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/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/gso.h>
107#include <net/xfrm.h>
108#include <trace/events/udp.h>
109#include <linux/static_key.h>
110#include <linux/btf_ids.h>
111#include <trace/events/skb.h>
112#include <net/busy_poll.h>
113#include "udp_impl.h"
114#include <net/sock_reuseport.h>
115#include <net/addrconf.h>
116#include <net/udp_tunnel.h>
117#include <net/gro.h>
118#if IS_ENABLED(CONFIG_IPV6)
119#include <net/ipv6_stubs.h>
120#endif
121
122struct udp_table udp_table __read_mostly;
123EXPORT_SYMBOL(udp_table);
124
125long sysctl_udp_mem[3] __read_mostly;
126EXPORT_SYMBOL(sysctl_udp_mem);
127
128atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
129EXPORT_SYMBOL(udp_memory_allocated);
130DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
131EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
132
133#define MAX_UDP_PORTS 65536
134#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
135
136static struct udp_table *udp_get_table_prot(struct sock *sk)
137{
138 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
139}
140
141static int udp_lib_lport_inuse(struct net *net, __u16 num,
142 const struct udp_hslot *hslot,
143 unsigned long *bitmap,
144 struct sock *sk, unsigned int log)
145{
146 struct sock *sk2;
147 kuid_t uid = sock_i_uid(sk);
148
149 sk_for_each(sk2, &hslot->head) {
150 if (net_eq(sock_net(sk2), net) &&
151 sk2 != sk &&
152 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
153 (!sk2->sk_reuse || !sk->sk_reuse) &&
154 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
155 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
156 inet_rcv_saddr_equal(sk, sk2, true)) {
157 if (sk2->sk_reuseport && sk->sk_reuseport &&
158 !rcu_access_pointer(sk->sk_reuseport_cb) &&
159 uid_eq(uid, sock_i_uid(sk2))) {
160 if (!bitmap)
161 return 0;
162 } else {
163 if (!bitmap)
164 return 1;
165 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
166 bitmap);
167 }
168 }
169 }
170 return 0;
171}
172
173/*
174 * Note: we still hold spinlock of primary hash chain, so no other writer
175 * can insert/delete a socket with local_port == num
176 */
177static int udp_lib_lport_inuse2(struct net *net, __u16 num,
178 struct udp_hslot *hslot2,
179 struct sock *sk)
180{
181 struct sock *sk2;
182 kuid_t uid = sock_i_uid(sk);
183 int res = 0;
184
185 spin_lock(&hslot2->lock);
186 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
187 if (net_eq(sock_net(sk2), net) &&
188 sk2 != sk &&
189 (udp_sk(sk2)->udp_port_hash == num) &&
190 (!sk2->sk_reuse || !sk->sk_reuse) &&
191 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
192 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
193 inet_rcv_saddr_equal(sk, sk2, true)) {
194 if (sk2->sk_reuseport && sk->sk_reuseport &&
195 !rcu_access_pointer(sk->sk_reuseport_cb) &&
196 uid_eq(uid, sock_i_uid(sk2))) {
197 res = 0;
198 } else {
199 res = 1;
200 }
201 break;
202 }
203 }
204 spin_unlock(&hslot2->lock);
205 return res;
206}
207
208static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
209{
210 struct net *net = sock_net(sk);
211 kuid_t uid = sock_i_uid(sk);
212 struct sock *sk2;
213
214 sk_for_each(sk2, &hslot->head) {
215 if (net_eq(sock_net(sk2), net) &&
216 sk2 != sk &&
217 sk2->sk_family == sk->sk_family &&
218 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
219 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
220 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
221 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
222 inet_rcv_saddr_equal(sk, sk2, false)) {
223 return reuseport_add_sock(sk, sk2,
224 inet_rcv_saddr_any(sk));
225 }
226 }
227
228 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
229}
230
231/**
232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
233 *
234 * @sk: socket struct in question
235 * @snum: port number to look up
236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237 * with NULL address
238 */
239int udp_lib_get_port(struct sock *sk, unsigned short snum,
240 unsigned int hash2_nulladdr)
241{
242 struct udp_table *udptable = udp_get_table_prot(sk);
243 struct udp_hslot *hslot, *hslot2;
244 struct net *net = sock_net(sk);
245 int error = -EADDRINUSE;
246
247 if (!snum) {
248 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
249 unsigned short first, last;
250 int low, high, remaining;
251 unsigned int rand;
252
253 inet_sk_get_local_port_range(sk, &low, &high);
254 remaining = (high - low) + 1;
255
256 rand = get_random_u32();
257 first = reciprocal_scale(rand, remaining) + low;
258 /*
259 * force rand to be an odd multiple of UDP_HTABLE_SIZE
260 */
261 rand = (rand | 1) * (udptable->mask + 1);
262 last = first + udptable->mask + 1;
263 do {
264 hslot = udp_hashslot(udptable, net, first);
265 bitmap_zero(bitmap, PORTS_PER_CHAIN);
266 spin_lock_bh(&hslot->lock);
267 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
268 udptable->log);
269
270 snum = first;
271 /*
272 * Iterate on all possible values of snum for this hash.
273 * Using steps of an odd multiple of UDP_HTABLE_SIZE
274 * give us randomization and full range coverage.
275 */
276 do {
277 if (low <= snum && snum <= high &&
278 !test_bit(snum >> udptable->log, bitmap) &&
279 !inet_is_local_reserved_port(net, snum))
280 goto found;
281 snum += rand;
282 } while (snum != first);
283 spin_unlock_bh(&hslot->lock);
284 cond_resched();
285 } while (++first != last);
286 goto fail;
287 } else {
288 hslot = udp_hashslot(udptable, net, snum);
289 spin_lock_bh(&hslot->lock);
290 if (hslot->count > 10) {
291 int exist;
292 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
293
294 slot2 &= udptable->mask;
295 hash2_nulladdr &= udptable->mask;
296
297 hslot2 = udp_hashslot2(udptable, slot2);
298 if (hslot->count < hslot2->count)
299 goto scan_primary_hash;
300
301 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
302 if (!exist && (hash2_nulladdr != slot2)) {
303 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
304 exist = udp_lib_lport_inuse2(net, snum, hslot2,
305 sk);
306 }
307 if (exist)
308 goto fail_unlock;
309 else
310 goto found;
311 }
312scan_primary_hash:
313 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
314 goto fail_unlock;
315 }
316found:
317 inet_sk(sk)->inet_num = snum;
318 udp_sk(sk)->udp_port_hash = snum;
319 udp_sk(sk)->udp_portaddr_hash ^= snum;
320 if (sk_unhashed(sk)) {
321 if (sk->sk_reuseport &&
322 udp_reuseport_add_sock(sk, hslot)) {
323 inet_sk(sk)->inet_num = 0;
324 udp_sk(sk)->udp_port_hash = 0;
325 udp_sk(sk)->udp_portaddr_hash ^= snum;
326 goto fail_unlock;
327 }
328
329 sk_add_node_rcu(sk, &hslot->head);
330 hslot->count++;
331 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
332
333 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
334 spin_lock(&hslot2->lock);
335 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
336 sk->sk_family == AF_INET6)
337 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
338 &hslot2->head);
339 else
340 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
341 &hslot2->head);
342 hslot2->count++;
343 spin_unlock(&hslot2->lock);
344 }
345 sock_set_flag(sk, SOCK_RCU_FREE);
346 error = 0;
347fail_unlock:
348 spin_unlock_bh(&hslot->lock);
349fail:
350 return error;
351}
352EXPORT_SYMBOL(udp_lib_get_port);
353
354int udp_v4_get_port(struct sock *sk, unsigned short snum)
355{
356 unsigned int hash2_nulladdr =
357 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
358 unsigned int hash2_partial =
359 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
360
361 /* precompute partial secondary hash */
362 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
363 return udp_lib_get_port(sk, snum, hash2_nulladdr);
364}
365
366static int compute_score(struct sock *sk, struct net *net,
367 __be32 saddr, __be16 sport,
368 __be32 daddr, unsigned short hnum,
369 int dif, int sdif)
370{
371 int score;
372 struct inet_sock *inet;
373 bool dev_match;
374
375 if (!net_eq(sock_net(sk), net) ||
376 udp_sk(sk)->udp_port_hash != hnum ||
377 ipv6_only_sock(sk))
378 return -1;
379
380 if (sk->sk_rcv_saddr != daddr)
381 return -1;
382
383 score = (sk->sk_family == PF_INET) ? 2 : 1;
384
385 inet = inet_sk(sk);
386 if (inet->inet_daddr) {
387 if (inet->inet_daddr != saddr)
388 return -1;
389 score += 4;
390 }
391
392 if (inet->inet_dport) {
393 if (inet->inet_dport != sport)
394 return -1;
395 score += 4;
396 }
397
398 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
399 dif, sdif);
400 if (!dev_match)
401 return -1;
402 if (sk->sk_bound_dev_if)
403 score += 4;
404
405 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
406 score++;
407 return score;
408}
409
410INDIRECT_CALLABLE_SCOPE
411u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
412 const __be32 faddr, const __be16 fport)
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
420/* called with rcu_read_lock() */
421static struct sock *udp4_lib_lookup2(struct net *net,
422 __be32 saddr, __be16 sport,
423 __be32 daddr, unsigned int hnum,
424 int dif, int sdif,
425 struct udp_hslot *hslot2,
426 struct sk_buff *skb)
427{
428 struct sock *sk, *result;
429 int score, badness;
430 bool need_rescore;
431
432 result = NULL;
433 badness = 0;
434 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
435 need_rescore = false;
436rescore:
437 score = compute_score(need_rescore ? result : sk, net, saddr,
438 sport, daddr, hnum, dif, sdif);
439 if (score > badness) {
440 badness = score;
441
442 if (need_rescore)
443 continue;
444
445 if (sk->sk_state == TCP_ESTABLISHED) {
446 result = sk;
447 continue;
448 }
449
450 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
451 saddr, sport, daddr, hnum, udp_ehashfn);
452 if (!result) {
453 result = sk;
454 continue;
455 }
456
457 /* Fall back to scoring if group has connections */
458 if (!reuseport_has_conns(sk))
459 return result;
460
461 /* Reuseport logic returned an error, keep original score. */
462 if (IS_ERR(result))
463 continue;
464
465 /* compute_score is too long of a function to be
466 * inlined, and calling it again here yields
467 * measureable overhead for some
468 * workloads. Work around it by jumping
469 * backwards to rescore 'result'.
470 */
471 need_rescore = true;
472 goto rescore;
473 }
474 }
475 return result;
476}
477
478/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
479 * harder than this. -DaveM
480 */
481struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
482 __be16 sport, __be32 daddr, __be16 dport, int dif,
483 int sdif, struct udp_table *udptable, struct sk_buff *skb)
484{
485 unsigned short hnum = ntohs(dport);
486 unsigned int hash2, slot2;
487 struct udp_hslot *hslot2;
488 struct sock *result, *sk;
489
490 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
491 slot2 = hash2 & udptable->mask;
492 hslot2 = &udptable->hash2[slot2];
493
494 /* Lookup connected or non-wildcard socket */
495 result = udp4_lib_lookup2(net, saddr, sport,
496 daddr, hnum, dif, sdif,
497 hslot2, skb);
498 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
499 goto done;
500
501 /* Lookup redirect from BPF */
502 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
503 udptable == net->ipv4.udp_table) {
504 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
505 saddr, sport, daddr, hnum, dif,
506 udp_ehashfn);
507 if (sk) {
508 result = sk;
509 goto done;
510 }
511 }
512
513 /* Got non-wildcard socket or error on first lookup */
514 if (result)
515 goto done;
516
517 /* Lookup wildcard sockets */
518 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
519 slot2 = hash2 & udptable->mask;
520 hslot2 = &udptable->hash2[slot2];
521
522 result = udp4_lib_lookup2(net, saddr, sport,
523 htonl(INADDR_ANY), hnum, dif, sdif,
524 hslot2, skb);
525done:
526 if (IS_ERR(result))
527 return NULL;
528 return result;
529}
530EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
531
532static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
533 __be16 sport, __be16 dport,
534 struct udp_table *udptable)
535{
536 const struct iphdr *iph = ip_hdr(skb);
537
538 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
539 iph->daddr, dport, inet_iif(skb),
540 inet_sdif(skb), udptable, skb);
541}
542
543struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
544 __be16 sport, __be16 dport)
545{
546 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
547 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
548 struct net *net = dev_net(skb->dev);
549 int iif, sdif;
550
551 inet_get_iif_sdif(skb, &iif, &sdif);
552
553 return __udp4_lib_lookup(net, iph->saddr, sport,
554 iph->daddr, dport, iif,
555 sdif, net->ipv4.udp_table, NULL);
556}
557
558/* Must be called under rcu_read_lock().
559 * Does increment socket refcount.
560 */
561#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
562struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
563 __be32 daddr, __be16 dport, int dif)
564{
565 struct sock *sk;
566
567 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
568 dif, 0, net->ipv4.udp_table, NULL);
569 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
570 sk = NULL;
571 return sk;
572}
573EXPORT_SYMBOL_GPL(udp4_lib_lookup);
574#endif
575
576static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
577 __be16 loc_port, __be32 loc_addr,
578 __be16 rmt_port, __be32 rmt_addr,
579 int dif, int sdif, unsigned short hnum)
580{
581 const struct inet_sock *inet = inet_sk(sk);
582
583 if (!net_eq(sock_net(sk), net) ||
584 udp_sk(sk)->udp_port_hash != hnum ||
585 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
586 (inet->inet_dport != rmt_port && inet->inet_dport) ||
587 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
588 ipv6_only_sock(sk) ||
589 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
590 return false;
591 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
592 return false;
593 return true;
594}
595
596DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
597EXPORT_SYMBOL(udp_encap_needed_key);
598
599#if IS_ENABLED(CONFIG_IPV6)
600DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
601EXPORT_SYMBOL(udpv6_encap_needed_key);
602#endif
603
604void udp_encap_enable(void)
605{
606 static_branch_inc(&udp_encap_needed_key);
607}
608EXPORT_SYMBOL(udp_encap_enable);
609
610void udp_encap_disable(void)
611{
612 static_branch_dec(&udp_encap_needed_key);
613}
614EXPORT_SYMBOL(udp_encap_disable);
615
616/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
617 * through error handlers in encapsulations looking for a match.
618 */
619static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
620{
621 int i;
622
623 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
624 int (*handler)(struct sk_buff *skb, u32 info);
625 const struct ip_tunnel_encap_ops *encap;
626
627 encap = rcu_dereference(iptun_encaps[i]);
628 if (!encap)
629 continue;
630 handler = encap->err_handler;
631 if (handler && !handler(skb, info))
632 return 0;
633 }
634
635 return -ENOENT;
636}
637
638/* Try to match ICMP errors to UDP tunnels by looking up a socket without
639 * reversing source and destination port: this will match tunnels that force the
640 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
641 * lwtunnels might actually break this assumption by being configured with
642 * different destination ports on endpoints, in this case we won't be able to
643 * trace ICMP messages back to them.
644 *
645 * If this doesn't match any socket, probe tunnels with arbitrary destination
646 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
647 * we've sent packets to won't necessarily match the local destination port.
648 *
649 * Then ask the tunnel implementation to match the error against a valid
650 * association.
651 *
652 * Return an error if we can't find a match, the socket if we need further
653 * processing, zero otherwise.
654 */
655static struct sock *__udp4_lib_err_encap(struct net *net,
656 const struct iphdr *iph,
657 struct udphdr *uh,
658 struct udp_table *udptable,
659 struct sock *sk,
660 struct sk_buff *skb, u32 info)
661{
662 int (*lookup)(struct sock *sk, struct sk_buff *skb);
663 int network_offset, transport_offset;
664 struct udp_sock *up;
665
666 network_offset = skb_network_offset(skb);
667 transport_offset = skb_transport_offset(skb);
668
669 /* Network header needs to point to the outer IPv4 header inside ICMP */
670 skb_reset_network_header(skb);
671
672 /* Transport header needs to point to the UDP header */
673 skb_set_transport_header(skb, iph->ihl << 2);
674
675 if (sk) {
676 up = udp_sk(sk);
677
678 lookup = READ_ONCE(up->encap_err_lookup);
679 if (lookup && lookup(sk, skb))
680 sk = NULL;
681
682 goto out;
683 }
684
685 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
686 iph->saddr, uh->dest, skb->dev->ifindex, 0,
687 udptable, NULL);
688 if (sk) {
689 up = udp_sk(sk);
690
691 lookup = READ_ONCE(up->encap_err_lookup);
692 if (!lookup || lookup(sk, skb))
693 sk = NULL;
694 }
695
696out:
697 if (!sk)
698 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
699
700 skb_set_transport_header(skb, transport_offset);
701 skb_set_network_header(skb, network_offset);
702
703 return sk;
704}
705
706/*
707 * This routine is called by the ICMP module when it gets some
708 * sort of error condition. If err < 0 then the socket should
709 * be closed and the error returned to the user. If err > 0
710 * it's just the icmp type << 8 | icmp code.
711 * Header points to the ip header of the error packet. We move
712 * on past this. Then (as it used to claim before adjustment)
713 * header points to the first 8 bytes of the udp header. We need
714 * to find the appropriate port.
715 */
716
717int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
718{
719 struct inet_sock *inet;
720 const struct iphdr *iph = (const struct iphdr *)skb->data;
721 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
722 const int type = icmp_hdr(skb)->type;
723 const int code = icmp_hdr(skb)->code;
724 bool tunnel = false;
725 struct sock *sk;
726 int harderr;
727 int err;
728 struct net *net = dev_net(skb->dev);
729
730 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
731 iph->saddr, uh->source, skb->dev->ifindex,
732 inet_sdif(skb), udptable, NULL);
733
734 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
735 /* No socket for error: try tunnels before discarding */
736 if (static_branch_unlikely(&udp_encap_needed_key)) {
737 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
738 info);
739 if (!sk)
740 return 0;
741 } else
742 sk = ERR_PTR(-ENOENT);
743
744 if (IS_ERR(sk)) {
745 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
746 return PTR_ERR(sk);
747 }
748
749 tunnel = true;
750 }
751
752 err = 0;
753 harderr = 0;
754 inet = inet_sk(sk);
755
756 switch (type) {
757 default:
758 case ICMP_TIME_EXCEEDED:
759 err = EHOSTUNREACH;
760 break;
761 case ICMP_SOURCE_QUENCH:
762 goto out;
763 case ICMP_PARAMETERPROB:
764 err = EPROTO;
765 harderr = 1;
766 break;
767 case ICMP_DEST_UNREACH:
768 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
769 ipv4_sk_update_pmtu(skb, sk, info);
770 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
771 err = EMSGSIZE;
772 harderr = 1;
773 break;
774 }
775 goto out;
776 }
777 err = EHOSTUNREACH;
778 if (code <= NR_ICMP_UNREACH) {
779 harderr = icmp_err_convert[code].fatal;
780 err = icmp_err_convert[code].errno;
781 }
782 break;
783 case ICMP_REDIRECT:
784 ipv4_sk_redirect(skb, sk);
785 goto out;
786 }
787
788 /*
789 * RFC1122: OK. Passes ICMP errors back to application, as per
790 * 4.1.3.3.
791 */
792 if (tunnel) {
793 /* ...not for tunnels though: we don't have a sending socket */
794 if (udp_sk(sk)->encap_err_rcv)
795 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
796 (u8 *)(uh+1));
797 goto out;
798 }
799 if (!inet_test_bit(RECVERR, sk)) {
800 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
801 goto out;
802 } else
803 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
804
805 sk->sk_err = err;
806 sk_error_report(sk);
807out:
808 return 0;
809}
810
811int udp_err(struct sk_buff *skb, u32 info)
812{
813 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
814}
815
816/*
817 * Throw away all pending data and cancel the corking. Socket is locked.
818 */
819void udp_flush_pending_frames(struct sock *sk)
820{
821 struct udp_sock *up = udp_sk(sk);
822
823 if (up->pending) {
824 up->len = 0;
825 WRITE_ONCE(up->pending, 0);
826 ip_flush_pending_frames(sk);
827 }
828}
829EXPORT_SYMBOL(udp_flush_pending_frames);
830
831/**
832 * udp4_hwcsum - handle outgoing HW checksumming
833 * @skb: sk_buff containing the filled-in UDP header
834 * (checksum field must be zeroed out)
835 * @src: source IP address
836 * @dst: destination IP address
837 */
838void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
839{
840 struct udphdr *uh = udp_hdr(skb);
841 int offset = skb_transport_offset(skb);
842 int len = skb->len - offset;
843 int hlen = len;
844 __wsum csum = 0;
845
846 if (!skb_has_frag_list(skb)) {
847 /*
848 * Only one fragment on the socket.
849 */
850 skb->csum_start = skb_transport_header(skb) - skb->head;
851 skb->csum_offset = offsetof(struct udphdr, check);
852 uh->check = ~csum_tcpudp_magic(src, dst, len,
853 IPPROTO_UDP, 0);
854 } else {
855 struct sk_buff *frags;
856
857 /*
858 * HW-checksum won't work as there are two or more
859 * fragments on the socket so that all csums of sk_buffs
860 * should be together
861 */
862 skb_walk_frags(skb, frags) {
863 csum = csum_add(csum, frags->csum);
864 hlen -= frags->len;
865 }
866
867 csum = skb_checksum(skb, offset, hlen, csum);
868 skb->ip_summed = CHECKSUM_NONE;
869
870 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
871 if (uh->check == 0)
872 uh->check = CSUM_MANGLED_0;
873 }
874}
875EXPORT_SYMBOL_GPL(udp4_hwcsum);
876
877/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
878 * for the simple case like when setting the checksum for a UDP tunnel.
879 */
880void udp_set_csum(bool nocheck, struct sk_buff *skb,
881 __be32 saddr, __be32 daddr, int len)
882{
883 struct udphdr *uh = udp_hdr(skb);
884
885 if (nocheck) {
886 uh->check = 0;
887 } else if (skb_is_gso(skb)) {
888 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
889 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
890 uh->check = 0;
891 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
892 if (uh->check == 0)
893 uh->check = CSUM_MANGLED_0;
894 } else {
895 skb->ip_summed = CHECKSUM_PARTIAL;
896 skb->csum_start = skb_transport_header(skb) - skb->head;
897 skb->csum_offset = offsetof(struct udphdr, check);
898 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
899 }
900}
901EXPORT_SYMBOL(udp_set_csum);
902
903static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
904 struct inet_cork *cork)
905{
906 struct sock *sk = skb->sk;
907 struct inet_sock *inet = inet_sk(sk);
908 struct udphdr *uh;
909 int err;
910 int is_udplite = IS_UDPLITE(sk);
911 int offset = skb_transport_offset(skb);
912 int len = skb->len - offset;
913 int datalen = len - sizeof(*uh);
914 __wsum csum = 0;
915
916 /*
917 * Create a UDP header
918 */
919 uh = udp_hdr(skb);
920 uh->source = inet->inet_sport;
921 uh->dest = fl4->fl4_dport;
922 uh->len = htons(len);
923 uh->check = 0;
924
925 if (cork->gso_size) {
926 const int hlen = skb_network_header_len(skb) +
927 sizeof(struct udphdr);
928
929 if (hlen + cork->gso_size > cork->fragsize) {
930 kfree_skb(skb);
931 return -EINVAL;
932 }
933 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
934 kfree_skb(skb);
935 return -EINVAL;
936 }
937 if (sk->sk_no_check_tx) {
938 kfree_skb(skb);
939 return -EINVAL;
940 }
941 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
942 dst_xfrm(skb_dst(skb))) {
943 kfree_skb(skb);
944 return -EIO;
945 }
946
947 if (datalen > cork->gso_size) {
948 skb_shinfo(skb)->gso_size = cork->gso_size;
949 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
950 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
951 cork->gso_size);
952 }
953 goto csum_partial;
954 }
955
956 if (is_udplite) /* UDP-Lite */
957 csum = udplite_csum(skb);
958
959 else if (sk->sk_no_check_tx) { /* UDP csum off */
960
961 skb->ip_summed = CHECKSUM_NONE;
962 goto send;
963
964 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
965csum_partial:
966
967 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
968 goto send;
969
970 } else
971 csum = udp_csum(skb);
972
973 /* add protocol-dependent pseudo-header */
974 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
975 sk->sk_protocol, csum);
976 if (uh->check == 0)
977 uh->check = CSUM_MANGLED_0;
978
979send:
980 err = ip_send_skb(sock_net(sk), skb);
981 if (err) {
982 if (err == -ENOBUFS &&
983 !inet_test_bit(RECVERR, sk)) {
984 UDP_INC_STATS(sock_net(sk),
985 UDP_MIB_SNDBUFERRORS, is_udplite);
986 err = 0;
987 }
988 } else
989 UDP_INC_STATS(sock_net(sk),
990 UDP_MIB_OUTDATAGRAMS, is_udplite);
991 return err;
992}
993
994/*
995 * Push out all pending data as one UDP datagram. Socket is locked.
996 */
997int udp_push_pending_frames(struct sock *sk)
998{
999 struct udp_sock *up = udp_sk(sk);
1000 struct inet_sock *inet = inet_sk(sk);
1001 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1002 struct sk_buff *skb;
1003 int err = 0;
1004
1005 skb = ip_finish_skb(sk, fl4);
1006 if (!skb)
1007 goto out;
1008
1009 err = udp_send_skb(skb, fl4, &inet->cork.base);
1010
1011out:
1012 up->len = 0;
1013 WRITE_ONCE(up->pending, 0);
1014 return err;
1015}
1016EXPORT_SYMBOL(udp_push_pending_frames);
1017
1018static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1019{
1020 switch (cmsg->cmsg_type) {
1021 case UDP_SEGMENT:
1022 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1023 return -EINVAL;
1024 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1025 return 0;
1026 default:
1027 return -EINVAL;
1028 }
1029}
1030
1031int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1032{
1033 struct cmsghdr *cmsg;
1034 bool need_ip = false;
1035 int err;
1036
1037 for_each_cmsghdr(cmsg, msg) {
1038 if (!CMSG_OK(msg, cmsg))
1039 return -EINVAL;
1040
1041 if (cmsg->cmsg_level != SOL_UDP) {
1042 need_ip = true;
1043 continue;
1044 }
1045
1046 err = __udp_cmsg_send(cmsg, gso_size);
1047 if (err)
1048 return err;
1049 }
1050
1051 return need_ip;
1052}
1053EXPORT_SYMBOL_GPL(udp_cmsg_send);
1054
1055int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1056{
1057 struct inet_sock *inet = inet_sk(sk);
1058 struct udp_sock *up = udp_sk(sk);
1059 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1060 struct flowi4 fl4_stack;
1061 struct flowi4 *fl4;
1062 int ulen = len;
1063 struct ipcm_cookie ipc;
1064 struct rtable *rt = NULL;
1065 int free = 0;
1066 int connected = 0;
1067 __be32 daddr, faddr, saddr;
1068 u8 tos, scope;
1069 __be16 dport;
1070 int err, is_udplite = IS_UDPLITE(sk);
1071 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1072 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1073 struct sk_buff *skb;
1074 struct ip_options_data opt_copy;
1075 int uc_index;
1076
1077 if (len > 0xFFFF)
1078 return -EMSGSIZE;
1079
1080 /*
1081 * Check the flags.
1082 */
1083
1084 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1085 return -EOPNOTSUPP;
1086
1087 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1088
1089 fl4 = &inet->cork.fl.u.ip4;
1090 if (READ_ONCE(up->pending)) {
1091 /*
1092 * There are pending frames.
1093 * The socket lock must be held while it's corked.
1094 */
1095 lock_sock(sk);
1096 if (likely(up->pending)) {
1097 if (unlikely(up->pending != AF_INET)) {
1098 release_sock(sk);
1099 return -EINVAL;
1100 }
1101 goto do_append_data;
1102 }
1103 release_sock(sk);
1104 }
1105 ulen += sizeof(struct udphdr);
1106
1107 /*
1108 * Get and verify the address.
1109 */
1110 if (usin) {
1111 if (msg->msg_namelen < sizeof(*usin))
1112 return -EINVAL;
1113 if (usin->sin_family != AF_INET) {
1114 if (usin->sin_family != AF_UNSPEC)
1115 return -EAFNOSUPPORT;
1116 }
1117
1118 daddr = usin->sin_addr.s_addr;
1119 dport = usin->sin_port;
1120 if (dport == 0)
1121 return -EINVAL;
1122 } else {
1123 if (sk->sk_state != TCP_ESTABLISHED)
1124 return -EDESTADDRREQ;
1125 daddr = inet->inet_daddr;
1126 dport = inet->inet_dport;
1127 /* Open fast path for connected socket.
1128 Route will not be used, if at least one option is set.
1129 */
1130 connected = 1;
1131 }
1132
1133 ipcm_init_sk(&ipc, inet);
1134 ipc.gso_size = READ_ONCE(up->gso_size);
1135
1136 if (msg->msg_controllen) {
1137 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1138 if (err > 0) {
1139 err = ip_cmsg_send(sk, msg, &ipc,
1140 sk->sk_family == AF_INET6);
1141 connected = 0;
1142 }
1143 if (unlikely(err < 0)) {
1144 kfree(ipc.opt);
1145 return err;
1146 }
1147 if (ipc.opt)
1148 free = 1;
1149 }
1150 if (!ipc.opt) {
1151 struct ip_options_rcu *inet_opt;
1152
1153 rcu_read_lock();
1154 inet_opt = rcu_dereference(inet->inet_opt);
1155 if (inet_opt) {
1156 memcpy(&opt_copy, inet_opt,
1157 sizeof(*inet_opt) + inet_opt->opt.optlen);
1158 ipc.opt = &opt_copy.opt;
1159 }
1160 rcu_read_unlock();
1161 }
1162
1163 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1164 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1165 (struct sockaddr *)usin,
1166 &msg->msg_namelen,
1167 &ipc.addr);
1168 if (err)
1169 goto out_free;
1170 if (usin) {
1171 if (usin->sin_port == 0) {
1172 /* BPF program set invalid port. Reject it. */
1173 err = -EINVAL;
1174 goto out_free;
1175 }
1176 daddr = usin->sin_addr.s_addr;
1177 dport = usin->sin_port;
1178 }
1179 }
1180
1181 saddr = ipc.addr;
1182 ipc.addr = faddr = daddr;
1183
1184 if (ipc.opt && ipc.opt->opt.srr) {
1185 if (!daddr) {
1186 err = -EINVAL;
1187 goto out_free;
1188 }
1189 faddr = ipc.opt->opt.faddr;
1190 connected = 0;
1191 }
1192 tos = get_rttos(&ipc, inet);
1193 scope = ip_sendmsg_scope(inet, &ipc, msg);
1194 if (scope == RT_SCOPE_LINK)
1195 connected = 0;
1196
1197 uc_index = READ_ONCE(inet->uc_index);
1198 if (ipv4_is_multicast(daddr)) {
1199 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1200 ipc.oif = READ_ONCE(inet->mc_index);
1201 if (!saddr)
1202 saddr = READ_ONCE(inet->mc_addr);
1203 connected = 0;
1204 } else if (!ipc.oif) {
1205 ipc.oif = uc_index;
1206 } else if (ipv4_is_lbcast(daddr) && uc_index) {
1207 /* oif is set, packet is to local broadcast and
1208 * uc_index is set. oif is most likely set
1209 * by sk_bound_dev_if. If uc_index != oif check if the
1210 * oif is an L3 master and uc_index is an L3 slave.
1211 * If so, we want to allow the send using the uc_index.
1212 */
1213 if (ipc.oif != uc_index &&
1214 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1215 uc_index)) {
1216 ipc.oif = uc_index;
1217 }
1218 }
1219
1220 if (connected)
1221 rt = dst_rtable(sk_dst_check(sk, 0));
1222
1223 if (!rt) {
1224 struct net *net = sock_net(sk);
1225 __u8 flow_flags = inet_sk_flowi_flags(sk);
1226
1227 fl4 = &fl4_stack;
1228
1229 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1230 sk->sk_protocol, flow_flags, faddr, saddr,
1231 dport, inet->inet_sport, sk->sk_uid);
1232
1233 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1234 rt = ip_route_output_flow(net, fl4, sk);
1235 if (IS_ERR(rt)) {
1236 err = PTR_ERR(rt);
1237 rt = NULL;
1238 if (err == -ENETUNREACH)
1239 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1240 goto out;
1241 }
1242
1243 err = -EACCES;
1244 if ((rt->rt_flags & RTCF_BROADCAST) &&
1245 !sock_flag(sk, SOCK_BROADCAST))
1246 goto out;
1247 if (connected)
1248 sk_dst_set(sk, dst_clone(&rt->dst));
1249 }
1250
1251 if (msg->msg_flags&MSG_CONFIRM)
1252 goto do_confirm;
1253back_from_confirm:
1254
1255 saddr = fl4->saddr;
1256 if (!ipc.addr)
1257 daddr = ipc.addr = fl4->daddr;
1258
1259 /* Lockless fast path for the non-corking case. */
1260 if (!corkreq) {
1261 struct inet_cork cork;
1262
1263 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1264 sizeof(struct udphdr), &ipc, &rt,
1265 &cork, msg->msg_flags);
1266 err = PTR_ERR(skb);
1267 if (!IS_ERR_OR_NULL(skb))
1268 err = udp_send_skb(skb, fl4, &cork);
1269 goto out;
1270 }
1271
1272 lock_sock(sk);
1273 if (unlikely(up->pending)) {
1274 /* The socket is already corked while preparing it. */
1275 /* ... which is an evident application bug. --ANK */
1276 release_sock(sk);
1277
1278 net_dbg_ratelimited("socket already corked\n");
1279 err = -EINVAL;
1280 goto out;
1281 }
1282 /*
1283 * Now cork the socket to pend data.
1284 */
1285 fl4 = &inet->cork.fl.u.ip4;
1286 fl4->daddr = daddr;
1287 fl4->saddr = saddr;
1288 fl4->fl4_dport = dport;
1289 fl4->fl4_sport = inet->inet_sport;
1290 WRITE_ONCE(up->pending, AF_INET);
1291
1292do_append_data:
1293 up->len += ulen;
1294 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1295 sizeof(struct udphdr), &ipc, &rt,
1296 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1297 if (err)
1298 udp_flush_pending_frames(sk);
1299 else if (!corkreq)
1300 err = udp_push_pending_frames(sk);
1301 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1302 WRITE_ONCE(up->pending, 0);
1303 release_sock(sk);
1304
1305out:
1306 ip_rt_put(rt);
1307out_free:
1308 if (free)
1309 kfree(ipc.opt);
1310 if (!err)
1311 return len;
1312 /*
1313 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1314 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1315 * we don't have a good statistic (IpOutDiscards but it can be too many
1316 * things). We could add another new stat but at least for now that
1317 * seems like overkill.
1318 */
1319 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1320 UDP_INC_STATS(sock_net(sk),
1321 UDP_MIB_SNDBUFERRORS, is_udplite);
1322 }
1323 return err;
1324
1325do_confirm:
1326 if (msg->msg_flags & MSG_PROBE)
1327 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1328 if (!(msg->msg_flags&MSG_PROBE) || len)
1329 goto back_from_confirm;
1330 err = 0;
1331 goto out;
1332}
1333EXPORT_SYMBOL(udp_sendmsg);
1334
1335void udp_splice_eof(struct socket *sock)
1336{
1337 struct sock *sk = sock->sk;
1338 struct udp_sock *up = udp_sk(sk);
1339
1340 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1341 return;
1342
1343 lock_sock(sk);
1344 if (up->pending && !udp_test_bit(CORK, sk))
1345 udp_push_pending_frames(sk);
1346 release_sock(sk);
1347}
1348EXPORT_SYMBOL_GPL(udp_splice_eof);
1349
1350#define UDP_SKB_IS_STATELESS 0x80000000
1351
1352/* all head states (dst, sk, nf conntrack) except skb extensions are
1353 * cleared by udp_rcv().
1354 *
1355 * We need to preserve secpath, if present, to eventually process
1356 * IP_CMSG_PASSSEC at recvmsg() time.
1357 *
1358 * Other extensions can be cleared.
1359 */
1360static bool udp_try_make_stateless(struct sk_buff *skb)
1361{
1362 if (!skb_has_extensions(skb))
1363 return true;
1364
1365 if (!secpath_exists(skb)) {
1366 skb_ext_reset(skb);
1367 return true;
1368 }
1369
1370 return false;
1371}
1372
1373static void udp_set_dev_scratch(struct sk_buff *skb)
1374{
1375 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1376
1377 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1378 scratch->_tsize_state = skb->truesize;
1379#if BITS_PER_LONG == 64
1380 scratch->len = skb->len;
1381 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1382 scratch->is_linear = !skb_is_nonlinear(skb);
1383#endif
1384 if (udp_try_make_stateless(skb))
1385 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1386}
1387
1388static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1389{
1390 /* We come here after udp_lib_checksum_complete() returned 0.
1391 * This means that __skb_checksum_complete() might have
1392 * set skb->csum_valid to 1.
1393 * On 64bit platforms, we can set csum_unnecessary
1394 * to true, but only if the skb is not shared.
1395 */
1396#if BITS_PER_LONG == 64
1397 if (!skb_shared(skb))
1398 udp_skb_scratch(skb)->csum_unnecessary = true;
1399#endif
1400}
1401
1402static int udp_skb_truesize(struct sk_buff *skb)
1403{
1404 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1405}
1406
1407static bool udp_skb_has_head_state(struct sk_buff *skb)
1408{
1409 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1410}
1411
1412/* fully reclaim rmem/fwd memory allocated for skb */
1413static void udp_rmem_release(struct sock *sk, int size, int partial,
1414 bool rx_queue_lock_held)
1415{
1416 struct udp_sock *up = udp_sk(sk);
1417 struct sk_buff_head *sk_queue;
1418 int amt;
1419
1420 if (likely(partial)) {
1421 up->forward_deficit += size;
1422 size = up->forward_deficit;
1423 if (size < READ_ONCE(up->forward_threshold) &&
1424 !skb_queue_empty(&up->reader_queue))
1425 return;
1426 } else {
1427 size += up->forward_deficit;
1428 }
1429 up->forward_deficit = 0;
1430
1431 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1432 * if the called don't held it already
1433 */
1434 sk_queue = &sk->sk_receive_queue;
1435 if (!rx_queue_lock_held)
1436 spin_lock(&sk_queue->lock);
1437
1438
1439 sk_forward_alloc_add(sk, size);
1440 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1441 sk_forward_alloc_add(sk, -amt);
1442
1443 if (amt)
1444 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1445
1446 atomic_sub(size, &sk->sk_rmem_alloc);
1447
1448 /* this can save us from acquiring the rx queue lock on next receive */
1449 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1450
1451 if (!rx_queue_lock_held)
1452 spin_unlock(&sk_queue->lock);
1453}
1454
1455/* Note: called with reader_queue.lock held.
1456 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1457 * This avoids a cache line miss while receive_queue lock is held.
1458 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1459 */
1460void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1461{
1462 prefetch(&skb->data);
1463 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1464}
1465EXPORT_SYMBOL(udp_skb_destructor);
1466
1467/* as above, but the caller held the rx queue lock, too */
1468static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1469{
1470 prefetch(&skb->data);
1471 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1472}
1473
1474/* Idea of busylocks is to let producers grab an extra spinlock
1475 * to relieve pressure on the receive_queue spinlock shared by consumer.
1476 * Under flood, this means that only one producer can be in line
1477 * trying to acquire the receive_queue spinlock.
1478 * These busylock can be allocated on a per cpu manner, instead of a
1479 * per socket one (that would consume a cache line per socket)
1480 */
1481static int udp_busylocks_log __read_mostly;
1482static spinlock_t *udp_busylocks __read_mostly;
1483
1484static spinlock_t *busylock_acquire(void *ptr)
1485{
1486 spinlock_t *busy;
1487
1488 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1489 spin_lock(busy);
1490 return busy;
1491}
1492
1493static void busylock_release(spinlock_t *busy)
1494{
1495 if (busy)
1496 spin_unlock(busy);
1497}
1498
1499static int udp_rmem_schedule(struct sock *sk, int size)
1500{
1501 int delta;
1502
1503 delta = size - sk->sk_forward_alloc;
1504 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1505 return -ENOBUFS;
1506
1507 return 0;
1508}
1509
1510int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1511{
1512 struct sk_buff_head *list = &sk->sk_receive_queue;
1513 int rmem, err = -ENOMEM;
1514 spinlock_t *busy = NULL;
1515 int size;
1516
1517 /* try to avoid the costly atomic add/sub pair when the receive
1518 * queue is full; always allow at least a packet
1519 */
1520 rmem = atomic_read(&sk->sk_rmem_alloc);
1521 if (rmem > sk->sk_rcvbuf)
1522 goto drop;
1523
1524 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1525 * having linear skbs :
1526 * - Reduce memory overhead and thus increase receive queue capacity
1527 * - Less cache line misses at copyout() time
1528 * - Less work at consume_skb() (less alien page frag freeing)
1529 */
1530 if (rmem > (sk->sk_rcvbuf >> 1)) {
1531 skb_condense(skb);
1532
1533 busy = busylock_acquire(sk);
1534 }
1535 size = skb->truesize;
1536 udp_set_dev_scratch(skb);
1537
1538 /* we drop only if the receive buf is full and the receive
1539 * queue contains some other skb
1540 */
1541 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1542 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1543 goto uncharge_drop;
1544
1545 spin_lock(&list->lock);
1546 err = udp_rmem_schedule(sk, size);
1547 if (err) {
1548 spin_unlock(&list->lock);
1549 goto uncharge_drop;
1550 }
1551
1552 sk_forward_alloc_add(sk, -size);
1553
1554 /* no need to setup a destructor, we will explicitly release the
1555 * forward allocated memory on dequeue
1556 */
1557 sock_skb_set_dropcount(sk, skb);
1558
1559 __skb_queue_tail(list, skb);
1560 spin_unlock(&list->lock);
1561
1562 if (!sock_flag(sk, SOCK_DEAD))
1563 INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
1564
1565 busylock_release(busy);
1566 return 0;
1567
1568uncharge_drop:
1569 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1570
1571drop:
1572 atomic_inc(&sk->sk_drops);
1573 busylock_release(busy);
1574 return err;
1575}
1576EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1577
1578void udp_destruct_common(struct sock *sk)
1579{
1580 /* reclaim completely the forward allocated memory */
1581 struct udp_sock *up = udp_sk(sk);
1582 unsigned int total = 0;
1583 struct sk_buff *skb;
1584
1585 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1586 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1587 total += skb->truesize;
1588 kfree_skb(skb);
1589 }
1590 udp_rmem_release(sk, total, 0, true);
1591}
1592EXPORT_SYMBOL_GPL(udp_destruct_common);
1593
1594static void udp_destruct_sock(struct sock *sk)
1595{
1596 udp_destruct_common(sk);
1597 inet_sock_destruct(sk);
1598}
1599
1600int udp_init_sock(struct sock *sk)
1601{
1602 udp_lib_init_sock(sk);
1603 sk->sk_destruct = udp_destruct_sock;
1604 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1605 return 0;
1606}
1607
1608void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1609{
1610 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
1611 sk_peek_offset_bwd(sk, len);
1612
1613 if (!skb_unref(skb))
1614 return;
1615
1616 /* In the more common cases we cleared the head states previously,
1617 * see __udp_queue_rcv_skb().
1618 */
1619 if (unlikely(udp_skb_has_head_state(skb)))
1620 skb_release_head_state(skb);
1621 __consume_stateless_skb(skb);
1622}
1623EXPORT_SYMBOL_GPL(skb_consume_udp);
1624
1625static struct sk_buff *__first_packet_length(struct sock *sk,
1626 struct sk_buff_head *rcvq,
1627 int *total)
1628{
1629 struct sk_buff *skb;
1630
1631 while ((skb = skb_peek(rcvq)) != NULL) {
1632 if (udp_lib_checksum_complete(skb)) {
1633 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1634 IS_UDPLITE(sk));
1635 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1636 IS_UDPLITE(sk));
1637 atomic_inc(&sk->sk_drops);
1638 __skb_unlink(skb, rcvq);
1639 *total += skb->truesize;
1640 kfree_skb(skb);
1641 } else {
1642 udp_skb_csum_unnecessary_set(skb);
1643 break;
1644 }
1645 }
1646 return skb;
1647}
1648
1649/**
1650 * first_packet_length - return length of first packet in receive queue
1651 * @sk: socket
1652 *
1653 * Drops all bad checksum frames, until a valid one is found.
1654 * Returns the length of found skb, or -1 if none is found.
1655 */
1656static int first_packet_length(struct sock *sk)
1657{
1658 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1659 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1660 struct sk_buff *skb;
1661 int total = 0;
1662 int res;
1663
1664 spin_lock_bh(&rcvq->lock);
1665 skb = __first_packet_length(sk, rcvq, &total);
1666 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1667 spin_lock(&sk_queue->lock);
1668 skb_queue_splice_tail_init(sk_queue, rcvq);
1669 spin_unlock(&sk_queue->lock);
1670
1671 skb = __first_packet_length(sk, rcvq, &total);
1672 }
1673 res = skb ? skb->len : -1;
1674 if (total)
1675 udp_rmem_release(sk, total, 1, false);
1676 spin_unlock_bh(&rcvq->lock);
1677 return res;
1678}
1679
1680/*
1681 * IOCTL requests applicable to the UDP protocol
1682 */
1683
1684int udp_ioctl(struct sock *sk, int cmd, int *karg)
1685{
1686 switch (cmd) {
1687 case SIOCOUTQ:
1688 {
1689 *karg = sk_wmem_alloc_get(sk);
1690 return 0;
1691 }
1692
1693 case SIOCINQ:
1694 {
1695 *karg = max_t(int, 0, first_packet_length(sk));
1696 return 0;
1697 }
1698
1699 default:
1700 return -ENOIOCTLCMD;
1701 }
1702
1703 return 0;
1704}
1705EXPORT_SYMBOL(udp_ioctl);
1706
1707struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1708 int *off, int *err)
1709{
1710 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1711 struct sk_buff_head *queue;
1712 struct sk_buff *last;
1713 long timeo;
1714 int error;
1715
1716 queue = &udp_sk(sk)->reader_queue;
1717 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1718 do {
1719 struct sk_buff *skb;
1720
1721 error = sock_error(sk);
1722 if (error)
1723 break;
1724
1725 error = -EAGAIN;
1726 do {
1727 spin_lock_bh(&queue->lock);
1728 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1729 err, &last);
1730 if (skb) {
1731 if (!(flags & MSG_PEEK))
1732 udp_skb_destructor(sk, skb);
1733 spin_unlock_bh(&queue->lock);
1734 return skb;
1735 }
1736
1737 if (skb_queue_empty_lockless(sk_queue)) {
1738 spin_unlock_bh(&queue->lock);
1739 goto busy_check;
1740 }
1741
1742 /* refill the reader queue and walk it again
1743 * keep both queues locked to avoid re-acquiring
1744 * the sk_receive_queue lock if fwd memory scheduling
1745 * is needed.
1746 */
1747 spin_lock(&sk_queue->lock);
1748 skb_queue_splice_tail_init(sk_queue, queue);
1749
1750 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1751 err, &last);
1752 if (skb && !(flags & MSG_PEEK))
1753 udp_skb_dtor_locked(sk, skb);
1754 spin_unlock(&sk_queue->lock);
1755 spin_unlock_bh(&queue->lock);
1756 if (skb)
1757 return skb;
1758
1759busy_check:
1760 if (!sk_can_busy_loop(sk))
1761 break;
1762
1763 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1764 } while (!skb_queue_empty_lockless(sk_queue));
1765
1766 /* sk_queue is empty, reader_queue may contain peeked packets */
1767 } while (timeo &&
1768 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1769 &error, &timeo,
1770 (struct sk_buff *)sk_queue));
1771
1772 *err = error;
1773 return NULL;
1774}
1775EXPORT_SYMBOL(__skb_recv_udp);
1776
1777int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1778{
1779 struct sk_buff *skb;
1780 int err;
1781
1782try_again:
1783 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1784 if (!skb)
1785 return err;
1786
1787 if (udp_lib_checksum_complete(skb)) {
1788 int is_udplite = IS_UDPLITE(sk);
1789 struct net *net = sock_net(sk);
1790
1791 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1792 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1793 atomic_inc(&sk->sk_drops);
1794 kfree_skb(skb);
1795 goto try_again;
1796 }
1797
1798 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1799 return recv_actor(sk, skb);
1800}
1801EXPORT_SYMBOL(udp_read_skb);
1802
1803/*
1804 * This should be easy, if there is something there we
1805 * return it, otherwise we block.
1806 */
1807
1808int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1809 int *addr_len)
1810{
1811 struct inet_sock *inet = inet_sk(sk);
1812 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1813 struct sk_buff *skb;
1814 unsigned int ulen, copied;
1815 int off, err, peeking = flags & MSG_PEEK;
1816 int is_udplite = IS_UDPLITE(sk);
1817 bool checksum_valid = false;
1818
1819 if (flags & MSG_ERRQUEUE)
1820 return ip_recv_error(sk, msg, len, addr_len);
1821
1822try_again:
1823 off = sk_peek_offset(sk, flags);
1824 skb = __skb_recv_udp(sk, flags, &off, &err);
1825 if (!skb)
1826 return err;
1827
1828 ulen = udp_skb_len(skb);
1829 copied = len;
1830 if (copied > ulen - off)
1831 copied = ulen - off;
1832 else if (copied < ulen)
1833 msg->msg_flags |= MSG_TRUNC;
1834
1835 /*
1836 * If checksum is needed at all, try to do it while copying the
1837 * data. If the data is truncated, or if we only want a partial
1838 * coverage checksum (UDP-Lite), do it before the copy.
1839 */
1840
1841 if (copied < ulen || peeking ||
1842 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1843 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1844 !__udp_lib_checksum_complete(skb);
1845 if (!checksum_valid)
1846 goto csum_copy_err;
1847 }
1848
1849 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1850 if (udp_skb_is_linear(skb))
1851 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1852 else
1853 err = skb_copy_datagram_msg(skb, off, msg, copied);
1854 } else {
1855 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1856
1857 if (err == -EINVAL)
1858 goto csum_copy_err;
1859 }
1860
1861 if (unlikely(err)) {
1862 if (!peeking) {
1863 atomic_inc(&sk->sk_drops);
1864 UDP_INC_STATS(sock_net(sk),
1865 UDP_MIB_INERRORS, is_udplite);
1866 }
1867 kfree_skb(skb);
1868 return err;
1869 }
1870
1871 if (!peeking)
1872 UDP_INC_STATS(sock_net(sk),
1873 UDP_MIB_INDATAGRAMS, is_udplite);
1874
1875 sock_recv_cmsgs(msg, sk, skb);
1876
1877 /* Copy the address. */
1878 if (sin) {
1879 sin->sin_family = AF_INET;
1880 sin->sin_port = udp_hdr(skb)->source;
1881 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1882 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1883 *addr_len = sizeof(*sin);
1884
1885 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1886 (struct sockaddr *)sin,
1887 addr_len);
1888 }
1889
1890 if (udp_test_bit(GRO_ENABLED, sk))
1891 udp_cmsg_recv(msg, sk, skb);
1892
1893 if (inet_cmsg_flags(inet))
1894 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1895
1896 err = copied;
1897 if (flags & MSG_TRUNC)
1898 err = ulen;
1899
1900 skb_consume_udp(sk, skb, peeking ? -err : err);
1901 return err;
1902
1903csum_copy_err:
1904 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1905 udp_skb_destructor)) {
1906 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1907 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1908 }
1909 kfree_skb(skb);
1910
1911 /* starting over for a new packet, but check if we need to yield */
1912 cond_resched();
1913 msg->msg_flags &= ~MSG_TRUNC;
1914 goto try_again;
1915}
1916
1917int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1918{
1919 /* This check is replicated from __ip4_datagram_connect() and
1920 * intended to prevent BPF program called below from accessing bytes
1921 * that are out of the bound specified by user in addr_len.
1922 */
1923 if (addr_len < sizeof(struct sockaddr_in))
1924 return -EINVAL;
1925
1926 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
1927}
1928EXPORT_SYMBOL(udp_pre_connect);
1929
1930int __udp_disconnect(struct sock *sk, int flags)
1931{
1932 struct inet_sock *inet = inet_sk(sk);
1933 /*
1934 * 1003.1g - break association.
1935 */
1936
1937 sk->sk_state = TCP_CLOSE;
1938 inet->inet_daddr = 0;
1939 inet->inet_dport = 0;
1940 sock_rps_reset_rxhash(sk);
1941 sk->sk_bound_dev_if = 0;
1942 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1943 inet_reset_saddr(sk);
1944 if (sk->sk_prot->rehash &&
1945 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1946 sk->sk_prot->rehash(sk);
1947 }
1948
1949 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1950 sk->sk_prot->unhash(sk);
1951 inet->inet_sport = 0;
1952 }
1953 sk_dst_reset(sk);
1954 return 0;
1955}
1956EXPORT_SYMBOL(__udp_disconnect);
1957
1958int udp_disconnect(struct sock *sk, int flags)
1959{
1960 lock_sock(sk);
1961 __udp_disconnect(sk, flags);
1962 release_sock(sk);
1963 return 0;
1964}
1965EXPORT_SYMBOL(udp_disconnect);
1966
1967void udp_lib_unhash(struct sock *sk)
1968{
1969 if (sk_hashed(sk)) {
1970 struct udp_table *udptable = udp_get_table_prot(sk);
1971 struct udp_hslot *hslot, *hslot2;
1972
1973 hslot = udp_hashslot(udptable, sock_net(sk),
1974 udp_sk(sk)->udp_port_hash);
1975 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1976
1977 spin_lock_bh(&hslot->lock);
1978 if (rcu_access_pointer(sk->sk_reuseport_cb))
1979 reuseport_detach_sock(sk);
1980 if (sk_del_node_init_rcu(sk)) {
1981 hslot->count--;
1982 inet_sk(sk)->inet_num = 0;
1983 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1984
1985 spin_lock(&hslot2->lock);
1986 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1987 hslot2->count--;
1988 spin_unlock(&hslot2->lock);
1989 }
1990 spin_unlock_bh(&hslot->lock);
1991 }
1992}
1993EXPORT_SYMBOL(udp_lib_unhash);
1994
1995/*
1996 * inet_rcv_saddr was changed, we must rehash secondary hash
1997 */
1998void udp_lib_rehash(struct sock *sk, u16 newhash)
1999{
2000 if (sk_hashed(sk)) {
2001 struct udp_table *udptable = udp_get_table_prot(sk);
2002 struct udp_hslot *hslot, *hslot2, *nhslot2;
2003
2004 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2005 nhslot2 = udp_hashslot2(udptable, newhash);
2006 udp_sk(sk)->udp_portaddr_hash = newhash;
2007
2008 if (hslot2 != nhslot2 ||
2009 rcu_access_pointer(sk->sk_reuseport_cb)) {
2010 hslot = udp_hashslot(udptable, sock_net(sk),
2011 udp_sk(sk)->udp_port_hash);
2012 /* we must lock primary chain too */
2013 spin_lock_bh(&hslot->lock);
2014 if (rcu_access_pointer(sk->sk_reuseport_cb))
2015 reuseport_detach_sock(sk);
2016
2017 if (hslot2 != nhslot2) {
2018 spin_lock(&hslot2->lock);
2019 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2020 hslot2->count--;
2021 spin_unlock(&hslot2->lock);
2022
2023 spin_lock(&nhslot2->lock);
2024 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2025 &nhslot2->head);
2026 nhslot2->count++;
2027 spin_unlock(&nhslot2->lock);
2028 }
2029
2030 spin_unlock_bh(&hslot->lock);
2031 }
2032 }
2033}
2034EXPORT_SYMBOL(udp_lib_rehash);
2035
2036void udp_v4_rehash(struct sock *sk)
2037{
2038 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2039 inet_sk(sk)->inet_rcv_saddr,
2040 inet_sk(sk)->inet_num);
2041 udp_lib_rehash(sk, new_hash);
2042}
2043
2044static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2045{
2046 int rc;
2047
2048 if (inet_sk(sk)->inet_daddr) {
2049 sock_rps_save_rxhash(sk, skb);
2050 sk_mark_napi_id(sk, skb);
2051 sk_incoming_cpu_update(sk);
2052 } else {
2053 sk_mark_napi_id_once(sk, skb);
2054 }
2055
2056 rc = __udp_enqueue_schedule_skb(sk, skb);
2057 if (rc < 0) {
2058 int is_udplite = IS_UDPLITE(sk);
2059 int drop_reason;
2060
2061 /* Note that an ENOMEM error is charged twice */
2062 if (rc == -ENOMEM) {
2063 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2064 is_udplite);
2065 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2066 } else {
2067 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2068 is_udplite);
2069 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2070 }
2071 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2072 kfree_skb_reason(skb, drop_reason);
2073 trace_udp_fail_queue_rcv_skb(rc, sk);
2074 return -1;
2075 }
2076
2077 return 0;
2078}
2079
2080/* returns:
2081 * -1: error
2082 * 0: success
2083 * >0: "udp encap" protocol resubmission
2084 *
2085 * Note that in the success and error cases, the skb is assumed to
2086 * have either been requeued or freed.
2087 */
2088static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2089{
2090 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2091 struct udp_sock *up = udp_sk(sk);
2092 int is_udplite = IS_UDPLITE(sk);
2093
2094 /*
2095 * Charge it to the socket, dropping if the queue is full.
2096 */
2097 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2098 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2099 goto drop;
2100 }
2101 nf_reset_ct(skb);
2102
2103 if (static_branch_unlikely(&udp_encap_needed_key) &&
2104 READ_ONCE(up->encap_type)) {
2105 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2106
2107 /*
2108 * This is an encapsulation socket so pass the skb to
2109 * the socket's udp_encap_rcv() hook. Otherwise, just
2110 * fall through and pass this up the UDP socket.
2111 * up->encap_rcv() returns the following value:
2112 * =0 if skb was successfully passed to the encap
2113 * handler or was discarded by it.
2114 * >0 if skb should be passed on to UDP.
2115 * <0 if skb should be resubmitted as proto -N
2116 */
2117
2118 /* if we're overly short, let UDP handle it */
2119 encap_rcv = READ_ONCE(up->encap_rcv);
2120 if (encap_rcv) {
2121 int ret;
2122
2123 /* Verify checksum before giving to encap */
2124 if (udp_lib_checksum_complete(skb))
2125 goto csum_error;
2126
2127 ret = encap_rcv(sk, skb);
2128 if (ret <= 0) {
2129 __UDP_INC_STATS(sock_net(sk),
2130 UDP_MIB_INDATAGRAMS,
2131 is_udplite);
2132 return -ret;
2133 }
2134 }
2135
2136 /* FALLTHROUGH -- it's a UDP Packet */
2137 }
2138
2139 /*
2140 * UDP-Lite specific tests, ignored on UDP sockets
2141 */
2142 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2143 u16 pcrlen = READ_ONCE(up->pcrlen);
2144
2145 /*
2146 * MIB statistics other than incrementing the error count are
2147 * disabled for the following two types of errors: these depend
2148 * on the application settings, not on the functioning of the
2149 * protocol stack as such.
2150 *
2151 * RFC 3828 here recommends (sec 3.3): "There should also be a
2152 * way ... to ... at least let the receiving application block
2153 * delivery of packets with coverage values less than a value
2154 * provided by the application."
2155 */
2156 if (pcrlen == 0) { /* full coverage was set */
2157 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2158 UDP_SKB_CB(skb)->cscov, skb->len);
2159 goto drop;
2160 }
2161 /* The next case involves violating the min. coverage requested
2162 * by the receiver. This is subtle: if receiver wants x and x is
2163 * greater than the buffersize/MTU then receiver will complain
2164 * that it wants x while sender emits packets of smaller size y.
2165 * Therefore the above ...()->partial_cov statement is essential.
2166 */
2167 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2168 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2169 UDP_SKB_CB(skb)->cscov, pcrlen);
2170 goto drop;
2171 }
2172 }
2173
2174 prefetch(&sk->sk_rmem_alloc);
2175 if (rcu_access_pointer(sk->sk_filter) &&
2176 udp_lib_checksum_complete(skb))
2177 goto csum_error;
2178
2179 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2180 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2181 goto drop;
2182 }
2183
2184 udp_csum_pull_header(skb);
2185
2186 ipv4_pktinfo_prepare(sk, skb, true);
2187 return __udp_queue_rcv_skb(sk, skb);
2188
2189csum_error:
2190 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2191 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2192drop:
2193 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2194 atomic_inc(&sk->sk_drops);
2195 kfree_skb_reason(skb, drop_reason);
2196 return -1;
2197}
2198
2199static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2200{
2201 struct sk_buff *next, *segs;
2202 int ret;
2203
2204 if (likely(!udp_unexpected_gso(sk, skb)))
2205 return udp_queue_rcv_one_skb(sk, skb);
2206
2207 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2208 __skb_push(skb, -skb_mac_offset(skb));
2209 segs = udp_rcv_segment(sk, skb, true);
2210 skb_list_walk_safe(segs, skb, next) {
2211 __skb_pull(skb, skb_transport_offset(skb));
2212
2213 udp_post_segment_fix_csum(skb);
2214 ret = udp_queue_rcv_one_skb(sk, skb);
2215 if (ret > 0)
2216 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2217 }
2218 return 0;
2219}
2220
2221/* For TCP sockets, sk_rx_dst is protected by socket lock
2222 * For UDP, we use xchg() to guard against concurrent changes.
2223 */
2224bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2225{
2226 struct dst_entry *old;
2227
2228 if (dst_hold_safe(dst)) {
2229 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2230 dst_release(old);
2231 return old != dst;
2232 }
2233 return false;
2234}
2235EXPORT_SYMBOL(udp_sk_rx_dst_set);
2236
2237/*
2238 * Multicasts and broadcasts go to each listener.
2239 *
2240 * Note: called only from the BH handler context.
2241 */
2242static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2243 struct udphdr *uh,
2244 __be32 saddr, __be32 daddr,
2245 struct udp_table *udptable,
2246 int proto)
2247{
2248 struct sock *sk, *first = NULL;
2249 unsigned short hnum = ntohs(uh->dest);
2250 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2251 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2252 unsigned int offset = offsetof(typeof(*sk), sk_node);
2253 int dif = skb->dev->ifindex;
2254 int sdif = inet_sdif(skb);
2255 struct hlist_node *node;
2256 struct sk_buff *nskb;
2257
2258 if (use_hash2) {
2259 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2260 udptable->mask;
2261 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2262start_lookup:
2263 hslot = &udptable->hash2[hash2];
2264 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2265 }
2266
2267 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2268 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2269 uh->source, saddr, dif, sdif, hnum))
2270 continue;
2271
2272 if (!first) {
2273 first = sk;
2274 continue;
2275 }
2276 nskb = skb_clone(skb, GFP_ATOMIC);
2277
2278 if (unlikely(!nskb)) {
2279 atomic_inc(&sk->sk_drops);
2280 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2281 IS_UDPLITE(sk));
2282 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2283 IS_UDPLITE(sk));
2284 continue;
2285 }
2286 if (udp_queue_rcv_skb(sk, nskb) > 0)
2287 consume_skb(nskb);
2288 }
2289
2290 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2291 if (use_hash2 && hash2 != hash2_any) {
2292 hash2 = hash2_any;
2293 goto start_lookup;
2294 }
2295
2296 if (first) {
2297 if (udp_queue_rcv_skb(first, skb) > 0)
2298 consume_skb(skb);
2299 } else {
2300 kfree_skb(skb);
2301 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2302 proto == IPPROTO_UDPLITE);
2303 }
2304 return 0;
2305}
2306
2307/* Initialize UDP checksum. If exited with zero value (success),
2308 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2309 * Otherwise, csum completion requires checksumming packet body,
2310 * including udp header and folding it to skb->csum.
2311 */
2312static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2313 int proto)
2314{
2315 int err;
2316
2317 UDP_SKB_CB(skb)->partial_cov = 0;
2318 UDP_SKB_CB(skb)->cscov = skb->len;
2319
2320 if (proto == IPPROTO_UDPLITE) {
2321 err = udplite_checksum_init(skb, uh);
2322 if (err)
2323 return err;
2324
2325 if (UDP_SKB_CB(skb)->partial_cov) {
2326 skb->csum = inet_compute_pseudo(skb, proto);
2327 return 0;
2328 }
2329 }
2330
2331 /* Note, we are only interested in != 0 or == 0, thus the
2332 * force to int.
2333 */
2334 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2335 inet_compute_pseudo);
2336 if (err)
2337 return err;
2338
2339 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2340 /* If SW calculated the value, we know it's bad */
2341 if (skb->csum_complete_sw)
2342 return 1;
2343
2344 /* HW says the value is bad. Let's validate that.
2345 * skb->csum is no longer the full packet checksum,
2346 * so don't treat it as such.
2347 */
2348 skb_checksum_complete_unset(skb);
2349 }
2350
2351 return 0;
2352}
2353
2354/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2355 * return code conversion for ip layer consumption
2356 */
2357static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2358 struct udphdr *uh)
2359{
2360 int ret;
2361
2362 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2363 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2364
2365 ret = udp_queue_rcv_skb(sk, skb);
2366
2367 /* a return value > 0 means to resubmit the input, but
2368 * it wants the return to be -protocol, or 0
2369 */
2370 if (ret > 0)
2371 return -ret;
2372 return 0;
2373}
2374
2375/*
2376 * All we need to do is get the socket, and then do a checksum.
2377 */
2378
2379int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2380 int proto)
2381{
2382 struct sock *sk;
2383 struct udphdr *uh;
2384 unsigned short ulen;
2385 struct rtable *rt = skb_rtable(skb);
2386 __be32 saddr, daddr;
2387 struct net *net = dev_net(skb->dev);
2388 bool refcounted;
2389 int drop_reason;
2390
2391 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2392
2393 /*
2394 * Validate the packet.
2395 */
2396 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2397 goto drop; /* No space for header. */
2398
2399 uh = udp_hdr(skb);
2400 ulen = ntohs(uh->len);
2401 saddr = ip_hdr(skb)->saddr;
2402 daddr = ip_hdr(skb)->daddr;
2403
2404 if (ulen > skb->len)
2405 goto short_packet;
2406
2407 if (proto == IPPROTO_UDP) {
2408 /* UDP validates ulen. */
2409 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2410 goto short_packet;
2411 uh = udp_hdr(skb);
2412 }
2413
2414 if (udp4_csum_init(skb, uh, proto))
2415 goto csum_error;
2416
2417 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2418 &refcounted, udp_ehashfn);
2419 if (IS_ERR(sk))
2420 goto no_sk;
2421
2422 if (sk) {
2423 struct dst_entry *dst = skb_dst(skb);
2424 int ret;
2425
2426 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2427 udp_sk_rx_dst_set(sk, dst);
2428
2429 ret = udp_unicast_rcv_skb(sk, skb, uh);
2430 if (refcounted)
2431 sock_put(sk);
2432 return ret;
2433 }
2434
2435 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2436 return __udp4_lib_mcast_deliver(net, skb, uh,
2437 saddr, daddr, udptable, proto);
2438
2439 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2440 if (sk)
2441 return udp_unicast_rcv_skb(sk, skb, uh);
2442no_sk:
2443 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2444 goto drop;
2445 nf_reset_ct(skb);
2446
2447 /* No socket. Drop packet silently, if checksum is wrong */
2448 if (udp_lib_checksum_complete(skb))
2449 goto csum_error;
2450
2451 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2452 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2453 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2454
2455 /*
2456 * Hmm. We got an UDP packet to a port to which we
2457 * don't wanna listen. Ignore it.
2458 */
2459 kfree_skb_reason(skb, drop_reason);
2460 return 0;
2461
2462short_packet:
2463 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2464 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2465 proto == IPPROTO_UDPLITE ? "Lite" : "",
2466 &saddr, ntohs(uh->source),
2467 ulen, skb->len,
2468 &daddr, ntohs(uh->dest));
2469 goto drop;
2470
2471csum_error:
2472 /*
2473 * RFC1122: OK. Discards the bad packet silently (as far as
2474 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2475 */
2476 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2477 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2478 proto == IPPROTO_UDPLITE ? "Lite" : "",
2479 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2480 ulen);
2481 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2482drop:
2483 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2484 kfree_skb_reason(skb, drop_reason);
2485 return 0;
2486}
2487
2488/* We can only early demux multicast if there is a single matching socket.
2489 * If more than one socket found returns NULL
2490 */
2491static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2492 __be16 loc_port, __be32 loc_addr,
2493 __be16 rmt_port, __be32 rmt_addr,
2494 int dif, int sdif)
2495{
2496 struct udp_table *udptable = net->ipv4.udp_table;
2497 unsigned short hnum = ntohs(loc_port);
2498 struct sock *sk, *result;
2499 struct udp_hslot *hslot;
2500 unsigned int slot;
2501
2502 slot = udp_hashfn(net, hnum, udptable->mask);
2503 hslot = &udptable->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 struct udp_table *udptable = net->ipv4.udp_table;
2532 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2533 unsigned short hnum = ntohs(loc_port);
2534 unsigned int hash2, slot2;
2535 struct udp_hslot *hslot2;
2536 __portpair ports;
2537 struct sock *sk;
2538
2539 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2540 slot2 = hash2 & udptable->mask;
2541 hslot2 = &udptable->hash2[slot2];
2542 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2543
2544 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2545 if (inet_match(net, sk, acookie, ports, dif, sdif))
2546 return sk;
2547 /* Only check first socket in chain */
2548 break;
2549 }
2550 return NULL;
2551}
2552
2553int udp_v4_early_demux(struct sk_buff *skb)
2554{
2555 struct net *net = dev_net(skb->dev);
2556 struct in_device *in_dev = NULL;
2557 const struct iphdr *iph;
2558 const struct udphdr *uh;
2559 struct sock *sk = NULL;
2560 struct dst_entry *dst;
2561 int dif = skb->dev->ifindex;
2562 int sdif = inet_sdif(skb);
2563 int ours;
2564
2565 /* validate the packet */
2566 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2567 return 0;
2568
2569 iph = ip_hdr(skb);
2570 uh = udp_hdr(skb);
2571
2572 if (skb->pkt_type == PACKET_MULTICAST) {
2573 in_dev = __in_dev_get_rcu(skb->dev);
2574
2575 if (!in_dev)
2576 return 0;
2577
2578 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2579 iph->protocol);
2580 if (!ours)
2581 return 0;
2582
2583 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2584 uh->source, iph->saddr,
2585 dif, sdif);
2586 } else if (skb->pkt_type == PACKET_HOST) {
2587 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2588 uh->source, iph->saddr, dif, sdif);
2589 }
2590
2591 if (!sk)
2592 return 0;
2593
2594 skb->sk = sk;
2595 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2596 skb->destructor = sock_pfree;
2597 dst = rcu_dereference(sk->sk_rx_dst);
2598
2599 if (dst)
2600 dst = dst_check(dst, 0);
2601 if (dst) {
2602 u32 itag = 0;
2603
2604 /* set noref for now.
2605 * any place which wants to hold dst has to call
2606 * dst_hold_safe()
2607 */
2608 skb_dst_set_noref(skb, dst);
2609
2610 /* for unconnected multicast sockets we need to validate
2611 * the source on each packet
2612 */
2613 if (!inet_sk(sk)->inet_daddr && in_dev)
2614 return ip_mc_validate_source(skb, iph->daddr,
2615 iph->saddr,
2616 iph->tos & IPTOS_RT_MASK,
2617 skb->dev, in_dev, &itag);
2618 }
2619 return 0;
2620}
2621
2622int udp_rcv(struct sk_buff *skb)
2623{
2624 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2625}
2626
2627void udp_destroy_sock(struct sock *sk)
2628{
2629 struct udp_sock *up = udp_sk(sk);
2630 bool slow = lock_sock_fast(sk);
2631
2632 /* protects from races with udp_abort() */
2633 sock_set_flag(sk, SOCK_DEAD);
2634 udp_flush_pending_frames(sk);
2635 unlock_sock_fast(sk, slow);
2636 if (static_branch_unlikely(&udp_encap_needed_key)) {
2637 if (up->encap_type) {
2638 void (*encap_destroy)(struct sock *sk);
2639 encap_destroy = READ_ONCE(up->encap_destroy);
2640 if (encap_destroy)
2641 encap_destroy(sk);
2642 }
2643 if (udp_test_bit(ENCAP_ENABLED, sk))
2644 static_branch_dec(&udp_encap_needed_key);
2645 }
2646}
2647
2648static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2649 struct sock *sk)
2650{
2651#ifdef CONFIG_XFRM
2652 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2653 if (family == AF_INET)
2654 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2655 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2656 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2657 }
2658#endif
2659}
2660
2661/*
2662 * Socket option code for UDP
2663 */
2664int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2665 sockptr_t optval, unsigned int optlen,
2666 int (*push_pending_frames)(struct sock *))
2667{
2668 struct udp_sock *up = udp_sk(sk);
2669 int val, valbool;
2670 int err = 0;
2671 int is_udplite = IS_UDPLITE(sk);
2672
2673 if (level == SOL_SOCKET) {
2674 err = sk_setsockopt(sk, level, optname, optval, optlen);
2675
2676 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2677 sockopt_lock_sock(sk);
2678 /* paired with READ_ONCE in udp_rmem_release() */
2679 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2680 sockopt_release_sock(sk);
2681 }
2682 return err;
2683 }
2684
2685 if (optlen < sizeof(int))
2686 return -EINVAL;
2687
2688 if (copy_from_sockptr(&val, optval, sizeof(val)))
2689 return -EFAULT;
2690
2691 valbool = val ? 1 : 0;
2692
2693 switch (optname) {
2694 case UDP_CORK:
2695 if (val != 0) {
2696 udp_set_bit(CORK, sk);
2697 } else {
2698 udp_clear_bit(CORK, sk);
2699 lock_sock(sk);
2700 push_pending_frames(sk);
2701 release_sock(sk);
2702 }
2703 break;
2704
2705 case UDP_ENCAP:
2706 switch (val) {
2707 case 0:
2708#ifdef CONFIG_XFRM
2709 case UDP_ENCAP_ESPINUDP:
2710 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2711 fallthrough;
2712 case UDP_ENCAP_ESPINUDP_NON_IKE:
2713#if IS_ENABLED(CONFIG_IPV6)
2714 if (sk->sk_family == AF_INET6)
2715 WRITE_ONCE(up->encap_rcv,
2716 ipv6_stub->xfrm6_udp_encap_rcv);
2717 else
2718#endif
2719 WRITE_ONCE(up->encap_rcv,
2720 xfrm4_udp_encap_rcv);
2721#endif
2722 fallthrough;
2723 case UDP_ENCAP_L2TPINUDP:
2724 WRITE_ONCE(up->encap_type, val);
2725 udp_tunnel_encap_enable(sk);
2726 break;
2727 default:
2728 err = -ENOPROTOOPT;
2729 break;
2730 }
2731 break;
2732
2733 case UDP_NO_CHECK6_TX:
2734 udp_set_no_check6_tx(sk, valbool);
2735 break;
2736
2737 case UDP_NO_CHECK6_RX:
2738 udp_set_no_check6_rx(sk, valbool);
2739 break;
2740
2741 case UDP_SEGMENT:
2742 if (val < 0 || val > USHRT_MAX)
2743 return -EINVAL;
2744 WRITE_ONCE(up->gso_size, val);
2745 break;
2746
2747 case UDP_GRO:
2748
2749 /* when enabling GRO, accept the related GSO packet type */
2750 if (valbool)
2751 udp_tunnel_encap_enable(sk);
2752 udp_assign_bit(GRO_ENABLED, sk, valbool);
2753 udp_assign_bit(ACCEPT_L4, sk, valbool);
2754 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
2755 break;
2756
2757 /*
2758 * UDP-Lite's partial checksum coverage (RFC 3828).
2759 */
2760 /* The sender sets actual checksum coverage length via this option.
2761 * The case coverage > packet length is handled by send module. */
2762 case UDPLITE_SEND_CSCOV:
2763 if (!is_udplite) /* Disable the option on UDP sockets */
2764 return -ENOPROTOOPT;
2765 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2766 val = 8;
2767 else if (val > USHRT_MAX)
2768 val = USHRT_MAX;
2769 WRITE_ONCE(up->pcslen, val);
2770 udp_set_bit(UDPLITE_SEND_CC, sk);
2771 break;
2772
2773 /* The receiver specifies a minimum checksum coverage value. To make
2774 * sense, this should be set to at least 8 (as done below). If zero is
2775 * used, this again means full checksum coverage. */
2776 case UDPLITE_RECV_CSCOV:
2777 if (!is_udplite) /* Disable the option on UDP sockets */
2778 return -ENOPROTOOPT;
2779 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2780 val = 8;
2781 else if (val > USHRT_MAX)
2782 val = USHRT_MAX;
2783 WRITE_ONCE(up->pcrlen, val);
2784 udp_set_bit(UDPLITE_RECV_CC, sk);
2785 break;
2786
2787 default:
2788 err = -ENOPROTOOPT;
2789 break;
2790 }
2791
2792 return err;
2793}
2794EXPORT_SYMBOL(udp_lib_setsockopt);
2795
2796int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2797 unsigned int optlen)
2798{
2799 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2800 return udp_lib_setsockopt(sk, level, optname,
2801 optval, optlen,
2802 udp_push_pending_frames);
2803 return ip_setsockopt(sk, level, optname, optval, optlen);
2804}
2805
2806int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2807 char __user *optval, int __user *optlen)
2808{
2809 struct udp_sock *up = udp_sk(sk);
2810 int val, len;
2811
2812 if (get_user(len, optlen))
2813 return -EFAULT;
2814
2815 if (len < 0)
2816 return -EINVAL;
2817
2818 len = min_t(unsigned int, len, sizeof(int));
2819
2820 switch (optname) {
2821 case UDP_CORK:
2822 val = udp_test_bit(CORK, sk);
2823 break;
2824
2825 case UDP_ENCAP:
2826 val = READ_ONCE(up->encap_type);
2827 break;
2828
2829 case UDP_NO_CHECK6_TX:
2830 val = udp_get_no_check6_tx(sk);
2831 break;
2832
2833 case UDP_NO_CHECK6_RX:
2834 val = udp_get_no_check6_rx(sk);
2835 break;
2836
2837 case UDP_SEGMENT:
2838 val = READ_ONCE(up->gso_size);
2839 break;
2840
2841 case UDP_GRO:
2842 val = udp_test_bit(GRO_ENABLED, sk);
2843 break;
2844
2845 /* The following two cannot be changed on UDP sockets, the return is
2846 * always 0 (which corresponds to the full checksum coverage of UDP). */
2847 case UDPLITE_SEND_CSCOV:
2848 val = READ_ONCE(up->pcslen);
2849 break;
2850
2851 case UDPLITE_RECV_CSCOV:
2852 val = READ_ONCE(up->pcrlen);
2853 break;
2854
2855 default:
2856 return -ENOPROTOOPT;
2857 }
2858
2859 if (put_user(len, optlen))
2860 return -EFAULT;
2861 if (copy_to_user(optval, &val, len))
2862 return -EFAULT;
2863 return 0;
2864}
2865EXPORT_SYMBOL(udp_lib_getsockopt);
2866
2867int udp_getsockopt(struct sock *sk, int level, int optname,
2868 char __user *optval, int __user *optlen)
2869{
2870 if (level == SOL_UDP || level == SOL_UDPLITE)
2871 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2872 return ip_getsockopt(sk, level, optname, optval, optlen);
2873}
2874
2875/**
2876 * udp_poll - wait for a UDP event.
2877 * @file: - file struct
2878 * @sock: - socket
2879 * @wait: - poll table
2880 *
2881 * This is same as datagram poll, except for the special case of
2882 * blocking sockets. If application is using a blocking fd
2883 * and a packet with checksum error is in the queue;
2884 * then it could get return from select indicating data available
2885 * but then block when reading it. Add special case code
2886 * to work around these arguably broken applications.
2887 */
2888__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2889{
2890 __poll_t mask = datagram_poll(file, sock, wait);
2891 struct sock *sk = sock->sk;
2892
2893 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2894 mask |= EPOLLIN | EPOLLRDNORM;
2895
2896 /* Check for false positives due to checksum errors */
2897 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2898 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2899 mask &= ~(EPOLLIN | EPOLLRDNORM);
2900
2901 /* psock ingress_msg queue should not contain any bad checksum frames */
2902 if (sk_is_readable(sk))
2903 mask |= EPOLLIN | EPOLLRDNORM;
2904 return mask;
2905
2906}
2907EXPORT_SYMBOL(udp_poll);
2908
2909int udp_abort(struct sock *sk, int err)
2910{
2911 if (!has_current_bpf_ctx())
2912 lock_sock(sk);
2913
2914 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2915 * with close()
2916 */
2917 if (sock_flag(sk, SOCK_DEAD))
2918 goto out;
2919
2920 sk->sk_err = err;
2921 sk_error_report(sk);
2922 __udp_disconnect(sk, 0);
2923
2924out:
2925 if (!has_current_bpf_ctx())
2926 release_sock(sk);
2927
2928 return 0;
2929}
2930EXPORT_SYMBOL_GPL(udp_abort);
2931
2932struct proto udp_prot = {
2933 .name = "UDP",
2934 .owner = THIS_MODULE,
2935 .close = udp_lib_close,
2936 .pre_connect = udp_pre_connect,
2937 .connect = ip4_datagram_connect,
2938 .disconnect = udp_disconnect,
2939 .ioctl = udp_ioctl,
2940 .init = udp_init_sock,
2941 .destroy = udp_destroy_sock,
2942 .setsockopt = udp_setsockopt,
2943 .getsockopt = udp_getsockopt,
2944 .sendmsg = udp_sendmsg,
2945 .recvmsg = udp_recvmsg,
2946 .splice_eof = udp_splice_eof,
2947 .release_cb = ip4_datagram_release_cb,
2948 .hash = udp_lib_hash,
2949 .unhash = udp_lib_unhash,
2950 .rehash = udp_v4_rehash,
2951 .get_port = udp_v4_get_port,
2952 .put_port = udp_lib_unhash,
2953#ifdef CONFIG_BPF_SYSCALL
2954 .psock_update_sk_prot = udp_bpf_update_proto,
2955#endif
2956 .memory_allocated = &udp_memory_allocated,
2957 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2958
2959 .sysctl_mem = sysctl_udp_mem,
2960 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2961 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2962 .obj_size = sizeof(struct udp_sock),
2963 .h.udp_table = NULL,
2964 .diag_destroy = udp_abort,
2965};
2966EXPORT_SYMBOL(udp_prot);
2967
2968/* ------------------------------------------------------------------------ */
2969#ifdef CONFIG_PROC_FS
2970
2971static unsigned short seq_file_family(const struct seq_file *seq);
2972static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2973{
2974 unsigned short family = seq_file_family(seq);
2975
2976 /* AF_UNSPEC is used as a match all */
2977 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2978 net_eq(sock_net(sk), seq_file_net(seq)));
2979}
2980
2981#ifdef CONFIG_BPF_SYSCALL
2982static const struct seq_operations bpf_iter_udp_seq_ops;
2983#endif
2984static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2985 struct net *net)
2986{
2987 const struct udp_seq_afinfo *afinfo;
2988
2989#ifdef CONFIG_BPF_SYSCALL
2990 if (seq->op == &bpf_iter_udp_seq_ops)
2991 return net->ipv4.udp_table;
2992#endif
2993
2994 afinfo = pde_data(file_inode(seq->file));
2995 return afinfo->udp_table ? : net->ipv4.udp_table;
2996}
2997
2998static struct sock *udp_get_first(struct seq_file *seq, int start)
2999{
3000 struct udp_iter_state *state = seq->private;
3001 struct net *net = seq_file_net(seq);
3002 struct udp_table *udptable;
3003 struct sock *sk;
3004
3005 udptable = udp_get_table_seq(seq, 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 (seq_sk_match(seq, sk))
3017 goto found;
3018 }
3019 spin_unlock_bh(&hslot->lock);
3020 }
3021 sk = NULL;
3022found:
3023 return sk;
3024}
3025
3026static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3027{
3028 struct udp_iter_state *state = seq->private;
3029 struct net *net = seq_file_net(seq);
3030 struct udp_table *udptable;
3031
3032 do {
3033 sk = sk_next(sk);
3034 } while (sk && !seq_sk_match(seq, sk));
3035
3036 if (!sk) {
3037 udptable = udp_get_table_seq(seq, net);
3038
3039 if (state->bucket <= udptable->mask)
3040 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3041
3042 return udp_get_first(seq, state->bucket + 1);
3043 }
3044 return sk;
3045}
3046
3047static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3048{
3049 struct sock *sk = udp_get_first(seq, 0);
3050
3051 if (sk)
3052 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3053 --pos;
3054 return pos ? NULL : sk;
3055}
3056
3057void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3058{
3059 struct udp_iter_state *state = seq->private;
3060 state->bucket = MAX_UDP_PORTS;
3061
3062 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3063}
3064EXPORT_SYMBOL(udp_seq_start);
3065
3066void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3067{
3068 struct sock *sk;
3069
3070 if (v == SEQ_START_TOKEN)
3071 sk = udp_get_idx(seq, 0);
3072 else
3073 sk = udp_get_next(seq, v);
3074
3075 ++*pos;
3076 return sk;
3077}
3078EXPORT_SYMBOL(udp_seq_next);
3079
3080void udp_seq_stop(struct seq_file *seq, void *v)
3081{
3082 struct udp_iter_state *state = seq->private;
3083 struct udp_table *udptable;
3084
3085 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3086
3087 if (state->bucket <= udptable->mask)
3088 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3089}
3090EXPORT_SYMBOL(udp_seq_stop);
3091
3092/* ------------------------------------------------------------------------ */
3093static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3094 int bucket)
3095{
3096 struct inet_sock *inet = inet_sk(sp);
3097 __be32 dest = inet->inet_daddr;
3098 __be32 src = inet->inet_rcv_saddr;
3099 __u16 destp = ntohs(inet->inet_dport);
3100 __u16 srcp = ntohs(inet->inet_sport);
3101
3102 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3103 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3104 bucket, src, srcp, dest, destp, sp->sk_state,
3105 sk_wmem_alloc_get(sp),
3106 udp_rqueue_get(sp),
3107 0, 0L, 0,
3108 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3109 0, sock_i_ino(sp),
3110 refcount_read(&sp->sk_refcnt), sp,
3111 atomic_read(&sp->sk_drops));
3112}
3113
3114int udp4_seq_show(struct seq_file *seq, void *v)
3115{
3116 seq_setwidth(seq, 127);
3117 if (v == SEQ_START_TOKEN)
3118 seq_puts(seq, " sl local_address rem_address st tx_queue "
3119 "rx_queue tr tm->when retrnsmt uid timeout "
3120 "inode ref pointer drops");
3121 else {
3122 struct udp_iter_state *state = seq->private;
3123
3124 udp4_format_sock(v, seq, state->bucket);
3125 }
3126 seq_pad(seq, '\n');
3127 return 0;
3128}
3129
3130#ifdef CONFIG_BPF_SYSCALL
3131struct bpf_iter__udp {
3132 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3133 __bpf_md_ptr(struct udp_sock *, udp_sk);
3134 uid_t uid __aligned(8);
3135 int bucket __aligned(8);
3136};
3137
3138struct bpf_udp_iter_state {
3139 struct udp_iter_state state;
3140 unsigned int cur_sk;
3141 unsigned int end_sk;
3142 unsigned int max_sk;
3143 int offset;
3144 struct sock **batch;
3145 bool st_bucket_done;
3146};
3147
3148static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3149 unsigned int new_batch_sz);
3150static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3151{
3152 struct bpf_udp_iter_state *iter = seq->private;
3153 struct udp_iter_state *state = &iter->state;
3154 struct net *net = seq_file_net(seq);
3155 int resume_bucket, resume_offset;
3156 struct udp_table *udptable;
3157 unsigned int batch_sks = 0;
3158 bool resized = false;
3159 struct sock *sk;
3160
3161 resume_bucket = state->bucket;
3162 resume_offset = iter->offset;
3163
3164 /* The current batch is done, so advance the bucket. */
3165 if (iter->st_bucket_done)
3166 state->bucket++;
3167
3168 udptable = udp_get_table_seq(seq, net);
3169
3170again:
3171 /* New batch for the next bucket.
3172 * Iterate over the hash table to find a bucket with sockets matching
3173 * the iterator attributes, and return the first matching socket from
3174 * the bucket. The remaining matched sockets from the bucket are batched
3175 * before releasing the bucket lock. This allows BPF programs that are
3176 * called in seq_show to acquire the bucket lock if needed.
3177 */
3178 iter->cur_sk = 0;
3179 iter->end_sk = 0;
3180 iter->st_bucket_done = false;
3181 batch_sks = 0;
3182
3183 for (; state->bucket <= udptable->mask; state->bucket++) {
3184 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3185
3186 if (hlist_empty(&hslot2->head))
3187 continue;
3188
3189 iter->offset = 0;
3190 spin_lock_bh(&hslot2->lock);
3191 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3192 if (seq_sk_match(seq, sk)) {
3193 /* Resume from the last iterated socket at the
3194 * offset in the bucket before iterator was stopped.
3195 */
3196 if (state->bucket == resume_bucket &&
3197 iter->offset < resume_offset) {
3198 ++iter->offset;
3199 continue;
3200 }
3201 if (iter->end_sk < iter->max_sk) {
3202 sock_hold(sk);
3203 iter->batch[iter->end_sk++] = sk;
3204 }
3205 batch_sks++;
3206 }
3207 }
3208 spin_unlock_bh(&hslot2->lock);
3209
3210 if (iter->end_sk)
3211 break;
3212 }
3213
3214 /* All done: no batch made. */
3215 if (!iter->end_sk)
3216 return NULL;
3217
3218 if (iter->end_sk == batch_sks) {
3219 /* Batching is done for the current bucket; return the first
3220 * socket to be iterated from the batch.
3221 */
3222 iter->st_bucket_done = true;
3223 goto done;
3224 }
3225 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3226 resized = true;
3227 /* After allocating a larger batch, retry one more time to grab
3228 * the whole bucket.
3229 */
3230 goto again;
3231 }
3232done:
3233 return iter->batch[0];
3234}
3235
3236static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3237{
3238 struct bpf_udp_iter_state *iter = seq->private;
3239 struct sock *sk;
3240
3241 /* Whenever seq_next() is called, the iter->cur_sk is
3242 * done with seq_show(), so unref the iter->cur_sk.
3243 */
3244 if (iter->cur_sk < iter->end_sk) {
3245 sock_put(iter->batch[iter->cur_sk++]);
3246 ++iter->offset;
3247 }
3248
3249 /* After updating iter->cur_sk, check if there are more sockets
3250 * available in the current bucket batch.
3251 */
3252 if (iter->cur_sk < iter->end_sk)
3253 sk = iter->batch[iter->cur_sk];
3254 else
3255 /* Prepare a new batch. */
3256 sk = bpf_iter_udp_batch(seq);
3257
3258 ++*pos;
3259 return sk;
3260}
3261
3262static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3263{
3264 /* bpf iter does not support lseek, so it always
3265 * continue from where it was stop()-ped.
3266 */
3267 if (*pos)
3268 return bpf_iter_udp_batch(seq);
3269
3270 return SEQ_START_TOKEN;
3271}
3272
3273static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3274 struct udp_sock *udp_sk, uid_t uid, int bucket)
3275{
3276 struct bpf_iter__udp ctx;
3277
3278 meta->seq_num--; /* skip SEQ_START_TOKEN */
3279 ctx.meta = meta;
3280 ctx.udp_sk = udp_sk;
3281 ctx.uid = uid;
3282 ctx.bucket = bucket;
3283 return bpf_iter_run_prog(prog, &ctx);
3284}
3285
3286static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3287{
3288 struct udp_iter_state *state = seq->private;
3289 struct bpf_iter_meta meta;
3290 struct bpf_prog *prog;
3291 struct sock *sk = v;
3292 uid_t uid;
3293 int ret;
3294
3295 if (v == SEQ_START_TOKEN)
3296 return 0;
3297
3298 lock_sock(sk);
3299
3300 if (unlikely(sk_unhashed(sk))) {
3301 ret = SEQ_SKIP;
3302 goto unlock;
3303 }
3304
3305 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3306 meta.seq = seq;
3307 prog = bpf_iter_get_info(&meta, false);
3308 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3309
3310unlock:
3311 release_sock(sk);
3312 return ret;
3313}
3314
3315static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3316{
3317 while (iter->cur_sk < iter->end_sk)
3318 sock_put(iter->batch[iter->cur_sk++]);
3319}
3320
3321static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3322{
3323 struct bpf_udp_iter_state *iter = seq->private;
3324 struct bpf_iter_meta meta;
3325 struct bpf_prog *prog;
3326
3327 if (!v) {
3328 meta.seq = seq;
3329 prog = bpf_iter_get_info(&meta, true);
3330 if (prog)
3331 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3332 }
3333
3334 if (iter->cur_sk < iter->end_sk) {
3335 bpf_iter_udp_put_batch(iter);
3336 iter->st_bucket_done = false;
3337 }
3338}
3339
3340static const struct seq_operations bpf_iter_udp_seq_ops = {
3341 .start = bpf_iter_udp_seq_start,
3342 .next = bpf_iter_udp_seq_next,
3343 .stop = bpf_iter_udp_seq_stop,
3344 .show = bpf_iter_udp_seq_show,
3345};
3346#endif
3347
3348static unsigned short seq_file_family(const struct seq_file *seq)
3349{
3350 const struct udp_seq_afinfo *afinfo;
3351
3352#ifdef CONFIG_BPF_SYSCALL
3353 /* BPF iterator: bpf programs to filter sockets. */
3354 if (seq->op == &bpf_iter_udp_seq_ops)
3355 return AF_UNSPEC;
3356#endif
3357
3358 /* Proc fs iterator */
3359 afinfo = pde_data(file_inode(seq->file));
3360 return afinfo->family;
3361}
3362
3363const struct seq_operations udp_seq_ops = {
3364 .start = udp_seq_start,
3365 .next = udp_seq_next,
3366 .stop = udp_seq_stop,
3367 .show = udp4_seq_show,
3368};
3369EXPORT_SYMBOL(udp_seq_ops);
3370
3371static struct udp_seq_afinfo udp4_seq_afinfo = {
3372 .family = AF_INET,
3373 .udp_table = NULL,
3374};
3375
3376static int __net_init udp4_proc_init_net(struct net *net)
3377{
3378 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3379 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3380 return -ENOMEM;
3381 return 0;
3382}
3383
3384static void __net_exit udp4_proc_exit_net(struct net *net)
3385{
3386 remove_proc_entry("udp", net->proc_net);
3387}
3388
3389static struct pernet_operations udp4_net_ops = {
3390 .init = udp4_proc_init_net,
3391 .exit = udp4_proc_exit_net,
3392};
3393
3394int __init udp4_proc_init(void)
3395{
3396 return register_pernet_subsys(&udp4_net_ops);
3397}
3398
3399void udp4_proc_exit(void)
3400{
3401 unregister_pernet_subsys(&udp4_net_ops);
3402}
3403#endif /* CONFIG_PROC_FS */
3404
3405static __initdata unsigned long uhash_entries;
3406static int __init set_uhash_entries(char *str)
3407{
3408 ssize_t ret;
3409
3410 if (!str)
3411 return 0;
3412
3413 ret = kstrtoul(str, 0, &uhash_entries);
3414 if (ret)
3415 return 0;
3416
3417 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3418 uhash_entries = UDP_HTABLE_SIZE_MIN;
3419 return 1;
3420}
3421__setup("uhash_entries=", set_uhash_entries);
3422
3423void __init udp_table_init(struct udp_table *table, const char *name)
3424{
3425 unsigned int i;
3426
3427 table->hash = alloc_large_system_hash(name,
3428 2 * sizeof(struct udp_hslot),
3429 uhash_entries,
3430 21, /* one slot per 2 MB */
3431 0,
3432 &table->log,
3433 &table->mask,
3434 UDP_HTABLE_SIZE_MIN,
3435 UDP_HTABLE_SIZE_MAX);
3436
3437 table->hash2 = table->hash + (table->mask + 1);
3438 for (i = 0; i <= table->mask; i++) {
3439 INIT_HLIST_HEAD(&table->hash[i].head);
3440 table->hash[i].count = 0;
3441 spin_lock_init(&table->hash[i].lock);
3442 }
3443 for (i = 0; i <= table->mask; i++) {
3444 INIT_HLIST_HEAD(&table->hash2[i].head);
3445 table->hash2[i].count = 0;
3446 spin_lock_init(&table->hash2[i].lock);
3447 }
3448}
3449
3450u32 udp_flow_hashrnd(void)
3451{
3452 static u32 hashrnd __read_mostly;
3453
3454 net_get_random_once(&hashrnd, sizeof(hashrnd));
3455
3456 return hashrnd;
3457}
3458EXPORT_SYMBOL(udp_flow_hashrnd);
3459
3460static void __net_init udp_sysctl_init(struct net *net)
3461{
3462 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3463 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3464
3465#ifdef CONFIG_NET_L3_MASTER_DEV
3466 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3467#endif
3468}
3469
3470static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3471{
3472 struct udp_table *udptable;
3473 int i;
3474
3475 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3476 if (!udptable)
3477 goto out;
3478
3479 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3480 GFP_KERNEL_ACCOUNT);
3481 if (!udptable->hash)
3482 goto free_table;
3483
3484 udptable->hash2 = udptable->hash + hash_entries;
3485 udptable->mask = hash_entries - 1;
3486 udptable->log = ilog2(hash_entries);
3487
3488 for (i = 0; i < hash_entries; i++) {
3489 INIT_HLIST_HEAD(&udptable->hash[i].head);
3490 udptable->hash[i].count = 0;
3491 spin_lock_init(&udptable->hash[i].lock);
3492
3493 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3494 udptable->hash2[i].count = 0;
3495 spin_lock_init(&udptable->hash2[i].lock);
3496 }
3497
3498 return udptable;
3499
3500free_table:
3501 kfree(udptable);
3502out:
3503 return NULL;
3504}
3505
3506static void __net_exit udp_pernet_table_free(struct net *net)
3507{
3508 struct udp_table *udptable = net->ipv4.udp_table;
3509
3510 if (udptable == &udp_table)
3511 return;
3512
3513 kvfree(udptable->hash);
3514 kfree(udptable);
3515}
3516
3517static void __net_init udp_set_table(struct net *net)
3518{
3519 struct udp_table *udptable;
3520 unsigned int hash_entries;
3521 struct net *old_net;
3522
3523 if (net_eq(net, &init_net))
3524 goto fallback;
3525
3526 old_net = current->nsproxy->net_ns;
3527 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3528 if (!hash_entries)
3529 goto fallback;
3530
3531 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3532 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3533 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3534 else
3535 hash_entries = roundup_pow_of_two(hash_entries);
3536
3537 udptable = udp_pernet_table_alloc(hash_entries);
3538 if (udptable) {
3539 net->ipv4.udp_table = udptable;
3540 } else {
3541 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3542 "for a netns, fallback to the global one\n",
3543 hash_entries);
3544fallback:
3545 net->ipv4.udp_table = &udp_table;
3546 }
3547}
3548
3549static int __net_init udp_pernet_init(struct net *net)
3550{
3551 udp_sysctl_init(net);
3552 udp_set_table(net);
3553
3554 return 0;
3555}
3556
3557static void __net_exit udp_pernet_exit(struct net *net)
3558{
3559 udp_pernet_table_free(net);
3560}
3561
3562static struct pernet_operations __net_initdata udp_sysctl_ops = {
3563 .init = udp_pernet_init,
3564 .exit = udp_pernet_exit,
3565};
3566
3567#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3568DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3569 struct udp_sock *udp_sk, uid_t uid, int bucket)
3570
3571static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3572 unsigned int new_batch_sz)
3573{
3574 struct sock **new_batch;
3575
3576 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3577 GFP_USER | __GFP_NOWARN);
3578 if (!new_batch)
3579 return -ENOMEM;
3580
3581 bpf_iter_udp_put_batch(iter);
3582 kvfree(iter->batch);
3583 iter->batch = new_batch;
3584 iter->max_sk = new_batch_sz;
3585
3586 return 0;
3587}
3588
3589#define INIT_BATCH_SZ 16
3590
3591static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3592{
3593 struct bpf_udp_iter_state *iter = priv_data;
3594 int ret;
3595
3596 ret = bpf_iter_init_seq_net(priv_data, aux);
3597 if (ret)
3598 return ret;
3599
3600 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3601 if (ret)
3602 bpf_iter_fini_seq_net(priv_data);
3603
3604 return ret;
3605}
3606
3607static void bpf_iter_fini_udp(void *priv_data)
3608{
3609 struct bpf_udp_iter_state *iter = priv_data;
3610
3611 bpf_iter_fini_seq_net(priv_data);
3612 kvfree(iter->batch);
3613}
3614
3615static const struct bpf_iter_seq_info udp_seq_info = {
3616 .seq_ops = &bpf_iter_udp_seq_ops,
3617 .init_seq_private = bpf_iter_init_udp,
3618 .fini_seq_private = bpf_iter_fini_udp,
3619 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3620};
3621
3622static struct bpf_iter_reg udp_reg_info = {
3623 .target = "udp",
3624 .ctx_arg_info_size = 1,
3625 .ctx_arg_info = {
3626 { offsetof(struct bpf_iter__udp, udp_sk),
3627 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3628 },
3629 .seq_info = &udp_seq_info,
3630};
3631
3632static void __init bpf_iter_register(void)
3633{
3634 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3635 if (bpf_iter_reg_target(&udp_reg_info))
3636 pr_warn("Warning: could not register bpf iterator udp\n");
3637}
3638#endif
3639
3640void __init udp_init(void)
3641{
3642 unsigned long limit;
3643 unsigned int i;
3644
3645 udp_table_init(&udp_table, "UDP");
3646 limit = nr_free_buffer_pages() / 8;
3647 limit = max(limit, 128UL);
3648 sysctl_udp_mem[0] = limit / 4 * 3;
3649 sysctl_udp_mem[1] = limit;
3650 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3651
3652 /* 16 spinlocks per cpu */
3653 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3654 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3655 GFP_KERNEL);
3656 if (!udp_busylocks)
3657 panic("UDP: failed to alloc udp_busylocks\n");
3658 for (i = 0; i < (1U << udp_busylocks_log); i++)
3659 spin_lock_init(udp_busylocks + i);
3660
3661 if (register_pernet_subsys(&udp_sysctl_ops))
3662 panic("UDP: failed to init sysctl parameters.\n");
3663
3664#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3665 bpf_iter_register();
3666#endif
3667}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * The User Datagram Protocol (UDP).
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
13 * Hirokazu Takahashi, <taka@valinux.co.jp>
14 *
15 * Fixes:
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
29 * does NOT close.
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
58 * for connect.
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * datagrams.
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
72 * James Chapman : Add L2TP encapsulation type.
73 */
74
75#define pr_fmt(fmt) "UDP: " fmt
76
77#include <linux/uaccess.h>
78#include <asm/ioctls.h>
79#include <linux/memblock.h>
80#include <linux/highmem.h>
81#include <linux/swap.h>
82#include <linux/types.h>
83#include <linux/fcntl.h>
84#include <linux/module.h>
85#include <linux/socket.h>
86#include <linux/sockios.h>
87#include <linux/igmp.h>
88#include <linux/inetdevice.h>
89#include <linux/in.h>
90#include <linux/errno.h>
91#include <linux/timer.h>
92#include <linux/mm.h>
93#include <linux/inet.h>
94#include <linux/netdevice.h>
95#include <linux/slab.h>
96#include <net/tcp_states.h>
97#include <linux/skbuff.h>
98#include <linux/proc_fs.h>
99#include <linux/seq_file.h>
100#include <net/net_namespace.h>
101#include <net/icmp.h>
102#include <net/inet_hashtables.h>
103#include <net/ip_tunnels.h>
104#include <net/route.h>
105#include <net/checksum.h>
106#include <net/xfrm.h>
107#include <trace/events/udp.h>
108#include <linux/static_key.h>
109#include <trace/events/skb.h>
110#include <net/busy_poll.h>
111#include "udp_impl.h"
112#include <net/sock_reuseport.h>
113#include <net/addrconf.h>
114#include <net/udp_tunnel.h>
115
116struct udp_table udp_table __read_mostly;
117EXPORT_SYMBOL(udp_table);
118
119long sysctl_udp_mem[3] __read_mostly;
120EXPORT_SYMBOL(sysctl_udp_mem);
121
122atomic_long_t udp_memory_allocated;
123EXPORT_SYMBOL(udp_memory_allocated);
124
125#define MAX_UDP_PORTS 65536
126#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
127
128static int udp_lib_lport_inuse(struct net *net, __u16 num,
129 const struct udp_hslot *hslot,
130 unsigned long *bitmap,
131 struct sock *sk, unsigned int log)
132{
133 struct sock *sk2;
134 kuid_t uid = sock_i_uid(sk);
135
136 sk_for_each(sk2, &hslot->head) {
137 if (net_eq(sock_net(sk2), net) &&
138 sk2 != sk &&
139 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
140 (!sk2->sk_reuse || !sk->sk_reuse) &&
141 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
142 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
143 inet_rcv_saddr_equal(sk, sk2, true)) {
144 if (sk2->sk_reuseport && sk->sk_reuseport &&
145 !rcu_access_pointer(sk->sk_reuseport_cb) &&
146 uid_eq(uid, sock_i_uid(sk2))) {
147 if (!bitmap)
148 return 0;
149 } else {
150 if (!bitmap)
151 return 1;
152 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
153 bitmap);
154 }
155 }
156 }
157 return 0;
158}
159
160/*
161 * Note: we still hold spinlock of primary hash chain, so no other writer
162 * can insert/delete a socket with local_port == num
163 */
164static int udp_lib_lport_inuse2(struct net *net, __u16 num,
165 struct udp_hslot *hslot2,
166 struct sock *sk)
167{
168 struct sock *sk2;
169 kuid_t uid = sock_i_uid(sk);
170 int res = 0;
171
172 spin_lock(&hslot2->lock);
173 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
174 if (net_eq(sock_net(sk2), net) &&
175 sk2 != sk &&
176 (udp_sk(sk2)->udp_port_hash == num) &&
177 (!sk2->sk_reuse || !sk->sk_reuse) &&
178 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
179 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
180 inet_rcv_saddr_equal(sk, sk2, true)) {
181 if (sk2->sk_reuseport && sk->sk_reuseport &&
182 !rcu_access_pointer(sk->sk_reuseport_cb) &&
183 uid_eq(uid, sock_i_uid(sk2))) {
184 res = 0;
185 } else {
186 res = 1;
187 }
188 break;
189 }
190 }
191 spin_unlock(&hslot2->lock);
192 return res;
193}
194
195static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
196{
197 struct net *net = sock_net(sk);
198 kuid_t uid = sock_i_uid(sk);
199 struct sock *sk2;
200
201 sk_for_each(sk2, &hslot->head) {
202 if (net_eq(sock_net(sk2), net) &&
203 sk2 != sk &&
204 sk2->sk_family == sk->sk_family &&
205 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
206 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
207 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
208 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
209 inet_rcv_saddr_equal(sk, sk2, false)) {
210 return reuseport_add_sock(sk, sk2,
211 inet_rcv_saddr_any(sk));
212 }
213 }
214
215 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
216}
217
218/**
219 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
220 *
221 * @sk: socket struct in question
222 * @snum: port number to look up
223 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
224 * with NULL address
225 */
226int udp_lib_get_port(struct sock *sk, unsigned short snum,
227 unsigned int hash2_nulladdr)
228{
229 struct udp_hslot *hslot, *hslot2;
230 struct udp_table *udptable = sk->sk_prot->h.udp_table;
231 int error = 1;
232 struct net *net = sock_net(sk);
233
234 if (!snum) {
235 int low, high, remaining;
236 unsigned int rand;
237 unsigned short first, last;
238 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
239
240 inet_get_local_port_range(net, &low, &high);
241 remaining = (high - low) + 1;
242
243 rand = prandom_u32();
244 first = reciprocal_scale(rand, remaining) + low;
245 /*
246 * force rand to be an odd multiple of UDP_HTABLE_SIZE
247 */
248 rand = (rand | 1) * (udptable->mask + 1);
249 last = first + udptable->mask + 1;
250 do {
251 hslot = udp_hashslot(udptable, net, first);
252 bitmap_zero(bitmap, PORTS_PER_CHAIN);
253 spin_lock_bh(&hslot->lock);
254 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
255 udptable->log);
256
257 snum = first;
258 /*
259 * Iterate on all possible values of snum for this hash.
260 * Using steps of an odd multiple of UDP_HTABLE_SIZE
261 * give us randomization and full range coverage.
262 */
263 do {
264 if (low <= snum && snum <= high &&
265 !test_bit(snum >> udptable->log, bitmap) &&
266 !inet_is_local_reserved_port(net, snum))
267 goto found;
268 snum += rand;
269 } while (snum != first);
270 spin_unlock_bh(&hslot->lock);
271 cond_resched();
272 } while (++first != last);
273 goto fail;
274 } else {
275 hslot = udp_hashslot(udptable, net, snum);
276 spin_lock_bh(&hslot->lock);
277 if (hslot->count > 10) {
278 int exist;
279 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
280
281 slot2 &= udptable->mask;
282 hash2_nulladdr &= udptable->mask;
283
284 hslot2 = udp_hashslot2(udptable, slot2);
285 if (hslot->count < hslot2->count)
286 goto scan_primary_hash;
287
288 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
289 if (!exist && (hash2_nulladdr != slot2)) {
290 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
291 exist = udp_lib_lport_inuse2(net, snum, hslot2,
292 sk);
293 }
294 if (exist)
295 goto fail_unlock;
296 else
297 goto found;
298 }
299scan_primary_hash:
300 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
301 goto fail_unlock;
302 }
303found:
304 inet_sk(sk)->inet_num = snum;
305 udp_sk(sk)->udp_port_hash = snum;
306 udp_sk(sk)->udp_portaddr_hash ^= snum;
307 if (sk_unhashed(sk)) {
308 if (sk->sk_reuseport &&
309 udp_reuseport_add_sock(sk, hslot)) {
310 inet_sk(sk)->inet_num = 0;
311 udp_sk(sk)->udp_port_hash = 0;
312 udp_sk(sk)->udp_portaddr_hash ^= snum;
313 goto fail_unlock;
314 }
315
316 sk_add_node_rcu(sk, &hslot->head);
317 hslot->count++;
318 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
319
320 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
321 spin_lock(&hslot2->lock);
322 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
323 sk->sk_family == AF_INET6)
324 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
325 &hslot2->head);
326 else
327 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
328 &hslot2->head);
329 hslot2->count++;
330 spin_unlock(&hslot2->lock);
331 }
332 sock_set_flag(sk, SOCK_RCU_FREE);
333 error = 0;
334fail_unlock:
335 spin_unlock_bh(&hslot->lock);
336fail:
337 return error;
338}
339EXPORT_SYMBOL(udp_lib_get_port);
340
341int udp_v4_get_port(struct sock *sk, unsigned short snum)
342{
343 unsigned int hash2_nulladdr =
344 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
345 unsigned int hash2_partial =
346 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
347
348 /* precompute partial secondary hash */
349 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
350 return udp_lib_get_port(sk, snum, hash2_nulladdr);
351}
352
353static int compute_score(struct sock *sk, struct net *net,
354 __be32 saddr, __be16 sport,
355 __be32 daddr, unsigned short hnum,
356 int dif, int sdif)
357{
358 int score;
359 struct inet_sock *inet;
360 bool dev_match;
361
362 if (!net_eq(sock_net(sk), net) ||
363 udp_sk(sk)->udp_port_hash != hnum ||
364 ipv6_only_sock(sk))
365 return -1;
366
367 if (sk->sk_rcv_saddr != daddr)
368 return -1;
369
370 score = (sk->sk_family == PF_INET) ? 2 : 1;
371
372 inet = inet_sk(sk);
373 if (inet->inet_daddr) {
374 if (inet->inet_daddr != saddr)
375 return -1;
376 score += 4;
377 }
378
379 if (inet->inet_dport) {
380 if (inet->inet_dport != sport)
381 return -1;
382 score += 4;
383 }
384
385 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
386 dif, sdif);
387 if (!dev_match)
388 return -1;
389 score += 4;
390
391 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
392 score++;
393 return score;
394}
395
396static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
397 const __u16 lport, const __be32 faddr,
398 const __be16 fport)
399{
400 static u32 udp_ehash_secret __read_mostly;
401
402 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
403
404 return __inet_ehashfn(laddr, lport, faddr, fport,
405 udp_ehash_secret + net_hash_mix(net));
406}
407
408/* called with rcu_read_lock() */
409static struct sock *udp4_lib_lookup2(struct net *net,
410 __be32 saddr, __be16 sport,
411 __be32 daddr, unsigned int hnum,
412 int dif, int sdif,
413 struct udp_hslot *hslot2,
414 struct sk_buff *skb)
415{
416 struct sock *sk, *result;
417 int score, badness;
418 u32 hash = 0;
419
420 result = NULL;
421 badness = 0;
422 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
423 score = compute_score(sk, net, saddr, sport,
424 daddr, hnum, dif, sdif);
425 if (score > badness) {
426 if (sk->sk_reuseport &&
427 sk->sk_state != TCP_ESTABLISHED) {
428 hash = udp_ehashfn(net, daddr, hnum,
429 saddr, sport);
430 result = reuseport_select_sock(sk, hash, skb,
431 sizeof(struct udphdr));
432 if (result && !reuseport_has_conns(sk, false))
433 return result;
434 }
435 badness = score;
436 result = sk;
437 }
438 }
439 return result;
440}
441
442/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
443 * harder than this. -DaveM
444 */
445struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
446 __be16 sport, __be32 daddr, __be16 dport, int dif,
447 int sdif, struct udp_table *udptable, struct sk_buff *skb)
448{
449 struct sock *result;
450 unsigned short hnum = ntohs(dport);
451 unsigned int hash2, slot2;
452 struct udp_hslot *hslot2;
453
454 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
455 slot2 = hash2 & udptable->mask;
456 hslot2 = &udptable->hash2[slot2];
457
458 result = udp4_lib_lookup2(net, saddr, sport,
459 daddr, hnum, dif, sdif,
460 hslot2, skb);
461 if (!result) {
462 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
463 slot2 = hash2 & udptable->mask;
464 hslot2 = &udptable->hash2[slot2];
465
466 result = udp4_lib_lookup2(net, saddr, sport,
467 htonl(INADDR_ANY), hnum, dif, sdif,
468 hslot2, skb);
469 }
470 if (IS_ERR(result))
471 return NULL;
472 return result;
473}
474EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
475
476static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
477 __be16 sport, __be16 dport,
478 struct udp_table *udptable)
479{
480 const struct iphdr *iph = ip_hdr(skb);
481
482 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
483 iph->daddr, dport, inet_iif(skb),
484 inet_sdif(skb), udptable, skb);
485}
486
487struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
488 __be16 sport, __be16 dport)
489{
490 const struct iphdr *iph = ip_hdr(skb);
491
492 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
493 iph->daddr, dport, inet_iif(skb),
494 inet_sdif(skb), &udp_table, NULL);
495}
496EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
497
498/* Must be called under rcu_read_lock().
499 * Does increment socket refcount.
500 */
501#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
502struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
503 __be32 daddr, __be16 dport, int dif)
504{
505 struct sock *sk;
506
507 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
508 dif, 0, &udp_table, NULL);
509 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
510 sk = NULL;
511 return sk;
512}
513EXPORT_SYMBOL_GPL(udp4_lib_lookup);
514#endif
515
516static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
517 __be16 loc_port, __be32 loc_addr,
518 __be16 rmt_port, __be32 rmt_addr,
519 int dif, int sdif, unsigned short hnum)
520{
521 struct inet_sock *inet = inet_sk(sk);
522
523 if (!net_eq(sock_net(sk), net) ||
524 udp_sk(sk)->udp_port_hash != hnum ||
525 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
526 (inet->inet_dport != rmt_port && inet->inet_dport) ||
527 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
528 ipv6_only_sock(sk) ||
529 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
530 return false;
531 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
532 return false;
533 return true;
534}
535
536DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
537void udp_encap_enable(void)
538{
539 static_branch_inc(&udp_encap_needed_key);
540}
541EXPORT_SYMBOL(udp_encap_enable);
542
543/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
544 * through error handlers in encapsulations looking for a match.
545 */
546static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
547{
548 int i;
549
550 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
551 int (*handler)(struct sk_buff *skb, u32 info);
552 const struct ip_tunnel_encap_ops *encap;
553
554 encap = rcu_dereference(iptun_encaps[i]);
555 if (!encap)
556 continue;
557 handler = encap->err_handler;
558 if (handler && !handler(skb, info))
559 return 0;
560 }
561
562 return -ENOENT;
563}
564
565/* Try to match ICMP errors to UDP tunnels by looking up a socket without
566 * reversing source and destination port: this will match tunnels that force the
567 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
568 * lwtunnels might actually break this assumption by being configured with
569 * different destination ports on endpoints, in this case we won't be able to
570 * trace ICMP messages back to them.
571 *
572 * If this doesn't match any socket, probe tunnels with arbitrary destination
573 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
574 * we've sent packets to won't necessarily match the local destination port.
575 *
576 * Then ask the tunnel implementation to match the error against a valid
577 * association.
578 *
579 * Return an error if we can't find a match, the socket if we need further
580 * processing, zero otherwise.
581 */
582static struct sock *__udp4_lib_err_encap(struct net *net,
583 const struct iphdr *iph,
584 struct udphdr *uh,
585 struct udp_table *udptable,
586 struct sk_buff *skb, u32 info)
587{
588 int network_offset, transport_offset;
589 struct sock *sk;
590
591 network_offset = skb_network_offset(skb);
592 transport_offset = skb_transport_offset(skb);
593
594 /* Network header needs to point to the outer IPv4 header inside ICMP */
595 skb_reset_network_header(skb);
596
597 /* Transport header needs to point to the UDP header */
598 skb_set_transport_header(skb, iph->ihl << 2);
599
600 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
601 iph->saddr, uh->dest, skb->dev->ifindex, 0,
602 udptable, NULL);
603 if (sk) {
604 int (*lookup)(struct sock *sk, struct sk_buff *skb);
605 struct udp_sock *up = udp_sk(sk);
606
607 lookup = READ_ONCE(up->encap_err_lookup);
608 if (!lookup || lookup(sk, skb))
609 sk = NULL;
610 }
611
612 if (!sk)
613 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
614
615 skb_set_transport_header(skb, transport_offset);
616 skb_set_network_header(skb, network_offset);
617
618 return sk;
619}
620
621/*
622 * This routine is called by the ICMP module when it gets some
623 * sort of error condition. If err < 0 then the socket should
624 * be closed and the error returned to the user. If err > 0
625 * it's just the icmp type << 8 | icmp code.
626 * Header points to the ip header of the error packet. We move
627 * on past this. Then (as it used to claim before adjustment)
628 * header points to the first 8 bytes of the udp header. We need
629 * to find the appropriate port.
630 */
631
632int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
633{
634 struct inet_sock *inet;
635 const struct iphdr *iph = (const struct iphdr *)skb->data;
636 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
637 const int type = icmp_hdr(skb)->type;
638 const int code = icmp_hdr(skb)->code;
639 bool tunnel = false;
640 struct sock *sk;
641 int harderr;
642 int err;
643 struct net *net = dev_net(skb->dev);
644
645 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
646 iph->saddr, uh->source, skb->dev->ifindex,
647 inet_sdif(skb), udptable, NULL);
648 if (!sk) {
649 /* No socket for error: try tunnels before discarding */
650 sk = ERR_PTR(-ENOENT);
651 if (static_branch_unlikely(&udp_encap_needed_key)) {
652 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
653 info);
654 if (!sk)
655 return 0;
656 }
657
658 if (IS_ERR(sk)) {
659 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
660 return PTR_ERR(sk);
661 }
662
663 tunnel = true;
664 }
665
666 err = 0;
667 harderr = 0;
668 inet = inet_sk(sk);
669
670 switch (type) {
671 default:
672 case ICMP_TIME_EXCEEDED:
673 err = EHOSTUNREACH;
674 break;
675 case ICMP_SOURCE_QUENCH:
676 goto out;
677 case ICMP_PARAMETERPROB:
678 err = EPROTO;
679 harderr = 1;
680 break;
681 case ICMP_DEST_UNREACH:
682 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
683 ipv4_sk_update_pmtu(skb, sk, info);
684 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
685 err = EMSGSIZE;
686 harderr = 1;
687 break;
688 }
689 goto out;
690 }
691 err = EHOSTUNREACH;
692 if (code <= NR_ICMP_UNREACH) {
693 harderr = icmp_err_convert[code].fatal;
694 err = icmp_err_convert[code].errno;
695 }
696 break;
697 case ICMP_REDIRECT:
698 ipv4_sk_redirect(skb, sk);
699 goto out;
700 }
701
702 /*
703 * RFC1122: OK. Passes ICMP errors back to application, as per
704 * 4.1.3.3.
705 */
706 if (tunnel) {
707 /* ...not for tunnels though: we don't have a sending socket */
708 goto out;
709 }
710 if (!inet->recverr) {
711 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
712 goto out;
713 } else
714 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
715
716 sk->sk_err = err;
717 sk->sk_error_report(sk);
718out:
719 return 0;
720}
721
722int udp_err(struct sk_buff *skb, u32 info)
723{
724 return __udp4_lib_err(skb, info, &udp_table);
725}
726
727/*
728 * Throw away all pending data and cancel the corking. Socket is locked.
729 */
730void udp_flush_pending_frames(struct sock *sk)
731{
732 struct udp_sock *up = udp_sk(sk);
733
734 if (up->pending) {
735 up->len = 0;
736 up->pending = 0;
737 ip_flush_pending_frames(sk);
738 }
739}
740EXPORT_SYMBOL(udp_flush_pending_frames);
741
742/**
743 * udp4_hwcsum - handle outgoing HW checksumming
744 * @skb: sk_buff containing the filled-in UDP header
745 * (checksum field must be zeroed out)
746 * @src: source IP address
747 * @dst: destination IP address
748 */
749void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
750{
751 struct udphdr *uh = udp_hdr(skb);
752 int offset = skb_transport_offset(skb);
753 int len = skb->len - offset;
754 int hlen = len;
755 __wsum csum = 0;
756
757 if (!skb_has_frag_list(skb)) {
758 /*
759 * Only one fragment on the socket.
760 */
761 skb->csum_start = skb_transport_header(skb) - skb->head;
762 skb->csum_offset = offsetof(struct udphdr, check);
763 uh->check = ~csum_tcpudp_magic(src, dst, len,
764 IPPROTO_UDP, 0);
765 } else {
766 struct sk_buff *frags;
767
768 /*
769 * HW-checksum won't work as there are two or more
770 * fragments on the socket so that all csums of sk_buffs
771 * should be together
772 */
773 skb_walk_frags(skb, frags) {
774 csum = csum_add(csum, frags->csum);
775 hlen -= frags->len;
776 }
777
778 csum = skb_checksum(skb, offset, hlen, csum);
779 skb->ip_summed = CHECKSUM_NONE;
780
781 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
782 if (uh->check == 0)
783 uh->check = CSUM_MANGLED_0;
784 }
785}
786EXPORT_SYMBOL_GPL(udp4_hwcsum);
787
788/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
789 * for the simple case like when setting the checksum for a UDP tunnel.
790 */
791void udp_set_csum(bool nocheck, struct sk_buff *skb,
792 __be32 saddr, __be32 daddr, int len)
793{
794 struct udphdr *uh = udp_hdr(skb);
795
796 if (nocheck) {
797 uh->check = 0;
798 } else if (skb_is_gso(skb)) {
799 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
800 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
801 uh->check = 0;
802 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
803 if (uh->check == 0)
804 uh->check = CSUM_MANGLED_0;
805 } else {
806 skb->ip_summed = CHECKSUM_PARTIAL;
807 skb->csum_start = skb_transport_header(skb) - skb->head;
808 skb->csum_offset = offsetof(struct udphdr, check);
809 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
810 }
811}
812EXPORT_SYMBOL(udp_set_csum);
813
814static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
815 struct inet_cork *cork)
816{
817 struct sock *sk = skb->sk;
818 struct inet_sock *inet = inet_sk(sk);
819 struct udphdr *uh;
820 int err = 0;
821 int is_udplite = IS_UDPLITE(sk);
822 int offset = skb_transport_offset(skb);
823 int len = skb->len - offset;
824 int datalen = len - sizeof(*uh);
825 __wsum csum = 0;
826
827 /*
828 * Create a UDP header
829 */
830 uh = udp_hdr(skb);
831 uh->source = inet->inet_sport;
832 uh->dest = fl4->fl4_dport;
833 uh->len = htons(len);
834 uh->check = 0;
835
836 if (cork->gso_size) {
837 const int hlen = skb_network_header_len(skb) +
838 sizeof(struct udphdr);
839
840 if (hlen + cork->gso_size > cork->fragsize) {
841 kfree_skb(skb);
842 return -EINVAL;
843 }
844 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
845 kfree_skb(skb);
846 return -EINVAL;
847 }
848 if (sk->sk_no_check_tx) {
849 kfree_skb(skb);
850 return -EINVAL;
851 }
852 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
853 dst_xfrm(skb_dst(skb))) {
854 kfree_skb(skb);
855 return -EIO;
856 }
857
858 if (datalen > cork->gso_size) {
859 skb_shinfo(skb)->gso_size = cork->gso_size;
860 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
861 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
862 cork->gso_size);
863 }
864 goto csum_partial;
865 }
866
867 if (is_udplite) /* UDP-Lite */
868 csum = udplite_csum(skb);
869
870 else if (sk->sk_no_check_tx) { /* UDP csum off */
871
872 skb->ip_summed = CHECKSUM_NONE;
873 goto send;
874
875 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
876csum_partial:
877
878 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
879 goto send;
880
881 } else
882 csum = udp_csum(skb);
883
884 /* add protocol-dependent pseudo-header */
885 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
886 sk->sk_protocol, csum);
887 if (uh->check == 0)
888 uh->check = CSUM_MANGLED_0;
889
890send:
891 err = ip_send_skb(sock_net(sk), skb);
892 if (err) {
893 if (err == -ENOBUFS && !inet->recverr) {
894 UDP_INC_STATS(sock_net(sk),
895 UDP_MIB_SNDBUFERRORS, is_udplite);
896 err = 0;
897 }
898 } else
899 UDP_INC_STATS(sock_net(sk),
900 UDP_MIB_OUTDATAGRAMS, is_udplite);
901 return err;
902}
903
904/*
905 * Push out all pending data as one UDP datagram. Socket is locked.
906 */
907int udp_push_pending_frames(struct sock *sk)
908{
909 struct udp_sock *up = udp_sk(sk);
910 struct inet_sock *inet = inet_sk(sk);
911 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
912 struct sk_buff *skb;
913 int err = 0;
914
915 skb = ip_finish_skb(sk, fl4);
916 if (!skb)
917 goto out;
918
919 err = udp_send_skb(skb, fl4, &inet->cork.base);
920
921out:
922 up->len = 0;
923 up->pending = 0;
924 return err;
925}
926EXPORT_SYMBOL(udp_push_pending_frames);
927
928static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
929{
930 switch (cmsg->cmsg_type) {
931 case UDP_SEGMENT:
932 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
933 return -EINVAL;
934 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
935 return 0;
936 default:
937 return -EINVAL;
938 }
939}
940
941int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
942{
943 struct cmsghdr *cmsg;
944 bool need_ip = false;
945 int err;
946
947 for_each_cmsghdr(cmsg, msg) {
948 if (!CMSG_OK(msg, cmsg))
949 return -EINVAL;
950
951 if (cmsg->cmsg_level != SOL_UDP) {
952 need_ip = true;
953 continue;
954 }
955
956 err = __udp_cmsg_send(cmsg, gso_size);
957 if (err)
958 return err;
959 }
960
961 return need_ip;
962}
963EXPORT_SYMBOL_GPL(udp_cmsg_send);
964
965int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
966{
967 struct inet_sock *inet = inet_sk(sk);
968 struct udp_sock *up = udp_sk(sk);
969 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
970 struct flowi4 fl4_stack;
971 struct flowi4 *fl4;
972 int ulen = len;
973 struct ipcm_cookie ipc;
974 struct rtable *rt = NULL;
975 int free = 0;
976 int connected = 0;
977 __be32 daddr, faddr, saddr;
978 __be16 dport;
979 u8 tos;
980 int err, is_udplite = IS_UDPLITE(sk);
981 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
982 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
983 struct sk_buff *skb;
984 struct ip_options_data opt_copy;
985
986 if (len > 0xFFFF)
987 return -EMSGSIZE;
988
989 /*
990 * Check the flags.
991 */
992
993 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
994 return -EOPNOTSUPP;
995
996 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
997
998 fl4 = &inet->cork.fl.u.ip4;
999 if (up->pending) {
1000 /*
1001 * There are pending frames.
1002 * The socket lock must be held while it's corked.
1003 */
1004 lock_sock(sk);
1005 if (likely(up->pending)) {
1006 if (unlikely(up->pending != AF_INET)) {
1007 release_sock(sk);
1008 return -EINVAL;
1009 }
1010 goto do_append_data;
1011 }
1012 release_sock(sk);
1013 }
1014 ulen += sizeof(struct udphdr);
1015
1016 /*
1017 * Get and verify the address.
1018 */
1019 if (usin) {
1020 if (msg->msg_namelen < sizeof(*usin))
1021 return -EINVAL;
1022 if (usin->sin_family != AF_INET) {
1023 if (usin->sin_family != AF_UNSPEC)
1024 return -EAFNOSUPPORT;
1025 }
1026
1027 daddr = usin->sin_addr.s_addr;
1028 dport = usin->sin_port;
1029 if (dport == 0)
1030 return -EINVAL;
1031 } else {
1032 if (sk->sk_state != TCP_ESTABLISHED)
1033 return -EDESTADDRREQ;
1034 daddr = inet->inet_daddr;
1035 dport = inet->inet_dport;
1036 /* Open fast path for connected socket.
1037 Route will not be used, if at least one option is set.
1038 */
1039 connected = 1;
1040 }
1041
1042 ipcm_init_sk(&ipc, inet);
1043 ipc.gso_size = up->gso_size;
1044
1045 if (msg->msg_controllen) {
1046 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1047 if (err > 0)
1048 err = ip_cmsg_send(sk, msg, &ipc,
1049 sk->sk_family == AF_INET6);
1050 if (unlikely(err < 0)) {
1051 kfree(ipc.opt);
1052 return err;
1053 }
1054 if (ipc.opt)
1055 free = 1;
1056 connected = 0;
1057 }
1058 if (!ipc.opt) {
1059 struct ip_options_rcu *inet_opt;
1060
1061 rcu_read_lock();
1062 inet_opt = rcu_dereference(inet->inet_opt);
1063 if (inet_opt) {
1064 memcpy(&opt_copy, inet_opt,
1065 sizeof(*inet_opt) + inet_opt->opt.optlen);
1066 ipc.opt = &opt_copy.opt;
1067 }
1068 rcu_read_unlock();
1069 }
1070
1071 if (cgroup_bpf_enabled && !connected) {
1072 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1073 (struct sockaddr *)usin, &ipc.addr);
1074 if (err)
1075 goto out_free;
1076 if (usin) {
1077 if (usin->sin_port == 0) {
1078 /* BPF program set invalid port. Reject it. */
1079 err = -EINVAL;
1080 goto out_free;
1081 }
1082 daddr = usin->sin_addr.s_addr;
1083 dport = usin->sin_port;
1084 }
1085 }
1086
1087 saddr = ipc.addr;
1088 ipc.addr = faddr = daddr;
1089
1090 if (ipc.opt && ipc.opt->opt.srr) {
1091 if (!daddr) {
1092 err = -EINVAL;
1093 goto out_free;
1094 }
1095 faddr = ipc.opt->opt.faddr;
1096 connected = 0;
1097 }
1098 tos = get_rttos(&ipc, inet);
1099 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1100 (msg->msg_flags & MSG_DONTROUTE) ||
1101 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1102 tos |= RTO_ONLINK;
1103 connected = 0;
1104 }
1105
1106 if (ipv4_is_multicast(daddr)) {
1107 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1108 ipc.oif = inet->mc_index;
1109 if (!saddr)
1110 saddr = inet->mc_addr;
1111 connected = 0;
1112 } else if (!ipc.oif) {
1113 ipc.oif = inet->uc_index;
1114 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1115 /* oif is set, packet is to local broadcast and
1116 * and uc_index is set. oif is most likely set
1117 * by sk_bound_dev_if. If uc_index != oif check if the
1118 * oif is an L3 master and uc_index is an L3 slave.
1119 * If so, we want to allow the send using the uc_index.
1120 */
1121 if (ipc.oif != inet->uc_index &&
1122 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1123 inet->uc_index)) {
1124 ipc.oif = inet->uc_index;
1125 }
1126 }
1127
1128 if (connected)
1129 rt = (struct rtable *)sk_dst_check(sk, 0);
1130
1131 if (!rt) {
1132 struct net *net = sock_net(sk);
1133 __u8 flow_flags = inet_sk_flowi_flags(sk);
1134
1135 fl4 = &fl4_stack;
1136
1137 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1138 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1139 flow_flags,
1140 faddr, saddr, dport, inet->inet_sport,
1141 sk->sk_uid);
1142
1143 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1144 rt = ip_route_output_flow(net, fl4, sk);
1145 if (IS_ERR(rt)) {
1146 err = PTR_ERR(rt);
1147 rt = NULL;
1148 if (err == -ENETUNREACH)
1149 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1150 goto out;
1151 }
1152
1153 err = -EACCES;
1154 if ((rt->rt_flags & RTCF_BROADCAST) &&
1155 !sock_flag(sk, SOCK_BROADCAST))
1156 goto out;
1157 if (connected)
1158 sk_dst_set(sk, dst_clone(&rt->dst));
1159 }
1160
1161 if (msg->msg_flags&MSG_CONFIRM)
1162 goto do_confirm;
1163back_from_confirm:
1164
1165 saddr = fl4->saddr;
1166 if (!ipc.addr)
1167 daddr = ipc.addr = fl4->daddr;
1168
1169 /* Lockless fast path for the non-corking case. */
1170 if (!corkreq) {
1171 struct inet_cork cork;
1172
1173 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1174 sizeof(struct udphdr), &ipc, &rt,
1175 &cork, msg->msg_flags);
1176 err = PTR_ERR(skb);
1177 if (!IS_ERR_OR_NULL(skb))
1178 err = udp_send_skb(skb, fl4, &cork);
1179 goto out;
1180 }
1181
1182 lock_sock(sk);
1183 if (unlikely(up->pending)) {
1184 /* The socket is already corked while preparing it. */
1185 /* ... which is an evident application bug. --ANK */
1186 release_sock(sk);
1187
1188 net_dbg_ratelimited("socket already corked\n");
1189 err = -EINVAL;
1190 goto out;
1191 }
1192 /*
1193 * Now cork the socket to pend data.
1194 */
1195 fl4 = &inet->cork.fl.u.ip4;
1196 fl4->daddr = daddr;
1197 fl4->saddr = saddr;
1198 fl4->fl4_dport = dport;
1199 fl4->fl4_sport = inet->inet_sport;
1200 up->pending = AF_INET;
1201
1202do_append_data:
1203 up->len += ulen;
1204 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1205 sizeof(struct udphdr), &ipc, &rt,
1206 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1207 if (err)
1208 udp_flush_pending_frames(sk);
1209 else if (!corkreq)
1210 err = udp_push_pending_frames(sk);
1211 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1212 up->pending = 0;
1213 release_sock(sk);
1214
1215out:
1216 ip_rt_put(rt);
1217out_free:
1218 if (free)
1219 kfree(ipc.opt);
1220 if (!err)
1221 return len;
1222 /*
1223 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1224 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1225 * we don't have a good statistic (IpOutDiscards but it can be too many
1226 * things). We could add another new stat but at least for now that
1227 * seems like overkill.
1228 */
1229 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1230 UDP_INC_STATS(sock_net(sk),
1231 UDP_MIB_SNDBUFERRORS, is_udplite);
1232 }
1233 return err;
1234
1235do_confirm:
1236 if (msg->msg_flags & MSG_PROBE)
1237 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1238 if (!(msg->msg_flags&MSG_PROBE) || len)
1239 goto back_from_confirm;
1240 err = 0;
1241 goto out;
1242}
1243EXPORT_SYMBOL(udp_sendmsg);
1244
1245int udp_sendpage(struct sock *sk, struct page *page, int offset,
1246 size_t size, int flags)
1247{
1248 struct inet_sock *inet = inet_sk(sk);
1249 struct udp_sock *up = udp_sk(sk);
1250 int ret;
1251
1252 if (flags & MSG_SENDPAGE_NOTLAST)
1253 flags |= MSG_MORE;
1254
1255 if (!up->pending) {
1256 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1257
1258 /* Call udp_sendmsg to specify destination address which
1259 * sendpage interface can't pass.
1260 * This will succeed only when the socket is connected.
1261 */
1262 ret = udp_sendmsg(sk, &msg, 0);
1263 if (ret < 0)
1264 return ret;
1265 }
1266
1267 lock_sock(sk);
1268
1269 if (unlikely(!up->pending)) {
1270 release_sock(sk);
1271
1272 net_dbg_ratelimited("cork failed\n");
1273 return -EINVAL;
1274 }
1275
1276 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1277 page, offset, size, flags);
1278 if (ret == -EOPNOTSUPP) {
1279 release_sock(sk);
1280 return sock_no_sendpage(sk->sk_socket, page, offset,
1281 size, flags);
1282 }
1283 if (ret < 0) {
1284 udp_flush_pending_frames(sk);
1285 goto out;
1286 }
1287
1288 up->len += size;
1289 if (!(up->corkflag || (flags&MSG_MORE)))
1290 ret = udp_push_pending_frames(sk);
1291 if (!ret)
1292 ret = size;
1293out:
1294 release_sock(sk);
1295 return ret;
1296}
1297
1298#define UDP_SKB_IS_STATELESS 0x80000000
1299
1300/* all head states (dst, sk, nf conntrack) except skb extensions are
1301 * cleared by udp_rcv().
1302 *
1303 * We need to preserve secpath, if present, to eventually process
1304 * IP_CMSG_PASSSEC at recvmsg() time.
1305 *
1306 * Other extensions can be cleared.
1307 */
1308static bool udp_try_make_stateless(struct sk_buff *skb)
1309{
1310 if (!skb_has_extensions(skb))
1311 return true;
1312
1313 if (!secpath_exists(skb)) {
1314 skb_ext_reset(skb);
1315 return true;
1316 }
1317
1318 return false;
1319}
1320
1321static void udp_set_dev_scratch(struct sk_buff *skb)
1322{
1323 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1324
1325 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1326 scratch->_tsize_state = skb->truesize;
1327#if BITS_PER_LONG == 64
1328 scratch->len = skb->len;
1329 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1330 scratch->is_linear = !skb_is_nonlinear(skb);
1331#endif
1332 if (udp_try_make_stateless(skb))
1333 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1334}
1335
1336static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1337{
1338 /* We come here after udp_lib_checksum_complete() returned 0.
1339 * This means that __skb_checksum_complete() might have
1340 * set skb->csum_valid to 1.
1341 * On 64bit platforms, we can set csum_unnecessary
1342 * to true, but only if the skb is not shared.
1343 */
1344#if BITS_PER_LONG == 64
1345 if (!skb_shared(skb))
1346 udp_skb_scratch(skb)->csum_unnecessary = true;
1347#endif
1348}
1349
1350static int udp_skb_truesize(struct sk_buff *skb)
1351{
1352 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1353}
1354
1355static bool udp_skb_has_head_state(struct sk_buff *skb)
1356{
1357 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1358}
1359
1360/* fully reclaim rmem/fwd memory allocated for skb */
1361static void udp_rmem_release(struct sock *sk, int size, int partial,
1362 bool rx_queue_lock_held)
1363{
1364 struct udp_sock *up = udp_sk(sk);
1365 struct sk_buff_head *sk_queue;
1366 int amt;
1367
1368 if (likely(partial)) {
1369 up->forward_deficit += size;
1370 size = up->forward_deficit;
1371 if (size < (sk->sk_rcvbuf >> 2))
1372 return;
1373 } else {
1374 size += up->forward_deficit;
1375 }
1376 up->forward_deficit = 0;
1377
1378 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1379 * if the called don't held it already
1380 */
1381 sk_queue = &sk->sk_receive_queue;
1382 if (!rx_queue_lock_held)
1383 spin_lock(&sk_queue->lock);
1384
1385
1386 sk->sk_forward_alloc += size;
1387 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1388 sk->sk_forward_alloc -= amt;
1389
1390 if (amt)
1391 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1392
1393 atomic_sub(size, &sk->sk_rmem_alloc);
1394
1395 /* this can save us from acquiring the rx queue lock on next receive */
1396 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1397
1398 if (!rx_queue_lock_held)
1399 spin_unlock(&sk_queue->lock);
1400}
1401
1402/* Note: called with reader_queue.lock held.
1403 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1404 * This avoids a cache line miss while receive_queue lock is held.
1405 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1406 */
1407void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1408{
1409 prefetch(&skb->data);
1410 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1411}
1412EXPORT_SYMBOL(udp_skb_destructor);
1413
1414/* as above, but the caller held the rx queue lock, too */
1415static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1416{
1417 prefetch(&skb->data);
1418 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1419}
1420
1421/* Idea of busylocks is to let producers grab an extra spinlock
1422 * to relieve pressure on the receive_queue spinlock shared by consumer.
1423 * Under flood, this means that only one producer can be in line
1424 * trying to acquire the receive_queue spinlock.
1425 * These busylock can be allocated on a per cpu manner, instead of a
1426 * per socket one (that would consume a cache line per socket)
1427 */
1428static int udp_busylocks_log __read_mostly;
1429static spinlock_t *udp_busylocks __read_mostly;
1430
1431static spinlock_t *busylock_acquire(void *ptr)
1432{
1433 spinlock_t *busy;
1434
1435 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1436 spin_lock(busy);
1437 return busy;
1438}
1439
1440static void busylock_release(spinlock_t *busy)
1441{
1442 if (busy)
1443 spin_unlock(busy);
1444}
1445
1446int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1447{
1448 struct sk_buff_head *list = &sk->sk_receive_queue;
1449 int rmem, delta, amt, err = -ENOMEM;
1450 spinlock_t *busy = NULL;
1451 int size;
1452
1453 /* try to avoid the costly atomic add/sub pair when the receive
1454 * queue is full; always allow at least a packet
1455 */
1456 rmem = atomic_read(&sk->sk_rmem_alloc);
1457 if (rmem > sk->sk_rcvbuf)
1458 goto drop;
1459
1460 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1461 * having linear skbs :
1462 * - Reduce memory overhead and thus increase receive queue capacity
1463 * - Less cache line misses at copyout() time
1464 * - Less work at consume_skb() (less alien page frag freeing)
1465 */
1466 if (rmem > (sk->sk_rcvbuf >> 1)) {
1467 skb_condense(skb);
1468
1469 busy = busylock_acquire(sk);
1470 }
1471 size = skb->truesize;
1472 udp_set_dev_scratch(skb);
1473
1474 /* we drop only if the receive buf is full and the receive
1475 * queue contains some other skb
1476 */
1477 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1478 if (rmem > (size + sk->sk_rcvbuf))
1479 goto uncharge_drop;
1480
1481 spin_lock(&list->lock);
1482 if (size >= sk->sk_forward_alloc) {
1483 amt = sk_mem_pages(size);
1484 delta = amt << SK_MEM_QUANTUM_SHIFT;
1485 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1486 err = -ENOBUFS;
1487 spin_unlock(&list->lock);
1488 goto uncharge_drop;
1489 }
1490
1491 sk->sk_forward_alloc += delta;
1492 }
1493
1494 sk->sk_forward_alloc -= size;
1495
1496 /* no need to setup a destructor, we will explicitly release the
1497 * forward allocated memory on dequeue
1498 */
1499 sock_skb_set_dropcount(sk, skb);
1500
1501 __skb_queue_tail(list, skb);
1502 spin_unlock(&list->lock);
1503
1504 if (!sock_flag(sk, SOCK_DEAD))
1505 sk->sk_data_ready(sk);
1506
1507 busylock_release(busy);
1508 return 0;
1509
1510uncharge_drop:
1511 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1512
1513drop:
1514 atomic_inc(&sk->sk_drops);
1515 busylock_release(busy);
1516 return err;
1517}
1518EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1519
1520void udp_destruct_sock(struct sock *sk)
1521{
1522 /* reclaim completely the forward allocated memory */
1523 struct udp_sock *up = udp_sk(sk);
1524 unsigned int total = 0;
1525 struct sk_buff *skb;
1526
1527 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1528 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1529 total += skb->truesize;
1530 kfree_skb(skb);
1531 }
1532 udp_rmem_release(sk, total, 0, true);
1533
1534 inet_sock_destruct(sk);
1535}
1536EXPORT_SYMBOL_GPL(udp_destruct_sock);
1537
1538int udp_init_sock(struct sock *sk)
1539{
1540 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1541 sk->sk_destruct = udp_destruct_sock;
1542 return 0;
1543}
1544EXPORT_SYMBOL_GPL(udp_init_sock);
1545
1546void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1547{
1548 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1549 bool slow = lock_sock_fast(sk);
1550
1551 sk_peek_offset_bwd(sk, len);
1552 unlock_sock_fast(sk, slow);
1553 }
1554
1555 if (!skb_unref(skb))
1556 return;
1557
1558 /* In the more common cases we cleared the head states previously,
1559 * see __udp_queue_rcv_skb().
1560 */
1561 if (unlikely(udp_skb_has_head_state(skb)))
1562 skb_release_head_state(skb);
1563 __consume_stateless_skb(skb);
1564}
1565EXPORT_SYMBOL_GPL(skb_consume_udp);
1566
1567static struct sk_buff *__first_packet_length(struct sock *sk,
1568 struct sk_buff_head *rcvq,
1569 int *total)
1570{
1571 struct sk_buff *skb;
1572
1573 while ((skb = skb_peek(rcvq)) != NULL) {
1574 if (udp_lib_checksum_complete(skb)) {
1575 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1576 IS_UDPLITE(sk));
1577 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1578 IS_UDPLITE(sk));
1579 atomic_inc(&sk->sk_drops);
1580 __skb_unlink(skb, rcvq);
1581 *total += skb->truesize;
1582 kfree_skb(skb);
1583 } else {
1584 udp_skb_csum_unnecessary_set(skb);
1585 break;
1586 }
1587 }
1588 return skb;
1589}
1590
1591/**
1592 * first_packet_length - return length of first packet in receive queue
1593 * @sk: socket
1594 *
1595 * Drops all bad checksum frames, until a valid one is found.
1596 * Returns the length of found skb, or -1 if none is found.
1597 */
1598static int first_packet_length(struct sock *sk)
1599{
1600 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1601 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1602 struct sk_buff *skb;
1603 int total = 0;
1604 int res;
1605
1606 spin_lock_bh(&rcvq->lock);
1607 skb = __first_packet_length(sk, rcvq, &total);
1608 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1609 spin_lock(&sk_queue->lock);
1610 skb_queue_splice_tail_init(sk_queue, rcvq);
1611 spin_unlock(&sk_queue->lock);
1612
1613 skb = __first_packet_length(sk, rcvq, &total);
1614 }
1615 res = skb ? skb->len : -1;
1616 if (total)
1617 udp_rmem_release(sk, total, 1, false);
1618 spin_unlock_bh(&rcvq->lock);
1619 return res;
1620}
1621
1622/*
1623 * IOCTL requests applicable to the UDP protocol
1624 */
1625
1626int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1627{
1628 switch (cmd) {
1629 case SIOCOUTQ:
1630 {
1631 int amount = sk_wmem_alloc_get(sk);
1632
1633 return put_user(amount, (int __user *)arg);
1634 }
1635
1636 case SIOCINQ:
1637 {
1638 int amount = max_t(int, 0, first_packet_length(sk));
1639
1640 return put_user(amount, (int __user *)arg);
1641 }
1642
1643 default:
1644 return -ENOIOCTLCMD;
1645 }
1646
1647 return 0;
1648}
1649EXPORT_SYMBOL(udp_ioctl);
1650
1651struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1652 int noblock, int *off, int *err)
1653{
1654 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1655 struct sk_buff_head *queue;
1656 struct sk_buff *last;
1657 long timeo;
1658 int error;
1659
1660 queue = &udp_sk(sk)->reader_queue;
1661 flags |= noblock ? MSG_DONTWAIT : 0;
1662 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1663 do {
1664 struct sk_buff *skb;
1665
1666 error = sock_error(sk);
1667 if (error)
1668 break;
1669
1670 error = -EAGAIN;
1671 do {
1672 spin_lock_bh(&queue->lock);
1673 skb = __skb_try_recv_from_queue(sk, queue, flags,
1674 udp_skb_destructor,
1675 off, err, &last);
1676 if (skb) {
1677 spin_unlock_bh(&queue->lock);
1678 return skb;
1679 }
1680
1681 if (skb_queue_empty_lockless(sk_queue)) {
1682 spin_unlock_bh(&queue->lock);
1683 goto busy_check;
1684 }
1685
1686 /* refill the reader queue and walk it again
1687 * keep both queues locked to avoid re-acquiring
1688 * the sk_receive_queue lock if fwd memory scheduling
1689 * is needed.
1690 */
1691 spin_lock(&sk_queue->lock);
1692 skb_queue_splice_tail_init(sk_queue, queue);
1693
1694 skb = __skb_try_recv_from_queue(sk, queue, flags,
1695 udp_skb_dtor_locked,
1696 off, err, &last);
1697 spin_unlock(&sk_queue->lock);
1698 spin_unlock_bh(&queue->lock);
1699 if (skb)
1700 return skb;
1701
1702busy_check:
1703 if (!sk_can_busy_loop(sk))
1704 break;
1705
1706 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1707 } while (!skb_queue_empty_lockless(sk_queue));
1708
1709 /* sk_queue is empty, reader_queue may contain peeked packets */
1710 } while (timeo &&
1711 !__skb_wait_for_more_packets(sk, &error, &timeo,
1712 (struct sk_buff *)sk_queue));
1713
1714 *err = error;
1715 return NULL;
1716}
1717EXPORT_SYMBOL(__skb_recv_udp);
1718
1719/*
1720 * This should be easy, if there is something there we
1721 * return it, otherwise we block.
1722 */
1723
1724int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1725 int flags, int *addr_len)
1726{
1727 struct inet_sock *inet = inet_sk(sk);
1728 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1729 struct sk_buff *skb;
1730 unsigned int ulen, copied;
1731 int off, err, peeking = flags & MSG_PEEK;
1732 int is_udplite = IS_UDPLITE(sk);
1733 bool checksum_valid = false;
1734
1735 if (flags & MSG_ERRQUEUE)
1736 return ip_recv_error(sk, msg, len, addr_len);
1737
1738try_again:
1739 off = sk_peek_offset(sk, flags);
1740 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1741 if (!skb)
1742 return err;
1743
1744 ulen = udp_skb_len(skb);
1745 copied = len;
1746 if (copied > ulen - off)
1747 copied = ulen - off;
1748 else if (copied < ulen)
1749 msg->msg_flags |= MSG_TRUNC;
1750
1751 /*
1752 * If checksum is needed at all, try to do it while copying the
1753 * data. If the data is truncated, or if we only want a partial
1754 * coverage checksum (UDP-Lite), do it before the copy.
1755 */
1756
1757 if (copied < ulen || peeking ||
1758 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1759 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1760 !__udp_lib_checksum_complete(skb);
1761 if (!checksum_valid)
1762 goto csum_copy_err;
1763 }
1764
1765 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1766 if (udp_skb_is_linear(skb))
1767 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1768 else
1769 err = skb_copy_datagram_msg(skb, off, msg, copied);
1770 } else {
1771 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1772
1773 if (err == -EINVAL)
1774 goto csum_copy_err;
1775 }
1776
1777 if (unlikely(err)) {
1778 if (!peeking) {
1779 atomic_inc(&sk->sk_drops);
1780 UDP_INC_STATS(sock_net(sk),
1781 UDP_MIB_INERRORS, is_udplite);
1782 }
1783 kfree_skb(skb);
1784 return err;
1785 }
1786
1787 if (!peeking)
1788 UDP_INC_STATS(sock_net(sk),
1789 UDP_MIB_INDATAGRAMS, is_udplite);
1790
1791 sock_recv_ts_and_drops(msg, sk, skb);
1792
1793 /* Copy the address. */
1794 if (sin) {
1795 sin->sin_family = AF_INET;
1796 sin->sin_port = udp_hdr(skb)->source;
1797 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1798 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1799 *addr_len = sizeof(*sin);
1800
1801 if (cgroup_bpf_enabled)
1802 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1803 (struct sockaddr *)sin);
1804 }
1805
1806 if (udp_sk(sk)->gro_enabled)
1807 udp_cmsg_recv(msg, sk, skb);
1808
1809 if (inet->cmsg_flags)
1810 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1811
1812 err = copied;
1813 if (flags & MSG_TRUNC)
1814 err = ulen;
1815
1816 skb_consume_udp(sk, skb, peeking ? -err : err);
1817 return err;
1818
1819csum_copy_err:
1820 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1821 udp_skb_destructor)) {
1822 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1823 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1824 }
1825 kfree_skb(skb);
1826
1827 /* starting over for a new packet, but check if we need to yield */
1828 cond_resched();
1829 msg->msg_flags &= ~MSG_TRUNC;
1830 goto try_again;
1831}
1832
1833int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1834{
1835 /* This check is replicated from __ip4_datagram_connect() and
1836 * intended to prevent BPF program called below from accessing bytes
1837 * that are out of the bound specified by user in addr_len.
1838 */
1839 if (addr_len < sizeof(struct sockaddr_in))
1840 return -EINVAL;
1841
1842 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1843}
1844EXPORT_SYMBOL(udp_pre_connect);
1845
1846int __udp_disconnect(struct sock *sk, int flags)
1847{
1848 struct inet_sock *inet = inet_sk(sk);
1849 /*
1850 * 1003.1g - break association.
1851 */
1852
1853 sk->sk_state = TCP_CLOSE;
1854 inet->inet_daddr = 0;
1855 inet->inet_dport = 0;
1856 sock_rps_reset_rxhash(sk);
1857 sk->sk_bound_dev_if = 0;
1858 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1859 inet_reset_saddr(sk);
1860
1861 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1862 sk->sk_prot->unhash(sk);
1863 inet->inet_sport = 0;
1864 }
1865 sk_dst_reset(sk);
1866 return 0;
1867}
1868EXPORT_SYMBOL(__udp_disconnect);
1869
1870int udp_disconnect(struct sock *sk, int flags)
1871{
1872 lock_sock(sk);
1873 __udp_disconnect(sk, flags);
1874 release_sock(sk);
1875 return 0;
1876}
1877EXPORT_SYMBOL(udp_disconnect);
1878
1879void udp_lib_unhash(struct sock *sk)
1880{
1881 if (sk_hashed(sk)) {
1882 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1883 struct udp_hslot *hslot, *hslot2;
1884
1885 hslot = udp_hashslot(udptable, sock_net(sk),
1886 udp_sk(sk)->udp_port_hash);
1887 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1888
1889 spin_lock_bh(&hslot->lock);
1890 if (rcu_access_pointer(sk->sk_reuseport_cb))
1891 reuseport_detach_sock(sk);
1892 if (sk_del_node_init_rcu(sk)) {
1893 hslot->count--;
1894 inet_sk(sk)->inet_num = 0;
1895 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1896
1897 spin_lock(&hslot2->lock);
1898 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1899 hslot2->count--;
1900 spin_unlock(&hslot2->lock);
1901 }
1902 spin_unlock_bh(&hslot->lock);
1903 }
1904}
1905EXPORT_SYMBOL(udp_lib_unhash);
1906
1907/*
1908 * inet_rcv_saddr was changed, we must rehash secondary hash
1909 */
1910void udp_lib_rehash(struct sock *sk, u16 newhash)
1911{
1912 if (sk_hashed(sk)) {
1913 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1914 struct udp_hslot *hslot, *hslot2, *nhslot2;
1915
1916 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1917 nhslot2 = udp_hashslot2(udptable, newhash);
1918 udp_sk(sk)->udp_portaddr_hash = newhash;
1919
1920 if (hslot2 != nhslot2 ||
1921 rcu_access_pointer(sk->sk_reuseport_cb)) {
1922 hslot = udp_hashslot(udptable, sock_net(sk),
1923 udp_sk(sk)->udp_port_hash);
1924 /* we must lock primary chain too */
1925 spin_lock_bh(&hslot->lock);
1926 if (rcu_access_pointer(sk->sk_reuseport_cb))
1927 reuseport_detach_sock(sk);
1928
1929 if (hslot2 != nhslot2) {
1930 spin_lock(&hslot2->lock);
1931 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1932 hslot2->count--;
1933 spin_unlock(&hslot2->lock);
1934
1935 spin_lock(&nhslot2->lock);
1936 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1937 &nhslot2->head);
1938 nhslot2->count++;
1939 spin_unlock(&nhslot2->lock);
1940 }
1941
1942 spin_unlock_bh(&hslot->lock);
1943 }
1944 }
1945}
1946EXPORT_SYMBOL(udp_lib_rehash);
1947
1948void udp_v4_rehash(struct sock *sk)
1949{
1950 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1951 inet_sk(sk)->inet_rcv_saddr,
1952 inet_sk(sk)->inet_num);
1953 udp_lib_rehash(sk, new_hash);
1954}
1955
1956static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1957{
1958 int rc;
1959
1960 if (inet_sk(sk)->inet_daddr) {
1961 sock_rps_save_rxhash(sk, skb);
1962 sk_mark_napi_id(sk, skb);
1963 sk_incoming_cpu_update(sk);
1964 } else {
1965 sk_mark_napi_id_once(sk, skb);
1966 }
1967
1968 rc = __udp_enqueue_schedule_skb(sk, skb);
1969 if (rc < 0) {
1970 int is_udplite = IS_UDPLITE(sk);
1971
1972 /* Note that an ENOMEM error is charged twice */
1973 if (rc == -ENOMEM)
1974 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1975 is_udplite);
1976 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1977 kfree_skb(skb);
1978 trace_udp_fail_queue_rcv_skb(rc, sk);
1979 return -1;
1980 }
1981
1982 return 0;
1983}
1984
1985/* returns:
1986 * -1: error
1987 * 0: success
1988 * >0: "udp encap" protocol resubmission
1989 *
1990 * Note that in the success and error cases, the skb is assumed to
1991 * have either been requeued or freed.
1992 */
1993static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
1994{
1995 struct udp_sock *up = udp_sk(sk);
1996 int is_udplite = IS_UDPLITE(sk);
1997
1998 /*
1999 * Charge it to the socket, dropping if the queue is full.
2000 */
2001 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2002 goto drop;
2003 nf_reset_ct(skb);
2004
2005 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2006 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2007
2008 /*
2009 * This is an encapsulation socket so pass the skb to
2010 * the socket's udp_encap_rcv() hook. Otherwise, just
2011 * fall through and pass this up the UDP socket.
2012 * up->encap_rcv() returns the following value:
2013 * =0 if skb was successfully passed to the encap
2014 * handler or was discarded by it.
2015 * >0 if skb should be passed on to UDP.
2016 * <0 if skb should be resubmitted as proto -N
2017 */
2018
2019 /* if we're overly short, let UDP handle it */
2020 encap_rcv = READ_ONCE(up->encap_rcv);
2021 if (encap_rcv) {
2022 int ret;
2023
2024 /* Verify checksum before giving to encap */
2025 if (udp_lib_checksum_complete(skb))
2026 goto csum_error;
2027
2028 ret = encap_rcv(sk, skb);
2029 if (ret <= 0) {
2030 __UDP_INC_STATS(sock_net(sk),
2031 UDP_MIB_INDATAGRAMS,
2032 is_udplite);
2033 return -ret;
2034 }
2035 }
2036
2037 /* FALLTHROUGH -- it's a UDP Packet */
2038 }
2039
2040 /*
2041 * UDP-Lite specific tests, ignored on UDP sockets
2042 */
2043 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2044
2045 /*
2046 * MIB statistics other than incrementing the error count are
2047 * disabled for the following two types of errors: these depend
2048 * on the application settings, not on the functioning of the
2049 * protocol stack as such.
2050 *
2051 * RFC 3828 here recommends (sec 3.3): "There should also be a
2052 * way ... to ... at least let the receiving application block
2053 * delivery of packets with coverage values less than a value
2054 * provided by the application."
2055 */
2056 if (up->pcrlen == 0) { /* full coverage was set */
2057 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2058 UDP_SKB_CB(skb)->cscov, skb->len);
2059 goto drop;
2060 }
2061 /* The next case involves violating the min. coverage requested
2062 * by the receiver. This is subtle: if receiver wants x and x is
2063 * greater than the buffersize/MTU then receiver will complain
2064 * that it wants x while sender emits packets of smaller size y.
2065 * Therefore the above ...()->partial_cov statement is essential.
2066 */
2067 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
2068 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2069 UDP_SKB_CB(skb)->cscov, up->pcrlen);
2070 goto drop;
2071 }
2072 }
2073
2074 prefetch(&sk->sk_rmem_alloc);
2075 if (rcu_access_pointer(sk->sk_filter) &&
2076 udp_lib_checksum_complete(skb))
2077 goto csum_error;
2078
2079 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2080 goto drop;
2081
2082 udp_csum_pull_header(skb);
2083
2084 ipv4_pktinfo_prepare(sk, skb);
2085 return __udp_queue_rcv_skb(sk, skb);
2086
2087csum_error:
2088 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2089drop:
2090 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2091 atomic_inc(&sk->sk_drops);
2092 kfree_skb(skb);
2093 return -1;
2094}
2095
2096static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2097{
2098 struct sk_buff *next, *segs;
2099 int ret;
2100
2101 if (likely(!udp_unexpected_gso(sk, skb)))
2102 return udp_queue_rcv_one_skb(sk, skb);
2103
2104 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET);
2105 __skb_push(skb, -skb_mac_offset(skb));
2106 segs = udp_rcv_segment(sk, skb, true);
2107 for (skb = segs; skb; skb = next) {
2108 next = skb->next;
2109 __skb_pull(skb, skb_transport_offset(skb));
2110 ret = udp_queue_rcv_one_skb(sk, skb);
2111 if (ret > 0)
2112 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2113 }
2114 return 0;
2115}
2116
2117/* For TCP sockets, sk_rx_dst is protected by socket lock
2118 * For UDP, we use xchg() to guard against concurrent changes.
2119 */
2120bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2121{
2122 struct dst_entry *old;
2123
2124 if (dst_hold_safe(dst)) {
2125 old = xchg(&sk->sk_rx_dst, dst);
2126 dst_release(old);
2127 return old != dst;
2128 }
2129 return false;
2130}
2131EXPORT_SYMBOL(udp_sk_rx_dst_set);
2132
2133/*
2134 * Multicasts and broadcasts go to each listener.
2135 *
2136 * Note: called only from the BH handler context.
2137 */
2138static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2139 struct udphdr *uh,
2140 __be32 saddr, __be32 daddr,
2141 struct udp_table *udptable,
2142 int proto)
2143{
2144 struct sock *sk, *first = NULL;
2145 unsigned short hnum = ntohs(uh->dest);
2146 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2147 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2148 unsigned int offset = offsetof(typeof(*sk), sk_node);
2149 int dif = skb->dev->ifindex;
2150 int sdif = inet_sdif(skb);
2151 struct hlist_node *node;
2152 struct sk_buff *nskb;
2153
2154 if (use_hash2) {
2155 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2156 udptable->mask;
2157 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2158start_lookup:
2159 hslot = &udptable->hash2[hash2];
2160 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2161 }
2162
2163 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2164 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2165 uh->source, saddr, dif, sdif, hnum))
2166 continue;
2167
2168 if (!first) {
2169 first = sk;
2170 continue;
2171 }
2172 nskb = skb_clone(skb, GFP_ATOMIC);
2173
2174 if (unlikely(!nskb)) {
2175 atomic_inc(&sk->sk_drops);
2176 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2177 IS_UDPLITE(sk));
2178 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2179 IS_UDPLITE(sk));
2180 continue;
2181 }
2182 if (udp_queue_rcv_skb(sk, nskb) > 0)
2183 consume_skb(nskb);
2184 }
2185
2186 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2187 if (use_hash2 && hash2 != hash2_any) {
2188 hash2 = hash2_any;
2189 goto start_lookup;
2190 }
2191
2192 if (first) {
2193 if (udp_queue_rcv_skb(first, skb) > 0)
2194 consume_skb(skb);
2195 } else {
2196 kfree_skb(skb);
2197 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2198 proto == IPPROTO_UDPLITE);
2199 }
2200 return 0;
2201}
2202
2203/* Initialize UDP checksum. If exited with zero value (success),
2204 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2205 * Otherwise, csum completion requires checksumming packet body,
2206 * including udp header and folding it to skb->csum.
2207 */
2208static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2209 int proto)
2210{
2211 int err;
2212
2213 UDP_SKB_CB(skb)->partial_cov = 0;
2214 UDP_SKB_CB(skb)->cscov = skb->len;
2215
2216 if (proto == IPPROTO_UDPLITE) {
2217 err = udplite_checksum_init(skb, uh);
2218 if (err)
2219 return err;
2220
2221 if (UDP_SKB_CB(skb)->partial_cov) {
2222 skb->csum = inet_compute_pseudo(skb, proto);
2223 return 0;
2224 }
2225 }
2226
2227 /* Note, we are only interested in != 0 or == 0, thus the
2228 * force to int.
2229 */
2230 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2231 inet_compute_pseudo);
2232 if (err)
2233 return err;
2234
2235 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2236 /* If SW calculated the value, we know it's bad */
2237 if (skb->csum_complete_sw)
2238 return 1;
2239
2240 /* HW says the value is bad. Let's validate that.
2241 * skb->csum is no longer the full packet checksum,
2242 * so don't treat it as such.
2243 */
2244 skb_checksum_complete_unset(skb);
2245 }
2246
2247 return 0;
2248}
2249
2250/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2251 * return code conversion for ip layer consumption
2252 */
2253static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2254 struct udphdr *uh)
2255{
2256 int ret;
2257
2258 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2259 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2260
2261 ret = udp_queue_rcv_skb(sk, skb);
2262
2263 /* a return value > 0 means to resubmit the input, but
2264 * it wants the return to be -protocol, or 0
2265 */
2266 if (ret > 0)
2267 return -ret;
2268 return 0;
2269}
2270
2271/*
2272 * All we need to do is get the socket, and then do a checksum.
2273 */
2274
2275int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2276 int proto)
2277{
2278 struct sock *sk;
2279 struct udphdr *uh;
2280 unsigned short ulen;
2281 struct rtable *rt = skb_rtable(skb);
2282 __be32 saddr, daddr;
2283 struct net *net = dev_net(skb->dev);
2284
2285 /*
2286 * Validate the packet.
2287 */
2288 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2289 goto drop; /* No space for header. */
2290
2291 uh = udp_hdr(skb);
2292 ulen = ntohs(uh->len);
2293 saddr = ip_hdr(skb)->saddr;
2294 daddr = ip_hdr(skb)->daddr;
2295
2296 if (ulen > skb->len)
2297 goto short_packet;
2298
2299 if (proto == IPPROTO_UDP) {
2300 /* UDP validates ulen. */
2301 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2302 goto short_packet;
2303 uh = udp_hdr(skb);
2304 }
2305
2306 if (udp4_csum_init(skb, uh, proto))
2307 goto csum_error;
2308
2309 sk = skb_steal_sock(skb);
2310 if (sk) {
2311 struct dst_entry *dst = skb_dst(skb);
2312 int ret;
2313
2314 if (unlikely(sk->sk_rx_dst != dst))
2315 udp_sk_rx_dst_set(sk, dst);
2316
2317 ret = udp_unicast_rcv_skb(sk, skb, uh);
2318 sock_put(sk);
2319 return ret;
2320 }
2321
2322 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2323 return __udp4_lib_mcast_deliver(net, skb, uh,
2324 saddr, daddr, udptable, proto);
2325
2326 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2327 if (sk)
2328 return udp_unicast_rcv_skb(sk, skb, uh);
2329
2330 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2331 goto drop;
2332 nf_reset_ct(skb);
2333
2334 /* No socket. Drop packet silently, if checksum is wrong */
2335 if (udp_lib_checksum_complete(skb))
2336 goto csum_error;
2337
2338 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2339 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2340
2341 /*
2342 * Hmm. We got an UDP packet to a port to which we
2343 * don't wanna listen. Ignore it.
2344 */
2345 kfree_skb(skb);
2346 return 0;
2347
2348short_packet:
2349 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2350 proto == IPPROTO_UDPLITE ? "Lite" : "",
2351 &saddr, ntohs(uh->source),
2352 ulen, skb->len,
2353 &daddr, ntohs(uh->dest));
2354 goto drop;
2355
2356csum_error:
2357 /*
2358 * RFC1122: OK. Discards the bad packet silently (as far as
2359 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2360 */
2361 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2362 proto == IPPROTO_UDPLITE ? "Lite" : "",
2363 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2364 ulen);
2365 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2366drop:
2367 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2368 kfree_skb(skb);
2369 return 0;
2370}
2371
2372/* We can only early demux multicast if there is a single matching socket.
2373 * If more than one socket found returns NULL
2374 */
2375static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2376 __be16 loc_port, __be32 loc_addr,
2377 __be16 rmt_port, __be32 rmt_addr,
2378 int dif, int sdif)
2379{
2380 struct sock *sk, *result;
2381 unsigned short hnum = ntohs(loc_port);
2382 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2383 struct udp_hslot *hslot = &udp_table.hash[slot];
2384
2385 /* Do not bother scanning a too big list */
2386 if (hslot->count > 10)
2387 return NULL;
2388
2389 result = NULL;
2390 sk_for_each_rcu(sk, &hslot->head) {
2391 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2392 rmt_port, rmt_addr, dif, sdif, hnum)) {
2393 if (result)
2394 return NULL;
2395 result = sk;
2396 }
2397 }
2398
2399 return result;
2400}
2401
2402/* For unicast we should only early demux connected sockets or we can
2403 * break forwarding setups. The chains here can be long so only check
2404 * if the first socket is an exact match and if not move on.
2405 */
2406static struct sock *__udp4_lib_demux_lookup(struct net *net,
2407 __be16 loc_port, __be32 loc_addr,
2408 __be16 rmt_port, __be32 rmt_addr,
2409 int dif, int sdif)
2410{
2411 unsigned short hnum = ntohs(loc_port);
2412 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2413 unsigned int slot2 = hash2 & udp_table.mask;
2414 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2415 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2416 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2417 struct sock *sk;
2418
2419 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2420 if (INET_MATCH(sk, net, acookie, rmt_addr,
2421 loc_addr, ports, dif, sdif))
2422 return sk;
2423 /* Only check first socket in chain */
2424 break;
2425 }
2426 return NULL;
2427}
2428
2429int udp_v4_early_demux(struct sk_buff *skb)
2430{
2431 struct net *net = dev_net(skb->dev);
2432 struct in_device *in_dev = NULL;
2433 const struct iphdr *iph;
2434 const struct udphdr *uh;
2435 struct sock *sk = NULL;
2436 struct dst_entry *dst;
2437 int dif = skb->dev->ifindex;
2438 int sdif = inet_sdif(skb);
2439 int ours;
2440
2441 /* validate the packet */
2442 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2443 return 0;
2444
2445 iph = ip_hdr(skb);
2446 uh = udp_hdr(skb);
2447
2448 if (skb->pkt_type == PACKET_MULTICAST) {
2449 in_dev = __in_dev_get_rcu(skb->dev);
2450
2451 if (!in_dev)
2452 return 0;
2453
2454 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2455 iph->protocol);
2456 if (!ours)
2457 return 0;
2458
2459 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2460 uh->source, iph->saddr,
2461 dif, sdif);
2462 } else if (skb->pkt_type == PACKET_HOST) {
2463 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2464 uh->source, iph->saddr, dif, sdif);
2465 }
2466
2467 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2468 return 0;
2469
2470 skb->sk = sk;
2471 skb->destructor = sock_efree;
2472 dst = READ_ONCE(sk->sk_rx_dst);
2473
2474 if (dst)
2475 dst = dst_check(dst, 0);
2476 if (dst) {
2477 u32 itag = 0;
2478
2479 /* set noref for now.
2480 * any place which wants to hold dst has to call
2481 * dst_hold_safe()
2482 */
2483 skb_dst_set_noref(skb, dst);
2484
2485 /* for unconnected multicast sockets we need to validate
2486 * the source on each packet
2487 */
2488 if (!inet_sk(sk)->inet_daddr && in_dev)
2489 return ip_mc_validate_source(skb, iph->daddr,
2490 iph->saddr, iph->tos,
2491 skb->dev, in_dev, &itag);
2492 }
2493 return 0;
2494}
2495
2496int udp_rcv(struct sk_buff *skb)
2497{
2498 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2499}
2500
2501void udp_destroy_sock(struct sock *sk)
2502{
2503 struct udp_sock *up = udp_sk(sk);
2504 bool slow = lock_sock_fast(sk);
2505 udp_flush_pending_frames(sk);
2506 unlock_sock_fast(sk, slow);
2507 if (static_branch_unlikely(&udp_encap_needed_key)) {
2508 if (up->encap_type) {
2509 void (*encap_destroy)(struct sock *sk);
2510 encap_destroy = READ_ONCE(up->encap_destroy);
2511 if (encap_destroy)
2512 encap_destroy(sk);
2513 }
2514 if (up->encap_enabled)
2515 static_branch_dec(&udp_encap_needed_key);
2516 }
2517}
2518
2519/*
2520 * Socket option code for UDP
2521 */
2522int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2523 char __user *optval, unsigned int optlen,
2524 int (*push_pending_frames)(struct sock *))
2525{
2526 struct udp_sock *up = udp_sk(sk);
2527 int val, valbool;
2528 int err = 0;
2529 int is_udplite = IS_UDPLITE(sk);
2530
2531 if (optlen < sizeof(int))
2532 return -EINVAL;
2533
2534 if (get_user(val, (int __user *)optval))
2535 return -EFAULT;
2536
2537 valbool = val ? 1 : 0;
2538
2539 switch (optname) {
2540 case UDP_CORK:
2541 if (val != 0) {
2542 up->corkflag = 1;
2543 } else {
2544 up->corkflag = 0;
2545 lock_sock(sk);
2546 push_pending_frames(sk);
2547 release_sock(sk);
2548 }
2549 break;
2550
2551 case UDP_ENCAP:
2552 switch (val) {
2553 case 0:
2554 case UDP_ENCAP_ESPINUDP:
2555 case UDP_ENCAP_ESPINUDP_NON_IKE:
2556 up->encap_rcv = xfrm4_udp_encap_rcv;
2557 /* FALLTHROUGH */
2558 case UDP_ENCAP_L2TPINUDP:
2559 up->encap_type = val;
2560 lock_sock(sk);
2561 udp_tunnel_encap_enable(sk->sk_socket);
2562 release_sock(sk);
2563 break;
2564 default:
2565 err = -ENOPROTOOPT;
2566 break;
2567 }
2568 break;
2569
2570 case UDP_NO_CHECK6_TX:
2571 up->no_check6_tx = valbool;
2572 break;
2573
2574 case UDP_NO_CHECK6_RX:
2575 up->no_check6_rx = valbool;
2576 break;
2577
2578 case UDP_SEGMENT:
2579 if (val < 0 || val > USHRT_MAX)
2580 return -EINVAL;
2581 up->gso_size = val;
2582 break;
2583
2584 case UDP_GRO:
2585 lock_sock(sk);
2586 if (valbool)
2587 udp_tunnel_encap_enable(sk->sk_socket);
2588 up->gro_enabled = valbool;
2589 release_sock(sk);
2590 break;
2591
2592 /*
2593 * UDP-Lite's partial checksum coverage (RFC 3828).
2594 */
2595 /* The sender sets actual checksum coverage length via this option.
2596 * The case coverage > packet length is handled by send module. */
2597 case UDPLITE_SEND_CSCOV:
2598 if (!is_udplite) /* Disable the option on UDP sockets */
2599 return -ENOPROTOOPT;
2600 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2601 val = 8;
2602 else if (val > USHRT_MAX)
2603 val = USHRT_MAX;
2604 up->pcslen = val;
2605 up->pcflag |= UDPLITE_SEND_CC;
2606 break;
2607
2608 /* The receiver specifies a minimum checksum coverage value. To make
2609 * sense, this should be set to at least 8 (as done below). If zero is
2610 * used, this again means full checksum coverage. */
2611 case UDPLITE_RECV_CSCOV:
2612 if (!is_udplite) /* Disable the option on UDP sockets */
2613 return -ENOPROTOOPT;
2614 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2615 val = 8;
2616 else if (val > USHRT_MAX)
2617 val = USHRT_MAX;
2618 up->pcrlen = val;
2619 up->pcflag |= UDPLITE_RECV_CC;
2620 break;
2621
2622 default:
2623 err = -ENOPROTOOPT;
2624 break;
2625 }
2626
2627 return err;
2628}
2629EXPORT_SYMBOL(udp_lib_setsockopt);
2630
2631int udp_setsockopt(struct sock *sk, int level, int optname,
2632 char __user *optval, unsigned int optlen)
2633{
2634 if (level == SOL_UDP || level == SOL_UDPLITE)
2635 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2636 udp_push_pending_frames);
2637 return ip_setsockopt(sk, level, optname, optval, optlen);
2638}
2639
2640#ifdef CONFIG_COMPAT
2641int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2642 char __user *optval, unsigned int optlen)
2643{
2644 if (level == SOL_UDP || level == SOL_UDPLITE)
2645 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2646 udp_push_pending_frames);
2647 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2648}
2649#endif
2650
2651int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2652 char __user *optval, int __user *optlen)
2653{
2654 struct udp_sock *up = udp_sk(sk);
2655 int val, len;
2656
2657 if (get_user(len, optlen))
2658 return -EFAULT;
2659
2660 len = min_t(unsigned int, len, sizeof(int));
2661
2662 if (len < 0)
2663 return -EINVAL;
2664
2665 switch (optname) {
2666 case UDP_CORK:
2667 val = up->corkflag;
2668 break;
2669
2670 case UDP_ENCAP:
2671 val = up->encap_type;
2672 break;
2673
2674 case UDP_NO_CHECK6_TX:
2675 val = up->no_check6_tx;
2676 break;
2677
2678 case UDP_NO_CHECK6_RX:
2679 val = up->no_check6_rx;
2680 break;
2681
2682 case UDP_SEGMENT:
2683 val = up->gso_size;
2684 break;
2685
2686 /* The following two cannot be changed on UDP sockets, the return is
2687 * always 0 (which corresponds to the full checksum coverage of UDP). */
2688 case UDPLITE_SEND_CSCOV:
2689 val = up->pcslen;
2690 break;
2691
2692 case UDPLITE_RECV_CSCOV:
2693 val = up->pcrlen;
2694 break;
2695
2696 default:
2697 return -ENOPROTOOPT;
2698 }
2699
2700 if (put_user(len, optlen))
2701 return -EFAULT;
2702 if (copy_to_user(optval, &val, len))
2703 return -EFAULT;
2704 return 0;
2705}
2706EXPORT_SYMBOL(udp_lib_getsockopt);
2707
2708int udp_getsockopt(struct sock *sk, int level, int optname,
2709 char __user *optval, int __user *optlen)
2710{
2711 if (level == SOL_UDP || level == SOL_UDPLITE)
2712 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2713 return ip_getsockopt(sk, level, optname, optval, optlen);
2714}
2715
2716#ifdef CONFIG_COMPAT
2717int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2718 char __user *optval, int __user *optlen)
2719{
2720 if (level == SOL_UDP || level == SOL_UDPLITE)
2721 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2722 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2723}
2724#endif
2725/**
2726 * udp_poll - wait for a UDP event.
2727 * @file - file struct
2728 * @sock - socket
2729 * @wait - poll table
2730 *
2731 * This is same as datagram poll, except for the special case of
2732 * blocking sockets. If application is using a blocking fd
2733 * and a packet with checksum error is in the queue;
2734 * then it could get return from select indicating data available
2735 * but then block when reading it. Add special case code
2736 * to work around these arguably broken applications.
2737 */
2738__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2739{
2740 __poll_t mask = datagram_poll(file, sock, wait);
2741 struct sock *sk = sock->sk;
2742
2743 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2744 mask |= EPOLLIN | EPOLLRDNORM;
2745
2746 /* Check for false positives due to checksum errors */
2747 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2748 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2749 mask &= ~(EPOLLIN | EPOLLRDNORM);
2750
2751 return mask;
2752
2753}
2754EXPORT_SYMBOL(udp_poll);
2755
2756int udp_abort(struct sock *sk, int err)
2757{
2758 lock_sock(sk);
2759
2760 sk->sk_err = err;
2761 sk->sk_error_report(sk);
2762 __udp_disconnect(sk, 0);
2763
2764 release_sock(sk);
2765
2766 return 0;
2767}
2768EXPORT_SYMBOL_GPL(udp_abort);
2769
2770struct proto udp_prot = {
2771 .name = "UDP",
2772 .owner = THIS_MODULE,
2773 .close = udp_lib_close,
2774 .pre_connect = udp_pre_connect,
2775 .connect = ip4_datagram_connect,
2776 .disconnect = udp_disconnect,
2777 .ioctl = udp_ioctl,
2778 .init = udp_init_sock,
2779 .destroy = udp_destroy_sock,
2780 .setsockopt = udp_setsockopt,
2781 .getsockopt = udp_getsockopt,
2782 .sendmsg = udp_sendmsg,
2783 .recvmsg = udp_recvmsg,
2784 .sendpage = udp_sendpage,
2785 .release_cb = ip4_datagram_release_cb,
2786 .hash = udp_lib_hash,
2787 .unhash = udp_lib_unhash,
2788 .rehash = udp_v4_rehash,
2789 .get_port = udp_v4_get_port,
2790 .memory_allocated = &udp_memory_allocated,
2791 .sysctl_mem = sysctl_udp_mem,
2792 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2793 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2794 .obj_size = sizeof(struct udp_sock),
2795 .h.udp_table = &udp_table,
2796#ifdef CONFIG_COMPAT
2797 .compat_setsockopt = compat_udp_setsockopt,
2798 .compat_getsockopt = compat_udp_getsockopt,
2799#endif
2800 .diag_destroy = udp_abort,
2801};
2802EXPORT_SYMBOL(udp_prot);
2803
2804/* ------------------------------------------------------------------------ */
2805#ifdef CONFIG_PROC_FS
2806
2807static struct sock *udp_get_first(struct seq_file *seq, int start)
2808{
2809 struct sock *sk;
2810 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2811 struct udp_iter_state *state = seq->private;
2812 struct net *net = seq_file_net(seq);
2813
2814 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2815 ++state->bucket) {
2816 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2817
2818 if (hlist_empty(&hslot->head))
2819 continue;
2820
2821 spin_lock_bh(&hslot->lock);
2822 sk_for_each(sk, &hslot->head) {
2823 if (!net_eq(sock_net(sk), net))
2824 continue;
2825 if (sk->sk_family == afinfo->family)
2826 goto found;
2827 }
2828 spin_unlock_bh(&hslot->lock);
2829 }
2830 sk = NULL;
2831found:
2832 return sk;
2833}
2834
2835static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2836{
2837 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2838 struct udp_iter_state *state = seq->private;
2839 struct net *net = seq_file_net(seq);
2840
2841 do {
2842 sk = sk_next(sk);
2843 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2844
2845 if (!sk) {
2846 if (state->bucket <= afinfo->udp_table->mask)
2847 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2848 return udp_get_first(seq, state->bucket + 1);
2849 }
2850 return sk;
2851}
2852
2853static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2854{
2855 struct sock *sk = udp_get_first(seq, 0);
2856
2857 if (sk)
2858 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2859 --pos;
2860 return pos ? NULL : sk;
2861}
2862
2863void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2864{
2865 struct udp_iter_state *state = seq->private;
2866 state->bucket = MAX_UDP_PORTS;
2867
2868 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2869}
2870EXPORT_SYMBOL(udp_seq_start);
2871
2872void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2873{
2874 struct sock *sk;
2875
2876 if (v == SEQ_START_TOKEN)
2877 sk = udp_get_idx(seq, 0);
2878 else
2879 sk = udp_get_next(seq, v);
2880
2881 ++*pos;
2882 return sk;
2883}
2884EXPORT_SYMBOL(udp_seq_next);
2885
2886void udp_seq_stop(struct seq_file *seq, void *v)
2887{
2888 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2889 struct udp_iter_state *state = seq->private;
2890
2891 if (state->bucket <= afinfo->udp_table->mask)
2892 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2893}
2894EXPORT_SYMBOL(udp_seq_stop);
2895
2896/* ------------------------------------------------------------------------ */
2897static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2898 int bucket)
2899{
2900 struct inet_sock *inet = inet_sk(sp);
2901 __be32 dest = inet->inet_daddr;
2902 __be32 src = inet->inet_rcv_saddr;
2903 __u16 destp = ntohs(inet->inet_dport);
2904 __u16 srcp = ntohs(inet->inet_sport);
2905
2906 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2907 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2908 bucket, src, srcp, dest, destp, sp->sk_state,
2909 sk_wmem_alloc_get(sp),
2910 udp_rqueue_get(sp),
2911 0, 0L, 0,
2912 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2913 0, sock_i_ino(sp),
2914 refcount_read(&sp->sk_refcnt), sp,
2915 atomic_read(&sp->sk_drops));
2916}
2917
2918int udp4_seq_show(struct seq_file *seq, void *v)
2919{
2920 seq_setwidth(seq, 127);
2921 if (v == SEQ_START_TOKEN)
2922 seq_puts(seq, " sl local_address rem_address st tx_queue "
2923 "rx_queue tr tm->when retrnsmt uid timeout "
2924 "inode ref pointer drops");
2925 else {
2926 struct udp_iter_state *state = seq->private;
2927
2928 udp4_format_sock(v, seq, state->bucket);
2929 }
2930 seq_pad(seq, '\n');
2931 return 0;
2932}
2933
2934const struct seq_operations udp_seq_ops = {
2935 .start = udp_seq_start,
2936 .next = udp_seq_next,
2937 .stop = udp_seq_stop,
2938 .show = udp4_seq_show,
2939};
2940EXPORT_SYMBOL(udp_seq_ops);
2941
2942static struct udp_seq_afinfo udp4_seq_afinfo = {
2943 .family = AF_INET,
2944 .udp_table = &udp_table,
2945};
2946
2947static int __net_init udp4_proc_init_net(struct net *net)
2948{
2949 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2950 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2951 return -ENOMEM;
2952 return 0;
2953}
2954
2955static void __net_exit udp4_proc_exit_net(struct net *net)
2956{
2957 remove_proc_entry("udp", net->proc_net);
2958}
2959
2960static struct pernet_operations udp4_net_ops = {
2961 .init = udp4_proc_init_net,
2962 .exit = udp4_proc_exit_net,
2963};
2964
2965int __init udp4_proc_init(void)
2966{
2967 return register_pernet_subsys(&udp4_net_ops);
2968}
2969
2970void udp4_proc_exit(void)
2971{
2972 unregister_pernet_subsys(&udp4_net_ops);
2973}
2974#endif /* CONFIG_PROC_FS */
2975
2976static __initdata unsigned long uhash_entries;
2977static int __init set_uhash_entries(char *str)
2978{
2979 ssize_t ret;
2980
2981 if (!str)
2982 return 0;
2983
2984 ret = kstrtoul(str, 0, &uhash_entries);
2985 if (ret)
2986 return 0;
2987
2988 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2989 uhash_entries = UDP_HTABLE_SIZE_MIN;
2990 return 1;
2991}
2992__setup("uhash_entries=", set_uhash_entries);
2993
2994void __init udp_table_init(struct udp_table *table, const char *name)
2995{
2996 unsigned int i;
2997
2998 table->hash = alloc_large_system_hash(name,
2999 2 * sizeof(struct udp_hslot),
3000 uhash_entries,
3001 21, /* one slot per 2 MB */
3002 0,
3003 &table->log,
3004 &table->mask,
3005 UDP_HTABLE_SIZE_MIN,
3006 64 * 1024);
3007
3008 table->hash2 = table->hash + (table->mask + 1);
3009 for (i = 0; i <= table->mask; i++) {
3010 INIT_HLIST_HEAD(&table->hash[i].head);
3011 table->hash[i].count = 0;
3012 spin_lock_init(&table->hash[i].lock);
3013 }
3014 for (i = 0; i <= table->mask; i++) {
3015 INIT_HLIST_HEAD(&table->hash2[i].head);
3016 table->hash2[i].count = 0;
3017 spin_lock_init(&table->hash2[i].lock);
3018 }
3019}
3020
3021u32 udp_flow_hashrnd(void)
3022{
3023 static u32 hashrnd __read_mostly;
3024
3025 net_get_random_once(&hashrnd, sizeof(hashrnd));
3026
3027 return hashrnd;
3028}
3029EXPORT_SYMBOL(udp_flow_hashrnd);
3030
3031static void __udp_sysctl_init(struct net *net)
3032{
3033 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3034 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3035
3036#ifdef CONFIG_NET_L3_MASTER_DEV
3037 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3038#endif
3039}
3040
3041static int __net_init udp_sysctl_init(struct net *net)
3042{
3043 __udp_sysctl_init(net);
3044 return 0;
3045}
3046
3047static struct pernet_operations __net_initdata udp_sysctl_ops = {
3048 .init = udp_sysctl_init,
3049};
3050
3051void __init udp_init(void)
3052{
3053 unsigned long limit;
3054 unsigned int i;
3055
3056 udp_table_init(&udp_table, "UDP");
3057 limit = nr_free_buffer_pages() / 8;
3058 limit = max(limit, 128UL);
3059 sysctl_udp_mem[0] = limit / 4 * 3;
3060 sysctl_udp_mem[1] = limit;
3061 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3062
3063 __udp_sysctl_init(&init_net);
3064
3065 /* 16 spinlocks per cpu */
3066 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3067 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3068 GFP_KERNEL);
3069 if (!udp_busylocks)
3070 panic("UDP: failed to alloc udp_busylocks\n");
3071 for (i = 0; i < (1U << udp_busylocks_log); i++)
3072 spin_lock_init(udp_busylocks + i);
3073
3074 if (register_pernet_subsys(&udp_sysctl_ops))
3075 panic("UDP: failed to init sysctl parameters.\n");
3076}