Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * The User Datagram Protocol (UDP).
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
13 * Hirokazu Takahashi, <taka@valinux.co.jp>
14 *
15 * Fixes:
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
29 * does NOT close.
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
58 * for connect.
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * datagrams.
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
72 * James Chapman : Add L2TP encapsulation type.
73 */
74
75#define pr_fmt(fmt) "UDP: " fmt
76
77#include <linux/uaccess.h>
78#include <asm/ioctls.h>
79#include <linux/memblock.h>
80#include <linux/highmem.h>
81#include <linux/swap.h>
82#include <linux/types.h>
83#include <linux/fcntl.h>
84#include <linux/module.h>
85#include <linux/socket.h>
86#include <linux/sockios.h>
87#include <linux/igmp.h>
88#include <linux/inetdevice.h>
89#include <linux/in.h>
90#include <linux/errno.h>
91#include <linux/timer.h>
92#include <linux/mm.h>
93#include <linux/inet.h>
94#include <linux/netdevice.h>
95#include <linux/slab.h>
96#include <net/tcp_states.h>
97#include <linux/skbuff.h>
98#include <linux/proc_fs.h>
99#include <linux/seq_file.h>
100#include <net/net_namespace.h>
101#include <net/icmp.h>
102#include <net/inet_hashtables.h>
103#include <net/ip_tunnels.h>
104#include <net/route.h>
105#include <net/checksum.h>
106#include <net/xfrm.h>
107#include <trace/events/udp.h>
108#include <linux/static_key.h>
109#include <trace/events/skb.h>
110#include <net/busy_poll.h>
111#include "udp_impl.h"
112#include <net/sock_reuseport.h>
113#include <net/addrconf.h>
114#include <net/udp_tunnel.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
122atomic_long_t udp_memory_allocated;
123EXPORT_SYMBOL(udp_memory_allocated);
124
125#define MAX_UDP_PORTS 65536
126#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
127
128static int udp_lib_lport_inuse(struct net *net, __u16 num,
129 const struct udp_hslot *hslot,
130 unsigned long *bitmap,
131 struct sock *sk, unsigned int log)
132{
133 struct sock *sk2;
134 kuid_t uid = sock_i_uid(sk);
135
136 sk_for_each(sk2, &hslot->head) {
137 if (net_eq(sock_net(sk2), net) &&
138 sk2 != sk &&
139 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
140 (!sk2->sk_reuse || !sk->sk_reuse) &&
141 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
142 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
143 inet_rcv_saddr_equal(sk, sk2, true)) {
144 if (sk2->sk_reuseport && sk->sk_reuseport &&
145 !rcu_access_pointer(sk->sk_reuseport_cb) &&
146 uid_eq(uid, sock_i_uid(sk2))) {
147 if (!bitmap)
148 return 0;
149 } else {
150 if (!bitmap)
151 return 1;
152 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
153 bitmap);
154 }
155 }
156 }
157 return 0;
158}
159
160/*
161 * Note: we still hold spinlock of primary hash chain, so no other writer
162 * can insert/delete a socket with local_port == num
163 */
164static int udp_lib_lport_inuse2(struct net *net, __u16 num,
165 struct udp_hslot *hslot2,
166 struct sock *sk)
167{
168 struct sock *sk2;
169 kuid_t uid = sock_i_uid(sk);
170 int res = 0;
171
172 spin_lock(&hslot2->lock);
173 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
174 if (net_eq(sock_net(sk2), net) &&
175 sk2 != sk &&
176 (udp_sk(sk2)->udp_port_hash == num) &&
177 (!sk2->sk_reuse || !sk->sk_reuse) &&
178 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
179 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
180 inet_rcv_saddr_equal(sk, sk2, true)) {
181 if (sk2->sk_reuseport && sk->sk_reuseport &&
182 !rcu_access_pointer(sk->sk_reuseport_cb) &&
183 uid_eq(uid, sock_i_uid(sk2))) {
184 res = 0;
185 } else {
186 res = 1;
187 }
188 break;
189 }
190 }
191 spin_unlock(&hslot2->lock);
192 return res;
193}
194
195static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
196{
197 struct net *net = sock_net(sk);
198 kuid_t uid = sock_i_uid(sk);
199 struct sock *sk2;
200
201 sk_for_each(sk2, &hslot->head) {
202 if (net_eq(sock_net(sk2), net) &&
203 sk2 != sk &&
204 sk2->sk_family == sk->sk_family &&
205 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
206 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
207 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
208 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
209 inet_rcv_saddr_equal(sk, sk2, false)) {
210 return reuseport_add_sock(sk, sk2,
211 inet_rcv_saddr_any(sk));
212 }
213 }
214
215 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
216}
217
218/**
219 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
220 *
221 * @sk: socket struct in question
222 * @snum: port number to look up
223 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
224 * with NULL address
225 */
226int udp_lib_get_port(struct sock *sk, unsigned short snum,
227 unsigned int hash2_nulladdr)
228{
229 struct udp_hslot *hslot, *hslot2;
230 struct udp_table *udptable = sk->sk_prot->h.udp_table;
231 int error = 1;
232 struct net *net = sock_net(sk);
233
234 if (!snum) {
235 int low, high, remaining;
236 unsigned int rand;
237 unsigned short first, last;
238 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
239
240 inet_get_local_port_range(net, &low, &high);
241 remaining = (high - low) + 1;
242
243 rand = prandom_u32();
244 first = reciprocal_scale(rand, remaining) + low;
245 /*
246 * force rand to be an odd multiple of UDP_HTABLE_SIZE
247 */
248 rand = (rand | 1) * (udptable->mask + 1);
249 last = first + udptable->mask + 1;
250 do {
251 hslot = udp_hashslot(udptable, net, first);
252 bitmap_zero(bitmap, PORTS_PER_CHAIN);
253 spin_lock_bh(&hslot->lock);
254 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
255 udptable->log);
256
257 snum = first;
258 /*
259 * Iterate on all possible values of snum for this hash.
260 * Using steps of an odd multiple of UDP_HTABLE_SIZE
261 * give us randomization and full range coverage.
262 */
263 do {
264 if (low <= snum && snum <= high &&
265 !test_bit(snum >> udptable->log, bitmap) &&
266 !inet_is_local_reserved_port(net, snum))
267 goto found;
268 snum += rand;
269 } while (snum != first);
270 spin_unlock_bh(&hslot->lock);
271 cond_resched();
272 } while (++first != last);
273 goto fail;
274 } else {
275 hslot = udp_hashslot(udptable, net, snum);
276 spin_lock_bh(&hslot->lock);
277 if (hslot->count > 10) {
278 int exist;
279 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
280
281 slot2 &= udptable->mask;
282 hash2_nulladdr &= udptable->mask;
283
284 hslot2 = udp_hashslot2(udptable, slot2);
285 if (hslot->count < hslot2->count)
286 goto scan_primary_hash;
287
288 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
289 if (!exist && (hash2_nulladdr != slot2)) {
290 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
291 exist = udp_lib_lport_inuse2(net, snum, hslot2,
292 sk);
293 }
294 if (exist)
295 goto fail_unlock;
296 else
297 goto found;
298 }
299scan_primary_hash:
300 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
301 goto fail_unlock;
302 }
303found:
304 inet_sk(sk)->inet_num = snum;
305 udp_sk(sk)->udp_port_hash = snum;
306 udp_sk(sk)->udp_portaddr_hash ^= snum;
307 if (sk_unhashed(sk)) {
308 if (sk->sk_reuseport &&
309 udp_reuseport_add_sock(sk, hslot)) {
310 inet_sk(sk)->inet_num = 0;
311 udp_sk(sk)->udp_port_hash = 0;
312 udp_sk(sk)->udp_portaddr_hash ^= snum;
313 goto fail_unlock;
314 }
315
316 sk_add_node_rcu(sk, &hslot->head);
317 hslot->count++;
318 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
319
320 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
321 spin_lock(&hslot2->lock);
322 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
323 sk->sk_family == AF_INET6)
324 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
325 &hslot2->head);
326 else
327 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
328 &hslot2->head);
329 hslot2->count++;
330 spin_unlock(&hslot2->lock);
331 }
332 sock_set_flag(sk, SOCK_RCU_FREE);
333 error = 0;
334fail_unlock:
335 spin_unlock_bh(&hslot->lock);
336fail:
337 return error;
338}
339EXPORT_SYMBOL(udp_lib_get_port);
340
341int udp_v4_get_port(struct sock *sk, unsigned short snum)
342{
343 unsigned int hash2_nulladdr =
344 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
345 unsigned int hash2_partial =
346 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
347
348 /* precompute partial secondary hash */
349 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
350 return udp_lib_get_port(sk, snum, hash2_nulladdr);
351}
352
353static int compute_score(struct sock *sk, struct net *net,
354 __be32 saddr, __be16 sport,
355 __be32 daddr, unsigned short hnum,
356 int dif, int sdif)
357{
358 int score;
359 struct inet_sock *inet;
360 bool dev_match;
361
362 if (!net_eq(sock_net(sk), net) ||
363 udp_sk(sk)->udp_port_hash != hnum ||
364 ipv6_only_sock(sk))
365 return -1;
366
367 if (sk->sk_rcv_saddr != daddr)
368 return -1;
369
370 score = (sk->sk_family == PF_INET) ? 2 : 1;
371
372 inet = inet_sk(sk);
373 if (inet->inet_daddr) {
374 if (inet->inet_daddr != saddr)
375 return -1;
376 score += 4;
377 }
378
379 if (inet->inet_dport) {
380 if (inet->inet_dport != sport)
381 return -1;
382 score += 4;
383 }
384
385 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
386 dif, sdif);
387 if (!dev_match)
388 return -1;
389 score += 4;
390
391 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
392 score++;
393 return score;
394}
395
396static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
397 const __u16 lport, const __be32 faddr,
398 const __be16 fport)
399{
400 static u32 udp_ehash_secret __read_mostly;
401
402 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
403
404 return __inet_ehashfn(laddr, lport, faddr, fport,
405 udp_ehash_secret + net_hash_mix(net));
406}
407
408/* called with rcu_read_lock() */
409static struct sock *udp4_lib_lookup2(struct net *net,
410 __be32 saddr, __be16 sport,
411 __be32 daddr, unsigned int hnum,
412 int dif, int sdif,
413 struct udp_hslot *hslot2,
414 struct sk_buff *skb)
415{
416 struct sock *sk, *result;
417 int score, badness;
418 u32 hash = 0;
419
420 result = NULL;
421 badness = 0;
422 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
423 score = compute_score(sk, net, saddr, sport,
424 daddr, hnum, dif, sdif);
425 if (score > badness) {
426 if (sk->sk_reuseport &&
427 sk->sk_state != TCP_ESTABLISHED) {
428 hash = udp_ehashfn(net, daddr, hnum,
429 saddr, sport);
430 result = reuseport_select_sock(sk, hash, skb,
431 sizeof(struct udphdr));
432 if (result && !reuseport_has_conns(sk, false))
433 return result;
434 }
435 badness = score;
436 result = sk;
437 }
438 }
439 return result;
440}
441
442/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
443 * harder than this. -DaveM
444 */
445struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
446 __be16 sport, __be32 daddr, __be16 dport, int dif,
447 int sdif, struct udp_table *udptable, struct sk_buff *skb)
448{
449 struct sock *result;
450 unsigned short hnum = ntohs(dport);
451 unsigned int hash2, slot2;
452 struct udp_hslot *hslot2;
453
454 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
455 slot2 = hash2 & udptable->mask;
456 hslot2 = &udptable->hash2[slot2];
457
458 result = udp4_lib_lookup2(net, saddr, sport,
459 daddr, hnum, dif, sdif,
460 hslot2, skb);
461 if (!result) {
462 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
463 slot2 = hash2 & udptable->mask;
464 hslot2 = &udptable->hash2[slot2];
465
466 result = udp4_lib_lookup2(net, saddr, sport,
467 htonl(INADDR_ANY), hnum, dif, sdif,
468 hslot2, skb);
469 }
470 if (IS_ERR(result))
471 return NULL;
472 return result;
473}
474EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
475
476static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
477 __be16 sport, __be16 dport,
478 struct udp_table *udptable)
479{
480 const struct iphdr *iph = ip_hdr(skb);
481
482 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
483 iph->daddr, dport, inet_iif(skb),
484 inet_sdif(skb), udptable, skb);
485}
486
487struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
488 __be16 sport, __be16 dport)
489{
490 const struct iphdr *iph = ip_hdr(skb);
491
492 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
493 iph->daddr, dport, inet_iif(skb),
494 inet_sdif(skb), &udp_table, NULL);
495}
496EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
497
498/* Must be called under rcu_read_lock().
499 * Does increment socket refcount.
500 */
501#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
502struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
503 __be32 daddr, __be16 dport, int dif)
504{
505 struct sock *sk;
506
507 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
508 dif, 0, &udp_table, NULL);
509 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
510 sk = NULL;
511 return sk;
512}
513EXPORT_SYMBOL_GPL(udp4_lib_lookup);
514#endif
515
516static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
517 __be16 loc_port, __be32 loc_addr,
518 __be16 rmt_port, __be32 rmt_addr,
519 int dif, int sdif, unsigned short hnum)
520{
521 struct inet_sock *inet = inet_sk(sk);
522
523 if (!net_eq(sock_net(sk), net) ||
524 udp_sk(sk)->udp_port_hash != hnum ||
525 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
526 (inet->inet_dport != rmt_port && inet->inet_dport) ||
527 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
528 ipv6_only_sock(sk) ||
529 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
530 return false;
531 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
532 return false;
533 return true;
534}
535
536DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
537void udp_encap_enable(void)
538{
539 static_branch_inc(&udp_encap_needed_key);
540}
541EXPORT_SYMBOL(udp_encap_enable);
542
543/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
544 * through error handlers in encapsulations looking for a match.
545 */
546static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
547{
548 int i;
549
550 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
551 int (*handler)(struct sk_buff *skb, u32 info);
552 const struct ip_tunnel_encap_ops *encap;
553
554 encap = rcu_dereference(iptun_encaps[i]);
555 if (!encap)
556 continue;
557 handler = encap->err_handler;
558 if (handler && !handler(skb, info))
559 return 0;
560 }
561
562 return -ENOENT;
563}
564
565/* Try to match ICMP errors to UDP tunnels by looking up a socket without
566 * reversing source and destination port: this will match tunnels that force the
567 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
568 * lwtunnels might actually break this assumption by being configured with
569 * different destination ports on endpoints, in this case we won't be able to
570 * trace ICMP messages back to them.
571 *
572 * If this doesn't match any socket, probe tunnels with arbitrary destination
573 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
574 * we've sent packets to won't necessarily match the local destination port.
575 *
576 * Then ask the tunnel implementation to match the error against a valid
577 * association.
578 *
579 * Return an error if we can't find a match, the socket if we need further
580 * processing, zero otherwise.
581 */
582static struct sock *__udp4_lib_err_encap(struct net *net,
583 const struct iphdr *iph,
584 struct udphdr *uh,
585 struct udp_table *udptable,
586 struct sk_buff *skb, u32 info)
587{
588 int network_offset, transport_offset;
589 struct sock *sk;
590
591 network_offset = skb_network_offset(skb);
592 transport_offset = skb_transport_offset(skb);
593
594 /* Network header needs to point to the outer IPv4 header inside ICMP */
595 skb_reset_network_header(skb);
596
597 /* Transport header needs to point to the UDP header */
598 skb_set_transport_header(skb, iph->ihl << 2);
599
600 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
601 iph->saddr, uh->dest, skb->dev->ifindex, 0,
602 udptable, NULL);
603 if (sk) {
604 int (*lookup)(struct sock *sk, struct sk_buff *skb);
605 struct udp_sock *up = udp_sk(sk);
606
607 lookup = READ_ONCE(up->encap_err_lookup);
608 if (!lookup || lookup(sk, skb))
609 sk = NULL;
610 }
611
612 if (!sk)
613 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
614
615 skb_set_transport_header(skb, transport_offset);
616 skb_set_network_header(skb, network_offset);
617
618 return sk;
619}
620
621/*
622 * This routine is called by the ICMP module when it gets some
623 * sort of error condition. If err < 0 then the socket should
624 * be closed and the error returned to the user. If err > 0
625 * it's just the icmp type << 8 | icmp code.
626 * Header points to the ip header of the error packet. We move
627 * on past this. Then (as it used to claim before adjustment)
628 * header points to the first 8 bytes of the udp header. We need
629 * to find the appropriate port.
630 */
631
632int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
633{
634 struct inet_sock *inet;
635 const struct iphdr *iph = (const struct iphdr *)skb->data;
636 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
637 const int type = icmp_hdr(skb)->type;
638 const int code = icmp_hdr(skb)->code;
639 bool tunnel = false;
640 struct sock *sk;
641 int harderr;
642 int err;
643 struct net *net = dev_net(skb->dev);
644
645 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
646 iph->saddr, uh->source, skb->dev->ifindex,
647 inet_sdif(skb), udptable, NULL);
648 if (!sk) {
649 /* No socket for error: try tunnels before discarding */
650 sk = ERR_PTR(-ENOENT);
651 if (static_branch_unlikely(&udp_encap_needed_key)) {
652 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
653 info);
654 if (!sk)
655 return 0;
656 }
657
658 if (IS_ERR(sk)) {
659 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
660 return PTR_ERR(sk);
661 }
662
663 tunnel = true;
664 }
665
666 err = 0;
667 harderr = 0;
668 inet = inet_sk(sk);
669
670 switch (type) {
671 default:
672 case ICMP_TIME_EXCEEDED:
673 err = EHOSTUNREACH;
674 break;
675 case ICMP_SOURCE_QUENCH:
676 goto out;
677 case ICMP_PARAMETERPROB:
678 err = EPROTO;
679 harderr = 1;
680 break;
681 case ICMP_DEST_UNREACH:
682 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
683 ipv4_sk_update_pmtu(skb, sk, info);
684 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
685 err = EMSGSIZE;
686 harderr = 1;
687 break;
688 }
689 goto out;
690 }
691 err = EHOSTUNREACH;
692 if (code <= NR_ICMP_UNREACH) {
693 harderr = icmp_err_convert[code].fatal;
694 err = icmp_err_convert[code].errno;
695 }
696 break;
697 case ICMP_REDIRECT:
698 ipv4_sk_redirect(skb, sk);
699 goto out;
700 }
701
702 /*
703 * RFC1122: OK. Passes ICMP errors back to application, as per
704 * 4.1.3.3.
705 */
706 if (tunnel) {
707 /* ...not for tunnels though: we don't have a sending socket */
708 goto out;
709 }
710 if (!inet->recverr) {
711 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
712 goto out;
713 } else
714 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
715
716 sk->sk_err = err;
717 sk->sk_error_report(sk);
718out:
719 return 0;
720}
721
722int udp_err(struct sk_buff *skb, u32 info)
723{
724 return __udp4_lib_err(skb, info, &udp_table);
725}
726
727/*
728 * Throw away all pending data and cancel the corking. Socket is locked.
729 */
730void udp_flush_pending_frames(struct sock *sk)
731{
732 struct udp_sock *up = udp_sk(sk);
733
734 if (up->pending) {
735 up->len = 0;
736 up->pending = 0;
737 ip_flush_pending_frames(sk);
738 }
739}
740EXPORT_SYMBOL(udp_flush_pending_frames);
741
742/**
743 * udp4_hwcsum - handle outgoing HW checksumming
744 * @skb: sk_buff containing the filled-in UDP header
745 * (checksum field must be zeroed out)
746 * @src: source IP address
747 * @dst: destination IP address
748 */
749void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
750{
751 struct udphdr *uh = udp_hdr(skb);
752 int offset = skb_transport_offset(skb);
753 int len = skb->len - offset;
754 int hlen = len;
755 __wsum csum = 0;
756
757 if (!skb_has_frag_list(skb)) {
758 /*
759 * Only one fragment on the socket.
760 */
761 skb->csum_start = skb_transport_header(skb) - skb->head;
762 skb->csum_offset = offsetof(struct udphdr, check);
763 uh->check = ~csum_tcpudp_magic(src, dst, len,
764 IPPROTO_UDP, 0);
765 } else {
766 struct sk_buff *frags;
767
768 /*
769 * HW-checksum won't work as there are two or more
770 * fragments on the socket so that all csums of sk_buffs
771 * should be together
772 */
773 skb_walk_frags(skb, frags) {
774 csum = csum_add(csum, frags->csum);
775 hlen -= frags->len;
776 }
777
778 csum = skb_checksum(skb, offset, hlen, csum);
779 skb->ip_summed = CHECKSUM_NONE;
780
781 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
782 if (uh->check == 0)
783 uh->check = CSUM_MANGLED_0;
784 }
785}
786EXPORT_SYMBOL_GPL(udp4_hwcsum);
787
788/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
789 * for the simple case like when setting the checksum for a UDP tunnel.
790 */
791void udp_set_csum(bool nocheck, struct sk_buff *skb,
792 __be32 saddr, __be32 daddr, int len)
793{
794 struct udphdr *uh = udp_hdr(skb);
795
796 if (nocheck) {
797 uh->check = 0;
798 } else if (skb_is_gso(skb)) {
799 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
800 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
801 uh->check = 0;
802 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
803 if (uh->check == 0)
804 uh->check = CSUM_MANGLED_0;
805 } else {
806 skb->ip_summed = CHECKSUM_PARTIAL;
807 skb->csum_start = skb_transport_header(skb) - skb->head;
808 skb->csum_offset = offsetof(struct udphdr, check);
809 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
810 }
811}
812EXPORT_SYMBOL(udp_set_csum);
813
814static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
815 struct inet_cork *cork)
816{
817 struct sock *sk = skb->sk;
818 struct inet_sock *inet = inet_sk(sk);
819 struct udphdr *uh;
820 int err = 0;
821 int is_udplite = IS_UDPLITE(sk);
822 int offset = skb_transport_offset(skb);
823 int len = skb->len - offset;
824 int datalen = len - sizeof(*uh);
825 __wsum csum = 0;
826
827 /*
828 * Create a UDP header
829 */
830 uh = udp_hdr(skb);
831 uh->source = inet->inet_sport;
832 uh->dest = fl4->fl4_dport;
833 uh->len = htons(len);
834 uh->check = 0;
835
836 if (cork->gso_size) {
837 const int hlen = skb_network_header_len(skb) +
838 sizeof(struct udphdr);
839
840 if (hlen + cork->gso_size > cork->fragsize) {
841 kfree_skb(skb);
842 return -EINVAL;
843 }
844 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
845 kfree_skb(skb);
846 return -EINVAL;
847 }
848 if (sk->sk_no_check_tx) {
849 kfree_skb(skb);
850 return -EINVAL;
851 }
852 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
853 dst_xfrm(skb_dst(skb))) {
854 kfree_skb(skb);
855 return -EIO;
856 }
857
858 if (datalen > cork->gso_size) {
859 skb_shinfo(skb)->gso_size = cork->gso_size;
860 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
861 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
862 cork->gso_size);
863 }
864 goto csum_partial;
865 }
866
867 if (is_udplite) /* UDP-Lite */
868 csum = udplite_csum(skb);
869
870 else if (sk->sk_no_check_tx) { /* UDP csum off */
871
872 skb->ip_summed = CHECKSUM_NONE;
873 goto send;
874
875 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
876csum_partial:
877
878 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
879 goto send;
880
881 } else
882 csum = udp_csum(skb);
883
884 /* add protocol-dependent pseudo-header */
885 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
886 sk->sk_protocol, csum);
887 if (uh->check == 0)
888 uh->check = CSUM_MANGLED_0;
889
890send:
891 err = ip_send_skb(sock_net(sk), skb);
892 if (err) {
893 if (err == -ENOBUFS && !inet->recverr) {
894 UDP_INC_STATS(sock_net(sk),
895 UDP_MIB_SNDBUFERRORS, is_udplite);
896 err = 0;
897 }
898 } else
899 UDP_INC_STATS(sock_net(sk),
900 UDP_MIB_OUTDATAGRAMS, is_udplite);
901 return err;
902}
903
904/*
905 * Push out all pending data as one UDP datagram. Socket is locked.
906 */
907int udp_push_pending_frames(struct sock *sk)
908{
909 struct udp_sock *up = udp_sk(sk);
910 struct inet_sock *inet = inet_sk(sk);
911 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
912 struct sk_buff *skb;
913 int err = 0;
914
915 skb = ip_finish_skb(sk, fl4);
916 if (!skb)
917 goto out;
918
919 err = udp_send_skb(skb, fl4, &inet->cork.base);
920
921out:
922 up->len = 0;
923 up->pending = 0;
924 return err;
925}
926EXPORT_SYMBOL(udp_push_pending_frames);
927
928static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
929{
930 switch (cmsg->cmsg_type) {
931 case UDP_SEGMENT:
932 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
933 return -EINVAL;
934 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
935 return 0;
936 default:
937 return -EINVAL;
938 }
939}
940
941int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
942{
943 struct cmsghdr *cmsg;
944 bool need_ip = false;
945 int err;
946
947 for_each_cmsghdr(cmsg, msg) {
948 if (!CMSG_OK(msg, cmsg))
949 return -EINVAL;
950
951 if (cmsg->cmsg_level != SOL_UDP) {
952 need_ip = true;
953 continue;
954 }
955
956 err = __udp_cmsg_send(cmsg, gso_size);
957 if (err)
958 return err;
959 }
960
961 return need_ip;
962}
963EXPORT_SYMBOL_GPL(udp_cmsg_send);
964
965int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
966{
967 struct inet_sock *inet = inet_sk(sk);
968 struct udp_sock *up = udp_sk(sk);
969 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
970 struct flowi4 fl4_stack;
971 struct flowi4 *fl4;
972 int ulen = len;
973 struct ipcm_cookie ipc;
974 struct rtable *rt = NULL;
975 int free = 0;
976 int connected = 0;
977 __be32 daddr, faddr, saddr;
978 __be16 dport;
979 u8 tos;
980 int err, is_udplite = IS_UDPLITE(sk);
981 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
982 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
983 struct sk_buff *skb;
984 struct ip_options_data opt_copy;
985
986 if (len > 0xFFFF)
987 return -EMSGSIZE;
988
989 /*
990 * Check the flags.
991 */
992
993 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
994 return -EOPNOTSUPP;
995
996 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
997
998 fl4 = &inet->cork.fl.u.ip4;
999 if (up->pending) {
1000 /*
1001 * There are pending frames.
1002 * The socket lock must be held while it's corked.
1003 */
1004 lock_sock(sk);
1005 if (likely(up->pending)) {
1006 if (unlikely(up->pending != AF_INET)) {
1007 release_sock(sk);
1008 return -EINVAL;
1009 }
1010 goto do_append_data;
1011 }
1012 release_sock(sk);
1013 }
1014 ulen += sizeof(struct udphdr);
1015
1016 /*
1017 * Get and verify the address.
1018 */
1019 if (usin) {
1020 if (msg->msg_namelen < sizeof(*usin))
1021 return -EINVAL;
1022 if (usin->sin_family != AF_INET) {
1023 if (usin->sin_family != AF_UNSPEC)
1024 return -EAFNOSUPPORT;
1025 }
1026
1027 daddr = usin->sin_addr.s_addr;
1028 dport = usin->sin_port;
1029 if (dport == 0)
1030 return -EINVAL;
1031 } else {
1032 if (sk->sk_state != TCP_ESTABLISHED)
1033 return -EDESTADDRREQ;
1034 daddr = inet->inet_daddr;
1035 dport = inet->inet_dport;
1036 /* Open fast path for connected socket.
1037 Route will not be used, if at least one option is set.
1038 */
1039 connected = 1;
1040 }
1041
1042 ipcm_init_sk(&ipc, inet);
1043 ipc.gso_size = up->gso_size;
1044
1045 if (msg->msg_controllen) {
1046 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1047 if (err > 0)
1048 err = ip_cmsg_send(sk, msg, &ipc,
1049 sk->sk_family == AF_INET6);
1050 if (unlikely(err < 0)) {
1051 kfree(ipc.opt);
1052 return err;
1053 }
1054 if (ipc.opt)
1055 free = 1;
1056 connected = 0;
1057 }
1058 if (!ipc.opt) {
1059 struct ip_options_rcu *inet_opt;
1060
1061 rcu_read_lock();
1062 inet_opt = rcu_dereference(inet->inet_opt);
1063 if (inet_opt) {
1064 memcpy(&opt_copy, inet_opt,
1065 sizeof(*inet_opt) + inet_opt->opt.optlen);
1066 ipc.opt = &opt_copy.opt;
1067 }
1068 rcu_read_unlock();
1069 }
1070
1071 if (cgroup_bpf_enabled && !connected) {
1072 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1073 (struct sockaddr *)usin, &ipc.addr);
1074 if (err)
1075 goto out_free;
1076 if (usin) {
1077 if (usin->sin_port == 0) {
1078 /* BPF program set invalid port. Reject it. */
1079 err = -EINVAL;
1080 goto out_free;
1081 }
1082 daddr = usin->sin_addr.s_addr;
1083 dport = usin->sin_port;
1084 }
1085 }
1086
1087 saddr = ipc.addr;
1088 ipc.addr = faddr = daddr;
1089
1090 if (ipc.opt && ipc.opt->opt.srr) {
1091 if (!daddr) {
1092 err = -EINVAL;
1093 goto out_free;
1094 }
1095 faddr = ipc.opt->opt.faddr;
1096 connected = 0;
1097 }
1098 tos = get_rttos(&ipc, inet);
1099 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1100 (msg->msg_flags & MSG_DONTROUTE) ||
1101 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1102 tos |= RTO_ONLINK;
1103 connected = 0;
1104 }
1105
1106 if (ipv4_is_multicast(daddr)) {
1107 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1108 ipc.oif = inet->mc_index;
1109 if (!saddr)
1110 saddr = inet->mc_addr;
1111 connected = 0;
1112 } else if (!ipc.oif) {
1113 ipc.oif = inet->uc_index;
1114 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1115 /* oif is set, packet is to local broadcast and
1116 * and uc_index is set. oif is most likely set
1117 * by sk_bound_dev_if. If uc_index != oif check if the
1118 * oif is an L3 master and uc_index is an L3 slave.
1119 * If so, we want to allow the send using the uc_index.
1120 */
1121 if (ipc.oif != inet->uc_index &&
1122 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1123 inet->uc_index)) {
1124 ipc.oif = inet->uc_index;
1125 }
1126 }
1127
1128 if (connected)
1129 rt = (struct rtable *)sk_dst_check(sk, 0);
1130
1131 if (!rt) {
1132 struct net *net = sock_net(sk);
1133 __u8 flow_flags = inet_sk_flowi_flags(sk);
1134
1135 fl4 = &fl4_stack;
1136
1137 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1138 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1139 flow_flags,
1140 faddr, saddr, dport, inet->inet_sport,
1141 sk->sk_uid);
1142
1143 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1144 rt = ip_route_output_flow(net, fl4, sk);
1145 if (IS_ERR(rt)) {
1146 err = PTR_ERR(rt);
1147 rt = NULL;
1148 if (err == -ENETUNREACH)
1149 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1150 goto out;
1151 }
1152
1153 err = -EACCES;
1154 if ((rt->rt_flags & RTCF_BROADCAST) &&
1155 !sock_flag(sk, SOCK_BROADCAST))
1156 goto out;
1157 if (connected)
1158 sk_dst_set(sk, dst_clone(&rt->dst));
1159 }
1160
1161 if (msg->msg_flags&MSG_CONFIRM)
1162 goto do_confirm;
1163back_from_confirm:
1164
1165 saddr = fl4->saddr;
1166 if (!ipc.addr)
1167 daddr = ipc.addr = fl4->daddr;
1168
1169 /* Lockless fast path for the non-corking case. */
1170 if (!corkreq) {
1171 struct inet_cork cork;
1172
1173 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1174 sizeof(struct udphdr), &ipc, &rt,
1175 &cork, msg->msg_flags);
1176 err = PTR_ERR(skb);
1177 if (!IS_ERR_OR_NULL(skb))
1178 err = udp_send_skb(skb, fl4, &cork);
1179 goto out;
1180 }
1181
1182 lock_sock(sk);
1183 if (unlikely(up->pending)) {
1184 /* The socket is already corked while preparing it. */
1185 /* ... which is an evident application bug. --ANK */
1186 release_sock(sk);
1187
1188 net_dbg_ratelimited("socket already corked\n");
1189 err = -EINVAL;
1190 goto out;
1191 }
1192 /*
1193 * Now cork the socket to pend data.
1194 */
1195 fl4 = &inet->cork.fl.u.ip4;
1196 fl4->daddr = daddr;
1197 fl4->saddr = saddr;
1198 fl4->fl4_dport = dport;
1199 fl4->fl4_sport = inet->inet_sport;
1200 up->pending = AF_INET;
1201
1202do_append_data:
1203 up->len += ulen;
1204 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1205 sizeof(struct udphdr), &ipc, &rt,
1206 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1207 if (err)
1208 udp_flush_pending_frames(sk);
1209 else if (!corkreq)
1210 err = udp_push_pending_frames(sk);
1211 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1212 up->pending = 0;
1213 release_sock(sk);
1214
1215out:
1216 ip_rt_put(rt);
1217out_free:
1218 if (free)
1219 kfree(ipc.opt);
1220 if (!err)
1221 return len;
1222 /*
1223 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1224 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1225 * we don't have a good statistic (IpOutDiscards but it can be too many
1226 * things). We could add another new stat but at least for now that
1227 * seems like overkill.
1228 */
1229 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1230 UDP_INC_STATS(sock_net(sk),
1231 UDP_MIB_SNDBUFERRORS, is_udplite);
1232 }
1233 return err;
1234
1235do_confirm:
1236 if (msg->msg_flags & MSG_PROBE)
1237 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1238 if (!(msg->msg_flags&MSG_PROBE) || len)
1239 goto back_from_confirm;
1240 err = 0;
1241 goto out;
1242}
1243EXPORT_SYMBOL(udp_sendmsg);
1244
1245int udp_sendpage(struct sock *sk, struct page *page, int offset,
1246 size_t size, int flags)
1247{
1248 struct inet_sock *inet = inet_sk(sk);
1249 struct udp_sock *up = udp_sk(sk);
1250 int ret;
1251
1252 if (flags & MSG_SENDPAGE_NOTLAST)
1253 flags |= MSG_MORE;
1254
1255 if (!up->pending) {
1256 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1257
1258 /* Call udp_sendmsg to specify destination address which
1259 * sendpage interface can't pass.
1260 * This will succeed only when the socket is connected.
1261 */
1262 ret = udp_sendmsg(sk, &msg, 0);
1263 if (ret < 0)
1264 return ret;
1265 }
1266
1267 lock_sock(sk);
1268
1269 if (unlikely(!up->pending)) {
1270 release_sock(sk);
1271
1272 net_dbg_ratelimited("cork failed\n");
1273 return -EINVAL;
1274 }
1275
1276 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1277 page, offset, size, flags);
1278 if (ret == -EOPNOTSUPP) {
1279 release_sock(sk);
1280 return sock_no_sendpage(sk->sk_socket, page, offset,
1281 size, flags);
1282 }
1283 if (ret < 0) {
1284 udp_flush_pending_frames(sk);
1285 goto out;
1286 }
1287
1288 up->len += size;
1289 if (!(up->corkflag || (flags&MSG_MORE)))
1290 ret = udp_push_pending_frames(sk);
1291 if (!ret)
1292 ret = size;
1293out:
1294 release_sock(sk);
1295 return ret;
1296}
1297
1298#define UDP_SKB_IS_STATELESS 0x80000000
1299
1300/* all head states (dst, sk, nf conntrack) except skb extensions are
1301 * cleared by udp_rcv().
1302 *
1303 * We need to preserve secpath, if present, to eventually process
1304 * IP_CMSG_PASSSEC at recvmsg() time.
1305 *
1306 * Other extensions can be cleared.
1307 */
1308static bool udp_try_make_stateless(struct sk_buff *skb)
1309{
1310 if (!skb_has_extensions(skb))
1311 return true;
1312
1313 if (!secpath_exists(skb)) {
1314 skb_ext_reset(skb);
1315 return true;
1316 }
1317
1318 return false;
1319}
1320
1321static void udp_set_dev_scratch(struct sk_buff *skb)
1322{
1323 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1324
1325 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1326 scratch->_tsize_state = skb->truesize;
1327#if BITS_PER_LONG == 64
1328 scratch->len = skb->len;
1329 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1330 scratch->is_linear = !skb_is_nonlinear(skb);
1331#endif
1332 if (udp_try_make_stateless(skb))
1333 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1334}
1335
1336static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1337{
1338 /* We come here after udp_lib_checksum_complete() returned 0.
1339 * This means that __skb_checksum_complete() might have
1340 * set skb->csum_valid to 1.
1341 * On 64bit platforms, we can set csum_unnecessary
1342 * to true, but only if the skb is not shared.
1343 */
1344#if BITS_PER_LONG == 64
1345 if (!skb_shared(skb))
1346 udp_skb_scratch(skb)->csum_unnecessary = true;
1347#endif
1348}
1349
1350static int udp_skb_truesize(struct sk_buff *skb)
1351{
1352 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1353}
1354
1355static bool udp_skb_has_head_state(struct sk_buff *skb)
1356{
1357 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1358}
1359
1360/* fully reclaim rmem/fwd memory allocated for skb */
1361static void udp_rmem_release(struct sock *sk, int size, int partial,
1362 bool rx_queue_lock_held)
1363{
1364 struct udp_sock *up = udp_sk(sk);
1365 struct sk_buff_head *sk_queue;
1366 int amt;
1367
1368 if (likely(partial)) {
1369 up->forward_deficit += size;
1370 size = up->forward_deficit;
1371 if (size < (sk->sk_rcvbuf >> 2))
1372 return;
1373 } else {
1374 size += up->forward_deficit;
1375 }
1376 up->forward_deficit = 0;
1377
1378 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1379 * if the called don't held it already
1380 */
1381 sk_queue = &sk->sk_receive_queue;
1382 if (!rx_queue_lock_held)
1383 spin_lock(&sk_queue->lock);
1384
1385
1386 sk->sk_forward_alloc += size;
1387 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1388 sk->sk_forward_alloc -= amt;
1389
1390 if (amt)
1391 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1392
1393 atomic_sub(size, &sk->sk_rmem_alloc);
1394
1395 /* this can save us from acquiring the rx queue lock on next receive */
1396 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1397
1398 if (!rx_queue_lock_held)
1399 spin_unlock(&sk_queue->lock);
1400}
1401
1402/* Note: called with reader_queue.lock held.
1403 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1404 * This avoids a cache line miss while receive_queue lock is held.
1405 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1406 */
1407void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1408{
1409 prefetch(&skb->data);
1410 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1411}
1412EXPORT_SYMBOL(udp_skb_destructor);
1413
1414/* as above, but the caller held the rx queue lock, too */
1415static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1416{
1417 prefetch(&skb->data);
1418 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1419}
1420
1421/* Idea of busylocks is to let producers grab an extra spinlock
1422 * to relieve pressure on the receive_queue spinlock shared by consumer.
1423 * Under flood, this means that only one producer can be in line
1424 * trying to acquire the receive_queue spinlock.
1425 * These busylock can be allocated on a per cpu manner, instead of a
1426 * per socket one (that would consume a cache line per socket)
1427 */
1428static int udp_busylocks_log __read_mostly;
1429static spinlock_t *udp_busylocks __read_mostly;
1430
1431static spinlock_t *busylock_acquire(void *ptr)
1432{
1433 spinlock_t *busy;
1434
1435 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1436 spin_lock(busy);
1437 return busy;
1438}
1439
1440static void busylock_release(spinlock_t *busy)
1441{
1442 if (busy)
1443 spin_unlock(busy);
1444}
1445
1446int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1447{
1448 struct sk_buff_head *list = &sk->sk_receive_queue;
1449 int rmem, delta, amt, err = -ENOMEM;
1450 spinlock_t *busy = NULL;
1451 int size;
1452
1453 /* try to avoid the costly atomic add/sub pair when the receive
1454 * queue is full; always allow at least a packet
1455 */
1456 rmem = atomic_read(&sk->sk_rmem_alloc);
1457 if (rmem > sk->sk_rcvbuf)
1458 goto drop;
1459
1460 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1461 * having linear skbs :
1462 * - Reduce memory overhead and thus increase receive queue capacity
1463 * - Less cache line misses at copyout() time
1464 * - Less work at consume_skb() (less alien page frag freeing)
1465 */
1466 if (rmem > (sk->sk_rcvbuf >> 1)) {
1467 skb_condense(skb);
1468
1469 busy = busylock_acquire(sk);
1470 }
1471 size = skb->truesize;
1472 udp_set_dev_scratch(skb);
1473
1474 /* we drop only if the receive buf is full and the receive
1475 * queue contains some other skb
1476 */
1477 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1478 if (rmem > (size + sk->sk_rcvbuf))
1479 goto uncharge_drop;
1480
1481 spin_lock(&list->lock);
1482 if (size >= sk->sk_forward_alloc) {
1483 amt = sk_mem_pages(size);
1484 delta = amt << SK_MEM_QUANTUM_SHIFT;
1485 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1486 err = -ENOBUFS;
1487 spin_unlock(&list->lock);
1488 goto uncharge_drop;
1489 }
1490
1491 sk->sk_forward_alloc += delta;
1492 }
1493
1494 sk->sk_forward_alloc -= size;
1495
1496 /* no need to setup a destructor, we will explicitly release the
1497 * forward allocated memory on dequeue
1498 */
1499 sock_skb_set_dropcount(sk, skb);
1500
1501 __skb_queue_tail(list, skb);
1502 spin_unlock(&list->lock);
1503
1504 if (!sock_flag(sk, SOCK_DEAD))
1505 sk->sk_data_ready(sk);
1506
1507 busylock_release(busy);
1508 return 0;
1509
1510uncharge_drop:
1511 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1512
1513drop:
1514 atomic_inc(&sk->sk_drops);
1515 busylock_release(busy);
1516 return err;
1517}
1518EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1519
1520void udp_destruct_sock(struct sock *sk)
1521{
1522 /* reclaim completely the forward allocated memory */
1523 struct udp_sock *up = udp_sk(sk);
1524 unsigned int total = 0;
1525 struct sk_buff *skb;
1526
1527 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1528 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1529 total += skb->truesize;
1530 kfree_skb(skb);
1531 }
1532 udp_rmem_release(sk, total, 0, true);
1533
1534 inet_sock_destruct(sk);
1535}
1536EXPORT_SYMBOL_GPL(udp_destruct_sock);
1537
1538int udp_init_sock(struct sock *sk)
1539{
1540 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1541 sk->sk_destruct = udp_destruct_sock;
1542 return 0;
1543}
1544EXPORT_SYMBOL_GPL(udp_init_sock);
1545
1546void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1547{
1548 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1549 bool slow = lock_sock_fast(sk);
1550
1551 sk_peek_offset_bwd(sk, len);
1552 unlock_sock_fast(sk, slow);
1553 }
1554
1555 if (!skb_unref(skb))
1556 return;
1557
1558 /* In the more common cases we cleared the head states previously,
1559 * see __udp_queue_rcv_skb().
1560 */
1561 if (unlikely(udp_skb_has_head_state(skb)))
1562 skb_release_head_state(skb);
1563 __consume_stateless_skb(skb);
1564}
1565EXPORT_SYMBOL_GPL(skb_consume_udp);
1566
1567static struct sk_buff *__first_packet_length(struct sock *sk,
1568 struct sk_buff_head *rcvq,
1569 int *total)
1570{
1571 struct sk_buff *skb;
1572
1573 while ((skb = skb_peek(rcvq)) != NULL) {
1574 if (udp_lib_checksum_complete(skb)) {
1575 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1576 IS_UDPLITE(sk));
1577 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1578 IS_UDPLITE(sk));
1579 atomic_inc(&sk->sk_drops);
1580 __skb_unlink(skb, rcvq);
1581 *total += skb->truesize;
1582 kfree_skb(skb);
1583 } else {
1584 udp_skb_csum_unnecessary_set(skb);
1585 break;
1586 }
1587 }
1588 return skb;
1589}
1590
1591/**
1592 * first_packet_length - return length of first packet in receive queue
1593 * @sk: socket
1594 *
1595 * Drops all bad checksum frames, until a valid one is found.
1596 * Returns the length of found skb, or -1 if none is found.
1597 */
1598static int first_packet_length(struct sock *sk)
1599{
1600 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1601 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1602 struct sk_buff *skb;
1603 int total = 0;
1604 int res;
1605
1606 spin_lock_bh(&rcvq->lock);
1607 skb = __first_packet_length(sk, rcvq, &total);
1608 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1609 spin_lock(&sk_queue->lock);
1610 skb_queue_splice_tail_init(sk_queue, rcvq);
1611 spin_unlock(&sk_queue->lock);
1612
1613 skb = __first_packet_length(sk, rcvq, &total);
1614 }
1615 res = skb ? skb->len : -1;
1616 if (total)
1617 udp_rmem_release(sk, total, 1, false);
1618 spin_unlock_bh(&rcvq->lock);
1619 return res;
1620}
1621
1622/*
1623 * IOCTL requests applicable to the UDP protocol
1624 */
1625
1626int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1627{
1628 switch (cmd) {
1629 case SIOCOUTQ:
1630 {
1631 int amount = sk_wmem_alloc_get(sk);
1632
1633 return put_user(amount, (int __user *)arg);
1634 }
1635
1636 case SIOCINQ:
1637 {
1638 int amount = max_t(int, 0, first_packet_length(sk));
1639
1640 return put_user(amount, (int __user *)arg);
1641 }
1642
1643 default:
1644 return -ENOIOCTLCMD;
1645 }
1646
1647 return 0;
1648}
1649EXPORT_SYMBOL(udp_ioctl);
1650
1651struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1652 int noblock, int *off, int *err)
1653{
1654 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1655 struct sk_buff_head *queue;
1656 struct sk_buff *last;
1657 long timeo;
1658 int error;
1659
1660 queue = &udp_sk(sk)->reader_queue;
1661 flags |= noblock ? MSG_DONTWAIT : 0;
1662 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1663 do {
1664 struct sk_buff *skb;
1665
1666 error = sock_error(sk);
1667 if (error)
1668 break;
1669
1670 error = -EAGAIN;
1671 do {
1672 spin_lock_bh(&queue->lock);
1673 skb = __skb_try_recv_from_queue(sk, queue, flags,
1674 udp_skb_destructor,
1675 off, err, &last);
1676 if (skb) {
1677 spin_unlock_bh(&queue->lock);
1678 return skb;
1679 }
1680
1681 if (skb_queue_empty_lockless(sk_queue)) {
1682 spin_unlock_bh(&queue->lock);
1683 goto busy_check;
1684 }
1685
1686 /* refill the reader queue and walk it again
1687 * keep both queues locked to avoid re-acquiring
1688 * the sk_receive_queue lock if fwd memory scheduling
1689 * is needed.
1690 */
1691 spin_lock(&sk_queue->lock);
1692 skb_queue_splice_tail_init(sk_queue, queue);
1693
1694 skb = __skb_try_recv_from_queue(sk, queue, flags,
1695 udp_skb_dtor_locked,
1696 off, err, &last);
1697 spin_unlock(&sk_queue->lock);
1698 spin_unlock_bh(&queue->lock);
1699 if (skb)
1700 return skb;
1701
1702busy_check:
1703 if (!sk_can_busy_loop(sk))
1704 break;
1705
1706 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1707 } while (!skb_queue_empty_lockless(sk_queue));
1708
1709 /* sk_queue is empty, reader_queue may contain peeked packets */
1710 } while (timeo &&
1711 !__skb_wait_for_more_packets(sk, &error, &timeo,
1712 (struct sk_buff *)sk_queue));
1713
1714 *err = error;
1715 return NULL;
1716}
1717EXPORT_SYMBOL(__skb_recv_udp);
1718
1719/*
1720 * This should be easy, if there is something there we
1721 * return it, otherwise we block.
1722 */
1723
1724int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1725 int flags, int *addr_len)
1726{
1727 struct inet_sock *inet = inet_sk(sk);
1728 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1729 struct sk_buff *skb;
1730 unsigned int ulen, copied;
1731 int off, err, peeking = flags & MSG_PEEK;
1732 int is_udplite = IS_UDPLITE(sk);
1733 bool checksum_valid = false;
1734
1735 if (flags & MSG_ERRQUEUE)
1736 return ip_recv_error(sk, msg, len, addr_len);
1737
1738try_again:
1739 off = sk_peek_offset(sk, flags);
1740 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1741 if (!skb)
1742 return err;
1743
1744 ulen = udp_skb_len(skb);
1745 copied = len;
1746 if (copied > ulen - off)
1747 copied = ulen - off;
1748 else if (copied < ulen)
1749 msg->msg_flags |= MSG_TRUNC;
1750
1751 /*
1752 * If checksum is needed at all, try to do it while copying the
1753 * data. If the data is truncated, or if we only want a partial
1754 * coverage checksum (UDP-Lite), do it before the copy.
1755 */
1756
1757 if (copied < ulen || peeking ||
1758 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1759 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1760 !__udp_lib_checksum_complete(skb);
1761 if (!checksum_valid)
1762 goto csum_copy_err;
1763 }
1764
1765 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1766 if (udp_skb_is_linear(skb))
1767 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1768 else
1769 err = skb_copy_datagram_msg(skb, off, msg, copied);
1770 } else {
1771 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1772
1773 if (err == -EINVAL)
1774 goto csum_copy_err;
1775 }
1776
1777 if (unlikely(err)) {
1778 if (!peeking) {
1779 atomic_inc(&sk->sk_drops);
1780 UDP_INC_STATS(sock_net(sk),
1781 UDP_MIB_INERRORS, is_udplite);
1782 }
1783 kfree_skb(skb);
1784 return err;
1785 }
1786
1787 if (!peeking)
1788 UDP_INC_STATS(sock_net(sk),
1789 UDP_MIB_INDATAGRAMS, is_udplite);
1790
1791 sock_recv_ts_and_drops(msg, sk, skb);
1792
1793 /* Copy the address. */
1794 if (sin) {
1795 sin->sin_family = AF_INET;
1796 sin->sin_port = udp_hdr(skb)->source;
1797 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1798 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1799 *addr_len = sizeof(*sin);
1800
1801 if (cgroup_bpf_enabled)
1802 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1803 (struct sockaddr *)sin);
1804 }
1805
1806 if (udp_sk(sk)->gro_enabled)
1807 udp_cmsg_recv(msg, sk, skb);
1808
1809 if (inet->cmsg_flags)
1810 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1811
1812 err = copied;
1813 if (flags & MSG_TRUNC)
1814 err = ulen;
1815
1816 skb_consume_udp(sk, skb, peeking ? -err : err);
1817 return err;
1818
1819csum_copy_err:
1820 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1821 udp_skb_destructor)) {
1822 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1823 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1824 }
1825 kfree_skb(skb);
1826
1827 /* starting over for a new packet, but check if we need to yield */
1828 cond_resched();
1829 msg->msg_flags &= ~MSG_TRUNC;
1830 goto try_again;
1831}
1832
1833int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1834{
1835 /* This check is replicated from __ip4_datagram_connect() and
1836 * intended to prevent BPF program called below from accessing bytes
1837 * that are out of the bound specified by user in addr_len.
1838 */
1839 if (addr_len < sizeof(struct sockaddr_in))
1840 return -EINVAL;
1841
1842 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1843}
1844EXPORT_SYMBOL(udp_pre_connect);
1845
1846int __udp_disconnect(struct sock *sk, int flags)
1847{
1848 struct inet_sock *inet = inet_sk(sk);
1849 /*
1850 * 1003.1g - break association.
1851 */
1852
1853 sk->sk_state = TCP_CLOSE;
1854 inet->inet_daddr = 0;
1855 inet->inet_dport = 0;
1856 sock_rps_reset_rxhash(sk);
1857 sk->sk_bound_dev_if = 0;
1858 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1859 inet_reset_saddr(sk);
1860
1861 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1862 sk->sk_prot->unhash(sk);
1863 inet->inet_sport = 0;
1864 }
1865 sk_dst_reset(sk);
1866 return 0;
1867}
1868EXPORT_SYMBOL(__udp_disconnect);
1869
1870int udp_disconnect(struct sock *sk, int flags)
1871{
1872 lock_sock(sk);
1873 __udp_disconnect(sk, flags);
1874 release_sock(sk);
1875 return 0;
1876}
1877EXPORT_SYMBOL(udp_disconnect);
1878
1879void udp_lib_unhash(struct sock *sk)
1880{
1881 if (sk_hashed(sk)) {
1882 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1883 struct udp_hslot *hslot, *hslot2;
1884
1885 hslot = udp_hashslot(udptable, sock_net(sk),
1886 udp_sk(sk)->udp_port_hash);
1887 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1888
1889 spin_lock_bh(&hslot->lock);
1890 if (rcu_access_pointer(sk->sk_reuseport_cb))
1891 reuseport_detach_sock(sk);
1892 if (sk_del_node_init_rcu(sk)) {
1893 hslot->count--;
1894 inet_sk(sk)->inet_num = 0;
1895 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1896
1897 spin_lock(&hslot2->lock);
1898 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1899 hslot2->count--;
1900 spin_unlock(&hslot2->lock);
1901 }
1902 spin_unlock_bh(&hslot->lock);
1903 }
1904}
1905EXPORT_SYMBOL(udp_lib_unhash);
1906
1907/*
1908 * inet_rcv_saddr was changed, we must rehash secondary hash
1909 */
1910void udp_lib_rehash(struct sock *sk, u16 newhash)
1911{
1912 if (sk_hashed(sk)) {
1913 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1914 struct udp_hslot *hslot, *hslot2, *nhslot2;
1915
1916 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1917 nhslot2 = udp_hashslot2(udptable, newhash);
1918 udp_sk(sk)->udp_portaddr_hash = newhash;
1919
1920 if (hslot2 != nhslot2 ||
1921 rcu_access_pointer(sk->sk_reuseport_cb)) {
1922 hslot = udp_hashslot(udptable, sock_net(sk),
1923 udp_sk(sk)->udp_port_hash);
1924 /* we must lock primary chain too */
1925 spin_lock_bh(&hslot->lock);
1926 if (rcu_access_pointer(sk->sk_reuseport_cb))
1927 reuseport_detach_sock(sk);
1928
1929 if (hslot2 != nhslot2) {
1930 spin_lock(&hslot2->lock);
1931 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1932 hslot2->count--;
1933 spin_unlock(&hslot2->lock);
1934
1935 spin_lock(&nhslot2->lock);
1936 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1937 &nhslot2->head);
1938 nhslot2->count++;
1939 spin_unlock(&nhslot2->lock);
1940 }
1941
1942 spin_unlock_bh(&hslot->lock);
1943 }
1944 }
1945}
1946EXPORT_SYMBOL(udp_lib_rehash);
1947
1948void udp_v4_rehash(struct sock *sk)
1949{
1950 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1951 inet_sk(sk)->inet_rcv_saddr,
1952 inet_sk(sk)->inet_num);
1953 udp_lib_rehash(sk, new_hash);
1954}
1955
1956static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1957{
1958 int rc;
1959
1960 if (inet_sk(sk)->inet_daddr) {
1961 sock_rps_save_rxhash(sk, skb);
1962 sk_mark_napi_id(sk, skb);
1963 sk_incoming_cpu_update(sk);
1964 } else {
1965 sk_mark_napi_id_once(sk, skb);
1966 }
1967
1968 rc = __udp_enqueue_schedule_skb(sk, skb);
1969 if (rc < 0) {
1970 int is_udplite = IS_UDPLITE(sk);
1971
1972 /* Note that an ENOMEM error is charged twice */
1973 if (rc == -ENOMEM)
1974 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1975 is_udplite);
1976 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1977 kfree_skb(skb);
1978 trace_udp_fail_queue_rcv_skb(rc, sk);
1979 return -1;
1980 }
1981
1982 return 0;
1983}
1984
1985/* returns:
1986 * -1: error
1987 * 0: success
1988 * >0: "udp encap" protocol resubmission
1989 *
1990 * Note that in the success and error cases, the skb is assumed to
1991 * have either been requeued or freed.
1992 */
1993static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
1994{
1995 struct udp_sock *up = udp_sk(sk);
1996 int is_udplite = IS_UDPLITE(sk);
1997
1998 /*
1999 * Charge it to the socket, dropping if the queue is full.
2000 */
2001 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2002 goto drop;
2003 nf_reset_ct(skb);
2004
2005 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2006 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2007
2008 /*
2009 * This is an encapsulation socket so pass the skb to
2010 * the socket's udp_encap_rcv() hook. Otherwise, just
2011 * fall through and pass this up the UDP socket.
2012 * up->encap_rcv() returns the following value:
2013 * =0 if skb was successfully passed to the encap
2014 * handler or was discarded by it.
2015 * >0 if skb should be passed on to UDP.
2016 * <0 if skb should be resubmitted as proto -N
2017 */
2018
2019 /* if we're overly short, let UDP handle it */
2020 encap_rcv = READ_ONCE(up->encap_rcv);
2021 if (encap_rcv) {
2022 int ret;
2023
2024 /* Verify checksum before giving to encap */
2025 if (udp_lib_checksum_complete(skb))
2026 goto csum_error;
2027
2028 ret = encap_rcv(sk, skb);
2029 if (ret <= 0) {
2030 __UDP_INC_STATS(sock_net(sk),
2031 UDP_MIB_INDATAGRAMS,
2032 is_udplite);
2033 return -ret;
2034 }
2035 }
2036
2037 /* FALLTHROUGH -- it's a UDP Packet */
2038 }
2039
2040 /*
2041 * UDP-Lite specific tests, ignored on UDP sockets
2042 */
2043 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2044
2045 /*
2046 * MIB statistics other than incrementing the error count are
2047 * disabled for the following two types of errors: these depend
2048 * on the application settings, not on the functioning of the
2049 * protocol stack as such.
2050 *
2051 * RFC 3828 here recommends (sec 3.3): "There should also be a
2052 * way ... to ... at least let the receiving application block
2053 * delivery of packets with coverage values less than a value
2054 * provided by the application."
2055 */
2056 if (up->pcrlen == 0) { /* full coverage was set */
2057 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2058 UDP_SKB_CB(skb)->cscov, skb->len);
2059 goto drop;
2060 }
2061 /* The next case involves violating the min. coverage requested
2062 * by the receiver. This is subtle: if receiver wants x and x is
2063 * greater than the buffersize/MTU then receiver will complain
2064 * that it wants x while sender emits packets of smaller size y.
2065 * Therefore the above ...()->partial_cov statement is essential.
2066 */
2067 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2068 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2069 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2070 goto drop;
2071 }
2072 }
2073
2074 prefetch(&sk->sk_rmem_alloc);
2075 if (rcu_access_pointer(sk->sk_filter) &&
2076 udp_lib_checksum_complete(skb))
2077 goto csum_error;
2078
2079 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2080 goto drop;
2081
2082 udp_csum_pull_header(skb);
2083
2084 ipv4_pktinfo_prepare(sk, skb);
2085 return __udp_queue_rcv_skb(sk, skb);
2086
2087csum_error:
2088 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2089drop:
2090 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2091 atomic_inc(&sk->sk_drops);
2092 kfree_skb(skb);
2093 return -1;
2094}
2095
2096static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2097{
2098 struct sk_buff *next, *segs;
2099 int ret;
2100
2101 if (likely(!udp_unexpected_gso(sk, skb)))
2102 return udp_queue_rcv_one_skb(sk, skb);
2103
2104 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET);
2105 __skb_push(skb, -skb_mac_offset(skb));
2106 segs = udp_rcv_segment(sk, skb, true);
2107 for (skb = segs; skb; skb = next) {
2108 next = skb->next;
2109 __skb_pull(skb, skb_transport_offset(skb));
2110 ret = udp_queue_rcv_one_skb(sk, skb);
2111 if (ret > 0)
2112 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2113 }
2114 return 0;
2115}
2116
2117/* For TCP sockets, sk_rx_dst is protected by socket lock
2118 * For UDP, we use xchg() to guard against concurrent changes.
2119 */
2120bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2121{
2122 struct dst_entry *old;
2123
2124 if (dst_hold_safe(dst)) {
2125 old = xchg(&sk->sk_rx_dst, dst);
2126 dst_release(old);
2127 return old != dst;
2128 }
2129 return false;
2130}
2131EXPORT_SYMBOL(udp_sk_rx_dst_set);
2132
2133/*
2134 * Multicasts and broadcasts go to each listener.
2135 *
2136 * Note: called only from the BH handler context.
2137 */
2138static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2139 struct udphdr *uh,
2140 __be32 saddr, __be32 daddr,
2141 struct udp_table *udptable,
2142 int proto)
2143{
2144 struct sock *sk, *first = NULL;
2145 unsigned short hnum = ntohs(uh->dest);
2146 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2147 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2148 unsigned int offset = offsetof(typeof(*sk), sk_node);
2149 int dif = skb->dev->ifindex;
2150 int sdif = inet_sdif(skb);
2151 struct hlist_node *node;
2152 struct sk_buff *nskb;
2153
2154 if (use_hash2) {
2155 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2156 udptable->mask;
2157 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2158start_lookup:
2159 hslot = &udptable->hash2[hash2];
2160 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2161 }
2162
2163 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2164 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2165 uh->source, saddr, dif, sdif, hnum))
2166 continue;
2167
2168 if (!first) {
2169 first = sk;
2170 continue;
2171 }
2172 nskb = skb_clone(skb, GFP_ATOMIC);
2173
2174 if (unlikely(!nskb)) {
2175 atomic_inc(&sk->sk_drops);
2176 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2177 IS_UDPLITE(sk));
2178 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2179 IS_UDPLITE(sk));
2180 continue;
2181 }
2182 if (udp_queue_rcv_skb(sk, nskb) > 0)
2183 consume_skb(nskb);
2184 }
2185
2186 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2187 if (use_hash2 && hash2 != hash2_any) {
2188 hash2 = hash2_any;
2189 goto start_lookup;
2190 }
2191
2192 if (first) {
2193 if (udp_queue_rcv_skb(first, skb) > 0)
2194 consume_skb(skb);
2195 } else {
2196 kfree_skb(skb);
2197 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2198 proto == IPPROTO_UDPLITE);
2199 }
2200 return 0;
2201}
2202
2203/* Initialize UDP checksum. If exited with zero value (success),
2204 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2205 * Otherwise, csum completion requires checksumming packet body,
2206 * including udp header and folding it to skb->csum.
2207 */
2208static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2209 int proto)
2210{
2211 int err;
2212
2213 UDP_SKB_CB(skb)->partial_cov = 0;
2214 UDP_SKB_CB(skb)->cscov = skb->len;
2215
2216 if (proto == IPPROTO_UDPLITE) {
2217 err = udplite_checksum_init(skb, uh);
2218 if (err)
2219 return err;
2220
2221 if (UDP_SKB_CB(skb)->partial_cov) {
2222 skb->csum = inet_compute_pseudo(skb, proto);
2223 return 0;
2224 }
2225 }
2226
2227 /* Note, we are only interested in != 0 or == 0, thus the
2228 * force to int.
2229 */
2230 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2231 inet_compute_pseudo);
2232 if (err)
2233 return err;
2234
2235 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2236 /* If SW calculated the value, we know it's bad */
2237 if (skb->csum_complete_sw)
2238 return 1;
2239
2240 /* HW says the value is bad. Let's validate that.
2241 * skb->csum is no longer the full packet checksum,
2242 * so don't treat it as such.
2243 */
2244 skb_checksum_complete_unset(skb);
2245 }
2246
2247 return 0;
2248}
2249
2250/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2251 * return code conversion for ip layer consumption
2252 */
2253static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2254 struct udphdr *uh)
2255{
2256 int ret;
2257
2258 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2259 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2260
2261 ret = udp_queue_rcv_skb(sk, skb);
2262
2263 /* a return value > 0 means to resubmit the input, but
2264 * it wants the return to be -protocol, or 0
2265 */
2266 if (ret > 0)
2267 return -ret;
2268 return 0;
2269}
2270
2271/*
2272 * All we need to do is get the socket, and then do a checksum.
2273 */
2274
2275int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2276 int proto)
2277{
2278 struct sock *sk;
2279 struct udphdr *uh;
2280 unsigned short ulen;
2281 struct rtable *rt = skb_rtable(skb);
2282 __be32 saddr, daddr;
2283 struct net *net = dev_net(skb->dev);
2284
2285 /*
2286 * Validate the packet.
2287 */
2288 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2289 goto drop; /* No space for header. */
2290
2291 uh = udp_hdr(skb);
2292 ulen = ntohs(uh->len);
2293 saddr = ip_hdr(skb)->saddr;
2294 daddr = ip_hdr(skb)->daddr;
2295
2296 if (ulen > skb->len)
2297 goto short_packet;
2298
2299 if (proto == IPPROTO_UDP) {
2300 /* UDP validates ulen. */
2301 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2302 goto short_packet;
2303 uh = udp_hdr(skb);
2304 }
2305
2306 if (udp4_csum_init(skb, uh, proto))
2307 goto csum_error;
2308
2309 sk = skb_steal_sock(skb);
2310 if (sk) {
2311 struct dst_entry *dst = skb_dst(skb);
2312 int ret;
2313
2314 if (unlikely(sk->sk_rx_dst != dst))
2315 udp_sk_rx_dst_set(sk, dst);
2316
2317 ret = udp_unicast_rcv_skb(sk, skb, uh);
2318 sock_put(sk);
2319 return ret;
2320 }
2321
2322 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2323 return __udp4_lib_mcast_deliver(net, skb, uh,
2324 saddr, daddr, udptable, proto);
2325
2326 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2327 if (sk)
2328 return udp_unicast_rcv_skb(sk, skb, uh);
2329
2330 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2331 goto drop;
2332 nf_reset_ct(skb);
2333
2334 /* No socket. Drop packet silently, if checksum is wrong */
2335 if (udp_lib_checksum_complete(skb))
2336 goto csum_error;
2337
2338 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2339 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2340
2341 /*
2342 * Hmm. We got an UDP packet to a port to which we
2343 * don't wanna listen. Ignore it.
2344 */
2345 kfree_skb(skb);
2346 return 0;
2347
2348short_packet:
2349 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2350 proto == IPPROTO_UDPLITE ? "Lite" : "",
2351 &saddr, ntohs(uh->source),
2352 ulen, skb->len,
2353 &daddr, ntohs(uh->dest));
2354 goto drop;
2355
2356csum_error:
2357 /*
2358 * RFC1122: OK. Discards the bad packet silently (as far as
2359 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2360 */
2361 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2362 proto == IPPROTO_UDPLITE ? "Lite" : "",
2363 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2364 ulen);
2365 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2366drop:
2367 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2368 kfree_skb(skb);
2369 return 0;
2370}
2371
2372/* We can only early demux multicast if there is a single matching socket.
2373 * If more than one socket found returns NULL
2374 */
2375static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2376 __be16 loc_port, __be32 loc_addr,
2377 __be16 rmt_port, __be32 rmt_addr,
2378 int dif, int sdif)
2379{
2380 struct sock *sk, *result;
2381 unsigned short hnum = ntohs(loc_port);
2382 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2383 struct udp_hslot *hslot = &udp_table.hash[slot];
2384
2385 /* Do not bother scanning a too big list */
2386 if (hslot->count > 10)
2387 return NULL;
2388
2389 result = NULL;
2390 sk_for_each_rcu(sk, &hslot->head) {
2391 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2392 rmt_port, rmt_addr, dif, sdif, hnum)) {
2393 if (result)
2394 return NULL;
2395 result = sk;
2396 }
2397 }
2398
2399 return result;
2400}
2401
2402/* For unicast we should only early demux connected sockets or we can
2403 * break forwarding setups. The chains here can be long so only check
2404 * if the first socket is an exact match and if not move on.
2405 */
2406static struct sock *__udp4_lib_demux_lookup(struct net *net,
2407 __be16 loc_port, __be32 loc_addr,
2408 __be16 rmt_port, __be32 rmt_addr,
2409 int dif, int sdif)
2410{
2411 unsigned short hnum = ntohs(loc_port);
2412 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2413 unsigned int slot2 = hash2 & udp_table.mask;
2414 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2415 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2416 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2417 struct sock *sk;
2418
2419 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2420 if (INET_MATCH(sk, net, acookie, rmt_addr,
2421 loc_addr, ports, dif, sdif))
2422 return sk;
2423 /* Only check first socket in chain */
2424 break;
2425 }
2426 return NULL;
2427}
2428
2429int udp_v4_early_demux(struct sk_buff *skb)
2430{
2431 struct net *net = dev_net(skb->dev);
2432 struct in_device *in_dev = NULL;
2433 const struct iphdr *iph;
2434 const struct udphdr *uh;
2435 struct sock *sk = NULL;
2436 struct dst_entry *dst;
2437 int dif = skb->dev->ifindex;
2438 int sdif = inet_sdif(skb);
2439 int ours;
2440
2441 /* validate the packet */
2442 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2443 return 0;
2444
2445 iph = ip_hdr(skb);
2446 uh = udp_hdr(skb);
2447
2448 if (skb->pkt_type == PACKET_MULTICAST) {
2449 in_dev = __in_dev_get_rcu(skb->dev);
2450
2451 if (!in_dev)
2452 return 0;
2453
2454 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2455 iph->protocol);
2456 if (!ours)
2457 return 0;
2458
2459 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2460 uh->source, iph->saddr,
2461 dif, sdif);
2462 } else if (skb->pkt_type == PACKET_HOST) {
2463 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2464 uh->source, iph->saddr, dif, sdif);
2465 }
2466
2467 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2468 return 0;
2469
2470 skb->sk = sk;
2471 skb->destructor = sock_efree;
2472 dst = READ_ONCE(sk->sk_rx_dst);
2473
2474 if (dst)
2475 dst = dst_check(dst, 0);
2476 if (dst) {
2477 u32 itag = 0;
2478
2479 /* set noref for now.
2480 * any place which wants to hold dst has to call
2481 * dst_hold_safe()
2482 */
2483 skb_dst_set_noref(skb, dst);
2484
2485 /* for unconnected multicast sockets we need to validate
2486 * the source on each packet
2487 */
2488 if (!inet_sk(sk)->inet_daddr && in_dev)
2489 return ip_mc_validate_source(skb, iph->daddr,
2490 iph->saddr, iph->tos,
2491 skb->dev, in_dev, &itag);
2492 }
2493 return 0;
2494}
2495
2496int udp_rcv(struct sk_buff *skb)
2497{
2498 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2499}
2500
2501void udp_destroy_sock(struct sock *sk)
2502{
2503 struct udp_sock *up = udp_sk(sk);
2504 bool slow = lock_sock_fast(sk);
2505 udp_flush_pending_frames(sk);
2506 unlock_sock_fast(sk, slow);
2507 if (static_branch_unlikely(&udp_encap_needed_key)) {
2508 if (up->encap_type) {
2509 void (*encap_destroy)(struct sock *sk);
2510 encap_destroy = READ_ONCE(up->encap_destroy);
2511 if (encap_destroy)
2512 encap_destroy(sk);
2513 }
2514 if (up->encap_enabled)
2515 static_branch_dec(&udp_encap_needed_key);
2516 }
2517}
2518
2519/*
2520 * Socket option code for UDP
2521 */
2522int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2523 char __user *optval, unsigned int optlen,
2524 int (*push_pending_frames)(struct sock *))
2525{
2526 struct udp_sock *up = udp_sk(sk);
2527 int val, valbool;
2528 int err = 0;
2529 int is_udplite = IS_UDPLITE(sk);
2530
2531 if (optlen < sizeof(int))
2532 return -EINVAL;
2533
2534 if (get_user(val, (int __user *)optval))
2535 return -EFAULT;
2536
2537 valbool = val ? 1 : 0;
2538
2539 switch (optname) {
2540 case UDP_CORK:
2541 if (val != 0) {
2542 up->corkflag = 1;
2543 } else {
2544 up->corkflag = 0;
2545 lock_sock(sk);
2546 push_pending_frames(sk);
2547 release_sock(sk);
2548 }
2549 break;
2550
2551 case UDP_ENCAP:
2552 switch (val) {
2553 case 0:
2554 case UDP_ENCAP_ESPINUDP:
2555 case UDP_ENCAP_ESPINUDP_NON_IKE:
2556 up->encap_rcv = xfrm4_udp_encap_rcv;
2557 /* FALLTHROUGH */
2558 case UDP_ENCAP_L2TPINUDP:
2559 up->encap_type = val;
2560 lock_sock(sk);
2561 udp_tunnel_encap_enable(sk->sk_socket);
2562 release_sock(sk);
2563 break;
2564 default:
2565 err = -ENOPROTOOPT;
2566 break;
2567 }
2568 break;
2569
2570 case UDP_NO_CHECK6_TX:
2571 up->no_check6_tx = valbool;
2572 break;
2573
2574 case UDP_NO_CHECK6_RX:
2575 up->no_check6_rx = valbool;
2576 break;
2577
2578 case UDP_SEGMENT:
2579 if (val < 0 || val > USHRT_MAX)
2580 return -EINVAL;
2581 up->gso_size = val;
2582 break;
2583
2584 case UDP_GRO:
2585 lock_sock(sk);
2586 if (valbool)
2587 udp_tunnel_encap_enable(sk->sk_socket);
2588 up->gro_enabled = valbool;
2589 release_sock(sk);
2590 break;
2591
2592 /*
2593 * UDP-Lite's partial checksum coverage (RFC 3828).
2594 */
2595 /* The sender sets actual checksum coverage length via this option.
2596 * The case coverage > packet length is handled by send module. */
2597 case UDPLITE_SEND_CSCOV:
2598 if (!is_udplite) /* Disable the option on UDP sockets */
2599 return -ENOPROTOOPT;
2600 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2601 val = 8;
2602 else if (val > USHRT_MAX)
2603 val = USHRT_MAX;
2604 up->pcslen = val;
2605 up->pcflag |= UDPLITE_SEND_CC;
2606 break;
2607
2608 /* The receiver specifies a minimum checksum coverage value. To make
2609 * sense, this should be set to at least 8 (as done below). If zero is
2610 * used, this again means full checksum coverage. */
2611 case UDPLITE_RECV_CSCOV:
2612 if (!is_udplite) /* Disable the option on UDP sockets */
2613 return -ENOPROTOOPT;
2614 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2615 val = 8;
2616 else if (val > USHRT_MAX)
2617 val = USHRT_MAX;
2618 up->pcrlen = val;
2619 up->pcflag |= UDPLITE_RECV_CC;
2620 break;
2621
2622 default:
2623 err = -ENOPROTOOPT;
2624 break;
2625 }
2626
2627 return err;
2628}
2629EXPORT_SYMBOL(udp_lib_setsockopt);
2630
2631int udp_setsockopt(struct sock *sk, int level, int optname,
2632 char __user *optval, unsigned int optlen)
2633{
2634 if (level == SOL_UDP || level == SOL_UDPLITE)
2635 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2636 udp_push_pending_frames);
2637 return ip_setsockopt(sk, level, optname, optval, optlen);
2638}
2639
2640#ifdef CONFIG_COMPAT
2641int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2642 char __user *optval, unsigned int optlen)
2643{
2644 if (level == SOL_UDP || level == SOL_UDPLITE)
2645 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2646 udp_push_pending_frames);
2647 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2648}
2649#endif
2650
2651int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2652 char __user *optval, int __user *optlen)
2653{
2654 struct udp_sock *up = udp_sk(sk);
2655 int val, len;
2656
2657 if (get_user(len, optlen))
2658 return -EFAULT;
2659
2660 len = min_t(unsigned int, len, sizeof(int));
2661
2662 if (len < 0)
2663 return -EINVAL;
2664
2665 switch (optname) {
2666 case UDP_CORK:
2667 val = up->corkflag;
2668 break;
2669
2670 case UDP_ENCAP:
2671 val = up->encap_type;
2672 break;
2673
2674 case UDP_NO_CHECK6_TX:
2675 val = up->no_check6_tx;
2676 break;
2677
2678 case UDP_NO_CHECK6_RX:
2679 val = up->no_check6_rx;
2680 break;
2681
2682 case UDP_SEGMENT:
2683 val = up->gso_size;
2684 break;
2685
2686 /* The following two cannot be changed on UDP sockets, the return is
2687 * always 0 (which corresponds to the full checksum coverage of UDP). */
2688 case UDPLITE_SEND_CSCOV:
2689 val = up->pcslen;
2690 break;
2691
2692 case UDPLITE_RECV_CSCOV:
2693 val = up->pcrlen;
2694 break;
2695
2696 default:
2697 return -ENOPROTOOPT;
2698 }
2699
2700 if (put_user(len, optlen))
2701 return -EFAULT;
2702 if (copy_to_user(optval, &val, len))
2703 return -EFAULT;
2704 return 0;
2705}
2706EXPORT_SYMBOL(udp_lib_getsockopt);
2707
2708int udp_getsockopt(struct sock *sk, int level, int optname,
2709 char __user *optval, int __user *optlen)
2710{
2711 if (level == SOL_UDP || level == SOL_UDPLITE)
2712 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2713 return ip_getsockopt(sk, level, optname, optval, optlen);
2714}
2715
2716#ifdef CONFIG_COMPAT
2717int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2718 char __user *optval, int __user *optlen)
2719{
2720 if (level == SOL_UDP || level == SOL_UDPLITE)
2721 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2722 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2723}
2724#endif
2725/**
2726 * udp_poll - wait for a UDP event.
2727 * @file - file struct
2728 * @sock - socket
2729 * @wait - poll table
2730 *
2731 * This is same as datagram poll, except for the special case of
2732 * blocking sockets. If application is using a blocking fd
2733 * and a packet with checksum error is in the queue;
2734 * then it could get return from select indicating data available
2735 * but then block when reading it. Add special case code
2736 * to work around these arguably broken applications.
2737 */
2738__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2739{
2740 __poll_t mask = datagram_poll(file, sock, wait);
2741 struct sock *sk = sock->sk;
2742
2743 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2744 mask |= EPOLLIN | EPOLLRDNORM;
2745
2746 /* Check for false positives due to checksum errors */
2747 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2748 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2749 mask &= ~(EPOLLIN | EPOLLRDNORM);
2750
2751 return mask;
2752
2753}
2754EXPORT_SYMBOL(udp_poll);
2755
2756int udp_abort(struct sock *sk, int err)
2757{
2758 lock_sock(sk);
2759
2760 sk->sk_err = err;
2761 sk->sk_error_report(sk);
2762 __udp_disconnect(sk, 0);
2763
2764 release_sock(sk);
2765
2766 return 0;
2767}
2768EXPORT_SYMBOL_GPL(udp_abort);
2769
2770struct proto udp_prot = {
2771 .name = "UDP",
2772 .owner = THIS_MODULE,
2773 .close = udp_lib_close,
2774 .pre_connect = udp_pre_connect,
2775 .connect = ip4_datagram_connect,
2776 .disconnect = udp_disconnect,
2777 .ioctl = udp_ioctl,
2778 .init = udp_init_sock,
2779 .destroy = udp_destroy_sock,
2780 .setsockopt = udp_setsockopt,
2781 .getsockopt = udp_getsockopt,
2782 .sendmsg = udp_sendmsg,
2783 .recvmsg = udp_recvmsg,
2784 .sendpage = udp_sendpage,
2785 .release_cb = ip4_datagram_release_cb,
2786 .hash = udp_lib_hash,
2787 .unhash = udp_lib_unhash,
2788 .rehash = udp_v4_rehash,
2789 .get_port = udp_v4_get_port,
2790 .memory_allocated = &udp_memory_allocated,
2791 .sysctl_mem = sysctl_udp_mem,
2792 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2793 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2794 .obj_size = sizeof(struct udp_sock),
2795 .h.udp_table = &udp_table,
2796#ifdef CONFIG_COMPAT
2797 .compat_setsockopt = compat_udp_setsockopt,
2798 .compat_getsockopt = compat_udp_getsockopt,
2799#endif
2800 .diag_destroy = udp_abort,
2801};
2802EXPORT_SYMBOL(udp_prot);
2803
2804/* ------------------------------------------------------------------------ */
2805#ifdef CONFIG_PROC_FS
2806
2807static struct sock *udp_get_first(struct seq_file *seq, int start)
2808{
2809 struct sock *sk;
2810 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2811 struct udp_iter_state *state = seq->private;
2812 struct net *net = seq_file_net(seq);
2813
2814 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2815 ++state->bucket) {
2816 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2817
2818 if (hlist_empty(&hslot->head))
2819 continue;
2820
2821 spin_lock_bh(&hslot->lock);
2822 sk_for_each(sk, &hslot->head) {
2823 if (!net_eq(sock_net(sk), net))
2824 continue;
2825 if (sk->sk_family == afinfo->family)
2826 goto found;
2827 }
2828 spin_unlock_bh(&hslot->lock);
2829 }
2830 sk = NULL;
2831found:
2832 return sk;
2833}
2834
2835static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2836{
2837 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2838 struct udp_iter_state *state = seq->private;
2839 struct net *net = seq_file_net(seq);
2840
2841 do {
2842 sk = sk_next(sk);
2843 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2844
2845 if (!sk) {
2846 if (state->bucket <= afinfo->udp_table->mask)
2847 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2848 return udp_get_first(seq, state->bucket + 1);
2849 }
2850 return sk;
2851}
2852
2853static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2854{
2855 struct sock *sk = udp_get_first(seq, 0);
2856
2857 if (sk)
2858 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2859 --pos;
2860 return pos ? NULL : sk;
2861}
2862
2863void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2864{
2865 struct udp_iter_state *state = seq->private;
2866 state->bucket = MAX_UDP_PORTS;
2867
2868 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2869}
2870EXPORT_SYMBOL(udp_seq_start);
2871
2872void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2873{
2874 struct sock *sk;
2875
2876 if (v == SEQ_START_TOKEN)
2877 sk = udp_get_idx(seq, 0);
2878 else
2879 sk = udp_get_next(seq, v);
2880
2881 ++*pos;
2882 return sk;
2883}
2884EXPORT_SYMBOL(udp_seq_next);
2885
2886void udp_seq_stop(struct seq_file *seq, void *v)
2887{
2888 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2889 struct udp_iter_state *state = seq->private;
2890
2891 if (state->bucket <= afinfo->udp_table->mask)
2892 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2893}
2894EXPORT_SYMBOL(udp_seq_stop);
2895
2896/* ------------------------------------------------------------------------ */
2897static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2898 int bucket)
2899{
2900 struct inet_sock *inet = inet_sk(sp);
2901 __be32 dest = inet->inet_daddr;
2902 __be32 src = inet->inet_rcv_saddr;
2903 __u16 destp = ntohs(inet->inet_dport);
2904 __u16 srcp = ntohs(inet->inet_sport);
2905
2906 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2907 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2908 bucket, src, srcp, dest, destp, sp->sk_state,
2909 sk_wmem_alloc_get(sp),
2910 udp_rqueue_get(sp),
2911 0, 0L, 0,
2912 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2913 0, sock_i_ino(sp),
2914 refcount_read(&sp->sk_refcnt), sp,
2915 atomic_read(&sp->sk_drops));
2916}
2917
2918int udp4_seq_show(struct seq_file *seq, void *v)
2919{
2920 seq_setwidth(seq, 127);
2921 if (v == SEQ_START_TOKEN)
2922 seq_puts(seq, " sl local_address rem_address st tx_queue "
2923 "rx_queue tr tm->when retrnsmt uid timeout "
2924 "inode ref pointer drops");
2925 else {
2926 struct udp_iter_state *state = seq->private;
2927
2928 udp4_format_sock(v, seq, state->bucket);
2929 }
2930 seq_pad(seq, '\n');
2931 return 0;
2932}
2933
2934const struct seq_operations udp_seq_ops = {
2935 .start = udp_seq_start,
2936 .next = udp_seq_next,
2937 .stop = udp_seq_stop,
2938 .show = udp4_seq_show,
2939};
2940EXPORT_SYMBOL(udp_seq_ops);
2941
2942static struct udp_seq_afinfo udp4_seq_afinfo = {
2943 .family = AF_INET,
2944 .udp_table = &udp_table,
2945};
2946
2947static int __net_init udp4_proc_init_net(struct net *net)
2948{
2949 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2950 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2951 return -ENOMEM;
2952 return 0;
2953}
2954
2955static void __net_exit udp4_proc_exit_net(struct net *net)
2956{
2957 remove_proc_entry("udp", net->proc_net);
2958}
2959
2960static struct pernet_operations udp4_net_ops = {
2961 .init = udp4_proc_init_net,
2962 .exit = udp4_proc_exit_net,
2963};
2964
2965int __init udp4_proc_init(void)
2966{
2967 return register_pernet_subsys(&udp4_net_ops);
2968}
2969
2970void udp4_proc_exit(void)
2971{
2972 unregister_pernet_subsys(&udp4_net_ops);
2973}
2974#endif /* CONFIG_PROC_FS */
2975
2976static __initdata unsigned long uhash_entries;
2977static int __init set_uhash_entries(char *str)
2978{
2979 ssize_t ret;
2980
2981 if (!str)
2982 return 0;
2983
2984 ret = kstrtoul(str, 0, &uhash_entries);
2985 if (ret)
2986 return 0;
2987
2988 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2989 uhash_entries = UDP_HTABLE_SIZE_MIN;
2990 return 1;
2991}
2992__setup("uhash_entries=", set_uhash_entries);
2993
2994void __init udp_table_init(struct udp_table *table, const char *name)
2995{
2996 unsigned int i;
2997
2998 table->hash = alloc_large_system_hash(name,
2999 2 * sizeof(struct udp_hslot),
3000 uhash_entries,
3001 21, /* one slot per 2 MB */
3002 0,
3003 &table->log,
3004 &table->mask,
3005 UDP_HTABLE_SIZE_MIN,
3006 64 * 1024);
3007
3008 table->hash2 = table->hash + (table->mask + 1);
3009 for (i = 0; i <= table->mask; i++) {
3010 INIT_HLIST_HEAD(&table->hash[i].head);
3011 table->hash[i].count = 0;
3012 spin_lock_init(&table->hash[i].lock);
3013 }
3014 for (i = 0; i <= table->mask; i++) {
3015 INIT_HLIST_HEAD(&table->hash2[i].head);
3016 table->hash2[i].count = 0;
3017 spin_lock_init(&table->hash2[i].lock);
3018 }
3019}
3020
3021u32 udp_flow_hashrnd(void)
3022{
3023 static u32 hashrnd __read_mostly;
3024
3025 net_get_random_once(&hashrnd, sizeof(hashrnd));
3026
3027 return hashrnd;
3028}
3029EXPORT_SYMBOL(udp_flow_hashrnd);
3030
3031static void __udp_sysctl_init(struct net *net)
3032{
3033 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3034 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3035
3036#ifdef CONFIG_NET_L3_MASTER_DEV
3037 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3038#endif
3039}
3040
3041static int __net_init udp_sysctl_init(struct net *net)
3042{
3043 __udp_sysctl_init(net);
3044 return 0;
3045}
3046
3047static struct pernet_operations __net_initdata udp_sysctl_ops = {
3048 .init = udp_sysctl_init,
3049};
3050
3051void __init udp_init(void)
3052{
3053 unsigned long limit;
3054 unsigned int i;
3055
3056 udp_table_init(&udp_table, "UDP");
3057 limit = nr_free_buffer_pages() / 8;
3058 limit = max(limit, 128UL);
3059 sysctl_udp_mem[0] = limit / 4 * 3;
3060 sysctl_udp_mem[1] = limit;
3061 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3062
3063 __udp_sysctl_init(&init_net);
3064
3065 /* 16 spinlocks per cpu */
3066 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3067 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3068 GFP_KERNEL);
3069 if (!udp_busylocks)
3070 panic("UDP: failed to alloc udp_busylocks\n");
3071 for (i = 0; i < (1U << udp_busylocks_log); i++)
3072 spin_lock_init(udp_busylocks + i);
3073
3074 if (register_pernet_subsys(&udp_sysctl_ops))
3075 panic("UDP: failed to init sysctl parameters.\n");
3076}
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}