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.h>
104#include <net/ip_tunnels.h>
105#include <net/route.h>
106#include <net/checksum.h>
107#include <net/gso.h>
108#include <net/xfrm.h>
109#include <trace/events/udp.h>
110#include <linux/static_key.h>
111#include <linux/btf_ids.h>
112#include <trace/events/skb.h>
113#include <net/busy_poll.h>
114#include "udp_impl.h"
115#include <net/sock_reuseport.h>
116#include <net/addrconf.h>
117#include <net/udp_tunnel.h>
118#include <net/gro.h>
119#if IS_ENABLED(CONFIG_IPV6)
120#include <net/ipv6_stubs.h>
121#endif
122
123struct udp_table udp_table __read_mostly;
124EXPORT_SYMBOL(udp_table);
125
126long sysctl_udp_mem[3] __read_mostly;
127EXPORT_SYMBOL(sysctl_udp_mem);
128
129atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
130EXPORT_SYMBOL(udp_memory_allocated);
131DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
132EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
133
134#define MAX_UDP_PORTS 65536
135#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
136
137static struct udp_table *udp_get_table_prot(struct sock *sk)
138{
139 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
140}
141
142static int udp_lib_lport_inuse(struct net *net, __u16 num,
143 const struct udp_hslot *hslot,
144 unsigned long *bitmap,
145 struct sock *sk, unsigned int log)
146{
147 struct sock *sk2;
148 kuid_t uid = sock_i_uid(sk);
149
150 sk_for_each(sk2, &hslot->head) {
151 if (net_eq(sock_net(sk2), net) &&
152 sk2 != sk &&
153 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
154 (!sk2->sk_reuse || !sk->sk_reuse) &&
155 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
156 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
157 inet_rcv_saddr_equal(sk, sk2, true)) {
158 if (sk2->sk_reuseport && sk->sk_reuseport &&
159 !rcu_access_pointer(sk->sk_reuseport_cb) &&
160 uid_eq(uid, sock_i_uid(sk2))) {
161 if (!bitmap)
162 return 0;
163 } else {
164 if (!bitmap)
165 return 1;
166 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
167 bitmap);
168 }
169 }
170 }
171 return 0;
172}
173
174/*
175 * Note: we still hold spinlock of primary hash chain, so no other writer
176 * can insert/delete a socket with local_port == num
177 */
178static int udp_lib_lport_inuse2(struct net *net, __u16 num,
179 struct udp_hslot *hslot2,
180 struct sock *sk)
181{
182 struct sock *sk2;
183 kuid_t uid = sock_i_uid(sk);
184 int res = 0;
185
186 spin_lock(&hslot2->lock);
187 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
188 if (net_eq(sock_net(sk2), net) &&
189 sk2 != sk &&
190 (udp_sk(sk2)->udp_port_hash == num) &&
191 (!sk2->sk_reuse || !sk->sk_reuse) &&
192 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
193 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
194 inet_rcv_saddr_equal(sk, sk2, true)) {
195 if (sk2->sk_reuseport && sk->sk_reuseport &&
196 !rcu_access_pointer(sk->sk_reuseport_cb) &&
197 uid_eq(uid, sock_i_uid(sk2))) {
198 res = 0;
199 } else {
200 res = 1;
201 }
202 break;
203 }
204 }
205 spin_unlock(&hslot2->lock);
206 return res;
207}
208
209static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
210{
211 struct net *net = sock_net(sk);
212 kuid_t uid = sock_i_uid(sk);
213 struct sock *sk2;
214
215 sk_for_each(sk2, &hslot->head) {
216 if (net_eq(sock_net(sk2), net) &&
217 sk2 != sk &&
218 sk2->sk_family == sk->sk_family &&
219 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
220 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
221 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
222 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
223 inet_rcv_saddr_equal(sk, sk2, false)) {
224 return reuseport_add_sock(sk, sk2,
225 inet_rcv_saddr_any(sk));
226 }
227 }
228
229 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
230}
231
232/**
233 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
234 *
235 * @sk: socket struct in question
236 * @snum: port number to look up
237 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
238 * with NULL address
239 */
240int udp_lib_get_port(struct sock *sk, unsigned short snum,
241 unsigned int hash2_nulladdr)
242{
243 struct udp_table *udptable = udp_get_table_prot(sk);
244 struct udp_hslot *hslot, *hslot2;
245 struct net *net = sock_net(sk);
246 int error = -EADDRINUSE;
247
248 if (!snum) {
249 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
250 unsigned short first, last;
251 int low, high, remaining;
252 unsigned int rand;
253
254 inet_sk_get_local_port_range(sk, &low, &high);
255 remaining = (high - low) + 1;
256
257 rand = get_random_u32();
258 first = reciprocal_scale(rand, remaining) + low;
259 /*
260 * force rand to be an odd multiple of UDP_HTABLE_SIZE
261 */
262 rand = (rand | 1) * (udptable->mask + 1);
263 last = first + udptable->mask + 1;
264 do {
265 hslot = udp_hashslot(udptable, net, first);
266 bitmap_zero(bitmap, PORTS_PER_CHAIN);
267 spin_lock_bh(&hslot->lock);
268 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
269 udptable->log);
270
271 snum = first;
272 /*
273 * Iterate on all possible values of snum for this hash.
274 * Using steps of an odd multiple of UDP_HTABLE_SIZE
275 * give us randomization and full range coverage.
276 */
277 do {
278 if (low <= snum && snum <= high &&
279 !test_bit(snum >> udptable->log, bitmap) &&
280 !inet_is_local_reserved_port(net, snum))
281 goto found;
282 snum += rand;
283 } while (snum != first);
284 spin_unlock_bh(&hslot->lock);
285 cond_resched();
286 } while (++first != last);
287 goto fail;
288 } else {
289 hslot = udp_hashslot(udptable, net, snum);
290 spin_lock_bh(&hslot->lock);
291 if (hslot->count > 10) {
292 int exist;
293 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
294
295 slot2 &= udptable->mask;
296 hash2_nulladdr &= udptable->mask;
297
298 hslot2 = udp_hashslot2(udptable, slot2);
299 if (hslot->count < hslot2->count)
300 goto scan_primary_hash;
301
302 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
303 if (!exist && (hash2_nulladdr != slot2)) {
304 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
305 exist = udp_lib_lport_inuse2(net, snum, hslot2,
306 sk);
307 }
308 if (exist)
309 goto fail_unlock;
310 else
311 goto found;
312 }
313scan_primary_hash:
314 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
315 goto fail_unlock;
316 }
317found:
318 inet_sk(sk)->inet_num = snum;
319 udp_sk(sk)->udp_port_hash = snum;
320 udp_sk(sk)->udp_portaddr_hash ^= snum;
321 if (sk_unhashed(sk)) {
322 if (sk->sk_reuseport &&
323 udp_reuseport_add_sock(sk, hslot)) {
324 inet_sk(sk)->inet_num = 0;
325 udp_sk(sk)->udp_port_hash = 0;
326 udp_sk(sk)->udp_portaddr_hash ^= snum;
327 goto fail_unlock;
328 }
329
330 sock_set_flag(sk, SOCK_RCU_FREE);
331
332 sk_add_node_rcu(sk, &hslot->head);
333 hslot->count++;
334 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
335
336 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
337 spin_lock(&hslot2->lock);
338 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
339 sk->sk_family == AF_INET6)
340 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
341 &hslot2->head);
342 else
343 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
344 &hslot2->head);
345 hslot2->count++;
346 spin_unlock(&hslot2->lock);
347 }
348
349 error = 0;
350fail_unlock:
351 spin_unlock_bh(&hslot->lock);
352fail:
353 return error;
354}
355EXPORT_SYMBOL(udp_lib_get_port);
356
357int udp_v4_get_port(struct sock *sk, unsigned short snum)
358{
359 unsigned int hash2_nulladdr =
360 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
361 unsigned int hash2_partial =
362 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
363
364 /* precompute partial secondary hash */
365 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
366 return udp_lib_get_port(sk, snum, hash2_nulladdr);
367}
368
369static int compute_score(struct sock *sk, const struct net *net,
370 __be32 saddr, __be16 sport,
371 __be32 daddr, unsigned short hnum,
372 int dif, int sdif)
373{
374 int score;
375 struct inet_sock *inet;
376 bool dev_match;
377
378 if (!net_eq(sock_net(sk), net) ||
379 udp_sk(sk)->udp_port_hash != hnum ||
380 ipv6_only_sock(sk))
381 return -1;
382
383 if (sk->sk_rcv_saddr != daddr)
384 return -1;
385
386 score = (sk->sk_family == PF_INET) ? 2 : 1;
387
388 inet = inet_sk(sk);
389 if (inet->inet_daddr) {
390 if (inet->inet_daddr != saddr)
391 return -1;
392 score += 4;
393 }
394
395 if (inet->inet_dport) {
396 if (inet->inet_dport != sport)
397 return -1;
398 score += 4;
399 }
400
401 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
402 dif, sdif);
403 if (!dev_match)
404 return -1;
405 if (sk->sk_bound_dev_if)
406 score += 4;
407
408 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
409 score++;
410 return score;
411}
412
413u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
414 const __be32 faddr, const __be16 fport)
415{
416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
417
418 return __inet_ehashfn(laddr, lport, faddr, fport,
419 udp_ehash_secret + net_hash_mix(net));
420}
421EXPORT_SYMBOL(udp_ehashfn);
422
423/**
424 * udp4_lib_lookup1() - Simplified lookup using primary hash (destination port)
425 * @net: Network namespace
426 * @saddr: Source address, network order
427 * @sport: Source port, network order
428 * @daddr: Destination address, network order
429 * @hnum: Destination port, host order
430 * @dif: Destination interface index
431 * @sdif: Destination bridge port index, if relevant
432 * @udptable: Set of UDP hash tables
433 *
434 * Simplified lookup to be used as fallback if no sockets are found due to a
435 * potential race between (receive) address change, and lookup happening before
436 * the rehash operation. This function ignores SO_REUSEPORT groups while scoring
437 * result sockets, because if we have one, we don't need the fallback at all.
438 *
439 * Called under rcu_read_lock().
440 *
441 * Return: socket with highest matching score if any, NULL if none
442 */
443static struct sock *udp4_lib_lookup1(const struct net *net,
444 __be32 saddr, __be16 sport,
445 __be32 daddr, unsigned int hnum,
446 int dif, int sdif,
447 const struct udp_table *udptable)
448{
449 unsigned int slot = udp_hashfn(net, hnum, udptable->mask);
450 struct udp_hslot *hslot = &udptable->hash[slot];
451 struct sock *sk, *result = NULL;
452 int score, badness = 0;
453
454 sk_for_each_rcu(sk, &hslot->head) {
455 score = compute_score(sk, net,
456 saddr, sport, daddr, hnum, dif, sdif);
457 if (score > badness) {
458 result = sk;
459 badness = score;
460 }
461 }
462
463 return result;
464}
465
466/* called with rcu_read_lock() */
467static struct sock *udp4_lib_lookup2(const struct net *net,
468 __be32 saddr, __be16 sport,
469 __be32 daddr, unsigned int hnum,
470 int dif, int sdif,
471 struct udp_hslot *hslot2,
472 struct sk_buff *skb)
473{
474 struct sock *sk, *result;
475 int score, badness;
476 bool need_rescore;
477
478 result = NULL;
479 badness = 0;
480 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
481 need_rescore = false;
482rescore:
483 score = compute_score(need_rescore ? result : sk, net, saddr,
484 sport, daddr, hnum, dif, sdif);
485 if (score > badness) {
486 badness = score;
487
488 if (need_rescore)
489 continue;
490
491 if (sk->sk_state == TCP_ESTABLISHED) {
492 result = sk;
493 continue;
494 }
495
496 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
497 saddr, sport, daddr, hnum, udp_ehashfn);
498 if (!result) {
499 result = sk;
500 continue;
501 }
502
503 /* Fall back to scoring if group has connections */
504 if (!reuseport_has_conns(sk))
505 return result;
506
507 /* Reuseport logic returned an error, keep original score. */
508 if (IS_ERR(result))
509 continue;
510
511 /* compute_score is too long of a function to be
512 * inlined, and calling it again here yields
513 * measureable overhead for some
514 * workloads. Work around it by jumping
515 * backwards to rescore 'result'.
516 */
517 need_rescore = true;
518 goto rescore;
519 }
520 }
521 return result;
522}
523
524#if IS_ENABLED(CONFIG_BASE_SMALL)
525static struct sock *udp4_lib_lookup4(const struct net *net,
526 __be32 saddr, __be16 sport,
527 __be32 daddr, unsigned int hnum,
528 int dif, int sdif,
529 struct udp_table *udptable)
530{
531 return NULL;
532}
533
534static void udp_rehash4(struct udp_table *udptable, struct sock *sk,
535 u16 newhash4)
536{
537}
538
539static void udp_unhash4(struct udp_table *udptable, struct sock *sk)
540{
541}
542#else /* !CONFIG_BASE_SMALL */
543static struct sock *udp4_lib_lookup4(const struct net *net,
544 __be32 saddr, __be16 sport,
545 __be32 daddr, unsigned int hnum,
546 int dif, int sdif,
547 struct udp_table *udptable)
548{
549 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
550 const struct hlist_nulls_node *node;
551 struct udp_hslot *hslot4;
552 unsigned int hash4, slot;
553 struct udp_sock *up;
554 struct sock *sk;
555
556 hash4 = udp_ehashfn(net, daddr, hnum, saddr, sport);
557 slot = hash4 & udptable->mask;
558 hslot4 = &udptable->hash4[slot];
559 INET_ADDR_COOKIE(acookie, saddr, daddr);
560
561begin:
562 /* SLAB_TYPESAFE_BY_RCU not used, so we don't need to touch sk_refcnt */
563 udp_lrpa_for_each_entry_rcu(up, node, &hslot4->nulls_head) {
564 sk = (struct sock *)up;
565 if (inet_match(net, sk, acookie, ports, dif, sdif))
566 return sk;
567 }
568
569 /* if the nulls value we got at the end of this lookup is not the
570 * expected one, we must restart lookup. We probably met an item that
571 * was moved to another chain due to rehash.
572 */
573 if (get_nulls_value(node) != slot)
574 goto begin;
575
576 return NULL;
577}
578
579/* udp_rehash4() only checks hslot4, and hash4_cnt is not processed. */
580static void udp_rehash4(struct udp_table *udptable, struct sock *sk,
581 u16 newhash4)
582{
583 struct udp_hslot *hslot4, *nhslot4;
584
585 hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash);
586 nhslot4 = udp_hashslot4(udptable, newhash4);
587 udp_sk(sk)->udp_lrpa_hash = newhash4;
588
589 if (hslot4 != nhslot4) {
590 spin_lock_bh(&hslot4->lock);
591 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node);
592 hslot4->count--;
593 spin_unlock_bh(&hslot4->lock);
594
595 spin_lock_bh(&nhslot4->lock);
596 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node,
597 &nhslot4->nulls_head);
598 nhslot4->count++;
599 spin_unlock_bh(&nhslot4->lock);
600 }
601}
602
603static void udp_unhash4(struct udp_table *udptable, struct sock *sk)
604{
605 struct udp_hslot *hslot2, *hslot4;
606
607 if (udp_hashed4(sk)) {
608 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
609 hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash);
610
611 spin_lock(&hslot4->lock);
612 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node);
613 hslot4->count--;
614 spin_unlock(&hslot4->lock);
615
616 spin_lock(&hslot2->lock);
617 udp_hash4_dec(hslot2);
618 spin_unlock(&hslot2->lock);
619 }
620}
621
622void udp_lib_hash4(struct sock *sk, u16 hash)
623{
624 struct udp_hslot *hslot, *hslot2, *hslot4;
625 struct net *net = sock_net(sk);
626 struct udp_table *udptable;
627
628 /* Connected udp socket can re-connect to another remote address, which
629 * will be handled by rehash. Thus no need to redo hash4 here.
630 */
631 if (udp_hashed4(sk))
632 return;
633
634 udptable = net->ipv4.udp_table;
635 hslot = udp_hashslot(udptable, net, udp_sk(sk)->udp_port_hash);
636 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
637 hslot4 = udp_hashslot4(udptable, hash);
638 udp_sk(sk)->udp_lrpa_hash = hash;
639
640 spin_lock_bh(&hslot->lock);
641 if (rcu_access_pointer(sk->sk_reuseport_cb))
642 reuseport_detach_sock(sk);
643
644 spin_lock(&hslot4->lock);
645 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node,
646 &hslot4->nulls_head);
647 hslot4->count++;
648 spin_unlock(&hslot4->lock);
649
650 spin_lock(&hslot2->lock);
651 udp_hash4_inc(hslot2);
652 spin_unlock(&hslot2->lock);
653
654 spin_unlock_bh(&hslot->lock);
655}
656EXPORT_SYMBOL(udp_lib_hash4);
657
658/* call with sock lock */
659void udp4_hash4(struct sock *sk)
660{
661 struct net *net = sock_net(sk);
662 unsigned int hash;
663
664 if (sk_unhashed(sk) || sk->sk_rcv_saddr == htonl(INADDR_ANY))
665 return;
666
667 hash = udp_ehashfn(net, sk->sk_rcv_saddr, sk->sk_num,
668 sk->sk_daddr, sk->sk_dport);
669
670 udp_lib_hash4(sk, hash);
671}
672EXPORT_SYMBOL(udp4_hash4);
673#endif /* CONFIG_BASE_SMALL */
674
675/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
676 * harder than this. -DaveM
677 */
678struct sock *__udp4_lib_lookup(const struct net *net, __be32 saddr,
679 __be16 sport, __be32 daddr, __be16 dport, int dif,
680 int sdif, struct udp_table *udptable, struct sk_buff *skb)
681{
682 unsigned short hnum = ntohs(dport);
683 struct udp_hslot *hslot2;
684 struct sock *result, *sk;
685 unsigned int hash2;
686
687 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
688 hslot2 = udp_hashslot2(udptable, hash2);
689
690 if (udp_has_hash4(hslot2)) {
691 result = udp4_lib_lookup4(net, saddr, sport, daddr, hnum,
692 dif, sdif, udptable);
693 if (result) /* udp4_lib_lookup4 return sk or NULL */
694 return result;
695 }
696
697 /* Lookup connected or non-wildcard socket */
698 result = udp4_lib_lookup2(net, saddr, sport,
699 daddr, hnum, dif, sdif,
700 hslot2, skb);
701 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
702 goto done;
703
704 /* Lookup redirect from BPF */
705 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
706 udptable == net->ipv4.udp_table) {
707 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
708 saddr, sport, daddr, hnum, dif,
709 udp_ehashfn);
710 if (sk) {
711 result = sk;
712 goto done;
713 }
714 }
715
716 /* Got non-wildcard socket or error on first lookup */
717 if (result)
718 goto done;
719
720 /* Lookup wildcard sockets */
721 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
722 hslot2 = udp_hashslot2(udptable, hash2);
723
724 result = udp4_lib_lookup2(net, saddr, sport,
725 htonl(INADDR_ANY), hnum, dif, sdif,
726 hslot2, skb);
727 if (!IS_ERR_OR_NULL(result))
728 goto done;
729
730 /* Primary hash (destination port) lookup as fallback for this race:
731 * 1. __ip4_datagram_connect() sets sk_rcv_saddr
732 * 2. lookup (this function): new sk_rcv_saddr, hashes not updated yet
733 * 3. rehash operation updating _secondary and four-tuple_ hashes
734 * The primary hash doesn't need an update after 1., so, thanks to this
735 * further step, 1. and 3. don't need to be atomic against the lookup.
736 */
737 result = udp4_lib_lookup1(net, saddr, sport, daddr, hnum, dif, sdif,
738 udptable);
739
740done:
741 if (IS_ERR(result))
742 return NULL;
743 return result;
744}
745EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
746
747static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
748 __be16 sport, __be16 dport,
749 struct udp_table *udptable)
750{
751 const struct iphdr *iph = ip_hdr(skb);
752
753 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
754 iph->daddr, dport, inet_iif(skb),
755 inet_sdif(skb), udptable, skb);
756}
757
758struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
759 __be16 sport, __be16 dport)
760{
761 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
762 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
763 struct net *net = dev_net(skb->dev);
764 int iif, sdif;
765
766 inet_get_iif_sdif(skb, &iif, &sdif);
767
768 return __udp4_lib_lookup(net, iph->saddr, sport,
769 iph->daddr, dport, iif,
770 sdif, net->ipv4.udp_table, NULL);
771}
772
773/* Must be called under rcu_read_lock().
774 * Does increment socket refcount.
775 */
776#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
777struct sock *udp4_lib_lookup(const struct net *net, __be32 saddr, __be16 sport,
778 __be32 daddr, __be16 dport, int dif)
779{
780 struct sock *sk;
781
782 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
783 dif, 0, net->ipv4.udp_table, NULL);
784 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
785 sk = NULL;
786 return sk;
787}
788EXPORT_SYMBOL_GPL(udp4_lib_lookup);
789#endif
790
791static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
792 __be16 loc_port, __be32 loc_addr,
793 __be16 rmt_port, __be32 rmt_addr,
794 int dif, int sdif, unsigned short hnum)
795{
796 const struct inet_sock *inet = inet_sk(sk);
797
798 if (!net_eq(sock_net(sk), net) ||
799 udp_sk(sk)->udp_port_hash != hnum ||
800 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
801 (inet->inet_dport != rmt_port && inet->inet_dport) ||
802 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
803 ipv6_only_sock(sk) ||
804 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
805 return false;
806 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
807 return false;
808 return true;
809}
810
811DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
812EXPORT_SYMBOL(udp_encap_needed_key);
813
814#if IS_ENABLED(CONFIG_IPV6)
815DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
816EXPORT_SYMBOL(udpv6_encap_needed_key);
817#endif
818
819void udp_encap_enable(void)
820{
821 static_branch_inc(&udp_encap_needed_key);
822}
823EXPORT_SYMBOL(udp_encap_enable);
824
825void udp_encap_disable(void)
826{
827 static_branch_dec(&udp_encap_needed_key);
828}
829EXPORT_SYMBOL(udp_encap_disable);
830
831/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
832 * through error handlers in encapsulations looking for a match.
833 */
834static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
835{
836 int i;
837
838 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
839 int (*handler)(struct sk_buff *skb, u32 info);
840 const struct ip_tunnel_encap_ops *encap;
841
842 encap = rcu_dereference(iptun_encaps[i]);
843 if (!encap)
844 continue;
845 handler = encap->err_handler;
846 if (handler && !handler(skb, info))
847 return 0;
848 }
849
850 return -ENOENT;
851}
852
853/* Try to match ICMP errors to UDP tunnels by looking up a socket without
854 * reversing source and destination port: this will match tunnels that force the
855 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
856 * lwtunnels might actually break this assumption by being configured with
857 * different destination ports on endpoints, in this case we won't be able to
858 * trace ICMP messages back to them.
859 *
860 * If this doesn't match any socket, probe tunnels with arbitrary destination
861 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
862 * we've sent packets to won't necessarily match the local destination port.
863 *
864 * Then ask the tunnel implementation to match the error against a valid
865 * association.
866 *
867 * Return an error if we can't find a match, the socket if we need further
868 * processing, zero otherwise.
869 */
870static struct sock *__udp4_lib_err_encap(struct net *net,
871 const struct iphdr *iph,
872 struct udphdr *uh,
873 struct udp_table *udptable,
874 struct sock *sk,
875 struct sk_buff *skb, u32 info)
876{
877 int (*lookup)(struct sock *sk, struct sk_buff *skb);
878 int network_offset, transport_offset;
879 struct udp_sock *up;
880
881 network_offset = skb_network_offset(skb);
882 transport_offset = skb_transport_offset(skb);
883
884 /* Network header needs to point to the outer IPv4 header inside ICMP */
885 skb_reset_network_header(skb);
886
887 /* Transport header needs to point to the UDP header */
888 skb_set_transport_header(skb, iph->ihl << 2);
889
890 if (sk) {
891 up = udp_sk(sk);
892
893 lookup = READ_ONCE(up->encap_err_lookup);
894 if (lookup && lookup(sk, skb))
895 sk = NULL;
896
897 goto out;
898 }
899
900 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
901 iph->saddr, uh->dest, skb->dev->ifindex, 0,
902 udptable, NULL);
903 if (sk) {
904 up = udp_sk(sk);
905
906 lookup = READ_ONCE(up->encap_err_lookup);
907 if (!lookup || lookup(sk, skb))
908 sk = NULL;
909 }
910
911out:
912 if (!sk)
913 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
914
915 skb_set_transport_header(skb, transport_offset);
916 skb_set_network_header(skb, network_offset);
917
918 return sk;
919}
920
921/*
922 * This routine is called by the ICMP module when it gets some
923 * sort of error condition. If err < 0 then the socket should
924 * be closed and the error returned to the user. If err > 0
925 * it's just the icmp type << 8 | icmp code.
926 * Header points to the ip header of the error packet. We move
927 * on past this. Then (as it used to claim before adjustment)
928 * header points to the first 8 bytes of the udp header. We need
929 * to find the appropriate port.
930 */
931
932int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
933{
934 struct inet_sock *inet;
935 const struct iphdr *iph = (const struct iphdr *)skb->data;
936 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
937 const int type = icmp_hdr(skb)->type;
938 const int code = icmp_hdr(skb)->code;
939 bool tunnel = false;
940 struct sock *sk;
941 int harderr;
942 int err;
943 struct net *net = dev_net(skb->dev);
944
945 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
946 iph->saddr, uh->source, skb->dev->ifindex,
947 inet_sdif(skb), udptable, NULL);
948
949 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
950 /* No socket for error: try tunnels before discarding */
951 if (static_branch_unlikely(&udp_encap_needed_key)) {
952 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
953 info);
954 if (!sk)
955 return 0;
956 } else
957 sk = ERR_PTR(-ENOENT);
958
959 if (IS_ERR(sk)) {
960 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
961 return PTR_ERR(sk);
962 }
963
964 tunnel = true;
965 }
966
967 err = 0;
968 harderr = 0;
969 inet = inet_sk(sk);
970
971 switch (type) {
972 default:
973 case ICMP_TIME_EXCEEDED:
974 err = EHOSTUNREACH;
975 break;
976 case ICMP_SOURCE_QUENCH:
977 goto out;
978 case ICMP_PARAMETERPROB:
979 err = EPROTO;
980 harderr = 1;
981 break;
982 case ICMP_DEST_UNREACH:
983 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
984 ipv4_sk_update_pmtu(skb, sk, info);
985 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
986 err = EMSGSIZE;
987 harderr = 1;
988 break;
989 }
990 goto out;
991 }
992 err = EHOSTUNREACH;
993 if (code <= NR_ICMP_UNREACH) {
994 harderr = icmp_err_convert[code].fatal;
995 err = icmp_err_convert[code].errno;
996 }
997 break;
998 case ICMP_REDIRECT:
999 ipv4_sk_redirect(skb, sk);
1000 goto out;
1001 }
1002
1003 /*
1004 * RFC1122: OK. Passes ICMP errors back to application, as per
1005 * 4.1.3.3.
1006 */
1007 if (tunnel) {
1008 /* ...not for tunnels though: we don't have a sending socket */
1009 if (udp_sk(sk)->encap_err_rcv)
1010 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
1011 (u8 *)(uh+1));
1012 goto out;
1013 }
1014 if (!inet_test_bit(RECVERR, sk)) {
1015 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
1016 goto out;
1017 } else
1018 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
1019
1020 sk->sk_err = err;
1021 sk_error_report(sk);
1022out:
1023 return 0;
1024}
1025
1026int udp_err(struct sk_buff *skb, u32 info)
1027{
1028 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
1029}
1030
1031/*
1032 * Throw away all pending data and cancel the corking. Socket is locked.
1033 */
1034void udp_flush_pending_frames(struct sock *sk)
1035{
1036 struct udp_sock *up = udp_sk(sk);
1037
1038 if (up->pending) {
1039 up->len = 0;
1040 WRITE_ONCE(up->pending, 0);
1041 ip_flush_pending_frames(sk);
1042 }
1043}
1044EXPORT_SYMBOL(udp_flush_pending_frames);
1045
1046/**
1047 * udp4_hwcsum - handle outgoing HW checksumming
1048 * @skb: sk_buff containing the filled-in UDP header
1049 * (checksum field must be zeroed out)
1050 * @src: source IP address
1051 * @dst: destination IP address
1052 */
1053void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
1054{
1055 struct udphdr *uh = udp_hdr(skb);
1056 int offset = skb_transport_offset(skb);
1057 int len = skb->len - offset;
1058 int hlen = len;
1059 __wsum csum = 0;
1060
1061 if (!skb_has_frag_list(skb)) {
1062 /*
1063 * Only one fragment on the socket.
1064 */
1065 skb->csum_start = skb_transport_header(skb) - skb->head;
1066 skb->csum_offset = offsetof(struct udphdr, check);
1067 uh->check = ~csum_tcpudp_magic(src, dst, len,
1068 IPPROTO_UDP, 0);
1069 } else {
1070 struct sk_buff *frags;
1071
1072 /*
1073 * HW-checksum won't work as there are two or more
1074 * fragments on the socket so that all csums of sk_buffs
1075 * should be together
1076 */
1077 skb_walk_frags(skb, frags) {
1078 csum = csum_add(csum, frags->csum);
1079 hlen -= frags->len;
1080 }
1081
1082 csum = skb_checksum(skb, offset, hlen, csum);
1083 skb->ip_summed = CHECKSUM_NONE;
1084
1085 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
1086 if (uh->check == 0)
1087 uh->check = CSUM_MANGLED_0;
1088 }
1089}
1090EXPORT_SYMBOL_GPL(udp4_hwcsum);
1091
1092/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
1093 * for the simple case like when setting the checksum for a UDP tunnel.
1094 */
1095void udp_set_csum(bool nocheck, struct sk_buff *skb,
1096 __be32 saddr, __be32 daddr, int len)
1097{
1098 struct udphdr *uh = udp_hdr(skb);
1099
1100 if (nocheck) {
1101 uh->check = 0;
1102 } else if (skb_is_gso(skb)) {
1103 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
1104 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
1105 uh->check = 0;
1106 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
1107 if (uh->check == 0)
1108 uh->check = CSUM_MANGLED_0;
1109 } else {
1110 skb->ip_summed = CHECKSUM_PARTIAL;
1111 skb->csum_start = skb_transport_header(skb) - skb->head;
1112 skb->csum_offset = offsetof(struct udphdr, check);
1113 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
1114 }
1115}
1116EXPORT_SYMBOL(udp_set_csum);
1117
1118static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
1119 struct inet_cork *cork)
1120{
1121 struct sock *sk = skb->sk;
1122 struct inet_sock *inet = inet_sk(sk);
1123 struct udphdr *uh;
1124 int err;
1125 int is_udplite = IS_UDPLITE(sk);
1126 int offset = skb_transport_offset(skb);
1127 int len = skb->len - offset;
1128 int datalen = len - sizeof(*uh);
1129 __wsum csum = 0;
1130
1131 /*
1132 * Create a UDP header
1133 */
1134 uh = udp_hdr(skb);
1135 uh->source = inet->inet_sport;
1136 uh->dest = fl4->fl4_dport;
1137 uh->len = htons(len);
1138 uh->check = 0;
1139
1140 if (cork->gso_size) {
1141 const int hlen = skb_network_header_len(skb) +
1142 sizeof(struct udphdr);
1143
1144 if (hlen + min(datalen, cork->gso_size) > cork->fragsize) {
1145 kfree_skb(skb);
1146 return -EMSGSIZE;
1147 }
1148 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1149 kfree_skb(skb);
1150 return -EINVAL;
1151 }
1152 if (sk->sk_no_check_tx) {
1153 kfree_skb(skb);
1154 return -EINVAL;
1155 }
1156 if (is_udplite || dst_xfrm(skb_dst(skb))) {
1157 kfree_skb(skb);
1158 return -EIO;
1159 }
1160
1161 if (datalen > cork->gso_size) {
1162 skb_shinfo(skb)->gso_size = cork->gso_size;
1163 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1164 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1165 cork->gso_size);
1166
1167 /* Don't checksum the payload, skb will get segmented */
1168 goto csum_partial;
1169 }
1170 }
1171
1172 if (is_udplite) /* UDP-Lite */
1173 csum = udplite_csum(skb);
1174
1175 else if (sk->sk_no_check_tx) { /* UDP csum off */
1176
1177 skb->ip_summed = CHECKSUM_NONE;
1178 goto send;
1179
1180 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1181csum_partial:
1182
1183 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
1184 goto send;
1185
1186 } else
1187 csum = udp_csum(skb);
1188
1189 /* add protocol-dependent pseudo-header */
1190 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
1191 sk->sk_protocol, csum);
1192 if (uh->check == 0)
1193 uh->check = CSUM_MANGLED_0;
1194
1195send:
1196 err = ip_send_skb(sock_net(sk), skb);
1197 if (err) {
1198 if (err == -ENOBUFS &&
1199 !inet_test_bit(RECVERR, sk)) {
1200 UDP_INC_STATS(sock_net(sk),
1201 UDP_MIB_SNDBUFERRORS, is_udplite);
1202 err = 0;
1203 }
1204 } else
1205 UDP_INC_STATS(sock_net(sk),
1206 UDP_MIB_OUTDATAGRAMS, is_udplite);
1207 return err;
1208}
1209
1210/*
1211 * Push out all pending data as one UDP datagram. Socket is locked.
1212 */
1213int udp_push_pending_frames(struct sock *sk)
1214{
1215 struct udp_sock *up = udp_sk(sk);
1216 struct inet_sock *inet = inet_sk(sk);
1217 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1218 struct sk_buff *skb;
1219 int err = 0;
1220
1221 skb = ip_finish_skb(sk, fl4);
1222 if (!skb)
1223 goto out;
1224
1225 err = udp_send_skb(skb, fl4, &inet->cork.base);
1226
1227out:
1228 up->len = 0;
1229 WRITE_ONCE(up->pending, 0);
1230 return err;
1231}
1232EXPORT_SYMBOL(udp_push_pending_frames);
1233
1234static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1235{
1236 switch (cmsg->cmsg_type) {
1237 case UDP_SEGMENT:
1238 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1239 return -EINVAL;
1240 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1241 return 0;
1242 default:
1243 return -EINVAL;
1244 }
1245}
1246
1247int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1248{
1249 struct cmsghdr *cmsg;
1250 bool need_ip = false;
1251 int err;
1252
1253 for_each_cmsghdr(cmsg, msg) {
1254 if (!CMSG_OK(msg, cmsg))
1255 return -EINVAL;
1256
1257 if (cmsg->cmsg_level != SOL_UDP) {
1258 need_ip = true;
1259 continue;
1260 }
1261
1262 err = __udp_cmsg_send(cmsg, gso_size);
1263 if (err)
1264 return err;
1265 }
1266
1267 return need_ip;
1268}
1269EXPORT_SYMBOL_GPL(udp_cmsg_send);
1270
1271int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1272{
1273 struct inet_sock *inet = inet_sk(sk);
1274 struct udp_sock *up = udp_sk(sk);
1275 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1276 struct flowi4 fl4_stack;
1277 struct flowi4 *fl4;
1278 int ulen = len;
1279 struct ipcm_cookie ipc;
1280 struct rtable *rt = NULL;
1281 int free = 0;
1282 int connected = 0;
1283 __be32 daddr, faddr, saddr;
1284 u8 tos, scope;
1285 __be16 dport;
1286 int err, is_udplite = IS_UDPLITE(sk);
1287 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1288 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1289 struct sk_buff *skb;
1290 struct ip_options_data opt_copy;
1291 int uc_index;
1292
1293 if (len > 0xFFFF)
1294 return -EMSGSIZE;
1295
1296 /*
1297 * Check the flags.
1298 */
1299
1300 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1301 return -EOPNOTSUPP;
1302
1303 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1304
1305 fl4 = &inet->cork.fl.u.ip4;
1306 if (READ_ONCE(up->pending)) {
1307 /*
1308 * There are pending frames.
1309 * The socket lock must be held while it's corked.
1310 */
1311 lock_sock(sk);
1312 if (likely(up->pending)) {
1313 if (unlikely(up->pending != AF_INET)) {
1314 release_sock(sk);
1315 return -EINVAL;
1316 }
1317 goto do_append_data;
1318 }
1319 release_sock(sk);
1320 }
1321 ulen += sizeof(struct udphdr);
1322
1323 /*
1324 * Get and verify the address.
1325 */
1326 if (usin) {
1327 if (msg->msg_namelen < sizeof(*usin))
1328 return -EINVAL;
1329 if (usin->sin_family != AF_INET) {
1330 if (usin->sin_family != AF_UNSPEC)
1331 return -EAFNOSUPPORT;
1332 }
1333
1334 daddr = usin->sin_addr.s_addr;
1335 dport = usin->sin_port;
1336 if (dport == 0)
1337 return -EINVAL;
1338 } else {
1339 if (sk->sk_state != TCP_ESTABLISHED)
1340 return -EDESTADDRREQ;
1341 daddr = inet->inet_daddr;
1342 dport = inet->inet_dport;
1343 /* Open fast path for connected socket.
1344 Route will not be used, if at least one option is set.
1345 */
1346 connected = 1;
1347 }
1348
1349 ipcm_init_sk(&ipc, inet);
1350 ipc.gso_size = READ_ONCE(up->gso_size);
1351
1352 if (msg->msg_controllen) {
1353 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1354 if (err > 0) {
1355 err = ip_cmsg_send(sk, msg, &ipc,
1356 sk->sk_family == AF_INET6);
1357 connected = 0;
1358 }
1359 if (unlikely(err < 0)) {
1360 kfree(ipc.opt);
1361 return err;
1362 }
1363 if (ipc.opt)
1364 free = 1;
1365 }
1366 if (!ipc.opt) {
1367 struct ip_options_rcu *inet_opt;
1368
1369 rcu_read_lock();
1370 inet_opt = rcu_dereference(inet->inet_opt);
1371 if (inet_opt) {
1372 memcpy(&opt_copy, inet_opt,
1373 sizeof(*inet_opt) + inet_opt->opt.optlen);
1374 ipc.opt = &opt_copy.opt;
1375 }
1376 rcu_read_unlock();
1377 }
1378
1379 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1380 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1381 (struct sockaddr *)usin,
1382 &msg->msg_namelen,
1383 &ipc.addr);
1384 if (err)
1385 goto out_free;
1386 if (usin) {
1387 if (usin->sin_port == 0) {
1388 /* BPF program set invalid port. Reject it. */
1389 err = -EINVAL;
1390 goto out_free;
1391 }
1392 daddr = usin->sin_addr.s_addr;
1393 dport = usin->sin_port;
1394 }
1395 }
1396
1397 saddr = ipc.addr;
1398 ipc.addr = faddr = daddr;
1399
1400 if (ipc.opt && ipc.opt->opt.srr) {
1401 if (!daddr) {
1402 err = -EINVAL;
1403 goto out_free;
1404 }
1405 faddr = ipc.opt->opt.faddr;
1406 connected = 0;
1407 }
1408 tos = get_rttos(&ipc, inet);
1409 scope = ip_sendmsg_scope(inet, &ipc, msg);
1410 if (scope == RT_SCOPE_LINK)
1411 connected = 0;
1412
1413 uc_index = READ_ONCE(inet->uc_index);
1414 if (ipv4_is_multicast(daddr)) {
1415 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1416 ipc.oif = READ_ONCE(inet->mc_index);
1417 if (!saddr)
1418 saddr = READ_ONCE(inet->mc_addr);
1419 connected = 0;
1420 } else if (!ipc.oif) {
1421 ipc.oif = uc_index;
1422 } else if (ipv4_is_lbcast(daddr) && uc_index) {
1423 /* oif is set, packet is to local broadcast and
1424 * uc_index is set. oif is most likely set
1425 * by sk_bound_dev_if. If uc_index != oif check if the
1426 * oif is an L3 master and uc_index is an L3 slave.
1427 * If so, we want to allow the send using the uc_index.
1428 */
1429 if (ipc.oif != uc_index &&
1430 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1431 uc_index)) {
1432 ipc.oif = uc_index;
1433 }
1434 }
1435
1436 if (connected)
1437 rt = dst_rtable(sk_dst_check(sk, 0));
1438
1439 if (!rt) {
1440 struct net *net = sock_net(sk);
1441 __u8 flow_flags = inet_sk_flowi_flags(sk);
1442
1443 fl4 = &fl4_stack;
1444
1445 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1446 sk->sk_protocol, flow_flags, faddr, saddr,
1447 dport, inet->inet_sport, sk->sk_uid);
1448
1449 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1450 rt = ip_route_output_flow(net, fl4, sk);
1451 if (IS_ERR(rt)) {
1452 err = PTR_ERR(rt);
1453 rt = NULL;
1454 if (err == -ENETUNREACH)
1455 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1456 goto out;
1457 }
1458
1459 err = -EACCES;
1460 if ((rt->rt_flags & RTCF_BROADCAST) &&
1461 !sock_flag(sk, SOCK_BROADCAST))
1462 goto out;
1463 if (connected)
1464 sk_dst_set(sk, dst_clone(&rt->dst));
1465 }
1466
1467 if (msg->msg_flags&MSG_CONFIRM)
1468 goto do_confirm;
1469back_from_confirm:
1470
1471 saddr = fl4->saddr;
1472 if (!ipc.addr)
1473 daddr = ipc.addr = fl4->daddr;
1474
1475 /* Lockless fast path for the non-corking case. */
1476 if (!corkreq) {
1477 struct inet_cork cork;
1478
1479 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1480 sizeof(struct udphdr), &ipc, &rt,
1481 &cork, msg->msg_flags);
1482 err = PTR_ERR(skb);
1483 if (!IS_ERR_OR_NULL(skb))
1484 err = udp_send_skb(skb, fl4, &cork);
1485 goto out;
1486 }
1487
1488 lock_sock(sk);
1489 if (unlikely(up->pending)) {
1490 /* The socket is already corked while preparing it. */
1491 /* ... which is an evident application bug. --ANK */
1492 release_sock(sk);
1493
1494 net_dbg_ratelimited("socket already corked\n");
1495 err = -EINVAL;
1496 goto out;
1497 }
1498 /*
1499 * Now cork the socket to pend data.
1500 */
1501 fl4 = &inet->cork.fl.u.ip4;
1502 fl4->daddr = daddr;
1503 fl4->saddr = saddr;
1504 fl4->fl4_dport = dport;
1505 fl4->fl4_sport = inet->inet_sport;
1506 WRITE_ONCE(up->pending, AF_INET);
1507
1508do_append_data:
1509 up->len += ulen;
1510 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1511 sizeof(struct udphdr), &ipc, &rt,
1512 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1513 if (err)
1514 udp_flush_pending_frames(sk);
1515 else if (!corkreq)
1516 err = udp_push_pending_frames(sk);
1517 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1518 WRITE_ONCE(up->pending, 0);
1519 release_sock(sk);
1520
1521out:
1522 ip_rt_put(rt);
1523out_free:
1524 if (free)
1525 kfree(ipc.opt);
1526 if (!err)
1527 return len;
1528 /*
1529 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1530 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1531 * we don't have a good statistic (IpOutDiscards but it can be too many
1532 * things). We could add another new stat but at least for now that
1533 * seems like overkill.
1534 */
1535 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1536 UDP_INC_STATS(sock_net(sk),
1537 UDP_MIB_SNDBUFERRORS, is_udplite);
1538 }
1539 return err;
1540
1541do_confirm:
1542 if (msg->msg_flags & MSG_PROBE)
1543 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1544 if (!(msg->msg_flags&MSG_PROBE) || len)
1545 goto back_from_confirm;
1546 err = 0;
1547 goto out;
1548}
1549EXPORT_SYMBOL(udp_sendmsg);
1550
1551void udp_splice_eof(struct socket *sock)
1552{
1553 struct sock *sk = sock->sk;
1554 struct udp_sock *up = udp_sk(sk);
1555
1556 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1557 return;
1558
1559 lock_sock(sk);
1560 if (up->pending && !udp_test_bit(CORK, sk))
1561 udp_push_pending_frames(sk);
1562 release_sock(sk);
1563}
1564EXPORT_SYMBOL_GPL(udp_splice_eof);
1565
1566#define UDP_SKB_IS_STATELESS 0x80000000
1567
1568/* all head states (dst, sk, nf conntrack) except skb extensions are
1569 * cleared by udp_rcv().
1570 *
1571 * We need to preserve secpath, if present, to eventually process
1572 * IP_CMSG_PASSSEC at recvmsg() time.
1573 *
1574 * Other extensions can be cleared.
1575 */
1576static bool udp_try_make_stateless(struct sk_buff *skb)
1577{
1578 if (!skb_has_extensions(skb))
1579 return true;
1580
1581 if (!secpath_exists(skb)) {
1582 skb_ext_reset(skb);
1583 return true;
1584 }
1585
1586 return false;
1587}
1588
1589static void udp_set_dev_scratch(struct sk_buff *skb)
1590{
1591 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1592
1593 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1594 scratch->_tsize_state = skb->truesize;
1595#if BITS_PER_LONG == 64
1596 scratch->len = skb->len;
1597 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1598 scratch->is_linear = !skb_is_nonlinear(skb);
1599#endif
1600 if (udp_try_make_stateless(skb))
1601 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1602}
1603
1604static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1605{
1606 /* We come here after udp_lib_checksum_complete() returned 0.
1607 * This means that __skb_checksum_complete() might have
1608 * set skb->csum_valid to 1.
1609 * On 64bit platforms, we can set csum_unnecessary
1610 * to true, but only if the skb is not shared.
1611 */
1612#if BITS_PER_LONG == 64
1613 if (!skb_shared(skb))
1614 udp_skb_scratch(skb)->csum_unnecessary = true;
1615#endif
1616}
1617
1618static int udp_skb_truesize(struct sk_buff *skb)
1619{
1620 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1621}
1622
1623static bool udp_skb_has_head_state(struct sk_buff *skb)
1624{
1625 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1626}
1627
1628/* fully reclaim rmem/fwd memory allocated for skb */
1629static void udp_rmem_release(struct sock *sk, int size, int partial,
1630 bool rx_queue_lock_held)
1631{
1632 struct udp_sock *up = udp_sk(sk);
1633 struct sk_buff_head *sk_queue;
1634 int amt;
1635
1636 if (likely(partial)) {
1637 up->forward_deficit += size;
1638 size = up->forward_deficit;
1639 if (size < READ_ONCE(up->forward_threshold) &&
1640 !skb_queue_empty(&up->reader_queue))
1641 return;
1642 } else {
1643 size += up->forward_deficit;
1644 }
1645 up->forward_deficit = 0;
1646
1647 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1648 * if the called don't held it already
1649 */
1650 sk_queue = &sk->sk_receive_queue;
1651 if (!rx_queue_lock_held)
1652 spin_lock(&sk_queue->lock);
1653
1654
1655 sk_forward_alloc_add(sk, size);
1656 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1657 sk_forward_alloc_add(sk, -amt);
1658
1659 if (amt)
1660 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1661
1662 atomic_sub(size, &sk->sk_rmem_alloc);
1663
1664 /* this can save us from acquiring the rx queue lock on next receive */
1665 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1666
1667 if (!rx_queue_lock_held)
1668 spin_unlock(&sk_queue->lock);
1669}
1670
1671/* Note: called with reader_queue.lock held.
1672 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1673 * This avoids a cache line miss while receive_queue lock is held.
1674 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1675 */
1676void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1677{
1678 prefetch(&skb->data);
1679 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1680}
1681EXPORT_SYMBOL(udp_skb_destructor);
1682
1683/* as above, but the caller held the rx queue lock, too */
1684static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1685{
1686 prefetch(&skb->data);
1687 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1688}
1689
1690/* Idea of busylocks is to let producers grab an extra spinlock
1691 * to relieve pressure on the receive_queue spinlock shared by consumer.
1692 * Under flood, this means that only one producer can be in line
1693 * trying to acquire the receive_queue spinlock.
1694 * These busylock can be allocated on a per cpu manner, instead of a
1695 * per socket one (that would consume a cache line per socket)
1696 */
1697static int udp_busylocks_log __read_mostly;
1698static spinlock_t *udp_busylocks __read_mostly;
1699
1700static spinlock_t *busylock_acquire(void *ptr)
1701{
1702 spinlock_t *busy;
1703
1704 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1705 spin_lock(busy);
1706 return busy;
1707}
1708
1709static void busylock_release(spinlock_t *busy)
1710{
1711 if (busy)
1712 spin_unlock(busy);
1713}
1714
1715static int udp_rmem_schedule(struct sock *sk, int size)
1716{
1717 int delta;
1718
1719 delta = size - sk->sk_forward_alloc;
1720 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1721 return -ENOBUFS;
1722
1723 return 0;
1724}
1725
1726int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1727{
1728 struct sk_buff_head *list = &sk->sk_receive_queue;
1729 int rmem, err = -ENOMEM;
1730 spinlock_t *busy = NULL;
1731 int size, rcvbuf;
1732
1733 /* Immediately drop when the receive queue is full.
1734 * Always allow at least one packet.
1735 */
1736 rmem = atomic_read(&sk->sk_rmem_alloc);
1737 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1738 if (rmem > rcvbuf)
1739 goto drop;
1740
1741 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1742 * having linear skbs :
1743 * - Reduce memory overhead and thus increase receive queue capacity
1744 * - Less cache line misses at copyout() time
1745 * - Less work at consume_skb() (less alien page frag freeing)
1746 */
1747 if (rmem > (rcvbuf >> 1)) {
1748 skb_condense(skb);
1749
1750 busy = busylock_acquire(sk);
1751 }
1752 size = skb->truesize;
1753 udp_set_dev_scratch(skb);
1754
1755 atomic_add(size, &sk->sk_rmem_alloc);
1756
1757 spin_lock(&list->lock);
1758 err = udp_rmem_schedule(sk, size);
1759 if (err) {
1760 spin_unlock(&list->lock);
1761 goto uncharge_drop;
1762 }
1763
1764 sk_forward_alloc_add(sk, -size);
1765
1766 /* no need to setup a destructor, we will explicitly release the
1767 * forward allocated memory on dequeue
1768 */
1769 sock_skb_set_dropcount(sk, skb);
1770
1771 __skb_queue_tail(list, skb);
1772 spin_unlock(&list->lock);
1773
1774 if (!sock_flag(sk, SOCK_DEAD))
1775 INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
1776
1777 busylock_release(busy);
1778 return 0;
1779
1780uncharge_drop:
1781 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1782
1783drop:
1784 atomic_inc(&sk->sk_drops);
1785 busylock_release(busy);
1786 return err;
1787}
1788EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1789
1790void udp_destruct_common(struct sock *sk)
1791{
1792 /* reclaim completely the forward allocated memory */
1793 struct udp_sock *up = udp_sk(sk);
1794 unsigned int total = 0;
1795 struct sk_buff *skb;
1796
1797 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1798 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1799 total += skb->truesize;
1800 kfree_skb(skb);
1801 }
1802 udp_rmem_release(sk, total, 0, true);
1803}
1804EXPORT_SYMBOL_GPL(udp_destruct_common);
1805
1806static void udp_destruct_sock(struct sock *sk)
1807{
1808 udp_destruct_common(sk);
1809 inet_sock_destruct(sk);
1810}
1811
1812int udp_init_sock(struct sock *sk)
1813{
1814 udp_lib_init_sock(sk);
1815 sk->sk_destruct = udp_destruct_sock;
1816 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1817 return 0;
1818}
1819
1820void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1821{
1822 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
1823 sk_peek_offset_bwd(sk, len);
1824
1825 if (!skb_unref(skb))
1826 return;
1827
1828 /* In the more common cases we cleared the head states previously,
1829 * see __udp_queue_rcv_skb().
1830 */
1831 if (unlikely(udp_skb_has_head_state(skb)))
1832 skb_release_head_state(skb);
1833 __consume_stateless_skb(skb);
1834}
1835EXPORT_SYMBOL_GPL(skb_consume_udp);
1836
1837static struct sk_buff *__first_packet_length(struct sock *sk,
1838 struct sk_buff_head *rcvq,
1839 int *total)
1840{
1841 struct sk_buff *skb;
1842
1843 while ((skb = skb_peek(rcvq)) != NULL) {
1844 if (udp_lib_checksum_complete(skb)) {
1845 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1846 IS_UDPLITE(sk));
1847 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1848 IS_UDPLITE(sk));
1849 atomic_inc(&sk->sk_drops);
1850 __skb_unlink(skb, rcvq);
1851 *total += skb->truesize;
1852 kfree_skb(skb);
1853 } else {
1854 udp_skb_csum_unnecessary_set(skb);
1855 break;
1856 }
1857 }
1858 return skb;
1859}
1860
1861/**
1862 * first_packet_length - return length of first packet in receive queue
1863 * @sk: socket
1864 *
1865 * Drops all bad checksum frames, until a valid one is found.
1866 * Returns the length of found skb, or -1 if none is found.
1867 */
1868static int first_packet_length(struct sock *sk)
1869{
1870 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1871 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1872 struct sk_buff *skb;
1873 int total = 0;
1874 int res;
1875
1876 spin_lock_bh(&rcvq->lock);
1877 skb = __first_packet_length(sk, rcvq, &total);
1878 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1879 spin_lock(&sk_queue->lock);
1880 skb_queue_splice_tail_init(sk_queue, rcvq);
1881 spin_unlock(&sk_queue->lock);
1882
1883 skb = __first_packet_length(sk, rcvq, &total);
1884 }
1885 res = skb ? skb->len : -1;
1886 if (total)
1887 udp_rmem_release(sk, total, 1, false);
1888 spin_unlock_bh(&rcvq->lock);
1889 return res;
1890}
1891
1892/*
1893 * IOCTL requests applicable to the UDP protocol
1894 */
1895
1896int udp_ioctl(struct sock *sk, int cmd, int *karg)
1897{
1898 switch (cmd) {
1899 case SIOCOUTQ:
1900 {
1901 *karg = sk_wmem_alloc_get(sk);
1902 return 0;
1903 }
1904
1905 case SIOCINQ:
1906 {
1907 *karg = max_t(int, 0, first_packet_length(sk));
1908 return 0;
1909 }
1910
1911 default:
1912 return -ENOIOCTLCMD;
1913 }
1914
1915 return 0;
1916}
1917EXPORT_SYMBOL(udp_ioctl);
1918
1919struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1920 int *off, int *err)
1921{
1922 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1923 struct sk_buff_head *queue;
1924 struct sk_buff *last;
1925 long timeo;
1926 int error;
1927
1928 queue = &udp_sk(sk)->reader_queue;
1929 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1930 do {
1931 struct sk_buff *skb;
1932
1933 error = sock_error(sk);
1934 if (error)
1935 break;
1936
1937 error = -EAGAIN;
1938 do {
1939 spin_lock_bh(&queue->lock);
1940 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1941 err, &last);
1942 if (skb) {
1943 if (!(flags & MSG_PEEK))
1944 udp_skb_destructor(sk, skb);
1945 spin_unlock_bh(&queue->lock);
1946 return skb;
1947 }
1948
1949 if (skb_queue_empty_lockless(sk_queue)) {
1950 spin_unlock_bh(&queue->lock);
1951 goto busy_check;
1952 }
1953
1954 /* refill the reader queue and walk it again
1955 * keep both queues locked to avoid re-acquiring
1956 * the sk_receive_queue lock if fwd memory scheduling
1957 * is needed.
1958 */
1959 spin_lock(&sk_queue->lock);
1960 skb_queue_splice_tail_init(sk_queue, queue);
1961
1962 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1963 err, &last);
1964 if (skb && !(flags & MSG_PEEK))
1965 udp_skb_dtor_locked(sk, skb);
1966 spin_unlock(&sk_queue->lock);
1967 spin_unlock_bh(&queue->lock);
1968 if (skb)
1969 return skb;
1970
1971busy_check:
1972 if (!sk_can_busy_loop(sk))
1973 break;
1974
1975 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1976 } while (!skb_queue_empty_lockless(sk_queue));
1977
1978 /* sk_queue is empty, reader_queue may contain peeked packets */
1979 } while (timeo &&
1980 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1981 &error, &timeo,
1982 (struct sk_buff *)sk_queue));
1983
1984 *err = error;
1985 return NULL;
1986}
1987EXPORT_SYMBOL(__skb_recv_udp);
1988
1989int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1990{
1991 struct sk_buff *skb;
1992 int err;
1993
1994try_again:
1995 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1996 if (!skb)
1997 return err;
1998
1999 if (udp_lib_checksum_complete(skb)) {
2000 int is_udplite = IS_UDPLITE(sk);
2001 struct net *net = sock_net(sk);
2002
2003 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
2004 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
2005 atomic_inc(&sk->sk_drops);
2006 kfree_skb(skb);
2007 goto try_again;
2008 }
2009
2010 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
2011 return recv_actor(sk, skb);
2012}
2013EXPORT_SYMBOL(udp_read_skb);
2014
2015/*
2016 * This should be easy, if there is something there we
2017 * return it, otherwise we block.
2018 */
2019
2020int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
2021 int *addr_len)
2022{
2023 struct inet_sock *inet = inet_sk(sk);
2024 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
2025 struct sk_buff *skb;
2026 unsigned int ulen, copied;
2027 int off, err, peeking = flags & MSG_PEEK;
2028 int is_udplite = IS_UDPLITE(sk);
2029 bool checksum_valid = false;
2030
2031 if (flags & MSG_ERRQUEUE)
2032 return ip_recv_error(sk, msg, len, addr_len);
2033
2034try_again:
2035 off = sk_peek_offset(sk, flags);
2036 skb = __skb_recv_udp(sk, flags, &off, &err);
2037 if (!skb)
2038 return err;
2039
2040 ulen = udp_skb_len(skb);
2041 copied = len;
2042 if (copied > ulen - off)
2043 copied = ulen - off;
2044 else if (copied < ulen)
2045 msg->msg_flags |= MSG_TRUNC;
2046
2047 /*
2048 * If checksum is needed at all, try to do it while copying the
2049 * data. If the data is truncated, or if we only want a partial
2050 * coverage checksum (UDP-Lite), do it before the copy.
2051 */
2052
2053 if (copied < ulen || peeking ||
2054 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
2055 checksum_valid = udp_skb_csum_unnecessary(skb) ||
2056 !__udp_lib_checksum_complete(skb);
2057 if (!checksum_valid)
2058 goto csum_copy_err;
2059 }
2060
2061 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
2062 if (udp_skb_is_linear(skb))
2063 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
2064 else
2065 err = skb_copy_datagram_msg(skb, off, msg, copied);
2066 } else {
2067 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
2068
2069 if (err == -EINVAL)
2070 goto csum_copy_err;
2071 }
2072
2073 if (unlikely(err)) {
2074 if (!peeking) {
2075 atomic_inc(&sk->sk_drops);
2076 UDP_INC_STATS(sock_net(sk),
2077 UDP_MIB_INERRORS, is_udplite);
2078 }
2079 kfree_skb(skb);
2080 return err;
2081 }
2082
2083 if (!peeking)
2084 UDP_INC_STATS(sock_net(sk),
2085 UDP_MIB_INDATAGRAMS, is_udplite);
2086
2087 sock_recv_cmsgs(msg, sk, skb);
2088
2089 /* Copy the address. */
2090 if (sin) {
2091 sin->sin_family = AF_INET;
2092 sin->sin_port = udp_hdr(skb)->source;
2093 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
2094 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
2095 *addr_len = sizeof(*sin);
2096
2097 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
2098 (struct sockaddr *)sin,
2099 addr_len);
2100 }
2101
2102 if (udp_test_bit(GRO_ENABLED, sk))
2103 udp_cmsg_recv(msg, sk, skb);
2104
2105 if (inet_cmsg_flags(inet))
2106 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
2107
2108 err = copied;
2109 if (flags & MSG_TRUNC)
2110 err = ulen;
2111
2112 skb_consume_udp(sk, skb, peeking ? -err : err);
2113 return err;
2114
2115csum_copy_err:
2116 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
2117 udp_skb_destructor)) {
2118 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2119 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2120 }
2121 kfree_skb(skb);
2122
2123 /* starting over for a new packet, but check if we need to yield */
2124 cond_resched();
2125 msg->msg_flags &= ~MSG_TRUNC;
2126 goto try_again;
2127}
2128
2129int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
2130{
2131 /* This check is replicated from __ip4_datagram_connect() and
2132 * intended to prevent BPF program called below from accessing bytes
2133 * that are out of the bound specified by user in addr_len.
2134 */
2135 if (addr_len < sizeof(struct sockaddr_in))
2136 return -EINVAL;
2137
2138 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
2139}
2140EXPORT_SYMBOL(udp_pre_connect);
2141
2142static int udp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
2143{
2144 int res;
2145
2146 lock_sock(sk);
2147 res = __ip4_datagram_connect(sk, uaddr, addr_len);
2148 if (!res)
2149 udp4_hash4(sk);
2150 release_sock(sk);
2151 return res;
2152}
2153
2154int __udp_disconnect(struct sock *sk, int flags)
2155{
2156 struct inet_sock *inet = inet_sk(sk);
2157 /*
2158 * 1003.1g - break association.
2159 */
2160
2161 sk->sk_state = TCP_CLOSE;
2162 inet->inet_daddr = 0;
2163 inet->inet_dport = 0;
2164 sock_rps_reset_rxhash(sk);
2165 sk->sk_bound_dev_if = 0;
2166 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
2167 inet_reset_saddr(sk);
2168 if (sk->sk_prot->rehash &&
2169 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2170 sk->sk_prot->rehash(sk);
2171 }
2172
2173 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
2174 sk->sk_prot->unhash(sk);
2175 inet->inet_sport = 0;
2176 }
2177 sk_dst_reset(sk);
2178 return 0;
2179}
2180EXPORT_SYMBOL(__udp_disconnect);
2181
2182int udp_disconnect(struct sock *sk, int flags)
2183{
2184 lock_sock(sk);
2185 __udp_disconnect(sk, flags);
2186 release_sock(sk);
2187 return 0;
2188}
2189EXPORT_SYMBOL(udp_disconnect);
2190
2191void udp_lib_unhash(struct sock *sk)
2192{
2193 if (sk_hashed(sk)) {
2194 struct udp_table *udptable = udp_get_table_prot(sk);
2195 struct udp_hslot *hslot, *hslot2;
2196
2197 hslot = udp_hashslot(udptable, sock_net(sk),
2198 udp_sk(sk)->udp_port_hash);
2199 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2200
2201 spin_lock_bh(&hslot->lock);
2202 if (rcu_access_pointer(sk->sk_reuseport_cb))
2203 reuseport_detach_sock(sk);
2204 if (sk_del_node_init_rcu(sk)) {
2205 hslot->count--;
2206 inet_sk(sk)->inet_num = 0;
2207 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2208
2209 spin_lock(&hslot2->lock);
2210 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2211 hslot2->count--;
2212 spin_unlock(&hslot2->lock);
2213
2214 udp_unhash4(udptable, sk);
2215 }
2216 spin_unlock_bh(&hslot->lock);
2217 }
2218}
2219EXPORT_SYMBOL(udp_lib_unhash);
2220
2221/*
2222 * inet_rcv_saddr was changed, we must rehash secondary hash
2223 */
2224void udp_lib_rehash(struct sock *sk, u16 newhash, u16 newhash4)
2225{
2226 if (sk_hashed(sk)) {
2227 struct udp_table *udptable = udp_get_table_prot(sk);
2228 struct udp_hslot *hslot, *hslot2, *nhslot2;
2229
2230 hslot = udp_hashslot(udptable, sock_net(sk),
2231 udp_sk(sk)->udp_port_hash);
2232 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2233 nhslot2 = udp_hashslot2(udptable, newhash);
2234 udp_sk(sk)->udp_portaddr_hash = newhash;
2235
2236 if (hslot2 != nhslot2 ||
2237 rcu_access_pointer(sk->sk_reuseport_cb)) {
2238 /* we must lock primary chain too */
2239 spin_lock_bh(&hslot->lock);
2240 if (rcu_access_pointer(sk->sk_reuseport_cb))
2241 reuseport_detach_sock(sk);
2242
2243 if (hslot2 != nhslot2) {
2244 spin_lock(&hslot2->lock);
2245 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2246 hslot2->count--;
2247 spin_unlock(&hslot2->lock);
2248
2249 spin_lock(&nhslot2->lock);
2250 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2251 &nhslot2->head);
2252 nhslot2->count++;
2253 spin_unlock(&nhslot2->lock);
2254 }
2255
2256 spin_unlock_bh(&hslot->lock);
2257 }
2258
2259 /* Now process hash4 if necessary:
2260 * (1) update hslot4;
2261 * (2) update hslot2->hash4_cnt.
2262 * Note that hslot2/hslot4 should be checked separately, as
2263 * either of them may change with the other unchanged.
2264 */
2265 if (udp_hashed4(sk)) {
2266 spin_lock_bh(&hslot->lock);
2267
2268 udp_rehash4(udptable, sk, newhash4);
2269 if (hslot2 != nhslot2) {
2270 spin_lock(&hslot2->lock);
2271 udp_hash4_dec(hslot2);
2272 spin_unlock(&hslot2->lock);
2273
2274 spin_lock(&nhslot2->lock);
2275 udp_hash4_inc(nhslot2);
2276 spin_unlock(&nhslot2->lock);
2277 }
2278
2279 spin_unlock_bh(&hslot->lock);
2280 }
2281 }
2282}
2283EXPORT_SYMBOL(udp_lib_rehash);
2284
2285void udp_v4_rehash(struct sock *sk)
2286{
2287 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2288 inet_sk(sk)->inet_rcv_saddr,
2289 inet_sk(sk)->inet_num);
2290 u16 new_hash4 = udp_ehashfn(sock_net(sk),
2291 sk->sk_rcv_saddr, sk->sk_num,
2292 sk->sk_daddr, sk->sk_dport);
2293
2294 udp_lib_rehash(sk, new_hash, new_hash4);
2295}
2296
2297static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2298{
2299 int rc;
2300
2301 if (inet_sk(sk)->inet_daddr) {
2302 sock_rps_save_rxhash(sk, skb);
2303 sk_mark_napi_id(sk, skb);
2304 sk_incoming_cpu_update(sk);
2305 } else {
2306 sk_mark_napi_id_once(sk, skb);
2307 }
2308
2309 rc = __udp_enqueue_schedule_skb(sk, skb);
2310 if (rc < 0) {
2311 int is_udplite = IS_UDPLITE(sk);
2312 int drop_reason;
2313
2314 /* Note that an ENOMEM error is charged twice */
2315 if (rc == -ENOMEM) {
2316 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2317 is_udplite);
2318 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2319 } else {
2320 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2321 is_udplite);
2322 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2323 }
2324 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2325 trace_udp_fail_queue_rcv_skb(rc, sk, skb);
2326 sk_skb_reason_drop(sk, skb, drop_reason);
2327 return -1;
2328 }
2329
2330 return 0;
2331}
2332
2333/* returns:
2334 * -1: error
2335 * 0: success
2336 * >0: "udp encap" protocol resubmission
2337 *
2338 * Note that in the success and error cases, the skb is assumed to
2339 * have either been requeued or freed.
2340 */
2341static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2342{
2343 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2344 struct udp_sock *up = udp_sk(sk);
2345 int is_udplite = IS_UDPLITE(sk);
2346
2347 /*
2348 * Charge it to the socket, dropping if the queue is full.
2349 */
2350 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2351 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2352 goto drop;
2353 }
2354 nf_reset_ct(skb);
2355
2356 if (static_branch_unlikely(&udp_encap_needed_key) &&
2357 READ_ONCE(up->encap_type)) {
2358 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2359
2360 /*
2361 * This is an encapsulation socket so pass the skb to
2362 * the socket's udp_encap_rcv() hook. Otherwise, just
2363 * fall through and pass this up the UDP socket.
2364 * up->encap_rcv() returns the following value:
2365 * =0 if skb was successfully passed to the encap
2366 * handler or was discarded by it.
2367 * >0 if skb should be passed on to UDP.
2368 * <0 if skb should be resubmitted as proto -N
2369 */
2370
2371 /* if we're overly short, let UDP handle it */
2372 encap_rcv = READ_ONCE(up->encap_rcv);
2373 if (encap_rcv) {
2374 int ret;
2375
2376 /* Verify checksum before giving to encap */
2377 if (udp_lib_checksum_complete(skb))
2378 goto csum_error;
2379
2380 ret = encap_rcv(sk, skb);
2381 if (ret <= 0) {
2382 __UDP_INC_STATS(sock_net(sk),
2383 UDP_MIB_INDATAGRAMS,
2384 is_udplite);
2385 return -ret;
2386 }
2387 }
2388
2389 /* FALLTHROUGH -- it's a UDP Packet */
2390 }
2391
2392 /*
2393 * UDP-Lite specific tests, ignored on UDP sockets
2394 */
2395 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2396 u16 pcrlen = READ_ONCE(up->pcrlen);
2397
2398 /*
2399 * MIB statistics other than incrementing the error count are
2400 * disabled for the following two types of errors: these depend
2401 * on the application settings, not on the functioning of the
2402 * protocol stack as such.
2403 *
2404 * RFC 3828 here recommends (sec 3.3): "There should also be a
2405 * way ... to ... at least let the receiving application block
2406 * delivery of packets with coverage values less than a value
2407 * provided by the application."
2408 */
2409 if (pcrlen == 0) { /* full coverage was set */
2410 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2411 UDP_SKB_CB(skb)->cscov, skb->len);
2412 goto drop;
2413 }
2414 /* The next case involves violating the min. coverage requested
2415 * by the receiver. This is subtle: if receiver wants x and x is
2416 * greater than the buffersize/MTU then receiver will complain
2417 * that it wants x while sender emits packets of smaller size y.
2418 * Therefore the above ...()->partial_cov statement is essential.
2419 */
2420 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2421 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2422 UDP_SKB_CB(skb)->cscov, pcrlen);
2423 goto drop;
2424 }
2425 }
2426
2427 prefetch(&sk->sk_rmem_alloc);
2428 if (rcu_access_pointer(sk->sk_filter) &&
2429 udp_lib_checksum_complete(skb))
2430 goto csum_error;
2431
2432 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2433 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2434 goto drop;
2435 }
2436
2437 udp_csum_pull_header(skb);
2438
2439 ipv4_pktinfo_prepare(sk, skb, true);
2440 return __udp_queue_rcv_skb(sk, skb);
2441
2442csum_error:
2443 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2444 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2445drop:
2446 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2447 atomic_inc(&sk->sk_drops);
2448 sk_skb_reason_drop(sk, skb, drop_reason);
2449 return -1;
2450}
2451
2452static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2453{
2454 struct sk_buff *next, *segs;
2455 int ret;
2456
2457 if (likely(!udp_unexpected_gso(sk, skb)))
2458 return udp_queue_rcv_one_skb(sk, skb);
2459
2460 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2461 __skb_push(skb, -skb_mac_offset(skb));
2462 segs = udp_rcv_segment(sk, skb, true);
2463 skb_list_walk_safe(segs, skb, next) {
2464 __skb_pull(skb, skb_transport_offset(skb));
2465
2466 udp_post_segment_fix_csum(skb);
2467 ret = udp_queue_rcv_one_skb(sk, skb);
2468 if (ret > 0)
2469 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2470 }
2471 return 0;
2472}
2473
2474/* For TCP sockets, sk_rx_dst is protected by socket lock
2475 * For UDP, we use xchg() to guard against concurrent changes.
2476 */
2477bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2478{
2479 struct dst_entry *old;
2480
2481 if (dst_hold_safe(dst)) {
2482 old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst)));
2483 dst_release(old);
2484 return old != dst;
2485 }
2486 return false;
2487}
2488EXPORT_SYMBOL(udp_sk_rx_dst_set);
2489
2490/*
2491 * Multicasts and broadcasts go to each listener.
2492 *
2493 * Note: called only from the BH handler context.
2494 */
2495static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2496 struct udphdr *uh,
2497 __be32 saddr, __be32 daddr,
2498 struct udp_table *udptable,
2499 int proto)
2500{
2501 struct sock *sk, *first = NULL;
2502 unsigned short hnum = ntohs(uh->dest);
2503 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2504 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2505 unsigned int offset = offsetof(typeof(*sk), sk_node);
2506 int dif = skb->dev->ifindex;
2507 int sdif = inet_sdif(skb);
2508 struct hlist_node *node;
2509 struct sk_buff *nskb;
2510
2511 if (use_hash2) {
2512 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2513 udptable->mask;
2514 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2515start_lookup:
2516 hslot = &udptable->hash2[hash2].hslot;
2517 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2518 }
2519
2520 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2521 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2522 uh->source, saddr, dif, sdif, hnum))
2523 continue;
2524
2525 if (!first) {
2526 first = sk;
2527 continue;
2528 }
2529 nskb = skb_clone(skb, GFP_ATOMIC);
2530
2531 if (unlikely(!nskb)) {
2532 atomic_inc(&sk->sk_drops);
2533 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2534 IS_UDPLITE(sk));
2535 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2536 IS_UDPLITE(sk));
2537 continue;
2538 }
2539 if (udp_queue_rcv_skb(sk, nskb) > 0)
2540 consume_skb(nskb);
2541 }
2542
2543 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2544 if (use_hash2 && hash2 != hash2_any) {
2545 hash2 = hash2_any;
2546 goto start_lookup;
2547 }
2548
2549 if (first) {
2550 if (udp_queue_rcv_skb(first, skb) > 0)
2551 consume_skb(skb);
2552 } else {
2553 kfree_skb(skb);
2554 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2555 proto == IPPROTO_UDPLITE);
2556 }
2557 return 0;
2558}
2559
2560/* Initialize UDP checksum. If exited with zero value (success),
2561 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2562 * Otherwise, csum completion requires checksumming packet body,
2563 * including udp header and folding it to skb->csum.
2564 */
2565static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2566 int proto)
2567{
2568 int err;
2569
2570 UDP_SKB_CB(skb)->partial_cov = 0;
2571 UDP_SKB_CB(skb)->cscov = skb->len;
2572
2573 if (proto == IPPROTO_UDPLITE) {
2574 err = udplite_checksum_init(skb, uh);
2575 if (err)
2576 return err;
2577
2578 if (UDP_SKB_CB(skb)->partial_cov) {
2579 skb->csum = inet_compute_pseudo(skb, proto);
2580 return 0;
2581 }
2582 }
2583
2584 /* Note, we are only interested in != 0 or == 0, thus the
2585 * force to int.
2586 */
2587 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2588 inet_compute_pseudo);
2589 if (err)
2590 return err;
2591
2592 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2593 /* If SW calculated the value, we know it's bad */
2594 if (skb->csum_complete_sw)
2595 return 1;
2596
2597 /* HW says the value is bad. Let's validate that.
2598 * skb->csum is no longer the full packet checksum,
2599 * so don't treat it as such.
2600 */
2601 skb_checksum_complete_unset(skb);
2602 }
2603
2604 return 0;
2605}
2606
2607/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2608 * return code conversion for ip layer consumption
2609 */
2610static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2611 struct udphdr *uh)
2612{
2613 int ret;
2614
2615 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2616 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2617
2618 ret = udp_queue_rcv_skb(sk, skb);
2619
2620 /* a return value > 0 means to resubmit the input, but
2621 * it wants the return to be -protocol, or 0
2622 */
2623 if (ret > 0)
2624 return -ret;
2625 return 0;
2626}
2627
2628/*
2629 * All we need to do is get the socket, and then do a checksum.
2630 */
2631
2632int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2633 int proto)
2634{
2635 struct sock *sk = NULL;
2636 struct udphdr *uh;
2637 unsigned short ulen;
2638 struct rtable *rt = skb_rtable(skb);
2639 __be32 saddr, daddr;
2640 struct net *net = dev_net(skb->dev);
2641 bool refcounted;
2642 int drop_reason;
2643
2644 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2645
2646 /*
2647 * Validate the packet.
2648 */
2649 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2650 goto drop; /* No space for header. */
2651
2652 uh = udp_hdr(skb);
2653 ulen = ntohs(uh->len);
2654 saddr = ip_hdr(skb)->saddr;
2655 daddr = ip_hdr(skb)->daddr;
2656
2657 if (ulen > skb->len)
2658 goto short_packet;
2659
2660 if (proto == IPPROTO_UDP) {
2661 /* UDP validates ulen. */
2662 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2663 goto short_packet;
2664 uh = udp_hdr(skb);
2665 }
2666
2667 if (udp4_csum_init(skb, uh, proto))
2668 goto csum_error;
2669
2670 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2671 &refcounted, udp_ehashfn);
2672 if (IS_ERR(sk))
2673 goto no_sk;
2674
2675 if (sk) {
2676 struct dst_entry *dst = skb_dst(skb);
2677 int ret;
2678
2679 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2680 udp_sk_rx_dst_set(sk, dst);
2681
2682 ret = udp_unicast_rcv_skb(sk, skb, uh);
2683 if (refcounted)
2684 sock_put(sk);
2685 return ret;
2686 }
2687
2688 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2689 return __udp4_lib_mcast_deliver(net, skb, uh,
2690 saddr, daddr, udptable, proto);
2691
2692 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2693 if (sk)
2694 return udp_unicast_rcv_skb(sk, skb, uh);
2695no_sk:
2696 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2697 goto drop;
2698 nf_reset_ct(skb);
2699
2700 /* No socket. Drop packet silently, if checksum is wrong */
2701 if (udp_lib_checksum_complete(skb))
2702 goto csum_error;
2703
2704 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2705 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2706 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2707
2708 /*
2709 * Hmm. We got an UDP packet to a port to which we
2710 * don't wanna listen. Ignore it.
2711 */
2712 sk_skb_reason_drop(sk, skb, drop_reason);
2713 return 0;
2714
2715short_packet:
2716 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2717 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2718 proto == IPPROTO_UDPLITE ? "Lite" : "",
2719 &saddr, ntohs(uh->source),
2720 ulen, skb->len,
2721 &daddr, ntohs(uh->dest));
2722 goto drop;
2723
2724csum_error:
2725 /*
2726 * RFC1122: OK. Discards the bad packet silently (as far as
2727 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2728 */
2729 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2730 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2731 proto == IPPROTO_UDPLITE ? "Lite" : "",
2732 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2733 ulen);
2734 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2735drop:
2736 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2737 sk_skb_reason_drop(sk, skb, drop_reason);
2738 return 0;
2739}
2740
2741/* We can only early demux multicast if there is a single matching socket.
2742 * If more than one socket found returns NULL
2743 */
2744static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2745 __be16 loc_port, __be32 loc_addr,
2746 __be16 rmt_port, __be32 rmt_addr,
2747 int dif, int sdif)
2748{
2749 struct udp_table *udptable = net->ipv4.udp_table;
2750 unsigned short hnum = ntohs(loc_port);
2751 struct sock *sk, *result;
2752 struct udp_hslot *hslot;
2753 unsigned int slot;
2754
2755 slot = udp_hashfn(net, hnum, udptable->mask);
2756 hslot = &udptable->hash[slot];
2757
2758 /* Do not bother scanning a too big list */
2759 if (hslot->count > 10)
2760 return NULL;
2761
2762 result = NULL;
2763 sk_for_each_rcu(sk, &hslot->head) {
2764 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2765 rmt_port, rmt_addr, dif, sdif, hnum)) {
2766 if (result)
2767 return NULL;
2768 result = sk;
2769 }
2770 }
2771
2772 return result;
2773}
2774
2775/* For unicast we should only early demux connected sockets or we can
2776 * break forwarding setups. The chains here can be long so only check
2777 * if the first socket is an exact match and if not move on.
2778 */
2779static struct sock *__udp4_lib_demux_lookup(struct net *net,
2780 __be16 loc_port, __be32 loc_addr,
2781 __be16 rmt_port, __be32 rmt_addr,
2782 int dif, int sdif)
2783{
2784 struct udp_table *udptable = net->ipv4.udp_table;
2785 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2786 unsigned short hnum = ntohs(loc_port);
2787 struct udp_hslot *hslot2;
2788 unsigned int hash2;
2789 __portpair ports;
2790 struct sock *sk;
2791
2792 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2793 hslot2 = udp_hashslot2(udptable, hash2);
2794 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2795
2796 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2797 if (inet_match(net, sk, acookie, ports, dif, sdif))
2798 return sk;
2799 /* Only check first socket in chain */
2800 break;
2801 }
2802 return NULL;
2803}
2804
2805int udp_v4_early_demux(struct sk_buff *skb)
2806{
2807 struct net *net = dev_net(skb->dev);
2808 struct in_device *in_dev = NULL;
2809 const struct iphdr *iph;
2810 const struct udphdr *uh;
2811 struct sock *sk = NULL;
2812 struct dst_entry *dst;
2813 int dif = skb->dev->ifindex;
2814 int sdif = inet_sdif(skb);
2815 int ours;
2816
2817 /* validate the packet */
2818 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2819 return 0;
2820
2821 iph = ip_hdr(skb);
2822 uh = udp_hdr(skb);
2823
2824 if (skb->pkt_type == PACKET_MULTICAST) {
2825 in_dev = __in_dev_get_rcu(skb->dev);
2826
2827 if (!in_dev)
2828 return 0;
2829
2830 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2831 iph->protocol);
2832 if (!ours)
2833 return 0;
2834
2835 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2836 uh->source, iph->saddr,
2837 dif, sdif);
2838 } else if (skb->pkt_type == PACKET_HOST) {
2839 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2840 uh->source, iph->saddr, dif, sdif);
2841 }
2842
2843 if (!sk)
2844 return 0;
2845
2846 skb->sk = sk;
2847 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2848 skb->destructor = sock_pfree;
2849 dst = rcu_dereference(sk->sk_rx_dst);
2850
2851 if (dst)
2852 dst = dst_check(dst, 0);
2853 if (dst) {
2854 u32 itag = 0;
2855
2856 /* set noref for now.
2857 * any place which wants to hold dst has to call
2858 * dst_hold_safe()
2859 */
2860 skb_dst_set_noref(skb, dst);
2861
2862 /* for unconnected multicast sockets we need to validate
2863 * the source on each packet
2864 */
2865 if (!inet_sk(sk)->inet_daddr && in_dev)
2866 return ip_mc_validate_source(skb, iph->daddr,
2867 iph->saddr,
2868 ip4h_dscp(iph),
2869 skb->dev, in_dev, &itag);
2870 }
2871 return 0;
2872}
2873
2874int udp_rcv(struct sk_buff *skb)
2875{
2876 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2877}
2878
2879void udp_destroy_sock(struct sock *sk)
2880{
2881 struct udp_sock *up = udp_sk(sk);
2882 bool slow = lock_sock_fast(sk);
2883
2884 /* protects from races with udp_abort() */
2885 sock_set_flag(sk, SOCK_DEAD);
2886 udp_flush_pending_frames(sk);
2887 unlock_sock_fast(sk, slow);
2888 if (static_branch_unlikely(&udp_encap_needed_key)) {
2889 if (up->encap_type) {
2890 void (*encap_destroy)(struct sock *sk);
2891 encap_destroy = READ_ONCE(up->encap_destroy);
2892 if (encap_destroy)
2893 encap_destroy(sk);
2894 }
2895 if (udp_test_bit(ENCAP_ENABLED, sk))
2896 static_branch_dec(&udp_encap_needed_key);
2897 }
2898}
2899
2900static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2901 struct sock *sk)
2902{
2903#ifdef CONFIG_XFRM
2904 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2905 if (family == AF_INET)
2906 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2907 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2908 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2909 }
2910#endif
2911}
2912
2913/*
2914 * Socket option code for UDP
2915 */
2916int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2917 sockptr_t optval, unsigned int optlen,
2918 int (*push_pending_frames)(struct sock *))
2919{
2920 struct udp_sock *up = udp_sk(sk);
2921 int val, valbool;
2922 int err = 0;
2923 int is_udplite = IS_UDPLITE(sk);
2924
2925 if (level == SOL_SOCKET) {
2926 err = sk_setsockopt(sk, level, optname, optval, optlen);
2927
2928 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2929 sockopt_lock_sock(sk);
2930 /* paired with READ_ONCE in udp_rmem_release() */
2931 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2932 sockopt_release_sock(sk);
2933 }
2934 return err;
2935 }
2936
2937 if (optlen < sizeof(int))
2938 return -EINVAL;
2939
2940 if (copy_from_sockptr(&val, optval, sizeof(val)))
2941 return -EFAULT;
2942
2943 valbool = val ? 1 : 0;
2944
2945 switch (optname) {
2946 case UDP_CORK:
2947 if (val != 0) {
2948 udp_set_bit(CORK, sk);
2949 } else {
2950 udp_clear_bit(CORK, sk);
2951 lock_sock(sk);
2952 push_pending_frames(sk);
2953 release_sock(sk);
2954 }
2955 break;
2956
2957 case UDP_ENCAP:
2958 switch (val) {
2959 case 0:
2960#ifdef CONFIG_XFRM
2961 case UDP_ENCAP_ESPINUDP:
2962 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2963#if IS_ENABLED(CONFIG_IPV6)
2964 if (sk->sk_family == AF_INET6)
2965 WRITE_ONCE(up->encap_rcv,
2966 ipv6_stub->xfrm6_udp_encap_rcv);
2967 else
2968#endif
2969 WRITE_ONCE(up->encap_rcv,
2970 xfrm4_udp_encap_rcv);
2971#endif
2972 fallthrough;
2973 case UDP_ENCAP_L2TPINUDP:
2974 WRITE_ONCE(up->encap_type, val);
2975 udp_tunnel_encap_enable(sk);
2976 break;
2977 default:
2978 err = -ENOPROTOOPT;
2979 break;
2980 }
2981 break;
2982
2983 case UDP_NO_CHECK6_TX:
2984 udp_set_no_check6_tx(sk, valbool);
2985 break;
2986
2987 case UDP_NO_CHECK6_RX:
2988 udp_set_no_check6_rx(sk, valbool);
2989 break;
2990
2991 case UDP_SEGMENT:
2992 if (val < 0 || val > USHRT_MAX)
2993 return -EINVAL;
2994 WRITE_ONCE(up->gso_size, val);
2995 break;
2996
2997 case UDP_GRO:
2998
2999 /* when enabling GRO, accept the related GSO packet type */
3000 if (valbool)
3001 udp_tunnel_encap_enable(sk);
3002 udp_assign_bit(GRO_ENABLED, sk, valbool);
3003 udp_assign_bit(ACCEPT_L4, sk, valbool);
3004 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
3005 break;
3006
3007 /*
3008 * UDP-Lite's partial checksum coverage (RFC 3828).
3009 */
3010 /* The sender sets actual checksum coverage length via this option.
3011 * The case coverage > packet length is handled by send module. */
3012 case UDPLITE_SEND_CSCOV:
3013 if (!is_udplite) /* Disable the option on UDP sockets */
3014 return -ENOPROTOOPT;
3015 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
3016 val = 8;
3017 else if (val > USHRT_MAX)
3018 val = USHRT_MAX;
3019 WRITE_ONCE(up->pcslen, val);
3020 udp_set_bit(UDPLITE_SEND_CC, sk);
3021 break;
3022
3023 /* The receiver specifies a minimum checksum coverage value. To make
3024 * sense, this should be set to at least 8 (as done below). If zero is
3025 * used, this again means full checksum coverage. */
3026 case UDPLITE_RECV_CSCOV:
3027 if (!is_udplite) /* Disable the option on UDP sockets */
3028 return -ENOPROTOOPT;
3029 if (val != 0 && val < 8) /* Avoid silly minimal values. */
3030 val = 8;
3031 else if (val > USHRT_MAX)
3032 val = USHRT_MAX;
3033 WRITE_ONCE(up->pcrlen, val);
3034 udp_set_bit(UDPLITE_RECV_CC, sk);
3035 break;
3036
3037 default:
3038 err = -ENOPROTOOPT;
3039 break;
3040 }
3041
3042 return err;
3043}
3044EXPORT_SYMBOL(udp_lib_setsockopt);
3045
3046int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
3047 unsigned int optlen)
3048{
3049 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
3050 return udp_lib_setsockopt(sk, level, optname,
3051 optval, optlen,
3052 udp_push_pending_frames);
3053 return ip_setsockopt(sk, level, optname, optval, optlen);
3054}
3055
3056int udp_lib_getsockopt(struct sock *sk, int level, int optname,
3057 char __user *optval, int __user *optlen)
3058{
3059 struct udp_sock *up = udp_sk(sk);
3060 int val, len;
3061
3062 if (get_user(len, optlen))
3063 return -EFAULT;
3064
3065 if (len < 0)
3066 return -EINVAL;
3067
3068 len = min_t(unsigned int, len, sizeof(int));
3069
3070 switch (optname) {
3071 case UDP_CORK:
3072 val = udp_test_bit(CORK, sk);
3073 break;
3074
3075 case UDP_ENCAP:
3076 val = READ_ONCE(up->encap_type);
3077 break;
3078
3079 case UDP_NO_CHECK6_TX:
3080 val = udp_get_no_check6_tx(sk);
3081 break;
3082
3083 case UDP_NO_CHECK6_RX:
3084 val = udp_get_no_check6_rx(sk);
3085 break;
3086
3087 case UDP_SEGMENT:
3088 val = READ_ONCE(up->gso_size);
3089 break;
3090
3091 case UDP_GRO:
3092 val = udp_test_bit(GRO_ENABLED, sk);
3093 break;
3094
3095 /* The following two cannot be changed on UDP sockets, the return is
3096 * always 0 (which corresponds to the full checksum coverage of UDP). */
3097 case UDPLITE_SEND_CSCOV:
3098 val = READ_ONCE(up->pcslen);
3099 break;
3100
3101 case UDPLITE_RECV_CSCOV:
3102 val = READ_ONCE(up->pcrlen);
3103 break;
3104
3105 default:
3106 return -ENOPROTOOPT;
3107 }
3108
3109 if (put_user(len, optlen))
3110 return -EFAULT;
3111 if (copy_to_user(optval, &val, len))
3112 return -EFAULT;
3113 return 0;
3114}
3115EXPORT_SYMBOL(udp_lib_getsockopt);
3116
3117int udp_getsockopt(struct sock *sk, int level, int optname,
3118 char __user *optval, int __user *optlen)
3119{
3120 if (level == SOL_UDP || level == SOL_UDPLITE)
3121 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
3122 return ip_getsockopt(sk, level, optname, optval, optlen);
3123}
3124
3125/**
3126 * udp_poll - wait for a UDP event.
3127 * @file: - file struct
3128 * @sock: - socket
3129 * @wait: - poll table
3130 *
3131 * This is same as datagram poll, except for the special case of
3132 * blocking sockets. If application is using a blocking fd
3133 * and a packet with checksum error is in the queue;
3134 * then it could get return from select indicating data available
3135 * but then block when reading it. Add special case code
3136 * to work around these arguably broken applications.
3137 */
3138__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
3139{
3140 __poll_t mask = datagram_poll(file, sock, wait);
3141 struct sock *sk = sock->sk;
3142
3143 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
3144 mask |= EPOLLIN | EPOLLRDNORM;
3145
3146 /* Check for false positives due to checksum errors */
3147 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
3148 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
3149 mask &= ~(EPOLLIN | EPOLLRDNORM);
3150
3151 /* psock ingress_msg queue should not contain any bad checksum frames */
3152 if (sk_is_readable(sk))
3153 mask |= EPOLLIN | EPOLLRDNORM;
3154 return mask;
3155
3156}
3157EXPORT_SYMBOL(udp_poll);
3158
3159int udp_abort(struct sock *sk, int err)
3160{
3161 if (!has_current_bpf_ctx())
3162 lock_sock(sk);
3163
3164 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
3165 * with close()
3166 */
3167 if (sock_flag(sk, SOCK_DEAD))
3168 goto out;
3169
3170 sk->sk_err = err;
3171 sk_error_report(sk);
3172 __udp_disconnect(sk, 0);
3173
3174out:
3175 if (!has_current_bpf_ctx())
3176 release_sock(sk);
3177
3178 return 0;
3179}
3180EXPORT_SYMBOL_GPL(udp_abort);
3181
3182struct proto udp_prot = {
3183 .name = "UDP",
3184 .owner = THIS_MODULE,
3185 .close = udp_lib_close,
3186 .pre_connect = udp_pre_connect,
3187 .connect = udp_connect,
3188 .disconnect = udp_disconnect,
3189 .ioctl = udp_ioctl,
3190 .init = udp_init_sock,
3191 .destroy = udp_destroy_sock,
3192 .setsockopt = udp_setsockopt,
3193 .getsockopt = udp_getsockopt,
3194 .sendmsg = udp_sendmsg,
3195 .recvmsg = udp_recvmsg,
3196 .splice_eof = udp_splice_eof,
3197 .release_cb = ip4_datagram_release_cb,
3198 .hash = udp_lib_hash,
3199 .unhash = udp_lib_unhash,
3200 .rehash = udp_v4_rehash,
3201 .get_port = udp_v4_get_port,
3202 .put_port = udp_lib_unhash,
3203#ifdef CONFIG_BPF_SYSCALL
3204 .psock_update_sk_prot = udp_bpf_update_proto,
3205#endif
3206 .memory_allocated = &udp_memory_allocated,
3207 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
3208
3209 .sysctl_mem = sysctl_udp_mem,
3210 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
3211 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
3212 .obj_size = sizeof(struct udp_sock),
3213 .h.udp_table = NULL,
3214 .diag_destroy = udp_abort,
3215};
3216EXPORT_SYMBOL(udp_prot);
3217
3218/* ------------------------------------------------------------------------ */
3219#ifdef CONFIG_PROC_FS
3220
3221static unsigned short seq_file_family(const struct seq_file *seq);
3222static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
3223{
3224 unsigned short family = seq_file_family(seq);
3225
3226 /* AF_UNSPEC is used as a match all */
3227 return ((family == AF_UNSPEC || family == sk->sk_family) &&
3228 net_eq(sock_net(sk), seq_file_net(seq)));
3229}
3230
3231#ifdef CONFIG_BPF_SYSCALL
3232static const struct seq_operations bpf_iter_udp_seq_ops;
3233#endif
3234static struct udp_table *udp_get_table_seq(struct seq_file *seq,
3235 struct net *net)
3236{
3237 const struct udp_seq_afinfo *afinfo;
3238
3239#ifdef CONFIG_BPF_SYSCALL
3240 if (seq->op == &bpf_iter_udp_seq_ops)
3241 return net->ipv4.udp_table;
3242#endif
3243
3244 afinfo = pde_data(file_inode(seq->file));
3245 return afinfo->udp_table ? : net->ipv4.udp_table;
3246}
3247
3248static struct sock *udp_get_first(struct seq_file *seq, int start)
3249{
3250 struct udp_iter_state *state = seq->private;
3251 struct net *net = seq_file_net(seq);
3252 struct udp_table *udptable;
3253 struct sock *sk;
3254
3255 udptable = udp_get_table_seq(seq, net);
3256
3257 for (state->bucket = start; state->bucket <= udptable->mask;
3258 ++state->bucket) {
3259 struct udp_hslot *hslot = &udptable->hash[state->bucket];
3260
3261 if (hlist_empty(&hslot->head))
3262 continue;
3263
3264 spin_lock_bh(&hslot->lock);
3265 sk_for_each(sk, &hslot->head) {
3266 if (seq_sk_match(seq, sk))
3267 goto found;
3268 }
3269 spin_unlock_bh(&hslot->lock);
3270 }
3271 sk = NULL;
3272found:
3273 return sk;
3274}
3275
3276static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3277{
3278 struct udp_iter_state *state = seq->private;
3279 struct net *net = seq_file_net(seq);
3280 struct udp_table *udptable;
3281
3282 do {
3283 sk = sk_next(sk);
3284 } while (sk && !seq_sk_match(seq, sk));
3285
3286 if (!sk) {
3287 udptable = udp_get_table_seq(seq, net);
3288
3289 if (state->bucket <= udptable->mask)
3290 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3291
3292 return udp_get_first(seq, state->bucket + 1);
3293 }
3294 return sk;
3295}
3296
3297static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3298{
3299 struct sock *sk = udp_get_first(seq, 0);
3300
3301 if (sk)
3302 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3303 --pos;
3304 return pos ? NULL : sk;
3305}
3306
3307void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3308{
3309 struct udp_iter_state *state = seq->private;
3310 state->bucket = MAX_UDP_PORTS;
3311
3312 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3313}
3314EXPORT_SYMBOL(udp_seq_start);
3315
3316void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3317{
3318 struct sock *sk;
3319
3320 if (v == SEQ_START_TOKEN)
3321 sk = udp_get_idx(seq, 0);
3322 else
3323 sk = udp_get_next(seq, v);
3324
3325 ++*pos;
3326 return sk;
3327}
3328EXPORT_SYMBOL(udp_seq_next);
3329
3330void udp_seq_stop(struct seq_file *seq, void *v)
3331{
3332 struct udp_iter_state *state = seq->private;
3333 struct udp_table *udptable;
3334
3335 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3336
3337 if (state->bucket <= udptable->mask)
3338 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3339}
3340EXPORT_SYMBOL(udp_seq_stop);
3341
3342/* ------------------------------------------------------------------------ */
3343static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3344 int bucket)
3345{
3346 struct inet_sock *inet = inet_sk(sp);
3347 __be32 dest = inet->inet_daddr;
3348 __be32 src = inet->inet_rcv_saddr;
3349 __u16 destp = ntohs(inet->inet_dport);
3350 __u16 srcp = ntohs(inet->inet_sport);
3351
3352 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3353 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3354 bucket, src, srcp, dest, destp, sp->sk_state,
3355 sk_wmem_alloc_get(sp),
3356 udp_rqueue_get(sp),
3357 0, 0L, 0,
3358 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3359 0, sock_i_ino(sp),
3360 refcount_read(&sp->sk_refcnt), sp,
3361 atomic_read(&sp->sk_drops));
3362}
3363
3364int udp4_seq_show(struct seq_file *seq, void *v)
3365{
3366 seq_setwidth(seq, 127);
3367 if (v == SEQ_START_TOKEN)
3368 seq_puts(seq, " sl local_address rem_address st tx_queue "
3369 "rx_queue tr tm->when retrnsmt uid timeout "
3370 "inode ref pointer drops");
3371 else {
3372 struct udp_iter_state *state = seq->private;
3373
3374 udp4_format_sock(v, seq, state->bucket);
3375 }
3376 seq_pad(seq, '\n');
3377 return 0;
3378}
3379
3380#ifdef CONFIG_BPF_SYSCALL
3381struct bpf_iter__udp {
3382 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3383 __bpf_md_ptr(struct udp_sock *, udp_sk);
3384 uid_t uid __aligned(8);
3385 int bucket __aligned(8);
3386};
3387
3388struct bpf_udp_iter_state {
3389 struct udp_iter_state state;
3390 unsigned int cur_sk;
3391 unsigned int end_sk;
3392 unsigned int max_sk;
3393 int offset;
3394 struct sock **batch;
3395 bool st_bucket_done;
3396};
3397
3398static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3399 unsigned int new_batch_sz);
3400static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3401{
3402 struct bpf_udp_iter_state *iter = seq->private;
3403 struct udp_iter_state *state = &iter->state;
3404 struct net *net = seq_file_net(seq);
3405 int resume_bucket, resume_offset;
3406 struct udp_table *udptable;
3407 unsigned int batch_sks = 0;
3408 bool resized = false;
3409 struct sock *sk;
3410
3411 resume_bucket = state->bucket;
3412 resume_offset = iter->offset;
3413
3414 /* The current batch is done, so advance the bucket. */
3415 if (iter->st_bucket_done)
3416 state->bucket++;
3417
3418 udptable = udp_get_table_seq(seq, net);
3419
3420again:
3421 /* New batch for the next bucket.
3422 * Iterate over the hash table to find a bucket with sockets matching
3423 * the iterator attributes, and return the first matching socket from
3424 * the bucket. The remaining matched sockets from the bucket are batched
3425 * before releasing the bucket lock. This allows BPF programs that are
3426 * called in seq_show to acquire the bucket lock if needed.
3427 */
3428 iter->cur_sk = 0;
3429 iter->end_sk = 0;
3430 iter->st_bucket_done = false;
3431 batch_sks = 0;
3432
3433 for (; state->bucket <= udptable->mask; state->bucket++) {
3434 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket].hslot;
3435
3436 if (hlist_empty(&hslot2->head))
3437 continue;
3438
3439 iter->offset = 0;
3440 spin_lock_bh(&hslot2->lock);
3441 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3442 if (seq_sk_match(seq, sk)) {
3443 /* Resume from the last iterated socket at the
3444 * offset in the bucket before iterator was stopped.
3445 */
3446 if (state->bucket == resume_bucket &&
3447 iter->offset < resume_offset) {
3448 ++iter->offset;
3449 continue;
3450 }
3451 if (iter->end_sk < iter->max_sk) {
3452 sock_hold(sk);
3453 iter->batch[iter->end_sk++] = sk;
3454 }
3455 batch_sks++;
3456 }
3457 }
3458 spin_unlock_bh(&hslot2->lock);
3459
3460 if (iter->end_sk)
3461 break;
3462 }
3463
3464 /* All done: no batch made. */
3465 if (!iter->end_sk)
3466 return NULL;
3467
3468 if (iter->end_sk == batch_sks) {
3469 /* Batching is done for the current bucket; return the first
3470 * socket to be iterated from the batch.
3471 */
3472 iter->st_bucket_done = true;
3473 goto done;
3474 }
3475 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3476 resized = true;
3477 /* After allocating a larger batch, retry one more time to grab
3478 * the whole bucket.
3479 */
3480 goto again;
3481 }
3482done:
3483 return iter->batch[0];
3484}
3485
3486static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3487{
3488 struct bpf_udp_iter_state *iter = seq->private;
3489 struct sock *sk;
3490
3491 /* Whenever seq_next() is called, the iter->cur_sk is
3492 * done with seq_show(), so unref the iter->cur_sk.
3493 */
3494 if (iter->cur_sk < iter->end_sk) {
3495 sock_put(iter->batch[iter->cur_sk++]);
3496 ++iter->offset;
3497 }
3498
3499 /* After updating iter->cur_sk, check if there are more sockets
3500 * available in the current bucket batch.
3501 */
3502 if (iter->cur_sk < iter->end_sk)
3503 sk = iter->batch[iter->cur_sk];
3504 else
3505 /* Prepare a new batch. */
3506 sk = bpf_iter_udp_batch(seq);
3507
3508 ++*pos;
3509 return sk;
3510}
3511
3512static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3513{
3514 /* bpf iter does not support lseek, so it always
3515 * continue from where it was stop()-ped.
3516 */
3517 if (*pos)
3518 return bpf_iter_udp_batch(seq);
3519
3520 return SEQ_START_TOKEN;
3521}
3522
3523static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3524 struct udp_sock *udp_sk, uid_t uid, int bucket)
3525{
3526 struct bpf_iter__udp ctx;
3527
3528 meta->seq_num--; /* skip SEQ_START_TOKEN */
3529 ctx.meta = meta;
3530 ctx.udp_sk = udp_sk;
3531 ctx.uid = uid;
3532 ctx.bucket = bucket;
3533 return bpf_iter_run_prog(prog, &ctx);
3534}
3535
3536static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3537{
3538 struct udp_iter_state *state = seq->private;
3539 struct bpf_iter_meta meta;
3540 struct bpf_prog *prog;
3541 struct sock *sk = v;
3542 uid_t uid;
3543 int ret;
3544
3545 if (v == SEQ_START_TOKEN)
3546 return 0;
3547
3548 lock_sock(sk);
3549
3550 if (unlikely(sk_unhashed(sk))) {
3551 ret = SEQ_SKIP;
3552 goto unlock;
3553 }
3554
3555 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3556 meta.seq = seq;
3557 prog = bpf_iter_get_info(&meta, false);
3558 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3559
3560unlock:
3561 release_sock(sk);
3562 return ret;
3563}
3564
3565static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3566{
3567 while (iter->cur_sk < iter->end_sk)
3568 sock_put(iter->batch[iter->cur_sk++]);
3569}
3570
3571static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3572{
3573 struct bpf_udp_iter_state *iter = seq->private;
3574 struct bpf_iter_meta meta;
3575 struct bpf_prog *prog;
3576
3577 if (!v) {
3578 meta.seq = seq;
3579 prog = bpf_iter_get_info(&meta, true);
3580 if (prog)
3581 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3582 }
3583
3584 if (iter->cur_sk < iter->end_sk) {
3585 bpf_iter_udp_put_batch(iter);
3586 iter->st_bucket_done = false;
3587 }
3588}
3589
3590static const struct seq_operations bpf_iter_udp_seq_ops = {
3591 .start = bpf_iter_udp_seq_start,
3592 .next = bpf_iter_udp_seq_next,
3593 .stop = bpf_iter_udp_seq_stop,
3594 .show = bpf_iter_udp_seq_show,
3595};
3596#endif
3597
3598static unsigned short seq_file_family(const struct seq_file *seq)
3599{
3600 const struct udp_seq_afinfo *afinfo;
3601
3602#ifdef CONFIG_BPF_SYSCALL
3603 /* BPF iterator: bpf programs to filter sockets. */
3604 if (seq->op == &bpf_iter_udp_seq_ops)
3605 return AF_UNSPEC;
3606#endif
3607
3608 /* Proc fs iterator */
3609 afinfo = pde_data(file_inode(seq->file));
3610 return afinfo->family;
3611}
3612
3613const struct seq_operations udp_seq_ops = {
3614 .start = udp_seq_start,
3615 .next = udp_seq_next,
3616 .stop = udp_seq_stop,
3617 .show = udp4_seq_show,
3618};
3619EXPORT_SYMBOL(udp_seq_ops);
3620
3621static struct udp_seq_afinfo udp4_seq_afinfo = {
3622 .family = AF_INET,
3623 .udp_table = NULL,
3624};
3625
3626static int __net_init udp4_proc_init_net(struct net *net)
3627{
3628 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3629 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3630 return -ENOMEM;
3631 return 0;
3632}
3633
3634static void __net_exit udp4_proc_exit_net(struct net *net)
3635{
3636 remove_proc_entry("udp", net->proc_net);
3637}
3638
3639static struct pernet_operations udp4_net_ops = {
3640 .init = udp4_proc_init_net,
3641 .exit = udp4_proc_exit_net,
3642};
3643
3644int __init udp4_proc_init(void)
3645{
3646 return register_pernet_subsys(&udp4_net_ops);
3647}
3648
3649void udp4_proc_exit(void)
3650{
3651 unregister_pernet_subsys(&udp4_net_ops);
3652}
3653#endif /* CONFIG_PROC_FS */
3654
3655static __initdata unsigned long uhash_entries;
3656static int __init set_uhash_entries(char *str)
3657{
3658 ssize_t ret;
3659
3660 if (!str)
3661 return 0;
3662
3663 ret = kstrtoul(str, 0, &uhash_entries);
3664 if (ret)
3665 return 0;
3666
3667 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3668 uhash_entries = UDP_HTABLE_SIZE_MIN;
3669 return 1;
3670}
3671__setup("uhash_entries=", set_uhash_entries);
3672
3673void __init udp_table_init(struct udp_table *table, const char *name)
3674{
3675 unsigned int i, slot_size;
3676
3677 slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3678 udp_hash4_slot_size();
3679 table->hash = alloc_large_system_hash(name,
3680 slot_size,
3681 uhash_entries,
3682 21, /* one slot per 2 MB */
3683 0,
3684 &table->log,
3685 &table->mask,
3686 UDP_HTABLE_SIZE_MIN,
3687 UDP_HTABLE_SIZE_MAX);
3688
3689 table->hash2 = (void *)(table->hash + (table->mask + 1));
3690 for (i = 0; i <= table->mask; i++) {
3691 INIT_HLIST_HEAD(&table->hash[i].head);
3692 table->hash[i].count = 0;
3693 spin_lock_init(&table->hash[i].lock);
3694 }
3695 for (i = 0; i <= table->mask; i++) {
3696 INIT_HLIST_HEAD(&table->hash2[i].hslot.head);
3697 table->hash2[i].hslot.count = 0;
3698 spin_lock_init(&table->hash2[i].hslot.lock);
3699 }
3700 udp_table_hash4_init(table);
3701}
3702
3703u32 udp_flow_hashrnd(void)
3704{
3705 static u32 hashrnd __read_mostly;
3706
3707 net_get_random_once(&hashrnd, sizeof(hashrnd));
3708
3709 return hashrnd;
3710}
3711EXPORT_SYMBOL(udp_flow_hashrnd);
3712
3713static void __net_init udp_sysctl_init(struct net *net)
3714{
3715 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3716 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3717
3718#ifdef CONFIG_NET_L3_MASTER_DEV
3719 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3720#endif
3721}
3722
3723static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3724{
3725 struct udp_table *udptable;
3726 unsigned int slot_size;
3727 int i;
3728
3729 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3730 if (!udptable)
3731 goto out;
3732
3733 slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3734 udp_hash4_slot_size();
3735 udptable->hash = vmalloc_huge(hash_entries * slot_size,
3736 GFP_KERNEL_ACCOUNT);
3737 if (!udptable->hash)
3738 goto free_table;
3739
3740 udptable->hash2 = (void *)(udptable->hash + hash_entries);
3741 udptable->mask = hash_entries - 1;
3742 udptable->log = ilog2(hash_entries);
3743
3744 for (i = 0; i < hash_entries; i++) {
3745 INIT_HLIST_HEAD(&udptable->hash[i].head);
3746 udptable->hash[i].count = 0;
3747 spin_lock_init(&udptable->hash[i].lock);
3748
3749 INIT_HLIST_HEAD(&udptable->hash2[i].hslot.head);
3750 udptable->hash2[i].hslot.count = 0;
3751 spin_lock_init(&udptable->hash2[i].hslot.lock);
3752 }
3753 udp_table_hash4_init(udptable);
3754
3755 return udptable;
3756
3757free_table:
3758 kfree(udptable);
3759out:
3760 return NULL;
3761}
3762
3763static void __net_exit udp_pernet_table_free(struct net *net)
3764{
3765 struct udp_table *udptable = net->ipv4.udp_table;
3766
3767 if (udptable == &udp_table)
3768 return;
3769
3770 kvfree(udptable->hash);
3771 kfree(udptable);
3772}
3773
3774static void __net_init udp_set_table(struct net *net)
3775{
3776 struct udp_table *udptable;
3777 unsigned int hash_entries;
3778 struct net *old_net;
3779
3780 if (net_eq(net, &init_net))
3781 goto fallback;
3782
3783 old_net = current->nsproxy->net_ns;
3784 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3785 if (!hash_entries)
3786 goto fallback;
3787
3788 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3789 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3790 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3791 else
3792 hash_entries = roundup_pow_of_two(hash_entries);
3793
3794 udptable = udp_pernet_table_alloc(hash_entries);
3795 if (udptable) {
3796 net->ipv4.udp_table = udptable;
3797 } else {
3798 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3799 "for a netns, fallback to the global one\n",
3800 hash_entries);
3801fallback:
3802 net->ipv4.udp_table = &udp_table;
3803 }
3804}
3805
3806static int __net_init udp_pernet_init(struct net *net)
3807{
3808 udp_sysctl_init(net);
3809 udp_set_table(net);
3810
3811 return 0;
3812}
3813
3814static void __net_exit udp_pernet_exit(struct net *net)
3815{
3816 udp_pernet_table_free(net);
3817}
3818
3819static struct pernet_operations __net_initdata udp_sysctl_ops = {
3820 .init = udp_pernet_init,
3821 .exit = udp_pernet_exit,
3822};
3823
3824#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3825DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3826 struct udp_sock *udp_sk, uid_t uid, int bucket)
3827
3828static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3829 unsigned int new_batch_sz)
3830{
3831 struct sock **new_batch;
3832
3833 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3834 GFP_USER | __GFP_NOWARN);
3835 if (!new_batch)
3836 return -ENOMEM;
3837
3838 bpf_iter_udp_put_batch(iter);
3839 kvfree(iter->batch);
3840 iter->batch = new_batch;
3841 iter->max_sk = new_batch_sz;
3842
3843 return 0;
3844}
3845
3846#define INIT_BATCH_SZ 16
3847
3848static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3849{
3850 struct bpf_udp_iter_state *iter = priv_data;
3851 int ret;
3852
3853 ret = bpf_iter_init_seq_net(priv_data, aux);
3854 if (ret)
3855 return ret;
3856
3857 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3858 if (ret)
3859 bpf_iter_fini_seq_net(priv_data);
3860
3861 return ret;
3862}
3863
3864static void bpf_iter_fini_udp(void *priv_data)
3865{
3866 struct bpf_udp_iter_state *iter = priv_data;
3867
3868 bpf_iter_fini_seq_net(priv_data);
3869 kvfree(iter->batch);
3870}
3871
3872static const struct bpf_iter_seq_info udp_seq_info = {
3873 .seq_ops = &bpf_iter_udp_seq_ops,
3874 .init_seq_private = bpf_iter_init_udp,
3875 .fini_seq_private = bpf_iter_fini_udp,
3876 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3877};
3878
3879static struct bpf_iter_reg udp_reg_info = {
3880 .target = "udp",
3881 .ctx_arg_info_size = 1,
3882 .ctx_arg_info = {
3883 { offsetof(struct bpf_iter__udp, udp_sk),
3884 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3885 },
3886 .seq_info = &udp_seq_info,
3887};
3888
3889static void __init bpf_iter_register(void)
3890{
3891 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3892 if (bpf_iter_reg_target(&udp_reg_info))
3893 pr_warn("Warning: could not register bpf iterator udp\n");
3894}
3895#endif
3896
3897void __init udp_init(void)
3898{
3899 unsigned long limit;
3900 unsigned int i;
3901
3902 udp_table_init(&udp_table, "UDP");
3903 limit = nr_free_buffer_pages() / 8;
3904 limit = max(limit, 128UL);
3905 sysctl_udp_mem[0] = limit / 4 * 3;
3906 sysctl_udp_mem[1] = limit;
3907 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3908
3909 /* 16 spinlocks per cpu */
3910 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3911 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3912 GFP_KERNEL);
3913 if (!udp_busylocks)
3914 panic("UDP: failed to alloc udp_busylocks\n");
3915 for (i = 0; i < (1U << udp_busylocks_log); i++)
3916 spin_lock_init(udp_busylocks + i);
3917
3918 if (register_pernet_subsys(&udp_sysctl_ops))
3919 panic("UDP: failed to init sysctl parameters.\n");
3920
3921#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3922 bpf_iter_register();
3923#endif
3924}
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 static u32 udp_ehash_secret __read_mostly;
415
416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
417
418 return __inet_ehashfn(laddr, lport, faddr, fport,
419 udp_ehash_secret + net_hash_mix(net));
420}
421
422/* called with rcu_read_lock() */
423static struct sock *udp4_lib_lookup2(struct net *net,
424 __be32 saddr, __be16 sport,
425 __be32 daddr, unsigned int hnum,
426 int dif, int sdif,
427 struct udp_hslot *hslot2,
428 struct sk_buff *skb)
429{
430 struct sock *sk, *result;
431 int score, badness;
432
433 result = NULL;
434 badness = 0;
435 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
436 score = compute_score(sk, net, saddr, sport,
437 daddr, hnum, dif, sdif);
438 if (score > badness) {
439 badness = score;
440
441 if (sk->sk_state == TCP_ESTABLISHED) {
442 result = sk;
443 continue;
444 }
445
446 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
447 saddr, sport, daddr, hnum, udp_ehashfn);
448 if (!result) {
449 result = sk;
450 continue;
451 }
452
453 /* Fall back to scoring if group has connections */
454 if (!reuseport_has_conns(sk))
455 return result;
456
457 /* Reuseport logic returned an error, keep original score. */
458 if (IS_ERR(result))
459 continue;
460
461 badness = compute_score(result, net, saddr, sport,
462 daddr, hnum, dif, sdif);
463
464 }
465 }
466 return result;
467}
468
469/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
470 * harder than this. -DaveM
471 */
472struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
473 __be16 sport, __be32 daddr, __be16 dport, int dif,
474 int sdif, struct udp_table *udptable, struct sk_buff *skb)
475{
476 unsigned short hnum = ntohs(dport);
477 unsigned int hash2, slot2;
478 struct udp_hslot *hslot2;
479 struct sock *result, *sk;
480
481 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
482 slot2 = hash2 & udptable->mask;
483 hslot2 = &udptable->hash2[slot2];
484
485 /* Lookup connected or non-wildcard socket */
486 result = udp4_lib_lookup2(net, saddr, sport,
487 daddr, hnum, dif, sdif,
488 hslot2, skb);
489 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
490 goto done;
491
492 /* Lookup redirect from BPF */
493 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
494 udptable == net->ipv4.udp_table) {
495 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
496 saddr, sport, daddr, hnum, dif,
497 udp_ehashfn);
498 if (sk) {
499 result = sk;
500 goto done;
501 }
502 }
503
504 /* Got non-wildcard socket or error on first lookup */
505 if (result)
506 goto done;
507
508 /* Lookup wildcard sockets */
509 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
510 slot2 = hash2 & udptable->mask;
511 hslot2 = &udptable->hash2[slot2];
512
513 result = udp4_lib_lookup2(net, saddr, sport,
514 htonl(INADDR_ANY), hnum, dif, sdif,
515 hslot2, skb);
516done:
517 if (IS_ERR(result))
518 return NULL;
519 return result;
520}
521EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
522
523static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
524 __be16 sport, __be16 dport,
525 struct udp_table *udptable)
526{
527 const struct iphdr *iph = ip_hdr(skb);
528
529 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
530 iph->daddr, dport, inet_iif(skb),
531 inet_sdif(skb), udptable, skb);
532}
533
534struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
535 __be16 sport, __be16 dport)
536{
537 const struct iphdr *iph = ip_hdr(skb);
538 struct net *net = dev_net(skb->dev);
539 int iif, sdif;
540
541 inet_get_iif_sdif(skb, &iif, &sdif);
542
543 return __udp4_lib_lookup(net, iph->saddr, sport,
544 iph->daddr, dport, iif,
545 sdif, net->ipv4.udp_table, NULL);
546}
547
548/* Must be called under rcu_read_lock().
549 * Does increment socket refcount.
550 */
551#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
552struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
553 __be32 daddr, __be16 dport, int dif)
554{
555 struct sock *sk;
556
557 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
558 dif, 0, net->ipv4.udp_table, NULL);
559 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
560 sk = NULL;
561 return sk;
562}
563EXPORT_SYMBOL_GPL(udp4_lib_lookup);
564#endif
565
566static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
567 __be16 loc_port, __be32 loc_addr,
568 __be16 rmt_port, __be32 rmt_addr,
569 int dif, int sdif, unsigned short hnum)
570{
571 const struct inet_sock *inet = inet_sk(sk);
572
573 if (!net_eq(sock_net(sk), net) ||
574 udp_sk(sk)->udp_port_hash != hnum ||
575 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
576 (inet->inet_dport != rmt_port && inet->inet_dport) ||
577 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
578 ipv6_only_sock(sk) ||
579 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
580 return false;
581 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
582 return false;
583 return true;
584}
585
586DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
587void udp_encap_enable(void)
588{
589 static_branch_inc(&udp_encap_needed_key);
590}
591EXPORT_SYMBOL(udp_encap_enable);
592
593void udp_encap_disable(void)
594{
595 static_branch_dec(&udp_encap_needed_key);
596}
597EXPORT_SYMBOL(udp_encap_disable);
598
599/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
600 * through error handlers in encapsulations looking for a match.
601 */
602static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
603{
604 int i;
605
606 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
607 int (*handler)(struct sk_buff *skb, u32 info);
608 const struct ip_tunnel_encap_ops *encap;
609
610 encap = rcu_dereference(iptun_encaps[i]);
611 if (!encap)
612 continue;
613 handler = encap->err_handler;
614 if (handler && !handler(skb, info))
615 return 0;
616 }
617
618 return -ENOENT;
619}
620
621/* Try to match ICMP errors to UDP tunnels by looking up a socket without
622 * reversing source and destination port: this will match tunnels that force the
623 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
624 * lwtunnels might actually break this assumption by being configured with
625 * different destination ports on endpoints, in this case we won't be able to
626 * trace ICMP messages back to them.
627 *
628 * If this doesn't match any socket, probe tunnels with arbitrary destination
629 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
630 * we've sent packets to won't necessarily match the local destination port.
631 *
632 * Then ask the tunnel implementation to match the error against a valid
633 * association.
634 *
635 * Return an error if we can't find a match, the socket if we need further
636 * processing, zero otherwise.
637 */
638static struct sock *__udp4_lib_err_encap(struct net *net,
639 const struct iphdr *iph,
640 struct udphdr *uh,
641 struct udp_table *udptable,
642 struct sock *sk,
643 struct sk_buff *skb, u32 info)
644{
645 int (*lookup)(struct sock *sk, struct sk_buff *skb);
646 int network_offset, transport_offset;
647 struct udp_sock *up;
648
649 network_offset = skb_network_offset(skb);
650 transport_offset = skb_transport_offset(skb);
651
652 /* Network header needs to point to the outer IPv4 header inside ICMP */
653 skb_reset_network_header(skb);
654
655 /* Transport header needs to point to the UDP header */
656 skb_set_transport_header(skb, iph->ihl << 2);
657
658 if (sk) {
659 up = udp_sk(sk);
660
661 lookup = READ_ONCE(up->encap_err_lookup);
662 if (lookup && lookup(sk, skb))
663 sk = NULL;
664
665 goto out;
666 }
667
668 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
669 iph->saddr, uh->dest, skb->dev->ifindex, 0,
670 udptable, NULL);
671 if (sk) {
672 up = udp_sk(sk);
673
674 lookup = READ_ONCE(up->encap_err_lookup);
675 if (!lookup || lookup(sk, skb))
676 sk = NULL;
677 }
678
679out:
680 if (!sk)
681 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
682
683 skb_set_transport_header(skb, transport_offset);
684 skb_set_network_header(skb, network_offset);
685
686 return sk;
687}
688
689/*
690 * This routine is called by the ICMP module when it gets some
691 * sort of error condition. If err < 0 then the socket should
692 * be closed and the error returned to the user. If err > 0
693 * it's just the icmp type << 8 | icmp code.
694 * Header points to the ip header of the error packet. We move
695 * on past this. Then (as it used to claim before adjustment)
696 * header points to the first 8 bytes of the udp header. We need
697 * to find the appropriate port.
698 */
699
700int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
701{
702 struct inet_sock *inet;
703 const struct iphdr *iph = (const struct iphdr *)skb->data;
704 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
705 const int type = icmp_hdr(skb)->type;
706 const int code = icmp_hdr(skb)->code;
707 bool tunnel = false;
708 struct sock *sk;
709 int harderr;
710 int err;
711 struct net *net = dev_net(skb->dev);
712
713 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
714 iph->saddr, uh->source, skb->dev->ifindex,
715 inet_sdif(skb), udptable, NULL);
716
717 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
718 /* No socket for error: try tunnels before discarding */
719 if (static_branch_unlikely(&udp_encap_needed_key)) {
720 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
721 info);
722 if (!sk)
723 return 0;
724 } else
725 sk = ERR_PTR(-ENOENT);
726
727 if (IS_ERR(sk)) {
728 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
729 return PTR_ERR(sk);
730 }
731
732 tunnel = true;
733 }
734
735 err = 0;
736 harderr = 0;
737 inet = inet_sk(sk);
738
739 switch (type) {
740 default:
741 case ICMP_TIME_EXCEEDED:
742 err = EHOSTUNREACH;
743 break;
744 case ICMP_SOURCE_QUENCH:
745 goto out;
746 case ICMP_PARAMETERPROB:
747 err = EPROTO;
748 harderr = 1;
749 break;
750 case ICMP_DEST_UNREACH:
751 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
752 ipv4_sk_update_pmtu(skb, sk, info);
753 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
754 err = EMSGSIZE;
755 harderr = 1;
756 break;
757 }
758 goto out;
759 }
760 err = EHOSTUNREACH;
761 if (code <= NR_ICMP_UNREACH) {
762 harderr = icmp_err_convert[code].fatal;
763 err = icmp_err_convert[code].errno;
764 }
765 break;
766 case ICMP_REDIRECT:
767 ipv4_sk_redirect(skb, sk);
768 goto out;
769 }
770
771 /*
772 * RFC1122: OK. Passes ICMP errors back to application, as per
773 * 4.1.3.3.
774 */
775 if (tunnel) {
776 /* ...not for tunnels though: we don't have a sending socket */
777 if (udp_sk(sk)->encap_err_rcv)
778 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
779 (u8 *)(uh+1));
780 goto out;
781 }
782 if (!inet_test_bit(RECVERR, sk)) {
783 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
784 goto out;
785 } else
786 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
787
788 sk->sk_err = err;
789 sk_error_report(sk);
790out:
791 return 0;
792}
793
794int udp_err(struct sk_buff *skb, u32 info)
795{
796 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
797}
798
799/*
800 * Throw away all pending data and cancel the corking. Socket is locked.
801 */
802void udp_flush_pending_frames(struct sock *sk)
803{
804 struct udp_sock *up = udp_sk(sk);
805
806 if (up->pending) {
807 up->len = 0;
808 WRITE_ONCE(up->pending, 0);
809 ip_flush_pending_frames(sk);
810 }
811}
812EXPORT_SYMBOL(udp_flush_pending_frames);
813
814/**
815 * udp4_hwcsum - handle outgoing HW checksumming
816 * @skb: sk_buff containing the filled-in UDP header
817 * (checksum field must be zeroed out)
818 * @src: source IP address
819 * @dst: destination IP address
820 */
821void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
822{
823 struct udphdr *uh = udp_hdr(skb);
824 int offset = skb_transport_offset(skb);
825 int len = skb->len - offset;
826 int hlen = len;
827 __wsum csum = 0;
828
829 if (!skb_has_frag_list(skb)) {
830 /*
831 * Only one fragment on the socket.
832 */
833 skb->csum_start = skb_transport_header(skb) - skb->head;
834 skb->csum_offset = offsetof(struct udphdr, check);
835 uh->check = ~csum_tcpudp_magic(src, dst, len,
836 IPPROTO_UDP, 0);
837 } else {
838 struct sk_buff *frags;
839
840 /*
841 * HW-checksum won't work as there are two or more
842 * fragments on the socket so that all csums of sk_buffs
843 * should be together
844 */
845 skb_walk_frags(skb, frags) {
846 csum = csum_add(csum, frags->csum);
847 hlen -= frags->len;
848 }
849
850 csum = skb_checksum(skb, offset, hlen, csum);
851 skb->ip_summed = CHECKSUM_NONE;
852
853 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
854 if (uh->check == 0)
855 uh->check = CSUM_MANGLED_0;
856 }
857}
858EXPORT_SYMBOL_GPL(udp4_hwcsum);
859
860/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
861 * for the simple case like when setting the checksum for a UDP tunnel.
862 */
863void udp_set_csum(bool nocheck, struct sk_buff *skb,
864 __be32 saddr, __be32 daddr, int len)
865{
866 struct udphdr *uh = udp_hdr(skb);
867
868 if (nocheck) {
869 uh->check = 0;
870 } else if (skb_is_gso(skb)) {
871 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
872 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
873 uh->check = 0;
874 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
875 if (uh->check == 0)
876 uh->check = CSUM_MANGLED_0;
877 } else {
878 skb->ip_summed = CHECKSUM_PARTIAL;
879 skb->csum_start = skb_transport_header(skb) - skb->head;
880 skb->csum_offset = offsetof(struct udphdr, check);
881 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
882 }
883}
884EXPORT_SYMBOL(udp_set_csum);
885
886static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
887 struct inet_cork *cork)
888{
889 struct sock *sk = skb->sk;
890 struct inet_sock *inet = inet_sk(sk);
891 struct udphdr *uh;
892 int err;
893 int is_udplite = IS_UDPLITE(sk);
894 int offset = skb_transport_offset(skb);
895 int len = skb->len - offset;
896 int datalen = len - sizeof(*uh);
897 __wsum csum = 0;
898
899 /*
900 * Create a UDP header
901 */
902 uh = udp_hdr(skb);
903 uh->source = inet->inet_sport;
904 uh->dest = fl4->fl4_dport;
905 uh->len = htons(len);
906 uh->check = 0;
907
908 if (cork->gso_size) {
909 const int hlen = skb_network_header_len(skb) +
910 sizeof(struct udphdr);
911
912 if (hlen + cork->gso_size > cork->fragsize) {
913 kfree_skb(skb);
914 return -EINVAL;
915 }
916 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
917 kfree_skb(skb);
918 return -EINVAL;
919 }
920 if (sk->sk_no_check_tx) {
921 kfree_skb(skb);
922 return -EINVAL;
923 }
924 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
925 dst_xfrm(skb_dst(skb))) {
926 kfree_skb(skb);
927 return -EIO;
928 }
929
930 if (datalen > cork->gso_size) {
931 skb_shinfo(skb)->gso_size = cork->gso_size;
932 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
933 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
934 cork->gso_size);
935 }
936 goto csum_partial;
937 }
938
939 if (is_udplite) /* UDP-Lite */
940 csum = udplite_csum(skb);
941
942 else if (sk->sk_no_check_tx) { /* UDP csum off */
943
944 skb->ip_summed = CHECKSUM_NONE;
945 goto send;
946
947 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
948csum_partial:
949
950 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
951 goto send;
952
953 } else
954 csum = udp_csum(skb);
955
956 /* add protocol-dependent pseudo-header */
957 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
958 sk->sk_protocol, csum);
959 if (uh->check == 0)
960 uh->check = CSUM_MANGLED_0;
961
962send:
963 err = ip_send_skb(sock_net(sk), skb);
964 if (err) {
965 if (err == -ENOBUFS &&
966 !inet_test_bit(RECVERR, sk)) {
967 UDP_INC_STATS(sock_net(sk),
968 UDP_MIB_SNDBUFERRORS, is_udplite);
969 err = 0;
970 }
971 } else
972 UDP_INC_STATS(sock_net(sk),
973 UDP_MIB_OUTDATAGRAMS, is_udplite);
974 return err;
975}
976
977/*
978 * Push out all pending data as one UDP datagram. Socket is locked.
979 */
980int udp_push_pending_frames(struct sock *sk)
981{
982 struct udp_sock *up = udp_sk(sk);
983 struct inet_sock *inet = inet_sk(sk);
984 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
985 struct sk_buff *skb;
986 int err = 0;
987
988 skb = ip_finish_skb(sk, fl4);
989 if (!skb)
990 goto out;
991
992 err = udp_send_skb(skb, fl4, &inet->cork.base);
993
994out:
995 up->len = 0;
996 WRITE_ONCE(up->pending, 0);
997 return err;
998}
999EXPORT_SYMBOL(udp_push_pending_frames);
1000
1001static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1002{
1003 switch (cmsg->cmsg_type) {
1004 case UDP_SEGMENT:
1005 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1006 return -EINVAL;
1007 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1008 return 0;
1009 default:
1010 return -EINVAL;
1011 }
1012}
1013
1014int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1015{
1016 struct cmsghdr *cmsg;
1017 bool need_ip = false;
1018 int err;
1019
1020 for_each_cmsghdr(cmsg, msg) {
1021 if (!CMSG_OK(msg, cmsg))
1022 return -EINVAL;
1023
1024 if (cmsg->cmsg_level != SOL_UDP) {
1025 need_ip = true;
1026 continue;
1027 }
1028
1029 err = __udp_cmsg_send(cmsg, gso_size);
1030 if (err)
1031 return err;
1032 }
1033
1034 return need_ip;
1035}
1036EXPORT_SYMBOL_GPL(udp_cmsg_send);
1037
1038int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1039{
1040 struct inet_sock *inet = inet_sk(sk);
1041 struct udp_sock *up = udp_sk(sk);
1042 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1043 struct flowi4 fl4_stack;
1044 struct flowi4 *fl4;
1045 int ulen = len;
1046 struct ipcm_cookie ipc;
1047 struct rtable *rt = NULL;
1048 int free = 0;
1049 int connected = 0;
1050 __be32 daddr, faddr, saddr;
1051 u8 tos, scope;
1052 __be16 dport;
1053 int err, is_udplite = IS_UDPLITE(sk);
1054 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1055 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1056 struct sk_buff *skb;
1057 struct ip_options_data opt_copy;
1058 int uc_index;
1059
1060 if (len > 0xFFFF)
1061 return -EMSGSIZE;
1062
1063 /*
1064 * Check the flags.
1065 */
1066
1067 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1068 return -EOPNOTSUPP;
1069
1070 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1071
1072 fl4 = &inet->cork.fl.u.ip4;
1073 if (READ_ONCE(up->pending)) {
1074 /*
1075 * There are pending frames.
1076 * The socket lock must be held while it's corked.
1077 */
1078 lock_sock(sk);
1079 if (likely(up->pending)) {
1080 if (unlikely(up->pending != AF_INET)) {
1081 release_sock(sk);
1082 return -EINVAL;
1083 }
1084 goto do_append_data;
1085 }
1086 release_sock(sk);
1087 }
1088 ulen += sizeof(struct udphdr);
1089
1090 /*
1091 * Get and verify the address.
1092 */
1093 if (usin) {
1094 if (msg->msg_namelen < sizeof(*usin))
1095 return -EINVAL;
1096 if (usin->sin_family != AF_INET) {
1097 if (usin->sin_family != AF_UNSPEC)
1098 return -EAFNOSUPPORT;
1099 }
1100
1101 daddr = usin->sin_addr.s_addr;
1102 dport = usin->sin_port;
1103 if (dport == 0)
1104 return -EINVAL;
1105 } else {
1106 if (sk->sk_state != TCP_ESTABLISHED)
1107 return -EDESTADDRREQ;
1108 daddr = inet->inet_daddr;
1109 dport = inet->inet_dport;
1110 /* Open fast path for connected socket.
1111 Route will not be used, if at least one option is set.
1112 */
1113 connected = 1;
1114 }
1115
1116 ipcm_init_sk(&ipc, inet);
1117 ipc.gso_size = READ_ONCE(up->gso_size);
1118
1119 if (msg->msg_controllen) {
1120 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1121 if (err > 0)
1122 err = ip_cmsg_send(sk, msg, &ipc,
1123 sk->sk_family == AF_INET6);
1124 if (unlikely(err < 0)) {
1125 kfree(ipc.opt);
1126 return err;
1127 }
1128 if (ipc.opt)
1129 free = 1;
1130 connected = 0;
1131 }
1132 if (!ipc.opt) {
1133 struct ip_options_rcu *inet_opt;
1134
1135 rcu_read_lock();
1136 inet_opt = rcu_dereference(inet->inet_opt);
1137 if (inet_opt) {
1138 memcpy(&opt_copy, inet_opt,
1139 sizeof(*inet_opt) + inet_opt->opt.optlen);
1140 ipc.opt = &opt_copy.opt;
1141 }
1142 rcu_read_unlock();
1143 }
1144
1145 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1146 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1147 (struct sockaddr *)usin,
1148 &msg->msg_namelen,
1149 &ipc.addr);
1150 if (err)
1151 goto out_free;
1152 if (usin) {
1153 if (usin->sin_port == 0) {
1154 /* BPF program set invalid port. Reject it. */
1155 err = -EINVAL;
1156 goto out_free;
1157 }
1158 daddr = usin->sin_addr.s_addr;
1159 dport = usin->sin_port;
1160 }
1161 }
1162
1163 saddr = ipc.addr;
1164 ipc.addr = faddr = daddr;
1165
1166 if (ipc.opt && ipc.opt->opt.srr) {
1167 if (!daddr) {
1168 err = -EINVAL;
1169 goto out_free;
1170 }
1171 faddr = ipc.opt->opt.faddr;
1172 connected = 0;
1173 }
1174 tos = get_rttos(&ipc, inet);
1175 scope = ip_sendmsg_scope(inet, &ipc, msg);
1176 if (scope == RT_SCOPE_LINK)
1177 connected = 0;
1178
1179 uc_index = READ_ONCE(inet->uc_index);
1180 if (ipv4_is_multicast(daddr)) {
1181 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1182 ipc.oif = READ_ONCE(inet->mc_index);
1183 if (!saddr)
1184 saddr = READ_ONCE(inet->mc_addr);
1185 connected = 0;
1186 } else if (!ipc.oif) {
1187 ipc.oif = uc_index;
1188 } else if (ipv4_is_lbcast(daddr) && uc_index) {
1189 /* oif is set, packet is to local broadcast and
1190 * uc_index is set. oif is most likely set
1191 * by sk_bound_dev_if. If uc_index != oif check if the
1192 * oif is an L3 master and uc_index is an L3 slave.
1193 * If so, we want to allow the send using the uc_index.
1194 */
1195 if (ipc.oif != uc_index &&
1196 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1197 uc_index)) {
1198 ipc.oif = uc_index;
1199 }
1200 }
1201
1202 if (connected)
1203 rt = (struct rtable *)sk_dst_check(sk, 0);
1204
1205 if (!rt) {
1206 struct net *net = sock_net(sk);
1207 __u8 flow_flags = inet_sk_flowi_flags(sk);
1208
1209 fl4 = &fl4_stack;
1210
1211 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1212 sk->sk_protocol, flow_flags, faddr, saddr,
1213 dport, inet->inet_sport, sk->sk_uid);
1214
1215 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1216 rt = ip_route_output_flow(net, fl4, sk);
1217 if (IS_ERR(rt)) {
1218 err = PTR_ERR(rt);
1219 rt = NULL;
1220 if (err == -ENETUNREACH)
1221 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1222 goto out;
1223 }
1224
1225 err = -EACCES;
1226 if ((rt->rt_flags & RTCF_BROADCAST) &&
1227 !sock_flag(sk, SOCK_BROADCAST))
1228 goto out;
1229 if (connected)
1230 sk_dst_set(sk, dst_clone(&rt->dst));
1231 }
1232
1233 if (msg->msg_flags&MSG_CONFIRM)
1234 goto do_confirm;
1235back_from_confirm:
1236
1237 saddr = fl4->saddr;
1238 if (!ipc.addr)
1239 daddr = ipc.addr = fl4->daddr;
1240
1241 /* Lockless fast path for the non-corking case. */
1242 if (!corkreq) {
1243 struct inet_cork cork;
1244
1245 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1246 sizeof(struct udphdr), &ipc, &rt,
1247 &cork, msg->msg_flags);
1248 err = PTR_ERR(skb);
1249 if (!IS_ERR_OR_NULL(skb))
1250 err = udp_send_skb(skb, fl4, &cork);
1251 goto out;
1252 }
1253
1254 lock_sock(sk);
1255 if (unlikely(up->pending)) {
1256 /* The socket is already corked while preparing it. */
1257 /* ... which is an evident application bug. --ANK */
1258 release_sock(sk);
1259
1260 net_dbg_ratelimited("socket already corked\n");
1261 err = -EINVAL;
1262 goto out;
1263 }
1264 /*
1265 * Now cork the socket to pend data.
1266 */
1267 fl4 = &inet->cork.fl.u.ip4;
1268 fl4->daddr = daddr;
1269 fl4->saddr = saddr;
1270 fl4->fl4_dport = dport;
1271 fl4->fl4_sport = inet->inet_sport;
1272 WRITE_ONCE(up->pending, AF_INET);
1273
1274do_append_data:
1275 up->len += ulen;
1276 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1277 sizeof(struct udphdr), &ipc, &rt,
1278 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1279 if (err)
1280 udp_flush_pending_frames(sk);
1281 else if (!corkreq)
1282 err = udp_push_pending_frames(sk);
1283 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1284 WRITE_ONCE(up->pending, 0);
1285 release_sock(sk);
1286
1287out:
1288 ip_rt_put(rt);
1289out_free:
1290 if (free)
1291 kfree(ipc.opt);
1292 if (!err)
1293 return len;
1294 /*
1295 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1296 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1297 * we don't have a good statistic (IpOutDiscards but it can be too many
1298 * things). We could add another new stat but at least for now that
1299 * seems like overkill.
1300 */
1301 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1302 UDP_INC_STATS(sock_net(sk),
1303 UDP_MIB_SNDBUFERRORS, is_udplite);
1304 }
1305 return err;
1306
1307do_confirm:
1308 if (msg->msg_flags & MSG_PROBE)
1309 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1310 if (!(msg->msg_flags&MSG_PROBE) || len)
1311 goto back_from_confirm;
1312 err = 0;
1313 goto out;
1314}
1315EXPORT_SYMBOL(udp_sendmsg);
1316
1317void udp_splice_eof(struct socket *sock)
1318{
1319 struct sock *sk = sock->sk;
1320 struct udp_sock *up = udp_sk(sk);
1321
1322 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1323 return;
1324
1325 lock_sock(sk);
1326 if (up->pending && !udp_test_bit(CORK, sk))
1327 udp_push_pending_frames(sk);
1328 release_sock(sk);
1329}
1330EXPORT_SYMBOL_GPL(udp_splice_eof);
1331
1332#define UDP_SKB_IS_STATELESS 0x80000000
1333
1334/* all head states (dst, sk, nf conntrack) except skb extensions are
1335 * cleared by udp_rcv().
1336 *
1337 * We need to preserve secpath, if present, to eventually process
1338 * IP_CMSG_PASSSEC at recvmsg() time.
1339 *
1340 * Other extensions can be cleared.
1341 */
1342static bool udp_try_make_stateless(struct sk_buff *skb)
1343{
1344 if (!skb_has_extensions(skb))
1345 return true;
1346
1347 if (!secpath_exists(skb)) {
1348 skb_ext_reset(skb);
1349 return true;
1350 }
1351
1352 return false;
1353}
1354
1355static void udp_set_dev_scratch(struct sk_buff *skb)
1356{
1357 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1358
1359 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1360 scratch->_tsize_state = skb->truesize;
1361#if BITS_PER_LONG == 64
1362 scratch->len = skb->len;
1363 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1364 scratch->is_linear = !skb_is_nonlinear(skb);
1365#endif
1366 if (udp_try_make_stateless(skb))
1367 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1368}
1369
1370static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1371{
1372 /* We come here after udp_lib_checksum_complete() returned 0.
1373 * This means that __skb_checksum_complete() might have
1374 * set skb->csum_valid to 1.
1375 * On 64bit platforms, we can set csum_unnecessary
1376 * to true, but only if the skb is not shared.
1377 */
1378#if BITS_PER_LONG == 64
1379 if (!skb_shared(skb))
1380 udp_skb_scratch(skb)->csum_unnecessary = true;
1381#endif
1382}
1383
1384static int udp_skb_truesize(struct sk_buff *skb)
1385{
1386 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1387}
1388
1389static bool udp_skb_has_head_state(struct sk_buff *skb)
1390{
1391 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1392}
1393
1394/* fully reclaim rmem/fwd memory allocated for skb */
1395static void udp_rmem_release(struct sock *sk, int size, int partial,
1396 bool rx_queue_lock_held)
1397{
1398 struct udp_sock *up = udp_sk(sk);
1399 struct sk_buff_head *sk_queue;
1400 int amt;
1401
1402 if (likely(partial)) {
1403 up->forward_deficit += size;
1404 size = up->forward_deficit;
1405 if (size < READ_ONCE(up->forward_threshold) &&
1406 !skb_queue_empty(&up->reader_queue))
1407 return;
1408 } else {
1409 size += up->forward_deficit;
1410 }
1411 up->forward_deficit = 0;
1412
1413 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1414 * if the called don't held it already
1415 */
1416 sk_queue = &sk->sk_receive_queue;
1417 if (!rx_queue_lock_held)
1418 spin_lock(&sk_queue->lock);
1419
1420
1421 sk_forward_alloc_add(sk, size);
1422 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1423 sk_forward_alloc_add(sk, -amt);
1424
1425 if (amt)
1426 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1427
1428 atomic_sub(size, &sk->sk_rmem_alloc);
1429
1430 /* this can save us from acquiring the rx queue lock on next receive */
1431 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1432
1433 if (!rx_queue_lock_held)
1434 spin_unlock(&sk_queue->lock);
1435}
1436
1437/* Note: called with reader_queue.lock held.
1438 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1439 * This avoids a cache line miss while receive_queue lock is held.
1440 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1441 */
1442void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1443{
1444 prefetch(&skb->data);
1445 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1446}
1447EXPORT_SYMBOL(udp_skb_destructor);
1448
1449/* as above, but the caller held the rx queue lock, too */
1450static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1451{
1452 prefetch(&skb->data);
1453 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1454}
1455
1456/* Idea of busylocks is to let producers grab an extra spinlock
1457 * to relieve pressure on the receive_queue spinlock shared by consumer.
1458 * Under flood, this means that only one producer can be in line
1459 * trying to acquire the receive_queue spinlock.
1460 * These busylock can be allocated on a per cpu manner, instead of a
1461 * per socket one (that would consume a cache line per socket)
1462 */
1463static int udp_busylocks_log __read_mostly;
1464static spinlock_t *udp_busylocks __read_mostly;
1465
1466static spinlock_t *busylock_acquire(void *ptr)
1467{
1468 spinlock_t *busy;
1469
1470 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1471 spin_lock(busy);
1472 return busy;
1473}
1474
1475static void busylock_release(spinlock_t *busy)
1476{
1477 if (busy)
1478 spin_unlock(busy);
1479}
1480
1481static int udp_rmem_schedule(struct sock *sk, int size)
1482{
1483 int delta;
1484
1485 delta = size - sk->sk_forward_alloc;
1486 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1487 return -ENOBUFS;
1488
1489 return 0;
1490}
1491
1492int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1493{
1494 struct sk_buff_head *list = &sk->sk_receive_queue;
1495 int rmem, err = -ENOMEM;
1496 spinlock_t *busy = NULL;
1497 int size;
1498
1499 /* try to avoid the costly atomic add/sub pair when the receive
1500 * queue is full; always allow at least a packet
1501 */
1502 rmem = atomic_read(&sk->sk_rmem_alloc);
1503 if (rmem > sk->sk_rcvbuf)
1504 goto drop;
1505
1506 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1507 * having linear skbs :
1508 * - Reduce memory overhead and thus increase receive queue capacity
1509 * - Less cache line misses at copyout() time
1510 * - Less work at consume_skb() (less alien page frag freeing)
1511 */
1512 if (rmem > (sk->sk_rcvbuf >> 1)) {
1513 skb_condense(skb);
1514
1515 busy = busylock_acquire(sk);
1516 }
1517 size = skb->truesize;
1518 udp_set_dev_scratch(skb);
1519
1520 /* we drop only if the receive buf is full and the receive
1521 * queue contains some other skb
1522 */
1523 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1524 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1525 goto uncharge_drop;
1526
1527 spin_lock(&list->lock);
1528 err = udp_rmem_schedule(sk, size);
1529 if (err) {
1530 spin_unlock(&list->lock);
1531 goto uncharge_drop;
1532 }
1533
1534 sk_forward_alloc_add(sk, -size);
1535
1536 /* no need to setup a destructor, we will explicitly release the
1537 * forward allocated memory on dequeue
1538 */
1539 sock_skb_set_dropcount(sk, skb);
1540
1541 __skb_queue_tail(list, skb);
1542 spin_unlock(&list->lock);
1543
1544 if (!sock_flag(sk, SOCK_DEAD))
1545 INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
1546
1547 busylock_release(busy);
1548 return 0;
1549
1550uncharge_drop:
1551 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1552
1553drop:
1554 atomic_inc(&sk->sk_drops);
1555 busylock_release(busy);
1556 return err;
1557}
1558EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1559
1560void udp_destruct_common(struct sock *sk)
1561{
1562 /* reclaim completely the forward allocated memory */
1563 struct udp_sock *up = udp_sk(sk);
1564 unsigned int total = 0;
1565 struct sk_buff *skb;
1566
1567 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1568 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1569 total += skb->truesize;
1570 kfree_skb(skb);
1571 }
1572 udp_rmem_release(sk, total, 0, true);
1573}
1574EXPORT_SYMBOL_GPL(udp_destruct_common);
1575
1576static void udp_destruct_sock(struct sock *sk)
1577{
1578 udp_destruct_common(sk);
1579 inet_sock_destruct(sk);
1580}
1581
1582int udp_init_sock(struct sock *sk)
1583{
1584 udp_lib_init_sock(sk);
1585 sk->sk_destruct = udp_destruct_sock;
1586 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1587 return 0;
1588}
1589
1590void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1591{
1592 sk_peek_offset_bwd(sk, len);
1593
1594 if (!skb_unref(skb))
1595 return;
1596
1597 /* In the more common cases we cleared the head states previously,
1598 * see __udp_queue_rcv_skb().
1599 */
1600 if (unlikely(udp_skb_has_head_state(skb)))
1601 skb_release_head_state(skb);
1602 __consume_stateless_skb(skb);
1603}
1604EXPORT_SYMBOL_GPL(skb_consume_udp);
1605
1606static struct sk_buff *__first_packet_length(struct sock *sk,
1607 struct sk_buff_head *rcvq,
1608 int *total)
1609{
1610 struct sk_buff *skb;
1611
1612 while ((skb = skb_peek(rcvq)) != NULL) {
1613 if (udp_lib_checksum_complete(skb)) {
1614 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1615 IS_UDPLITE(sk));
1616 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1617 IS_UDPLITE(sk));
1618 atomic_inc(&sk->sk_drops);
1619 __skb_unlink(skb, rcvq);
1620 *total += skb->truesize;
1621 kfree_skb(skb);
1622 } else {
1623 udp_skb_csum_unnecessary_set(skb);
1624 break;
1625 }
1626 }
1627 return skb;
1628}
1629
1630/**
1631 * first_packet_length - return length of first packet in receive queue
1632 * @sk: socket
1633 *
1634 * Drops all bad checksum frames, until a valid one is found.
1635 * Returns the length of found skb, or -1 if none is found.
1636 */
1637static int first_packet_length(struct sock *sk)
1638{
1639 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1640 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1641 struct sk_buff *skb;
1642 int total = 0;
1643 int res;
1644
1645 spin_lock_bh(&rcvq->lock);
1646 skb = __first_packet_length(sk, rcvq, &total);
1647 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1648 spin_lock(&sk_queue->lock);
1649 skb_queue_splice_tail_init(sk_queue, rcvq);
1650 spin_unlock(&sk_queue->lock);
1651
1652 skb = __first_packet_length(sk, rcvq, &total);
1653 }
1654 res = skb ? skb->len : -1;
1655 if (total)
1656 udp_rmem_release(sk, total, 1, false);
1657 spin_unlock_bh(&rcvq->lock);
1658 return res;
1659}
1660
1661/*
1662 * IOCTL requests applicable to the UDP protocol
1663 */
1664
1665int udp_ioctl(struct sock *sk, int cmd, int *karg)
1666{
1667 switch (cmd) {
1668 case SIOCOUTQ:
1669 {
1670 *karg = sk_wmem_alloc_get(sk);
1671 return 0;
1672 }
1673
1674 case SIOCINQ:
1675 {
1676 *karg = max_t(int, 0, first_packet_length(sk));
1677 return 0;
1678 }
1679
1680 default:
1681 return -ENOIOCTLCMD;
1682 }
1683
1684 return 0;
1685}
1686EXPORT_SYMBOL(udp_ioctl);
1687
1688struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1689 int *off, int *err)
1690{
1691 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1692 struct sk_buff_head *queue;
1693 struct sk_buff *last;
1694 long timeo;
1695 int error;
1696
1697 queue = &udp_sk(sk)->reader_queue;
1698 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1699 do {
1700 struct sk_buff *skb;
1701
1702 error = sock_error(sk);
1703 if (error)
1704 break;
1705
1706 error = -EAGAIN;
1707 do {
1708 spin_lock_bh(&queue->lock);
1709 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1710 err, &last);
1711 if (skb) {
1712 if (!(flags & MSG_PEEK))
1713 udp_skb_destructor(sk, skb);
1714 spin_unlock_bh(&queue->lock);
1715 return skb;
1716 }
1717
1718 if (skb_queue_empty_lockless(sk_queue)) {
1719 spin_unlock_bh(&queue->lock);
1720 goto busy_check;
1721 }
1722
1723 /* refill the reader queue and walk it again
1724 * keep both queues locked to avoid re-acquiring
1725 * the sk_receive_queue lock if fwd memory scheduling
1726 * is needed.
1727 */
1728 spin_lock(&sk_queue->lock);
1729 skb_queue_splice_tail_init(sk_queue, queue);
1730
1731 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1732 err, &last);
1733 if (skb && !(flags & MSG_PEEK))
1734 udp_skb_dtor_locked(sk, skb);
1735 spin_unlock(&sk_queue->lock);
1736 spin_unlock_bh(&queue->lock);
1737 if (skb)
1738 return skb;
1739
1740busy_check:
1741 if (!sk_can_busy_loop(sk))
1742 break;
1743
1744 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1745 } while (!skb_queue_empty_lockless(sk_queue));
1746
1747 /* sk_queue is empty, reader_queue may contain peeked packets */
1748 } while (timeo &&
1749 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1750 &error, &timeo,
1751 (struct sk_buff *)sk_queue));
1752
1753 *err = error;
1754 return NULL;
1755}
1756EXPORT_SYMBOL(__skb_recv_udp);
1757
1758int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1759{
1760 struct sk_buff *skb;
1761 int err;
1762
1763try_again:
1764 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1765 if (!skb)
1766 return err;
1767
1768 if (udp_lib_checksum_complete(skb)) {
1769 int is_udplite = IS_UDPLITE(sk);
1770 struct net *net = sock_net(sk);
1771
1772 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1773 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1774 atomic_inc(&sk->sk_drops);
1775 kfree_skb(skb);
1776 goto try_again;
1777 }
1778
1779 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1780 return recv_actor(sk, skb);
1781}
1782EXPORT_SYMBOL(udp_read_skb);
1783
1784/*
1785 * This should be easy, if there is something there we
1786 * return it, otherwise we block.
1787 */
1788
1789int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1790 int *addr_len)
1791{
1792 struct inet_sock *inet = inet_sk(sk);
1793 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1794 struct sk_buff *skb;
1795 unsigned int ulen, copied;
1796 int off, err, peeking = flags & MSG_PEEK;
1797 int is_udplite = IS_UDPLITE(sk);
1798 bool checksum_valid = false;
1799
1800 if (flags & MSG_ERRQUEUE)
1801 return ip_recv_error(sk, msg, len, addr_len);
1802
1803try_again:
1804 off = sk_peek_offset(sk, flags);
1805 skb = __skb_recv_udp(sk, flags, &off, &err);
1806 if (!skb)
1807 return err;
1808
1809 ulen = udp_skb_len(skb);
1810 copied = len;
1811 if (copied > ulen - off)
1812 copied = ulen - off;
1813 else if (copied < ulen)
1814 msg->msg_flags |= MSG_TRUNC;
1815
1816 /*
1817 * If checksum is needed at all, try to do it while copying the
1818 * data. If the data is truncated, or if we only want a partial
1819 * coverage checksum (UDP-Lite), do it before the copy.
1820 */
1821
1822 if (copied < ulen || peeking ||
1823 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1824 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1825 !__udp_lib_checksum_complete(skb);
1826 if (!checksum_valid)
1827 goto csum_copy_err;
1828 }
1829
1830 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1831 if (udp_skb_is_linear(skb))
1832 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1833 else
1834 err = skb_copy_datagram_msg(skb, off, msg, copied);
1835 } else {
1836 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1837
1838 if (err == -EINVAL)
1839 goto csum_copy_err;
1840 }
1841
1842 if (unlikely(err)) {
1843 if (!peeking) {
1844 atomic_inc(&sk->sk_drops);
1845 UDP_INC_STATS(sock_net(sk),
1846 UDP_MIB_INERRORS, is_udplite);
1847 }
1848 kfree_skb(skb);
1849 return err;
1850 }
1851
1852 if (!peeking)
1853 UDP_INC_STATS(sock_net(sk),
1854 UDP_MIB_INDATAGRAMS, is_udplite);
1855
1856 sock_recv_cmsgs(msg, sk, skb);
1857
1858 /* Copy the address. */
1859 if (sin) {
1860 sin->sin_family = AF_INET;
1861 sin->sin_port = udp_hdr(skb)->source;
1862 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1863 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1864 *addr_len = sizeof(*sin);
1865
1866 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1867 (struct sockaddr *)sin,
1868 addr_len);
1869 }
1870
1871 if (udp_test_bit(GRO_ENABLED, sk))
1872 udp_cmsg_recv(msg, sk, skb);
1873
1874 if (inet_cmsg_flags(inet))
1875 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1876
1877 err = copied;
1878 if (flags & MSG_TRUNC)
1879 err = ulen;
1880
1881 skb_consume_udp(sk, skb, peeking ? -err : err);
1882 return err;
1883
1884csum_copy_err:
1885 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1886 udp_skb_destructor)) {
1887 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1888 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1889 }
1890 kfree_skb(skb);
1891
1892 /* starting over for a new packet, but check if we need to yield */
1893 cond_resched();
1894 msg->msg_flags &= ~MSG_TRUNC;
1895 goto try_again;
1896}
1897
1898int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1899{
1900 /* This check is replicated from __ip4_datagram_connect() and
1901 * intended to prevent BPF program called below from accessing bytes
1902 * that are out of the bound specified by user in addr_len.
1903 */
1904 if (addr_len < sizeof(struct sockaddr_in))
1905 return -EINVAL;
1906
1907 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
1908}
1909EXPORT_SYMBOL(udp_pre_connect);
1910
1911int __udp_disconnect(struct sock *sk, int flags)
1912{
1913 struct inet_sock *inet = inet_sk(sk);
1914 /*
1915 * 1003.1g - break association.
1916 */
1917
1918 sk->sk_state = TCP_CLOSE;
1919 inet->inet_daddr = 0;
1920 inet->inet_dport = 0;
1921 sock_rps_reset_rxhash(sk);
1922 sk->sk_bound_dev_if = 0;
1923 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1924 inet_reset_saddr(sk);
1925 if (sk->sk_prot->rehash &&
1926 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1927 sk->sk_prot->rehash(sk);
1928 }
1929
1930 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1931 sk->sk_prot->unhash(sk);
1932 inet->inet_sport = 0;
1933 }
1934 sk_dst_reset(sk);
1935 return 0;
1936}
1937EXPORT_SYMBOL(__udp_disconnect);
1938
1939int udp_disconnect(struct sock *sk, int flags)
1940{
1941 lock_sock(sk);
1942 __udp_disconnect(sk, flags);
1943 release_sock(sk);
1944 return 0;
1945}
1946EXPORT_SYMBOL(udp_disconnect);
1947
1948void udp_lib_unhash(struct sock *sk)
1949{
1950 if (sk_hashed(sk)) {
1951 struct udp_table *udptable = udp_get_table_prot(sk);
1952 struct udp_hslot *hslot, *hslot2;
1953
1954 hslot = udp_hashslot(udptable, sock_net(sk),
1955 udp_sk(sk)->udp_port_hash);
1956 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1957
1958 spin_lock_bh(&hslot->lock);
1959 if (rcu_access_pointer(sk->sk_reuseport_cb))
1960 reuseport_detach_sock(sk);
1961 if (sk_del_node_init_rcu(sk)) {
1962 hslot->count--;
1963 inet_sk(sk)->inet_num = 0;
1964 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1965
1966 spin_lock(&hslot2->lock);
1967 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1968 hslot2->count--;
1969 spin_unlock(&hslot2->lock);
1970 }
1971 spin_unlock_bh(&hslot->lock);
1972 }
1973}
1974EXPORT_SYMBOL(udp_lib_unhash);
1975
1976/*
1977 * inet_rcv_saddr was changed, we must rehash secondary hash
1978 */
1979void udp_lib_rehash(struct sock *sk, u16 newhash)
1980{
1981 if (sk_hashed(sk)) {
1982 struct udp_table *udptable = udp_get_table_prot(sk);
1983 struct udp_hslot *hslot, *hslot2, *nhslot2;
1984
1985 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1986 nhslot2 = udp_hashslot2(udptable, newhash);
1987 udp_sk(sk)->udp_portaddr_hash = newhash;
1988
1989 if (hslot2 != nhslot2 ||
1990 rcu_access_pointer(sk->sk_reuseport_cb)) {
1991 hslot = udp_hashslot(udptable, sock_net(sk),
1992 udp_sk(sk)->udp_port_hash);
1993 /* we must lock primary chain too */
1994 spin_lock_bh(&hslot->lock);
1995 if (rcu_access_pointer(sk->sk_reuseport_cb))
1996 reuseport_detach_sock(sk);
1997
1998 if (hslot2 != nhslot2) {
1999 spin_lock(&hslot2->lock);
2000 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2001 hslot2->count--;
2002 spin_unlock(&hslot2->lock);
2003
2004 spin_lock(&nhslot2->lock);
2005 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2006 &nhslot2->head);
2007 nhslot2->count++;
2008 spin_unlock(&nhslot2->lock);
2009 }
2010
2011 spin_unlock_bh(&hslot->lock);
2012 }
2013 }
2014}
2015EXPORT_SYMBOL(udp_lib_rehash);
2016
2017void udp_v4_rehash(struct sock *sk)
2018{
2019 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2020 inet_sk(sk)->inet_rcv_saddr,
2021 inet_sk(sk)->inet_num);
2022 udp_lib_rehash(sk, new_hash);
2023}
2024
2025static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2026{
2027 int rc;
2028
2029 if (inet_sk(sk)->inet_daddr) {
2030 sock_rps_save_rxhash(sk, skb);
2031 sk_mark_napi_id(sk, skb);
2032 sk_incoming_cpu_update(sk);
2033 } else {
2034 sk_mark_napi_id_once(sk, skb);
2035 }
2036
2037 rc = __udp_enqueue_schedule_skb(sk, skb);
2038 if (rc < 0) {
2039 int is_udplite = IS_UDPLITE(sk);
2040 int drop_reason;
2041
2042 /* Note that an ENOMEM error is charged twice */
2043 if (rc == -ENOMEM) {
2044 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2045 is_udplite);
2046 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2047 } else {
2048 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2049 is_udplite);
2050 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2051 }
2052 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2053 kfree_skb_reason(skb, drop_reason);
2054 trace_udp_fail_queue_rcv_skb(rc, sk);
2055 return -1;
2056 }
2057
2058 return 0;
2059}
2060
2061/* returns:
2062 * -1: error
2063 * 0: success
2064 * >0: "udp encap" protocol resubmission
2065 *
2066 * Note that in the success and error cases, the skb is assumed to
2067 * have either been requeued or freed.
2068 */
2069static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2070{
2071 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2072 struct udp_sock *up = udp_sk(sk);
2073 int is_udplite = IS_UDPLITE(sk);
2074
2075 /*
2076 * Charge it to the socket, dropping if the queue is full.
2077 */
2078 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2079 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2080 goto drop;
2081 }
2082 nf_reset_ct(skb);
2083
2084 if (static_branch_unlikely(&udp_encap_needed_key) &&
2085 READ_ONCE(up->encap_type)) {
2086 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2087
2088 /*
2089 * This is an encapsulation socket so pass the skb to
2090 * the socket's udp_encap_rcv() hook. Otherwise, just
2091 * fall through and pass this up the UDP socket.
2092 * up->encap_rcv() returns the following value:
2093 * =0 if skb was successfully passed to the encap
2094 * handler or was discarded by it.
2095 * >0 if skb should be passed on to UDP.
2096 * <0 if skb should be resubmitted as proto -N
2097 */
2098
2099 /* if we're overly short, let UDP handle it */
2100 encap_rcv = READ_ONCE(up->encap_rcv);
2101 if (encap_rcv) {
2102 int ret;
2103
2104 /* Verify checksum before giving to encap */
2105 if (udp_lib_checksum_complete(skb))
2106 goto csum_error;
2107
2108 ret = encap_rcv(sk, skb);
2109 if (ret <= 0) {
2110 __UDP_INC_STATS(sock_net(sk),
2111 UDP_MIB_INDATAGRAMS,
2112 is_udplite);
2113 return -ret;
2114 }
2115 }
2116
2117 /* FALLTHROUGH -- it's a UDP Packet */
2118 }
2119
2120 /*
2121 * UDP-Lite specific tests, ignored on UDP sockets
2122 */
2123 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2124 u16 pcrlen = READ_ONCE(up->pcrlen);
2125
2126 /*
2127 * MIB statistics other than incrementing the error count are
2128 * disabled for the following two types of errors: these depend
2129 * on the application settings, not on the functioning of the
2130 * protocol stack as such.
2131 *
2132 * RFC 3828 here recommends (sec 3.3): "There should also be a
2133 * way ... to ... at least let the receiving application block
2134 * delivery of packets with coverage values less than a value
2135 * provided by the application."
2136 */
2137 if (pcrlen == 0) { /* full coverage was set */
2138 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2139 UDP_SKB_CB(skb)->cscov, skb->len);
2140 goto drop;
2141 }
2142 /* The next case involves violating the min. coverage requested
2143 * by the receiver. This is subtle: if receiver wants x and x is
2144 * greater than the buffersize/MTU then receiver will complain
2145 * that it wants x while sender emits packets of smaller size y.
2146 * Therefore the above ...()->partial_cov statement is essential.
2147 */
2148 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2149 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2150 UDP_SKB_CB(skb)->cscov, pcrlen);
2151 goto drop;
2152 }
2153 }
2154
2155 prefetch(&sk->sk_rmem_alloc);
2156 if (rcu_access_pointer(sk->sk_filter) &&
2157 udp_lib_checksum_complete(skb))
2158 goto csum_error;
2159
2160 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2161 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2162 goto drop;
2163 }
2164
2165 udp_csum_pull_header(skb);
2166
2167 ipv4_pktinfo_prepare(sk, skb, true);
2168 return __udp_queue_rcv_skb(sk, skb);
2169
2170csum_error:
2171 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2172 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2173drop:
2174 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2175 atomic_inc(&sk->sk_drops);
2176 kfree_skb_reason(skb, drop_reason);
2177 return -1;
2178}
2179
2180static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2181{
2182 struct sk_buff *next, *segs;
2183 int ret;
2184
2185 if (likely(!udp_unexpected_gso(sk, skb)))
2186 return udp_queue_rcv_one_skb(sk, skb);
2187
2188 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2189 __skb_push(skb, -skb_mac_offset(skb));
2190 segs = udp_rcv_segment(sk, skb, true);
2191 skb_list_walk_safe(segs, skb, next) {
2192 __skb_pull(skb, skb_transport_offset(skb));
2193
2194 udp_post_segment_fix_csum(skb);
2195 ret = udp_queue_rcv_one_skb(sk, skb);
2196 if (ret > 0)
2197 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2198 }
2199 return 0;
2200}
2201
2202/* For TCP sockets, sk_rx_dst is protected by socket lock
2203 * For UDP, we use xchg() to guard against concurrent changes.
2204 */
2205bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2206{
2207 struct dst_entry *old;
2208
2209 if (dst_hold_safe(dst)) {
2210 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2211 dst_release(old);
2212 return old != dst;
2213 }
2214 return false;
2215}
2216EXPORT_SYMBOL(udp_sk_rx_dst_set);
2217
2218/*
2219 * Multicasts and broadcasts go to each listener.
2220 *
2221 * Note: called only from the BH handler context.
2222 */
2223static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2224 struct udphdr *uh,
2225 __be32 saddr, __be32 daddr,
2226 struct udp_table *udptable,
2227 int proto)
2228{
2229 struct sock *sk, *first = NULL;
2230 unsigned short hnum = ntohs(uh->dest);
2231 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2232 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2233 unsigned int offset = offsetof(typeof(*sk), sk_node);
2234 int dif = skb->dev->ifindex;
2235 int sdif = inet_sdif(skb);
2236 struct hlist_node *node;
2237 struct sk_buff *nskb;
2238
2239 if (use_hash2) {
2240 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2241 udptable->mask;
2242 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2243start_lookup:
2244 hslot = &udptable->hash2[hash2];
2245 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2246 }
2247
2248 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2249 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2250 uh->source, saddr, dif, sdif, hnum))
2251 continue;
2252
2253 if (!first) {
2254 first = sk;
2255 continue;
2256 }
2257 nskb = skb_clone(skb, GFP_ATOMIC);
2258
2259 if (unlikely(!nskb)) {
2260 atomic_inc(&sk->sk_drops);
2261 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2262 IS_UDPLITE(sk));
2263 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2264 IS_UDPLITE(sk));
2265 continue;
2266 }
2267 if (udp_queue_rcv_skb(sk, nskb) > 0)
2268 consume_skb(nskb);
2269 }
2270
2271 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2272 if (use_hash2 && hash2 != hash2_any) {
2273 hash2 = hash2_any;
2274 goto start_lookup;
2275 }
2276
2277 if (first) {
2278 if (udp_queue_rcv_skb(first, skb) > 0)
2279 consume_skb(skb);
2280 } else {
2281 kfree_skb(skb);
2282 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2283 proto == IPPROTO_UDPLITE);
2284 }
2285 return 0;
2286}
2287
2288/* Initialize UDP checksum. If exited with zero value (success),
2289 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2290 * Otherwise, csum completion requires checksumming packet body,
2291 * including udp header and folding it to skb->csum.
2292 */
2293static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2294 int proto)
2295{
2296 int err;
2297
2298 UDP_SKB_CB(skb)->partial_cov = 0;
2299 UDP_SKB_CB(skb)->cscov = skb->len;
2300
2301 if (proto == IPPROTO_UDPLITE) {
2302 err = udplite_checksum_init(skb, uh);
2303 if (err)
2304 return err;
2305
2306 if (UDP_SKB_CB(skb)->partial_cov) {
2307 skb->csum = inet_compute_pseudo(skb, proto);
2308 return 0;
2309 }
2310 }
2311
2312 /* Note, we are only interested in != 0 or == 0, thus the
2313 * force to int.
2314 */
2315 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2316 inet_compute_pseudo);
2317 if (err)
2318 return err;
2319
2320 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2321 /* If SW calculated the value, we know it's bad */
2322 if (skb->csum_complete_sw)
2323 return 1;
2324
2325 /* HW says the value is bad. Let's validate that.
2326 * skb->csum is no longer the full packet checksum,
2327 * so don't treat it as such.
2328 */
2329 skb_checksum_complete_unset(skb);
2330 }
2331
2332 return 0;
2333}
2334
2335/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2336 * return code conversion for ip layer consumption
2337 */
2338static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2339 struct udphdr *uh)
2340{
2341 int ret;
2342
2343 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2344 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2345
2346 ret = udp_queue_rcv_skb(sk, skb);
2347
2348 /* a return value > 0 means to resubmit the input, but
2349 * it wants the return to be -protocol, or 0
2350 */
2351 if (ret > 0)
2352 return -ret;
2353 return 0;
2354}
2355
2356/*
2357 * All we need to do is get the socket, and then do a checksum.
2358 */
2359
2360int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2361 int proto)
2362{
2363 struct sock *sk;
2364 struct udphdr *uh;
2365 unsigned short ulen;
2366 struct rtable *rt = skb_rtable(skb);
2367 __be32 saddr, daddr;
2368 struct net *net = dev_net(skb->dev);
2369 bool refcounted;
2370 int drop_reason;
2371
2372 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2373
2374 /*
2375 * Validate the packet.
2376 */
2377 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2378 goto drop; /* No space for header. */
2379
2380 uh = udp_hdr(skb);
2381 ulen = ntohs(uh->len);
2382 saddr = ip_hdr(skb)->saddr;
2383 daddr = ip_hdr(skb)->daddr;
2384
2385 if (ulen > skb->len)
2386 goto short_packet;
2387
2388 if (proto == IPPROTO_UDP) {
2389 /* UDP validates ulen. */
2390 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2391 goto short_packet;
2392 uh = udp_hdr(skb);
2393 }
2394
2395 if (udp4_csum_init(skb, uh, proto))
2396 goto csum_error;
2397
2398 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2399 &refcounted, udp_ehashfn);
2400 if (IS_ERR(sk))
2401 goto no_sk;
2402
2403 if (sk) {
2404 struct dst_entry *dst = skb_dst(skb);
2405 int ret;
2406
2407 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2408 udp_sk_rx_dst_set(sk, dst);
2409
2410 ret = udp_unicast_rcv_skb(sk, skb, uh);
2411 if (refcounted)
2412 sock_put(sk);
2413 return ret;
2414 }
2415
2416 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2417 return __udp4_lib_mcast_deliver(net, skb, uh,
2418 saddr, daddr, udptable, proto);
2419
2420 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2421 if (sk)
2422 return udp_unicast_rcv_skb(sk, skb, uh);
2423no_sk:
2424 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2425 goto drop;
2426 nf_reset_ct(skb);
2427
2428 /* No socket. Drop packet silently, if checksum is wrong */
2429 if (udp_lib_checksum_complete(skb))
2430 goto csum_error;
2431
2432 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2433 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2434 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2435
2436 /*
2437 * Hmm. We got an UDP packet to a port to which we
2438 * don't wanna listen. Ignore it.
2439 */
2440 kfree_skb_reason(skb, drop_reason);
2441 return 0;
2442
2443short_packet:
2444 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2445 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2446 proto == IPPROTO_UDPLITE ? "Lite" : "",
2447 &saddr, ntohs(uh->source),
2448 ulen, skb->len,
2449 &daddr, ntohs(uh->dest));
2450 goto drop;
2451
2452csum_error:
2453 /*
2454 * RFC1122: OK. Discards the bad packet silently (as far as
2455 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2456 */
2457 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2458 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2459 proto == IPPROTO_UDPLITE ? "Lite" : "",
2460 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2461 ulen);
2462 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2463drop:
2464 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2465 kfree_skb_reason(skb, drop_reason);
2466 return 0;
2467}
2468
2469/* We can only early demux multicast if there is a single matching socket.
2470 * If more than one socket found returns NULL
2471 */
2472static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2473 __be16 loc_port, __be32 loc_addr,
2474 __be16 rmt_port, __be32 rmt_addr,
2475 int dif, int sdif)
2476{
2477 struct udp_table *udptable = net->ipv4.udp_table;
2478 unsigned short hnum = ntohs(loc_port);
2479 struct sock *sk, *result;
2480 struct udp_hslot *hslot;
2481 unsigned int slot;
2482
2483 slot = udp_hashfn(net, hnum, udptable->mask);
2484 hslot = &udptable->hash[slot];
2485
2486 /* Do not bother scanning a too big list */
2487 if (hslot->count > 10)
2488 return NULL;
2489
2490 result = NULL;
2491 sk_for_each_rcu(sk, &hslot->head) {
2492 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2493 rmt_port, rmt_addr, dif, sdif, hnum)) {
2494 if (result)
2495 return NULL;
2496 result = sk;
2497 }
2498 }
2499
2500 return result;
2501}
2502
2503/* For unicast we should only early demux connected sockets or we can
2504 * break forwarding setups. The chains here can be long so only check
2505 * if the first socket is an exact match and if not move on.
2506 */
2507static struct sock *__udp4_lib_demux_lookup(struct net *net,
2508 __be16 loc_port, __be32 loc_addr,
2509 __be16 rmt_port, __be32 rmt_addr,
2510 int dif, int sdif)
2511{
2512 struct udp_table *udptable = net->ipv4.udp_table;
2513 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2514 unsigned short hnum = ntohs(loc_port);
2515 unsigned int hash2, slot2;
2516 struct udp_hslot *hslot2;
2517 __portpair ports;
2518 struct sock *sk;
2519
2520 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2521 slot2 = hash2 & udptable->mask;
2522 hslot2 = &udptable->hash2[slot2];
2523 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2524
2525 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2526 if (inet_match(net, sk, acookie, ports, dif, sdif))
2527 return sk;
2528 /* Only check first socket in chain */
2529 break;
2530 }
2531 return NULL;
2532}
2533
2534int udp_v4_early_demux(struct sk_buff *skb)
2535{
2536 struct net *net = dev_net(skb->dev);
2537 struct in_device *in_dev = NULL;
2538 const struct iphdr *iph;
2539 const struct udphdr *uh;
2540 struct sock *sk = NULL;
2541 struct dst_entry *dst;
2542 int dif = skb->dev->ifindex;
2543 int sdif = inet_sdif(skb);
2544 int ours;
2545
2546 /* validate the packet */
2547 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2548 return 0;
2549
2550 iph = ip_hdr(skb);
2551 uh = udp_hdr(skb);
2552
2553 if (skb->pkt_type == PACKET_MULTICAST) {
2554 in_dev = __in_dev_get_rcu(skb->dev);
2555
2556 if (!in_dev)
2557 return 0;
2558
2559 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2560 iph->protocol);
2561 if (!ours)
2562 return 0;
2563
2564 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2565 uh->source, iph->saddr,
2566 dif, sdif);
2567 } else if (skb->pkt_type == PACKET_HOST) {
2568 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2569 uh->source, iph->saddr, dif, sdif);
2570 }
2571
2572 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2573 return 0;
2574
2575 skb->sk = sk;
2576 skb->destructor = sock_efree;
2577 dst = rcu_dereference(sk->sk_rx_dst);
2578
2579 if (dst)
2580 dst = dst_check(dst, 0);
2581 if (dst) {
2582 u32 itag = 0;
2583
2584 /* set noref for now.
2585 * any place which wants to hold dst has to call
2586 * dst_hold_safe()
2587 */
2588 skb_dst_set_noref(skb, dst);
2589
2590 /* for unconnected multicast sockets we need to validate
2591 * the source on each packet
2592 */
2593 if (!inet_sk(sk)->inet_daddr && in_dev)
2594 return ip_mc_validate_source(skb, iph->daddr,
2595 iph->saddr,
2596 iph->tos & IPTOS_RT_MASK,
2597 skb->dev, in_dev, &itag);
2598 }
2599 return 0;
2600}
2601
2602int udp_rcv(struct sk_buff *skb)
2603{
2604 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2605}
2606
2607void udp_destroy_sock(struct sock *sk)
2608{
2609 struct udp_sock *up = udp_sk(sk);
2610 bool slow = lock_sock_fast(sk);
2611
2612 /* protects from races with udp_abort() */
2613 sock_set_flag(sk, SOCK_DEAD);
2614 udp_flush_pending_frames(sk);
2615 unlock_sock_fast(sk, slow);
2616 if (static_branch_unlikely(&udp_encap_needed_key)) {
2617 if (up->encap_type) {
2618 void (*encap_destroy)(struct sock *sk);
2619 encap_destroy = READ_ONCE(up->encap_destroy);
2620 if (encap_destroy)
2621 encap_destroy(sk);
2622 }
2623 if (udp_test_bit(ENCAP_ENABLED, sk))
2624 static_branch_dec(&udp_encap_needed_key);
2625 }
2626}
2627
2628static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2629 struct sock *sk)
2630{
2631#ifdef CONFIG_XFRM
2632 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2633 if (family == AF_INET)
2634 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2635 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2636 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2637 }
2638#endif
2639}
2640
2641/*
2642 * Socket option code for UDP
2643 */
2644int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2645 sockptr_t optval, unsigned int optlen,
2646 int (*push_pending_frames)(struct sock *))
2647{
2648 struct udp_sock *up = udp_sk(sk);
2649 int val, valbool;
2650 int err = 0;
2651 int is_udplite = IS_UDPLITE(sk);
2652
2653 if (level == SOL_SOCKET) {
2654 err = sk_setsockopt(sk, level, optname, optval, optlen);
2655
2656 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2657 sockopt_lock_sock(sk);
2658 /* paired with READ_ONCE in udp_rmem_release() */
2659 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2660 sockopt_release_sock(sk);
2661 }
2662 return err;
2663 }
2664
2665 if (optlen < sizeof(int))
2666 return -EINVAL;
2667
2668 if (copy_from_sockptr(&val, optval, sizeof(val)))
2669 return -EFAULT;
2670
2671 valbool = val ? 1 : 0;
2672
2673 switch (optname) {
2674 case UDP_CORK:
2675 if (val != 0) {
2676 udp_set_bit(CORK, sk);
2677 } else {
2678 udp_clear_bit(CORK, sk);
2679 lock_sock(sk);
2680 push_pending_frames(sk);
2681 release_sock(sk);
2682 }
2683 break;
2684
2685 case UDP_ENCAP:
2686 switch (val) {
2687 case 0:
2688#ifdef CONFIG_XFRM
2689 case UDP_ENCAP_ESPINUDP:
2690 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2691 fallthrough;
2692 case UDP_ENCAP_ESPINUDP_NON_IKE:
2693#if IS_ENABLED(CONFIG_IPV6)
2694 if (sk->sk_family == AF_INET6)
2695 WRITE_ONCE(up->encap_rcv,
2696 ipv6_stub->xfrm6_udp_encap_rcv);
2697 else
2698#endif
2699 WRITE_ONCE(up->encap_rcv,
2700 xfrm4_udp_encap_rcv);
2701#endif
2702 fallthrough;
2703 case UDP_ENCAP_L2TPINUDP:
2704 WRITE_ONCE(up->encap_type, val);
2705 udp_tunnel_encap_enable(sk);
2706 break;
2707 default:
2708 err = -ENOPROTOOPT;
2709 break;
2710 }
2711 break;
2712
2713 case UDP_NO_CHECK6_TX:
2714 udp_set_no_check6_tx(sk, valbool);
2715 break;
2716
2717 case UDP_NO_CHECK6_RX:
2718 udp_set_no_check6_rx(sk, valbool);
2719 break;
2720
2721 case UDP_SEGMENT:
2722 if (val < 0 || val > USHRT_MAX)
2723 return -EINVAL;
2724 WRITE_ONCE(up->gso_size, val);
2725 break;
2726
2727 case UDP_GRO:
2728
2729 /* when enabling GRO, accept the related GSO packet type */
2730 if (valbool)
2731 udp_tunnel_encap_enable(sk);
2732 udp_assign_bit(GRO_ENABLED, sk, valbool);
2733 udp_assign_bit(ACCEPT_L4, sk, valbool);
2734 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
2735 break;
2736
2737 /*
2738 * UDP-Lite's partial checksum coverage (RFC 3828).
2739 */
2740 /* The sender sets actual checksum coverage length via this option.
2741 * The case coverage > packet length is handled by send module. */
2742 case UDPLITE_SEND_CSCOV:
2743 if (!is_udplite) /* Disable the option on UDP sockets */
2744 return -ENOPROTOOPT;
2745 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2746 val = 8;
2747 else if (val > USHRT_MAX)
2748 val = USHRT_MAX;
2749 WRITE_ONCE(up->pcslen, val);
2750 udp_set_bit(UDPLITE_SEND_CC, sk);
2751 break;
2752
2753 /* The receiver specifies a minimum checksum coverage value. To make
2754 * sense, this should be set to at least 8 (as done below). If zero is
2755 * used, this again means full checksum coverage. */
2756 case UDPLITE_RECV_CSCOV:
2757 if (!is_udplite) /* Disable the option on UDP sockets */
2758 return -ENOPROTOOPT;
2759 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2760 val = 8;
2761 else if (val > USHRT_MAX)
2762 val = USHRT_MAX;
2763 WRITE_ONCE(up->pcrlen, val);
2764 udp_set_bit(UDPLITE_RECV_CC, sk);
2765 break;
2766
2767 default:
2768 err = -ENOPROTOOPT;
2769 break;
2770 }
2771
2772 return err;
2773}
2774EXPORT_SYMBOL(udp_lib_setsockopt);
2775
2776int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2777 unsigned int optlen)
2778{
2779 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2780 return udp_lib_setsockopt(sk, level, optname,
2781 optval, optlen,
2782 udp_push_pending_frames);
2783 return ip_setsockopt(sk, level, optname, optval, optlen);
2784}
2785
2786int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2787 char __user *optval, int __user *optlen)
2788{
2789 struct udp_sock *up = udp_sk(sk);
2790 int val, len;
2791
2792 if (get_user(len, optlen))
2793 return -EFAULT;
2794
2795 len = min_t(unsigned int, len, sizeof(int));
2796
2797 if (len < 0)
2798 return -EINVAL;
2799
2800 switch (optname) {
2801 case UDP_CORK:
2802 val = udp_test_bit(CORK, sk);
2803 break;
2804
2805 case UDP_ENCAP:
2806 val = READ_ONCE(up->encap_type);
2807 break;
2808
2809 case UDP_NO_CHECK6_TX:
2810 val = udp_get_no_check6_tx(sk);
2811 break;
2812
2813 case UDP_NO_CHECK6_RX:
2814 val = udp_get_no_check6_rx(sk);
2815 break;
2816
2817 case UDP_SEGMENT:
2818 val = READ_ONCE(up->gso_size);
2819 break;
2820
2821 case UDP_GRO:
2822 val = udp_test_bit(GRO_ENABLED, sk);
2823 break;
2824
2825 /* The following two cannot be changed on UDP sockets, the return is
2826 * always 0 (which corresponds to the full checksum coverage of UDP). */
2827 case UDPLITE_SEND_CSCOV:
2828 val = READ_ONCE(up->pcslen);
2829 break;
2830
2831 case UDPLITE_RECV_CSCOV:
2832 val = READ_ONCE(up->pcrlen);
2833 break;
2834
2835 default:
2836 return -ENOPROTOOPT;
2837 }
2838
2839 if (put_user(len, optlen))
2840 return -EFAULT;
2841 if (copy_to_user(optval, &val, len))
2842 return -EFAULT;
2843 return 0;
2844}
2845EXPORT_SYMBOL(udp_lib_getsockopt);
2846
2847int udp_getsockopt(struct sock *sk, int level, int optname,
2848 char __user *optval, int __user *optlen)
2849{
2850 if (level == SOL_UDP || level == SOL_UDPLITE)
2851 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2852 return ip_getsockopt(sk, level, optname, optval, optlen);
2853}
2854
2855/**
2856 * udp_poll - wait for a UDP event.
2857 * @file: - file struct
2858 * @sock: - socket
2859 * @wait: - poll table
2860 *
2861 * This is same as datagram poll, except for the special case of
2862 * blocking sockets. If application is using a blocking fd
2863 * and a packet with checksum error is in the queue;
2864 * then it could get return from select indicating data available
2865 * but then block when reading it. Add special case code
2866 * to work around these arguably broken applications.
2867 */
2868__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2869{
2870 __poll_t mask = datagram_poll(file, sock, wait);
2871 struct sock *sk = sock->sk;
2872
2873 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2874 mask |= EPOLLIN | EPOLLRDNORM;
2875
2876 /* Check for false positives due to checksum errors */
2877 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2878 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2879 mask &= ~(EPOLLIN | EPOLLRDNORM);
2880
2881 /* psock ingress_msg queue should not contain any bad checksum frames */
2882 if (sk_is_readable(sk))
2883 mask |= EPOLLIN | EPOLLRDNORM;
2884 return mask;
2885
2886}
2887EXPORT_SYMBOL(udp_poll);
2888
2889int udp_abort(struct sock *sk, int err)
2890{
2891 if (!has_current_bpf_ctx())
2892 lock_sock(sk);
2893
2894 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2895 * with close()
2896 */
2897 if (sock_flag(sk, SOCK_DEAD))
2898 goto out;
2899
2900 sk->sk_err = err;
2901 sk_error_report(sk);
2902 __udp_disconnect(sk, 0);
2903
2904out:
2905 if (!has_current_bpf_ctx())
2906 release_sock(sk);
2907
2908 return 0;
2909}
2910EXPORT_SYMBOL_GPL(udp_abort);
2911
2912struct proto udp_prot = {
2913 .name = "UDP",
2914 .owner = THIS_MODULE,
2915 .close = udp_lib_close,
2916 .pre_connect = udp_pre_connect,
2917 .connect = ip4_datagram_connect,
2918 .disconnect = udp_disconnect,
2919 .ioctl = udp_ioctl,
2920 .init = udp_init_sock,
2921 .destroy = udp_destroy_sock,
2922 .setsockopt = udp_setsockopt,
2923 .getsockopt = udp_getsockopt,
2924 .sendmsg = udp_sendmsg,
2925 .recvmsg = udp_recvmsg,
2926 .splice_eof = udp_splice_eof,
2927 .release_cb = ip4_datagram_release_cb,
2928 .hash = udp_lib_hash,
2929 .unhash = udp_lib_unhash,
2930 .rehash = udp_v4_rehash,
2931 .get_port = udp_v4_get_port,
2932 .put_port = udp_lib_unhash,
2933#ifdef CONFIG_BPF_SYSCALL
2934 .psock_update_sk_prot = udp_bpf_update_proto,
2935#endif
2936 .memory_allocated = &udp_memory_allocated,
2937 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2938
2939 .sysctl_mem = sysctl_udp_mem,
2940 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2941 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2942 .obj_size = sizeof(struct udp_sock),
2943 .h.udp_table = NULL,
2944 .diag_destroy = udp_abort,
2945};
2946EXPORT_SYMBOL(udp_prot);
2947
2948/* ------------------------------------------------------------------------ */
2949#ifdef CONFIG_PROC_FS
2950
2951static unsigned short seq_file_family(const struct seq_file *seq);
2952static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2953{
2954 unsigned short family = seq_file_family(seq);
2955
2956 /* AF_UNSPEC is used as a match all */
2957 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2958 net_eq(sock_net(sk), seq_file_net(seq)));
2959}
2960
2961#ifdef CONFIG_BPF_SYSCALL
2962static const struct seq_operations bpf_iter_udp_seq_ops;
2963#endif
2964static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2965 struct net *net)
2966{
2967 const struct udp_seq_afinfo *afinfo;
2968
2969#ifdef CONFIG_BPF_SYSCALL
2970 if (seq->op == &bpf_iter_udp_seq_ops)
2971 return net->ipv4.udp_table;
2972#endif
2973
2974 afinfo = pde_data(file_inode(seq->file));
2975 return afinfo->udp_table ? : net->ipv4.udp_table;
2976}
2977
2978static struct sock *udp_get_first(struct seq_file *seq, int start)
2979{
2980 struct udp_iter_state *state = seq->private;
2981 struct net *net = seq_file_net(seq);
2982 struct udp_table *udptable;
2983 struct sock *sk;
2984
2985 udptable = udp_get_table_seq(seq, net);
2986
2987 for (state->bucket = start; state->bucket <= udptable->mask;
2988 ++state->bucket) {
2989 struct udp_hslot *hslot = &udptable->hash[state->bucket];
2990
2991 if (hlist_empty(&hslot->head))
2992 continue;
2993
2994 spin_lock_bh(&hslot->lock);
2995 sk_for_each(sk, &hslot->head) {
2996 if (seq_sk_match(seq, sk))
2997 goto found;
2998 }
2999 spin_unlock_bh(&hslot->lock);
3000 }
3001 sk = NULL;
3002found:
3003 return sk;
3004}
3005
3006static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3007{
3008 struct udp_iter_state *state = seq->private;
3009 struct net *net = seq_file_net(seq);
3010 struct udp_table *udptable;
3011
3012 do {
3013 sk = sk_next(sk);
3014 } while (sk && !seq_sk_match(seq, sk));
3015
3016 if (!sk) {
3017 udptable = udp_get_table_seq(seq, net);
3018
3019 if (state->bucket <= udptable->mask)
3020 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3021
3022 return udp_get_first(seq, state->bucket + 1);
3023 }
3024 return sk;
3025}
3026
3027static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3028{
3029 struct sock *sk = udp_get_first(seq, 0);
3030
3031 if (sk)
3032 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3033 --pos;
3034 return pos ? NULL : sk;
3035}
3036
3037void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3038{
3039 struct udp_iter_state *state = seq->private;
3040 state->bucket = MAX_UDP_PORTS;
3041
3042 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3043}
3044EXPORT_SYMBOL(udp_seq_start);
3045
3046void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3047{
3048 struct sock *sk;
3049
3050 if (v == SEQ_START_TOKEN)
3051 sk = udp_get_idx(seq, 0);
3052 else
3053 sk = udp_get_next(seq, v);
3054
3055 ++*pos;
3056 return sk;
3057}
3058EXPORT_SYMBOL(udp_seq_next);
3059
3060void udp_seq_stop(struct seq_file *seq, void *v)
3061{
3062 struct udp_iter_state *state = seq->private;
3063 struct udp_table *udptable;
3064
3065 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3066
3067 if (state->bucket <= udptable->mask)
3068 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3069}
3070EXPORT_SYMBOL(udp_seq_stop);
3071
3072/* ------------------------------------------------------------------------ */
3073static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3074 int bucket)
3075{
3076 struct inet_sock *inet = inet_sk(sp);
3077 __be32 dest = inet->inet_daddr;
3078 __be32 src = inet->inet_rcv_saddr;
3079 __u16 destp = ntohs(inet->inet_dport);
3080 __u16 srcp = ntohs(inet->inet_sport);
3081
3082 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3083 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3084 bucket, src, srcp, dest, destp, sp->sk_state,
3085 sk_wmem_alloc_get(sp),
3086 udp_rqueue_get(sp),
3087 0, 0L, 0,
3088 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3089 0, sock_i_ino(sp),
3090 refcount_read(&sp->sk_refcnt), sp,
3091 atomic_read(&sp->sk_drops));
3092}
3093
3094int udp4_seq_show(struct seq_file *seq, void *v)
3095{
3096 seq_setwidth(seq, 127);
3097 if (v == SEQ_START_TOKEN)
3098 seq_puts(seq, " sl local_address rem_address st tx_queue "
3099 "rx_queue tr tm->when retrnsmt uid timeout "
3100 "inode ref pointer drops");
3101 else {
3102 struct udp_iter_state *state = seq->private;
3103
3104 udp4_format_sock(v, seq, state->bucket);
3105 }
3106 seq_pad(seq, '\n');
3107 return 0;
3108}
3109
3110#ifdef CONFIG_BPF_SYSCALL
3111struct bpf_iter__udp {
3112 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3113 __bpf_md_ptr(struct udp_sock *, udp_sk);
3114 uid_t uid __aligned(8);
3115 int bucket __aligned(8);
3116};
3117
3118struct bpf_udp_iter_state {
3119 struct udp_iter_state state;
3120 unsigned int cur_sk;
3121 unsigned int end_sk;
3122 unsigned int max_sk;
3123 int offset;
3124 struct sock **batch;
3125 bool st_bucket_done;
3126};
3127
3128static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3129 unsigned int new_batch_sz);
3130static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3131{
3132 struct bpf_udp_iter_state *iter = seq->private;
3133 struct udp_iter_state *state = &iter->state;
3134 struct net *net = seq_file_net(seq);
3135 int resume_bucket, resume_offset;
3136 struct udp_table *udptable;
3137 unsigned int batch_sks = 0;
3138 bool resized = false;
3139 struct sock *sk;
3140
3141 resume_bucket = state->bucket;
3142 resume_offset = iter->offset;
3143
3144 /* The current batch is done, so advance the bucket. */
3145 if (iter->st_bucket_done)
3146 state->bucket++;
3147
3148 udptable = udp_get_table_seq(seq, net);
3149
3150again:
3151 /* New batch for the next bucket.
3152 * Iterate over the hash table to find a bucket with sockets matching
3153 * the iterator attributes, and return the first matching socket from
3154 * the bucket. The remaining matched sockets from the bucket are batched
3155 * before releasing the bucket lock. This allows BPF programs that are
3156 * called in seq_show to acquire the bucket lock if needed.
3157 */
3158 iter->cur_sk = 0;
3159 iter->end_sk = 0;
3160 iter->st_bucket_done = false;
3161 batch_sks = 0;
3162
3163 for (; state->bucket <= udptable->mask; state->bucket++) {
3164 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3165
3166 if (hlist_empty(&hslot2->head))
3167 continue;
3168
3169 iter->offset = 0;
3170 spin_lock_bh(&hslot2->lock);
3171 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3172 if (seq_sk_match(seq, sk)) {
3173 /* Resume from the last iterated socket at the
3174 * offset in the bucket before iterator was stopped.
3175 */
3176 if (state->bucket == resume_bucket &&
3177 iter->offset < resume_offset) {
3178 ++iter->offset;
3179 continue;
3180 }
3181 if (iter->end_sk < iter->max_sk) {
3182 sock_hold(sk);
3183 iter->batch[iter->end_sk++] = sk;
3184 }
3185 batch_sks++;
3186 }
3187 }
3188 spin_unlock_bh(&hslot2->lock);
3189
3190 if (iter->end_sk)
3191 break;
3192 }
3193
3194 /* All done: no batch made. */
3195 if (!iter->end_sk)
3196 return NULL;
3197
3198 if (iter->end_sk == batch_sks) {
3199 /* Batching is done for the current bucket; return the first
3200 * socket to be iterated from the batch.
3201 */
3202 iter->st_bucket_done = true;
3203 goto done;
3204 }
3205 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3206 resized = true;
3207 /* After allocating a larger batch, retry one more time to grab
3208 * the whole bucket.
3209 */
3210 goto again;
3211 }
3212done:
3213 return iter->batch[0];
3214}
3215
3216static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3217{
3218 struct bpf_udp_iter_state *iter = seq->private;
3219 struct sock *sk;
3220
3221 /* Whenever seq_next() is called, the iter->cur_sk is
3222 * done with seq_show(), so unref the iter->cur_sk.
3223 */
3224 if (iter->cur_sk < iter->end_sk) {
3225 sock_put(iter->batch[iter->cur_sk++]);
3226 ++iter->offset;
3227 }
3228
3229 /* After updating iter->cur_sk, check if there are more sockets
3230 * available in the current bucket batch.
3231 */
3232 if (iter->cur_sk < iter->end_sk)
3233 sk = iter->batch[iter->cur_sk];
3234 else
3235 /* Prepare a new batch. */
3236 sk = bpf_iter_udp_batch(seq);
3237
3238 ++*pos;
3239 return sk;
3240}
3241
3242static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3243{
3244 /* bpf iter does not support lseek, so it always
3245 * continue from where it was stop()-ped.
3246 */
3247 if (*pos)
3248 return bpf_iter_udp_batch(seq);
3249
3250 return SEQ_START_TOKEN;
3251}
3252
3253static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3254 struct udp_sock *udp_sk, uid_t uid, int bucket)
3255{
3256 struct bpf_iter__udp ctx;
3257
3258 meta->seq_num--; /* skip SEQ_START_TOKEN */
3259 ctx.meta = meta;
3260 ctx.udp_sk = udp_sk;
3261 ctx.uid = uid;
3262 ctx.bucket = bucket;
3263 return bpf_iter_run_prog(prog, &ctx);
3264}
3265
3266static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3267{
3268 struct udp_iter_state *state = seq->private;
3269 struct bpf_iter_meta meta;
3270 struct bpf_prog *prog;
3271 struct sock *sk = v;
3272 uid_t uid;
3273 int ret;
3274
3275 if (v == SEQ_START_TOKEN)
3276 return 0;
3277
3278 lock_sock(sk);
3279
3280 if (unlikely(sk_unhashed(sk))) {
3281 ret = SEQ_SKIP;
3282 goto unlock;
3283 }
3284
3285 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3286 meta.seq = seq;
3287 prog = bpf_iter_get_info(&meta, false);
3288 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3289
3290unlock:
3291 release_sock(sk);
3292 return ret;
3293}
3294
3295static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3296{
3297 while (iter->cur_sk < iter->end_sk)
3298 sock_put(iter->batch[iter->cur_sk++]);
3299}
3300
3301static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3302{
3303 struct bpf_udp_iter_state *iter = seq->private;
3304 struct bpf_iter_meta meta;
3305 struct bpf_prog *prog;
3306
3307 if (!v) {
3308 meta.seq = seq;
3309 prog = bpf_iter_get_info(&meta, true);
3310 if (prog)
3311 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3312 }
3313
3314 if (iter->cur_sk < iter->end_sk) {
3315 bpf_iter_udp_put_batch(iter);
3316 iter->st_bucket_done = false;
3317 }
3318}
3319
3320static const struct seq_operations bpf_iter_udp_seq_ops = {
3321 .start = bpf_iter_udp_seq_start,
3322 .next = bpf_iter_udp_seq_next,
3323 .stop = bpf_iter_udp_seq_stop,
3324 .show = bpf_iter_udp_seq_show,
3325};
3326#endif
3327
3328static unsigned short seq_file_family(const struct seq_file *seq)
3329{
3330 const struct udp_seq_afinfo *afinfo;
3331
3332#ifdef CONFIG_BPF_SYSCALL
3333 /* BPF iterator: bpf programs to filter sockets. */
3334 if (seq->op == &bpf_iter_udp_seq_ops)
3335 return AF_UNSPEC;
3336#endif
3337
3338 /* Proc fs iterator */
3339 afinfo = pde_data(file_inode(seq->file));
3340 return afinfo->family;
3341}
3342
3343const struct seq_operations udp_seq_ops = {
3344 .start = udp_seq_start,
3345 .next = udp_seq_next,
3346 .stop = udp_seq_stop,
3347 .show = udp4_seq_show,
3348};
3349EXPORT_SYMBOL(udp_seq_ops);
3350
3351static struct udp_seq_afinfo udp4_seq_afinfo = {
3352 .family = AF_INET,
3353 .udp_table = NULL,
3354};
3355
3356static int __net_init udp4_proc_init_net(struct net *net)
3357{
3358 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3359 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3360 return -ENOMEM;
3361 return 0;
3362}
3363
3364static void __net_exit udp4_proc_exit_net(struct net *net)
3365{
3366 remove_proc_entry("udp", net->proc_net);
3367}
3368
3369static struct pernet_operations udp4_net_ops = {
3370 .init = udp4_proc_init_net,
3371 .exit = udp4_proc_exit_net,
3372};
3373
3374int __init udp4_proc_init(void)
3375{
3376 return register_pernet_subsys(&udp4_net_ops);
3377}
3378
3379void udp4_proc_exit(void)
3380{
3381 unregister_pernet_subsys(&udp4_net_ops);
3382}
3383#endif /* CONFIG_PROC_FS */
3384
3385static __initdata unsigned long uhash_entries;
3386static int __init set_uhash_entries(char *str)
3387{
3388 ssize_t ret;
3389
3390 if (!str)
3391 return 0;
3392
3393 ret = kstrtoul(str, 0, &uhash_entries);
3394 if (ret)
3395 return 0;
3396
3397 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3398 uhash_entries = UDP_HTABLE_SIZE_MIN;
3399 return 1;
3400}
3401__setup("uhash_entries=", set_uhash_entries);
3402
3403void __init udp_table_init(struct udp_table *table, const char *name)
3404{
3405 unsigned int i;
3406
3407 table->hash = alloc_large_system_hash(name,
3408 2 * sizeof(struct udp_hslot),
3409 uhash_entries,
3410 21, /* one slot per 2 MB */
3411 0,
3412 &table->log,
3413 &table->mask,
3414 UDP_HTABLE_SIZE_MIN,
3415 UDP_HTABLE_SIZE_MAX);
3416
3417 table->hash2 = table->hash + (table->mask + 1);
3418 for (i = 0; i <= table->mask; i++) {
3419 INIT_HLIST_HEAD(&table->hash[i].head);
3420 table->hash[i].count = 0;
3421 spin_lock_init(&table->hash[i].lock);
3422 }
3423 for (i = 0; i <= table->mask; i++) {
3424 INIT_HLIST_HEAD(&table->hash2[i].head);
3425 table->hash2[i].count = 0;
3426 spin_lock_init(&table->hash2[i].lock);
3427 }
3428}
3429
3430u32 udp_flow_hashrnd(void)
3431{
3432 static u32 hashrnd __read_mostly;
3433
3434 net_get_random_once(&hashrnd, sizeof(hashrnd));
3435
3436 return hashrnd;
3437}
3438EXPORT_SYMBOL(udp_flow_hashrnd);
3439
3440static void __net_init udp_sysctl_init(struct net *net)
3441{
3442 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3443 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3444
3445#ifdef CONFIG_NET_L3_MASTER_DEV
3446 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3447#endif
3448}
3449
3450static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3451{
3452 struct udp_table *udptable;
3453 int i;
3454
3455 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3456 if (!udptable)
3457 goto out;
3458
3459 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3460 GFP_KERNEL_ACCOUNT);
3461 if (!udptable->hash)
3462 goto free_table;
3463
3464 udptable->hash2 = udptable->hash + hash_entries;
3465 udptable->mask = hash_entries - 1;
3466 udptable->log = ilog2(hash_entries);
3467
3468 for (i = 0; i < hash_entries; i++) {
3469 INIT_HLIST_HEAD(&udptable->hash[i].head);
3470 udptable->hash[i].count = 0;
3471 spin_lock_init(&udptable->hash[i].lock);
3472
3473 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3474 udptable->hash2[i].count = 0;
3475 spin_lock_init(&udptable->hash2[i].lock);
3476 }
3477
3478 return udptable;
3479
3480free_table:
3481 kfree(udptable);
3482out:
3483 return NULL;
3484}
3485
3486static void __net_exit udp_pernet_table_free(struct net *net)
3487{
3488 struct udp_table *udptable = net->ipv4.udp_table;
3489
3490 if (udptable == &udp_table)
3491 return;
3492
3493 kvfree(udptable->hash);
3494 kfree(udptable);
3495}
3496
3497static void __net_init udp_set_table(struct net *net)
3498{
3499 struct udp_table *udptable;
3500 unsigned int hash_entries;
3501 struct net *old_net;
3502
3503 if (net_eq(net, &init_net))
3504 goto fallback;
3505
3506 old_net = current->nsproxy->net_ns;
3507 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3508 if (!hash_entries)
3509 goto fallback;
3510
3511 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3512 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3513 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3514 else
3515 hash_entries = roundup_pow_of_two(hash_entries);
3516
3517 udptable = udp_pernet_table_alloc(hash_entries);
3518 if (udptable) {
3519 net->ipv4.udp_table = udptable;
3520 } else {
3521 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3522 "for a netns, fallback to the global one\n",
3523 hash_entries);
3524fallback:
3525 net->ipv4.udp_table = &udp_table;
3526 }
3527}
3528
3529static int __net_init udp_pernet_init(struct net *net)
3530{
3531 udp_sysctl_init(net);
3532 udp_set_table(net);
3533
3534 return 0;
3535}
3536
3537static void __net_exit udp_pernet_exit(struct net *net)
3538{
3539 udp_pernet_table_free(net);
3540}
3541
3542static struct pernet_operations __net_initdata udp_sysctl_ops = {
3543 .init = udp_pernet_init,
3544 .exit = udp_pernet_exit,
3545};
3546
3547#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3548DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3549 struct udp_sock *udp_sk, uid_t uid, int bucket)
3550
3551static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3552 unsigned int new_batch_sz)
3553{
3554 struct sock **new_batch;
3555
3556 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3557 GFP_USER | __GFP_NOWARN);
3558 if (!new_batch)
3559 return -ENOMEM;
3560
3561 bpf_iter_udp_put_batch(iter);
3562 kvfree(iter->batch);
3563 iter->batch = new_batch;
3564 iter->max_sk = new_batch_sz;
3565
3566 return 0;
3567}
3568
3569#define INIT_BATCH_SZ 16
3570
3571static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3572{
3573 struct bpf_udp_iter_state *iter = priv_data;
3574 int ret;
3575
3576 ret = bpf_iter_init_seq_net(priv_data, aux);
3577 if (ret)
3578 return ret;
3579
3580 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3581 if (ret)
3582 bpf_iter_fini_seq_net(priv_data);
3583
3584 return ret;
3585}
3586
3587static void bpf_iter_fini_udp(void *priv_data)
3588{
3589 struct bpf_udp_iter_state *iter = priv_data;
3590
3591 bpf_iter_fini_seq_net(priv_data);
3592 kvfree(iter->batch);
3593}
3594
3595static const struct bpf_iter_seq_info udp_seq_info = {
3596 .seq_ops = &bpf_iter_udp_seq_ops,
3597 .init_seq_private = bpf_iter_init_udp,
3598 .fini_seq_private = bpf_iter_fini_udp,
3599 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3600};
3601
3602static struct bpf_iter_reg udp_reg_info = {
3603 .target = "udp",
3604 .ctx_arg_info_size = 1,
3605 .ctx_arg_info = {
3606 { offsetof(struct bpf_iter__udp, udp_sk),
3607 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3608 },
3609 .seq_info = &udp_seq_info,
3610};
3611
3612static void __init bpf_iter_register(void)
3613{
3614 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3615 if (bpf_iter_reg_target(&udp_reg_info))
3616 pr_warn("Warning: could not register bpf iterator udp\n");
3617}
3618#endif
3619
3620void __init udp_init(void)
3621{
3622 unsigned long limit;
3623 unsigned int i;
3624
3625 udp_table_init(&udp_table, "UDP");
3626 limit = nr_free_buffer_pages() / 8;
3627 limit = max(limit, 128UL);
3628 sysctl_udp_mem[0] = limit / 4 * 3;
3629 sysctl_udp_mem[1] = limit;
3630 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3631
3632 /* 16 spinlocks per cpu */
3633 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3634 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3635 GFP_KERNEL);
3636 if (!udp_busylocks)
3637 panic("UDP: failed to alloc udp_busylocks\n");
3638 for (i = 0; i < (1U << udp_busylocks_log); i++)
3639 spin_lock_init(udp_busylocks + i);
3640
3641 if (register_pernet_subsys(&udp_sysctl_ops))
3642 panic("UDP: failed to init sysctl parameters.\n");
3643
3644#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3645 bpf_iter_register();
3646#endif
3647}