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1184}
1185
1186static bool udp_skb_has_head_state(struct sk_buff *skb)
1187{
1188 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1189}
1190
1191/* fully reclaim rmem/fwd memory allocated for skb */
1192static void udp_rmem_release(struct sock *sk, int size, int partial,
1193 bool rx_queue_lock_held)
1194{
1195 struct udp_sock *up = udp_sk(sk);
1196 struct sk_buff_head *sk_queue;
1197 int amt;
1198
1199 if (likely(partial)) {
1200 up->forward_deficit += size;
1201 size = up->forward_deficit;
1202 if (size < (sk->sk_rcvbuf >> 2))
1203 return;
1204 } else {
1205 size += up->forward_deficit;
1206 }
1207 up->forward_deficit = 0;
1208
1209 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1210 * if the called don't held it already
1211 */
1212 sk_queue = &sk->sk_receive_queue;
1213 if (!rx_queue_lock_held)
1214 spin_lock(&sk_queue->lock);
1215
1216
1217 sk->sk_forward_alloc += size;
1218 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1219 sk->sk_forward_alloc -= amt;
1220
1221 if (amt)
1222 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1223
1224 atomic_sub(size, &sk->sk_rmem_alloc);
1225
1226 /* this can save us from acquiring the rx queue lock on next receive */
1227 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1228
1229 if (!rx_queue_lock_held)
1230 spin_unlock(&sk_queue->lock);
1231}
1232
1233/* Note: called with reader_queue.lock held.
1234 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1235 * This avoids a cache line miss while receive_queue lock is held.
1236 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1237 */
1238void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1239{
1240 prefetch(&skb->data);
1241 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1242}
1243EXPORT_SYMBOL(udp_skb_destructor);
1244
1245/* as above, but the caller held the rx queue lock, too */
1246static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1247{
1248 prefetch(&skb->data);
1249 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1250}
1251
1252/* Idea of busylocks is to let producers grab an extra spinlock
1253 * to relieve pressure on the receive_queue spinlock shared by consumer.
1254 * Under flood, this means that only one producer can be in line
1255 * trying to acquire the receive_queue spinlock.
1256 * These busylock can be allocated on a per cpu manner, instead of a
1257 * per socket one (that would consume a cache line per socket)
1258 */
1259static int udp_busylocks_log __read_mostly;
1260static spinlock_t *udp_busylocks __read_mostly;
1261
1262static spinlock_t *busylock_acquire(void *ptr)
1263{
1264 spinlock_t *busy;
1265
1266 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1267 spin_lock(busy);
1268 return busy;
1269}
1270
1271static void busylock_release(spinlock_t *busy)
1272{
1273 if (busy)
1274 spin_unlock(busy);
1275}
1276
1277int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1278{
1279 struct sk_buff_head *list = &sk->sk_receive_queue;
1280 int rmem, delta, amt, err = -ENOMEM;
1281 spinlock_t *busy = NULL;
1282 int size;
1283
1284 /* try to avoid the costly atomic add/sub pair when the receive
1285 * queue is full; always allow at least a packet
1286 */
1287 rmem = atomic_read(&sk->sk_rmem_alloc);
1288 if (rmem > sk->sk_rcvbuf)
1289 goto drop;
1290
1291 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1292 * having linear skbs :
1293 * - Reduce memory overhead and thus increase receive queue capacity
1294 * - Less cache line misses at copyout() time
1295 * - Less work at consume_skb() (less alien page frag freeing)
1296 */
1297 if (rmem > (sk->sk_rcvbuf >> 1)) {
1298 skb_condense(skb);
1299
1300 busy = busylock_acquire(sk);
1301 }
1302 size = skb->truesize;
1303 udp_set_dev_scratch(skb);
1304
1305 /* we drop only if the receive buf is full and the receive
1306 * queue contains some other skb
1307 */
1308 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1309 if (rmem > (size + sk->sk_rcvbuf))
1310 goto uncharge_drop;
1311
1312 spin_lock(&list->lock);
1313 if (size >= sk->sk_forward_alloc) {
1314 amt = sk_mem_pages(size);
1315 delta = amt << SK_MEM_QUANTUM_SHIFT;
1316 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1317 err = -ENOBUFS;
1318 spin_unlock(&list->lock);
1319 goto uncharge_drop;
1320 }
1321
1322 sk->sk_forward_alloc += delta;
1323 }
1324
1325 sk->sk_forward_alloc -= size;
1326
1327 /* no need to setup a destructor, we will explicitly release the
1328 * forward allocated memory on dequeue
1329 */
1330 sock_skb_set_dropcount(sk, skb);
1331
1332 __skb_queue_tail(list, skb);
1333 spin_unlock(&list->lock);
1334
1335 if (!sock_flag(sk, SOCK_DEAD))
1336 sk->sk_data_ready(sk);
1337
1338 busylock_release(busy);
1339 return 0;
1340
1341uncharge_drop:
1342 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1343
1344drop:
1345 atomic_inc(&sk->sk_drops);
1346 busylock_release(busy);
1347 return err;
1348}
1349EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1350
1351void udp_destruct_sock(struct sock *sk)
1352{
1353 /* reclaim completely the forward allocated memory */
1354 struct udp_sock *up = udp_sk(sk);
1355 unsigned int total = 0;
1356 struct sk_buff *skb;
1357
1358 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1359 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1360 total += skb->truesize;
1361 kfree_skb(skb);
1362 }
1363 udp_rmem_release(sk, total, 0, true);
1364
1365 inet_sock_destruct(sk);
1366}
1367EXPORT_SYMBOL_GPL(udp_destruct_sock);
1368
1369int udp_init_sock(struct sock *sk)
1370{
1371 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1372 sk->sk_destruct = udp_destruct_sock;
1373 return 0;
1374}
1375EXPORT_SYMBOL_GPL(udp_init_sock);
1376
1377void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1378{
1379 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1380 bool slow = lock_sock_fast(sk);
1381
1382 sk_peek_offset_bwd(sk, len);
1383 unlock_sock_fast(sk, slow);
1384 }
1385
1386 if (!skb_unref(skb))
1387 return;
1388
1389 /* In the more common cases we cleared the head states previously,
1390 * see __udp_queue_rcv_skb().
1391 */
1392 if (unlikely(udp_skb_has_head_state(skb)))
1393 skb_release_head_state(skb);
1394 __consume_stateless_skb(skb);
1395}
1396EXPORT_SYMBOL_GPL(skb_consume_udp);
1397
1398static struct sk_buff *__first_packet_length(struct sock *sk,
1399 struct sk_buff_head *rcvq,
1400 int *total)
1401{
1402 struct sk_buff *skb;
1403
1404 while ((skb = skb_peek(rcvq)) != NULL) {
1405 if (udp_lib_checksum_complete(skb)) {
1406 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1407 IS_UDPLITE(sk));
1408 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1409 IS_UDPLITE(sk));
1410 atomic_inc(&sk->sk_drops);
1411 __skb_unlink(skb, rcvq);
1412 *total += skb->truesize;
1413 kfree_skb(skb);
1414 } else {
1415 /* the csum related bits could be changed, refresh
1416 * the scratch area
1417 */
1418 udp_set_dev_scratch(skb);
1419 break;
1420 }
1421 }
1422 return skb;
1423}
1424
1425/**
1426 * first_packet_length - return length of first packet in receive queue
1427 * @sk: socket
1428 *
1429 * Drops all bad checksum frames, until a valid one is found.
1430 * Returns the length of found skb, or -1 if none is found.
1431 */
1432static int first_packet_length(struct sock *sk)
1433{
1434 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1435 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1436 struct sk_buff *skb;
1437 int total = 0;
1438 int res;
1439
1440 spin_lock_bh(&rcvq->lock);
1441 skb = __first_packet_length(sk, rcvq, &total);
1442 if (!skb && !skb_queue_empty(sk_queue)) {
1443 spin_lock(&sk_queue->lock);
1444 skb_queue_splice_tail_init(sk_queue, rcvq);
1445 spin_unlock(&sk_queue->lock);
1446
1447 skb = __first_packet_length(sk, rcvq, &total);
1448 }
1449 res = skb ? skb->len : -1;
1450 if (total)
1451 udp_rmem_release(sk, total, 1, false);
1452 spin_unlock_bh(&rcvq->lock);
1453 return res;
1454}
1455
1456/*
1457 * IOCTL requests applicable to the UDP protocol
1458 */
1459
1460int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1461{
1462 switch (cmd) {
1463 case SIOCOUTQ:
1464 {
1465 int amount = sk_wmem_alloc_get(sk);
1466
1467 return put_user(amount, (int __user *)arg);
1468 }
1469
1470 case SIOCINQ:
1471 {
1472 int amount = max_t(int, 0, first_packet_length(sk));
1473
1474 return put_user(amount, (int __user *)arg);
1475 }
1476
1477 default:
1478 return -ENOIOCTLCMD;
1479 }
1480
1481 return 0;
1482}
1483EXPORT_SYMBOL(udp_ioctl);
1484
1485struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1486 int noblock, int *peeked, int *off, int *err)
1487{
1488 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1489 struct sk_buff_head *queue;
1490 struct sk_buff *last;
1491 long timeo;
1492 int error;
1493
1494 queue = &udp_sk(sk)->reader_queue;
1495 flags |= noblock ? MSG_DONTWAIT : 0;
1496 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1497 do {
1498 struct sk_buff *skb;
1499
1500 error = sock_error(sk);
1501 if (error)
1502 break;
1503
1504 error = -EAGAIN;
1505 *peeked = 0;
1506 do {
1507 spin_lock_bh(&queue->lock);
1508 skb = __skb_try_recv_from_queue(sk, queue, flags,
1509 udp_skb_destructor,
1510 peeked, off, err,
1511 &last);
1512 if (skb) {
1513 spin_unlock_bh(&queue->lock);
1514 return skb;
1515 }
1516
1517 if (skb_queue_empty(sk_queue)) {
1518 spin_unlock_bh(&queue->lock);
1519 goto busy_check;
1520 }
1521
1522 /* refill the reader queue and walk it again
1523 * keep both queues locked to avoid re-acquiring
1524 * the sk_receive_queue lock if fwd memory scheduling
1525 * is needed.
1526 */
1527 spin_lock(&sk_queue->lock);
1528 skb_queue_splice_tail_init(sk_queue, queue);
1529
1530 skb = __skb_try_recv_from_queue(sk, queue, flags,
1531 udp_skb_dtor_locked,
1532 peeked, off, err,
1533 &last);
1534 spin_unlock(&sk_queue->lock);
1535 spin_unlock_bh(&queue->lock);
1536 if (skb)
1537 return skb;
1538
1539busy_check:
1540 if (!sk_can_busy_loop(sk))
1541 break;
1542
1543 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1544 } while (!skb_queue_empty(sk_queue));
1545
1546 /* sk_queue is empty, reader_queue may contain peeked packets */
1547 } while (timeo &&
1548 !__skb_wait_for_more_packets(sk, &error, &timeo,
1549 (struct sk_buff *)sk_queue));
1550
1551 *err = error;
1552 return NULL;
1553}
1554EXPORT_SYMBOL_GPL(__skb_recv_udp);
1555
1556/*
1557 * This should be easy, if there is something there we
1558 * return it, otherwise we block.
1559 */
1560
1561int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1562 int flags, int *addr_len)
1563{
1564 struct inet_sock *inet = inet_sk(sk);
1565 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1566 struct sk_buff *skb;
1567 unsigned int ulen, copied;
1568 int peeked, peeking, off;
1569 int err;
1570 int is_udplite = IS_UDPLITE(sk);
1571 bool checksum_valid = false;
1572
1573 if (flags & MSG_ERRQUEUE)
1574 return ip_recv_error(sk, msg, len, addr_len);
1575
1576try_again:
1577 peeking = flags & MSG_PEEK;
1578 off = sk_peek_offset(sk, flags);
1579 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1580 if (!skb)
1581 return err;
1582
1583 ulen = udp_skb_len(skb);
1584 copied = len;
1585 if (copied > ulen - off)
1586 copied = ulen - off;
1587 else if (copied < ulen)
1588 msg->msg_flags |= MSG_TRUNC;
1589
1590 /*
1591 * If checksum is needed at all, try to do it while copying the
1592 * data. If the data is truncated, or if we only want a partial
1593 * coverage checksum (UDP-Lite), do it before the copy.
1594 */
1595
1596 if (copied < ulen || peeking ||
1597 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1598 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1599 !__udp_lib_checksum_complete(skb);
1600 if (!checksum_valid)
1601 goto csum_copy_err;
1602 }
1603
1604 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1605 if (udp_skb_is_linear(skb))
1606 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1607 else
1608 err = skb_copy_datagram_msg(skb, off, msg, copied);
1609 } else {
1610 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1611
1612 if (err == -EINVAL)
1613 goto csum_copy_err;
1614 }
1615
1616 if (unlikely(err)) {
1617 if (!peeked) {
1618 atomic_inc(&sk->sk_drops);
1619 UDP_INC_STATS(sock_net(sk),
1620 UDP_MIB_INERRORS, is_udplite);
1621 }
1622 kfree_skb(skb);
1623 return err;
1624 }
1625
1626 if (!peeked)
1627 UDP_INC_STATS(sock_net(sk),
1628 UDP_MIB_INDATAGRAMS, is_udplite);
1629
1630 sock_recv_ts_and_drops(msg, sk, skb);
1631
1632 /* Copy the address. */
1633 if (sin) {
1634 sin->sin_family = AF_INET;
1635 sin->sin_port = udp_hdr(skb)->source;
1636 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1637 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1638 *addr_len = sizeof(*sin);
1639 }
1640 if (inet->cmsg_flags)
1641 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1642
1643 err = copied;
1644 if (flags & MSG_TRUNC)
1645 err = ulen;
1646
1647 skb_consume_udp(sk, skb, peeking ? -err : err);
1648 return err;
1649
1650csum_copy_err:
1651 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1652 udp_skb_destructor)) {
1653 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1654 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1655 }
1656 kfree_skb(skb);
1657
1658 /* starting over for a new packet, but check if we need to yield */
1659 cond_resched();
1660 msg->msg_flags &= ~MSG_TRUNC;
1661 goto try_again;
1662}
1663
1664int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1665{
1666 /* This check is replicated from __ip4_datagram_connect() and
1667 * intended to prevent BPF program called below from accessing bytes
1668 * that are out of the bound specified by user in addr_len.
1669 */
1670 if (addr_len < sizeof(struct sockaddr_in))
1671 return -EINVAL;
1672
1673 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1674}
1675EXPORT_SYMBOL(udp_pre_connect);
1676
1677int __udp_disconnect(struct sock *sk, int flags)
1678{
1679 struct inet_sock *inet = inet_sk(sk);
1680 /*
1681 * 1003.1g - break association.
1682 */
1683
1684 sk->sk_state = TCP_CLOSE;
1685 inet->inet_daddr = 0;
1686 inet->inet_dport = 0;
1687 sock_rps_reset_rxhash(sk);
1688 sk->sk_bound_dev_if = 0;
1689 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1690 inet_reset_saddr(sk);
1691
1692 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1693 sk->sk_prot->unhash(sk);
1694 inet->inet_sport = 0;
1695 }
1696 sk_dst_reset(sk);
1697 return 0;
1698}
1699EXPORT_SYMBOL(__udp_disconnect);
1700
1701int udp_disconnect(struct sock *sk, int flags)
1702{
1703 lock_sock(sk);
1704 __udp_disconnect(sk, flags);
1705 release_sock(sk);
1706 return 0;
1707}
1708EXPORT_SYMBOL(udp_disconnect);
1709
1710void udp_lib_unhash(struct sock *sk)
1711{
1712 if (sk_hashed(sk)) {
1713 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1714 struct udp_hslot *hslot, *hslot2;
1715
1716 hslot = udp_hashslot(udptable, sock_net(sk),
1717 udp_sk(sk)->udp_port_hash);
1718 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1719
1720 spin_lock_bh(&hslot->lock);
1721 if (rcu_access_pointer(sk->sk_reuseport_cb))
1722 reuseport_detach_sock(sk);
1723 if (sk_del_node_init_rcu(sk)) {
1724 hslot->count--;
1725 inet_sk(sk)->inet_num = 0;
1726 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1727
1728 spin_lock(&hslot2->lock);
1729 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1730 hslot2->count--;
1731 spin_unlock(&hslot2->lock);
1732 }
1733 spin_unlock_bh(&hslot->lock);
1734 }
1735}
1736EXPORT_SYMBOL(udp_lib_unhash);
1737
1738/*
1739 * inet_rcv_saddr was changed, we must rehash secondary hash
1740 */
1741void udp_lib_rehash(struct sock *sk, u16 newhash)
1742{
1743 if (sk_hashed(sk)) {
1744 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1745 struct udp_hslot *hslot, *hslot2, *nhslot2;
1746
1747 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1748 nhslot2 = udp_hashslot2(udptable, newhash);
1749 udp_sk(sk)->udp_portaddr_hash = newhash;
1750
1751 if (hslot2 != nhslot2 ||
1752 rcu_access_pointer(sk->sk_reuseport_cb)) {
1753 hslot = udp_hashslot(udptable, sock_net(sk),
1754 udp_sk(sk)->udp_port_hash);
1755 /* we must lock primary chain too */
1756 spin_lock_bh(&hslot->lock);
1757 if (rcu_access_pointer(sk->sk_reuseport_cb))
1758 reuseport_detach_sock(sk);
1759
1760 if (hslot2 != nhslot2) {
1761 spin_lock(&hslot2->lock);
1762 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1763 hslot2->count--;
1764 spin_unlock(&hslot2->lock);
1765
1766 spin_lock(&nhslot2->lock);
1767 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1768 &nhslot2->head);
1769 nhslot2->count++;
1770 spin_unlock(&nhslot2->lock);
1771 }
1772
1773 spin_unlock_bh(&hslot->lock);
1774 }
1775 }
1776}
1777EXPORT_SYMBOL(udp_lib_rehash);
1778
1779static void udp_v4_rehash(struct sock *sk)
1780{
1781 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1782 inet_sk(sk)->inet_rcv_saddr,
1783 inet_sk(sk)->inet_num);
1784 udp_lib_rehash(sk, new_hash);
1785}
1786
1787static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1788{
1789 int rc;
1790
1791 if (inet_sk(sk)->inet_daddr) {
1792 sock_rps_save_rxhash(sk, skb);
1793 sk_mark_napi_id(sk, skb);
1794 sk_incoming_cpu_update(sk);
1795 } else {
1796 sk_mark_napi_id_once(sk, skb);
1797 }
1798
1799 rc = __udp_enqueue_schedule_skb(sk, skb);
1800 if (rc < 0) {
1801 int is_udplite = IS_UDPLITE(sk);
1802
1803 /* Note that an ENOMEM error is charged twice */
1804 if (rc == -ENOMEM)
1805 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1806 is_udplite);
1807 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1808 kfree_skb(skb);
1809 trace_udp_fail_queue_rcv_skb(rc, sk);
1810 return -1;
1811 }
1812
1813 return 0;
1814}
1815
1816static struct static_key udp_encap_needed __read_mostly;
1817void udp_encap_enable(void)
1818{
1819 static_key_enable(&udp_encap_needed);
1820}
1821EXPORT_SYMBOL(udp_encap_enable);
1822
1823/* returns:
1824 * -1: error
1825 * 0: success
1826 * >0: "udp encap" protocol resubmission
1827 *
1828 * Note that in the success and error cases, the skb is assumed to
1829 * have either been requeued or freed.
1830 */
1831static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1832{
1833 struct udp_sock *up = udp_sk(sk);
1834 int is_udplite = IS_UDPLITE(sk);
1835
1836 /*
1837 * Charge it to the socket, dropping if the queue is full.
1838 */
1839 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1840 goto drop;
1841 nf_reset(skb);
1842
1843 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1844 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1845
1846 /*
1847 * This is an encapsulation socket so pass the skb to
1848 * the socket's udp_encap_rcv() hook. Otherwise, just
1849 * fall through and pass this up the UDP socket.
1850 * up->encap_rcv() returns the following value:
1851 * =0 if skb was successfully passed to the encap
1852 * handler or was discarded by it.
1853 * >0 if skb should be passed on to UDP.
1854 * <0 if skb should be resubmitted as proto -N
1855 */
1856
1857 /* if we're overly short, let UDP handle it */
1858 encap_rcv = READ_ONCE(up->encap_rcv);
1859 if (encap_rcv) {
1860 int ret;
1861
1862 /* Verify checksum before giving to encap */
1863 if (udp_lib_checksum_complete(skb))
1864 goto csum_error;
1865
1866 ret = encap_rcv(sk, skb);
1867 if (ret <= 0) {
1868 __UDP_INC_STATS(sock_net(sk),
1869 UDP_MIB_INDATAGRAMS,
1870 is_udplite);
1871 return -ret;
1872 }
1873 }
1874
1875 /* FALLTHROUGH -- it's a UDP Packet */
1876 }
1877
1878 /*
1879 * UDP-Lite specific tests, ignored on UDP sockets
1880 */
1881 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1882
1883 /*
1884 * MIB statistics other than incrementing the error count are
1885 * disabled for the following two types of errors: these depend
1886 * on the application settings, not on the functioning of the
1887 * protocol stack as such.
1888 *
1889 * RFC 3828 here recommends (sec 3.3): "There should also be a
1890 * way ... to ... at least let the receiving application block
1891 * delivery of packets with coverage values less than a value
1892 * provided by the application."
1893 */
1894 if (up->pcrlen == 0) { /* full coverage was set */
1895 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1896 UDP_SKB_CB(skb)->cscov, skb->len);
1897 goto drop;
1898 }
1899 /* The next case involves violating the min. coverage requested
1900 * by the receiver. This is subtle: if receiver wants x and x is
1901 * greater than the buffersize/MTU then receiver will complain
1902 * that it wants x while sender emits packets of smaller size y.
1903 * Therefore the above ...()->partial_cov statement is essential.
1904 */
1905 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1906 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1907 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1908 goto drop;
1909 }
1910 }
1911
1912 prefetch(&sk->sk_rmem_alloc);
1913 if (rcu_access_pointer(sk->sk_filter) &&
1914 udp_lib_checksum_complete(skb))
1915 goto csum_error;
1916
1917 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1918 goto drop;
1919
1920 udp_csum_pull_header(skb);
1921
1922 ipv4_pktinfo_prepare(sk, skb);
1923 return __udp_queue_rcv_skb(sk, skb);
1924
1925csum_error:
1926 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1927drop:
1928 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1929 atomic_inc(&sk->sk_drops);
1930 kfree_skb(skb);
1931 return -1;
1932}
1933
1934/* For TCP sockets, sk_rx_dst is protected by socket lock
1935 * For UDP, we use xchg() to guard against concurrent changes.
1936 */
1937bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1938{
1939 struct dst_entry *old;
1940
1941 if (dst_hold_safe(dst)) {
1942 old = xchg(&sk->sk_rx_dst, dst);
1943 dst_release(old);
1944 return old != dst;
1945 }
1946 return false;
1947}
1948EXPORT_SYMBOL(udp_sk_rx_dst_set);
1949
1950/*
1951 * Multicasts and broadcasts go to each listener.
1952 *
1953 * Note: called only from the BH handler context.
1954 */
1955static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1956 struct udphdr *uh,
1957 __be32 saddr, __be32 daddr,
1958 struct udp_table *udptable,
1959 int proto)
1960{
1961 struct sock *sk, *first = NULL;
1962 unsigned short hnum = ntohs(uh->dest);
1963 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1964 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1965 unsigned int offset = offsetof(typeof(*sk), sk_node);
1966 int dif = skb->dev->ifindex;
1967 int sdif = inet_sdif(skb);
1968 struct hlist_node *node;
1969 struct sk_buff *nskb;
1970
1971 if (use_hash2) {
1972 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1973 udptable->mask;
1974 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1975start_lookup:
1976 hslot = &udptable->hash2[hash2];
1977 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1978 }
1979
1980 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1981 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1982 uh->source, saddr, dif, sdif, hnum))
1983 continue;
1984
1985 if (!first) {
1986 first = sk;
1987 continue;
1988 }
1989 nskb = skb_clone(skb, GFP_ATOMIC);
1990
1991 if (unlikely(!nskb)) {
1992 atomic_inc(&sk->sk_drops);
1993 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1994 IS_UDPLITE(sk));
1995 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1996 IS_UDPLITE(sk));
1997 continue;
1998 }
1999 if (udp_queue_rcv_skb(sk, nskb) > 0)
2000 consume_skb(nskb);
2001 }
2002
2003 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2004 if (use_hash2 && hash2 != hash2_any) {
2005 hash2 = hash2_any;
2006 goto start_lookup;
2007 }
2008
2009 if (first) {
2010 if (udp_queue_rcv_skb(first, skb) > 0)
2011 consume_skb(skb);
2012 } else {
2013 kfree_skb(skb);
2014 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2015 proto == IPPROTO_UDPLITE);
2016 }
2017 return 0;
2018}
2019
2020/* Initialize UDP checksum. If exited with zero value (success),
2021 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2022 * Otherwise, csum completion requires chacksumming packet body,
2023 * including udp header and folding it to skb->csum.
2024 */
2025static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2026 int proto)
2027{
2028 int err;
2029
2030 UDP_SKB_CB(skb)->partial_cov = 0;
2031 UDP_SKB_CB(skb)->cscov = skb->len;
2032
2033 if (proto == IPPROTO_UDPLITE) {
2034 err = udplite_checksum_init(skb, uh);
2035 if (err)
2036 return err;
2037
2038 if (UDP_SKB_CB(skb)->partial_cov) {
2039 skb->csum = inet_compute_pseudo(skb, proto);
2040 return 0;
2041 }
2042 }
2043
2044 /* Note, we are only interested in != 0 or == 0, thus the
2045 * force to int.
2046 */
2047 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2048 inet_compute_pseudo);
2049}
2050
2051/*
2052 * All we need to do is get the socket, and then do a checksum.
2053 */
2054
2055int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2056 int proto)
2057{
2058 struct sock *sk;
2059 struct udphdr *uh;
2060 unsigned short ulen;
2061 struct rtable *rt = skb_rtable(skb);
2062 __be32 saddr, daddr;
2063 struct net *net = dev_net(skb->dev);
2064
2065 /*
2066 * Validate the packet.
2067 */
2068 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2069 goto drop; /* No space for header. */
2070
2071 uh = udp_hdr(skb);
2072 ulen = ntohs(uh->len);
2073 saddr = ip_hdr(skb)->saddr;
2074 daddr = ip_hdr(skb)->daddr;
2075
2076 if (ulen > skb->len)
2077 goto short_packet;
2078
2079 if (proto == IPPROTO_UDP) {
2080 /* UDP validates ulen. */
2081 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2082 goto short_packet;
2083 uh = udp_hdr(skb);
2084 }
2085
2086 if (udp4_csum_init(skb, uh, proto))
2087 goto csum_error;
2088
2089 sk = skb_steal_sock(skb);
2090 if (sk) {
2091 struct dst_entry *dst = skb_dst(skb);
2092 int ret;
2093
2094 if (unlikely(sk->sk_rx_dst != dst))
2095 udp_sk_rx_dst_set(sk, dst);
2096
2097 ret = udp_queue_rcv_skb(sk, skb);
2098 sock_put(sk);
2099 /* a return value > 0 means to resubmit the input, but
2100 * it wants the return to be -protocol, or 0
2101 */
2102 if (ret > 0)
2103 return -ret;
2104 return 0;
2105 }
2106
2107 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2108 return __udp4_lib_mcast_deliver(net, skb, uh,
2109 saddr, daddr, udptable, proto);
2110
2111 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2112 if (sk) {
2113 int ret;
2114
2115 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2116 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2117 inet_compute_pseudo);
2118
2119 ret = udp_queue_rcv_skb(sk, skb);
2120
2121 /* a return value > 0 means to resubmit the input, but
2122 * it wants the return to be -protocol, or 0
2123 */
2124 if (ret > 0)
2125 return -ret;
2126 return 0;
2127 }
2128
2129 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2130 goto drop;
2131 nf_reset(skb);
2132
2133 /* No socket. Drop packet silently, if checksum is wrong */
2134 if (udp_lib_checksum_complete(skb))
2135 goto csum_error;
2136
2137 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2138 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2139
2140 /*
2141 * Hmm. We got an UDP packet to a port to which we
2142 * don't wanna listen. Ignore it.
2143 */
2144 kfree_skb(skb);
2145 return 0;
2146
2147short_packet:
2148 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2149 proto == IPPROTO_UDPLITE ? "Lite" : "",
2150 &saddr, ntohs(uh->source),
2151 ulen, skb->len,
2152 &daddr, ntohs(uh->dest));
2153 goto drop;
2154
2155csum_error:
2156 /*
2157 * RFC1122: OK. Discards the bad packet silently (as far as
2158 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2159 */
2160 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2161 proto == IPPROTO_UDPLITE ? "Lite" : "",
2162 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2163 ulen);
2164 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2165drop:
2166 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2167 kfree_skb(skb);
2168 return 0;
2169}
2170
2171/* We can only early demux multicast if there is a single matching socket.
2172 * If more than one socket found returns NULL
2173 */
2174static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2175 __be16 loc_port, __be32 loc_addr,
2176 __be16 rmt_port, __be32 rmt_addr,
2177 int dif, int sdif)
2178{
2179 struct sock *sk, *result;
2180 unsigned short hnum = ntohs(loc_port);
2181 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2182 struct udp_hslot *hslot = &udp_table.hash[slot];
2183
2184 /* Do not bother scanning a too big list */
2185 if (hslot->count > 10)
2186 return NULL;
2187
2188 result = NULL;
2189 sk_for_each_rcu(sk, &hslot->head) {
2190 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2191 rmt_port, rmt_addr, dif, sdif, hnum)) {
2192 if (result)
2193 return NULL;
2194 result = sk;
2195 }
2196 }
2197
2198 return result;
2199}
2200
2201/* For unicast we should only early demux connected sockets or we can
2202 * break forwarding setups. The chains here can be long so only check
2203 * if the first socket is an exact match and if not move on.
2204 */
2205static struct sock *__udp4_lib_demux_lookup(struct net *net,
2206 __be16 loc_port, __be32 loc_addr,
2207 __be16 rmt_port, __be32 rmt_addr,
2208 int dif, int sdif)
2209{
2210 unsigned short hnum = ntohs(loc_port);
2211 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2212 unsigned int slot2 = hash2 & udp_table.mask;
2213 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2214 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2215 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2216 struct sock *sk;
2217
2218 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2219 if (INET_MATCH(sk, net, acookie, rmt_addr,
2220 loc_addr, ports, dif, sdif))
2221 return sk;
2222 /* Only check first socket in chain */
2223 break;
2224 }
2225 return NULL;
2226}
2227
2228int udp_v4_early_demux(struct sk_buff *skb)
2229{
2230 struct net *net = dev_net(skb->dev);
2231 struct in_device *in_dev = NULL;
2232 const struct iphdr *iph;
2233 const struct udphdr *uh;
2234 struct sock *sk = NULL;
2235 struct dst_entry *dst;
2236 int dif = skb->dev->ifindex;
2237 int sdif = inet_sdif(skb);
2238 int ours;
2239
2240 /* validate the packet */
2241 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2242 return 0;
2243
2244 iph = ip_hdr(skb);
2245 uh = udp_hdr(skb);
2246
2247 if (skb->pkt_type == PACKET_MULTICAST) {
2248 in_dev = __in_dev_get_rcu(skb->dev);
2249
2250 if (!in_dev)
2251 return 0;
2252
2253 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2254 iph->protocol);
2255 if (!ours)
2256 return 0;
2257
2258 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2259 uh->source, iph->saddr,
2260 dif, sdif);
2261 } else if (skb->pkt_type == PACKET_HOST) {
2262 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2263 uh->source, iph->saddr, dif, sdif);
2264 }
2265
2266 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2267 return 0;
2268
2269 skb->sk = sk;
2270 skb->destructor = sock_efree;
2271 dst = READ_ONCE(sk->sk_rx_dst);
2272
2273 if (dst)
2274 dst = dst_check(dst, 0);
2275 if (dst) {
2276 u32 itag = 0;
2277
2278 /* set noref for now.
2279 * any place which wants to hold dst has to call
2280 * dst_hold_safe()
2281 */
2282 skb_dst_set_noref(skb, dst);
2283
2284 /* for unconnected multicast sockets we need to validate
2285 * the source on each packet
2286 */
2287 if (!inet_sk(sk)->inet_daddr && in_dev)
2288 return ip_mc_validate_source(skb, iph->daddr,
2289 iph->saddr, iph->tos,
2290 skb->dev, in_dev, &itag);
2291 }
2292 return 0;
2293}
2294
2295int udp_rcv(struct sk_buff *skb)
2296{
2297 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2298}
2299
2300void udp_destroy_sock(struct sock *sk)
2301{
2302 struct udp_sock *up = udp_sk(sk);
2303 bool slow = lock_sock_fast(sk);
2304 udp_flush_pending_frames(sk);
2305 unlock_sock_fast(sk, slow);
2306 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2307 void (*encap_destroy)(struct sock *sk);
2308 encap_destroy = READ_ONCE(up->encap_destroy);
2309 if (encap_destroy)
2310 encap_destroy(sk);
2311 }
2312}
2313
2314/*
2315 * Socket option code for UDP
2316 */
2317int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2318 char __user *optval, unsigned int optlen,
2319 int (*push_pending_frames)(struct sock *))
2320{
2321 struct udp_sock *up = udp_sk(sk);
2322 int val, valbool;
2323 int err = 0;
2324 int is_udplite = IS_UDPLITE(sk);
2325
2326 if (optlen < sizeof(int))
2327 return -EINVAL;
2328
2329 if (get_user(val, (int __user *)optval))
2330 return -EFAULT;
2331
2332 valbool = val ? 1 : 0;
2333
2334 switch (optname) {
2335 case UDP_CORK:
2336 if (val != 0) {
2337 up->corkflag = 1;
2338 } else {
2339 up->corkflag = 0;
2340 lock_sock(sk);
2341 push_pending_frames(sk);
2342 release_sock(sk);
2343 }
2344 break;
2345
2346 case UDP_ENCAP:
2347 switch (val) {
2348 case 0:
2349 case UDP_ENCAP_ESPINUDP:
2350 case UDP_ENCAP_ESPINUDP_NON_IKE:
2351 up->encap_rcv = xfrm4_udp_encap_rcv;
2352 /* FALLTHROUGH */
2353 case UDP_ENCAP_L2TPINUDP:
2354 up->encap_type = val;
2355 udp_encap_enable();
2356 break;
2357 default:
2358 err = -ENOPROTOOPT;
2359 break;
2360 }
2361 break;
2362
2363 case UDP_NO_CHECK6_TX:
2364 up->no_check6_tx = valbool;
2365 break;
2366
2367 case UDP_NO_CHECK6_RX:
2368 up->no_check6_rx = valbool;
2369 break;
2370
2371 /*
2372 * UDP-Lite's partial checksum coverage (RFC 3828).
2373 */
2374 /* The sender sets actual checksum coverage length via this option.
2375 * The case coverage > packet length is handled by send module. */
2376 case UDPLITE_SEND_CSCOV:
2377 if (!is_udplite) /* Disable the option on UDP sockets */
2378 return -ENOPROTOOPT;
2379 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2380 val = 8;
2381 else if (val > USHRT_MAX)
2382 val = USHRT_MAX;
2383 up->pcslen = val;
2384 up->pcflag |= UDPLITE_SEND_CC;
2385 break;
2386
2387 /* The receiver specifies a minimum checksum coverage value. To make
2388 * sense, this should be set to at least 8 (as done below). If zero is
2389 * used, this again means full checksum coverage. */
2390 case UDPLITE_RECV_CSCOV:
2391 if (!is_udplite) /* Disable the option on UDP sockets */
2392 return -ENOPROTOOPT;
2393 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2394 val = 8;
2395 else if (val > USHRT_MAX)
2396 val = USHRT_MAX;
2397 up->pcrlen = val;
2398 up->pcflag |= UDPLITE_RECV_CC;
2399 break;
2400
2401 default:
2402 err = -ENOPROTOOPT;
2403 break;
2404 }
2405
2406 return err;
2407}
2408EXPORT_SYMBOL(udp_lib_setsockopt);
2409
2410int udp_setsockopt(struct sock *sk, int level, int optname,
2411 char __user *optval, unsigned int optlen)
2412{
2413 if (level == SOL_UDP || level == SOL_UDPLITE)
2414 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2415 udp_push_pending_frames);
2416 return ip_setsockopt(sk, level, optname, optval, optlen);
2417}
2418
2419#ifdef CONFIG_COMPAT
2420int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2421 char __user *optval, unsigned int optlen)
2422{
2423 if (level == SOL_UDP || level == SOL_UDPLITE)
2424 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2425 udp_push_pending_frames);
2426 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2427}
2428#endif
2429
2430int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2431 char __user *optval, int __user *optlen)
2432{
2433 struct udp_sock *up = udp_sk(sk);
2434 int val, len;
2435
2436 if (get_user(len, optlen))
2437 return -EFAULT;
2438
2439 len = min_t(unsigned int, len, sizeof(int));
2440
2441 if (len < 0)
2442 return -EINVAL;
2443
2444 switch (optname) {
2445 case UDP_CORK:
2446 val = up->corkflag;
2447 break;
2448
2449 case UDP_ENCAP:
2450 val = up->encap_type;
2451 break;
2452
2453 case UDP_NO_CHECK6_TX:
2454 val = up->no_check6_tx;
2455 break;
2456
2457 case UDP_NO_CHECK6_RX:
2458 val = up->no_check6_rx;
2459 break;
2460
2461 /* The following two cannot be changed on UDP sockets, the return is
2462 * always 0 (which corresponds to the full checksum coverage of UDP). */
2463 case UDPLITE_SEND_CSCOV:
2464 val = up->pcslen;
2465 break;
2466
2467 case UDPLITE_RECV_CSCOV:
2468 val = up->pcrlen;
2469 break;
2470
2471 default:
2472 return -ENOPROTOOPT;
2473 }
2474
2475 if (put_user(len, optlen))
2476 return -EFAULT;
2477 if (copy_to_user(optval, &val, len))
2478 return -EFAULT;
2479 return 0;
2480}
2481EXPORT_SYMBOL(udp_lib_getsockopt);
2482
2483int udp_getsockopt(struct sock *sk, int level, int optname,
2484 char __user *optval, int __user *optlen)
2485{
2486 if (level == SOL_UDP || level == SOL_UDPLITE)
2487 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2488 return ip_getsockopt(sk, level, optname, optval, optlen);
2489}
2490
2491#ifdef CONFIG_COMPAT
2492int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2493 char __user *optval, int __user *optlen)
2494{
2495 if (level == SOL_UDP || level == SOL_UDPLITE)
2496 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2497 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2498}
2499#endif
2500/**
2501 * udp_poll - wait for a UDP event.
2502 * @file - file struct
2503 * @sock - socket
2504 * @wait - poll table
2505 *
2506 * This is same as datagram poll, except for the special case of
2507 * blocking sockets. If application is using a blocking fd
2508 * and a packet with checksum error is in the queue;
2509 * then it could get return from select indicating data available
2510 * but then block when reading it. Add special case code
2511 * to work around these arguably broken applications.
2512 */
2513__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2514{
2515 __poll_t mask = datagram_poll(file, sock, wait);
2516 struct sock *sk = sock->sk;
2517
2518 if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2519 mask |= EPOLLIN | EPOLLRDNORM;
2520
2521 /* Check for false positives due to checksum errors */
2522 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2523 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2524 mask &= ~(EPOLLIN | EPOLLRDNORM);
2525
2526 return mask;
2527
2528}
2529EXPORT_SYMBOL(udp_poll);
2530
2531int udp_abort(struct sock *sk, int err)
2532{
2533 lock_sock(sk);
2534
2535 sk->sk_err = err;
2536 sk->sk_error_report(sk);
2537 __udp_disconnect(sk, 0);
2538
2539 release_sock(sk);
2540
2541 return 0;
2542}
2543EXPORT_SYMBOL_GPL(udp_abort);
2544
2545struct proto udp_prot = {
2546 .name = "UDP",
2547 .owner = THIS_MODULE,
2548 .close = udp_lib_close,
2549 .pre_connect = udp_pre_connect,
2550 .connect = ip4_datagram_connect,
2551 .disconnect = udp_disconnect,
2552 .ioctl = udp_ioctl,
2553 .init = udp_init_sock,
2554 .destroy = udp_destroy_sock,
2555 .setsockopt = udp_setsockopt,
2556 .getsockopt = udp_getsockopt,
2557 .sendmsg = udp_sendmsg,
2558 .recvmsg = udp_recvmsg,
2559 .sendpage = udp_sendpage,
2560 .release_cb = ip4_datagram_release_cb,
2561 .hash = udp_lib_hash,
2562 .unhash = udp_lib_unhash,
2563 .rehash = udp_v4_rehash,
2564 .get_port = udp_v4_get_port,
2565 .memory_allocated = &udp_memory_allocated,
2566 .sysctl_mem = sysctl_udp_mem,
2567 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2568 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2569 .obj_size = sizeof(struct udp_sock),
2570 .h.udp_table = &udp_table,
2571#ifdef CONFIG_COMPAT
2572 .compat_setsockopt = compat_udp_setsockopt,
2573 .compat_getsockopt = compat_udp_getsockopt,
2574#endif
2575 .diag_destroy = udp_abort,
2576};
2577EXPORT_SYMBOL(udp_prot);
2578
2579/* ------------------------------------------------------------------------ */
2580#ifdef CONFIG_PROC_FS
2581
2582static struct sock *udp_get_first(struct seq_file *seq, int start)
2583{
2584 struct sock *sk;
2585 struct udp_iter_state *state = seq->private;
2586 struct net *net = seq_file_net(seq);
2587
2588 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2589 ++state->bucket) {
2590 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2591
2592 if (hlist_empty(&hslot->head))
2593 continue;
2594
2595 spin_lock_bh(&hslot->lock);
2596 sk_for_each(sk, &hslot->head) {
2597 if (!net_eq(sock_net(sk), net))
2598 continue;
2599 if (sk->sk_family == state->family)
2600 goto found;
2601 }
2602 spin_unlock_bh(&hslot->lock);
2603 }
2604 sk = NULL;
2605found:
2606 return sk;
2607}
2608
2609static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2610{
2611 struct udp_iter_state *state = seq->private;
2612 struct net *net = seq_file_net(seq);
2613
2614 do {
2615 sk = sk_next(sk);
2616 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2617
2618 if (!sk) {
2619 if (state->bucket <= state->udp_table->mask)
2620 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2621 return udp_get_first(seq, state->bucket + 1);
2622 }
2623 return sk;
2624}
2625
2626static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2627{
2628 struct sock *sk = udp_get_first(seq, 0);
2629
2630 if (sk)
2631 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2632 --pos;
2633 return pos ? NULL : sk;
2634}
2635
2636static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2637{
2638 struct udp_iter_state *state = seq->private;
2639 state->bucket = MAX_UDP_PORTS;
2640
2641 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2642}
2643
2644static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2645{
2646 struct sock *sk;
2647
2648 if (v == SEQ_START_TOKEN)
2649 sk = udp_get_idx(seq, 0);
2650 else
2651 sk = udp_get_next(seq, v);
2652
2653 ++*pos;
2654 return sk;
2655}
2656
2657static void udp_seq_stop(struct seq_file *seq, void *v)
2658{
2659 struct udp_iter_state *state = seq->private;
2660
2661 if (state->bucket <= state->udp_table->mask)
2662 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2663}
2664
2665int udp_seq_open(struct inode *inode, struct file *file)
2666{
2667 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2668 struct udp_iter_state *s;
2669 int err;
2670
2671 err = seq_open_net(inode, file, &afinfo->seq_ops,
2672 sizeof(struct udp_iter_state));
2673 if (err < 0)
2674 return err;
2675
2676 s = ((struct seq_file *)file->private_data)->private;
2677 s->family = afinfo->family;
2678 s->udp_table = afinfo->udp_table;
2679 return err;
2680}
2681EXPORT_SYMBOL(udp_seq_open);
2682
2683/* ------------------------------------------------------------------------ */
2684int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2685{
2686 struct proc_dir_entry *p;
2687 int rc = 0;
2688
2689 afinfo->seq_ops.start = udp_seq_start;
2690 afinfo->seq_ops.next = udp_seq_next;
2691 afinfo->seq_ops.stop = udp_seq_stop;
2692
2693 p = proc_create_data(afinfo->name, 0444, net->proc_net,
2694 afinfo->seq_fops, afinfo);
2695 if (!p)
2696 rc = -ENOMEM;
2697 return rc;
2698}
2699EXPORT_SYMBOL(udp_proc_register);
2700
2701void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2702{
2703 remove_proc_entry(afinfo->name, net->proc_net);
2704}
2705EXPORT_SYMBOL(udp_proc_unregister);
2706
2707/* ------------------------------------------------------------------------ */
2708static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2709 int bucket)
2710{
2711 struct inet_sock *inet = inet_sk(sp);
2712 __be32 dest = inet->inet_daddr;
2713 __be32 src = inet->inet_rcv_saddr;
2714 __u16 destp = ntohs(inet->inet_dport);
2715 __u16 srcp = ntohs(inet->inet_sport);
2716
2717 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2718 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2719 bucket, src, srcp, dest, destp, sp->sk_state,
2720 sk_wmem_alloc_get(sp),
2721 sk_rmem_alloc_get(sp),
2722 0, 0L, 0,
2723 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2724 0, sock_i_ino(sp),
2725 refcount_read(&sp->sk_refcnt), sp,
2726 atomic_read(&sp->sk_drops));
2727}
2728
2729int udp4_seq_show(struct seq_file *seq, void *v)
2730{
2731 seq_setwidth(seq, 127);
2732 if (v == SEQ_START_TOKEN)
2733 seq_puts(seq, " sl local_address rem_address st tx_queue "
2734 "rx_queue tr tm->when retrnsmt uid timeout "
2735 "inode ref pointer drops");
2736 else {
2737 struct udp_iter_state *state = seq->private;
2738
2739 udp4_format_sock(v, seq, state->bucket);
2740 }
2741 seq_pad(seq, '\n');
2742 return 0;
2743}
2744
2745static const struct file_operations udp_afinfo_seq_fops = {
2746 .open = udp_seq_open,
2747 .read = seq_read,
2748 .llseek = seq_lseek,
2749 .release = seq_release_net
2750};
2751
2752/* ------------------------------------------------------------------------ */
2753static struct udp_seq_afinfo udp4_seq_afinfo = {
2754 .name = "udp",
2755 .family = AF_INET,
2756 .udp_table = &udp_table,
2757 .seq_fops = &udp_afinfo_seq_fops,
2758 .seq_ops = {
2759 .show = udp4_seq_show,
2760 },
2761};
2762
2763static int __net_init udp4_proc_init_net(struct net *net)
2764{
2765 return udp_proc_register(net, &udp4_seq_afinfo);
2766}
2767
2768static void __net_exit udp4_proc_exit_net(struct net *net)
2769{
2770 udp_proc_unregister(net, &udp4_seq_afinfo);
2771}
2772
2773static struct pernet_operations udp4_net_ops = {
2774 .init = udp4_proc_init_net,
2775 .exit = udp4_proc_exit_net,
2776};
2777
2778int __init udp4_proc_init(void)
2779{
2780 return register_pernet_subsys(&udp4_net_ops);
2781}
2782
2783void udp4_proc_exit(void)
2784{
2785 unregister_pernet_subsys(&udp4_net_ops);
2786}
2787#endif /* CONFIG_PROC_FS */
2788
2789static __initdata unsigned long uhash_entries;
2790static int __init set_uhash_entries(char *str)
2791{
2792 ssize_t ret;
2793
2794 if (!str)
2795 return 0;
2796
2797 ret = kstrtoul(str, 0, &uhash_entries);
2798 if (ret)
2799 return 0;
2800
2801 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2802 uhash_entries = UDP_HTABLE_SIZE_MIN;
2803 return 1;
2804}
2805__setup("uhash_entries=", set_uhash_entries);
2806
2807void __init udp_table_init(struct udp_table *table, const char *name)
2808{
2809 unsigned int i;
2810
2811 table->hash = alloc_large_system_hash(name,
2812 2 * sizeof(struct udp_hslot),
2813 uhash_entries,
2814 21, /* one slot per 2 MB */
2815 0,
2816 &table->log,
2817 &table->mask,
2818 UDP_HTABLE_SIZE_MIN,
2819 64 * 1024);
2820
2821 table->hash2 = table->hash + (table->mask + 1);
2822 for (i = 0; i <= table->mask; i++) {
2823 INIT_HLIST_HEAD(&table->hash[i].head);
2824 table->hash[i].count = 0;
2825 spin_lock_init(&table->hash[i].lock);
2826 }
2827 for (i = 0; i <= table->mask; i++) {
2828 INIT_HLIST_HEAD(&table->hash2[i].head);
2829 table->hash2[i].count = 0;
2830 spin_lock_init(&table->hash2[i].lock);
2831 }
2832}
2833
2834u32 udp_flow_hashrnd(void)
2835{
2836 static u32 hashrnd __read_mostly;
2837
2838 net_get_random_once(&hashrnd, sizeof(hashrnd));
2839
2840 return hashrnd;
2841}
2842EXPORT_SYMBOL(udp_flow_hashrnd);
2843
2844static void __udp_sysctl_init(struct net *net)
2845{
2846 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2847 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2848
2849#ifdef CONFIG_NET_L3_MASTER_DEV
2850 net->ipv4.sysctl_udp_l3mdev_accept = 0;
2851#endif
2852}
2853
2854static int __net_init udp_sysctl_init(struct net *net)
2855{
2856 __udp_sysctl_init(net);
2857 return 0;
2858}
2859
2860static struct pernet_operations __net_initdata udp_sysctl_ops = {
2861 .init = udp_sysctl_init,
2862};
2863
2864void __init udp_init(void)
2865{
2866 unsigned long limit;
2867 unsigned int i;
2868
2869 udp_table_init(&udp_table, "UDP");
2870 limit = nr_free_buffer_pages() / 8;
2871 limit = max(limit, 128UL);
2872 sysctl_udp_mem[0] = limit / 4 * 3;
2873 sysctl_udp_mem[1] = limit;
2874 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2875
2876 __udp_sysctl_init(&init_net);
2877
2878 /* 16 spinlocks per cpu */
2879 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2880 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2881 GFP_KERNEL);
2882 if (!udp_busylocks)
2883 panic("UDP: failed to alloc udp_busylocks\n");
2884 for (i = 0; i < (1U << udp_busylocks_log); i++)
2885 spin_lock_init(udp_busylocks + i);
2886
2887 if (register_pernet_subsys(&udp_sysctl_ops))
2888 panic("UDP: failed to init sysctl parameters.\n");
2889}
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.
5 *
6 * The User Datagram Protocol (UDP).
7 *
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>
13 *
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.
72 *
73 *
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.
78 */
79
80#define pr_fmt(fmt) "UDP: " fmt
81
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/in.h>
94#include <linux/errno.h>
95#include <linux/timer.h>
96#include <linux/mm.h>
97#include <linux/inet.h>
98#include <linux/netdevice.h>
99#include <linux/slab.h>
100#include <net/tcp_states.h>
101#include <linux/skbuff.h>
102#include <linux/proc_fs.h>
103#include <linux/seq_file.h>
104#include <net/net_namespace.h>
105#include <net/icmp.h>
106#include <net/inet_hashtables.h>
107#include <net/route.h>
108#include <net/checksum.h>
109#include <net/xfrm.h>
110#include <trace/events/udp.h>
111#include <linux/static_key.h>
112#include <trace/events/skb.h>
113#include <net/busy_poll.h>
114#include "udp_impl.h"
115
116struct udp_table udp_table __read_mostly;
117EXPORT_SYMBOL(udp_table);
118
119long sysctl_udp_mem[3] __read_mostly;
120EXPORT_SYMBOL(sysctl_udp_mem);
121
122int sysctl_udp_rmem_min __read_mostly;
123EXPORT_SYMBOL(sysctl_udp_rmem_min);
124
125int sysctl_udp_wmem_min __read_mostly;
126EXPORT_SYMBOL(sysctl_udp_wmem_min);
127
128atomic_long_t udp_memory_allocated;
129EXPORT_SYMBOL(udp_memory_allocated);
130
131#define MAX_UDP_PORTS 65536
132#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
133
134static int udp_lib_lport_inuse(struct net *net, __u16 num,
135 const struct udp_hslot *hslot,
136 unsigned long *bitmap,
137 struct sock *sk,
138 int (*saddr_comp)(const struct sock *sk1,
139 const struct sock *sk2),
140 unsigned int log)
141{
142 struct sock *sk2;
143 struct hlist_nulls_node *node;
144 kuid_t uid = sock_i_uid(sk);
145
146 sk_nulls_for_each(sk2, node, &hslot->head)
147 if (net_eq(sock_net(sk2), net) &&
148 sk2 != sk &&
149 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
150 (!sk2->sk_reuse || !sk->sk_reuse) &&
151 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
152 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
153 (!sk2->sk_reuseport || !sk->sk_reuseport ||
154 !uid_eq(uid, sock_i_uid(sk2))) &&
155 (*saddr_comp)(sk, sk2)) {
156 if (bitmap)
157 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
158 bitmap);
159 else
160 return 1;
161 }
162 return 0;
163}
164
165/*
166 * Note: we still hold spinlock of primary hash chain, so no other writer
167 * can insert/delete a socket with local_port == num
168 */
169static int udp_lib_lport_inuse2(struct net *net, __u16 num,
170 struct udp_hslot *hslot2,
171 struct sock *sk,
172 int (*saddr_comp)(const struct sock *sk1,
173 const struct sock *sk2))
174{
175 struct sock *sk2;
176 struct hlist_nulls_node *node;
177 kuid_t uid = sock_i_uid(sk);
178 int res = 0;
179
180 spin_lock(&hslot2->lock);
181 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
182 if (net_eq(sock_net(sk2), net) &&
183 sk2 != sk &&
184 (udp_sk(sk2)->udp_port_hash == num) &&
185 (!sk2->sk_reuse || !sk->sk_reuse) &&
186 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
187 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
188 (!sk2->sk_reuseport || !sk->sk_reuseport ||
189 !uid_eq(uid, sock_i_uid(sk2))) &&
190 (*saddr_comp)(sk, sk2)) {
191 res = 1;
192 break;
193 }
194 spin_unlock(&hslot2->lock);
195 return res;
196}
197
198/**
199 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
200 *
201 * @sk: socket struct in question
202 * @snum: port number to look up
203 * @saddr_comp: AF-dependent comparison of bound local IP addresses
204 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
205 * with NULL address
206 */
207int udp_lib_get_port(struct sock *sk, unsigned short snum,
208 int (*saddr_comp)(const struct sock *sk1,
209 const struct sock *sk2),
210 unsigned int hash2_nulladdr)
211{
212 struct udp_hslot *hslot, *hslot2;
213 struct udp_table *udptable = sk->sk_prot->h.udp_table;
214 int error = 1;
215 struct net *net = sock_net(sk);
216
217 if (!snum) {
218 int low, high, remaining;
219 unsigned int rand;
220 unsigned short first, last;
221 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
222
223 inet_get_local_port_range(net, &low, &high);
224 remaining = (high - low) + 1;
225
226 rand = prandom_u32();
227 first = (((u64)rand * remaining) >> 32) + low;
228 /*
229 * force rand to be an odd multiple of UDP_HTABLE_SIZE
230 */
231 rand = (rand | 1) * (udptable->mask + 1);
232 last = first + udptable->mask + 1;
233 do {
234 hslot = udp_hashslot(udptable, net, first);
235 bitmap_zero(bitmap, PORTS_PER_CHAIN);
236 spin_lock_bh(&hslot->lock);
237 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
238 saddr_comp, udptable->log);
239
240 snum = first;
241 /*
242 * Iterate on all possible values of snum for this hash.
243 * Using steps of an odd multiple of UDP_HTABLE_SIZE
244 * give us randomization and full range coverage.
245 */
246 do {
247 if (low <= snum && snum <= high &&
248 !test_bit(snum >> udptable->log, bitmap) &&
249 !inet_is_reserved_local_port(snum))
250 goto found;
251 snum += rand;
252 } while (snum != first);
253 spin_unlock_bh(&hslot->lock);
254 } while (++first != last);
255 goto fail;
256 } else {
257 hslot = udp_hashslot(udptable, net, snum);
258 spin_lock_bh(&hslot->lock);
259 if (hslot->count > 10) {
260 int exist;
261 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
262
263 slot2 &= udptable->mask;
264 hash2_nulladdr &= udptable->mask;
265
266 hslot2 = udp_hashslot2(udptable, slot2);
267 if (hslot->count < hslot2->count)
268 goto scan_primary_hash;
269
270 exist = udp_lib_lport_inuse2(net, snum, hslot2,
271 sk, saddr_comp);
272 if (!exist && (hash2_nulladdr != slot2)) {
273 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
274 exist = udp_lib_lport_inuse2(net, snum, hslot2,
275 sk, saddr_comp);
276 }
277 if (exist)
278 goto fail_unlock;
279 else
280 goto found;
281 }
282scan_primary_hash:
283 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
284 saddr_comp, 0))
285 goto fail_unlock;
286 }
287found:
288 inet_sk(sk)->inet_num = snum;
289 udp_sk(sk)->udp_port_hash = snum;
290 udp_sk(sk)->udp_portaddr_hash ^= snum;
291 if (sk_unhashed(sk)) {
292 sk_nulls_add_node_rcu(sk, &hslot->head);
293 hslot->count++;
294 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
295
296 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
297 spin_lock(&hslot2->lock);
298 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
299 &hslot2->head);
300 hslot2->count++;
301 spin_unlock(&hslot2->lock);
302 }
303 error = 0;
304fail_unlock:
305 spin_unlock_bh(&hslot->lock);
306fail:
307 return error;
308}
309EXPORT_SYMBOL(udp_lib_get_port);
310
311static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
312{
313 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
314
315 return (!ipv6_only_sock(sk2) &&
316 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
317 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
318}
319
320static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
321 unsigned int port)
322{
323 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
324}
325
326int udp_v4_get_port(struct sock *sk, unsigned short snum)
327{
328 unsigned int hash2_nulladdr =
329 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
330 unsigned int hash2_partial =
331 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
332
333 /* precompute partial secondary hash */
334 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
335 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
336}
337
338static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
339 unsigned short hnum,
340 __be16 sport, __be32 daddr, __be16 dport, int dif)
341{
342 int score = -1;
343
344 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
345 !ipv6_only_sock(sk)) {
346 struct inet_sock *inet = inet_sk(sk);
347
348 score = (sk->sk_family == PF_INET ? 2 : 1);
349 if (inet->inet_rcv_saddr) {
350 if (inet->inet_rcv_saddr != daddr)
351 return -1;
352 score += 4;
353 }
354 if (inet->inet_daddr) {
355 if (inet->inet_daddr != saddr)
356 return -1;
357 score += 4;
358 }
359 if (inet->inet_dport) {
360 if (inet->inet_dport != sport)
361 return -1;
362 score += 4;
363 }
364 if (sk->sk_bound_dev_if) {
365 if (sk->sk_bound_dev_if != dif)
366 return -1;
367 score += 4;
368 }
369 }
370 return score;
371}
372
373/*
374 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
375 */
376static inline int compute_score2(struct sock *sk, struct net *net,
377 __be32 saddr, __be16 sport,
378 __be32 daddr, unsigned int hnum, int dif)
379{
380 int score = -1;
381
382 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
383 struct inet_sock *inet = inet_sk(sk);
384
385 if (inet->inet_rcv_saddr != daddr)
386 return -1;
387 if (inet->inet_num != hnum)
388 return -1;
389
390 score = (sk->sk_family == PF_INET ? 2 : 1);
391 if (inet->inet_daddr) {
392 if (inet->inet_daddr != saddr)
393 return -1;
394 score += 4;
395 }
396 if (inet->inet_dport) {
397 if (inet->inet_dport != sport)
398 return -1;
399 score += 4;
400 }
401 if (sk->sk_bound_dev_if) {
402 if (sk->sk_bound_dev_if != dif)
403 return -1;
404 score += 4;
405 }
406 }
407 return score;
408}
409
410static unsigned int udp_ehashfn(struct net *net, const __be32 laddr,
411 const __u16 lport, const __be32 faddr,
412 const __be16 fport)
413{
414 static u32 udp_ehash_secret __read_mostly;
415
416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
417
418 return __inet_ehashfn(laddr, lport, faddr, fport,
419 udp_ehash_secret + net_hash_mix(net));
420}
421
422
423/* called with read_rcu_lock() */
424static struct sock *udp4_lib_lookup2(struct net *net,
425 __be32 saddr, __be16 sport,
426 __be32 daddr, unsigned int hnum, int dif,
427 struct udp_hslot *hslot2, unsigned int slot2)
428{
429 struct sock *sk, *result;
430 struct hlist_nulls_node *node;
431 int score, badness, matches = 0, reuseport = 0;
432 u32 hash = 0;
433
434begin:
435 result = NULL;
436 badness = 0;
437 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
438 score = compute_score2(sk, net, saddr, sport,
439 daddr, hnum, dif);
440 if (score > badness) {
441 result = sk;
442 badness = score;
443 reuseport = sk->sk_reuseport;
444 if (reuseport) {
445 hash = udp_ehashfn(net, daddr, hnum,
446 saddr, sport);
447 matches = 1;
448 }
449 } else if (score == badness && reuseport) {
450 matches++;
451 if (((u64)hash * matches) >> 32 == 0)
452 result = sk;
453 hash = next_pseudo_random32(hash);
454 }
455 }
456 /*
457 * if the nulls value we got at the end of this lookup is
458 * not the expected one, we must restart lookup.
459 * We probably met an item that was moved to another chain.
460 */
461 if (get_nulls_value(node) != slot2)
462 goto begin;
463 if (result) {
464 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
465 result = NULL;
466 else if (unlikely(compute_score2(result, net, saddr, sport,
467 daddr, hnum, dif) < badness)) {
468 sock_put(result);
469 goto begin;
470 }
471 }
472 return result;
473}
474
475/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
476 * harder than this. -DaveM
477 */
478struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
479 __be16 sport, __be32 daddr, __be16 dport,
480 int dif, struct udp_table *udptable)
481{
482 struct sock *sk, *result;
483 struct hlist_nulls_node *node;
484 unsigned short hnum = ntohs(dport);
485 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
486 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
487 int score, badness, matches = 0, reuseport = 0;
488 u32 hash = 0;
489
490 rcu_read_lock();
491 if (hslot->count > 10) {
492 hash2 = udp4_portaddr_hash(net, daddr, hnum);
493 slot2 = hash2 & udptable->mask;
494 hslot2 = &udptable->hash2[slot2];
495 if (hslot->count < hslot2->count)
496 goto begin;
497
498 result = udp4_lib_lookup2(net, saddr, sport,
499 daddr, hnum, dif,
500 hslot2, slot2);
501 if (!result) {
502 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
503 slot2 = hash2 & udptable->mask;
504 hslot2 = &udptable->hash2[slot2];
505 if (hslot->count < hslot2->count)
506 goto begin;
507
508 result = udp4_lib_lookup2(net, saddr, sport,
509 htonl(INADDR_ANY), hnum, dif,
510 hslot2, slot2);
511 }
512 rcu_read_unlock();
513 return result;
514 }
515begin:
516 result = NULL;
517 badness = 0;
518 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
519 score = compute_score(sk, net, saddr, hnum, sport,
520 daddr, dport, dif);
521 if (score > badness) {
522 result = sk;
523 badness = score;
524 reuseport = sk->sk_reuseport;
525 if (reuseport) {
526 hash = udp_ehashfn(net, daddr, hnum,
527 saddr, sport);
528 matches = 1;
529 }
530 } else if (score == badness && reuseport) {
531 matches++;
532 if (((u64)hash * matches) >> 32 == 0)
533 result = sk;
534 hash = next_pseudo_random32(hash);
535 }
536 }
537 /*
538 * if the nulls value we got at the end of this lookup is
539 * not the expected one, we must restart lookup.
540 * We probably met an item that was moved to another chain.
541 */
542 if (get_nulls_value(node) != slot)
543 goto begin;
544
545 if (result) {
546 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
547 result = NULL;
548 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
549 daddr, dport, dif) < badness)) {
550 sock_put(result);
551 goto begin;
552 }
553 }
554 rcu_read_unlock();
555 return result;
556}
557EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
558
559static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
560 __be16 sport, __be16 dport,
561 struct udp_table *udptable)
562{
563 const struct iphdr *iph = ip_hdr(skb);
564
565 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
566 iph->daddr, dport, inet_iif(skb),
567 udptable);
568}
569
570struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
571 __be32 daddr, __be16 dport, int dif)
572{
573 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
574}
575EXPORT_SYMBOL_GPL(udp4_lib_lookup);
576
577static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
578 __be16 loc_port, __be32 loc_addr,
579 __be16 rmt_port, __be32 rmt_addr,
580 int dif, unsigned short hnum)
581{
582 struct inet_sock *inet = inet_sk(sk);
583
584 if (!net_eq(sock_net(sk), net) ||
585 udp_sk(sk)->udp_port_hash != hnum ||
586 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
587 (inet->inet_dport != rmt_port && inet->inet_dport) ||
588 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
589 ipv6_only_sock(sk) ||
590 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
591 return false;
592 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
593 return false;
594 return true;
595}
596
597static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
598 __be16 loc_port, __be32 loc_addr,
599 __be16 rmt_port, __be32 rmt_addr,
600 int dif)
601{
602 struct hlist_nulls_node *node;
603 struct sock *s = sk;
604 unsigned short hnum = ntohs(loc_port);
605
606 sk_nulls_for_each_from(s, node) {
607 if (__udp_is_mcast_sock(net, s,
608 loc_port, loc_addr,
609 rmt_port, rmt_addr,
610 dif, hnum))
611 goto found;
612 }
613 s = NULL;
614found:
615 return s;
616}
617
618/*
619 * This routine is called by the ICMP module when it gets some
620 * sort of error condition. If err < 0 then the socket should
621 * be closed and the error returned to the user. If err > 0
622 * it's just the icmp type << 8 | icmp code.
623 * Header points to the ip header of the error packet. We move
624 * on past this. Then (as it used to claim before adjustment)
625 * header points to the first 8 bytes of the udp header. We need
626 * to find the appropriate port.
627 */
628
629void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
630{
631 struct inet_sock *inet;
632 const struct iphdr *iph = (const struct iphdr *)skb->data;
633 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
634 const int type = icmp_hdr(skb)->type;
635 const int code = icmp_hdr(skb)->code;
636 struct sock *sk;
637 int harderr;
638 int err;
639 struct net *net = dev_net(skb->dev);
640
641 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
642 iph->saddr, uh->source, skb->dev->ifindex, udptable);
643 if (sk == NULL) {
644 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
645 return; /* No socket for error */
646 }
647
648 err = 0;
649 harderr = 0;
650 inet = inet_sk(sk);
651
652 switch (type) {
653 default:
654 case ICMP_TIME_EXCEEDED:
655 err = EHOSTUNREACH;
656 break;
657 case ICMP_SOURCE_QUENCH:
658 goto out;
659 case ICMP_PARAMETERPROB:
660 err = EPROTO;
661 harderr = 1;
662 break;
663 case ICMP_DEST_UNREACH:
664 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
665 ipv4_sk_update_pmtu(skb, sk, info);
666 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
667 err = EMSGSIZE;
668 harderr = 1;
669 break;
670 }
671 goto out;
672 }
673 err = EHOSTUNREACH;
674 if (code <= NR_ICMP_UNREACH) {
675 harderr = icmp_err_convert[code].fatal;
676 err = icmp_err_convert[code].errno;
677 }
678 break;
679 case ICMP_REDIRECT:
680 ipv4_sk_redirect(skb, sk);
681 goto out;
682 }
683
684 /*
685 * RFC1122: OK. Passes ICMP errors back to application, as per
686 * 4.1.3.3.
687 */
688 if (!inet->recverr) {
689 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
690 goto out;
691 } else
692 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
693
694 sk->sk_err = err;
695 sk->sk_error_report(sk);
696out:
697 sock_put(sk);
698}
699
700void udp_err(struct sk_buff *skb, u32 info)
701{
702 __udp4_lib_err(skb, info, &udp_table);
703}
704
705/*
706 * Throw away all pending data and cancel the corking. Socket is locked.
707 */
708void udp_flush_pending_frames(struct sock *sk)
709{
710 struct udp_sock *up = udp_sk(sk);
711
712 if (up->pending) {
713 up->len = 0;
714 up->pending = 0;
715 ip_flush_pending_frames(sk);
716 }
717}
718EXPORT_SYMBOL(udp_flush_pending_frames);
719
720/**
721 * udp4_hwcsum - handle outgoing HW checksumming
722 * @skb: sk_buff containing the filled-in UDP header
723 * (checksum field must be zeroed out)
724 * @src: source IP address
725 * @dst: destination IP address
726 */
727void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
728{
729 struct udphdr *uh = udp_hdr(skb);
730 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
731 int offset = skb_transport_offset(skb);
732 int len = skb->len - offset;
733 int hlen = len;
734 __wsum csum = 0;
735
736 if (!frags) {
737 /*
738 * Only one fragment on the socket.
739 */
740 skb->csum_start = skb_transport_header(skb) - skb->head;
741 skb->csum_offset = offsetof(struct udphdr, check);
742 uh->check = ~csum_tcpudp_magic(src, dst, len,
743 IPPROTO_UDP, 0);
744 } else {
745 /*
746 * HW-checksum won't work as there are two or more
747 * fragments on the socket so that all csums of sk_buffs
748 * should be together
749 */
750 do {
751 csum = csum_add(csum, frags->csum);
752 hlen -= frags->len;
753 } while ((frags = frags->next));
754
755 csum = skb_checksum(skb, offset, hlen, csum);
756 skb->ip_summed = CHECKSUM_NONE;
757
758 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
759 if (uh->check == 0)
760 uh->check = CSUM_MANGLED_0;
761 }
762}
763EXPORT_SYMBOL_GPL(udp4_hwcsum);
764
765static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
766{
767 struct sock *sk = skb->sk;
768 struct inet_sock *inet = inet_sk(sk);
769 struct udphdr *uh;
770 int err = 0;
771 int is_udplite = IS_UDPLITE(sk);
772 int offset = skb_transport_offset(skb);
773 int len = skb->len - offset;
774 __wsum csum = 0;
775
776 /*
777 * Create a UDP header
778 */
779 uh = udp_hdr(skb);
780 uh->source = inet->inet_sport;
781 uh->dest = fl4->fl4_dport;
782 uh->len = htons(len);
783 uh->check = 0;
784
785 if (is_udplite) /* UDP-Lite */
786 csum = udplite_csum(skb);
787
788 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
789
790 skb->ip_summed = CHECKSUM_NONE;
791 goto send;
792
793 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
794
795 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
796 goto send;
797
798 } else
799 csum = udp_csum(skb);
800
801 /* add protocol-dependent pseudo-header */
802 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
803 sk->sk_protocol, csum);
804 if (uh->check == 0)
805 uh->check = CSUM_MANGLED_0;
806
807send:
808 err = ip_send_skb(sock_net(sk), skb);
809 if (err) {
810 if (err == -ENOBUFS && !inet->recverr) {
811 UDP_INC_STATS_USER(sock_net(sk),
812 UDP_MIB_SNDBUFERRORS, is_udplite);
813 err = 0;
814 }
815 } else
816 UDP_INC_STATS_USER(sock_net(sk),
817 UDP_MIB_OUTDATAGRAMS, is_udplite);
818 return err;
819}
820
821/*
822 * Push out all pending data as one UDP datagram. Socket is locked.
823 */
824int udp_push_pending_frames(struct sock *sk)
825{
826 struct udp_sock *up = udp_sk(sk);
827 struct inet_sock *inet = inet_sk(sk);
828 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
829 struct sk_buff *skb;
830 int err = 0;
831
832 skb = ip_finish_skb(sk, fl4);
833 if (!skb)
834 goto out;
835
836 err = udp_send_skb(skb, fl4);
837
838out:
839 up->len = 0;
840 up->pending = 0;
841 return err;
842}
843EXPORT_SYMBOL(udp_push_pending_frames);
844
845int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
846 size_t len)
847{
848 struct inet_sock *inet = inet_sk(sk);
849 struct udp_sock *up = udp_sk(sk);
850 struct flowi4 fl4_stack;
851 struct flowi4 *fl4;
852 int ulen = len;
853 struct ipcm_cookie ipc;
854 struct rtable *rt = NULL;
855 int free = 0;
856 int connected = 0;
857 __be32 daddr, faddr, saddr;
858 __be16 dport;
859 u8 tos;
860 int err, is_udplite = IS_UDPLITE(sk);
861 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
862 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
863 struct sk_buff *skb;
864 struct ip_options_data opt_copy;
865
866 if (len > 0xFFFF)
867 return -EMSGSIZE;
868
869 /*
870 * Check the flags.
871 */
872
873 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
874 return -EOPNOTSUPP;
875
876 ipc.opt = NULL;
877 ipc.tx_flags = 0;
878 ipc.ttl = 0;
879 ipc.tos = -1;
880
881 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
882
883 fl4 = &inet->cork.fl.u.ip4;
884 if (up->pending) {
885 /*
886 * There are pending frames.
887 * The socket lock must be held while it's corked.
888 */
889 lock_sock(sk);
890 if (likely(up->pending)) {
891 if (unlikely(up->pending != AF_INET)) {
892 release_sock(sk);
893 return -EINVAL;
894 }
895 goto do_append_data;
896 }
897 release_sock(sk);
898 }
899 ulen += sizeof(struct udphdr);
900
901 /*
902 * Get and verify the address.
903 */
904 if (msg->msg_name) {
905 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
906 if (msg->msg_namelen < sizeof(*usin))
907 return -EINVAL;
908 if (usin->sin_family != AF_INET) {
909 if (usin->sin_family != AF_UNSPEC)
910 return -EAFNOSUPPORT;
911 }
912
913 daddr = usin->sin_addr.s_addr;
914 dport = usin->sin_port;
915 if (dport == 0)
916 return -EINVAL;
917 } else {
918 if (sk->sk_state != TCP_ESTABLISHED)
919 return -EDESTADDRREQ;
920 daddr = inet->inet_daddr;
921 dport = inet->inet_dport;
922 /* Open fast path for connected socket.
923 Route will not be used, if at least one option is set.
924 */
925 connected = 1;
926 }
927 ipc.addr = inet->inet_saddr;
928
929 ipc.oif = sk->sk_bound_dev_if;
930
931 sock_tx_timestamp(sk, &ipc.tx_flags);
932
933 if (msg->msg_controllen) {
934 err = ip_cmsg_send(sock_net(sk), msg, &ipc,
935 sk->sk_family == AF_INET6);
936 if (err)
937 return err;
938 if (ipc.opt)
939 free = 1;
940 connected = 0;
941 }
942 if (!ipc.opt) {
943 struct ip_options_rcu *inet_opt;
944
945 rcu_read_lock();
946 inet_opt = rcu_dereference(inet->inet_opt);
947 if (inet_opt) {
948 memcpy(&opt_copy, inet_opt,
949 sizeof(*inet_opt) + inet_opt->opt.optlen);
950 ipc.opt = &opt_copy.opt;
951 }
952 rcu_read_unlock();
953 }
954
955 saddr = ipc.addr;
956 ipc.addr = faddr = daddr;
957
958 if (ipc.opt && ipc.opt->opt.srr) {
959 if (!daddr)
960 return -EINVAL;
961 faddr = ipc.opt->opt.faddr;
962 connected = 0;
963 }
964 tos = get_rttos(&ipc, inet);
965 if (sock_flag(sk, SOCK_LOCALROUTE) ||
966 (msg->msg_flags & MSG_DONTROUTE) ||
967 (ipc.opt && ipc.opt->opt.is_strictroute)) {
968 tos |= RTO_ONLINK;
969 connected = 0;
970 }
971
972 if (ipv4_is_multicast(daddr)) {
973 if (!ipc.oif)
974 ipc.oif = inet->mc_index;
975 if (!saddr)
976 saddr = inet->mc_addr;
977 connected = 0;
978 } else if (!ipc.oif)
979 ipc.oif = inet->uc_index;
980
981 if (connected)
982 rt = (struct rtable *)sk_dst_check(sk, 0);
983
984 if (rt == NULL) {
985 struct net *net = sock_net(sk);
986
987 fl4 = &fl4_stack;
988 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
989 RT_SCOPE_UNIVERSE, sk->sk_protocol,
990 inet_sk_flowi_flags(sk),
991 faddr, saddr, dport, inet->inet_sport);
992
993 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
994 rt = ip_route_output_flow(net, fl4, sk);
995 if (IS_ERR(rt)) {
996 err = PTR_ERR(rt);
997 rt = NULL;
998 if (err == -ENETUNREACH)
999 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1000 goto out;
1001 }
1002
1003 err = -EACCES;
1004 if ((rt->rt_flags & RTCF_BROADCAST) &&
1005 !sock_flag(sk, SOCK_BROADCAST))
1006 goto out;
1007 if (connected)
1008 sk_dst_set(sk, dst_clone(&rt->dst));
1009 }
1010
1011 if (msg->msg_flags&MSG_CONFIRM)
1012 goto do_confirm;
1013back_from_confirm:
1014
1015 saddr = fl4->saddr;
1016 if (!ipc.addr)
1017 daddr = ipc.addr = fl4->daddr;
1018
1019 /* Lockless fast path for the non-corking case. */
1020 if (!corkreq) {
1021 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
1022 sizeof(struct udphdr), &ipc, &rt,
1023 msg->msg_flags);
1024 err = PTR_ERR(skb);
1025 if (!IS_ERR_OR_NULL(skb))
1026 err = udp_send_skb(skb, fl4);
1027 goto out;
1028 }
1029
1030 lock_sock(sk);
1031 if (unlikely(up->pending)) {
1032 /* The socket is already corked while preparing it. */
1033 /* ... which is an evident application bug. --ANK */
1034 release_sock(sk);
1035
1036 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1037 err = -EINVAL;
1038 goto out;
1039 }
1040 /*
1041 * Now cork the socket to pend data.
1042 */
1043 fl4 = &inet->cork.fl.u.ip4;
1044 fl4->daddr = daddr;
1045 fl4->saddr = saddr;
1046 fl4->fl4_dport = dport;
1047 fl4->fl4_sport = inet->inet_sport;
1048 up->pending = AF_INET;
1049
1050do_append_data:
1051 up->len += ulen;
1052 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1053 sizeof(struct udphdr), &ipc, &rt,
1054 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1055 if (err)
1056 udp_flush_pending_frames(sk);
1057 else if (!corkreq)
1058 err = udp_push_pending_frames(sk);
1059 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1060 up->pending = 0;
1061 release_sock(sk);
1062
1063out:
1064 ip_rt_put(rt);
1065 if (free)
1066 kfree(ipc.opt);
1067 if (!err)
1068 return len;
1069 /*
1070 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1071 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1072 * we don't have a good statistic (IpOutDiscards but it can be too many
1073 * things). We could add another new stat but at least for now that
1074 * seems like overkill.
1075 */
1076 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1077 UDP_INC_STATS_USER(sock_net(sk),
1078 UDP_MIB_SNDBUFERRORS, is_udplite);
1079 }
1080 return err;
1081
1082do_confirm:
1083 dst_confirm(&rt->dst);
1084 if (!(msg->msg_flags&MSG_PROBE) || len)
1085 goto back_from_confirm;
1086 err = 0;
1087 goto out;
1088}
1089EXPORT_SYMBOL(udp_sendmsg);
1090
1091int udp_sendpage(struct sock *sk, struct page *page, int offset,
1092 size_t size, int flags)
1093{
1094 struct inet_sock *inet = inet_sk(sk);
1095 struct udp_sock *up = udp_sk(sk);
1096 int ret;
1097
1098 if (flags & MSG_SENDPAGE_NOTLAST)
1099 flags |= MSG_MORE;
1100
1101 if (!up->pending) {
1102 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1103
1104 /* Call udp_sendmsg to specify destination address which
1105 * sendpage interface can't pass.
1106 * This will succeed only when the socket is connected.
1107 */
1108 ret = udp_sendmsg(NULL, sk, &msg, 0);
1109 if (ret < 0)
1110 return ret;
1111 }
1112
1113 lock_sock(sk);
1114
1115 if (unlikely(!up->pending)) {
1116 release_sock(sk);
1117
1118 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1119 return -EINVAL;
1120 }
1121
1122 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1123 page, offset, size, flags);
1124 if (ret == -EOPNOTSUPP) {
1125 release_sock(sk);
1126 return sock_no_sendpage(sk->sk_socket, page, offset,
1127 size, flags);
1128 }
1129 if (ret < 0) {
1130 udp_flush_pending_frames(sk);
1131 goto out;
1132 }
1133
1134 up->len += size;
1135 if (!(up->corkflag || (flags&MSG_MORE)))
1136 ret = udp_push_pending_frames(sk);
1137 if (!ret)
1138 ret = size;
1139out:
1140 release_sock(sk);
1141 return ret;
1142}
1143
1144
1145/**
1146 * first_packet_length - return length of first packet in receive queue
1147 * @sk: socket
1148 *
1149 * Drops all bad checksum frames, until a valid one is found.
1150 * Returns the length of found skb, or 0 if none is found.
1151 */
1152static unsigned int first_packet_length(struct sock *sk)
1153{
1154 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1155 struct sk_buff *skb;
1156 unsigned int res;
1157
1158 __skb_queue_head_init(&list_kill);
1159
1160 spin_lock_bh(&rcvq->lock);
1161 while ((skb = skb_peek(rcvq)) != NULL &&
1162 udp_lib_checksum_complete(skb)) {
1163 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1164 IS_UDPLITE(sk));
1165 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1166 IS_UDPLITE(sk));
1167 atomic_inc(&sk->sk_drops);
1168 __skb_unlink(skb, rcvq);
1169 __skb_queue_tail(&list_kill, skb);
1170 }
1171 res = skb ? skb->len : 0;
1172 spin_unlock_bh(&rcvq->lock);
1173
1174 if (!skb_queue_empty(&list_kill)) {
1175 bool slow = lock_sock_fast(sk);
1176
1177 __skb_queue_purge(&list_kill);
1178 sk_mem_reclaim_partial(sk);
1179 unlock_sock_fast(sk, slow);
1180 }
1181 return res;
1182}
1183
1184/*
1185 * IOCTL requests applicable to the UDP protocol
1186 */
1187
1188int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1189{
1190 switch (cmd) {
1191 case SIOCOUTQ:
1192 {
1193 int amount = sk_wmem_alloc_get(sk);
1194
1195 return put_user(amount, (int __user *)arg);
1196 }
1197
1198 case SIOCINQ:
1199 {
1200 unsigned int amount = first_packet_length(sk);
1201
1202 if (amount)
1203 /*
1204 * We will only return the amount
1205 * of this packet since that is all
1206 * that will be read.
1207 */
1208 amount -= sizeof(struct udphdr);
1209
1210 return put_user(amount, (int __user *)arg);
1211 }
1212
1213 default:
1214 return -ENOIOCTLCMD;
1215 }
1216
1217 return 0;
1218}
1219EXPORT_SYMBOL(udp_ioctl);
1220
1221/*
1222 * This should be easy, if there is something there we
1223 * return it, otherwise we block.
1224 */
1225
1226int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1227 size_t len, int noblock, int flags, int *addr_len)
1228{
1229 struct inet_sock *inet = inet_sk(sk);
1230 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1231 struct sk_buff *skb;
1232 unsigned int ulen, copied;
1233 int peeked, off = 0;
1234 int err;
1235 int is_udplite = IS_UDPLITE(sk);
1236 bool slow;
1237
1238 if (flags & MSG_ERRQUEUE)
1239 return ip_recv_error(sk, msg, len, addr_len);
1240
1241try_again:
1242 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1243 &peeked, &off, &err);
1244 if (!skb)
1245 goto out;
1246
1247 ulen = skb->len - sizeof(struct udphdr);
1248 copied = len;
1249 if (copied > ulen)
1250 copied = ulen;
1251 else if (copied < ulen)
1252 msg->msg_flags |= MSG_TRUNC;
1253
1254 /*
1255 * If checksum is needed at all, try to do it while copying the
1256 * data. If the data is truncated, or if we only want a partial
1257 * coverage checksum (UDP-Lite), do it before the copy.
1258 */
1259
1260 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1261 if (udp_lib_checksum_complete(skb))
1262 goto csum_copy_err;
1263 }
1264
1265 if (skb_csum_unnecessary(skb))
1266 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1267 msg->msg_iov, copied);
1268 else {
1269 err = skb_copy_and_csum_datagram_iovec(skb,
1270 sizeof(struct udphdr),
1271 msg->msg_iov);
1272
1273 if (err == -EINVAL)
1274 goto csum_copy_err;
1275 }
1276
1277 if (unlikely(err)) {
1278 trace_kfree_skb(skb, udp_recvmsg);
1279 if (!peeked) {
1280 atomic_inc(&sk->sk_drops);
1281 UDP_INC_STATS_USER(sock_net(sk),
1282 UDP_MIB_INERRORS, is_udplite);
1283 }
1284 goto out_free;
1285 }
1286
1287 if (!peeked)
1288 UDP_INC_STATS_USER(sock_net(sk),
1289 UDP_MIB_INDATAGRAMS, is_udplite);
1290
1291 sock_recv_ts_and_drops(msg, sk, skb);
1292
1293 /* Copy the address. */
1294 if (sin) {
1295 sin->sin_family = AF_INET;
1296 sin->sin_port = udp_hdr(skb)->source;
1297 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1298 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1299 *addr_len = sizeof(*sin);
1300 }
1301 if (inet->cmsg_flags)
1302 ip_cmsg_recv(msg, skb);
1303
1304 err = copied;
1305 if (flags & MSG_TRUNC)
1306 err = ulen;
1307
1308out_free:
1309 skb_free_datagram_locked(sk, skb);
1310out:
1311 return err;
1312
1313csum_copy_err:
1314 slow = lock_sock_fast(sk);
1315 if (!skb_kill_datagram(sk, skb, flags)) {
1316 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1317 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1318 }
1319 unlock_sock_fast(sk, slow);
1320
1321 if (noblock)
1322 return -EAGAIN;
1323
1324 /* starting over for a new packet */
1325 msg->msg_flags &= ~MSG_TRUNC;
1326 goto try_again;
1327}
1328
1329
1330int udp_disconnect(struct sock *sk, int flags)
1331{
1332 struct inet_sock *inet = inet_sk(sk);
1333 /*
1334 * 1003.1g - break association.
1335 */
1336
1337 sk->sk_state = TCP_CLOSE;
1338 inet->inet_daddr = 0;
1339 inet->inet_dport = 0;
1340 sock_rps_reset_rxhash(sk);
1341 sk->sk_bound_dev_if = 0;
1342 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1343 inet_reset_saddr(sk);
1344
1345 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1346 sk->sk_prot->unhash(sk);
1347 inet->inet_sport = 0;
1348 }
1349 sk_dst_reset(sk);
1350 return 0;
1351}
1352EXPORT_SYMBOL(udp_disconnect);
1353
1354void udp_lib_unhash(struct sock *sk)
1355{
1356 if (sk_hashed(sk)) {
1357 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1358 struct udp_hslot *hslot, *hslot2;
1359
1360 hslot = udp_hashslot(udptable, sock_net(sk),
1361 udp_sk(sk)->udp_port_hash);
1362 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1363
1364 spin_lock_bh(&hslot->lock);
1365 if (sk_nulls_del_node_init_rcu(sk)) {
1366 hslot->count--;
1367 inet_sk(sk)->inet_num = 0;
1368 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1369
1370 spin_lock(&hslot2->lock);
1371 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1372 hslot2->count--;
1373 spin_unlock(&hslot2->lock);
1374 }
1375 spin_unlock_bh(&hslot->lock);
1376 }
1377}
1378EXPORT_SYMBOL(udp_lib_unhash);
1379
1380/*
1381 * inet_rcv_saddr was changed, we must rehash secondary hash
1382 */
1383void udp_lib_rehash(struct sock *sk, u16 newhash)
1384{
1385 if (sk_hashed(sk)) {
1386 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1387 struct udp_hslot *hslot, *hslot2, *nhslot2;
1388
1389 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1390 nhslot2 = udp_hashslot2(udptable, newhash);
1391 udp_sk(sk)->udp_portaddr_hash = newhash;
1392 if (hslot2 != nhslot2) {
1393 hslot = udp_hashslot(udptable, sock_net(sk),
1394 udp_sk(sk)->udp_port_hash);
1395 /* we must lock primary chain too */
1396 spin_lock_bh(&hslot->lock);
1397
1398 spin_lock(&hslot2->lock);
1399 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1400 hslot2->count--;
1401 spin_unlock(&hslot2->lock);
1402
1403 spin_lock(&nhslot2->lock);
1404 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1405 &nhslot2->head);
1406 nhslot2->count++;
1407 spin_unlock(&nhslot2->lock);
1408
1409 spin_unlock_bh(&hslot->lock);
1410 }
1411 }
1412}
1413EXPORT_SYMBOL(udp_lib_rehash);
1414
1415static void udp_v4_rehash(struct sock *sk)
1416{
1417 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1418 inet_sk(sk)->inet_rcv_saddr,
1419 inet_sk(sk)->inet_num);
1420 udp_lib_rehash(sk, new_hash);
1421}
1422
1423static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1424{
1425 int rc;
1426
1427 if (inet_sk(sk)->inet_daddr) {
1428 sock_rps_save_rxhash(sk, skb);
1429 sk_mark_napi_id(sk, skb);
1430 }
1431
1432 rc = sock_queue_rcv_skb(sk, skb);
1433 if (rc < 0) {
1434 int is_udplite = IS_UDPLITE(sk);
1435
1436 /* Note that an ENOMEM error is charged twice */
1437 if (rc == -ENOMEM)
1438 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1439 is_udplite);
1440 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1441 kfree_skb(skb);
1442 trace_udp_fail_queue_rcv_skb(rc, sk);
1443 return -1;
1444 }
1445
1446 return 0;
1447
1448}
1449
1450static struct static_key udp_encap_needed __read_mostly;
1451void udp_encap_enable(void)
1452{
1453 if (!static_key_enabled(&udp_encap_needed))
1454 static_key_slow_inc(&udp_encap_needed);
1455}
1456EXPORT_SYMBOL(udp_encap_enable);
1457
1458/* returns:
1459 * -1: error
1460 * 0: success
1461 * >0: "udp encap" protocol resubmission
1462 *
1463 * Note that in the success and error cases, the skb is assumed to
1464 * have either been requeued or freed.
1465 */
1466int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1467{
1468 struct udp_sock *up = udp_sk(sk);
1469 int rc;
1470 int is_udplite = IS_UDPLITE(sk);
1471
1472 /*
1473 * Charge it to the socket, dropping if the queue is full.
1474 */
1475 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1476 goto drop;
1477 nf_reset(skb);
1478
1479 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1480 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1481
1482 /*
1483 * This is an encapsulation socket so pass the skb to
1484 * the socket's udp_encap_rcv() hook. Otherwise, just
1485 * fall through and pass this up the UDP socket.
1486 * up->encap_rcv() returns the following value:
1487 * =0 if skb was successfully passed to the encap
1488 * handler or was discarded by it.
1489 * >0 if skb should be passed on to UDP.
1490 * <0 if skb should be resubmitted as proto -N
1491 */
1492
1493 /* if we're overly short, let UDP handle it */
1494 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1495 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1496 int ret;
1497
1498 ret = encap_rcv(sk, skb);
1499 if (ret <= 0) {
1500 UDP_INC_STATS_BH(sock_net(sk),
1501 UDP_MIB_INDATAGRAMS,
1502 is_udplite);
1503 return -ret;
1504 }
1505 }
1506
1507 /* FALLTHROUGH -- it's a UDP Packet */
1508 }
1509
1510 /*
1511 * UDP-Lite specific tests, ignored on UDP sockets
1512 */
1513 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1514
1515 /*
1516 * MIB statistics other than incrementing the error count are
1517 * disabled for the following two types of errors: these depend
1518 * on the application settings, not on the functioning of the
1519 * protocol stack as such.
1520 *
1521 * RFC 3828 here recommends (sec 3.3): "There should also be a
1522 * way ... to ... at least let the receiving application block
1523 * delivery of packets with coverage values less than a value
1524 * provided by the application."
1525 */
1526 if (up->pcrlen == 0) { /* full coverage was set */
1527 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1528 UDP_SKB_CB(skb)->cscov, skb->len);
1529 goto drop;
1530 }
1531 /* The next case involves violating the min. coverage requested
1532 * by the receiver. This is subtle: if receiver wants x and x is
1533 * greater than the buffersize/MTU then receiver will complain
1534 * that it wants x while sender emits packets of smaller size y.
1535 * Therefore the above ...()->partial_cov statement is essential.
1536 */
1537 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1538 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1539 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1540 goto drop;
1541 }
1542 }
1543
1544 if (rcu_access_pointer(sk->sk_filter) &&
1545 udp_lib_checksum_complete(skb))
1546 goto csum_error;
1547
1548
1549 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1550 goto drop;
1551
1552 rc = 0;
1553
1554 ipv4_pktinfo_prepare(sk, skb);
1555 bh_lock_sock(sk);
1556 if (!sock_owned_by_user(sk))
1557 rc = __udp_queue_rcv_skb(sk, skb);
1558 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1559 bh_unlock_sock(sk);
1560 goto drop;
1561 }
1562 bh_unlock_sock(sk);
1563
1564 return rc;
1565
1566csum_error:
1567 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1568drop:
1569 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1570 atomic_inc(&sk->sk_drops);
1571 kfree_skb(skb);
1572 return -1;
1573}
1574
1575
1576static void flush_stack(struct sock **stack, unsigned int count,
1577 struct sk_buff *skb, unsigned int final)
1578{
1579 unsigned int i;
1580 struct sk_buff *skb1 = NULL;
1581 struct sock *sk;
1582
1583 for (i = 0; i < count; i++) {
1584 sk = stack[i];
1585 if (likely(skb1 == NULL))
1586 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1587
1588 if (!skb1) {
1589 atomic_inc(&sk->sk_drops);
1590 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1591 IS_UDPLITE(sk));
1592 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1593 IS_UDPLITE(sk));
1594 }
1595
1596 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1597 skb1 = NULL;
1598 }
1599 if (unlikely(skb1))
1600 kfree_skb(skb1);
1601}
1602
1603/* For TCP sockets, sk_rx_dst is protected by socket lock
1604 * For UDP, we use xchg() to guard against concurrent changes.
1605 */
1606static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1607{
1608 struct dst_entry *old;
1609
1610 dst_hold(dst);
1611 old = xchg(&sk->sk_rx_dst, dst);
1612 dst_release(old);
1613}
1614
1615/*
1616 * Multicasts and broadcasts go to each listener.
1617 *
1618 * Note: called only from the BH handler context.
1619 */
1620static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1621 struct udphdr *uh,
1622 __be32 saddr, __be32 daddr,
1623 struct udp_table *udptable)
1624{
1625 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1626 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1627 int dif;
1628 unsigned int i, count = 0;
1629
1630 spin_lock(&hslot->lock);
1631 sk = sk_nulls_head(&hslot->head);
1632 dif = skb->dev->ifindex;
1633 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1634 while (sk) {
1635 stack[count++] = sk;
1636 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1637 daddr, uh->source, saddr, dif);
1638 if (unlikely(count == ARRAY_SIZE(stack))) {
1639 if (!sk)
1640 break;
1641 flush_stack(stack, count, skb, ~0);
1642 count = 0;
1643 }
1644 }
1645 /*
1646 * before releasing chain lock, we must take a reference on sockets
1647 */
1648 for (i = 0; i < count; i++)
1649 sock_hold(stack[i]);
1650
1651 spin_unlock(&hslot->lock);
1652
1653 /*
1654 * do the slow work with no lock held
1655 */
1656 if (count) {
1657 flush_stack(stack, count, skb, count - 1);
1658
1659 for (i = 0; i < count; i++)
1660 sock_put(stack[i]);
1661 } else {
1662 kfree_skb(skb);
1663 }
1664 return 0;
1665}
1666
1667/* Initialize UDP checksum. If exited with zero value (success),
1668 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1669 * Otherwise, csum completion requires chacksumming packet body,
1670 * including udp header and folding it to skb->csum.
1671 */
1672static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1673 int proto)
1674{
1675 const struct iphdr *iph;
1676 int err;
1677
1678 UDP_SKB_CB(skb)->partial_cov = 0;
1679 UDP_SKB_CB(skb)->cscov = skb->len;
1680
1681 if (proto == IPPROTO_UDPLITE) {
1682 err = udplite_checksum_init(skb, uh);
1683 if (err)
1684 return err;
1685 }
1686
1687 iph = ip_hdr(skb);
1688 if (uh->check == 0) {
1689 skb->ip_summed = CHECKSUM_UNNECESSARY;
1690 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1691 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1692 proto, skb->csum))
1693 skb->ip_summed = CHECKSUM_UNNECESSARY;
1694 }
1695 if (!skb_csum_unnecessary(skb))
1696 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1697 skb->len, proto, 0);
1698 /* Probably, we should checksum udp header (it should be in cache
1699 * in any case) and data in tiny packets (< rx copybreak).
1700 */
1701
1702 return 0;
1703}
1704
1705/*
1706 * All we need to do is get the socket, and then do a checksum.
1707 */
1708
1709int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1710 int proto)
1711{
1712 struct sock *sk;
1713 struct udphdr *uh;
1714 unsigned short ulen;
1715 struct rtable *rt = skb_rtable(skb);
1716 __be32 saddr, daddr;
1717 struct net *net = dev_net(skb->dev);
1718
1719 /*
1720 * Validate the packet.
1721 */
1722 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1723 goto drop; /* No space for header. */
1724
1725 uh = udp_hdr(skb);
1726 ulen = ntohs(uh->len);
1727 saddr = ip_hdr(skb)->saddr;
1728 daddr = ip_hdr(skb)->daddr;
1729
1730 if (ulen > skb->len)
1731 goto short_packet;
1732
1733 if (proto == IPPROTO_UDP) {
1734 /* UDP validates ulen. */
1735 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1736 goto short_packet;
1737 uh = udp_hdr(skb);
1738 }
1739
1740 if (udp4_csum_init(skb, uh, proto))
1741 goto csum_error;
1742
1743 sk = skb_steal_sock(skb);
1744 if (sk) {
1745 struct dst_entry *dst = skb_dst(skb);
1746 int ret;
1747
1748 if (unlikely(sk->sk_rx_dst != dst))
1749 udp_sk_rx_dst_set(sk, dst);
1750
1751 ret = udp_queue_rcv_skb(sk, skb);
1752 sock_put(sk);
1753 /* a return value > 0 means to resubmit the input, but
1754 * it wants the return to be -protocol, or 0
1755 */
1756 if (ret > 0)
1757 return -ret;
1758 return 0;
1759 } else {
1760 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1761 return __udp4_lib_mcast_deliver(net, skb, uh,
1762 saddr, daddr, udptable);
1763
1764 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1765 }
1766
1767 if (sk != NULL) {
1768 int ret;
1769
1770 ret = udp_queue_rcv_skb(sk, skb);
1771 sock_put(sk);
1772
1773 /* a return value > 0 means to resubmit the input, but
1774 * it wants the return to be -protocol, or 0
1775 */
1776 if (ret > 0)
1777 return -ret;
1778 return 0;
1779 }
1780
1781 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1782 goto drop;
1783 nf_reset(skb);
1784
1785 /* No socket. Drop packet silently, if checksum is wrong */
1786 if (udp_lib_checksum_complete(skb))
1787 goto csum_error;
1788
1789 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1790 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1791
1792 /*
1793 * Hmm. We got an UDP packet to a port to which we
1794 * don't wanna listen. Ignore it.
1795 */
1796 kfree_skb(skb);
1797 return 0;
1798
1799short_packet:
1800 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1801 proto == IPPROTO_UDPLITE ? "Lite" : "",
1802 &saddr, ntohs(uh->source),
1803 ulen, skb->len,
1804 &daddr, ntohs(uh->dest));
1805 goto drop;
1806
1807csum_error:
1808 /*
1809 * RFC1122: OK. Discards the bad packet silently (as far as
1810 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1811 */
1812 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1813 proto == IPPROTO_UDPLITE ? "Lite" : "",
1814 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1815 ulen);
1816 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1817drop:
1818 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1819 kfree_skb(skb);
1820 return 0;
1821}
1822
1823/* We can only early demux multicast if there is a single matching socket.
1824 * If more than one socket found returns NULL
1825 */
1826static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1827 __be16 loc_port, __be32 loc_addr,
1828 __be16 rmt_port, __be32 rmt_addr,
1829 int dif)
1830{
1831 struct sock *sk, *result;
1832 struct hlist_nulls_node *node;
1833 unsigned short hnum = ntohs(loc_port);
1834 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1835 struct udp_hslot *hslot = &udp_table.hash[slot];
1836
1837 rcu_read_lock();
1838begin:
1839 count = 0;
1840 result = NULL;
1841 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1842 if (__udp_is_mcast_sock(net, sk,
1843 loc_port, loc_addr,
1844 rmt_port, rmt_addr,
1845 dif, hnum)) {
1846 result = sk;
1847 ++count;
1848 }
1849 }
1850 /*
1851 * if the nulls value we got at the end of this lookup is
1852 * not the expected one, we must restart lookup.
1853 * We probably met an item that was moved to another chain.
1854 */
1855 if (get_nulls_value(node) != slot)
1856 goto begin;
1857
1858 if (result) {
1859 if (count != 1 ||
1860 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1861 result = NULL;
1862 else if (unlikely(!__udp_is_mcast_sock(net, result,
1863 loc_port, loc_addr,
1864 rmt_port, rmt_addr,
1865 dif, hnum))) {
1866 sock_put(result);
1867 result = NULL;
1868 }
1869 }
1870 rcu_read_unlock();
1871 return result;
1872}
1873
1874/* For unicast we should only early demux connected sockets or we can
1875 * break forwarding setups. The chains here can be long so only check
1876 * if the first socket is an exact match and if not move on.
1877 */
1878static struct sock *__udp4_lib_demux_lookup(struct net *net,
1879 __be16 loc_port, __be32 loc_addr,
1880 __be16 rmt_port, __be32 rmt_addr,
1881 int dif)
1882{
1883 struct sock *sk, *result;
1884 struct hlist_nulls_node *node;
1885 unsigned short hnum = ntohs(loc_port);
1886 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1887 unsigned int slot2 = hash2 & udp_table.mask;
1888 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1889 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr)
1890 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1891
1892 rcu_read_lock();
1893 result = NULL;
1894 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
1895 if (INET_MATCH(sk, net, acookie,
1896 rmt_addr, loc_addr, ports, dif))
1897 result = sk;
1898 /* Only check first socket in chain */
1899 break;
1900 }
1901
1902 if (result) {
1903 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1904 result = NULL;
1905 else if (unlikely(!INET_MATCH(sk, net, acookie,
1906 rmt_addr, loc_addr,
1907 ports, dif))) {
1908 sock_put(result);
1909 result = NULL;
1910 }
1911 }
1912 rcu_read_unlock();
1913 return result;
1914}
1915
1916void udp_v4_early_demux(struct sk_buff *skb)
1917{
1918 struct net *net = dev_net(skb->dev);
1919 const struct iphdr *iph;
1920 const struct udphdr *uh;
1921 struct sock *sk;
1922 struct dst_entry *dst;
1923 int dif = skb->dev->ifindex;
1924
1925 /* validate the packet */
1926 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1927 return;
1928
1929 iph = ip_hdr(skb);
1930 uh = udp_hdr(skb);
1931
1932 if (skb->pkt_type == PACKET_BROADCAST ||
1933 skb->pkt_type == PACKET_MULTICAST)
1934 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1935 uh->source, iph->saddr, dif);
1936 else if (skb->pkt_type == PACKET_HOST)
1937 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1938 uh->source, iph->saddr, dif);
1939 else
1940 return;
1941
1942 if (!sk)
1943 return;
1944
1945 skb->sk = sk;
1946 skb->destructor = sock_edemux;
1947 dst = sk->sk_rx_dst;
1948
1949 if (dst)
1950 dst = dst_check(dst, 0);
1951 if (dst)
1952 skb_dst_set_noref(skb, dst);
1953}
1954
1955int udp_rcv(struct sk_buff *skb)
1956{
1957 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1958}
1959
1960void udp_destroy_sock(struct sock *sk)
1961{
1962 struct udp_sock *up = udp_sk(sk);
1963 bool slow = lock_sock_fast(sk);
1964 udp_flush_pending_frames(sk);
1965 unlock_sock_fast(sk, slow);
1966 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1967 void (*encap_destroy)(struct sock *sk);
1968 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1969 if (encap_destroy)
1970 encap_destroy(sk);
1971 }
1972}
1973
1974/*
1975 * Socket option code for UDP
1976 */
1977int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1978 char __user *optval, unsigned int optlen,
1979 int (*push_pending_frames)(struct sock *))
1980{
1981 struct udp_sock *up = udp_sk(sk);
1982 int val;
1983 int err = 0;
1984 int is_udplite = IS_UDPLITE(sk);
1985
1986 if (optlen < sizeof(int))
1987 return -EINVAL;
1988
1989 if (get_user(val, (int __user *)optval))
1990 return -EFAULT;
1991
1992 switch (optname) {
1993 case UDP_CORK:
1994 if (val != 0) {
1995 up->corkflag = 1;
1996 } else {
1997 up->corkflag = 0;
1998 lock_sock(sk);
1999 (*push_pending_frames)(sk);
2000 release_sock(sk);
2001 }
2002 break;
2003
2004 case UDP_ENCAP:
2005 switch (val) {
2006 case 0:
2007 case UDP_ENCAP_ESPINUDP:
2008 case UDP_ENCAP_ESPINUDP_NON_IKE:
2009 up->encap_rcv = xfrm4_udp_encap_rcv;
2010 /* FALLTHROUGH */
2011 case UDP_ENCAP_L2TPINUDP:
2012 up->encap_type = val;
2013 udp_encap_enable();
2014 break;
2015 default:
2016 err = -ENOPROTOOPT;
2017 break;
2018 }
2019 break;
2020
2021 /*
2022 * UDP-Lite's partial checksum coverage (RFC 3828).
2023 */
2024 /* The sender sets actual checksum coverage length via this option.
2025 * The case coverage > packet length is handled by send module. */
2026 case UDPLITE_SEND_CSCOV:
2027 if (!is_udplite) /* Disable the option on UDP sockets */
2028 return -ENOPROTOOPT;
2029 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2030 val = 8;
2031 else if (val > USHRT_MAX)
2032 val = USHRT_MAX;
2033 up->pcslen = val;
2034 up->pcflag |= UDPLITE_SEND_CC;
2035 break;
2036
2037 /* The receiver specifies a minimum checksum coverage value. To make
2038 * sense, this should be set to at least 8 (as done below). If zero is
2039 * used, this again means full checksum coverage. */
2040 case UDPLITE_RECV_CSCOV:
2041 if (!is_udplite) /* Disable the option on UDP sockets */
2042 return -ENOPROTOOPT;
2043 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2044 val = 8;
2045 else if (val > USHRT_MAX)
2046 val = USHRT_MAX;
2047 up->pcrlen = val;
2048 up->pcflag |= UDPLITE_RECV_CC;
2049 break;
2050
2051 default:
2052 err = -ENOPROTOOPT;
2053 break;
2054 }
2055
2056 return err;
2057}
2058EXPORT_SYMBOL(udp_lib_setsockopt);
2059
2060int udp_setsockopt(struct sock *sk, int level, int optname,
2061 char __user *optval, unsigned int optlen)
2062{
2063 if (level == SOL_UDP || level == SOL_UDPLITE)
2064 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2065 udp_push_pending_frames);
2066 return ip_setsockopt(sk, level, optname, optval, optlen);
2067}
2068
2069#ifdef CONFIG_COMPAT
2070int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2071 char __user *optval, unsigned int optlen)
2072{
2073 if (level == SOL_UDP || level == SOL_UDPLITE)
2074 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2075 udp_push_pending_frames);
2076 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2077}
2078#endif
2079
2080int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2081 char __user *optval, int __user *optlen)
2082{
2083 struct udp_sock *up = udp_sk(sk);
2084 int val, len;
2085
2086 if (get_user(len, optlen))
2087 return -EFAULT;
2088
2089 len = min_t(unsigned int, len, sizeof(int));
2090
2091 if (len < 0)
2092 return -EINVAL;
2093
2094 switch (optname) {
2095 case UDP_CORK:
2096 val = up->corkflag;
2097 break;
2098
2099 case UDP_ENCAP:
2100 val = up->encap_type;
2101 break;
2102
2103 /* The following two cannot be changed on UDP sockets, the return is
2104 * always 0 (which corresponds to the full checksum coverage of UDP). */
2105 case UDPLITE_SEND_CSCOV:
2106 val = up->pcslen;
2107 break;
2108
2109 case UDPLITE_RECV_CSCOV:
2110 val = up->pcrlen;
2111 break;
2112
2113 default:
2114 return -ENOPROTOOPT;
2115 }
2116
2117 if (put_user(len, optlen))
2118 return -EFAULT;
2119 if (copy_to_user(optval, &val, len))
2120 return -EFAULT;
2121 return 0;
2122}
2123EXPORT_SYMBOL(udp_lib_getsockopt);
2124
2125int udp_getsockopt(struct sock *sk, int level, int optname,
2126 char __user *optval, int __user *optlen)
2127{
2128 if (level == SOL_UDP || level == SOL_UDPLITE)
2129 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2130 return ip_getsockopt(sk, level, optname, optval, optlen);
2131}
2132
2133#ifdef CONFIG_COMPAT
2134int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2135 char __user *optval, int __user *optlen)
2136{
2137 if (level == SOL_UDP || level == SOL_UDPLITE)
2138 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2139 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2140}
2141#endif
2142/**
2143 * udp_poll - wait for a UDP event.
2144 * @file - file struct
2145 * @sock - socket
2146 * @wait - poll table
2147 *
2148 * This is same as datagram poll, except for the special case of
2149 * blocking sockets. If application is using a blocking fd
2150 * and a packet with checksum error is in the queue;
2151 * then it could get return from select indicating data available
2152 * but then block when reading it. Add special case code
2153 * to work around these arguably broken applications.
2154 */
2155unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2156{
2157 unsigned int mask = datagram_poll(file, sock, wait);
2158 struct sock *sk = sock->sk;
2159
2160 sock_rps_record_flow(sk);
2161
2162 /* Check for false positives due to checksum errors */
2163 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2164 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2165 mask &= ~(POLLIN | POLLRDNORM);
2166
2167 return mask;
2168
2169}
2170EXPORT_SYMBOL(udp_poll);
2171
2172struct proto udp_prot = {
2173 .name = "UDP",
2174 .owner = THIS_MODULE,
2175 .close = udp_lib_close,
2176 .connect = ip4_datagram_connect,
2177 .disconnect = udp_disconnect,
2178 .ioctl = udp_ioctl,
2179 .destroy = udp_destroy_sock,
2180 .setsockopt = udp_setsockopt,
2181 .getsockopt = udp_getsockopt,
2182 .sendmsg = udp_sendmsg,
2183 .recvmsg = udp_recvmsg,
2184 .sendpage = udp_sendpage,
2185 .backlog_rcv = __udp_queue_rcv_skb,
2186 .release_cb = ip4_datagram_release_cb,
2187 .hash = udp_lib_hash,
2188 .unhash = udp_lib_unhash,
2189 .rehash = udp_v4_rehash,
2190 .get_port = udp_v4_get_port,
2191 .memory_allocated = &udp_memory_allocated,
2192 .sysctl_mem = sysctl_udp_mem,
2193 .sysctl_wmem = &sysctl_udp_wmem_min,
2194 .sysctl_rmem = &sysctl_udp_rmem_min,
2195 .obj_size = sizeof(struct udp_sock),
2196 .slab_flags = SLAB_DESTROY_BY_RCU,
2197 .h.udp_table = &udp_table,
2198#ifdef CONFIG_COMPAT
2199 .compat_setsockopt = compat_udp_setsockopt,
2200 .compat_getsockopt = compat_udp_getsockopt,
2201#endif
2202 .clear_sk = sk_prot_clear_portaddr_nulls,
2203};
2204EXPORT_SYMBOL(udp_prot);
2205
2206/* ------------------------------------------------------------------------ */
2207#ifdef CONFIG_PROC_FS
2208
2209static struct sock *udp_get_first(struct seq_file *seq, int start)
2210{
2211 struct sock *sk;
2212 struct udp_iter_state *state = seq->private;
2213 struct net *net = seq_file_net(seq);
2214
2215 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2216 ++state->bucket) {
2217 struct hlist_nulls_node *node;
2218 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2219
2220 if (hlist_nulls_empty(&hslot->head))
2221 continue;
2222
2223 spin_lock_bh(&hslot->lock);
2224 sk_nulls_for_each(sk, node, &hslot->head) {
2225 if (!net_eq(sock_net(sk), net))
2226 continue;
2227 if (sk->sk_family == state->family)
2228 goto found;
2229 }
2230 spin_unlock_bh(&hslot->lock);
2231 }
2232 sk = NULL;
2233found:
2234 return sk;
2235}
2236
2237static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2238{
2239 struct udp_iter_state *state = seq->private;
2240 struct net *net = seq_file_net(seq);
2241
2242 do {
2243 sk = sk_nulls_next(sk);
2244 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2245
2246 if (!sk) {
2247 if (state->bucket <= state->udp_table->mask)
2248 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2249 return udp_get_first(seq, state->bucket + 1);
2250 }
2251 return sk;
2252}
2253
2254static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2255{
2256 struct sock *sk = udp_get_first(seq, 0);
2257
2258 if (sk)
2259 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2260 --pos;
2261 return pos ? NULL : sk;
2262}
2263
2264static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2265{
2266 struct udp_iter_state *state = seq->private;
2267 state->bucket = MAX_UDP_PORTS;
2268
2269 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2270}
2271
2272static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2273{
2274 struct sock *sk;
2275
2276 if (v == SEQ_START_TOKEN)
2277 sk = udp_get_idx(seq, 0);
2278 else
2279 sk = udp_get_next(seq, v);
2280
2281 ++*pos;
2282 return sk;
2283}
2284
2285static void udp_seq_stop(struct seq_file *seq, void *v)
2286{
2287 struct udp_iter_state *state = seq->private;
2288
2289 if (state->bucket <= state->udp_table->mask)
2290 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2291}
2292
2293int udp_seq_open(struct inode *inode, struct file *file)
2294{
2295 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2296 struct udp_iter_state *s;
2297 int err;
2298
2299 err = seq_open_net(inode, file, &afinfo->seq_ops,
2300 sizeof(struct udp_iter_state));
2301 if (err < 0)
2302 return err;
2303
2304 s = ((struct seq_file *)file->private_data)->private;
2305 s->family = afinfo->family;
2306 s->udp_table = afinfo->udp_table;
2307 return err;
2308}
2309EXPORT_SYMBOL(udp_seq_open);
2310
2311/* ------------------------------------------------------------------------ */
2312int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2313{
2314 struct proc_dir_entry *p;
2315 int rc = 0;
2316
2317 afinfo->seq_ops.start = udp_seq_start;
2318 afinfo->seq_ops.next = udp_seq_next;
2319 afinfo->seq_ops.stop = udp_seq_stop;
2320
2321 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2322 afinfo->seq_fops, afinfo);
2323 if (!p)
2324 rc = -ENOMEM;
2325 return rc;
2326}
2327EXPORT_SYMBOL(udp_proc_register);
2328
2329void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2330{
2331 remove_proc_entry(afinfo->name, net->proc_net);
2332}
2333EXPORT_SYMBOL(udp_proc_unregister);
2334
2335/* ------------------------------------------------------------------------ */
2336static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2337 int bucket)
2338{
2339 struct inet_sock *inet = inet_sk(sp);
2340 __be32 dest = inet->inet_daddr;
2341 __be32 src = inet->inet_rcv_saddr;
2342 __u16 destp = ntohs(inet->inet_dport);
2343 __u16 srcp = ntohs(inet->inet_sport);
2344
2345 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2346 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2347 bucket, src, srcp, dest, destp, sp->sk_state,
2348 sk_wmem_alloc_get(sp),
2349 sk_rmem_alloc_get(sp),
2350 0, 0L, 0,
2351 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2352 0, sock_i_ino(sp),
2353 atomic_read(&sp->sk_refcnt), sp,
2354 atomic_read(&sp->sk_drops));
2355}
2356
2357int udp4_seq_show(struct seq_file *seq, void *v)
2358{
2359 seq_setwidth(seq, 127);
2360 if (v == SEQ_START_TOKEN)
2361 seq_puts(seq, " sl local_address rem_address st tx_queue "
2362 "rx_queue tr tm->when retrnsmt uid timeout "
2363 "inode ref pointer drops");
2364 else {
2365 struct udp_iter_state *state = seq->private;
2366
2367 udp4_format_sock(v, seq, state->bucket);
2368 }
2369 seq_pad(seq, '\n');
2370 return 0;
2371}
2372
2373static const struct file_operations udp_afinfo_seq_fops = {
2374 .owner = THIS_MODULE,
2375 .open = udp_seq_open,
2376 .read = seq_read,
2377 .llseek = seq_lseek,
2378 .release = seq_release_net
2379};
2380
2381/* ------------------------------------------------------------------------ */
2382static struct udp_seq_afinfo udp4_seq_afinfo = {
2383 .name = "udp",
2384 .family = AF_INET,
2385 .udp_table = &udp_table,
2386 .seq_fops = &udp_afinfo_seq_fops,
2387 .seq_ops = {
2388 .show = udp4_seq_show,
2389 },
2390};
2391
2392static int __net_init udp4_proc_init_net(struct net *net)
2393{
2394 return udp_proc_register(net, &udp4_seq_afinfo);
2395}
2396
2397static void __net_exit udp4_proc_exit_net(struct net *net)
2398{
2399 udp_proc_unregister(net, &udp4_seq_afinfo);
2400}
2401
2402static struct pernet_operations udp4_net_ops = {
2403 .init = udp4_proc_init_net,
2404 .exit = udp4_proc_exit_net,
2405};
2406
2407int __init udp4_proc_init(void)
2408{
2409 return register_pernet_subsys(&udp4_net_ops);
2410}
2411
2412void udp4_proc_exit(void)
2413{
2414 unregister_pernet_subsys(&udp4_net_ops);
2415}
2416#endif /* CONFIG_PROC_FS */
2417
2418static __initdata unsigned long uhash_entries;
2419static int __init set_uhash_entries(char *str)
2420{
2421 ssize_t ret;
2422
2423 if (!str)
2424 return 0;
2425
2426 ret = kstrtoul(str, 0, &uhash_entries);
2427 if (ret)
2428 return 0;
2429
2430 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2431 uhash_entries = UDP_HTABLE_SIZE_MIN;
2432 return 1;
2433}
2434__setup("uhash_entries=", set_uhash_entries);
2435
2436void __init udp_table_init(struct udp_table *table, const char *name)
2437{
2438 unsigned int i;
2439
2440 table->hash = alloc_large_system_hash(name,
2441 2 * sizeof(struct udp_hslot),
2442 uhash_entries,
2443 21, /* one slot per 2 MB */
2444 0,
2445 &table->log,
2446 &table->mask,
2447 UDP_HTABLE_SIZE_MIN,
2448 64 * 1024);
2449
2450 table->hash2 = table->hash + (table->mask + 1);
2451 for (i = 0; i <= table->mask; i++) {
2452 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2453 table->hash[i].count = 0;
2454 spin_lock_init(&table->hash[i].lock);
2455 }
2456 for (i = 0; i <= table->mask; i++) {
2457 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2458 table->hash2[i].count = 0;
2459 spin_lock_init(&table->hash2[i].lock);
2460 }
2461}
2462
2463void __init udp_init(void)
2464{
2465 unsigned long limit;
2466
2467 udp_table_init(&udp_table, "UDP");
2468 limit = nr_free_buffer_pages() / 8;
2469 limit = max(limit, 128UL);
2470 sysctl_udp_mem[0] = limit / 4 * 3;
2471 sysctl_udp_mem[1] = limit;
2472 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2473
2474 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2475 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2476}
2477
2478struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
2479 netdev_features_t features)
2480{
2481 struct sk_buff *segs = ERR_PTR(-EINVAL);
2482 u16 mac_offset = skb->mac_header;
2483 int mac_len = skb->mac_len;
2484 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
2485 __be16 protocol = skb->protocol;
2486 netdev_features_t enc_features;
2487 int outer_hlen;
2488
2489 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
2490 goto out;
2491
2492 skb->encapsulation = 0;
2493 __skb_pull(skb, tnl_hlen);
2494 skb_reset_mac_header(skb);
2495 skb_set_network_header(skb, skb_inner_network_offset(skb));
2496 skb->mac_len = skb_inner_network_offset(skb);
2497 skb->protocol = htons(ETH_P_TEB);
2498
2499 /* segment inner packet. */
2500 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb);
2501 segs = skb_mac_gso_segment(skb, enc_features);
2502 if (!segs || IS_ERR(segs)) {
2503 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
2504 mac_len);
2505 goto out;
2506 }
2507
2508 outer_hlen = skb_tnl_header_len(skb);
2509 skb = segs;
2510 do {
2511 struct udphdr *uh;
2512 int udp_offset = outer_hlen - tnl_hlen;
2513
2514 skb_reset_inner_headers(skb);
2515 skb->encapsulation = 1;
2516
2517 skb->mac_len = mac_len;
2518
2519 skb_push(skb, outer_hlen);
2520 skb_reset_mac_header(skb);
2521 skb_set_network_header(skb, mac_len);
2522 skb_set_transport_header(skb, udp_offset);
2523 uh = udp_hdr(skb);
2524 uh->len = htons(skb->len - udp_offset);
2525
2526 /* csum segment if tunnel sets skb with csum. */
2527 if (protocol == htons(ETH_P_IP) && unlikely(uh->check)) {
2528 struct iphdr *iph = ip_hdr(skb);
2529
2530 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
2531 skb->len - udp_offset,
2532 IPPROTO_UDP, 0);
2533 uh->check = csum_fold(skb_checksum(skb, udp_offset,
2534 skb->len - udp_offset, 0));
2535 if (uh->check == 0)
2536 uh->check = CSUM_MANGLED_0;
2537
2538 } else if (protocol == htons(ETH_P_IPV6)) {
2539 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
2540 u32 len = skb->len - udp_offset;
2541
2542 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
2543 len, IPPROTO_UDP, 0);
2544 uh->check = csum_fold(skb_checksum(skb, udp_offset, len, 0));
2545 if (uh->check == 0)
2546 uh->check = CSUM_MANGLED_0;
2547 skb->ip_summed = CHECKSUM_NONE;
2548 }
2549
2550 skb->protocol = protocol;
2551 } while ((skb = skb->next));
2552out:
2553 return segs;
2554}