1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The User Datagram Protocol (UDP).
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
13 *
14 * Fixes:
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
28 * does NOT close.
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
57 * for connect.
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
64 * datagrams.
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
72 *
73 *
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
78 */
79
80 #define pr_fmt(fmt) "UDP: " fmt
81
82 #include <linux/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/memblock.h>
85 #include <linux/highmem.h>
86 #include <linux/swap.h>
87 #include <linux/types.h>
88 #include <linux/fcntl.h>
89 #include <linux/module.h>
90 #include <linux/socket.h>
91 #include <linux/sockios.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
94 #include <linux/in.h>
95 #include <linux/errno.h>
96 #include <linux/timer.h>
97 #include <linux/mm.h>
98 #include <linux/inet.h>
99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/route.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <trace/events/udp.h>
112 #include <linux/static_key.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
119 struct udp_table udp_table __read_mostly;
120 EXPORT_SYMBOL(udp_table);
121
122 long sysctl_udp_mem[3] __read_mostly;
123 EXPORT_SYMBOL(sysctl_udp_mem);
124
125 atomic_long_t udp_memory_allocated;
126 EXPORT_SYMBOL(udp_memory_allocated);
127
128 #define MAX_UDP_PORTS 65536
129 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
130
131 /* IPCB reference means this can not be used from early demux */
132 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
133 {
134 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
135 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
136 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
137 return true;
138 #endif
139 return false;
140 }
141
142 static int udp_lib_lport_inuse(struct net *net, __u16 num,
143 const struct udp_hslot *hslot,
144 unsigned long *bitmap,
145 struct sock *sk, unsigned int log)
146 {
147 struct sock *sk2;
148 kuid_t uid = sock_i_uid(sk);
149
150 sk_for_each(sk2, &hslot->head) {
151 if (net_eq(sock_net(sk2), net) &&
152 sk2 != sk &&
153 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
154 (!sk2->sk_reuse || !sk->sk_reuse) &&
155 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
156 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
157 inet_rcv_saddr_equal(sk, sk2, true)) {
158 if (sk2->sk_reuseport && sk->sk_reuseport &&
159 !rcu_access_pointer(sk->sk_reuseport_cb) &&
160 uid_eq(uid, sock_i_uid(sk2))) {
161 if (!bitmap)
162 return 0;
163 } else {
164 if (!bitmap)
165 return 1;
166 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
167 bitmap);
168 }
169 }
170 }
171 return 0;
172 }
173
174 /*
175 * Note: we still hold spinlock of primary hash chain, so no other writer
176 * can insert/delete a socket with local_port == num
177 */
178 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
179 struct udp_hslot *hslot2,
180 struct sock *sk)
181 {
182 struct sock *sk2;
183 kuid_t uid = sock_i_uid(sk);
184 int res = 0;
185
186 spin_lock(&hslot2->lock);
187 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
188 if (net_eq(sock_net(sk2), net) &&
189 sk2 != sk &&
190 (udp_sk(sk2)->udp_port_hash == num) &&
191 (!sk2->sk_reuse || !sk->sk_reuse) &&
192 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
193 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
194 inet_rcv_saddr_equal(sk, sk2, true)) {
195 if (sk2->sk_reuseport && sk->sk_reuseport &&
196 !rcu_access_pointer(sk->sk_reuseport_cb) &&
197 uid_eq(uid, sock_i_uid(sk2))) {
198 res = 0;
199 } else {
200 res = 1;
201 }
202 break;
203 }
204 }
205 spin_unlock(&hslot2->lock);
206 return res;
207 }
208
209 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
210 {
211 struct net *net = sock_net(sk);
212 kuid_t uid = sock_i_uid(sk);
213 struct sock *sk2;
214
215 sk_for_each(sk2, &hslot->head) {
216 if (net_eq(sock_net(sk2), net) &&
217 sk2 != sk &&
218 sk2->sk_family == sk->sk_family &&
219 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
220 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
221 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
222 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
223 inet_rcv_saddr_equal(sk, sk2, false)) {
224 return reuseport_add_sock(sk, sk2,
225 inet_rcv_saddr_any(sk));
226 }
227 }
228
229 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
230 }
231
232 /**
233 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
234 *
235 * @sk: socket struct in question
236 * @snum: port number to look up
237 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
238 * with NULL address
239 */
240 int udp_lib_get_port(struct sock *sk, unsigned short snum,
241 unsigned int hash2_nulladdr)
242 {
243 struct udp_hslot *hslot, *hslot2;
244 struct udp_table *udptable = sk->sk_prot->h.udp_table;
245 int error = 1;
246 struct net *net = sock_net(sk);
247
248 if (!snum) {
249 int low, high, remaining;
250 unsigned int rand;
251 unsigned short first, last;
252 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
253
254 inet_get_local_port_range(net, &low, &high);
255 remaining = (high - low) + 1;
256
257 rand = prandom_u32();
258 first = reciprocal_scale(rand, remaining) + low;
259 /*
260 * force rand to be an odd multiple of UDP_HTABLE_SIZE
261 */
262 rand = (rand | 1) * (udptable->mask + 1);
263 last = first + udptable->mask + 1;
264 do {
265 hslot = udp_hashslot(udptable, net, first);
266 bitmap_zero(bitmap, PORTS_PER_CHAIN);
267 spin_lock_bh(&hslot->lock);
268 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
269 udptable->log);
270
271 snum = first;
272 /*
273 * Iterate on all possible values of snum for this hash.
274 * Using steps of an odd multiple of UDP_HTABLE_SIZE
275 * give us randomization and full range coverage.
276 */
277 do {
278 if (low <= snum && snum <= high &&
279 !test_bit(snum >> udptable->log, bitmap) &&
280 !inet_is_local_reserved_port(net, snum))
281 goto found;
282 snum += rand;
283 } while (snum != first);
284 spin_unlock_bh(&hslot->lock);
285 cond_resched();
286 } while (++first != last);
287 goto fail;
288 } else {
289 hslot = udp_hashslot(udptable, net, snum);
290 spin_lock_bh(&hslot->lock);
291 if (hslot->count > 10) {
292 int exist;
293 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
294
295 slot2 &= udptable->mask;
296 hash2_nulladdr &= udptable->mask;
297
298 hslot2 = udp_hashslot2(udptable, slot2);
299 if (hslot->count < hslot2->count)
300 goto scan_primary_hash;
301
302 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
303 if (!exist && (hash2_nulladdr != slot2)) {
304 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
305 exist = udp_lib_lport_inuse2(net, snum, hslot2,
306 sk);
307 }
308 if (exist)
309 goto fail_unlock;
310 else
311 goto found;
312 }
313 scan_primary_hash:
314 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
315 goto fail_unlock;
316 }
317 found:
318 inet_sk(sk)->inet_num = snum;
319 udp_sk(sk)->udp_port_hash = snum;
320 udp_sk(sk)->udp_portaddr_hash ^= snum;
321 if (sk_unhashed(sk)) {
322 if (sk->sk_reuseport &&
323 udp_reuseport_add_sock(sk, hslot)) {
324 inet_sk(sk)->inet_num = 0;
325 udp_sk(sk)->udp_port_hash = 0;
326 udp_sk(sk)->udp_portaddr_hash ^= snum;
327 goto fail_unlock;
328 }
329
330 sk_add_node_rcu(sk, &hslot->head);
331 hslot->count++;
332 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
333
334 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
335 spin_lock(&hslot2->lock);
336 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
337 sk->sk_family == AF_INET6)
338 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
339 &hslot2->head);
340 else
341 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
342 &hslot2->head);
343 hslot2->count++;
344 spin_unlock(&hslot2->lock);
345 }
346 sock_set_flag(sk, SOCK_RCU_FREE);
347 error = 0;
348 fail_unlock:
349 spin_unlock_bh(&hslot->lock);
350 fail:
351 return error;
352 }
353 EXPORT_SYMBOL(udp_lib_get_port);
354
355 int udp_v4_get_port(struct sock *sk, unsigned short snum)
356 {
357 unsigned int hash2_nulladdr =
358 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
359 unsigned int hash2_partial =
360 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
361
362 /* precompute partial secondary hash */
363 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
364 return udp_lib_get_port(sk, snum, hash2_nulladdr);
365 }
366
367 static int compute_score(struct sock *sk, struct net *net,
368 __be32 saddr, __be16 sport,
369 __be32 daddr, unsigned short hnum,
370 int dif, int sdif, bool exact_dif)
371 {
372 int score;
373 struct inet_sock *inet;
374
375 if (!net_eq(sock_net(sk), net) ||
376 udp_sk(sk)->udp_port_hash != hnum ||
377 ipv6_only_sock(sk))
378 return -1;
379
380 score = (sk->sk_family == PF_INET) ? 2 : 1;
381 inet = inet_sk(sk);
382
383 if (inet->inet_rcv_saddr) {
384 if (inet->inet_rcv_saddr != daddr)
385 return -1;
386 score += 4;
387 }
388
389 if (inet->inet_daddr) {
390 if (inet->inet_daddr != saddr)
391 return -1;
392 score += 4;
393 }
394
395 if (inet->inet_dport) {
396 if (inet->inet_dport != sport)
397 return -1;
398 score += 4;
399 }
400
401 if (sk->sk_bound_dev_if || exact_dif) {
402 bool dev_match = (sk->sk_bound_dev_if == dif ||
403 sk->sk_bound_dev_if == sdif);
404
405 if (!dev_match)
406 return -1;
407 if (sk->sk_bound_dev_if)
408 score += 4;
409 }
410
411 if (sk->sk_incoming_cpu == raw_smp_processor_id())
412 score++;
413 return score;
414 }
415
416 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
417 const __u16 lport, const __be32 faddr,
418 const __be16 fport)
419 {
420 static u32 udp_ehash_secret __read_mostly;
421
422 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
423
424 return __inet_ehashfn(laddr, lport, faddr, fport,
425 udp_ehash_secret + net_hash_mix(net));
426 }
427
428 /* called with rcu_read_lock() */
429 static struct sock *udp4_lib_lookup2(struct net *net,
430 __be32 saddr, __be16 sport,
431 __be32 daddr, unsigned int hnum,
432 int dif, int sdif, bool exact_dif,
433 struct udp_hslot *hslot2,
434 struct sk_buff *skb)
435 {
436 struct sock *sk, *result;
437 int score, badness;
438 u32 hash = 0;
439
440 result = NULL;
441 badness = 0;
442 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
443 score = compute_score(sk, net, saddr, sport,
444 daddr, hnum, dif, sdif, exact_dif);
445 if (score > badness) {
446 if (sk->sk_reuseport) {
447 hash = udp_ehashfn(net, daddr, hnum,
448 saddr, sport);
449 result = reuseport_select_sock(sk, hash, skb,
450 sizeof(struct udphdr));
451 if (result)
452 return result;
453 }
454 badness = score;
455 result = sk;
456 }
457 }
458 return result;
459 }
460
461 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
462 * harder than this. -DaveM
463 */
464 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
465 __be16 sport, __be32 daddr, __be16 dport, int dif,
466 int sdif, struct udp_table *udptable, struct sk_buff *skb)
467 {
468 struct sock *sk, *result;
469 unsigned short hnum = ntohs(dport);
470 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
471 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
472 bool exact_dif = udp_lib_exact_dif_match(net, skb);
473 int score, badness;
474 u32 hash = 0;
475
476 if (hslot->count > 10) {
477 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
478 slot2 = hash2 & udptable->mask;
479 hslot2 = &udptable->hash2[slot2];
480 if (hslot->count < hslot2->count)
481 goto begin;
482
483 result = udp4_lib_lookup2(net, saddr, sport,
484 daddr, hnum, dif, sdif,
485 exact_dif, hslot2, skb);
486 if (!result) {
487 unsigned int old_slot2 = slot2;
488 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
489 slot2 = hash2 & udptable->mask;
490 /* avoid searching the same slot again. */
491 if (unlikely(slot2 == old_slot2))
492 return result;
493
494 hslot2 = &udptable->hash2[slot2];
495 if (hslot->count < hslot2->count)
496 goto begin;
497
498 result = udp4_lib_lookup2(net, saddr, sport,
499 daddr, hnum, dif, sdif,
500 exact_dif, hslot2, skb);
501 }
502 if (unlikely(IS_ERR(result)))
503 return NULL;
504 return result;
505 }
506 begin:
507 result = NULL;
508 badness = 0;
509 sk_for_each_rcu(sk, &hslot->head) {
510 score = compute_score(sk, net, saddr, sport,
511 daddr, hnum, dif, sdif, exact_dif);
512 if (score > badness) {
513 if (sk->sk_reuseport) {
514 hash = udp_ehashfn(net, daddr, hnum,
515 saddr, sport);
516 result = reuseport_select_sock(sk, hash, skb,
517 sizeof(struct udphdr));
518 if (unlikely(IS_ERR(result)))
519 return NULL;
520 if (result)
521 return result;
522 }
523 result = sk;
524 badness = score;
525 }
526 }
527 return result;
528 }
529 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
530
531 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
532 __be16 sport, __be16 dport,
533 struct udp_table *udptable)
534 {
535 const struct iphdr *iph = ip_hdr(skb);
536
537 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
538 iph->daddr, dport, inet_iif(skb),
539 inet_sdif(skb), udptable, skb);
540 }
541
542 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
543 __be16 sport, __be16 dport)
544 {
545 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
546 }
547 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
548
549 /* Must be called under rcu_read_lock().
550 * Does increment socket refcount.
551 */
552 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
553 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
554 __be32 daddr, __be16 dport, int dif)
555 {
556 struct sock *sk;
557
558 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
559 dif, 0, &udp_table, NULL);
560 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
561 sk = NULL;
562 return sk;
563 }
564 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
565 #endif
566
567 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
568 __be16 loc_port, __be32 loc_addr,
569 __be16 rmt_port, __be32 rmt_addr,
570 int dif, int sdif, unsigned short hnum)
571 {
572 struct inet_sock *inet = inet_sk(sk);
573
574 if (!net_eq(sock_net(sk), net) ||
575 udp_sk(sk)->udp_port_hash != hnum ||
576 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
577 (inet->inet_dport != rmt_port && inet->inet_dport) ||
578 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
579 ipv6_only_sock(sk) ||
580 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
581 sk->sk_bound_dev_if != sdif))
582 return false;
583 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
584 return false;
585 return true;
586 }
587
588 /*
589 * This routine is called by the ICMP module when it gets some
590 * sort of error condition. If err < 0 then the socket should
591 * be closed and the error returned to the user. If err > 0
592 * it's just the icmp type << 8 | icmp code.
593 * Header points to the ip header of the error packet. We move
594 * on past this. Then (as it used to claim before adjustment)
595 * header points to the first 8 bytes of the udp header. We need
596 * to find the appropriate port.
597 */
598
599 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
600 {
601 struct inet_sock *inet;
602 const struct iphdr *iph = (const struct iphdr *)skb->data;
603 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
604 const int type = icmp_hdr(skb)->type;
605 const int code = icmp_hdr(skb)->code;
606 struct sock *sk;
607 int harderr;
608 int err;
609 struct net *net = dev_net(skb->dev);
610
611 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
612 iph->saddr, uh->source, skb->dev->ifindex,
613 inet_sdif(skb), udptable, NULL);
614 if (!sk) {
615 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
616 return; /* No socket for error */
617 }
618
619 err = 0;
620 harderr = 0;
621 inet = inet_sk(sk);
622
623 switch (type) {
624 default:
625 case ICMP_TIME_EXCEEDED:
626 err = EHOSTUNREACH;
627 break;
628 case ICMP_SOURCE_QUENCH:
629 goto out;
630 case ICMP_PARAMETERPROB:
631 err = EPROTO;
632 harderr = 1;
633 break;
634 case ICMP_DEST_UNREACH:
635 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
636 ipv4_sk_update_pmtu(skb, sk, info);
637 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
638 err = EMSGSIZE;
639 harderr = 1;
640 break;
641 }
642 goto out;
643 }
644 err = EHOSTUNREACH;
645 if (code <= NR_ICMP_UNREACH) {
646 harderr = icmp_err_convert[code].fatal;
647 err = icmp_err_convert[code].errno;
648 }
649 break;
650 case ICMP_REDIRECT:
651 ipv4_sk_redirect(skb, sk);
652 goto out;
653 }
654
655 /*
656 * RFC1122: OK. Passes ICMP errors back to application, as per
657 * 4.1.3.3.
658 */
659 if (!inet->recverr) {
660 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
661 goto out;
662 } else
663 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
664
665 sk->sk_err = err;
666 sk->sk_error_report(sk);
667 out:
668 return;
669 }
670
671 void udp_err(struct sk_buff *skb, u32 info)
672 {
673 __udp4_lib_err(skb, info, &udp_table);
674 }
675
676 /*
677 * Throw away all pending data and cancel the corking. Socket is locked.
678 */
679 void udp_flush_pending_frames(struct sock *sk)
680 {
681 struct udp_sock *up = udp_sk(sk);
682
683 if (up->pending) {
684 up->len = 0;
685 up->pending = 0;
686 ip_flush_pending_frames(sk);
687 }
688 }
689 EXPORT_SYMBOL(udp_flush_pending_frames);
690
691 /**
692 * udp4_hwcsum - handle outgoing HW checksumming
693 * @skb: sk_buff containing the filled-in UDP header
694 * (checksum field must be zeroed out)
695 * @src: source IP address
696 * @dst: destination IP address
697 */
698 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
699 {
700 struct udphdr *uh = udp_hdr(skb);
701 int offset = skb_transport_offset(skb);
702 int len = skb->len - offset;
703 int hlen = len;
704 __wsum csum = 0;
705
706 if (!skb_has_frag_list(skb)) {
707 /*
708 * Only one fragment on the socket.
709 */
710 skb->csum_start = skb_transport_header(skb) - skb->head;
711 skb->csum_offset = offsetof(struct udphdr, check);
712 uh->check = ~csum_tcpudp_magic(src, dst, len,
713 IPPROTO_UDP, 0);
714 } else {
715 struct sk_buff *frags;
716
717 /*
718 * HW-checksum won't work as there are two or more
719 * fragments on the socket so that all csums of sk_buffs
720 * should be together
721 */
722 skb_walk_frags(skb, frags) {
723 csum = csum_add(csum, frags->csum);
724 hlen -= frags->len;
725 }
726
727 csum = skb_checksum(skb, offset, hlen, csum);
728 skb->ip_summed = CHECKSUM_NONE;
729
730 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
731 if (uh->check == 0)
732 uh->check = CSUM_MANGLED_0;
733 }
734 }
735 EXPORT_SYMBOL_GPL(udp4_hwcsum);
736
737 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
738 * for the simple case like when setting the checksum for a UDP tunnel.
739 */
740 void udp_set_csum(bool nocheck, struct sk_buff *skb,
741 __be32 saddr, __be32 daddr, int len)
742 {
743 struct udphdr *uh = udp_hdr(skb);
744
745 if (nocheck) {
746 uh->check = 0;
747 } else if (skb_is_gso(skb)) {
748 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
749 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
750 uh->check = 0;
751 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
752 if (uh->check == 0)
753 uh->check = CSUM_MANGLED_0;
754 } else {
755 skb->ip_summed = CHECKSUM_PARTIAL;
756 skb->csum_start = skb_transport_header(skb) - skb->head;
757 skb->csum_offset = offsetof(struct udphdr, check);
758 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
759 }
760 }
761 EXPORT_SYMBOL(udp_set_csum);
762
763 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
764 struct inet_cork *cork)
765 {
766 struct sock *sk = skb->sk;
767 struct inet_sock *inet = inet_sk(sk);
768 struct udphdr *uh;
769 int err = 0;
770 int is_udplite = IS_UDPLITE(sk);
771 int offset = skb_transport_offset(skb);
772 int len = skb->len - offset;
773 __wsum csum = 0;
774
775 /*
776 * Create a UDP header
777 */
778 uh = udp_hdr(skb);
779 uh->source = inet->inet_sport;
780 uh->dest = fl4->fl4_dport;
781 uh->len = htons(len);
782 uh->check = 0;
783
784 if (cork->gso_size) {
785 const int hlen = skb_network_header_len(skb) +
786 sizeof(struct udphdr);
787
788 if (hlen + cork->gso_size > cork->fragsize) {
789 kfree_skb(skb);
790 return -EINVAL;
791 }
792 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
793 kfree_skb(skb);
794 return -EINVAL;
795 }
796 if (sk->sk_no_check_tx) {
797 kfree_skb(skb);
798 return -EINVAL;
799 }
800 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
801 dst_xfrm(skb_dst(skb))) {
802 kfree_skb(skb);
803 return -EIO;
804 }
805
806 skb_shinfo(skb)->gso_size = cork->gso_size;
807 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
808 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh),
809 cork->gso_size);
810 goto csum_partial;
811 }
812
813 if (is_udplite) /* UDP-Lite */
814 csum = udplite_csum(skb);
815
816 else if (sk->sk_no_check_tx) { /* UDP csum off */
817
818 skb->ip_summed = CHECKSUM_NONE;
819 goto send;
820
821 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
822 csum_partial:
823
824 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
825 goto send;
826
827 } else
828 csum = udp_csum(skb);
829
830 /* add protocol-dependent pseudo-header */
831 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
832 sk->sk_protocol, csum);
833 if (uh->check == 0)
834 uh->check = CSUM_MANGLED_0;
835
836 send:
837 err = ip_send_skb(sock_net(sk), skb);
838 if (err) {
839 if (err == -ENOBUFS && !inet->recverr) {
840 UDP_INC_STATS(sock_net(sk),
841 UDP_MIB_SNDBUFERRORS, is_udplite);
842 err = 0;
843 }
844 } else
845 UDP_INC_STATS(sock_net(sk),
846 UDP_MIB_OUTDATAGRAMS, is_udplite);
847 return err;
848 }
849
850 /*
851 * Push out all pending data as one UDP datagram. Socket is locked.
852 */
853 int udp_push_pending_frames(struct sock *sk)
854 {
855 struct udp_sock *up = udp_sk(sk);
856 struct inet_sock *inet = inet_sk(sk);
857 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
858 struct sk_buff *skb;
859 int err = 0;
860
861 skb = ip_finish_skb(sk, fl4);
862 if (!skb)
863 goto out;
864
865 err = udp_send_skb(skb, fl4, &inet->cork.base);
866
867 out:
868 up->len = 0;
869 up->pending = 0;
870 return err;
871 }
872 EXPORT_SYMBOL(udp_push_pending_frames);
873
874 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
875 {
876 switch (cmsg->cmsg_type) {
877 case UDP_SEGMENT:
878 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
879 return -EINVAL;
880 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
881 return 0;
882 default:
883 return -EINVAL;
884 }
885 }
886
887 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
888 {
889 struct cmsghdr *cmsg;
890 bool need_ip = false;
891 int err;
892
893 for_each_cmsghdr(cmsg, msg) {
894 if (!CMSG_OK(msg, cmsg))
895 return -EINVAL;
896
897 if (cmsg->cmsg_level != SOL_UDP) {
898 need_ip = true;
899 continue;
900 }
901
902 err = __udp_cmsg_send(cmsg, gso_size);
903 if (err)
904 return err;
905 }
906
907 return need_ip;
908 }
909 EXPORT_SYMBOL_GPL(udp_cmsg_send);
910
911 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
912 {
913 struct inet_sock *inet = inet_sk(sk);
914 struct udp_sock *up = udp_sk(sk);
915 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
916 struct flowi4 fl4_stack;
917 struct flowi4 *fl4;
918 int ulen = len;
919 struct ipcm_cookie ipc;
920 struct rtable *rt = NULL;
921 int free = 0;
922 int connected = 0;
923 __be32 daddr, faddr, saddr;
924 __be16 dport;
925 u8 tos;
926 int err, is_udplite = IS_UDPLITE(sk);
927 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
928 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
929 struct sk_buff *skb;
930 struct ip_options_data opt_copy;
931
932 if (len > 0xFFFF)
933 return -EMSGSIZE;
934
935 /*
936 * Check the flags.
937 */
938
939 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
940 return -EOPNOTSUPP;
941
942 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
943
944 fl4 = &inet->cork.fl.u.ip4;
945 if (up->pending) {
946 /*
947 * There are pending frames.
948 * The socket lock must be held while it's corked.
949 */
950 lock_sock(sk);
951 if (likely(up->pending)) {
952 if (unlikely(up->pending != AF_INET)) {
953 release_sock(sk);
954 return -EINVAL;
955 }
956 goto do_append_data;
957 }
958 release_sock(sk);
959 }
960 ulen += sizeof(struct udphdr);
961
962 /*
963 * Get and verify the address.
964 */
965 if (usin) {
966 if (msg->msg_namelen < sizeof(*usin))
967 return -EINVAL;
968 if (usin->sin_family != AF_INET) {
969 if (usin->sin_family != AF_UNSPEC)
970 return -EAFNOSUPPORT;
971 }
972
973 daddr = usin->sin_addr.s_addr;
974 dport = usin->sin_port;
975 if (dport == 0)
976 return -EINVAL;
977 } else {
978 if (sk->sk_state != TCP_ESTABLISHED)
979 return -EDESTADDRREQ;
980 daddr = inet->inet_daddr;
981 dport = inet->inet_dport;
982 /* Open fast path for connected socket.
983 Route will not be used, if at least one option is set.
984 */
985 connected = 1;
986 }
987
988 ipcm_init_sk(&ipc, inet);
989 ipc.gso_size = up->gso_size;
990
991 if (msg->msg_controllen) {
992 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
993 if (err > 0)
994 err = ip_cmsg_send(sk, msg, &ipc,
995 sk->sk_family == AF_INET6);
996 if (unlikely(err < 0)) {
997 kfree(ipc.opt);
998 return err;
999 }
1000 if (ipc.opt)
1001 free = 1;
1002 connected = 0;
1003 }
1004 if (!ipc.opt) {
1005 struct ip_options_rcu *inet_opt;
1006
1007 rcu_read_lock();
1008 inet_opt = rcu_dereference(inet->inet_opt);
1009 if (inet_opt) {
1010 memcpy(&opt_copy, inet_opt,
1011 sizeof(*inet_opt) + inet_opt->opt.optlen);
1012 ipc.opt = &opt_copy.opt;
1013 }
1014 rcu_read_unlock();
1015 }
1016
1017 if (cgroup_bpf_enabled && !connected) {
1018 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1019 (struct sockaddr *)usin, &ipc.addr);
1020 if (err)
1021 goto out_free;
1022 if (usin) {
1023 if (usin->sin_port == 0) {
1024 /* BPF program set invalid port. Reject it. */
1025 err = -EINVAL;
1026 goto out_free;
1027 }
1028 daddr = usin->sin_addr.s_addr;
1029 dport = usin->sin_port;
1030 }
1031 }
1032
1033 saddr = ipc.addr;
1034 ipc.addr = faddr = daddr;
1035
1036 if (ipc.opt && ipc.opt->opt.srr) {
1037 if (!daddr) {
1038 err = -EINVAL;
1039 goto out_free;
1040 }
1041 faddr = ipc.opt->opt.faddr;
1042 connected = 0;
1043 }
1044 tos = get_rttos(&ipc, inet);
1045 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1046 (msg->msg_flags & MSG_DONTROUTE) ||
1047 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1048 tos |= RTO_ONLINK;
1049 connected = 0;
1050 }
1051
1052 if (ipv4_is_multicast(daddr)) {
1053 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1054 ipc.oif = inet->mc_index;
1055 if (!saddr)
1056 saddr = inet->mc_addr;
1057 connected = 0;
1058 } else if (!ipc.oif) {
1059 ipc.oif = inet->uc_index;
1060 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1061 /* oif is set, packet is to local broadcast and
1062 * and uc_index is set. oif is most likely set
1063 * by sk_bound_dev_if. If uc_index != oif check if the
1064 * oif is an L3 master and uc_index is an L3 slave.
1065 * If so, we want to allow the send using the uc_index.
1066 */
1067 if (ipc.oif != inet->uc_index &&
1068 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1069 inet->uc_index)) {
1070 ipc.oif = inet->uc_index;
1071 }
1072 }
1073
1074 if (connected)
1075 rt = (struct rtable *)sk_dst_check(sk, 0);
1076
1077 if (!rt) {
1078 struct net *net = sock_net(sk);
1079 __u8 flow_flags = inet_sk_flowi_flags(sk);
1080
1081 fl4 = &fl4_stack;
1082
1083 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1084 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1085 flow_flags,
1086 faddr, saddr, dport, inet->inet_sport,
1087 sk->sk_uid);
1088
1089 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1090 rt = ip_route_output_flow(net, fl4, sk);
1091 if (IS_ERR(rt)) {
1092 err = PTR_ERR(rt);
1093 rt = NULL;
1094 if (err == -ENETUNREACH)
1095 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1096 goto out;
1097 }
1098
1099 err = -EACCES;
1100 if ((rt->rt_flags & RTCF_BROADCAST) &&
1101 !sock_flag(sk, SOCK_BROADCAST))
1102 goto out;
1103 if (connected)
1104 sk_dst_set(sk, dst_clone(&rt->dst));
1105 }
1106
1107 if (msg->msg_flags&MSG_CONFIRM)
1108 goto do_confirm;
1109 back_from_confirm:
1110
1111 saddr = fl4->saddr;
1112 if (!ipc.addr)
1113 daddr = ipc.addr = fl4->daddr;
1114
1115 /* Lockless fast path for the non-corking case. */
1116 if (!corkreq) {
1117 struct inet_cork cork;
1118
1119 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1120 sizeof(struct udphdr), &ipc, &rt,
1121 &cork, msg->msg_flags);
1122 err = PTR_ERR(skb);
1123 if (!IS_ERR_OR_NULL(skb))
1124 err = udp_send_skb(skb, fl4, &cork);
1125 goto out;
1126 }
1127
1128 lock_sock(sk);
1129 if (unlikely(up->pending)) {
1130 /* The socket is already corked while preparing it. */
1131 /* ... which is an evident application bug. --ANK */
1132 release_sock(sk);
1133
1134 net_dbg_ratelimited("socket already corked\n");
1135 err = -EINVAL;
1136 goto out;
1137 }
1138 /*
1139 * Now cork the socket to pend data.
1140 */
1141 fl4 = &inet->cork.fl.u.ip4;
1142 fl4->daddr = daddr;
1143 fl4->saddr = saddr;
1144 fl4->fl4_dport = dport;
1145 fl4->fl4_sport = inet->inet_sport;
1146 up->pending = AF_INET;
1147
1148 do_append_data:
1149 up->len += ulen;
1150 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1151 sizeof(struct udphdr), &ipc, &rt,
1152 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1153 if (err)
1154 udp_flush_pending_frames(sk);
1155 else if (!corkreq)
1156 err = udp_push_pending_frames(sk);
1157 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1158 up->pending = 0;
1159 release_sock(sk);
1160
1161 out:
1162 ip_rt_put(rt);
1163 out_free:
1164 if (free)
1165 kfree(ipc.opt);
1166 if (!err)
1167 return len;
1168 /*
1169 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1170 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1171 * we don't have a good statistic (IpOutDiscards but it can be too many
1172 * things). We could add another new stat but at least for now that
1173 * seems like overkill.
1174 */
1175 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1176 UDP_INC_STATS(sock_net(sk),
1177 UDP_MIB_SNDBUFERRORS, is_udplite);
1178 }
1179 return err;
1180
1181 do_confirm:
1182 if (msg->msg_flags & MSG_PROBE)
1183 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1184 if (!(msg->msg_flags&MSG_PROBE) || len)
1185 goto back_from_confirm;
1186 err = 0;
1187 goto out;
1188 }
1189 EXPORT_SYMBOL(udp_sendmsg);
1190
1191 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1192 size_t size, int flags)
1193 {
1194 struct inet_sock *inet = inet_sk(sk);
1195 struct udp_sock *up = udp_sk(sk);
1196 int ret;
1197
1198 if (flags & MSG_SENDPAGE_NOTLAST)
1199 flags |= MSG_MORE;
1200
1201 if (!up->pending) {
1202 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1203
1204 /* Call udp_sendmsg to specify destination address which
1205 * sendpage interface can't pass.
1206 * This will succeed only when the socket is connected.
1207 */
1208 ret = udp_sendmsg(sk, &msg, 0);
1209 if (ret < 0)
1210 return ret;
1211 }
1212
1213 lock_sock(sk);
1214
1215 if (unlikely(!up->pending)) {
1216 release_sock(sk);
1217
1218 net_dbg_ratelimited("cork failed\n");
1219 return -EINVAL;
1220 }
1221
1222 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1223 page, offset, size, flags);
1224 if (ret == -EOPNOTSUPP) {
1225 release_sock(sk);
1226 return sock_no_sendpage(sk->sk_socket, page, offset,
1227 size, flags);
1228 }
1229 if (ret < 0) {
1230 udp_flush_pending_frames(sk);
1231 goto out;
1232 }
1233
1234 up->len += size;
1235 if (!(up->corkflag || (flags&MSG_MORE)))
1236 ret = udp_push_pending_frames(sk);
1237 if (!ret)
1238 ret = size;
1239 out:
1240 release_sock(sk);
1241 return ret;
1242 }
1243
1244 #define UDP_SKB_IS_STATELESS 0x80000000
1245
1246 static void udp_set_dev_scratch(struct sk_buff *skb)
1247 {
1248 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1249
1250 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1251 scratch->_tsize_state = skb->truesize;
1252 #if BITS_PER_LONG == 64
1253 scratch->len = skb->len;
1254 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1255 scratch->is_linear = !skb_is_nonlinear(skb);
1256 #endif
1257 /* all head states execept sp (dst, sk, nf) are always cleared by
1258 * udp_rcv() and we need to preserve secpath, if present, to eventually
1259 * process IP_CMSG_PASSSEC at recvmsg() time
1260 */
1261 if (likely(!skb_sec_path(skb)))
1262 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1263 }
1264
1265 static int udp_skb_truesize(struct sk_buff *skb)
1266 {
1267 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1268 }
1269
1270 static bool udp_skb_has_head_state(struct sk_buff *skb)
1271 {
1272 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1273 }
1274
1275 /* fully reclaim rmem/fwd memory allocated for skb */
1276 static void udp_rmem_release(struct sock *sk, int size, int partial,
1277 bool rx_queue_lock_held)
1278 {
1279 struct udp_sock *up = udp_sk(sk);
1280 struct sk_buff_head *sk_queue;
1281 int amt;
1282
1283 if (likely(partial)) {
1284 up->forward_deficit += size;
1285 size = up->forward_deficit;
1286 if (size < (sk->sk_rcvbuf >> 2))
1287 return;
1288 } else {
1289 size += up->forward_deficit;
1290 }
1291 up->forward_deficit = 0;
1292
1293 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1294 * if the called don't held it already
1295 */
1296 sk_queue = &sk->sk_receive_queue;
1297 if (!rx_queue_lock_held)
1298 spin_lock(&sk_queue->lock);
1299
1300
1301 sk->sk_forward_alloc += size;
1302 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1303 sk->sk_forward_alloc -= amt;
1304
1305 if (amt)
1306 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1307
1308 atomic_sub(size, &sk->sk_rmem_alloc);
1309
1310 /* this can save us from acquiring the rx queue lock on next receive */
1311 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1312
1313 if (!rx_queue_lock_held)
1314 spin_unlock(&sk_queue->lock);
1315 }
1316
1317 /* Note: called with reader_queue.lock held.
1318 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1319 * This avoids a cache line miss while receive_queue lock is held.
1320 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1321 */
1322 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1323 {
1324 prefetch(&skb->data);
1325 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1326 }
1327 EXPORT_SYMBOL(udp_skb_destructor);
1328
1329 /* as above, but the caller held the rx queue lock, too */
1330 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1331 {
1332 prefetch(&skb->data);
1333 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1334 }
1335
1336 /* Idea of busylocks is to let producers grab an extra spinlock
1337 * to relieve pressure on the receive_queue spinlock shared by consumer.
1338 * Under flood, this means that only one producer can be in line
1339 * trying to acquire the receive_queue spinlock.
1340 * These busylock can be allocated on a per cpu manner, instead of a
1341 * per socket one (that would consume a cache line per socket)
1342 */
1343 static int udp_busylocks_log __read_mostly;
1344 static spinlock_t *udp_busylocks __read_mostly;
1345
1346 static spinlock_t *busylock_acquire(void *ptr)
1347 {
1348 spinlock_t *busy;
1349
1350 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1351 spin_lock(busy);
1352 return busy;
1353 }
1354
1355 static void busylock_release(spinlock_t *busy)
1356 {
1357 if (busy)
1358 spin_unlock(busy);
1359 }
1360
1361 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1362 {
1363 struct sk_buff_head *list = &sk->sk_receive_queue;
1364 int rmem, delta, amt, err = -ENOMEM;
1365 spinlock_t *busy = NULL;
1366 int size;
1367
1368 /* try to avoid the costly atomic add/sub pair when the receive
1369 * queue is full; always allow at least a packet
1370 */
1371 rmem = atomic_read(&sk->sk_rmem_alloc);
1372 if (rmem > sk->sk_rcvbuf)
1373 goto drop;
1374
1375 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1376 * having linear skbs :
1377 * - Reduce memory overhead and thus increase receive queue capacity
1378 * - Less cache line misses at copyout() time
1379 * - Less work at consume_skb() (less alien page frag freeing)
1380 */
1381 if (rmem > (sk->sk_rcvbuf >> 1)) {
1382 skb_condense(skb);
1383
1384 busy = busylock_acquire(sk);
1385 }
1386 size = skb->truesize;
1387 udp_set_dev_scratch(skb);
1388
1389 /* we drop only if the receive buf is full and the receive
1390 * queue contains some other skb
1391 */
1392 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1393 if (rmem > (size + sk->sk_rcvbuf))
1394 goto uncharge_drop;
1395
1396 spin_lock(&list->lock);
1397 if (size >= sk->sk_forward_alloc) {
1398 amt = sk_mem_pages(size);
1399 delta = amt << SK_MEM_QUANTUM_SHIFT;
1400 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1401 err = -ENOBUFS;
1402 spin_unlock(&list->lock);
1403 goto uncharge_drop;
1404 }
1405
1406 sk->sk_forward_alloc += delta;
1407 }
1408
1409 sk->sk_forward_alloc -= size;
1410
1411 /* no need to setup a destructor, we will explicitly release the
1412 * forward allocated memory on dequeue
1413 */
1414 sock_skb_set_dropcount(sk, skb);
1415
1416 __skb_queue_tail(list, skb);
1417 spin_unlock(&list->lock);
1418
1419 if (!sock_flag(sk, SOCK_DEAD))
1420 sk->sk_data_ready(sk);
1421
1422 busylock_release(busy);
1423 return 0;
1424
1425 uncharge_drop:
1426 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1427
1428 drop:
1429 atomic_inc(&sk->sk_drops);
1430 busylock_release(busy);
1431 return err;
1432 }
1433 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1434
1435 void udp_destruct_sock(struct sock *sk)
1436 {
1437 /* reclaim completely the forward allocated memory */
1438 struct udp_sock *up = udp_sk(sk);
1439 unsigned int total = 0;
1440 struct sk_buff *skb;
1441
1442 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1443 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1444 total += skb->truesize;
1445 kfree_skb(skb);
1446 }
1447 udp_rmem_release(sk, total, 0, true);
1448
1449 inet_sock_destruct(sk);
1450 }
1451 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1452
1453 int udp_init_sock(struct sock *sk)
1454 {
1455 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1456 sk->sk_destruct = udp_destruct_sock;
1457 return 0;
1458 }
1459 EXPORT_SYMBOL_GPL(udp_init_sock);
1460
1461 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1462 {
1463 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1464 bool slow = lock_sock_fast(sk);
1465
1466 sk_peek_offset_bwd(sk, len);
1467 unlock_sock_fast(sk, slow);
1468 }
1469
1470 if (!skb_unref(skb))
1471 return;
1472
1473 /* In the more common cases we cleared the head states previously,
1474 * see __udp_queue_rcv_skb().
1475 */
1476 if (unlikely(udp_skb_has_head_state(skb)))
1477 skb_release_head_state(skb);
1478 __consume_stateless_skb(skb);
1479 }
1480 EXPORT_SYMBOL_GPL(skb_consume_udp);
1481
1482 static struct sk_buff *__first_packet_length(struct sock *sk,
1483 struct sk_buff_head *rcvq,
1484 int *total)
1485 {
1486 struct sk_buff *skb;
1487
1488 while ((skb = skb_peek(rcvq)) != NULL) {
1489 if (udp_lib_checksum_complete(skb)) {
1490 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1491 IS_UDPLITE(sk));
1492 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1493 IS_UDPLITE(sk));
1494 atomic_inc(&sk->sk_drops);
1495 __skb_unlink(skb, rcvq);
1496 *total += skb->truesize;
1497 kfree_skb(skb);
1498 } else {
1499 /* the csum related bits could be changed, refresh
1500 * the scratch area
1501 */
1502 udp_set_dev_scratch(skb);
1503 break;
1504 }
1505 }
1506 return skb;
1507 }
1508
1509 /**
1510 * first_packet_length - return length of first packet in receive queue
1511 * @sk: socket
1512 *
1513 * Drops all bad checksum frames, until a valid one is found.
1514 * Returns the length of found skb, or -1 if none is found.
1515 */
1516 static int first_packet_length(struct sock *sk)
1517 {
1518 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1519 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1520 struct sk_buff *skb;
1521 int total = 0;
1522 int res;
1523
1524 spin_lock_bh(&rcvq->lock);
1525 skb = __first_packet_length(sk, rcvq, &total);
1526 if (!skb && !skb_queue_empty(sk_queue)) {
1527 spin_lock(&sk_queue->lock);
1528 skb_queue_splice_tail_init(sk_queue, rcvq);
1529 spin_unlock(&sk_queue->lock);
1530
1531 skb = __first_packet_length(sk, rcvq, &total);
1532 }
1533 res = skb ? skb->len : -1;
1534 if (total)
1535 udp_rmem_release(sk, total, 1, false);
1536 spin_unlock_bh(&rcvq->lock);
1537 return res;
1538 }
1539
1540 /*
1541 * IOCTL requests applicable to the UDP protocol
1542 */
1543
1544 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1545 {
1546 switch (cmd) {
1547 case SIOCOUTQ:
1548 {
1549 int amount = sk_wmem_alloc_get(sk);
1550
1551 return put_user(amount, (int __user *)arg);
1552 }
1553
1554 case SIOCINQ:
1555 {
1556 int amount = max_t(int, 0, first_packet_length(sk));
1557
1558 return put_user(amount, (int __user *)arg);
1559 }
1560
1561 default:
1562 return -ENOIOCTLCMD;
1563 }
1564
1565 return 0;
1566 }
1567 EXPORT_SYMBOL(udp_ioctl);
1568
1569 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1570 int noblock, int *peeked, int *off, int *err)
1571 {
1572 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1573 struct sk_buff_head *queue;
1574 struct sk_buff *last;
1575 long timeo;
1576 int error;
1577
1578 queue = &udp_sk(sk)->reader_queue;
1579 flags |= noblock ? MSG_DONTWAIT : 0;
1580 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1581 do {
1582 struct sk_buff *skb;
1583
1584 error = sock_error(sk);
1585 if (error)
1586 break;
1587
1588 error = -EAGAIN;
1589 *peeked = 0;
1590 do {
1591 spin_lock_bh(&queue->lock);
1592 skb = __skb_try_recv_from_queue(sk, queue, flags,
1593 udp_skb_destructor,
1594 peeked, off, err,
1595 &last);
1596 if (skb) {
1597 spin_unlock_bh(&queue->lock);
1598 return skb;
1599 }
1600
1601 if (skb_queue_empty(sk_queue)) {
1602 spin_unlock_bh(&queue->lock);
1603 goto busy_check;
1604 }
1605
1606 /* refill the reader queue and walk it again
1607 * keep both queues locked to avoid re-acquiring
1608 * the sk_receive_queue lock if fwd memory scheduling
1609 * is needed.
1610 */
1611 spin_lock(&sk_queue->lock);
1612 skb_queue_splice_tail_init(sk_queue, queue);
1613
1614 skb = __skb_try_recv_from_queue(sk, queue, flags,
1615 udp_skb_dtor_locked,
1616 peeked, off, err,
1617 &last);
1618 spin_unlock(&sk_queue->lock);
1619 spin_unlock_bh(&queue->lock);
1620 if (skb)
1621 return skb;
1622
1623 busy_check:
1624 if (!sk_can_busy_loop(sk))
1625 break;
1626
1627 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1628 } while (!skb_queue_empty(sk_queue));
1629
1630 /* sk_queue is empty, reader_queue may contain peeked packets */
1631 } while (timeo &&
1632 !__skb_wait_for_more_packets(sk, &error, &timeo,
1633 (struct sk_buff *)sk_queue));
1634
1635 *err = error;
1636 return NULL;
1637 }
1638 EXPORT_SYMBOL(__skb_recv_udp);
1639
1640 /*
1641 * This should be easy, if there is something there we
1642 * return it, otherwise we block.
1643 */
1644
1645 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1646 int flags, int *addr_len)
1647 {
1648 struct inet_sock *inet = inet_sk(sk);
1649 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1650 struct sk_buff *skb;
1651 unsigned int ulen, copied;
1652 int peeked, peeking, off;
1653 int err;
1654 int is_udplite = IS_UDPLITE(sk);
1655 bool checksum_valid = false;
1656
1657 if (flags & MSG_ERRQUEUE)
1658 return ip_recv_error(sk, msg, len, addr_len);
1659
1660 try_again:
1661 peeking = flags & MSG_PEEK;
1662 off = sk_peek_offset(sk, flags);
1663 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1664 if (!skb)
1665 return err;
1666 if (ccs_socket_post_recvmsg_permission(sk, skb, flags))
1667 return -EAGAIN; /* Hope less harmful than -EPERM. */
1668
1669 ulen = udp_skb_len(skb);
1670 copied = len;
1671 if (copied > ulen - off)
1672 copied = ulen - off;
1673 else if (copied < ulen)
1674 msg->msg_flags |= MSG_TRUNC;
1675
1676 /*
1677 * If checksum is needed at all, try to do it while copying the
1678 * data. If the data is truncated, or if we only want a partial
1679 * coverage checksum (UDP-Lite), do it before the copy.
1680 */
1681
1682 if (copied < ulen || peeking ||
1683 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1684 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1685 !__udp_lib_checksum_complete(skb);
1686 if (!checksum_valid)
1687 goto csum_copy_err;
1688 }
1689
1690 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1691 if (udp_skb_is_linear(skb))
1692 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1693 else
1694 err = skb_copy_datagram_msg(skb, off, msg, copied);
1695 } else {
1696 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1697
1698 if (err == -EINVAL)
1699 goto csum_copy_err;
1700 }
1701
1702 if (unlikely(err)) {
1703 if (!peeked) {
1704 atomic_inc(&sk->sk_drops);
1705 UDP_INC_STATS(sock_net(sk),
1706 UDP_MIB_INERRORS, is_udplite);
1707 }
1708 kfree_skb(skb);
1709 return err;
1710 }
1711
1712 if (!peeked)
1713 UDP_INC_STATS(sock_net(sk),
1714 UDP_MIB_INDATAGRAMS, is_udplite);
1715
1716 sock_recv_ts_and_drops(msg, sk, skb);
1717
1718 /* Copy the address. */
1719 if (sin) {
1720 sin->sin_family = AF_INET;
1721 sin->sin_port = udp_hdr(skb)->source;
1722 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1723 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1724 *addr_len = sizeof(*sin);
1725 }
1726 if (inet->cmsg_flags)
1727 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1728
1729 err = copied;
1730 if (flags & MSG_TRUNC)
1731 err = ulen;
1732
1733 skb_consume_udp(sk, skb, peeking ? -err : err);
1734 return err;
1735
1736 csum_copy_err:
1737 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1738 udp_skb_destructor)) {
1739 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1740 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1741 }
1742 kfree_skb(skb);
1743
1744 /* starting over for a new packet, but check if we need to yield */
1745 cond_resched();
1746 msg->msg_flags &= ~MSG_TRUNC;
1747 goto try_again;
1748 }
1749
1750 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1751 {
1752 /* This check is replicated from __ip4_datagram_connect() and
1753 * intended to prevent BPF program called below from accessing bytes
1754 * that are out of the bound specified by user in addr_len.
1755 */
1756 if (addr_len < sizeof(struct sockaddr_in))
1757 return -EINVAL;
1758
1759 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1760 }
1761 EXPORT_SYMBOL(udp_pre_connect);
1762
1763 int __udp_disconnect(struct sock *sk, int flags)
1764 {
1765 struct inet_sock *inet = inet_sk(sk);
1766 /*
1767 * 1003.1g - break association.
1768 */
1769
1770 sk->sk_state = TCP_CLOSE;
1771 inet->inet_daddr = 0;
1772 inet->inet_dport = 0;
1773 sock_rps_reset_rxhash(sk);
1774 sk->sk_bound_dev_if = 0;
1775 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1776 inet_reset_saddr(sk);
1777
1778 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1779 sk->sk_prot->unhash(sk);
1780 inet->inet_sport = 0;
1781 }
1782 sk_dst_reset(sk);
1783 return 0;
1784 }
1785 EXPORT_SYMBOL(__udp_disconnect);
1786
1787 int udp_disconnect(struct sock *sk, int flags)
1788 {
1789 lock_sock(sk);
1790 __udp_disconnect(sk, flags);
1791 release_sock(sk);
1792 return 0;
1793 }
1794 EXPORT_SYMBOL(udp_disconnect);
1795
1796 void udp_lib_unhash(struct sock *sk)
1797 {
1798 if (sk_hashed(sk)) {
1799 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1800 struct udp_hslot *hslot, *hslot2;
1801
1802 hslot = udp_hashslot(udptable, sock_net(sk),
1803 udp_sk(sk)->udp_port_hash);
1804 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1805
1806 spin_lock_bh(&hslot->lock);
1807 if (rcu_access_pointer(sk->sk_reuseport_cb))
1808 reuseport_detach_sock(sk);
1809 if (sk_del_node_init_rcu(sk)) {
1810 hslot->count--;
1811 inet_sk(sk)->inet_num = 0;
1812 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1813
1814 spin_lock(&hslot2->lock);
1815 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1816 hslot2->count--;
1817 spin_unlock(&hslot2->lock);
1818 }
1819 spin_unlock_bh(&hslot->lock);
1820 }
1821 }
1822 EXPORT_SYMBOL(udp_lib_unhash);
1823
1824 /*
1825 * inet_rcv_saddr was changed, we must rehash secondary hash
1826 */
1827 void udp_lib_rehash(struct sock *sk, u16 newhash)
1828 {
1829 if (sk_hashed(sk)) {
1830 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1831 struct udp_hslot *hslot, *hslot2, *nhslot2;
1832
1833 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1834 nhslot2 = udp_hashslot2(udptable, newhash);
1835 udp_sk(sk)->udp_portaddr_hash = newhash;
1836
1837 if (hslot2 != nhslot2 ||
1838 rcu_access_pointer(sk->sk_reuseport_cb)) {
1839 hslot = udp_hashslot(udptable, sock_net(sk),
1840 udp_sk(sk)->udp_port_hash);
1841 /* we must lock primary chain too */
1842 spin_lock_bh(&hslot->lock);
1843 if (rcu_access_pointer(sk->sk_reuseport_cb))
1844 reuseport_detach_sock(sk);
1845
1846 if (hslot2 != nhslot2) {
1847 spin_lock(&hslot2->lock);
1848 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1849 hslot2->count--;
1850 spin_unlock(&hslot2->lock);
1851
1852 spin_lock(&nhslot2->lock);
1853 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1854 &nhslot2->head);
1855 nhslot2->count++;
1856 spin_unlock(&nhslot2->lock);
1857 }
1858
1859 spin_unlock_bh(&hslot->lock);
1860 }
1861 }
1862 }
1863 EXPORT_SYMBOL(udp_lib_rehash);
1864
1865 static void udp_v4_rehash(struct sock *sk)
1866 {
1867 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1868 inet_sk(sk)->inet_rcv_saddr,
1869 inet_sk(sk)->inet_num);
1870 udp_lib_rehash(sk, new_hash);
1871 }
1872
1873 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1874 {
1875 int rc;
1876
1877 if (inet_sk(sk)->inet_daddr) {
1878 sock_rps_save_rxhash(sk, skb);
1879 sk_mark_napi_id(sk, skb);
1880 sk_incoming_cpu_update(sk);
1881 } else {
1882 sk_mark_napi_id_once(sk, skb);
1883 }
1884
1885 rc = __udp_enqueue_schedule_skb(sk, skb);
1886 if (rc < 0) {
1887 int is_udplite = IS_UDPLITE(sk);
1888
1889 /* Note that an ENOMEM error is charged twice */
1890 if (rc == -ENOMEM)
1891 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1892 is_udplite);
1893 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1894 kfree_skb(skb);
1895 trace_udp_fail_queue_rcv_skb(rc, sk);
1896 return -1;
1897 }
1898
1899 return 0;
1900 }
1901
1902 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
1903 void udp_encap_enable(void)
1904 {
1905 static_branch_enable(&udp_encap_needed_key);
1906 }
1907 EXPORT_SYMBOL(udp_encap_enable);
1908
1909 /* returns:
1910 * -1: error
1911 * 0: success
1912 * >0: "udp encap" protocol resubmission
1913 *
1914 * Note that in the success and error cases, the skb is assumed to
1915 * have either been requeued or freed.
1916 */
1917 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1918 {
1919 struct udp_sock *up = udp_sk(sk);
1920 int is_udplite = IS_UDPLITE(sk);
1921
1922 /*
1923 * Charge it to the socket, dropping if the queue is full.
1924 */
1925 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1926 goto drop;
1927 nf_reset(skb);
1928
1929 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
1930 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1931
1932 /*
1933 * This is an encapsulation socket so pass the skb to
1934 * the socket's udp_encap_rcv() hook. Otherwise, just
1935 * fall through and pass this up the UDP socket.
1936 * up->encap_rcv() returns the following value:
1937 * =0 if skb was successfully passed to the encap
1938 * handler or was discarded by it.
1939 * >0 if skb should be passed on to UDP.
1940 * <0 if skb should be resubmitted as proto -N
1941 */
1942
1943 /* if we're overly short, let UDP handle it */
1944 encap_rcv = READ_ONCE(up->encap_rcv);
1945 if (encap_rcv) {
1946 int ret;
1947
1948 /* Verify checksum before giving to encap */
1949 if (udp_lib_checksum_complete(skb))
1950 goto csum_error;
1951
1952 ret = encap_rcv(sk, skb);
1953 if (ret <= 0) {
1954 __UDP_INC_STATS(sock_net(sk),
1955 UDP_MIB_INDATAGRAMS,
1956 is_udplite);
1957 return -ret;
1958 }
1959 }
1960
1961 /* FALLTHROUGH -- it's a UDP Packet */
1962 }
1963
1964 /*
1965 * UDP-Lite specific tests, ignored on UDP sockets
1966 */
1967 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1968
1969 /*
1970 * MIB statistics other than incrementing the error count are
1971 * disabled for the following two types of errors: these depend
1972 * on the application settings, not on the functioning of the
1973 * protocol stack as such.
1974 *
1975 * RFC 3828 here recommends (sec 3.3): "There should also be a
1976 * way ... to ... at least let the receiving application block
1977 * delivery of packets with coverage values less than a value
1978 * provided by the application."
1979 */
1980 if (up->pcrlen == 0) { /* full coverage was set */
1981 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1982 UDP_SKB_CB(skb)->cscov, skb->len);
1983 goto drop;
1984 }
1985 /* The next case involves violating the min. coverage requested
1986 * by the receiver. This is subtle: if receiver wants x and x is
1987 * greater than the buffersize/MTU then receiver will complain
1988 * that it wants x while sender emits packets of smaller size y.
1989 * Therefore the above ...()->partial_cov statement is essential.
1990 */
1991 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1992 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1993 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1994 goto drop;
1995 }
1996 }
1997
1998 prefetch(&sk->sk_rmem_alloc);
1999 if (rcu_access_pointer(sk->sk_filter) &&
2000 udp_lib_checksum_complete(skb))
2001 goto csum_error;
2002
2003 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2004 goto drop;
2005
2006 udp_csum_pull_header(skb);
2007
2008 ipv4_pktinfo_prepare(sk, skb);
2009 return __udp_queue_rcv_skb(sk, skb);
2010
2011 csum_error:
2012 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2013 drop:
2014 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2015 atomic_inc(&sk->sk_drops);
2016 kfree_skb(skb);
2017 return -1;
2018 }
2019
2020 /* For TCP sockets, sk_rx_dst is protected by socket lock
2021 * For UDP, we use xchg() to guard against concurrent changes.
2022 */
2023 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2024 {
2025 struct dst_entry *old;
2026
2027 if (dst_hold_safe(dst)) {
2028 old = xchg(&sk->sk_rx_dst, dst);
2029 dst_release(old);
2030 return old != dst;
2031 }
2032 return false;
2033 }
2034 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2035
2036 /*
2037 * Multicasts and broadcasts go to each listener.
2038 *
2039 * Note: called only from the BH handler context.
2040 */
2041 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2042 struct udphdr *uh,
2043 __be32 saddr, __be32 daddr,
2044 struct udp_table *udptable,
2045 int proto)
2046 {
2047 struct sock *sk, *first = NULL;
2048 unsigned short hnum = ntohs(uh->dest);
2049 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2050 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2051 unsigned int offset = offsetof(typeof(*sk), sk_node);
2052 int dif = skb->dev->ifindex;
2053 int sdif = inet_sdif(skb);
2054 struct hlist_node *node;
2055 struct sk_buff *nskb;
2056
2057 if (use_hash2) {
2058 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2059 udptable->mask;
2060 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2061 start_lookup:
2062 hslot = &udptable->hash2[hash2];
2063 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2064 }
2065
2066 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2067 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2068 uh->source, saddr, dif, sdif, hnum))
2069 continue;
2070
2071 if (!first) {
2072 first = sk;
2073 continue;
2074 }
2075 nskb = skb_clone(skb, GFP_ATOMIC);
2076
2077 if (unlikely(!nskb)) {
2078 atomic_inc(&sk->sk_drops);
2079 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2080 IS_UDPLITE(sk));
2081 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2082 IS_UDPLITE(sk));
2083 continue;
2084 }
2085 if (udp_queue_rcv_skb(sk, nskb) > 0)
2086 consume_skb(nskb);
2087 }
2088
2089 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2090 if (use_hash2 && hash2 != hash2_any) {
2091 hash2 = hash2_any;
2092 goto start_lookup;
2093 }
2094
2095 if (first) {
2096 if (udp_queue_rcv_skb(first, skb) > 0)
2097 consume_skb(skb);
2098 } else {
2099 kfree_skb(skb);
2100 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2101 proto == IPPROTO_UDPLITE);
2102 }
2103 return 0;
2104 }
2105
2106 /* Initialize UDP checksum. If exited with zero value (success),
2107 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2108 * Otherwise, csum completion requires chacksumming packet body,
2109 * including udp header and folding it to skb->csum.
2110 */
2111 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2112 int proto)
2113 {
2114 int err;
2115
2116 UDP_SKB_CB(skb)->partial_cov = 0;
2117 UDP_SKB_CB(skb)->cscov = skb->len;
2118
2119 if (proto == IPPROTO_UDPLITE) {
2120 err = udplite_checksum_init(skb, uh);
2121 if (err)
2122 return err;
2123
2124 if (UDP_SKB_CB(skb)->partial_cov) {
2125 skb->csum = inet_compute_pseudo(skb, proto);
2126 return 0;
2127 }
2128 }
2129
2130 /* Note, we are only interested in != 0 or == 0, thus the
2131 * force to int.
2132 */
2133 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2134 inet_compute_pseudo);
2135 if (err)
2136 return err;
2137
2138 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2139 /* If SW calculated the value, we know it's bad */
2140 if (skb->csum_complete_sw)
2141 return 1;
2142
2143 /* HW says the value is bad. Let's validate that.
2144 * skb->csum is no longer the full packet checksum,
2145 * so don't treat it as such.
2146 */
2147 skb_checksum_complete_unset(skb);
2148 }
2149
2150 return 0;
2151 }
2152
2153 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2154 * return code conversion for ip layer consumption
2155 */
2156 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2157 struct udphdr *uh)
2158 {
2159 int ret;
2160
2161 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2162 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2163 inet_compute_pseudo);
2164
2165 ret = udp_queue_rcv_skb(sk, skb);
2166
2167 /* a return value > 0 means to resubmit the input, but
2168 * it wants the return to be -protocol, or 0
2169 */
2170 if (ret > 0)
2171 return -ret;
2172 return 0;
2173 }
2174
2175 /*
2176 * All we need to do is get the socket, and then do a checksum.
2177 */
2178
2179 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2180 int proto)
2181 {
2182 struct sock *sk;
2183 struct udphdr *uh;
2184 unsigned short ulen;
2185 struct rtable *rt = skb_rtable(skb);
2186 __be32 saddr, daddr;
2187 struct net *net = dev_net(skb->dev);
2188
2189 /*
2190 * Validate the packet.
2191 */
2192 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2193 goto drop; /* No space for header. */
2194
2195 uh = udp_hdr(skb);
2196 ulen = ntohs(uh->len);
2197 saddr = ip_hdr(skb)->saddr;
2198 daddr = ip_hdr(skb)->daddr;
2199
2200 if (ulen > skb->len)
2201 goto short_packet;
2202
2203 if (proto == IPPROTO_UDP) {
2204 /* UDP validates ulen. */
2205 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2206 goto short_packet;
2207 uh = udp_hdr(skb);
2208 }
2209
2210 if (udp4_csum_init(skb, uh, proto))
2211 goto csum_error;
2212
2213 sk = skb_steal_sock(skb);
2214 if (sk) {
2215 struct dst_entry *dst = skb_dst(skb);
2216 int ret;
2217
2218 if (unlikely(sk->sk_rx_dst != dst))
2219 udp_sk_rx_dst_set(sk, dst);
2220
2221 ret = udp_unicast_rcv_skb(sk, skb, uh);
2222 sock_put(sk);
2223 return ret;
2224 }
2225
2226 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2227 return __udp4_lib_mcast_deliver(net, skb, uh,
2228 saddr, daddr, udptable, proto);
2229
2230 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2231 if (sk)
2232 return udp_unicast_rcv_skb(sk, skb, uh);
2233
2234 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2235 goto drop;
2236 nf_reset(skb);
2237
2238 /* No socket. Drop packet silently, if checksum is wrong */
2239 if (udp_lib_checksum_complete(skb))
2240 goto csum_error;
2241
2242 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2243 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2244
2245 /*
2246 * Hmm. We got an UDP packet to a port to which we
2247 * don't wanna listen. Ignore it.
2248 */
2249 kfree_skb(skb);
2250 return 0;
2251
2252 short_packet:
2253 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2254 proto == IPPROTO_UDPLITE ? "Lite" : "",
2255 &saddr, ntohs(uh->source),
2256 ulen, skb->len,
2257 &daddr, ntohs(uh->dest));
2258 goto drop;
2259
2260 csum_error:
2261 /*
2262 * RFC1122: OK. Discards the bad packet silently (as far as
2263 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2264 */
2265 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2266 proto == IPPROTO_UDPLITE ? "Lite" : "",
2267 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2268 ulen);
2269 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2270 drop:
2271 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2272 kfree_skb(skb);
2273 return 0;
2274 }
2275
2276 /* We can only early demux multicast if there is a single matching socket.
2277 * If more than one socket found returns NULL
2278 */
2279 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2280 __be16 loc_port, __be32 loc_addr,
2281 __be16 rmt_port, __be32 rmt_addr,
2282 int dif, int sdif)
2283 {
2284 struct sock *sk, *result;
2285 unsigned short hnum = ntohs(loc_port);
2286 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2287 struct udp_hslot *hslot = &udp_table.hash[slot];
2288
2289 /* Do not bother scanning a too big list */
2290 if (hslot->count > 10)
2291 return NULL;
2292
2293 result = NULL;
2294 sk_for_each_rcu(sk, &hslot->head) {
2295 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2296 rmt_port, rmt_addr, dif, sdif, hnum)) {
2297 if (result)
2298 return NULL;
2299 result = sk;
2300 }
2301 }
2302
2303 return result;
2304 }
2305
2306 /* For unicast we should only early demux connected sockets or we can
2307 * break forwarding setups. The chains here can be long so only check
2308 * if the first socket is an exact match and if not move on.
2309 */
2310 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2311 __be16 loc_port, __be32 loc_addr,
2312 __be16 rmt_port, __be32 rmt_addr,
2313 int dif, int sdif)
2314 {
2315 unsigned short hnum = ntohs(loc_port);
2316 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2317 unsigned int slot2 = hash2 & udp_table.mask;
2318 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2319 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2320 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2321 struct sock *sk;
2322
2323 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2324 if (INET_MATCH(sk, net, acookie, rmt_addr,
2325 loc_addr, ports, dif, sdif))
2326 return sk;
2327 /* Only check first socket in chain */
2328 break;
2329 }
2330 return NULL;
2331 }
2332
2333 int udp_v4_early_demux(struct sk_buff *skb)
2334 {
2335 struct net *net = dev_net(skb->dev);
2336 struct in_device *in_dev = NULL;
2337 const struct iphdr *iph;
2338 const struct udphdr *uh;
2339 struct sock *sk = NULL;
2340 struct dst_entry *dst;
2341 int dif = skb->dev->ifindex;
2342 int sdif = inet_sdif(skb);
2343 int ours;
2344
2345 /* validate the packet */
2346 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2347 return 0;
2348
2349 iph = ip_hdr(skb);
2350 uh = udp_hdr(skb);
2351
2352 if (skb->pkt_type == PACKET_MULTICAST) {
2353 in_dev = __in_dev_get_rcu(skb->dev);
2354
2355 if (!in_dev)
2356 return 0;
2357
2358 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2359 iph->protocol);
2360 if (!ours)
2361 return 0;
2362
2363 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2364 uh->source, iph->saddr,
2365 dif, sdif);
2366 } else if (skb->pkt_type == PACKET_HOST) {
2367 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2368 uh->source, iph->saddr, dif, sdif);
2369 }
2370
2371 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2372 return 0;
2373
2374 skb->sk = sk;
2375 skb->destructor = sock_efree;
2376 dst = READ_ONCE(sk->sk_rx_dst);
2377
2378 if (dst)
2379 dst = dst_check(dst, 0);
2380 if (dst) {
2381 u32 itag = 0;
2382
2383 /* set noref for now.
2384 * any place which wants to hold dst has to call
2385 * dst_hold_safe()
2386 */
2387 skb_dst_set_noref(skb, dst);
2388
2389 /* for unconnected multicast sockets we need to validate
2390 * the source on each packet
2391 */
2392 if (!inet_sk(sk)->inet_daddr && in_dev)
2393 return ip_mc_validate_source(skb, iph->daddr,
2394 iph->saddr, iph->tos,
2395 skb->dev, in_dev, &itag);
2396 }
2397 return 0;
2398 }
2399
2400 int udp_rcv(struct sk_buff *skb)
2401 {
2402 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2403 }
2404
2405 void udp_destroy_sock(struct sock *sk)
2406 {
2407 struct udp_sock *up = udp_sk(sk);
2408 bool slow = lock_sock_fast(sk);
2409 udp_flush_pending_frames(sk);
2410 unlock_sock_fast(sk, slow);
2411 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2412 void (*encap_destroy)(struct sock *sk);
2413 encap_destroy = READ_ONCE(up->encap_destroy);
2414 if (encap_destroy)
2415 encap_destroy(sk);
2416 }
2417 }
2418
2419 /*
2420 * Socket option code for UDP
2421 */
2422 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2423 char __user *optval, unsigned int optlen,
2424 int (*push_pending_frames)(struct sock *))
2425 {
2426 struct udp_sock *up = udp_sk(sk);
2427 int val, valbool;
2428 int err = 0;
2429 int is_udplite = IS_UDPLITE(sk);
2430
2431 if (optlen < sizeof(int))
2432 return -EINVAL;
2433
2434 if (get_user(val, (int __user *)optval))
2435 return -EFAULT;
2436
2437 valbool = val ? 1 : 0;
2438
2439 switch (optname) {
2440 case UDP_CORK:
2441 if (val != 0) {
2442 up->corkflag = 1;
2443 } else {
2444 up->corkflag = 0;
2445 lock_sock(sk);
2446 push_pending_frames(sk);
2447 release_sock(sk);
2448 }
2449 break;
2450
2451 case UDP_ENCAP:
2452 switch (val) {
2453 case 0:
2454 case UDP_ENCAP_ESPINUDP:
2455 case UDP_ENCAP_ESPINUDP_NON_IKE:
2456 up->encap_rcv = xfrm4_udp_encap_rcv;
2457 /* FALLTHROUGH */
2458 case UDP_ENCAP_L2TPINUDP:
2459 up->encap_type = val;
2460 udp_encap_enable();
2461 break;
2462 default:
2463 err = -ENOPROTOOPT;
2464 break;
2465 }
2466 break;
2467
2468 case UDP_NO_CHECK6_TX:
2469 up->no_check6_tx = valbool;
2470 break;
2471
2472 case UDP_NO_CHECK6_RX:
2473 up->no_check6_rx = valbool;
2474 break;
2475
2476 case UDP_SEGMENT:
2477 if (val < 0 || val > USHRT_MAX)
2478 return -EINVAL;
2479 up->gso_size = val;
2480 break;
2481
2482 /*
2483 * UDP-Lite's partial checksum coverage (RFC 3828).
2484 */
2485 /* The sender sets actual checksum coverage length via this option.
2486 * The case coverage > packet length is handled by send module. */
2487 case UDPLITE_SEND_CSCOV:
2488 if (!is_udplite) /* Disable the option on UDP sockets */
2489 return -ENOPROTOOPT;
2490 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2491 val = 8;
2492 else if (val > USHRT_MAX)
2493 val = USHRT_MAX;
2494 up->pcslen = val;
2495 up->pcflag |= UDPLITE_SEND_CC;
2496 break;
2497
2498 /* The receiver specifies a minimum checksum coverage value. To make
2499 * sense, this should be set to at least 8 (as done below). If zero is
2500 * used, this again means full checksum coverage. */
2501 case UDPLITE_RECV_CSCOV:
2502 if (!is_udplite) /* Disable the option on UDP sockets */
2503 return -ENOPROTOOPT;
2504 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2505 val = 8;
2506 else if (val > USHRT_MAX)
2507 val = USHRT_MAX;
2508 up->pcrlen = val;
2509 up->pcflag |= UDPLITE_RECV_CC;
2510 break;
2511
2512 default:
2513 err = -ENOPROTOOPT;
2514 break;
2515 }
2516
2517 return err;
2518 }
2519 EXPORT_SYMBOL(udp_lib_setsockopt);
2520
2521 int udp_setsockopt(struct sock *sk, int level, int optname,
2522 char __user *optval, unsigned int optlen)
2523 {
2524 if (level == SOL_UDP || level == SOL_UDPLITE)
2525 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2526 udp_push_pending_frames);
2527 return ip_setsockopt(sk, level, optname, optval, optlen);
2528 }
2529
2530 #ifdef CONFIG_COMPAT
2531 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2532 char __user *optval, unsigned int optlen)
2533 {
2534 if (level == SOL_UDP || level == SOL_UDPLITE)
2535 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2536 udp_push_pending_frames);
2537 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2538 }
2539 #endif
2540
2541 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2542 char __user *optval, int __user *optlen)
2543 {
2544 struct udp_sock *up = udp_sk(sk);
2545 int val, len;
2546
2547 if (get_user(len, optlen))
2548 return -EFAULT;
2549
2550 len = min_t(unsigned int, len, sizeof(int));
2551
2552 if (len < 0)
2553 return -EINVAL;
2554
2555 switch (optname) {
2556 case UDP_CORK:
2557 val = up->corkflag;
2558 break;
2559
2560 case UDP_ENCAP:
2561 val = up->encap_type;
2562 break;
2563
2564 case UDP_NO_CHECK6_TX:
2565 val = up->no_check6_tx;
2566 break;
2567
2568 case UDP_NO_CHECK6_RX:
2569 val = up->no_check6_rx;
2570 break;
2571
2572 case UDP_SEGMENT:
2573 val = up->gso_size;
2574 break;
2575
2576 /* The following two cannot be changed on UDP sockets, the return is
2577 * always 0 (which corresponds to the full checksum coverage of UDP). */
2578 case UDPLITE_SEND_CSCOV:
2579 val = up->pcslen;
2580 break;
2581
2582 case UDPLITE_RECV_CSCOV:
2583 val = up->pcrlen;
2584 break;
2585
2586 default:
2587 return -ENOPROTOOPT;
2588 }
2589
2590 if (put_user(len, optlen))
2591 return -EFAULT;
2592 if (copy_to_user(optval, &val, len))
2593 return -EFAULT;
2594 return 0;
2595 }
2596 EXPORT_SYMBOL(udp_lib_getsockopt);
2597
2598 int udp_getsockopt(struct sock *sk, int level, int optname,
2599 char __user *optval, int __user *optlen)
2600 {
2601 if (level == SOL_UDP || level == SOL_UDPLITE)
2602 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2603 return ip_getsockopt(sk, level, optname, optval, optlen);
2604 }
2605
2606 #ifdef CONFIG_COMPAT
2607 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2608 char __user *optval, int __user *optlen)
2609 {
2610 if (level == SOL_UDP || level == SOL_UDPLITE)
2611 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2612 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2613 }
2614 #endif
2615 /**
2616 * udp_poll - wait for a UDP event.
2617 * @file - file struct
2618 * @sock - socket
2619 * @wait - poll table
2620 *
2621 * This is same as datagram poll, except for the special case of
2622 * blocking sockets. If application is using a blocking fd
2623 * and a packet with checksum error is in the queue;
2624 * then it could get return from select indicating data available
2625 * but then block when reading it. Add special case code
2626 * to work around these arguably broken applications.
2627 */
2628 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2629 {
2630 __poll_t mask = datagram_poll(file, sock, wait);
2631 struct sock *sk = sock->sk;
2632
2633 if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2634 mask |= EPOLLIN | EPOLLRDNORM;
2635
2636 /* Check for false positives due to checksum errors */
2637 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2638 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2639 mask &= ~(EPOLLIN | EPOLLRDNORM);
2640
2641 return mask;
2642
2643 }
2644 EXPORT_SYMBOL(udp_poll);
2645
2646 int udp_abort(struct sock *sk, int err)
2647 {
2648 lock_sock(sk);
2649
2650 sk->sk_err = err;
2651 sk->sk_error_report(sk);
2652 __udp_disconnect(sk, 0);
2653
2654 release_sock(sk);
2655
2656 return 0;
2657 }
2658 EXPORT_SYMBOL_GPL(udp_abort);
2659
2660 struct proto udp_prot = {
2661 .name = "UDP",
2662 .owner = THIS_MODULE,
2663 .close = udp_lib_close,
2664 .pre_connect = udp_pre_connect,
2665 .connect = ip4_datagram_connect,
2666 .disconnect = udp_disconnect,
2667 .ioctl = udp_ioctl,
2668 .init = udp_init_sock,
2669 .destroy = udp_destroy_sock,
2670 .setsockopt = udp_setsockopt,
2671 .getsockopt = udp_getsockopt,
2672 .sendmsg = udp_sendmsg,
2673 .recvmsg = udp_recvmsg,
2674 .sendpage = udp_sendpage,
2675 .release_cb = ip4_datagram_release_cb,
2676 .hash = udp_lib_hash,
2677 .unhash = udp_lib_unhash,
2678 .rehash = udp_v4_rehash,
2679 .get_port = udp_v4_get_port,
2680 .memory_allocated = &udp_memory_allocated,
2681 .sysctl_mem = sysctl_udp_mem,
2682 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2683 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2684 .obj_size = sizeof(struct udp_sock),
2685 .h.udp_table = &udp_table,
2686 #ifdef CONFIG_COMPAT
2687 .compat_setsockopt = compat_udp_setsockopt,
2688 .compat_getsockopt = compat_udp_getsockopt,
2689 #endif
2690 .diag_destroy = udp_abort,
2691 };
2692 EXPORT_SYMBOL(udp_prot);
2693
2694 /* ------------------------------------------------------------------------ */
2695 #ifdef CONFIG_PROC_FS
2696
2697 static struct sock *udp_get_first(struct seq_file *seq, int start)
2698 {
2699 struct sock *sk;
2700 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2701 struct udp_iter_state *state = seq->private;
2702 struct net *net = seq_file_net(seq);
2703
2704 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2705 ++state->bucket) {
2706 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2707
2708 if (hlist_empty(&hslot->head))
2709 continue;
2710
2711 spin_lock_bh(&hslot->lock);
2712 sk_for_each(sk, &hslot->head) {
2713 if (!net_eq(sock_net(sk), net))
2714 continue;
2715 if (sk->sk_family == afinfo->family)
2716 goto found;
2717 }
2718 spin_unlock_bh(&hslot->lock);
2719 }
2720 sk = NULL;
2721 found:
2722 return sk;
2723 }
2724
2725 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2726 {
2727 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2728 struct udp_iter_state *state = seq->private;
2729 struct net *net = seq_file_net(seq);
2730
2731 do {
2732 sk = sk_next(sk);
2733 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2734
2735 if (!sk) {
2736 if (state->bucket <= afinfo->udp_table->mask)
2737 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2738 return udp_get_first(seq, state->bucket + 1);
2739 }
2740 return sk;
2741 }
2742
2743 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2744 {
2745 struct sock *sk = udp_get_first(seq, 0);
2746
2747 if (sk)
2748 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2749 --pos;
2750 return pos ? NULL : sk;
2751 }
2752
2753 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2754 {
2755 struct udp_iter_state *state = seq->private;
2756 state->bucket = MAX_UDP_PORTS;
2757
2758 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2759 }
2760 EXPORT_SYMBOL(udp_seq_start);
2761
2762 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2763 {
2764 struct sock *sk;
2765
2766 if (v == SEQ_START_TOKEN)
2767 sk = udp_get_idx(seq, 0);
2768 else
2769 sk = udp_get_next(seq, v);
2770
2771 ++*pos;
2772 return sk;
2773 }
2774 EXPORT_SYMBOL(udp_seq_next);
2775
2776 void udp_seq_stop(struct seq_file *seq, void *v)
2777 {
2778 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2779 struct udp_iter_state *state = seq->private;
2780
2781 if (state->bucket <= afinfo->udp_table->mask)
2782 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2783 }
2784 EXPORT_SYMBOL(udp_seq_stop);
2785
2786 /* ------------------------------------------------------------------------ */
2787 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2788 int bucket)
2789 {
2790 struct inet_sock *inet = inet_sk(sp);
2791 __be32 dest = inet->inet_daddr;
2792 __be32 src = inet->inet_rcv_saddr;
2793 __u16 destp = ntohs(inet->inet_dport);
2794 __u16 srcp = ntohs(inet->inet_sport);
2795
2796 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2797 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2798 bucket, src, srcp, dest, destp, sp->sk_state,
2799 sk_wmem_alloc_get(sp),
2800 udp_rqueue_get(sp),
2801 0, 0L, 0,
2802 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2803 0, sock_i_ino(sp),
2804 refcount_read(&sp->sk_refcnt), sp,
2805 atomic_read(&sp->sk_drops));
2806 }
2807
2808 int udp4_seq_show(struct seq_file *seq, void *v)
2809 {
2810 seq_setwidth(seq, 127);
2811 if (v == SEQ_START_TOKEN)
2812 seq_puts(seq, " sl local_address rem_address st tx_queue "
2813 "rx_queue tr tm->when retrnsmt uid timeout "
2814 "inode ref pointer drops");
2815 else {
2816 struct udp_iter_state *state = seq->private;
2817
2818 udp4_format_sock(v, seq, state->bucket);
2819 }
2820 seq_pad(seq, '\n');
2821 return 0;
2822 }
2823
2824 const struct seq_operations udp_seq_ops = {
2825 .start = udp_seq_start,
2826 .next = udp_seq_next,
2827 .stop = udp_seq_stop,
2828 .show = udp4_seq_show,
2829 };
2830 EXPORT_SYMBOL(udp_seq_ops);
2831
2832 static struct udp_seq_afinfo udp4_seq_afinfo = {
2833 .family = AF_INET,
2834 .udp_table = &udp_table,
2835 };
2836
2837 static int __net_init udp4_proc_init_net(struct net *net)
2838 {
2839 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2840 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2841 return -ENOMEM;
2842 return 0;
2843 }
2844
2845 static void __net_exit udp4_proc_exit_net(struct net *net)
2846 {
2847 remove_proc_entry("udp", net->proc_net);
2848 }
2849
2850 static struct pernet_operations udp4_net_ops = {
2851 .init = udp4_proc_init_net,
2852 .exit = udp4_proc_exit_net,
2853 };
2854
2855 int __init udp4_proc_init(void)
2856 {
2857 return register_pernet_subsys(&udp4_net_ops);
2858 }
2859
2860 void udp4_proc_exit(void)
2861 {
2862 unregister_pernet_subsys(&udp4_net_ops);
2863 }
2864 #endif /* CONFIG_PROC_FS */
2865
2866 static __initdata unsigned long uhash_entries;
2867 static int __init set_uhash_entries(char *str)
2868 {
2869 ssize_t ret;
2870
2871 if (!str)
2872 return 0;
2873
2874 ret = kstrtoul(str, 0, &uhash_entries);
2875 if (ret)
2876 return 0;
2877
2878 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2879 uhash_entries = UDP_HTABLE_SIZE_MIN;
2880 return 1;
2881 }
2882 __setup("uhash_entries=", set_uhash_entries);
2883
2884 void __init udp_table_init(struct udp_table *table, const char *name)
2885 {
2886 unsigned int i;
2887
2888 table->hash = alloc_large_system_hash(name,
2889 2 * sizeof(struct udp_hslot),
2890 uhash_entries,
2891 21, /* one slot per 2 MB */
2892 0,
2893 &table->log,
2894 &table->mask,
2895 UDP_HTABLE_SIZE_MIN,
2896 64 * 1024);
2897
2898 table->hash2 = table->hash + (table->mask + 1);
2899 for (i = 0; i <= table->mask; i++) {
2900 INIT_HLIST_HEAD(&table->hash[i].head);
2901 table->hash[i].count = 0;
2902 spin_lock_init(&table->hash[i].lock);
2903 }
2904 for (i = 0; i <= table->mask; i++) {
2905 INIT_HLIST_HEAD(&table->hash2[i].head);
2906 table->hash2[i].count = 0;
2907 spin_lock_init(&table->hash2[i].lock);
2908 }
2909 }
2910
2911 u32 udp_flow_hashrnd(void)
2912 {
2913 static u32 hashrnd __read_mostly;
2914
2915 net_get_random_once(&hashrnd, sizeof(hashrnd));
2916
2917 return hashrnd;
2918 }
2919 EXPORT_SYMBOL(udp_flow_hashrnd);
2920
2921 static void __udp_sysctl_init(struct net *net)
2922 {
2923 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2924 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2925
2926 #ifdef CONFIG_NET_L3_MASTER_DEV
2927 net->ipv4.sysctl_udp_l3mdev_accept = 0;
2928 #endif
2929 }
2930
2931 static int __net_init udp_sysctl_init(struct net *net)
2932 {
2933 __udp_sysctl_init(net);
2934 return 0;
2935 }
2936
2937 static struct pernet_operations __net_initdata udp_sysctl_ops = {
2938 .init = udp_sysctl_init,
2939 };
2940
2941 void __init udp_init(void)
2942 {
2943 unsigned long limit;
2944 unsigned int i;
2945
2946 udp_table_init(&udp_table, "UDP");
2947 limit = nr_free_buffer_pages() / 8;
2948 limit = max(limit, 128UL);
2949 sysctl_udp_mem[0] = limit / 4 * 3;
2950 sysctl_udp_mem[1] = limit;
2951 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2952
2953 __udp_sysctl_init(&init_net);
2954
2955 /* 16 spinlocks per cpu */
2956 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2957 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2958 GFP_KERNEL);
2959 if (!udp_busylocks)
2960 panic("UDP: failed to alloc udp_busylocks\n");
2961 for (i = 0; i < (1U << udp_busylocks_log); i++)
2962 spin_lock_init(udp_busylocks + i);
2963
2964 if (register_pernet_subsys(&udp_sysctl_ops))
2965 panic("UDP: failed to init sysctl parameters.\n");
2966 }
2967
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.