projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net/ipv6: switch do_ipv6_setsockopt to sockptr_t
[linux-2.6-block.git] / net / ipv4 / udp.c
CommitLineData
2874c5fd 1// SPDX-License-Identifier: GPL-2.0-or-later
1da177e4
LT		 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 *
02c30a84 9 * Authors: Ross Biro
1da177e4
LT		 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
113aa838 12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
1da177e4
LT		 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
e905a9ed
YH		 22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
1da177e4
LT		 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
e905a9ed 58 * for connect.
1da177e4
LT		 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
342f0234 72 * James Chapman : Add L2TP encapsulation type.
1da177e4 73 */
e905a9ed 74
afd46503
JP		 75#define pr_fmt(fmt) "UDP: " fmt
76
7c0f6ba6 77#include <linux/uaccess.h>
1da177e4 78#include <asm/ioctls.h>
57c8a661 79#include <linux/memblock.h>
8203efb3
ED		 80#include <linux/highmem.h>
81#include <linux/swap.h>
1da177e4
LT		 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>
14c85021 87#include <linux/igmp.h>
6e540309 88#include <linux/inetdevice.h>
1da177e4
LT		 89#include <linux/in.h>
90#include <linux/errno.h>
91#include <linux/timer.h>
92#include <linux/mm.h>
1da177e4 93#include <linux/inet.h>
1da177e4 94#include <linux/netdevice.h>
5a0e3ad6 95#include <linux/slab.h>
c752f073 96#include <net/tcp_states.h>
1da177e4
LT		 97#include <linux/skbuff.h>
98#include <linux/proc_fs.h>
99#include <linux/seq_file.h>
457c4cbc 100#include <net/net_namespace.h>
1da177e4 101#include <net/icmp.h>
421b3885 102#include <net/inet_hashtables.h>
e7cc0824 103#include <net/ip_tunnels.h>
1da177e4 104#include <net/route.h>
1da177e4
LT		 105#include <net/checksum.h>
106#include <net/xfrm.h>
296f7ea7 107#include <trace/events/udp.h>
447167bf 108#include <linux/static_key.h>
951cf368 109#include <linux/btf_ids.h>
22911fc5 110#include <trace/events/skb.h>
076bb0c8 111#include <net/busy_poll.h>
ba4e58ec 112#include "udp_impl.h"
e32ea7e7 113#include <net/sock_reuseport.h>
217375a0 114#include <net/addrconf.h>
60fb9567 115#include <net/udp_tunnel.h>
0146dca7
SD		 116#if IS_ENABLED(CONFIG_IPV6)
117#include <net/ipv6_stubs.h>
118#endif
1da177e4 119
f86dcc5a 120struct udp_table udp_table __read_mostly;
645ca708 121EXPORT_SYMBOL(udp_table);
1da177e4 122
8d987e5c 123long sysctl_udp_mem[3] __read_mostly;
95766fff 124EXPORT_SYMBOL(sysctl_udp_mem);
c482c568 125
8d987e5c 126atomic_long_t udp_memory_allocated;
95766fff
HA		 127EXPORT_SYMBOL(udp_memory_allocated);
128
f86dcc5a
ED		 129#define MAX_UDP_PORTS 65536
130#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
98322f22 131
f24d43c0 132static int udp_lib_lport_inuse(struct net *net, __u16 num,
645ca708 133 const struct udp_hslot *hslot,
98322f22 134 unsigned long *bitmap,
fe38d2a1 135 struct sock *sk, unsigned int log)
1da177e4 136{
f24d43c0 137 struct sock *sk2;
ba418fa3 138 kuid_t uid = sock_i_uid(sk);
25030a7f 139
ca065d0c 140 sk_for_each(sk2, &hslot->head) {
9d4fb27d
JP		 141 if (net_eq(sock_net(sk2), net) &&
142 sk2 != sk &&
d4cada4a 143 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
9d4fb27d
JP		 144 (!sk2->sk_reuse || !sk->sk_reuse) &&
145 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
146 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
fe38d2a1 147 inet_rcv_saddr_equal(sk, sk2, true)) {
df560056
EG		 148 if (sk2->sk_reuseport && sk->sk_reuseport &&
149 !rcu_access_pointer(sk->sk_reuseport_cb) &&
150 uid_eq(uid, sock_i_uid(sk2))) {
151 if (!bitmap)
152 return 0;
153 } else {
154 if (!bitmap)
155 return 1;
156 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
157 bitmap);
158 }
98322f22 159 }
4243cdc2 160 }
25030a7f
GR		 161 return 0;
162}
163
30fff923
ED		 164/*
165 * Note: we still hold spinlock of primary hash chain, so no other writer
166 * can insert/delete a socket with local_port == num
167 */
168static int udp_lib_lport_inuse2(struct net *net, __u16 num,
4243cdc2 169 struct udp_hslot *hslot2,
fe38d2a1 170 struct sock *sk)
30fff923
ED		 171{
172 struct sock *sk2;
ba418fa3 173 kuid_t uid = sock_i_uid(sk);
30fff923
ED		 174 int res = 0;
175
176 spin_lock(&hslot2->lock);
ca065d0c 177 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
9d4fb27d
JP		 178 if (net_eq(sock_net(sk2), net) &&
179 sk2 != sk &&
180 (udp_sk(sk2)->udp_port_hash == num) &&
181 (!sk2->sk_reuse || !sk->sk_reuse) &&
182 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
183 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
fe38d2a1 184 inet_rcv_saddr_equal(sk, sk2, true)) {
df560056
EG		 185 if (sk2->sk_reuseport && sk->sk_reuseport &&
186 !rcu_access_pointer(sk->sk_reuseport_cb) &&
187 uid_eq(uid, sock_i_uid(sk2))) {
188 res = 0;
189 } else {
190 res = 1;
191 }
30fff923
ED		 192 break;
193 }
4243cdc2 194 }
30fff923
ED		 195 spin_unlock(&hslot2->lock);
196 return res;
197}
198
fe38d2a1 199static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
e32ea7e7
CG		 200{
201 struct net *net = sock_net(sk);
e32ea7e7
CG		 202 kuid_t uid = sock_i_uid(sk);
203 struct sock *sk2;
204
ca065d0c 205 sk_for_each(sk2, &hslot->head) {
e32ea7e7
CG		 206 if (net_eq(sock_net(sk2), net) &&
207 sk2 != sk &&
208 sk2->sk_family == sk->sk_family &&
209 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
210 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
211 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
212 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
fe38d2a1 213 inet_rcv_saddr_equal(sk, sk2, false)) {
2dbb9b9e
MKL		 214 return reuseport_add_sock(sk, sk2,
215 inet_rcv_saddr_any(sk));
e32ea7e7
CG		 216 }
217 }
218
2dbb9b9e 219 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
e32ea7e7
CG		 220}
221
25030a7f 222/**
6ba5a3c5 223 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
25030a7f
GR		 224 *
225 * @sk: socket struct in question
226 * @snum: port number to look up
25985edc 227 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
30fff923 228 * with NULL address
25030a7f 229 */
6ba5a3c5 230int udp_lib_get_port(struct sock *sk, unsigned short snum,
30fff923 231 unsigned int hash2_nulladdr)
25030a7f 232{
512615b6 233 struct udp_hslot *hslot, *hslot2;
645ca708 234 struct udp_table *udptable = sk->sk_prot->h.udp_table;
25030a7f 235 int error = 1;
3b1e0a65 236 struct net *net = sock_net(sk);
1da177e4 237
32c1da70 238 if (!snum) {
9088c560 239 int low, high, remaining;
95c96174 240 unsigned int rand;
98322f22
ED		 241 unsigned short first, last;
242 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
32c1da70 243
0bbf87d8 244 inet_get_local_port_range(net, &low, &high);
a25de534 245 remaining = (high - low) + 1;
227b60f5 246
63862b5b 247 rand = prandom_u32();
8fc54f68 248 first = reciprocal_scale(rand, remaining) + low;
98322f22
ED		 249 /*
250 * force rand to be an odd multiple of UDP_HTABLE_SIZE
251 */
f86dcc5a 252 rand = (rand | 1) * (udptable->mask + 1);
5781b235
ED		 253 last = first + udptable->mask + 1;
254 do {
f86dcc5a 255 hslot = udp_hashslot(udptable, net, first);
98322f22 256 bitmap_zero(bitmap, PORTS_PER_CHAIN);
645ca708 257 spin_lock_bh(&hslot->lock);
98322f22 258 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
fe38d2a1 259 udptable->log);
98322f22
ED		 260
261 snum = first;
262 /*
263 * Iterate on all possible values of snum for this hash.
264 * Using steps of an odd multiple of UDP_HTABLE_SIZE
265 * give us randomization and full range coverage.
266 */
9088c560 267 do {
98322f22 268 if (low <= snum && snum <= high &&
e3826f1e 269 !test_bit(snum >> udptable->log, bitmap) &&
122ff243 270 !inet_is_local_reserved_port(net, snum))
98322f22
ED		 271 goto found;
272 snum += rand;
273 } while (snum != first);
274 spin_unlock_bh(&hslot->lock);
df560056 275 cond_resched();
5781b235 276 } while (++first != last);
98322f22 277 goto fail;
645ca708 278 } else {
f86dcc5a 279 hslot = udp_hashslot(udptable, net, snum);
645ca708 280 spin_lock_bh(&hslot->lock);
30fff923
ED		 281 if (hslot->count > 10) {
282 int exist;
283 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
284
285 slot2 &= udptable->mask;
286 hash2_nulladdr &= udptable->mask;
287
288 hslot2 = udp_hashslot2(udptable, slot2);
289 if (hslot->count < hslot2->count)
290 goto scan_primary_hash;
291
fe38d2a1 292 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
30fff923
ED		 293 if (!exist && (hash2_nulladdr != slot2)) {
294 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
295 exist = udp_lib_lport_inuse2(net, snum, hslot2,
fe38d2a1 296 sk);
30fff923
ED		 297 }
298 if (exist)
299 goto fail_unlock;
300 else
301 goto found;
302 }
303scan_primary_hash:
fe38d2a1 304 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
645ca708
ED		 305 goto fail_unlock;
306 }
98322f22 307found:
c720c7e8 308 inet_sk(sk)->inet_num = snum;
d4cada4a
ED		 309 udp_sk(sk)->udp_port_hash = snum;
310 udp_sk(sk)->udp_portaddr_hash ^= snum;
1da177e4 311 if (sk_unhashed(sk)) {
e32ea7e7 312 if (sk->sk_reuseport &&
fe38d2a1 313 udp_reuseport_add_sock(sk, hslot)) {
e32ea7e7
CG		 314 inet_sk(sk)->inet_num = 0;
315 udp_sk(sk)->udp_port_hash = 0;
316 udp_sk(sk)->udp_portaddr_hash ^= snum;
317 goto fail_unlock;
318 }
319
ca065d0c 320 sk_add_node_rcu(sk, &hslot->head);
fdcc8aa9 321 hslot->count++;
c29a0bc4 322 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
512615b6
ED		 323
324 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
325 spin_lock(&hslot2->lock);
d894ba18 326 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
1602f49b
DM		 327 sk->sk_family == AF_INET6)
328 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
329 &hslot2->head);
d894ba18 330 else
1602f49b
DM		 331 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
332 &hslot2->head);
512615b6
ED		 333 hslot2->count++;
334 spin_unlock(&hslot2->lock);
1da177e4 335 }
ca065d0c 336 sock_set_flag(sk, SOCK_RCU_FREE);
25030a7f 337 error = 0;
645ca708
ED		 338fail_unlock:
339 spin_unlock_bh(&hslot->lock);
1da177e4 340fail:
25030a7f
GR		 341 return error;
342}
c482c568 343EXPORT_SYMBOL(udp_lib_get_port);
25030a7f 344
6ba5a3c5 345int udp_v4_get_port(struct sock *sk, unsigned short snum)
db8dac20 346{
30fff923 347 unsigned int hash2_nulladdr =
f0b1e64c 348 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
30fff923 349 unsigned int hash2_partial =
f0b1e64c 350 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
30fff923 351
d4cada4a 352 /* precompute partial secondary hash */
30fff923 353 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
fe38d2a1 354 return udp_lib_get_port(sk, snum, hash2_nulladdr);
db8dac20
DM		 355}
356
d1e37288
SX		 357static int compute_score(struct sock *sk, struct net *net,
358 __be32 saddr, __be16 sport,
fb74c277 359 __be32 daddr, unsigned short hnum,
73545373 360 int dif, int sdif)
645ca708 361{
60c04aec
JP		 362 int score;
363 struct inet_sock *inet;
6da5b0f0 364 bool dev_match;
645ca708 365
60c04aec
JP		 366 if (!net_eq(sock_net(sk), net) ||
367 udp_sk(sk)->udp_port_hash != hnum ||
368 ipv6_only_sock(sk))
369 return -1;
645ca708 370
4cdeeee9
PO		 371 if (sk->sk_rcv_saddr != daddr)
372 return -1;
60c04aec 373
4cdeeee9 374 score = (sk->sk_family == PF_INET) ? 2 : 1;
60c04aec 375
4cdeeee9 376 inet = inet_sk(sk);
60c04aec
JP		 377 if (inet->inet_daddr) {
378 if (inet->inet_daddr != saddr)
379 return -1;
380 score += 4;
381 }
382
383 if (inet->inet_dport) {
384 if (inet->inet_dport != sport)
385 return -1;
386 score += 4;
387 }
388
6da5b0f0
MM		 389 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
390 dif, sdif);
391 if (!dev_match)
392 return -1;
393 score += 4;
fb74c277 394
7170a977 395 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
70da268b 396 score++;
645ca708
ED		 397 return score;
398}
399
6eada011
ED		 400static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
401 const __u16 lport, const __be32 faddr,
402 const __be16 fport)
65cd8033 403{
1bbdceef
HFS		 404 static u32 udp_ehash_secret __read_mostly;
405
406 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
407
65cd8033 408 return __inet_ehashfn(laddr, lport, faddr, fport,
1bbdceef 409 udp_ehash_secret + net_hash_mix(net));
65cd8033
HFS		 410}
411
7629c73a
JS		 412static inline struct sock *lookup_reuseport(struct net *net, struct sock *sk,
413 struct sk_buff *skb,
414 __be32 saddr, __be16 sport,
415 __be32 daddr, unsigned short hnum)
416{
417 struct sock *reuse_sk = NULL;
418 u32 hash;
419
420 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
421 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
422 reuse_sk = reuseport_select_sock(sk, hash, skb,
423 sizeof(struct udphdr));
424 /* Fall back to scoring if group has connections */
425 if (reuseport_has_conns(sk, false))
426 return NULL;
427 }
428 return reuse_sk;
429}
430
d1e37288 431/* called with rcu_read_lock() */
5051ebd2 432static struct sock *udp4_lib_lookup2(struct net *net,
fb74c277
DA		 433 __be32 saddr, __be16 sport,
434 __be32 daddr, unsigned int hnum,
73545373 435 int dif, int sdif,
fb74c277
DA		 436 struct udp_hslot *hslot2,
437 struct sk_buff *skb)
5051ebd2
ED		 438{
439 struct sock *sk, *result;
e94a62f5 440 int score, badness;
5051ebd2 441
5051ebd2 442 result = NULL;
ba418fa3 443 badness = 0;
ca065d0c 444 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
d1e37288 445 score = compute_score(sk, net, saddr, sport,
73545373 446 daddr, hnum, dif, sdif);
5051ebd2 447 if (score > badness) {
7629c73a
JS		 448 result = lookup_reuseport(net, sk, skb,
449 saddr, sport, daddr, hnum);
450 if (result)
451 return result;
452
ca065d0c
ED		 453 badness = score;
454 result = sk;
5051ebd2
ED		 455 }
456 }
5051ebd2
ED		 457 return result;
458}
459
72f7e944
JS		 460static inline struct sock *udp4_lookup_run_bpf(struct net *net,
461 struct udp_table *udptable,
462 struct sk_buff *skb,
463 __be32 saddr, __be16 sport,
464 __be32 daddr, u16 hnum)
465{
466 struct sock *sk, *reuse_sk;
467 bool no_reuseport;
468
469 if (udptable != &udp_table)
470 return NULL; /* only UDP is supported */
471
472 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP,
473 saddr, sport, daddr, hnum, &sk);
474 if (no_reuseport || IS_ERR_OR_NULL(sk))
475 return sk;
476
477 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
478 if (reuse_sk)
479 sk = reuse_sk;
480 return sk;
481}
482
db8dac20
DM		 483/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
484 * harder than this. -DaveM
485 */
fce82338 486struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
fb74c277
DA		 487 __be16 sport, __be32 daddr, __be16 dport, int dif,
488 int sdif, struct udp_table *udptable, struct sk_buff *skb)
db8dac20 489{
db8dac20 490 unsigned short hnum = ntohs(dport);
4cdeeee9
PO		 491 unsigned int hash2, slot2;
492 struct udp_hslot *hslot2;
72f7e944 493 struct sock *result, *sk;
645ca708 494
4cdeeee9
PO		 495 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
496 slot2 = hash2 & udptable->mask;
497 hslot2 = &udptable->hash2[slot2];
498
72f7e944 499 /* Lookup connected or non-wildcard socket */
4cdeeee9
PO		 500 result = udp4_lib_lookup2(net, saddr, sport,
501 daddr, hnum, dif, sdif,
73545373 502 hslot2, skb);
72f7e944
JS		 503 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
504 goto done;
505
506 /* Lookup redirect from BPF */
507 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
508 sk = udp4_lookup_run_bpf(net, udptable, skb,
509 saddr, sport, daddr, hnum);
510 if (sk) {
511 result = sk;
512 goto done;
513 }
db8dac20 514 }
72f7e944
JS		 515
516 /* Got non-wildcard socket or error on first lookup */
517 if (result)
518 goto done;
519
520 /* Lookup wildcard sockets */
521 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
522 slot2 = hash2 & udptable->mask;
523 hslot2 = &udptable->hash2[slot2];
524
525 result = udp4_lib_lookup2(net, saddr, sport,
526 htonl(INADDR_ANY), hnum, dif, sdif,
527 hslot2, skb);
528done:
88e235b8 529 if (IS_ERR(result))
4cdeeee9 530 return NULL;
db8dac20
DM		 531 return result;
532}
fce82338 533EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
db8dac20 534
607c4aaf
KK		 535static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
536 __be16 sport, __be16 dport,
645ca708 537 struct udp_table *udptable)
607c4aaf
KK		 538{
539 const struct iphdr *iph = ip_hdr(skb);
540
ed7cbbce 541 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
8afdd99a 542 iph->daddr, dport, inet_iif(skb),
fb74c277 543 inet_sdif(skb), udptable, skb);
607c4aaf
KK		 544}
545
63058308
TH		 546struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
547 __be16 sport, __be16 dport)
548{
257a525f
MKL		 549 const struct iphdr *iph = ip_hdr(skb);
550
551 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
552 iph->daddr, dport, inet_iif(skb),
553 inet_sdif(skb), &udp_table, NULL);
63058308
TH		 554}
555EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
556
ca065d0c
ED		 557/* Must be called under rcu_read_lock().
558 * Does increment socket refcount.
559 */
6e86000c 560#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
bcd41303
KK		 561struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
562 __be32 daddr, __be16 dport, int dif)
563{
ca065d0c
ED		 564 struct sock *sk;
565
566 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
fb74c277 567 dif, 0, &udp_table, NULL);
41c6d650 568 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
ca065d0c
ED		 569 sk = NULL;
570 return sk;
bcd41303
KK		 571}
572EXPORT_SYMBOL_GPL(udp4_lib_lookup);
ca065d0c 573#endif
bcd41303 574
421b3885
SB		 575static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
576 __be16 loc_port, __be32 loc_addr,
577 __be16 rmt_port, __be32 rmt_addr,
fb74c277 578 int dif, int sdif, unsigned short hnum)
421b3885
SB		 579{
580 struct inet_sock *inet = inet_sk(sk);
581
582 if (!net_eq(sock_net(sk), net) ||
583 udp_sk(sk)->udp_port_hash != hnum ||
584 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
585 (inet->inet_dport != rmt_port && inet->inet_dport) ||
586 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
587 ipv6_only_sock(sk) ||
82ba25c6 588 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
421b3885 589 return false;
60d9b031 590 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
421b3885
SB		 591 return false;
592 return true;
593}
594
a36e185e
SB		 595DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
596void udp_encap_enable(void)
597{
9c480601 598 static_branch_inc(&udp_encap_needed_key);
a36e185e
SB		 599}
600EXPORT_SYMBOL(udp_encap_enable);
601
e7cc0824
SB		 602/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
603 * through error handlers in encapsulations looking for a match.
604 */
605static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
606{
607 int i;
608
609 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
610 int (*handler)(struct sk_buff *skb, u32 info);
92b95364 611 const struct ip_tunnel_encap_ops *encap;
e7cc0824 612
92b95364
PA		 613 encap = rcu_dereference(iptun_encaps[i]);
614 if (!encap)
e7cc0824 615 continue;
92b95364 616 handler = encap->err_handler;
e7cc0824
SB		 617 if (handler && !handler(skb, info))
618 return 0;
619 }
620
621 return -ENOENT;
622}
623
a36e185e
SB		 624/* Try to match ICMP errors to UDP tunnels by looking up a socket without
625 * reversing source and destination port: this will match tunnels that force the
626 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
627 * lwtunnels might actually break this assumption by being configured with
628 * different destination ports on endpoints, in this case we won't be able to
629 * trace ICMP messages back to them.
630 *
e7cc0824
SB		 631 * If this doesn't match any socket, probe tunnels with arbitrary destination
632 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
633 * we've sent packets to won't necessarily match the local destination port.
634 *
a36e185e
SB		 635 * Then ask the tunnel implementation to match the error against a valid
636 * association.
637 *
e7cc0824
SB		 638 * Return an error if we can't find a match, the socket if we need further
639 * processing, zero otherwise.
a36e185e
SB		 640 */
641static struct sock *__udp4_lib_err_encap(struct net *net,
642 const struct iphdr *iph,
643 struct udphdr *uh,
644 struct udp_table *udptable,
e7cc0824 645 struct sk_buff *skb, u32 info)
a36e185e 646{
a36e185e 647 int network_offset, transport_offset;
a36e185e
SB		 648 struct sock *sk;
649
a36e185e
SB		 650 network_offset = skb_network_offset(skb);
651 transport_offset = skb_transport_offset(skb);
652
653 /* Network header needs to point to the outer IPv4 header inside ICMP */
654 skb_reset_network_header(skb);
655
656 /* Transport header needs to point to the UDP header */
657 skb_set_transport_header(skb, iph->ihl << 2);
658
e7cc0824
SB		 659 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
660 iph->saddr, uh->dest, skb->dev->ifindex, 0,
661 udptable, NULL);
662 if (sk) {
663 int (*lookup)(struct sock *sk, struct sk_buff *skb);
664 struct udp_sock *up = udp_sk(sk);
665
666 lookup = READ_ONCE(up->encap_err_lookup);
667 if (!lookup || lookup(sk, skb))
668 sk = NULL;
669 }
670
671 if (!sk)
672 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
a36e185e
SB		 673
674 skb_set_transport_header(skb, transport_offset);
675 skb_set_network_header(skb, network_offset);
676
677 return sk;
678}
679
db8dac20
DM		 680/*
681 * This routine is called by the ICMP module when it gets some
682 * sort of error condition. If err < 0 then the socket should
683 * be closed and the error returned to the user. If err > 0
684 * it's just the icmp type << 8 | icmp code.
685 * Header points to the ip header of the error packet. We move
686 * on past this. Then (as it used to claim before adjustment)
687 * header points to the first 8 bytes of the udp header. We need
688 * to find the appropriate port.
689 */
690
32bbd879 691int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
db8dac20
DM		 692{
693 struct inet_sock *inet;
b71d1d42 694 const struct iphdr *iph = (const struct iphdr *)skb->data;
c482c568 695 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
db8dac20
DM		 696 const int type = icmp_hdr(skb)->type;
697 const int code = icmp_hdr(skb)->code;
a36e185e 698 bool tunnel = false;
db8dac20
DM		 699 struct sock *sk;
700 int harderr;
701 int err;
fd54d716 702 struct net *net = dev_net(skb->dev);
db8dac20 703
fd54d716 704 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
f64bf6b8
MM		 705 iph->saddr, uh->source, skb->dev->ifindex,
706 inet_sdif(skb), udptable, NULL);
51456b29 707 if (!sk) {
a36e185e 708 /* No socket for error: try tunnels before discarding */
e7cc0824
SB		 709 sk = ERR_PTR(-ENOENT);
710 if (static_branch_unlikely(&udp_encap_needed_key)) {
711 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
712 info);
713 if (!sk)
714 return 0;
715 }
a36e185e 716
e7cc0824 717 if (IS_ERR(sk)) {
a36e185e 718 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
e7cc0824 719 return PTR_ERR(sk);
a36e185e 720 }
e7cc0824 721
a36e185e 722 tunnel = true;
db8dac20
DM		 723 }
724
725 err = 0;
726 harderr = 0;
727 inet = inet_sk(sk);
728
729 switch (type) {
730 default:
731 case ICMP_TIME_EXCEEDED:
732 err = EHOSTUNREACH;
733 break;
734 case ICMP_SOURCE_QUENCH:
735 goto out;
736 case ICMP_PARAMETERPROB:
737 err = EPROTO;
738 harderr = 1;
739 break;
740 case ICMP_DEST_UNREACH:
741 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
36393395 742 ipv4_sk_update_pmtu(skb, sk, info);
db8dac20
DM		 743 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
744 err = EMSGSIZE;
745 harderr = 1;
746 break;
747 }
748 goto out;
749 }
750 err = EHOSTUNREACH;
751 if (code <= NR_ICMP_UNREACH) {
752 harderr = icmp_err_convert[code].fatal;
753 err = icmp_err_convert[code].errno;
754 }
755 break;
55be7a9c
DM		 756 case ICMP_REDIRECT:
757 ipv4_sk_redirect(skb, sk);
1a462d18 758 goto out;
db8dac20
DM		 759 }
760
761 /*
762 * RFC1122: OK. Passes ICMP errors back to application, as per
763 * 4.1.3.3.
764 */
a36e185e
SB		 765 if (tunnel) {
766 /* ...not for tunnels though: we don't have a sending socket */
767 goto out;
768 }
db8dac20
DM		 769 if (!inet->recverr) {
770 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
771 goto out;
b1faf566 772 } else
c482c568 773 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
b1faf566 774
db8dac20
DM		 775 sk->sk_err = err;
776 sk->sk_error_report(sk);
777out:
32bbd879 778 return 0;
db8dac20
DM		 779}
780
32bbd879 781int udp_err(struct sk_buff *skb, u32 info)
db8dac20 782{
32bbd879 783 return __udp4_lib_err(skb, info, &udp_table);
db8dac20
DM		 784}
785
786/*
787 * Throw away all pending data and cancel the corking. Socket is locked.
788 */
36d926b9 789void udp_flush_pending_frames(struct sock *sk)
db8dac20
DM		 790{
791 struct udp_sock *up = udp_sk(sk);
792
793 if (up->pending) {
794 up->len = 0;
795 up->pending = 0;
796 ip_flush_pending_frames(sk);
797 }
798}
36d926b9 799EXPORT_SYMBOL(udp_flush_pending_frames);
db8dac20
DM		 800
801/**
f6b9664f 802 * udp4_hwcsum - handle outgoing HW checksumming
db8dac20
DM		 803 * @skb: sk_buff containing the filled-in UDP header
804 * (checksum field must be zeroed out)
f6b9664f
HX		 805 * @src: source IP address
806 * @dst: destination IP address
db8dac20 807 */
c26bf4a5 808void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
db8dac20 809{
db8dac20 810 struct udphdr *uh = udp_hdr(skb);
f6b9664f
HX		 811 int offset = skb_transport_offset(skb);
812 int len = skb->len - offset;
813 int hlen = len;
db8dac20
DM		 814 __wsum csum = 0;
815
ebbe495f 816 if (!skb_has_frag_list(skb)) {
db8dac20
DM		 817 /*
818 * Only one fragment on the socket.
819 */
820 skb->csum_start = skb_transport_header(skb) - skb->head;
821 skb->csum_offset = offsetof(struct udphdr, check);
f6b9664f
HX		 822 uh->check = ~csum_tcpudp_magic(src, dst, len,
823 IPPROTO_UDP, 0);
db8dac20 824 } else {
ebbe495f
WC		 825 struct sk_buff *frags;
826
db8dac20
DM		 827 /*
828 * HW-checksum won't work as there are two or more
829 * fragments on the socket so that all csums of sk_buffs
830 * should be together
831 */
ebbe495f 832 skb_walk_frags(skb, frags) {
f6b9664f
HX		 833 csum = csum_add(csum, frags->csum);
834 hlen -= frags->len;
ebbe495f 835 }
db8dac20 836
f6b9664f 837 csum = skb_checksum(skb, offset, hlen, csum);
db8dac20
DM		 838 skb->ip_summed = CHECKSUM_NONE;
839
db8dac20
DM		 840 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
841 if (uh->check == 0)
842 uh->check = CSUM_MANGLED_0;
843 }
844}
c26bf4a5 845EXPORT_SYMBOL_GPL(udp4_hwcsum);
db8dac20 846
af5fcba7
TH		 847/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
848 * for the simple case like when setting the checksum for a UDP tunnel.
849 */
850void udp_set_csum(bool nocheck, struct sk_buff *skb,
851 __be32 saddr, __be32 daddr, int len)
852{
853 struct udphdr *uh = udp_hdr(skb);
854
179bc67f 855 if (nocheck) {
af5fcba7 856 uh->check = 0;
179bc67f 857 } else if (skb_is_gso(skb)) {
af5fcba7 858 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
179bc67f
EC		 859 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
860 uh->check = 0;
861 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
862 if (uh->check == 0)
863 uh->check = CSUM_MANGLED_0;
d75f1306 864 } else {
af5fcba7
TH		 865 skb->ip_summed = CHECKSUM_PARTIAL;
866 skb->csum_start = skb_transport_header(skb) - skb->head;
867 skb->csum_offset = offsetof(struct udphdr, check);
868 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
af5fcba7
TH		 869 }
870}
871EXPORT_SYMBOL(udp_set_csum);
872
bec1f6f6
WB		 873static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
874 struct inet_cork *cork)
db8dac20 875{
f6b9664f 876 struct sock *sk = skb->sk;
db8dac20 877 struct inet_sock *inet = inet_sk(sk);
db8dac20
DM		 878 struct udphdr *uh;
879 int err = 0;
880 int is_udplite = IS_UDPLITE(sk);
f6b9664f
HX		 881 int offset = skb_transport_offset(skb);
882 int len = skb->len - offset;
4094871d 883 int datalen = len - sizeof(*uh);
db8dac20
DM		 884 __wsum csum = 0;
885
db8dac20
DM		 886 /*
887 * Create a UDP header
888 */
889 uh = udp_hdr(skb);
f6b9664f 890 uh->source = inet->inet_sport;
79ab0531 891 uh->dest = fl4->fl4_dport;
f6b9664f 892 uh->len = htons(len);
db8dac20
DM		 893 uh->check = 0;
894
bec1f6f6
WB		 895 if (cork->gso_size) {
896 const int hlen = skb_network_header_len(skb) +
897 sizeof(struct udphdr);
898
0f149c9f
WB		 899 if (hlen + cork->gso_size > cork->fragsize) {
900 kfree_skb(skb);
bec1f6f6 901 return -EINVAL;
0f149c9f
WB		 902 }
903 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
904 kfree_skb(skb);
bec1f6f6 905 return -EINVAL;
0f149c9f
WB		 906 }
907 if (sk->sk_no_check_tx) {
908 kfree_skb(skb);
a8c744a8 909 return -EINVAL;
0f149c9f 910 }
ff06342c 911 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
0f149c9f
WB		 912 dst_xfrm(skb_dst(skb))) {
913 kfree_skb(skb);
bec1f6f6 914 return -EIO;
0f149c9f 915 }
bec1f6f6 916
4094871d
JH		 917 if (datalen > cork->gso_size) {
918 skb_shinfo(skb)->gso_size = cork->gso_size;
919 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
920 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
921 cork->gso_size);
922 }
a8c744a8 923 goto csum_partial;
bec1f6f6
WB		 924 }
925
db8dac20 926 if (is_udplite) /* UDP-Lite */
f6b9664f 927 csum = udplite_csum(skb);
db8dac20 928
ab2fb7e3 929 else if (sk->sk_no_check_tx) { /* UDP csum off */
db8dac20
DM		 930
931 skb->ip_summed = CHECKSUM_NONE;
932 goto send;
933
934 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
a8c744a8 935csum_partial:
db8dac20 936
79ab0531 937 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
db8dac20
DM		 938 goto send;
939
f6b9664f
HX		 940 } else
941 csum = udp_csum(skb);
db8dac20
DM		 942
943 /* add protocol-dependent pseudo-header */
79ab0531 944 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
c482c568 945 sk->sk_protocol, csum);
db8dac20
DM		 946 if (uh->check == 0)
947 uh->check = CSUM_MANGLED_0;
948
949send:
b5ec8eea 950 err = ip_send_skb(sock_net(sk), skb);
6ce9e7b5
ED		 951 if (err) {
952 if (err == -ENOBUFS && !inet->recverr) {
6aef70a8
ED		 953 UDP_INC_STATS(sock_net(sk),
954 UDP_MIB_SNDBUFERRORS, is_udplite);
6ce9e7b5
ED		 955 err = 0;
956 }
957 } else
6aef70a8
ED		 958 UDP_INC_STATS(sock_net(sk),
959 UDP_MIB_OUTDATAGRAMS, is_udplite);
f6b9664f
HX		 960 return err;
961}
962
963/*
964 * Push out all pending data as one UDP datagram. Socket is locked.
965 */
8822b64a 966int udp_push_pending_frames(struct sock *sk)
f6b9664f
HX		 967{
968 struct udp_sock *up = udp_sk(sk);
969 struct inet_sock *inet = inet_sk(sk);
b6f21b26 970 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
f6b9664f
HX		 971 struct sk_buff *skb;
972 int err = 0;
973
77968b78 974 skb = ip_finish_skb(sk, fl4);
f6b9664f
HX		 975 if (!skb)
976 goto out;
977
bec1f6f6 978 err = udp_send_skb(skb, fl4, &inet->cork.base);
f6b9664f 979
db8dac20
DM		 980out:
981 up->len = 0;
982 up->pending = 0;
db8dac20
DM		 983 return err;
984}
8822b64a 985EXPORT_SYMBOL(udp_push_pending_frames);
db8dac20 986
2e8de857
WB		 987static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
988{
989 switch (cmsg->cmsg_type) {
990 case UDP_SEGMENT:
991 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
992 return -EINVAL;
993 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
994 return 0;
995 default:
996 return -EINVAL;
997 }
998}
999
1000int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1001{
1002 struct cmsghdr *cmsg;
1003 bool need_ip = false;
1004 int err;
1005
1006 for_each_cmsghdr(cmsg, msg) {
1007 if (!CMSG_OK(msg, cmsg))
1008 return -EINVAL;
1009
1010 if (cmsg->cmsg_level != SOL_UDP) {
1011 need_ip = true;
1012 continue;
1013 }
1014
1015 err = __udp_cmsg_send(cmsg, gso_size);
1016 if (err)
1017 return err;
1018 }
1019
1020 return need_ip;
1021}
1022EXPORT_SYMBOL_GPL(udp_cmsg_send);
1023
1b784140 1024int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
db8dac20
DM		 1025{
1026 struct inet_sock *inet = inet_sk(sk);
1027 struct udp_sock *up = udp_sk(sk);
1cedee13 1028 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
e474995f 1029 struct flowi4 fl4_stack;
b6f21b26 1030 struct flowi4 *fl4;
db8dac20
DM		 1031 int ulen = len;
1032 struct ipcm_cookie ipc;
1033 struct rtable *rt = NULL;
1034 int free = 0;
1035 int connected = 0;
1036 __be32 daddr, faddr, saddr;
1037 __be16 dport;
1038 u8 tos;
1039 int err, is_udplite = IS_UDPLITE(sk);
1040 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1041 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
903ab86d 1042 struct sk_buff *skb;
f6d8bd05 1043 struct ip_options_data opt_copy;
db8dac20
DM		 1044
1045 if (len > 0xFFFF)
1046 return -EMSGSIZE;
1047
1048 /*
1049 * Check the flags.
1050 */
1051
c482c568 1052 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
db8dac20
DM		 1053 return -EOPNOTSUPP;
1054
903ab86d
HX		 1055 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1056
f5fca608 1057 fl4 = &inet->cork.fl.u.ip4;
db8dac20
DM		 1058 if (up->pending) {
1059 /*
1060 * There are pending frames.
1061 * The socket lock must be held while it's corked.
1062 */
1063 lock_sock(sk);
1064 if (likely(up->pending)) {
1065 if (unlikely(up->pending != AF_INET)) {
1066 release_sock(sk);
1067 return -EINVAL;
1068 }
1069 goto do_append_data;
1070 }
1071 release_sock(sk);
1072 }
1073 ulen += sizeof(struct udphdr);
1074
1075 /*
1076 * Get and verify the address.
1077 */
1cedee13 1078 if (usin) {
db8dac20
DM		 1079 if (msg->msg_namelen < sizeof(*usin))
1080 return -EINVAL;
1081 if (usin->sin_family != AF_INET) {
1082 if (usin->sin_family != AF_UNSPEC)
1083 return -EAFNOSUPPORT;
1084 }
1085
1086 daddr = usin->sin_addr.s_addr;
1087 dport = usin->sin_port;
1088 if (dport == 0)
1089 return -EINVAL;
1090 } else {
1091 if (sk->sk_state != TCP_ESTABLISHED)
1092 return -EDESTADDRREQ;
c720c7e8
ED		 1093 daddr = inet->inet_daddr;
1094 dport = inet->inet_dport;
db8dac20
DM		 1095 /* Open fast path for connected socket.
1096 Route will not be used, if at least one option is set.
1097 */
1098 connected = 1;
1099 }
db8dac20 1100
35178206 1101 ipcm_init_sk(&ipc, inet);
bec1f6f6 1102 ipc.gso_size = up->gso_size;
bf84a010 1103
db8dac20 1104 if (msg->msg_controllen) {
2e8de857
WB		 1105 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1106 if (err > 0)
1107 err = ip_cmsg_send(sk, msg, &ipc,
1108 sk->sk_family == AF_INET6);
1109 if (unlikely(err < 0)) {
91948309 1110 kfree(ipc.opt);
db8dac20 1111 return err;
91948309 1112 }
db8dac20
DM		 1113 if (ipc.opt)
1114 free = 1;
1115 connected = 0;
1116 }
f6d8bd05
ED		 1117 if (!ipc.opt) {
1118 struct ip_options_rcu *inet_opt;
1119
1120 rcu_read_lock();
1121 inet_opt = rcu_dereference(inet->inet_opt);
1122 if (inet_opt) {
1123 memcpy(&opt_copy, inet_opt,
1124 sizeof(*inet_opt) + inet_opt->opt.optlen);
1125 ipc.opt = &opt_copy.opt;
1126 }
1127 rcu_read_unlock();
1128 }
db8dac20 1129
1cedee13
AI		 1130 if (cgroup_bpf_enabled && !connected) {
1131 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1132 (struct sockaddr *)usin, &ipc.addr);
1133 if (err)
1134 goto out_free;
1135 if (usin) {
1136 if (usin->sin_port == 0) {
1137 /* BPF program set invalid port. Reject it. */
1138 err = -EINVAL;
1139 goto out_free;
1140 }
1141 daddr = usin->sin_addr.s_addr;
1142 dport = usin->sin_port;
1143 }
1144 }
1145
db8dac20
DM		 1146 saddr = ipc.addr;
1147 ipc.addr = faddr = daddr;
1148
f6d8bd05 1149 if (ipc.opt && ipc.opt->opt.srr) {
1b97013b
AI		 1150 if (!daddr) {
1151 err = -EINVAL;
1152 goto out_free;
1153 }
f6d8bd05 1154 faddr = ipc.opt->opt.faddr;
db8dac20
DM		 1155 connected = 0;
1156 }
aa661581 1157 tos = get_rttos(&ipc, inet);
db8dac20
DM		 1158 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1159 (msg->msg_flags & MSG_DONTROUTE) ||
f6d8bd05 1160 (ipc.opt && ipc.opt->opt.is_strictroute)) {
db8dac20
DM		 1161 tos |= RTO_ONLINK;
1162 connected = 0;
1163 }
1164
1165 if (ipv4_is_multicast(daddr)) {
854da991 1166 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
db8dac20
DM		 1167 ipc.oif = inet->mc_index;
1168 if (!saddr)
1169 saddr = inet->mc_addr;
1170 connected = 0;
9515a2e0 1171 } else if (!ipc.oif) {
76e21053 1172 ipc.oif = inet->uc_index;
9515a2e0
DA		 1173 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1174 /* oif is set, packet is to local broadcast and
1175 * and uc_index is set. oif is most likely set
1176 * by sk_bound_dev_if. If uc_index != oif check if the
1177 * oif is an L3 master and uc_index is an L3 slave.
1178 * If so, we want to allow the send using the uc_index.
1179 */
1180 if (ipc.oif != inet->uc_index &&
1181 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1182 inet->uc_index)) {
1183 ipc.oif = inet->uc_index;
1184 }
1185 }
db8dac20
DM		 1186
1187 if (connected)
c482c568 1188 rt = (struct rtable *)sk_dst_check(sk, 0);
db8dac20 1189
51456b29 1190 if (!rt) {
84a3aa00 1191 struct net *net = sock_net(sk);
9a24abfa 1192 __u8 flow_flags = inet_sk_flowi_flags(sk);
84a3aa00 1193
e474995f 1194 fl4 = &fl4_stack;
9a24abfa 1195
c6af0c22 1196 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
c0951cbc 1197 RT_SCOPE_UNIVERSE, sk->sk_protocol,
9a24abfa 1198 flow_flags,
e2d118a1
LC		 1199 faddr, saddr, dport, inet->inet_sport,
1200 sk->sk_uid);
c0951cbc 1201
e474995f
DM		 1202 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1203 rt = ip_route_output_flow(net, fl4, sk);
b23dd4fe
DM		 1204 if (IS_ERR(rt)) {
1205 err = PTR_ERR(rt);
06dc94b1 1206 rt = NULL;
db8dac20 1207 if (err == -ENETUNREACH)
f1d8cba6 1208 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
db8dac20
DM		 1209 goto out;
1210 }
1211
1212 err = -EACCES;
1213 if ((rt->rt_flags & RTCF_BROADCAST) &&
1214 !sock_flag(sk, SOCK_BROADCAST))
1215 goto out;
1216 if (connected)
d8d1f30b 1217 sk_dst_set(sk, dst_clone(&rt->dst));
db8dac20
DM		 1218 }
1219
1220 if (msg->msg_flags&MSG_CONFIRM)
1221 goto do_confirm;
1222back_from_confirm:
1223
e474995f 1224 saddr = fl4->saddr;
db8dac20 1225 if (!ipc.addr)
e474995f 1226 daddr = ipc.addr = fl4->daddr;
db8dac20 1227
903ab86d
HX		 1228 /* Lockless fast path for the non-corking case. */
1229 if (!corkreq) {
1cd7884d
WB		 1230 struct inet_cork cork;
1231
f69e6d13 1232 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
903ab86d 1233 sizeof(struct udphdr), &ipc, &rt,
1cd7884d 1234 &cork, msg->msg_flags);
903ab86d 1235 err = PTR_ERR(skb);
50c3a487 1236 if (!IS_ERR_OR_NULL(skb))
bec1f6f6 1237 err = udp_send_skb(skb, fl4, &cork);
903ab86d
HX		 1238 goto out;
1239 }
1240
db8dac20
DM		 1241 lock_sock(sk);
1242 if (unlikely(up->pending)) {
1243 /* The socket is already corked while preparing it. */
1244 /* ... which is an evident application bug. --ANK */
1245 release_sock(sk);
1246
197df02c 1247 net_dbg_ratelimited("socket already corked\n");
db8dac20
DM		 1248 err = -EINVAL;
1249 goto out;
1250 }
1251 /*
1252 * Now cork the socket to pend data.
1253 */
b6f21b26
DM		 1254 fl4 = &inet->cork.fl.u.ip4;
1255 fl4->daddr = daddr;
1256 fl4->saddr = saddr;
9cce96df
DM		 1257 fl4->fl4_dport = dport;
1258 fl4->fl4_sport = inet->inet_sport;
db8dac20
DM		 1259 up->pending = AF_INET;
1260
1261do_append_data:
1262 up->len += ulen;
f69e6d13 1263 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
f5fca608
DM		 1264 sizeof(struct udphdr), &ipc, &rt,
1265 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
db8dac20
DM		 1266 if (err)
1267 udp_flush_pending_frames(sk);
1268 else if (!corkreq)
1269 err = udp_push_pending_frames(sk);
1270 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1271 up->pending = 0;
1272 release_sock(sk);
1273
1274out:
1275 ip_rt_put(rt);
1b97013b 1276out_free:
db8dac20
DM		 1277 if (free)
1278 kfree(ipc.opt);
1279 if (!err)
1280 return len;
1281 /*
1282 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1283 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1284 * we don't have a good statistic (IpOutDiscards but it can be too many
1285 * things). We could add another new stat but at least for now that
1286 * seems like overkill.
1287 */
1288 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
6aef70a8
ED		 1289 UDP_INC_STATS(sock_net(sk),
1290 UDP_MIB_SNDBUFERRORS, is_udplite);
db8dac20
DM		 1291 }
1292 return err;
1293
1294do_confirm:
0dec879f
JA		 1295 if (msg->msg_flags & MSG_PROBE)
1296 dst_confirm_neigh(&rt->dst, &fl4->daddr);
db8dac20
DM		 1297 if (!(msg->msg_flags&MSG_PROBE) || len)
1298 goto back_from_confirm;
1299 err = 0;
1300 goto out;
1301}
c482c568 1302EXPORT_SYMBOL(udp_sendmsg);
db8dac20
DM		 1303
1304int udp_sendpage(struct sock *sk, struct page *page, int offset,
1305 size_t size, int flags)
1306{
f5fca608 1307 struct inet_sock *inet = inet_sk(sk);
db8dac20
DM		 1308 struct udp_sock *up = udp_sk(sk);
1309 int ret;
1310
d3f7d56a
SL		 1311 if (flags & MSG_SENDPAGE_NOTLAST)
1312 flags |= MSG_MORE;
1313
db8dac20
DM		 1314 if (!up->pending) {
1315 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1316
1317 /* Call udp_sendmsg to specify destination address which
1318 * sendpage interface can't pass.
1319 * This will succeed only when the socket is connected.
1320 */
1b784140 1321 ret = udp_sendmsg(sk, &msg, 0);
db8dac20
DM		 1322 if (ret < 0)
1323 return ret;
1324 }
1325
1326 lock_sock(sk);
1327
1328 if (unlikely(!up->pending)) {
1329 release_sock(sk);
1330
197df02c 1331 net_dbg_ratelimited("cork failed\n");
db8dac20
DM		 1332 return -EINVAL;
1333 }
1334
f5fca608
DM		 1335 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1336 page, offset, size, flags);
db8dac20
DM		 1337 if (ret == -EOPNOTSUPP) {
1338 release_sock(sk);
1339 return sock_no_sendpage(sk->sk_socket, page, offset,
1340 size, flags);
1341 }
1342 if (ret < 0) {
1343 udp_flush_pending_frames(sk);
1344 goto out;
1345 }
1346
1347 up->len += size;
1348 if (!(up->corkflag || (flags&MSG_MORE)))
1349 ret = udp_push_pending_frames(sk);
1350 if (!ret)
1351 ret = size;
1352out:
1353 release_sock(sk);
1354 return ret;
1355}
1356
dce4551c
PA		 1357#define UDP_SKB_IS_STATELESS 0x80000000
1358
677bf08c
FW		 1359/* all head states (dst, sk, nf conntrack) except skb extensions are
1360 * cleared by udp_rcv().
1361 *
1362 * We need to preserve secpath, if present, to eventually process
1363 * IP_CMSG_PASSSEC at recvmsg() time.
1364 *
1365 * Other extensions can be cleared.
1366 */
1367static bool udp_try_make_stateless(struct sk_buff *skb)
1368{
1369 if (!skb_has_extensions(skb))
1370 return true;
1371
1372 if (!secpath_exists(skb)) {
1373 skb_ext_reset(skb);
1374 return true;
1375 }
1376
1377 return false;
1378}
1379
b65ac446
PA		 1380static void udp_set_dev_scratch(struct sk_buff *skb)
1381{
dce4551c 1382 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
b65ac446
PA		 1383
1384 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
dce4551c
PA		 1385 scratch->_tsize_state = skb->truesize;
1386#if BITS_PER_LONG == 64
b65ac446
PA		 1387 scratch->len = skb->len;
1388 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1389 scratch->is_linear = !skb_is_nonlinear(skb);
dce4551c 1390#endif
677bf08c 1391 if (udp_try_make_stateless(skb))
dce4551c 1392 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
b65ac446
PA		 1393}
1394
a793183c
ED		 1395static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1396{
1397 /* We come here after udp_lib_checksum_complete() returned 0.
1398 * This means that __skb_checksum_complete() might have
1399 * set skb->csum_valid to 1.
1400 * On 64bit platforms, we can set csum_unnecessary
1401 * to true, but only if the skb is not shared.
1402 */
1403#if BITS_PER_LONG == 64
1404 if (!skb_shared(skb))
1405 udp_skb_scratch(skb)->csum_unnecessary = true;
1406#endif
1407}
1408
b65ac446
PA		 1409static int udp_skb_truesize(struct sk_buff *skb)
1410{
dce4551c 1411 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
b65ac446
PA		 1412}
1413
dce4551c 1414static bool udp_skb_has_head_state(struct sk_buff *skb)
b65ac446 1415{
dce4551c 1416 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
b65ac446 1417}
b65ac446 1418
7c13f97f 1419/* fully reclaim rmem/fwd memory allocated for skb */
6dfb4367
PA		 1420static void udp_rmem_release(struct sock *sk, int size, int partial,
1421 bool rx_queue_lock_held)
f970bd9e 1422{
6b229cf7 1423 struct udp_sock *up = udp_sk(sk);
2276f58a 1424 struct sk_buff_head *sk_queue;
f970bd9e
PA		 1425 int amt;
1426
6b229cf7
ED		 1427 if (likely(partial)) {
1428 up->forward_deficit += size;
1429 size = up->forward_deficit;
d39ca259
PA		 1430 if (size < (sk->sk_rcvbuf >> 2) &&
1431 !skb_queue_empty(&up->reader_queue))
6b229cf7
ED		 1432 return;
1433 } else {
1434 size += up->forward_deficit;
1435 }
1436 up->forward_deficit = 0;
1437
6dfb4367
PA		 1438 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1439 * if the called don't held it already
1440 */
2276f58a 1441 sk_queue = &sk->sk_receive_queue;
6dfb4367
PA		 1442 if (!rx_queue_lock_held)
1443 spin_lock(&sk_queue->lock);
1444
2276f58a 1445
f970bd9e
PA		 1446 sk->sk_forward_alloc += size;
1447 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1448 sk->sk_forward_alloc -= amt;
f970bd9e
PA		 1449
1450 if (amt)
1451 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
02ab0d13
ED		 1452
1453 atomic_sub(size, &sk->sk_rmem_alloc);
2276f58a
PA		 1454
1455 /* this can save us from acquiring the rx queue lock on next receive */
1456 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1457
6dfb4367
PA		 1458 if (!rx_queue_lock_held)
1459 spin_unlock(&sk_queue->lock);
f970bd9e
PA		 1460}
1461
2276f58a 1462/* Note: called with reader_queue.lock held.
c84d9490
ED		 1463 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1464 * This avoids a cache line miss while receive_queue lock is held.
1465 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1466 */
7c13f97f 1467void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
f970bd9e 1468{
b65ac446
PA		 1469 prefetch(&skb->data);
1470 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
f970bd9e 1471}
7c13f97f 1472EXPORT_SYMBOL(udp_skb_destructor);
f970bd9e 1473
6dfb4367 1474/* as above, but the caller held the rx queue lock, too */
64f5102d 1475static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
6dfb4367 1476{
b65ac446
PA		 1477 prefetch(&skb->data);
1478 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
6dfb4367
PA		 1479}
1480
4b272750
ED		 1481/* Idea of busylocks is to let producers grab an extra spinlock
1482 * to relieve pressure on the receive_queue spinlock shared by consumer.
1483 * Under flood, this means that only one producer can be in line
1484 * trying to acquire the receive_queue spinlock.
1485 * These busylock can be allocated on a per cpu manner, instead of a
1486 * per socket one (that would consume a cache line per socket)
1487 */
1488static int udp_busylocks_log __read_mostly;
1489static spinlock_t *udp_busylocks __read_mostly;
1490
1491static spinlock_t *busylock_acquire(void *ptr)
1492{
1493 spinlock_t *busy;
1494
1495 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1496 spin_lock(busy);
1497 return busy;
1498}
1499
1500static void busylock_release(spinlock_t *busy)
1501{
1502 if (busy)
1503 spin_unlock(busy);
1504}
1505
f970bd9e
PA		 1506int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1507{
1508 struct sk_buff_head *list = &sk->sk_receive_queue;
1509 int rmem, delta, amt, err = -ENOMEM;
4b272750 1510 spinlock_t *busy = NULL;
c8c8b127 1511 int size;
f970bd9e
PA		 1512
1513 /* try to avoid the costly atomic add/sub pair when the receive
1514 * queue is full; always allow at least a packet
1515 */
1516 rmem = atomic_read(&sk->sk_rmem_alloc);
363dc73a 1517 if (rmem > sk->sk_rcvbuf)
f970bd9e
PA		 1518 goto drop;
1519
c8c8b127
ED		 1520 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1521 * having linear skbs :
1522 * - Reduce memory overhead and thus increase receive queue capacity
1523 * - Less cache line misses at copyout() time
1524 * - Less work at consume_skb() (less alien page frag freeing)
1525 */
4b272750 1526 if (rmem > (sk->sk_rcvbuf >> 1)) {
c8c8b127 1527 skb_condense(skb);
4b272750
ED		 1528
1529 busy = busylock_acquire(sk);
1530 }
c8c8b127 1531 size = skb->truesize;
b65ac446 1532 udp_set_dev_scratch(skb);
c8c8b127 1533
f970bd9e
PA		 1534 /* we drop only if the receive buf is full and the receive
1535 * queue contains some other skb
1536 */
1537 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
feed8a4f 1538 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
f970bd9e
PA		 1539 goto uncharge_drop;
1540
1541 spin_lock(&list->lock);
1542 if (size >= sk->sk_forward_alloc) {
1543 amt = sk_mem_pages(size);
1544 delta = amt << SK_MEM_QUANTUM_SHIFT;
1545 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1546 err = -ENOBUFS;
1547 spin_unlock(&list->lock);
1548 goto uncharge_drop;
1549 }
1550
1551 sk->sk_forward_alloc += delta;
1552 }
1553
1554 sk->sk_forward_alloc -= size;
1555
7c13f97f
PA		 1556 /* no need to setup a destructor, we will explicitly release the
1557 * forward allocated memory on dequeue
1558 */
f970bd9e
PA		 1559 sock_skb_set_dropcount(sk, skb);
1560
1561 __skb_queue_tail(list, skb);
1562 spin_unlock(&list->lock);
1563
1564 if (!sock_flag(sk, SOCK_DEAD))
1565 sk->sk_data_ready(sk);
1566
4b272750 1567 busylock_release(busy);
f970bd9e
PA		 1568 return 0;
1569
1570uncharge_drop:
1571 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1572
1573drop:
1574 atomic_inc(&sk->sk_drops);
4b272750 1575 busylock_release(busy);
f970bd9e
PA		 1576 return err;
1577}
1578EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1579
c915fe13 1580void udp_destruct_sock(struct sock *sk)
f970bd9e
PA		 1581{
1582 /* reclaim completely the forward allocated memory */
2276f58a 1583 struct udp_sock *up = udp_sk(sk);
7c13f97f
PA		 1584 unsigned int total = 0;
1585 struct sk_buff *skb;
1586
2276f58a
PA		 1587 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1588 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
7c13f97f
PA		 1589 total += skb->truesize;
1590 kfree_skb(skb);
1591 }
6dfb4367 1592 udp_rmem_release(sk, total, 0, true);
7c13f97f 1593
f970bd9e
PA		 1594 inet_sock_destruct(sk);
1595}
c915fe13 1596EXPORT_SYMBOL_GPL(udp_destruct_sock);
f970bd9e
PA		 1597
1598int udp_init_sock(struct sock *sk)
1599{
2276f58a 1600 skb_queue_head_init(&udp_sk(sk)->reader_queue);
f970bd9e
PA		 1601 sk->sk_destruct = udp_destruct_sock;
1602 return 0;
1603}
1604EXPORT_SYMBOL_GPL(udp_init_sock);
1605
1606void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1607{
1608 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1609 bool slow = lock_sock_fast(sk);
1610
1611 sk_peek_offset_bwd(sk, len);
1612 unlock_sock_fast(sk, slow);
1613 }
0a463c78 1614
ca2c1418
PA		 1615 if (!skb_unref(skb))
1616 return;
1617
dce4551c
PA		 1618 /* In the more common cases we cleared the head states previously,
1619 * see __udp_queue_rcv_skb().
0ddf3fb2 1620 */
dce4551c 1621 if (unlikely(udp_skb_has_head_state(skb)))
0ddf3fb2 1622 skb_release_head_state(skb);
ca2c1418 1623 __consume_stateless_skb(skb);
f970bd9e
PA		 1624}
1625EXPORT_SYMBOL_GPL(skb_consume_udp);
1626
2276f58a
PA		 1627static struct sk_buff *__first_packet_length(struct sock *sk,
1628 struct sk_buff_head *rcvq,
1629 int *total)
1630{
1631 struct sk_buff *skb;
1632
9bd780f5
PA		 1633 while ((skb = skb_peek(rcvq)) != NULL) {
1634 if (udp_lib_checksum_complete(skb)) {
1635 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1636 IS_UDPLITE(sk));
1637 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1638 IS_UDPLITE(sk));
1639 atomic_inc(&sk->sk_drops);
1640 __skb_unlink(skb, rcvq);
1641 *total += skb->truesize;
1642 kfree_skb(skb);
1643 } else {
a793183c 1644 udp_skb_csum_unnecessary_set(skb);
9bd780f5
PA		 1645 break;
1646 }
2276f58a
PA		 1647 }
1648 return skb;
1649}
1650
85584672
ED		 1651/**
1652 * first_packet_length - return length of first packet in receive queue
1653 * @sk: socket
1654 *
1655 * Drops all bad checksum frames, until a valid one is found.
e83c6744 1656 * Returns the length of found skb, or -1 if none is found.
85584672 1657 */
e83c6744 1658static int first_packet_length(struct sock *sk)
85584672 1659{
2276f58a
PA		 1660 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1661 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
85584672 1662 struct sk_buff *skb;
7c13f97f 1663 int total = 0;
e83c6744 1664 int res;
85584672 1665
85584672 1666 spin_lock_bh(&rcvq->lock);
2276f58a 1667 skb = __first_packet_length(sk, rcvq, &total);
137a0dbe 1668 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
2276f58a
PA		 1669 spin_lock(&sk_queue->lock);
1670 skb_queue_splice_tail_init(sk_queue, rcvq);
1671 spin_unlock(&sk_queue->lock);
1672
1673 skb = __first_packet_length(sk, rcvq, &total);
85584672 1674 }
e83c6744 1675 res = skb ? skb->len : -1;
7c13f97f 1676 if (total)
6dfb4367 1677 udp_rmem_release(sk, total, 1, false);
85584672 1678 spin_unlock_bh(&rcvq->lock);
85584672
ED		 1679 return res;
1680}
1681
1da177e4
LT		 1682/*
1683 * IOCTL requests applicable to the UDP protocol
1684 */
e905a9ed 1685
1da177e4
LT		 1686int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1687{
6516c655
SH		 1688 switch (cmd) {
1689 case SIOCOUTQ:
1da177e4 1690 {
31e6d363
ED		 1691 int amount = sk_wmem_alloc_get(sk);
1692
6516c655
SH		 1693 return put_user(amount, (int __user *)arg);
1694 }
1da177e4 1695
6516c655
SH		 1696 case SIOCINQ:
1697 {
e83c6744 1698 int amount = max_t(int, 0, first_packet_length(sk));
6516c655 1699
6516c655
SH		 1700 return put_user(amount, (int __user *)arg);
1701 }
1da177e4 1702
6516c655
SH		 1703 default:
1704 return -ENOIOCTLCMD;
1da177e4 1705 }
6516c655
SH		 1706
1707 return 0;
1da177e4 1708}
c482c568 1709EXPORT_SYMBOL(udp_ioctl);
1da177e4 1710
2276f58a 1711struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
fd69c399 1712 int noblock, int *off, int *err)
2276f58a
PA		 1713{
1714 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1715 struct sk_buff_head *queue;
1716 struct sk_buff *last;
1717 long timeo;
1718 int error;
1719
1720 queue = &udp_sk(sk)->reader_queue;
1721 flags |= noblock ? MSG_DONTWAIT : 0;
1722 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1723 do {
1724 struct sk_buff *skb;
1725
1726 error = sock_error(sk);
1727 if (error)
1728 break;
1729
1730 error = -EAGAIN;
2276f58a 1731 do {
2276f58a 1732 spin_lock_bh(&queue->lock);
e427cad6
PA		 1733 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1734 err, &last);
2276f58a 1735 if (skb) {
e427cad6
PA		 1736 if (!(flags & MSG_PEEK))
1737 udp_skb_destructor(sk, skb);
2276f58a 1738 spin_unlock_bh(&queue->lock);
2276f58a
PA		 1739 return skb;
1740 }
1741
137a0dbe 1742 if (skb_queue_empty_lockless(sk_queue)) {
2276f58a
PA		 1743 spin_unlock_bh(&queue->lock);
1744 goto busy_check;
1745 }
1746
6dfb4367
PA		 1747 /* refill the reader queue and walk it again
1748 * keep both queues locked to avoid re-acquiring
1749 * the sk_receive_queue lock if fwd memory scheduling
1750 * is needed.
1751 */
2276f58a
PA		 1752 spin_lock(&sk_queue->lock);
1753 skb_queue_splice_tail_init(sk_queue, queue);
2276f58a 1754
e427cad6
PA		 1755 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1756 err, &last);
1757 if (skb && !(flags & MSG_PEEK))
1758 udp_skb_dtor_locked(sk, skb);
6dfb4367 1759 spin_unlock(&sk_queue->lock);
2276f58a 1760 spin_unlock_bh(&queue->lock);
de321ed3 1761 if (skb)
2276f58a 1762 return skb;
2276f58a
PA		 1763
1764busy_check:
1765 if (!sk_can_busy_loop(sk))
1766 break;
1767
1768 sk_busy_loop(sk, flags & MSG_DONTWAIT);
137a0dbe 1769 } while (!skb_queue_empty_lockless(sk_queue));
2276f58a
PA		 1770
1771 /* sk_queue is empty, reader_queue may contain peeked packets */
1772 } while (timeo &&
b50b0580
SD		 1773 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1774 &error, &timeo,
2276f58a
PA		 1775 (struct sk_buff *)sk_queue));
1776
1777 *err = error;
1778 return NULL;
1779}
7e823644 1780EXPORT_SYMBOL(__skb_recv_udp);
2276f58a 1781
db8dac20
DM		 1782/*
1783 * This should be easy, if there is something there we
1784 * return it, otherwise we block.
1785 */
1786
1b784140
YX		 1787int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1788 int flags, int *addr_len)
db8dac20
DM		 1789{
1790 struct inet_sock *inet = inet_sk(sk);
342dfc30 1791 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
db8dac20 1792 struct sk_buff *skb;
59c2cdae 1793 unsigned int ulen, copied;
fd69c399 1794 int off, err, peeking = flags & MSG_PEEK;
db8dac20 1795 int is_udplite = IS_UDPLITE(sk);
197c949e 1796 bool checksum_valid = false;
db8dac20 1797
db8dac20 1798 if (flags & MSG_ERRQUEUE)
85fbaa75 1799 return ip_recv_error(sk, msg, len, addr_len);
db8dac20
DM		 1800
1801try_again:
a0917e0b 1802 off = sk_peek_offset(sk, flags);
fd69c399 1803 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
db8dac20 1804 if (!skb)
627d2d6b 1805 return err;
db8dac20 1806
b65ac446 1807 ulen = udp_skb_len(skb);
59c2cdae 1808 copied = len;
627d2d6b 1809 if (copied > ulen - off)
1810 copied = ulen - off;
59c2cdae 1811 else if (copied < ulen)
db8dac20
DM		 1812 msg->msg_flags |= MSG_TRUNC;
1813
1814 /*
1815 * If checksum is needed at all, try to do it while copying the
1816 * data. If the data is truncated, or if we only want a partial
1817 * coverage checksum (UDP-Lite), do it before the copy.
1818 */
1819
d21dbdfe
ED		 1820 if (copied < ulen || peeking ||
1821 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
b65ac446
PA		 1822 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1823 !__udp_lib_checksum_complete(skb);
197c949e 1824 if (!checksum_valid)
db8dac20
DM		 1825 goto csum_copy_err;
1826 }
1827
b65ac446
PA		 1828 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1829 if (udp_skb_is_linear(skb))
1830 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1831 else
1832 err = skb_copy_datagram_msg(skb, off, msg, copied);
1833 } else {
627d2d6b 1834 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
db8dac20
DM		 1835
1836 if (err == -EINVAL)
1837 goto csum_copy_err;
1838 }
1839
22911fc5 1840 if (unlikely(err)) {
fd69c399 1841 if (!peeking) {
979402b1 1842 atomic_inc(&sk->sk_drops);
6aef70a8
ED		 1843 UDP_INC_STATS(sock_net(sk),
1844 UDP_MIB_INERRORS, is_udplite);
979402b1 1845 }
850cbadd 1846 kfree_skb(skb);
627d2d6b 1847 return err;
22911fc5 1848 }
db8dac20 1849
fd69c399 1850 if (!peeking)
6aef70a8
ED		 1851 UDP_INC_STATS(sock_net(sk),
1852 UDP_MIB_INDATAGRAMS, is_udplite);
db8dac20 1853
3b885787 1854 sock_recv_ts_and_drops(msg, sk, skb);
db8dac20
DM		 1855
1856 /* Copy the address. */
c482c568 1857 if (sin) {
db8dac20
DM		 1858 sin->sin_family = AF_INET;
1859 sin->sin_port = udp_hdr(skb)->source;
1860 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1861 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
bceaa902 1862 *addr_len = sizeof(*sin);
983695fa
DB		 1863
1864 if (cgroup_bpf_enabled)
1865 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1866 (struct sockaddr *)sin);
db8dac20 1867 }
bcd1665e
PA		 1868
1869 if (udp_sk(sk)->gro_enabled)
1870 udp_cmsg_recv(msg, sk, skb);
1871
db8dac20 1872 if (inet->cmsg_flags)
ad959036 1873 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
db8dac20 1874
59c2cdae 1875 err = copied;
db8dac20
DM		 1876 if (flags & MSG_TRUNC)
1877 err = ulen;
1878
850cbadd 1879 skb_consume_udp(sk, skb, peeking ? -err : err);
db8dac20
DM		 1880 return err;
1881
1882csum_copy_err:
2276f58a
PA		 1883 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1884 udp_skb_destructor)) {
6aef70a8
ED		 1885 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1886 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
6a5dc9e5 1887 }
850cbadd 1888 kfree_skb(skb);
db8dac20 1889
beb39db5
ED		 1890 /* starting over for a new packet, but check if we need to yield */
1891 cond_resched();
9cfaa8de 1892 msg->msg_flags &= ~MSG_TRUNC;
db8dac20
DM		 1893 goto try_again;
1894}
1895
d74bad4e
AI		 1896int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1897{
1898 /* This check is replicated from __ip4_datagram_connect() and
1899 * intended to prevent BPF program called below from accessing bytes
1900 * that are out of the bound specified by user in addr_len.
1901 */
1902 if (addr_len < sizeof(struct sockaddr_in))
1903 return -EINVAL;
1904
1905 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1906}
1907EXPORT_SYMBOL(udp_pre_connect);
1908
286c72de 1909int __udp_disconnect(struct sock *sk, int flags)
1da177e4
LT		 1910{
1911 struct inet_sock *inet = inet_sk(sk);
1912 /*
1913 * 1003.1g - break association.
1914 */
e905a9ed 1915
1da177e4 1916 sk->sk_state = TCP_CLOSE;
c720c7e8
ED		 1917 inet->inet_daddr = 0;
1918 inet->inet_dport = 0;
bdeab991 1919 sock_rps_reset_rxhash(sk);
1da177e4 1920 sk->sk_bound_dev_if = 0;
303d0403 1921 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1da177e4 1922 inet_reset_saddr(sk);
303d0403
WB		 1923 if (sk->sk_prot->rehash &&
1924 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1925 sk->sk_prot->rehash(sk);
1926 }
1da177e4
LT		 1927
1928 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1929 sk->sk_prot->unhash(sk);
c720c7e8 1930 inet->inet_sport = 0;
1da177e4
LT		 1931 }
1932 sk_dst_reset(sk);
1933 return 0;
1934}
286c72de
ED		 1935EXPORT_SYMBOL(__udp_disconnect);
1936
1937int udp_disconnect(struct sock *sk, int flags)
1938{
1939 lock_sock(sk);
1940 __udp_disconnect(sk, flags);
1941 release_sock(sk);
1942 return 0;
1943}
c482c568 1944EXPORT_SYMBOL(udp_disconnect);
1da177e4 1945
645ca708
ED		 1946void udp_lib_unhash(struct sock *sk)
1947{
723b4610
ED		 1948 if (sk_hashed(sk)) {
1949 struct udp_table *udptable = sk->sk_prot->h.udp_table;
512615b6
ED		 1950 struct udp_hslot *hslot, *hslot2;
1951
1952 hslot = udp_hashslot(udptable, sock_net(sk),
1953 udp_sk(sk)->udp_port_hash);
1954 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
645ca708 1955
723b4610 1956 spin_lock_bh(&hslot->lock);
e32ea7e7
CG		 1957 if (rcu_access_pointer(sk->sk_reuseport_cb))
1958 reuseport_detach_sock(sk);
ca065d0c 1959 if (sk_del_node_init_rcu(sk)) {
fdcc8aa9 1960 hslot->count--;
c720c7e8 1961 inet_sk(sk)->inet_num = 0;
723b4610 1962 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
512615b6
ED		 1963
1964 spin_lock(&hslot2->lock);
ca065d0c 1965 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
512615b6
ED		 1966 hslot2->count--;
1967 spin_unlock(&hslot2->lock);
723b4610
ED		 1968 }
1969 spin_unlock_bh(&hslot->lock);
645ca708 1970 }
645ca708
ED		 1971}
1972EXPORT_SYMBOL(udp_lib_unhash);
1973
719f8358
ED		 1974/*
1975 * inet_rcv_saddr was changed, we must rehash secondary hash
1976 */
1977void udp_lib_rehash(struct sock *sk, u16 newhash)
1978{
1979 if (sk_hashed(sk)) {
1980 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1981 struct udp_hslot *hslot, *hslot2, *nhslot2;
1982
1983 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1984 nhslot2 = udp_hashslot2(udptable, newhash);
1985 udp_sk(sk)->udp_portaddr_hash = newhash;
e32ea7e7
CG		 1986
1987 if (hslot2 != nhslot2 ||
1988 rcu_access_pointer(sk->sk_reuseport_cb)) {
719f8358
ED		 1989 hslot = udp_hashslot(udptable, sock_net(sk),
1990 udp_sk(sk)->udp_port_hash);
1991 /* we must lock primary chain too */
1992 spin_lock_bh(&hslot->lock);
e32ea7e7
CG		 1993 if (rcu_access_pointer(sk->sk_reuseport_cb))
1994 reuseport_detach_sock(sk);
1995
1996 if (hslot2 != nhslot2) {
1997 spin_lock(&hslot2->lock);
ca065d0c 1998 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
e32ea7e7
CG		 1999 hslot2->count--;
2000 spin_unlock(&hslot2->lock);
2001
2002 spin_lock(&nhslot2->lock);
ca065d0c 2003 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
e32ea7e7
CG		 2004 &nhslot2->head);
2005 nhslot2->count++;
2006 spin_unlock(&nhslot2->lock);
2007 }
719f8358
ED		 2008
2009 spin_unlock_bh(&hslot->lock);
2010 }
2011 }
2012}
2013EXPORT_SYMBOL(udp_lib_rehash);
2014
8f6b5392 2015void udp_v4_rehash(struct sock *sk)
719f8358 2016{
f0b1e64c 2017 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
719f8358
ED		 2018 inet_sk(sk)->inet_rcv_saddr,
2019 inet_sk(sk)->inet_num);
2020 udp_lib_rehash(sk, new_hash);
2021}
2022
a3f96c47 2023static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
93821778 2024{
fec5e652 2025 int rc;
766e9037 2026
005ec974 2027 if (inet_sk(sk)->inet_daddr) {
bdeab991 2028 sock_rps_save_rxhash(sk, skb);
005ec974 2029 sk_mark_napi_id(sk, skb);
2c8c56e1 2030 sk_incoming_cpu_update(sk);
e68b6e50
ED		 2031 } else {
2032 sk_mark_napi_id_once(sk, skb);
005ec974 2033 }
fec5e652 2034
850cbadd 2035 rc = __udp_enqueue_schedule_skb(sk, skb);
766e9037
ED		 2036 if (rc < 0) {
2037 int is_udplite = IS_UDPLITE(sk);
93821778 2038
93821778 2039 /* Note that an ENOMEM error is charged twice */
766e9037 2040 if (rc == -ENOMEM)
e61da9e2 2041 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
02c22347 2042 is_udplite);
e61da9e2 2043 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
766e9037 2044 kfree_skb(skb);
296f7ea7 2045 trace_udp_fail_queue_rcv_skb(rc, sk);
766e9037 2046 return -1;
93821778
HX		 2047 }
2048
2049 return 0;
93821778
HX		 2050}
2051
db8dac20
DM		 2052/* returns:
2053 * -1: error
2054 * 0: success
2055 * >0: "udp encap" protocol resubmission
2056 *
2057 * Note that in the success and error cases, the skb is assumed to
2058 * have either been requeued or freed.
2059 */
cf329aa4 2060static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
db8dac20
DM		 2061{
2062 struct udp_sock *up = udp_sk(sk);
db8dac20
DM		 2063 int is_udplite = IS_UDPLITE(sk);
2064
2065 /*
2066 * Charge it to the socket, dropping if the queue is full.
2067 */
2068 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2069 goto drop;
895b5c9f 2070 nf_reset_ct(skb);
db8dac20 2071
88ab3108 2072 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
0ad92ad0
ED		 2073 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2074
db8dac20
DM		 2075 /*
2076 * This is an encapsulation socket so pass the skb to
2077 * the socket's udp_encap_rcv() hook. Otherwise, just
2078 * fall through and pass this up the UDP socket.
2079 * up->encap_rcv() returns the following value:
2080 * =0 if skb was successfully passed to the encap
2081 * handler or was discarded by it.
2082 * >0 if skb should be passed on to UDP.
2083 * <0 if skb should be resubmitted as proto -N
2084 */
2085
2086 /* if we're overly short, let UDP handle it */
6aa7de05 2087 encap_rcv = READ_ONCE(up->encap_rcv);
e5aed006 2088 if (encap_rcv) {
db8dac20
DM		 2089 int ret;
2090
0a80966b
TH		 2091 /* Verify checksum before giving to encap */
2092 if (udp_lib_checksum_complete(skb))
2093 goto csum_error;
2094
0ad92ad0 2095 ret = encap_rcv(sk, skb);
db8dac20 2096 if (ret <= 0) {
02c22347
ED		 2097 __UDP_INC_STATS(sock_net(sk),
2098 UDP_MIB_INDATAGRAMS,
2099 is_udplite);
db8dac20
DM		 2100 return -ret;
2101 }
2102 }
2103
2104 /* FALLTHROUGH -- it's a UDP Packet */
2105 }
2106
2107 /*
2108 * UDP-Lite specific tests, ignored on UDP sockets
2109 */
2110 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
2111
2112 /*
2113 * MIB statistics other than incrementing the error count are
2114 * disabled for the following two types of errors: these depend
2115 * on the application settings, not on the functioning of the
2116 * protocol stack as such.
2117 *
2118 * RFC 3828 here recommends (sec 3.3): "There should also be a
2119 * way ... to ... at least let the receiving application block
2120 * delivery of packets with coverage values less than a value
2121 * provided by the application."
2122 */
2123 if (up->pcrlen == 0) { /* full coverage was set */
ba7a46f1
JP		 2124 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2125 UDP_SKB_CB(skb)->cscov, skb->len);
db8dac20
DM		 2126 goto drop;
2127 }
2128 /* The next case involves violating the min. coverage requested
2129 * by the receiver. This is subtle: if receiver wants x and x is
2130 * greater than the buffersize/MTU then receiver will complain
2131 * that it wants x while sender emits packets of smaller size y.
2132 * Therefore the above ...()->partial_cov statement is essential.
2133 */
2134 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
ba7a46f1
JP		 2135 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2136 UDP_SKB_CB(skb)->cscov, up->pcrlen);
db8dac20
DM		 2137 goto drop;
2138 }
2139 }
2140
dd99e425 2141 prefetch(&sk->sk_rmem_alloc);
ce25d66a
ED		 2142 if (rcu_access_pointer(sk->sk_filter) &&
2143 udp_lib_checksum_complete(skb))
e6afc8ac 2144 goto csum_error;
ce25d66a 2145
ba66bbe5 2146 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
a6127697 2147 goto drop;
db8dac20 2148
e6afc8ac 2149 udp_csum_pull_header(skb);
db8dac20 2150
fbf8866d 2151 ipv4_pktinfo_prepare(sk, skb);
850cbadd 2152 return __udp_queue_rcv_skb(sk, skb);
db8dac20 2153
6a5dc9e5 2154csum_error:
02c22347 2155 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
db8dac20 2156drop:
02c22347 2157 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
8edf19c2 2158 atomic_inc(&sk->sk_drops);
db8dac20
DM		 2159 kfree_skb(skb);
2160 return -1;
2161}
2162
cf329aa4
PA		 2163static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2164{
2165 struct sk_buff *next, *segs;
2166 int ret;
2167
2168 if (likely(!udp_unexpected_gso(sk, skb)))
2169 return udp_queue_rcv_one_skb(sk, skb);
2170
a08e7fd9 2171 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
cf329aa4
PA		 2172 __skb_push(skb, -skb_mac_offset(skb));
2173 segs = udp_rcv_segment(sk, skb, true);
1a186c14 2174 skb_list_walk_safe(segs, skb, next) {
cf329aa4
PA		 2175 __skb_pull(skb, skb_transport_offset(skb));
2176 ret = udp_queue_rcv_one_skb(sk, skb);
2177 if (ret > 0)
2178 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2179 }
2180 return 0;
2181}
2182
97502231 2183/* For TCP sockets, sk_rx_dst is protected by socket lock
e47eb5df 2184 * For UDP, we use xchg() to guard against concurrent changes.
97502231 2185 */
64f0f5d1 2186bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
421b3885 2187{
97502231
ED		 2188 struct dst_entry *old;
2189
d24406c8
WW		 2190 if (dst_hold_safe(dst)) {
2191 old = xchg(&sk->sk_rx_dst, dst);
2192 dst_release(old);
64f0f5d1 2193 return old != dst;
d24406c8 2194 }
64f0f5d1 2195 return false;
421b3885 2196}
c9f2c1ae 2197EXPORT_SYMBOL(udp_sk_rx_dst_set);
421b3885 2198
db8dac20
DM		 2199/*
2200 * Multicasts and broadcasts go to each listener.
2201 *
1240d137 2202 * Note: called only from the BH handler context.
db8dac20 2203 */
e3163493 2204static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
db8dac20
DM		 2205 struct udphdr *uh,
2206 __be32 saddr, __be32 daddr,
36cbb245
RJ		 2207 struct udp_table *udptable,
2208 int proto)
db8dac20 2209{
ca065d0c 2210 struct sock *sk, *first = NULL;
5cf3d461
DH		 2211 unsigned short hnum = ntohs(uh->dest);
2212 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2dc41cff 2213 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
ca065d0c
ED		 2214 unsigned int offset = offsetof(typeof(*sk), sk_node);
2215 int dif = skb->dev->ifindex;
fb74c277 2216 int sdif = inet_sdif(skb);
ca065d0c
ED		 2217 struct hlist_node *node;
2218 struct sk_buff *nskb;
2dc41cff
DH		 2219
2220 if (use_hash2) {
f0b1e64c 2221 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
73e2d5e3 2222 udptable->mask;
f0b1e64c 2223 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2dc41cff 2224start_lookup:
73e2d5e3 2225 hslot = &udptable->hash2[hash2];
2dc41cff
DH		 2226 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2227 }
db8dac20 2228
ca065d0c
ED		 2229 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2230 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
fb74c277 2231 uh->source, saddr, dif, sdif, hnum))
ca065d0c
ED		 2232 continue;
2233
2234 if (!first) {
2235 first = sk;
2236 continue;
1240d137 2237 }
ca065d0c 2238 nskb = skb_clone(skb, GFP_ATOMIC);
1240d137 2239
ca065d0c
ED		 2240 if (unlikely(!nskb)) {
2241 atomic_inc(&sk->sk_drops);
02c22347
ED		 2242 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2243 IS_UDPLITE(sk));
2244 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2245 IS_UDPLITE(sk));
ca065d0c
ED		 2246 continue;
2247 }
2248 if (udp_queue_rcv_skb(sk, nskb) > 0)
2249 consume_skb(nskb);
2250 }
1240d137 2251
2dc41cff
DH		 2252 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2253 if (use_hash2 && hash2 != hash2_any) {
2254 hash2 = hash2_any;
2255 goto start_lookup;
2256 }
2257
ca065d0c
ED		 2258 if (first) {
2259 if (udp_queue_rcv_skb(first, skb) > 0)
2260 consume_skb(skb);
1240d137 2261 } else {
ca065d0c 2262 kfree_skb(skb);
02c22347
ED		 2263 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2264 proto == IPPROTO_UDPLITE);
1240d137 2265 }
db8dac20
DM		 2266 return 0;
2267}
2268
2269/* Initialize UDP checksum. If exited with zero value (success),
2270 * CHECKSUM_UNNECESSARY means, that no more checks are required.
666a3d6e 2271 * Otherwise, csum completion requires checksumming packet body,
db8dac20
DM		 2272 * including udp header and folding it to skb->csum.
2273 */
2274static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2275 int proto)
2276{
db8dac20
DM		 2277 int err;
2278
2279 UDP_SKB_CB(skb)->partial_cov = 0;
2280 UDP_SKB_CB(skb)->cscov = skb->len;
2281
2282 if (proto == IPPROTO_UDPLITE) {
2283 err = udplite_checksum_init(skb, uh);
2284 if (err)
2285 return err;
15f35d49
AK		 2286
2287 if (UDP_SKB_CB(skb)->partial_cov) {
2288 skb->csum = inet_compute_pseudo(skb, proto);
2289 return 0;
2290 }
db8dac20
DM		 2291 }
2292
b46d9f62
HFS		 2293 /* Note, we are only interested in != 0 or == 0, thus the
2294 * force to int.
2295 */
db4f1be3
ST		 2296 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2297 inet_compute_pseudo);
2298 if (err)
2299 return err;
2300
2301 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2302 /* If SW calculated the value, we know it's bad */
2303 if (skb->csum_complete_sw)
2304 return 1;
2305
2306 /* HW says the value is bad. Let's validate that.
2307 * skb->csum is no longer the full packet checksum,
2308 * so don't treat it as such.
2309 */
2310 skb_checksum_complete_unset(skb);
2311 }
2312
2313 return 0;
db8dac20
DM		 2314}
2315
2b5a9217
PA		 2316/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2317 * return code conversion for ip layer consumption
2318 */
2319static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2320 struct udphdr *uh)
2321{
2322 int ret;
2323
2324 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
e4aa33ad 2325 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2b5a9217
PA		 2326
2327 ret = udp_queue_rcv_skb(sk, skb);
2328
2329 /* a return value > 0 means to resubmit the input, but
2330 * it wants the return to be -protocol, or 0
2331 */
2332 if (ret > 0)
2333 return -ret;
2334 return 0;
2335}
2336
db8dac20
DM		 2337/*
2338 * All we need to do is get the socket, and then do a checksum.
2339 */
2340
645ca708 2341int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
db8dac20
DM		 2342 int proto)
2343{
2344 struct sock *sk;
7b5e56f9 2345 struct udphdr *uh;
db8dac20 2346 unsigned short ulen;
adf30907 2347 struct rtable *rt = skb_rtable(skb);
2783ef23 2348 __be32 saddr, daddr;
0283328e 2349 struct net *net = dev_net(skb->dev);
71489e21 2350 bool refcounted;
db8dac20
DM		 2351
2352 /*
2353 * Validate the packet.
2354 */
2355 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2356 goto drop; /* No space for header. */
2357
7b5e56f9 2358 uh = udp_hdr(skb);
db8dac20 2359 ulen = ntohs(uh->len);
ccc2d97c
BM		 2360 saddr = ip_hdr(skb)->saddr;
2361 daddr = ip_hdr(skb)->daddr;
2362
db8dac20
DM		 2363 if (ulen > skb->len)
2364 goto short_packet;
2365
2366 if (proto == IPPROTO_UDP) {
2367 /* UDP validates ulen. */
2368 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2369 goto short_packet;
2370 uh = udp_hdr(skb);
2371 }
2372
2373 if (udp4_csum_init(skb, uh, proto))
2374 goto csum_error;
2375
71489e21 2376 sk = skb_steal_sock(skb, &refcounted);
8afdd99a 2377 if (sk) {
97502231 2378 struct dst_entry *dst = skb_dst(skb);
421b3885 2379 int ret;
421b3885 2380
97502231
ED		 2381 if (unlikely(sk->sk_rx_dst != dst))
2382 udp_sk_rx_dst_set(sk, dst);
db8dac20 2383
2b5a9217 2384 ret = udp_unicast_rcv_skb(sk, skb, uh);
71489e21
JS		 2385 if (refcounted)
2386 sock_put(sk);
2b5a9217 2387 return ret;
421b3885 2388 }
db8dac20 2389
c18450a5
FF		 2390 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2391 return __udp4_lib_mcast_deliver(net, skb, uh,
36cbb245 2392 saddr, daddr, udptable, proto);
c18450a5
FF		 2393
2394 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2b5a9217
PA		 2395 if (sk)
2396 return udp_unicast_rcv_skb(sk, skb, uh);
db8dac20
DM		 2397
2398 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2399 goto drop;
895b5c9f 2400 nf_reset_ct(skb);
db8dac20
DM		 2401
2402 /* No socket. Drop packet silently, if checksum is wrong */
2403 if (udp_lib_checksum_complete(skb))
2404 goto csum_error;
2405
02c22347 2406 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
db8dac20
DM		 2407 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2408
2409 /*
2410 * Hmm. We got an UDP packet to a port to which we
2411 * don't wanna listen. Ignore it.
2412 */
2413 kfree_skb(skb);
2414 return 0;
2415
2416short_packet:
ba7a46f1
JP		 2417 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2418 proto == IPPROTO_UDPLITE ? "Lite" : "",
2419 &saddr, ntohs(uh->source),
2420 ulen, skb->len,
2421 &daddr, ntohs(uh->dest));
db8dac20
DM		 2422 goto drop;
2423
2424csum_error:
2425 /*
2426 * RFC1122: OK. Discards the bad packet silently (as far as
2427 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2428 */
ba7a46f1
JP		 2429 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2430 proto == IPPROTO_UDPLITE ? "Lite" : "",
2431 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2432 ulen);
02c22347 2433 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
db8dac20 2434drop:
02c22347 2435 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
db8dac20
DM		 2436 kfree_skb(skb);
2437 return 0;
2438}
2439
421b3885
SB		 2440/* We can only early demux multicast if there is a single matching socket.
2441 * If more than one socket found returns NULL
2442 */
2443static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2444 __be16 loc_port, __be32 loc_addr,
2445 __be16 rmt_port, __be32 rmt_addr,
fb74c277 2446 int dif, int sdif)
421b3885
SB		 2447{
2448 struct sock *sk, *result;
421b3885 2449 unsigned short hnum = ntohs(loc_port);
ca065d0c 2450 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
421b3885
SB		 2451 struct udp_hslot *hslot = &udp_table.hash[slot];
2452
63c6f81c
ED		 2453 /* Do not bother scanning a too big list */
2454 if (hslot->count > 10)
2455 return NULL;
2456
421b3885 2457 result = NULL;
ca065d0c
ED		 2458 sk_for_each_rcu(sk, &hslot->head) {
2459 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
fb74c277 2460 rmt_port, rmt_addr, dif, sdif, hnum)) {
ca065d0c
ED		 2461 if (result)
2462 return NULL;
421b3885 2463 result = sk;
421b3885
SB		 2464 }
2465 }
ca065d0c 2466
421b3885
SB		 2467 return result;
2468}
2469
2470/* For unicast we should only early demux connected sockets or we can
2471 * break forwarding setups. The chains here can be long so only check
2472 * if the first socket is an exact match and if not move on.
2473 */
2474static struct sock *__udp4_lib_demux_lookup(struct net *net,
2475 __be16 loc_port, __be32 loc_addr,
2476 __be16 rmt_port, __be32 rmt_addr,
3fa6f616 2477 int dif, int sdif)
421b3885 2478{
421b3885 2479 unsigned short hnum = ntohs(loc_port);
f0b1e64c 2480 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
421b3885
SB		 2481 unsigned int slot2 = hash2 & udp_table.mask;
2482 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
c7228317 2483 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
421b3885 2484 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
ca065d0c 2485 struct sock *sk;
421b3885 2486
ca065d0c
ED		 2487 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2488 if (INET_MATCH(sk, net, acookie, rmt_addr,
3fa6f616 2489 loc_addr, ports, dif, sdif))
ca065d0c 2490 return sk;
421b3885
SB		 2491 /* Only check first socket in chain */
2492 break;
2493 }
ca065d0c 2494 return NULL;
421b3885
SB		 2495}
2496
7487449c 2497int udp_v4_early_demux(struct sk_buff *skb)
421b3885 2498{
610438b7 2499 struct net *net = dev_net(skb->dev);
bc044e8d 2500 struct in_device *in_dev = NULL;
610438b7
ED		 2501 const struct iphdr *iph;
2502 const struct udphdr *uh;
ca065d0c 2503 struct sock *sk = NULL;
421b3885 2504 struct dst_entry *dst;
421b3885 2505 int dif = skb->dev->ifindex;
fb74c277 2506 int sdif = inet_sdif(skb);
6e540309 2507 int ours;
421b3885
SB		 2508
2509 /* validate the packet */
2510 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
7487449c 2511 return 0;
421b3885 2512
610438b7
ED		 2513 iph = ip_hdr(skb);
2514 uh = udp_hdr(skb);
2515
996b44fc 2516 if (skb->pkt_type == PACKET_MULTICAST) {
bc044e8d 2517 in_dev = __in_dev_get_rcu(skb->dev);
6e540309
SB		 2518
2519 if (!in_dev)
7487449c 2520 return 0;
6e540309 2521
996b44fc
PA		 2522 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2523 iph->protocol);
2524 if (!ours)
2525 return 0;
ad0ea198 2526
421b3885 2527 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
fb74c277
DA		 2528 uh->source, iph->saddr,
2529 dif, sdif);
6e540309 2530 } else if (skb->pkt_type == PACKET_HOST) {
421b3885 2531 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
3fa6f616 2532 uh->source, iph->saddr, dif, sdif);
6e540309 2533 }
421b3885 2534
41c6d650 2535 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
7487449c 2536 return 0;
421b3885
SB		 2537
2538 skb->sk = sk;
82eabd9e 2539 skb->destructor = sock_efree;
10e2eb87 2540 dst = READ_ONCE(sk->sk_rx_dst);
421b3885
SB		 2541
2542 if (dst)
2543 dst = dst_check(dst, 0);
10e2eb87 2544 if (dst) {
bc044e8d
PA		 2545 u32 itag = 0;
2546
d24406c8
WW		 2547 /* set noref for now.
2548 * any place which wants to hold dst has to call
2549 * dst_hold_safe()
2550 */
2551 skb_dst_set_noref(skb, dst);
bc044e8d
PA		 2552
2553 /* for unconnected multicast sockets we need to validate
2554 * the source on each packet
2555 */
2556 if (!inet_sk(sk)->inet_daddr && in_dev)
2557 return ip_mc_validate_source(skb, iph->daddr,
2558 iph->saddr, iph->tos,
2559 skb->dev, in_dev, &itag);
10e2eb87 2560 }
7487449c 2561 return 0;
421b3885
SB		 2562}
2563
db8dac20
DM		 2564int udp_rcv(struct sk_buff *skb)
2565{
645ca708 2566 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
db8dac20
DM		 2567}
2568
7d06b2e0 2569void udp_destroy_sock(struct sock *sk)
db8dac20 2570{
44046a59 2571 struct udp_sock *up = udp_sk(sk);
8a74ad60 2572 bool slow = lock_sock_fast(sk);
db8dac20 2573 udp_flush_pending_frames(sk);
8a74ad60 2574 unlock_sock_fast(sk, slow);
60fb9567
PA		 2575 if (static_branch_unlikely(&udp_encap_needed_key)) {
2576 if (up->encap_type) {
2577 void (*encap_destroy)(struct sock *sk);
2578 encap_destroy = READ_ONCE(up->encap_destroy);
2579 if (encap_destroy)
2580 encap_destroy(sk);
2581 }
2582 if (up->encap_enabled)
9c480601 2583 static_branch_dec(&udp_encap_needed_key);
44046a59 2584 }
db8dac20
DM		 2585}
2586
1da177e4
LT		 2587/*
2588 * Socket option code for UDP
2589 */
4c0a6cb0 2590int udp_lib_setsockopt(struct sock *sk, int level, int optname,
b7058842 2591 char __user *optval, unsigned int optlen,
4c0a6cb0 2592 int (*push_pending_frames)(struct sock *))
1da177e4
LT		 2593{
2594 struct udp_sock *up = udp_sk(sk);
1c19448c 2595 int val, valbool;
1da177e4 2596 int err = 0;
b2bf1e26 2597 int is_udplite = IS_UDPLITE(sk);
1da177e4 2598
c482c568 2599 if (optlen < sizeof(int))
1da177e4
LT		 2600 return -EINVAL;
2601
2602 if (get_user(val, (int __user *)optval))
2603 return -EFAULT;
2604
1c19448c
TH		 2605 valbool = val ? 1 : 0;
2606
6516c655 2607 switch (optname) {
1da177e4
LT		 2608 case UDP_CORK:
2609 if (val != 0) {
2610 up->corkflag = 1;
2611 } else {
2612 up->corkflag = 0;
2613 lock_sock(sk);
4243cdc2 2614 push_pending_frames(sk);
1da177e4
LT		 2615 release_sock(sk);
2616 }
2617 break;
e905a9ed 2618
1da177e4
LT		 2619 case UDP_ENCAP:
2620 switch (val) {
2621 case 0:
fd1ac07f 2622#ifdef CONFIG_XFRM
1da177e4
LT		 2623 case UDP_ENCAP_ESPINUDP:
2624 case UDP_ENCAP_ESPINUDP_NON_IKE:
0146dca7
SD		 2625#if IS_ENABLED(CONFIG_IPV6)
2626 if (sk->sk_family == AF_INET6)
2627 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2628 else
2629#endif
2630 up->encap_rcv = xfrm4_udp_encap_rcv;
fd1ac07f 2631#endif
a8eceea8 2632 fallthrough;
342f0234 2633 case UDP_ENCAP_L2TPINUDP:
1da177e4 2634 up->encap_type = val;
60fb9567
PA		 2635 lock_sock(sk);
2636 udp_tunnel_encap_enable(sk->sk_socket);
2637 release_sock(sk);
1da177e4
LT		 2638 break;
2639 default:
2640 err = -ENOPROTOOPT;
2641 break;
2642 }
2643 break;
2644
1c19448c
TH		 2645 case UDP_NO_CHECK6_TX:
2646 up->no_check6_tx = valbool;
2647 break;
2648
2649 case UDP_NO_CHECK6_RX:
2650 up->no_check6_rx = valbool;
2651 break;
2652
bec1f6f6
WB		 2653 case UDP_SEGMENT:
2654 if (val < 0 || val > USHRT_MAX)
2655 return -EINVAL;
2656 up->gso_size = val;
2657 break;
2658
e20cf8d3
PA		 2659 case UDP_GRO:
2660 lock_sock(sk);
2661 if (valbool)
2662 udp_tunnel_encap_enable(sk->sk_socket);
2663 up->gro_enabled = valbool;
2664 release_sock(sk);
2665 break;
2666
ba4e58ec
GR		 2667 /*
2668 * UDP-Lite's partial checksum coverage (RFC 3828).
2669 */
2670 /* The sender sets actual checksum coverage length via this option.
2671 * The case coverage > packet length is handled by send module. */
2672 case UDPLITE_SEND_CSCOV:
b2bf1e26 2673 if (!is_udplite) /* Disable the option on UDP sockets */
ba4e58ec
GR		 2674 return -ENOPROTOOPT;
2675 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2676 val = 8;
4be929be
AD		 2677 else if (val > USHRT_MAX)
2678 val = USHRT_MAX;
ba4e58ec
GR		 2679 up->pcslen = val;
2680 up->pcflag |= UDPLITE_SEND_CC;
2681 break;
2682
e905a9ed
YH		 2683 /* The receiver specifies a minimum checksum coverage value. To make
2684 * sense, this should be set to at least 8 (as done below). If zero is
ba4e58ec
GR		 2685 * used, this again means full checksum coverage. */
2686 case UDPLITE_RECV_CSCOV:
b2bf1e26 2687 if (!is_udplite) /* Disable the option on UDP sockets */
ba4e58ec
GR		 2688 return -ENOPROTOOPT;
2689 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2690 val = 8;
4be929be
AD		 2691 else if (val > USHRT_MAX)
2692 val = USHRT_MAX;
ba4e58ec
GR		 2693 up->pcrlen = val;
2694 up->pcflag |= UDPLITE_RECV_CC;
2695 break;
2696
1da177e4
LT		 2697 default:
2698 err = -ENOPROTOOPT;
2699 break;
6516c655 2700 }
1da177e4
LT		 2701
2702 return err;
2703}
c482c568 2704EXPORT_SYMBOL(udp_lib_setsockopt);
1da177e4 2705
db8dac20 2706int udp_setsockopt(struct sock *sk, int level, int optname,
b7058842 2707 char __user *optval, unsigned int optlen)
db8dac20
DM		 2708{
2709 if (level == SOL_UDP || level == SOL_UDPLITE)
2710 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2711 udp_push_pending_frames);
2712 return ip_setsockopt(sk, level, optname, optval, optlen);
2713}
2714
4c0a6cb0
GR		 2715int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2716 char __user *optval, int __user *optlen)
1da177e4
LT		 2717{
2718 struct udp_sock *up = udp_sk(sk);
2719 int val, len;
2720
c482c568 2721 if (get_user(len, optlen))
1da177e4
LT		 2722 return -EFAULT;
2723
2724 len = min_t(unsigned int, len, sizeof(int));
e905a9ed 2725
6516c655 2726 if (len < 0)
1da177e4
LT		 2727 return -EINVAL;
2728
6516c655 2729 switch (optname) {
1da177e4
LT		 2730 case UDP_CORK:
2731 val = up->corkflag;
2732 break;
2733
2734 case UDP_ENCAP:
2735 val = up->encap_type;
2736 break;
2737
1c19448c
TH		 2738 case UDP_NO_CHECK6_TX:
2739 val = up->no_check6_tx;
2740 break;
2741
2742 case UDP_NO_CHECK6_RX:
2743 val = up->no_check6_rx;
2744 break;
2745
bec1f6f6
WB		 2746 case UDP_SEGMENT:
2747 val = up->gso_size;
2748 break;
2749
ba4e58ec
GR		 2750 /* The following two cannot be changed on UDP sockets, the return is
2751 * always 0 (which corresponds to the full checksum coverage of UDP). */
2752 case UDPLITE_SEND_CSCOV:
2753 val = up->pcslen;
2754 break;
2755
2756 case UDPLITE_RECV_CSCOV:
2757 val = up->pcrlen;
2758 break;
2759
1da177e4
LT		 2760 default:
2761 return -ENOPROTOOPT;
6516c655 2762 }
1da177e4 2763
6516c655 2764 if (put_user(len, optlen))
e905a9ed 2765 return -EFAULT;
c482c568 2766 if (copy_to_user(optval, &val, len))
1da177e4 2767 return -EFAULT;
e905a9ed 2768 return 0;
1da177e4 2769}
c482c568 2770EXPORT_SYMBOL(udp_lib_getsockopt);
1da177e4 2771
db8dac20
DM		 2772int udp_getsockopt(struct sock *sk, int level, int optname,
2773 char __user *optval, int __user *optlen)
2774{
2775 if (level == SOL_UDP || level == SOL_UDPLITE)
2776 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2777 return ip_getsockopt(sk, level, optname, optval, optlen);
2778}
2779
1da177e4
LT		 2780/**
2781 * udp_poll - wait for a UDP event.
3628e3cb
AL		 2782 * @file: - file struct
2783 * @sock: - socket
2784 * @wait: - poll table
1da177e4 2785 *
e905a9ed 2786 * This is same as datagram poll, except for the special case of
1da177e4
LT		 2787 * blocking sockets. If application is using a blocking fd
2788 * and a packet with checksum error is in the queue;
2789 * then it could get return from select indicating data available
2790 * but then block when reading it. Add special case code
2791 * to work around these arguably broken applications.
2792 */
a11e1d43 2793__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1da177e4 2794{
a11e1d43 2795 __poll_t mask = datagram_poll(file, sock, wait);
1da177e4 2796 struct sock *sk = sock->sk;
ba4e58ec 2797
3ef7cf57 2798 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
a9a08845 2799 mask |= EPOLLIN | EPOLLRDNORM;
2276f58a 2800
1da177e4 2801 /* Check for false positives due to checksum errors */
a11e1d43 2802 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
e83c6744 2803 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
a9a08845 2804 mask &= ~(EPOLLIN | EPOLLRDNORM);
1da177e4
LT		 2805
2806 return mask;
e905a9ed 2807
1da177e4 2808}
a11e1d43 2809EXPORT_SYMBOL(udp_poll);
1da177e4 2810
5d77dca8
DA		 2811int udp_abort(struct sock *sk, int err)
2812{
2813 lock_sock(sk);
2814
2815 sk->sk_err = err;
2816 sk->sk_error_report(sk);
286c72de 2817 __udp_disconnect(sk, 0);
5d77dca8
DA		 2818
2819 release_sock(sk);
2820
2821 return 0;
2822}
2823EXPORT_SYMBOL_GPL(udp_abort);
2824
db8dac20 2825struct proto udp_prot = {
1e802951
TZ		 2826 .name = "UDP",
2827 .owner = THIS_MODULE,
2828 .close = udp_lib_close,
d74bad4e 2829 .pre_connect = udp_pre_connect,
1e802951
TZ		 2830 .connect = ip4_datagram_connect,
2831 .disconnect = udp_disconnect,
2832 .ioctl = udp_ioctl,
2833 .init = udp_init_sock,
2834 .destroy = udp_destroy_sock,
2835 .setsockopt = udp_setsockopt,
2836 .getsockopt = udp_getsockopt,
2837 .sendmsg = udp_sendmsg,
2838 .recvmsg = udp_recvmsg,
2839 .sendpage = udp_sendpage,
2840 .release_cb = ip4_datagram_release_cb,
2841 .hash = udp_lib_hash,
2842 .unhash = udp_lib_unhash,
2843 .rehash = udp_v4_rehash,
2844 .get_port = udp_v4_get_port,
2845 .memory_allocated = &udp_memory_allocated,
2846 .sysctl_mem = sysctl_udp_mem,
2847 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2848 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2849 .obj_size = sizeof(struct udp_sock),
2850 .h.udp_table = &udp_table,
1e802951 2851 .diag_destroy = udp_abort,
db8dac20 2852};
c482c568 2853EXPORT_SYMBOL(udp_prot);
1da177e4
LT		 2854
2855/* ------------------------------------------------------------------------ */
2856#ifdef CONFIG_PROC_FS
2857
645ca708 2858static struct sock *udp_get_first(struct seq_file *seq, int start)
1da177e4
LT		 2859{
2860 struct sock *sk;
9e8ca27a 2861 struct udp_seq_afinfo *afinfo;
1da177e4 2862 struct udp_iter_state *state = seq->private;
6f191efe 2863 struct net *net = seq_file_net(seq);
1da177e4 2864
9e8ca27a
YS		 2865 if (state->bpf_seq_afinfo)
2866 afinfo = state->bpf_seq_afinfo;
2867 else
2868 afinfo = PDE_DATA(file_inode(seq->file));
2869
a3d2599b 2870 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
f86dcc5a 2871 ++state->bucket) {
a3d2599b 2872 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
f86dcc5a 2873
ca065d0c 2874 if (hlist_empty(&hslot->head))
f86dcc5a
ED		 2875 continue;
2876
645ca708 2877 spin_lock_bh(&hslot->lock);
ca065d0c 2878 sk_for_each(sk, &hslot->head) {
878628fb 2879 if (!net_eq(sock_net(sk), net))
a91275ef 2880 continue;
9e8ca27a
YS		 2881 if (afinfo->family == AF_UNSPEC ||
2882 sk->sk_family == afinfo->family)
1da177e4
LT		 2883 goto found;
2884 }
645ca708 2885 spin_unlock_bh(&hslot->lock);
1da177e4
LT		 2886 }
2887 sk = NULL;
2888found:
2889 return sk;
2890}
2891
2892static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2893{
9e8ca27a 2894 struct udp_seq_afinfo *afinfo;
1da177e4 2895 struct udp_iter_state *state = seq->private;
6f191efe 2896 struct net *net = seq_file_net(seq);
1da177e4 2897
9e8ca27a
YS		 2898 if (state->bpf_seq_afinfo)
2899 afinfo = state->bpf_seq_afinfo;
2900 else
2901 afinfo = PDE_DATA(file_inode(seq->file));
2902
1da177e4 2903 do {
ca065d0c 2904 sk = sk_next(sk);
9e8ca27a
YS		 2905 } while (sk && (!net_eq(sock_net(sk), net) ||
2906 (afinfo->family != AF_UNSPEC &&
2907 sk->sk_family != afinfo->family)));
1da177e4 2908
645ca708 2909 if (!sk) {
a3d2599b
CH		 2910 if (state->bucket <= afinfo->udp_table->mask)
2911 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
645ca708 2912 return udp_get_first(seq, state->bucket + 1);
1da177e4
LT		 2913 }
2914 return sk;
2915}
2916
2917static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2918{
645ca708 2919 struct sock *sk = udp_get_first(seq, 0);
1da177e4
LT		 2920
2921 if (sk)
6516c655 2922 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1da177e4
LT		 2923 --pos;
2924 return pos ? NULL : sk;
2925}
2926
a3d2599b 2927void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1da177e4 2928{
30842f29 2929 struct udp_iter_state *state = seq->private;
f86dcc5a 2930 state->bucket = MAX_UDP_PORTS;
30842f29 2931
b50660f1 2932 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1da177e4 2933}
a3d2599b 2934EXPORT_SYMBOL(udp_seq_start);
1da177e4 2935
a3d2599b 2936void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1da177e4
LT		 2937{
2938 struct sock *sk;
2939
b50660f1 2940 if (v == SEQ_START_TOKEN)
1da177e4
LT		 2941 sk = udp_get_idx(seq, 0);
2942 else
2943 sk = udp_get_next(seq, v);
2944
2945 ++*pos;
2946 return sk;
2947}
a3d2599b 2948EXPORT_SYMBOL(udp_seq_next);
1da177e4 2949
a3d2599b 2950void udp_seq_stop(struct seq_file *seq, void *v)
1da177e4 2951{
9e8ca27a 2952 struct udp_seq_afinfo *afinfo;
645ca708
ED		 2953 struct udp_iter_state *state = seq->private;
2954
9e8ca27a
YS		 2955 if (state->bpf_seq_afinfo)
2956 afinfo = state->bpf_seq_afinfo;
2957 else
2958 afinfo = PDE_DATA(file_inode(seq->file));
2959
a3d2599b
CH		 2960 if (state->bucket <= afinfo->udp_table->mask)
2961 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
1da177e4 2962}
a3d2599b 2963EXPORT_SYMBOL(udp_seq_stop);
db8dac20
DM		 2964
2965/* ------------------------------------------------------------------------ */
5e659e4c 2966static void udp4_format_sock(struct sock *sp, struct seq_file *f,
652586df 2967 int bucket)
db8dac20
DM		 2968{
2969 struct inet_sock *inet = inet_sk(sp);
c720c7e8
ED		 2970 __be32 dest = inet->inet_daddr;
2971 __be32 src = inet->inet_rcv_saddr;
2972 __u16 destp = ntohs(inet->inet_dport);
2973 __u16 srcp = ntohs(inet->inet_sport);
db8dac20 2974
f86dcc5a 2975 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
ea9a0379 2976 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
db8dac20 2977 bucket, src, srcp, dest, destp, sp->sk_state,
31e6d363 2978 sk_wmem_alloc_get(sp),
6c206b20 2979 udp_rqueue_get(sp),
a7cb5a49
EB		 2980 0, 0L, 0,
2981 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2982 0, sock_i_ino(sp),
41c6d650 2983 refcount_read(&sp->sk_refcnt), sp,
652586df 2984 atomic_read(&sp->sk_drops));
db8dac20
DM		 2985}
2986
2987int udp4_seq_show(struct seq_file *seq, void *v)
2988{
652586df 2989 seq_setwidth(seq, 127);
db8dac20 2990 if (v == SEQ_START_TOKEN)
652586df 2991 seq_puts(seq, " sl local_address rem_address st tx_queue "
db8dac20 2992 "rx_queue tr tm->when retrnsmt uid timeout "
cb61cb9b 2993 "inode ref pointer drops");
db8dac20 2994 else {
db8dac20
DM		 2995 struct udp_iter_state *state = seq->private;
2996
652586df 2997 udp4_format_sock(v, seq, state->bucket);
db8dac20 2998 }
652586df 2999 seq_pad(seq, '\n');
db8dac20
DM		 3000 return 0;
3001}
3002
5788b3a0
YS		 3003#ifdef CONFIG_BPF_SYSCALL
3004struct bpf_iter__udp {
3005 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3006 __bpf_md_ptr(struct udp_sock *, udp_sk);
3007 uid_t uid __aligned(8);
3008 int bucket __aligned(8);
3009};
3010
3011static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3012 struct udp_sock *udp_sk, uid_t uid, int bucket)
3013{
3014 struct bpf_iter__udp ctx;
3015
3016 meta->seq_num--; /* skip SEQ_START_TOKEN */
3017 ctx.meta = meta;
3018 ctx.udp_sk = udp_sk;
3019 ctx.uid = uid;
3020 ctx.bucket = bucket;
3021 return bpf_iter_run_prog(prog, &ctx);
3022}
3023
3024static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3025{
3026 struct udp_iter_state *state = seq->private;
3027 struct bpf_iter_meta meta;
3028 struct bpf_prog *prog;
3029 struct sock *sk = v;
3030 uid_t uid;
3031
3032 if (v == SEQ_START_TOKEN)
3033 return 0;
3034
3035 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3036 meta.seq = seq;
3037 prog = bpf_iter_get_info(&meta, false);
3038 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3039}
3040
3041static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3042{
3043 struct bpf_iter_meta meta;
3044 struct bpf_prog *prog;
3045
3046 if (!v) {
3047 meta.seq = seq;
3048 prog = bpf_iter_get_info(&meta, true);
3049 if (prog)
3050 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3051 }
3052
3053 udp_seq_stop(seq, v);
3054}
3055
3056static const struct seq_operations bpf_iter_udp_seq_ops = {
3057 .start = udp_seq_start,
3058 .next = udp_seq_next,
3059 .stop = bpf_iter_udp_seq_stop,
3060 .show = bpf_iter_udp_seq_show,
3061};
3062#endif
3063
c3506372 3064const struct seq_operations udp_seq_ops = {
a3d2599b
CH		 3065 .start = udp_seq_start,
3066 .next = udp_seq_next,
3067 .stop = udp_seq_stop,
3068 .show = udp4_seq_show,
3069};
c3506372 3070EXPORT_SYMBOL(udp_seq_ops);
73cb88ec 3071
db8dac20 3072static struct udp_seq_afinfo udp4_seq_afinfo = {
db8dac20 3073 .family = AF_INET,
645ca708 3074 .udp_table = &udp_table,
db8dac20
DM		 3075};
3076
2c8c1e72 3077static int __net_init udp4_proc_init_net(struct net *net)
15439feb 3078{
c3506372
CH		 3079 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3080 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
a3d2599b
CH		 3081 return -ENOMEM;
3082 return 0;
15439feb
PE		 3083}
3084
2c8c1e72 3085static void __net_exit udp4_proc_exit_net(struct net *net)
15439feb 3086{
a3d2599b 3087 remove_proc_entry("udp", net->proc_net);
15439feb
PE		 3088}
3089
3090static struct pernet_operations udp4_net_ops = {
3091 .init = udp4_proc_init_net,
3092 .exit = udp4_proc_exit_net,
3093};
3094
db8dac20
DM		 3095int __init udp4_proc_init(void)
3096{
15439feb 3097 return register_pernet_subsys(&udp4_net_ops);
db8dac20
DM		 3098}
3099
3100void udp4_proc_exit(void)
3101{
15439feb 3102 unregister_pernet_subsys(&udp4_net_ops);
db8dac20 3103}
1da177e4
LT		 3104#endif /* CONFIG_PROC_FS */
3105
f86dcc5a
ED		 3106static __initdata unsigned long uhash_entries;
3107static int __init set_uhash_entries(char *str)
645ca708 3108{
413c27d8
EZ		 3109 ssize_t ret;
3110
f86dcc5a
ED		 3111 if (!str)
3112 return 0;
413c27d8
EZ		 3113
3114 ret = kstrtoul(str, 0, &uhash_entries);
3115 if (ret)
3116 return 0;
3117
f86dcc5a
ED		 3118 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3119 uhash_entries = UDP_HTABLE_SIZE_MIN;
3120 return 1;
3121}
3122__setup("uhash_entries=", set_uhash_entries);
645ca708 3123
f86dcc5a
ED		 3124void __init udp_table_init(struct udp_table *table, const char *name)
3125{
3126 unsigned int i;
3127
31fe62b9
TB		 3128 table->hash = alloc_large_system_hash(name,
3129 2 * sizeof(struct udp_hslot),
3130 uhash_entries,
3131 21, /* one slot per 2 MB */
3132 0,
3133 &table->log,
3134 &table->mask,
3135 UDP_HTABLE_SIZE_MIN,
3136 64 * 1024);
3137
512615b6 3138 table->hash2 = table->hash + (table->mask + 1);
f86dcc5a 3139 for (i = 0; i <= table->mask; i++) {
ca065d0c 3140 INIT_HLIST_HEAD(&table->hash[i].head);
fdcc8aa9 3141 table->hash[i].count = 0;
645ca708
ED		 3142 spin_lock_init(&table->hash[i].lock);
3143 }
512615b6 3144 for (i = 0; i <= table->mask; i++) {
ca065d0c 3145 INIT_HLIST_HEAD(&table->hash2[i].head);
512615b6
ED		 3146 table->hash2[i].count = 0;
3147 spin_lock_init(&table->hash2[i].lock);
3148 }
645ca708
ED		 3149}
3150
723b8e46
TH		 3151u32 udp_flow_hashrnd(void)
3152{
3153 static u32 hashrnd __read_mostly;
3154
3155 net_get_random_once(&hashrnd, sizeof(hashrnd));
3156
3157 return hashrnd;
3158}
3159EXPORT_SYMBOL(udp_flow_hashrnd);
3160
1e802951
TZ		 3161static void __udp_sysctl_init(struct net *net)
3162{
3163 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3164 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3165
3166#ifdef CONFIG_NET_L3_MASTER_DEV
3167 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3168#endif
3169}
3170
3171static int __net_init udp_sysctl_init(struct net *net)
3172{
3173 __udp_sysctl_init(net);
3174 return 0;
3175}
3176
3177static struct pernet_operations __net_initdata udp_sysctl_ops = {
fc18999e 3178 .init = udp_sysctl_init,
1e802951
TZ		 3179};
3180
5788b3a0
YS		 3181#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3182DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3183 struct udp_sock *udp_sk, uid_t uid, int bucket)
3184
3185static int bpf_iter_init_udp(void *priv_data)
3186{
3187 struct udp_iter_state *st = priv_data;
3188 struct udp_seq_afinfo *afinfo;
3189 int ret;
3190
3191 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3192 if (!afinfo)
3193 return -ENOMEM;
3194
3195 afinfo->family = AF_UNSPEC;
3196 afinfo->udp_table = &udp_table;
3197 st->bpf_seq_afinfo = afinfo;
3198 ret = bpf_iter_init_seq_net(priv_data);
3199 if (ret)
3200 kfree(afinfo);
3201 return ret;
3202}
3203
3204static void bpf_iter_fini_udp(void *priv_data)
3205{
3206 struct udp_iter_state *st = priv_data;
3207
3208 kfree(st->bpf_seq_afinfo);
3209 bpf_iter_fini_seq_net(priv_data);
3210}
3211
951cf368 3212static struct bpf_iter_reg udp_reg_info = {
5788b3a0
YS		 3213 .target = "udp",
3214 .seq_ops = &bpf_iter_udp_seq_ops,
3215 .init_seq_private = bpf_iter_init_udp,
3216 .fini_seq_private = bpf_iter_fini_udp,
3217 .seq_priv_size = sizeof(struct udp_iter_state),
3218 .ctx_arg_info_size = 1,
3219 .ctx_arg_info = {
3220 { offsetof(struct bpf_iter__udp, udp_sk),
3221 PTR_TO_BTF_ID_OR_NULL },
3222 },
3223};
3224
3225static void __init bpf_iter_register(void)
3226{
951cf368 3227 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
5788b3a0
YS		 3228 if (bpf_iter_reg_target(&udp_reg_info))
3229 pr_warn("Warning: could not register bpf iterator udp\n");
3230}
3231#endif
3232
95766fff
HA		 3233void __init udp_init(void)
3234{
f03d78db 3235 unsigned long limit;
4b272750 3236 unsigned int i;
95766fff 3237
f86dcc5a 3238 udp_table_init(&udp_table, "UDP");
f03d78db 3239 limit = nr_free_buffer_pages() / 8;
95766fff
HA		 3240 limit = max(limit, 128UL);
3241 sysctl_udp_mem[0] = limit / 4 * 3;
3242 sysctl_udp_mem[1] = limit;
3243 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3244
1e802951 3245 __udp_sysctl_init(&init_net);
4b272750
ED		 3246
3247 /* 16 spinlocks per cpu */
3248 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3249 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3250 GFP_KERNEL);
3251 if (!udp_busylocks)
3252 panic("UDP: failed to alloc udp_busylocks\n");
3253 for (i = 0; i < (1U << udp_busylocks_log); i++)
3254 spin_lock_init(udp_busylocks + i);
1e802951
TZ		 3255
3256 if (register_pernet_subsys(&udp_sysctl_ops))
3257 panic("UDP: failed to init sysctl parameters.\n");
5788b3a0
YS		 3258
3259#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3260 bpf_iter_register();
3261#endif
95766fff 3262}
Linux 6.x block layer and io_uring tree(s)