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 <linux/btf_ids.h>
110#include <trace/events/skb.h>
111#include <net/busy_poll.h>
112#include "udp_impl.h"
113#include <net/sock_reuseport.h>
114#include <net/addrconf.h>
115#include <net/udp_tunnel.h>
116#if IS_ENABLED(CONFIG_IPV6)
117#include <net/ipv6_stubs.h>
118#endif
119
120struct udp_table udp_table __read_mostly;
121EXPORT_SYMBOL(udp_table);
122
123long sysctl_udp_mem[3] __read_mostly;
124EXPORT_SYMBOL(sysctl_udp_mem);
125
126atomic_long_t udp_memory_allocated;
127EXPORT_SYMBOL(udp_memory_allocated);
128
129#define MAX_UDP_PORTS 65536
130#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
131
132static int udp_lib_lport_inuse(struct net *net, __u16 num,
133 const struct udp_hslot *hslot,
134 unsigned long *bitmap,
135 struct sock *sk, unsigned int log)
136{
137 struct sock *sk2;
138 kuid_t uid = sock_i_uid(sk);
139
140 sk_for_each(sk2, &hslot->head) {
141 if (net_eq(sock_net(sk2), net) &&
142 sk2 != sk &&
143 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
144 (!sk2->sk_reuse || !sk->sk_reuse) &&
145 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
146 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
147 inet_rcv_saddr_equal(sk, sk2, true)) {
148 if (sk2->sk_reuseport && sk->sk_reuseport &&
149 !rcu_access_pointer(sk->sk_reuseport_cb) &&
150 uid_eq(uid, sock_i_uid(sk2))) {
151 if (!bitmap)
152 return 0;
153 } else {
154 if (!bitmap)
155 return 1;
156 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
157 bitmap);
158 }
159 }
160 }
161 return 0;
162}
163
164/*
165 * Note: we still hold spinlock of primary hash chain, so no other writer
166 * can insert/delete a socket with local_port == num
167 */
168static int udp_lib_lport_inuse2(struct net *net, __u16 num,
169 struct udp_hslot *hslot2,
170 struct sock *sk)
171{
172 struct sock *sk2;
173 kuid_t uid = sock_i_uid(sk);
174 int res = 0;
175
176 spin_lock(&hslot2->lock);
177 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
178 if (net_eq(sock_net(sk2), net) &&
179 sk2 != sk &&
180 (udp_sk(sk2)->udp_port_hash == num) &&
181 (!sk2->sk_reuse || !sk->sk_reuse) &&
182 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
183 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
184 inet_rcv_saddr_equal(sk, sk2, true)) {
185 if (sk2->sk_reuseport && sk->sk_reuseport &&
186 !rcu_access_pointer(sk->sk_reuseport_cb) &&
187 uid_eq(uid, sock_i_uid(sk2))) {
188 res = 0;
189 } else {
190 res = 1;
191 }
192 break;
193 }
194 }
195 spin_unlock(&hslot2->lock);
196 return res;
197}
198
199static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
200{
201 struct net *net = sock_net(sk);
202 kuid_t uid = sock_i_uid(sk);
203 struct sock *sk2;
204
205 sk_for_each(sk2, &hslot->head) {
206 if (net_eq(sock_net(sk2), net) &&
207 sk2 != sk &&
208 sk2->sk_family == sk->sk_family &&
209 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
210 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
211 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
212 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
213 inet_rcv_saddr_equal(sk, sk2, false)) {
214 return reuseport_add_sock(sk, sk2,
215 inet_rcv_saddr_any(sk));
216 }
217 }
218
219 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
220}
221
222/**
223 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
224 *
225 * @sk: socket struct in question
226 * @snum: port number to look up
227 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
228 * with NULL address
229 */
230int udp_lib_get_port(struct sock *sk, unsigned short snum,
231 unsigned int hash2_nulladdr)
232{
233 struct udp_hslot *hslot, *hslot2;
234 struct udp_table *udptable = sk->sk_prot->h.udp_table;
235 int error = 1;
236 struct net *net = sock_net(sk);
237
238 if (!snum) {
239 int low, high, remaining;
240 unsigned int rand;
241 unsigned short first, last;
242 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
243
244 inet_get_local_port_range(net, &low, &high);
245 remaining = (high - low) + 1;
246
247 rand = prandom_u32();
248 first = reciprocal_scale(rand, remaining) + low;
249 /*
250 * force rand to be an odd multiple of UDP_HTABLE_SIZE
251 */
252 rand = (rand | 1) * (udptable->mask + 1);
253 last = first + udptable->mask + 1;
254 do {
255 hslot = udp_hashslot(udptable, net, first);
256 bitmap_zero(bitmap, PORTS_PER_CHAIN);
257 spin_lock_bh(&hslot->lock);
258 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
259 udptable->log);
260
261 snum = first;
262 /*
263 * Iterate on all possible values of snum for this hash.
264 * Using steps of an odd multiple of UDP_HTABLE_SIZE
265 * give us randomization and full range coverage.
266 */
267 do {
268 if (low <= snum && snum <= high &&
269 !test_bit(snum >> udptable->log, bitmap) &&
270 !inet_is_local_reserved_port(net, snum))
271 goto found;
272 snum += rand;
273 } while (snum != first);
274 spin_unlock_bh(&hslot->lock);
275 cond_resched();
276 } while (++first != last);
277 goto fail;
278 } else {
279 hslot = udp_hashslot(udptable, net, snum);
280 spin_lock_bh(&hslot->lock);
281 if (hslot->count > 10) {
282 int exist;
283 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
284
285 slot2 &= udptable->mask;
286 hash2_nulladdr &= udptable->mask;
287
288 hslot2 = udp_hashslot2(udptable, slot2);
289 if (hslot->count < hslot2->count)
290 goto scan_primary_hash;
291
292 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
293 if (!exist && (hash2_nulladdr != slot2)) {
294 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
295 exist = udp_lib_lport_inuse2(net, snum, hslot2,
296 sk);
297 }
298 if (exist)
299 goto fail_unlock;
300 else
301 goto found;
302 }
303scan_primary_hash:
304 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
305 goto fail_unlock;
306 }
307found:
308 inet_sk(sk)->inet_num = snum;
309 udp_sk(sk)->udp_port_hash = snum;
310 udp_sk(sk)->udp_portaddr_hash ^= snum;
311 if (sk_unhashed(sk)) {
312 if (sk->sk_reuseport &&
313 udp_reuseport_add_sock(sk, hslot)) {
314 inet_sk(sk)->inet_num = 0;
315 udp_sk(sk)->udp_port_hash = 0;
316 udp_sk(sk)->udp_portaddr_hash ^= snum;
317 goto fail_unlock;
318 }
319
320 sk_add_node_rcu(sk, &hslot->head);
321 hslot->count++;
322 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
323
324 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
325 spin_lock(&hslot2->lock);
326 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
327 sk->sk_family == AF_INET6)
328 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
329 &hslot2->head);
330 else
331 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
332 &hslot2->head);
333 hslot2->count++;
334 spin_unlock(&hslot2->lock);
335 }
336 sock_set_flag(sk, SOCK_RCU_FREE);
337 error = 0;
338fail_unlock:
339 spin_unlock_bh(&hslot->lock);
340fail:
341 return error;
342}
343EXPORT_SYMBOL(udp_lib_get_port);
344
345int udp_v4_get_port(struct sock *sk, unsigned short snum)
346{
347 unsigned int hash2_nulladdr =
348 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
349 unsigned int hash2_partial =
350 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
351
352 /* precompute partial secondary hash */
353 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
354 return udp_lib_get_port(sk, snum, hash2_nulladdr);
355}
356
357static int compute_score(struct sock *sk, struct net *net,
358 __be32 saddr, __be16 sport,
359 __be32 daddr, unsigned short hnum,
360 int dif, int sdif)
361{
362 int score;
363 struct inet_sock *inet;
364 bool dev_match;
365
366 if (!net_eq(sock_net(sk), net) ||
367 udp_sk(sk)->udp_port_hash != hnum ||
368 ipv6_only_sock(sk))
369 return -1;
370
371 if (sk->sk_rcv_saddr != daddr)
372 return -1;
373
374 score = (sk->sk_family == PF_INET) ? 2 : 1;
375
376 inet = inet_sk(sk);
377 if (inet->inet_daddr) {
378 if (inet->inet_daddr != saddr)
379 return -1;
380 score += 4;
381 }
382
383 if (inet->inet_dport) {
384 if (inet->inet_dport != sport)
385 return -1;
386 score += 4;
387 }
388
389 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
390 dif, sdif);
391 if (!dev_match)
392 return -1;
393 if (sk->sk_bound_dev_if)
394 score += 4;
395
396 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
397 score++;
398 return score;
399}
400
401static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
402 const __u16 lport, const __be32 faddr,
403 const __be16 fport)
404{
405 static u32 udp_ehash_secret __read_mostly;
406
407 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
408
409 return __inet_ehashfn(laddr, lport, faddr, fport,
410 udp_ehash_secret + net_hash_mix(net));
411}
412
413static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
414 struct sk_buff *skb,
415 __be32 saddr, __be16 sport,
416 __be32 daddr, unsigned short hnum)
417{
418 struct sock *reuse_sk = NULL;
419 u32 hash;
420
421 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
422 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
423 reuse_sk = reuseport_select_sock(sk, hash, skb,
424 sizeof(struct udphdr));
425 }
426 return reuse_sk;
427}
428
429/* called with rcu_read_lock() */
430static struct sock *udp4_lib_lookup2(struct net *net,
431 __be32 saddr, __be16 sport,
432 __be32 daddr, unsigned int hnum,
433 int dif, int sdif,
434 struct udp_hslot *hslot2,
435 struct sk_buff *skb)
436{
437 struct sock *sk, *result;
438 int score, badness;
439
440 result = NULL;
441 badness = 0;
442 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
443 score = compute_score(sk, net, saddr, sport,
444 daddr, hnum, dif, sdif);
445 if (score > badness) {
446 result = lookup_reuseport(net, sk, skb,
447 saddr, sport, daddr, hnum);
448 /* Fall back to scoring if group has connections */
449 if (result && !reuseport_has_conns(sk, false))
450 return result;
451
452 result = result ? : sk;
453 badness = score;
454 }
455 }
456 return result;
457}
458
459static struct sock *udp4_lookup_run_bpf(struct net *net,
460 struct udp_table *udptable,
461 struct sk_buff *skb,
462 __be32 saddr, __be16 sport,
463 __be32 daddr, u16 hnum)
464{
465 struct sock *sk, *reuse_sk;
466 bool no_reuseport;
467
468 if (udptable != &udp_table)
469 return NULL; /* only UDP is supported */
470
471 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP,
472 saddr, sport, daddr, hnum, &sk);
473 if (no_reuseport || IS_ERR_OR_NULL(sk))
474 return sk;
475
476 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
477 if (reuse_sk)
478 sk = reuse_sk;
479 return sk;
480}
481
482/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
483 * harder than this. -DaveM
484 */
485struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
486 __be16 sport, __be32 daddr, __be16 dport, int dif,
487 int sdif, struct udp_table *udptable, struct sk_buff *skb)
488{
489 unsigned short hnum = ntohs(dport);
490 unsigned int hash2, slot2;
491 struct udp_hslot *hslot2;
492 struct sock *result, *sk;
493
494 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
495 slot2 = hash2 & udptable->mask;
496 hslot2 = &udptable->hash2[slot2];
497
498 /* Lookup connected or non-wildcard socket */
499 result = udp4_lib_lookup2(net, saddr, sport,
500 daddr, hnum, dif, sdif,
501 hslot2, skb);
502 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
503 goto done;
504
505 /* Lookup redirect from BPF */
506 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
507 sk = udp4_lookup_run_bpf(net, udptable, skb,
508 saddr, sport, daddr, hnum);
509 if (sk) {
510 result = sk;
511 goto done;
512 }
513 }
514
515 /* Got non-wildcard socket or error on first lookup */
516 if (result)
517 goto done;
518
519 /* Lookup wildcard sockets */
520 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
521 slot2 = hash2 & udptable->mask;
522 hslot2 = &udptable->hash2[slot2];
523
524 result = udp4_lib_lookup2(net, saddr, sport,
525 htonl(INADDR_ANY), hnum, dif, sdif,
526 hslot2, skb);
527done:
528 if (IS_ERR(result))
529 return NULL;
530 return result;
531}
532EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
533
534static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
535 __be16 sport, __be16 dport,
536 struct udp_table *udptable)
537{
538 const struct iphdr *iph = ip_hdr(skb);
539
540 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
541 iph->daddr, dport, inet_iif(skb),
542 inet_sdif(skb), udptable, skb);
543}
544
545struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
546 __be16 sport, __be16 dport)
547{
548 const struct iphdr *iph = ip_hdr(skb);
549
550 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
551 iph->daddr, dport, inet_iif(skb),
552 inet_sdif(skb), &udp_table, NULL);
553}
554
555/* Must be called under rcu_read_lock().
556 * Does increment socket refcount.
557 */
558#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
559struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
560 __be32 daddr, __be16 dport, int dif)
561{
562 struct sock *sk;
563
564 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
565 dif, 0, &udp_table, NULL);
566 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
567 sk = NULL;
568 return sk;
569}
570EXPORT_SYMBOL_GPL(udp4_lib_lookup);
571#endif
572
573static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
574 __be16 loc_port, __be32 loc_addr,
575 __be16 rmt_port, __be32 rmt_addr,
576 int dif, int sdif, unsigned short hnum)
577{
578 struct inet_sock *inet = inet_sk(sk);
579
580 if (!net_eq(sock_net(sk), net) ||
581 udp_sk(sk)->udp_port_hash != hnum ||
582 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
583 (inet->inet_dport != rmt_port && inet->inet_dport) ||
584 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
585 ipv6_only_sock(sk) ||
586 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
587 return false;
588 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
589 return false;
590 return true;
591}
592
593DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
594void udp_encap_enable(void)
595{
596 static_branch_inc(&udp_encap_needed_key);
597}
598EXPORT_SYMBOL(udp_encap_enable);
599
600void udp_encap_disable(void)
601{
602 static_branch_dec(&udp_encap_needed_key);
603}
604EXPORT_SYMBOL(udp_encap_disable);
605
606/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
607 * through error handlers in encapsulations looking for a match.
608 */
609static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
610{
611 int i;
612
613 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
614 int (*handler)(struct sk_buff *skb, u32 info);
615 const struct ip_tunnel_encap_ops *encap;
616
617 encap = rcu_dereference(iptun_encaps[i]);
618 if (!encap)
619 continue;
620 handler = encap->err_handler;
621 if (handler && !handler(skb, info))
622 return 0;
623 }
624
625 return -ENOENT;
626}
627
628/* Try to match ICMP errors to UDP tunnels by looking up a socket without
629 * reversing source and destination port: this will match tunnels that force the
630 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
631 * lwtunnels might actually break this assumption by being configured with
632 * different destination ports on endpoints, in this case we won't be able to
633 * trace ICMP messages back to them.
634 *
635 * If this doesn't match any socket, probe tunnels with arbitrary destination
636 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
637 * we've sent packets to won't necessarily match the local destination port.
638 *
639 * Then ask the tunnel implementation to match the error against a valid
640 * association.
641 *
642 * Return an error if we can't find a match, the socket if we need further
643 * processing, zero otherwise.
644 */
645static struct sock *__udp4_lib_err_encap(struct net *net,
646 const struct iphdr *iph,
647 struct udphdr *uh,
648 struct udp_table *udptable,
649 struct sock *sk,
650 struct sk_buff *skb, u32 info)
651{
652 int (*lookup)(struct sock *sk, struct sk_buff *skb);
653 int network_offset, transport_offset;
654 struct udp_sock *up;
655
656 network_offset = skb_network_offset(skb);
657 transport_offset = skb_transport_offset(skb);
658
659 /* Network header needs to point to the outer IPv4 header inside ICMP */
660 skb_reset_network_header(skb);
661
662 /* Transport header needs to point to the UDP header */
663 skb_set_transport_header(skb, iph->ihl << 2);
664
665 if (sk) {
666 up = udp_sk(sk);
667
668 lookup = READ_ONCE(up->encap_err_lookup);
669 if (lookup && lookup(sk, skb))
670 sk = NULL;
671
672 goto out;
673 }
674
675 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
676 iph->saddr, uh->dest, skb->dev->ifindex, 0,
677 udptable, NULL);
678 if (sk) {
679 up = udp_sk(sk);
680
681 lookup = READ_ONCE(up->encap_err_lookup);
682 if (!lookup || lookup(sk, skb))
683 sk = NULL;
684 }
685
686out:
687 if (!sk)
688 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
689
690 skb_set_transport_header(skb, transport_offset);
691 skb_set_network_header(skb, network_offset);
692
693 return sk;
694}
695
696/*
697 * This routine is called by the ICMP module when it gets some
698 * sort of error condition. If err < 0 then the socket should
699 * be closed and the error returned to the user. If err > 0
700 * it's just the icmp type << 8 | icmp code.
701 * Header points to the ip header of the error packet. We move
702 * on past this. Then (as it used to claim before adjustment)
703 * header points to the first 8 bytes of the udp header. We need
704 * to find the appropriate port.
705 */
706
707int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
708{
709 struct inet_sock *inet;
710 const struct iphdr *iph = (const struct iphdr *)skb->data;
711 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
712 const int type = icmp_hdr(skb)->type;
713 const int code = icmp_hdr(skb)->code;
714 bool tunnel = false;
715 struct sock *sk;
716 int harderr;
717 int err;
718 struct net *net = dev_net(skb->dev);
719
720 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
721 iph->saddr, uh->source, skb->dev->ifindex,
722 inet_sdif(skb), udptable, NULL);
723
724 if (!sk || udp_sk(sk)->encap_type) {
725 /* No socket for error: try tunnels before discarding */
726 if (static_branch_unlikely(&udp_encap_needed_key)) {
727 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
728 info);
729 if (!sk)
730 return 0;
731 } else
732 sk = ERR_PTR(-ENOENT);
733
734 if (IS_ERR(sk)) {
735 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
736 return PTR_ERR(sk);
737 }
738
739 tunnel = true;
740 }
741
742 err = 0;
743 harderr = 0;
744 inet = inet_sk(sk);
745
746 switch (type) {
747 default:
748 case ICMP_TIME_EXCEEDED:
749 err = EHOSTUNREACH;
750 break;
751 case ICMP_SOURCE_QUENCH:
752 goto out;
753 case ICMP_PARAMETERPROB:
754 err = EPROTO;
755 harderr = 1;
756 break;
757 case ICMP_DEST_UNREACH:
758 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
759 ipv4_sk_update_pmtu(skb, sk, info);
760 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
761 err = EMSGSIZE;
762 harderr = 1;
763 break;
764 }
765 goto out;
766 }
767 err = EHOSTUNREACH;
768 if (code <= NR_ICMP_UNREACH) {
769 harderr = icmp_err_convert[code].fatal;
770 err = icmp_err_convert[code].errno;
771 }
772 break;
773 case ICMP_REDIRECT:
774 ipv4_sk_redirect(skb, sk);
775 goto out;
776 }
777
778 /*
779 * RFC1122: OK. Passes ICMP errors back to application, as per
780 * 4.1.3.3.
781 */
782 if (tunnel) {
783 /* ...not for tunnels though: we don't have a sending socket */
784 goto out;
785 }
786 if (!inet->recverr) {
787 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
788 goto out;
789 } else
790 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
791
792 sk->sk_err = err;
793 sk_error_report(sk);
794out:
795 return 0;
796}
797
798int udp_err(struct sk_buff *skb, u32 info)
799{
800 return __udp4_lib_err(skb, info, &udp_table);
801}
802
803/*
804 * Throw away all pending data and cancel the corking. Socket is locked.
805 */
806void udp_flush_pending_frames(struct sock *sk)
807{
808 struct udp_sock *up = udp_sk(sk);
809
810 if (up->pending) {
811 up->len = 0;
812 up->pending = 0;
813 ip_flush_pending_frames(sk);
814 }
815}
816EXPORT_SYMBOL(udp_flush_pending_frames);
817
818/**
819 * udp4_hwcsum - handle outgoing HW checksumming
820 * @skb: sk_buff containing the filled-in UDP header
821 * (checksum field must be zeroed out)
822 * @src: source IP address
823 * @dst: destination IP address
824 */
825void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
826{
827 struct udphdr *uh = udp_hdr(skb);
828 int offset = skb_transport_offset(skb);
829 int len = skb->len - offset;
830 int hlen = len;
831 __wsum csum = 0;
832
833 if (!skb_has_frag_list(skb)) {
834 /*
835 * Only one fragment on the socket.
836 */
837 skb->csum_start = skb_transport_header(skb) - skb->head;
838 skb->csum_offset = offsetof(struct udphdr, check);
839 uh->check = ~csum_tcpudp_magic(src, dst, len,
840 IPPROTO_UDP, 0);
841 } else {
842 struct sk_buff *frags;
843
844 /*
845 * HW-checksum won't work as there are two or more
846 * fragments on the socket so that all csums of sk_buffs
847 * should be together
848 */
849 skb_walk_frags(skb, frags) {
850 csum = csum_add(csum, frags->csum);
851 hlen -= frags->len;
852 }
853
854 csum = skb_checksum(skb, offset, hlen, csum);
855 skb->ip_summed = CHECKSUM_NONE;
856
857 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
858 if (uh->check == 0)
859 uh->check = CSUM_MANGLED_0;
860 }
861}
862EXPORT_SYMBOL_GPL(udp4_hwcsum);
863
864/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
865 * for the simple case like when setting the checksum for a UDP tunnel.
866 */
867void udp_set_csum(bool nocheck, struct sk_buff *skb,
868 __be32 saddr, __be32 daddr, int len)
869{
870 struct udphdr *uh = udp_hdr(skb);
871
872 if (nocheck) {
873 uh->check = 0;
874 } else if (skb_is_gso(skb)) {
875 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
876 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
877 uh->check = 0;
878 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
879 if (uh->check == 0)
880 uh->check = CSUM_MANGLED_0;
881 } else {
882 skb->ip_summed = CHECKSUM_PARTIAL;
883 skb->csum_start = skb_transport_header(skb) - skb->head;
884 skb->csum_offset = offsetof(struct udphdr, check);
885 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
886 }
887}
888EXPORT_SYMBOL(udp_set_csum);
889
890static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
891 struct inet_cork *cork)
892{
893 struct sock *sk = skb->sk;
894 struct inet_sock *inet = inet_sk(sk);
895 struct udphdr *uh;
896 int err;
897 int is_udplite = IS_UDPLITE(sk);
898 int offset = skb_transport_offset(skb);
899 int len = skb->len - offset;
900 int datalen = len - sizeof(*uh);
901 __wsum csum = 0;
902
903 /*
904 * Create a UDP header
905 */
906 uh = udp_hdr(skb);
907 uh->source = inet->inet_sport;
908 uh->dest = fl4->fl4_dport;
909 uh->len = htons(len);
910 uh->check = 0;
911
912 if (cork->gso_size) {
913 const int hlen = skb_network_header_len(skb) +
914 sizeof(struct udphdr);
915
916 if (hlen + cork->gso_size > cork->fragsize) {
917 kfree_skb(skb);
918 return -EINVAL;
919 }
920 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
921 kfree_skb(skb);
922 return -EINVAL;
923 }
924 if (sk->sk_no_check_tx) {
925 kfree_skb(skb);
926 return -EINVAL;
927 }
928 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
929 dst_xfrm(skb_dst(skb))) {
930 kfree_skb(skb);
931 return -EIO;
932 }
933
934 if (datalen > cork->gso_size) {
935 skb_shinfo(skb)->gso_size = cork->gso_size;
936 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
937 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
938 cork->gso_size);
939 }
940 goto csum_partial;
941 }
942
943 if (is_udplite) /* UDP-Lite */
944 csum = udplite_csum(skb);
945
946 else if (sk->sk_no_check_tx) { /* UDP csum off */
947
948 skb->ip_summed = CHECKSUM_NONE;
949 goto send;
950
951 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
952csum_partial:
953
954 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
955 goto send;
956
957 } else
958 csum = udp_csum(skb);
959
960 /* add protocol-dependent pseudo-header */
961 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
962 sk->sk_protocol, csum);
963 if (uh->check == 0)
964 uh->check = CSUM_MANGLED_0;
965
966send:
967 err = ip_send_skb(sock_net(sk), skb);
968 if (err) {
969 if (err == -ENOBUFS && !inet->recverr) {
970 UDP_INC_STATS(sock_net(sk),
971 UDP_MIB_SNDBUFERRORS, is_udplite);
972 err = 0;
973 }
974 } else
975 UDP_INC_STATS(sock_net(sk),
976 UDP_MIB_OUTDATAGRAMS, is_udplite);
977 return err;
978}
979
980/*
981 * Push out all pending data as one UDP datagram. Socket is locked.
982 */
983int udp_push_pending_frames(struct sock *sk)
984{
985 struct udp_sock *up = udp_sk(sk);
986 struct inet_sock *inet = inet_sk(sk);
987 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
988 struct sk_buff *skb;
989 int err = 0;
990
991 skb = ip_finish_skb(sk, fl4);
992 if (!skb)
993 goto out;
994
995 err = udp_send_skb(skb, fl4, &inet->cork.base);
996
997out:
998 up->len = 0;
999 up->pending = 0;
1000 return err;
1001}
1002EXPORT_SYMBOL(udp_push_pending_frames);
1003
1004static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1005{
1006 switch (cmsg->cmsg_type) {
1007 case UDP_SEGMENT:
1008 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1009 return -EINVAL;
1010 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1011 return 0;
1012 default:
1013 return -EINVAL;
1014 }
1015}
1016
1017int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1018{
1019 struct cmsghdr *cmsg;
1020 bool need_ip = false;
1021 int err;
1022
1023 for_each_cmsghdr(cmsg, msg) {
1024 if (!CMSG_OK(msg, cmsg))
1025 return -EINVAL;
1026
1027 if (cmsg->cmsg_level != SOL_UDP) {
1028 need_ip = true;
1029 continue;
1030 }
1031
1032 err = __udp_cmsg_send(cmsg, gso_size);
1033 if (err)
1034 return err;
1035 }
1036
1037 return need_ip;
1038}
1039EXPORT_SYMBOL_GPL(udp_cmsg_send);
1040
1041int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1042{
1043 struct inet_sock *inet = inet_sk(sk);
1044 struct udp_sock *up = udp_sk(sk);
1045 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1046 struct flowi4 fl4_stack;
1047 struct flowi4 *fl4;
1048 int ulen = len;
1049 struct ipcm_cookie ipc;
1050 struct rtable *rt = NULL;
1051 int free = 0;
1052 int connected = 0;
1053 __be32 daddr, faddr, saddr;
1054 __be16 dport;
1055 u8 tos;
1056 int err, is_udplite = IS_UDPLITE(sk);
1057 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1058 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1059 struct sk_buff *skb;
1060 struct ip_options_data opt_copy;
1061
1062 if (len > 0xFFFF)
1063 return -EMSGSIZE;
1064
1065 /*
1066 * Check the flags.
1067 */
1068
1069 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1070 return -EOPNOTSUPP;
1071
1072 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1073
1074 fl4 = &inet->cork.fl.u.ip4;
1075 if (up->pending) {
1076 /*
1077 * There are pending frames.
1078 * The socket lock must be held while it's corked.
1079 */
1080 lock_sock(sk);
1081 if (likely(up->pending)) {
1082 if (unlikely(up->pending != AF_INET)) {
1083 release_sock(sk);
1084 return -EINVAL;
1085 }
1086 goto do_append_data;
1087 }
1088 release_sock(sk);
1089 }
1090 ulen += sizeof(struct udphdr);
1091
1092 /*
1093 * Get and verify the address.
1094 */
1095 if (usin) {
1096 if (msg->msg_namelen < sizeof(*usin))
1097 return -EINVAL;
1098 if (usin->sin_family != AF_INET) {
1099 if (usin->sin_family != AF_UNSPEC)
1100 return -EAFNOSUPPORT;
1101 }
1102
1103 daddr = usin->sin_addr.s_addr;
1104 dport = usin->sin_port;
1105 if (dport == 0)
1106 return -EINVAL;
1107 } else {
1108 if (sk->sk_state != TCP_ESTABLISHED)
1109 return -EDESTADDRREQ;
1110 daddr = inet->inet_daddr;
1111 dport = inet->inet_dport;
1112 /* Open fast path for connected socket.
1113 Route will not be used, if at least one option is set.
1114 */
1115 connected = 1;
1116 }
1117
1118 ipcm_init_sk(&ipc, inet);
1119 ipc.gso_size = READ_ONCE(up->gso_size);
1120
1121 if (msg->msg_controllen) {
1122 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1123 if (err > 0)
1124 err = ip_cmsg_send(sk, msg, &ipc,
1125 sk->sk_family == AF_INET6);
1126 if (unlikely(err < 0)) {
1127 kfree(ipc.opt);
1128 return err;
1129 }
1130 if (ipc.opt)
1131 free = 1;
1132 connected = 0;
1133 }
1134 if (!ipc.opt) {
1135 struct ip_options_rcu *inet_opt;
1136
1137 rcu_read_lock();
1138 inet_opt = rcu_dereference(inet->inet_opt);
1139 if (inet_opt) {
1140 memcpy(&opt_copy, inet_opt,
1141 sizeof(*inet_opt) + inet_opt->opt.optlen);
1142 ipc.opt = &opt_copy.opt;
1143 }
1144 rcu_read_unlock();
1145 }
1146
1147 if (cgroup_bpf_enabled(BPF_CGROUP_UDP4_SENDMSG) && !connected) {
1148 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1149 (struct sockaddr *)usin, &ipc.addr);
1150 if (err)
1151 goto out_free;
1152 if (usin) {
1153 if (usin->sin_port == 0) {
1154 /* BPF program set invalid port. Reject it. */
1155 err = -EINVAL;
1156 goto out_free;
1157 }
1158 daddr = usin->sin_addr.s_addr;
1159 dport = usin->sin_port;
1160 }
1161 }
1162
1163 saddr = ipc.addr;
1164 ipc.addr = faddr = daddr;
1165
1166 if (ipc.opt && ipc.opt->opt.srr) {
1167 if (!daddr) {
1168 err = -EINVAL;
1169 goto out_free;
1170 }
1171 faddr = ipc.opt->opt.faddr;
1172 connected = 0;
1173 }
1174 tos = get_rttos(&ipc, inet);
1175 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1176 (msg->msg_flags & MSG_DONTROUTE) ||
1177 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1178 tos |= RTO_ONLINK;
1179 connected = 0;
1180 }
1181
1182 if (ipv4_is_multicast(daddr)) {
1183 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1184 ipc.oif = inet->mc_index;
1185 if (!saddr)
1186 saddr = inet->mc_addr;
1187 connected = 0;
1188 } else if (!ipc.oif) {
1189 ipc.oif = inet->uc_index;
1190 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1191 /* oif is set, packet is to local broadcast and
1192 * uc_index is set. oif is most likely set
1193 * by sk_bound_dev_if. If uc_index != oif check if the
1194 * oif is an L3 master and uc_index is an L3 slave.
1195 * If so, we want to allow the send using the uc_index.
1196 */
1197 if (ipc.oif != inet->uc_index &&
1198 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1199 inet->uc_index)) {
1200 ipc.oif = inet->uc_index;
1201 }
1202 }
1203
1204 if (connected)
1205 rt = (struct rtable *)sk_dst_check(sk, 0);
1206
1207 if (!rt) {
1208 struct net *net = sock_net(sk);
1209 __u8 flow_flags = inet_sk_flowi_flags(sk);
1210
1211 fl4 = &fl4_stack;
1212
1213 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1214 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1215 flow_flags,
1216 faddr, saddr, dport, inet->inet_sport,
1217 sk->sk_uid);
1218
1219 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1220 rt = ip_route_output_flow(net, fl4, sk);
1221 if (IS_ERR(rt)) {
1222 err = PTR_ERR(rt);
1223 rt = NULL;
1224 if (err == -ENETUNREACH)
1225 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1226 goto out;
1227 }
1228
1229 err = -EACCES;
1230 if ((rt->rt_flags & RTCF_BROADCAST) &&
1231 !sock_flag(sk, SOCK_BROADCAST))
1232 goto out;
1233 if (connected)
1234 sk_dst_set(sk, dst_clone(&rt->dst));
1235 }
1236
1237 if (msg->msg_flags&MSG_CONFIRM)
1238 goto do_confirm;
1239back_from_confirm:
1240
1241 saddr = fl4->saddr;
1242 if (!ipc.addr)
1243 daddr = ipc.addr = fl4->daddr;
1244
1245 /* Lockless fast path for the non-corking case. */
1246 if (!corkreq) {
1247 struct inet_cork cork;
1248
1249 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1250 sizeof(struct udphdr), &ipc, &rt,
1251 &cork, msg->msg_flags);
1252 err = PTR_ERR(skb);
1253 if (!IS_ERR_OR_NULL(skb))
1254 err = udp_send_skb(skb, fl4, &cork);
1255 goto out;
1256 }
1257
1258 lock_sock(sk);
1259 if (unlikely(up->pending)) {
1260 /* The socket is already corked while preparing it. */
1261 /* ... which is an evident application bug. --ANK */
1262 release_sock(sk);
1263
1264 net_dbg_ratelimited("socket already corked\n");
1265 err = -EINVAL;
1266 goto out;
1267 }
1268 /*
1269 * Now cork the socket to pend data.
1270 */
1271 fl4 = &inet->cork.fl.u.ip4;
1272 fl4->daddr = daddr;
1273 fl4->saddr = saddr;
1274 fl4->fl4_dport = dport;
1275 fl4->fl4_sport = inet->inet_sport;
1276 up->pending = AF_INET;
1277
1278do_append_data:
1279 up->len += ulen;
1280 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1281 sizeof(struct udphdr), &ipc, &rt,
1282 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1283 if (err)
1284 udp_flush_pending_frames(sk);
1285 else if (!corkreq)
1286 err = udp_push_pending_frames(sk);
1287 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1288 up->pending = 0;
1289 release_sock(sk);
1290
1291out:
1292 ip_rt_put(rt);
1293out_free:
1294 if (free)
1295 kfree(ipc.opt);
1296 if (!err)
1297 return len;
1298 /*
1299 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1300 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1301 * we don't have a good statistic (IpOutDiscards but it can be too many
1302 * things). We could add another new stat but at least for now that
1303 * seems like overkill.
1304 */
1305 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1306 UDP_INC_STATS(sock_net(sk),
1307 UDP_MIB_SNDBUFERRORS, is_udplite);
1308 }
1309 return err;
1310
1311do_confirm:
1312 if (msg->msg_flags & MSG_PROBE)
1313 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1314 if (!(msg->msg_flags&MSG_PROBE) || len)
1315 goto back_from_confirm;
1316 err = 0;
1317 goto out;
1318}
1319EXPORT_SYMBOL(udp_sendmsg);
1320
1321int udp_sendpage(struct sock *sk, struct page *page, int offset,
1322 size_t size, int flags)
1323{
1324 struct inet_sock *inet = inet_sk(sk);
1325 struct udp_sock *up = udp_sk(sk);
1326 int ret;
1327
1328 if (flags & MSG_SENDPAGE_NOTLAST)
1329 flags |= MSG_MORE;
1330
1331 if (!up->pending) {
1332 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1333
1334 /* Call udp_sendmsg to specify destination address which
1335 * sendpage interface can't pass.
1336 * This will succeed only when the socket is connected.
1337 */
1338 ret = udp_sendmsg(sk, &msg, 0);
1339 if (ret < 0)
1340 return ret;
1341 }
1342
1343 lock_sock(sk);
1344
1345 if (unlikely(!up->pending)) {
1346 release_sock(sk);
1347
1348 net_dbg_ratelimited("cork failed\n");
1349 return -EINVAL;
1350 }
1351
1352 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1353 page, offset, size, flags);
1354 if (ret == -EOPNOTSUPP) {
1355 release_sock(sk);
1356 return sock_no_sendpage(sk->sk_socket, page, offset,
1357 size, flags);
1358 }
1359 if (ret < 0) {
1360 udp_flush_pending_frames(sk);
1361 goto out;
1362 }
1363
1364 up->len += size;
1365 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1366 ret = udp_push_pending_frames(sk);
1367 if (!ret)
1368 ret = size;
1369out:
1370 release_sock(sk);
1371 return ret;
1372}
1373
1374#define UDP_SKB_IS_STATELESS 0x80000000
1375
1376/* all head states (dst, sk, nf conntrack) except skb extensions are
1377 * cleared by udp_rcv().
1378 *
1379 * We need to preserve secpath, if present, to eventually process
1380 * IP_CMSG_PASSSEC at recvmsg() time.
1381 *
1382 * Other extensions can be cleared.
1383 */
1384static bool udp_try_make_stateless(struct sk_buff *skb)
1385{
1386 if (!skb_has_extensions(skb))
1387 return true;
1388
1389 if (!secpath_exists(skb)) {
1390 skb_ext_reset(skb);
1391 return true;
1392 }
1393
1394 return false;
1395}
1396
1397static void udp_set_dev_scratch(struct sk_buff *skb)
1398{
1399 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1400
1401 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1402 scratch->_tsize_state = skb->truesize;
1403#if BITS_PER_LONG == 64
1404 scratch->len = skb->len;
1405 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1406 scratch->is_linear = !skb_is_nonlinear(skb);
1407#endif
1408 if (udp_try_make_stateless(skb))
1409 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1410}
1411
1412static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1413{
1414 /* We come here after udp_lib_checksum_complete() returned 0.
1415 * This means that __skb_checksum_complete() might have
1416 * set skb->csum_valid to 1.
1417 * On 64bit platforms, we can set csum_unnecessary
1418 * to true, but only if the skb is not shared.
1419 */
1420#if BITS_PER_LONG == 64
1421 if (!skb_shared(skb))
1422 udp_skb_scratch(skb)->csum_unnecessary = true;
1423#endif
1424}
1425
1426static int udp_skb_truesize(struct sk_buff *skb)
1427{
1428 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1429}
1430
1431static bool udp_skb_has_head_state(struct sk_buff *skb)
1432{
1433 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1434}
1435
1436/* fully reclaim rmem/fwd memory allocated for skb */
1437static void udp_rmem_release(struct sock *sk, int size, int partial,
1438 bool rx_queue_lock_held)
1439{
1440 struct udp_sock *up = udp_sk(sk);
1441 struct sk_buff_head *sk_queue;
1442 int amt;
1443
1444 if (likely(partial)) {
1445 up->forward_deficit += size;
1446 size = up->forward_deficit;
1447 if (size < (sk->sk_rcvbuf >> 2) &&
1448 !skb_queue_empty(&up->reader_queue))
1449 return;
1450 } else {
1451 size += up->forward_deficit;
1452 }
1453 up->forward_deficit = 0;
1454
1455 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1456 * if the called don't held it already
1457 */
1458 sk_queue = &sk->sk_receive_queue;
1459 if (!rx_queue_lock_held)
1460 spin_lock(&sk_queue->lock);
1461
1462
1463 sk->sk_forward_alloc += size;
1464 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1465 sk->sk_forward_alloc -= amt;
1466
1467 if (amt)
1468 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1469
1470 atomic_sub(size, &sk->sk_rmem_alloc);
1471
1472 /* this can save us from acquiring the rx queue lock on next receive */
1473 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1474
1475 if (!rx_queue_lock_held)
1476 spin_unlock(&sk_queue->lock);
1477}
1478
1479/* Note: called with reader_queue.lock held.
1480 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1481 * This avoids a cache line miss while receive_queue lock is held.
1482 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1483 */
1484void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1485{
1486 prefetch(&skb->data);
1487 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1488}
1489EXPORT_SYMBOL(udp_skb_destructor);
1490
1491/* as above, but the caller held the rx queue lock, too */
1492static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1493{
1494 prefetch(&skb->data);
1495 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1496}
1497
1498/* Idea of busylocks is to let producers grab an extra spinlock
1499 * to relieve pressure on the receive_queue spinlock shared by consumer.
1500 * Under flood, this means that only one producer can be in line
1501 * trying to acquire the receive_queue spinlock.
1502 * These busylock can be allocated on a per cpu manner, instead of a
1503 * per socket one (that would consume a cache line per socket)
1504 */
1505static int udp_busylocks_log __read_mostly;
1506static spinlock_t *udp_busylocks __read_mostly;
1507
1508static spinlock_t *busylock_acquire(void *ptr)
1509{
1510 spinlock_t *busy;
1511
1512 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1513 spin_lock(busy);
1514 return busy;
1515}
1516
1517static void busylock_release(spinlock_t *busy)
1518{
1519 if (busy)
1520 spin_unlock(busy);
1521}
1522
1523int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1524{
1525 struct sk_buff_head *list = &sk->sk_receive_queue;
1526 int rmem, delta, amt, err = -ENOMEM;
1527 spinlock_t *busy = NULL;
1528 int size;
1529
1530 /* try to avoid the costly atomic add/sub pair when the receive
1531 * queue is full; always allow at least a packet
1532 */
1533 rmem = atomic_read(&sk->sk_rmem_alloc);
1534 if (rmem > sk->sk_rcvbuf)
1535 goto drop;
1536
1537 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1538 * having linear skbs :
1539 * - Reduce memory overhead and thus increase receive queue capacity
1540 * - Less cache line misses at copyout() time
1541 * - Less work at consume_skb() (less alien page frag freeing)
1542 */
1543 if (rmem > (sk->sk_rcvbuf >> 1)) {
1544 skb_condense(skb);
1545
1546 busy = busylock_acquire(sk);
1547 }
1548 size = skb->truesize;
1549 udp_set_dev_scratch(skb);
1550
1551 /* we drop only if the receive buf is full and the receive
1552 * queue contains some other skb
1553 */
1554 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1555 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1556 goto uncharge_drop;
1557
1558 spin_lock(&list->lock);
1559 if (size >= sk->sk_forward_alloc) {
1560 amt = sk_mem_pages(size);
1561 delta = amt << SK_MEM_QUANTUM_SHIFT;
1562 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1563 err = -ENOBUFS;
1564 spin_unlock(&list->lock);
1565 goto uncharge_drop;
1566 }
1567
1568 sk->sk_forward_alloc += delta;
1569 }
1570
1571 sk->sk_forward_alloc -= size;
1572
1573 /* no need to setup a destructor, we will explicitly release the
1574 * forward allocated memory on dequeue
1575 */
1576 sock_skb_set_dropcount(sk, skb);
1577
1578 __skb_queue_tail(list, skb);
1579 spin_unlock(&list->lock);
1580
1581 if (!sock_flag(sk, SOCK_DEAD))
1582 sk->sk_data_ready(sk);
1583
1584 busylock_release(busy);
1585 return 0;
1586
1587uncharge_drop:
1588 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1589
1590drop:
1591 atomic_inc(&sk->sk_drops);
1592 busylock_release(busy);
1593 return err;
1594}
1595EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1596
1597void udp_destruct_sock(struct sock *sk)
1598{
1599 /* reclaim completely the forward allocated memory */
1600 struct udp_sock *up = udp_sk(sk);
1601 unsigned int total = 0;
1602 struct sk_buff *skb;
1603
1604 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1605 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1606 total += skb->truesize;
1607 kfree_skb(skb);
1608 }
1609 udp_rmem_release(sk, total, 0, true);
1610
1611 inet_sock_destruct(sk);
1612}
1613EXPORT_SYMBOL_GPL(udp_destruct_sock);
1614
1615int udp_init_sock(struct sock *sk)
1616{
1617 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1618 sk->sk_destruct = udp_destruct_sock;
1619 return 0;
1620}
1621EXPORT_SYMBOL_GPL(udp_init_sock);
1622
1623void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1624{
1625 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1626 bool slow = lock_sock_fast(sk);
1627
1628 sk_peek_offset_bwd(sk, len);
1629 unlock_sock_fast(sk, slow);
1630 }
1631
1632 if (!skb_unref(skb))
1633 return;
1634
1635 /* In the more common cases we cleared the head states previously,
1636 * see __udp_queue_rcv_skb().
1637 */
1638 if (unlikely(udp_skb_has_head_state(skb)))
1639 skb_release_head_state(skb);
1640 __consume_stateless_skb(skb);
1641}
1642EXPORT_SYMBOL_GPL(skb_consume_udp);
1643
1644static struct sk_buff *__first_packet_length(struct sock *sk,
1645 struct sk_buff_head *rcvq,
1646 int *total)
1647{
1648 struct sk_buff *skb;
1649
1650 while ((skb = skb_peek(rcvq)) != NULL) {
1651 if (udp_lib_checksum_complete(skb)) {
1652 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1653 IS_UDPLITE(sk));
1654 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1655 IS_UDPLITE(sk));
1656 atomic_inc(&sk->sk_drops);
1657 __skb_unlink(skb, rcvq);
1658 *total += skb->truesize;
1659 kfree_skb(skb);
1660 } else {
1661 udp_skb_csum_unnecessary_set(skb);
1662 break;
1663 }
1664 }
1665 return skb;
1666}
1667
1668/**
1669 * first_packet_length - return length of first packet in receive queue
1670 * @sk: socket
1671 *
1672 * Drops all bad checksum frames, until a valid one is found.
1673 * Returns the length of found skb, or -1 if none is found.
1674 */
1675static int first_packet_length(struct sock *sk)
1676{
1677 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1678 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1679 struct sk_buff *skb;
1680 int total = 0;
1681 int res;
1682
1683 spin_lock_bh(&rcvq->lock);
1684 skb = __first_packet_length(sk, rcvq, &total);
1685 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1686 spin_lock(&sk_queue->lock);
1687 skb_queue_splice_tail_init(sk_queue, rcvq);
1688 spin_unlock(&sk_queue->lock);
1689
1690 skb = __first_packet_length(sk, rcvq, &total);
1691 }
1692 res = skb ? skb->len : -1;
1693 if (total)
1694 udp_rmem_release(sk, total, 1, false);
1695 spin_unlock_bh(&rcvq->lock);
1696 return res;
1697}
1698
1699/*
1700 * IOCTL requests applicable to the UDP protocol
1701 */
1702
1703int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1704{
1705 switch (cmd) {
1706 case SIOCOUTQ:
1707 {
1708 int amount = sk_wmem_alloc_get(sk);
1709
1710 return put_user(amount, (int __user *)arg);
1711 }
1712
1713 case SIOCINQ:
1714 {
1715 int amount = max_t(int, 0, first_packet_length(sk));
1716
1717 return put_user(amount, (int __user *)arg);
1718 }
1719
1720 default:
1721 return -ENOIOCTLCMD;
1722 }
1723
1724 return 0;
1725}
1726EXPORT_SYMBOL(udp_ioctl);
1727
1728struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1729 int noblock, int *off, int *err)
1730{
1731 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1732 struct sk_buff_head *queue;
1733 struct sk_buff *last;
1734 long timeo;
1735 int error;
1736
1737 queue = &udp_sk(sk)->reader_queue;
1738 flags |= noblock ? MSG_DONTWAIT : 0;
1739 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1740 do {
1741 struct sk_buff *skb;
1742
1743 error = sock_error(sk);
1744 if (error)
1745 break;
1746
1747 error = -EAGAIN;
1748 do {
1749 spin_lock_bh(&queue->lock);
1750 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1751 err, &last);
1752 if (skb) {
1753 if (!(flags & MSG_PEEK))
1754 udp_skb_destructor(sk, skb);
1755 spin_unlock_bh(&queue->lock);
1756 return skb;
1757 }
1758
1759 if (skb_queue_empty_lockless(sk_queue)) {
1760 spin_unlock_bh(&queue->lock);
1761 goto busy_check;
1762 }
1763
1764 /* refill the reader queue and walk it again
1765 * keep both queues locked to avoid re-acquiring
1766 * the sk_receive_queue lock if fwd memory scheduling
1767 * is needed.
1768 */
1769 spin_lock(&sk_queue->lock);
1770 skb_queue_splice_tail_init(sk_queue, queue);
1771
1772 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1773 err, &last);
1774 if (skb && !(flags & MSG_PEEK))
1775 udp_skb_dtor_locked(sk, skb);
1776 spin_unlock(&sk_queue->lock);
1777 spin_unlock_bh(&queue->lock);
1778 if (skb)
1779 return skb;
1780
1781busy_check:
1782 if (!sk_can_busy_loop(sk))
1783 break;
1784
1785 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1786 } while (!skb_queue_empty_lockless(sk_queue));
1787
1788 /* sk_queue is empty, reader_queue may contain peeked packets */
1789 } while (timeo &&
1790 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1791 &error, &timeo,
1792 (struct sk_buff *)sk_queue));
1793
1794 *err = error;
1795 return NULL;
1796}
1797EXPORT_SYMBOL(__skb_recv_udp);
1798
1799int udp_read_sock(struct sock *sk, read_descriptor_t *desc,
1800 sk_read_actor_t recv_actor)
1801{
1802 int copied = 0;
1803
1804 while (1) {
1805 struct sk_buff *skb;
1806 int err, used;
1807
1808 skb = skb_recv_udp(sk, 0, 1, &err);
1809 if (!skb)
1810 return err;
1811 used = recv_actor(desc, skb, 0, skb->len);
1812 if (used <= 0) {
1813 if (!copied)
1814 copied = used;
1815 kfree_skb(skb);
1816 break;
1817 } else if (used <= skb->len) {
1818 copied += used;
1819 }
1820
1821 kfree_skb(skb);
1822 if (!desc->count)
1823 break;
1824 }
1825
1826 return copied;
1827}
1828EXPORT_SYMBOL(udp_read_sock);
1829
1830/*
1831 * This should be easy, if there is something there we
1832 * return it, otherwise we block.
1833 */
1834
1835int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1836 int flags, int *addr_len)
1837{
1838 struct inet_sock *inet = inet_sk(sk);
1839 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1840 struct sk_buff *skb;
1841 unsigned int ulen, copied;
1842 int off, err, peeking = flags & MSG_PEEK;
1843 int is_udplite = IS_UDPLITE(sk);
1844 bool checksum_valid = false;
1845
1846 if (flags & MSG_ERRQUEUE)
1847 return ip_recv_error(sk, msg, len, addr_len);
1848
1849try_again:
1850 off = sk_peek_offset(sk, flags);
1851 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1852 if (!skb)
1853 return err;
1854
1855 ulen = udp_skb_len(skb);
1856 copied = len;
1857 if (copied > ulen - off)
1858 copied = ulen - off;
1859 else if (copied < ulen)
1860 msg->msg_flags |= MSG_TRUNC;
1861
1862 /*
1863 * If checksum is needed at all, try to do it while copying the
1864 * data. If the data is truncated, or if we only want a partial
1865 * coverage checksum (UDP-Lite), do it before the copy.
1866 */
1867
1868 if (copied < ulen || peeking ||
1869 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1870 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1871 !__udp_lib_checksum_complete(skb);
1872 if (!checksum_valid)
1873 goto csum_copy_err;
1874 }
1875
1876 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1877 if (udp_skb_is_linear(skb))
1878 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1879 else
1880 err = skb_copy_datagram_msg(skb, off, msg, copied);
1881 } else {
1882 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1883
1884 if (err == -EINVAL)
1885 goto csum_copy_err;
1886 }
1887
1888 if (unlikely(err)) {
1889 if (!peeking) {
1890 atomic_inc(&sk->sk_drops);
1891 UDP_INC_STATS(sock_net(sk),
1892 UDP_MIB_INERRORS, is_udplite);
1893 }
1894 kfree_skb(skb);
1895 return err;
1896 }
1897
1898 if (!peeking)
1899 UDP_INC_STATS(sock_net(sk),
1900 UDP_MIB_INDATAGRAMS, is_udplite);
1901
1902 sock_recv_ts_and_drops(msg, sk, skb);
1903
1904 /* Copy the address. */
1905 if (sin) {
1906 sin->sin_family = AF_INET;
1907 sin->sin_port = udp_hdr(skb)->source;
1908 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1909 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1910 *addr_len = sizeof(*sin);
1911
1912 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1913 (struct sockaddr *)sin);
1914 }
1915
1916 if (udp_sk(sk)->gro_enabled)
1917 udp_cmsg_recv(msg, sk, skb);
1918
1919 if (inet->cmsg_flags)
1920 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1921
1922 err = copied;
1923 if (flags & MSG_TRUNC)
1924 err = ulen;
1925
1926 skb_consume_udp(sk, skb, peeking ? -err : err);
1927 return err;
1928
1929csum_copy_err:
1930 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1931 udp_skb_destructor)) {
1932 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1933 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1934 }
1935 kfree_skb(skb);
1936
1937 /* starting over for a new packet, but check if we need to yield */
1938 cond_resched();
1939 msg->msg_flags &= ~MSG_TRUNC;
1940 goto try_again;
1941}
1942
1943int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1944{
1945 /* This check is replicated from __ip4_datagram_connect() and
1946 * intended to prevent BPF program called below from accessing bytes
1947 * that are out of the bound specified by user in addr_len.
1948 */
1949 if (addr_len < sizeof(struct sockaddr_in))
1950 return -EINVAL;
1951
1952 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1953}
1954EXPORT_SYMBOL(udp_pre_connect);
1955
1956int __udp_disconnect(struct sock *sk, int flags)
1957{
1958 struct inet_sock *inet = inet_sk(sk);
1959 /*
1960 * 1003.1g - break association.
1961 */
1962
1963 sk->sk_state = TCP_CLOSE;
1964 inet->inet_daddr = 0;
1965 inet->inet_dport = 0;
1966 sock_rps_reset_rxhash(sk);
1967 sk->sk_bound_dev_if = 0;
1968 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1969 inet_reset_saddr(sk);
1970 if (sk->sk_prot->rehash &&
1971 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1972 sk->sk_prot->rehash(sk);
1973 }
1974
1975 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1976 sk->sk_prot->unhash(sk);
1977 inet->inet_sport = 0;
1978 }
1979 sk_dst_reset(sk);
1980 return 0;
1981}
1982EXPORT_SYMBOL(__udp_disconnect);
1983
1984int udp_disconnect(struct sock *sk, int flags)
1985{
1986 lock_sock(sk);
1987 __udp_disconnect(sk, flags);
1988 release_sock(sk);
1989 return 0;
1990}
1991EXPORT_SYMBOL(udp_disconnect);
1992
1993void udp_lib_unhash(struct sock *sk)
1994{
1995 if (sk_hashed(sk)) {
1996 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1997 struct udp_hslot *hslot, *hslot2;
1998
1999 hslot = udp_hashslot(udptable, sock_net(sk),
2000 udp_sk(sk)->udp_port_hash);
2001 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2002
2003 spin_lock_bh(&hslot->lock);
2004 if (rcu_access_pointer(sk->sk_reuseport_cb))
2005 reuseport_detach_sock(sk);
2006 if (sk_del_node_init_rcu(sk)) {
2007 hslot->count--;
2008 inet_sk(sk)->inet_num = 0;
2009 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2010
2011 spin_lock(&hslot2->lock);
2012 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2013 hslot2->count--;
2014 spin_unlock(&hslot2->lock);
2015 }
2016 spin_unlock_bh(&hslot->lock);
2017 }
2018}
2019EXPORT_SYMBOL(udp_lib_unhash);
2020
2021/*
2022 * inet_rcv_saddr was changed, we must rehash secondary hash
2023 */
2024void udp_lib_rehash(struct sock *sk, u16 newhash)
2025{
2026 if (sk_hashed(sk)) {
2027 struct udp_table *udptable = sk->sk_prot->h.udp_table;
2028 struct udp_hslot *hslot, *hslot2, *nhslot2;
2029
2030 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2031 nhslot2 = udp_hashslot2(udptable, newhash);
2032 udp_sk(sk)->udp_portaddr_hash = newhash;
2033
2034 if (hslot2 != nhslot2 ||
2035 rcu_access_pointer(sk->sk_reuseport_cb)) {
2036 hslot = udp_hashslot(udptable, sock_net(sk),
2037 udp_sk(sk)->udp_port_hash);
2038 /* we must lock primary chain too */
2039 spin_lock_bh(&hslot->lock);
2040 if (rcu_access_pointer(sk->sk_reuseport_cb))
2041 reuseport_detach_sock(sk);
2042
2043 if (hslot2 != nhslot2) {
2044 spin_lock(&hslot2->lock);
2045 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2046 hslot2->count--;
2047 spin_unlock(&hslot2->lock);
2048
2049 spin_lock(&nhslot2->lock);
2050 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2051 &nhslot2->head);
2052 nhslot2->count++;
2053 spin_unlock(&nhslot2->lock);
2054 }
2055
2056 spin_unlock_bh(&hslot->lock);
2057 }
2058 }
2059}
2060EXPORT_SYMBOL(udp_lib_rehash);
2061
2062void udp_v4_rehash(struct sock *sk)
2063{
2064 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2065 inet_sk(sk)->inet_rcv_saddr,
2066 inet_sk(sk)->inet_num);
2067 udp_lib_rehash(sk, new_hash);
2068}
2069
2070static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2071{
2072 int rc;
2073
2074 if (inet_sk(sk)->inet_daddr) {
2075 sock_rps_save_rxhash(sk, skb);
2076 sk_mark_napi_id(sk, skb);
2077 sk_incoming_cpu_update(sk);
2078 } else {
2079 sk_mark_napi_id_once(sk, skb);
2080 }
2081
2082 rc = __udp_enqueue_schedule_skb(sk, skb);
2083 if (rc < 0) {
2084 int is_udplite = IS_UDPLITE(sk);
2085
2086 /* Note that an ENOMEM error is charged twice */
2087 if (rc == -ENOMEM)
2088 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2089 is_udplite);
2090 else
2091 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2092 is_udplite);
2093 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2094 kfree_skb(skb);
2095 trace_udp_fail_queue_rcv_skb(rc, sk);
2096 return -1;
2097 }
2098
2099 return 0;
2100}
2101
2102/* returns:
2103 * -1: error
2104 * 0: success
2105 * >0: "udp encap" protocol resubmission
2106 *
2107 * Note that in the success and error cases, the skb is assumed to
2108 * have either been requeued or freed.
2109 */
2110static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2111{
2112 struct udp_sock *up = udp_sk(sk);
2113 int is_udplite = IS_UDPLITE(sk);
2114
2115 /*
2116 * Charge it to the socket, dropping if the queue is full.
2117 */
2118 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2119 goto drop;
2120 nf_reset_ct(skb);
2121
2122 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2123 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2124
2125 /*
2126 * This is an encapsulation socket so pass the skb to
2127 * the socket's udp_encap_rcv() hook. Otherwise, just
2128 * fall through and pass this up the UDP socket.
2129 * up->encap_rcv() returns the following value:
2130 * =0 if skb was successfully passed to the encap
2131 * handler or was discarded by it.
2132 * >0 if skb should be passed on to UDP.
2133 * <0 if skb should be resubmitted as proto -N
2134 */
2135
2136 /* if we're overly short, let UDP handle it */
2137 encap_rcv = READ_ONCE(up->encap_rcv);
2138 if (encap_rcv) {
2139 int ret;
2140
2141 /* Verify checksum before giving to encap */
2142 if (udp_lib_checksum_complete(skb))
2143 goto csum_error;
2144
2145 ret = encap_rcv(sk, skb);
2146 if (ret <= 0) {
2147 __UDP_INC_STATS(sock_net(sk),
2148 UDP_MIB_INDATAGRAMS,
2149 is_udplite);
2150 return -ret;
2151 }
2152 }
2153
2154 /* FALLTHROUGH -- it's a UDP Packet */
2155 }
2156
2157 /*
2158 * UDP-Lite specific tests, ignored on UDP sockets
2159 */
2160 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2161
2162 /*
2163 * MIB statistics other than incrementing the error count are
2164 * disabled for the following two types of errors: these depend
2165 * on the application settings, not on the functioning of the
2166 * protocol stack as such.
2167 *
2168 * RFC 3828 here recommends (sec 3.3): "There should also be a
2169 * way ... to ... at least let the receiving application block
2170 * delivery of packets with coverage values less than a value
2171 * provided by the application."
2172 */
2173 if (up->pcrlen == 0) { /* full coverage was set */
2174 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2175 UDP_SKB_CB(skb)->cscov, skb->len);
2176 goto drop;
2177 }
2178 /* The next case involves violating the min. coverage requested
2179 * by the receiver. This is subtle: if receiver wants x and x is
2180 * greater than the buffersize/MTU then receiver will complain
2181 * that it wants x while sender emits packets of smaller size y.
2182 * Therefore the above ...()->partial_cov statement is essential.
2183 */
2184 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2185 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2186 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2187 goto drop;
2188 }
2189 }
2190
2191 prefetch(&sk->sk_rmem_alloc);
2192 if (rcu_access_pointer(sk->sk_filter) &&
2193 udp_lib_checksum_complete(skb))
2194 goto csum_error;
2195
2196 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2197 goto drop;
2198
2199 udp_csum_pull_header(skb);
2200
2201 ipv4_pktinfo_prepare(sk, skb);
2202 return __udp_queue_rcv_skb(sk, skb);
2203
2204csum_error:
2205 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2206drop:
2207 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2208 atomic_inc(&sk->sk_drops);
2209 kfree_skb(skb);
2210 return -1;
2211}
2212
2213static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2214{
2215 struct sk_buff *next, *segs;
2216 int ret;
2217
2218 if (likely(!udp_unexpected_gso(sk, skb)))
2219 return udp_queue_rcv_one_skb(sk, skb);
2220
2221 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2222 __skb_push(skb, -skb_mac_offset(skb));
2223 segs = udp_rcv_segment(sk, skb, true);
2224 skb_list_walk_safe(segs, skb, next) {
2225 __skb_pull(skb, skb_transport_offset(skb));
2226
2227 udp_post_segment_fix_csum(skb);
2228 ret = udp_queue_rcv_one_skb(sk, skb);
2229 if (ret > 0)
2230 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2231 }
2232 return 0;
2233}
2234
2235/* For TCP sockets, sk_rx_dst is protected by socket lock
2236 * For UDP, we use xchg() to guard against concurrent changes.
2237 */
2238bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2239{
2240 struct dst_entry *old;
2241
2242 if (dst_hold_safe(dst)) {
2243 old = xchg(&sk->sk_rx_dst, dst);
2244 dst_release(old);
2245 return old != dst;
2246 }
2247 return false;
2248}
2249EXPORT_SYMBOL(udp_sk_rx_dst_set);
2250
2251/*
2252 * Multicasts and broadcasts go to each listener.
2253 *
2254 * Note: called only from the BH handler context.
2255 */
2256static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2257 struct udphdr *uh,
2258 __be32 saddr, __be32 daddr,
2259 struct udp_table *udptable,
2260 int proto)
2261{
2262 struct sock *sk, *first = NULL;
2263 unsigned short hnum = ntohs(uh->dest);
2264 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2265 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2266 unsigned int offset = offsetof(typeof(*sk), sk_node);
2267 int dif = skb->dev->ifindex;
2268 int sdif = inet_sdif(skb);
2269 struct hlist_node *node;
2270 struct sk_buff *nskb;
2271
2272 if (use_hash2) {
2273 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2274 udptable->mask;
2275 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2276start_lookup:
2277 hslot = &udptable->hash2[hash2];
2278 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2279 }
2280
2281 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2282 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2283 uh->source, saddr, dif, sdif, hnum))
2284 continue;
2285
2286 if (!first) {
2287 first = sk;
2288 continue;
2289 }
2290 nskb = skb_clone(skb, GFP_ATOMIC);
2291
2292 if (unlikely(!nskb)) {
2293 atomic_inc(&sk->sk_drops);
2294 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2295 IS_UDPLITE(sk));
2296 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2297 IS_UDPLITE(sk));
2298 continue;
2299 }
2300 if (udp_queue_rcv_skb(sk, nskb) > 0)
2301 consume_skb(nskb);
2302 }
2303
2304 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2305 if (use_hash2 && hash2 != hash2_any) {
2306 hash2 = hash2_any;
2307 goto start_lookup;
2308 }
2309
2310 if (first) {
2311 if (udp_queue_rcv_skb(first, skb) > 0)
2312 consume_skb(skb);
2313 } else {
2314 kfree_skb(skb);
2315 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2316 proto == IPPROTO_UDPLITE);
2317 }
2318 return 0;
2319}
2320
2321/* Initialize UDP checksum. If exited with zero value (success),
2322 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2323 * Otherwise, csum completion requires checksumming packet body,
2324 * including udp header and folding it to skb->csum.
2325 */
2326static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2327 int proto)
2328{
2329 int err;
2330
2331 UDP_SKB_CB(skb)->partial_cov = 0;
2332 UDP_SKB_CB(skb)->cscov = skb->len;
2333
2334 if (proto == IPPROTO_UDPLITE) {
2335 err = udplite_checksum_init(skb, uh);
2336 if (err)
2337 return err;
2338
2339 if (UDP_SKB_CB(skb)->partial_cov) {
2340 skb->csum = inet_compute_pseudo(skb, proto);
2341 return 0;
2342 }
2343 }
2344
2345 /* Note, we are only interested in != 0 or == 0, thus the
2346 * force to int.
2347 */
2348 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2349 inet_compute_pseudo);
2350 if (err)
2351 return err;
2352
2353 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2354 /* If SW calculated the value, we know it's bad */
2355 if (skb->csum_complete_sw)
2356 return 1;
2357
2358 /* HW says the value is bad. Let's validate that.
2359 * skb->csum is no longer the full packet checksum,
2360 * so don't treat it as such.
2361 */
2362 skb_checksum_complete_unset(skb);
2363 }
2364
2365 return 0;
2366}
2367
2368/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2369 * return code conversion for ip layer consumption
2370 */
2371static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2372 struct udphdr *uh)
2373{
2374 int ret;
2375
2376 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2377 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2378
2379 ret = udp_queue_rcv_skb(sk, skb);
2380
2381 /* a return value > 0 means to resubmit the input, but
2382 * it wants the return to be -protocol, or 0
2383 */
2384 if (ret > 0)
2385 return -ret;
2386 return 0;
2387}
2388
2389/*
2390 * All we need to do is get the socket, and then do a checksum.
2391 */
2392
2393int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2394 int proto)
2395{
2396 struct sock *sk;
2397 struct udphdr *uh;
2398 unsigned short ulen;
2399 struct rtable *rt = skb_rtable(skb);
2400 __be32 saddr, daddr;
2401 struct net *net = dev_net(skb->dev);
2402 bool refcounted;
2403
2404 /*
2405 * Validate the packet.
2406 */
2407 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2408 goto drop; /* No space for header. */
2409
2410 uh = udp_hdr(skb);
2411 ulen = ntohs(uh->len);
2412 saddr = ip_hdr(skb)->saddr;
2413 daddr = ip_hdr(skb)->daddr;
2414
2415 if (ulen > skb->len)
2416 goto short_packet;
2417
2418 if (proto == IPPROTO_UDP) {
2419 /* UDP validates ulen. */
2420 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2421 goto short_packet;
2422 uh = udp_hdr(skb);
2423 }
2424
2425 if (udp4_csum_init(skb, uh, proto))
2426 goto csum_error;
2427
2428 sk = skb_steal_sock(skb, &refcounted);
2429 if (sk) {
2430 struct dst_entry *dst = skb_dst(skb);
2431 int ret;
2432
2433 if (unlikely(sk->sk_rx_dst != dst))
2434 udp_sk_rx_dst_set(sk, dst);
2435
2436 ret = udp_unicast_rcv_skb(sk, skb, uh);
2437 if (refcounted)
2438 sock_put(sk);
2439 return ret;
2440 }
2441
2442 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2443 return __udp4_lib_mcast_deliver(net, skb, uh,
2444 saddr, daddr, udptable, proto);
2445
2446 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2447 if (sk)
2448 return udp_unicast_rcv_skb(sk, skb, uh);
2449
2450 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2451 goto drop;
2452 nf_reset_ct(skb);
2453
2454 /* No socket. Drop packet silently, if checksum is wrong */
2455 if (udp_lib_checksum_complete(skb))
2456 goto csum_error;
2457
2458 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2459 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2460
2461 /*
2462 * Hmm. We got an UDP packet to a port to which we
2463 * don't wanna listen. Ignore it.
2464 */
2465 kfree_skb(skb);
2466 return 0;
2467
2468short_packet:
2469 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2470 proto == IPPROTO_UDPLITE ? "Lite" : "",
2471 &saddr, ntohs(uh->source),
2472 ulen, skb->len,
2473 &daddr, ntohs(uh->dest));
2474 goto drop;
2475
2476csum_error:
2477 /*
2478 * RFC1122: OK. Discards the bad packet silently (as far as
2479 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2480 */
2481 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2482 proto == IPPROTO_UDPLITE ? "Lite" : "",
2483 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2484 ulen);
2485 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2486drop:
2487 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2488 kfree_skb(skb);
2489 return 0;
2490}
2491
2492/* We can only early demux multicast if there is a single matching socket.
2493 * If more than one socket found returns NULL
2494 */
2495static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2496 __be16 loc_port, __be32 loc_addr,
2497 __be16 rmt_port, __be32 rmt_addr,
2498 int dif, int sdif)
2499{
2500 struct sock *sk, *result;
2501 unsigned short hnum = ntohs(loc_port);
2502 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2503 struct udp_hslot *hslot = &udp_table.hash[slot];
2504
2505 /* Do not bother scanning a too big list */
2506 if (hslot->count > 10)
2507 return NULL;
2508
2509 result = NULL;
2510 sk_for_each_rcu(sk, &hslot->head) {
2511 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2512 rmt_port, rmt_addr, dif, sdif, hnum)) {
2513 if (result)
2514 return NULL;
2515 result = sk;
2516 }
2517 }
2518
2519 return result;
2520}
2521
2522/* For unicast we should only early demux connected sockets or we can
2523 * break forwarding setups. The chains here can be long so only check
2524 * if the first socket is an exact match and if not move on.
2525 */
2526static struct sock *__udp4_lib_demux_lookup(struct net *net,
2527 __be16 loc_port, __be32 loc_addr,
2528 __be16 rmt_port, __be32 rmt_addr,
2529 int dif, int sdif)
2530{
2531 unsigned short hnum = ntohs(loc_port);
2532 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2533 unsigned int slot2 = hash2 & udp_table.mask;
2534 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2535 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2536 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2537 struct sock *sk;
2538
2539 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2540 if (INET_MATCH(sk, net, acookie, rmt_addr,
2541 loc_addr, ports, dif, sdif))
2542 return sk;
2543 /* Only check first socket in chain */
2544 break;
2545 }
2546 return NULL;
2547}
2548
2549int udp_v4_early_demux(struct sk_buff *skb)
2550{
2551 struct net *net = dev_net(skb->dev);
2552 struct in_device *in_dev = NULL;
2553 const struct iphdr *iph;
2554 const struct udphdr *uh;
2555 struct sock *sk = NULL;
2556 struct dst_entry *dst;
2557 int dif = skb->dev->ifindex;
2558 int sdif = inet_sdif(skb);
2559 int ours;
2560
2561 /* validate the packet */
2562 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2563 return 0;
2564
2565 iph = ip_hdr(skb);
2566 uh = udp_hdr(skb);
2567
2568 if (skb->pkt_type == PACKET_MULTICAST) {
2569 in_dev = __in_dev_get_rcu(skb->dev);
2570
2571 if (!in_dev)
2572 return 0;
2573
2574 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2575 iph->protocol);
2576 if (!ours)
2577 return 0;
2578
2579 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2580 uh->source, iph->saddr,
2581 dif, sdif);
2582 } else if (skb->pkt_type == PACKET_HOST) {
2583 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2584 uh->source, iph->saddr, dif, sdif);
2585 }
2586
2587 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2588 return 0;
2589
2590 skb->sk = sk;
2591 skb->destructor = sock_efree;
2592 dst = READ_ONCE(sk->sk_rx_dst);
2593
2594 if (dst)
2595 dst = dst_check(dst, 0);
2596 if (dst) {
2597 u32 itag = 0;
2598
2599 /* set noref for now.
2600 * any place which wants to hold dst has to call
2601 * dst_hold_safe()
2602 */
2603 skb_dst_set_noref(skb, dst);
2604
2605 /* for unconnected multicast sockets we need to validate
2606 * the source on each packet
2607 */
2608 if (!inet_sk(sk)->inet_daddr && in_dev)
2609 return ip_mc_validate_source(skb, iph->daddr,
2610 iph->saddr,
2611 iph->tos & IPTOS_RT_MASK,
2612 skb->dev, in_dev, &itag);
2613 }
2614 return 0;
2615}
2616
2617int udp_rcv(struct sk_buff *skb)
2618{
2619 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2620}
2621
2622void udp_destroy_sock(struct sock *sk)
2623{
2624 struct udp_sock *up = udp_sk(sk);
2625 bool slow = lock_sock_fast(sk);
2626
2627 /* protects from races with udp_abort() */
2628 sock_set_flag(sk, SOCK_DEAD);
2629 udp_flush_pending_frames(sk);
2630 unlock_sock_fast(sk, slow);
2631 if (static_branch_unlikely(&udp_encap_needed_key)) {
2632 if (up->encap_type) {
2633 void (*encap_destroy)(struct sock *sk);
2634 encap_destroy = READ_ONCE(up->encap_destroy);
2635 if (encap_destroy)
2636 encap_destroy(sk);
2637 }
2638 if (up->encap_enabled)
2639 static_branch_dec(&udp_encap_needed_key);
2640 }
2641}
2642
2643/*
2644 * Socket option code for UDP
2645 */
2646int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2647 sockptr_t optval, unsigned int optlen,
2648 int (*push_pending_frames)(struct sock *))
2649{
2650 struct udp_sock *up = udp_sk(sk);
2651 int val, valbool;
2652 int err = 0;
2653 int is_udplite = IS_UDPLITE(sk);
2654
2655 if (optlen < sizeof(int))
2656 return -EINVAL;
2657
2658 if (copy_from_sockptr(&val, optval, sizeof(val)))
2659 return -EFAULT;
2660
2661 valbool = val ? 1 : 0;
2662
2663 switch (optname) {
2664 case UDP_CORK:
2665 if (val != 0) {
2666 WRITE_ONCE(up->corkflag, 1);
2667 } else {
2668 WRITE_ONCE(up->corkflag, 0);
2669 lock_sock(sk);
2670 push_pending_frames(sk);
2671 release_sock(sk);
2672 }
2673 break;
2674
2675 case UDP_ENCAP:
2676 switch (val) {
2677 case 0:
2678#ifdef CONFIG_XFRM
2679 case UDP_ENCAP_ESPINUDP:
2680 case UDP_ENCAP_ESPINUDP_NON_IKE:
2681#if IS_ENABLED(CONFIG_IPV6)
2682 if (sk->sk_family == AF_INET6)
2683 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2684 else
2685#endif
2686 up->encap_rcv = xfrm4_udp_encap_rcv;
2687#endif
2688 fallthrough;
2689 case UDP_ENCAP_L2TPINUDP:
2690 up->encap_type = val;
2691 lock_sock(sk);
2692 udp_tunnel_encap_enable(sk->sk_socket);
2693 release_sock(sk);
2694 break;
2695 default:
2696 err = -ENOPROTOOPT;
2697 break;
2698 }
2699 break;
2700
2701 case UDP_NO_CHECK6_TX:
2702 up->no_check6_tx = valbool;
2703 break;
2704
2705 case UDP_NO_CHECK6_RX:
2706 up->no_check6_rx = valbool;
2707 break;
2708
2709 case UDP_SEGMENT:
2710 if (val < 0 || val > USHRT_MAX)
2711 return -EINVAL;
2712 WRITE_ONCE(up->gso_size, val);
2713 break;
2714
2715 case UDP_GRO:
2716 lock_sock(sk);
2717
2718 /* when enabling GRO, accept the related GSO packet type */
2719 if (valbool)
2720 udp_tunnel_encap_enable(sk->sk_socket);
2721 up->gro_enabled = valbool;
2722 up->accept_udp_l4 = valbool;
2723 release_sock(sk);
2724 break;
2725
2726 /*
2727 * UDP-Lite's partial checksum coverage (RFC 3828).
2728 */
2729 /* The sender sets actual checksum coverage length via this option.
2730 * The case coverage > packet length is handled by send module. */
2731 case UDPLITE_SEND_CSCOV:
2732 if (!is_udplite) /* Disable the option on UDP sockets */
2733 return -ENOPROTOOPT;
2734 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2735 val = 8;
2736 else if (val > USHRT_MAX)
2737 val = USHRT_MAX;
2738 up->pcslen = val;
2739 up->pcflag |= UDPLITE_SEND_CC;
2740 break;
2741
2742 /* The receiver specifies a minimum checksum coverage value. To make
2743 * sense, this should be set to at least 8 (as done below). If zero is
2744 * used, this again means full checksum coverage. */
2745 case UDPLITE_RECV_CSCOV:
2746 if (!is_udplite) /* Disable the option on UDP sockets */
2747 return -ENOPROTOOPT;
2748 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2749 val = 8;
2750 else if (val > USHRT_MAX)
2751 val = USHRT_MAX;
2752 up->pcrlen = val;
2753 up->pcflag |= UDPLITE_RECV_CC;
2754 break;
2755
2756 default:
2757 err = -ENOPROTOOPT;
2758 break;
2759 }
2760
2761 return err;
2762}
2763EXPORT_SYMBOL(udp_lib_setsockopt);
2764
2765int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2766 unsigned int optlen)
2767{
2768 if (level == SOL_UDP || level == SOL_UDPLITE)
2769 return udp_lib_setsockopt(sk, level, optname,
2770 optval, optlen,
2771 udp_push_pending_frames);
2772 return ip_setsockopt(sk, level, optname, optval, optlen);
2773}
2774
2775int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2776 char __user *optval, int __user *optlen)
2777{
2778 struct udp_sock *up = udp_sk(sk);
2779 int val, len;
2780
2781 if (get_user(len, optlen))
2782 return -EFAULT;
2783
2784 len = min_t(unsigned int, len, sizeof(int));
2785
2786 if (len < 0)
2787 return -EINVAL;
2788
2789 switch (optname) {
2790 case UDP_CORK:
2791 val = READ_ONCE(up->corkflag);
2792 break;
2793
2794 case UDP_ENCAP:
2795 val = up->encap_type;
2796 break;
2797
2798 case UDP_NO_CHECK6_TX:
2799 val = up->no_check6_tx;
2800 break;
2801
2802 case UDP_NO_CHECK6_RX:
2803 val = up->no_check6_rx;
2804 break;
2805
2806 case UDP_SEGMENT:
2807 val = READ_ONCE(up->gso_size);
2808 break;
2809
2810 case UDP_GRO:
2811 val = up->gro_enabled;
2812 break;
2813
2814 /* The following two cannot be changed on UDP sockets, the return is
2815 * always 0 (which corresponds to the full checksum coverage of UDP). */
2816 case UDPLITE_SEND_CSCOV:
2817 val = up->pcslen;
2818 break;
2819
2820 case UDPLITE_RECV_CSCOV:
2821 val = up->pcrlen;
2822 break;
2823
2824 default:
2825 return -ENOPROTOOPT;
2826 }
2827
2828 if (put_user(len, optlen))
2829 return -EFAULT;
2830 if (copy_to_user(optval, &val, len))
2831 return -EFAULT;
2832 return 0;
2833}
2834EXPORT_SYMBOL(udp_lib_getsockopt);
2835
2836int udp_getsockopt(struct sock *sk, int level, int optname,
2837 char __user *optval, int __user *optlen)
2838{
2839 if (level == SOL_UDP || level == SOL_UDPLITE)
2840 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2841 return ip_getsockopt(sk, level, optname, optval, optlen);
2842}
2843
2844/**
2845 * udp_poll - wait for a UDP event.
2846 * @file: - file struct
2847 * @sock: - socket
2848 * @wait: - poll table
2849 *
2850 * This is same as datagram poll, except for the special case of
2851 * blocking sockets. If application is using a blocking fd
2852 * and a packet with checksum error is in the queue;
2853 * then it could get return from select indicating data available
2854 * but then block when reading it. Add special case code
2855 * to work around these arguably broken applications.
2856 */
2857__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2858{
2859 __poll_t mask = datagram_poll(file, sock, wait);
2860 struct sock *sk = sock->sk;
2861
2862 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2863 mask |= EPOLLIN | EPOLLRDNORM;
2864
2865 /* Check for false positives due to checksum errors */
2866 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2867 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2868 mask &= ~(EPOLLIN | EPOLLRDNORM);
2869
2870 return mask;
2871
2872}
2873EXPORT_SYMBOL(udp_poll);
2874
2875int udp_abort(struct sock *sk, int err)
2876{
2877 lock_sock(sk);
2878
2879 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2880 * with close()
2881 */
2882 if (sock_flag(sk, SOCK_DEAD))
2883 goto out;
2884
2885 sk->sk_err = err;
2886 sk_error_report(sk);
2887 __udp_disconnect(sk, 0);
2888
2889out:
2890 release_sock(sk);
2891
2892 return 0;
2893}
2894EXPORT_SYMBOL_GPL(udp_abort);
2895
2896struct proto udp_prot = {
2897 .name = "UDP",
2898 .owner = THIS_MODULE,
2899 .close = udp_lib_close,
2900 .pre_connect = udp_pre_connect,
2901 .connect = ip4_datagram_connect,
2902 .disconnect = udp_disconnect,
2903 .ioctl = udp_ioctl,
2904 .init = udp_init_sock,
2905 .destroy = udp_destroy_sock,
2906 .setsockopt = udp_setsockopt,
2907 .getsockopt = udp_getsockopt,
2908 .sendmsg = udp_sendmsg,
2909 .recvmsg = udp_recvmsg,
2910 .sendpage = udp_sendpage,
2911 .release_cb = ip4_datagram_release_cb,
2912 .hash = udp_lib_hash,
2913 .unhash = udp_lib_unhash,
2914 .rehash = udp_v4_rehash,
2915 .get_port = udp_v4_get_port,
2916#ifdef CONFIG_BPF_SYSCALL
2917 .psock_update_sk_prot = udp_bpf_update_proto,
2918#endif
2919 .memory_allocated = &udp_memory_allocated,
2920 .sysctl_mem = sysctl_udp_mem,
2921 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2922 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2923 .obj_size = sizeof(struct udp_sock),
2924 .h.udp_table = &udp_table,
2925 .diag_destroy = udp_abort,
2926};
2927EXPORT_SYMBOL(udp_prot);
2928
2929/* ------------------------------------------------------------------------ */
2930#ifdef CONFIG_PROC_FS
2931
2932static struct sock *udp_get_first(struct seq_file *seq, int start)
2933{
2934 struct sock *sk;
2935 struct udp_seq_afinfo *afinfo;
2936 struct udp_iter_state *state = seq->private;
2937 struct net *net = seq_file_net(seq);
2938
2939 if (state->bpf_seq_afinfo)
2940 afinfo = state->bpf_seq_afinfo;
2941 else
2942 afinfo = PDE_DATA(file_inode(seq->file));
2943
2944 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2945 ++state->bucket) {
2946 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2947
2948 if (hlist_empty(&hslot->head))
2949 continue;
2950
2951 spin_lock_bh(&hslot->lock);
2952 sk_for_each(sk, &hslot->head) {
2953 if (!net_eq(sock_net(sk), net))
2954 continue;
2955 if (afinfo->family == AF_UNSPEC ||
2956 sk->sk_family == afinfo->family)
2957 goto found;
2958 }
2959 spin_unlock_bh(&hslot->lock);
2960 }
2961 sk = NULL;
2962found:
2963 return sk;
2964}
2965
2966static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2967{
2968 struct udp_seq_afinfo *afinfo;
2969 struct udp_iter_state *state = seq->private;
2970 struct net *net = seq_file_net(seq);
2971
2972 if (state->bpf_seq_afinfo)
2973 afinfo = state->bpf_seq_afinfo;
2974 else
2975 afinfo = PDE_DATA(file_inode(seq->file));
2976
2977 do {
2978 sk = sk_next(sk);
2979 } while (sk && (!net_eq(sock_net(sk), net) ||
2980 (afinfo->family != AF_UNSPEC &&
2981 sk->sk_family != afinfo->family)));
2982
2983 if (!sk) {
2984 if (state->bucket <= afinfo->udp_table->mask)
2985 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2986 return udp_get_first(seq, state->bucket + 1);
2987 }
2988 return sk;
2989}
2990
2991static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2992{
2993 struct sock *sk = udp_get_first(seq, 0);
2994
2995 if (sk)
2996 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2997 --pos;
2998 return pos ? NULL : sk;
2999}
3000
3001void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3002{
3003 struct udp_iter_state *state = seq->private;
3004 state->bucket = MAX_UDP_PORTS;
3005
3006 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3007}
3008EXPORT_SYMBOL(udp_seq_start);
3009
3010void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3011{
3012 struct sock *sk;
3013
3014 if (v == SEQ_START_TOKEN)
3015 sk = udp_get_idx(seq, 0);
3016 else
3017 sk = udp_get_next(seq, v);
3018
3019 ++*pos;
3020 return sk;
3021}
3022EXPORT_SYMBOL(udp_seq_next);
3023
3024void udp_seq_stop(struct seq_file *seq, void *v)
3025{
3026 struct udp_seq_afinfo *afinfo;
3027 struct udp_iter_state *state = seq->private;
3028
3029 if (state->bpf_seq_afinfo)
3030 afinfo = state->bpf_seq_afinfo;
3031 else
3032 afinfo = PDE_DATA(file_inode(seq->file));
3033
3034 if (state->bucket <= afinfo->udp_table->mask)
3035 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
3036}
3037EXPORT_SYMBOL(udp_seq_stop);
3038
3039/* ------------------------------------------------------------------------ */
3040static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3041 int bucket)
3042{
3043 struct inet_sock *inet = inet_sk(sp);
3044 __be32 dest = inet->inet_daddr;
3045 __be32 src = inet->inet_rcv_saddr;
3046 __u16 destp = ntohs(inet->inet_dport);
3047 __u16 srcp = ntohs(inet->inet_sport);
3048
3049 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3050 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3051 bucket, src, srcp, dest, destp, sp->sk_state,
3052 sk_wmem_alloc_get(sp),
3053 udp_rqueue_get(sp),
3054 0, 0L, 0,
3055 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3056 0, sock_i_ino(sp),
3057 refcount_read(&sp->sk_refcnt), sp,
3058 atomic_read(&sp->sk_drops));
3059}
3060
3061int udp4_seq_show(struct seq_file *seq, void *v)
3062{
3063 seq_setwidth(seq, 127);
3064 if (v == SEQ_START_TOKEN)
3065 seq_puts(seq, " sl local_address rem_address st tx_queue "
3066 "rx_queue tr tm->when retrnsmt uid timeout "
3067 "inode ref pointer drops");
3068 else {
3069 struct udp_iter_state *state = seq->private;
3070
3071 udp4_format_sock(v, seq, state->bucket);
3072 }
3073 seq_pad(seq, '\n');
3074 return 0;
3075}
3076
3077#ifdef CONFIG_BPF_SYSCALL
3078struct bpf_iter__udp {
3079 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3080 __bpf_md_ptr(struct udp_sock *, udp_sk);
3081 uid_t uid __aligned(8);
3082 int bucket __aligned(8);
3083};
3084
3085static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3086 struct udp_sock *udp_sk, uid_t uid, int bucket)
3087{
3088 struct bpf_iter__udp ctx;
3089
3090 meta->seq_num--; /* skip SEQ_START_TOKEN */
3091 ctx.meta = meta;
3092 ctx.udp_sk = udp_sk;
3093 ctx.uid = uid;
3094 ctx.bucket = bucket;
3095 return bpf_iter_run_prog(prog, &ctx);
3096}
3097
3098static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3099{
3100 struct udp_iter_state *state = seq->private;
3101 struct bpf_iter_meta meta;
3102 struct bpf_prog *prog;
3103 struct sock *sk = v;
3104 uid_t uid;
3105
3106 if (v == SEQ_START_TOKEN)
3107 return 0;
3108
3109 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3110 meta.seq = seq;
3111 prog = bpf_iter_get_info(&meta, false);
3112 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3113}
3114
3115static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3116{
3117 struct bpf_iter_meta meta;
3118 struct bpf_prog *prog;
3119
3120 if (!v) {
3121 meta.seq = seq;
3122 prog = bpf_iter_get_info(&meta, true);
3123 if (prog)
3124 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3125 }
3126
3127 udp_seq_stop(seq, v);
3128}
3129
3130static const struct seq_operations bpf_iter_udp_seq_ops = {
3131 .start = udp_seq_start,
3132 .next = udp_seq_next,
3133 .stop = bpf_iter_udp_seq_stop,
3134 .show = bpf_iter_udp_seq_show,
3135};
3136#endif
3137
3138const struct seq_operations udp_seq_ops = {
3139 .start = udp_seq_start,
3140 .next = udp_seq_next,
3141 .stop = udp_seq_stop,
3142 .show = udp4_seq_show,
3143};
3144EXPORT_SYMBOL(udp_seq_ops);
3145
3146static struct udp_seq_afinfo udp4_seq_afinfo = {
3147 .family = AF_INET,
3148 .udp_table = &udp_table,
3149};
3150
3151static int __net_init udp4_proc_init_net(struct net *net)
3152{
3153 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3154 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3155 return -ENOMEM;
3156 return 0;
3157}
3158
3159static void __net_exit udp4_proc_exit_net(struct net *net)
3160{
3161 remove_proc_entry("udp", net->proc_net);
3162}
3163
3164static struct pernet_operations udp4_net_ops = {
3165 .init = udp4_proc_init_net,
3166 .exit = udp4_proc_exit_net,
3167};
3168
3169int __init udp4_proc_init(void)
3170{
3171 return register_pernet_subsys(&udp4_net_ops);
3172}
3173
3174void udp4_proc_exit(void)
3175{
3176 unregister_pernet_subsys(&udp4_net_ops);
3177}
3178#endif /* CONFIG_PROC_FS */
3179
3180static __initdata unsigned long uhash_entries;
3181static int __init set_uhash_entries(char *str)
3182{
3183 ssize_t ret;
3184
3185 if (!str)
3186 return 0;
3187
3188 ret = kstrtoul(str, 0, &uhash_entries);
3189 if (ret)
3190 return 0;
3191
3192 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3193 uhash_entries = UDP_HTABLE_SIZE_MIN;
3194 return 1;
3195}
3196__setup("uhash_entries=", set_uhash_entries);
3197
3198void __init udp_table_init(struct udp_table *table, const char *name)
3199{
3200 unsigned int i;
3201
3202 table->hash = alloc_large_system_hash(name,
3203 2 * sizeof(struct udp_hslot),
3204 uhash_entries,
3205 21, /* one slot per 2 MB */
3206 0,
3207 &table->log,
3208 &table->mask,
3209 UDP_HTABLE_SIZE_MIN,
3210 64 * 1024);
3211
3212 table->hash2 = table->hash + (table->mask + 1);
3213 for (i = 0; i <= table->mask; i++) {
3214 INIT_HLIST_HEAD(&table->hash[i].head);
3215 table->hash[i].count = 0;
3216 spin_lock_init(&table->hash[i].lock);
3217 }
3218 for (i = 0; i <= table->mask; i++) {
3219 INIT_HLIST_HEAD(&table->hash2[i].head);
3220 table->hash2[i].count = 0;
3221 spin_lock_init(&table->hash2[i].lock);
3222 }
3223}
3224
3225u32 udp_flow_hashrnd(void)
3226{
3227 static u32 hashrnd __read_mostly;
3228
3229 net_get_random_once(&hashrnd, sizeof(hashrnd));
3230
3231 return hashrnd;
3232}
3233EXPORT_SYMBOL(udp_flow_hashrnd);
3234
3235static void __udp_sysctl_init(struct net *net)
3236{
3237 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3238 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3239
3240#ifdef CONFIG_NET_L3_MASTER_DEV
3241 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3242#endif
3243}
3244
3245static int __net_init udp_sysctl_init(struct net *net)
3246{
3247 __udp_sysctl_init(net);
3248 return 0;
3249}
3250
3251static struct pernet_operations __net_initdata udp_sysctl_ops = {
3252 .init = udp_sysctl_init,
3253};
3254
3255#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3256DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3257 struct udp_sock *udp_sk, uid_t uid, int bucket)
3258
3259static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3260{
3261 struct udp_iter_state *st = priv_data;
3262 struct udp_seq_afinfo *afinfo;
3263 int ret;
3264
3265 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3266 if (!afinfo)
3267 return -ENOMEM;
3268
3269 afinfo->family = AF_UNSPEC;
3270 afinfo->udp_table = &udp_table;
3271 st->bpf_seq_afinfo = afinfo;
3272 ret = bpf_iter_init_seq_net(priv_data, aux);
3273 if (ret)
3274 kfree(afinfo);
3275 return ret;
3276}
3277
3278static void bpf_iter_fini_udp(void *priv_data)
3279{
3280 struct udp_iter_state *st = priv_data;
3281
3282 kfree(st->bpf_seq_afinfo);
3283 bpf_iter_fini_seq_net(priv_data);
3284}
3285
3286static const struct bpf_iter_seq_info udp_seq_info = {
3287 .seq_ops = &bpf_iter_udp_seq_ops,
3288 .init_seq_private = bpf_iter_init_udp,
3289 .fini_seq_private = bpf_iter_fini_udp,
3290 .seq_priv_size = sizeof(struct udp_iter_state),
3291};
3292
3293static struct bpf_iter_reg udp_reg_info = {
3294 .target = "udp",
3295 .ctx_arg_info_size = 1,
3296 .ctx_arg_info = {
3297 { offsetof(struct bpf_iter__udp, udp_sk),
3298 PTR_TO_BTF_ID_OR_NULL },
3299 },
3300 .seq_info = &udp_seq_info,
3301};
3302
3303static void __init bpf_iter_register(void)
3304{
3305 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3306 if (bpf_iter_reg_target(&udp_reg_info))
3307 pr_warn("Warning: could not register bpf iterator udp\n");
3308}
3309#endif
3310
3311void __init udp_init(void)
3312{
3313 unsigned long limit;
3314 unsigned int i;
3315
3316 udp_table_init(&udp_table, "UDP");
3317 limit = nr_free_buffer_pages() / 8;
3318 limit = max(limit, 128UL);
3319 sysctl_udp_mem[0] = limit / 4 * 3;
3320 sysctl_udp_mem[1] = limit;
3321 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3322
3323 __udp_sysctl_init(&init_net);
3324
3325 /* 16 spinlocks per cpu */
3326 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3327 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3328 GFP_KERNEL);
3329 if (!udp_busylocks)
3330 panic("UDP: failed to alloc udp_busylocks\n");
3331 for (i = 0; i < (1U << udp_busylocks_log); i++)
3332 spin_lock_init(udp_busylocks + i);
3333
3334 if (register_pernet_subsys(&udp_sysctl_ops))
3335 panic("UDP: failed to init sysctl parameters.\n");
3336
3337#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3338 bpf_iter_register();
3339#endif
3340}
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/bpf-cgroup.h>
78#include <linux/uaccess.h>
79#include <asm/ioctls.h>
80#include <linux/memblock.h>
81#include <linux/highmem.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 <linux/btf_ids.h>
110#include <trace/events/skb.h>
111#include <net/busy_poll.h>
112#include "udp_impl.h"
113#include <net/sock_reuseport.h>
114#include <net/addrconf.h>
115#include <net/udp_tunnel.h>
116#if IS_ENABLED(CONFIG_IPV6)
117#include <net/ipv6_stubs.h>
118#endif
119
120struct udp_table udp_table __read_mostly;
121EXPORT_SYMBOL(udp_table);
122
123long sysctl_udp_mem[3] __read_mostly;
124EXPORT_SYMBOL(sysctl_udp_mem);
125
126atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
127EXPORT_SYMBOL(udp_memory_allocated);
128DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
129EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
130
131#define MAX_UDP_PORTS 65536
132#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
133
134static struct udp_table *udp_get_table_prot(struct sock *sk)
135{
136 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
137}
138
139static int udp_lib_lport_inuse(struct net *net, __u16 num,
140 const struct udp_hslot *hslot,
141 unsigned long *bitmap,
142 struct sock *sk, unsigned int log)
143{
144 struct sock *sk2;
145 kuid_t uid = sock_i_uid(sk);
146
147 sk_for_each(sk2, &hslot->head) {
148 if (net_eq(sock_net(sk2), net) &&
149 sk2 != sk &&
150 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
151 (!sk2->sk_reuse || !sk->sk_reuse) &&
152 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
153 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
154 inet_rcv_saddr_equal(sk, sk2, true)) {
155 if (sk2->sk_reuseport && sk->sk_reuseport &&
156 !rcu_access_pointer(sk->sk_reuseport_cb) &&
157 uid_eq(uid, sock_i_uid(sk2))) {
158 if (!bitmap)
159 return 0;
160 } else {
161 if (!bitmap)
162 return 1;
163 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
164 bitmap);
165 }
166 }
167 }
168 return 0;
169}
170
171/*
172 * Note: we still hold spinlock of primary hash chain, so no other writer
173 * can insert/delete a socket with local_port == num
174 */
175static int udp_lib_lport_inuse2(struct net *net, __u16 num,
176 struct udp_hslot *hslot2,
177 struct sock *sk)
178{
179 struct sock *sk2;
180 kuid_t uid = sock_i_uid(sk);
181 int res = 0;
182
183 spin_lock(&hslot2->lock);
184 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
185 if (net_eq(sock_net(sk2), net) &&
186 sk2 != sk &&
187 (udp_sk(sk2)->udp_port_hash == num) &&
188 (!sk2->sk_reuse || !sk->sk_reuse) &&
189 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
190 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
191 inet_rcv_saddr_equal(sk, sk2, true)) {
192 if (sk2->sk_reuseport && sk->sk_reuseport &&
193 !rcu_access_pointer(sk->sk_reuseport_cb) &&
194 uid_eq(uid, sock_i_uid(sk2))) {
195 res = 0;
196 } else {
197 res = 1;
198 }
199 break;
200 }
201 }
202 spin_unlock(&hslot2->lock);
203 return res;
204}
205
206static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
207{
208 struct net *net = sock_net(sk);
209 kuid_t uid = sock_i_uid(sk);
210 struct sock *sk2;
211
212 sk_for_each(sk2, &hslot->head) {
213 if (net_eq(sock_net(sk2), net) &&
214 sk2 != sk &&
215 sk2->sk_family == sk->sk_family &&
216 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
217 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
218 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
219 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
220 inet_rcv_saddr_equal(sk, sk2, false)) {
221 return reuseport_add_sock(sk, sk2,
222 inet_rcv_saddr_any(sk));
223 }
224 }
225
226 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
227}
228
229/**
230 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
231 *
232 * @sk: socket struct in question
233 * @snum: port number to look up
234 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
235 * with NULL address
236 */
237int udp_lib_get_port(struct sock *sk, unsigned short snum,
238 unsigned int hash2_nulladdr)
239{
240 struct udp_table *udptable = udp_get_table_prot(sk);
241 struct udp_hslot *hslot, *hslot2;
242 struct net *net = sock_net(sk);
243 int error = -EADDRINUSE;
244
245 if (!snum) {
246 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
247 unsigned short first, last;
248 int low, high, remaining;
249 unsigned int rand;
250
251 inet_get_local_port_range(net, &low, &high);
252 remaining = (high - low) + 1;
253
254 rand = get_random_u32();
255 first = reciprocal_scale(rand, remaining) + low;
256 /*
257 * force rand to be an odd multiple of UDP_HTABLE_SIZE
258 */
259 rand = (rand | 1) * (udptable->mask + 1);
260 last = first + udptable->mask + 1;
261 do {
262 hslot = udp_hashslot(udptable, net, first);
263 bitmap_zero(bitmap, PORTS_PER_CHAIN);
264 spin_lock_bh(&hslot->lock);
265 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
266 udptable->log);
267
268 snum = first;
269 /*
270 * Iterate on all possible values of snum for this hash.
271 * Using steps of an odd multiple of UDP_HTABLE_SIZE
272 * give us randomization and full range coverage.
273 */
274 do {
275 if (low <= snum && snum <= high &&
276 !test_bit(snum >> udptable->log, bitmap) &&
277 !inet_is_local_reserved_port(net, snum))
278 goto found;
279 snum += rand;
280 } while (snum != first);
281 spin_unlock_bh(&hslot->lock);
282 cond_resched();
283 } while (++first != last);
284 goto fail;
285 } else {
286 hslot = udp_hashslot(udptable, net, snum);
287 spin_lock_bh(&hslot->lock);
288 if (hslot->count > 10) {
289 int exist;
290 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
291
292 slot2 &= udptable->mask;
293 hash2_nulladdr &= udptable->mask;
294
295 hslot2 = udp_hashslot2(udptable, slot2);
296 if (hslot->count < hslot2->count)
297 goto scan_primary_hash;
298
299 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
300 if (!exist && (hash2_nulladdr != slot2)) {
301 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
302 exist = udp_lib_lport_inuse2(net, snum, hslot2,
303 sk);
304 }
305 if (exist)
306 goto fail_unlock;
307 else
308 goto found;
309 }
310scan_primary_hash:
311 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
312 goto fail_unlock;
313 }
314found:
315 inet_sk(sk)->inet_num = snum;
316 udp_sk(sk)->udp_port_hash = snum;
317 udp_sk(sk)->udp_portaddr_hash ^= snum;
318 if (sk_unhashed(sk)) {
319 if (sk->sk_reuseport &&
320 udp_reuseport_add_sock(sk, hslot)) {
321 inet_sk(sk)->inet_num = 0;
322 udp_sk(sk)->udp_port_hash = 0;
323 udp_sk(sk)->udp_portaddr_hash ^= snum;
324 goto fail_unlock;
325 }
326
327 sk_add_node_rcu(sk, &hslot->head);
328 hslot->count++;
329 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
330
331 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
332 spin_lock(&hslot2->lock);
333 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
334 sk->sk_family == AF_INET6)
335 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
336 &hslot2->head);
337 else
338 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
339 &hslot2->head);
340 hslot2->count++;
341 spin_unlock(&hslot2->lock);
342 }
343 sock_set_flag(sk, SOCK_RCU_FREE);
344 error = 0;
345fail_unlock:
346 spin_unlock_bh(&hslot->lock);
347fail:
348 return error;
349}
350EXPORT_SYMBOL(udp_lib_get_port);
351
352int udp_v4_get_port(struct sock *sk, unsigned short snum)
353{
354 unsigned int hash2_nulladdr =
355 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
356 unsigned int hash2_partial =
357 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
358
359 /* precompute partial secondary hash */
360 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
361 return udp_lib_get_port(sk, snum, hash2_nulladdr);
362}
363
364static int compute_score(struct sock *sk, struct net *net,
365 __be32 saddr, __be16 sport,
366 __be32 daddr, unsigned short hnum,
367 int dif, int sdif)
368{
369 int score;
370 struct inet_sock *inet;
371 bool dev_match;
372
373 if (!net_eq(sock_net(sk), net) ||
374 udp_sk(sk)->udp_port_hash != hnum ||
375 ipv6_only_sock(sk))
376 return -1;
377
378 if (sk->sk_rcv_saddr != daddr)
379 return -1;
380
381 score = (sk->sk_family == PF_INET) ? 2 : 1;
382
383 inet = inet_sk(sk);
384 if (inet->inet_daddr) {
385 if (inet->inet_daddr != saddr)
386 return -1;
387 score += 4;
388 }
389
390 if (inet->inet_dport) {
391 if (inet->inet_dport != sport)
392 return -1;
393 score += 4;
394 }
395
396 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
397 dif, sdif);
398 if (!dev_match)
399 return -1;
400 if (sk->sk_bound_dev_if)
401 score += 4;
402
403 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
404 score++;
405 return score;
406}
407
408static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
409 const __u16 lport, const __be32 faddr,
410 const __be16 fport)
411{
412 static u32 udp_ehash_secret __read_mostly;
413
414 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
415
416 return __inet_ehashfn(laddr, lport, faddr, fport,
417 udp_ehash_secret + net_hash_mix(net));
418}
419
420static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
421 struct sk_buff *skb,
422 __be32 saddr, __be16 sport,
423 __be32 daddr, unsigned short hnum)
424{
425 struct sock *reuse_sk = NULL;
426 u32 hash;
427
428 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
429 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
430 reuse_sk = reuseport_select_sock(sk, hash, skb,
431 sizeof(struct udphdr));
432 }
433 return reuse_sk;
434}
435
436/* called with rcu_read_lock() */
437static struct sock *udp4_lib_lookup2(struct net *net,
438 __be32 saddr, __be16 sport,
439 __be32 daddr, unsigned int hnum,
440 int dif, int sdif,
441 struct udp_hslot *hslot2,
442 struct sk_buff *skb)
443{
444 struct sock *sk, *result;
445 int score, badness;
446
447 result = NULL;
448 badness = 0;
449 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
450 score = compute_score(sk, net, saddr, sport,
451 daddr, hnum, dif, sdif);
452 if (score > badness) {
453 result = lookup_reuseport(net, sk, skb,
454 saddr, sport, daddr, hnum);
455 /* Fall back to scoring if group has connections */
456 if (result && !reuseport_has_conns(sk))
457 return result;
458
459 result = result ? : sk;
460 badness = score;
461 }
462 }
463 return result;
464}
465
466static struct sock *udp4_lookup_run_bpf(struct net *net,
467 struct udp_table *udptable,
468 struct sk_buff *skb,
469 __be32 saddr, __be16 sport,
470 __be32 daddr, u16 hnum, const int dif)
471{
472 struct sock *sk, *reuse_sk;
473 bool no_reuseport;
474
475 if (udptable != net->ipv4.udp_table)
476 return NULL; /* only UDP is supported */
477
478 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport,
479 daddr, hnum, dif, &sk);
480 if (no_reuseport || IS_ERR_OR_NULL(sk))
481 return sk;
482
483 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
484 if (reuse_sk)
485 sk = reuse_sk;
486 return sk;
487}
488
489/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
490 * harder than this. -DaveM
491 */
492struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
493 __be16 sport, __be32 daddr, __be16 dport, int dif,
494 int sdif, struct udp_table *udptable, struct sk_buff *skb)
495{
496 unsigned short hnum = ntohs(dport);
497 unsigned int hash2, slot2;
498 struct udp_hslot *hslot2;
499 struct sock *result, *sk;
500
501 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
502 slot2 = hash2 & udptable->mask;
503 hslot2 = &udptable->hash2[slot2];
504
505 /* Lookup connected or non-wildcard socket */
506 result = udp4_lib_lookup2(net, saddr, sport,
507 daddr, hnum, dif, sdif,
508 hslot2, skb);
509 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
510 goto done;
511
512 /* Lookup redirect from BPF */
513 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
514 sk = udp4_lookup_run_bpf(net, udptable, skb,
515 saddr, sport, daddr, hnum, dif);
516 if (sk) {
517 result = sk;
518 goto done;
519 }
520 }
521
522 /* Got non-wildcard socket or error on first lookup */
523 if (result)
524 goto done;
525
526 /* Lookup wildcard sockets */
527 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
528 slot2 = hash2 & udptable->mask;
529 hslot2 = &udptable->hash2[slot2];
530
531 result = udp4_lib_lookup2(net, saddr, sport,
532 htonl(INADDR_ANY), hnum, dif, sdif,
533 hslot2, skb);
534done:
535 if (IS_ERR(result))
536 return NULL;
537 return result;
538}
539EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
540
541static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
542 __be16 sport, __be16 dport,
543 struct udp_table *udptable)
544{
545 const struct iphdr *iph = ip_hdr(skb);
546
547 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
548 iph->daddr, dport, inet_iif(skb),
549 inet_sdif(skb), udptable, skb);
550}
551
552struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
553 __be16 sport, __be16 dport)
554{
555 const struct iphdr *iph = ip_hdr(skb);
556 struct net *net = dev_net(skb->dev);
557
558 return __udp4_lib_lookup(net, iph->saddr, sport,
559 iph->daddr, dport, inet_iif(skb),
560 inet_sdif(skb), net->ipv4.udp_table, NULL);
561}
562
563/* Must be called under rcu_read_lock().
564 * Does increment socket refcount.
565 */
566#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
567struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
568 __be32 daddr, __be16 dport, int dif)
569{
570 struct sock *sk;
571
572 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
573 dif, 0, net->ipv4.udp_table, NULL);
574 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
575 sk = NULL;
576 return sk;
577}
578EXPORT_SYMBOL_GPL(udp4_lib_lookup);
579#endif
580
581static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
582 __be16 loc_port, __be32 loc_addr,
583 __be16 rmt_port, __be32 rmt_addr,
584 int dif, int sdif, unsigned short hnum)
585{
586 struct inet_sock *inet = inet_sk(sk);
587
588 if (!net_eq(sock_net(sk), net) ||
589 udp_sk(sk)->udp_port_hash != hnum ||
590 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
591 (inet->inet_dport != rmt_port && inet->inet_dport) ||
592 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
593 ipv6_only_sock(sk) ||
594 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
595 return false;
596 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
597 return false;
598 return true;
599}
600
601DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
602void udp_encap_enable(void)
603{
604 static_branch_inc(&udp_encap_needed_key);
605}
606EXPORT_SYMBOL(udp_encap_enable);
607
608void udp_encap_disable(void)
609{
610 static_branch_dec(&udp_encap_needed_key);
611}
612EXPORT_SYMBOL(udp_encap_disable);
613
614/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
615 * through error handlers in encapsulations looking for a match.
616 */
617static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
618{
619 int i;
620
621 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
622 int (*handler)(struct sk_buff *skb, u32 info);
623 const struct ip_tunnel_encap_ops *encap;
624
625 encap = rcu_dereference(iptun_encaps[i]);
626 if (!encap)
627 continue;
628 handler = encap->err_handler;
629 if (handler && !handler(skb, info))
630 return 0;
631 }
632
633 return -ENOENT;
634}
635
636/* Try to match ICMP errors to UDP tunnels by looking up a socket without
637 * reversing source and destination port: this will match tunnels that force the
638 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
639 * lwtunnels might actually break this assumption by being configured with
640 * different destination ports on endpoints, in this case we won't be able to
641 * trace ICMP messages back to them.
642 *
643 * If this doesn't match any socket, probe tunnels with arbitrary destination
644 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
645 * we've sent packets to won't necessarily match the local destination port.
646 *
647 * Then ask the tunnel implementation to match the error against a valid
648 * association.
649 *
650 * Return an error if we can't find a match, the socket if we need further
651 * processing, zero otherwise.
652 */
653static struct sock *__udp4_lib_err_encap(struct net *net,
654 const struct iphdr *iph,
655 struct udphdr *uh,
656 struct udp_table *udptable,
657 struct sock *sk,
658 struct sk_buff *skb, u32 info)
659{
660 int (*lookup)(struct sock *sk, struct sk_buff *skb);
661 int network_offset, transport_offset;
662 struct udp_sock *up;
663
664 network_offset = skb_network_offset(skb);
665 transport_offset = skb_transport_offset(skb);
666
667 /* Network header needs to point to the outer IPv4 header inside ICMP */
668 skb_reset_network_header(skb);
669
670 /* Transport header needs to point to the UDP header */
671 skb_set_transport_header(skb, iph->ihl << 2);
672
673 if (sk) {
674 up = udp_sk(sk);
675
676 lookup = READ_ONCE(up->encap_err_lookup);
677 if (lookup && lookup(sk, skb))
678 sk = NULL;
679
680 goto out;
681 }
682
683 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
684 iph->saddr, uh->dest, skb->dev->ifindex, 0,
685 udptable, NULL);
686 if (sk) {
687 up = udp_sk(sk);
688
689 lookup = READ_ONCE(up->encap_err_lookup);
690 if (!lookup || lookup(sk, skb))
691 sk = NULL;
692 }
693
694out:
695 if (!sk)
696 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
697
698 skb_set_transport_header(skb, transport_offset);
699 skb_set_network_header(skb, network_offset);
700
701 return sk;
702}
703
704/*
705 * This routine is called by the ICMP module when it gets some
706 * sort of error condition. If err < 0 then the socket should
707 * be closed and the error returned to the user. If err > 0
708 * it's just the icmp type << 8 | icmp code.
709 * Header points to the ip header of the error packet. We move
710 * on past this. Then (as it used to claim before adjustment)
711 * header points to the first 8 bytes of the udp header. We need
712 * to find the appropriate port.
713 */
714
715int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
716{
717 struct inet_sock *inet;
718 const struct iphdr *iph = (const struct iphdr *)skb->data;
719 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
720 const int type = icmp_hdr(skb)->type;
721 const int code = icmp_hdr(skb)->code;
722 bool tunnel = false;
723 struct sock *sk;
724 int harderr;
725 int err;
726 struct net *net = dev_net(skb->dev);
727
728 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
729 iph->saddr, uh->source, skb->dev->ifindex,
730 inet_sdif(skb), udptable, NULL);
731
732 if (!sk || udp_sk(sk)->encap_type) {
733 /* No socket for error: try tunnels before discarding */
734 if (static_branch_unlikely(&udp_encap_needed_key)) {
735 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
736 info);
737 if (!sk)
738 return 0;
739 } else
740 sk = ERR_PTR(-ENOENT);
741
742 if (IS_ERR(sk)) {
743 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
744 return PTR_ERR(sk);
745 }
746
747 tunnel = true;
748 }
749
750 err = 0;
751 harderr = 0;
752 inet = inet_sk(sk);
753
754 switch (type) {
755 default:
756 case ICMP_TIME_EXCEEDED:
757 err = EHOSTUNREACH;
758 break;
759 case ICMP_SOURCE_QUENCH:
760 goto out;
761 case ICMP_PARAMETERPROB:
762 err = EPROTO;
763 harderr = 1;
764 break;
765 case ICMP_DEST_UNREACH:
766 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
767 ipv4_sk_update_pmtu(skb, sk, info);
768 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
769 err = EMSGSIZE;
770 harderr = 1;
771 break;
772 }
773 goto out;
774 }
775 err = EHOSTUNREACH;
776 if (code <= NR_ICMP_UNREACH) {
777 harderr = icmp_err_convert[code].fatal;
778 err = icmp_err_convert[code].errno;
779 }
780 break;
781 case ICMP_REDIRECT:
782 ipv4_sk_redirect(skb, sk);
783 goto out;
784 }
785
786 /*
787 * RFC1122: OK. Passes ICMP errors back to application, as per
788 * 4.1.3.3.
789 */
790 if (tunnel) {
791 /* ...not for tunnels though: we don't have a sending socket */
792 if (udp_sk(sk)->encap_err_rcv)
793 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
794 (u8 *)(uh+1));
795 goto out;
796 }
797 if (!inet->recverr) {
798 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
799 goto out;
800 } else
801 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
802
803 sk->sk_err = err;
804 sk_error_report(sk);
805out:
806 return 0;
807}
808
809int udp_err(struct sk_buff *skb, u32 info)
810{
811 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
812}
813
814/*
815 * Throw away all pending data and cancel the corking. Socket is locked.
816 */
817void udp_flush_pending_frames(struct sock *sk)
818{
819 struct udp_sock *up = udp_sk(sk);
820
821 if (up->pending) {
822 up->len = 0;
823 up->pending = 0;
824 ip_flush_pending_frames(sk);
825 }
826}
827EXPORT_SYMBOL(udp_flush_pending_frames);
828
829/**
830 * udp4_hwcsum - handle outgoing HW checksumming
831 * @skb: sk_buff containing the filled-in UDP header
832 * (checksum field must be zeroed out)
833 * @src: source IP address
834 * @dst: destination IP address
835 */
836void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
837{
838 struct udphdr *uh = udp_hdr(skb);
839 int offset = skb_transport_offset(skb);
840 int len = skb->len - offset;
841 int hlen = len;
842 __wsum csum = 0;
843
844 if (!skb_has_frag_list(skb)) {
845 /*
846 * Only one fragment on the socket.
847 */
848 skb->csum_start = skb_transport_header(skb) - skb->head;
849 skb->csum_offset = offsetof(struct udphdr, check);
850 uh->check = ~csum_tcpudp_magic(src, dst, len,
851 IPPROTO_UDP, 0);
852 } else {
853 struct sk_buff *frags;
854
855 /*
856 * HW-checksum won't work as there are two or more
857 * fragments on the socket so that all csums of sk_buffs
858 * should be together
859 */
860 skb_walk_frags(skb, frags) {
861 csum = csum_add(csum, frags->csum);
862 hlen -= frags->len;
863 }
864
865 csum = skb_checksum(skb, offset, hlen, csum);
866 skb->ip_summed = CHECKSUM_NONE;
867
868 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
869 if (uh->check == 0)
870 uh->check = CSUM_MANGLED_0;
871 }
872}
873EXPORT_SYMBOL_GPL(udp4_hwcsum);
874
875/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
876 * for the simple case like when setting the checksum for a UDP tunnel.
877 */
878void udp_set_csum(bool nocheck, struct sk_buff *skb,
879 __be32 saddr, __be32 daddr, int len)
880{
881 struct udphdr *uh = udp_hdr(skb);
882
883 if (nocheck) {
884 uh->check = 0;
885 } else if (skb_is_gso(skb)) {
886 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
887 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
888 uh->check = 0;
889 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
890 if (uh->check == 0)
891 uh->check = CSUM_MANGLED_0;
892 } else {
893 skb->ip_summed = CHECKSUM_PARTIAL;
894 skb->csum_start = skb_transport_header(skb) - skb->head;
895 skb->csum_offset = offsetof(struct udphdr, check);
896 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
897 }
898}
899EXPORT_SYMBOL(udp_set_csum);
900
901static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
902 struct inet_cork *cork)
903{
904 struct sock *sk = skb->sk;
905 struct inet_sock *inet = inet_sk(sk);
906 struct udphdr *uh;
907 int err;
908 int is_udplite = IS_UDPLITE(sk);
909 int offset = skb_transport_offset(skb);
910 int len = skb->len - offset;
911 int datalen = len - sizeof(*uh);
912 __wsum csum = 0;
913
914 /*
915 * Create a UDP header
916 */
917 uh = udp_hdr(skb);
918 uh->source = inet->inet_sport;
919 uh->dest = fl4->fl4_dport;
920 uh->len = htons(len);
921 uh->check = 0;
922
923 if (cork->gso_size) {
924 const int hlen = skb_network_header_len(skb) +
925 sizeof(struct udphdr);
926
927 if (hlen + cork->gso_size > cork->fragsize) {
928 kfree_skb(skb);
929 return -EINVAL;
930 }
931 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
932 kfree_skb(skb);
933 return -EINVAL;
934 }
935 if (sk->sk_no_check_tx) {
936 kfree_skb(skb);
937 return -EINVAL;
938 }
939 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
940 dst_xfrm(skb_dst(skb))) {
941 kfree_skb(skb);
942 return -EIO;
943 }
944
945 if (datalen > cork->gso_size) {
946 skb_shinfo(skb)->gso_size = cork->gso_size;
947 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
948 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
949 cork->gso_size);
950 }
951 goto csum_partial;
952 }
953
954 if (is_udplite) /* UDP-Lite */
955 csum = udplite_csum(skb);
956
957 else if (sk->sk_no_check_tx) { /* UDP csum off */
958
959 skb->ip_summed = CHECKSUM_NONE;
960 goto send;
961
962 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
963csum_partial:
964
965 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
966 goto send;
967
968 } else
969 csum = udp_csum(skb);
970
971 /* add protocol-dependent pseudo-header */
972 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
973 sk->sk_protocol, csum);
974 if (uh->check == 0)
975 uh->check = CSUM_MANGLED_0;
976
977send:
978 err = ip_send_skb(sock_net(sk), skb);
979 if (err) {
980 if (err == -ENOBUFS && !inet->recverr) {
981 UDP_INC_STATS(sock_net(sk),
982 UDP_MIB_SNDBUFERRORS, is_udplite);
983 err = 0;
984 }
985 } else
986 UDP_INC_STATS(sock_net(sk),
987 UDP_MIB_OUTDATAGRAMS, is_udplite);
988 return err;
989}
990
991/*
992 * Push out all pending data as one UDP datagram. Socket is locked.
993 */
994int udp_push_pending_frames(struct sock *sk)
995{
996 struct udp_sock *up = udp_sk(sk);
997 struct inet_sock *inet = inet_sk(sk);
998 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
999 struct sk_buff *skb;
1000 int err = 0;
1001
1002 skb = ip_finish_skb(sk, fl4);
1003 if (!skb)
1004 goto out;
1005
1006 err = udp_send_skb(skb, fl4, &inet->cork.base);
1007
1008out:
1009 up->len = 0;
1010 up->pending = 0;
1011 return err;
1012}
1013EXPORT_SYMBOL(udp_push_pending_frames);
1014
1015static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1016{
1017 switch (cmsg->cmsg_type) {
1018 case UDP_SEGMENT:
1019 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1020 return -EINVAL;
1021 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1022 return 0;
1023 default:
1024 return -EINVAL;
1025 }
1026}
1027
1028int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1029{
1030 struct cmsghdr *cmsg;
1031 bool need_ip = false;
1032 int err;
1033
1034 for_each_cmsghdr(cmsg, msg) {
1035 if (!CMSG_OK(msg, cmsg))
1036 return -EINVAL;
1037
1038 if (cmsg->cmsg_level != SOL_UDP) {
1039 need_ip = true;
1040 continue;
1041 }
1042
1043 err = __udp_cmsg_send(cmsg, gso_size);
1044 if (err)
1045 return err;
1046 }
1047
1048 return need_ip;
1049}
1050EXPORT_SYMBOL_GPL(udp_cmsg_send);
1051
1052int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1053{
1054 struct inet_sock *inet = inet_sk(sk);
1055 struct udp_sock *up = udp_sk(sk);
1056 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1057 struct flowi4 fl4_stack;
1058 struct flowi4 *fl4;
1059 int ulen = len;
1060 struct ipcm_cookie ipc;
1061 struct rtable *rt = NULL;
1062 int free = 0;
1063 int connected = 0;
1064 __be32 daddr, faddr, saddr;
1065 __be16 dport;
1066 u8 tos;
1067 int err, is_udplite = IS_UDPLITE(sk);
1068 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1069 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1070 struct sk_buff *skb;
1071 struct ip_options_data opt_copy;
1072
1073 if (len > 0xFFFF)
1074 return -EMSGSIZE;
1075
1076 /*
1077 * Check the flags.
1078 */
1079
1080 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1081 return -EOPNOTSUPP;
1082
1083 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1084
1085 fl4 = &inet->cork.fl.u.ip4;
1086 if (up->pending) {
1087 /*
1088 * There are pending frames.
1089 * The socket lock must be held while it's corked.
1090 */
1091 lock_sock(sk);
1092 if (likely(up->pending)) {
1093 if (unlikely(up->pending != AF_INET)) {
1094 release_sock(sk);
1095 return -EINVAL;
1096 }
1097 goto do_append_data;
1098 }
1099 release_sock(sk);
1100 }
1101 ulen += sizeof(struct udphdr);
1102
1103 /*
1104 * Get and verify the address.
1105 */
1106 if (usin) {
1107 if (msg->msg_namelen < sizeof(*usin))
1108 return -EINVAL;
1109 if (usin->sin_family != AF_INET) {
1110 if (usin->sin_family != AF_UNSPEC)
1111 return -EAFNOSUPPORT;
1112 }
1113
1114 daddr = usin->sin_addr.s_addr;
1115 dport = usin->sin_port;
1116 if (dport == 0)
1117 return -EINVAL;
1118 } else {
1119 if (sk->sk_state != TCP_ESTABLISHED)
1120 return -EDESTADDRREQ;
1121 daddr = inet->inet_daddr;
1122 dport = inet->inet_dport;
1123 /* Open fast path for connected socket.
1124 Route will not be used, if at least one option is set.
1125 */
1126 connected = 1;
1127 }
1128
1129 ipcm_init_sk(&ipc, inet);
1130 ipc.gso_size = READ_ONCE(up->gso_size);
1131
1132 if (msg->msg_controllen) {
1133 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1134 if (err > 0)
1135 err = ip_cmsg_send(sk, msg, &ipc,
1136 sk->sk_family == AF_INET6);
1137 if (unlikely(err < 0)) {
1138 kfree(ipc.opt);
1139 return err;
1140 }
1141 if (ipc.opt)
1142 free = 1;
1143 connected = 0;
1144 }
1145 if (!ipc.opt) {
1146 struct ip_options_rcu *inet_opt;
1147
1148 rcu_read_lock();
1149 inet_opt = rcu_dereference(inet->inet_opt);
1150 if (inet_opt) {
1151 memcpy(&opt_copy, inet_opt,
1152 sizeof(*inet_opt) + inet_opt->opt.optlen);
1153 ipc.opt = &opt_copy.opt;
1154 }
1155 rcu_read_unlock();
1156 }
1157
1158 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1159 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1160 (struct sockaddr *)usin, &ipc.addr);
1161 if (err)
1162 goto out_free;
1163 if (usin) {
1164 if (usin->sin_port == 0) {
1165 /* BPF program set invalid port. Reject it. */
1166 err = -EINVAL;
1167 goto out_free;
1168 }
1169 daddr = usin->sin_addr.s_addr;
1170 dport = usin->sin_port;
1171 }
1172 }
1173
1174 saddr = ipc.addr;
1175 ipc.addr = faddr = daddr;
1176
1177 if (ipc.opt && ipc.opt->opt.srr) {
1178 if (!daddr) {
1179 err = -EINVAL;
1180 goto out_free;
1181 }
1182 faddr = ipc.opt->opt.faddr;
1183 connected = 0;
1184 }
1185 tos = get_rttos(&ipc, inet);
1186 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1187 (msg->msg_flags & MSG_DONTROUTE) ||
1188 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1189 tos |= RTO_ONLINK;
1190 connected = 0;
1191 }
1192
1193 if (ipv4_is_multicast(daddr)) {
1194 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1195 ipc.oif = inet->mc_index;
1196 if (!saddr)
1197 saddr = inet->mc_addr;
1198 connected = 0;
1199 } else if (!ipc.oif) {
1200 ipc.oif = inet->uc_index;
1201 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1202 /* oif is set, packet is to local broadcast and
1203 * uc_index is set. oif is most likely set
1204 * by sk_bound_dev_if. If uc_index != oif check if the
1205 * oif is an L3 master and uc_index is an L3 slave.
1206 * If so, we want to allow the send using the uc_index.
1207 */
1208 if (ipc.oif != inet->uc_index &&
1209 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1210 inet->uc_index)) {
1211 ipc.oif = inet->uc_index;
1212 }
1213 }
1214
1215 if (connected)
1216 rt = (struct rtable *)sk_dst_check(sk, 0);
1217
1218 if (!rt) {
1219 struct net *net = sock_net(sk);
1220 __u8 flow_flags = inet_sk_flowi_flags(sk);
1221
1222 fl4 = &fl4_stack;
1223
1224 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1225 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1226 flow_flags,
1227 faddr, saddr, dport, inet->inet_sport,
1228 sk->sk_uid);
1229
1230 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1231 rt = ip_route_output_flow(net, fl4, sk);
1232 if (IS_ERR(rt)) {
1233 err = PTR_ERR(rt);
1234 rt = NULL;
1235 if (err == -ENETUNREACH)
1236 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1237 goto out;
1238 }
1239
1240 err = -EACCES;
1241 if ((rt->rt_flags & RTCF_BROADCAST) &&
1242 !sock_flag(sk, SOCK_BROADCAST))
1243 goto out;
1244 if (connected)
1245 sk_dst_set(sk, dst_clone(&rt->dst));
1246 }
1247
1248 if (msg->msg_flags&MSG_CONFIRM)
1249 goto do_confirm;
1250back_from_confirm:
1251
1252 saddr = fl4->saddr;
1253 if (!ipc.addr)
1254 daddr = ipc.addr = fl4->daddr;
1255
1256 /* Lockless fast path for the non-corking case. */
1257 if (!corkreq) {
1258 struct inet_cork cork;
1259
1260 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1261 sizeof(struct udphdr), &ipc, &rt,
1262 &cork, msg->msg_flags);
1263 err = PTR_ERR(skb);
1264 if (!IS_ERR_OR_NULL(skb))
1265 err = udp_send_skb(skb, fl4, &cork);
1266 goto out;
1267 }
1268
1269 lock_sock(sk);
1270 if (unlikely(up->pending)) {
1271 /* The socket is already corked while preparing it. */
1272 /* ... which is an evident application bug. --ANK */
1273 release_sock(sk);
1274
1275 net_dbg_ratelimited("socket already corked\n");
1276 err = -EINVAL;
1277 goto out;
1278 }
1279 /*
1280 * Now cork the socket to pend data.
1281 */
1282 fl4 = &inet->cork.fl.u.ip4;
1283 fl4->daddr = daddr;
1284 fl4->saddr = saddr;
1285 fl4->fl4_dport = dport;
1286 fl4->fl4_sport = inet->inet_sport;
1287 up->pending = AF_INET;
1288
1289do_append_data:
1290 up->len += ulen;
1291 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1292 sizeof(struct udphdr), &ipc, &rt,
1293 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1294 if (err)
1295 udp_flush_pending_frames(sk);
1296 else if (!corkreq)
1297 err = udp_push_pending_frames(sk);
1298 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1299 up->pending = 0;
1300 release_sock(sk);
1301
1302out:
1303 ip_rt_put(rt);
1304out_free:
1305 if (free)
1306 kfree(ipc.opt);
1307 if (!err)
1308 return len;
1309 /*
1310 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1311 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1312 * we don't have a good statistic (IpOutDiscards but it can be too many
1313 * things). We could add another new stat but at least for now that
1314 * seems like overkill.
1315 */
1316 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1317 UDP_INC_STATS(sock_net(sk),
1318 UDP_MIB_SNDBUFERRORS, is_udplite);
1319 }
1320 return err;
1321
1322do_confirm:
1323 if (msg->msg_flags & MSG_PROBE)
1324 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1325 if (!(msg->msg_flags&MSG_PROBE) || len)
1326 goto back_from_confirm;
1327 err = 0;
1328 goto out;
1329}
1330EXPORT_SYMBOL(udp_sendmsg);
1331
1332int udp_sendpage(struct sock *sk, struct page *page, int offset,
1333 size_t size, int flags)
1334{
1335 struct inet_sock *inet = inet_sk(sk);
1336 struct udp_sock *up = udp_sk(sk);
1337 int ret;
1338
1339 if (flags & MSG_SENDPAGE_NOTLAST)
1340 flags |= MSG_MORE;
1341
1342 if (!up->pending) {
1343 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1344
1345 /* Call udp_sendmsg to specify destination address which
1346 * sendpage interface can't pass.
1347 * This will succeed only when the socket is connected.
1348 */
1349 ret = udp_sendmsg(sk, &msg, 0);
1350 if (ret < 0)
1351 return ret;
1352 }
1353
1354 lock_sock(sk);
1355
1356 if (unlikely(!up->pending)) {
1357 release_sock(sk);
1358
1359 net_dbg_ratelimited("cork failed\n");
1360 return -EINVAL;
1361 }
1362
1363 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1364 page, offset, size, flags);
1365 if (ret == -EOPNOTSUPP) {
1366 release_sock(sk);
1367 return sock_no_sendpage(sk->sk_socket, page, offset,
1368 size, flags);
1369 }
1370 if (ret < 0) {
1371 udp_flush_pending_frames(sk);
1372 goto out;
1373 }
1374
1375 up->len += size;
1376 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1377 ret = udp_push_pending_frames(sk);
1378 if (!ret)
1379 ret = size;
1380out:
1381 release_sock(sk);
1382 return ret;
1383}
1384
1385#define UDP_SKB_IS_STATELESS 0x80000000
1386
1387/* all head states (dst, sk, nf conntrack) except skb extensions are
1388 * cleared by udp_rcv().
1389 *
1390 * We need to preserve secpath, if present, to eventually process
1391 * IP_CMSG_PASSSEC at recvmsg() time.
1392 *
1393 * Other extensions can be cleared.
1394 */
1395static bool udp_try_make_stateless(struct sk_buff *skb)
1396{
1397 if (!skb_has_extensions(skb))
1398 return true;
1399
1400 if (!secpath_exists(skb)) {
1401 skb_ext_reset(skb);
1402 return true;
1403 }
1404
1405 return false;
1406}
1407
1408static void udp_set_dev_scratch(struct sk_buff *skb)
1409{
1410 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1411
1412 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1413 scratch->_tsize_state = skb->truesize;
1414#if BITS_PER_LONG == 64
1415 scratch->len = skb->len;
1416 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1417 scratch->is_linear = !skb_is_nonlinear(skb);
1418#endif
1419 if (udp_try_make_stateless(skb))
1420 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1421}
1422
1423static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1424{
1425 /* We come here after udp_lib_checksum_complete() returned 0.
1426 * This means that __skb_checksum_complete() might have
1427 * set skb->csum_valid to 1.
1428 * On 64bit platforms, we can set csum_unnecessary
1429 * to true, but only if the skb is not shared.
1430 */
1431#if BITS_PER_LONG == 64
1432 if (!skb_shared(skb))
1433 udp_skb_scratch(skb)->csum_unnecessary = true;
1434#endif
1435}
1436
1437static int udp_skb_truesize(struct sk_buff *skb)
1438{
1439 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1440}
1441
1442static bool udp_skb_has_head_state(struct sk_buff *skb)
1443{
1444 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1445}
1446
1447/* fully reclaim rmem/fwd memory allocated for skb */
1448static void udp_rmem_release(struct sock *sk, int size, int partial,
1449 bool rx_queue_lock_held)
1450{
1451 struct udp_sock *up = udp_sk(sk);
1452 struct sk_buff_head *sk_queue;
1453 int amt;
1454
1455 if (likely(partial)) {
1456 up->forward_deficit += size;
1457 size = up->forward_deficit;
1458 if (size < READ_ONCE(up->forward_threshold) &&
1459 !skb_queue_empty(&up->reader_queue))
1460 return;
1461 } else {
1462 size += up->forward_deficit;
1463 }
1464 up->forward_deficit = 0;
1465
1466 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1467 * if the called don't held it already
1468 */
1469 sk_queue = &sk->sk_receive_queue;
1470 if (!rx_queue_lock_held)
1471 spin_lock(&sk_queue->lock);
1472
1473
1474 sk->sk_forward_alloc += size;
1475 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1476 sk->sk_forward_alloc -= amt;
1477
1478 if (amt)
1479 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1480
1481 atomic_sub(size, &sk->sk_rmem_alloc);
1482
1483 /* this can save us from acquiring the rx queue lock on next receive */
1484 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1485
1486 if (!rx_queue_lock_held)
1487 spin_unlock(&sk_queue->lock);
1488}
1489
1490/* Note: called with reader_queue.lock held.
1491 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1492 * This avoids a cache line miss while receive_queue lock is held.
1493 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1494 */
1495void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1496{
1497 prefetch(&skb->data);
1498 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1499}
1500EXPORT_SYMBOL(udp_skb_destructor);
1501
1502/* as above, but the caller held the rx queue lock, too */
1503static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1504{
1505 prefetch(&skb->data);
1506 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1507}
1508
1509/* Idea of busylocks is to let producers grab an extra spinlock
1510 * to relieve pressure on the receive_queue spinlock shared by consumer.
1511 * Under flood, this means that only one producer can be in line
1512 * trying to acquire the receive_queue spinlock.
1513 * These busylock can be allocated on a per cpu manner, instead of a
1514 * per socket one (that would consume a cache line per socket)
1515 */
1516static int udp_busylocks_log __read_mostly;
1517static spinlock_t *udp_busylocks __read_mostly;
1518
1519static spinlock_t *busylock_acquire(void *ptr)
1520{
1521 spinlock_t *busy;
1522
1523 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1524 spin_lock(busy);
1525 return busy;
1526}
1527
1528static void busylock_release(spinlock_t *busy)
1529{
1530 if (busy)
1531 spin_unlock(busy);
1532}
1533
1534int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1535{
1536 struct sk_buff_head *list = &sk->sk_receive_queue;
1537 int rmem, delta, amt, err = -ENOMEM;
1538 spinlock_t *busy = NULL;
1539 int size;
1540
1541 /* try to avoid the costly atomic add/sub pair when the receive
1542 * queue is full; always allow at least a packet
1543 */
1544 rmem = atomic_read(&sk->sk_rmem_alloc);
1545 if (rmem > sk->sk_rcvbuf)
1546 goto drop;
1547
1548 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1549 * having linear skbs :
1550 * - Reduce memory overhead and thus increase receive queue capacity
1551 * - Less cache line misses at copyout() time
1552 * - Less work at consume_skb() (less alien page frag freeing)
1553 */
1554 if (rmem > (sk->sk_rcvbuf >> 1)) {
1555 skb_condense(skb);
1556
1557 busy = busylock_acquire(sk);
1558 }
1559 size = skb->truesize;
1560 udp_set_dev_scratch(skb);
1561
1562 /* we drop only if the receive buf is full and the receive
1563 * queue contains some other skb
1564 */
1565 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1566 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1567 goto uncharge_drop;
1568
1569 spin_lock(&list->lock);
1570 if (size >= sk->sk_forward_alloc) {
1571 amt = sk_mem_pages(size);
1572 delta = amt << PAGE_SHIFT;
1573 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1574 err = -ENOBUFS;
1575 spin_unlock(&list->lock);
1576 goto uncharge_drop;
1577 }
1578
1579 sk->sk_forward_alloc += delta;
1580 }
1581
1582 sk->sk_forward_alloc -= size;
1583
1584 /* no need to setup a destructor, we will explicitly release the
1585 * forward allocated memory on dequeue
1586 */
1587 sock_skb_set_dropcount(sk, skb);
1588
1589 __skb_queue_tail(list, skb);
1590 spin_unlock(&list->lock);
1591
1592 if (!sock_flag(sk, SOCK_DEAD))
1593 sk->sk_data_ready(sk);
1594
1595 busylock_release(busy);
1596 return 0;
1597
1598uncharge_drop:
1599 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1600
1601drop:
1602 atomic_inc(&sk->sk_drops);
1603 busylock_release(busy);
1604 return err;
1605}
1606EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1607
1608void udp_destruct_common(struct sock *sk)
1609{
1610 /* reclaim completely the forward allocated memory */
1611 struct udp_sock *up = udp_sk(sk);
1612 unsigned int total = 0;
1613 struct sk_buff *skb;
1614
1615 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1616 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1617 total += skb->truesize;
1618 kfree_skb(skb);
1619 }
1620 udp_rmem_release(sk, total, 0, true);
1621}
1622EXPORT_SYMBOL_GPL(udp_destruct_common);
1623
1624static void udp_destruct_sock(struct sock *sk)
1625{
1626 udp_destruct_common(sk);
1627 inet_sock_destruct(sk);
1628}
1629
1630int udp_init_sock(struct sock *sk)
1631{
1632 udp_lib_init_sock(sk);
1633 sk->sk_destruct = udp_destruct_sock;
1634 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1635 return 0;
1636}
1637
1638void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1639{
1640 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1641 bool slow = lock_sock_fast(sk);
1642
1643 sk_peek_offset_bwd(sk, len);
1644 unlock_sock_fast(sk, slow);
1645 }
1646
1647 if (!skb_unref(skb))
1648 return;
1649
1650 /* In the more common cases we cleared the head states previously,
1651 * see __udp_queue_rcv_skb().
1652 */
1653 if (unlikely(udp_skb_has_head_state(skb)))
1654 skb_release_head_state(skb);
1655 __consume_stateless_skb(skb);
1656}
1657EXPORT_SYMBOL_GPL(skb_consume_udp);
1658
1659static struct sk_buff *__first_packet_length(struct sock *sk,
1660 struct sk_buff_head *rcvq,
1661 int *total)
1662{
1663 struct sk_buff *skb;
1664
1665 while ((skb = skb_peek(rcvq)) != NULL) {
1666 if (udp_lib_checksum_complete(skb)) {
1667 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1668 IS_UDPLITE(sk));
1669 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1670 IS_UDPLITE(sk));
1671 atomic_inc(&sk->sk_drops);
1672 __skb_unlink(skb, rcvq);
1673 *total += skb->truesize;
1674 kfree_skb(skb);
1675 } else {
1676 udp_skb_csum_unnecessary_set(skb);
1677 break;
1678 }
1679 }
1680 return skb;
1681}
1682
1683/**
1684 * first_packet_length - return length of first packet in receive queue
1685 * @sk: socket
1686 *
1687 * Drops all bad checksum frames, until a valid one is found.
1688 * Returns the length of found skb, or -1 if none is found.
1689 */
1690static int first_packet_length(struct sock *sk)
1691{
1692 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1693 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1694 struct sk_buff *skb;
1695 int total = 0;
1696 int res;
1697
1698 spin_lock_bh(&rcvq->lock);
1699 skb = __first_packet_length(sk, rcvq, &total);
1700 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1701 spin_lock(&sk_queue->lock);
1702 skb_queue_splice_tail_init(sk_queue, rcvq);
1703 spin_unlock(&sk_queue->lock);
1704
1705 skb = __first_packet_length(sk, rcvq, &total);
1706 }
1707 res = skb ? skb->len : -1;
1708 if (total)
1709 udp_rmem_release(sk, total, 1, false);
1710 spin_unlock_bh(&rcvq->lock);
1711 return res;
1712}
1713
1714/*
1715 * IOCTL requests applicable to the UDP protocol
1716 */
1717
1718int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1719{
1720 switch (cmd) {
1721 case SIOCOUTQ:
1722 {
1723 int amount = sk_wmem_alloc_get(sk);
1724
1725 return put_user(amount, (int __user *)arg);
1726 }
1727
1728 case SIOCINQ:
1729 {
1730 int amount = max_t(int, 0, first_packet_length(sk));
1731
1732 return put_user(amount, (int __user *)arg);
1733 }
1734
1735 default:
1736 return -ENOIOCTLCMD;
1737 }
1738
1739 return 0;
1740}
1741EXPORT_SYMBOL(udp_ioctl);
1742
1743struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1744 int *off, int *err)
1745{
1746 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1747 struct sk_buff_head *queue;
1748 struct sk_buff *last;
1749 long timeo;
1750 int error;
1751
1752 queue = &udp_sk(sk)->reader_queue;
1753 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1754 do {
1755 struct sk_buff *skb;
1756
1757 error = sock_error(sk);
1758 if (error)
1759 break;
1760
1761 error = -EAGAIN;
1762 do {
1763 spin_lock_bh(&queue->lock);
1764 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1765 err, &last);
1766 if (skb) {
1767 if (!(flags & MSG_PEEK))
1768 udp_skb_destructor(sk, skb);
1769 spin_unlock_bh(&queue->lock);
1770 return skb;
1771 }
1772
1773 if (skb_queue_empty_lockless(sk_queue)) {
1774 spin_unlock_bh(&queue->lock);
1775 goto busy_check;
1776 }
1777
1778 /* refill the reader queue and walk it again
1779 * keep both queues locked to avoid re-acquiring
1780 * the sk_receive_queue lock if fwd memory scheduling
1781 * is needed.
1782 */
1783 spin_lock(&sk_queue->lock);
1784 skb_queue_splice_tail_init(sk_queue, queue);
1785
1786 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1787 err, &last);
1788 if (skb && !(flags & MSG_PEEK))
1789 udp_skb_dtor_locked(sk, skb);
1790 spin_unlock(&sk_queue->lock);
1791 spin_unlock_bh(&queue->lock);
1792 if (skb)
1793 return skb;
1794
1795busy_check:
1796 if (!sk_can_busy_loop(sk))
1797 break;
1798
1799 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1800 } while (!skb_queue_empty_lockless(sk_queue));
1801
1802 /* sk_queue is empty, reader_queue may contain peeked packets */
1803 } while (timeo &&
1804 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1805 &error, &timeo,
1806 (struct sk_buff *)sk_queue));
1807
1808 *err = error;
1809 return NULL;
1810}
1811EXPORT_SYMBOL(__skb_recv_udp);
1812
1813int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1814{
1815 struct sk_buff *skb;
1816 int err, copied;
1817
1818try_again:
1819 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1820 if (!skb)
1821 return err;
1822
1823 if (udp_lib_checksum_complete(skb)) {
1824 int is_udplite = IS_UDPLITE(sk);
1825 struct net *net = sock_net(sk);
1826
1827 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1828 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1829 atomic_inc(&sk->sk_drops);
1830 kfree_skb(skb);
1831 goto try_again;
1832 }
1833
1834 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1835 copied = recv_actor(sk, skb);
1836 kfree_skb(skb);
1837
1838 return copied;
1839}
1840EXPORT_SYMBOL(udp_read_skb);
1841
1842/*
1843 * This should be easy, if there is something there we
1844 * return it, otherwise we block.
1845 */
1846
1847int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1848 int *addr_len)
1849{
1850 struct inet_sock *inet = inet_sk(sk);
1851 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1852 struct sk_buff *skb;
1853 unsigned int ulen, copied;
1854 int off, err, peeking = flags & MSG_PEEK;
1855 int is_udplite = IS_UDPLITE(sk);
1856 bool checksum_valid = false;
1857
1858 if (flags & MSG_ERRQUEUE)
1859 return ip_recv_error(sk, msg, len, addr_len);
1860
1861try_again:
1862 off = sk_peek_offset(sk, flags);
1863 skb = __skb_recv_udp(sk, flags, &off, &err);
1864 if (!skb)
1865 return err;
1866
1867 ulen = udp_skb_len(skb);
1868 copied = len;
1869 if (copied > ulen - off)
1870 copied = ulen - off;
1871 else if (copied < ulen)
1872 msg->msg_flags |= MSG_TRUNC;
1873
1874 /*
1875 * If checksum is needed at all, try to do it while copying the
1876 * data. If the data is truncated, or if we only want a partial
1877 * coverage checksum (UDP-Lite), do it before the copy.
1878 */
1879
1880 if (copied < ulen || peeking ||
1881 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1882 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1883 !__udp_lib_checksum_complete(skb);
1884 if (!checksum_valid)
1885 goto csum_copy_err;
1886 }
1887
1888 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1889 if (udp_skb_is_linear(skb))
1890 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1891 else
1892 err = skb_copy_datagram_msg(skb, off, msg, copied);
1893 } else {
1894 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1895
1896 if (err == -EINVAL)
1897 goto csum_copy_err;
1898 }
1899
1900 if (unlikely(err)) {
1901 if (!peeking) {
1902 atomic_inc(&sk->sk_drops);
1903 UDP_INC_STATS(sock_net(sk),
1904 UDP_MIB_INERRORS, is_udplite);
1905 }
1906 kfree_skb(skb);
1907 return err;
1908 }
1909
1910 if (!peeking)
1911 UDP_INC_STATS(sock_net(sk),
1912 UDP_MIB_INDATAGRAMS, is_udplite);
1913
1914 sock_recv_cmsgs(msg, sk, skb);
1915
1916 /* Copy the address. */
1917 if (sin) {
1918 sin->sin_family = AF_INET;
1919 sin->sin_port = udp_hdr(skb)->source;
1920 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1921 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1922 *addr_len = sizeof(*sin);
1923
1924 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1925 (struct sockaddr *)sin);
1926 }
1927
1928 if (udp_sk(sk)->gro_enabled)
1929 udp_cmsg_recv(msg, sk, skb);
1930
1931 if (inet->cmsg_flags)
1932 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1933
1934 err = copied;
1935 if (flags & MSG_TRUNC)
1936 err = ulen;
1937
1938 skb_consume_udp(sk, skb, peeking ? -err : err);
1939 return err;
1940
1941csum_copy_err:
1942 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1943 udp_skb_destructor)) {
1944 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1945 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1946 }
1947 kfree_skb(skb);
1948
1949 /* starting over for a new packet, but check if we need to yield */
1950 cond_resched();
1951 msg->msg_flags &= ~MSG_TRUNC;
1952 goto try_again;
1953}
1954
1955int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1956{
1957 /* This check is replicated from __ip4_datagram_connect() and
1958 * intended to prevent BPF program called below from accessing bytes
1959 * that are out of the bound specified by user in addr_len.
1960 */
1961 if (addr_len < sizeof(struct sockaddr_in))
1962 return -EINVAL;
1963
1964 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1965}
1966EXPORT_SYMBOL(udp_pre_connect);
1967
1968int __udp_disconnect(struct sock *sk, int flags)
1969{
1970 struct inet_sock *inet = inet_sk(sk);
1971 /*
1972 * 1003.1g - break association.
1973 */
1974
1975 sk->sk_state = TCP_CLOSE;
1976 inet->inet_daddr = 0;
1977 inet->inet_dport = 0;
1978 sock_rps_reset_rxhash(sk);
1979 sk->sk_bound_dev_if = 0;
1980 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1981 inet_reset_saddr(sk);
1982 if (sk->sk_prot->rehash &&
1983 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1984 sk->sk_prot->rehash(sk);
1985 }
1986
1987 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1988 sk->sk_prot->unhash(sk);
1989 inet->inet_sport = 0;
1990 }
1991 sk_dst_reset(sk);
1992 return 0;
1993}
1994EXPORT_SYMBOL(__udp_disconnect);
1995
1996int udp_disconnect(struct sock *sk, int flags)
1997{
1998 lock_sock(sk);
1999 __udp_disconnect(sk, flags);
2000 release_sock(sk);
2001 return 0;
2002}
2003EXPORT_SYMBOL(udp_disconnect);
2004
2005void udp_lib_unhash(struct sock *sk)
2006{
2007 if (sk_hashed(sk)) {
2008 struct udp_table *udptable = udp_get_table_prot(sk);
2009 struct udp_hslot *hslot, *hslot2;
2010
2011 hslot = udp_hashslot(udptable, sock_net(sk),
2012 udp_sk(sk)->udp_port_hash);
2013 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2014
2015 spin_lock_bh(&hslot->lock);
2016 if (rcu_access_pointer(sk->sk_reuseport_cb))
2017 reuseport_detach_sock(sk);
2018 if (sk_del_node_init_rcu(sk)) {
2019 hslot->count--;
2020 inet_sk(sk)->inet_num = 0;
2021 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2022
2023 spin_lock(&hslot2->lock);
2024 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2025 hslot2->count--;
2026 spin_unlock(&hslot2->lock);
2027 }
2028 spin_unlock_bh(&hslot->lock);
2029 }
2030}
2031EXPORT_SYMBOL(udp_lib_unhash);
2032
2033/*
2034 * inet_rcv_saddr was changed, we must rehash secondary hash
2035 */
2036void udp_lib_rehash(struct sock *sk, u16 newhash)
2037{
2038 if (sk_hashed(sk)) {
2039 struct udp_table *udptable = udp_get_table_prot(sk);
2040 struct udp_hslot *hslot, *hslot2, *nhslot2;
2041
2042 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2043 nhslot2 = udp_hashslot2(udptable, newhash);
2044 udp_sk(sk)->udp_portaddr_hash = newhash;
2045
2046 if (hslot2 != nhslot2 ||
2047 rcu_access_pointer(sk->sk_reuseport_cb)) {
2048 hslot = udp_hashslot(udptable, sock_net(sk),
2049 udp_sk(sk)->udp_port_hash);
2050 /* we must lock primary chain too */
2051 spin_lock_bh(&hslot->lock);
2052 if (rcu_access_pointer(sk->sk_reuseport_cb))
2053 reuseport_detach_sock(sk);
2054
2055 if (hslot2 != nhslot2) {
2056 spin_lock(&hslot2->lock);
2057 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2058 hslot2->count--;
2059 spin_unlock(&hslot2->lock);
2060
2061 spin_lock(&nhslot2->lock);
2062 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2063 &nhslot2->head);
2064 nhslot2->count++;
2065 spin_unlock(&nhslot2->lock);
2066 }
2067
2068 spin_unlock_bh(&hslot->lock);
2069 }
2070 }
2071}
2072EXPORT_SYMBOL(udp_lib_rehash);
2073
2074void udp_v4_rehash(struct sock *sk)
2075{
2076 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2077 inet_sk(sk)->inet_rcv_saddr,
2078 inet_sk(sk)->inet_num);
2079 udp_lib_rehash(sk, new_hash);
2080}
2081
2082static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2083{
2084 int rc;
2085
2086 if (inet_sk(sk)->inet_daddr) {
2087 sock_rps_save_rxhash(sk, skb);
2088 sk_mark_napi_id(sk, skb);
2089 sk_incoming_cpu_update(sk);
2090 } else {
2091 sk_mark_napi_id_once(sk, skb);
2092 }
2093
2094 rc = __udp_enqueue_schedule_skb(sk, skb);
2095 if (rc < 0) {
2096 int is_udplite = IS_UDPLITE(sk);
2097 int drop_reason;
2098
2099 /* Note that an ENOMEM error is charged twice */
2100 if (rc == -ENOMEM) {
2101 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2102 is_udplite);
2103 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2104 } else {
2105 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2106 is_udplite);
2107 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2108 }
2109 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2110 kfree_skb_reason(skb, drop_reason);
2111 trace_udp_fail_queue_rcv_skb(rc, sk);
2112 return -1;
2113 }
2114
2115 return 0;
2116}
2117
2118/* returns:
2119 * -1: error
2120 * 0: success
2121 * >0: "udp encap" protocol resubmission
2122 *
2123 * Note that in the success and error cases, the skb is assumed to
2124 * have either been requeued or freed.
2125 */
2126static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2127{
2128 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2129 struct udp_sock *up = udp_sk(sk);
2130 int is_udplite = IS_UDPLITE(sk);
2131
2132 /*
2133 * Charge it to the socket, dropping if the queue is full.
2134 */
2135 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2136 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2137 goto drop;
2138 }
2139 nf_reset_ct(skb);
2140
2141 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2142 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2143
2144 /*
2145 * This is an encapsulation socket so pass the skb to
2146 * the socket's udp_encap_rcv() hook. Otherwise, just
2147 * fall through and pass this up the UDP socket.
2148 * up->encap_rcv() returns the following value:
2149 * =0 if skb was successfully passed to the encap
2150 * handler or was discarded by it.
2151 * >0 if skb should be passed on to UDP.
2152 * <0 if skb should be resubmitted as proto -N
2153 */
2154
2155 /* if we're overly short, let UDP handle it */
2156 encap_rcv = READ_ONCE(up->encap_rcv);
2157 if (encap_rcv) {
2158 int ret;
2159
2160 /* Verify checksum before giving to encap */
2161 if (udp_lib_checksum_complete(skb))
2162 goto csum_error;
2163
2164 ret = encap_rcv(sk, skb);
2165 if (ret <= 0) {
2166 __UDP_INC_STATS(sock_net(sk),
2167 UDP_MIB_INDATAGRAMS,
2168 is_udplite);
2169 return -ret;
2170 }
2171 }
2172
2173 /* FALLTHROUGH -- it's a UDP Packet */
2174 }
2175
2176 /*
2177 * UDP-Lite specific tests, ignored on UDP sockets
2178 */
2179 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2180
2181 /*
2182 * MIB statistics other than incrementing the error count are
2183 * disabled for the following two types of errors: these depend
2184 * on the application settings, not on the functioning of the
2185 * protocol stack as such.
2186 *
2187 * RFC 3828 here recommends (sec 3.3): "There should also be a
2188 * way ... to ... at least let the receiving application block
2189 * delivery of packets with coverage values less than a value
2190 * provided by the application."
2191 */
2192 if (up->pcrlen == 0) { /* full coverage was set */
2193 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2194 UDP_SKB_CB(skb)->cscov, skb->len);
2195 goto drop;
2196 }
2197 /* The next case involves violating the min. coverage requested
2198 * by the receiver. This is subtle: if receiver wants x and x is
2199 * greater than the buffersize/MTU then receiver will complain
2200 * that it wants x while sender emits packets of smaller size y.
2201 * Therefore the above ...()->partial_cov statement is essential.
2202 */
2203 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2204 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2205 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2206 goto drop;
2207 }
2208 }
2209
2210 prefetch(&sk->sk_rmem_alloc);
2211 if (rcu_access_pointer(sk->sk_filter) &&
2212 udp_lib_checksum_complete(skb))
2213 goto csum_error;
2214
2215 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2216 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2217 goto drop;
2218 }
2219
2220 udp_csum_pull_header(skb);
2221
2222 ipv4_pktinfo_prepare(sk, skb);
2223 return __udp_queue_rcv_skb(sk, skb);
2224
2225csum_error:
2226 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2227 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2228drop:
2229 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2230 atomic_inc(&sk->sk_drops);
2231 kfree_skb_reason(skb, drop_reason);
2232 return -1;
2233}
2234
2235static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2236{
2237 struct sk_buff *next, *segs;
2238 int ret;
2239
2240 if (likely(!udp_unexpected_gso(sk, skb)))
2241 return udp_queue_rcv_one_skb(sk, skb);
2242
2243 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2244 __skb_push(skb, -skb_mac_offset(skb));
2245 segs = udp_rcv_segment(sk, skb, true);
2246 skb_list_walk_safe(segs, skb, next) {
2247 __skb_pull(skb, skb_transport_offset(skb));
2248
2249 udp_post_segment_fix_csum(skb);
2250 ret = udp_queue_rcv_one_skb(sk, skb);
2251 if (ret > 0)
2252 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2253 }
2254 return 0;
2255}
2256
2257/* For TCP sockets, sk_rx_dst is protected by socket lock
2258 * For UDP, we use xchg() to guard against concurrent changes.
2259 */
2260bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2261{
2262 struct dst_entry *old;
2263
2264 if (dst_hold_safe(dst)) {
2265 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2266 dst_release(old);
2267 return old != dst;
2268 }
2269 return false;
2270}
2271EXPORT_SYMBOL(udp_sk_rx_dst_set);
2272
2273/*
2274 * Multicasts and broadcasts go to each listener.
2275 *
2276 * Note: called only from the BH handler context.
2277 */
2278static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2279 struct udphdr *uh,
2280 __be32 saddr, __be32 daddr,
2281 struct udp_table *udptable,
2282 int proto)
2283{
2284 struct sock *sk, *first = NULL;
2285 unsigned short hnum = ntohs(uh->dest);
2286 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2287 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2288 unsigned int offset = offsetof(typeof(*sk), sk_node);
2289 int dif = skb->dev->ifindex;
2290 int sdif = inet_sdif(skb);
2291 struct hlist_node *node;
2292 struct sk_buff *nskb;
2293
2294 if (use_hash2) {
2295 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2296 udptable->mask;
2297 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2298start_lookup:
2299 hslot = &udptable->hash2[hash2];
2300 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2301 }
2302
2303 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2304 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2305 uh->source, saddr, dif, sdif, hnum))
2306 continue;
2307
2308 if (!first) {
2309 first = sk;
2310 continue;
2311 }
2312 nskb = skb_clone(skb, GFP_ATOMIC);
2313
2314 if (unlikely(!nskb)) {
2315 atomic_inc(&sk->sk_drops);
2316 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2317 IS_UDPLITE(sk));
2318 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2319 IS_UDPLITE(sk));
2320 continue;
2321 }
2322 if (udp_queue_rcv_skb(sk, nskb) > 0)
2323 consume_skb(nskb);
2324 }
2325
2326 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2327 if (use_hash2 && hash2 != hash2_any) {
2328 hash2 = hash2_any;
2329 goto start_lookup;
2330 }
2331
2332 if (first) {
2333 if (udp_queue_rcv_skb(first, skb) > 0)
2334 consume_skb(skb);
2335 } else {
2336 kfree_skb(skb);
2337 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2338 proto == IPPROTO_UDPLITE);
2339 }
2340 return 0;
2341}
2342
2343/* Initialize UDP checksum. If exited with zero value (success),
2344 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2345 * Otherwise, csum completion requires checksumming packet body,
2346 * including udp header and folding it to skb->csum.
2347 */
2348static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2349 int proto)
2350{
2351 int err;
2352
2353 UDP_SKB_CB(skb)->partial_cov = 0;
2354 UDP_SKB_CB(skb)->cscov = skb->len;
2355
2356 if (proto == IPPROTO_UDPLITE) {
2357 err = udplite_checksum_init(skb, uh);
2358 if (err)
2359 return err;
2360
2361 if (UDP_SKB_CB(skb)->partial_cov) {
2362 skb->csum = inet_compute_pseudo(skb, proto);
2363 return 0;
2364 }
2365 }
2366
2367 /* Note, we are only interested in != 0 or == 0, thus the
2368 * force to int.
2369 */
2370 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2371 inet_compute_pseudo);
2372 if (err)
2373 return err;
2374
2375 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2376 /* If SW calculated the value, we know it's bad */
2377 if (skb->csum_complete_sw)
2378 return 1;
2379
2380 /* HW says the value is bad. Let's validate that.
2381 * skb->csum is no longer the full packet checksum,
2382 * so don't treat it as such.
2383 */
2384 skb_checksum_complete_unset(skb);
2385 }
2386
2387 return 0;
2388}
2389
2390/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2391 * return code conversion for ip layer consumption
2392 */
2393static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2394 struct udphdr *uh)
2395{
2396 int ret;
2397
2398 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2399 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2400
2401 ret = udp_queue_rcv_skb(sk, skb);
2402
2403 /* a return value > 0 means to resubmit the input, but
2404 * it wants the return to be -protocol, or 0
2405 */
2406 if (ret > 0)
2407 return -ret;
2408 return 0;
2409}
2410
2411/*
2412 * All we need to do is get the socket, and then do a checksum.
2413 */
2414
2415int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2416 int proto)
2417{
2418 struct sock *sk;
2419 struct udphdr *uh;
2420 unsigned short ulen;
2421 struct rtable *rt = skb_rtable(skb);
2422 __be32 saddr, daddr;
2423 struct net *net = dev_net(skb->dev);
2424 bool refcounted;
2425 int drop_reason;
2426
2427 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2428
2429 /*
2430 * Validate the packet.
2431 */
2432 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2433 goto drop; /* No space for header. */
2434
2435 uh = udp_hdr(skb);
2436 ulen = ntohs(uh->len);
2437 saddr = ip_hdr(skb)->saddr;
2438 daddr = ip_hdr(skb)->daddr;
2439
2440 if (ulen > skb->len)
2441 goto short_packet;
2442
2443 if (proto == IPPROTO_UDP) {
2444 /* UDP validates ulen. */
2445 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2446 goto short_packet;
2447 uh = udp_hdr(skb);
2448 }
2449
2450 if (udp4_csum_init(skb, uh, proto))
2451 goto csum_error;
2452
2453 sk = skb_steal_sock(skb, &refcounted);
2454 if (sk) {
2455 struct dst_entry *dst = skb_dst(skb);
2456 int ret;
2457
2458 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2459 udp_sk_rx_dst_set(sk, dst);
2460
2461 ret = udp_unicast_rcv_skb(sk, skb, uh);
2462 if (refcounted)
2463 sock_put(sk);
2464 return ret;
2465 }
2466
2467 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2468 return __udp4_lib_mcast_deliver(net, skb, uh,
2469 saddr, daddr, udptable, proto);
2470
2471 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2472 if (sk)
2473 return udp_unicast_rcv_skb(sk, skb, uh);
2474
2475 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2476 goto drop;
2477 nf_reset_ct(skb);
2478
2479 /* No socket. Drop packet silently, if checksum is wrong */
2480 if (udp_lib_checksum_complete(skb))
2481 goto csum_error;
2482
2483 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2484 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2485 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2486
2487 /*
2488 * Hmm. We got an UDP packet to a port to which we
2489 * don't wanna listen. Ignore it.
2490 */
2491 kfree_skb_reason(skb, drop_reason);
2492 return 0;
2493
2494short_packet:
2495 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2496 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2497 proto == IPPROTO_UDPLITE ? "Lite" : "",
2498 &saddr, ntohs(uh->source),
2499 ulen, skb->len,
2500 &daddr, ntohs(uh->dest));
2501 goto drop;
2502
2503csum_error:
2504 /*
2505 * RFC1122: OK. Discards the bad packet silently (as far as
2506 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2507 */
2508 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2509 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2510 proto == IPPROTO_UDPLITE ? "Lite" : "",
2511 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2512 ulen);
2513 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2514drop:
2515 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2516 kfree_skb_reason(skb, drop_reason);
2517 return 0;
2518}
2519
2520/* We can only early demux multicast if there is a single matching socket.
2521 * If more than one socket found returns NULL
2522 */
2523static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2524 __be16 loc_port, __be32 loc_addr,
2525 __be16 rmt_port, __be32 rmt_addr,
2526 int dif, int sdif)
2527{
2528 struct udp_table *udptable = net->ipv4.udp_table;
2529 unsigned short hnum = ntohs(loc_port);
2530 struct sock *sk, *result;
2531 struct udp_hslot *hslot;
2532 unsigned int slot;
2533
2534 slot = udp_hashfn(net, hnum, udptable->mask);
2535 hslot = &udptable->hash[slot];
2536
2537 /* Do not bother scanning a too big list */
2538 if (hslot->count > 10)
2539 return NULL;
2540
2541 result = NULL;
2542 sk_for_each_rcu(sk, &hslot->head) {
2543 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2544 rmt_port, rmt_addr, dif, sdif, hnum)) {
2545 if (result)
2546 return NULL;
2547 result = sk;
2548 }
2549 }
2550
2551 return result;
2552}
2553
2554/* For unicast we should only early demux connected sockets or we can
2555 * break forwarding setups. The chains here can be long so only check
2556 * if the first socket is an exact match and if not move on.
2557 */
2558static struct sock *__udp4_lib_demux_lookup(struct net *net,
2559 __be16 loc_port, __be32 loc_addr,
2560 __be16 rmt_port, __be32 rmt_addr,
2561 int dif, int sdif)
2562{
2563 struct udp_table *udptable = net->ipv4.udp_table;
2564 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2565 unsigned short hnum = ntohs(loc_port);
2566 unsigned int hash2, slot2;
2567 struct udp_hslot *hslot2;
2568 __portpair ports;
2569 struct sock *sk;
2570
2571 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2572 slot2 = hash2 & udptable->mask;
2573 hslot2 = &udptable->hash2[slot2];
2574 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2575
2576 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2577 if (inet_match(net, sk, acookie, ports, dif, sdif))
2578 return sk;
2579 /* Only check first socket in chain */
2580 break;
2581 }
2582 return NULL;
2583}
2584
2585int udp_v4_early_demux(struct sk_buff *skb)
2586{
2587 struct net *net = dev_net(skb->dev);
2588 struct in_device *in_dev = NULL;
2589 const struct iphdr *iph;
2590 const struct udphdr *uh;
2591 struct sock *sk = NULL;
2592 struct dst_entry *dst;
2593 int dif = skb->dev->ifindex;
2594 int sdif = inet_sdif(skb);
2595 int ours;
2596
2597 /* validate the packet */
2598 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2599 return 0;
2600
2601 iph = ip_hdr(skb);
2602 uh = udp_hdr(skb);
2603
2604 if (skb->pkt_type == PACKET_MULTICAST) {
2605 in_dev = __in_dev_get_rcu(skb->dev);
2606
2607 if (!in_dev)
2608 return 0;
2609
2610 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2611 iph->protocol);
2612 if (!ours)
2613 return 0;
2614
2615 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2616 uh->source, iph->saddr,
2617 dif, sdif);
2618 } else if (skb->pkt_type == PACKET_HOST) {
2619 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2620 uh->source, iph->saddr, dif, sdif);
2621 }
2622
2623 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2624 return 0;
2625
2626 skb->sk = sk;
2627 skb->destructor = sock_efree;
2628 dst = rcu_dereference(sk->sk_rx_dst);
2629
2630 if (dst)
2631 dst = dst_check(dst, 0);
2632 if (dst) {
2633 u32 itag = 0;
2634
2635 /* set noref for now.
2636 * any place which wants to hold dst has to call
2637 * dst_hold_safe()
2638 */
2639 skb_dst_set_noref(skb, dst);
2640
2641 /* for unconnected multicast sockets we need to validate
2642 * the source on each packet
2643 */
2644 if (!inet_sk(sk)->inet_daddr && in_dev)
2645 return ip_mc_validate_source(skb, iph->daddr,
2646 iph->saddr,
2647 iph->tos & IPTOS_RT_MASK,
2648 skb->dev, in_dev, &itag);
2649 }
2650 return 0;
2651}
2652
2653int udp_rcv(struct sk_buff *skb)
2654{
2655 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2656}
2657
2658void udp_destroy_sock(struct sock *sk)
2659{
2660 struct udp_sock *up = udp_sk(sk);
2661 bool slow = lock_sock_fast(sk);
2662
2663 /* protects from races with udp_abort() */
2664 sock_set_flag(sk, SOCK_DEAD);
2665 udp_flush_pending_frames(sk);
2666 unlock_sock_fast(sk, slow);
2667 if (static_branch_unlikely(&udp_encap_needed_key)) {
2668 if (up->encap_type) {
2669 void (*encap_destroy)(struct sock *sk);
2670 encap_destroy = READ_ONCE(up->encap_destroy);
2671 if (encap_destroy)
2672 encap_destroy(sk);
2673 }
2674 if (up->encap_enabled)
2675 static_branch_dec(&udp_encap_needed_key);
2676 }
2677}
2678
2679/*
2680 * Socket option code for UDP
2681 */
2682int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2683 sockptr_t optval, unsigned int optlen,
2684 int (*push_pending_frames)(struct sock *))
2685{
2686 struct udp_sock *up = udp_sk(sk);
2687 int val, valbool;
2688 int err = 0;
2689 int is_udplite = IS_UDPLITE(sk);
2690
2691 if (level == SOL_SOCKET) {
2692 err = sk_setsockopt(sk, level, optname, optval, optlen);
2693
2694 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2695 sockopt_lock_sock(sk);
2696 /* paired with READ_ONCE in udp_rmem_release() */
2697 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2698 sockopt_release_sock(sk);
2699 }
2700 return err;
2701 }
2702
2703 if (optlen < sizeof(int))
2704 return -EINVAL;
2705
2706 if (copy_from_sockptr(&val, optval, sizeof(val)))
2707 return -EFAULT;
2708
2709 valbool = val ? 1 : 0;
2710
2711 switch (optname) {
2712 case UDP_CORK:
2713 if (val != 0) {
2714 WRITE_ONCE(up->corkflag, 1);
2715 } else {
2716 WRITE_ONCE(up->corkflag, 0);
2717 lock_sock(sk);
2718 push_pending_frames(sk);
2719 release_sock(sk);
2720 }
2721 break;
2722
2723 case UDP_ENCAP:
2724 switch (val) {
2725 case 0:
2726#ifdef CONFIG_XFRM
2727 case UDP_ENCAP_ESPINUDP:
2728 case UDP_ENCAP_ESPINUDP_NON_IKE:
2729#if IS_ENABLED(CONFIG_IPV6)
2730 if (sk->sk_family == AF_INET6)
2731 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2732 else
2733#endif
2734 up->encap_rcv = xfrm4_udp_encap_rcv;
2735#endif
2736 fallthrough;
2737 case UDP_ENCAP_L2TPINUDP:
2738 up->encap_type = val;
2739 lock_sock(sk);
2740 udp_tunnel_encap_enable(sk->sk_socket);
2741 release_sock(sk);
2742 break;
2743 default:
2744 err = -ENOPROTOOPT;
2745 break;
2746 }
2747 break;
2748
2749 case UDP_NO_CHECK6_TX:
2750 up->no_check6_tx = valbool;
2751 break;
2752
2753 case UDP_NO_CHECK6_RX:
2754 up->no_check6_rx = valbool;
2755 break;
2756
2757 case UDP_SEGMENT:
2758 if (val < 0 || val > USHRT_MAX)
2759 return -EINVAL;
2760 WRITE_ONCE(up->gso_size, val);
2761 break;
2762
2763 case UDP_GRO:
2764 lock_sock(sk);
2765
2766 /* when enabling GRO, accept the related GSO packet type */
2767 if (valbool)
2768 udp_tunnel_encap_enable(sk->sk_socket);
2769 up->gro_enabled = valbool;
2770 up->accept_udp_l4 = valbool;
2771 release_sock(sk);
2772 break;
2773
2774 /*
2775 * UDP-Lite's partial checksum coverage (RFC 3828).
2776 */
2777 /* The sender sets actual checksum coverage length via this option.
2778 * The case coverage > packet length is handled by send module. */
2779 case UDPLITE_SEND_CSCOV:
2780 if (!is_udplite) /* Disable the option on UDP sockets */
2781 return -ENOPROTOOPT;
2782 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2783 val = 8;
2784 else if (val > USHRT_MAX)
2785 val = USHRT_MAX;
2786 up->pcslen = val;
2787 up->pcflag |= UDPLITE_SEND_CC;
2788 break;
2789
2790 /* The receiver specifies a minimum checksum coverage value. To make
2791 * sense, this should be set to at least 8 (as done below). If zero is
2792 * used, this again means full checksum coverage. */
2793 case UDPLITE_RECV_CSCOV:
2794 if (!is_udplite) /* Disable the option on UDP sockets */
2795 return -ENOPROTOOPT;
2796 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2797 val = 8;
2798 else if (val > USHRT_MAX)
2799 val = USHRT_MAX;
2800 up->pcrlen = val;
2801 up->pcflag |= UDPLITE_RECV_CC;
2802 break;
2803
2804 default:
2805 err = -ENOPROTOOPT;
2806 break;
2807 }
2808
2809 return err;
2810}
2811EXPORT_SYMBOL(udp_lib_setsockopt);
2812
2813int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2814 unsigned int optlen)
2815{
2816 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2817 return udp_lib_setsockopt(sk, level, optname,
2818 optval, optlen,
2819 udp_push_pending_frames);
2820 return ip_setsockopt(sk, level, optname, optval, optlen);
2821}
2822
2823int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2824 char __user *optval, int __user *optlen)
2825{
2826 struct udp_sock *up = udp_sk(sk);
2827 int val, len;
2828
2829 if (get_user(len, optlen))
2830 return -EFAULT;
2831
2832 len = min_t(unsigned int, len, sizeof(int));
2833
2834 if (len < 0)
2835 return -EINVAL;
2836
2837 switch (optname) {
2838 case UDP_CORK:
2839 val = READ_ONCE(up->corkflag);
2840 break;
2841
2842 case UDP_ENCAP:
2843 val = up->encap_type;
2844 break;
2845
2846 case UDP_NO_CHECK6_TX:
2847 val = up->no_check6_tx;
2848 break;
2849
2850 case UDP_NO_CHECK6_RX:
2851 val = up->no_check6_rx;
2852 break;
2853
2854 case UDP_SEGMENT:
2855 val = READ_ONCE(up->gso_size);
2856 break;
2857
2858 case UDP_GRO:
2859 val = up->gro_enabled;
2860 break;
2861
2862 /* The following two cannot be changed on UDP sockets, the return is
2863 * always 0 (which corresponds to the full checksum coverage of UDP). */
2864 case UDPLITE_SEND_CSCOV:
2865 val = up->pcslen;
2866 break;
2867
2868 case UDPLITE_RECV_CSCOV:
2869 val = up->pcrlen;
2870 break;
2871
2872 default:
2873 return -ENOPROTOOPT;
2874 }
2875
2876 if (put_user(len, optlen))
2877 return -EFAULT;
2878 if (copy_to_user(optval, &val, len))
2879 return -EFAULT;
2880 return 0;
2881}
2882EXPORT_SYMBOL(udp_lib_getsockopt);
2883
2884int udp_getsockopt(struct sock *sk, int level, int optname,
2885 char __user *optval, int __user *optlen)
2886{
2887 if (level == SOL_UDP || level == SOL_UDPLITE)
2888 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2889 return ip_getsockopt(sk, level, optname, optval, optlen);
2890}
2891
2892/**
2893 * udp_poll - wait for a UDP event.
2894 * @file: - file struct
2895 * @sock: - socket
2896 * @wait: - poll table
2897 *
2898 * This is same as datagram poll, except for the special case of
2899 * blocking sockets. If application is using a blocking fd
2900 * and a packet with checksum error is in the queue;
2901 * then it could get return from select indicating data available
2902 * but then block when reading it. Add special case code
2903 * to work around these arguably broken applications.
2904 */
2905__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2906{
2907 __poll_t mask = datagram_poll(file, sock, wait);
2908 struct sock *sk = sock->sk;
2909
2910 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2911 mask |= EPOLLIN | EPOLLRDNORM;
2912
2913 /* Check for false positives due to checksum errors */
2914 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2915 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2916 mask &= ~(EPOLLIN | EPOLLRDNORM);
2917
2918 /* psock ingress_msg queue should not contain any bad checksum frames */
2919 if (sk_is_readable(sk))
2920 mask |= EPOLLIN | EPOLLRDNORM;
2921 return mask;
2922
2923}
2924EXPORT_SYMBOL(udp_poll);
2925
2926int udp_abort(struct sock *sk, int err)
2927{
2928 lock_sock(sk);
2929
2930 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2931 * with close()
2932 */
2933 if (sock_flag(sk, SOCK_DEAD))
2934 goto out;
2935
2936 sk->sk_err = err;
2937 sk_error_report(sk);
2938 __udp_disconnect(sk, 0);
2939
2940out:
2941 release_sock(sk);
2942
2943 return 0;
2944}
2945EXPORT_SYMBOL_GPL(udp_abort);
2946
2947struct proto udp_prot = {
2948 .name = "UDP",
2949 .owner = THIS_MODULE,
2950 .close = udp_lib_close,
2951 .pre_connect = udp_pre_connect,
2952 .connect = ip4_datagram_connect,
2953 .disconnect = udp_disconnect,
2954 .ioctl = udp_ioctl,
2955 .init = udp_init_sock,
2956 .destroy = udp_destroy_sock,
2957 .setsockopt = udp_setsockopt,
2958 .getsockopt = udp_getsockopt,
2959 .sendmsg = udp_sendmsg,
2960 .recvmsg = udp_recvmsg,
2961 .sendpage = udp_sendpage,
2962 .release_cb = ip4_datagram_release_cb,
2963 .hash = udp_lib_hash,
2964 .unhash = udp_lib_unhash,
2965 .rehash = udp_v4_rehash,
2966 .get_port = udp_v4_get_port,
2967 .put_port = udp_lib_unhash,
2968#ifdef CONFIG_BPF_SYSCALL
2969 .psock_update_sk_prot = udp_bpf_update_proto,
2970#endif
2971 .memory_allocated = &udp_memory_allocated,
2972 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2973
2974 .sysctl_mem = sysctl_udp_mem,
2975 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2976 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2977 .obj_size = sizeof(struct udp_sock),
2978 .h.udp_table = NULL,
2979 .diag_destroy = udp_abort,
2980};
2981EXPORT_SYMBOL(udp_prot);
2982
2983/* ------------------------------------------------------------------------ */
2984#ifdef CONFIG_PROC_FS
2985
2986static struct udp_table *udp_get_table_afinfo(struct udp_seq_afinfo *afinfo,
2987 struct net *net)
2988{
2989 return afinfo->udp_table ? : net->ipv4.udp_table;
2990}
2991
2992static struct sock *udp_get_first(struct seq_file *seq, int start)
2993{
2994 struct udp_iter_state *state = seq->private;
2995 struct net *net = seq_file_net(seq);
2996 struct udp_seq_afinfo *afinfo;
2997 struct udp_table *udptable;
2998 struct sock *sk;
2999
3000 if (state->bpf_seq_afinfo)
3001 afinfo = state->bpf_seq_afinfo;
3002 else
3003 afinfo = pde_data(file_inode(seq->file));
3004
3005 udptable = udp_get_table_afinfo(afinfo, net);
3006
3007 for (state->bucket = start; state->bucket <= udptable->mask;
3008 ++state->bucket) {
3009 struct udp_hslot *hslot = &udptable->hash[state->bucket];
3010
3011 if (hlist_empty(&hslot->head))
3012 continue;
3013
3014 spin_lock_bh(&hslot->lock);
3015 sk_for_each(sk, &hslot->head) {
3016 if (!net_eq(sock_net(sk), net))
3017 continue;
3018 if (afinfo->family == AF_UNSPEC ||
3019 sk->sk_family == afinfo->family)
3020 goto found;
3021 }
3022 spin_unlock_bh(&hslot->lock);
3023 }
3024 sk = NULL;
3025found:
3026 return sk;
3027}
3028
3029static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3030{
3031 struct udp_iter_state *state = seq->private;
3032 struct net *net = seq_file_net(seq);
3033 struct udp_seq_afinfo *afinfo;
3034 struct udp_table *udptable;
3035
3036 if (state->bpf_seq_afinfo)
3037 afinfo = state->bpf_seq_afinfo;
3038 else
3039 afinfo = pde_data(file_inode(seq->file));
3040
3041 do {
3042 sk = sk_next(sk);
3043 } while (sk && (!net_eq(sock_net(sk), net) ||
3044 (afinfo->family != AF_UNSPEC &&
3045 sk->sk_family != afinfo->family)));
3046
3047 if (!sk) {
3048 udptable = udp_get_table_afinfo(afinfo, net);
3049
3050 if (state->bucket <= udptable->mask)
3051 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3052
3053 return udp_get_first(seq, state->bucket + 1);
3054 }
3055 return sk;
3056}
3057
3058static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3059{
3060 struct sock *sk = udp_get_first(seq, 0);
3061
3062 if (sk)
3063 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3064 --pos;
3065 return pos ? NULL : sk;
3066}
3067
3068void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3069{
3070 struct udp_iter_state *state = seq->private;
3071 state->bucket = MAX_UDP_PORTS;
3072
3073 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3074}
3075EXPORT_SYMBOL(udp_seq_start);
3076
3077void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3078{
3079 struct sock *sk;
3080
3081 if (v == SEQ_START_TOKEN)
3082 sk = udp_get_idx(seq, 0);
3083 else
3084 sk = udp_get_next(seq, v);
3085
3086 ++*pos;
3087 return sk;
3088}
3089EXPORT_SYMBOL(udp_seq_next);
3090
3091void udp_seq_stop(struct seq_file *seq, void *v)
3092{
3093 struct udp_iter_state *state = seq->private;
3094 struct udp_seq_afinfo *afinfo;
3095 struct udp_table *udptable;
3096
3097 if (state->bpf_seq_afinfo)
3098 afinfo = state->bpf_seq_afinfo;
3099 else
3100 afinfo = pde_data(file_inode(seq->file));
3101
3102 udptable = udp_get_table_afinfo(afinfo, seq_file_net(seq));
3103
3104 if (state->bucket <= udptable->mask)
3105 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3106}
3107EXPORT_SYMBOL(udp_seq_stop);
3108
3109/* ------------------------------------------------------------------------ */
3110static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3111 int bucket)
3112{
3113 struct inet_sock *inet = inet_sk(sp);
3114 __be32 dest = inet->inet_daddr;
3115 __be32 src = inet->inet_rcv_saddr;
3116 __u16 destp = ntohs(inet->inet_dport);
3117 __u16 srcp = ntohs(inet->inet_sport);
3118
3119 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3120 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3121 bucket, src, srcp, dest, destp, sp->sk_state,
3122 sk_wmem_alloc_get(sp),
3123 udp_rqueue_get(sp),
3124 0, 0L, 0,
3125 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3126 0, sock_i_ino(sp),
3127 refcount_read(&sp->sk_refcnt), sp,
3128 atomic_read(&sp->sk_drops));
3129}
3130
3131int udp4_seq_show(struct seq_file *seq, void *v)
3132{
3133 seq_setwidth(seq, 127);
3134 if (v == SEQ_START_TOKEN)
3135 seq_puts(seq, " sl local_address rem_address st tx_queue "
3136 "rx_queue tr tm->when retrnsmt uid timeout "
3137 "inode ref pointer drops");
3138 else {
3139 struct udp_iter_state *state = seq->private;
3140
3141 udp4_format_sock(v, seq, state->bucket);
3142 }
3143 seq_pad(seq, '\n');
3144 return 0;
3145}
3146
3147#ifdef CONFIG_BPF_SYSCALL
3148struct bpf_iter__udp {
3149 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3150 __bpf_md_ptr(struct udp_sock *, udp_sk);
3151 uid_t uid __aligned(8);
3152 int bucket __aligned(8);
3153};
3154
3155static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3156 struct udp_sock *udp_sk, uid_t uid, int bucket)
3157{
3158 struct bpf_iter__udp ctx;
3159
3160 meta->seq_num--; /* skip SEQ_START_TOKEN */
3161 ctx.meta = meta;
3162 ctx.udp_sk = udp_sk;
3163 ctx.uid = uid;
3164 ctx.bucket = bucket;
3165 return bpf_iter_run_prog(prog, &ctx);
3166}
3167
3168static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3169{
3170 struct udp_iter_state *state = seq->private;
3171 struct bpf_iter_meta meta;
3172 struct bpf_prog *prog;
3173 struct sock *sk = v;
3174 uid_t uid;
3175
3176 if (v == SEQ_START_TOKEN)
3177 return 0;
3178
3179 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3180 meta.seq = seq;
3181 prog = bpf_iter_get_info(&meta, false);
3182 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3183}
3184
3185static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3186{
3187 struct bpf_iter_meta meta;
3188 struct bpf_prog *prog;
3189
3190 if (!v) {
3191 meta.seq = seq;
3192 prog = bpf_iter_get_info(&meta, true);
3193 if (prog)
3194 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3195 }
3196
3197 udp_seq_stop(seq, v);
3198}
3199
3200static const struct seq_operations bpf_iter_udp_seq_ops = {
3201 .start = udp_seq_start,
3202 .next = udp_seq_next,
3203 .stop = bpf_iter_udp_seq_stop,
3204 .show = bpf_iter_udp_seq_show,
3205};
3206#endif
3207
3208const struct seq_operations udp_seq_ops = {
3209 .start = udp_seq_start,
3210 .next = udp_seq_next,
3211 .stop = udp_seq_stop,
3212 .show = udp4_seq_show,
3213};
3214EXPORT_SYMBOL(udp_seq_ops);
3215
3216static struct udp_seq_afinfo udp4_seq_afinfo = {
3217 .family = AF_INET,
3218 .udp_table = NULL,
3219};
3220
3221static int __net_init udp4_proc_init_net(struct net *net)
3222{
3223 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3224 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3225 return -ENOMEM;
3226 return 0;
3227}
3228
3229static void __net_exit udp4_proc_exit_net(struct net *net)
3230{
3231 remove_proc_entry("udp", net->proc_net);
3232}
3233
3234static struct pernet_operations udp4_net_ops = {
3235 .init = udp4_proc_init_net,
3236 .exit = udp4_proc_exit_net,
3237};
3238
3239int __init udp4_proc_init(void)
3240{
3241 return register_pernet_subsys(&udp4_net_ops);
3242}
3243
3244void udp4_proc_exit(void)
3245{
3246 unregister_pernet_subsys(&udp4_net_ops);
3247}
3248#endif /* CONFIG_PROC_FS */
3249
3250static __initdata unsigned long uhash_entries;
3251static int __init set_uhash_entries(char *str)
3252{
3253 ssize_t ret;
3254
3255 if (!str)
3256 return 0;
3257
3258 ret = kstrtoul(str, 0, &uhash_entries);
3259 if (ret)
3260 return 0;
3261
3262 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3263 uhash_entries = UDP_HTABLE_SIZE_MIN;
3264 return 1;
3265}
3266__setup("uhash_entries=", set_uhash_entries);
3267
3268void __init udp_table_init(struct udp_table *table, const char *name)
3269{
3270 unsigned int i;
3271
3272 table->hash = alloc_large_system_hash(name,
3273 2 * sizeof(struct udp_hslot),
3274 uhash_entries,
3275 21, /* one slot per 2 MB */
3276 0,
3277 &table->log,
3278 &table->mask,
3279 UDP_HTABLE_SIZE_MIN,
3280 UDP_HTABLE_SIZE_MAX);
3281
3282 table->hash2 = table->hash + (table->mask + 1);
3283 for (i = 0; i <= table->mask; i++) {
3284 INIT_HLIST_HEAD(&table->hash[i].head);
3285 table->hash[i].count = 0;
3286 spin_lock_init(&table->hash[i].lock);
3287 }
3288 for (i = 0; i <= table->mask; i++) {
3289 INIT_HLIST_HEAD(&table->hash2[i].head);
3290 table->hash2[i].count = 0;
3291 spin_lock_init(&table->hash2[i].lock);
3292 }
3293}
3294
3295u32 udp_flow_hashrnd(void)
3296{
3297 static u32 hashrnd __read_mostly;
3298
3299 net_get_random_once(&hashrnd, sizeof(hashrnd));
3300
3301 return hashrnd;
3302}
3303EXPORT_SYMBOL(udp_flow_hashrnd);
3304
3305static void __net_init udp_sysctl_init(struct net *net)
3306{
3307 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3308 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3309
3310#ifdef CONFIG_NET_L3_MASTER_DEV
3311 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3312#endif
3313}
3314
3315static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3316{
3317 struct udp_table *udptable;
3318 int i;
3319
3320 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3321 if (!udptable)
3322 goto out;
3323
3324 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3325 GFP_KERNEL_ACCOUNT);
3326 if (!udptable->hash)
3327 goto free_table;
3328
3329 udptable->hash2 = udptable->hash + hash_entries;
3330 udptable->mask = hash_entries - 1;
3331 udptable->log = ilog2(hash_entries);
3332
3333 for (i = 0; i < hash_entries; i++) {
3334 INIT_HLIST_HEAD(&udptable->hash[i].head);
3335 udptable->hash[i].count = 0;
3336 spin_lock_init(&udptable->hash[i].lock);
3337
3338 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3339 udptable->hash2[i].count = 0;
3340 spin_lock_init(&udptable->hash2[i].lock);
3341 }
3342
3343 return udptable;
3344
3345free_table:
3346 kfree(udptable);
3347out:
3348 return NULL;
3349}
3350
3351static void __net_exit udp_pernet_table_free(struct net *net)
3352{
3353 struct udp_table *udptable = net->ipv4.udp_table;
3354
3355 if (udptable == &udp_table)
3356 return;
3357
3358 kvfree(udptable->hash);
3359 kfree(udptable);
3360}
3361
3362static void __net_init udp_set_table(struct net *net)
3363{
3364 struct udp_table *udptable;
3365 unsigned int hash_entries;
3366 struct net *old_net;
3367
3368 if (net_eq(net, &init_net))
3369 goto fallback;
3370
3371 old_net = current->nsproxy->net_ns;
3372 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3373 if (!hash_entries)
3374 goto fallback;
3375
3376 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3377 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3378 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3379 else
3380 hash_entries = roundup_pow_of_two(hash_entries);
3381
3382 udptable = udp_pernet_table_alloc(hash_entries);
3383 if (udptable) {
3384 net->ipv4.udp_table = udptable;
3385 } else {
3386 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3387 "for a netns, fallback to the global one\n",
3388 hash_entries);
3389fallback:
3390 net->ipv4.udp_table = &udp_table;
3391 }
3392}
3393
3394static int __net_init udp_pernet_init(struct net *net)
3395{
3396 udp_sysctl_init(net);
3397 udp_set_table(net);
3398
3399 return 0;
3400}
3401
3402static void __net_exit udp_pernet_exit(struct net *net)
3403{
3404 udp_pernet_table_free(net);
3405}
3406
3407static struct pernet_operations __net_initdata udp_sysctl_ops = {
3408 .init = udp_pernet_init,
3409 .exit = udp_pernet_exit,
3410};
3411
3412#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3413DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3414 struct udp_sock *udp_sk, uid_t uid, int bucket)
3415
3416static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3417{
3418 struct udp_iter_state *st = priv_data;
3419 struct udp_seq_afinfo *afinfo;
3420 int ret;
3421
3422 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3423 if (!afinfo)
3424 return -ENOMEM;
3425
3426 afinfo->family = AF_UNSPEC;
3427 afinfo->udp_table = NULL;
3428 st->bpf_seq_afinfo = afinfo;
3429 ret = bpf_iter_init_seq_net(priv_data, aux);
3430 if (ret)
3431 kfree(afinfo);
3432 return ret;
3433}
3434
3435static void bpf_iter_fini_udp(void *priv_data)
3436{
3437 struct udp_iter_state *st = priv_data;
3438
3439 kfree(st->bpf_seq_afinfo);
3440 bpf_iter_fini_seq_net(priv_data);
3441}
3442
3443static const struct bpf_iter_seq_info udp_seq_info = {
3444 .seq_ops = &bpf_iter_udp_seq_ops,
3445 .init_seq_private = bpf_iter_init_udp,
3446 .fini_seq_private = bpf_iter_fini_udp,
3447 .seq_priv_size = sizeof(struct udp_iter_state),
3448};
3449
3450static struct bpf_iter_reg udp_reg_info = {
3451 .target = "udp",
3452 .ctx_arg_info_size = 1,
3453 .ctx_arg_info = {
3454 { offsetof(struct bpf_iter__udp, udp_sk),
3455 PTR_TO_BTF_ID_OR_NULL },
3456 },
3457 .seq_info = &udp_seq_info,
3458};
3459
3460static void __init bpf_iter_register(void)
3461{
3462 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3463 if (bpf_iter_reg_target(&udp_reg_info))
3464 pr_warn("Warning: could not register bpf iterator udp\n");
3465}
3466#endif
3467
3468void __init udp_init(void)
3469{
3470 unsigned long limit;
3471 unsigned int i;
3472
3473 udp_table_init(&udp_table, "UDP");
3474 limit = nr_free_buffer_pages() / 8;
3475 limit = max(limit, 128UL);
3476 sysctl_udp_mem[0] = limit / 4 * 3;
3477 sysctl_udp_mem[1] = limit;
3478 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3479
3480 /* 16 spinlocks per cpu */
3481 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3482 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3483 GFP_KERNEL);
3484 if (!udp_busylocks)
3485 panic("UDP: failed to alloc udp_busylocks\n");
3486 for (i = 0; i < (1U << udp_busylocks_log); i++)
3487 spin_lock_init(udp_busylocks + i);
3488
3489 if (register_pernet_subsys(&udp_sysctl_ops))
3490 panic("UDP: failed to init sysctl parameters.\n");
3491
3492#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3493 bpf_iter_register();
3494#endif
3495}