?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
printf: Remove unused 'bprintf'
[drm/drm-misc.git] / net / ipv4 / udp.c
blob6a01905d379fdd5c1815c3a71694071f6e5e06d3
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
123 struct udp_table udp_table __read_mostly;
124 EXPORT_SYMBOL(udp_table);
125
126 long sysctl_udp_mem[3] __read_mostly;
127 EXPORT_SYMBOL(sysctl_udp_mem);
128
129 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
130 EXPORT_SYMBOL(udp_memory_allocated);
131 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
132 EXPORT_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
137 static 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
142 static 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 */
178 static 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
209 static 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 */
240 int 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 }
313 scan_primary_hash:
314 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
315 goto fail_unlock;
316 }
317 found:
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;
350 fail_unlock:
351 spin_unlock_bh(&hslot->lock);
352 fail:
353 return error;
354 }
355 EXPORT_SYMBOL(udp_lib_get_port);
356
357 int 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
369 static 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
413 u32 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 }
421 EXPORT_SYMBOL(udp_ehashfn);
422
423 /* called with rcu_read_lock() */
424 static struct sock *udp4_lib_lookup2(const struct net *net,
425 __be32 saddr, __be16 sport,
426 __be32 daddr, unsigned int hnum,
427 int dif, int sdif,
428 struct udp_hslot *hslot2,
429 struct sk_buff *skb)
430 {
431 struct sock *sk, *result;
432 int score, badness;
433 bool need_rescore;
434
435 result = NULL;
436 badness = 0;
437 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
438 need_rescore = false;
439 rescore:
440 score = compute_score(need_rescore ? result : sk, net, saddr,
441 sport, daddr, hnum, dif, sdif);
442 if (score > badness) {
443 badness = score;
444
445 if (need_rescore)
446 continue;
447
448 if (sk->sk_state == TCP_ESTABLISHED) {
449 result = sk;
450 continue;
451 }
452
453 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
454 saddr, sport, daddr, hnum, udp_ehashfn);
455 if (!result) {
456 result = sk;
457 continue;
458 }
459
460 /* Fall back to scoring if group has connections */
461 if (!reuseport_has_conns(sk))
462 return result;
463
464 /* Reuseport logic returned an error, keep original score. */
465 if (IS_ERR(result))
466 continue;
467
468 /* compute_score is too long of a function to be
469 * inlined, and calling it again here yields
470 * measureable overhead for some
471 * workloads. Work around it by jumping
472 * backwards to rescore 'result'.
473 */
474 need_rescore = true;
475 goto rescore;
476 }
477 }
478 return result;
479 }
480
481 #if IS_ENABLED(CONFIG_BASE_SMALL)
482 static struct sock *udp4_lib_lookup4(const struct net *net,
483 __be32 saddr, __be16 sport,
484 __be32 daddr, unsigned int hnum,
485 int dif, int sdif,
486 struct udp_table *udptable)
487 {
488 return NULL;
489 }
490
491 static void udp_rehash4(struct udp_table *udptable, struct sock *sk,
492 u16 newhash4)
493 {
494 }
495
496 static void udp_unhash4(struct udp_table *udptable, struct sock *sk)
497 {
498 }
499 #else /* !CONFIG_BASE_SMALL */
500 static struct sock *udp4_lib_lookup4(const struct net *net,
501 __be32 saddr, __be16 sport,
502 __be32 daddr, unsigned int hnum,
503 int dif, int sdif,
504 struct udp_table *udptable)
505 {
506 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
507 const struct hlist_nulls_node *node;
508 struct udp_hslot *hslot4;
509 unsigned int hash4, slot;
510 struct udp_sock *up;
511 struct sock *sk;
512
513 hash4 = udp_ehashfn(net, daddr, hnum, saddr, sport);
514 slot = hash4 & udptable->mask;
515 hslot4 = &udptable->hash4[slot];
516 INET_ADDR_COOKIE(acookie, saddr, daddr);
517
518 begin:
519 /* SLAB_TYPESAFE_BY_RCU not used, so we don't need to touch sk_refcnt */
520 udp_lrpa_for_each_entry_rcu(up, node, &hslot4->nulls_head) {
521 sk = (struct sock *)up;
522 if (inet_match(net, sk, acookie, ports, dif, sdif))
523 return sk;
524 }
525
526 /* if the nulls value we got at the end of this lookup is not the
527 * expected one, we must restart lookup. We probably met an item that
528 * was moved to another chain due to rehash.
529 */
530 if (get_nulls_value(node) != slot)
531 goto begin;
532
533 return NULL;
534 }
535
536 /* In hash4, rehash can happen in connect(), where hash4_cnt keeps unchanged. */
537 static void udp_rehash4(struct udp_table *udptable, struct sock *sk,
538 u16 newhash4)
539 {
540 struct udp_hslot *hslot4, *nhslot4;
541
542 hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash);
543 nhslot4 = udp_hashslot4(udptable, newhash4);
544 udp_sk(sk)->udp_lrpa_hash = newhash4;
545
546 if (hslot4 != nhslot4) {
547 spin_lock_bh(&hslot4->lock);
548 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node);
549 hslot4->count--;
550 spin_unlock_bh(&hslot4->lock);
551
552 spin_lock_bh(&nhslot4->lock);
553 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node,
554 &nhslot4->nulls_head);
555 nhslot4->count++;
556 spin_unlock_bh(&nhslot4->lock);
557 }
558 }
559
560 static void udp_unhash4(struct udp_table *udptable, struct sock *sk)
561 {
562 struct udp_hslot *hslot2, *hslot4;
563
564 if (udp_hashed4(sk)) {
565 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
566 hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash);
567
568 spin_lock(&hslot4->lock);
569 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node);
570 hslot4->count--;
571 spin_unlock(&hslot4->lock);
572
573 spin_lock(&hslot2->lock);
574 udp_hash4_dec(hslot2);
575 spin_unlock(&hslot2->lock);
576 }
577 }
578
579 void udp_lib_hash4(struct sock *sk, u16 hash)
580 {
581 struct udp_hslot *hslot, *hslot2, *hslot4;
582 struct net *net = sock_net(sk);
583 struct udp_table *udptable;
584
585 /* Connected udp socket can re-connect to another remote address,
586 * so rehash4 is needed.
587 */
588 udptable = net->ipv4.udp_table;
589 if (udp_hashed4(sk)) {
590 udp_rehash4(udptable, sk, hash);
591 return;
592 }
593
594 hslot = udp_hashslot(udptable, net, udp_sk(sk)->udp_port_hash);
595 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
596 hslot4 = udp_hashslot4(udptable, hash);
597 udp_sk(sk)->udp_lrpa_hash = hash;
598
599 spin_lock_bh(&hslot->lock);
600 if (rcu_access_pointer(sk->sk_reuseport_cb))
601 reuseport_detach_sock(sk);
602
603 spin_lock(&hslot4->lock);
604 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node,
605 &hslot4->nulls_head);
606 hslot4->count++;
607 spin_unlock(&hslot4->lock);
608
609 spin_lock(&hslot2->lock);
610 udp_hash4_inc(hslot2);
611 spin_unlock(&hslot2->lock);
612
613 spin_unlock_bh(&hslot->lock);
614 }
615 EXPORT_SYMBOL(udp_lib_hash4);
616
617 /* call with sock lock */
618 void udp4_hash4(struct sock *sk)
619 {
620 struct net *net = sock_net(sk);
621 unsigned int hash;
622
623 if (sk_unhashed(sk) || sk->sk_rcv_saddr == htonl(INADDR_ANY))
624 return;
625
626 hash = udp_ehashfn(net, sk->sk_rcv_saddr, sk->sk_num,
627 sk->sk_daddr, sk->sk_dport);
628
629 udp_lib_hash4(sk, hash);
630 }
631 EXPORT_SYMBOL(udp4_hash4);
632 #endif /* CONFIG_BASE_SMALL */
633
634 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
635 * harder than this. -DaveM
636 */
637 struct sock *__udp4_lib_lookup(const struct net *net, __be32 saddr,
638 __be16 sport, __be32 daddr, __be16 dport, int dif,
639 int sdif, struct udp_table *udptable, struct sk_buff *skb)
640 {
641 unsigned short hnum = ntohs(dport);
642 struct udp_hslot *hslot2;
643 struct sock *result, *sk;
644 unsigned int hash2;
645
646 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
647 hslot2 = udp_hashslot2(udptable, hash2);
648
649 if (udp_has_hash4(hslot2)) {
650 result = udp4_lib_lookup4(net, saddr, sport, daddr, hnum,
651 dif, sdif, udptable);
652 if (result) /* udp4_lib_lookup4 return sk or NULL */
653 return result;
654 }
655
656 /* Lookup connected or non-wildcard socket */
657 result = udp4_lib_lookup2(net, saddr, sport,
658 daddr, hnum, dif, sdif,
659 hslot2, skb);
660 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
661 goto done;
662
663 /* Lookup redirect from BPF */
664 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
665 udptable == net->ipv4.udp_table) {
666 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
667 saddr, sport, daddr, hnum, dif,
668 udp_ehashfn);
669 if (sk) {
670 result = sk;
671 goto done;
672 }
673 }
674
675 /* Got non-wildcard socket or error on first lookup */
676 if (result)
677 goto done;
678
679 /* Lookup wildcard sockets */
680 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
681 hslot2 = udp_hashslot2(udptable, hash2);
682
683 result = udp4_lib_lookup2(net, saddr, sport,
684 htonl(INADDR_ANY), hnum, dif, sdif,
685 hslot2, skb);
686 done:
687 if (IS_ERR(result))
688 return NULL;
689 return result;
690 }
691 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
692
693 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
694 __be16 sport, __be16 dport,
695 struct udp_table *udptable)
696 {
697 const struct iphdr *iph = ip_hdr(skb);
698
699 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
700 iph->daddr, dport, inet_iif(skb),
701 inet_sdif(skb), udptable, skb);
702 }
703
704 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
705 __be16 sport, __be16 dport)
706 {
707 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
708 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
709 struct net *net = dev_net(skb->dev);
710 int iif, sdif;
711
712 inet_get_iif_sdif(skb, &iif, &sdif);
713
714 return __udp4_lib_lookup(net, iph->saddr, sport,
715 iph->daddr, dport, iif,
716 sdif, net->ipv4.udp_table, NULL);
717 }
718
719 /* Must be called under rcu_read_lock().
720 * Does increment socket refcount.
721 */
722 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
723 struct sock *udp4_lib_lookup(const struct net *net, __be32 saddr, __be16 sport,
724 __be32 daddr, __be16 dport, int dif)
725 {
726 struct sock *sk;
727
728 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
729 dif, 0, net->ipv4.udp_table, NULL);
730 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
731 sk = NULL;
732 return sk;
733 }
734 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
735 #endif
736
737 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
738 __be16 loc_port, __be32 loc_addr,
739 __be16 rmt_port, __be32 rmt_addr,
740 int dif, int sdif, unsigned short hnum)
741 {
742 const struct inet_sock *inet = inet_sk(sk);
743
744 if (!net_eq(sock_net(sk), net) ||
745 udp_sk(sk)->udp_port_hash != hnum ||
746 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
747 (inet->inet_dport != rmt_port && inet->inet_dport) ||
748 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
749 ipv6_only_sock(sk) ||
750 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
751 return false;
752 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
753 return false;
754 return true;
755 }
756
757 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
758 EXPORT_SYMBOL(udp_encap_needed_key);
759
760 #if IS_ENABLED(CONFIG_IPV6)
761 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
762 EXPORT_SYMBOL(udpv6_encap_needed_key);
763 #endif
764
765 void udp_encap_enable(void)
766 {
767 static_branch_inc(&udp_encap_needed_key);
768 }
769 EXPORT_SYMBOL(udp_encap_enable);
770
771 void udp_encap_disable(void)
772 {
773 static_branch_dec(&udp_encap_needed_key);
774 }
775 EXPORT_SYMBOL(udp_encap_disable);
776
777 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
778 * through error handlers in encapsulations looking for a match.
779 */
780 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
781 {
782 int i;
783
784 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
785 int (*handler)(struct sk_buff *skb, u32 info);
786 const struct ip_tunnel_encap_ops *encap;
787
788 encap = rcu_dereference(iptun_encaps[i]);
789 if (!encap)
790 continue;
791 handler = encap->err_handler;
792 if (handler && !handler(skb, info))
793 return 0;
794 }
795
796 return -ENOENT;
797 }
798
799 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
800 * reversing source and destination port: this will match tunnels that force the
801 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
802 * lwtunnels might actually break this assumption by being configured with
803 * different destination ports on endpoints, in this case we won't be able to
804 * trace ICMP messages back to them.
805 *
806 * If this doesn't match any socket, probe tunnels with arbitrary destination
807 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
808 * we've sent packets to won't necessarily match the local destination port.
809 *
810 * Then ask the tunnel implementation to match the error against a valid
811 * association.
812 *
813 * Return an error if we can't find a match, the socket if we need further
814 * processing, zero otherwise.
815 */
816 static struct sock *__udp4_lib_err_encap(struct net *net,
817 const struct iphdr *iph,
818 struct udphdr *uh,
819 struct udp_table *udptable,
820 struct sock *sk,
821 struct sk_buff *skb, u32 info)
822 {
823 int (*lookup)(struct sock *sk, struct sk_buff *skb);
824 int network_offset, transport_offset;
825 struct udp_sock *up;
826
827 network_offset = skb_network_offset(skb);
828 transport_offset = skb_transport_offset(skb);
829
830 /* Network header needs to point to the outer IPv4 header inside ICMP */
831 skb_reset_network_header(skb);
832
833 /* Transport header needs to point to the UDP header */
834 skb_set_transport_header(skb, iph->ihl << 2);
835
836 if (sk) {
837 up = udp_sk(sk);
838
839 lookup = READ_ONCE(up->encap_err_lookup);
840 if (lookup && lookup(sk, skb))
841 sk = NULL;
842
843 goto out;
844 }
845
846 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
847 iph->saddr, uh->dest, skb->dev->ifindex, 0,
848 udptable, NULL);
849 if (sk) {
850 up = udp_sk(sk);
851
852 lookup = READ_ONCE(up->encap_err_lookup);
853 if (!lookup || lookup(sk, skb))
854 sk = NULL;
855 }
856
857 out:
858 if (!sk)
859 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
860
861 skb_set_transport_header(skb, transport_offset);
862 skb_set_network_header(skb, network_offset);
863
864 return sk;
865 }
866
867 /*
868 * This routine is called by the ICMP module when it gets some
869 * sort of error condition. If err < 0 then the socket should
870 * be closed and the error returned to the user. If err > 0
871 * it's just the icmp type << 8 | icmp code.
872 * Header points to the ip header of the error packet. We move
873 * on past this. Then (as it used to claim before adjustment)
874 * header points to the first 8 bytes of the udp header. We need
875 * to find the appropriate port.
876 */
877
878 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
879 {
880 struct inet_sock *inet;
881 const struct iphdr *iph = (const struct iphdr *)skb->data;
882 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
883 const int type = icmp_hdr(skb)->type;
884 const int code = icmp_hdr(skb)->code;
885 bool tunnel = false;
886 struct sock *sk;
887 int harderr;
888 int err;
889 struct net *net = dev_net(skb->dev);
890
891 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
892 iph->saddr, uh->source, skb->dev->ifindex,
893 inet_sdif(skb), udptable, NULL);
894
895 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
896 /* No socket for error: try tunnels before discarding */
897 if (static_branch_unlikely(&udp_encap_needed_key)) {
898 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
899 info);
900 if (!sk)
901 return 0;
902 } else
903 sk = ERR_PTR(-ENOENT);
904
905 if (IS_ERR(sk)) {
906 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
907 return PTR_ERR(sk);
908 }
909
910 tunnel = true;
911 }
912
913 err = 0;
914 harderr = 0;
915 inet = inet_sk(sk);
916
917 switch (type) {
918 default:
919 case ICMP_TIME_EXCEEDED:
920 err = EHOSTUNREACH;
921 break;
922 case ICMP_SOURCE_QUENCH:
923 goto out;
924 case ICMP_PARAMETERPROB:
925 err = EPROTO;
926 harderr = 1;
927 break;
928 case ICMP_DEST_UNREACH:
929 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
930 ipv4_sk_update_pmtu(skb, sk, info);
931 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
932 err = EMSGSIZE;
933 harderr = 1;
934 break;
935 }
936 goto out;
937 }
938 err = EHOSTUNREACH;
939 if (code <= NR_ICMP_UNREACH) {
940 harderr = icmp_err_convert[code].fatal;
941 err = icmp_err_convert[code].errno;
942 }
943 break;
944 case ICMP_REDIRECT:
945 ipv4_sk_redirect(skb, sk);
946 goto out;
947 }
948
949 /*
950 * RFC1122: OK. Passes ICMP errors back to application, as per
951 * 4.1.3.3.
952 */
953 if (tunnel) {
954 /* ...not for tunnels though: we don't have a sending socket */
955 if (udp_sk(sk)->encap_err_rcv)
956 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
957 (u8 *)(uh+1));
958 goto out;
959 }
960 if (!inet_test_bit(RECVERR, sk)) {
961 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
962 goto out;
963 } else
964 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
965
966 sk->sk_err = err;
967 sk_error_report(sk);
968 out:
969 return 0;
970 }
971
972 int udp_err(struct sk_buff *skb, u32 info)
973 {
974 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
975 }
976
977 /*
978 * Throw away all pending data and cancel the corking. Socket is locked.
979 */
980 void udp_flush_pending_frames(struct sock *sk)
981 {
982 struct udp_sock *up = udp_sk(sk);
983
984 if (up->pending) {
985 up->len = 0;
986 WRITE_ONCE(up->pending, 0);
987 ip_flush_pending_frames(sk);
988 }
989 }
990 EXPORT_SYMBOL(udp_flush_pending_frames);
991
992 /**
993 * udp4_hwcsum - handle outgoing HW checksumming
994 * @skb: sk_buff containing the filled-in UDP header
995 * (checksum field must be zeroed out)
996 * @src: source IP address
997 * @dst: destination IP address
998 */
999 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
1000 {
1001 struct udphdr *uh = udp_hdr(skb);
1002 int offset = skb_transport_offset(skb);
1003 int len = skb->len - offset;
1004 int hlen = len;
1005 __wsum csum = 0;
1006
1007 if (!skb_has_frag_list(skb)) {
1008 /*
1009 * Only one fragment on the socket.
1010 */
1011 skb->csum_start = skb_transport_header(skb) - skb->head;
1012 skb->csum_offset = offsetof(struct udphdr, check);
1013 uh->check = ~csum_tcpudp_magic(src, dst, len,
1014 IPPROTO_UDP, 0);
1015 } else {
1016 struct sk_buff *frags;
1017
1018 /*
1019 * HW-checksum won't work as there are two or more
1020 * fragments on the socket so that all csums of sk_buffs
1021 * should be together
1022 */
1023 skb_walk_frags(skb, frags) {
1024 csum = csum_add(csum, frags->csum);
1025 hlen -= frags->len;
1026 }
1027
1028 csum = skb_checksum(skb, offset, hlen, csum);
1029 skb->ip_summed = CHECKSUM_NONE;
1030
1031 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
1032 if (uh->check == 0)
1033 uh->check = CSUM_MANGLED_0;
1034 }
1035 }
1036 EXPORT_SYMBOL_GPL(udp4_hwcsum);
1037
1038 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
1039 * for the simple case like when setting the checksum for a UDP tunnel.
1040 */
1041 void udp_set_csum(bool nocheck, struct sk_buff *skb,
1042 __be32 saddr, __be32 daddr, int len)
1043 {
1044 struct udphdr *uh = udp_hdr(skb);
1045
1046 if (nocheck) {
1047 uh->check = 0;
1048 } else if (skb_is_gso(skb)) {
1049 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
1050 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
1051 uh->check = 0;
1052 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
1053 if (uh->check == 0)
1054 uh->check = CSUM_MANGLED_0;
1055 } else {
1056 skb->ip_summed = CHECKSUM_PARTIAL;
1057 skb->csum_start = skb_transport_header(skb) - skb->head;
1058 skb->csum_offset = offsetof(struct udphdr, check);
1059 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
1060 }
1061 }
1062 EXPORT_SYMBOL(udp_set_csum);
1063
1064 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
1065 struct inet_cork *cork)
1066 {
1067 struct sock *sk = skb->sk;
1068 struct inet_sock *inet = inet_sk(sk);
1069 struct udphdr *uh;
1070 int err;
1071 int is_udplite = IS_UDPLITE(sk);
1072 int offset = skb_transport_offset(skb);
1073 int len = skb->len - offset;
1074 int datalen = len - sizeof(*uh);
1075 __wsum csum = 0;
1076
1077 /*
1078 * Create a UDP header
1079 */
1080 uh = udp_hdr(skb);
1081 uh->source = inet->inet_sport;
1082 uh->dest = fl4->fl4_dport;
1083 uh->len = htons(len);
1084 uh->check = 0;
1085
1086 if (cork->gso_size) {
1087 const int hlen = skb_network_header_len(skb) +
1088 sizeof(struct udphdr);
1089
1090 if (hlen + cork->gso_size > cork->fragsize) {
1091 kfree_skb(skb);
1092 return -EINVAL;
1093 }
1094 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1095 kfree_skb(skb);
1096 return -EINVAL;
1097 }
1098 if (sk->sk_no_check_tx) {
1099 kfree_skb(skb);
1100 return -EINVAL;
1101 }
1102 if (is_udplite || dst_xfrm(skb_dst(skb))) {
1103 kfree_skb(skb);
1104 return -EIO;
1105 }
1106
1107 if (datalen > cork->gso_size) {
1108 skb_shinfo(skb)->gso_size = cork->gso_size;
1109 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1110 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1111 cork->gso_size);
1112
1113 /* Don't checksum the payload, skb will get segmented */
1114 goto csum_partial;
1115 }
1116 }
1117
1118 if (is_udplite) /* UDP-Lite */
1119 csum = udplite_csum(skb);
1120
1121 else if (sk->sk_no_check_tx) { /* UDP csum off */
1122
1123 skb->ip_summed = CHECKSUM_NONE;
1124 goto send;
1125
1126 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1127 csum_partial:
1128
1129 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
1130 goto send;
1131
1132 } else
1133 csum = udp_csum(skb);
1134
1135 /* add protocol-dependent pseudo-header */
1136 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
1137 sk->sk_protocol, csum);
1138 if (uh->check == 0)
1139 uh->check = CSUM_MANGLED_0;
1140
1141 send:
1142 err = ip_send_skb(sock_net(sk), skb);
1143 if (err) {
1144 if (err == -ENOBUFS &&
1145 !inet_test_bit(RECVERR, sk)) {
1146 UDP_INC_STATS(sock_net(sk),
1147 UDP_MIB_SNDBUFERRORS, is_udplite);
1148 err = 0;
1149 }
1150 } else
1151 UDP_INC_STATS(sock_net(sk),
1152 UDP_MIB_OUTDATAGRAMS, is_udplite);
1153 return err;
1154 }
1155
1156 /*
1157 * Push out all pending data as one UDP datagram. Socket is locked.
1158 */
1159 int udp_push_pending_frames(struct sock *sk)
1160 {
1161 struct udp_sock *up = udp_sk(sk);
1162 struct inet_sock *inet = inet_sk(sk);
1163 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1164 struct sk_buff *skb;
1165 int err = 0;
1166
1167 skb = ip_finish_skb(sk, fl4);
1168 if (!skb)
1169 goto out;
1170
1171 err = udp_send_skb(skb, fl4, &inet->cork.base);
1172
1173 out:
1174 up->len = 0;
1175 WRITE_ONCE(up->pending, 0);
1176 return err;
1177 }
1178 EXPORT_SYMBOL(udp_push_pending_frames);
1179
1180 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1181 {
1182 switch (cmsg->cmsg_type) {
1183 case UDP_SEGMENT:
1184 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1185 return -EINVAL;
1186 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1187 return 0;
1188 default:
1189 return -EINVAL;
1190 }
1191 }
1192
1193 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1194 {
1195 struct cmsghdr *cmsg;
1196 bool need_ip = false;
1197 int err;
1198
1199 for_each_cmsghdr(cmsg, msg) {
1200 if (!CMSG_OK(msg, cmsg))
1201 return -EINVAL;
1202
1203 if (cmsg->cmsg_level != SOL_UDP) {
1204 need_ip = true;
1205 continue;
1206 }
1207
1208 err = __udp_cmsg_send(cmsg, gso_size);
1209 if (err)
1210 return err;
1211 }
1212
1213 return need_ip;
1214 }
1215 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1216
1217 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1218 {
1219 struct inet_sock *inet = inet_sk(sk);
1220 struct udp_sock *up = udp_sk(sk);
1221 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1222 struct flowi4 fl4_stack;
1223 struct flowi4 *fl4;
1224 int ulen = len;
1225 struct ipcm_cookie ipc;
1226 struct rtable *rt = NULL;
1227 int free = 0;
1228 int connected = 0;
1229 __be32 daddr, faddr, saddr;
1230 u8 tos, scope;
1231 __be16 dport;
1232 int err, is_udplite = IS_UDPLITE(sk);
1233 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1234 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1235 struct sk_buff *skb;
1236 struct ip_options_data opt_copy;
1237 int uc_index;
1238
1239 if (len > 0xFFFF)
1240 return -EMSGSIZE;
1241
1242 /*
1243 * Check the flags.
1244 */
1245
1246 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1247 return -EOPNOTSUPP;
1248
1249 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1250
1251 fl4 = &inet->cork.fl.u.ip4;
1252 if (READ_ONCE(up->pending)) {
1253 /*
1254 * There are pending frames.
1255 * The socket lock must be held while it's corked.
1256 */
1257 lock_sock(sk);
1258 if (likely(up->pending)) {
1259 if (unlikely(up->pending != AF_INET)) {
1260 release_sock(sk);
1261 return -EINVAL;
1262 }
1263 goto do_append_data;
1264 }
1265 release_sock(sk);
1266 }
1267 ulen += sizeof(struct udphdr);
1268
1269 /*
1270 * Get and verify the address.
1271 */
1272 if (usin) {
1273 if (msg->msg_namelen < sizeof(*usin))
1274 return -EINVAL;
1275 if (usin->sin_family != AF_INET) {
1276 if (usin->sin_family != AF_UNSPEC)
1277 return -EAFNOSUPPORT;
1278 }
1279
1280 daddr = usin->sin_addr.s_addr;
1281 dport = usin->sin_port;
1282 if (dport == 0)
1283 return -EINVAL;
1284 } else {
1285 if (sk->sk_state != TCP_ESTABLISHED)
1286 return -EDESTADDRREQ;
1287 daddr = inet->inet_daddr;
1288 dport = inet->inet_dport;
1289 /* Open fast path for connected socket.
1290 Route will not be used, if at least one option is set.
1291 */
1292 connected = 1;
1293 }
1294
1295 ipcm_init_sk(&ipc, inet);
1296 ipc.gso_size = READ_ONCE(up->gso_size);
1297
1298 if (msg->msg_controllen) {
1299 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1300 if (err > 0) {
1301 err = ip_cmsg_send(sk, msg, &ipc,
1302 sk->sk_family == AF_INET6);
1303 connected = 0;
1304 }
1305 if (unlikely(err < 0)) {
1306 kfree(ipc.opt);
1307 return err;
1308 }
1309 if (ipc.opt)
1310 free = 1;
1311 }
1312 if (!ipc.opt) {
1313 struct ip_options_rcu *inet_opt;
1314
1315 rcu_read_lock();
1316 inet_opt = rcu_dereference(inet->inet_opt);
1317 if (inet_opt) {
1318 memcpy(&opt_copy, inet_opt,
1319 sizeof(*inet_opt) + inet_opt->opt.optlen);
1320 ipc.opt = &opt_copy.opt;
1321 }
1322 rcu_read_unlock();
1323 }
1324
1325 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1326 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1327 (struct sockaddr *)usin,
1328 &msg->msg_namelen,
1329 &ipc.addr);
1330 if (err)
1331 goto out_free;
1332 if (usin) {
1333 if (usin->sin_port == 0) {
1334 /* BPF program set invalid port. Reject it. */
1335 err = -EINVAL;
1336 goto out_free;
1337 }
1338 daddr = usin->sin_addr.s_addr;
1339 dport = usin->sin_port;
1340 }
1341 }
1342
1343 saddr = ipc.addr;
1344 ipc.addr = faddr = daddr;
1345
1346 if (ipc.opt && ipc.opt->opt.srr) {
1347 if (!daddr) {
1348 err = -EINVAL;
1349 goto out_free;
1350 }
1351 faddr = ipc.opt->opt.faddr;
1352 connected = 0;
1353 }
1354 tos = get_rttos(&ipc, inet);
1355 scope = ip_sendmsg_scope(inet, &ipc, msg);
1356 if (scope == RT_SCOPE_LINK)
1357 connected = 0;
1358
1359 uc_index = READ_ONCE(inet->uc_index);
1360 if (ipv4_is_multicast(daddr)) {
1361 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1362 ipc.oif = READ_ONCE(inet->mc_index);
1363 if (!saddr)
1364 saddr = READ_ONCE(inet->mc_addr);
1365 connected = 0;
1366 } else if (!ipc.oif) {
1367 ipc.oif = uc_index;
1368 } else if (ipv4_is_lbcast(daddr) && uc_index) {
1369 /* oif is set, packet is to local broadcast and
1370 * uc_index is set. oif is most likely set
1371 * by sk_bound_dev_if. If uc_index != oif check if the
1372 * oif is an L3 master and uc_index is an L3 slave.
1373 * If so, we want to allow the send using the uc_index.
1374 */
1375 if (ipc.oif != uc_index &&
1376 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1377 uc_index)) {
1378 ipc.oif = uc_index;
1379 }
1380 }
1381
1382 if (connected)
1383 rt = dst_rtable(sk_dst_check(sk, 0));
1384
1385 if (!rt) {
1386 struct net *net = sock_net(sk);
1387 __u8 flow_flags = inet_sk_flowi_flags(sk);
1388
1389 fl4 = &fl4_stack;
1390
1391 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1392 sk->sk_protocol, flow_flags, faddr, saddr,
1393 dport, inet->inet_sport, sk->sk_uid);
1394
1395 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1396 rt = ip_route_output_flow(net, fl4, sk);
1397 if (IS_ERR(rt)) {
1398 err = PTR_ERR(rt);
1399 rt = NULL;
1400 if (err == -ENETUNREACH)
1401 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1402 goto out;
1403 }
1404
1405 err = -EACCES;
1406 if ((rt->rt_flags & RTCF_BROADCAST) &&
1407 !sock_flag(sk, SOCK_BROADCAST))
1408 goto out;
1409 if (connected)
1410 sk_dst_set(sk, dst_clone(&rt->dst));
1411 }
1412
1413 if (msg->msg_flags&MSG_CONFIRM)
1414 goto do_confirm;
1415 back_from_confirm:
1416
1417 saddr = fl4->saddr;
1418 if (!ipc.addr)
1419 daddr = ipc.addr = fl4->daddr;
1420
1421 /* Lockless fast path for the non-corking case. */
1422 if (!corkreq) {
1423 struct inet_cork cork;
1424
1425 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1426 sizeof(struct udphdr), &ipc, &rt,
1427 &cork, msg->msg_flags);
1428 err = PTR_ERR(skb);
1429 if (!IS_ERR_OR_NULL(skb))
1430 err = udp_send_skb(skb, fl4, &cork);
1431 goto out;
1432 }
1433
1434 lock_sock(sk);
1435 if (unlikely(up->pending)) {
1436 /* The socket is already corked while preparing it. */
1437 /* ... which is an evident application bug. --ANK */
1438 release_sock(sk);
1439
1440 net_dbg_ratelimited("socket already corked\n");
1441 err = -EINVAL;
1442 goto out;
1443 }
1444 /*
1445 * Now cork the socket to pend data.
1446 */
1447 fl4 = &inet->cork.fl.u.ip4;
1448 fl4->daddr = daddr;
1449 fl4->saddr = saddr;
1450 fl4->fl4_dport = dport;
1451 fl4->fl4_sport = inet->inet_sport;
1452 WRITE_ONCE(up->pending, AF_INET);
1453
1454 do_append_data:
1455 up->len += ulen;
1456 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1457 sizeof(struct udphdr), &ipc, &rt,
1458 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1459 if (err)
1460 udp_flush_pending_frames(sk);
1461 else if (!corkreq)
1462 err = udp_push_pending_frames(sk);
1463 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1464 WRITE_ONCE(up->pending, 0);
1465 release_sock(sk);
1466
1467 out:
1468 ip_rt_put(rt);
1469 out_free:
1470 if (free)
1471 kfree(ipc.opt);
1472 if (!err)
1473 return len;
1474 /*
1475 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1476 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1477 * we don't have a good statistic (IpOutDiscards but it can be too many
1478 * things). We could add another new stat but at least for now that
1479 * seems like overkill.
1480 */
1481 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1482 UDP_INC_STATS(sock_net(sk),
1483 UDP_MIB_SNDBUFERRORS, is_udplite);
1484 }
1485 return err;
1486
1487 do_confirm:
1488 if (msg->msg_flags & MSG_PROBE)
1489 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1490 if (!(msg->msg_flags&MSG_PROBE) || len)
1491 goto back_from_confirm;
1492 err = 0;
1493 goto out;
1494 }
1495 EXPORT_SYMBOL(udp_sendmsg);
1496
1497 void udp_splice_eof(struct socket *sock)
1498 {
1499 struct sock *sk = sock->sk;
1500 struct udp_sock *up = udp_sk(sk);
1501
1502 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1503 return;
1504
1505 lock_sock(sk);
1506 if (up->pending && !udp_test_bit(CORK, sk))
1507 udp_push_pending_frames(sk);
1508 release_sock(sk);
1509 }
1510 EXPORT_SYMBOL_GPL(udp_splice_eof);
1511
1512 #define UDP_SKB_IS_STATELESS 0x80000000
1513
1514 /* all head states (dst, sk, nf conntrack) except skb extensions are
1515 * cleared by udp_rcv().
1516 *
1517 * We need to preserve secpath, if present, to eventually process
1518 * IP_CMSG_PASSSEC at recvmsg() time.
1519 *
1520 * Other extensions can be cleared.
1521 */
1522 static bool udp_try_make_stateless(struct sk_buff *skb)
1523 {
1524 if (!skb_has_extensions(skb))
1525 return true;
1526
1527 if (!secpath_exists(skb)) {
1528 skb_ext_reset(skb);
1529 return true;
1530 }
1531
1532 return false;
1533 }
1534
1535 static void udp_set_dev_scratch(struct sk_buff *skb)
1536 {
1537 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1538
1539 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1540 scratch->_tsize_state = skb->truesize;
1541 #if BITS_PER_LONG == 64
1542 scratch->len = skb->len;
1543 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1544 scratch->is_linear = !skb_is_nonlinear(skb);
1545 #endif
1546 if (udp_try_make_stateless(skb))
1547 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1548 }
1549
1550 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1551 {
1552 /* We come here after udp_lib_checksum_complete() returned 0.
1553 * This means that __skb_checksum_complete() might have
1554 * set skb->csum_valid to 1.
1555 * On 64bit platforms, we can set csum_unnecessary
1556 * to true, but only if the skb is not shared.
1557 */
1558 #if BITS_PER_LONG == 64
1559 if (!skb_shared(skb))
1560 udp_skb_scratch(skb)->csum_unnecessary = true;
1561 #endif
1562 }
1563
1564 static int udp_skb_truesize(struct sk_buff *skb)
1565 {
1566 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1567 }
1568
1569 static bool udp_skb_has_head_state(struct sk_buff *skb)
1570 {
1571 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1572 }
1573
1574 /* fully reclaim rmem/fwd memory allocated for skb */
1575 static void udp_rmem_release(struct sock *sk, int size, int partial,
1576 bool rx_queue_lock_held)
1577 {
1578 struct udp_sock *up = udp_sk(sk);
1579 struct sk_buff_head *sk_queue;
1580 int amt;
1581
1582 if (likely(partial)) {
1583 up->forward_deficit += size;
1584 size = up->forward_deficit;
1585 if (size < READ_ONCE(up->forward_threshold) &&
1586 !skb_queue_empty(&up->reader_queue))
1587 return;
1588 } else {
1589 size += up->forward_deficit;
1590 }
1591 up->forward_deficit = 0;
1592
1593 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1594 * if the called don't held it already
1595 */
1596 sk_queue = &sk->sk_receive_queue;
1597 if (!rx_queue_lock_held)
1598 spin_lock(&sk_queue->lock);
1599
1600
1601 sk_forward_alloc_add(sk, size);
1602 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1603 sk_forward_alloc_add(sk, -amt);
1604
1605 if (amt)
1606 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1607
1608 atomic_sub(size, &sk->sk_rmem_alloc);
1609
1610 /* this can save us from acquiring the rx queue lock on next receive */
1611 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1612
1613 if (!rx_queue_lock_held)
1614 spin_unlock(&sk_queue->lock);
1615 }
1616
1617 /* Note: called with reader_queue.lock held.
1618 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1619 * This avoids a cache line miss while receive_queue lock is held.
1620 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1621 */
1622 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1623 {
1624 prefetch(&skb->data);
1625 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1626 }
1627 EXPORT_SYMBOL(udp_skb_destructor);
1628
1629 /* as above, but the caller held the rx queue lock, too */
1630 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1631 {
1632 prefetch(&skb->data);
1633 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1634 }
1635
1636 /* Idea of busylocks is to let producers grab an extra spinlock
1637 * to relieve pressure on the receive_queue spinlock shared by consumer.
1638 * Under flood, this means that only one producer can be in line
1639 * trying to acquire the receive_queue spinlock.
1640 * These busylock can be allocated on a per cpu manner, instead of a
1641 * per socket one (that would consume a cache line per socket)
1642 */
1643 static int udp_busylocks_log __read_mostly;
1644 static spinlock_t *udp_busylocks __read_mostly;
1645
1646 static spinlock_t *busylock_acquire(void *ptr)
1647 {
1648 spinlock_t *busy;
1649
1650 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1651 spin_lock(busy);
1652 return busy;
1653 }
1654
1655 static void busylock_release(spinlock_t *busy)
1656 {
1657 if (busy)
1658 spin_unlock(busy);
1659 }
1660
1661 static int udp_rmem_schedule(struct sock *sk, int size)
1662 {
1663 int delta;
1664
1665 delta = size - sk->sk_forward_alloc;
1666 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1667 return -ENOBUFS;
1668
1669 return 0;
1670 }
1671
1672 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1673 {
1674 struct sk_buff_head *list = &sk->sk_receive_queue;
1675 int rmem, err = -ENOMEM;
1676 spinlock_t *busy = NULL;
1677 bool becomes_readable;
1678 int size, rcvbuf;
1679
1680 /* Immediately drop when the receive queue is full.
1681 * Always allow at least one packet.
1682 */
1683 rmem = atomic_read(&sk->sk_rmem_alloc);
1684 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1685 if (rmem > rcvbuf)
1686 goto drop;
1687
1688 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1689 * having linear skbs :
1690 * - Reduce memory overhead and thus increase receive queue capacity
1691 * - Less cache line misses at copyout() time
1692 * - Less work at consume_skb() (less alien page frag freeing)
1693 */
1694 if (rmem > (rcvbuf >> 1)) {
1695 skb_condense(skb);
1696
1697 busy = busylock_acquire(sk);
1698 }
1699 size = skb->truesize;
1700 udp_set_dev_scratch(skb);
1701
1702 atomic_add(size, &sk->sk_rmem_alloc);
1703
1704 spin_lock(&list->lock);
1705 err = udp_rmem_schedule(sk, size);
1706 if (err) {
1707 spin_unlock(&list->lock);
1708 goto uncharge_drop;
1709 }
1710
1711 sk_forward_alloc_add(sk, -size);
1712
1713 /* no need to setup a destructor, we will explicitly release the
1714 * forward allocated memory on dequeue
1715 */
1716 sock_skb_set_dropcount(sk, skb);
1717
1718 becomes_readable = skb_queue_empty(list);
1719 __skb_queue_tail(list, skb);
1720 spin_unlock(&list->lock);
1721
1722 if (!sock_flag(sk, SOCK_DEAD)) {
1723 if (becomes_readable ||
1724 sk->sk_data_ready != sock_def_readable ||
1725 READ_ONCE(sk->sk_peek_off) >= 0)
1726 INDIRECT_CALL_1(sk->sk_data_ready,
1727 sock_def_readable, sk);
1728 else
1729 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
1730 }
1731 busylock_release(busy);
1732 return 0;
1733
1734 uncharge_drop:
1735 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1736
1737 drop:
1738 atomic_inc(&sk->sk_drops);
1739 busylock_release(busy);
1740 return err;
1741 }
1742 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1743
1744 void udp_destruct_common(struct sock *sk)
1745 {
1746 /* reclaim completely the forward allocated memory */
1747 struct udp_sock *up = udp_sk(sk);
1748 unsigned int total = 0;
1749 struct sk_buff *skb;
1750
1751 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1752 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1753 total += skb->truesize;
1754 kfree_skb(skb);
1755 }
1756 udp_rmem_release(sk, total, 0, true);
1757 }
1758 EXPORT_SYMBOL_GPL(udp_destruct_common);
1759
1760 static void udp_destruct_sock(struct sock *sk)
1761 {
1762 udp_destruct_common(sk);
1763 inet_sock_destruct(sk);
1764 }
1765
1766 int udp_init_sock(struct sock *sk)
1767 {
1768 udp_lib_init_sock(sk);
1769 sk->sk_destruct = udp_destruct_sock;
1770 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1771 return 0;
1772 }
1773
1774 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1775 {
1776 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
1777 sk_peek_offset_bwd(sk, len);
1778
1779 if (!skb_unref(skb))
1780 return;
1781
1782 /* In the more common cases we cleared the head states previously,
1783 * see __udp_queue_rcv_skb().
1784 */
1785 if (unlikely(udp_skb_has_head_state(skb)))
1786 skb_release_head_state(skb);
1787 __consume_stateless_skb(skb);
1788 }
1789 EXPORT_SYMBOL_GPL(skb_consume_udp);
1790
1791 static struct sk_buff *__first_packet_length(struct sock *sk,
1792 struct sk_buff_head *rcvq,
1793 int *total)
1794 {
1795 struct sk_buff *skb;
1796
1797 while ((skb = skb_peek(rcvq)) != NULL) {
1798 if (udp_lib_checksum_complete(skb)) {
1799 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1800 IS_UDPLITE(sk));
1801 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1802 IS_UDPLITE(sk));
1803 atomic_inc(&sk->sk_drops);
1804 __skb_unlink(skb, rcvq);
1805 *total += skb->truesize;
1806 kfree_skb(skb);
1807 } else {
1808 udp_skb_csum_unnecessary_set(skb);
1809 break;
1810 }
1811 }
1812 return skb;
1813 }
1814
1815 /**
1816 * first_packet_length - return length of first packet in receive queue
1817 * @sk: socket
1818 *
1819 * Drops all bad checksum frames, until a valid one is found.
1820 * Returns the length of found skb, or -1 if none is found.
1821 */
1822 static int first_packet_length(struct sock *sk)
1823 {
1824 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1825 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1826 struct sk_buff *skb;
1827 int total = 0;
1828 int res;
1829
1830 spin_lock_bh(&rcvq->lock);
1831 skb = __first_packet_length(sk, rcvq, &total);
1832 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1833 spin_lock(&sk_queue->lock);
1834 skb_queue_splice_tail_init(sk_queue, rcvq);
1835 spin_unlock(&sk_queue->lock);
1836
1837 skb = __first_packet_length(sk, rcvq, &total);
1838 }
1839 res = skb ? skb->len : -1;
1840 if (total)
1841 udp_rmem_release(sk, total, 1, false);
1842 spin_unlock_bh(&rcvq->lock);
1843 return res;
1844 }
1845
1846 /*
1847 * IOCTL requests applicable to the UDP protocol
1848 */
1849
1850 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1851 {
1852 switch (cmd) {
1853 case SIOCOUTQ:
1854 {
1855 *karg = sk_wmem_alloc_get(sk);
1856 return 0;
1857 }
1858
1859 case SIOCINQ:
1860 {
1861 *karg = max_t(int, 0, first_packet_length(sk));
1862 return 0;
1863 }
1864
1865 default:
1866 return -ENOIOCTLCMD;
1867 }
1868
1869 return 0;
1870 }
1871 EXPORT_SYMBOL(udp_ioctl);
1872
1873 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1874 int *off, int *err)
1875 {
1876 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1877 struct sk_buff_head *queue;
1878 struct sk_buff *last;
1879 long timeo;
1880 int error;
1881
1882 queue = &udp_sk(sk)->reader_queue;
1883 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1884 do {
1885 struct sk_buff *skb;
1886
1887 error = sock_error(sk);
1888 if (error)
1889 break;
1890
1891 error = -EAGAIN;
1892 do {
1893 spin_lock_bh(&queue->lock);
1894 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1895 err, &last);
1896 if (skb) {
1897 if (!(flags & MSG_PEEK))
1898 udp_skb_destructor(sk, skb);
1899 spin_unlock_bh(&queue->lock);
1900 return skb;
1901 }
1902
1903 if (skb_queue_empty_lockless(sk_queue)) {
1904 spin_unlock_bh(&queue->lock);
1905 goto busy_check;
1906 }
1907
1908 /* refill the reader queue and walk it again
1909 * keep both queues locked to avoid re-acquiring
1910 * the sk_receive_queue lock if fwd memory scheduling
1911 * is needed.
1912 */
1913 spin_lock(&sk_queue->lock);
1914 skb_queue_splice_tail_init(sk_queue, queue);
1915
1916 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1917 err, &last);
1918 if (skb && !(flags & MSG_PEEK))
1919 udp_skb_dtor_locked(sk, skb);
1920 spin_unlock(&sk_queue->lock);
1921 spin_unlock_bh(&queue->lock);
1922 if (skb)
1923 return skb;
1924
1925 busy_check:
1926 if (!sk_can_busy_loop(sk))
1927 break;
1928
1929 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1930 } while (!skb_queue_empty_lockless(sk_queue));
1931
1932 /* sk_queue is empty, reader_queue may contain peeked packets */
1933 } while (timeo &&
1934 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1935 &error, &timeo,
1936 (struct sk_buff *)sk_queue));
1937
1938 *err = error;
1939 return NULL;
1940 }
1941 EXPORT_SYMBOL(__skb_recv_udp);
1942
1943 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1944 {
1945 struct sk_buff *skb;
1946 int err;
1947
1948 try_again:
1949 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1950 if (!skb)
1951 return err;
1952
1953 if (udp_lib_checksum_complete(skb)) {
1954 int is_udplite = IS_UDPLITE(sk);
1955 struct net *net = sock_net(sk);
1956
1957 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1958 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1959 atomic_inc(&sk->sk_drops);
1960 kfree_skb(skb);
1961 goto try_again;
1962 }
1963
1964 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1965 return recv_actor(sk, skb);
1966 }
1967 EXPORT_SYMBOL(udp_read_skb);
1968
1969 /*
1970 * This should be easy, if there is something there we
1971 * return it, otherwise we block.
1972 */
1973
1974 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1975 int *addr_len)
1976 {
1977 struct inet_sock *inet = inet_sk(sk);
1978 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1979 struct sk_buff *skb;
1980 unsigned int ulen, copied;
1981 int off, err, peeking = flags & MSG_PEEK;
1982 int is_udplite = IS_UDPLITE(sk);
1983 bool checksum_valid = false;
1984
1985 if (flags & MSG_ERRQUEUE)
1986 return ip_recv_error(sk, msg, len, addr_len);
1987
1988 try_again:
1989 off = sk_peek_offset(sk, flags);
1990 skb = __skb_recv_udp(sk, flags, &off, &err);
1991 if (!skb)
1992 return err;
1993
1994 ulen = udp_skb_len(skb);
1995 copied = len;
1996 if (copied > ulen - off)
1997 copied = ulen - off;
1998 else if (copied < ulen)
1999 msg->msg_flags |= MSG_TRUNC;
2000
2001 /*
2002 * If checksum is needed at all, try to do it while copying the
2003 * data. If the data is truncated, or if we only want a partial
2004 * coverage checksum (UDP-Lite), do it before the copy.
2005 */
2006
2007 if (copied < ulen || peeking ||
2008 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
2009 checksum_valid = udp_skb_csum_unnecessary(skb) ||
2010 !__udp_lib_checksum_complete(skb);
2011 if (!checksum_valid)
2012 goto csum_copy_err;
2013 }
2014
2015 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
2016 if (udp_skb_is_linear(skb))
2017 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
2018 else
2019 err = skb_copy_datagram_msg(skb, off, msg, copied);
2020 } else {
2021 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
2022
2023 if (err == -EINVAL)
2024 goto csum_copy_err;
2025 }
2026
2027 if (unlikely(err)) {
2028 if (!peeking) {
2029 atomic_inc(&sk->sk_drops);
2030 UDP_INC_STATS(sock_net(sk),
2031 UDP_MIB_INERRORS, is_udplite);
2032 }
2033 kfree_skb(skb);
2034 return err;
2035 }
2036
2037 if (!peeking)
2038 UDP_INC_STATS(sock_net(sk),
2039 UDP_MIB_INDATAGRAMS, is_udplite);
2040
2041 sock_recv_cmsgs(msg, sk, skb);
2042
2043 /* Copy the address. */
2044 if (sin) {
2045 sin->sin_family = AF_INET;
2046 sin->sin_port = udp_hdr(skb)->source;
2047 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
2048 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
2049 *addr_len = sizeof(*sin);
2050
2051 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
2052 (struct sockaddr *)sin,
2053 addr_len);
2054 }
2055
2056 if (udp_test_bit(GRO_ENABLED, sk))
2057 udp_cmsg_recv(msg, sk, skb);
2058
2059 if (inet_cmsg_flags(inet))
2060 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
2061
2062 err = copied;
2063 if (flags & MSG_TRUNC)
2064 err = ulen;
2065
2066 skb_consume_udp(sk, skb, peeking ? -err : err);
2067 return err;
2068
2069 csum_copy_err:
2070 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
2071 udp_skb_destructor)) {
2072 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2073 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2074 }
2075 kfree_skb(skb);
2076
2077 /* starting over for a new packet, but check if we need to yield */
2078 cond_resched();
2079 msg->msg_flags &= ~MSG_TRUNC;
2080 goto try_again;
2081 }
2082
2083 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
2084 {
2085 /* This check is replicated from __ip4_datagram_connect() and
2086 * intended to prevent BPF program called below from accessing bytes
2087 * that are out of the bound specified by user in addr_len.
2088 */
2089 if (addr_len < sizeof(struct sockaddr_in))
2090 return -EINVAL;
2091
2092 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
2093 }
2094 EXPORT_SYMBOL(udp_pre_connect);
2095
2096 static int udp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
2097 {
2098 int res;
2099
2100 lock_sock(sk);
2101 res = __ip4_datagram_connect(sk, uaddr, addr_len);
2102 if (!res)
2103 udp4_hash4(sk);
2104 release_sock(sk);
2105 return res;
2106 }
2107
2108 int __udp_disconnect(struct sock *sk, int flags)
2109 {
2110 struct inet_sock *inet = inet_sk(sk);
2111 /*
2112 * 1003.1g - break association.
2113 */
2114
2115 sk->sk_state = TCP_CLOSE;
2116 inet->inet_daddr = 0;
2117 inet->inet_dport = 0;
2118 sock_rps_reset_rxhash(sk);
2119 sk->sk_bound_dev_if = 0;
2120 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
2121 inet_reset_saddr(sk);
2122 if (sk->sk_prot->rehash &&
2123 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2124 sk->sk_prot->rehash(sk);
2125 }
2126
2127 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
2128 sk->sk_prot->unhash(sk);
2129 inet->inet_sport = 0;
2130 }
2131 sk_dst_reset(sk);
2132 return 0;
2133 }
2134 EXPORT_SYMBOL(__udp_disconnect);
2135
2136 int udp_disconnect(struct sock *sk, int flags)
2137 {
2138 lock_sock(sk);
2139 __udp_disconnect(sk, flags);
2140 release_sock(sk);
2141 return 0;
2142 }
2143 EXPORT_SYMBOL(udp_disconnect);
2144
2145 void udp_lib_unhash(struct sock *sk)
2146 {
2147 if (sk_hashed(sk)) {
2148 struct udp_table *udptable = udp_get_table_prot(sk);
2149 struct udp_hslot *hslot, *hslot2;
2150
2151 hslot = udp_hashslot(udptable, sock_net(sk),
2152 udp_sk(sk)->udp_port_hash);
2153 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2154
2155 spin_lock_bh(&hslot->lock);
2156 if (rcu_access_pointer(sk->sk_reuseport_cb))
2157 reuseport_detach_sock(sk);
2158 if (sk_del_node_init_rcu(sk)) {
2159 hslot->count--;
2160 inet_sk(sk)->inet_num = 0;
2161 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2162
2163 spin_lock(&hslot2->lock);
2164 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2165 hslot2->count--;
2166 spin_unlock(&hslot2->lock);
2167
2168 udp_unhash4(udptable, sk);
2169 }
2170 spin_unlock_bh(&hslot->lock);
2171 }
2172 }
2173 EXPORT_SYMBOL(udp_lib_unhash);
2174
2175 /*
2176 * inet_rcv_saddr was changed, we must rehash secondary hash
2177 */
2178 void udp_lib_rehash(struct sock *sk, u16 newhash, u16 newhash4)
2179 {
2180 if (sk_hashed(sk)) {
2181 struct udp_table *udptable = udp_get_table_prot(sk);
2182 struct udp_hslot *hslot, *hslot2, *nhslot2;
2183
2184 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2185 nhslot2 = udp_hashslot2(udptable, newhash);
2186 udp_sk(sk)->udp_portaddr_hash = newhash;
2187
2188 if (hslot2 != nhslot2 ||
2189 rcu_access_pointer(sk->sk_reuseport_cb)) {
2190 hslot = udp_hashslot(udptable, sock_net(sk),
2191 udp_sk(sk)->udp_port_hash);
2192 /* we must lock primary chain too */
2193 spin_lock_bh(&hslot->lock);
2194 if (rcu_access_pointer(sk->sk_reuseport_cb))
2195 reuseport_detach_sock(sk);
2196
2197 if (hslot2 != nhslot2) {
2198 spin_lock(&hslot2->lock);
2199 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2200 hslot2->count--;
2201 spin_unlock(&hslot2->lock);
2202
2203 spin_lock(&nhslot2->lock);
2204 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2205 &nhslot2->head);
2206 nhslot2->count++;
2207 spin_unlock(&nhslot2->lock);
2208 }
2209
2210 if (udp_hashed4(sk)) {
2211 udp_rehash4(udptable, sk, newhash4);
2212
2213 if (hslot2 != nhslot2) {
2214 spin_lock(&hslot2->lock);
2215 udp_hash4_dec(hslot2);
2216 spin_unlock(&hslot2->lock);
2217
2218 spin_lock(&nhslot2->lock);
2219 udp_hash4_inc(nhslot2);
2220 spin_unlock(&nhslot2->lock);
2221 }
2222 }
2223 spin_unlock_bh(&hslot->lock);
2224 }
2225 }
2226 }
2227 EXPORT_SYMBOL(udp_lib_rehash);
2228
2229 void udp_v4_rehash(struct sock *sk)
2230 {
2231 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2232 inet_sk(sk)->inet_rcv_saddr,
2233 inet_sk(sk)->inet_num);
2234 u16 new_hash4 = udp_ehashfn(sock_net(sk),
2235 sk->sk_rcv_saddr, sk->sk_num,
2236 sk->sk_daddr, sk->sk_dport);
2237
2238 udp_lib_rehash(sk, new_hash, new_hash4);
2239 }
2240
2241 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2242 {
2243 int rc;
2244
2245 if (inet_sk(sk)->inet_daddr) {
2246 sock_rps_save_rxhash(sk, skb);
2247 sk_mark_napi_id(sk, skb);
2248 sk_incoming_cpu_update(sk);
2249 } else {
2250 sk_mark_napi_id_once(sk, skb);
2251 }
2252
2253 rc = __udp_enqueue_schedule_skb(sk, skb);
2254 if (rc < 0) {
2255 int is_udplite = IS_UDPLITE(sk);
2256 int drop_reason;
2257
2258 /* Note that an ENOMEM error is charged twice */
2259 if (rc == -ENOMEM) {
2260 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2261 is_udplite);
2262 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2263 } else {
2264 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2265 is_udplite);
2266 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2267 }
2268 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2269 trace_udp_fail_queue_rcv_skb(rc, sk, skb);
2270 sk_skb_reason_drop(sk, skb, drop_reason);
2271 return -1;
2272 }
2273
2274 return 0;
2275 }
2276
2277 /* returns:
2278 * -1: error
2279 * 0: success
2280 * >0: "udp encap" protocol resubmission
2281 *
2282 * Note that in the success and error cases, the skb is assumed to
2283 * have either been requeued or freed.
2284 */
2285 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2286 {
2287 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2288 struct udp_sock *up = udp_sk(sk);
2289 int is_udplite = IS_UDPLITE(sk);
2290
2291 /*
2292 * Charge it to the socket, dropping if the queue is full.
2293 */
2294 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2295 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2296 goto drop;
2297 }
2298 nf_reset_ct(skb);
2299
2300 if (static_branch_unlikely(&udp_encap_needed_key) &&
2301 READ_ONCE(up->encap_type)) {
2302 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2303
2304 /*
2305 * This is an encapsulation socket so pass the skb to
2306 * the socket's udp_encap_rcv() hook. Otherwise, just
2307 * fall through and pass this up the UDP socket.
2308 * up->encap_rcv() returns the following value:
2309 * =0 if skb was successfully passed to the encap
2310 * handler or was discarded by it.
2311 * >0 if skb should be passed on to UDP.
2312 * <0 if skb should be resubmitted as proto -N
2313 */
2314
2315 /* if we're overly short, let UDP handle it */
2316 encap_rcv = READ_ONCE(up->encap_rcv);
2317 if (encap_rcv) {
2318 int ret;
2319
2320 /* Verify checksum before giving to encap */
2321 if (udp_lib_checksum_complete(skb))
2322 goto csum_error;
2323
2324 ret = encap_rcv(sk, skb);
2325 if (ret <= 0) {
2326 __UDP_INC_STATS(sock_net(sk),
2327 UDP_MIB_INDATAGRAMS,
2328 is_udplite);
2329 return -ret;
2330 }
2331 }
2332
2333 /* FALLTHROUGH -- it's a UDP Packet */
2334 }
2335
2336 /*
2337 * UDP-Lite specific tests, ignored on UDP sockets
2338 */
2339 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2340 u16 pcrlen = READ_ONCE(up->pcrlen);
2341
2342 /*
2343 * MIB statistics other than incrementing the error count are
2344 * disabled for the following two types of errors: these depend
2345 * on the application settings, not on the functioning of the
2346 * protocol stack as such.
2347 *
2348 * RFC 3828 here recommends (sec 3.3): "There should also be a
2349 * way ... to ... at least let the receiving application block
2350 * delivery of packets with coverage values less than a value
2351 * provided by the application."
2352 */
2353 if (pcrlen == 0) { /* full coverage was set */
2354 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2355 UDP_SKB_CB(skb)->cscov, skb->len);
2356 goto drop;
2357 }
2358 /* The next case involves violating the min. coverage requested
2359 * by the receiver. This is subtle: if receiver wants x and x is
2360 * greater than the buffersize/MTU then receiver will complain
2361 * that it wants x while sender emits packets of smaller size y.
2362 * Therefore the above ...()->partial_cov statement is essential.
2363 */
2364 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2365 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2366 UDP_SKB_CB(skb)->cscov, pcrlen);
2367 goto drop;
2368 }
2369 }
2370
2371 prefetch(&sk->sk_rmem_alloc);
2372 if (rcu_access_pointer(sk->sk_filter) &&
2373 udp_lib_checksum_complete(skb))
2374 goto csum_error;
2375
2376 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2377 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2378 goto drop;
2379 }
2380
2381 udp_csum_pull_header(skb);
2382
2383 ipv4_pktinfo_prepare(sk, skb, true);
2384 return __udp_queue_rcv_skb(sk, skb);
2385
2386 csum_error:
2387 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2388 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2389 drop:
2390 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2391 atomic_inc(&sk->sk_drops);
2392 sk_skb_reason_drop(sk, skb, drop_reason);
2393 return -1;
2394 }
2395
2396 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2397 {
2398 struct sk_buff *next, *segs;
2399 int ret;
2400
2401 if (likely(!udp_unexpected_gso(sk, skb)))
2402 return udp_queue_rcv_one_skb(sk, skb);
2403
2404 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2405 __skb_push(skb, -skb_mac_offset(skb));
2406 segs = udp_rcv_segment(sk, skb, true);
2407 skb_list_walk_safe(segs, skb, next) {
2408 __skb_pull(skb, skb_transport_offset(skb));
2409
2410 udp_post_segment_fix_csum(skb);
2411 ret = udp_queue_rcv_one_skb(sk, skb);
2412 if (ret > 0)
2413 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2414 }
2415 return 0;
2416 }
2417
2418 /* For TCP sockets, sk_rx_dst is protected by socket lock
2419 * For UDP, we use xchg() to guard against concurrent changes.
2420 */
2421 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2422 {
2423 struct dst_entry *old;
2424
2425 if (dst_hold_safe(dst)) {
2426 old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst)));
2427 dst_release(old);
2428 return old != dst;
2429 }
2430 return false;
2431 }
2432 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2433
2434 /*
2435 * Multicasts and broadcasts go to each listener.
2436 *
2437 * Note: called only from the BH handler context.
2438 */
2439 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2440 struct udphdr *uh,
2441 __be32 saddr, __be32 daddr,
2442 struct udp_table *udptable,
2443 int proto)
2444 {
2445 struct sock *sk, *first = NULL;
2446 unsigned short hnum = ntohs(uh->dest);
2447 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2448 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2449 unsigned int offset = offsetof(typeof(*sk), sk_node);
2450 int dif = skb->dev->ifindex;
2451 int sdif = inet_sdif(skb);
2452 struct hlist_node *node;
2453 struct sk_buff *nskb;
2454
2455 if (use_hash2) {
2456 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2457 udptable->mask;
2458 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2459 start_lookup:
2460 hslot = &udptable->hash2[hash2].hslot;
2461 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2462 }
2463
2464 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2465 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2466 uh->source, saddr, dif, sdif, hnum))
2467 continue;
2468
2469 if (!first) {
2470 first = sk;
2471 continue;
2472 }
2473 nskb = skb_clone(skb, GFP_ATOMIC);
2474
2475 if (unlikely(!nskb)) {
2476 atomic_inc(&sk->sk_drops);
2477 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2478 IS_UDPLITE(sk));
2479 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2480 IS_UDPLITE(sk));
2481 continue;
2482 }
2483 if (udp_queue_rcv_skb(sk, nskb) > 0)
2484 consume_skb(nskb);
2485 }
2486
2487 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2488 if (use_hash2 && hash2 != hash2_any) {
2489 hash2 = hash2_any;
2490 goto start_lookup;
2491 }
2492
2493 if (first) {
2494 if (udp_queue_rcv_skb(first, skb) > 0)
2495 consume_skb(skb);
2496 } else {
2497 kfree_skb(skb);
2498 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2499 proto == IPPROTO_UDPLITE);
2500 }
2501 return 0;
2502 }
2503
2504 /* Initialize UDP checksum. If exited with zero value (success),
2505 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2506 * Otherwise, csum completion requires checksumming packet body,
2507 * including udp header and folding it to skb->csum.
2508 */
2509 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2510 int proto)
2511 {
2512 int err;
2513
2514 UDP_SKB_CB(skb)->partial_cov = 0;
2515 UDP_SKB_CB(skb)->cscov = skb->len;
2516
2517 if (proto == IPPROTO_UDPLITE) {
2518 err = udplite_checksum_init(skb, uh);
2519 if (err)
2520 return err;
2521
2522 if (UDP_SKB_CB(skb)->partial_cov) {
2523 skb->csum = inet_compute_pseudo(skb, proto);
2524 return 0;
2525 }
2526 }
2527
2528 /* Note, we are only interested in != 0 or == 0, thus the
2529 * force to int.
2530 */
2531 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2532 inet_compute_pseudo);
2533 if (err)
2534 return err;
2535
2536 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2537 /* If SW calculated the value, we know it's bad */
2538 if (skb->csum_complete_sw)
2539 return 1;
2540
2541 /* HW says the value is bad. Let's validate that.
2542 * skb->csum is no longer the full packet checksum,
2543 * so don't treat it as such.
2544 */
2545 skb_checksum_complete_unset(skb);
2546 }
2547
2548 return 0;
2549 }
2550
2551 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2552 * return code conversion for ip layer consumption
2553 */
2554 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2555 struct udphdr *uh)
2556 {
2557 int ret;
2558
2559 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2560 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2561
2562 ret = udp_queue_rcv_skb(sk, skb);
2563
2564 /* a return value > 0 means to resubmit the input, but
2565 * it wants the return to be -protocol, or 0
2566 */
2567 if (ret > 0)
2568 return -ret;
2569 return 0;
2570 }
2571
2572 /*
2573 * All we need to do is get the socket, and then do a checksum.
2574 */
2575
2576 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2577 int proto)
2578 {
2579 struct sock *sk = NULL;
2580 struct udphdr *uh;
2581 unsigned short ulen;
2582 struct rtable *rt = skb_rtable(skb);
2583 __be32 saddr, daddr;
2584 struct net *net = dev_net(skb->dev);
2585 bool refcounted;
2586 int drop_reason;
2587
2588 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2589
2590 /*
2591 * Validate the packet.
2592 */
2593 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2594 goto drop; /* No space for header. */
2595
2596 uh = udp_hdr(skb);
2597 ulen = ntohs(uh->len);
2598 saddr = ip_hdr(skb)->saddr;
2599 daddr = ip_hdr(skb)->daddr;
2600
2601 if (ulen > skb->len)
2602 goto short_packet;
2603
2604 if (proto == IPPROTO_UDP) {
2605 /* UDP validates ulen. */
2606 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2607 goto short_packet;
2608 uh = udp_hdr(skb);
2609 }
2610
2611 if (udp4_csum_init(skb, uh, proto))
2612 goto csum_error;
2613
2614 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2615 &refcounted, udp_ehashfn);
2616 if (IS_ERR(sk))
2617 goto no_sk;
2618
2619 if (sk) {
2620 struct dst_entry *dst = skb_dst(skb);
2621 int ret;
2622
2623 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2624 udp_sk_rx_dst_set(sk, dst);
2625
2626 ret = udp_unicast_rcv_skb(sk, skb, uh);
2627 if (refcounted)
2628 sock_put(sk);
2629 return ret;
2630 }
2631
2632 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2633 return __udp4_lib_mcast_deliver(net, skb, uh,
2634 saddr, daddr, udptable, proto);
2635
2636 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2637 if (sk)
2638 return udp_unicast_rcv_skb(sk, skb, uh);
2639 no_sk:
2640 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2641 goto drop;
2642 nf_reset_ct(skb);
2643
2644 /* No socket. Drop packet silently, if checksum is wrong */
2645 if (udp_lib_checksum_complete(skb))
2646 goto csum_error;
2647
2648 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2649 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2650 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2651
2652 /*
2653 * Hmm. We got an UDP packet to a port to which we
2654 * don't wanna listen. Ignore it.
2655 */
2656 sk_skb_reason_drop(sk, skb, drop_reason);
2657 return 0;
2658
2659 short_packet:
2660 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2661 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2662 proto == IPPROTO_UDPLITE ? "Lite" : "",
2663 &saddr, ntohs(uh->source),
2664 ulen, skb->len,
2665 &daddr, ntohs(uh->dest));
2666 goto drop;
2667
2668 csum_error:
2669 /*
2670 * RFC1122: OK. Discards the bad packet silently (as far as
2671 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2672 */
2673 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2674 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2675 proto == IPPROTO_UDPLITE ? "Lite" : "",
2676 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2677 ulen);
2678 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2679 drop:
2680 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2681 sk_skb_reason_drop(sk, skb, drop_reason);
2682 return 0;
2683 }
2684
2685 /* We can only early demux multicast if there is a single matching socket.
2686 * If more than one socket found returns NULL
2687 */
2688 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2689 __be16 loc_port, __be32 loc_addr,
2690 __be16 rmt_port, __be32 rmt_addr,
2691 int dif, int sdif)
2692 {
2693 struct udp_table *udptable = net->ipv4.udp_table;
2694 unsigned short hnum = ntohs(loc_port);
2695 struct sock *sk, *result;
2696 struct udp_hslot *hslot;
2697 unsigned int slot;
2698
2699 slot = udp_hashfn(net, hnum, udptable->mask);
2700 hslot = &udptable->hash[slot];
2701
2702 /* Do not bother scanning a too big list */
2703 if (hslot->count > 10)
2704 return NULL;
2705
2706 result = NULL;
2707 sk_for_each_rcu(sk, &hslot->head) {
2708 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2709 rmt_port, rmt_addr, dif, sdif, hnum)) {
2710 if (result)
2711 return NULL;
2712 result = sk;
2713 }
2714 }
2715
2716 return result;
2717 }
2718
2719 /* For unicast we should only early demux connected sockets or we can
2720 * break forwarding setups. The chains here can be long so only check
2721 * if the first socket is an exact match and if not move on.
2722 */
2723 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2724 __be16 loc_port, __be32 loc_addr,
2725 __be16 rmt_port, __be32 rmt_addr,
2726 int dif, int sdif)
2727 {
2728 struct udp_table *udptable = net->ipv4.udp_table;
2729 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2730 unsigned short hnum = ntohs(loc_port);
2731 struct udp_hslot *hslot2;
2732 unsigned int hash2;
2733 __portpair ports;
2734 struct sock *sk;
2735
2736 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2737 hslot2 = udp_hashslot2(udptable, hash2);
2738 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2739
2740 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2741 if (inet_match(net, sk, acookie, ports, dif, sdif))
2742 return sk;
2743 /* Only check first socket in chain */
2744 break;
2745 }
2746 return NULL;
2747 }
2748
2749 int udp_v4_early_demux(struct sk_buff *skb)
2750 {
2751 struct net *net = dev_net(skb->dev);
2752 struct in_device *in_dev = NULL;
2753 const struct iphdr *iph;
2754 const struct udphdr *uh;
2755 struct sock *sk = NULL;
2756 struct dst_entry *dst;
2757 int dif = skb->dev->ifindex;
2758 int sdif = inet_sdif(skb);
2759 int ours;
2760
2761 /* validate the packet */
2762 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2763 return 0;
2764
2765 iph = ip_hdr(skb);
2766 uh = udp_hdr(skb);
2767
2768 if (skb->pkt_type == PACKET_MULTICAST) {
2769 in_dev = __in_dev_get_rcu(skb->dev);
2770
2771 if (!in_dev)
2772 return 0;
2773
2774 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2775 iph->protocol);
2776 if (!ours)
2777 return 0;
2778
2779 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2780 uh->source, iph->saddr,
2781 dif, sdif);
2782 } else if (skb->pkt_type == PACKET_HOST) {
2783 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2784 uh->source, iph->saddr, dif, sdif);
2785 }
2786
2787 if (!sk)
2788 return 0;
2789
2790 skb->sk = sk;
2791 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2792 skb->destructor = sock_pfree;
2793 dst = rcu_dereference(sk->sk_rx_dst);
2794
2795 if (dst)
2796 dst = dst_check(dst, 0);
2797 if (dst) {
2798 u32 itag = 0;
2799
2800 /* set noref for now.
2801 * any place which wants to hold dst has to call
2802 * dst_hold_safe()
2803 */
2804 skb_dst_set_noref(skb, dst);
2805
2806 /* for unconnected multicast sockets we need to validate
2807 * the source on each packet
2808 */
2809 if (!inet_sk(sk)->inet_daddr && in_dev)
2810 return ip_mc_validate_source(skb, iph->daddr,
2811 iph->saddr,
2812 ip4h_dscp(iph),
2813 skb->dev, in_dev, &itag);
2814 }
2815 return 0;
2816 }
2817
2818 int udp_rcv(struct sk_buff *skb)
2819 {
2820 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2821 }
2822
2823 void udp_destroy_sock(struct sock *sk)
2824 {
2825 struct udp_sock *up = udp_sk(sk);
2826 bool slow = lock_sock_fast(sk);
2827
2828 /* protects from races with udp_abort() */
2829 sock_set_flag(sk, SOCK_DEAD);
2830 udp_flush_pending_frames(sk);
2831 unlock_sock_fast(sk, slow);
2832 if (static_branch_unlikely(&udp_encap_needed_key)) {
2833 if (up->encap_type) {
2834 void (*encap_destroy)(struct sock *sk);
2835 encap_destroy = READ_ONCE(up->encap_destroy);
2836 if (encap_destroy)
2837 encap_destroy(sk);
2838 }
2839 if (udp_test_bit(ENCAP_ENABLED, sk))
2840 static_branch_dec(&udp_encap_needed_key);
2841 }
2842 }
2843
2844 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2845 struct sock *sk)
2846 {
2847 #ifdef CONFIG_XFRM
2848 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2849 if (family == AF_INET)
2850 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2851 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2852 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2853 }
2854 #endif
2855 }
2856
2857 /*
2858 * Socket option code for UDP
2859 */
2860 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2861 sockptr_t optval, unsigned int optlen,
2862 int (*push_pending_frames)(struct sock *))
2863 {
2864 struct udp_sock *up = udp_sk(sk);
2865 int val, valbool;
2866 int err = 0;
2867 int is_udplite = IS_UDPLITE(sk);
2868
2869 if (level == SOL_SOCKET) {
2870 err = sk_setsockopt(sk, level, optname, optval, optlen);
2871
2872 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2873 sockopt_lock_sock(sk);
2874 /* paired with READ_ONCE in udp_rmem_release() */
2875 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2876 sockopt_release_sock(sk);
2877 }
2878 return err;
2879 }
2880
2881 if (optlen < sizeof(int))
2882 return -EINVAL;
2883
2884 if (copy_from_sockptr(&val, optval, sizeof(val)))
2885 return -EFAULT;
2886
2887 valbool = val ? 1 : 0;
2888
2889 switch (optname) {
2890 case UDP_CORK:
2891 if (val != 0) {
2892 udp_set_bit(CORK, sk);
2893 } else {
2894 udp_clear_bit(CORK, sk);
2895 lock_sock(sk);
2896 push_pending_frames(sk);
2897 release_sock(sk);
2898 }
2899 break;
2900
2901 case UDP_ENCAP:
2902 switch (val) {
2903 case 0:
2904 #ifdef CONFIG_XFRM
2905 case UDP_ENCAP_ESPINUDP:
2906 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2907 #if IS_ENABLED(CONFIG_IPV6)
2908 if (sk->sk_family == AF_INET6)
2909 WRITE_ONCE(up->encap_rcv,
2910 ipv6_stub->xfrm6_udp_encap_rcv);
2911 else
2912 #endif
2913 WRITE_ONCE(up->encap_rcv,
2914 xfrm4_udp_encap_rcv);
2915 #endif
2916 fallthrough;
2917 case UDP_ENCAP_L2TPINUDP:
2918 WRITE_ONCE(up->encap_type, val);
2919 udp_tunnel_encap_enable(sk);
2920 break;
2921 default:
2922 err = -ENOPROTOOPT;
2923 break;
2924 }
2925 break;
2926
2927 case UDP_NO_CHECK6_TX:
2928 udp_set_no_check6_tx(sk, valbool);
2929 break;
2930
2931 case UDP_NO_CHECK6_RX:
2932 udp_set_no_check6_rx(sk, valbool);
2933 break;
2934
2935 case UDP_SEGMENT:
2936 if (val < 0 || val > USHRT_MAX)
2937 return -EINVAL;
2938 WRITE_ONCE(up->gso_size, val);
2939 break;
2940
2941 case UDP_GRO:
2942
2943 /* when enabling GRO, accept the related GSO packet type */
2944 if (valbool)
2945 udp_tunnel_encap_enable(sk);
2946 udp_assign_bit(GRO_ENABLED, sk, valbool);
2947 udp_assign_bit(ACCEPT_L4, sk, valbool);
2948 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
2949 break;
2950
2951 /*
2952 * UDP-Lite's partial checksum coverage (RFC 3828).
2953 */
2954 /* The sender sets actual checksum coverage length via this option.
2955 * The case coverage > packet length is handled by send module. */
2956 case UDPLITE_SEND_CSCOV:
2957 if (!is_udplite) /* Disable the option on UDP sockets */
2958 return -ENOPROTOOPT;
2959 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2960 val = 8;
2961 else if (val > USHRT_MAX)
2962 val = USHRT_MAX;
2963 WRITE_ONCE(up->pcslen, val);
2964 udp_set_bit(UDPLITE_SEND_CC, sk);
2965 break;
2966
2967 /* The receiver specifies a minimum checksum coverage value. To make
2968 * sense, this should be set to at least 8 (as done below). If zero is
2969 * used, this again means full checksum coverage. */
2970 case UDPLITE_RECV_CSCOV:
2971 if (!is_udplite) /* Disable the option on UDP sockets */
2972 return -ENOPROTOOPT;
2973 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2974 val = 8;
2975 else if (val > USHRT_MAX)
2976 val = USHRT_MAX;
2977 WRITE_ONCE(up->pcrlen, val);
2978 udp_set_bit(UDPLITE_RECV_CC, sk);
2979 break;
2980
2981 default:
2982 err = -ENOPROTOOPT;
2983 break;
2984 }
2985
2986 return err;
2987 }
2988 EXPORT_SYMBOL(udp_lib_setsockopt);
2989
2990 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2991 unsigned int optlen)
2992 {
2993 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2994 return udp_lib_setsockopt(sk, level, optname,
2995 optval, optlen,
2996 udp_push_pending_frames);
2997 return ip_setsockopt(sk, level, optname, optval, optlen);
2998 }
2999
3000 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
3001 char __user *optval, int __user *optlen)
3002 {
3003 struct udp_sock *up = udp_sk(sk);
3004 int val, len;
3005
3006 if (get_user(len, optlen))
3007 return -EFAULT;
3008
3009 if (len < 0)
3010 return -EINVAL;
3011
3012 len = min_t(unsigned int, len, sizeof(int));
3013
3014 switch (optname) {
3015 case UDP_CORK:
3016 val = udp_test_bit(CORK, sk);
3017 break;
3018
3019 case UDP_ENCAP:
3020 val = READ_ONCE(up->encap_type);
3021 break;
3022
3023 case UDP_NO_CHECK6_TX:
3024 val = udp_get_no_check6_tx(sk);
3025 break;
3026
3027 case UDP_NO_CHECK6_RX:
3028 val = udp_get_no_check6_rx(sk);
3029 break;
3030
3031 case UDP_SEGMENT:
3032 val = READ_ONCE(up->gso_size);
3033 break;
3034
3035 case UDP_GRO:
3036 val = udp_test_bit(GRO_ENABLED, sk);
3037 break;
3038
3039 /* The following two cannot be changed on UDP sockets, the return is
3040 * always 0 (which corresponds to the full checksum coverage of UDP). */
3041 case UDPLITE_SEND_CSCOV:
3042 val = READ_ONCE(up->pcslen);
3043 break;
3044
3045 case UDPLITE_RECV_CSCOV:
3046 val = READ_ONCE(up->pcrlen);
3047 break;
3048
3049 default:
3050 return -ENOPROTOOPT;
3051 }
3052
3053 if (put_user(len, optlen))
3054 return -EFAULT;
3055 if (copy_to_user(optval, &val, len))
3056 return -EFAULT;
3057 return 0;
3058 }
3059 EXPORT_SYMBOL(udp_lib_getsockopt);
3060
3061 int udp_getsockopt(struct sock *sk, int level, int optname,
3062 char __user *optval, int __user *optlen)
3063 {
3064 if (level == SOL_UDP || level == SOL_UDPLITE)
3065 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
3066 return ip_getsockopt(sk, level, optname, optval, optlen);
3067 }
3068
3069 /**
3070 * udp_poll - wait for a UDP event.
3071 * @file: - file struct
3072 * @sock: - socket
3073 * @wait: - poll table
3074 *
3075 * This is same as datagram poll, except for the special case of
3076 * blocking sockets. If application is using a blocking fd
3077 * and a packet with checksum error is in the queue;
3078 * then it could get return from select indicating data available
3079 * but then block when reading it. Add special case code
3080 * to work around these arguably broken applications.
3081 */
3082 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
3083 {
3084 __poll_t mask = datagram_poll(file, sock, wait);
3085 struct sock *sk = sock->sk;
3086
3087 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
3088 mask |= EPOLLIN | EPOLLRDNORM;
3089
3090 /* Check for false positives due to checksum errors */
3091 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
3092 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
3093 mask &= ~(EPOLLIN | EPOLLRDNORM);
3094
3095 /* psock ingress_msg queue should not contain any bad checksum frames */
3096 if (sk_is_readable(sk))
3097 mask |= EPOLLIN | EPOLLRDNORM;
3098 return mask;
3099
3100 }
3101 EXPORT_SYMBOL(udp_poll);
3102
3103 int udp_abort(struct sock *sk, int err)
3104 {
3105 if (!has_current_bpf_ctx())
3106 lock_sock(sk);
3107
3108 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
3109 * with close()
3110 */
3111 if (sock_flag(sk, SOCK_DEAD))
3112 goto out;
3113
3114 sk->sk_err = err;
3115 sk_error_report(sk);
3116 __udp_disconnect(sk, 0);
3117
3118 out:
3119 if (!has_current_bpf_ctx())
3120 release_sock(sk);
3121
3122 return 0;
3123 }
3124 EXPORT_SYMBOL_GPL(udp_abort);
3125
3126 struct proto udp_prot = {
3127 .name = "UDP",
3128 .owner = THIS_MODULE,
3129 .close = udp_lib_close,
3130 .pre_connect = udp_pre_connect,
3131 .connect = udp_connect,
3132 .disconnect = udp_disconnect,
3133 .ioctl = udp_ioctl,
3134 .init = udp_init_sock,
3135 .destroy = udp_destroy_sock,
3136 .setsockopt = udp_setsockopt,
3137 .getsockopt = udp_getsockopt,
3138 .sendmsg = udp_sendmsg,
3139 .recvmsg = udp_recvmsg,
3140 .splice_eof = udp_splice_eof,
3141 .release_cb = ip4_datagram_release_cb,
3142 .hash = udp_lib_hash,
3143 .unhash = udp_lib_unhash,
3144 .rehash = udp_v4_rehash,
3145 .get_port = udp_v4_get_port,
3146 .put_port = udp_lib_unhash,
3147 #ifdef CONFIG_BPF_SYSCALL
3148 .psock_update_sk_prot = udp_bpf_update_proto,
3149 #endif
3150 .memory_allocated = &udp_memory_allocated,
3151 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
3152
3153 .sysctl_mem = sysctl_udp_mem,
3154 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
3155 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
3156 .obj_size = sizeof(struct udp_sock),
3157 .h.udp_table = NULL,
3158 .diag_destroy = udp_abort,
3159 };
3160 EXPORT_SYMBOL(udp_prot);
3161
3162 /* ------------------------------------------------------------------------ */
3163 #ifdef CONFIG_PROC_FS
3164
3165 static unsigned short seq_file_family(const struct seq_file *seq);
3166 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
3167 {
3168 unsigned short family = seq_file_family(seq);
3169
3170 /* AF_UNSPEC is used as a match all */
3171 return ((family == AF_UNSPEC || family == sk->sk_family) &&
3172 net_eq(sock_net(sk), seq_file_net(seq)));
3173 }
3174
3175 #ifdef CONFIG_BPF_SYSCALL
3176 static const struct seq_operations bpf_iter_udp_seq_ops;
3177 #endif
3178 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
3179 struct net *net)
3180 {
3181 const struct udp_seq_afinfo *afinfo;
3182
3183 #ifdef CONFIG_BPF_SYSCALL
3184 if (seq->op == &bpf_iter_udp_seq_ops)
3185 return net->ipv4.udp_table;
3186 #endif
3187
3188 afinfo = pde_data(file_inode(seq->file));
3189 return afinfo->udp_table ? : net->ipv4.udp_table;
3190 }
3191
3192 static struct sock *udp_get_first(struct seq_file *seq, int start)
3193 {
3194 struct udp_iter_state *state = seq->private;
3195 struct net *net = seq_file_net(seq);
3196 struct udp_table *udptable;
3197 struct sock *sk;
3198
3199 udptable = udp_get_table_seq(seq, net);
3200
3201 for (state->bucket = start; state->bucket <= udptable->mask;
3202 ++state->bucket) {
3203 struct udp_hslot *hslot = &udptable->hash[state->bucket];
3204
3205 if (hlist_empty(&hslot->head))
3206 continue;
3207
3208 spin_lock_bh(&hslot->lock);
3209 sk_for_each(sk, &hslot->head) {
3210 if (seq_sk_match(seq, sk))
3211 goto found;
3212 }
3213 spin_unlock_bh(&hslot->lock);
3214 }
3215 sk = NULL;
3216 found:
3217 return sk;
3218 }
3219
3220 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3221 {
3222 struct udp_iter_state *state = seq->private;
3223 struct net *net = seq_file_net(seq);
3224 struct udp_table *udptable;
3225
3226 do {
3227 sk = sk_next(sk);
3228 } while (sk && !seq_sk_match(seq, sk));
3229
3230 if (!sk) {
3231 udptable = udp_get_table_seq(seq, net);
3232
3233 if (state->bucket <= udptable->mask)
3234 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3235
3236 return udp_get_first(seq, state->bucket + 1);
3237 }
3238 return sk;
3239 }
3240
3241 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3242 {
3243 struct sock *sk = udp_get_first(seq, 0);
3244
3245 if (sk)
3246 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3247 --pos;
3248 return pos ? NULL : sk;
3249 }
3250
3251 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3252 {
3253 struct udp_iter_state *state = seq->private;
3254 state->bucket = MAX_UDP_PORTS;
3255
3256 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3257 }
3258 EXPORT_SYMBOL(udp_seq_start);
3259
3260 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3261 {
3262 struct sock *sk;
3263
3264 if (v == SEQ_START_TOKEN)
3265 sk = udp_get_idx(seq, 0);
3266 else
3267 sk = udp_get_next(seq, v);
3268
3269 ++*pos;
3270 return sk;
3271 }
3272 EXPORT_SYMBOL(udp_seq_next);
3273
3274 void udp_seq_stop(struct seq_file *seq, void *v)
3275 {
3276 struct udp_iter_state *state = seq->private;
3277 struct udp_table *udptable;
3278
3279 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3280
3281 if (state->bucket <= udptable->mask)
3282 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3283 }
3284 EXPORT_SYMBOL(udp_seq_stop);
3285
3286 /* ------------------------------------------------------------------------ */
3287 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3288 int bucket)
3289 {
3290 struct inet_sock *inet = inet_sk(sp);
3291 __be32 dest = inet->inet_daddr;
3292 __be32 src = inet->inet_rcv_saddr;
3293 __u16 destp = ntohs(inet->inet_dport);
3294 __u16 srcp = ntohs(inet->inet_sport);
3295
3296 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3297 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3298 bucket, src, srcp, dest, destp, sp->sk_state,
3299 sk_wmem_alloc_get(sp),
3300 udp_rqueue_get(sp),
3301 0, 0L, 0,
3302 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3303 0, sock_i_ino(sp),
3304 refcount_read(&sp->sk_refcnt), sp,
3305 atomic_read(&sp->sk_drops));
3306 }
3307
3308 int udp4_seq_show(struct seq_file *seq, void *v)
3309 {
3310 seq_setwidth(seq, 127);
3311 if (v == SEQ_START_TOKEN)
3312 seq_puts(seq, " sl local_address rem_address st tx_queue "
3313 "rx_queue tr tm->when retrnsmt uid timeout "
3314 "inode ref pointer drops");
3315 else {
3316 struct udp_iter_state *state = seq->private;
3317
3318 udp4_format_sock(v, seq, state->bucket);
3319 }
3320 seq_pad(seq, '\n');
3321 return 0;
3322 }
3323
3324 #ifdef CONFIG_BPF_SYSCALL
3325 struct bpf_iter__udp {
3326 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3327 __bpf_md_ptr(struct udp_sock *, udp_sk);
3328 uid_t uid __aligned(8);
3329 int bucket __aligned(8);
3330 };
3331
3332 struct bpf_udp_iter_state {
3333 struct udp_iter_state state;
3334 unsigned int cur_sk;
3335 unsigned int end_sk;
3336 unsigned int max_sk;
3337 int offset;
3338 struct sock **batch;
3339 bool st_bucket_done;
3340 };
3341
3342 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3343 unsigned int new_batch_sz);
3344 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3345 {
3346 struct bpf_udp_iter_state *iter = seq->private;
3347 struct udp_iter_state *state = &iter->state;
3348 struct net *net = seq_file_net(seq);
3349 int resume_bucket, resume_offset;
3350 struct udp_table *udptable;
3351 unsigned int batch_sks = 0;
3352 bool resized = false;
3353 struct sock *sk;
3354
3355 resume_bucket = state->bucket;
3356 resume_offset = iter->offset;
3357
3358 /* The current batch is done, so advance the bucket. */
3359 if (iter->st_bucket_done)
3360 state->bucket++;
3361
3362 udptable = udp_get_table_seq(seq, net);
3363
3364 again:
3365 /* New batch for the next bucket.
3366 * Iterate over the hash table to find a bucket with sockets matching
3367 * the iterator attributes, and return the first matching socket from
3368 * the bucket. The remaining matched sockets from the bucket are batched
3369 * before releasing the bucket lock. This allows BPF programs that are
3370 * called in seq_show to acquire the bucket lock if needed.
3371 */
3372 iter->cur_sk = 0;
3373 iter->end_sk = 0;
3374 iter->st_bucket_done = false;
3375 batch_sks = 0;
3376
3377 for (; state->bucket <= udptable->mask; state->bucket++) {
3378 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket].hslot;
3379
3380 if (hlist_empty(&hslot2->head))
3381 continue;
3382
3383 iter->offset = 0;
3384 spin_lock_bh(&hslot2->lock);
3385 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3386 if (seq_sk_match(seq, sk)) {
3387 /* Resume from the last iterated socket at the
3388 * offset in the bucket before iterator was stopped.
3389 */
3390 if (state->bucket == resume_bucket &&
3391 iter->offset < resume_offset) {
3392 ++iter->offset;
3393 continue;
3394 }
3395 if (iter->end_sk < iter->max_sk) {
3396 sock_hold(sk);
3397 iter->batch[iter->end_sk++] = sk;
3398 }
3399 batch_sks++;
3400 }
3401 }
3402 spin_unlock_bh(&hslot2->lock);
3403
3404 if (iter->end_sk)
3405 break;
3406 }
3407
3408 /* All done: no batch made. */
3409 if (!iter->end_sk)
3410 return NULL;
3411
3412 if (iter->end_sk == batch_sks) {
3413 /* Batching is done for the current bucket; return the first
3414 * socket to be iterated from the batch.
3415 */
3416 iter->st_bucket_done = true;
3417 goto done;
3418 }
3419 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3420 resized = true;
3421 /* After allocating a larger batch, retry one more time to grab
3422 * the whole bucket.
3423 */
3424 goto again;
3425 }
3426 done:
3427 return iter->batch[0];
3428 }
3429
3430 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3431 {
3432 struct bpf_udp_iter_state *iter = seq->private;
3433 struct sock *sk;
3434
3435 /* Whenever seq_next() is called, the iter->cur_sk is
3436 * done with seq_show(), so unref the iter->cur_sk.
3437 */
3438 if (iter->cur_sk < iter->end_sk) {
3439 sock_put(iter->batch[iter->cur_sk++]);
3440 ++iter->offset;
3441 }
3442
3443 /* After updating iter->cur_sk, check if there are more sockets
3444 * available in the current bucket batch.
3445 */
3446 if (iter->cur_sk < iter->end_sk)
3447 sk = iter->batch[iter->cur_sk];
3448 else
3449 /* Prepare a new batch. */
3450 sk = bpf_iter_udp_batch(seq);
3451
3452 ++*pos;
3453 return sk;
3454 }
3455
3456 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3457 {
3458 /* bpf iter does not support lseek, so it always
3459 * continue from where it was stop()-ped.
3460 */
3461 if (*pos)
3462 return bpf_iter_udp_batch(seq);
3463
3464 return SEQ_START_TOKEN;
3465 }
3466
3467 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3468 struct udp_sock *udp_sk, uid_t uid, int bucket)
3469 {
3470 struct bpf_iter__udp ctx;
3471
3472 meta->seq_num--; /* skip SEQ_START_TOKEN */
3473 ctx.meta = meta;
3474 ctx.udp_sk = udp_sk;
3475 ctx.uid = uid;
3476 ctx.bucket = bucket;
3477 return bpf_iter_run_prog(prog, &ctx);
3478 }
3479
3480 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3481 {
3482 struct udp_iter_state *state = seq->private;
3483 struct bpf_iter_meta meta;
3484 struct bpf_prog *prog;
3485 struct sock *sk = v;
3486 uid_t uid;
3487 int ret;
3488
3489 if (v == SEQ_START_TOKEN)
3490 return 0;
3491
3492 lock_sock(sk);
3493
3494 if (unlikely(sk_unhashed(sk))) {
3495 ret = SEQ_SKIP;
3496 goto unlock;
3497 }
3498
3499 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3500 meta.seq = seq;
3501 prog = bpf_iter_get_info(&meta, false);
3502 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3503
3504 unlock:
3505 release_sock(sk);
3506 return ret;
3507 }
3508
3509 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3510 {
3511 while (iter->cur_sk < iter->end_sk)
3512 sock_put(iter->batch[iter->cur_sk++]);
3513 }
3514
3515 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3516 {
3517 struct bpf_udp_iter_state *iter = seq->private;
3518 struct bpf_iter_meta meta;
3519 struct bpf_prog *prog;
3520
3521 if (!v) {
3522 meta.seq = seq;
3523 prog = bpf_iter_get_info(&meta, true);
3524 if (prog)
3525 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3526 }
3527
3528 if (iter->cur_sk < iter->end_sk) {
3529 bpf_iter_udp_put_batch(iter);
3530 iter->st_bucket_done = false;
3531 }
3532 }
3533
3534 static const struct seq_operations bpf_iter_udp_seq_ops = {
3535 .start = bpf_iter_udp_seq_start,
3536 .next = bpf_iter_udp_seq_next,
3537 .stop = bpf_iter_udp_seq_stop,
3538 .show = bpf_iter_udp_seq_show,
3539 };
3540 #endif
3541
3542 static unsigned short seq_file_family(const struct seq_file *seq)
3543 {
3544 const struct udp_seq_afinfo *afinfo;
3545
3546 #ifdef CONFIG_BPF_SYSCALL
3547 /* BPF iterator: bpf programs to filter sockets. */
3548 if (seq->op == &bpf_iter_udp_seq_ops)
3549 return AF_UNSPEC;
3550 #endif
3551
3552 /* Proc fs iterator */
3553 afinfo = pde_data(file_inode(seq->file));
3554 return afinfo->family;
3555 }
3556
3557 const struct seq_operations udp_seq_ops = {
3558 .start = udp_seq_start,
3559 .next = udp_seq_next,
3560 .stop = udp_seq_stop,
3561 .show = udp4_seq_show,
3562 };
3563 EXPORT_SYMBOL(udp_seq_ops);
3564
3565 static struct udp_seq_afinfo udp4_seq_afinfo = {
3566 .family = AF_INET,
3567 .udp_table = NULL,
3568 };
3569
3570 static int __net_init udp4_proc_init_net(struct net *net)
3571 {
3572 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3573 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3574 return -ENOMEM;
3575 return 0;
3576 }
3577
3578 static void __net_exit udp4_proc_exit_net(struct net *net)
3579 {
3580 remove_proc_entry("udp", net->proc_net);
3581 }
3582
3583 static struct pernet_operations udp4_net_ops = {
3584 .init = udp4_proc_init_net,
3585 .exit = udp4_proc_exit_net,
3586 };
3587
3588 int __init udp4_proc_init(void)
3589 {
3590 return register_pernet_subsys(&udp4_net_ops);
3591 }
3592
3593 void udp4_proc_exit(void)
3594 {
3595 unregister_pernet_subsys(&udp4_net_ops);
3596 }
3597 #endif /* CONFIG_PROC_FS */
3598
3599 static __initdata unsigned long uhash_entries;
3600 static int __init set_uhash_entries(char *str)
3601 {
3602 ssize_t ret;
3603
3604 if (!str)
3605 return 0;
3606
3607 ret = kstrtoul(str, 0, &uhash_entries);
3608 if (ret)
3609 return 0;
3610
3611 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3612 uhash_entries = UDP_HTABLE_SIZE_MIN;
3613 return 1;
3614 }
3615 __setup("uhash_entries=", set_uhash_entries);
3616
3617 void __init udp_table_init(struct udp_table *table, const char *name)
3618 {
3619 unsigned int i, slot_size;
3620
3621 slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3622 udp_hash4_slot_size();
3623 table->hash = alloc_large_system_hash(name,
3624 slot_size,
3625 uhash_entries,
3626 21, /* one slot per 2 MB */
3627 0,
3628 &table->log,
3629 &table->mask,
3630 UDP_HTABLE_SIZE_MIN,
3631 UDP_HTABLE_SIZE_MAX);
3632
3633 table->hash2 = (void *)(table->hash + (table->mask + 1));
3634 for (i = 0; i <= table->mask; i++) {
3635 INIT_HLIST_HEAD(&table->hash[i].head);
3636 table->hash[i].count = 0;
3637 spin_lock_init(&table->hash[i].lock);
3638 }
3639 for (i = 0; i <= table->mask; i++) {
3640 INIT_HLIST_HEAD(&table->hash2[i].hslot.head);
3641 table->hash2[i].hslot.count = 0;
3642 spin_lock_init(&table->hash2[i].hslot.lock);
3643 }
3644 udp_table_hash4_init(table);
3645 }
3646
3647 u32 udp_flow_hashrnd(void)
3648 {
3649 static u32 hashrnd __read_mostly;
3650
3651 net_get_random_once(&hashrnd, sizeof(hashrnd));
3652
3653 return hashrnd;
3654 }
3655 EXPORT_SYMBOL(udp_flow_hashrnd);
3656
3657 static void __net_init udp_sysctl_init(struct net *net)
3658 {
3659 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3660 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3661
3662 #ifdef CONFIG_NET_L3_MASTER_DEV
3663 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3664 #endif
3665 }
3666
3667 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3668 {
3669 struct udp_table *udptable;
3670 unsigned int slot_size;
3671 int i;
3672
3673 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3674 if (!udptable)
3675 goto out;
3676
3677 slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3678 udp_hash4_slot_size();
3679 udptable->hash = vmalloc_huge(hash_entries * slot_size,
3680 GFP_KERNEL_ACCOUNT);
3681 if (!udptable->hash)
3682 goto free_table;
3683
3684 udptable->hash2 = (void *)(udptable->hash + hash_entries);
3685 udptable->mask = hash_entries - 1;
3686 udptable->log = ilog2(hash_entries);
3687
3688 for (i = 0; i < hash_entries; i++) {
3689 INIT_HLIST_HEAD(&udptable->hash[i].head);
3690 udptable->hash[i].count = 0;
3691 spin_lock_init(&udptable->hash[i].lock);
3692
3693 INIT_HLIST_HEAD(&udptable->hash2[i].hslot.head);
3694 udptable->hash2[i].hslot.count = 0;
3695 spin_lock_init(&udptable->hash2[i].hslot.lock);
3696 }
3697 udp_table_hash4_init(udptable);
3698
3699 return udptable;
3700
3701 free_table:
3702 kfree(udptable);
3703 out:
3704 return NULL;
3705 }
3706
3707 static void __net_exit udp_pernet_table_free(struct net *net)
3708 {
3709 struct udp_table *udptable = net->ipv4.udp_table;
3710
3711 if (udptable == &udp_table)
3712 return;
3713
3714 kvfree(udptable->hash);
3715 kfree(udptable);
3716 }
3717
3718 static void __net_init udp_set_table(struct net *net)
3719 {
3720 struct udp_table *udptable;
3721 unsigned int hash_entries;
3722 struct net *old_net;
3723
3724 if (net_eq(net, &init_net))
3725 goto fallback;
3726
3727 old_net = current->nsproxy->net_ns;
3728 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3729 if (!hash_entries)
3730 goto fallback;
3731
3732 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3733 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3734 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3735 else
3736 hash_entries = roundup_pow_of_two(hash_entries);
3737
3738 udptable = udp_pernet_table_alloc(hash_entries);
3739 if (udptable) {
3740 net->ipv4.udp_table = udptable;
3741 } else {
3742 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3743 "for a netns, fallback to the global one\n",
3744 hash_entries);
3745 fallback:
3746 net->ipv4.udp_table = &udp_table;
3747 }
3748 }
3749
3750 static int __net_init udp_pernet_init(struct net *net)
3751 {
3752 udp_sysctl_init(net);
3753 udp_set_table(net);
3754
3755 return 0;
3756 }
3757
3758 static void __net_exit udp_pernet_exit(struct net *net)
3759 {
3760 udp_pernet_table_free(net);
3761 }
3762
3763 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3764 .init = udp_pernet_init,
3765 .exit = udp_pernet_exit,
3766 };
3767
3768 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3769 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3770 struct udp_sock *udp_sk, uid_t uid, int bucket)
3771
3772 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3773 unsigned int new_batch_sz)
3774 {
3775 struct sock **new_batch;
3776
3777 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3778 GFP_USER | __GFP_NOWARN);
3779 if (!new_batch)
3780 return -ENOMEM;
3781
3782 bpf_iter_udp_put_batch(iter);
3783 kvfree(iter->batch);
3784 iter->batch = new_batch;
3785 iter->max_sk = new_batch_sz;
3786
3787 return 0;
3788 }
3789
3790 #define INIT_BATCH_SZ 16
3791
3792 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3793 {
3794 struct bpf_udp_iter_state *iter = priv_data;
3795 int ret;
3796
3797 ret = bpf_iter_init_seq_net(priv_data, aux);
3798 if (ret)
3799 return ret;
3800
3801 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3802 if (ret)
3803 bpf_iter_fini_seq_net(priv_data);
3804
3805 return ret;
3806 }
3807
3808 static void bpf_iter_fini_udp(void *priv_data)
3809 {
3810 struct bpf_udp_iter_state *iter = priv_data;
3811
3812 bpf_iter_fini_seq_net(priv_data);
3813 kvfree(iter->batch);
3814 }
3815
3816 static const struct bpf_iter_seq_info udp_seq_info = {
3817 .seq_ops = &bpf_iter_udp_seq_ops,
3818 .init_seq_private = bpf_iter_init_udp,
3819 .fini_seq_private = bpf_iter_fini_udp,
3820 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3821 };
3822
3823 static struct bpf_iter_reg udp_reg_info = {
3824 .target = "udp",
3825 .ctx_arg_info_size = 1,
3826 .ctx_arg_info = {
3827 { offsetof(struct bpf_iter__udp, udp_sk),
3828 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3829 },
3830 .seq_info = &udp_seq_info,
3831 };
3832
3833 static void __init bpf_iter_register(void)
3834 {
3835 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3836 if (bpf_iter_reg_target(&udp_reg_info))
3837 pr_warn("Warning: could not register bpf iterator udp\n");
3838 }
3839 #endif
3840
3841 void __init udp_init(void)
3842 {
3843 unsigned long limit;
3844 unsigned int i;
3845
3846 udp_table_init(&udp_table, "UDP");
3847 limit = nr_free_buffer_pages() / 8;
3848 limit = max(limit, 128UL);
3849 sysctl_udp_mem[0] = limit / 4 * 3;
3850 sysctl_udp_mem[1] = limit;
3851 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3852
3853 /* 16 spinlocks per cpu */
3854 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3855 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3856 GFP_KERNEL);
3857 if (!udp_busylocks)
3858 panic("UDP: failed to alloc udp_busylocks\n");
3859 for (i = 0; i < (1U << udp_busylocks_log); i++)
3860 spin_lock_init(udp_busylocks + i);
3861
3862 if (register_pernet_subsys(&udp_sysctl_ops))
3863 panic("UDP: failed to init sysctl parameters.\n");
3864
3865 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3866 bpf_iter_register();
3867 #endif
3868 }
Kernel DRM miscellaneous fixes and cross-tree changes