1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * The User Datagram Protocol (UDP).
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
13 * Hirokazu Takahashi, <taka@valinux.co.jp>
14 *
15 * Fixes:
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
29 * does NOT close.
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
58 * for connect.
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * datagrams.
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
72 * James Chapman : Add L2TP encapsulation type.
73 */
74
75 #define pr_fmt(fmt) "UDP: " fmt
76
77 #include <linux/bpf-cgroup.h>
78 #include <linux/uaccess.h>
79 #include <asm/ioctls.h>
80 #include <linux/memblock.h>
81 #include <linux/highmem.h>
82 #include <linux/types.h>
83 #include <linux/fcntl.h>
84 #include <linux/module.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/igmp.h>
88 #include <linux/inetdevice.h>
89 #include <linux/in.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
92 #include <linux/mm.h>
93 #include <linux/inet.h>
94 #include <linux/netdevice.h>
95 #include <linux/slab.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/inet_hashtables.h>
103 #include <net/ip.h>
104 #include <net/ip_tunnels.h>
105 #include <net/route.h>
106 #include <net/checksum.h>
107 #include <net/gso.h>
108 #include <net/xfrm.h>
109 #include <trace/events/udp.h>
110 #include <linux/static_key.h>
111 #include <linux/btf_ids.h>
112 #include <trace/events/skb.h>
113 #include <net/busy_poll.h>
114 #include "udp_impl.h"
115 #include <net/sock_reuseport.h>
116 #include <net/addrconf.h>
117 #include <net/udp_tunnel.h>
118 #include <net/gro.h>
119 #if IS_ENABLED(CONFIG_IPV6)
120 #include <net/ipv6_stubs.h>
121 #endif
122
123 struct udp_table udp_table __read_mostly;
124
125 long sysctl_udp_mem[3] __read_mostly;
126 EXPORT_IPV6_MOD(sysctl_udp_mem);
127
128 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
129 EXPORT_IPV6_MOD(udp_memory_allocated);
130 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
131 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
132
133 #define MAX_UDP_PORTS 65536
134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
135
udp_get_table_prot(struct sock * sk)136 static struct udp_table *udp_get_table_prot(struct sock *sk)
137 {
138 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
139 }
140
udp_lib_lport_inuse(struct net * net,__u16 num,const struct udp_hslot * hslot,unsigned long * bitmap,struct sock * sk,unsigned int log)141 static int udp_lib_lport_inuse(struct net *net, __u16 num,
142 const struct udp_hslot *hslot,
143 unsigned long *bitmap,
144 struct sock *sk, unsigned int log)
145 {
146 struct sock *sk2;
147 kuid_t uid = sock_i_uid(sk);
148
149 sk_for_each(sk2, &hslot->head) {
150 if (net_eq(sock_net(sk2), net) &&
151 sk2 != sk &&
152 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
153 (!sk2->sk_reuse || !sk->sk_reuse) &&
154 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
155 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
156 inet_rcv_saddr_equal(sk, sk2, true)) {
157 if (sk2->sk_reuseport && sk->sk_reuseport &&
158 !rcu_access_pointer(sk->sk_reuseport_cb) &&
159 uid_eq(uid, sock_i_uid(sk2))) {
160 if (!bitmap)
161 return 0;
162 } else {
163 if (!bitmap)
164 return 1;
165 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
166 bitmap);
167 }
168 }
169 }
170 return 0;
171 }
172
173 /*
174 * Note: we still hold spinlock of primary hash chain, so no other writer
175 * can insert/delete a socket with local_port == num
176 */
udp_lib_lport_inuse2(struct net * net,__u16 num,struct udp_hslot * hslot2,struct sock * sk)177 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
178 struct udp_hslot *hslot2,
179 struct sock *sk)
180 {
181 struct sock *sk2;
182 kuid_t uid = sock_i_uid(sk);
183 int res = 0;
184
185 spin_lock(&hslot2->lock);
186 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
187 if (net_eq(sock_net(sk2), net) &&
188 sk2 != sk &&
189 (udp_sk(sk2)->udp_port_hash == num) &&
190 (!sk2->sk_reuse || !sk->sk_reuse) &&
191 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
192 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
193 inet_rcv_saddr_equal(sk, sk2, true)) {
194 if (sk2->sk_reuseport && sk->sk_reuseport &&
195 !rcu_access_pointer(sk->sk_reuseport_cb) &&
196 uid_eq(uid, sock_i_uid(sk2))) {
197 res = 0;
198 } else {
199 res = 1;
200 }
201 break;
202 }
203 }
204 spin_unlock(&hslot2->lock);
205 return res;
206 }
207
udp_reuseport_add_sock(struct sock * sk,struct udp_hslot * hslot)208 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
209 {
210 struct net *net = sock_net(sk);
211 kuid_t uid = sock_i_uid(sk);
212 struct sock *sk2;
213
214 sk_for_each(sk2, &hslot->head) {
215 if (net_eq(sock_net(sk2), net) &&
216 sk2 != sk &&
217 sk2->sk_family == sk->sk_family &&
218 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
219 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
220 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
221 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
222 inet_rcv_saddr_equal(sk, sk2, false)) {
223 return reuseport_add_sock(sk, sk2,
224 inet_rcv_saddr_any(sk));
225 }
226 }
227
228 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
229 }
230
231 /**
232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
233 *
234 * @sk: socket struct in question
235 * @snum: port number to look up
236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237 * with NULL address
238 */
udp_lib_get_port(struct sock * sk,unsigned short snum,unsigned int hash2_nulladdr)239 int udp_lib_get_port(struct sock *sk, unsigned short snum,
240 unsigned int hash2_nulladdr)
241 {
242 struct udp_table *udptable = udp_get_table_prot(sk);
243 struct udp_hslot *hslot, *hslot2;
244 struct net *net = sock_net(sk);
245 int error = -EADDRINUSE;
246
247 if (!snum) {
248 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
249 unsigned short first, last;
250 int low, high, remaining;
251 unsigned int rand;
252
253 inet_sk_get_local_port_range(sk, &low, &high);
254 remaining = (high - low) + 1;
255
256 rand = get_random_u32();
257 first = reciprocal_scale(rand, remaining) + low;
258 /*
259 * force rand to be an odd multiple of UDP_HTABLE_SIZE
260 */
261 rand = (rand | 1) * (udptable->mask + 1);
262 last = first + udptable->mask + 1;
263 do {
264 hslot = udp_hashslot(udptable, net, first);
265 bitmap_zero(bitmap, PORTS_PER_CHAIN);
266 spin_lock_bh(&hslot->lock);
267 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
268 udptable->log);
269
270 snum = first;
271 /*
272 * Iterate on all possible values of snum for this hash.
273 * Using steps of an odd multiple of UDP_HTABLE_SIZE
274 * give us randomization and full range coverage.
275 */
276 do {
277 if (low <= snum && snum <= high &&
278 !test_bit(snum >> udptable->log, bitmap) &&
279 !inet_is_local_reserved_port(net, snum))
280 goto found;
281 snum += rand;
282 } while (snum != first);
283 spin_unlock_bh(&hslot->lock);
284 cond_resched();
285 } while (++first != last);
286 goto fail;
287 } else {
288 hslot = udp_hashslot(udptable, net, snum);
289 spin_lock_bh(&hslot->lock);
290 if (hslot->count > 10) {
291 int exist;
292 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
293
294 slot2 &= udptable->mask;
295 hash2_nulladdr &= udptable->mask;
296
297 hslot2 = udp_hashslot2(udptable, slot2);
298 if (hslot->count < hslot2->count)
299 goto scan_primary_hash;
300
301 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
302 if (!exist && (hash2_nulladdr != slot2)) {
303 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
304 exist = udp_lib_lport_inuse2(net, snum, hslot2,
305 sk);
306 }
307 if (exist)
308 goto fail_unlock;
309 else
310 goto found;
311 }
312 scan_primary_hash:
313 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
314 goto fail_unlock;
315 }
316 found:
317 inet_sk(sk)->inet_num = snum;
318 udp_sk(sk)->udp_port_hash = snum;
319 udp_sk(sk)->udp_portaddr_hash ^= snum;
320 if (sk_unhashed(sk)) {
321 if (sk->sk_reuseport &&
322 udp_reuseport_add_sock(sk, hslot)) {
323 inet_sk(sk)->inet_num = 0;
324 udp_sk(sk)->udp_port_hash = 0;
325 udp_sk(sk)->udp_portaddr_hash ^= snum;
326 goto fail_unlock;
327 }
328
329 sock_set_flag(sk, SOCK_RCU_FREE);
330
331 sk_add_node_rcu(sk, &hslot->head);
332 hslot->count++;
333 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
334
335 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
336 spin_lock(&hslot2->lock);
337 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
338 sk->sk_family == AF_INET6)
339 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
340 &hslot2->head);
341 else
342 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
343 &hslot2->head);
344 hslot2->count++;
345 spin_unlock(&hslot2->lock);
346 }
347
348 error = 0;
349 fail_unlock:
350 spin_unlock_bh(&hslot->lock);
351 fail:
352 return error;
353 }
354 EXPORT_IPV6_MOD(udp_lib_get_port);
355
udp_v4_get_port(struct sock * sk,unsigned short snum)356 int udp_v4_get_port(struct sock *sk, unsigned short snum)
357 {
358 unsigned int hash2_nulladdr =
359 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
360 unsigned int hash2_partial =
361 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
362
363 /* precompute partial secondary hash */
364 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
365 return udp_lib_get_port(sk, snum, hash2_nulladdr);
366 }
367
compute_score(struct sock * sk,const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,unsigned short hnum,int dif,int sdif)368 static int compute_score(struct sock *sk, const struct net *net,
369 __be32 saddr, __be16 sport,
370 __be32 daddr, unsigned short hnum,
371 int dif, int sdif)
372 {
373 int score;
374 struct inet_sock *inet;
375 bool dev_match;
376
377 if (!net_eq(sock_net(sk), net) ||
378 udp_sk(sk)->udp_port_hash != hnum ||
379 ipv6_only_sock(sk))
380 return -1;
381
382 if (sk->sk_rcv_saddr != daddr)
383 return -1;
384
385 score = (sk->sk_family == PF_INET) ? 2 : 1;
386
387 inet = inet_sk(sk);
388 if (inet->inet_daddr) {
389 if (inet->inet_daddr != saddr)
390 return -1;
391 score += 4;
392 }
393
394 if (inet->inet_dport) {
395 if (inet->inet_dport != sport)
396 return -1;
397 score += 4;
398 }
399
400 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
401 dif, sdif);
402 if (!dev_match)
403 return -1;
404 if (sk->sk_bound_dev_if)
405 score += 4;
406
407 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
408 score++;
409 return score;
410 }
411
udp_ehashfn(const struct net * net,const __be32 laddr,const __u16 lport,const __be32 faddr,const __be16 fport)412 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
413 const __be32 faddr, const __be16 fport)
414 {
415 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
416
417 return __inet_ehashfn(laddr, lport, faddr, fport,
418 udp_ehash_secret + net_hash_mix(net));
419 }
420 EXPORT_IPV6_MOD(udp_ehashfn);
421
422 /**
423 * udp4_lib_lookup1() - Simplified lookup using primary hash (destination port)
424 * @net: Network namespace
425 * @saddr: Source address, network order
426 * @sport: Source port, network order
427 * @daddr: Destination address, network order
428 * @hnum: Destination port, host order
429 * @dif: Destination interface index
430 * @sdif: Destination bridge port index, if relevant
431 * @udptable: Set of UDP hash tables
432 *
433 * Simplified lookup to be used as fallback if no sockets are found due to a
434 * potential race between (receive) address change, and lookup happening before
435 * the rehash operation. This function ignores SO_REUSEPORT groups while scoring
436 * result sockets, because if we have one, we don't need the fallback at all.
437 *
438 * Called under rcu_read_lock().
439 *
440 * Return: socket with highest matching score if any, NULL if none
441 */
udp4_lib_lookup1(const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,unsigned int hnum,int dif,int sdif,const struct udp_table * udptable)442 static struct sock *udp4_lib_lookup1(const struct net *net,
443 __be32 saddr, __be16 sport,
444 __be32 daddr, unsigned int hnum,
445 int dif, int sdif,
446 const struct udp_table *udptable)
447 {
448 unsigned int slot = udp_hashfn(net, hnum, udptable->mask);
449 struct udp_hslot *hslot = &udptable->hash[slot];
450 struct sock *sk, *result = NULL;
451 int score, badness = 0;
452
453 sk_for_each_rcu(sk, &hslot->head) {
454 score = compute_score(sk, net,
455 saddr, sport, daddr, hnum, dif, sdif);
456 if (score > badness) {
457 result = sk;
458 badness = score;
459 }
460 }
461
462 return result;
463 }
464
465 /* called with rcu_read_lock() */
udp4_lib_lookup2(const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,unsigned int hnum,int dif,int sdif,struct udp_hslot * hslot2,struct sk_buff * skb)466 static struct sock *udp4_lib_lookup2(const struct net *net,
467 __be32 saddr, __be16 sport,
468 __be32 daddr, unsigned int hnum,
469 int dif, int sdif,
470 struct udp_hslot *hslot2,
471 struct sk_buff *skb)
472 {
473 struct sock *sk, *result;
474 int score, badness;
475 bool need_rescore;
476
477 result = NULL;
478 badness = 0;
479 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
480 need_rescore = false;
481 rescore:
482 score = compute_score(need_rescore ? result : sk, net, saddr,
483 sport, daddr, hnum, dif, sdif);
484 if (score > badness) {
485 badness = score;
486
487 if (need_rescore)
488 continue;
489
490 if (sk->sk_state == TCP_ESTABLISHED) {
491 result = sk;
492 continue;
493 }
494
495 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
496 saddr, sport, daddr, hnum, udp_ehashfn);
497 if (!result) {
498 result = sk;
499 continue;
500 }
501
502 /* Fall back to scoring if group has connections */
503 if (!reuseport_has_conns(sk))
504 return result;
505
506 /* Reuseport logic returned an error, keep original score. */
507 if (IS_ERR(result))
508 continue;
509
510 /* compute_score is too long of a function to be
511 * inlined, and calling it again here yields
512 * measureable overhead for some
513 * workloads. Work around it by jumping
514 * backwards to rescore 'result'.
515 */
516 need_rescore = true;
517 goto rescore;
518 }
519 }
520 return result;
521 }
522
523 #if IS_ENABLED(CONFIG_BASE_SMALL)
udp4_lib_lookup4(const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,unsigned int hnum,int dif,int sdif,struct udp_table * udptable)524 static struct sock *udp4_lib_lookup4(const struct net *net,
525 __be32 saddr, __be16 sport,
526 __be32 daddr, unsigned int hnum,
527 int dif, int sdif,
528 struct udp_table *udptable)
529 {
530 return NULL;
531 }
532
udp_rehash4(struct udp_table * udptable,struct sock * sk,u16 newhash4)533 static void udp_rehash4(struct udp_table *udptable, struct sock *sk,
534 u16 newhash4)
535 {
536 }
537
udp_unhash4(struct udp_table * udptable,struct sock * sk)538 static void udp_unhash4(struct udp_table *udptable, struct sock *sk)
539 {
540 }
541 #else /* !CONFIG_BASE_SMALL */
udp4_lib_lookup4(const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,unsigned int hnum,int dif,int sdif,struct udp_table * udptable)542 static struct sock *udp4_lib_lookup4(const struct net *net,
543 __be32 saddr, __be16 sport,
544 __be32 daddr, unsigned int hnum,
545 int dif, int sdif,
546 struct udp_table *udptable)
547 {
548 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
549 const struct hlist_nulls_node *node;
550 struct udp_hslot *hslot4;
551 unsigned int hash4, slot;
552 struct udp_sock *up;
553 struct sock *sk;
554
555 hash4 = udp_ehashfn(net, daddr, hnum, saddr, sport);
556 slot = hash4 & udptable->mask;
557 hslot4 = &udptable->hash4[slot];
558 INET_ADDR_COOKIE(acookie, saddr, daddr);
559
560 begin:
561 /* SLAB_TYPESAFE_BY_RCU not used, so we don't need to touch sk_refcnt */
562 udp_lrpa_for_each_entry_rcu(up, node, &hslot4->nulls_head) {
563 sk = (struct sock *)up;
564 if (inet_match(net, sk, acookie, ports, dif, sdif))
565 return sk;
566 }
567
568 /* if the nulls value we got at the end of this lookup is not the
569 * expected one, we must restart lookup. We probably met an item that
570 * was moved to another chain due to rehash.
571 */
572 if (get_nulls_value(node) != slot)
573 goto begin;
574
575 return NULL;
576 }
577
578 /* udp_rehash4() only checks hslot4, and hash4_cnt is not processed. */
udp_rehash4(struct udp_table * udptable,struct sock * sk,u16 newhash4)579 static void udp_rehash4(struct udp_table *udptable, struct sock *sk,
580 u16 newhash4)
581 {
582 struct udp_hslot *hslot4, *nhslot4;
583
584 hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash);
585 nhslot4 = udp_hashslot4(udptable, newhash4);
586 udp_sk(sk)->udp_lrpa_hash = newhash4;
587
588 if (hslot4 != nhslot4) {
589 spin_lock_bh(&hslot4->lock);
590 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node);
591 hslot4->count--;
592 spin_unlock_bh(&hslot4->lock);
593
594 spin_lock_bh(&nhslot4->lock);
595 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node,
596 &nhslot4->nulls_head);
597 nhslot4->count++;
598 spin_unlock_bh(&nhslot4->lock);
599 }
600 }
601
udp_unhash4(struct udp_table * udptable,struct sock * sk)602 static void udp_unhash4(struct udp_table *udptable, struct sock *sk)
603 {
604 struct udp_hslot *hslot2, *hslot4;
605
606 if (udp_hashed4(sk)) {
607 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
608 hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash);
609
610 spin_lock(&hslot4->lock);
611 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node);
612 hslot4->count--;
613 spin_unlock(&hslot4->lock);
614
615 spin_lock(&hslot2->lock);
616 udp_hash4_dec(hslot2);
617 spin_unlock(&hslot2->lock);
618 }
619 }
620
udp_lib_hash4(struct sock * sk,u16 hash)621 void udp_lib_hash4(struct sock *sk, u16 hash)
622 {
623 struct udp_hslot *hslot, *hslot2, *hslot4;
624 struct net *net = sock_net(sk);
625 struct udp_table *udptable;
626
627 /* Connected udp socket can re-connect to another remote address, which
628 * will be handled by rehash. Thus no need to redo hash4 here.
629 */
630 if (udp_hashed4(sk))
631 return;
632
633 udptable = net->ipv4.udp_table;
634 hslot = udp_hashslot(udptable, net, udp_sk(sk)->udp_port_hash);
635 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
636 hslot4 = udp_hashslot4(udptable, hash);
637 udp_sk(sk)->udp_lrpa_hash = hash;
638
639 spin_lock_bh(&hslot->lock);
640 if (rcu_access_pointer(sk->sk_reuseport_cb))
641 reuseport_detach_sock(sk);
642
643 spin_lock(&hslot4->lock);
644 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node,
645 &hslot4->nulls_head);
646 hslot4->count++;
647 spin_unlock(&hslot4->lock);
648
649 spin_lock(&hslot2->lock);
650 udp_hash4_inc(hslot2);
651 spin_unlock(&hslot2->lock);
652
653 spin_unlock_bh(&hslot->lock);
654 }
655 EXPORT_IPV6_MOD(udp_lib_hash4);
656
657 /* call with sock lock */
udp4_hash4(struct sock * sk)658 void udp4_hash4(struct sock *sk)
659 {
660 struct net *net = sock_net(sk);
661 unsigned int hash;
662
663 if (sk_unhashed(sk) || sk->sk_rcv_saddr == htonl(INADDR_ANY))
664 return;
665
666 hash = udp_ehashfn(net, sk->sk_rcv_saddr, sk->sk_num,
667 sk->sk_daddr, sk->sk_dport);
668
669 udp_lib_hash4(sk, hash);
670 }
671 EXPORT_IPV6_MOD(udp4_hash4);
672 #endif /* CONFIG_BASE_SMALL */
673
674 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
675 * harder than this. -DaveM
676 */
__udp4_lib_lookup(const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,__be16 dport,int dif,int sdif,struct udp_table * udptable,struct sk_buff * skb)677 struct sock *__udp4_lib_lookup(const struct net *net, __be32 saddr,
678 __be16 sport, __be32 daddr, __be16 dport, int dif,
679 int sdif, struct udp_table *udptable, struct sk_buff *skb)
680 {
681 unsigned short hnum = ntohs(dport);
682 struct udp_hslot *hslot2;
683 struct sock *result, *sk;
684 unsigned int hash2;
685
686 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
687 hslot2 = udp_hashslot2(udptable, hash2);
688
689 if (udp_has_hash4(hslot2)) {
690 result = udp4_lib_lookup4(net, saddr, sport, daddr, hnum,
691 dif, sdif, udptable);
692 if (result) /* udp4_lib_lookup4 return sk or NULL */
693 return result;
694 }
695
696 /* Lookup connected or non-wildcard socket */
697 result = udp4_lib_lookup2(net, saddr, sport,
698 daddr, hnum, dif, sdif,
699 hslot2, skb);
700 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
701 goto done;
702
703 /* Lookup redirect from BPF */
704 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
705 udptable == net->ipv4.udp_table) {
706 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
707 saddr, sport, daddr, hnum, dif,
708 udp_ehashfn);
709 if (sk) {
710 result = sk;
711 goto done;
712 }
713 }
714
715 /* Got non-wildcard socket or error on first lookup */
716 if (result)
717 goto done;
718
719 /* Lookup wildcard sockets */
720 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
721 hslot2 = udp_hashslot2(udptable, hash2);
722
723 result = udp4_lib_lookup2(net, saddr, sport,
724 htonl(INADDR_ANY), hnum, dif, sdif,
725 hslot2, skb);
726 if (!IS_ERR_OR_NULL(result))
727 goto done;
728
729 /* Primary hash (destination port) lookup as fallback for this race:
730 * 1. __ip4_datagram_connect() sets sk_rcv_saddr
731 * 2. lookup (this function): new sk_rcv_saddr, hashes not updated yet
732 * 3. rehash operation updating _secondary and four-tuple_ hashes
733 * The primary hash doesn't need an update after 1., so, thanks to this
734 * further step, 1. and 3. don't need to be atomic against the lookup.
735 */
736 result = udp4_lib_lookup1(net, saddr, sport, daddr, hnum, dif, sdif,
737 udptable);
738
739 done:
740 if (IS_ERR(result))
741 return NULL;
742 return result;
743 }
744 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
745
__udp4_lib_lookup_skb(struct sk_buff * skb,__be16 sport,__be16 dport,struct udp_table * udptable)746 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
747 __be16 sport, __be16 dport,
748 struct udp_table *udptable)
749 {
750 const struct iphdr *iph = ip_hdr(skb);
751
752 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
753 iph->daddr, dport, inet_iif(skb),
754 inet_sdif(skb), udptable, skb);
755 }
756
udp4_lib_lookup_skb(const struct sk_buff * skb,__be16 sport,__be16 dport)757 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
758 __be16 sport, __be16 dport)
759 {
760 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
761 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
762 struct net *net = dev_net(skb->dev);
763 int iif, sdif;
764
765 inet_get_iif_sdif(skb, &iif, &sdif);
766
767 return __udp4_lib_lookup(net, iph->saddr, sport,
768 iph->daddr, dport, iif,
769 sdif, net->ipv4.udp_table, NULL);
770 }
771
772 /* Must be called under rcu_read_lock().
773 * Does increment socket refcount.
774 */
775 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
udp4_lib_lookup(const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,__be16 dport,int dif)776 struct sock *udp4_lib_lookup(const struct net *net, __be32 saddr, __be16 sport,
777 __be32 daddr, __be16 dport, int dif)
778 {
779 struct sock *sk;
780
781 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
782 dif, 0, net->ipv4.udp_table, NULL);
783 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
784 sk = NULL;
785 return sk;
786 }
787 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
788 #endif
789
__udp_is_mcast_sock(struct net * net,const struct sock * sk,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif,unsigned short hnum)790 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
791 __be16 loc_port, __be32 loc_addr,
792 __be16 rmt_port, __be32 rmt_addr,
793 int dif, int sdif, unsigned short hnum)
794 {
795 const struct inet_sock *inet = inet_sk(sk);
796
797 if (!net_eq(sock_net(sk), net) ||
798 udp_sk(sk)->udp_port_hash != hnum ||
799 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
800 (inet->inet_dport != rmt_port && inet->inet_dport) ||
801 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
802 ipv6_only_sock(sk) ||
803 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
804 return false;
805 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
806 return false;
807 return true;
808 }
809
810 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
811 EXPORT_IPV6_MOD(udp_encap_needed_key);
812
813 #if IS_ENABLED(CONFIG_IPV6)
814 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
815 EXPORT_IPV6_MOD(udpv6_encap_needed_key);
816 #endif
817
udp_encap_enable(void)818 void udp_encap_enable(void)
819 {
820 static_branch_inc(&udp_encap_needed_key);
821 }
822 EXPORT_SYMBOL(udp_encap_enable);
823
udp_encap_disable(void)824 void udp_encap_disable(void)
825 {
826 static_branch_dec(&udp_encap_needed_key);
827 }
828 EXPORT_SYMBOL(udp_encap_disable);
829
830 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
831 * through error handlers in encapsulations looking for a match.
832 */
__udp4_lib_err_encap_no_sk(struct sk_buff * skb,u32 info)833 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
834 {
835 int i;
836
837 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
838 int (*handler)(struct sk_buff *skb, u32 info);
839 const struct ip_tunnel_encap_ops *encap;
840
841 encap = rcu_dereference(iptun_encaps[i]);
842 if (!encap)
843 continue;
844 handler = encap->err_handler;
845 if (handler && !handler(skb, info))
846 return 0;
847 }
848
849 return -ENOENT;
850 }
851
852 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
853 * reversing source and destination port: this will match tunnels that force the
854 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
855 * lwtunnels might actually break this assumption by being configured with
856 * different destination ports on endpoints, in this case we won't be able to
857 * trace ICMP messages back to them.
858 *
859 * If this doesn't match any socket, probe tunnels with arbitrary destination
860 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
861 * we've sent packets to won't necessarily match the local destination port.
862 *
863 * Then ask the tunnel implementation to match the error against a valid
864 * association.
865 *
866 * Return an error if we can't find a match, the socket if we need further
867 * processing, zero otherwise.
868 */
__udp4_lib_err_encap(struct net * net,const struct iphdr * iph,struct udphdr * uh,struct udp_table * udptable,struct sock * sk,struct sk_buff * skb,u32 info)869 static struct sock *__udp4_lib_err_encap(struct net *net,
870 const struct iphdr *iph,
871 struct udphdr *uh,
872 struct udp_table *udptable,
873 struct sock *sk,
874 struct sk_buff *skb, u32 info)
875 {
876 int (*lookup)(struct sock *sk, struct sk_buff *skb);
877 int network_offset, transport_offset;
878 struct udp_sock *up;
879
880 network_offset = skb_network_offset(skb);
881 transport_offset = skb_transport_offset(skb);
882
883 /* Network header needs to point to the outer IPv4 header inside ICMP */
884 skb_reset_network_header(skb);
885
886 /* Transport header needs to point to the UDP header */
887 skb_set_transport_header(skb, iph->ihl << 2);
888
889 if (sk) {
890 up = udp_sk(sk);
891
892 lookup = READ_ONCE(up->encap_err_lookup);
893 if (lookup && lookup(sk, skb))
894 sk = NULL;
895
896 goto out;
897 }
898
899 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
900 iph->saddr, uh->dest, skb->dev->ifindex, 0,
901 udptable, NULL);
902 if (sk) {
903 up = udp_sk(sk);
904
905 lookup = READ_ONCE(up->encap_err_lookup);
906 if (!lookup || lookup(sk, skb))
907 sk = NULL;
908 }
909
910 out:
911 if (!sk)
912 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
913
914 skb_set_transport_header(skb, transport_offset);
915 skb_set_network_header(skb, network_offset);
916
917 return sk;
918 }
919
920 /*
921 * This routine is called by the ICMP module when it gets some
922 * sort of error condition. If err < 0 then the socket should
923 * be closed and the error returned to the user. If err > 0
924 * it's just the icmp type << 8 | icmp code.
925 * Header points to the ip header of the error packet. We move
926 * on past this. Then (as it used to claim before adjustment)
927 * header points to the first 8 bytes of the udp header. We need
928 * to find the appropriate port.
929 */
930
__udp4_lib_err(struct sk_buff * skb,u32 info,struct udp_table * udptable)931 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
932 {
933 struct inet_sock *inet;
934 const struct iphdr *iph = (const struct iphdr *)skb->data;
935 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
936 const int type = icmp_hdr(skb)->type;
937 const int code = icmp_hdr(skb)->code;
938 bool tunnel = false;
939 struct sock *sk;
940 int harderr;
941 int err;
942 struct net *net = dev_net(skb->dev);
943
944 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
945 iph->saddr, uh->source, skb->dev->ifindex,
946 inet_sdif(skb), udptable, NULL);
947
948 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
949 /* No socket for error: try tunnels before discarding */
950 if (static_branch_unlikely(&udp_encap_needed_key)) {
951 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
952 info);
953 if (!sk)
954 return 0;
955 } else
956 sk = ERR_PTR(-ENOENT);
957
958 if (IS_ERR(sk)) {
959 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
960 return PTR_ERR(sk);
961 }
962
963 tunnel = true;
964 }
965
966 err = 0;
967 harderr = 0;
968 inet = inet_sk(sk);
969
970 switch (type) {
971 default:
972 case ICMP_TIME_EXCEEDED:
973 err = EHOSTUNREACH;
974 break;
975 case ICMP_SOURCE_QUENCH:
976 goto out;
977 case ICMP_PARAMETERPROB:
978 err = EPROTO;
979 harderr = 1;
980 break;
981 case ICMP_DEST_UNREACH:
982 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
983 ipv4_sk_update_pmtu(skb, sk, info);
984 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
985 err = EMSGSIZE;
986 harderr = 1;
987 break;
988 }
989 goto out;
990 }
991 err = EHOSTUNREACH;
992 if (code <= NR_ICMP_UNREACH) {
993 harderr = icmp_err_convert[code].fatal;
994 err = icmp_err_convert[code].errno;
995 }
996 break;
997 case ICMP_REDIRECT:
998 ipv4_sk_redirect(skb, sk);
999 goto out;
1000 }
1001
1002 /*
1003 * RFC1122: OK. Passes ICMP errors back to application, as per
1004 * 4.1.3.3.
1005 */
1006 if (tunnel) {
1007 /* ...not for tunnels though: we don't have a sending socket */
1008 if (udp_sk(sk)->encap_err_rcv)
1009 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
1010 (u8 *)(uh+1));
1011 goto out;
1012 }
1013 if (!inet_test_bit(RECVERR, sk)) {
1014 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
1015 goto out;
1016 } else
1017 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
1018
1019 sk->sk_err = err;
1020 sk_error_report(sk);
1021 out:
1022 return 0;
1023 }
1024
udp_err(struct sk_buff * skb,u32 info)1025 int udp_err(struct sk_buff *skb, u32 info)
1026 {
1027 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
1028 }
1029
1030 /*
1031 * Throw away all pending data and cancel the corking. Socket is locked.
1032 */
udp_flush_pending_frames(struct sock * sk)1033 void udp_flush_pending_frames(struct sock *sk)
1034 {
1035 struct udp_sock *up = udp_sk(sk);
1036
1037 if (up->pending) {
1038 up->len = 0;
1039 WRITE_ONCE(up->pending, 0);
1040 ip_flush_pending_frames(sk);
1041 }
1042 }
1043 EXPORT_IPV6_MOD(udp_flush_pending_frames);
1044
1045 /**
1046 * udp4_hwcsum - handle outgoing HW checksumming
1047 * @skb: sk_buff containing the filled-in UDP header
1048 * (checksum field must be zeroed out)
1049 * @src: source IP address
1050 * @dst: destination IP address
1051 */
udp4_hwcsum(struct sk_buff * skb,__be32 src,__be32 dst)1052 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
1053 {
1054 struct udphdr *uh = udp_hdr(skb);
1055 int offset = skb_transport_offset(skb);
1056 int len = skb->len - offset;
1057 int hlen = len;
1058 __wsum csum = 0;
1059
1060 if (!skb_has_frag_list(skb)) {
1061 /*
1062 * Only one fragment on the socket.
1063 */
1064 skb->csum_start = skb_transport_header(skb) - skb->head;
1065 skb->csum_offset = offsetof(struct udphdr, check);
1066 uh->check = ~csum_tcpudp_magic(src, dst, len,
1067 IPPROTO_UDP, 0);
1068 } else {
1069 struct sk_buff *frags;
1070
1071 /*
1072 * HW-checksum won't work as there are two or more
1073 * fragments on the socket so that all csums of sk_buffs
1074 * should be together
1075 */
1076 skb_walk_frags(skb, frags) {
1077 csum = csum_add(csum, frags->csum);
1078 hlen -= frags->len;
1079 }
1080
1081 csum = skb_checksum(skb, offset, hlen, csum);
1082 skb->ip_summed = CHECKSUM_NONE;
1083
1084 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
1085 if (uh->check == 0)
1086 uh->check = CSUM_MANGLED_0;
1087 }
1088 }
1089 EXPORT_SYMBOL_GPL(udp4_hwcsum);
1090
1091 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
1092 * for the simple case like when setting the checksum for a UDP tunnel.
1093 */
udp_set_csum(bool nocheck,struct sk_buff * skb,__be32 saddr,__be32 daddr,int len)1094 void udp_set_csum(bool nocheck, struct sk_buff *skb,
1095 __be32 saddr, __be32 daddr, int len)
1096 {
1097 struct udphdr *uh = udp_hdr(skb);
1098
1099 if (nocheck) {
1100 uh->check = 0;
1101 } else if (skb_is_gso(skb)) {
1102 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
1103 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
1104 uh->check = 0;
1105 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
1106 if (uh->check == 0)
1107 uh->check = CSUM_MANGLED_0;
1108 } else {
1109 skb->ip_summed = CHECKSUM_PARTIAL;
1110 skb->csum_start = skb_transport_header(skb) - skb->head;
1111 skb->csum_offset = offsetof(struct udphdr, check);
1112 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
1113 }
1114 }
1115 EXPORT_SYMBOL(udp_set_csum);
1116
udp_send_skb(struct sk_buff * skb,struct flowi4 * fl4,struct inet_cork * cork)1117 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
1118 struct inet_cork *cork)
1119 {
1120 struct sock *sk = skb->sk;
1121 struct inet_sock *inet = inet_sk(sk);
1122 struct udphdr *uh;
1123 int err;
1124 int is_udplite = IS_UDPLITE(sk);
1125 int offset = skb_transport_offset(skb);
1126 int len = skb->len - offset;
1127 int datalen = len - sizeof(*uh);
1128 __wsum csum = 0;
1129
1130 /*
1131 * Create a UDP header
1132 */
1133 uh = udp_hdr(skb);
1134 uh->source = inet->inet_sport;
1135 uh->dest = fl4->fl4_dport;
1136 uh->len = htons(len);
1137 uh->check = 0;
1138
1139 if (cork->gso_size) {
1140 const int hlen = skb_network_header_len(skb) +
1141 sizeof(struct udphdr);
1142
1143 if (hlen + min(datalen, cork->gso_size) > cork->fragsize) {
1144 kfree_skb(skb);
1145 return -EMSGSIZE;
1146 }
1147 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1148 kfree_skb(skb);
1149 return -EINVAL;
1150 }
1151 if (sk->sk_no_check_tx) {
1152 kfree_skb(skb);
1153 return -EINVAL;
1154 }
1155 if (is_udplite || dst_xfrm(skb_dst(skb))) {
1156 kfree_skb(skb);
1157 return -EIO;
1158 }
1159
1160 if (datalen > cork->gso_size) {
1161 skb_shinfo(skb)->gso_size = cork->gso_size;
1162 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1163 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1164 cork->gso_size);
1165
1166 /* Don't checksum the payload, skb will get segmented */
1167 goto csum_partial;
1168 }
1169 }
1170
1171 if (is_udplite) /* UDP-Lite */
1172 csum = udplite_csum(skb);
1173
1174 else if (sk->sk_no_check_tx) { /* UDP csum off */
1175
1176 skb->ip_summed = CHECKSUM_NONE;
1177 goto send;
1178
1179 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1180 csum_partial:
1181
1182 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
1183 goto send;
1184
1185 } else
1186 csum = udp_csum(skb);
1187
1188 /* add protocol-dependent pseudo-header */
1189 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
1190 sk->sk_protocol, csum);
1191 if (uh->check == 0)
1192 uh->check = CSUM_MANGLED_0;
1193
1194 send:
1195 err = ip_send_skb(sock_net(sk), skb);
1196 if (err) {
1197 if (err == -ENOBUFS &&
1198 !inet_test_bit(RECVERR, sk)) {
1199 UDP_INC_STATS(sock_net(sk),
1200 UDP_MIB_SNDBUFERRORS, is_udplite);
1201 err = 0;
1202 }
1203 } else
1204 UDP_INC_STATS(sock_net(sk),
1205 UDP_MIB_OUTDATAGRAMS, is_udplite);
1206 return err;
1207 }
1208
1209 /*
1210 * Push out all pending data as one UDP datagram. Socket is locked.
1211 */
udp_push_pending_frames(struct sock * sk)1212 int udp_push_pending_frames(struct sock *sk)
1213 {
1214 struct udp_sock *up = udp_sk(sk);
1215 struct inet_sock *inet = inet_sk(sk);
1216 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1217 struct sk_buff *skb;
1218 int err = 0;
1219
1220 skb = ip_finish_skb(sk, fl4);
1221 if (!skb)
1222 goto out;
1223
1224 err = udp_send_skb(skb, fl4, &inet->cork.base);
1225
1226 out:
1227 up->len = 0;
1228 WRITE_ONCE(up->pending, 0);
1229 return err;
1230 }
1231 EXPORT_IPV6_MOD(udp_push_pending_frames);
1232
__udp_cmsg_send(struct cmsghdr * cmsg,u16 * gso_size)1233 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1234 {
1235 switch (cmsg->cmsg_type) {
1236 case UDP_SEGMENT:
1237 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1238 return -EINVAL;
1239 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1240 return 0;
1241 default:
1242 return -EINVAL;
1243 }
1244 }
1245
udp_cmsg_send(struct sock * sk,struct msghdr * msg,u16 * gso_size)1246 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1247 {
1248 struct cmsghdr *cmsg;
1249 bool need_ip = false;
1250 int err;
1251
1252 for_each_cmsghdr(cmsg, msg) {
1253 if (!CMSG_OK(msg, cmsg))
1254 return -EINVAL;
1255
1256 if (cmsg->cmsg_level != SOL_UDP) {
1257 need_ip = true;
1258 continue;
1259 }
1260
1261 err = __udp_cmsg_send(cmsg, gso_size);
1262 if (err)
1263 return err;
1264 }
1265
1266 return need_ip;
1267 }
1268 EXPORT_IPV6_MOD_GPL(udp_cmsg_send);
1269
udp_sendmsg(struct sock * sk,struct msghdr * msg,size_t len)1270 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1271 {
1272 struct inet_sock *inet = inet_sk(sk);
1273 struct udp_sock *up = udp_sk(sk);
1274 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1275 struct flowi4 fl4_stack;
1276 struct flowi4 *fl4;
1277 int ulen = len;
1278 struct ipcm_cookie ipc;
1279 struct rtable *rt = NULL;
1280 int free = 0;
1281 int connected = 0;
1282 __be32 daddr, faddr, saddr;
1283 u8 scope;
1284 __be16 dport;
1285 int err, is_udplite = IS_UDPLITE(sk);
1286 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1287 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1288 struct sk_buff *skb;
1289 struct ip_options_data opt_copy;
1290 int uc_index;
1291
1292 if (len > 0xFFFF)
1293 return -EMSGSIZE;
1294
1295 /*
1296 * Check the flags.
1297 */
1298
1299 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1300 return -EOPNOTSUPP;
1301
1302 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1303
1304 fl4 = &inet->cork.fl.u.ip4;
1305 if (READ_ONCE(up->pending)) {
1306 /*
1307 * There are pending frames.
1308 * The socket lock must be held while it's corked.
1309 */
1310 lock_sock(sk);
1311 if (likely(up->pending)) {
1312 if (unlikely(up->pending != AF_INET)) {
1313 release_sock(sk);
1314 return -EINVAL;
1315 }
1316 goto do_append_data;
1317 }
1318 release_sock(sk);
1319 }
1320 ulen += sizeof(struct udphdr);
1321
1322 /*
1323 * Get and verify the address.
1324 */
1325 if (usin) {
1326 if (msg->msg_namelen < sizeof(*usin))
1327 return -EINVAL;
1328 if (usin->sin_family != AF_INET) {
1329 if (usin->sin_family != AF_UNSPEC)
1330 return -EAFNOSUPPORT;
1331 }
1332
1333 daddr = usin->sin_addr.s_addr;
1334 dport = usin->sin_port;
1335 if (dport == 0)
1336 return -EINVAL;
1337 } else {
1338 if (sk->sk_state != TCP_ESTABLISHED)
1339 return -EDESTADDRREQ;
1340 daddr = inet->inet_daddr;
1341 dport = inet->inet_dport;
1342 /* Open fast path for connected socket.
1343 Route will not be used, if at least one option is set.
1344 */
1345 connected = 1;
1346 }
1347
1348 ipcm_init_sk(&ipc, inet);
1349 ipc.gso_size = READ_ONCE(up->gso_size);
1350
1351 if (msg->msg_controllen) {
1352 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1353 if (err > 0) {
1354 err = ip_cmsg_send(sk, msg, &ipc,
1355 sk->sk_family == AF_INET6);
1356 connected = 0;
1357 }
1358 if (unlikely(err < 0)) {
1359 kfree(ipc.opt);
1360 return err;
1361 }
1362 if (ipc.opt)
1363 free = 1;
1364 }
1365 if (!ipc.opt) {
1366 struct ip_options_rcu *inet_opt;
1367
1368 rcu_read_lock();
1369 inet_opt = rcu_dereference(inet->inet_opt);
1370 if (inet_opt) {
1371 memcpy(&opt_copy, inet_opt,
1372 sizeof(*inet_opt) + inet_opt->opt.optlen);
1373 ipc.opt = &opt_copy.opt;
1374 }
1375 rcu_read_unlock();
1376 }
1377
1378 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1379 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1380 (struct sockaddr *)usin,
1381 &msg->msg_namelen,
1382 &ipc.addr);
1383 if (err)
1384 goto out_free;
1385 if (usin) {
1386 if (usin->sin_port == 0) {
1387 /* BPF program set invalid port. Reject it. */
1388 err = -EINVAL;
1389 goto out_free;
1390 }
1391 daddr = usin->sin_addr.s_addr;
1392 dport = usin->sin_port;
1393 }
1394 }
1395
1396 saddr = ipc.addr;
1397 ipc.addr = faddr = daddr;
1398
1399 if (ipc.opt && ipc.opt->opt.srr) {
1400 if (!daddr) {
1401 err = -EINVAL;
1402 goto out_free;
1403 }
1404 faddr = ipc.opt->opt.faddr;
1405 connected = 0;
1406 }
1407 scope = ip_sendmsg_scope(inet, &ipc, msg);
1408 if (scope == RT_SCOPE_LINK)
1409 connected = 0;
1410
1411 uc_index = READ_ONCE(inet->uc_index);
1412 if (ipv4_is_multicast(daddr)) {
1413 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1414 ipc.oif = READ_ONCE(inet->mc_index);
1415 if (!saddr)
1416 saddr = READ_ONCE(inet->mc_addr);
1417 connected = 0;
1418 } else if (!ipc.oif) {
1419 ipc.oif = uc_index;
1420 } else if (ipv4_is_lbcast(daddr) && uc_index) {
1421 /* oif is set, packet is to local broadcast and
1422 * uc_index is set. oif is most likely set
1423 * by sk_bound_dev_if. If uc_index != oif check if the
1424 * oif is an L3 master and uc_index is an L3 slave.
1425 * If so, we want to allow the send using the uc_index.
1426 */
1427 if (ipc.oif != uc_index &&
1428 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1429 uc_index)) {
1430 ipc.oif = uc_index;
1431 }
1432 }
1433
1434 if (connected)
1435 rt = dst_rtable(sk_dst_check(sk, 0));
1436
1437 if (!rt) {
1438 struct net *net = sock_net(sk);
1439 __u8 flow_flags = inet_sk_flowi_flags(sk);
1440
1441 fl4 = &fl4_stack;
1442
1443 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark,
1444 ipc.tos & INET_DSCP_MASK, scope,
1445 sk->sk_protocol, flow_flags, faddr, saddr,
1446 dport, inet->inet_sport, sk->sk_uid);
1447
1448 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1449 rt = ip_route_output_flow(net, fl4, sk);
1450 if (IS_ERR(rt)) {
1451 err = PTR_ERR(rt);
1452 rt = NULL;
1453 if (err == -ENETUNREACH)
1454 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1455 goto out;
1456 }
1457
1458 err = -EACCES;
1459 if ((rt->rt_flags & RTCF_BROADCAST) &&
1460 !sock_flag(sk, SOCK_BROADCAST))
1461 goto out;
1462 if (connected)
1463 sk_dst_set(sk, dst_clone(&rt->dst));
1464 }
1465
1466 if (msg->msg_flags&MSG_CONFIRM)
1467 goto do_confirm;
1468 back_from_confirm:
1469
1470 saddr = fl4->saddr;
1471 if (!ipc.addr)
1472 daddr = ipc.addr = fl4->daddr;
1473
1474 /* Lockless fast path for the non-corking case. */
1475 if (!corkreq) {
1476 struct inet_cork cork;
1477
1478 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1479 sizeof(struct udphdr), &ipc, &rt,
1480 &cork, msg->msg_flags);
1481 err = PTR_ERR(skb);
1482 if (!IS_ERR_OR_NULL(skb))
1483 err = udp_send_skb(skb, fl4, &cork);
1484 goto out;
1485 }
1486
1487 lock_sock(sk);
1488 if (unlikely(up->pending)) {
1489 /* The socket is already corked while preparing it. */
1490 /* ... which is an evident application bug. --ANK */
1491 release_sock(sk);
1492
1493 net_dbg_ratelimited("socket already corked\n");
1494 err = -EINVAL;
1495 goto out;
1496 }
1497 /*
1498 * Now cork the socket to pend data.
1499 */
1500 fl4 = &inet->cork.fl.u.ip4;
1501 fl4->daddr = daddr;
1502 fl4->saddr = saddr;
1503 fl4->fl4_dport = dport;
1504 fl4->fl4_sport = inet->inet_sport;
1505 WRITE_ONCE(up->pending, AF_INET);
1506
1507 do_append_data:
1508 up->len += ulen;
1509 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1510 sizeof(struct udphdr), &ipc, &rt,
1511 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1512 if (err)
1513 udp_flush_pending_frames(sk);
1514 else if (!corkreq)
1515 err = udp_push_pending_frames(sk);
1516 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1517 WRITE_ONCE(up->pending, 0);
1518 release_sock(sk);
1519
1520 out:
1521 ip_rt_put(rt);
1522 out_free:
1523 if (free)
1524 kfree(ipc.opt);
1525 if (!err)
1526 return len;
1527 /*
1528 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1529 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1530 * we don't have a good statistic (IpOutDiscards but it can be too many
1531 * things). We could add another new stat but at least for now that
1532 * seems like overkill.
1533 */
1534 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1535 UDP_INC_STATS(sock_net(sk),
1536 UDP_MIB_SNDBUFERRORS, is_udplite);
1537 }
1538 return err;
1539
1540 do_confirm:
1541 if (msg->msg_flags & MSG_PROBE)
1542 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1543 if (!(msg->msg_flags&MSG_PROBE) || len)
1544 goto back_from_confirm;
1545 err = 0;
1546 goto out;
1547 }
1548 EXPORT_SYMBOL(udp_sendmsg);
1549
udp_splice_eof(struct socket * sock)1550 void udp_splice_eof(struct socket *sock)
1551 {
1552 struct sock *sk = sock->sk;
1553 struct udp_sock *up = udp_sk(sk);
1554
1555 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1556 return;
1557
1558 lock_sock(sk);
1559 if (up->pending && !udp_test_bit(CORK, sk))
1560 udp_push_pending_frames(sk);
1561 release_sock(sk);
1562 }
1563 EXPORT_IPV6_MOD_GPL(udp_splice_eof);
1564
1565 #define UDP_SKB_IS_STATELESS 0x80000000
1566
1567 /* all head states (dst, sk, nf conntrack) except skb extensions are
1568 * cleared by udp_rcv().
1569 *
1570 * We need to preserve secpath, if present, to eventually process
1571 * IP_CMSG_PASSSEC at recvmsg() time.
1572 *
1573 * Other extensions can be cleared.
1574 */
udp_try_make_stateless(struct sk_buff * skb)1575 static bool udp_try_make_stateless(struct sk_buff *skb)
1576 {
1577 if (!skb_has_extensions(skb))
1578 return true;
1579
1580 if (!secpath_exists(skb)) {
1581 skb_ext_reset(skb);
1582 return true;
1583 }
1584
1585 return false;
1586 }
1587
udp_set_dev_scratch(struct sk_buff * skb)1588 static void udp_set_dev_scratch(struct sk_buff *skb)
1589 {
1590 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1591
1592 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1593 scratch->_tsize_state = skb->truesize;
1594 #if BITS_PER_LONG == 64
1595 scratch->len = skb->len;
1596 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1597 scratch->is_linear = !skb_is_nonlinear(skb);
1598 #endif
1599 if (udp_try_make_stateless(skb))
1600 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1601 }
1602
udp_skb_csum_unnecessary_set(struct sk_buff * skb)1603 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1604 {
1605 /* We come here after udp_lib_checksum_complete() returned 0.
1606 * This means that __skb_checksum_complete() might have
1607 * set skb->csum_valid to 1.
1608 * On 64bit platforms, we can set csum_unnecessary
1609 * to true, but only if the skb is not shared.
1610 */
1611 #if BITS_PER_LONG == 64
1612 if (!skb_shared(skb))
1613 udp_skb_scratch(skb)->csum_unnecessary = true;
1614 #endif
1615 }
1616
udp_skb_truesize(struct sk_buff * skb)1617 static int udp_skb_truesize(struct sk_buff *skb)
1618 {
1619 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1620 }
1621
udp_skb_has_head_state(struct sk_buff * skb)1622 static bool udp_skb_has_head_state(struct sk_buff *skb)
1623 {
1624 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1625 }
1626
1627 /* fully reclaim rmem/fwd memory allocated for skb */
udp_rmem_release(struct sock * sk,int size,int partial,bool rx_queue_lock_held)1628 static void udp_rmem_release(struct sock *sk, int size, int partial,
1629 bool rx_queue_lock_held)
1630 {
1631 struct udp_sock *up = udp_sk(sk);
1632 struct sk_buff_head *sk_queue;
1633 int amt;
1634
1635 if (likely(partial)) {
1636 up->forward_deficit += size;
1637 size = up->forward_deficit;
1638 if (size < READ_ONCE(up->forward_threshold) &&
1639 !skb_queue_empty(&up->reader_queue))
1640 return;
1641 } else {
1642 size += up->forward_deficit;
1643 }
1644 up->forward_deficit = 0;
1645
1646 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1647 * if the called don't held it already
1648 */
1649 sk_queue = &sk->sk_receive_queue;
1650 if (!rx_queue_lock_held)
1651 spin_lock(&sk_queue->lock);
1652
1653
1654 sk_forward_alloc_add(sk, size);
1655 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1656 sk_forward_alloc_add(sk, -amt);
1657
1658 if (amt)
1659 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1660
1661 atomic_sub(size, &sk->sk_rmem_alloc);
1662
1663 /* this can save us from acquiring the rx queue lock on next receive */
1664 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1665
1666 if (!rx_queue_lock_held)
1667 spin_unlock(&sk_queue->lock);
1668 }
1669
1670 /* Note: called with reader_queue.lock held.
1671 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1672 * This avoids a cache line miss while receive_queue lock is held.
1673 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1674 */
udp_skb_destructor(struct sock * sk,struct sk_buff * skb)1675 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1676 {
1677 prefetch(&skb->data);
1678 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1679 }
1680 EXPORT_IPV6_MOD(udp_skb_destructor);
1681
1682 /* as above, but the caller held the rx queue lock, too */
udp_skb_dtor_locked(struct sock * sk,struct sk_buff * skb)1683 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1684 {
1685 prefetch(&skb->data);
1686 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1687 }
1688
1689 /* Idea of busylocks is to let producers grab an extra spinlock
1690 * to relieve pressure on the receive_queue spinlock shared by consumer.
1691 * Under flood, this means that only one producer can be in line
1692 * trying to acquire the receive_queue spinlock.
1693 * These busylock can be allocated on a per cpu manner, instead of a
1694 * per socket one (that would consume a cache line per socket)
1695 */
1696 static int udp_busylocks_log __read_mostly;
1697 static spinlock_t *udp_busylocks __read_mostly;
1698
busylock_acquire(void * ptr)1699 static spinlock_t *busylock_acquire(void *ptr)
1700 {
1701 spinlock_t *busy;
1702
1703 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1704 spin_lock(busy);
1705 return busy;
1706 }
1707
busylock_release(spinlock_t * busy)1708 static void busylock_release(spinlock_t *busy)
1709 {
1710 if (busy)
1711 spin_unlock(busy);
1712 }
1713
udp_rmem_schedule(struct sock * sk,int size)1714 static int udp_rmem_schedule(struct sock *sk, int size)
1715 {
1716 int delta;
1717
1718 delta = size - sk->sk_forward_alloc;
1719 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1720 return -ENOBUFS;
1721
1722 return 0;
1723 }
1724
__udp_enqueue_schedule_skb(struct sock * sk,struct sk_buff * skb)1725 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1726 {
1727 struct sk_buff_head *list = &sk->sk_receive_queue;
1728 int rmem, err = -ENOMEM;
1729 spinlock_t *busy = NULL;
1730 int size, rcvbuf;
1731
1732 /* Immediately drop when the receive queue is full.
1733 * Always allow at least one packet.
1734 */
1735 rmem = atomic_read(&sk->sk_rmem_alloc);
1736 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1737 if (rmem > rcvbuf)
1738 goto drop;
1739
1740 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1741 * having linear skbs :
1742 * - Reduce memory overhead and thus increase receive queue capacity
1743 * - Less cache line misses at copyout() time
1744 * - Less work at consume_skb() (less alien page frag freeing)
1745 */
1746 if (rmem > (rcvbuf >> 1)) {
1747 skb_condense(skb);
1748
1749 busy = busylock_acquire(sk);
1750 }
1751 size = skb->truesize;
1752 udp_set_dev_scratch(skb);
1753
1754 atomic_add(size, &sk->sk_rmem_alloc);
1755
1756 spin_lock(&list->lock);
1757 err = udp_rmem_schedule(sk, size);
1758 if (err) {
1759 spin_unlock(&list->lock);
1760 goto uncharge_drop;
1761 }
1762
1763 sk_forward_alloc_add(sk, -size);
1764
1765 /* no need to setup a destructor, we will explicitly release the
1766 * forward allocated memory on dequeue
1767 */
1768 sock_skb_set_dropcount(sk, skb);
1769
1770 __skb_queue_tail(list, skb);
1771 spin_unlock(&list->lock);
1772
1773 if (!sock_flag(sk, SOCK_DEAD))
1774 INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
1775
1776 busylock_release(busy);
1777 return 0;
1778
1779 uncharge_drop:
1780 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1781
1782 drop:
1783 atomic_inc(&sk->sk_drops);
1784 busylock_release(busy);
1785 return err;
1786 }
1787 EXPORT_IPV6_MOD_GPL(__udp_enqueue_schedule_skb);
1788
udp_destruct_common(struct sock * sk)1789 void udp_destruct_common(struct sock *sk)
1790 {
1791 /* reclaim completely the forward allocated memory */
1792 struct udp_sock *up = udp_sk(sk);
1793 unsigned int total = 0;
1794 struct sk_buff *skb;
1795
1796 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1797 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1798 total += skb->truesize;
1799 kfree_skb(skb);
1800 }
1801 udp_rmem_release(sk, total, 0, true);
1802 }
1803 EXPORT_IPV6_MOD_GPL(udp_destruct_common);
1804
udp_destruct_sock(struct sock * sk)1805 static void udp_destruct_sock(struct sock *sk)
1806 {
1807 udp_destruct_common(sk);
1808 inet_sock_destruct(sk);
1809 }
1810
udp_init_sock(struct sock * sk)1811 int udp_init_sock(struct sock *sk)
1812 {
1813 udp_lib_init_sock(sk);
1814 sk->sk_destruct = udp_destruct_sock;
1815 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1816 return 0;
1817 }
1818
skb_consume_udp(struct sock * sk,struct sk_buff * skb,int len)1819 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1820 {
1821 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
1822 sk_peek_offset_bwd(sk, len);
1823
1824 if (!skb_unref(skb))
1825 return;
1826
1827 /* In the more common cases we cleared the head states previously,
1828 * see __udp_queue_rcv_skb().
1829 */
1830 if (unlikely(udp_skb_has_head_state(skb)))
1831 skb_release_head_state(skb);
1832 __consume_stateless_skb(skb);
1833 }
1834 EXPORT_IPV6_MOD_GPL(skb_consume_udp);
1835
__first_packet_length(struct sock * sk,struct sk_buff_head * rcvq,int * total)1836 static struct sk_buff *__first_packet_length(struct sock *sk,
1837 struct sk_buff_head *rcvq,
1838 int *total)
1839 {
1840 struct sk_buff *skb;
1841
1842 while ((skb = skb_peek(rcvq)) != NULL) {
1843 if (udp_lib_checksum_complete(skb)) {
1844 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1845 IS_UDPLITE(sk));
1846 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1847 IS_UDPLITE(sk));
1848 atomic_inc(&sk->sk_drops);
1849 __skb_unlink(skb, rcvq);
1850 *total += skb->truesize;
1851 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM);
1852 } else {
1853 udp_skb_csum_unnecessary_set(skb);
1854 break;
1855 }
1856 }
1857 return skb;
1858 }
1859
1860 /**
1861 * first_packet_length - return length of first packet in receive queue
1862 * @sk: socket
1863 *
1864 * Drops all bad checksum frames, until a valid one is found.
1865 * Returns the length of found skb, or -1 if none is found.
1866 */
first_packet_length(struct sock * sk)1867 static int first_packet_length(struct sock *sk)
1868 {
1869 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1870 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1871 struct sk_buff *skb;
1872 int total = 0;
1873 int res;
1874
1875 spin_lock_bh(&rcvq->lock);
1876 skb = __first_packet_length(sk, rcvq, &total);
1877 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1878 spin_lock(&sk_queue->lock);
1879 skb_queue_splice_tail_init(sk_queue, rcvq);
1880 spin_unlock(&sk_queue->lock);
1881
1882 skb = __first_packet_length(sk, rcvq, &total);
1883 }
1884 res = skb ? skb->len : -1;
1885 if (total)
1886 udp_rmem_release(sk, total, 1, false);
1887 spin_unlock_bh(&rcvq->lock);
1888 return res;
1889 }
1890
1891 /*
1892 * IOCTL requests applicable to the UDP protocol
1893 */
1894
udp_ioctl(struct sock * sk,int cmd,int * karg)1895 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1896 {
1897 switch (cmd) {
1898 case SIOCOUTQ:
1899 {
1900 *karg = sk_wmem_alloc_get(sk);
1901 return 0;
1902 }
1903
1904 case SIOCINQ:
1905 {
1906 *karg = max_t(int, 0, first_packet_length(sk));
1907 return 0;
1908 }
1909
1910 default:
1911 return -ENOIOCTLCMD;
1912 }
1913
1914 return 0;
1915 }
1916 EXPORT_IPV6_MOD(udp_ioctl);
1917
__skb_recv_udp(struct sock * sk,unsigned int flags,int * off,int * err)1918 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1919 int *off, int *err)
1920 {
1921 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1922 struct sk_buff_head *queue;
1923 struct sk_buff *last;
1924 long timeo;
1925 int error;
1926
1927 queue = &udp_sk(sk)->reader_queue;
1928 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1929 do {
1930 struct sk_buff *skb;
1931
1932 error = sock_error(sk);
1933 if (error)
1934 break;
1935
1936 error = -EAGAIN;
1937 do {
1938 spin_lock_bh(&queue->lock);
1939 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1940 err, &last);
1941 if (skb) {
1942 if (!(flags & MSG_PEEK))
1943 udp_skb_destructor(sk, skb);
1944 spin_unlock_bh(&queue->lock);
1945 return skb;
1946 }
1947
1948 if (skb_queue_empty_lockless(sk_queue)) {
1949 spin_unlock_bh(&queue->lock);
1950 goto busy_check;
1951 }
1952
1953 /* refill the reader queue and walk it again
1954 * keep both queues locked to avoid re-acquiring
1955 * the sk_receive_queue lock if fwd memory scheduling
1956 * is needed.
1957 */
1958 spin_lock(&sk_queue->lock);
1959 skb_queue_splice_tail_init(sk_queue, queue);
1960
1961 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1962 err, &last);
1963 if (skb && !(flags & MSG_PEEK))
1964 udp_skb_dtor_locked(sk, skb);
1965 spin_unlock(&sk_queue->lock);
1966 spin_unlock_bh(&queue->lock);
1967 if (skb)
1968 return skb;
1969
1970 busy_check:
1971 if (!sk_can_busy_loop(sk))
1972 break;
1973
1974 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1975 } while (!skb_queue_empty_lockless(sk_queue));
1976
1977 /* sk_queue is empty, reader_queue may contain peeked packets */
1978 } while (timeo &&
1979 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1980 &error, &timeo,
1981 (struct sk_buff *)sk_queue));
1982
1983 *err = error;
1984 return NULL;
1985 }
1986 EXPORT_SYMBOL(__skb_recv_udp);
1987
udp_read_skb(struct sock * sk,skb_read_actor_t recv_actor)1988 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1989 {
1990 struct sk_buff *skb;
1991 int err;
1992
1993 try_again:
1994 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1995 if (!skb)
1996 return err;
1997
1998 if (udp_lib_checksum_complete(skb)) {
1999 int is_udplite = IS_UDPLITE(sk);
2000 struct net *net = sock_net(sk);
2001
2002 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
2003 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
2004 atomic_inc(&sk->sk_drops);
2005 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM);
2006 goto try_again;
2007 }
2008
2009 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
2010 return recv_actor(sk, skb);
2011 }
2012 EXPORT_IPV6_MOD(udp_read_skb);
2013
2014 /*
2015 * This should be easy, if there is something there we
2016 * return it, otherwise we block.
2017 */
2018
udp_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int flags,int * addr_len)2019 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
2020 int *addr_len)
2021 {
2022 struct inet_sock *inet = inet_sk(sk);
2023 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
2024 struct sk_buff *skb;
2025 unsigned int ulen, copied;
2026 int off, err, peeking = flags & MSG_PEEK;
2027 int is_udplite = IS_UDPLITE(sk);
2028 bool checksum_valid = false;
2029
2030 if (flags & MSG_ERRQUEUE)
2031 return ip_recv_error(sk, msg, len, addr_len);
2032
2033 try_again:
2034 off = sk_peek_offset(sk, flags);
2035 skb = __skb_recv_udp(sk, flags, &off, &err);
2036 if (!skb)
2037 return err;
2038
2039 ulen = udp_skb_len(skb);
2040 copied = len;
2041 if (copied > ulen - off)
2042 copied = ulen - off;
2043 else if (copied < ulen)
2044 msg->msg_flags |= MSG_TRUNC;
2045
2046 /*
2047 * If checksum is needed at all, try to do it while copying the
2048 * data. If the data is truncated, or if we only want a partial
2049 * coverage checksum (UDP-Lite), do it before the copy.
2050 */
2051
2052 if (copied < ulen || peeking ||
2053 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
2054 checksum_valid = udp_skb_csum_unnecessary(skb) ||
2055 !__udp_lib_checksum_complete(skb);
2056 if (!checksum_valid)
2057 goto csum_copy_err;
2058 }
2059
2060 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
2061 if (udp_skb_is_linear(skb))
2062 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
2063 else
2064 err = skb_copy_datagram_msg(skb, off, msg, copied);
2065 } else {
2066 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
2067
2068 if (err == -EINVAL)
2069 goto csum_copy_err;
2070 }
2071
2072 if (unlikely(err)) {
2073 if (!peeking) {
2074 atomic_inc(&sk->sk_drops);
2075 UDP_INC_STATS(sock_net(sk),
2076 UDP_MIB_INERRORS, is_udplite);
2077 }
2078 kfree_skb(skb);
2079 return err;
2080 }
2081
2082 if (!peeking)
2083 UDP_INC_STATS(sock_net(sk),
2084 UDP_MIB_INDATAGRAMS, is_udplite);
2085
2086 sock_recv_cmsgs(msg, sk, skb);
2087
2088 /* Copy the address. */
2089 if (sin) {
2090 sin->sin_family = AF_INET;
2091 sin->sin_port = udp_hdr(skb)->source;
2092 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
2093 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
2094 *addr_len = sizeof(*sin);
2095
2096 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
2097 (struct sockaddr *)sin,
2098 addr_len);
2099 }
2100
2101 if (udp_test_bit(GRO_ENABLED, sk))
2102 udp_cmsg_recv(msg, sk, skb);
2103
2104 if (inet_cmsg_flags(inet))
2105 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
2106
2107 err = copied;
2108 if (flags & MSG_TRUNC)
2109 err = ulen;
2110
2111 skb_consume_udp(sk, skb, peeking ? -err : err);
2112 return err;
2113
2114 csum_copy_err:
2115 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
2116 udp_skb_destructor)) {
2117 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2118 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2119 }
2120 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM);
2121
2122 /* starting over for a new packet, but check if we need to yield */
2123 cond_resched();
2124 msg->msg_flags &= ~MSG_TRUNC;
2125 goto try_again;
2126 }
2127
udp_pre_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)2128 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
2129 {
2130 /* This check is replicated from __ip4_datagram_connect() and
2131 * intended to prevent BPF program called below from accessing bytes
2132 * that are out of the bound specified by user in addr_len.
2133 */
2134 if (addr_len < sizeof(struct sockaddr_in))
2135 return -EINVAL;
2136
2137 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
2138 }
2139 EXPORT_IPV6_MOD(udp_pre_connect);
2140
udp_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)2141 static int udp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
2142 {
2143 int res;
2144
2145 lock_sock(sk);
2146 res = __ip4_datagram_connect(sk, uaddr, addr_len);
2147 if (!res)
2148 udp4_hash4(sk);
2149 release_sock(sk);
2150 return res;
2151 }
2152
__udp_disconnect(struct sock * sk,int flags)2153 int __udp_disconnect(struct sock *sk, int flags)
2154 {
2155 struct inet_sock *inet = inet_sk(sk);
2156 /*
2157 * 1003.1g - break association.
2158 */
2159
2160 sk->sk_state = TCP_CLOSE;
2161 inet->inet_daddr = 0;
2162 inet->inet_dport = 0;
2163 sock_rps_reset_rxhash(sk);
2164 sk->sk_bound_dev_if = 0;
2165 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
2166 inet_reset_saddr(sk);
2167 if (sk->sk_prot->rehash &&
2168 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2169 sk->sk_prot->rehash(sk);
2170 }
2171
2172 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
2173 sk->sk_prot->unhash(sk);
2174 inet->inet_sport = 0;
2175 }
2176 sk_dst_reset(sk);
2177 return 0;
2178 }
2179 EXPORT_SYMBOL(__udp_disconnect);
2180
udp_disconnect(struct sock * sk,int flags)2181 int udp_disconnect(struct sock *sk, int flags)
2182 {
2183 lock_sock(sk);
2184 __udp_disconnect(sk, flags);
2185 release_sock(sk);
2186 return 0;
2187 }
2188 EXPORT_IPV6_MOD(udp_disconnect);
2189
udp_lib_unhash(struct sock * sk)2190 void udp_lib_unhash(struct sock *sk)
2191 {
2192 if (sk_hashed(sk)) {
2193 struct udp_table *udptable = udp_get_table_prot(sk);
2194 struct udp_hslot *hslot, *hslot2;
2195
2196 hslot = udp_hashslot(udptable, sock_net(sk),
2197 udp_sk(sk)->udp_port_hash);
2198 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2199
2200 spin_lock_bh(&hslot->lock);
2201 if (rcu_access_pointer(sk->sk_reuseport_cb))
2202 reuseport_detach_sock(sk);
2203 if (sk_del_node_init_rcu(sk)) {
2204 hslot->count--;
2205 inet_sk(sk)->inet_num = 0;
2206 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2207
2208 spin_lock(&hslot2->lock);
2209 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2210 hslot2->count--;
2211 spin_unlock(&hslot2->lock);
2212
2213 udp_unhash4(udptable, sk);
2214 }
2215 spin_unlock_bh(&hslot->lock);
2216 }
2217 }
2218 EXPORT_IPV6_MOD(udp_lib_unhash);
2219
2220 /*
2221 * inet_rcv_saddr was changed, we must rehash secondary hash
2222 */
udp_lib_rehash(struct sock * sk,u16 newhash,u16 newhash4)2223 void udp_lib_rehash(struct sock *sk, u16 newhash, u16 newhash4)
2224 {
2225 if (sk_hashed(sk)) {
2226 struct udp_table *udptable = udp_get_table_prot(sk);
2227 struct udp_hslot *hslot, *hslot2, *nhslot2;
2228
2229 hslot = udp_hashslot(udptable, sock_net(sk),
2230 udp_sk(sk)->udp_port_hash);
2231 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2232 nhslot2 = udp_hashslot2(udptable, newhash);
2233 udp_sk(sk)->udp_portaddr_hash = newhash;
2234
2235 if (hslot2 != nhslot2 ||
2236 rcu_access_pointer(sk->sk_reuseport_cb)) {
2237 /* we must lock primary chain too */
2238 spin_lock_bh(&hslot->lock);
2239 if (rcu_access_pointer(sk->sk_reuseport_cb))
2240 reuseport_detach_sock(sk);
2241
2242 if (hslot2 != nhslot2) {
2243 spin_lock(&hslot2->lock);
2244 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2245 hslot2->count--;
2246 spin_unlock(&hslot2->lock);
2247
2248 spin_lock(&nhslot2->lock);
2249 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2250 &nhslot2->head);
2251 nhslot2->count++;
2252 spin_unlock(&nhslot2->lock);
2253 }
2254
2255 spin_unlock_bh(&hslot->lock);
2256 }
2257
2258 /* Now process hash4 if necessary:
2259 * (1) update hslot4;
2260 * (2) update hslot2->hash4_cnt.
2261 * Note that hslot2/hslot4 should be checked separately, as
2262 * either of them may change with the other unchanged.
2263 */
2264 if (udp_hashed4(sk)) {
2265 spin_lock_bh(&hslot->lock);
2266
2267 udp_rehash4(udptable, sk, newhash4);
2268 if (hslot2 != nhslot2) {
2269 spin_lock(&hslot2->lock);
2270 udp_hash4_dec(hslot2);
2271 spin_unlock(&hslot2->lock);
2272
2273 spin_lock(&nhslot2->lock);
2274 udp_hash4_inc(nhslot2);
2275 spin_unlock(&nhslot2->lock);
2276 }
2277
2278 spin_unlock_bh(&hslot->lock);
2279 }
2280 }
2281 }
2282 EXPORT_IPV6_MOD(udp_lib_rehash);
2283
udp_v4_rehash(struct sock * sk)2284 void udp_v4_rehash(struct sock *sk)
2285 {
2286 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2287 inet_sk(sk)->inet_rcv_saddr,
2288 inet_sk(sk)->inet_num);
2289 u16 new_hash4 = udp_ehashfn(sock_net(sk),
2290 sk->sk_rcv_saddr, sk->sk_num,
2291 sk->sk_daddr, sk->sk_dport);
2292
2293 udp_lib_rehash(sk, new_hash, new_hash4);
2294 }
2295
__udp_queue_rcv_skb(struct sock * sk,struct sk_buff * skb)2296 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2297 {
2298 int rc;
2299
2300 if (inet_sk(sk)->inet_daddr) {
2301 sock_rps_save_rxhash(sk, skb);
2302 sk_mark_napi_id(sk, skb);
2303 sk_incoming_cpu_update(sk);
2304 } else {
2305 sk_mark_napi_id_once(sk, skb);
2306 }
2307
2308 rc = __udp_enqueue_schedule_skb(sk, skb);
2309 if (rc < 0) {
2310 int is_udplite = IS_UDPLITE(sk);
2311 int drop_reason;
2312
2313 /* Note that an ENOMEM error is charged twice */
2314 if (rc == -ENOMEM) {
2315 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2316 is_udplite);
2317 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2318 } else {
2319 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2320 is_udplite);
2321 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2322 }
2323 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2324 trace_udp_fail_queue_rcv_skb(rc, sk, skb);
2325 sk_skb_reason_drop(sk, skb, drop_reason);
2326 return -1;
2327 }
2328
2329 return 0;
2330 }
2331
2332 /* returns:
2333 * -1: error
2334 * 0: success
2335 * >0: "udp encap" protocol resubmission
2336 *
2337 * Note that in the success and error cases, the skb is assumed to
2338 * have either been requeued or freed.
2339 */
udp_queue_rcv_one_skb(struct sock * sk,struct sk_buff * skb)2340 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2341 {
2342 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2343 struct udp_sock *up = udp_sk(sk);
2344 int is_udplite = IS_UDPLITE(sk);
2345
2346 /*
2347 * Charge it to the socket, dropping if the queue is full.
2348 */
2349 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2350 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2351 goto drop;
2352 }
2353 nf_reset_ct(skb);
2354
2355 if (static_branch_unlikely(&udp_encap_needed_key) &&
2356 READ_ONCE(up->encap_type)) {
2357 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2358
2359 /*
2360 * This is an encapsulation socket so pass the skb to
2361 * the socket's udp_encap_rcv() hook. Otherwise, just
2362 * fall through and pass this up the UDP socket.
2363 * up->encap_rcv() returns the following value:
2364 * =0 if skb was successfully passed to the encap
2365 * handler or was discarded by it.
2366 * >0 if skb should be passed on to UDP.
2367 * <0 if skb should be resubmitted as proto -N
2368 */
2369
2370 /* if we're overly short, let UDP handle it */
2371 encap_rcv = READ_ONCE(up->encap_rcv);
2372 if (encap_rcv) {
2373 int ret;
2374
2375 /* Verify checksum before giving to encap */
2376 if (udp_lib_checksum_complete(skb))
2377 goto csum_error;
2378
2379 ret = encap_rcv(sk, skb);
2380 if (ret <= 0) {
2381 __UDP_INC_STATS(sock_net(sk),
2382 UDP_MIB_INDATAGRAMS,
2383 is_udplite);
2384 return -ret;
2385 }
2386 }
2387
2388 /* FALLTHROUGH -- it's a UDP Packet */
2389 }
2390
2391 /*
2392 * UDP-Lite specific tests, ignored on UDP sockets
2393 */
2394 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2395 u16 pcrlen = READ_ONCE(up->pcrlen);
2396
2397 /*
2398 * MIB statistics other than incrementing the error count are
2399 * disabled for the following two types of errors: these depend
2400 * on the application settings, not on the functioning of the
2401 * protocol stack as such.
2402 *
2403 * RFC 3828 here recommends (sec 3.3): "There should also be a
2404 * way ... to ... at least let the receiving application block
2405 * delivery of packets with coverage values less than a value
2406 * provided by the application."
2407 */
2408 if (pcrlen == 0) { /* full coverage was set */
2409 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2410 UDP_SKB_CB(skb)->cscov, skb->len);
2411 goto drop;
2412 }
2413 /* The next case involves violating the min. coverage requested
2414 * by the receiver. This is subtle: if receiver wants x and x is
2415 * greater than the buffersize/MTU then receiver will complain
2416 * that it wants x while sender emits packets of smaller size y.
2417 * Therefore the above ...()->partial_cov statement is essential.
2418 */
2419 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2420 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2421 UDP_SKB_CB(skb)->cscov, pcrlen);
2422 goto drop;
2423 }
2424 }
2425
2426 prefetch(&sk->sk_rmem_alloc);
2427 if (rcu_access_pointer(sk->sk_filter) &&
2428 udp_lib_checksum_complete(skb))
2429 goto csum_error;
2430
2431 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2432 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2433 goto drop;
2434 }
2435
2436 udp_csum_pull_header(skb);
2437
2438 ipv4_pktinfo_prepare(sk, skb, true);
2439 return __udp_queue_rcv_skb(sk, skb);
2440
2441 csum_error:
2442 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2443 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2444 drop:
2445 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2446 atomic_inc(&sk->sk_drops);
2447 sk_skb_reason_drop(sk, skb, drop_reason);
2448 return -1;
2449 }
2450
udp_queue_rcv_skb(struct sock * sk,struct sk_buff * skb)2451 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2452 {
2453 struct sk_buff *next, *segs;
2454 int ret;
2455
2456 if (likely(!udp_unexpected_gso(sk, skb)))
2457 return udp_queue_rcv_one_skb(sk, skb);
2458
2459 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2460 __skb_push(skb, -skb_mac_offset(skb));
2461 segs = udp_rcv_segment(sk, skb, true);
2462 skb_list_walk_safe(segs, skb, next) {
2463 __skb_pull(skb, skb_transport_offset(skb));
2464
2465 udp_post_segment_fix_csum(skb);
2466 ret = udp_queue_rcv_one_skb(sk, skb);
2467 if (ret > 0)
2468 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2469 }
2470 return 0;
2471 }
2472
2473 /* For TCP sockets, sk_rx_dst is protected by socket lock
2474 * For UDP, we use xchg() to guard against concurrent changes.
2475 */
udp_sk_rx_dst_set(struct sock * sk,struct dst_entry * dst)2476 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2477 {
2478 struct dst_entry *old;
2479
2480 if (dst_hold_safe(dst)) {
2481 old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst)));
2482 dst_release(old);
2483 return old != dst;
2484 }
2485 return false;
2486 }
2487 EXPORT_IPV6_MOD(udp_sk_rx_dst_set);
2488
2489 /*
2490 * Multicasts and broadcasts go to each listener.
2491 *
2492 * Note: called only from the BH handler context.
2493 */
__udp4_lib_mcast_deliver(struct net * net,struct sk_buff * skb,struct udphdr * uh,__be32 saddr,__be32 daddr,struct udp_table * udptable,int proto)2494 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2495 struct udphdr *uh,
2496 __be32 saddr, __be32 daddr,
2497 struct udp_table *udptable,
2498 int proto)
2499 {
2500 struct sock *sk, *first = NULL;
2501 unsigned short hnum = ntohs(uh->dest);
2502 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2503 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2504 unsigned int offset = offsetof(typeof(*sk), sk_node);
2505 int dif = skb->dev->ifindex;
2506 int sdif = inet_sdif(skb);
2507 struct hlist_node *node;
2508 struct sk_buff *nskb;
2509
2510 if (use_hash2) {
2511 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2512 udptable->mask;
2513 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2514 start_lookup:
2515 hslot = &udptable->hash2[hash2].hslot;
2516 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2517 }
2518
2519 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2520 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2521 uh->source, saddr, dif, sdif, hnum))
2522 continue;
2523
2524 if (!first) {
2525 first = sk;
2526 continue;
2527 }
2528 nskb = skb_clone(skb, GFP_ATOMIC);
2529
2530 if (unlikely(!nskb)) {
2531 atomic_inc(&sk->sk_drops);
2532 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2533 IS_UDPLITE(sk));
2534 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2535 IS_UDPLITE(sk));
2536 continue;
2537 }
2538 if (udp_queue_rcv_skb(sk, nskb) > 0)
2539 consume_skb(nskb);
2540 }
2541
2542 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2543 if (use_hash2 && hash2 != hash2_any) {
2544 hash2 = hash2_any;
2545 goto start_lookup;
2546 }
2547
2548 if (first) {
2549 if (udp_queue_rcv_skb(first, skb) > 0)
2550 consume_skb(skb);
2551 } else {
2552 kfree_skb(skb);
2553 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2554 proto == IPPROTO_UDPLITE);
2555 }
2556 return 0;
2557 }
2558
2559 /* Initialize UDP checksum. If exited with zero value (success),
2560 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2561 * Otherwise, csum completion requires checksumming packet body,
2562 * including udp header and folding it to skb->csum.
2563 */
udp4_csum_init(struct sk_buff * skb,struct udphdr * uh,int proto)2564 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2565 int proto)
2566 {
2567 int err;
2568
2569 UDP_SKB_CB(skb)->partial_cov = 0;
2570 UDP_SKB_CB(skb)->cscov = skb->len;
2571
2572 if (proto == IPPROTO_UDPLITE) {
2573 err = udplite_checksum_init(skb, uh);
2574 if (err)
2575 return err;
2576
2577 if (UDP_SKB_CB(skb)->partial_cov) {
2578 skb->csum = inet_compute_pseudo(skb, proto);
2579 return 0;
2580 }
2581 }
2582
2583 /* Note, we are only interested in != 0 or == 0, thus the
2584 * force to int.
2585 */
2586 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2587 inet_compute_pseudo);
2588 if (err)
2589 return err;
2590
2591 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2592 /* If SW calculated the value, we know it's bad */
2593 if (skb->csum_complete_sw)
2594 return 1;
2595
2596 /* HW says the value is bad. Let's validate that.
2597 * skb->csum is no longer the full packet checksum,
2598 * so don't treat it as such.
2599 */
2600 skb_checksum_complete_unset(skb);
2601 }
2602
2603 return 0;
2604 }
2605
2606 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2607 * return code conversion for ip layer consumption
2608 */
udp_unicast_rcv_skb(struct sock * sk,struct sk_buff * skb,struct udphdr * uh)2609 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2610 struct udphdr *uh)
2611 {
2612 int ret;
2613
2614 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2615 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2616
2617 ret = udp_queue_rcv_skb(sk, skb);
2618
2619 /* a return value > 0 means to resubmit the input, but
2620 * it wants the return to be -protocol, or 0
2621 */
2622 if (ret > 0)
2623 return -ret;
2624 return 0;
2625 }
2626
2627 /*
2628 * All we need to do is get the socket, and then do a checksum.
2629 */
2630
__udp4_lib_rcv(struct sk_buff * skb,struct udp_table * udptable,int proto)2631 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2632 int proto)
2633 {
2634 struct sock *sk = NULL;
2635 struct udphdr *uh;
2636 unsigned short ulen;
2637 struct rtable *rt = skb_rtable(skb);
2638 __be32 saddr, daddr;
2639 struct net *net = dev_net(skb->dev);
2640 bool refcounted;
2641 int drop_reason;
2642
2643 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2644
2645 /*
2646 * Validate the packet.
2647 */
2648 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2649 goto drop; /* No space for header. */
2650
2651 uh = udp_hdr(skb);
2652 ulen = ntohs(uh->len);
2653 saddr = ip_hdr(skb)->saddr;
2654 daddr = ip_hdr(skb)->daddr;
2655
2656 if (ulen > skb->len)
2657 goto short_packet;
2658
2659 if (proto == IPPROTO_UDP) {
2660 /* UDP validates ulen. */
2661 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2662 goto short_packet;
2663 uh = udp_hdr(skb);
2664 }
2665
2666 if (udp4_csum_init(skb, uh, proto))
2667 goto csum_error;
2668
2669 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2670 &refcounted, udp_ehashfn);
2671 if (IS_ERR(sk))
2672 goto no_sk;
2673
2674 if (sk) {
2675 struct dst_entry *dst = skb_dst(skb);
2676 int ret;
2677
2678 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2679 udp_sk_rx_dst_set(sk, dst);
2680
2681 ret = udp_unicast_rcv_skb(sk, skb, uh);
2682 if (refcounted)
2683 sock_put(sk);
2684 return ret;
2685 }
2686
2687 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2688 return __udp4_lib_mcast_deliver(net, skb, uh,
2689 saddr, daddr, udptable, proto);
2690
2691 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2692 if (sk)
2693 return udp_unicast_rcv_skb(sk, skb, uh);
2694 no_sk:
2695 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2696 goto drop;
2697 nf_reset_ct(skb);
2698
2699 /* No socket. Drop packet silently, if checksum is wrong */
2700 if (udp_lib_checksum_complete(skb))
2701 goto csum_error;
2702
2703 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2704 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2705 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2706
2707 /*
2708 * Hmm. We got an UDP packet to a port to which we
2709 * don't wanna listen. Ignore it.
2710 */
2711 sk_skb_reason_drop(sk, skb, drop_reason);
2712 return 0;
2713
2714 short_packet:
2715 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2716 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2717 proto == IPPROTO_UDPLITE ? "Lite" : "",
2718 &saddr, ntohs(uh->source),
2719 ulen, skb->len,
2720 &daddr, ntohs(uh->dest));
2721 goto drop;
2722
2723 csum_error:
2724 /*
2725 * RFC1122: OK. Discards the bad packet silently (as far as
2726 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2727 */
2728 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2729 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2730 proto == IPPROTO_UDPLITE ? "Lite" : "",
2731 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2732 ulen);
2733 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2734 drop:
2735 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2736 sk_skb_reason_drop(sk, skb, drop_reason);
2737 return 0;
2738 }
2739
2740 /* We can only early demux multicast if there is a single matching socket.
2741 * If more than one socket found returns NULL
2742 */
__udp4_lib_mcast_demux_lookup(struct net * net,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif)2743 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2744 __be16 loc_port, __be32 loc_addr,
2745 __be16 rmt_port, __be32 rmt_addr,
2746 int dif, int sdif)
2747 {
2748 struct udp_table *udptable = net->ipv4.udp_table;
2749 unsigned short hnum = ntohs(loc_port);
2750 struct sock *sk, *result;
2751 struct udp_hslot *hslot;
2752 unsigned int slot;
2753
2754 slot = udp_hashfn(net, hnum, udptable->mask);
2755 hslot = &udptable->hash[slot];
2756
2757 /* Do not bother scanning a too big list */
2758 if (hslot->count > 10)
2759 return NULL;
2760
2761 result = NULL;
2762 sk_for_each_rcu(sk, &hslot->head) {
2763 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2764 rmt_port, rmt_addr, dif, sdif, hnum)) {
2765 if (result)
2766 return NULL;
2767 result = sk;
2768 }
2769 }
2770
2771 return result;
2772 }
2773
2774 /* For unicast we should only early demux connected sockets or we can
2775 * break forwarding setups. The chains here can be long so only check
2776 * if the first socket is an exact match and if not move on.
2777 */
__udp4_lib_demux_lookup(struct net * net,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif)2778 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2779 __be16 loc_port, __be32 loc_addr,
2780 __be16 rmt_port, __be32 rmt_addr,
2781 int dif, int sdif)
2782 {
2783 struct udp_table *udptable = net->ipv4.udp_table;
2784 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2785 unsigned short hnum = ntohs(loc_port);
2786 struct udp_hslot *hslot2;
2787 unsigned int hash2;
2788 __portpair ports;
2789 struct sock *sk;
2790
2791 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2792 hslot2 = udp_hashslot2(udptable, hash2);
2793 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2794
2795 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2796 if (inet_match(net, sk, acookie, ports, dif, sdif))
2797 return sk;
2798 /* Only check first socket in chain */
2799 break;
2800 }
2801 return NULL;
2802 }
2803
udp_v4_early_demux(struct sk_buff * skb)2804 int udp_v4_early_demux(struct sk_buff *skb)
2805 {
2806 struct net *net = dev_net(skb->dev);
2807 struct in_device *in_dev = NULL;
2808 const struct iphdr *iph;
2809 const struct udphdr *uh;
2810 struct sock *sk = NULL;
2811 struct dst_entry *dst;
2812 int dif = skb->dev->ifindex;
2813 int sdif = inet_sdif(skb);
2814 int ours;
2815
2816 /* validate the packet */
2817 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2818 return 0;
2819
2820 iph = ip_hdr(skb);
2821 uh = udp_hdr(skb);
2822
2823 if (skb->pkt_type == PACKET_MULTICAST) {
2824 in_dev = __in_dev_get_rcu(skb->dev);
2825
2826 if (!in_dev)
2827 return 0;
2828
2829 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2830 iph->protocol);
2831 if (!ours)
2832 return 0;
2833
2834 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2835 uh->source, iph->saddr,
2836 dif, sdif);
2837 } else if (skb->pkt_type == PACKET_HOST) {
2838 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2839 uh->source, iph->saddr, dif, sdif);
2840 }
2841
2842 if (!sk)
2843 return 0;
2844
2845 skb->sk = sk;
2846 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2847 skb->destructor = sock_pfree;
2848 dst = rcu_dereference(sk->sk_rx_dst);
2849
2850 if (dst)
2851 dst = dst_check(dst, 0);
2852 if (dst) {
2853 u32 itag = 0;
2854
2855 /* set noref for now.
2856 * any place which wants to hold dst has to call
2857 * dst_hold_safe()
2858 */
2859 skb_dst_set_noref(skb, dst);
2860
2861 /* for unconnected multicast sockets we need to validate
2862 * the source on each packet
2863 */
2864 if (!inet_sk(sk)->inet_daddr && in_dev)
2865 return ip_mc_validate_source(skb, iph->daddr,
2866 iph->saddr,
2867 ip4h_dscp(iph),
2868 skb->dev, in_dev, &itag);
2869 }
2870 return 0;
2871 }
2872
udp_rcv(struct sk_buff * skb)2873 int udp_rcv(struct sk_buff *skb)
2874 {
2875 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2876 }
2877
udp_destroy_sock(struct sock * sk)2878 void udp_destroy_sock(struct sock *sk)
2879 {
2880 struct udp_sock *up = udp_sk(sk);
2881 bool slow = lock_sock_fast(sk);
2882
2883 /* protects from races with udp_abort() */
2884 sock_set_flag(sk, SOCK_DEAD);
2885 udp_flush_pending_frames(sk);
2886 unlock_sock_fast(sk, slow);
2887 if (static_branch_unlikely(&udp_encap_needed_key)) {
2888 if (up->encap_type) {
2889 void (*encap_destroy)(struct sock *sk);
2890 encap_destroy = READ_ONCE(up->encap_destroy);
2891 if (encap_destroy)
2892 encap_destroy(sk);
2893 }
2894 if (udp_test_bit(ENCAP_ENABLED, sk))
2895 static_branch_dec(&udp_encap_needed_key);
2896 }
2897 }
2898
set_xfrm_gro_udp_encap_rcv(__u16 encap_type,unsigned short family,struct sock * sk)2899 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2900 struct sock *sk)
2901 {
2902 #ifdef CONFIG_XFRM
2903 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2904 if (family == AF_INET)
2905 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2906 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2907 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2908 }
2909 #endif
2910 }
2911
2912 /*
2913 * Socket option code for UDP
2914 */
udp_lib_setsockopt(struct sock * sk,int level,int optname,sockptr_t optval,unsigned int optlen,int (* push_pending_frames)(struct sock *))2915 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2916 sockptr_t optval, unsigned int optlen,
2917 int (*push_pending_frames)(struct sock *))
2918 {
2919 struct udp_sock *up = udp_sk(sk);
2920 int val, valbool;
2921 int err = 0;
2922 int is_udplite = IS_UDPLITE(sk);
2923
2924 if (level == SOL_SOCKET) {
2925 err = sk_setsockopt(sk, level, optname, optval, optlen);
2926
2927 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2928 sockopt_lock_sock(sk);
2929 /* paired with READ_ONCE in udp_rmem_release() */
2930 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2931 sockopt_release_sock(sk);
2932 }
2933 return err;
2934 }
2935
2936 if (optlen < sizeof(int))
2937 return -EINVAL;
2938
2939 if (copy_from_sockptr(&val, optval, sizeof(val)))
2940 return -EFAULT;
2941
2942 valbool = val ? 1 : 0;
2943
2944 switch (optname) {
2945 case UDP_CORK:
2946 if (val != 0) {
2947 udp_set_bit(CORK, sk);
2948 } else {
2949 udp_clear_bit(CORK, sk);
2950 lock_sock(sk);
2951 push_pending_frames(sk);
2952 release_sock(sk);
2953 }
2954 break;
2955
2956 case UDP_ENCAP:
2957 switch (val) {
2958 case 0:
2959 #ifdef CONFIG_XFRM
2960 case UDP_ENCAP_ESPINUDP:
2961 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2962 #if IS_ENABLED(CONFIG_IPV6)
2963 if (sk->sk_family == AF_INET6)
2964 WRITE_ONCE(up->encap_rcv,
2965 ipv6_stub->xfrm6_udp_encap_rcv);
2966 else
2967 #endif
2968 WRITE_ONCE(up->encap_rcv,
2969 xfrm4_udp_encap_rcv);
2970 #endif
2971 fallthrough;
2972 case UDP_ENCAP_L2TPINUDP:
2973 WRITE_ONCE(up->encap_type, val);
2974 udp_tunnel_encap_enable(sk);
2975 break;
2976 default:
2977 err = -ENOPROTOOPT;
2978 break;
2979 }
2980 break;
2981
2982 case UDP_NO_CHECK6_TX:
2983 udp_set_no_check6_tx(sk, valbool);
2984 break;
2985
2986 case UDP_NO_CHECK6_RX:
2987 udp_set_no_check6_rx(sk, valbool);
2988 break;
2989
2990 case UDP_SEGMENT:
2991 if (val < 0 || val > USHRT_MAX)
2992 return -EINVAL;
2993 WRITE_ONCE(up->gso_size, val);
2994 break;
2995
2996 case UDP_GRO:
2997
2998 /* when enabling GRO, accept the related GSO packet type */
2999 if (valbool)
3000 udp_tunnel_encap_enable(sk);
3001 udp_assign_bit(GRO_ENABLED, sk, valbool);
3002 udp_assign_bit(ACCEPT_L4, sk, valbool);
3003 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
3004 break;
3005
3006 /*
3007 * UDP-Lite's partial checksum coverage (RFC 3828).
3008 */
3009 /* The sender sets actual checksum coverage length via this option.
3010 * The case coverage > packet length is handled by send module. */
3011 case UDPLITE_SEND_CSCOV:
3012 if (!is_udplite) /* Disable the option on UDP sockets */
3013 return -ENOPROTOOPT;
3014 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
3015 val = 8;
3016 else if (val > USHRT_MAX)
3017 val = USHRT_MAX;
3018 WRITE_ONCE(up->pcslen, val);
3019 udp_set_bit(UDPLITE_SEND_CC, sk);
3020 break;
3021
3022 /* The receiver specifies a minimum checksum coverage value. To make
3023 * sense, this should be set to at least 8 (as done below). If zero is
3024 * used, this again means full checksum coverage. */
3025 case UDPLITE_RECV_CSCOV:
3026 if (!is_udplite) /* Disable the option on UDP sockets */
3027 return -ENOPROTOOPT;
3028 if (val != 0 && val < 8) /* Avoid silly minimal values. */
3029 val = 8;
3030 else if (val > USHRT_MAX)
3031 val = USHRT_MAX;
3032 WRITE_ONCE(up->pcrlen, val);
3033 udp_set_bit(UDPLITE_RECV_CC, sk);
3034 break;
3035
3036 default:
3037 err = -ENOPROTOOPT;
3038 break;
3039 }
3040
3041 return err;
3042 }
3043 EXPORT_IPV6_MOD(udp_lib_setsockopt);
3044
udp_setsockopt(struct sock * sk,int level,int optname,sockptr_t optval,unsigned int optlen)3045 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
3046 unsigned int optlen)
3047 {
3048 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
3049 return udp_lib_setsockopt(sk, level, optname,
3050 optval, optlen,
3051 udp_push_pending_frames);
3052 return ip_setsockopt(sk, level, optname, optval, optlen);
3053 }
3054
udp_lib_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * optlen)3055 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
3056 char __user *optval, int __user *optlen)
3057 {
3058 struct udp_sock *up = udp_sk(sk);
3059 int val, len;
3060
3061 if (get_user(len, optlen))
3062 return -EFAULT;
3063
3064 if (len < 0)
3065 return -EINVAL;
3066
3067 len = min_t(unsigned int, len, sizeof(int));
3068
3069 switch (optname) {
3070 case UDP_CORK:
3071 val = udp_test_bit(CORK, sk);
3072 break;
3073
3074 case UDP_ENCAP:
3075 val = READ_ONCE(up->encap_type);
3076 break;
3077
3078 case UDP_NO_CHECK6_TX:
3079 val = udp_get_no_check6_tx(sk);
3080 break;
3081
3082 case UDP_NO_CHECK6_RX:
3083 val = udp_get_no_check6_rx(sk);
3084 break;
3085
3086 case UDP_SEGMENT:
3087 val = READ_ONCE(up->gso_size);
3088 break;
3089
3090 case UDP_GRO:
3091 val = udp_test_bit(GRO_ENABLED, sk);
3092 break;
3093
3094 /* The following two cannot be changed on UDP sockets, the return is
3095 * always 0 (which corresponds to the full checksum coverage of UDP). */
3096 case UDPLITE_SEND_CSCOV:
3097 val = READ_ONCE(up->pcslen);
3098 break;
3099
3100 case UDPLITE_RECV_CSCOV:
3101 val = READ_ONCE(up->pcrlen);
3102 break;
3103
3104 default:
3105 return -ENOPROTOOPT;
3106 }
3107
3108 if (put_user(len, optlen))
3109 return -EFAULT;
3110 if (copy_to_user(optval, &val, len))
3111 return -EFAULT;
3112 return 0;
3113 }
3114 EXPORT_IPV6_MOD(udp_lib_getsockopt);
3115
udp_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * optlen)3116 int udp_getsockopt(struct sock *sk, int level, int optname,
3117 char __user *optval, int __user *optlen)
3118 {
3119 if (level == SOL_UDP || level == SOL_UDPLITE)
3120 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
3121 return ip_getsockopt(sk, level, optname, optval, optlen);
3122 }
3123
3124 /**
3125 * udp_poll - wait for a UDP event.
3126 * @file: - file struct
3127 * @sock: - socket
3128 * @wait: - poll table
3129 *
3130 * This is same as datagram poll, except for the special case of
3131 * blocking sockets. If application is using a blocking fd
3132 * and a packet with checksum error is in the queue;
3133 * then it could get return from select indicating data available
3134 * but then block when reading it. Add special case code
3135 * to work around these arguably broken applications.
3136 */
udp_poll(struct file * file,struct socket * sock,poll_table * wait)3137 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
3138 {
3139 __poll_t mask = datagram_poll(file, sock, wait);
3140 struct sock *sk = sock->sk;
3141
3142 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
3143 mask |= EPOLLIN | EPOLLRDNORM;
3144
3145 /* Check for false positives due to checksum errors */
3146 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
3147 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
3148 mask &= ~(EPOLLIN | EPOLLRDNORM);
3149
3150 /* psock ingress_msg queue should not contain any bad checksum frames */
3151 if (sk_is_readable(sk))
3152 mask |= EPOLLIN | EPOLLRDNORM;
3153 return mask;
3154
3155 }
3156 EXPORT_IPV6_MOD(udp_poll);
3157
udp_abort(struct sock * sk,int err)3158 int udp_abort(struct sock *sk, int err)
3159 {
3160 if (!has_current_bpf_ctx())
3161 lock_sock(sk);
3162
3163 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
3164 * with close()
3165 */
3166 if (sock_flag(sk, SOCK_DEAD))
3167 goto out;
3168
3169 sk->sk_err = err;
3170 sk_error_report(sk);
3171 __udp_disconnect(sk, 0);
3172
3173 out:
3174 if (!has_current_bpf_ctx())
3175 release_sock(sk);
3176
3177 return 0;
3178 }
3179 EXPORT_IPV6_MOD_GPL(udp_abort);
3180
3181 struct proto udp_prot = {
3182 .name = "UDP",
3183 .owner = THIS_MODULE,
3184 .close = udp_lib_close,
3185 .pre_connect = udp_pre_connect,
3186 .connect = udp_connect,
3187 .disconnect = udp_disconnect,
3188 .ioctl = udp_ioctl,
3189 .init = udp_init_sock,
3190 .destroy = udp_destroy_sock,
3191 .setsockopt = udp_setsockopt,
3192 .getsockopt = udp_getsockopt,
3193 .sendmsg = udp_sendmsg,
3194 .recvmsg = udp_recvmsg,
3195 .splice_eof = udp_splice_eof,
3196 .release_cb = ip4_datagram_release_cb,
3197 .hash = udp_lib_hash,
3198 .unhash = udp_lib_unhash,
3199 .rehash = udp_v4_rehash,
3200 .get_port = udp_v4_get_port,
3201 .put_port = udp_lib_unhash,
3202 #ifdef CONFIG_BPF_SYSCALL
3203 .psock_update_sk_prot = udp_bpf_update_proto,
3204 #endif
3205 .memory_allocated = &udp_memory_allocated,
3206 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
3207
3208 .sysctl_mem = sysctl_udp_mem,
3209 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
3210 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
3211 .obj_size = sizeof(struct udp_sock),
3212 .h.udp_table = NULL,
3213 .diag_destroy = udp_abort,
3214 };
3215 EXPORT_SYMBOL(udp_prot);
3216
3217 /* ------------------------------------------------------------------------ */
3218 #ifdef CONFIG_PROC_FS
3219
3220 static unsigned short seq_file_family(const struct seq_file *seq);
seq_sk_match(struct seq_file * seq,const struct sock * sk)3221 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
3222 {
3223 unsigned short family = seq_file_family(seq);
3224
3225 /* AF_UNSPEC is used as a match all */
3226 return ((family == AF_UNSPEC || family == sk->sk_family) &&
3227 net_eq(sock_net(sk), seq_file_net(seq)));
3228 }
3229
3230 #ifdef CONFIG_BPF_SYSCALL
3231 static const struct seq_operations bpf_iter_udp_seq_ops;
3232 #endif
udp_get_table_seq(struct seq_file * seq,struct net * net)3233 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
3234 struct net *net)
3235 {
3236 const struct udp_seq_afinfo *afinfo;
3237
3238 #ifdef CONFIG_BPF_SYSCALL
3239 if (seq->op == &bpf_iter_udp_seq_ops)
3240 return net->ipv4.udp_table;
3241 #endif
3242
3243 afinfo = pde_data(file_inode(seq->file));
3244 return afinfo->udp_table ? : net->ipv4.udp_table;
3245 }
3246
udp_get_first(struct seq_file * seq,int start)3247 static struct sock *udp_get_first(struct seq_file *seq, int start)
3248 {
3249 struct udp_iter_state *state = seq->private;
3250 struct net *net = seq_file_net(seq);
3251 struct udp_table *udptable;
3252 struct sock *sk;
3253
3254 udptable = udp_get_table_seq(seq, net);
3255
3256 for (state->bucket = start; state->bucket <= udptable->mask;
3257 ++state->bucket) {
3258 struct udp_hslot *hslot = &udptable->hash[state->bucket];
3259
3260 if (hlist_empty(&hslot->head))
3261 continue;
3262
3263 spin_lock_bh(&hslot->lock);
3264 sk_for_each(sk, &hslot->head) {
3265 if (seq_sk_match(seq, sk))
3266 goto found;
3267 }
3268 spin_unlock_bh(&hslot->lock);
3269 }
3270 sk = NULL;
3271 found:
3272 return sk;
3273 }
3274
udp_get_next(struct seq_file * seq,struct sock * sk)3275 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3276 {
3277 struct udp_iter_state *state = seq->private;
3278 struct net *net = seq_file_net(seq);
3279 struct udp_table *udptable;
3280
3281 do {
3282 sk = sk_next(sk);
3283 } while (sk && !seq_sk_match(seq, sk));
3284
3285 if (!sk) {
3286 udptable = udp_get_table_seq(seq, net);
3287
3288 if (state->bucket <= udptable->mask)
3289 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3290
3291 return udp_get_first(seq, state->bucket + 1);
3292 }
3293 return sk;
3294 }
3295
udp_get_idx(struct seq_file * seq,loff_t pos)3296 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3297 {
3298 struct sock *sk = udp_get_first(seq, 0);
3299
3300 if (sk)
3301 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3302 --pos;
3303 return pos ? NULL : sk;
3304 }
3305
udp_seq_start(struct seq_file * seq,loff_t * pos)3306 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3307 {
3308 struct udp_iter_state *state = seq->private;
3309 state->bucket = MAX_UDP_PORTS;
3310
3311 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3312 }
3313 EXPORT_IPV6_MOD(udp_seq_start);
3314
udp_seq_next(struct seq_file * seq,void * v,loff_t * pos)3315 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3316 {
3317 struct sock *sk;
3318
3319 if (v == SEQ_START_TOKEN)
3320 sk = udp_get_idx(seq, 0);
3321 else
3322 sk = udp_get_next(seq, v);
3323
3324 ++*pos;
3325 return sk;
3326 }
3327 EXPORT_IPV6_MOD(udp_seq_next);
3328
udp_seq_stop(struct seq_file * seq,void * v)3329 void udp_seq_stop(struct seq_file *seq, void *v)
3330 {
3331 struct udp_iter_state *state = seq->private;
3332 struct udp_table *udptable;
3333
3334 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3335
3336 if (state->bucket <= udptable->mask)
3337 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3338 }
3339 EXPORT_IPV6_MOD(udp_seq_stop);
3340
3341 /* ------------------------------------------------------------------------ */
udp4_format_sock(struct sock * sp,struct seq_file * f,int bucket)3342 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3343 int bucket)
3344 {
3345 struct inet_sock *inet = inet_sk(sp);
3346 __be32 dest = inet->inet_daddr;
3347 __be32 src = inet->inet_rcv_saddr;
3348 __u16 destp = ntohs(inet->inet_dport);
3349 __u16 srcp = ntohs(inet->inet_sport);
3350
3351 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3352 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3353 bucket, src, srcp, dest, destp, sp->sk_state,
3354 sk_wmem_alloc_get(sp),
3355 udp_rqueue_get(sp),
3356 0, 0L, 0,
3357 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3358 0, sock_i_ino(sp),
3359 refcount_read(&sp->sk_refcnt), sp,
3360 atomic_read(&sp->sk_drops));
3361 }
3362
udp4_seq_show(struct seq_file * seq,void * v)3363 int udp4_seq_show(struct seq_file *seq, void *v)
3364 {
3365 seq_setwidth(seq, 127);
3366 if (v == SEQ_START_TOKEN)
3367 seq_puts(seq, " sl local_address rem_address st tx_queue "
3368 "rx_queue tr tm->when retrnsmt uid timeout "
3369 "inode ref pointer drops");
3370 else {
3371 struct udp_iter_state *state = seq->private;
3372
3373 udp4_format_sock(v, seq, state->bucket);
3374 }
3375 seq_pad(seq, '\n');
3376 return 0;
3377 }
3378
3379 #ifdef CONFIG_BPF_SYSCALL
3380 struct bpf_iter__udp {
3381 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3382 __bpf_md_ptr(struct udp_sock *, udp_sk);
3383 uid_t uid __aligned(8);
3384 int bucket __aligned(8);
3385 };
3386
3387 struct bpf_udp_iter_state {
3388 struct udp_iter_state state;
3389 unsigned int cur_sk;
3390 unsigned int end_sk;
3391 unsigned int max_sk;
3392 int offset;
3393 struct sock **batch;
3394 bool st_bucket_done;
3395 };
3396
3397 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3398 unsigned int new_batch_sz);
bpf_iter_udp_batch(struct seq_file * seq)3399 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3400 {
3401 struct bpf_udp_iter_state *iter = seq->private;
3402 struct udp_iter_state *state = &iter->state;
3403 struct net *net = seq_file_net(seq);
3404 int resume_bucket, resume_offset;
3405 struct udp_table *udptable;
3406 unsigned int batch_sks = 0;
3407 bool resized = false;
3408 struct sock *sk;
3409
3410 resume_bucket = state->bucket;
3411 resume_offset = iter->offset;
3412
3413 /* The current batch is done, so advance the bucket. */
3414 if (iter->st_bucket_done)
3415 state->bucket++;
3416
3417 udptable = udp_get_table_seq(seq, net);
3418
3419 again:
3420 /* New batch for the next bucket.
3421 * Iterate over the hash table to find a bucket with sockets matching
3422 * the iterator attributes, and return the first matching socket from
3423 * the bucket. The remaining matched sockets from the bucket are batched
3424 * before releasing the bucket lock. This allows BPF programs that are
3425 * called in seq_show to acquire the bucket lock if needed.
3426 */
3427 iter->cur_sk = 0;
3428 iter->end_sk = 0;
3429 iter->st_bucket_done = false;
3430 batch_sks = 0;
3431
3432 for (; state->bucket <= udptable->mask; state->bucket++) {
3433 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket].hslot;
3434
3435 if (hlist_empty(&hslot2->head))
3436 continue;
3437
3438 iter->offset = 0;
3439 spin_lock_bh(&hslot2->lock);
3440 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3441 if (seq_sk_match(seq, sk)) {
3442 /* Resume from the last iterated socket at the
3443 * offset in the bucket before iterator was stopped.
3444 */
3445 if (state->bucket == resume_bucket &&
3446 iter->offset < resume_offset) {
3447 ++iter->offset;
3448 continue;
3449 }
3450 if (iter->end_sk < iter->max_sk) {
3451 sock_hold(sk);
3452 iter->batch[iter->end_sk++] = sk;
3453 }
3454 batch_sks++;
3455 }
3456 }
3457 spin_unlock_bh(&hslot2->lock);
3458
3459 if (iter->end_sk)
3460 break;
3461 }
3462
3463 /* All done: no batch made. */
3464 if (!iter->end_sk)
3465 return NULL;
3466
3467 if (iter->end_sk == batch_sks) {
3468 /* Batching is done for the current bucket; return the first
3469 * socket to be iterated from the batch.
3470 */
3471 iter->st_bucket_done = true;
3472 goto done;
3473 }
3474 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3475 resized = true;
3476 /* After allocating a larger batch, retry one more time to grab
3477 * the whole bucket.
3478 */
3479 goto again;
3480 }
3481 done:
3482 return iter->batch[0];
3483 }
3484
bpf_iter_udp_seq_next(struct seq_file * seq,void * v,loff_t * pos)3485 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3486 {
3487 struct bpf_udp_iter_state *iter = seq->private;
3488 struct sock *sk;
3489
3490 /* Whenever seq_next() is called, the iter->cur_sk is
3491 * done with seq_show(), so unref the iter->cur_sk.
3492 */
3493 if (iter->cur_sk < iter->end_sk) {
3494 sock_put(iter->batch[iter->cur_sk++]);
3495 ++iter->offset;
3496 }
3497
3498 /* After updating iter->cur_sk, check if there are more sockets
3499 * available in the current bucket batch.
3500 */
3501 if (iter->cur_sk < iter->end_sk)
3502 sk = iter->batch[iter->cur_sk];
3503 else
3504 /* Prepare a new batch. */
3505 sk = bpf_iter_udp_batch(seq);
3506
3507 ++*pos;
3508 return sk;
3509 }
3510
bpf_iter_udp_seq_start(struct seq_file * seq,loff_t * pos)3511 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3512 {
3513 /* bpf iter does not support lseek, so it always
3514 * continue from where it was stop()-ped.
3515 */
3516 if (*pos)
3517 return bpf_iter_udp_batch(seq);
3518
3519 return SEQ_START_TOKEN;
3520 }
3521
udp_prog_seq_show(struct bpf_prog * prog,struct bpf_iter_meta * meta,struct udp_sock * udp_sk,uid_t uid,int bucket)3522 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3523 struct udp_sock *udp_sk, uid_t uid, int bucket)
3524 {
3525 struct bpf_iter__udp ctx;
3526
3527 meta->seq_num--; /* skip SEQ_START_TOKEN */
3528 ctx.meta = meta;
3529 ctx.udp_sk = udp_sk;
3530 ctx.uid = uid;
3531 ctx.bucket = bucket;
3532 return bpf_iter_run_prog(prog, &ctx);
3533 }
3534
bpf_iter_udp_seq_show(struct seq_file * seq,void * v)3535 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3536 {
3537 struct udp_iter_state *state = seq->private;
3538 struct bpf_iter_meta meta;
3539 struct bpf_prog *prog;
3540 struct sock *sk = v;
3541 uid_t uid;
3542 int ret;
3543
3544 if (v == SEQ_START_TOKEN)
3545 return 0;
3546
3547 lock_sock(sk);
3548
3549 if (unlikely(sk_unhashed(sk))) {
3550 ret = SEQ_SKIP;
3551 goto unlock;
3552 }
3553
3554 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3555 meta.seq = seq;
3556 prog = bpf_iter_get_info(&meta, false);
3557 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3558
3559 unlock:
3560 release_sock(sk);
3561 return ret;
3562 }
3563
bpf_iter_udp_put_batch(struct bpf_udp_iter_state * iter)3564 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3565 {
3566 while (iter->cur_sk < iter->end_sk)
3567 sock_put(iter->batch[iter->cur_sk++]);
3568 }
3569
bpf_iter_udp_seq_stop(struct seq_file * seq,void * v)3570 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3571 {
3572 struct bpf_udp_iter_state *iter = seq->private;
3573 struct bpf_iter_meta meta;
3574 struct bpf_prog *prog;
3575
3576 if (!v) {
3577 meta.seq = seq;
3578 prog = bpf_iter_get_info(&meta, true);
3579 if (prog)
3580 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3581 }
3582
3583 if (iter->cur_sk < iter->end_sk) {
3584 bpf_iter_udp_put_batch(iter);
3585 iter->st_bucket_done = false;
3586 }
3587 }
3588
3589 static const struct seq_operations bpf_iter_udp_seq_ops = {
3590 .start = bpf_iter_udp_seq_start,
3591 .next = bpf_iter_udp_seq_next,
3592 .stop = bpf_iter_udp_seq_stop,
3593 .show = bpf_iter_udp_seq_show,
3594 };
3595 #endif
3596
seq_file_family(const struct seq_file * seq)3597 static unsigned short seq_file_family(const struct seq_file *seq)
3598 {
3599 const struct udp_seq_afinfo *afinfo;
3600
3601 #ifdef CONFIG_BPF_SYSCALL
3602 /* BPF iterator: bpf programs to filter sockets. */
3603 if (seq->op == &bpf_iter_udp_seq_ops)
3604 return AF_UNSPEC;
3605 #endif
3606
3607 /* Proc fs iterator */
3608 afinfo = pde_data(file_inode(seq->file));
3609 return afinfo->family;
3610 }
3611
3612 const struct seq_operations udp_seq_ops = {
3613 .start = udp_seq_start,
3614 .next = udp_seq_next,
3615 .stop = udp_seq_stop,
3616 .show = udp4_seq_show,
3617 };
3618 EXPORT_IPV6_MOD(udp_seq_ops);
3619
3620 static struct udp_seq_afinfo udp4_seq_afinfo = {
3621 .family = AF_INET,
3622 .udp_table = NULL,
3623 };
3624
udp4_proc_init_net(struct net * net)3625 static int __net_init udp4_proc_init_net(struct net *net)
3626 {
3627 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3628 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3629 return -ENOMEM;
3630 return 0;
3631 }
3632
udp4_proc_exit_net(struct net * net)3633 static void __net_exit udp4_proc_exit_net(struct net *net)
3634 {
3635 remove_proc_entry("udp", net->proc_net);
3636 }
3637
3638 static struct pernet_operations udp4_net_ops = {
3639 .init = udp4_proc_init_net,
3640 .exit = udp4_proc_exit_net,
3641 };
3642
udp4_proc_init(void)3643 int __init udp4_proc_init(void)
3644 {
3645 return register_pernet_subsys(&udp4_net_ops);
3646 }
3647
udp4_proc_exit(void)3648 void udp4_proc_exit(void)
3649 {
3650 unregister_pernet_subsys(&udp4_net_ops);
3651 }
3652 #endif /* CONFIG_PROC_FS */
3653
3654 static __initdata unsigned long uhash_entries;
set_uhash_entries(char * str)3655 static int __init set_uhash_entries(char *str)
3656 {
3657 ssize_t ret;
3658
3659 if (!str)
3660 return 0;
3661
3662 ret = kstrtoul(str, 0, &uhash_entries);
3663 if (ret)
3664 return 0;
3665
3666 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3667 uhash_entries = UDP_HTABLE_SIZE_MIN;
3668 return 1;
3669 }
3670 __setup("uhash_entries=", set_uhash_entries);
3671
udp_table_init(struct udp_table * table,const char * name)3672 void __init udp_table_init(struct udp_table *table, const char *name)
3673 {
3674 unsigned int i, slot_size;
3675
3676 slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3677 udp_hash4_slot_size();
3678 table->hash = alloc_large_system_hash(name,
3679 slot_size,
3680 uhash_entries,
3681 21, /* one slot per 2 MB */
3682 0,
3683 &table->log,
3684 &table->mask,
3685 UDP_HTABLE_SIZE_MIN,
3686 UDP_HTABLE_SIZE_MAX);
3687
3688 table->hash2 = (void *)(table->hash + (table->mask + 1));
3689 for (i = 0; i <= table->mask; i++) {
3690 INIT_HLIST_HEAD(&table->hash[i].head);
3691 table->hash[i].count = 0;
3692 spin_lock_init(&table->hash[i].lock);
3693 }
3694 for (i = 0; i <= table->mask; i++) {
3695 INIT_HLIST_HEAD(&table->hash2[i].hslot.head);
3696 table->hash2[i].hslot.count = 0;
3697 spin_lock_init(&table->hash2[i].hslot.lock);
3698 }
3699 udp_table_hash4_init(table);
3700 }
3701
udp_flow_hashrnd(void)3702 u32 udp_flow_hashrnd(void)
3703 {
3704 static u32 hashrnd __read_mostly;
3705
3706 net_get_random_once(&hashrnd, sizeof(hashrnd));
3707
3708 return hashrnd;
3709 }
3710 EXPORT_SYMBOL(udp_flow_hashrnd);
3711
udp_sysctl_init(struct net * net)3712 static void __net_init udp_sysctl_init(struct net *net)
3713 {
3714 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3715 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3716
3717 #ifdef CONFIG_NET_L3_MASTER_DEV
3718 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3719 #endif
3720 }
3721
udp_pernet_table_alloc(unsigned int hash_entries)3722 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3723 {
3724 struct udp_table *udptable;
3725 unsigned int slot_size;
3726 int i;
3727
3728 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3729 if (!udptable)
3730 goto out;
3731
3732 slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3733 udp_hash4_slot_size();
3734 udptable->hash = vmalloc_huge(hash_entries * slot_size,
3735 GFP_KERNEL_ACCOUNT);
3736 if (!udptable->hash)
3737 goto free_table;
3738
3739 udptable->hash2 = (void *)(udptable->hash + hash_entries);
3740 udptable->mask = hash_entries - 1;
3741 udptable->log = ilog2(hash_entries);
3742
3743 for (i = 0; i < hash_entries; i++) {
3744 INIT_HLIST_HEAD(&udptable->hash[i].head);
3745 udptable->hash[i].count = 0;
3746 spin_lock_init(&udptable->hash[i].lock);
3747
3748 INIT_HLIST_HEAD(&udptable->hash2[i].hslot.head);
3749 udptable->hash2[i].hslot.count = 0;
3750 spin_lock_init(&udptable->hash2[i].hslot.lock);
3751 }
3752 udp_table_hash4_init(udptable);
3753
3754 return udptable;
3755
3756 free_table:
3757 kfree(udptable);
3758 out:
3759 return NULL;
3760 }
3761
udp_pernet_table_free(struct net * net)3762 static void __net_exit udp_pernet_table_free(struct net *net)
3763 {
3764 struct udp_table *udptable = net->ipv4.udp_table;
3765
3766 if (udptable == &udp_table)
3767 return;
3768
3769 kvfree(udptable->hash);
3770 kfree(udptable);
3771 }
3772
udp_set_table(struct net * net)3773 static void __net_init udp_set_table(struct net *net)
3774 {
3775 struct udp_table *udptable;
3776 unsigned int hash_entries;
3777 struct net *old_net;
3778
3779 if (net_eq(net, &init_net))
3780 goto fallback;
3781
3782 old_net = current->nsproxy->net_ns;
3783 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3784 if (!hash_entries)
3785 goto fallback;
3786
3787 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3788 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3789 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3790 else
3791 hash_entries = roundup_pow_of_two(hash_entries);
3792
3793 udptable = udp_pernet_table_alloc(hash_entries);
3794 if (udptable) {
3795 net->ipv4.udp_table = udptable;
3796 } else {
3797 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3798 "for a netns, fallback to the global one\n",
3799 hash_entries);
3800 fallback:
3801 net->ipv4.udp_table = &udp_table;
3802 }
3803 }
3804
udp_pernet_init(struct net * net)3805 static int __net_init udp_pernet_init(struct net *net)
3806 {
3807 udp_sysctl_init(net);
3808 udp_set_table(net);
3809
3810 return 0;
3811 }
3812
udp_pernet_exit(struct net * net)3813 static void __net_exit udp_pernet_exit(struct net *net)
3814 {
3815 udp_pernet_table_free(net);
3816 }
3817
3818 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3819 .init = udp_pernet_init,
3820 .exit = udp_pernet_exit,
3821 };
3822
3823 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
DEFINE_BPF_ITER_FUNC(udp,struct bpf_iter_meta * meta,struct udp_sock * udp_sk,uid_t uid,int bucket)3824 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3825 struct udp_sock *udp_sk, uid_t uid, int bucket)
3826
3827 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3828 unsigned int new_batch_sz)
3829 {
3830 struct sock **new_batch;
3831
3832 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3833 GFP_USER | __GFP_NOWARN);
3834 if (!new_batch)
3835 return -ENOMEM;
3836
3837 bpf_iter_udp_put_batch(iter);
3838 kvfree(iter->batch);
3839 iter->batch = new_batch;
3840 iter->max_sk = new_batch_sz;
3841
3842 return 0;
3843 }
3844
3845 #define INIT_BATCH_SZ 16
3846
bpf_iter_init_udp(void * priv_data,struct bpf_iter_aux_info * aux)3847 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3848 {
3849 struct bpf_udp_iter_state *iter = priv_data;
3850 int ret;
3851
3852 ret = bpf_iter_init_seq_net(priv_data, aux);
3853 if (ret)
3854 return ret;
3855
3856 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3857 if (ret)
3858 bpf_iter_fini_seq_net(priv_data);
3859
3860 return ret;
3861 }
3862
bpf_iter_fini_udp(void * priv_data)3863 static void bpf_iter_fini_udp(void *priv_data)
3864 {
3865 struct bpf_udp_iter_state *iter = priv_data;
3866
3867 bpf_iter_fini_seq_net(priv_data);
3868 kvfree(iter->batch);
3869 }
3870
3871 static const struct bpf_iter_seq_info udp_seq_info = {
3872 .seq_ops = &bpf_iter_udp_seq_ops,
3873 .init_seq_private = bpf_iter_init_udp,
3874 .fini_seq_private = bpf_iter_fini_udp,
3875 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3876 };
3877
3878 static struct bpf_iter_reg udp_reg_info = {
3879 .target = "udp",
3880 .ctx_arg_info_size = 1,
3881 .ctx_arg_info = {
3882 { offsetof(struct bpf_iter__udp, udp_sk),
3883 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3884 },
3885 .seq_info = &udp_seq_info,
3886 };
3887
bpf_iter_register(void)3888 static void __init bpf_iter_register(void)
3889 {
3890 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3891 if (bpf_iter_reg_target(&udp_reg_info))
3892 pr_warn("Warning: could not register bpf iterator udp\n");
3893 }
3894 #endif
3895
udp_init(void)3896 void __init udp_init(void)
3897 {
3898 unsigned long limit;
3899 unsigned int i;
3900
3901 udp_table_init(&udp_table, "UDP");
3902 limit = nr_free_buffer_pages() / 8;
3903 limit = max(limit, 128UL);
3904 sysctl_udp_mem[0] = limit / 4 * 3;
3905 sysctl_udp_mem[1] = limit;
3906 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3907
3908 /* 16 spinlocks per cpu */
3909 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3910 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3911 GFP_KERNEL);
3912 if (!udp_busylocks)
3913 panic("UDP: failed to alloc udp_busylocks\n");
3914 for (i = 0; i < (1U << udp_busylocks_log); i++)
3915 spin_lock_init(udp_busylocks + i);
3916
3917 if (register_pernet_subsys(&udp_sysctl_ops))
3918 panic("UDP: failed to init sysctl parameters.\n");
3919
3920 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3921 bpf_iter_register();
3922 #endif
3923 }
3924