projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
treewide: use get_random_u32() when possible
[linux-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
aef2feda 77#include <linux/bpf-cgroup.h>
7c0f6ba6 78#include <linux/uaccess.h>
1da177e4 79#include <asm/ioctls.h>
57c8a661 80#include <linux/memblock.h>
8203efb3 81#include <linux/highmem.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
91b6d325 126atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
95766fff 127EXPORT_SYMBOL(udp_memory_allocated);
0defbb0a
ED		 128DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
129EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
95766fff 130
f86dcc5a
ED		 131#define MAX_UDP_PORTS 65536
132#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
98322f22 133
f24d43c0 134static int udp_lib_lport_inuse(struct net *net, __u16 num,
645ca708 135 const struct udp_hslot *hslot,
98322f22 136 unsigned long *bitmap,
fe38d2a1 137 struct sock *sk, unsigned int log)
1da177e4 138{
f24d43c0 139 struct sock *sk2;
ba418fa3 140 kuid_t uid = sock_i_uid(sk);
25030a7f 141
ca065d0c 142 sk_for_each(sk2, &hslot->head) {
9d4fb27d
JP		 143 if (net_eq(sock_net(sk2), net) &&
144 sk2 != sk &&
d4cada4a 145 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
9d4fb27d
JP		 146 (!sk2->sk_reuse || !sk->sk_reuse) &&
147 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
148 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
fe38d2a1 149 inet_rcv_saddr_equal(sk, sk2, true)) {
df560056
EG		 150 if (sk2->sk_reuseport && sk->sk_reuseport &&
151 !rcu_access_pointer(sk->sk_reuseport_cb) &&
152 uid_eq(uid, sock_i_uid(sk2))) {
153 if (!bitmap)
154 return 0;
155 } else {
156 if (!bitmap)
157 return 1;
158 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
159 bitmap);
160 }
98322f22 161 }
4243cdc2 162 }
25030a7f
GR		 163 return 0;
164}
165
30fff923
ED		 166/*
167 * Note: we still hold spinlock of primary hash chain, so no other writer
168 * can insert/delete a socket with local_port == num
169 */
170static int udp_lib_lport_inuse2(struct net *net, __u16 num,
4243cdc2 171 struct udp_hslot *hslot2,
fe38d2a1 172 struct sock *sk)
30fff923
ED		 173{
174 struct sock *sk2;
ba418fa3 175 kuid_t uid = sock_i_uid(sk);
30fff923
ED		 176 int res = 0;
177
178 spin_lock(&hslot2->lock);
ca065d0c 179 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
9d4fb27d
JP		 180 if (net_eq(sock_net(sk2), net) &&
181 sk2 != sk &&
182 (udp_sk(sk2)->udp_port_hash == num) &&
183 (!sk2->sk_reuse || !sk->sk_reuse) &&
184 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
185 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
fe38d2a1 186 inet_rcv_saddr_equal(sk, sk2, true)) {
df560056
EG		 187 if (sk2->sk_reuseport && sk->sk_reuseport &&
188 !rcu_access_pointer(sk->sk_reuseport_cb) &&
189 uid_eq(uid, sock_i_uid(sk2))) {
190 res = 0;
191 } else {
192 res = 1;
193 }
30fff923
ED		 194 break;
195 }
4243cdc2 196 }
30fff923
ED		 197 spin_unlock(&hslot2->lock);
198 return res;
199}
200
fe38d2a1 201static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
e32ea7e7
CG		 202{
203 struct net *net = sock_net(sk);
e32ea7e7
CG		 204 kuid_t uid = sock_i_uid(sk);
205 struct sock *sk2;
206
ca065d0c 207 sk_for_each(sk2, &hslot->head) {
e32ea7e7
CG		 208 if (net_eq(sock_net(sk2), net) &&
209 sk2 != sk &&
210 sk2->sk_family == sk->sk_family &&
211 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
212 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
213 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
214 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
fe38d2a1 215 inet_rcv_saddr_equal(sk, sk2, false)) {
2dbb9b9e
MKL		 216 return reuseport_add_sock(sk, sk2,
217 inet_rcv_saddr_any(sk));
e32ea7e7
CG		 218 }
219 }
220
2dbb9b9e 221 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
e32ea7e7
CG		 222}
223
25030a7f 224/**
6ba5a3c5 225 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
25030a7f
GR		 226 *
227 * @sk: socket struct in question
228 * @snum: port number to look up
25985edc 229 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
30fff923 230 * with NULL address
25030a7f 231 */
6ba5a3c5 232int udp_lib_get_port(struct sock *sk, unsigned short snum,
30fff923 233 unsigned int hash2_nulladdr)
25030a7f 234{
512615b6 235 struct udp_hslot *hslot, *hslot2;
645ca708 236 struct udp_table *udptable = sk->sk_prot->h.udp_table;
25030a7f 237 int error = 1;
3b1e0a65 238 struct net *net = sock_net(sk);
1da177e4 239
32c1da70 240 if (!snum) {
9088c560 241 int low, high, remaining;
95c96174 242 unsigned int rand;
98322f22
ED		 243 unsigned short first, last;
244 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
32c1da70 245
0bbf87d8 246 inet_get_local_port_range(net, &low, &high);
a25de534 247 remaining = (high - low) + 1;
227b60f5 248
a251c17a 249 rand = get_random_u32();
8fc54f68 250 first = reciprocal_scale(rand, remaining) + low;
98322f22
ED		 251 /*
252 * force rand to be an odd multiple of UDP_HTABLE_SIZE
253 */
f86dcc5a 254 rand = (rand | 1) * (udptable->mask + 1);
5781b235
ED		 255 last = first + udptable->mask + 1;
256 do {
f86dcc5a 257 hslot = udp_hashslot(udptable, net, first);
98322f22 258 bitmap_zero(bitmap, PORTS_PER_CHAIN);
645ca708 259 spin_lock_bh(&hslot->lock);
98322f22 260 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
fe38d2a1 261 udptable->log);
98322f22
ED		 262
263 snum = first;
264 /*
265 * Iterate on all possible values of snum for this hash.
266 * Using steps of an odd multiple of UDP_HTABLE_SIZE
267 * give us randomization and full range coverage.
268 */
9088c560 269 do {
98322f22 270 if (low <= snum && snum <= high &&
e3826f1e 271 !test_bit(snum >> udptable->log, bitmap) &&
122ff243 272 !inet_is_local_reserved_port(net, snum))
98322f22
ED		 273 goto found;
274 snum += rand;
275 } while (snum != first);
276 spin_unlock_bh(&hslot->lock);
df560056 277 cond_resched();
5781b235 278 } while (++first != last);
98322f22 279 goto fail;
645ca708 280 } else {
f86dcc5a 281 hslot = udp_hashslot(udptable, net, snum);
645ca708 282 spin_lock_bh(&hslot->lock);
30fff923
ED		 283 if (hslot->count > 10) {
284 int exist;
285 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
286
287 slot2 &= udptable->mask;
288 hash2_nulladdr &= udptable->mask;
289
290 hslot2 = udp_hashslot2(udptable, slot2);
291 if (hslot->count < hslot2->count)
292 goto scan_primary_hash;
293
fe38d2a1 294 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
30fff923
ED		 295 if (!exist && (hash2_nulladdr != slot2)) {
296 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
297 exist = udp_lib_lport_inuse2(net, snum, hslot2,
fe38d2a1 298 sk);
30fff923
ED		 299 }
300 if (exist)
301 goto fail_unlock;
302 else
303 goto found;
304 }
305scan_primary_hash:
fe38d2a1 306 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
645ca708
ED		 307 goto fail_unlock;
308 }
98322f22 309found:
c720c7e8 310 inet_sk(sk)->inet_num = snum;
d4cada4a
ED		 311 udp_sk(sk)->udp_port_hash = snum;
312 udp_sk(sk)->udp_portaddr_hash ^= snum;
1da177e4 313 if (sk_unhashed(sk)) {
e32ea7e7 314 if (sk->sk_reuseport &&
fe38d2a1 315 udp_reuseport_add_sock(sk, hslot)) {
e32ea7e7
CG		 316 inet_sk(sk)->inet_num = 0;
317 udp_sk(sk)->udp_port_hash = 0;
318 udp_sk(sk)->udp_portaddr_hash ^= snum;
319 goto fail_unlock;
320 }
321
ca065d0c 322 sk_add_node_rcu(sk, &hslot->head);
fdcc8aa9 323 hslot->count++;
c29a0bc4 324 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
512615b6
ED		 325
326 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
327 spin_lock(&hslot2->lock);
d894ba18 328 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
1602f49b
DM		 329 sk->sk_family == AF_INET6)
330 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
331 &hslot2->head);
d894ba18 332 else
1602f49b
DM		 333 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
334 &hslot2->head);
512615b6
ED		 335 hslot2->count++;
336 spin_unlock(&hslot2->lock);
1da177e4 337 }
ca065d0c 338 sock_set_flag(sk, SOCK_RCU_FREE);
25030a7f 339 error = 0;
645ca708
ED		 340fail_unlock:
341 spin_unlock_bh(&hslot->lock);
1da177e4 342fail:
25030a7f
GR		 343 return error;
344}
c482c568 345EXPORT_SYMBOL(udp_lib_get_port);
25030a7f 346
6ba5a3c5 347int udp_v4_get_port(struct sock *sk, unsigned short snum)
db8dac20 348{
30fff923 349 unsigned int hash2_nulladdr =
f0b1e64c 350 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
30fff923 351 unsigned int hash2_partial =
f0b1e64c 352 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
30fff923 353
d4cada4a 354 /* precompute partial secondary hash */
30fff923 355 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
fe38d2a1 356 return udp_lib_get_port(sk, snum, hash2_nulladdr);
db8dac20
DM		 357}
358
d1e37288
SX		 359static int compute_score(struct sock *sk, struct net *net,
360 __be32 saddr, __be16 sport,
fb74c277 361 __be32 daddr, unsigned short hnum,
73545373 362 int dif, int sdif)
645ca708 363{
60c04aec
JP		 364 int score;
365 struct inet_sock *inet;
6da5b0f0 366 bool dev_match;
645ca708 367
60c04aec
JP		 368 if (!net_eq(sock_net(sk), net) ||
369 udp_sk(sk)->udp_port_hash != hnum ||
370 ipv6_only_sock(sk))
371 return -1;
645ca708 372
4cdeeee9
PO		 373 if (sk->sk_rcv_saddr != daddr)
374 return -1;
60c04aec 375
4cdeeee9 376 score = (sk->sk_family == PF_INET) ? 2 : 1;
60c04aec 377
4cdeeee9 378 inet = inet_sk(sk);
60c04aec
JP		 379 if (inet->inet_daddr) {
380 if (inet->inet_daddr != saddr)
381 return -1;
382 score += 4;
383 }
384
385 if (inet->inet_dport) {
386 if (inet->inet_dport != sport)
387 return -1;
388 score += 4;
389 }
390
6da5b0f0
MM		 391 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
392 dif, sdif);
393 if (!dev_match)
394 return -1;
8d6c414c
MM		 395 if (sk->sk_bound_dev_if)
396 score += 4;
fb74c277 397
7170a977 398 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
70da268b 399 score++;
645ca708
ED		 400 return score;
401}
402
6eada011
ED		 403static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
404 const __u16 lport, const __be32 faddr,
405 const __be16 fport)
65cd8033 406{
1bbdceef
HFS		 407 static u32 udp_ehash_secret __read_mostly;
408
409 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
410
65cd8033 411 return __inet_ehashfn(laddr, lport, faddr, fport,
1bbdceef 412 udp_ehash_secret + net_hash_mix(net));
65cd8033
HFS		 413}
414
a57066b1
DM		 415static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
416 struct sk_buff *skb,
417 __be32 saddr, __be16 sport,
418 __be32 daddr, unsigned short hnum)
7629c73a
JS		 419{
420 struct sock *reuse_sk = NULL;
421 u32 hash;
422
423 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
424 hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
425 reuse_sk = reuseport_select_sock(sk, hash, skb,
426 sizeof(struct udphdr));
7629c73a
JS		 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);
a57066b1
DM		 450 /* Fall back to scoring if group has connections */
451 if (result && !reuseport_has_conns(sk, false))
7629c73a
JS		 452 return result;
453
a57066b1 454 result = result ? : sk;
ca065d0c 455 badness = score;
5051ebd2
ED		 456 }
457 }
5051ebd2
ED		 458 return result;
459}
460
a57066b1
DM		 461static struct sock *udp4_lookup_run_bpf(struct net *net,
462 struct udp_table *udptable,
463 struct sk_buff *skb,
464 __be32 saddr, __be16 sport,
f8931565 465 __be32 daddr, u16 hnum, const int dif)
72f7e944
JS		 466{
467 struct sock *sk, *reuse_sk;
468 bool no_reuseport;
469
470 if (udptable != &udp_table)
471 return NULL; /* only UDP is supported */
472
f8931565
MP		 473 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport,
474 daddr, hnum, dif, &sk);
72f7e944
JS		 475 if (no_reuseport || IS_ERR_OR_NULL(sk))
476 return sk;
477
478 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
c64c9c28 479 if (reuse_sk)
72f7e944
JS		 480 sk = reuse_sk;
481 return sk;
482}
483
db8dac20
DM		 484/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
485 * harder than this. -DaveM
486 */
fce82338 487struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
fb74c277
DA		 488 __be16 sport, __be32 daddr, __be16 dport, int dif,
489 int sdif, struct udp_table *udptable, struct sk_buff *skb)
db8dac20 490{
db8dac20 491 unsigned short hnum = ntohs(dport);
4cdeeee9
PO		 492 unsigned int hash2, slot2;
493 struct udp_hslot *hslot2;
72f7e944 494 struct sock *result, *sk;
645ca708 495
4cdeeee9
PO		 496 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
497 slot2 = hash2 & udptable->mask;
498 hslot2 = &udptable->hash2[slot2];
499
72f7e944 500 /* Lookup connected or non-wildcard socket */
4cdeeee9
PO		 501 result = udp4_lib_lookup2(net, saddr, sport,
502 daddr, hnum, dif, sdif,
73545373 503 hslot2, skb);
72f7e944
JS		 504 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
505 goto done;
506
507 /* Lookup redirect from BPF */
508 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
509 sk = udp4_lookup_run_bpf(net, udptable, skb,
f8931565 510 saddr, sport, daddr, hnum, dif);
72f7e944
JS		 511 if (sk) {
512 result = sk;
513 goto done;
514 }
db8dac20 515 }
72f7e944
JS		 516
517 /* Got non-wildcard socket or error on first lookup */
518 if (result)
519 goto done;
520
521 /* Lookup wildcard sockets */
522 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
523 slot2 = hash2 & udptable->mask;
524 hslot2 = &udptable->hash2[slot2];
525
526 result = udp4_lib_lookup2(net, saddr, sport,
527 htonl(INADDR_ANY), hnum, dif, sdif,
528 hslot2, skb);
529done:
88e235b8 530 if (IS_ERR(result))
4cdeeee9 531 return NULL;
db8dac20
DM		 532 return result;
533}
fce82338 534EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
db8dac20 535
607c4aaf
KK		 536static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
537 __be16 sport, __be16 dport,
645ca708 538 struct udp_table *udptable)
607c4aaf
KK		 539{
540 const struct iphdr *iph = ip_hdr(skb);
541
ed7cbbce 542 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
8afdd99a 543 iph->daddr, dport, inet_iif(skb),
fb74c277 544 inet_sdif(skb), udptable, skb);
607c4aaf
KK		 545}
546
7b58e63e 547struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
63058308
TH		 548 __be16 sport, __be16 dport)
549{
257a525f
MKL		 550 const struct iphdr *iph = ip_hdr(skb);
551
552 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
553 iph->daddr, dport, inet_iif(skb),
554 inet_sdif(skb), &udp_table, NULL);
63058308 555}
63058308 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
a4a600dd
XL		 602void udp_encap_disable(void)
603{
604 static_branch_dec(&udp_encap_needed_key);
605}
606EXPORT_SYMBOL(udp_encap_disable);
607
e7cc0824
SB		 608/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
609 * through error handlers in encapsulations looking for a match.
610 */
611static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
612{
613 int i;
614
615 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
616 int (*handler)(struct sk_buff *skb, u32 info);
92b95364 617 const struct ip_tunnel_encap_ops *encap;
e7cc0824 618
92b95364
PA		 619 encap = rcu_dereference(iptun_encaps[i]);
620 if (!encap)
e7cc0824 621 continue;
92b95364 622 handler = encap->err_handler;
e7cc0824
SB		 623 if (handler && !handler(skb, info))
624 return 0;
625 }
626
627 return -ENOENT;
628}
629
a36e185e
SB		 630/* Try to match ICMP errors to UDP tunnels by looking up a socket without
631 * reversing source and destination port: this will match tunnels that force the
632 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
633 * lwtunnels might actually break this assumption by being configured with
634 * different destination ports on endpoints, in this case we won't be able to
635 * trace ICMP messages back to them.
636 *
e7cc0824
SB		 637 * If this doesn't match any socket, probe tunnels with arbitrary destination
638 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
639 * we've sent packets to won't necessarily match the local destination port.
640 *
a36e185e
SB		 641 * Then ask the tunnel implementation to match the error against a valid
642 * association.
643 *
e7cc0824
SB		 644 * Return an error if we can't find a match, the socket if we need further
645 * processing, zero otherwise.
a36e185e
SB		 646 */
647static struct sock *__udp4_lib_err_encap(struct net *net,
648 const struct iphdr *iph,
649 struct udphdr *uh,
650 struct udp_table *udptable,
9bfce73c 651 struct sock *sk,
e7cc0824 652 struct sk_buff *skb, u32 info)
a36e185e 653{
9bfce73c 654 int (*lookup)(struct sock *sk, struct sk_buff *skb);
a36e185e 655 int network_offset, transport_offset;
9bfce73c 656 struct udp_sock *up;
a36e185e 657
a36e185e
SB		 658 network_offset = skb_network_offset(skb);
659 transport_offset = skb_transport_offset(skb);
660
661 /* Network header needs to point to the outer IPv4 header inside ICMP */
662 skb_reset_network_header(skb);
663
664 /* Transport header needs to point to the UDP header */
665 skb_set_transport_header(skb, iph->ihl << 2);
666
9bfce73c
VF		 667 if (sk) {
668 up = udp_sk(sk);
669
670 lookup = READ_ONCE(up->encap_err_lookup);
671 if (lookup && lookup(sk, skb))
672 sk = NULL;
673
674 goto out;
675 }
676
e7cc0824
SB		 677 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
678 iph->saddr, uh->dest, skb->dev->ifindex, 0,
679 udptable, NULL);
680 if (sk) {
9bfce73c 681 up = udp_sk(sk);
e7cc0824
SB		 682
683 lookup = READ_ONCE(up->encap_err_lookup);
684 if (!lookup || lookup(sk, skb))
685 sk = NULL;
686 }
687
9bfce73c 688out:
e7cc0824
SB		 689 if (!sk)
690 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
a36e185e
SB		 691
692 skb_set_transport_header(skb, transport_offset);
693 skb_set_network_header(skb, network_offset);
694
695 return sk;
696}
697
db8dac20
DM		 698/*
699 * This routine is called by the ICMP module when it gets some
700 * sort of error condition. If err < 0 then the socket should
701 * be closed and the error returned to the user. If err > 0
702 * it's just the icmp type << 8 | icmp code.
703 * Header points to the ip header of the error packet. We move
704 * on past this. Then (as it used to claim before adjustment)
705 * header points to the first 8 bytes of the udp header. We need
706 * to find the appropriate port.
707 */
708
32bbd879 709int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
db8dac20
DM		 710{
711 struct inet_sock *inet;
b71d1d42 712 const struct iphdr *iph = (const struct iphdr *)skb->data;
c482c568 713 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
db8dac20
DM		 714 const int type = icmp_hdr(skb)->type;
715 const int code = icmp_hdr(skb)->code;
a36e185e 716 bool tunnel = false;
db8dac20
DM		 717 struct sock *sk;
718 int harderr;
719 int err;
fd54d716 720 struct net *net = dev_net(skb->dev);
db8dac20 721
fd54d716 722 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
f64bf6b8
MM		 723 iph->saddr, uh->source, skb->dev->ifindex,
724 inet_sdif(skb), udptable, NULL);
9bfce73c 725
d26796ae 726 if (!sk || udp_sk(sk)->encap_type) {
a36e185e 727 /* No socket for error: try tunnels before discarding */
e7cc0824 728 if (static_branch_unlikely(&udp_encap_needed_key)) {
9bfce73c 729 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
e7cc0824
SB		 730 info);
731 if (!sk)
732 return 0;
9bfce73c
VF		 733 } else
734 sk = ERR_PTR(-ENOENT);
a36e185e 735
e7cc0824 736 if (IS_ERR(sk)) {
a36e185e 737 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
e7cc0824 738 return PTR_ERR(sk);
a36e185e 739 }
e7cc0824 740
a36e185e 741 tunnel = true;
db8dac20
DM		 742 }
743
744 err = 0;
745 harderr = 0;
746 inet = inet_sk(sk);
747
748 switch (type) {
749 default:
750 case ICMP_TIME_EXCEEDED:
751 err = EHOSTUNREACH;
752 break;
753 case ICMP_SOURCE_QUENCH:
754 goto out;
755 case ICMP_PARAMETERPROB:
756 err = EPROTO;
757 harderr = 1;
758 break;
759 case ICMP_DEST_UNREACH:
760 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
36393395 761 ipv4_sk_update_pmtu(skb, sk, info);
db8dac20
DM		 762 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
763 err = EMSGSIZE;
764 harderr = 1;
765 break;
766 }
767 goto out;
768 }
769 err = EHOSTUNREACH;
770 if (code <= NR_ICMP_UNREACH) {
771 harderr = icmp_err_convert[code].fatal;
772 err = icmp_err_convert[code].errno;
773 }
774 break;
55be7a9c
DM		 775 case ICMP_REDIRECT:
776 ipv4_sk_redirect(skb, sk);
1a462d18 777 goto out;
db8dac20
DM		 778 }
779
780 /*
781 * RFC1122: OK. Passes ICMP errors back to application, as per
782 * 4.1.3.3.
783 */
a36e185e
SB		 784 if (tunnel) {
785 /* ...not for tunnels though: we don't have a sending socket */
ac56a0b4
DH		 786 if (udp_sk(sk)->encap_err_rcv)
787 udp_sk(sk)->encap_err_rcv(sk, skb, iph->ihl << 2);
a36e185e
SB		 788 goto out;
789 }
db8dac20
DM		 790 if (!inet->recverr) {
791 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
792 goto out;
b1faf566 793 } else
c482c568 794 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
b1faf566 795
db8dac20 796 sk->sk_err = err;
e3ae2365 797 sk_error_report(sk);
db8dac20 798out:
32bbd879 799 return 0;
db8dac20
DM		 800}
801
32bbd879 802int udp_err(struct sk_buff *skb, u32 info)
db8dac20 803{
32bbd879 804 return __udp4_lib_err(skb, info, &udp_table);
db8dac20
DM		 805}
806
807/*
808 * Throw away all pending data and cancel the corking. Socket is locked.
809 */
36d926b9 810void udp_flush_pending_frames(struct sock *sk)
db8dac20
DM		 811{
812 struct udp_sock *up = udp_sk(sk);
813
814 if (up->pending) {
815 up->len = 0;
816 up->pending = 0;
817 ip_flush_pending_frames(sk);
818 }
819}
36d926b9 820EXPORT_SYMBOL(udp_flush_pending_frames);
db8dac20
DM		 821
822/**
f6b9664f 823 * udp4_hwcsum - handle outgoing HW checksumming
db8dac20
DM		 824 * @skb: sk_buff containing the filled-in UDP header
825 * (checksum field must be zeroed out)
f6b9664f
HX		 826 * @src: source IP address
827 * @dst: destination IP address
db8dac20 828 */
c26bf4a5 829void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
db8dac20 830{
db8dac20 831 struct udphdr *uh = udp_hdr(skb);
f6b9664f
HX		 832 int offset = skb_transport_offset(skb);
833 int len = skb->len - offset;
834 int hlen = len;
db8dac20
DM		 835 __wsum csum = 0;
836
ebbe495f 837 if (!skb_has_frag_list(skb)) {
db8dac20
DM		 838 /*
839 * Only one fragment on the socket.
840 */
841 skb->csum_start = skb_transport_header(skb) - skb->head;
842 skb->csum_offset = offsetof(struct udphdr, check);
f6b9664f
HX		 843 uh->check = ~csum_tcpudp_magic(src, dst, len,
844 IPPROTO_UDP, 0);
db8dac20 845 } else {
ebbe495f
WC		 846 struct sk_buff *frags;
847
db8dac20
DM		 848 /*
849 * HW-checksum won't work as there are two or more
850 * fragments on the socket so that all csums of sk_buffs
851 * should be together
852 */
ebbe495f 853 skb_walk_frags(skb, frags) {
f6b9664f
HX		 854 csum = csum_add(csum, frags->csum);
855 hlen -= frags->len;
ebbe495f 856 }
db8dac20 857
f6b9664f 858 csum = skb_checksum(skb, offset, hlen, csum);
db8dac20
DM		 859 skb->ip_summed = CHECKSUM_NONE;
860
db8dac20
DM		 861 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
862 if (uh->check == 0)
863 uh->check = CSUM_MANGLED_0;
864 }
865}
c26bf4a5 866EXPORT_SYMBOL_GPL(udp4_hwcsum);
db8dac20 867
af5fcba7
TH		 868/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
869 * for the simple case like when setting the checksum for a UDP tunnel.
870 */
871void udp_set_csum(bool nocheck, struct sk_buff *skb,
872 __be32 saddr, __be32 daddr, int len)
873{
874 struct udphdr *uh = udp_hdr(skb);
875
179bc67f 876 if (nocheck) {
af5fcba7 877 uh->check = 0;
179bc67f 878 } else if (skb_is_gso(skb)) {
af5fcba7 879 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
179bc67f
EC		 880 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
881 uh->check = 0;
882 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
883 if (uh->check == 0)
884 uh->check = CSUM_MANGLED_0;
d75f1306 885 } else {
af5fcba7
TH		 886 skb->ip_summed = CHECKSUM_PARTIAL;
887 skb->csum_start = skb_transport_header(skb) - skb->head;
888 skb->csum_offset = offsetof(struct udphdr, check);
889 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
af5fcba7
TH		 890 }
891}
892EXPORT_SYMBOL(udp_set_csum);
893
bec1f6f6
WB		 894static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
895 struct inet_cork *cork)
db8dac20 896{
f6b9664f 897 struct sock *sk = skb->sk;
db8dac20 898 struct inet_sock *inet = inet_sk(sk);
db8dac20 899 struct udphdr *uh;
cffb8f61 900 int err;
db8dac20 901 int is_udplite = IS_UDPLITE(sk);
f6b9664f
HX		 902 int offset = skb_transport_offset(skb);
903 int len = skb->len - offset;
4094871d 904 int datalen = len - sizeof(*uh);
db8dac20
DM		 905 __wsum csum = 0;
906
db8dac20
DM		 907 /*
908 * Create a UDP header
909 */
910 uh = udp_hdr(skb);
f6b9664f 911 uh->source = inet->inet_sport;
79ab0531 912 uh->dest = fl4->fl4_dport;
f6b9664f 913 uh->len = htons(len);
db8dac20
DM		 914 uh->check = 0;
915
bec1f6f6
WB		 916 if (cork->gso_size) {
917 const int hlen = skb_network_header_len(skb) +
918 sizeof(struct udphdr);
919
0f149c9f
WB		 920 if (hlen + cork->gso_size > cork->fragsize) {
921 kfree_skb(skb);
bec1f6f6 922 return -EINVAL;
0f149c9f 923 }
158390e4 924 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
0f149c9f 925 kfree_skb(skb);
bec1f6f6 926 return -EINVAL;
0f149c9f
WB		 927 }
928 if (sk->sk_no_check_tx) {
929 kfree_skb(skb);
a8c744a8 930 return -EINVAL;
0f149c9f 931 }
ff06342c 932 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
0f149c9f
WB		 933 dst_xfrm(skb_dst(skb))) {
934 kfree_skb(skb);
bec1f6f6 935 return -EIO;
0f149c9f 936 }
bec1f6f6 937
4094871d
JH		 938 if (datalen > cork->gso_size) {
939 skb_shinfo(skb)->gso_size = cork->gso_size;
940 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
941 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
942 cork->gso_size);
943 }
a8c744a8 944 goto csum_partial;
bec1f6f6
WB		 945 }
946
db8dac20 947 if (is_udplite) /* UDP-Lite */
f6b9664f 948 csum = udplite_csum(skb);
db8dac20 949
ab2fb7e3 950 else if (sk->sk_no_check_tx) { /* UDP csum off */
db8dac20
DM		 951
952 skb->ip_summed = CHECKSUM_NONE;
953 goto send;
954
955 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
a8c744a8 956csum_partial:
db8dac20 957
79ab0531 958 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
db8dac20
DM		 959 goto send;
960
f6b9664f
HX		 961 } else
962 csum = udp_csum(skb);
db8dac20
DM		 963
964 /* add protocol-dependent pseudo-header */
79ab0531 965 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
c482c568 966 sk->sk_protocol, csum);
db8dac20
DM		 967 if (uh->check == 0)
968 uh->check = CSUM_MANGLED_0;
969
970send:
b5ec8eea 971 err = ip_send_skb(sock_net(sk), skb);
6ce9e7b5
ED		 972 if (err) {
973 if (err == -ENOBUFS && !inet->recverr) {
6aef70a8
ED		 974 UDP_INC_STATS(sock_net(sk),
975 UDP_MIB_SNDBUFERRORS, is_udplite);
6ce9e7b5
ED		 976 err = 0;
977 }
978 } else
6aef70a8
ED		 979 UDP_INC_STATS(sock_net(sk),
980 UDP_MIB_OUTDATAGRAMS, is_udplite);
f6b9664f
HX		 981 return err;
982}
983
984/*
985 * Push out all pending data as one UDP datagram. Socket is locked.
986 */
8822b64a 987int udp_push_pending_frames(struct sock *sk)
f6b9664f
HX		 988{
989 struct udp_sock *up = udp_sk(sk);
990 struct inet_sock *inet = inet_sk(sk);
b6f21b26 991 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
f6b9664f
HX		 992 struct sk_buff *skb;
993 int err = 0;
994
77968b78 995 skb = ip_finish_skb(sk, fl4);
f6b9664f
HX		 996 if (!skb)
997 goto out;
998
bec1f6f6 999 err = udp_send_skb(skb, fl4, &inet->cork.base);
f6b9664f 1000
db8dac20
DM		 1001out:
1002 up->len = 0;
1003 up->pending = 0;
db8dac20
DM		 1004 return err;
1005}
8822b64a 1006EXPORT_SYMBOL(udp_push_pending_frames);
db8dac20 1007
2e8de857
WB		 1008static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1009{
1010 switch (cmsg->cmsg_type) {
1011 case UDP_SEGMENT:
1012 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1013 return -EINVAL;
1014 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1015 return 0;
1016 default:
1017 return -EINVAL;
1018 }
1019}
1020
1021int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1022{
1023 struct cmsghdr *cmsg;
1024 bool need_ip = false;
1025 int err;
1026
1027 for_each_cmsghdr(cmsg, msg) {
1028 if (!CMSG_OK(msg, cmsg))
1029 return -EINVAL;
1030
1031 if (cmsg->cmsg_level != SOL_UDP) {
1032 need_ip = true;
1033 continue;
1034 }
1035
1036 err = __udp_cmsg_send(cmsg, gso_size);
1037 if (err)
1038 return err;
1039 }
1040
1041 return need_ip;
1042}
1043EXPORT_SYMBOL_GPL(udp_cmsg_send);
1044
1b784140 1045int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
db8dac20
DM		 1046{
1047 struct inet_sock *inet = inet_sk(sk);
1048 struct udp_sock *up = udp_sk(sk);
1cedee13 1049 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
e474995f 1050 struct flowi4 fl4_stack;
b6f21b26 1051 struct flowi4 *fl4;
db8dac20
DM		 1052 int ulen = len;
1053 struct ipcm_cookie ipc;
1054 struct rtable *rt = NULL;
1055 int free = 0;
1056 int connected = 0;
1057 __be32 daddr, faddr, saddr;
1058 __be16 dport;
1059 u8 tos;
1060 int err, is_udplite = IS_UDPLITE(sk);
a9f59707 1061 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
db8dac20 1062 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
903ab86d 1063 struct sk_buff *skb;
f6d8bd05 1064 struct ip_options_data opt_copy;
db8dac20
DM		 1065
1066 if (len > 0xFFFF)
1067 return -EMSGSIZE;
1068
1069 /*
1070 * Check the flags.
1071 */
1072
c482c568 1073 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
db8dac20
DM		 1074 return -EOPNOTSUPP;
1075
903ab86d
HX		 1076 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1077
f5fca608 1078 fl4 = &inet->cork.fl.u.ip4;
db8dac20
DM		 1079 if (up->pending) {
1080 /*
1081 * There are pending frames.
1082 * The socket lock must be held while it's corked.
1083 */
1084 lock_sock(sk);
1085 if (likely(up->pending)) {
1086 if (unlikely(up->pending != AF_INET)) {
1087 release_sock(sk);
1088 return -EINVAL;
1089 }
1090 goto do_append_data;
1091 }
1092 release_sock(sk);
1093 }
1094 ulen += sizeof(struct udphdr);
1095
1096 /*
1097 * Get and verify the address.
1098 */
1cedee13 1099 if (usin) {
db8dac20
DM		 1100 if (msg->msg_namelen < sizeof(*usin))
1101 return -EINVAL;
1102 if (usin->sin_family != AF_INET) {
1103 if (usin->sin_family != AF_UNSPEC)
1104 return -EAFNOSUPPORT;
1105 }
1106
1107 daddr = usin->sin_addr.s_addr;
1108 dport = usin->sin_port;
1109 if (dport == 0)
1110 return -EINVAL;
1111 } else {
1112 if (sk->sk_state != TCP_ESTABLISHED)
1113 return -EDESTADDRREQ;
c720c7e8
ED		 1114 daddr = inet->inet_daddr;
1115 dport = inet->inet_dport;
db8dac20
DM		 1116 /* Open fast path for connected socket.
1117 Route will not be used, if at least one option is set.
1118 */
1119 connected = 1;
1120 }
db8dac20 1121
35178206 1122 ipcm_init_sk(&ipc, inet);
18a419ba 1123 ipc.gso_size = READ_ONCE(up->gso_size);
bf84a010 1124
db8dac20 1125 if (msg->msg_controllen) {
2e8de857
WB		 1126 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1127 if (err > 0)
1128 err = ip_cmsg_send(sk, msg, &ipc,
1129 sk->sk_family == AF_INET6);
1130 if (unlikely(err < 0)) {
91948309 1131 kfree(ipc.opt);
db8dac20 1132 return err;
91948309 1133 }
db8dac20
DM		 1134 if (ipc.opt)
1135 free = 1;
1136 connected = 0;
1137 }
f6d8bd05
ED		 1138 if (!ipc.opt) {
1139 struct ip_options_rcu *inet_opt;
1140
1141 rcu_read_lock();
1142 inet_opt = rcu_dereference(inet->inet_opt);
1143 if (inet_opt) {
1144 memcpy(&opt_copy, inet_opt,
1145 sizeof(*inet_opt) + inet_opt->opt.optlen);
1146 ipc.opt = &opt_copy.opt;
1147 }
1148 rcu_read_unlock();
1149 }
db8dac20 1150
6fc88c35 1151 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1cedee13
AI		 1152 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1153 (struct sockaddr *)usin, &ipc.addr);
1154 if (err)
1155 goto out_free;
1156 if (usin) {
1157 if (usin->sin_port == 0) {
1158 /* BPF program set invalid port. Reject it. */
1159 err = -EINVAL;
1160 goto out_free;
1161 }
1162 daddr = usin->sin_addr.s_addr;
1163 dport = usin->sin_port;
1164 }
1165 }
1166
db8dac20
DM		 1167 saddr = ipc.addr;
1168 ipc.addr = faddr = daddr;
1169
f6d8bd05 1170 if (ipc.opt && ipc.opt->opt.srr) {
1b97013b
AI		 1171 if (!daddr) {
1172 err = -EINVAL;
1173 goto out_free;
1174 }
f6d8bd05 1175 faddr = ipc.opt->opt.faddr;
db8dac20
DM		 1176 connected = 0;
1177 }
aa661581 1178 tos = get_rttos(&ipc, inet);
db8dac20
DM		 1179 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1180 (msg->msg_flags & MSG_DONTROUTE) ||
f6d8bd05 1181 (ipc.opt && ipc.opt->opt.is_strictroute)) {
db8dac20
DM		 1182 tos |= RTO_ONLINK;
1183 connected = 0;
1184 }
1185
1186 if (ipv4_is_multicast(daddr)) {
854da991 1187 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
db8dac20
DM		 1188 ipc.oif = inet->mc_index;
1189 if (!saddr)
1190 saddr = inet->mc_addr;
1191 connected = 0;
9515a2e0 1192 } else if (!ipc.oif) {
76e21053 1193 ipc.oif = inet->uc_index;
9515a2e0
DA		 1194 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1195 /* oif is set, packet is to local broadcast and
2bdcc73c 1196 * uc_index is set. oif is most likely set
9515a2e0
DA		 1197 * by sk_bound_dev_if. If uc_index != oif check if the
1198 * oif is an L3 master and uc_index is an L3 slave.
1199 * If so, we want to allow the send using the uc_index.
1200 */
1201 if (ipc.oif != inet->uc_index &&
1202 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1203 inet->uc_index)) {
1204 ipc.oif = inet->uc_index;
1205 }
1206 }
db8dac20
DM		 1207
1208 if (connected)
c482c568 1209 rt = (struct rtable *)sk_dst_check(sk, 0);
db8dac20 1210
51456b29 1211 if (!rt) {
84a3aa00 1212 struct net *net = sock_net(sk);
9a24abfa 1213 __u8 flow_flags = inet_sk_flowi_flags(sk);
84a3aa00 1214
e474995f 1215 fl4 = &fl4_stack;
9a24abfa 1216
c6af0c22 1217 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
c0951cbc 1218 RT_SCOPE_UNIVERSE, sk->sk_protocol,
9a24abfa 1219 flow_flags,
e2d118a1
LC		 1220 faddr, saddr, dport, inet->inet_sport,
1221 sk->sk_uid);
c0951cbc 1222
3df98d79 1223 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
e474995f 1224 rt = ip_route_output_flow(net, fl4, sk);
b23dd4fe
DM		 1225 if (IS_ERR(rt)) {
1226 err = PTR_ERR(rt);
06dc94b1 1227 rt = NULL;
db8dac20 1228 if (err == -ENETUNREACH)
f1d8cba6 1229 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
db8dac20
DM		 1230 goto out;
1231 }
1232
1233 err = -EACCES;
1234 if ((rt->rt_flags & RTCF_BROADCAST) &&
1235 !sock_flag(sk, SOCK_BROADCAST))
1236 goto out;
1237 if (connected)
d8d1f30b 1238 sk_dst_set(sk, dst_clone(&rt->dst));
db8dac20
DM		 1239 }
1240
1241 if (msg->msg_flags&MSG_CONFIRM)
1242 goto do_confirm;
1243back_from_confirm:
1244
e474995f 1245 saddr = fl4->saddr;
db8dac20 1246 if (!ipc.addr)
e474995f 1247 daddr = ipc.addr = fl4->daddr;
db8dac20 1248
903ab86d
HX		 1249 /* Lockless fast path for the non-corking case. */
1250 if (!corkreq) {
1cd7884d
WB		 1251 struct inet_cork cork;
1252
f69e6d13 1253 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
903ab86d 1254 sizeof(struct udphdr), &ipc, &rt,
1cd7884d 1255 &cork, msg->msg_flags);
903ab86d 1256 err = PTR_ERR(skb);
50c3a487 1257 if (!IS_ERR_OR_NULL(skb))
bec1f6f6 1258 err = udp_send_skb(skb, fl4, &cork);
903ab86d
HX		 1259 goto out;
1260 }
1261
db8dac20
DM		 1262 lock_sock(sk);
1263 if (unlikely(up->pending)) {
1264 /* The socket is already corked while preparing it. */
1265 /* ... which is an evident application bug. --ANK */
1266 release_sock(sk);
1267
197df02c 1268 net_dbg_ratelimited("socket already corked\n");
db8dac20
DM		 1269 err = -EINVAL;
1270 goto out;
1271 }
1272 /*
1273 * Now cork the socket to pend data.
1274 */
b6f21b26
DM		 1275 fl4 = &inet->cork.fl.u.ip4;
1276 fl4->daddr = daddr;
1277 fl4->saddr = saddr;
9cce96df
DM		 1278 fl4->fl4_dport = dport;
1279 fl4->fl4_sport = inet->inet_sport;
db8dac20
DM		 1280 up->pending = AF_INET;
1281
1282do_append_data:
1283 up->len += ulen;
f69e6d13 1284 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
f5fca608
DM		 1285 sizeof(struct udphdr), &ipc, &rt,
1286 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
db8dac20
DM		 1287 if (err)
1288 udp_flush_pending_frames(sk);
1289 else if (!corkreq)
1290 err = udp_push_pending_frames(sk);
1291 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1292 up->pending = 0;
1293 release_sock(sk);
1294
1295out:
1296 ip_rt_put(rt);
1b97013b 1297out_free:
db8dac20
DM		 1298 if (free)
1299 kfree(ipc.opt);
1300 if (!err)
1301 return len;
1302 /*
1303 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1304 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1305 * we don't have a good statistic (IpOutDiscards but it can be too many
1306 * things). We could add another new stat but at least for now that
1307 * seems like overkill.
1308 */
1309 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
6aef70a8
ED		 1310 UDP_INC_STATS(sock_net(sk),
1311 UDP_MIB_SNDBUFERRORS, is_udplite);
db8dac20
DM		 1312 }
1313 return err;
1314
1315do_confirm:
0dec879f
JA		 1316 if (msg->msg_flags & MSG_PROBE)
1317 dst_confirm_neigh(&rt->dst, &fl4->daddr);
db8dac20
DM		 1318 if (!(msg->msg_flags&MSG_PROBE) || len)
1319 goto back_from_confirm;
1320 err = 0;
1321 goto out;
1322}
c482c568 1323EXPORT_SYMBOL(udp_sendmsg);
db8dac20
DM		 1324
1325int udp_sendpage(struct sock *sk, struct page *page, int offset,
1326 size_t size, int flags)
1327{
f5fca608 1328 struct inet_sock *inet = inet_sk(sk);
db8dac20
DM		 1329 struct udp_sock *up = udp_sk(sk);
1330 int ret;
1331
d3f7d56a
SL		 1332 if (flags & MSG_SENDPAGE_NOTLAST)
1333 flags |= MSG_MORE;
1334
db8dac20
DM		 1335 if (!up->pending) {
1336 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1337
1338 /* Call udp_sendmsg to specify destination address which
1339 * sendpage interface can't pass.
1340 * This will succeed only when the socket is connected.
1341 */
1b784140 1342 ret = udp_sendmsg(sk, &msg, 0);
db8dac20
DM		 1343 if (ret < 0)
1344 return ret;
1345 }
1346
1347 lock_sock(sk);
1348
1349 if (unlikely(!up->pending)) {
1350 release_sock(sk);
1351
197df02c 1352 net_dbg_ratelimited("cork failed\n");
db8dac20
DM		 1353 return -EINVAL;
1354 }
1355
f5fca608
DM		 1356 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1357 page, offset, size, flags);
db8dac20
DM		 1358 if (ret == -EOPNOTSUPP) {
1359 release_sock(sk);
1360 return sock_no_sendpage(sk->sk_socket, page, offset,
1361 size, flags);
1362 }
1363 if (ret < 0) {
1364 udp_flush_pending_frames(sk);
1365 goto out;
1366 }
1367
1368 up->len += size;
a9f59707 1369 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
db8dac20
DM		 1370 ret = udp_push_pending_frames(sk);
1371 if (!ret)
1372 ret = size;
1373out:
1374 release_sock(sk);
1375 return ret;
1376}
1377
dce4551c
PA		 1378#define UDP_SKB_IS_STATELESS 0x80000000
1379
677bf08c
FW		 1380/* all head states (dst, sk, nf conntrack) except skb extensions are
1381 * cleared by udp_rcv().
1382 *
1383 * We need to preserve secpath, if present, to eventually process
1384 * IP_CMSG_PASSSEC at recvmsg() time.
1385 *
1386 * Other extensions can be cleared.
1387 */
1388static bool udp_try_make_stateless(struct sk_buff *skb)
1389{
1390 if (!skb_has_extensions(skb))
1391 return true;
1392
1393 if (!secpath_exists(skb)) {
1394 skb_ext_reset(skb);
1395 return true;
1396 }
1397
1398 return false;
1399}
1400
b65ac446
PA		 1401static void udp_set_dev_scratch(struct sk_buff *skb)
1402{
dce4551c 1403 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
b65ac446
PA		 1404
1405 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
dce4551c
PA		 1406 scratch->_tsize_state = skb->truesize;
1407#if BITS_PER_LONG == 64
b65ac446
PA		 1408 scratch->len = skb->len;
1409 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1410 scratch->is_linear = !skb_is_nonlinear(skb);
dce4551c 1411#endif
677bf08c 1412 if (udp_try_make_stateless(skb))
dce4551c 1413 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
b65ac446
PA		 1414}
1415
a793183c
ED		 1416static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1417{
1418 /* We come here after udp_lib_checksum_complete() returned 0.
1419 * This means that __skb_checksum_complete() might have
1420 * set skb->csum_valid to 1.
1421 * On 64bit platforms, we can set csum_unnecessary
1422 * to true, but only if the skb is not shared.
1423 */
1424#if BITS_PER_LONG == 64
1425 if (!skb_shared(skb))
1426 udp_skb_scratch(skb)->csum_unnecessary = true;
1427#endif
1428}
1429
b65ac446
PA		 1430static int udp_skb_truesize(struct sk_buff *skb)
1431{
dce4551c 1432 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
b65ac446
PA		 1433}
1434
dce4551c 1435static bool udp_skb_has_head_state(struct sk_buff *skb)
b65ac446 1436{
dce4551c 1437 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
b65ac446 1438}
b65ac446 1439
7c13f97f 1440/* fully reclaim rmem/fwd memory allocated for skb */
6dfb4367
PA		 1441static void udp_rmem_release(struct sock *sk, int size, int partial,
1442 bool rx_queue_lock_held)
f970bd9e 1443{
6b229cf7 1444 struct udp_sock *up = udp_sk(sk);
2276f58a 1445 struct sk_buff_head *sk_queue;
f970bd9e
PA		 1446 int amt;
1447
6b229cf7
ED		 1448 if (likely(partial)) {
1449 up->forward_deficit += size;
1450 size = up->forward_deficit;
d39ca259
PA		 1451 if (size < (sk->sk_rcvbuf >> 2) &&
1452 !skb_queue_empty(&up->reader_queue))
6b229cf7
ED		 1453 return;
1454 } else {
1455 size += up->forward_deficit;
1456 }
1457 up->forward_deficit = 0;
1458
6dfb4367
PA		 1459 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1460 * if the called don't held it already
1461 */
2276f58a 1462 sk_queue = &sk->sk_receive_queue;
6dfb4367
PA		 1463 if (!rx_queue_lock_held)
1464 spin_lock(&sk_queue->lock);
1465
2276f58a 1466
f970bd9e 1467 sk->sk_forward_alloc += size;
100fdd1f 1468 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
f970bd9e 1469 sk->sk_forward_alloc -= amt;
f970bd9e
PA		 1470
1471 if (amt)
100fdd1f 1472 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
02ab0d13
ED		 1473
1474 atomic_sub(size, &sk->sk_rmem_alloc);
2276f58a
PA		 1475
1476 /* this can save us from acquiring the rx queue lock on next receive */
1477 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1478
6dfb4367
PA		 1479 if (!rx_queue_lock_held)
1480 spin_unlock(&sk_queue->lock);
f970bd9e
PA		 1481}
1482
2276f58a 1483/* Note: called with reader_queue.lock held.
c84d9490
ED		 1484 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1485 * This avoids a cache line miss while receive_queue lock is held.
1486 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1487 */
7c13f97f 1488void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
f970bd9e 1489{
b65ac446
PA		 1490 prefetch(&skb->data);
1491 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
f970bd9e 1492}
7c13f97f 1493EXPORT_SYMBOL(udp_skb_destructor);
f970bd9e 1494
6dfb4367 1495/* as above, but the caller held the rx queue lock, too */
64f5102d 1496static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
6dfb4367 1497{
b65ac446
PA		 1498 prefetch(&skb->data);
1499 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
6dfb4367
PA		 1500}
1501
4b272750
ED		 1502/* Idea of busylocks is to let producers grab an extra spinlock
1503 * to relieve pressure on the receive_queue spinlock shared by consumer.
1504 * Under flood, this means that only one producer can be in line
1505 * trying to acquire the receive_queue spinlock.
1506 * These busylock can be allocated on a per cpu manner, instead of a
1507 * per socket one (that would consume a cache line per socket)
1508 */
1509static int udp_busylocks_log __read_mostly;
1510static spinlock_t *udp_busylocks __read_mostly;
1511
1512static spinlock_t *busylock_acquire(void *ptr)
1513{
1514 spinlock_t *busy;
1515
1516 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1517 spin_lock(busy);
1518 return busy;
1519}
1520
1521static void busylock_release(spinlock_t *busy)
1522{
1523 if (busy)
1524 spin_unlock(busy);
1525}
1526
f970bd9e
PA		 1527int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1528{
1529 struct sk_buff_head *list = &sk->sk_receive_queue;
1530 int rmem, delta, amt, err = -ENOMEM;
4b272750 1531 spinlock_t *busy = NULL;
c8c8b127 1532 int size;
f970bd9e
PA		 1533
1534 /* try to avoid the costly atomic add/sub pair when the receive
1535 * queue is full; always allow at least a packet
1536 */
1537 rmem = atomic_read(&sk->sk_rmem_alloc);
363dc73a 1538 if (rmem > sk->sk_rcvbuf)
f970bd9e
PA		 1539 goto drop;
1540
c8c8b127
ED		 1541 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1542 * having linear skbs :
1543 * - Reduce memory overhead and thus increase receive queue capacity
1544 * - Less cache line misses at copyout() time
1545 * - Less work at consume_skb() (less alien page frag freeing)
1546 */
4b272750 1547 if (rmem > (sk->sk_rcvbuf >> 1)) {
c8c8b127 1548 skb_condense(skb);
4b272750
ED		 1549
1550 busy = busylock_acquire(sk);
1551 }
c8c8b127 1552 size = skb->truesize;
b65ac446 1553 udp_set_dev_scratch(skb);
c8c8b127 1554
f970bd9e
PA		 1555 /* we drop only if the receive buf is full and the receive
1556 * queue contains some other skb
1557 */
1558 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
feed8a4f 1559 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
f970bd9e
PA		 1560 goto uncharge_drop;
1561
1562 spin_lock(&list->lock);
1563 if (size >= sk->sk_forward_alloc) {
1564 amt = sk_mem_pages(size);
100fdd1f 1565 delta = amt << PAGE_SHIFT;
f970bd9e
PA		 1566 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1567 err = -ENOBUFS;
1568 spin_unlock(&list->lock);
1569 goto uncharge_drop;
1570 }
1571
1572 sk->sk_forward_alloc += delta;
1573 }
1574
1575 sk->sk_forward_alloc -= size;
1576
7c13f97f
PA		 1577 /* no need to setup a destructor, we will explicitly release the
1578 * forward allocated memory on dequeue
1579 */
f970bd9e
PA		 1580 sock_skb_set_dropcount(sk, skb);
1581
1582 __skb_queue_tail(list, skb);
1583 spin_unlock(&list->lock);
1584
1585 if (!sock_flag(sk, SOCK_DEAD))
1586 sk->sk_data_ready(sk);
1587
4b272750 1588 busylock_release(busy);
f970bd9e
PA		 1589 return 0;
1590
1591uncharge_drop:
1592 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1593
1594drop:
1595 atomic_inc(&sk->sk_drops);
4b272750 1596 busylock_release(busy);
f970bd9e
PA		 1597 return err;
1598}
1599EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1600
c915fe13 1601void udp_destruct_sock(struct sock *sk)
f970bd9e
PA		 1602{
1603 /* reclaim completely the forward allocated memory */
2276f58a 1604 struct udp_sock *up = udp_sk(sk);
7c13f97f
PA		 1605 unsigned int total = 0;
1606 struct sk_buff *skb;
1607
2276f58a
PA		 1608 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1609 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
7c13f97f
PA		 1610 total += skb->truesize;
1611 kfree_skb(skb);
1612 }
6dfb4367 1613 udp_rmem_release(sk, total, 0, true);
7c13f97f 1614
f970bd9e
PA		 1615 inet_sock_destruct(sk);
1616}
c915fe13 1617EXPORT_SYMBOL_GPL(udp_destruct_sock);
f970bd9e
PA		 1618
1619int udp_init_sock(struct sock *sk)
1620{
2276f58a 1621 skb_queue_head_init(&udp_sk(sk)->reader_queue);
f970bd9e
PA		 1622 sk->sk_destruct = udp_destruct_sock;
1623 return 0;
1624}
1625EXPORT_SYMBOL_GPL(udp_init_sock);
1626
1627void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1628{
1629 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1630 bool slow = lock_sock_fast(sk);
1631
1632 sk_peek_offset_bwd(sk, len);
1633 unlock_sock_fast(sk, slow);
1634 }
0a463c78 1635
ca2c1418
PA		 1636 if (!skb_unref(skb))
1637 return;
1638
dce4551c
PA		 1639 /* In the more common cases we cleared the head states previously,
1640 * see __udp_queue_rcv_skb().
0ddf3fb2 1641 */
dce4551c 1642 if (unlikely(udp_skb_has_head_state(skb)))
0ddf3fb2 1643 skb_release_head_state(skb);
ca2c1418 1644 __consume_stateless_skb(skb);
f970bd9e
PA		 1645}
1646EXPORT_SYMBOL_GPL(skb_consume_udp);
1647
2276f58a
PA		 1648static struct sk_buff *__first_packet_length(struct sock *sk,
1649 struct sk_buff_head *rcvq,
1650 int *total)
1651{
1652 struct sk_buff *skb;
1653
9bd780f5
PA		 1654 while ((skb = skb_peek(rcvq)) != NULL) {
1655 if (udp_lib_checksum_complete(skb)) {
1656 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1657 IS_UDPLITE(sk));
1658 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1659 IS_UDPLITE(sk));
1660 atomic_inc(&sk->sk_drops);
1661 __skb_unlink(skb, rcvq);
1662 *total += skb->truesize;
1663 kfree_skb(skb);
1664 } else {
a793183c 1665 udp_skb_csum_unnecessary_set(skb);
9bd780f5
PA		 1666 break;
1667 }
2276f58a
PA		 1668 }
1669 return skb;
1670}
1671
85584672
ED		 1672/**
1673 * first_packet_length - return length of first packet in receive queue
1674 * @sk: socket
1675 *
1676 * Drops all bad checksum frames, until a valid one is found.
e83c6744 1677 * Returns the length of found skb, or -1 if none is found.
85584672 1678 */
e83c6744 1679static int first_packet_length(struct sock *sk)
85584672 1680{
2276f58a
PA		 1681 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1682 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
85584672 1683 struct sk_buff *skb;
7c13f97f 1684 int total = 0;
e83c6744 1685 int res;
85584672 1686
85584672 1687 spin_lock_bh(&rcvq->lock);
2276f58a 1688 skb = __first_packet_length(sk, rcvq, &total);
137a0dbe 1689 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
2276f58a
PA		 1690 spin_lock(&sk_queue->lock);
1691 skb_queue_splice_tail_init(sk_queue, rcvq);
1692 spin_unlock(&sk_queue->lock);
1693
1694 skb = __first_packet_length(sk, rcvq, &total);
85584672 1695 }
e83c6744 1696 res = skb ? skb->len : -1;
7c13f97f 1697 if (total)
6dfb4367 1698 udp_rmem_release(sk, total, 1, false);
85584672 1699 spin_unlock_bh(&rcvq->lock);
85584672
ED		 1700 return res;
1701}
1702
1da177e4
LT		 1703/*
1704 * IOCTL requests applicable to the UDP protocol
1705 */
e905a9ed 1706
1da177e4
LT		 1707int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1708{
6516c655
SH		 1709 switch (cmd) {
1710 case SIOCOUTQ:
1da177e4 1711 {
31e6d363
ED		 1712 int amount = sk_wmem_alloc_get(sk);
1713
6516c655
SH		 1714 return put_user(amount, (int __user *)arg);
1715 }
1da177e4 1716
6516c655
SH		 1717 case SIOCINQ:
1718 {
e83c6744 1719 int amount = max_t(int, 0, first_packet_length(sk));
6516c655 1720
6516c655
SH		 1721 return put_user(amount, (int __user *)arg);
1722 }
1da177e4 1723
6516c655
SH		 1724 default:
1725 return -ENOIOCTLCMD;
1da177e4 1726 }
6516c655
SH		 1727
1728 return 0;
1da177e4 1729}
c482c568 1730EXPORT_SYMBOL(udp_ioctl);
1da177e4 1731
2276f58a 1732struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
ec095263 1733 int *off, int *err)
2276f58a
PA		 1734{
1735 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1736 struct sk_buff_head *queue;
1737 struct sk_buff *last;
1738 long timeo;
1739 int error;
1740
1741 queue = &udp_sk(sk)->reader_queue;
2276f58a
PA		 1742 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1743 do {
1744 struct sk_buff *skb;
1745
1746 error = sock_error(sk);
1747 if (error)
1748 break;
1749
1750 error = -EAGAIN;
2276f58a 1751 do {
2276f58a 1752 spin_lock_bh(&queue->lock);
e427cad6
PA		 1753 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1754 err, &last);
2276f58a 1755 if (skb) {
e427cad6
PA		 1756 if (!(flags & MSG_PEEK))
1757 udp_skb_destructor(sk, skb);
2276f58a 1758 spin_unlock_bh(&queue->lock);
2276f58a
PA		 1759 return skb;
1760 }
1761
137a0dbe 1762 if (skb_queue_empty_lockless(sk_queue)) {
2276f58a
PA		 1763 spin_unlock_bh(&queue->lock);
1764 goto busy_check;
1765 }
1766
6dfb4367
PA		 1767 /* refill the reader queue and walk it again
1768 * keep both queues locked to avoid re-acquiring
1769 * the sk_receive_queue lock if fwd memory scheduling
1770 * is needed.
1771 */
2276f58a
PA		 1772 spin_lock(&sk_queue->lock);
1773 skb_queue_splice_tail_init(sk_queue, queue);
2276f58a 1774
e427cad6
PA		 1775 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1776 err, &last);
1777 if (skb && !(flags & MSG_PEEK))
1778 udp_skb_dtor_locked(sk, skb);
6dfb4367 1779 spin_unlock(&sk_queue->lock);
2276f58a 1780 spin_unlock_bh(&queue->lock);
de321ed3 1781 if (skb)
2276f58a 1782 return skb;
2276f58a
PA		 1783
1784busy_check:
1785 if (!sk_can_busy_loop(sk))
1786 break;
1787
1788 sk_busy_loop(sk, flags & MSG_DONTWAIT);
137a0dbe 1789 } while (!skb_queue_empty_lockless(sk_queue));
2276f58a
PA		 1790
1791 /* sk_queue is empty, reader_queue may contain peeked packets */
1792 } while (timeo &&
b50b0580
SD		 1793 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1794 &error, &timeo,
2276f58a
PA		 1795 (struct sk_buff *)sk_queue));
1796
1797 *err = error;
1798 return NULL;
1799}
7e823644 1800EXPORT_SYMBOL(__skb_recv_udp);
2276f58a 1801
965b57b4 1802int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
d7f57118 1803{
31f1fbcb
PY		 1804 struct sk_buff *skb;
1805 int err, copied;
099f896f 1806
31f1fbcb
PY		 1807try_again:
1808 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1809 if (!skb)
1810 return err;
099f896f 1811
31f1fbcb
PY		 1812 if (udp_lib_checksum_complete(skb)) {
1813 int is_udplite = IS_UDPLITE(sk);
1814 struct net *net = sock_net(sk);
d7f57118 1815
31f1fbcb
PY		 1816 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1817 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1818 atomic_inc(&sk->sk_drops);
e00a5c33 1819 kfree_skb(skb);
31f1fbcb 1820 goto try_again;
d7f57118
CW		 1821 }
1822
31f1fbcb
PY		 1823 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1824 copied = recv_actor(sk, skb);
1825 kfree_skb(skb);
1826
d7f57118
CW		 1827 return copied;
1828}
965b57b4 1829EXPORT_SYMBOL(udp_read_skb);
d7f57118 1830
db8dac20
DM		 1831/*
1832 * This should be easy, if there is something there we
1833 * return it, otherwise we block.
1834 */
1835
ec095263
OH		 1836int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1837 int *addr_len)
db8dac20
DM		 1838{
1839 struct inet_sock *inet = inet_sk(sk);
342dfc30 1840 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
db8dac20 1841 struct sk_buff *skb;
59c2cdae 1842 unsigned int ulen, copied;
fd69c399 1843 int off, err, peeking = flags & MSG_PEEK;
db8dac20 1844 int is_udplite = IS_UDPLITE(sk);
197c949e 1845 bool checksum_valid = false;
db8dac20 1846
db8dac20 1847 if (flags & MSG_ERRQUEUE)
85fbaa75 1848 return ip_recv_error(sk, msg, len, addr_len);
db8dac20
DM		 1849
1850try_again:
a0917e0b 1851 off = sk_peek_offset(sk, flags);
ec095263 1852 skb = __skb_recv_udp(sk, flags, &off, &err);
db8dac20 1853 if (!skb)
627d2d6b 1854 return err;
db8dac20 1855
b65ac446 1856 ulen = udp_skb_len(skb);
59c2cdae 1857 copied = len;
627d2d6b 1858 if (copied > ulen - off)
1859 copied = ulen - off;
59c2cdae 1860 else if (copied < ulen)
db8dac20
DM		 1861 msg->msg_flags |= MSG_TRUNC;
1862
1863 /*
1864 * If checksum is needed at all, try to do it while copying the
1865 * data. If the data is truncated, or if we only want a partial
1866 * coverage checksum (UDP-Lite), do it before the copy.
1867 */
1868
d21dbdfe
ED		 1869 if (copied < ulen || peeking ||
1870 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
b65ac446
PA		 1871 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1872 !__udp_lib_checksum_complete(skb);
197c949e 1873 if (!checksum_valid)
db8dac20
DM		 1874 goto csum_copy_err;
1875 }
1876
b65ac446
PA		 1877 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1878 if (udp_skb_is_linear(skb))
1879 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1880 else
1881 err = skb_copy_datagram_msg(skb, off, msg, copied);
1882 } else {
627d2d6b 1883 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
db8dac20
DM		 1884
1885 if (err == -EINVAL)
1886 goto csum_copy_err;
1887 }
1888
22911fc5 1889 if (unlikely(err)) {
fd69c399 1890 if (!peeking) {
979402b1 1891 atomic_inc(&sk->sk_drops);
6aef70a8
ED		 1892 UDP_INC_STATS(sock_net(sk),
1893 UDP_MIB_INERRORS, is_udplite);
979402b1 1894 }
850cbadd 1895 kfree_skb(skb);
627d2d6b 1896 return err;
22911fc5 1897 }
db8dac20 1898
fd69c399 1899 if (!peeking)
6aef70a8
ED		 1900 UDP_INC_STATS(sock_net(sk),
1901 UDP_MIB_INDATAGRAMS, is_udplite);
db8dac20 1902
6fd1d51c 1903 sock_recv_cmsgs(msg, sk, skb);
db8dac20
DM		 1904
1905 /* Copy the address. */
c482c568 1906 if (sin) {
db8dac20
DM		 1907 sin->sin_family = AF_INET;
1908 sin->sin_port = udp_hdr(skb)->source;
1909 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1910 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
bceaa902 1911 *addr_len = sizeof(*sin);
983695fa 1912
a9ed15da
SF		 1913 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1914 (struct sockaddr *)sin);
db8dac20 1915 }
bcd1665e
PA		 1916
1917 if (udp_sk(sk)->gro_enabled)
1918 udp_cmsg_recv(msg, sk, skb);
1919
db8dac20 1920 if (inet->cmsg_flags)
ad959036 1921 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
db8dac20 1922
59c2cdae 1923 err = copied;
db8dac20
DM		 1924 if (flags & MSG_TRUNC)
1925 err = ulen;
1926
850cbadd 1927 skb_consume_udp(sk, skb, peeking ? -err : err);
db8dac20
DM		 1928 return err;
1929
1930csum_copy_err:
2276f58a
PA		 1931 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1932 udp_skb_destructor)) {
6aef70a8
ED		 1933 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1934 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
6a5dc9e5 1935 }
850cbadd 1936 kfree_skb(skb);
db8dac20 1937
beb39db5
ED		 1938 /* starting over for a new packet, but check if we need to yield */
1939 cond_resched();
9cfaa8de 1940 msg->msg_flags &= ~MSG_TRUNC;
db8dac20
DM		 1941 goto try_again;
1942}
1943
d74bad4e
AI		 1944int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1945{
1946 /* This check is replicated from __ip4_datagram_connect() and
1947 * intended to prevent BPF program called below from accessing bytes
1948 * that are out of the bound specified by user in addr_len.
1949 */
1950 if (addr_len < sizeof(struct sockaddr_in))
1951 return -EINVAL;
1952
1953 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1954}
1955EXPORT_SYMBOL(udp_pre_connect);
1956
286c72de 1957int __udp_disconnect(struct sock *sk, int flags)
1da177e4
LT		 1958{
1959 struct inet_sock *inet = inet_sk(sk);
1960 /*
1961 * 1003.1g - break association.
1962 */
e905a9ed 1963
1da177e4 1964 sk->sk_state = TCP_CLOSE;
c720c7e8
ED		 1965 inet->inet_daddr = 0;
1966 inet->inet_dport = 0;
bdeab991 1967 sock_rps_reset_rxhash(sk);
1da177e4 1968 sk->sk_bound_dev_if = 0;
303d0403 1969 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1da177e4 1970 inet_reset_saddr(sk);
303d0403
WB		 1971 if (sk->sk_prot->rehash &&
1972 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1973 sk->sk_prot->rehash(sk);
1974 }
1da177e4
LT		 1975
1976 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1977 sk->sk_prot->unhash(sk);
c720c7e8 1978 inet->inet_sport = 0;
1da177e4
LT		 1979 }
1980 sk_dst_reset(sk);
1981 return 0;
1982}
286c72de
ED		 1983EXPORT_SYMBOL(__udp_disconnect);
1984
1985int udp_disconnect(struct sock *sk, int flags)
1986{
1987 lock_sock(sk);
1988 __udp_disconnect(sk, flags);
1989 release_sock(sk);
1990 return 0;
1991}
c482c568 1992EXPORT_SYMBOL(udp_disconnect);
1da177e4 1993
645ca708
ED		 1994void udp_lib_unhash(struct sock *sk)
1995{
723b4610
ED		 1996 if (sk_hashed(sk)) {
1997 struct udp_table *udptable = sk->sk_prot->h.udp_table;
512615b6
ED		 1998 struct udp_hslot *hslot, *hslot2;
1999
2000 hslot = udp_hashslot(udptable, sock_net(sk),
2001 udp_sk(sk)->udp_port_hash);
2002 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
645ca708 2003
723b4610 2004 spin_lock_bh(&hslot->lock);
e32ea7e7
CG		 2005 if (rcu_access_pointer(sk->sk_reuseport_cb))
2006 reuseport_detach_sock(sk);
ca065d0c 2007 if (sk_del_node_init_rcu(sk)) {
fdcc8aa9 2008 hslot->count--;
c720c7e8 2009 inet_sk(sk)->inet_num = 0;
723b4610 2010 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
512615b6
ED		 2011
2012 spin_lock(&hslot2->lock);
ca065d0c 2013 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
512615b6
ED		 2014 hslot2->count--;
2015 spin_unlock(&hslot2->lock);
723b4610
ED		 2016 }
2017 spin_unlock_bh(&hslot->lock);
645ca708 2018 }
645ca708
ED		 2019}
2020EXPORT_SYMBOL(udp_lib_unhash);
2021
719f8358
ED		 2022/*
2023 * inet_rcv_saddr was changed, we must rehash secondary hash
2024 */
2025void udp_lib_rehash(struct sock *sk, u16 newhash)
2026{
2027 if (sk_hashed(sk)) {
2028 struct udp_table *udptable = sk->sk_prot->h.udp_table;
2029 struct udp_hslot *hslot, *hslot2, *nhslot2;
2030
2031 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2032 nhslot2 = udp_hashslot2(udptable, newhash);
2033 udp_sk(sk)->udp_portaddr_hash = newhash;
e32ea7e7
CG		 2034
2035 if (hslot2 != nhslot2 ||
2036 rcu_access_pointer(sk->sk_reuseport_cb)) {
719f8358
ED		 2037 hslot = udp_hashslot(udptable, sock_net(sk),
2038 udp_sk(sk)->udp_port_hash);
2039 /* we must lock primary chain too */
2040 spin_lock_bh(&hslot->lock);
e32ea7e7
CG		 2041 if (rcu_access_pointer(sk->sk_reuseport_cb))
2042 reuseport_detach_sock(sk);
2043
2044 if (hslot2 != nhslot2) {
2045 spin_lock(&hslot2->lock);
ca065d0c 2046 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
e32ea7e7
CG		 2047 hslot2->count--;
2048 spin_unlock(&hslot2->lock);
2049
2050 spin_lock(&nhslot2->lock);
ca065d0c 2051 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
e32ea7e7
CG		 2052 &nhslot2->head);
2053 nhslot2->count++;
2054 spin_unlock(&nhslot2->lock);
2055 }
719f8358
ED		 2056
2057 spin_unlock_bh(&hslot->lock);
2058 }
2059 }
2060}
2061EXPORT_SYMBOL(udp_lib_rehash);
2062
8f6b5392 2063void udp_v4_rehash(struct sock *sk)
719f8358 2064{
f0b1e64c 2065 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
719f8358
ED		 2066 inet_sk(sk)->inet_rcv_saddr,
2067 inet_sk(sk)->inet_num);
2068 udp_lib_rehash(sk, new_hash);
2069}
2070
a3f96c47 2071static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
93821778 2072{
fec5e652 2073 int rc;
766e9037 2074
005ec974 2075 if (inet_sk(sk)->inet_daddr) {
bdeab991 2076 sock_rps_save_rxhash(sk, skb);
005ec974 2077 sk_mark_napi_id(sk, skb);
2c8c56e1 2078 sk_incoming_cpu_update(sk);
e68b6e50
ED		 2079 } else {
2080 sk_mark_napi_id_once(sk, skb);
005ec974 2081 }
fec5e652 2082
850cbadd 2083 rc = __udp_enqueue_schedule_skb(sk, skb);
766e9037
ED		 2084 if (rc < 0) {
2085 int is_udplite = IS_UDPLITE(sk);
08d4c037 2086 int drop_reason;
93821778 2087
93821778 2088 /* Note that an ENOMEM error is charged twice */
08d4c037 2089 if (rc == -ENOMEM) {
e61da9e2 2090 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
02c22347 2091 is_udplite);
08d4c037
MD		 2092 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2093 } else {
a3ce2b10
MD		 2094 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2095 is_udplite);
08d4c037
MD		 2096 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2097 }
e61da9e2 2098 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
08d4c037 2099 kfree_skb_reason(skb, drop_reason);
296f7ea7 2100 trace_udp_fail_queue_rcv_skb(rc, sk);
766e9037 2101 return -1;
93821778
HX		 2102 }
2103
2104 return 0;
93821778
HX		 2105}
2106
db8dac20
DM		 2107/* returns:
2108 * -1: error
2109 * 0: success
2110 * >0: "udp encap" protocol resubmission
2111 *
2112 * Note that in the success and error cases, the skb is assumed to
2113 * have either been requeued or freed.
2114 */
cf329aa4 2115static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
db8dac20 2116{
1379a92d 2117 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
db8dac20 2118 struct udp_sock *up = udp_sk(sk);
db8dac20
DM		 2119 int is_udplite = IS_UDPLITE(sk);
2120
2121 /*
2122 * Charge it to the socket, dropping if the queue is full.
2123 */
1379a92d
MD		 2124 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2125 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
db8dac20 2126 goto drop;
1379a92d 2127 }
895b5c9f 2128 nf_reset_ct(skb);
db8dac20 2129
88ab3108 2130 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
0ad92ad0
ED		 2131 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2132
db8dac20
DM		 2133 /*
2134 * This is an encapsulation socket so pass the skb to
2135 * the socket's udp_encap_rcv() hook. Otherwise, just
2136 * fall through and pass this up the UDP socket.
2137 * up->encap_rcv() returns the following value:
2138 * =0 if skb was successfully passed to the encap
2139 * handler or was discarded by it.
2140 * >0 if skb should be passed on to UDP.
2141 * <0 if skb should be resubmitted as proto -N
2142 */
2143
2144 /* if we're overly short, let UDP handle it */
6aa7de05 2145 encap_rcv = READ_ONCE(up->encap_rcv);
e5aed006 2146 if (encap_rcv) {
db8dac20
DM		 2147 int ret;
2148
0a80966b
TH		 2149 /* Verify checksum before giving to encap */
2150 if (udp_lib_checksum_complete(skb))
2151 goto csum_error;
2152
0ad92ad0 2153 ret = encap_rcv(sk, skb);
db8dac20 2154 if (ret <= 0) {
02c22347
ED		 2155 __UDP_INC_STATS(sock_net(sk),
2156 UDP_MIB_INDATAGRAMS,
2157 is_udplite);
db8dac20
DM		 2158 return -ret;
2159 }
2160 }
2161
2162 /* FALLTHROUGH -- it's a UDP Packet */
2163 }
2164
2165 /*
2166 * UDP-Lite specific tests, ignored on UDP sockets
2167 */
b0a42277 2168 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
db8dac20
DM		 2169
2170 /*
2171 * MIB statistics other than incrementing the error count are
2172 * disabled for the following two types of errors: these depend
2173 * on the application settings, not on the functioning of the
2174 * protocol stack as such.
2175 *
2176 * RFC 3828 here recommends (sec 3.3): "There should also be a
2177 * way ... to ... at least let the receiving application block
2178 * delivery of packets with coverage values less than a value
2179 * provided by the application."
2180 */
2181 if (up->pcrlen == 0) { /* full coverage was set */
ba7a46f1
JP		 2182 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2183 UDP_SKB_CB(skb)->cscov, skb->len);
db8dac20
DM		 2184 goto drop;
2185 }
2186 /* The next case involves violating the min. coverage requested
2187 * by the receiver. This is subtle: if receiver wants x and x is
2188 * greater than the buffersize/MTU then receiver will complain
2189 * that it wants x while sender emits packets of smaller size y.
2190 * Therefore the above ...()->partial_cov statement is essential.
2191 */
2192 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
ba7a46f1
JP		 2193 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2194 UDP_SKB_CB(skb)->cscov, up->pcrlen);
db8dac20
DM		 2195 goto drop;
2196 }
2197 }
2198
dd99e425 2199 prefetch(&sk->sk_rmem_alloc);
ce25d66a
ED		 2200 if (rcu_access_pointer(sk->sk_filter) &&
2201 udp_lib_checksum_complete(skb))
e6afc8ac 2202 goto csum_error;
ce25d66a 2203
1379a92d
MD		 2204 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2205 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
a6127697 2206 goto drop;
1379a92d 2207 }
db8dac20 2208
e6afc8ac 2209 udp_csum_pull_header(skb);
db8dac20 2210
fbf8866d 2211 ipv4_pktinfo_prepare(sk, skb);
850cbadd 2212 return __udp_queue_rcv_skb(sk, skb);
db8dac20 2213
6a5dc9e5 2214csum_error:
1379a92d 2215 drop_reason = SKB_DROP_REASON_UDP_CSUM;
02c22347 2216 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
db8dac20 2217drop:
02c22347 2218 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
8edf19c2 2219 atomic_inc(&sk->sk_drops);
1379a92d 2220 kfree_skb_reason(skb, drop_reason);
db8dac20
DM		 2221 return -1;
2222}
2223
cf329aa4
PA		 2224static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2225{
2226 struct sk_buff *next, *segs;
2227 int ret;
2228
2229 if (likely(!udp_unexpected_gso(sk, skb)))
2230 return udp_queue_rcv_one_skb(sk, skb);
2231
a08e7fd9 2232 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
cf329aa4
PA		 2233 __skb_push(skb, -skb_mac_offset(skb));
2234 segs = udp_rcv_segment(sk, skb, true);
1a186c14 2235 skb_list_walk_safe(segs, skb, next) {
cf329aa4 2236 __skb_pull(skb, skb_transport_offset(skb));
000ac44d
PA		 2237
2238 udp_post_segment_fix_csum(skb);
cf329aa4
PA		 2239 ret = udp_queue_rcv_one_skb(sk, skb);
2240 if (ret > 0)
10c678bd 2241 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
cf329aa4
PA		 2242 }
2243 return 0;
2244}
2245
97502231 2246/* For TCP sockets, sk_rx_dst is protected by socket lock
e47eb5df 2247 * For UDP, we use xchg() to guard against concurrent changes.
97502231 2248 */
64f0f5d1 2249bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
421b3885 2250{
97502231
ED		 2251 struct dst_entry *old;
2252
d24406c8 2253 if (dst_hold_safe(dst)) {
8f905c0e 2254 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
d24406c8 2255 dst_release(old);
64f0f5d1 2256 return old != dst;
d24406c8 2257 }
64f0f5d1 2258 return false;
421b3885 2259}
c9f2c1ae 2260EXPORT_SYMBOL(udp_sk_rx_dst_set);
421b3885 2261
db8dac20
DM		 2262/*
2263 * Multicasts and broadcasts go to each listener.
2264 *
1240d137 2265 * Note: called only from the BH handler context.
db8dac20 2266 */
e3163493 2267static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
db8dac20
DM		 2268 struct udphdr *uh,
2269 __be32 saddr, __be32 daddr,
36cbb245
RJ		 2270 struct udp_table *udptable,
2271 int proto)
db8dac20 2272{
ca065d0c 2273 struct sock *sk, *first = NULL;
5cf3d461
DH		 2274 unsigned short hnum = ntohs(uh->dest);
2275 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2dc41cff 2276 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
ca065d0c
ED		 2277 unsigned int offset = offsetof(typeof(*sk), sk_node);
2278 int dif = skb->dev->ifindex;
fb74c277 2279 int sdif = inet_sdif(skb);
ca065d0c
ED		 2280 struct hlist_node *node;
2281 struct sk_buff *nskb;
2dc41cff
DH		 2282
2283 if (use_hash2) {
f0b1e64c 2284 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
73e2d5e3 2285 udptable->mask;
f0b1e64c 2286 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2dc41cff 2287start_lookup:
73e2d5e3 2288 hslot = &udptable->hash2[hash2];
2dc41cff
DH		 2289 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2290 }
db8dac20 2291
ca065d0c
ED		 2292 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2293 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
fb74c277 2294 uh->source, saddr, dif, sdif, hnum))
ca065d0c
ED		 2295 continue;
2296
2297 if (!first) {
2298 first = sk;
2299 continue;
1240d137 2300 }
ca065d0c 2301 nskb = skb_clone(skb, GFP_ATOMIC);
1240d137 2302
ca065d0c
ED		 2303 if (unlikely(!nskb)) {
2304 atomic_inc(&sk->sk_drops);
02c22347
ED		 2305 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2306 IS_UDPLITE(sk));
2307 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2308 IS_UDPLITE(sk));
ca065d0c
ED		 2309 continue;
2310 }
2311 if (udp_queue_rcv_skb(sk, nskb) > 0)
2312 consume_skb(nskb);
2313 }
1240d137 2314
2dc41cff
DH		 2315 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2316 if (use_hash2 && hash2 != hash2_any) {
2317 hash2 = hash2_any;
2318 goto start_lookup;
2319 }
2320
ca065d0c
ED		 2321 if (first) {
2322 if (udp_queue_rcv_skb(first, skb) > 0)
2323 consume_skb(skb);
1240d137 2324 } else {
ca065d0c 2325 kfree_skb(skb);
02c22347
ED		 2326 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2327 proto == IPPROTO_UDPLITE);
1240d137 2328 }
db8dac20
DM		 2329 return 0;
2330}
2331
2332/* Initialize UDP checksum. If exited with zero value (success),
2333 * CHECKSUM_UNNECESSARY means, that no more checks are required.
666a3d6e 2334 * Otherwise, csum completion requires checksumming packet body,
db8dac20
DM		 2335 * including udp header and folding it to skb->csum.
2336 */
2337static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2338 int proto)
2339{
db8dac20
DM		 2340 int err;
2341
2342 UDP_SKB_CB(skb)->partial_cov = 0;
2343 UDP_SKB_CB(skb)->cscov = skb->len;
2344
2345 if (proto == IPPROTO_UDPLITE) {
2346 err = udplite_checksum_init(skb, uh);
2347 if (err)
2348 return err;
15f35d49
AK		 2349
2350 if (UDP_SKB_CB(skb)->partial_cov) {
2351 skb->csum = inet_compute_pseudo(skb, proto);
2352 return 0;
2353 }
db8dac20
DM		 2354 }
2355
b46d9f62
HFS		 2356 /* Note, we are only interested in != 0 or == 0, thus the
2357 * force to int.
2358 */
db4f1be3
ST		 2359 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2360 inet_compute_pseudo);
2361 if (err)
2362 return err;
2363
2364 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2365 /* If SW calculated the value, we know it's bad */
2366 if (skb->csum_complete_sw)
2367 return 1;
2368
2369 /* HW says the value is bad. Let's validate that.
2370 * skb->csum is no longer the full packet checksum,
2371 * so don't treat it as such.
2372 */
2373 skb_checksum_complete_unset(skb);
2374 }
2375
2376 return 0;
db8dac20
DM		 2377}
2378
2b5a9217
PA		 2379/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2380 * return code conversion for ip layer consumption
2381 */
2382static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2383 struct udphdr *uh)
2384{
2385 int ret;
2386
2387 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
e4aa33ad 2388 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2b5a9217
PA		 2389
2390 ret = udp_queue_rcv_skb(sk, skb);
2391
2392 /* a return value > 0 means to resubmit the input, but
2393 * it wants the return to be -protocol, or 0
2394 */
2395 if (ret > 0)
2396 return -ret;
2397 return 0;
2398}
2399
db8dac20
DM		 2400/*
2401 * All we need to do is get the socket, and then do a checksum.
2402 */
2403
645ca708 2404int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
db8dac20
DM		 2405 int proto)
2406{
2407 struct sock *sk;
7b5e56f9 2408 struct udphdr *uh;
db8dac20 2409 unsigned short ulen;
adf30907 2410 struct rtable *rt = skb_rtable(skb);
2783ef23 2411 __be32 saddr, daddr;
0283328e 2412 struct net *net = dev_net(skb->dev);
71489e21 2413 bool refcounted;
1c7fab70
MD		 2414 int drop_reason;
2415
2416 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
db8dac20
DM		 2417
2418 /*
2419 * Validate the packet.
2420 */
2421 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2422 goto drop; /* No space for header. */
2423
7b5e56f9 2424 uh = udp_hdr(skb);
db8dac20 2425 ulen = ntohs(uh->len);
ccc2d97c
BM		 2426 saddr = ip_hdr(skb)->saddr;
2427 daddr = ip_hdr(skb)->daddr;
2428
db8dac20
DM		 2429 if (ulen > skb->len)
2430 goto short_packet;
2431
2432 if (proto == IPPROTO_UDP) {
2433 /* UDP validates ulen. */
2434 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2435 goto short_packet;
2436 uh = udp_hdr(skb);
2437 }
2438
2439 if (udp4_csum_init(skb, uh, proto))
2440 goto csum_error;
2441
71489e21 2442 sk = skb_steal_sock(skb, &refcounted);
8afdd99a 2443 if (sk) {
97502231 2444 struct dst_entry *dst = skb_dst(skb);
421b3885 2445 int ret;
421b3885 2446
8f905c0e 2447 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
97502231 2448 udp_sk_rx_dst_set(sk, dst);
db8dac20 2449
2b5a9217 2450 ret = udp_unicast_rcv_skb(sk, skb, uh);
71489e21
JS		 2451 if (refcounted)
2452 sock_put(sk);
2b5a9217 2453 return ret;
421b3885 2454 }
db8dac20 2455
c18450a5
FF		 2456 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2457 return __udp4_lib_mcast_deliver(net, skb, uh,
36cbb245 2458 saddr, daddr, udptable, proto);
c18450a5
FF		 2459
2460 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2b5a9217
PA		 2461 if (sk)
2462 return udp_unicast_rcv_skb(sk, skb, uh);
db8dac20
DM		 2463
2464 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2465 goto drop;
895b5c9f 2466 nf_reset_ct(skb);
db8dac20
DM		 2467
2468 /* No socket. Drop packet silently, if checksum is wrong */
2469 if (udp_lib_checksum_complete(skb))
2470 goto csum_error;
2471
1c7fab70 2472 drop_reason = SKB_DROP_REASON_NO_SOCKET;
02c22347 2473 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
db8dac20
DM		 2474 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2475
2476 /*
2477 * Hmm. We got an UDP packet to a port to which we
2478 * don't wanna listen. Ignore it.
2479 */
1c7fab70 2480 kfree_skb_reason(skb, drop_reason);
db8dac20
DM		 2481 return 0;
2482
2483short_packet:
1c7fab70 2484 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
ba7a46f1
JP		 2485 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2486 proto == IPPROTO_UDPLITE ? "Lite" : "",
2487 &saddr, ntohs(uh->source),
2488 ulen, skb->len,
2489 &daddr, ntohs(uh->dest));
db8dac20
DM		 2490 goto drop;
2491
2492csum_error:
2493 /*
2494 * RFC1122: OK. Discards the bad packet silently (as far as
2495 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2496 */
1c7fab70 2497 drop_reason = SKB_DROP_REASON_UDP_CSUM;
ba7a46f1
JP		 2498 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2499 proto == IPPROTO_UDPLITE ? "Lite" : "",
2500 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2501 ulen);
02c22347 2502 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
db8dac20 2503drop:
02c22347 2504 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1c7fab70 2505 kfree_skb_reason(skb, drop_reason);
db8dac20
DM		 2506 return 0;
2507}
2508
421b3885
SB		 2509/* We can only early demux multicast if there is a single matching socket.
2510 * If more than one socket found returns NULL
2511 */
2512static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2513 __be16 loc_port, __be32 loc_addr,
2514 __be16 rmt_port, __be32 rmt_addr,
fb74c277 2515 int dif, int sdif)
421b3885
SB		 2516{
2517 struct sock *sk, *result;
421b3885 2518 unsigned short hnum = ntohs(loc_port);
ca065d0c 2519 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
421b3885
SB		 2520 struct udp_hslot *hslot = &udp_table.hash[slot];
2521
63c6f81c
ED		 2522 /* Do not bother scanning a too big list */
2523 if (hslot->count > 10)
2524 return NULL;
2525
421b3885 2526 result = NULL;
ca065d0c
ED		 2527 sk_for_each_rcu(sk, &hslot->head) {
2528 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
fb74c277 2529 rmt_port, rmt_addr, dif, sdif, hnum)) {
ca065d0c
ED		 2530 if (result)
2531 return NULL;
421b3885 2532 result = sk;
421b3885
SB		 2533 }
2534 }
ca065d0c 2535
421b3885
SB		 2536 return result;
2537}
2538
2539/* For unicast we should only early demux connected sockets or we can
2540 * break forwarding setups. The chains here can be long so only check
2541 * if the first socket is an exact match and if not move on.
2542 */
2543static struct sock *__udp4_lib_demux_lookup(struct net *net,
2544 __be16 loc_port, __be32 loc_addr,
2545 __be16 rmt_port, __be32 rmt_addr,
3fa6f616 2546 int dif, int sdif)
421b3885 2547{
421b3885 2548 unsigned short hnum = ntohs(loc_port);
f0b1e64c 2549 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
421b3885
SB		 2550 unsigned int slot2 = hash2 & udp_table.mask;
2551 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
c7228317 2552 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
421b3885 2553 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
ca065d0c 2554 struct sock *sk;
421b3885 2555
ca065d0c 2556 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
eda090c3 2557 if (inet_match(net, sk, acookie, ports, dif, sdif))
ca065d0c 2558 return sk;
421b3885
SB		 2559 /* Only check first socket in chain */
2560 break;
2561 }
ca065d0c 2562 return NULL;
421b3885
SB		 2563}
2564
7487449c 2565int udp_v4_early_demux(struct sk_buff *skb)
421b3885 2566{
610438b7 2567 struct net *net = dev_net(skb->dev);
bc044e8d 2568 struct in_device *in_dev = NULL;
610438b7
ED		 2569 const struct iphdr *iph;
2570 const struct udphdr *uh;
ca065d0c 2571 struct sock *sk = NULL;
421b3885 2572 struct dst_entry *dst;
421b3885 2573 int dif = skb->dev->ifindex;
fb74c277 2574 int sdif = inet_sdif(skb);
6e540309 2575 int ours;
421b3885
SB		 2576
2577 /* validate the packet */
2578 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
7487449c 2579 return 0;
421b3885 2580
610438b7
ED		 2581 iph = ip_hdr(skb);
2582 uh = udp_hdr(skb);
2583
996b44fc 2584 if (skb->pkt_type == PACKET_MULTICAST) {
bc044e8d 2585 in_dev = __in_dev_get_rcu(skb->dev);
6e540309
SB		 2586
2587 if (!in_dev)
7487449c 2588 return 0;
6e540309 2589
996b44fc
PA		 2590 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2591 iph->protocol);
2592 if (!ours)
2593 return 0;
ad0ea198 2594
421b3885 2595 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
fb74c277
DA		 2596 uh->source, iph->saddr,
2597 dif, sdif);
6e540309 2598 } else if (skb->pkt_type == PACKET_HOST) {
421b3885 2599 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
3fa6f616 2600 uh->source, iph->saddr, dif, sdif);
6e540309 2601 }
421b3885 2602
41c6d650 2603 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
7487449c 2604 return 0;
421b3885
SB		 2605
2606 skb->sk = sk;
82eabd9e 2607 skb->destructor = sock_efree;
8f905c0e 2608 dst = rcu_dereference(sk->sk_rx_dst);
421b3885
SB		 2609
2610 if (dst)
2611 dst = dst_check(dst, 0);
10e2eb87 2612 if (dst) {
bc044e8d
PA		 2613 u32 itag = 0;
2614
d24406c8
WW		 2615 /* set noref for now.
2616 * any place which wants to hold dst has to call
2617 * dst_hold_safe()
2618 */
2619 skb_dst_set_noref(skb, dst);
bc044e8d
PA		 2620
2621 /* for unconnected multicast sockets we need to validate
2622 * the source on each packet
2623 */
2624 if (!inet_sk(sk)->inet_daddr && in_dev)
2625 return ip_mc_validate_source(skb, iph->daddr,
8d2b51b0
GN		 2626 iph->saddr,
2627 iph->tos & IPTOS_RT_MASK,
bc044e8d 2628 skb->dev, in_dev, &itag);
10e2eb87 2629 }
7487449c 2630 return 0;
421b3885
SB		 2631}
2632
db8dac20
DM		 2633int udp_rcv(struct sk_buff *skb)
2634{
645ca708 2635 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
db8dac20
DM		 2636}
2637
7d06b2e0 2638void udp_destroy_sock(struct sock *sk)
db8dac20 2639{
44046a59 2640 struct udp_sock *up = udp_sk(sk);
8a74ad60 2641 bool slow = lock_sock_fast(sk);
a8b897c7
PA		 2642
2643 /* protects from races with udp_abort() */
2644 sock_set_flag(sk, SOCK_DEAD);
db8dac20 2645 udp_flush_pending_frames(sk);
8a74ad60 2646 unlock_sock_fast(sk, slow);
60fb9567
PA		 2647 if (static_branch_unlikely(&udp_encap_needed_key)) {
2648 if (up->encap_type) {
2649 void (*encap_destroy)(struct sock *sk);
2650 encap_destroy = READ_ONCE(up->encap_destroy);
2651 if (encap_destroy)
2652 encap_destroy(sk);
2653 }
2654 if (up->encap_enabled)
9c480601 2655 static_branch_dec(&udp_encap_needed_key);
44046a59 2656 }
db8dac20
DM		 2657}
2658
1da177e4
LT		 2659/*
2660 * Socket option code for UDP
2661 */
4c0a6cb0 2662int udp_lib_setsockopt(struct sock *sk, int level, int optname,
91ac1cca 2663 sockptr_t optval, unsigned int optlen,
4c0a6cb0 2664 int (*push_pending_frames)(struct sock *))
1da177e4
LT		 2665{
2666 struct udp_sock *up = udp_sk(sk);
1c19448c 2667 int val, valbool;
1da177e4 2668 int err = 0;
b2bf1e26 2669 int is_udplite = IS_UDPLITE(sk);
1da177e4 2670
c482c568 2671 if (optlen < sizeof(int))
1da177e4
LT		 2672 return -EINVAL;
2673
91ac1cca 2674 if (copy_from_sockptr(&val, optval, sizeof(val)))
1da177e4
LT		 2675 return -EFAULT;
2676
1c19448c
TH		 2677 valbool = val ? 1 : 0;
2678
6516c655 2679 switch (optname) {
1da177e4
LT		 2680 case UDP_CORK:
2681 if (val != 0) {
a9f59707 2682 WRITE_ONCE(up->corkflag, 1);
1da177e4 2683 } else {
a9f59707 2684 WRITE_ONCE(up->corkflag, 0);
1da177e4 2685 lock_sock(sk);
4243cdc2 2686 push_pending_frames(sk);
1da177e4
LT		 2687 release_sock(sk);
2688 }
2689 break;
e905a9ed 2690
1da177e4
LT		 2691 case UDP_ENCAP:
2692 switch (val) {
2693 case 0:
fd1ac07f 2694#ifdef CONFIG_XFRM
1da177e4
LT		 2695 case UDP_ENCAP_ESPINUDP:
2696 case UDP_ENCAP_ESPINUDP_NON_IKE:
0146dca7
SD		 2697#if IS_ENABLED(CONFIG_IPV6)
2698 if (sk->sk_family == AF_INET6)
2699 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2700 else
2701#endif
2702 up->encap_rcv = xfrm4_udp_encap_rcv;
fd1ac07f 2703#endif
a8eceea8 2704 fallthrough;
342f0234 2705 case UDP_ENCAP_L2TPINUDP:
1da177e4 2706 up->encap_type = val;
60fb9567
PA		 2707 lock_sock(sk);
2708 udp_tunnel_encap_enable(sk->sk_socket);
2709 release_sock(sk);
1da177e4
LT		 2710 break;
2711 default:
2712 err = -ENOPROTOOPT;
2713 break;
2714 }
2715 break;
2716
1c19448c
TH		 2717 case UDP_NO_CHECK6_TX:
2718 up->no_check6_tx = valbool;
2719 break;
2720
2721 case UDP_NO_CHECK6_RX:
2722 up->no_check6_rx = valbool;
2723 break;
2724
bec1f6f6
WB		 2725 case UDP_SEGMENT:
2726 if (val < 0 || val > USHRT_MAX)
2727 return -EINVAL;
18a419ba 2728 WRITE_ONCE(up->gso_size, val);
bec1f6f6
WB		 2729 break;
2730
e20cf8d3
PA		 2731 case UDP_GRO:
2732 lock_sock(sk);
78352f73
PA		 2733
2734 /* when enabling GRO, accept the related GSO packet type */
e20cf8d3
PA		 2735 if (valbool)
2736 udp_tunnel_encap_enable(sk->sk_socket);
2737 up->gro_enabled = valbool;
78352f73 2738 up->accept_udp_l4 = valbool;
e20cf8d3
PA		 2739 release_sock(sk);
2740 break;
2741
ba4e58ec
GR		 2742 /*
2743 * UDP-Lite's partial checksum coverage (RFC 3828).
2744 */
2745 /* The sender sets actual checksum coverage length via this option.
2746 * The case coverage > packet length is handled by send module. */
2747 case UDPLITE_SEND_CSCOV:
b2bf1e26 2748 if (!is_udplite) /* Disable the option on UDP sockets */
ba4e58ec
GR		 2749 return -ENOPROTOOPT;
2750 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2751 val = 8;
4be929be
AD		 2752 else if (val > USHRT_MAX)
2753 val = USHRT_MAX;
ba4e58ec
GR		 2754 up->pcslen = val;
2755 up->pcflag |= UDPLITE_SEND_CC;
2756 break;
2757
e905a9ed
YH		 2758 /* The receiver specifies a minimum checksum coverage value. To make
2759 * sense, this should be set to at least 8 (as done below). If zero is
ba4e58ec
GR		 2760 * used, this again means full checksum coverage. */
2761 case UDPLITE_RECV_CSCOV:
b2bf1e26 2762 if (!is_udplite) /* Disable the option on UDP sockets */
ba4e58ec
GR		 2763 return -ENOPROTOOPT;
2764 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2765 val = 8;
4be929be
AD		 2766 else if (val > USHRT_MAX)
2767 val = USHRT_MAX;
ba4e58ec
GR		 2768 up->pcrlen = val;
2769 up->pcflag |= UDPLITE_RECV_CC;
2770 break;
2771
1da177e4
LT		 2772 default:
2773 err = -ENOPROTOOPT;
2774 break;
6516c655 2775 }
1da177e4
LT		 2776
2777 return err;
2778}
c482c568 2779EXPORT_SYMBOL(udp_lib_setsockopt);
1da177e4 2780
a7b75c5a
CH		 2781int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2782 unsigned int optlen)
db8dac20
DM		 2783{
2784 if (level == SOL_UDP || level == SOL_UDPLITE)
91ac1cca 2785 return udp_lib_setsockopt(sk, level, optname,
a7b75c5a 2786 optval, optlen,
db8dac20
DM		 2787 udp_push_pending_frames);
2788 return ip_setsockopt(sk, level, optname, optval, optlen);
2789}
2790
4c0a6cb0
GR		 2791int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2792 char __user *optval, int __user *optlen)
1da177e4
LT		 2793{
2794 struct udp_sock *up = udp_sk(sk);
2795 int val, len;
2796
c482c568 2797 if (get_user(len, optlen))
1da177e4
LT		 2798 return -EFAULT;
2799
2800 len = min_t(unsigned int, len, sizeof(int));
e905a9ed 2801
6516c655 2802 if (len < 0)
1da177e4
LT		 2803 return -EINVAL;
2804
6516c655 2805 switch (optname) {
1da177e4 2806 case UDP_CORK:
a9f59707 2807 val = READ_ONCE(up->corkflag);
1da177e4
LT		 2808 break;
2809
2810 case UDP_ENCAP:
2811 val = up->encap_type;
2812 break;
2813
1c19448c
TH		 2814 case UDP_NO_CHECK6_TX:
2815 val = up->no_check6_tx;
2816 break;
2817
2818 case UDP_NO_CHECK6_RX:
2819 val = up->no_check6_rx;
2820 break;
2821
bec1f6f6 2822 case UDP_SEGMENT:
18a419ba 2823 val = READ_ONCE(up->gso_size);
bec1f6f6
WB		 2824 break;
2825
98184612
NM		 2826 case UDP_GRO:
2827 val = up->gro_enabled;
2828 break;
2829
ba4e58ec
GR		 2830 /* The following two cannot be changed on UDP sockets, the return is
2831 * always 0 (which corresponds to the full checksum coverage of UDP). */
2832 case UDPLITE_SEND_CSCOV:
2833 val = up->pcslen;
2834 break;
2835
2836 case UDPLITE_RECV_CSCOV:
2837 val = up->pcrlen;
2838 break;
2839
1da177e4
LT		 2840 default:
2841 return -ENOPROTOOPT;
6516c655 2842 }
1da177e4 2843
6516c655 2844 if (put_user(len, optlen))
e905a9ed 2845 return -EFAULT;
c482c568 2846 if (copy_to_user(optval, &val, len))
1da177e4 2847 return -EFAULT;
e905a9ed 2848 return 0;
1da177e4 2849}
c482c568 2850EXPORT_SYMBOL(udp_lib_getsockopt);
1da177e4 2851
db8dac20
DM		 2852int udp_getsockopt(struct sock *sk, int level, int optname,
2853 char __user *optval, int __user *optlen)
2854{
2855 if (level == SOL_UDP || level == SOL_UDPLITE)
2856 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2857 return ip_getsockopt(sk, level, optname, optval, optlen);
2858}
2859
1da177e4
LT		 2860/**
2861 * udp_poll - wait for a UDP event.
3628e3cb
AL		 2862 * @file: - file struct
2863 * @sock: - socket
2864 * @wait: - poll table
1da177e4 2865 *
e905a9ed 2866 * This is same as datagram poll, except for the special case of
1da177e4
LT		 2867 * blocking sockets. If application is using a blocking fd
2868 * and a packet with checksum error is in the queue;
2869 * then it could get return from select indicating data available
2870 * but then block when reading it. Add special case code
2871 * to work around these arguably broken applications.
2872 */
a11e1d43 2873__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1da177e4 2874{
a11e1d43 2875 __poll_t mask = datagram_poll(file, sock, wait);
1da177e4 2876 struct sock *sk = sock->sk;
ba4e58ec 2877
3ef7cf57 2878 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
a9a08845 2879 mask |= EPOLLIN | EPOLLRDNORM;
2276f58a 2880
1da177e4 2881 /* Check for false positives due to checksum errors */
a11e1d43 2882 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
e83c6744 2883 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
a9a08845 2884 mask &= ~(EPOLLIN | EPOLLRDNORM);
1da177e4 2885
af493388
CW		 2886 /* psock ingress_msg queue should not contain any bad checksum frames */
2887 if (sk_is_readable(sk))
2888 mask |= EPOLLIN | EPOLLRDNORM;
1da177e4 2889 return mask;
e905a9ed 2890
1da177e4 2891}
a11e1d43 2892EXPORT_SYMBOL(udp_poll);
1da177e4 2893
5d77dca8
DA		 2894int udp_abort(struct sock *sk, int err)
2895{
2896 lock_sock(sk);
2897
a8b897c7
PA		 2898 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2899 * with close()
2900 */
2901 if (sock_flag(sk, SOCK_DEAD))
2902 goto out;
2903
5d77dca8 2904 sk->sk_err = err;
e3ae2365 2905 sk_error_report(sk);
286c72de 2906 __udp_disconnect(sk, 0);
5d77dca8 2907
a8b897c7 2908out:
5d77dca8
DA		 2909 release_sock(sk);
2910
2911 return 0;
2912}
2913EXPORT_SYMBOL_GPL(udp_abort);
2914
db8dac20 2915struct proto udp_prot = {
1e802951
TZ		 2916 .name = "UDP",
2917 .owner = THIS_MODULE,
2918 .close = udp_lib_close,
d74bad4e 2919 .pre_connect = udp_pre_connect,
1e802951
TZ		 2920 .connect = ip4_datagram_connect,
2921 .disconnect = udp_disconnect,
2922 .ioctl = udp_ioctl,
2923 .init = udp_init_sock,
2924 .destroy = udp_destroy_sock,
2925 .setsockopt = udp_setsockopt,
2926 .getsockopt = udp_getsockopt,
2927 .sendmsg = udp_sendmsg,
2928 .recvmsg = udp_recvmsg,
2929 .sendpage = udp_sendpage,
2930 .release_cb = ip4_datagram_release_cb,
2931 .hash = udp_lib_hash,
2932 .unhash = udp_lib_unhash,
2933 .rehash = udp_v4_rehash,
2934 .get_port = udp_v4_get_port,
91a760b2 2935 .put_port = udp_lib_unhash,
8a59f9d1
CW		 2936#ifdef CONFIG_BPF_SYSCALL
2937 .psock_update_sk_prot = udp_bpf_update_proto,
2938#endif
1e802951 2939 .memory_allocated = &udp_memory_allocated,
0defbb0a
ED		 2940 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2941
1e802951
TZ		 2942 .sysctl_mem = sysctl_udp_mem,
2943 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2944 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2945 .obj_size = sizeof(struct udp_sock),
2946 .h.udp_table = &udp_table,
1e802951 2947 .diag_destroy = udp_abort,
db8dac20 2948};
c482c568 2949EXPORT_SYMBOL(udp_prot);
1da177e4
LT		 2950
2951/* ------------------------------------------------------------------------ */
2952#ifdef CONFIG_PROC_FS
2953
645ca708 2954static struct sock *udp_get_first(struct seq_file *seq, int start)
1da177e4
LT		 2955{
2956 struct sock *sk;
9e8ca27a 2957 struct udp_seq_afinfo *afinfo;
1da177e4 2958 struct udp_iter_state *state = seq->private;
6f191efe 2959 struct net *net = seq_file_net(seq);
1da177e4 2960
9e8ca27a
YS		 2961 if (state->bpf_seq_afinfo)
2962 afinfo = state->bpf_seq_afinfo;
2963 else
359745d7 2964 afinfo = pde_data(file_inode(seq->file));
9e8ca27a 2965
a3d2599b 2966 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
f86dcc5a 2967 ++state->bucket) {
a3d2599b 2968 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
f86dcc5a 2969
ca065d0c 2970 if (hlist_empty(&hslot->head))
f86dcc5a
ED		 2971 continue;
2972
645ca708 2973 spin_lock_bh(&hslot->lock);
ca065d0c 2974 sk_for_each(sk, &hslot->head) {
878628fb 2975 if (!net_eq(sock_net(sk), net))
a91275ef 2976 continue;
9e8ca27a
YS		 2977 if (afinfo->family == AF_UNSPEC ||
2978 sk->sk_family == afinfo->family)
1da177e4
LT		 2979 goto found;
2980 }
645ca708 2981 spin_unlock_bh(&hslot->lock);
1da177e4
LT		 2982 }
2983 sk = NULL;
2984found:
2985 return sk;
2986}
2987
2988static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2989{
9e8ca27a 2990 struct udp_seq_afinfo *afinfo;
1da177e4 2991 struct udp_iter_state *state = seq->private;
6f191efe 2992 struct net *net = seq_file_net(seq);
1da177e4 2993
9e8ca27a
YS		 2994 if (state->bpf_seq_afinfo)
2995 afinfo = state->bpf_seq_afinfo;
2996 else
359745d7 2997 afinfo = pde_data(file_inode(seq->file));
9e8ca27a 2998
1da177e4 2999 do {
ca065d0c 3000 sk = sk_next(sk);
9e8ca27a
YS		 3001 } while (sk && (!net_eq(sock_net(sk), net) ||
3002 (afinfo->family != AF_UNSPEC &&
3003 sk->sk_family != afinfo->family)));
1da177e4 3004
645ca708 3005 if (!sk) {
a3d2599b
CH		 3006 if (state->bucket <= afinfo->udp_table->mask)
3007 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
645ca708 3008 return udp_get_first(seq, state->bucket + 1);
1da177e4
LT		 3009 }
3010 return sk;
3011}
3012
3013static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3014{
645ca708 3015 struct sock *sk = udp_get_first(seq, 0);
1da177e4
LT		 3016
3017 if (sk)
6516c655 3018 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1da177e4
LT		 3019 --pos;
3020 return pos ? NULL : sk;
3021}
3022
a3d2599b 3023void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1da177e4 3024{
30842f29 3025 struct udp_iter_state *state = seq->private;
f86dcc5a 3026 state->bucket = MAX_UDP_PORTS;
30842f29 3027
b50660f1 3028 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1da177e4 3029}
a3d2599b 3030EXPORT_SYMBOL(udp_seq_start);
1da177e4 3031
a3d2599b 3032void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1da177e4
LT		 3033{
3034 struct sock *sk;
3035
b50660f1 3036 if (v == SEQ_START_TOKEN)
1da177e4
LT		 3037 sk = udp_get_idx(seq, 0);
3038 else
3039 sk = udp_get_next(seq, v);
3040
3041 ++*pos;
3042 return sk;
3043}
a3d2599b 3044EXPORT_SYMBOL(udp_seq_next);
1da177e4 3045
a3d2599b 3046void udp_seq_stop(struct seq_file *seq, void *v)
1da177e4 3047{
9e8ca27a 3048 struct udp_seq_afinfo *afinfo;
645ca708
ED		 3049 struct udp_iter_state *state = seq->private;
3050
9e8ca27a
YS		 3051 if (state->bpf_seq_afinfo)
3052 afinfo = state->bpf_seq_afinfo;
3053 else
359745d7 3054 afinfo = pde_data(file_inode(seq->file));
9e8ca27a 3055
a3d2599b
CH		 3056 if (state->bucket <= afinfo->udp_table->mask)
3057 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
1da177e4 3058}
a3d2599b 3059EXPORT_SYMBOL(udp_seq_stop);
db8dac20
DM		 3060
3061/* ------------------------------------------------------------------------ */
5e659e4c 3062static void udp4_format_sock(struct sock *sp, struct seq_file *f,
652586df 3063 int bucket)
db8dac20
DM		 3064{
3065 struct inet_sock *inet = inet_sk(sp);
c720c7e8
ED		 3066 __be32 dest = inet->inet_daddr;
3067 __be32 src = inet->inet_rcv_saddr;
3068 __u16 destp = ntohs(inet->inet_dport);
3069 __u16 srcp = ntohs(inet->inet_sport);
db8dac20 3070
f86dcc5a 3071 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
ea9a0379 3072 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
db8dac20 3073 bucket, src, srcp, dest, destp, sp->sk_state,
31e6d363 3074 sk_wmem_alloc_get(sp),
6c206b20 3075 udp_rqueue_get(sp),
a7cb5a49
EB		 3076 0, 0L, 0,
3077 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3078 0, sock_i_ino(sp),
41c6d650 3079 refcount_read(&sp->sk_refcnt), sp,
652586df 3080 atomic_read(&sp->sk_drops));
db8dac20
DM		 3081}
3082
3083int udp4_seq_show(struct seq_file *seq, void *v)
3084{
652586df 3085 seq_setwidth(seq, 127);
db8dac20 3086 if (v == SEQ_START_TOKEN)
6c25449e 3087 seq_puts(seq, " sl local_address rem_address st tx_queue "
db8dac20 3088 "rx_queue tr tm->when retrnsmt uid timeout "
cb61cb9b 3089 "inode ref pointer drops");
db8dac20 3090 else {
db8dac20
DM		 3091 struct udp_iter_state *state = seq->private;
3092
652586df 3093 udp4_format_sock(v, seq, state->bucket);
db8dac20 3094 }
652586df 3095 seq_pad(seq, '\n');
db8dac20
DM		 3096 return 0;
3097}
3098
5788b3a0
YS		 3099#ifdef CONFIG_BPF_SYSCALL
3100struct bpf_iter__udp {
3101 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3102 __bpf_md_ptr(struct udp_sock *, udp_sk);
3103 uid_t uid __aligned(8);
3104 int bucket __aligned(8);
3105};
3106
3107static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3108 struct udp_sock *udp_sk, uid_t uid, int bucket)
3109{
3110 struct bpf_iter__udp ctx;
3111
3112 meta->seq_num--; /* skip SEQ_START_TOKEN */
3113 ctx.meta = meta;
3114 ctx.udp_sk = udp_sk;
3115 ctx.uid = uid;
3116 ctx.bucket = bucket;
3117 return bpf_iter_run_prog(prog, &ctx);
3118}
3119
3120static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3121{
3122 struct udp_iter_state *state = seq->private;
3123 struct bpf_iter_meta meta;
3124 struct bpf_prog *prog;
3125 struct sock *sk = v;
3126 uid_t uid;
3127
3128 if (v == SEQ_START_TOKEN)
3129 return 0;
3130
3131 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3132 meta.seq = seq;
3133 prog = bpf_iter_get_info(&meta, false);
3134 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3135}
3136
3137static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3138{
3139 struct bpf_iter_meta meta;
3140 struct bpf_prog *prog;
3141
3142 if (!v) {
3143 meta.seq = seq;
3144 prog = bpf_iter_get_info(&meta, true);
3145 if (prog)
3146 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3147 }
3148
3149 udp_seq_stop(seq, v);
3150}
3151
3152static const struct seq_operations bpf_iter_udp_seq_ops = {
3153 .start = udp_seq_start,
3154 .next = udp_seq_next,
3155 .stop = bpf_iter_udp_seq_stop,
3156 .show = bpf_iter_udp_seq_show,
3157};
3158#endif
3159
c3506372 3160const struct seq_operations udp_seq_ops = {
a3d2599b
CH		 3161 .start = udp_seq_start,
3162 .next = udp_seq_next,
3163 .stop = udp_seq_stop,
3164 .show = udp4_seq_show,
3165};
c3506372 3166EXPORT_SYMBOL(udp_seq_ops);
73cb88ec 3167
db8dac20 3168static struct udp_seq_afinfo udp4_seq_afinfo = {
db8dac20 3169 .family = AF_INET,
645ca708 3170 .udp_table = &udp_table,
db8dac20
DM		 3171};
3172
2c8c1e72 3173static int __net_init udp4_proc_init_net(struct net *net)
15439feb 3174{
c3506372
CH		 3175 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3176 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
a3d2599b
CH		 3177 return -ENOMEM;
3178 return 0;
15439feb
PE		 3179}
3180
2c8c1e72 3181static void __net_exit udp4_proc_exit_net(struct net *net)
15439feb 3182{
a3d2599b 3183 remove_proc_entry("udp", net->proc_net);
15439feb
PE		 3184}
3185
3186static struct pernet_operations udp4_net_ops = {
3187 .init = udp4_proc_init_net,
3188 .exit = udp4_proc_exit_net,
3189};
3190
db8dac20
DM		 3191int __init udp4_proc_init(void)
3192{
15439feb 3193 return register_pernet_subsys(&udp4_net_ops);
db8dac20
DM		 3194}
3195
3196void udp4_proc_exit(void)
3197{
15439feb 3198 unregister_pernet_subsys(&udp4_net_ops);
db8dac20 3199}
1da177e4
LT		 3200#endif /* CONFIG_PROC_FS */
3201
f86dcc5a
ED		 3202static __initdata unsigned long uhash_entries;
3203static int __init set_uhash_entries(char *str)
645ca708 3204{
413c27d8
EZ		 3205 ssize_t ret;
3206
f86dcc5a
ED		 3207 if (!str)
3208 return 0;
413c27d8
EZ		 3209
3210 ret = kstrtoul(str, 0, &uhash_entries);
3211 if (ret)
3212 return 0;
3213
f86dcc5a
ED		 3214 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3215 uhash_entries = UDP_HTABLE_SIZE_MIN;
3216 return 1;
3217}
3218__setup("uhash_entries=", set_uhash_entries);
645ca708 3219
f86dcc5a
ED		 3220void __init udp_table_init(struct udp_table *table, const char *name)
3221{
3222 unsigned int i;
3223
31fe62b9
TB		 3224 table->hash = alloc_large_system_hash(name,
3225 2 * sizeof(struct udp_hslot),
3226 uhash_entries,
3227 21, /* one slot per 2 MB */
3228 0,
3229 &table->log,
3230 &table->mask,
3231 UDP_HTABLE_SIZE_MIN,
3232 64 * 1024);
3233
512615b6 3234 table->hash2 = table->hash + (table->mask + 1);
f86dcc5a 3235 for (i = 0; i <= table->mask; i++) {
ca065d0c 3236 INIT_HLIST_HEAD(&table->hash[i].head);
fdcc8aa9 3237 table->hash[i].count = 0;
645ca708
ED		 3238 spin_lock_init(&table->hash[i].lock);
3239 }
512615b6 3240 for (i = 0; i <= table->mask; i++) {
ca065d0c 3241 INIT_HLIST_HEAD(&table->hash2[i].head);
512615b6
ED		 3242 table->hash2[i].count = 0;
3243 spin_lock_init(&table->hash2[i].lock);
3244 }
645ca708
ED		 3245}
3246
723b8e46
TH		 3247u32 udp_flow_hashrnd(void)
3248{
3249 static u32 hashrnd __read_mostly;
3250
3251 net_get_random_once(&hashrnd, sizeof(hashrnd));
3252
3253 return hashrnd;
3254}
3255EXPORT_SYMBOL(udp_flow_hashrnd);
3256
02a7cb28 3257static int __net_init udp_sysctl_init(struct net *net)
1e802951 3258{
100fdd1f
ED		 3259 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3260 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
1e802951
TZ		 3261
3262#ifdef CONFIG_NET_L3_MASTER_DEV
3263 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3264#endif
1e802951 3265
1e802951
TZ		 3266 return 0;
3267}
3268
3269static struct pernet_operations __net_initdata udp_sysctl_ops = {
fc18999e 3270 .init = udp_sysctl_init,
1e802951
TZ		 3271};
3272
5788b3a0
YS		 3273#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3274DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3275 struct udp_sock *udp_sk, uid_t uid, int bucket)
3276
f9c79272 3277static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
5788b3a0
YS		 3278{
3279 struct udp_iter_state *st = priv_data;
3280 struct udp_seq_afinfo *afinfo;
3281 int ret;
3282
3283 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3284 if (!afinfo)
3285 return -ENOMEM;
3286
3287 afinfo->family = AF_UNSPEC;
3288 afinfo->udp_table = &udp_table;
3289 st->bpf_seq_afinfo = afinfo;
f9c79272 3290 ret = bpf_iter_init_seq_net(priv_data, aux);
5788b3a0
YS		 3291 if (ret)
3292 kfree(afinfo);
3293 return ret;
3294}
3295
3296static void bpf_iter_fini_udp(void *priv_data)
3297{
3298 struct udp_iter_state *st = priv_data;
3299
3300 kfree(st->bpf_seq_afinfo);
3301 bpf_iter_fini_seq_net(priv_data);
3302}
3303
14fc6bd6 3304static const struct bpf_iter_seq_info udp_seq_info = {
5788b3a0
YS		 3305 .seq_ops = &bpf_iter_udp_seq_ops,
3306 .init_seq_private = bpf_iter_init_udp,
3307 .fini_seq_private = bpf_iter_fini_udp,
3308 .seq_priv_size = sizeof(struct udp_iter_state),
14fc6bd6
YS		 3309};
3310
3311static struct bpf_iter_reg udp_reg_info = {
3312 .target = "udp",
5788b3a0
YS		 3313 .ctx_arg_info_size = 1,
3314 .ctx_arg_info = {
3315 { offsetof(struct bpf_iter__udp, udp_sk),
3316 PTR_TO_BTF_ID_OR_NULL },
3317 },
14fc6bd6 3318 .seq_info = &udp_seq_info,
5788b3a0
YS		 3319};
3320
3321static void __init bpf_iter_register(void)
3322{
951cf368 3323 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
5788b3a0
YS		 3324 if (bpf_iter_reg_target(&udp_reg_info))
3325 pr_warn("Warning: could not register bpf iterator udp\n");
3326}
3327#endif
3328
95766fff
HA		 3329void __init udp_init(void)
3330{
f03d78db 3331 unsigned long limit;
4b272750 3332 unsigned int i;
95766fff 3333
f86dcc5a 3334 udp_table_init(&udp_table, "UDP");
f03d78db 3335 limit = nr_free_buffer_pages() / 8;
95766fff
HA		 3336 limit = max(limit, 128UL);
3337 sysctl_udp_mem[0] = limit / 4 * 3;
3338 sysctl_udp_mem[1] = limit;
3339 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3340
4b272750
ED		 3341 /* 16 spinlocks per cpu */
3342 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3343 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3344 GFP_KERNEL);
3345 if (!udp_busylocks)
3346 panic("UDP: failed to alloc udp_busylocks\n");
3347 for (i = 0; i < (1U << udp_busylocks_log); i++)
3348 spin_lock_init(udp_busylocks + i);
1e802951
TZ		 3349
3350 if (register_pernet_subsys(&udp_sysctl_ops))
3351 panic("UDP: failed to init sysctl parameters.\n");
5788b3a0
YS		 3352
3353#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3354 bpf_iter_register();
3355#endif
95766fff 3356}
Linux 6.x block layer and io_uring tree(s)