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 Encapsulation 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 <linux/sock_diag.h>
97 #include <net/tcp_states.h>
98 #include <linux/skbuff.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <net/net_namespace.h>
102 #include <net/icmp.h>
103 #include <net/inet_hashtables.h>
104 #include <net/ip.h>
105 #include <net/ip_tunnels.h>
106 #include <net/route.h>
107 #include <net/checksum.h>
108 #include <net/gso.h>
109 #include <net/xfrm.h>
110 #include <trace/events/udp.h>
111 #include <linux/static_key.h>
112 #include <linux/btf_ids.h>
113 #include <trace/events/skb.h>
114 #include <net/busy_poll.h>
115 #include "udp_impl.h"
116 #include <net/sock_reuseport.h>
117 #include <net/addrconf.h>
118 #include <net/udp_tunnel.h>
119 #include <net/gro.h>
120 #if IS_ENABLED(CONFIG_IPV6)
121 #include <net/ipv6_stubs.h>
122 #endif
123 #include <net/rps.h>
124
125 struct udp_table udp_table __read_mostly;
126
127 long sysctl_udp_mem[3] __read_mostly;
128 EXPORT_IPV6_MOD(sysctl_udp_mem);
129
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 kuid_t uid = sk_uid(sk);
147 struct sock *sk2;
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, sk_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 kuid_t uid = sk_uid(sk);
182 struct sock *sk2;
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, sk_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 = sk_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, sk_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 * measurable 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,
1447 sk_uid(sk));
1448
1449 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1450 rt = ip_route_output_flow(net, fl4, sk);
1451 if (IS_ERR(rt)) {
1452 err = PTR_ERR(rt);
1453 rt = NULL;
1454 if (err == -ENETUNREACH)
1455 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1456 goto out;
1457 }
1458
1459 err = -EACCES;
1460 if ((rt->rt_flags & RTCF_BROADCAST) &&
1461 !sock_flag(sk, SOCK_BROADCAST))
1462 goto out;
1463 if (connected)
1464 sk_dst_set(sk, dst_clone(&rt->dst));
1465 }
1466
1467 if (msg->msg_flags&MSG_CONFIRM)
1468 goto do_confirm;
1469 back_from_confirm:
1470
1471 saddr = fl4->saddr;
1472 if (!ipc.addr)
1473 daddr = ipc.addr = fl4->daddr;
1474
1475 /* Lockless fast path for the non-corking case. */
1476 if (!corkreq) {
1477 struct inet_cork cork;
1478
1479 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1480 sizeof(struct udphdr), &ipc, &rt,
1481 &cork, msg->msg_flags);
1482 err = PTR_ERR(skb);
1483 if (!IS_ERR_OR_NULL(skb))
1484 err = udp_send_skb(skb, fl4, &cork);
1485 goto out;
1486 }
1487
1488 lock_sock(sk);
1489 if (unlikely(up->pending)) {
1490 /* The socket is already corked while preparing it. */
1491 /* ... which is an evident application bug. --ANK */
1492 release_sock(sk);
1493
1494 net_dbg_ratelimited("socket already corked\n");
1495 err = -EINVAL;
1496 goto out;
1497 }
1498 /*
1499 * Now cork the socket to pend data.
1500 */
1501 fl4 = &inet->cork.fl.u.ip4;
1502 fl4->daddr = daddr;
1503 fl4->saddr = saddr;
1504 fl4->fl4_dport = dport;
1505 fl4->fl4_sport = inet->inet_sport;
1506 WRITE_ONCE(up->pending, AF_INET);
1507
1508 do_append_data:
1509 up->len += ulen;
1510 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1511 sizeof(struct udphdr), &ipc, &rt,
1512 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1513 if (err)
1514 udp_flush_pending_frames(sk);
1515 else if (!corkreq)
1516 err = udp_push_pending_frames(sk);
1517 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1518 WRITE_ONCE(up->pending, 0);
1519 release_sock(sk);
1520
1521 out:
1522 ip_rt_put(rt);
1523 out_free:
1524 if (free)
1525 kfree(ipc.opt);
1526 if (!err)
1527 return len;
1528 /*
1529 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1530 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1531 * we don't have a good statistic (IpOutDiscards but it can be too many
1532 * things). We could add another new stat but at least for now that
1533 * seems like overkill.
1534 */
1535 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1536 UDP_INC_STATS(sock_net(sk),
1537 UDP_MIB_SNDBUFERRORS, is_udplite);
1538 }
1539 return err;
1540
1541 do_confirm:
1542 if (msg->msg_flags & MSG_PROBE)
1543 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1544 if (!(msg->msg_flags&MSG_PROBE) || len)
1545 goto back_from_confirm;
1546 err = 0;
1547 goto out;
1548 }
1549 EXPORT_SYMBOL(udp_sendmsg);
1550
udp_splice_eof(struct socket * sock)1551 void udp_splice_eof(struct socket *sock)
1552 {
1553 struct sock *sk = sock->sk;
1554 struct udp_sock *up = udp_sk(sk);
1555
1556 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1557 return;
1558
1559 lock_sock(sk);
1560 if (up->pending && !udp_test_bit(CORK, sk))
1561 udp_push_pending_frames(sk);
1562 release_sock(sk);
1563 }
1564 EXPORT_IPV6_MOD_GPL(udp_splice_eof);
1565
1566 #define UDP_SKB_IS_STATELESS 0x80000000
1567
1568 /* all head states (dst, sk, nf conntrack) except skb extensions are
1569 * cleared by udp_rcv().
1570 *
1571 * We need to preserve secpath, if present, to eventually process
1572 * IP_CMSG_PASSSEC at recvmsg() time.
1573 *
1574 * Other extensions can be cleared.
1575 */
udp_try_make_stateless(struct sk_buff * skb)1576 static bool udp_try_make_stateless(struct sk_buff *skb)
1577 {
1578 if (!skb_has_extensions(skb))
1579 return true;
1580
1581 if (!secpath_exists(skb)) {
1582 skb_ext_reset(skb);
1583 return true;
1584 }
1585
1586 return false;
1587 }
1588
udp_set_dev_scratch(struct sk_buff * skb)1589 static void udp_set_dev_scratch(struct sk_buff *skb)
1590 {
1591 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1592
1593 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1594 scratch->_tsize_state = skb->truesize;
1595 #if BITS_PER_LONG == 64
1596 scratch->len = skb->len;
1597 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1598 scratch->is_linear = !skb_is_nonlinear(skb);
1599 #endif
1600 if (udp_try_make_stateless(skb))
1601 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1602 }
1603
udp_skb_csum_unnecessary_set(struct sk_buff * skb)1604 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1605 {
1606 /* We come here after udp_lib_checksum_complete() returned 0.
1607 * This means that __skb_checksum_complete() might have
1608 * set skb->csum_valid to 1.
1609 * On 64bit platforms, we can set csum_unnecessary
1610 * to true, but only if the skb is not shared.
1611 */
1612 #if BITS_PER_LONG == 64
1613 if (!skb_shared(skb))
1614 udp_skb_scratch(skb)->csum_unnecessary = true;
1615 #endif
1616 }
1617
udp_skb_truesize(struct sk_buff * skb)1618 static int udp_skb_truesize(struct sk_buff *skb)
1619 {
1620 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1621 }
1622
udp_skb_has_head_state(struct sk_buff * skb)1623 static bool udp_skb_has_head_state(struct sk_buff *skb)
1624 {
1625 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1626 }
1627
1628 /* fully reclaim rmem/fwd memory allocated for skb */
udp_rmem_release(struct sock * sk,unsigned int size,int partial,bool rx_queue_lock_held)1629 static void udp_rmem_release(struct sock *sk, unsigned int size,
1630 int partial, bool rx_queue_lock_held)
1631 {
1632 struct udp_sock *up = udp_sk(sk);
1633 struct sk_buff_head *sk_queue;
1634 unsigned int amt;
1635
1636 if (likely(partial)) {
1637 up->forward_deficit += size;
1638 size = up->forward_deficit;
1639 if (size < READ_ONCE(up->forward_threshold) &&
1640 !skb_queue_empty(&up->reader_queue))
1641 return;
1642 } else {
1643 size += up->forward_deficit;
1644 }
1645 up->forward_deficit = 0;
1646
1647 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1648 * if the called don't held it already
1649 */
1650 sk_queue = &sk->sk_receive_queue;
1651 if (!rx_queue_lock_held)
1652 spin_lock(&sk_queue->lock);
1653
1654 amt = (size + sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1655 sk_forward_alloc_add(sk, size - amt);
1656
1657 if (amt)
1658 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1659
1660 atomic_sub(size, &sk->sk_rmem_alloc);
1661
1662 /* this can save us from acquiring the rx queue lock on next receive */
1663 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1664
1665 if (!rx_queue_lock_held)
1666 spin_unlock(&sk_queue->lock);
1667 }
1668
1669 /* Note: called with reader_queue.lock held.
1670 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1671 * This avoids a cache line miss while receive_queue lock is held.
1672 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1673 */
udp_skb_destructor(struct sock * sk,struct sk_buff * skb)1674 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1675 {
1676 prefetch(&skb->data);
1677 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1678 }
1679 EXPORT_IPV6_MOD(udp_skb_destructor);
1680
1681 /* as above, but the caller held the rx queue lock, too */
udp_skb_dtor_locked(struct sock * sk,struct sk_buff * skb)1682 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1683 {
1684 prefetch(&skb->data);
1685 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1686 }
1687
udp_rmem_schedule(struct sock * sk,int size)1688 static int udp_rmem_schedule(struct sock *sk, int size)
1689 {
1690 int delta;
1691
1692 delta = size - sk->sk_forward_alloc;
1693 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1694 return -ENOBUFS;
1695
1696 return 0;
1697 }
1698
__udp_enqueue_schedule_skb(struct sock * sk,struct sk_buff * skb)1699 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1700 {
1701 struct sk_buff_head *list = &sk->sk_receive_queue;
1702 struct udp_prod_queue *udp_prod_queue;
1703 struct sk_buff *next, *to_drop = NULL;
1704 struct llist_node *ll_list;
1705 unsigned int rmem, rcvbuf;
1706 int size, err = -ENOMEM;
1707 int total_size = 0;
1708 int q_size = 0;
1709 int dropcount;
1710 int nb = 0;
1711
1712 rmem = atomic_read(&sk->sk_rmem_alloc);
1713 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1714 size = skb->truesize;
1715
1716 udp_prod_queue = &udp_sk(sk)->udp_prod_queue[numa_node_id()];
1717
1718 rmem += atomic_read(&udp_prod_queue->rmem_alloc);
1719
1720 /* Immediately drop when the receive queue is full.
1721 * Cast to unsigned int performs the boundary check for INT_MAX.
1722 */
1723 if (rmem + size > rcvbuf) {
1724 if (rcvbuf > INT_MAX >> 1)
1725 goto drop;
1726
1727 /* Accept the packet if queue is empty. */
1728 if (rmem)
1729 goto drop;
1730 }
1731
1732 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1733 * having linear skbs :
1734 * - Reduce memory overhead and thus increase receive queue capacity
1735 * - Less cache line misses at copyout() time
1736 * - Less work at consume_skb() (less alien page frag freeing)
1737 */
1738 if (rmem > (rcvbuf >> 1)) {
1739 skb_condense(skb);
1740 size = skb->truesize;
1741 }
1742
1743 udp_set_dev_scratch(skb);
1744
1745 atomic_add(size, &udp_prod_queue->rmem_alloc);
1746
1747 if (!llist_add(&skb->ll_node, &udp_prod_queue->ll_root))
1748 return 0;
1749
1750 dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? sk_drops_read(sk) : 0;
1751
1752 spin_lock(&list->lock);
1753
1754 ll_list = llist_del_all(&udp_prod_queue->ll_root);
1755
1756 ll_list = llist_reverse_order(ll_list);
1757
1758 llist_for_each_entry_safe(skb, next, ll_list, ll_node) {
1759 size = udp_skb_truesize(skb);
1760 total_size += size;
1761 err = udp_rmem_schedule(sk, size);
1762 if (unlikely(err)) {
1763 /* Free the skbs outside of locked section. */
1764 skb->next = to_drop;
1765 to_drop = skb;
1766 continue;
1767 }
1768
1769 q_size += size;
1770 sk_forward_alloc_add(sk, -size);
1771
1772 /* no need to setup a destructor, we will explicitly release the
1773 * forward allocated memory on dequeue
1774 */
1775 SOCK_SKB_CB(skb)->dropcount = dropcount;
1776 nb++;
1777 __skb_queue_tail(list, skb);
1778 }
1779
1780 atomic_add(q_size, &sk->sk_rmem_alloc);
1781
1782 spin_unlock(&list->lock);
1783
1784 if (!sock_flag(sk, SOCK_DEAD)) {
1785 /* Multiple threads might be blocked in recvmsg(),
1786 * using prepare_to_wait_exclusive().
1787 */
1788 while (nb) {
1789 INDIRECT_CALL_1(sk->sk_data_ready,
1790 sock_def_readable, sk);
1791 nb--;
1792 }
1793 }
1794
1795 if (unlikely(to_drop)) {
1796 for (nb = 0; to_drop != NULL; nb++) {
1797 skb = to_drop;
1798 to_drop = skb->next;
1799 skb_mark_not_on_list(skb);
1800 /* TODO: update SNMP values. */
1801 sk_skb_reason_drop(sk, skb, SKB_DROP_REASON_PROTO_MEM);
1802 }
1803 numa_drop_add(&udp_sk(sk)->drop_counters, nb);
1804 }
1805
1806 atomic_sub(total_size, &udp_prod_queue->rmem_alloc);
1807
1808 return 0;
1809
1810 drop:
1811 udp_drops_inc(sk);
1812 return err;
1813 }
1814 EXPORT_IPV6_MOD_GPL(__udp_enqueue_schedule_skb);
1815
udp_destruct_common(struct sock * sk)1816 void udp_destruct_common(struct sock *sk)
1817 {
1818 /* reclaim completely the forward allocated memory */
1819 struct udp_sock *up = udp_sk(sk);
1820 unsigned int total = 0;
1821 struct sk_buff *skb;
1822
1823 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1824 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1825 total += skb->truesize;
1826 kfree_skb(skb);
1827 }
1828 udp_rmem_release(sk, total, 0, true);
1829 kfree(up->udp_prod_queue);
1830 }
1831 EXPORT_IPV6_MOD_GPL(udp_destruct_common);
1832
udp_destruct_sock(struct sock * sk)1833 static void udp_destruct_sock(struct sock *sk)
1834 {
1835 udp_destruct_common(sk);
1836 inet_sock_destruct(sk);
1837 }
1838
udp_init_sock(struct sock * sk)1839 int udp_init_sock(struct sock *sk)
1840 {
1841 int res = udp_lib_init_sock(sk);
1842
1843 sk->sk_destruct = udp_destruct_sock;
1844 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1845 return res;
1846 }
1847
skb_consume_udp(struct sock * sk,struct sk_buff * skb,int len)1848 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1849 {
1850 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
1851 sk_peek_offset_bwd(sk, len);
1852
1853 if (!skb_shared(skb)) {
1854 skb_attempt_defer_free(skb);
1855 return;
1856 }
1857
1858 if (!skb_unref(skb))
1859 return;
1860
1861 /* In the more common cases we cleared the head states previously,
1862 * see __udp_queue_rcv_skb().
1863 */
1864 if (unlikely(udp_skb_has_head_state(skb)))
1865 skb_release_head_state(skb);
1866 __consume_stateless_skb(skb);
1867 }
1868 EXPORT_IPV6_MOD_GPL(skb_consume_udp);
1869
__first_packet_length(struct sock * sk,struct sk_buff_head * rcvq,unsigned int * total)1870 static struct sk_buff *__first_packet_length(struct sock *sk,
1871 struct sk_buff_head *rcvq,
1872 unsigned int *total)
1873 {
1874 struct sk_buff *skb;
1875
1876 while ((skb = skb_peek(rcvq)) != NULL) {
1877 if (udp_lib_checksum_complete(skb)) {
1878 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1879 IS_UDPLITE(sk));
1880 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1881 IS_UDPLITE(sk));
1882 udp_drops_inc(sk);
1883 __skb_unlink(skb, rcvq);
1884 *total += skb->truesize;
1885 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM);
1886 } else {
1887 udp_skb_csum_unnecessary_set(skb);
1888 break;
1889 }
1890 }
1891 return skb;
1892 }
1893
1894 /**
1895 * first_packet_length - return length of first packet in receive queue
1896 * @sk: socket
1897 *
1898 * Drops all bad checksum frames, until a valid one is found.
1899 * Returns the length of found skb, or -1 if none is found.
1900 */
first_packet_length(struct sock * sk)1901 static int first_packet_length(struct sock *sk)
1902 {
1903 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1904 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1905 unsigned int total = 0;
1906 struct sk_buff *skb;
1907 int res;
1908
1909 spin_lock_bh(&rcvq->lock);
1910 skb = __first_packet_length(sk, rcvq, &total);
1911 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1912 spin_lock(&sk_queue->lock);
1913 skb_queue_splice_tail_init(sk_queue, rcvq);
1914 spin_unlock(&sk_queue->lock);
1915
1916 skb = __first_packet_length(sk, rcvq, &total);
1917 }
1918 res = skb ? skb->len : -1;
1919 if (total)
1920 udp_rmem_release(sk, total, 1, false);
1921 spin_unlock_bh(&rcvq->lock);
1922 return res;
1923 }
1924
1925 /*
1926 * IOCTL requests applicable to the UDP protocol
1927 */
1928
udp_ioctl(struct sock * sk,int cmd,int * karg)1929 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1930 {
1931 switch (cmd) {
1932 case SIOCOUTQ:
1933 {
1934 *karg = sk_wmem_alloc_get(sk);
1935 return 0;
1936 }
1937
1938 case SIOCINQ:
1939 {
1940 *karg = max_t(int, 0, first_packet_length(sk));
1941 return 0;
1942 }
1943
1944 default:
1945 return -ENOIOCTLCMD;
1946 }
1947
1948 return 0;
1949 }
1950 EXPORT_IPV6_MOD(udp_ioctl);
1951
__skb_recv_udp(struct sock * sk,unsigned int flags,int * off,int * err)1952 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1953 int *off, int *err)
1954 {
1955 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1956 struct sk_buff_head *queue;
1957 struct sk_buff *last;
1958 long timeo;
1959 int error;
1960
1961 queue = &udp_sk(sk)->reader_queue;
1962 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1963 do {
1964 struct sk_buff *skb;
1965
1966 error = sock_error(sk);
1967 if (error)
1968 break;
1969
1970 error = -EAGAIN;
1971 do {
1972 spin_lock_bh(&queue->lock);
1973 skb = __skb_try_recv_from_queue(queue, flags, off, err,
1974 &last);
1975 if (skb) {
1976 if (!(flags & MSG_PEEK))
1977 udp_skb_destructor(sk, skb);
1978 spin_unlock_bh(&queue->lock);
1979 return skb;
1980 }
1981
1982 if (skb_queue_empty_lockless(sk_queue)) {
1983 spin_unlock_bh(&queue->lock);
1984 goto busy_check;
1985 }
1986
1987 /* refill the reader queue and walk it again
1988 * keep both queues locked to avoid re-acquiring
1989 * the sk_receive_queue lock if fwd memory scheduling
1990 * is needed.
1991 */
1992 spin_lock(&sk_queue->lock);
1993 skb_queue_splice_tail_init(sk_queue, queue);
1994
1995 skb = __skb_try_recv_from_queue(queue, flags, off, err,
1996 &last);
1997 if (skb && !(flags & MSG_PEEK))
1998 udp_skb_dtor_locked(sk, skb);
1999 spin_unlock(&sk_queue->lock);
2000 spin_unlock_bh(&queue->lock);
2001 if (skb)
2002 return skb;
2003
2004 busy_check:
2005 if (!sk_can_busy_loop(sk))
2006 break;
2007
2008 sk_busy_loop(sk, flags & MSG_DONTWAIT);
2009 } while (!skb_queue_empty_lockless(sk_queue));
2010
2011 /* sk_queue is empty, reader_queue may contain peeked packets */
2012 } while (timeo &&
2013 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2014 &error, &timeo,
2015 (struct sk_buff *)sk_queue));
2016
2017 *err = error;
2018 return NULL;
2019 }
2020 EXPORT_SYMBOL(__skb_recv_udp);
2021
udp_read_skb(struct sock * sk,skb_read_actor_t recv_actor)2022 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2023 {
2024 struct sk_buff *skb;
2025 int err;
2026
2027 try_again:
2028 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
2029 if (!skb)
2030 return err;
2031
2032 if (udp_lib_checksum_complete(skb)) {
2033 int is_udplite = IS_UDPLITE(sk);
2034 struct net *net = sock_net(sk);
2035
2036 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
2037 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
2038 udp_drops_inc(sk);
2039 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM);
2040 goto try_again;
2041 }
2042
2043 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
2044 return recv_actor(sk, skb);
2045 }
2046 EXPORT_IPV6_MOD(udp_read_skb);
2047
2048 /*
2049 * This should be easy, if there is something there we
2050 * return it, otherwise we block.
2051 */
2052
udp_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int flags,int * addr_len)2053 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
2054 int *addr_len)
2055 {
2056 struct inet_sock *inet = inet_sk(sk);
2057 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
2058 struct sk_buff *skb;
2059 unsigned int ulen, copied;
2060 int off, err, peeking = flags & MSG_PEEK;
2061 int is_udplite = IS_UDPLITE(sk);
2062 bool checksum_valid = false;
2063
2064 if (flags & MSG_ERRQUEUE)
2065 return ip_recv_error(sk, msg, len, addr_len);
2066
2067 try_again:
2068 off = sk_peek_offset(sk, flags);
2069 skb = __skb_recv_udp(sk, flags, &off, &err);
2070 if (!skb)
2071 return err;
2072
2073 ulen = udp_skb_len(skb);
2074 copied = len;
2075 if (copied > ulen - off)
2076 copied = ulen - off;
2077 else if (copied < ulen)
2078 msg->msg_flags |= MSG_TRUNC;
2079
2080 /*
2081 * If checksum is needed at all, try to do it while copying the
2082 * data. If the data is truncated, or if we only want a partial
2083 * coverage checksum (UDP-Lite), do it before the copy.
2084 */
2085
2086 if (copied < ulen || peeking ||
2087 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
2088 checksum_valid = udp_skb_csum_unnecessary(skb) ||
2089 !__udp_lib_checksum_complete(skb);
2090 if (!checksum_valid)
2091 goto csum_copy_err;
2092 }
2093
2094 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
2095 if (udp_skb_is_linear(skb))
2096 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
2097 else
2098 err = skb_copy_datagram_msg(skb, off, msg, copied);
2099 } else {
2100 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
2101
2102 if (err == -EINVAL)
2103 goto csum_copy_err;
2104 }
2105
2106 if (unlikely(err)) {
2107 if (!peeking) {
2108 udp_drops_inc(sk);
2109 UDP_INC_STATS(sock_net(sk),
2110 UDP_MIB_INERRORS, is_udplite);
2111 }
2112 kfree_skb(skb);
2113 return err;
2114 }
2115
2116 if (!peeking)
2117 UDP_INC_STATS(sock_net(sk),
2118 UDP_MIB_INDATAGRAMS, is_udplite);
2119
2120 sock_recv_cmsgs(msg, sk, skb);
2121
2122 /* Copy the address. */
2123 if (sin) {
2124 sin->sin_family = AF_INET;
2125 sin->sin_port = udp_hdr(skb)->source;
2126 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
2127 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
2128 *addr_len = sizeof(*sin);
2129
2130 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
2131 (struct sockaddr *)sin,
2132 addr_len);
2133 }
2134
2135 if (udp_test_bit(GRO_ENABLED, sk))
2136 udp_cmsg_recv(msg, sk, skb);
2137
2138 if (inet_cmsg_flags(inet))
2139 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
2140
2141 err = copied;
2142 if (flags & MSG_TRUNC)
2143 err = ulen;
2144
2145 skb_consume_udp(sk, skb, peeking ? -err : err);
2146 return err;
2147
2148 csum_copy_err:
2149 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
2150 udp_skb_destructor)) {
2151 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2152 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2153 }
2154 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM);
2155
2156 /* starting over for a new packet, but check if we need to yield */
2157 cond_resched();
2158 msg->msg_flags &= ~MSG_TRUNC;
2159 goto try_again;
2160 }
2161
udp_pre_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)2162 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
2163 {
2164 /* This check is replicated from __ip4_datagram_connect() and
2165 * intended to prevent BPF program called below from accessing bytes
2166 * that are out of the bound specified by user in addr_len.
2167 */
2168 if (addr_len < sizeof(struct sockaddr_in))
2169 return -EINVAL;
2170
2171 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
2172 }
2173 EXPORT_IPV6_MOD(udp_pre_connect);
2174
udp_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)2175 static int udp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
2176 {
2177 int res;
2178
2179 lock_sock(sk);
2180 res = __ip4_datagram_connect(sk, uaddr, addr_len);
2181 if (!res)
2182 udp4_hash4(sk);
2183 release_sock(sk);
2184 return res;
2185 }
2186
__udp_disconnect(struct sock * sk,int flags)2187 int __udp_disconnect(struct sock *sk, int flags)
2188 {
2189 struct inet_sock *inet = inet_sk(sk);
2190 /*
2191 * 1003.1g - break association.
2192 */
2193
2194 sk->sk_state = TCP_CLOSE;
2195 inet->inet_daddr = 0;
2196 inet->inet_dport = 0;
2197 sock_rps_reset_rxhash(sk);
2198 sk->sk_bound_dev_if = 0;
2199 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
2200 inet_reset_saddr(sk);
2201 if (sk->sk_prot->rehash &&
2202 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2203 sk->sk_prot->rehash(sk);
2204 }
2205
2206 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
2207 sk->sk_prot->unhash(sk);
2208 inet->inet_sport = 0;
2209 }
2210 sk_dst_reset(sk);
2211 return 0;
2212 }
2213 EXPORT_SYMBOL(__udp_disconnect);
2214
udp_disconnect(struct sock * sk,int flags)2215 int udp_disconnect(struct sock *sk, int flags)
2216 {
2217 lock_sock(sk);
2218 __udp_disconnect(sk, flags);
2219 release_sock(sk);
2220 return 0;
2221 }
2222 EXPORT_IPV6_MOD(udp_disconnect);
2223
udp_lib_unhash(struct sock * sk)2224 void udp_lib_unhash(struct sock *sk)
2225 {
2226 if (sk_hashed(sk)) {
2227 struct udp_table *udptable = udp_get_table_prot(sk);
2228 struct udp_hslot *hslot, *hslot2;
2229
2230 sock_rps_delete_flow(sk);
2231 hslot = udp_hashslot(udptable, sock_net(sk),
2232 udp_sk(sk)->udp_port_hash);
2233 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2234
2235 spin_lock_bh(&hslot->lock);
2236 if (rcu_access_pointer(sk->sk_reuseport_cb))
2237 reuseport_detach_sock(sk);
2238 if (sk_del_node_init_rcu(sk)) {
2239 hslot->count--;
2240 inet_sk(sk)->inet_num = 0;
2241 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2242
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 udp_unhash4(udptable, sk);
2249 }
2250 spin_unlock_bh(&hslot->lock);
2251 }
2252 }
2253 EXPORT_IPV6_MOD(udp_lib_unhash);
2254
2255 /*
2256 * inet_rcv_saddr was changed, we must rehash secondary hash
2257 */
udp_lib_rehash(struct sock * sk,u16 newhash,u16 newhash4)2258 void udp_lib_rehash(struct sock *sk, u16 newhash, u16 newhash4)
2259 {
2260 if (sk_hashed(sk)) {
2261 struct udp_table *udptable = udp_get_table_prot(sk);
2262 struct udp_hslot *hslot, *hslot2, *nhslot2;
2263
2264 hslot = udp_hashslot(udptable, sock_net(sk),
2265 udp_sk(sk)->udp_port_hash);
2266 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2267 nhslot2 = udp_hashslot2(udptable, newhash);
2268 udp_sk(sk)->udp_portaddr_hash = newhash;
2269
2270 if (hslot2 != nhslot2 ||
2271 rcu_access_pointer(sk->sk_reuseport_cb)) {
2272 /* we must lock primary chain too */
2273 spin_lock_bh(&hslot->lock);
2274 if (rcu_access_pointer(sk->sk_reuseport_cb))
2275 reuseport_detach_sock(sk);
2276
2277 if (hslot2 != nhslot2) {
2278 spin_lock(&hslot2->lock);
2279 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2280 hslot2->count--;
2281 spin_unlock(&hslot2->lock);
2282
2283 spin_lock(&nhslot2->lock);
2284 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2285 &nhslot2->head);
2286 nhslot2->count++;
2287 spin_unlock(&nhslot2->lock);
2288 }
2289
2290 spin_unlock_bh(&hslot->lock);
2291 }
2292
2293 /* Now process hash4 if necessary:
2294 * (1) update hslot4;
2295 * (2) update hslot2->hash4_cnt.
2296 * Note that hslot2/hslot4 should be checked separately, as
2297 * either of them may change with the other unchanged.
2298 */
2299 if (udp_hashed4(sk)) {
2300 spin_lock_bh(&hslot->lock);
2301
2302 udp_rehash4(udptable, sk, newhash4);
2303 if (hslot2 != nhslot2) {
2304 spin_lock(&hslot2->lock);
2305 udp_hash4_dec(hslot2);
2306 spin_unlock(&hslot2->lock);
2307
2308 spin_lock(&nhslot2->lock);
2309 udp_hash4_inc(nhslot2);
2310 spin_unlock(&nhslot2->lock);
2311 }
2312
2313 spin_unlock_bh(&hslot->lock);
2314 }
2315 }
2316 }
2317 EXPORT_IPV6_MOD(udp_lib_rehash);
2318
udp_v4_rehash(struct sock * sk)2319 void udp_v4_rehash(struct sock *sk)
2320 {
2321 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2322 inet_sk(sk)->inet_rcv_saddr,
2323 inet_sk(sk)->inet_num);
2324 u16 new_hash4 = udp_ehashfn(sock_net(sk),
2325 sk->sk_rcv_saddr, sk->sk_num,
2326 sk->sk_daddr, sk->sk_dport);
2327
2328 udp_lib_rehash(sk, new_hash, new_hash4);
2329 }
2330
__udp_queue_rcv_skb(struct sock * sk,struct sk_buff * skb)2331 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2332 {
2333 int rc;
2334
2335 if (inet_sk(sk)->inet_daddr) {
2336 sock_rps_save_rxhash(sk, skb);
2337 sk_mark_napi_id(sk, skb);
2338 sk_incoming_cpu_update(sk);
2339 } else {
2340 sk_mark_napi_id_once(sk, skb);
2341 }
2342
2343 rc = __udp_enqueue_schedule_skb(sk, skb);
2344 if (rc < 0) {
2345 int is_udplite = IS_UDPLITE(sk);
2346 int drop_reason;
2347
2348 /* Note that an ENOMEM error is charged twice */
2349 if (rc == -ENOMEM) {
2350 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2351 is_udplite);
2352 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2353 } else {
2354 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2355 is_udplite);
2356 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2357 }
2358 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2359 trace_udp_fail_queue_rcv_skb(rc, sk, skb);
2360 sk_skb_reason_drop(sk, skb, drop_reason);
2361 return -1;
2362 }
2363
2364 return 0;
2365 }
2366
2367 /* returns:
2368 * -1: error
2369 * 0: success
2370 * >0: "udp encap" protocol resubmission
2371 *
2372 * Note that in the success and error cases, the skb is assumed to
2373 * have either been requeued or freed.
2374 */
udp_queue_rcv_one_skb(struct sock * sk,struct sk_buff * skb)2375 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2376 {
2377 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2378 struct udp_sock *up = udp_sk(sk);
2379 int is_udplite = IS_UDPLITE(sk);
2380
2381 /*
2382 * Charge it to the socket, dropping if the queue is full.
2383 */
2384 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2385 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2386 goto drop;
2387 }
2388 nf_reset_ct(skb);
2389
2390 if (static_branch_unlikely(&udp_encap_needed_key) &&
2391 READ_ONCE(up->encap_type)) {
2392 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2393
2394 /*
2395 * This is an encapsulation socket so pass the skb to
2396 * the socket's udp_encap_rcv() hook. Otherwise, just
2397 * fall through and pass this up the UDP socket.
2398 * up->encap_rcv() returns the following value:
2399 * =0 if skb was successfully passed to the encap
2400 * handler or was discarded by it.
2401 * >0 if skb should be passed on to UDP.
2402 * <0 if skb should be resubmitted as proto -N
2403 */
2404
2405 /* if we're overly short, let UDP handle it */
2406 encap_rcv = READ_ONCE(up->encap_rcv);
2407 if (encap_rcv) {
2408 int ret;
2409
2410 /* Verify checksum before giving to encap */
2411 if (udp_lib_checksum_complete(skb))
2412 goto csum_error;
2413
2414 ret = encap_rcv(sk, skb);
2415 if (ret <= 0) {
2416 __UDP_INC_STATS(sock_net(sk),
2417 UDP_MIB_INDATAGRAMS,
2418 is_udplite);
2419 return -ret;
2420 }
2421 }
2422
2423 /* FALLTHROUGH -- it's a UDP Packet */
2424 }
2425
2426 /*
2427 * UDP-Lite specific tests, ignored on UDP sockets
2428 */
2429 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2430 u16 pcrlen = READ_ONCE(up->pcrlen);
2431
2432 /*
2433 * MIB statistics other than incrementing the error count are
2434 * disabled for the following two types of errors: these depend
2435 * on the application settings, not on the functioning of the
2436 * protocol stack as such.
2437 *
2438 * RFC 3828 here recommends (sec 3.3): "There should also be a
2439 * way ... to ... at least let the receiving application block
2440 * delivery of packets with coverage values less than a value
2441 * provided by the application."
2442 */
2443 if (pcrlen == 0) { /* full coverage was set */
2444 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2445 UDP_SKB_CB(skb)->cscov, skb->len);
2446 goto drop;
2447 }
2448 /* The next case involves violating the min. coverage requested
2449 * by the receiver. This is subtle: if receiver wants x and x is
2450 * greater than the buffersize/MTU then receiver will complain
2451 * that it wants x while sender emits packets of smaller size y.
2452 * Therefore the above ...()->partial_cov statement is essential.
2453 */
2454 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2455 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2456 UDP_SKB_CB(skb)->cscov, pcrlen);
2457 goto drop;
2458 }
2459 }
2460
2461 prefetch(&sk->sk_rmem_alloc);
2462 if (rcu_access_pointer(sk->sk_filter) &&
2463 udp_lib_checksum_complete(skb))
2464 goto csum_error;
2465
2466 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr), &drop_reason))
2467 goto drop;
2468
2469 udp_csum_pull_header(skb);
2470
2471 ipv4_pktinfo_prepare(sk, skb, true);
2472 return __udp_queue_rcv_skb(sk, skb);
2473
2474 csum_error:
2475 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2476 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2477 drop:
2478 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2479 udp_drops_inc(sk);
2480 sk_skb_reason_drop(sk, skb, drop_reason);
2481 return -1;
2482 }
2483
udp_queue_rcv_skb(struct sock * sk,struct sk_buff * skb)2484 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2485 {
2486 struct sk_buff *next, *segs;
2487 int ret;
2488
2489 if (likely(!udp_unexpected_gso(sk, skb)))
2490 return udp_queue_rcv_one_skb(sk, skb);
2491
2492 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2493 __skb_push(skb, -skb_mac_offset(skb));
2494 segs = udp_rcv_segment(sk, skb, true);
2495 skb_list_walk_safe(segs, skb, next) {
2496 __skb_pull(skb, skb_transport_offset(skb));
2497
2498 udp_post_segment_fix_csum(skb);
2499 ret = udp_queue_rcv_one_skb(sk, skb);
2500 if (ret > 0)
2501 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2502 }
2503 return 0;
2504 }
2505
2506 /* For TCP sockets, sk_rx_dst is protected by socket lock
2507 * For UDP, we use xchg() to guard against concurrent changes.
2508 */
udp_sk_rx_dst_set(struct sock * sk,struct dst_entry * dst)2509 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2510 {
2511 struct dst_entry *old;
2512
2513 if (dst_hold_safe(dst)) {
2514 old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst)));
2515 dst_release(old);
2516 return old != dst;
2517 }
2518 return false;
2519 }
2520 EXPORT_IPV6_MOD(udp_sk_rx_dst_set);
2521
2522 /*
2523 * Multicasts and broadcasts go to each listener.
2524 *
2525 * Note: called only from the BH handler context.
2526 */
__udp4_lib_mcast_deliver(struct net * net,struct sk_buff * skb,struct udphdr * uh,__be32 saddr,__be32 daddr,struct udp_table * udptable,int proto)2527 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2528 struct udphdr *uh,
2529 __be32 saddr, __be32 daddr,
2530 struct udp_table *udptable,
2531 int proto)
2532 {
2533 struct sock *sk, *first = NULL;
2534 unsigned short hnum = ntohs(uh->dest);
2535 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2536 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2537 unsigned int offset = offsetof(typeof(*sk), sk_node);
2538 int dif = skb->dev->ifindex;
2539 int sdif = inet_sdif(skb);
2540 struct hlist_node *node;
2541 struct sk_buff *nskb;
2542
2543 if (use_hash2) {
2544 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2545 udptable->mask;
2546 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2547 start_lookup:
2548 hslot = &udptable->hash2[hash2].hslot;
2549 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2550 }
2551
2552 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2553 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2554 uh->source, saddr, dif, sdif, hnum))
2555 continue;
2556
2557 if (!first) {
2558 first = sk;
2559 continue;
2560 }
2561 nskb = skb_clone(skb, GFP_ATOMIC);
2562
2563 if (unlikely(!nskb)) {
2564 udp_drops_inc(sk);
2565 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2566 IS_UDPLITE(sk));
2567 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2568 IS_UDPLITE(sk));
2569 continue;
2570 }
2571 if (udp_queue_rcv_skb(sk, nskb) > 0)
2572 consume_skb(nskb);
2573 }
2574
2575 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2576 if (use_hash2 && hash2 != hash2_any) {
2577 hash2 = hash2_any;
2578 goto start_lookup;
2579 }
2580
2581 if (first) {
2582 if (udp_queue_rcv_skb(first, skb) > 0)
2583 consume_skb(skb);
2584 } else {
2585 kfree_skb(skb);
2586 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2587 proto == IPPROTO_UDPLITE);
2588 }
2589 return 0;
2590 }
2591
2592 /* Initialize UDP checksum. If exited with zero value (success),
2593 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2594 * Otherwise, csum completion requires checksumming packet body,
2595 * including udp header and folding it to skb->csum.
2596 */
udp4_csum_init(struct sk_buff * skb,struct udphdr * uh,int proto)2597 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2598 int proto)
2599 {
2600 int err;
2601
2602 UDP_SKB_CB(skb)->partial_cov = 0;
2603 UDP_SKB_CB(skb)->cscov = skb->len;
2604
2605 if (proto == IPPROTO_UDPLITE) {
2606 err = udplite_checksum_init(skb, uh);
2607 if (err)
2608 return err;
2609
2610 if (UDP_SKB_CB(skb)->partial_cov) {
2611 skb->csum = inet_compute_pseudo(skb, proto);
2612 return 0;
2613 }
2614 }
2615
2616 /* Note, we are only interested in != 0 or == 0, thus the
2617 * force to int.
2618 */
2619 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2620 inet_compute_pseudo);
2621 if (err)
2622 return err;
2623
2624 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2625 /* If SW calculated the value, we know it's bad */
2626 if (skb->csum_complete_sw)
2627 return 1;
2628
2629 /* HW says the value is bad. Let's validate that.
2630 * skb->csum is no longer the full packet checksum,
2631 * so don't treat it as such.
2632 */
2633 skb_checksum_complete_unset(skb);
2634 }
2635
2636 return 0;
2637 }
2638
2639 /* wrapper for udp_queue_rcv_skb taking care of csum conversion and
2640 * return code conversion for ip layer consumption
2641 */
udp_unicast_rcv_skb(struct sock * sk,struct sk_buff * skb,struct udphdr * uh)2642 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2643 struct udphdr *uh)
2644 {
2645 int ret;
2646
2647 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2648 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2649
2650 ret = udp_queue_rcv_skb(sk, skb);
2651
2652 /* a return value > 0 means to resubmit the input, but
2653 * it wants the return to be -protocol, or 0
2654 */
2655 if (ret > 0)
2656 return -ret;
2657 return 0;
2658 }
2659
2660 /*
2661 * All we need to do is get the socket, and then do a checksum.
2662 */
2663
__udp4_lib_rcv(struct sk_buff * skb,struct udp_table * udptable,int proto)2664 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2665 int proto)
2666 {
2667 struct sock *sk = NULL;
2668 struct udphdr *uh;
2669 unsigned short ulen;
2670 struct rtable *rt = skb_rtable(skb);
2671 __be32 saddr, daddr;
2672 struct net *net = dev_net(skb->dev);
2673 bool refcounted;
2674 int drop_reason;
2675
2676 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2677
2678 /*
2679 * Validate the packet.
2680 */
2681 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2682 goto drop; /* No space for header. */
2683
2684 uh = udp_hdr(skb);
2685 ulen = ntohs(uh->len);
2686 saddr = ip_hdr(skb)->saddr;
2687 daddr = ip_hdr(skb)->daddr;
2688
2689 if (ulen > skb->len)
2690 goto short_packet;
2691
2692 if (proto == IPPROTO_UDP) {
2693 /* UDP validates ulen. */
2694 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2695 goto short_packet;
2696 uh = udp_hdr(skb);
2697 }
2698
2699 if (udp4_csum_init(skb, uh, proto))
2700 goto csum_error;
2701
2702 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2703 &refcounted, udp_ehashfn);
2704 if (IS_ERR(sk))
2705 goto no_sk;
2706
2707 if (sk) {
2708 struct dst_entry *dst = skb_dst(skb);
2709 int ret;
2710
2711 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2712 udp_sk_rx_dst_set(sk, dst);
2713
2714 ret = udp_unicast_rcv_skb(sk, skb, uh);
2715 if (refcounted)
2716 sock_put(sk);
2717 return ret;
2718 }
2719
2720 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2721 return __udp4_lib_mcast_deliver(net, skb, uh,
2722 saddr, daddr, udptable, proto);
2723
2724 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2725 if (sk)
2726 return udp_unicast_rcv_skb(sk, skb, uh);
2727 no_sk:
2728 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2729 goto drop;
2730 nf_reset_ct(skb);
2731
2732 /* No socket. Drop packet silently, if checksum is wrong */
2733 if (udp_lib_checksum_complete(skb))
2734 goto csum_error;
2735
2736 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2737 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2738 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2739
2740 /*
2741 * Hmm. We got an UDP packet to a port to which we
2742 * don't wanna listen. Ignore it.
2743 */
2744 sk_skb_reason_drop(sk, skb, drop_reason);
2745 return 0;
2746
2747 short_packet:
2748 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2749 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2750 proto == IPPROTO_UDPLITE ? "Lite" : "",
2751 &saddr, ntohs(uh->source),
2752 ulen, skb->len,
2753 &daddr, ntohs(uh->dest));
2754 goto drop;
2755
2756 csum_error:
2757 /*
2758 * RFC1122: OK. Discards the bad packet silently (as far as
2759 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2760 */
2761 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2762 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2763 proto == IPPROTO_UDPLITE ? "Lite" : "",
2764 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2765 ulen);
2766 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2767 drop:
2768 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2769 sk_skb_reason_drop(sk, skb, drop_reason);
2770 return 0;
2771 }
2772
2773 /* We can only early demux multicast if there is a single matching socket.
2774 * If more than one socket found returns NULL
2775 */
__udp4_lib_mcast_demux_lookup(struct net * net,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif)2776 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2777 __be16 loc_port, __be32 loc_addr,
2778 __be16 rmt_port, __be32 rmt_addr,
2779 int dif, int sdif)
2780 {
2781 struct udp_table *udptable = net->ipv4.udp_table;
2782 unsigned short hnum = ntohs(loc_port);
2783 struct sock *sk, *result;
2784 struct udp_hslot *hslot;
2785 unsigned int slot;
2786
2787 slot = udp_hashfn(net, hnum, udptable->mask);
2788 hslot = &udptable->hash[slot];
2789
2790 /* Do not bother scanning a too big list */
2791 if (hslot->count > 10)
2792 return NULL;
2793
2794 result = NULL;
2795 sk_for_each_rcu(sk, &hslot->head) {
2796 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2797 rmt_port, rmt_addr, dif, sdif, hnum)) {
2798 if (result)
2799 return NULL;
2800 result = sk;
2801 }
2802 }
2803
2804 return result;
2805 }
2806
2807 /* For unicast we should only early demux connected sockets or we can
2808 * break forwarding setups. The chains here can be long so only check
2809 * if the first socket is an exact match and if not move on.
2810 */
__udp4_lib_demux_lookup(struct net * net,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif)2811 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2812 __be16 loc_port, __be32 loc_addr,
2813 __be16 rmt_port, __be32 rmt_addr,
2814 int dif, int sdif)
2815 {
2816 struct udp_table *udptable = net->ipv4.udp_table;
2817 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2818 unsigned short hnum = ntohs(loc_port);
2819 struct udp_hslot *hslot2;
2820 unsigned int hash2;
2821 __portpair ports;
2822 struct sock *sk;
2823
2824 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2825 hslot2 = udp_hashslot2(udptable, hash2);
2826 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2827
2828 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2829 if (inet_match(net, sk, acookie, ports, dif, sdif))
2830 return sk;
2831 /* Only check first socket in chain */
2832 break;
2833 }
2834 return NULL;
2835 }
2836
udp_v4_early_demux(struct sk_buff * skb)2837 enum skb_drop_reason udp_v4_early_demux(struct sk_buff *skb)
2838 {
2839 struct net *net = dev_net(skb->dev);
2840 struct in_device *in_dev = NULL;
2841 const struct iphdr *iph;
2842 const struct udphdr *uh;
2843 struct sock *sk = NULL;
2844 struct dst_entry *dst;
2845 int dif = skb->dev->ifindex;
2846 int sdif = inet_sdif(skb);
2847 int ours;
2848
2849 /* validate the packet */
2850 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2851 return SKB_NOT_DROPPED_YET;
2852
2853 iph = ip_hdr(skb);
2854 uh = udp_hdr(skb);
2855
2856 if (skb->pkt_type == PACKET_MULTICAST) {
2857 in_dev = __in_dev_get_rcu(skb->dev);
2858
2859 if (!in_dev)
2860 return SKB_NOT_DROPPED_YET;
2861
2862 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2863 iph->protocol);
2864 if (!ours)
2865 return SKB_NOT_DROPPED_YET;
2866
2867 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2868 uh->source, iph->saddr,
2869 dif, sdif);
2870 } else if (skb->pkt_type == PACKET_HOST) {
2871 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2872 uh->source, iph->saddr, dif, sdif);
2873 }
2874
2875 if (!sk)
2876 return SKB_NOT_DROPPED_YET;
2877
2878 skb->sk = sk;
2879 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2880 skb->destructor = sock_pfree;
2881 dst = rcu_dereference(sk->sk_rx_dst);
2882
2883 if (dst)
2884 dst = dst_check(dst, 0);
2885 if (dst) {
2886 u32 itag = 0;
2887
2888 /* set noref for now.
2889 * any place which wants to hold dst has to call
2890 * dst_hold_safe()
2891 */
2892 skb_dst_set_noref(skb, dst);
2893
2894 /* for unconnected multicast sockets we need to validate
2895 * the source on each packet
2896 */
2897 if (!inet_sk(sk)->inet_daddr && in_dev)
2898 return ip_mc_validate_source(skb, iph->daddr,
2899 iph->saddr,
2900 ip4h_dscp(iph),
2901 skb->dev, in_dev, &itag);
2902 }
2903 return SKB_NOT_DROPPED_YET;
2904 }
2905
udp_rcv(struct sk_buff * skb)2906 int udp_rcv(struct sk_buff *skb)
2907 {
2908 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2909 }
2910
udp_destroy_sock(struct sock * sk)2911 void udp_destroy_sock(struct sock *sk)
2912 {
2913 struct udp_sock *up = udp_sk(sk);
2914 bool slow = lock_sock_fast(sk);
2915
2916 /* protects from races with udp_abort() */
2917 sock_set_flag(sk, SOCK_DEAD);
2918 udp_flush_pending_frames(sk);
2919 unlock_sock_fast(sk, slow);
2920 if (static_branch_unlikely(&udp_encap_needed_key)) {
2921 if (up->encap_type) {
2922 void (*encap_destroy)(struct sock *sk);
2923 encap_destroy = READ_ONCE(up->encap_destroy);
2924 if (encap_destroy)
2925 encap_destroy(sk);
2926 }
2927 if (udp_test_bit(ENCAP_ENABLED, sk)) {
2928 static_branch_dec(&udp_encap_needed_key);
2929 udp_tunnel_cleanup_gro(sk);
2930 }
2931 }
2932 }
2933
2934 typedef struct sk_buff *(*udp_gro_receive_t)(struct sock *sk,
2935 struct list_head *head,
2936 struct sk_buff *skb);
2937
set_xfrm_gro_udp_encap_rcv(__u16 encap_type,unsigned short family,struct sock * sk)2938 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2939 struct sock *sk)
2940 {
2941 #ifdef CONFIG_XFRM
2942 udp_gro_receive_t new_gro_receive;
2943
2944 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2945 if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2946 new_gro_receive = ipv6_stub->xfrm6_gro_udp_encap_rcv;
2947 else
2948 new_gro_receive = xfrm4_gro_udp_encap_rcv;
2949
2950 if (udp_sk(sk)->gro_receive != new_gro_receive) {
2951 /*
2952 * With IPV6_ADDRFORM the gro callback could change
2953 * after being set, unregister the old one, if valid.
2954 */
2955 if (udp_sk(sk)->gro_receive)
2956 udp_tunnel_update_gro_rcv(sk, false);
2957
2958 WRITE_ONCE(udp_sk(sk)->gro_receive, new_gro_receive);
2959 udp_tunnel_update_gro_rcv(sk, true);
2960 }
2961 }
2962 #endif
2963 }
2964
2965 /*
2966 * Socket option code for UDP
2967 */
udp_lib_setsockopt(struct sock * sk,int level,int optname,sockptr_t optval,unsigned int optlen,int (* push_pending_frames)(struct sock *))2968 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2969 sockptr_t optval, unsigned int optlen,
2970 int (*push_pending_frames)(struct sock *))
2971 {
2972 struct udp_sock *up = udp_sk(sk);
2973 int val, valbool;
2974 int err = 0;
2975 int is_udplite = IS_UDPLITE(sk);
2976
2977 if (level == SOL_SOCKET) {
2978 err = sk_setsockopt(sk, level, optname, optval, optlen);
2979
2980 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2981 sockopt_lock_sock(sk);
2982 /* paired with READ_ONCE in udp_rmem_release() */
2983 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2984 sockopt_release_sock(sk);
2985 }
2986 return err;
2987 }
2988
2989 if (optlen < sizeof(int))
2990 return -EINVAL;
2991
2992 if (copy_from_sockptr(&val, optval, sizeof(val)))
2993 return -EFAULT;
2994
2995 valbool = val ? 1 : 0;
2996
2997 switch (optname) {
2998 case UDP_CORK:
2999 if (val != 0) {
3000 udp_set_bit(CORK, sk);
3001 } else {
3002 udp_clear_bit(CORK, sk);
3003 lock_sock(sk);
3004 push_pending_frames(sk);
3005 release_sock(sk);
3006 }
3007 break;
3008
3009 case UDP_ENCAP:
3010 sockopt_lock_sock(sk);
3011 switch (val) {
3012 case 0:
3013 #ifdef CONFIG_XFRM
3014 case UDP_ENCAP_ESPINUDP:
3015 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
3016 #if IS_ENABLED(CONFIG_IPV6)
3017 if (sk->sk_family == AF_INET6)
3018 WRITE_ONCE(up->encap_rcv,
3019 ipv6_stub->xfrm6_udp_encap_rcv);
3020 else
3021 #endif
3022 WRITE_ONCE(up->encap_rcv,
3023 xfrm4_udp_encap_rcv);
3024 #endif
3025 fallthrough;
3026 case UDP_ENCAP_L2TPINUDP:
3027 WRITE_ONCE(up->encap_type, val);
3028 udp_tunnel_encap_enable(sk);
3029 break;
3030 default:
3031 err = -ENOPROTOOPT;
3032 break;
3033 }
3034 sockopt_release_sock(sk);
3035 break;
3036
3037 case UDP_NO_CHECK6_TX:
3038 udp_set_no_check6_tx(sk, valbool);
3039 break;
3040
3041 case UDP_NO_CHECK6_RX:
3042 udp_set_no_check6_rx(sk, valbool);
3043 break;
3044
3045 case UDP_SEGMENT:
3046 if (val < 0 || val > USHRT_MAX)
3047 return -EINVAL;
3048 WRITE_ONCE(up->gso_size, val);
3049 break;
3050
3051 case UDP_GRO:
3052 sockopt_lock_sock(sk);
3053 /* when enabling GRO, accept the related GSO packet type */
3054 if (valbool)
3055 udp_tunnel_encap_enable(sk);
3056 udp_assign_bit(GRO_ENABLED, sk, valbool);
3057 udp_assign_bit(ACCEPT_L4, sk, valbool);
3058 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
3059 sockopt_release_sock(sk);
3060 break;
3061
3062 /*
3063 * UDP-Lite's partial checksum coverage (RFC 3828).
3064 */
3065 /* The sender sets actual checksum coverage length via this option.
3066 * The case coverage > packet length is handled by send module. */
3067 case UDPLITE_SEND_CSCOV:
3068 if (!is_udplite) /* Disable the option on UDP sockets */
3069 return -ENOPROTOOPT;
3070 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
3071 val = 8;
3072 else if (val > USHRT_MAX)
3073 val = USHRT_MAX;
3074 WRITE_ONCE(up->pcslen, val);
3075 udp_set_bit(UDPLITE_SEND_CC, sk);
3076 break;
3077
3078 /* The receiver specifies a minimum checksum coverage value. To make
3079 * sense, this should be set to at least 8 (as done below). If zero is
3080 * used, this again means full checksum coverage. */
3081 case UDPLITE_RECV_CSCOV:
3082 if (!is_udplite) /* Disable the option on UDP sockets */
3083 return -ENOPROTOOPT;
3084 if (val != 0 && val < 8) /* Avoid silly minimal values. */
3085 val = 8;
3086 else if (val > USHRT_MAX)
3087 val = USHRT_MAX;
3088 WRITE_ONCE(up->pcrlen, val);
3089 udp_set_bit(UDPLITE_RECV_CC, sk);
3090 break;
3091
3092 default:
3093 err = -ENOPROTOOPT;
3094 break;
3095 }
3096
3097 return err;
3098 }
3099 EXPORT_IPV6_MOD(udp_lib_setsockopt);
3100
udp_setsockopt(struct sock * sk,int level,int optname,sockptr_t optval,unsigned int optlen)3101 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
3102 unsigned int optlen)
3103 {
3104 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
3105 return udp_lib_setsockopt(sk, level, optname,
3106 optval, optlen,
3107 udp_push_pending_frames);
3108 return ip_setsockopt(sk, level, optname, optval, optlen);
3109 }
3110
udp_lib_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * optlen)3111 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
3112 char __user *optval, int __user *optlen)
3113 {
3114 struct udp_sock *up = udp_sk(sk);
3115 int val, len;
3116
3117 if (get_user(len, optlen))
3118 return -EFAULT;
3119
3120 if (len < 0)
3121 return -EINVAL;
3122
3123 len = min_t(unsigned int, len, sizeof(int));
3124
3125 switch (optname) {
3126 case UDP_CORK:
3127 val = udp_test_bit(CORK, sk);
3128 break;
3129
3130 case UDP_ENCAP:
3131 val = READ_ONCE(up->encap_type);
3132 break;
3133
3134 case UDP_NO_CHECK6_TX:
3135 val = udp_get_no_check6_tx(sk);
3136 break;
3137
3138 case UDP_NO_CHECK6_RX:
3139 val = udp_get_no_check6_rx(sk);
3140 break;
3141
3142 case UDP_SEGMENT:
3143 val = READ_ONCE(up->gso_size);
3144 break;
3145
3146 case UDP_GRO:
3147 val = udp_test_bit(GRO_ENABLED, sk);
3148 break;
3149
3150 /* The following two cannot be changed on UDP sockets, the return is
3151 * always 0 (which corresponds to the full checksum coverage of UDP). */
3152 case UDPLITE_SEND_CSCOV:
3153 val = READ_ONCE(up->pcslen);
3154 break;
3155
3156 case UDPLITE_RECV_CSCOV:
3157 val = READ_ONCE(up->pcrlen);
3158 break;
3159
3160 default:
3161 return -ENOPROTOOPT;
3162 }
3163
3164 if (put_user(len, optlen))
3165 return -EFAULT;
3166 if (copy_to_user(optval, &val, len))
3167 return -EFAULT;
3168 return 0;
3169 }
3170 EXPORT_IPV6_MOD(udp_lib_getsockopt);
3171
udp_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * optlen)3172 int udp_getsockopt(struct sock *sk, int level, int optname,
3173 char __user *optval, int __user *optlen)
3174 {
3175 if (level == SOL_UDP || level == SOL_UDPLITE)
3176 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
3177 return ip_getsockopt(sk, level, optname, optval, optlen);
3178 }
3179
3180 /**
3181 * udp_poll - wait for a UDP event.
3182 * @file: - file struct
3183 * @sock: - socket
3184 * @wait: - poll table
3185 *
3186 * This is same as datagram poll, except for the special case of
3187 * blocking sockets. If application is using a blocking fd
3188 * and a packet with checksum error is in the queue;
3189 * then it could get return from select indicating data available
3190 * but then block when reading it. Add special case code
3191 * to work around these arguably broken applications.
3192 */
udp_poll(struct file * file,struct socket * sock,poll_table * wait)3193 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
3194 {
3195 __poll_t mask = datagram_poll(file, sock, wait);
3196 struct sock *sk = sock->sk;
3197
3198 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
3199 mask |= EPOLLIN | EPOLLRDNORM;
3200
3201 /* Check for false positives due to checksum errors */
3202 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
3203 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
3204 mask &= ~(EPOLLIN | EPOLLRDNORM);
3205
3206 /* psock ingress_msg queue should not contain any bad checksum frames */
3207 if (sk_is_readable(sk))
3208 mask |= EPOLLIN | EPOLLRDNORM;
3209 return mask;
3210
3211 }
3212 EXPORT_IPV6_MOD(udp_poll);
3213
udp_abort(struct sock * sk,int err)3214 int udp_abort(struct sock *sk, int err)
3215 {
3216 if (!has_current_bpf_ctx())
3217 lock_sock(sk);
3218
3219 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
3220 * with close()
3221 */
3222 if (sock_flag(sk, SOCK_DEAD))
3223 goto out;
3224
3225 sk->sk_err = err;
3226 sk_error_report(sk);
3227 __udp_disconnect(sk, 0);
3228
3229 out:
3230 if (!has_current_bpf_ctx())
3231 release_sock(sk);
3232
3233 return 0;
3234 }
3235 EXPORT_IPV6_MOD_GPL(udp_abort);
3236
3237 struct proto udp_prot = {
3238 .name = "UDP",
3239 .owner = THIS_MODULE,
3240 .close = udp_lib_close,
3241 .pre_connect = udp_pre_connect,
3242 .connect = udp_connect,
3243 .disconnect = udp_disconnect,
3244 .ioctl = udp_ioctl,
3245 .init = udp_init_sock,
3246 .destroy = udp_destroy_sock,
3247 .setsockopt = udp_setsockopt,
3248 .getsockopt = udp_getsockopt,
3249 .sendmsg = udp_sendmsg,
3250 .recvmsg = udp_recvmsg,
3251 .splice_eof = udp_splice_eof,
3252 .release_cb = ip4_datagram_release_cb,
3253 .hash = udp_lib_hash,
3254 .unhash = udp_lib_unhash,
3255 .rehash = udp_v4_rehash,
3256 .get_port = udp_v4_get_port,
3257 .put_port = udp_lib_unhash,
3258 #ifdef CONFIG_BPF_SYSCALL
3259 .psock_update_sk_prot = udp_bpf_update_proto,
3260 #endif
3261 .memory_allocated = &net_aligned_data.udp_memory_allocated,
3262 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
3263
3264 .sysctl_mem = sysctl_udp_mem,
3265 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
3266 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
3267 .obj_size = sizeof(struct udp_sock),
3268 .h.udp_table = NULL,
3269 .diag_destroy = udp_abort,
3270 };
3271 EXPORT_SYMBOL(udp_prot);
3272
3273 /* ------------------------------------------------------------------------ */
3274 #ifdef CONFIG_PROC_FS
3275
3276 static unsigned short seq_file_family(const struct seq_file *seq);
seq_sk_match(struct seq_file * seq,const struct sock * sk)3277 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
3278 {
3279 unsigned short family = seq_file_family(seq);
3280
3281 /* AF_UNSPEC is used as a match all */
3282 return ((family == AF_UNSPEC || family == sk->sk_family) &&
3283 net_eq(sock_net(sk), seq_file_net(seq)));
3284 }
3285
3286 #ifdef CONFIG_BPF_SYSCALL
3287 static const struct seq_operations bpf_iter_udp_seq_ops;
3288 #endif
udp_get_table_seq(struct seq_file * seq,struct net * net)3289 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
3290 struct net *net)
3291 {
3292 const struct udp_seq_afinfo *afinfo;
3293
3294 #ifdef CONFIG_BPF_SYSCALL
3295 if (seq->op == &bpf_iter_udp_seq_ops)
3296 return net->ipv4.udp_table;
3297 #endif
3298
3299 afinfo = pde_data(file_inode(seq->file));
3300 return afinfo->udp_table ? : net->ipv4.udp_table;
3301 }
3302
udp_get_first(struct seq_file * seq,int start)3303 static struct sock *udp_get_first(struct seq_file *seq, int start)
3304 {
3305 struct udp_iter_state *state = seq->private;
3306 struct net *net = seq_file_net(seq);
3307 struct udp_table *udptable;
3308 struct sock *sk;
3309
3310 udptable = udp_get_table_seq(seq, net);
3311
3312 for (state->bucket = start; state->bucket <= udptable->mask;
3313 ++state->bucket) {
3314 struct udp_hslot *hslot = &udptable->hash[state->bucket];
3315
3316 if (hlist_empty(&hslot->head))
3317 continue;
3318
3319 spin_lock_bh(&hslot->lock);
3320 sk_for_each(sk, &hslot->head) {
3321 if (seq_sk_match(seq, sk))
3322 goto found;
3323 }
3324 spin_unlock_bh(&hslot->lock);
3325 }
3326 sk = NULL;
3327 found:
3328 return sk;
3329 }
3330
udp_get_next(struct seq_file * seq,struct sock * sk)3331 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3332 {
3333 struct udp_iter_state *state = seq->private;
3334 struct net *net = seq_file_net(seq);
3335 struct udp_table *udptable;
3336
3337 do {
3338 sk = sk_next(sk);
3339 } while (sk && !seq_sk_match(seq, sk));
3340
3341 if (!sk) {
3342 udptable = udp_get_table_seq(seq, net);
3343
3344 if (state->bucket <= udptable->mask)
3345 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3346
3347 return udp_get_first(seq, state->bucket + 1);
3348 }
3349 return sk;
3350 }
3351
udp_get_idx(struct seq_file * seq,loff_t pos)3352 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3353 {
3354 struct sock *sk = udp_get_first(seq, 0);
3355
3356 if (sk)
3357 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3358 --pos;
3359 return pos ? NULL : sk;
3360 }
3361
udp_seq_start(struct seq_file * seq,loff_t * pos)3362 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3363 {
3364 struct udp_iter_state *state = seq->private;
3365 state->bucket = MAX_UDP_PORTS;
3366
3367 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3368 }
3369 EXPORT_IPV6_MOD(udp_seq_start);
3370
udp_seq_next(struct seq_file * seq,void * v,loff_t * pos)3371 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3372 {
3373 struct sock *sk;
3374
3375 if (v == SEQ_START_TOKEN)
3376 sk = udp_get_idx(seq, 0);
3377 else
3378 sk = udp_get_next(seq, v);
3379
3380 ++*pos;
3381 return sk;
3382 }
3383 EXPORT_IPV6_MOD(udp_seq_next);
3384
udp_seq_stop(struct seq_file * seq,void * v)3385 void udp_seq_stop(struct seq_file *seq, void *v)
3386 {
3387 struct udp_iter_state *state = seq->private;
3388 struct udp_table *udptable;
3389
3390 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3391
3392 if (state->bucket <= udptable->mask)
3393 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3394 }
3395 EXPORT_IPV6_MOD(udp_seq_stop);
3396
3397 /* ------------------------------------------------------------------------ */
udp4_format_sock(struct sock * sp,struct seq_file * f,int bucket)3398 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3399 int bucket)
3400 {
3401 struct inet_sock *inet = inet_sk(sp);
3402 __be32 dest = inet->inet_daddr;
3403 __be32 src = inet->inet_rcv_saddr;
3404 __u16 destp = ntohs(inet->inet_dport);
3405 __u16 srcp = ntohs(inet->inet_sport);
3406
3407 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3408 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3409 bucket, src, srcp, dest, destp, sp->sk_state,
3410 sk_wmem_alloc_get(sp),
3411 udp_rqueue_get(sp),
3412 0, 0L, 0,
3413 from_kuid_munged(seq_user_ns(f), sk_uid(sp)),
3414 0, sock_i_ino(sp),
3415 refcount_read(&sp->sk_refcnt), sp,
3416 sk_drops_read(sp));
3417 }
3418
udp4_seq_show(struct seq_file * seq,void * v)3419 int udp4_seq_show(struct seq_file *seq, void *v)
3420 {
3421 seq_setwidth(seq, 127);
3422 if (v == SEQ_START_TOKEN)
3423 seq_puts(seq, " sl local_address rem_address st tx_queue "
3424 "rx_queue tr tm->when retrnsmt uid timeout "
3425 "inode ref pointer drops");
3426 else {
3427 struct udp_iter_state *state = seq->private;
3428
3429 udp4_format_sock(v, seq, state->bucket);
3430 }
3431 seq_pad(seq, '\n');
3432 return 0;
3433 }
3434
3435 #ifdef CONFIG_BPF_SYSCALL
3436 struct bpf_iter__udp {
3437 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3438 __bpf_md_ptr(struct udp_sock *, udp_sk);
3439 uid_t uid __aligned(8);
3440 int bucket __aligned(8);
3441 };
3442
3443 union bpf_udp_iter_batch_item {
3444 struct sock *sk;
3445 __u64 cookie;
3446 };
3447
3448 struct bpf_udp_iter_state {
3449 struct udp_iter_state state;
3450 unsigned int cur_sk;
3451 unsigned int end_sk;
3452 unsigned int max_sk;
3453 union bpf_udp_iter_batch_item *batch;
3454 };
3455
3456 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3457 unsigned int new_batch_sz, gfp_t flags);
bpf_iter_udp_resume(struct sock * first_sk,union bpf_udp_iter_batch_item * cookies,int n_cookies)3458 static struct sock *bpf_iter_udp_resume(struct sock *first_sk,
3459 union bpf_udp_iter_batch_item *cookies,
3460 int n_cookies)
3461 {
3462 struct sock *sk = NULL;
3463 int i;
3464
3465 for (i = 0; i < n_cookies; i++) {
3466 sk = first_sk;
3467 udp_portaddr_for_each_entry_from(sk)
3468 if (cookies[i].cookie == atomic64_read(&sk->sk_cookie))
3469 goto done;
3470 }
3471 done:
3472 return sk;
3473 }
3474
bpf_iter_udp_batch(struct seq_file * seq)3475 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3476 {
3477 struct bpf_udp_iter_state *iter = seq->private;
3478 struct udp_iter_state *state = &iter->state;
3479 unsigned int find_cookie, end_cookie;
3480 struct net *net = seq_file_net(seq);
3481 struct udp_table *udptable;
3482 unsigned int batch_sks = 0;
3483 int resume_bucket;
3484 int resizes = 0;
3485 struct sock *sk;
3486 int err = 0;
3487
3488 resume_bucket = state->bucket;
3489
3490 /* The current batch is done, so advance the bucket. */
3491 if (iter->cur_sk == iter->end_sk)
3492 state->bucket++;
3493
3494 udptable = udp_get_table_seq(seq, net);
3495
3496 again:
3497 /* New batch for the next bucket.
3498 * Iterate over the hash table to find a bucket with sockets matching
3499 * the iterator attributes, and return the first matching socket from
3500 * the bucket. The remaining matched sockets from the bucket are batched
3501 * before releasing the bucket lock. This allows BPF programs that are
3502 * called in seq_show to acquire the bucket lock if needed.
3503 */
3504 find_cookie = iter->cur_sk;
3505 end_cookie = iter->end_sk;
3506 iter->cur_sk = 0;
3507 iter->end_sk = 0;
3508 batch_sks = 0;
3509
3510 for (; state->bucket <= udptable->mask; state->bucket++) {
3511 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket].hslot;
3512
3513 if (hlist_empty(&hslot2->head))
3514 goto next_bucket;
3515
3516 spin_lock_bh(&hslot2->lock);
3517 sk = hlist_entry_safe(hslot2->head.first, struct sock,
3518 __sk_common.skc_portaddr_node);
3519 /* Resume from the first (in iteration order) unseen socket from
3520 * the last batch that still exists in resume_bucket. Most of
3521 * the time this will just be where the last iteration left off
3522 * in resume_bucket unless that socket disappeared between
3523 * reads.
3524 */
3525 if (state->bucket == resume_bucket)
3526 sk = bpf_iter_udp_resume(sk, &iter->batch[find_cookie],
3527 end_cookie - find_cookie);
3528 fill_batch:
3529 udp_portaddr_for_each_entry_from(sk) {
3530 if (seq_sk_match(seq, sk)) {
3531 if (iter->end_sk < iter->max_sk) {
3532 sock_hold(sk);
3533 iter->batch[iter->end_sk++].sk = sk;
3534 }
3535 batch_sks++;
3536 }
3537 }
3538
3539 /* Allocate a larger batch and try again. */
3540 if (unlikely(resizes <= 1 && iter->end_sk &&
3541 iter->end_sk != batch_sks)) {
3542 resizes++;
3543
3544 /* First, try with GFP_USER to maximize the chances of
3545 * grabbing more memory.
3546 */
3547 if (resizes == 1) {
3548 spin_unlock_bh(&hslot2->lock);
3549 err = bpf_iter_udp_realloc_batch(iter,
3550 batch_sks * 3 / 2,
3551 GFP_USER);
3552 if (err)
3553 return ERR_PTR(err);
3554 /* Start over. */
3555 goto again;
3556 }
3557
3558 /* Next, hold onto the lock, so the bucket doesn't
3559 * change while we get the rest of the sockets.
3560 */
3561 err = bpf_iter_udp_realloc_batch(iter, batch_sks,
3562 GFP_NOWAIT);
3563 if (err) {
3564 spin_unlock_bh(&hslot2->lock);
3565 return ERR_PTR(err);
3566 }
3567
3568 /* Pick up where we left off. */
3569 sk = iter->batch[iter->end_sk - 1].sk;
3570 sk = hlist_entry_safe(sk->__sk_common.skc_portaddr_node.next,
3571 struct sock,
3572 __sk_common.skc_portaddr_node);
3573 batch_sks = iter->end_sk;
3574 goto fill_batch;
3575 }
3576
3577 spin_unlock_bh(&hslot2->lock);
3578
3579 if (iter->end_sk)
3580 break;
3581 next_bucket:
3582 resizes = 0;
3583 }
3584
3585 WARN_ON_ONCE(iter->end_sk != batch_sks);
3586 return iter->end_sk ? iter->batch[0].sk : NULL;
3587 }
3588
bpf_iter_udp_seq_next(struct seq_file * seq,void * v,loff_t * pos)3589 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3590 {
3591 struct bpf_udp_iter_state *iter = seq->private;
3592 struct sock *sk;
3593
3594 /* Whenever seq_next() is called, the iter->cur_sk is
3595 * done with seq_show(), so unref the iter->cur_sk.
3596 */
3597 if (iter->cur_sk < iter->end_sk)
3598 sock_put(iter->batch[iter->cur_sk++].sk);
3599
3600 /* After updating iter->cur_sk, check if there are more sockets
3601 * available in the current bucket batch.
3602 */
3603 if (iter->cur_sk < iter->end_sk)
3604 sk = iter->batch[iter->cur_sk].sk;
3605 else
3606 /* Prepare a new batch. */
3607 sk = bpf_iter_udp_batch(seq);
3608
3609 ++*pos;
3610 return sk;
3611 }
3612
bpf_iter_udp_seq_start(struct seq_file * seq,loff_t * pos)3613 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3614 {
3615 /* bpf iter does not support lseek, so it always
3616 * continue from where it was stop()-ped.
3617 */
3618 if (*pos)
3619 return bpf_iter_udp_batch(seq);
3620
3621 return SEQ_START_TOKEN;
3622 }
3623
udp_prog_seq_show(struct bpf_prog * prog,struct bpf_iter_meta * meta,struct udp_sock * udp_sk,uid_t uid,int bucket)3624 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3625 struct udp_sock *udp_sk, uid_t uid, int bucket)
3626 {
3627 struct bpf_iter__udp ctx;
3628
3629 meta->seq_num--; /* skip SEQ_START_TOKEN */
3630 ctx.meta = meta;
3631 ctx.udp_sk = udp_sk;
3632 ctx.uid = uid;
3633 ctx.bucket = bucket;
3634 return bpf_iter_run_prog(prog, &ctx);
3635 }
3636
bpf_iter_udp_seq_show(struct seq_file * seq,void * v)3637 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3638 {
3639 struct udp_iter_state *state = seq->private;
3640 struct bpf_iter_meta meta;
3641 struct bpf_prog *prog;
3642 struct sock *sk = v;
3643 uid_t uid;
3644 int ret;
3645
3646 if (v == SEQ_START_TOKEN)
3647 return 0;
3648
3649 lock_sock(sk);
3650
3651 if (unlikely(sk_unhashed(sk))) {
3652 ret = SEQ_SKIP;
3653 goto unlock;
3654 }
3655
3656 uid = from_kuid_munged(seq_user_ns(seq), sk_uid(sk));
3657 meta.seq = seq;
3658 prog = bpf_iter_get_info(&meta, false);
3659 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3660
3661 unlock:
3662 release_sock(sk);
3663 return ret;
3664 }
3665
bpf_iter_udp_put_batch(struct bpf_udp_iter_state * iter)3666 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3667 {
3668 union bpf_udp_iter_batch_item *item;
3669 unsigned int cur_sk = iter->cur_sk;
3670 __u64 cookie;
3671
3672 /* Remember the cookies of the sockets we haven't seen yet, so we can
3673 * pick up where we left off next time around.
3674 */
3675 while (cur_sk < iter->end_sk) {
3676 item = &iter->batch[cur_sk++];
3677 cookie = sock_gen_cookie(item->sk);
3678 sock_put(item->sk);
3679 item->cookie = cookie;
3680 }
3681 }
3682
bpf_iter_udp_seq_stop(struct seq_file * seq,void * v)3683 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3684 {
3685 struct bpf_udp_iter_state *iter = seq->private;
3686 struct bpf_iter_meta meta;
3687 struct bpf_prog *prog;
3688
3689 if (!v) {
3690 meta.seq = seq;
3691 prog = bpf_iter_get_info(&meta, true);
3692 if (prog)
3693 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3694 }
3695
3696 if (iter->cur_sk < iter->end_sk)
3697 bpf_iter_udp_put_batch(iter);
3698 }
3699
3700 static const struct seq_operations bpf_iter_udp_seq_ops = {
3701 .start = bpf_iter_udp_seq_start,
3702 .next = bpf_iter_udp_seq_next,
3703 .stop = bpf_iter_udp_seq_stop,
3704 .show = bpf_iter_udp_seq_show,
3705 };
3706 #endif
3707
seq_file_family(const struct seq_file * seq)3708 static unsigned short seq_file_family(const struct seq_file *seq)
3709 {
3710 const struct udp_seq_afinfo *afinfo;
3711
3712 #ifdef CONFIG_BPF_SYSCALL
3713 /* BPF iterator: bpf programs to filter sockets. */
3714 if (seq->op == &bpf_iter_udp_seq_ops)
3715 return AF_UNSPEC;
3716 #endif
3717
3718 /* Proc fs iterator */
3719 afinfo = pde_data(file_inode(seq->file));
3720 return afinfo->family;
3721 }
3722
3723 const struct seq_operations udp_seq_ops = {
3724 .start = udp_seq_start,
3725 .next = udp_seq_next,
3726 .stop = udp_seq_stop,
3727 .show = udp4_seq_show,
3728 };
3729 EXPORT_IPV6_MOD(udp_seq_ops);
3730
3731 static struct udp_seq_afinfo udp4_seq_afinfo = {
3732 .family = AF_INET,
3733 .udp_table = NULL,
3734 };
3735
udp4_proc_init_net(struct net * net)3736 static int __net_init udp4_proc_init_net(struct net *net)
3737 {
3738 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3739 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3740 return -ENOMEM;
3741 return 0;
3742 }
3743
udp4_proc_exit_net(struct net * net)3744 static void __net_exit udp4_proc_exit_net(struct net *net)
3745 {
3746 remove_proc_entry("udp", net->proc_net);
3747 }
3748
3749 static struct pernet_operations udp4_net_ops = {
3750 .init = udp4_proc_init_net,
3751 .exit = udp4_proc_exit_net,
3752 };
3753
udp4_proc_init(void)3754 int __init udp4_proc_init(void)
3755 {
3756 return register_pernet_subsys(&udp4_net_ops);
3757 }
3758
udp4_proc_exit(void)3759 void udp4_proc_exit(void)
3760 {
3761 unregister_pernet_subsys(&udp4_net_ops);
3762 }
3763 #endif /* CONFIG_PROC_FS */
3764
3765 static __initdata unsigned long uhash_entries;
set_uhash_entries(char * str)3766 static int __init set_uhash_entries(char *str)
3767 {
3768 ssize_t ret;
3769
3770 if (!str)
3771 return 0;
3772
3773 ret = kstrtoul(str, 0, &uhash_entries);
3774 if (ret)
3775 return 0;
3776
3777 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3778 uhash_entries = UDP_HTABLE_SIZE_MIN;
3779 return 1;
3780 }
3781 __setup("uhash_entries=", set_uhash_entries);
3782
udp_table_init(struct udp_table * table,const char * name)3783 void __init udp_table_init(struct udp_table *table, const char *name)
3784 {
3785 unsigned int i, slot_size;
3786
3787 slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3788 udp_hash4_slot_size();
3789 table->hash = alloc_large_system_hash(name,
3790 slot_size,
3791 uhash_entries,
3792 21, /* one slot per 2 MB */
3793 0,
3794 &table->log,
3795 &table->mask,
3796 UDP_HTABLE_SIZE_MIN,
3797 UDP_HTABLE_SIZE_MAX);
3798
3799 table->hash2 = (void *)(table->hash + (table->mask + 1));
3800 for (i = 0; i <= table->mask; i++) {
3801 INIT_HLIST_HEAD(&table->hash[i].head);
3802 table->hash[i].count = 0;
3803 spin_lock_init(&table->hash[i].lock);
3804 }
3805 for (i = 0; i <= table->mask; i++) {
3806 INIT_HLIST_HEAD(&table->hash2[i].hslot.head);
3807 table->hash2[i].hslot.count = 0;
3808 spin_lock_init(&table->hash2[i].hslot.lock);
3809 }
3810 udp_table_hash4_init(table);
3811 }
3812
udp_flow_hashrnd(void)3813 u32 udp_flow_hashrnd(void)
3814 {
3815 static u32 hashrnd __read_mostly;
3816
3817 net_get_random_once(&hashrnd, sizeof(hashrnd));
3818
3819 return hashrnd;
3820 }
3821 EXPORT_SYMBOL(udp_flow_hashrnd);
3822
udp_sysctl_init(struct net * net)3823 static void __net_init udp_sysctl_init(struct net *net)
3824 {
3825 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3826 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3827
3828 #ifdef CONFIG_NET_L3_MASTER_DEV
3829 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3830 #endif
3831 }
3832
udp_pernet_table_alloc(unsigned int hash_entries)3833 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3834 {
3835 struct udp_table *udptable;
3836 unsigned int slot_size;
3837 int i;
3838
3839 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3840 if (!udptable)
3841 goto out;
3842
3843 slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3844 udp_hash4_slot_size();
3845 udptable->hash = vmalloc_huge(hash_entries * slot_size,
3846 GFP_KERNEL_ACCOUNT);
3847 if (!udptable->hash)
3848 goto free_table;
3849
3850 udptable->hash2 = (void *)(udptable->hash + hash_entries);
3851 udptable->mask = hash_entries - 1;
3852 udptable->log = ilog2(hash_entries);
3853
3854 for (i = 0; i < hash_entries; i++) {
3855 INIT_HLIST_HEAD(&udptable->hash[i].head);
3856 udptable->hash[i].count = 0;
3857 spin_lock_init(&udptable->hash[i].lock);
3858
3859 INIT_HLIST_HEAD(&udptable->hash2[i].hslot.head);
3860 udptable->hash2[i].hslot.count = 0;
3861 spin_lock_init(&udptable->hash2[i].hslot.lock);
3862 }
3863 udp_table_hash4_init(udptable);
3864
3865 return udptable;
3866
3867 free_table:
3868 kfree(udptable);
3869 out:
3870 return NULL;
3871 }
3872
udp_pernet_table_free(struct net * net)3873 static void __net_exit udp_pernet_table_free(struct net *net)
3874 {
3875 struct udp_table *udptable = net->ipv4.udp_table;
3876
3877 if (udptable == &udp_table)
3878 return;
3879
3880 kvfree(udptable->hash);
3881 kfree(udptable);
3882 }
3883
udp_set_table(struct net * net)3884 static void __net_init udp_set_table(struct net *net)
3885 {
3886 struct udp_table *udptable;
3887 unsigned int hash_entries;
3888 struct net *old_net;
3889
3890 if (net_eq(net, &init_net))
3891 goto fallback;
3892
3893 old_net = current->nsproxy->net_ns;
3894 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3895 if (!hash_entries)
3896 goto fallback;
3897
3898 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3899 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3900 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3901 else
3902 hash_entries = roundup_pow_of_two(hash_entries);
3903
3904 udptable = udp_pernet_table_alloc(hash_entries);
3905 if (udptable) {
3906 net->ipv4.udp_table = udptable;
3907 } else {
3908 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3909 "for a netns, fallback to the global one\n",
3910 hash_entries);
3911 fallback:
3912 net->ipv4.udp_table = &udp_table;
3913 }
3914 }
3915
udp_pernet_init(struct net * net)3916 static int __net_init udp_pernet_init(struct net *net)
3917 {
3918 #if IS_ENABLED(CONFIG_NET_UDP_TUNNEL)
3919 int i;
3920
3921 /* No tunnel is configured */
3922 for (i = 0; i < ARRAY_SIZE(net->ipv4.udp_tunnel_gro); ++i) {
3923 INIT_HLIST_HEAD(&net->ipv4.udp_tunnel_gro[i].list);
3924 RCU_INIT_POINTER(net->ipv4.udp_tunnel_gro[i].sk, NULL);
3925 }
3926 #endif
3927 udp_sysctl_init(net);
3928 udp_set_table(net);
3929
3930 return 0;
3931 }
3932
udp_pernet_exit(struct net * net)3933 static void __net_exit udp_pernet_exit(struct net *net)
3934 {
3935 udp_pernet_table_free(net);
3936 }
3937
3938 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3939 .init = udp_pernet_init,
3940 .exit = udp_pernet_exit,
3941 };
3942
3943 #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)3944 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3945 struct udp_sock *udp_sk, uid_t uid, int bucket)
3946
3947 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3948 unsigned int new_batch_sz, gfp_t flags)
3949 {
3950 union bpf_udp_iter_batch_item *new_batch;
3951
3952 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3953 flags | __GFP_NOWARN);
3954 if (!new_batch)
3955 return -ENOMEM;
3956
3957 if (flags != GFP_NOWAIT)
3958 bpf_iter_udp_put_batch(iter);
3959
3960 memcpy(new_batch, iter->batch, sizeof(*iter->batch) * iter->end_sk);
3961 kvfree(iter->batch);
3962 iter->batch = new_batch;
3963 iter->max_sk = new_batch_sz;
3964
3965 return 0;
3966 }
3967
3968 #define INIT_BATCH_SZ 16
3969
bpf_iter_init_udp(void * priv_data,struct bpf_iter_aux_info * aux)3970 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3971 {
3972 struct bpf_udp_iter_state *iter = priv_data;
3973 int ret;
3974
3975 ret = bpf_iter_init_seq_net(priv_data, aux);
3976 if (ret)
3977 return ret;
3978
3979 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ, GFP_USER);
3980 if (ret)
3981 bpf_iter_fini_seq_net(priv_data);
3982
3983 iter->state.bucket = -1;
3984
3985 return ret;
3986 }
3987
bpf_iter_fini_udp(void * priv_data)3988 static void bpf_iter_fini_udp(void *priv_data)
3989 {
3990 struct bpf_udp_iter_state *iter = priv_data;
3991
3992 bpf_iter_fini_seq_net(priv_data);
3993 kvfree(iter->batch);
3994 }
3995
3996 static const struct bpf_iter_seq_info udp_seq_info = {
3997 .seq_ops = &bpf_iter_udp_seq_ops,
3998 .init_seq_private = bpf_iter_init_udp,
3999 .fini_seq_private = bpf_iter_fini_udp,
4000 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
4001 };
4002
4003 static struct bpf_iter_reg udp_reg_info = {
4004 .target = "udp",
4005 .ctx_arg_info_size = 1,
4006 .ctx_arg_info = {
4007 { offsetof(struct bpf_iter__udp, udp_sk),
4008 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
4009 },
4010 .seq_info = &udp_seq_info,
4011 };
4012
bpf_iter_register(void)4013 static void __init bpf_iter_register(void)
4014 {
4015 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
4016 if (bpf_iter_reg_target(&udp_reg_info))
4017 pr_warn("Warning: could not register bpf iterator udp\n");
4018 }
4019 #endif
4020
udp_init(void)4021 void __init udp_init(void)
4022 {
4023 unsigned long limit;
4024
4025 udp_table_init(&udp_table, "UDP");
4026 limit = nr_free_buffer_pages() / 8;
4027 limit = max(limit, 128UL);
4028 sysctl_udp_mem[0] = limit / 4 * 3;
4029 sysctl_udp_mem[1] = limit;
4030 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
4031
4032 if (register_pernet_subsys(&udp_sysctl_ops))
4033 panic("UDP: failed to init sysctl parameters.\n");
4034
4035 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
4036 bpf_iter_register();
4037 #endif
4038 }
4039