?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/speculation/mds: Fix documentation typo
[linux/fpc-iii.git] / net / ipv4 / udp.c
blobb89920c0f22638d212784624aa698245c290f7c6
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/bootmem.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 int sysctl_udp_rmem_min __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_rmem_min);
127
128 int sysctl_udp_wmem_min __read_mostly;
129 EXPORT_SYMBOL(sysctl_udp_wmem_min);
130
131 atomic_long_t udp_memory_allocated;
132 EXPORT_SYMBOL(udp_memory_allocated);
133
134 #define MAX_UDP_PORTS 65536
135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
136
137 /* IPCB reference means this can not be used from early demux */
138 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
139 {
140 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
141 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
142 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
143 return true;
144 #endif
145 return false;
146 }
147
148 static int udp_lib_lport_inuse(struct net *net, __u16 num,
149 const struct udp_hslot *hslot,
150 unsigned long *bitmap,
151 struct sock *sk, unsigned int log)
152 {
153 struct sock *sk2;
154 kuid_t uid = sock_i_uid(sk);
155
156 sk_for_each(sk2, &hslot->head) {
157 if (net_eq(sock_net(sk2), net) &&
158 sk2 != sk &&
159 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
160 (!sk2->sk_reuse || !sk->sk_reuse) &&
161 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
162 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
163 inet_rcv_saddr_equal(sk, sk2, true)) {
164 if (sk2->sk_reuseport && sk->sk_reuseport &&
165 !rcu_access_pointer(sk->sk_reuseport_cb) &&
166 uid_eq(uid, sock_i_uid(sk2))) {
167 if (!bitmap)
168 return 0;
169 } else {
170 if (!bitmap)
171 return 1;
172 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
173 bitmap);
174 }
175 }
176 }
177 return 0;
178 }
179
180 /*
181 * Note: we still hold spinlock of primary hash chain, so no other writer
182 * can insert/delete a socket with local_port == num
183 */
184 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
185 struct udp_hslot *hslot2,
186 struct sock *sk)
187 {
188 struct sock *sk2;
189 kuid_t uid = sock_i_uid(sk);
190 int res = 0;
191
192 spin_lock(&hslot2->lock);
193 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
194 if (net_eq(sock_net(sk2), net) &&
195 sk2 != sk &&
196 (udp_sk(sk2)->udp_port_hash == num) &&
197 (!sk2->sk_reuse || !sk->sk_reuse) &&
198 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
199 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
200 inet_rcv_saddr_equal(sk, sk2, true)) {
201 if (sk2->sk_reuseport && sk->sk_reuseport &&
202 !rcu_access_pointer(sk->sk_reuseport_cb) &&
203 uid_eq(uid, sock_i_uid(sk2))) {
204 res = 0;
205 } else {
206 res = 1;
207 }
208 break;
209 }
210 }
211 spin_unlock(&hslot2->lock);
212 return res;
213 }
214
215 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
216 {
217 struct net *net = sock_net(sk);
218 kuid_t uid = sock_i_uid(sk);
219 struct sock *sk2;
220
221 sk_for_each(sk2, &hslot->head) {
222 if (net_eq(sock_net(sk2), net) &&
223 sk2 != sk &&
224 sk2->sk_family == sk->sk_family &&
225 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
226 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
227 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
228 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
229 inet_rcv_saddr_equal(sk, sk2, false)) {
230 return reuseport_add_sock(sk, sk2);
231 }
232 }
233
234 return reuseport_alloc(sk);
235 }
236
237 /**
238 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
239 *
240 * @sk: socket struct in question
241 * @snum: port number to look up
242 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
243 * with NULL address
244 */
245 int udp_lib_get_port(struct sock *sk, unsigned short snum,
246 unsigned int hash2_nulladdr)
247 {
248 struct udp_hslot *hslot, *hslot2;
249 struct udp_table *udptable = sk->sk_prot->h.udp_table;
250 int error = 1;
251 struct net *net = sock_net(sk);
252
253 if (!snum) {
254 int low, high, remaining;
255 unsigned int rand;
256 unsigned short first, last;
257 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
258
259 inet_get_local_port_range(net, &low, &high);
260 remaining = (high - low) + 1;
261
262 rand = prandom_u32();
263 first = reciprocal_scale(rand, remaining) + low;
264 /*
265 * force rand to be an odd multiple of UDP_HTABLE_SIZE
266 */
267 rand = (rand | 1) * (udptable->mask + 1);
268 last = first + udptable->mask + 1;
269 do {
270 hslot = udp_hashslot(udptable, net, first);
271 bitmap_zero(bitmap, PORTS_PER_CHAIN);
272 spin_lock_bh(&hslot->lock);
273 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
274 udptable->log);
275
276 snum = first;
277 /*
278 * Iterate on all possible values of snum for this hash.
279 * Using steps of an odd multiple of UDP_HTABLE_SIZE
280 * give us randomization and full range coverage.
281 */
282 do {
283 if (low <= snum && snum <= high &&
284 !test_bit(snum >> udptable->log, bitmap) &&
285 !inet_is_local_reserved_port(net, snum))
286 goto found;
287 snum += rand;
288 } while (snum != first);
289 spin_unlock_bh(&hslot->lock);
290 cond_resched();
291 } while (++first != last);
292 goto fail;
293 } else {
294 hslot = udp_hashslot(udptable, net, snum);
295 spin_lock_bh(&hslot->lock);
296 if (hslot->count > 10) {
297 int exist;
298 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
299
300 slot2 &= udptable->mask;
301 hash2_nulladdr &= udptable->mask;
302
303 hslot2 = udp_hashslot2(udptable, slot2);
304 if (hslot->count < hslot2->count)
305 goto scan_primary_hash;
306
307 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
308 if (!exist && (hash2_nulladdr != slot2)) {
309 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
310 exist = udp_lib_lport_inuse2(net, snum, hslot2,
311 sk);
312 }
313 if (exist)
314 goto fail_unlock;
315 else
316 goto found;
317 }
318 scan_primary_hash:
319 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
320 goto fail_unlock;
321 }
322 found:
323 inet_sk(sk)->inet_num = snum;
324 udp_sk(sk)->udp_port_hash = snum;
325 udp_sk(sk)->udp_portaddr_hash ^= snum;
326 if (sk_unhashed(sk)) {
327 if (sk->sk_reuseport &&
328 udp_reuseport_add_sock(sk, hslot)) {
329 inet_sk(sk)->inet_num = 0;
330 udp_sk(sk)->udp_port_hash = 0;
331 udp_sk(sk)->udp_portaddr_hash ^= snum;
332 goto fail_unlock;
333 }
334
335 sk_add_node_rcu(sk, &hslot->head);
336 hslot->count++;
337 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
338
339 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
340 spin_lock(&hslot2->lock);
341 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
342 sk->sk_family == AF_INET6)
343 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
344 &hslot2->head);
345 else
346 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
347 &hslot2->head);
348 hslot2->count++;
349 spin_unlock(&hslot2->lock);
350 }
351 sock_set_flag(sk, SOCK_RCU_FREE);
352 error = 0;
353 fail_unlock:
354 spin_unlock_bh(&hslot->lock);
355 fail:
356 return error;
357 }
358 EXPORT_SYMBOL(udp_lib_get_port);
359
360 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
361 unsigned int port)
362 {
363 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
364 }
365
366 int udp_v4_get_port(struct sock *sk, unsigned short snum)
367 {
368 unsigned int hash2_nulladdr =
369 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
370 unsigned int hash2_partial =
371 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
372
373 /* precompute partial secondary hash */
374 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
375 return udp_lib_get_port(sk, snum, hash2_nulladdr);
376 }
377
378 static int compute_score(struct sock *sk, struct net *net,
379 __be32 saddr, __be16 sport,
380 __be32 daddr, unsigned short hnum,
381 int dif, int sdif, bool exact_dif)
382 {
383 int score;
384 struct inet_sock *inet;
385
386 if (!net_eq(sock_net(sk), net) ||
387 udp_sk(sk)->udp_port_hash != hnum ||
388 ipv6_only_sock(sk))
389 return -1;
390
391 score = (sk->sk_family == PF_INET) ? 2 : 1;
392 inet = inet_sk(sk);
393
394 if (inet->inet_rcv_saddr) {
395 if (inet->inet_rcv_saddr != daddr)
396 return -1;
397 score += 4;
398 }
399
400 if (inet->inet_daddr) {
401 if (inet->inet_daddr != saddr)
402 return -1;
403 score += 4;
404 }
405
406 if (inet->inet_dport) {
407 if (inet->inet_dport != sport)
408 return -1;
409 score += 4;
410 }
411
412 if (sk->sk_bound_dev_if || exact_dif) {
413 bool dev_match = (sk->sk_bound_dev_if == dif ||
414 sk->sk_bound_dev_if == sdif);
415
416 if (!dev_match)
417 return -1;
418 if (sk->sk_bound_dev_if)
419 score += 4;
420 }
421
422 if (sk->sk_incoming_cpu == raw_smp_processor_id())
423 score++;
424 return score;
425 }
426
427 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
428 const __u16 lport, const __be32 faddr,
429 const __be16 fport)
430 {
431 static u32 udp_ehash_secret __read_mostly;
432
433 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
434
435 return __inet_ehashfn(laddr, lport, faddr, fport,
436 udp_ehash_secret + net_hash_mix(net));
437 }
438
439 /* called with rcu_read_lock() */
440 static struct sock *udp4_lib_lookup2(struct net *net,
441 __be32 saddr, __be16 sport,
442 __be32 daddr, unsigned int hnum,
443 int dif, int sdif, bool exact_dif,
444 struct udp_hslot *hslot2,
445 struct sk_buff *skb)
446 {
447 struct sock *sk, *result;
448 int score, badness, matches = 0, reuseport = 0;
449 u32 hash = 0;
450
451 result = NULL;
452 badness = 0;
453 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
454 score = compute_score(sk, net, saddr, sport,
455 daddr, hnum, dif, sdif, exact_dif);
456 if (score > badness) {
457 reuseport = sk->sk_reuseport;
458 if (reuseport) {
459 hash = udp_ehashfn(net, daddr, hnum,
460 saddr, sport);
461 result = reuseport_select_sock(sk, hash, skb,
462 sizeof(struct udphdr));
463 if (result)
464 return result;
465 matches = 1;
466 }
467 badness = score;
468 result = sk;
469 } else if (score == badness && reuseport) {
470 matches++;
471 if (reciprocal_scale(hash, matches) == 0)
472 result = sk;
473 hash = next_pseudo_random32(hash);
474 }
475 }
476 return result;
477 }
478
479 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
480 * harder than this. -DaveM
481 */
482 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
483 __be16 sport, __be32 daddr, __be16 dport, int dif,
484 int sdif, struct udp_table *udptable, struct sk_buff *skb)
485 {
486 struct sock *sk, *result;
487 unsigned short hnum = ntohs(dport);
488 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
489 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
490 bool exact_dif = udp_lib_exact_dif_match(net, skb);
491 int score, badness, matches = 0, reuseport = 0;
492 u32 hash = 0;
493
494 if (hslot->count > 10) {
495 hash2 = udp4_portaddr_hash(net, daddr, hnum);
496 slot2 = hash2 & udptable->mask;
497 hslot2 = &udptable->hash2[slot2];
498 if (hslot->count < hslot2->count)
499 goto begin;
500
501 result = udp4_lib_lookup2(net, saddr, sport,
502 daddr, hnum, dif, sdif,
503 exact_dif, hslot2, skb);
504 if (!result) {
505 unsigned int old_slot2 = slot2;
506 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
507 slot2 = hash2 & udptable->mask;
508 /* avoid searching the same slot again. */
509 if (unlikely(slot2 == old_slot2))
510 return result;
511
512 hslot2 = &udptable->hash2[slot2];
513 if (hslot->count < hslot2->count)
514 goto begin;
515
516 result = udp4_lib_lookup2(net, saddr, sport,
517 daddr, hnum, dif, sdif,
518 exact_dif, hslot2, skb);
519 }
520 return result;
521 }
522 begin:
523 result = NULL;
524 badness = 0;
525 sk_for_each_rcu(sk, &hslot->head) {
526 score = compute_score(sk, net, saddr, sport,
527 daddr, hnum, dif, sdif, exact_dif);
528 if (score > badness) {
529 reuseport = sk->sk_reuseport;
530 if (reuseport) {
531 hash = udp_ehashfn(net, daddr, hnum,
532 saddr, sport);
533 result = reuseport_select_sock(sk, hash, skb,
534 sizeof(struct udphdr));
535 if (result)
536 return result;
537 matches = 1;
538 }
539 result = sk;
540 badness = score;
541 } else if (score == badness && reuseport) {
542 matches++;
543 if (reciprocal_scale(hash, matches) == 0)
544 result = sk;
545 hash = next_pseudo_random32(hash);
546 }
547 }
548 return result;
549 }
550 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
551
552 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
553 __be16 sport, __be16 dport,
554 struct udp_table *udptable)
555 {
556 const struct iphdr *iph = ip_hdr(skb);
557
558 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
559 iph->daddr, dport, inet_iif(skb),
560 inet_sdif(skb), udptable, skb);
561 }
562
563 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
564 __be16 sport, __be16 dport)
565 {
566 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
567 }
568 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
569
570 /* Must be called under rcu_read_lock().
571 * Does increment socket refcount.
572 */
573 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
574 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
575 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
576 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
577 __be32 daddr, __be16 dport, int dif)
578 {
579 struct sock *sk;
580
581 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
582 dif, 0, &udp_table, NULL);
583 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
584 sk = NULL;
585 return sk;
586 }
587 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
588 #endif
589
590 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
591 __be16 loc_port, __be32 loc_addr,
592 __be16 rmt_port, __be32 rmt_addr,
593 int dif, int sdif, unsigned short hnum)
594 {
595 struct inet_sock *inet = inet_sk(sk);
596
597 if (!net_eq(sock_net(sk), net) ||
598 udp_sk(sk)->udp_port_hash != hnum ||
599 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
600 (inet->inet_dport != rmt_port && inet->inet_dport) ||
601 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
602 ipv6_only_sock(sk) ||
603 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
604 sk->sk_bound_dev_if != sdif))
605 return false;
606 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
607 return false;
608 return true;
609 }
610
611 /*
612 * This routine is called by the ICMP module when it gets some
613 * sort of error condition. If err < 0 then the socket should
614 * be closed and the error returned to the user. If err > 0
615 * it's just the icmp type << 8 | icmp code.
616 * Header points to the ip header of the error packet. We move
617 * on past this. Then (as it used to claim before adjustment)
618 * header points to the first 8 bytes of the udp header. We need
619 * to find the appropriate port.
620 */
621
622 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
623 {
624 struct inet_sock *inet;
625 const struct iphdr *iph = (const struct iphdr *)skb->data;
626 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
627 const int type = icmp_hdr(skb)->type;
628 const int code = icmp_hdr(skb)->code;
629 struct sock *sk;
630 int harderr;
631 int err;
632 struct net *net = dev_net(skb->dev);
633
634 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
635 iph->saddr, uh->source, skb->dev->ifindex, 0,
636 udptable, NULL);
637 if (!sk) {
638 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
639 return; /* No socket for error */
640 }
641
642 err = 0;
643 harderr = 0;
644 inet = inet_sk(sk);
645
646 switch (type) {
647 default:
648 case ICMP_TIME_EXCEEDED:
649 err = EHOSTUNREACH;
650 break;
651 case ICMP_SOURCE_QUENCH:
652 goto out;
653 case ICMP_PARAMETERPROB:
654 err = EPROTO;
655 harderr = 1;
656 break;
657 case ICMP_DEST_UNREACH:
658 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
659 ipv4_sk_update_pmtu(skb, sk, info);
660 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
661 err = EMSGSIZE;
662 harderr = 1;
663 break;
664 }
665 goto out;
666 }
667 err = EHOSTUNREACH;
668 if (code <= NR_ICMP_UNREACH) {
669 harderr = icmp_err_convert[code].fatal;
670 err = icmp_err_convert[code].errno;
671 }
672 break;
673 case ICMP_REDIRECT:
674 ipv4_sk_redirect(skb, sk);
675 goto out;
676 }
677
678 /*
679 * RFC1122: OK. Passes ICMP errors back to application, as per
680 * 4.1.3.3.
681 */
682 if (!inet->recverr) {
683 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
684 goto out;
685 } else
686 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
687
688 sk->sk_err = err;
689 sk->sk_error_report(sk);
690 out:
691 return;
692 }
693
694 void udp_err(struct sk_buff *skb, u32 info)
695 {
696 __udp4_lib_err(skb, info, &udp_table);
697 }
698
699 /*
700 * Throw away all pending data and cancel the corking. Socket is locked.
701 */
702 void udp_flush_pending_frames(struct sock *sk)
703 {
704 struct udp_sock *up = udp_sk(sk);
705
706 if (up->pending) {
707 up->len = 0;
708 up->pending = 0;
709 ip_flush_pending_frames(sk);
710 }
711 }
712 EXPORT_SYMBOL(udp_flush_pending_frames);
713
714 /**
715 * udp4_hwcsum - handle outgoing HW checksumming
716 * @skb: sk_buff containing the filled-in UDP header
717 * (checksum field must be zeroed out)
718 * @src: source IP address
719 * @dst: destination IP address
720 */
721 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
722 {
723 struct udphdr *uh = udp_hdr(skb);
724 int offset = skb_transport_offset(skb);
725 int len = skb->len - offset;
726 int hlen = len;
727 __wsum csum = 0;
728
729 if (!skb_has_frag_list(skb)) {
730 /*
731 * Only one fragment on the socket.
732 */
733 skb->csum_start = skb_transport_header(skb) - skb->head;
734 skb->csum_offset = offsetof(struct udphdr, check);
735 uh->check = ~csum_tcpudp_magic(src, dst, len,
736 IPPROTO_UDP, 0);
737 } else {
738 struct sk_buff *frags;
739
740 /*
741 * HW-checksum won't work as there are two or more
742 * fragments on the socket so that all csums of sk_buffs
743 * should be together
744 */
745 skb_walk_frags(skb, frags) {
746 csum = csum_add(csum, frags->csum);
747 hlen -= frags->len;
748 }
749
750 csum = skb_checksum(skb, offset, hlen, csum);
751 skb->ip_summed = CHECKSUM_NONE;
752
753 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
754 if (uh->check == 0)
755 uh->check = CSUM_MANGLED_0;
756 }
757 }
758 EXPORT_SYMBOL_GPL(udp4_hwcsum);
759
760 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
761 * for the simple case like when setting the checksum for a UDP tunnel.
762 */
763 void udp_set_csum(bool nocheck, struct sk_buff *skb,
764 __be32 saddr, __be32 daddr, int len)
765 {
766 struct udphdr *uh = udp_hdr(skb);
767
768 if (nocheck) {
769 uh->check = 0;
770 } else if (skb_is_gso(skb)) {
771 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
772 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
773 uh->check = 0;
774 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
775 if (uh->check == 0)
776 uh->check = CSUM_MANGLED_0;
777 } else {
778 skb->ip_summed = CHECKSUM_PARTIAL;
779 skb->csum_start = skb_transport_header(skb) - skb->head;
780 skb->csum_offset = offsetof(struct udphdr, check);
781 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
782 }
783 }
784 EXPORT_SYMBOL(udp_set_csum);
785
786 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
787 {
788 struct sock *sk = skb->sk;
789 struct inet_sock *inet = inet_sk(sk);
790 struct udphdr *uh;
791 int err = 0;
792 int is_udplite = IS_UDPLITE(sk);
793 int offset = skb_transport_offset(skb);
794 int len = skb->len - offset;
795 __wsum csum = 0;
796
797 /*
798 * Create a UDP header
799 */
800 uh = udp_hdr(skb);
801 uh->source = inet->inet_sport;
802 uh->dest = fl4->fl4_dport;
803 uh->len = htons(len);
804 uh->check = 0;
805
806 if (is_udplite) /* UDP-Lite */
807 csum = udplite_csum(skb);
808
809 else if (sk->sk_no_check_tx) { /* UDP csum off */
810
811 skb->ip_summed = CHECKSUM_NONE;
812 goto send;
813
814 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
815
816 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
817 goto send;
818
819 } else
820 csum = udp_csum(skb);
821
822 /* add protocol-dependent pseudo-header */
823 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
824 sk->sk_protocol, csum);
825 if (uh->check == 0)
826 uh->check = CSUM_MANGLED_0;
827
828 send:
829 err = ip_send_skb(sock_net(sk), skb);
830 if (err) {
831 if (err == -ENOBUFS && !inet->recverr) {
832 UDP_INC_STATS(sock_net(sk),
833 UDP_MIB_SNDBUFERRORS, is_udplite);
834 err = 0;
835 }
836 } else
837 UDP_INC_STATS(sock_net(sk),
838 UDP_MIB_OUTDATAGRAMS, is_udplite);
839 return err;
840 }
841
842 /*
843 * Push out all pending data as one UDP datagram. Socket is locked.
844 */
845 int udp_push_pending_frames(struct sock *sk)
846 {
847 struct udp_sock *up = udp_sk(sk);
848 struct inet_sock *inet = inet_sk(sk);
849 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
850 struct sk_buff *skb;
851 int err = 0;
852
853 skb = ip_finish_skb(sk, fl4);
854 if (!skb)
855 goto out;
856
857 err = udp_send_skb(skb, fl4);
858
859 out:
860 up->len = 0;
861 up->pending = 0;
862 return err;
863 }
864 EXPORT_SYMBOL(udp_push_pending_frames);
865
866 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
867 {
868 struct inet_sock *inet = inet_sk(sk);
869 struct udp_sock *up = udp_sk(sk);
870 struct flowi4 fl4_stack;
871 struct flowi4 *fl4;
872 int ulen = len;
873 struct ipcm_cookie ipc;
874 struct rtable *rt = NULL;
875 int free = 0;
876 int connected = 0;
877 __be32 daddr, faddr, saddr;
878 __be16 dport;
879 u8 tos;
880 int err, is_udplite = IS_UDPLITE(sk);
881 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
882 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
883 struct sk_buff *skb;
884 struct ip_options_data opt_copy;
885
886 if (len > 0xFFFF)
887 return -EMSGSIZE;
888
889 /*
890 * Check the flags.
891 */
892
893 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
894 return -EOPNOTSUPP;
895
896 ipc.opt = NULL;
897 ipc.tx_flags = 0;
898 ipc.ttl = 0;
899 ipc.tos = -1;
900
901 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
902
903 fl4 = &inet->cork.fl.u.ip4;
904 if (up->pending) {
905 /*
906 * There are pending frames.
907 * The socket lock must be held while it's corked.
908 */
909 lock_sock(sk);
910 if (likely(up->pending)) {
911 if (unlikely(up->pending != AF_INET)) {
912 release_sock(sk);
913 return -EINVAL;
914 }
915 goto do_append_data;
916 }
917 release_sock(sk);
918 }
919 ulen += sizeof(struct udphdr);
920
921 /*
922 * Get and verify the address.
923 */
924 if (msg->msg_name) {
925 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
926 if (msg->msg_namelen < sizeof(*usin))
927 return -EINVAL;
928 if (usin->sin_family != AF_INET) {
929 if (usin->sin_family != AF_UNSPEC)
930 return -EAFNOSUPPORT;
931 }
932
933 daddr = usin->sin_addr.s_addr;
934 dport = usin->sin_port;
935 if (dport == 0)
936 return -EINVAL;
937 } else {
938 if (sk->sk_state != TCP_ESTABLISHED)
939 return -EDESTADDRREQ;
940 daddr = inet->inet_daddr;
941 dport = inet->inet_dport;
942 /* Open fast path for connected socket.
943 Route will not be used, if at least one option is set.
944 */
945 connected = 1;
946 }
947
948 ipc.sockc.tsflags = sk->sk_tsflags;
949 ipc.addr = inet->inet_saddr;
950 ipc.oif = sk->sk_bound_dev_if;
951
952 if (msg->msg_controllen) {
953 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
954 if (unlikely(err)) {
955 kfree(ipc.opt);
956 return err;
957 }
958 if (ipc.opt)
959 free = 1;
960 connected = 0;
961 }
962 if (!ipc.opt) {
963 struct ip_options_rcu *inet_opt;
964
965 rcu_read_lock();
966 inet_opt = rcu_dereference(inet->inet_opt);
967 if (inet_opt) {
968 memcpy(&opt_copy, inet_opt,
969 sizeof(*inet_opt) + inet_opt->opt.optlen);
970 ipc.opt = &opt_copy.opt;
971 }
972 rcu_read_unlock();
973 }
974
975 saddr = ipc.addr;
976 ipc.addr = faddr = daddr;
977
978 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
979
980 if (ipc.opt && ipc.opt->opt.srr) {
981 if (!daddr) {
982 err = -EINVAL;
983 goto out_free;
984 }
985 faddr = ipc.opt->opt.faddr;
986 connected = 0;
987 }
988 tos = get_rttos(&ipc, inet);
989 if (sock_flag(sk, SOCK_LOCALROUTE) ||
990 (msg->msg_flags & MSG_DONTROUTE) ||
991 (ipc.opt && ipc.opt->opt.is_strictroute)) {
992 tos |= RTO_ONLINK;
993 connected = 0;
994 }
995
996 if (ipv4_is_multicast(daddr)) {
997 if (!ipc.oif)
998 ipc.oif = inet->mc_index;
999 if (!saddr)
1000 saddr = inet->mc_addr;
1001 connected = 0;
1002 } else if (!ipc.oif)
1003 ipc.oif = inet->uc_index;
1004
1005 if (connected)
1006 rt = (struct rtable *)sk_dst_check(sk, 0);
1007
1008 if (!rt) {
1009 struct net *net = sock_net(sk);
1010 __u8 flow_flags = inet_sk_flowi_flags(sk);
1011
1012 fl4 = &fl4_stack;
1013
1014 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1015 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1016 flow_flags,
1017 faddr, saddr, dport, inet->inet_sport,
1018 sk->sk_uid);
1019
1020 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1021 rt = ip_route_output_flow(net, fl4, sk);
1022 if (IS_ERR(rt)) {
1023 err = PTR_ERR(rt);
1024 rt = NULL;
1025 if (err == -ENETUNREACH)
1026 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1027 goto out;
1028 }
1029
1030 err = -EACCES;
1031 if ((rt->rt_flags & RTCF_BROADCAST) &&
1032 !sock_flag(sk, SOCK_BROADCAST))
1033 goto out;
1034 if (connected)
1035 sk_dst_set(sk, dst_clone(&rt->dst));
1036 }
1037
1038 if (msg->msg_flags&MSG_CONFIRM)
1039 goto do_confirm;
1040 back_from_confirm:
1041
1042 saddr = fl4->saddr;
1043 if (!ipc.addr)
1044 daddr = ipc.addr = fl4->daddr;
1045
1046 /* Lockless fast path for the non-corking case. */
1047 if (!corkreq) {
1048 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1049 sizeof(struct udphdr), &ipc, &rt,
1050 msg->msg_flags);
1051 err = PTR_ERR(skb);
1052 if (!IS_ERR_OR_NULL(skb))
1053 err = udp_send_skb(skb, fl4);
1054 goto out;
1055 }
1056
1057 lock_sock(sk);
1058 if (unlikely(up->pending)) {
1059 /* The socket is already corked while preparing it. */
1060 /* ... which is an evident application bug. --ANK */
1061 release_sock(sk);
1062
1063 net_dbg_ratelimited("socket already corked\n");
1064 err = -EINVAL;
1065 goto out;
1066 }
1067 /*
1068 * Now cork the socket to pend data.
1069 */
1070 fl4 = &inet->cork.fl.u.ip4;
1071 fl4->daddr = daddr;
1072 fl4->saddr = saddr;
1073 fl4->fl4_dport = dport;
1074 fl4->fl4_sport = inet->inet_sport;
1075 up->pending = AF_INET;
1076
1077 do_append_data:
1078 up->len += ulen;
1079 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1080 sizeof(struct udphdr), &ipc, &rt,
1081 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1082 if (err)
1083 udp_flush_pending_frames(sk);
1084 else if (!corkreq)
1085 err = udp_push_pending_frames(sk);
1086 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1087 up->pending = 0;
1088 release_sock(sk);
1089
1090 out:
1091 ip_rt_put(rt);
1092 out_free:
1093 if (free)
1094 kfree(ipc.opt);
1095 if (!err)
1096 return len;
1097 /*
1098 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1099 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1100 * we don't have a good statistic (IpOutDiscards but it can be too many
1101 * things). We could add another new stat but at least for now that
1102 * seems like overkill.
1103 */
1104 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1105 UDP_INC_STATS(sock_net(sk),
1106 UDP_MIB_SNDBUFERRORS, is_udplite);
1107 }
1108 return err;
1109
1110 do_confirm:
1111 if (msg->msg_flags & MSG_PROBE)
1112 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1113 if (!(msg->msg_flags&MSG_PROBE) || len)
1114 goto back_from_confirm;
1115 err = 0;
1116 goto out;
1117 }
1118 EXPORT_SYMBOL(udp_sendmsg);
1119
1120 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1121 size_t size, int flags)
1122 {
1123 struct inet_sock *inet = inet_sk(sk);
1124 struct udp_sock *up = udp_sk(sk);
1125 int ret;
1126
1127 if (flags & MSG_SENDPAGE_NOTLAST)
1128 flags |= MSG_MORE;
1129
1130 if (!up->pending) {
1131 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1132
1133 /* Call udp_sendmsg to specify destination address which
1134 * sendpage interface can't pass.
1135 * This will succeed only when the socket is connected.
1136 */
1137 ret = udp_sendmsg(sk, &msg, 0);
1138 if (ret < 0)
1139 return ret;
1140 }
1141
1142 lock_sock(sk);
1143
1144 if (unlikely(!up->pending)) {
1145 release_sock(sk);
1146
1147 net_dbg_ratelimited("cork failed\n");
1148 return -EINVAL;
1149 }
1150
1151 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1152 page, offset, size, flags);
1153 if (ret == -EOPNOTSUPP) {
1154 release_sock(sk);
1155 return sock_no_sendpage(sk->sk_socket, page, offset,
1156 size, flags);
1157 }
1158 if (ret < 0) {
1159 udp_flush_pending_frames(sk);
1160 goto out;
1161 }
1162
1163 up->len += size;
1164 if (!(up->corkflag || (flags&MSG_MORE)))
1165 ret = udp_push_pending_frames(sk);
1166 if (!ret)
1167 ret = size;
1168 out:
1169 release_sock(sk);
1170 return ret;
1171 }
1172
1173 #define UDP_SKB_IS_STATELESS 0x80000000
1174
1175 static void udp_set_dev_scratch(struct sk_buff *skb)
1176 {
1177 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1178
1179 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1180 scratch->_tsize_state = skb->truesize;
1181 #if BITS_PER_LONG == 64
1182 scratch->len = skb->len;
1183 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1184 scratch->is_linear = !skb_is_nonlinear(skb);
1185 #endif
1186 /* all head states execept sp (dst, sk, nf) are always cleared by
1187 * udp_rcv() and we need to preserve secpath, if present, to eventually
1188 * process IP_CMSG_PASSSEC at recvmsg() time
1189 */
1190 if (likely(!skb_sec_path(skb)))
1191 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1192 }
1193
1194 static int udp_skb_truesize(struct sk_buff *skb)
1195 {
1196 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1197 }
1198
1199 static bool udp_skb_has_head_state(struct sk_buff *skb)
1200 {
1201 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1202 }
1203
1204 /* fully reclaim rmem/fwd memory allocated for skb */
1205 static void udp_rmem_release(struct sock *sk, int size, int partial,
1206 bool rx_queue_lock_held)
1207 {
1208 struct udp_sock *up = udp_sk(sk);
1209 struct sk_buff_head *sk_queue;
1210 int amt;
1211
1212 if (likely(partial)) {
1213 up->forward_deficit += size;
1214 size = up->forward_deficit;
1215 if (size < (sk->sk_rcvbuf >> 2) &&
1216 !skb_queue_empty(&up->reader_queue))
1217 return;
1218 } else {
1219 size += up->forward_deficit;
1220 }
1221 up->forward_deficit = 0;
1222
1223 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1224 * if the called don't held it already
1225 */
1226 sk_queue = &sk->sk_receive_queue;
1227 if (!rx_queue_lock_held)
1228 spin_lock(&sk_queue->lock);
1229
1230
1231 sk->sk_forward_alloc += size;
1232 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1233 sk->sk_forward_alloc -= amt;
1234
1235 if (amt)
1236 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1237
1238 atomic_sub(size, &sk->sk_rmem_alloc);
1239
1240 /* this can save us from acquiring the rx queue lock on next receive */
1241 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1242
1243 if (!rx_queue_lock_held)
1244 spin_unlock(&sk_queue->lock);
1245 }
1246
1247 /* Note: called with reader_queue.lock held.
1248 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1249 * This avoids a cache line miss while receive_queue lock is held.
1250 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1251 */
1252 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1253 {
1254 prefetch(&skb->data);
1255 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1256 }
1257 EXPORT_SYMBOL(udp_skb_destructor);
1258
1259 /* as above, but the caller held the rx queue lock, too */
1260 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1261 {
1262 prefetch(&skb->data);
1263 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1264 }
1265
1266 /* Idea of busylocks is to let producers grab an extra spinlock
1267 * to relieve pressure on the receive_queue spinlock shared by consumer.
1268 * Under flood, this means that only one producer can be in line
1269 * trying to acquire the receive_queue spinlock.
1270 * These busylock can be allocated on a per cpu manner, instead of a
1271 * per socket one (that would consume a cache line per socket)
1272 */
1273 static int udp_busylocks_log __read_mostly;
1274 static spinlock_t *udp_busylocks __read_mostly;
1275
1276 static spinlock_t *busylock_acquire(void *ptr)
1277 {
1278 spinlock_t *busy;
1279
1280 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1281 spin_lock(busy);
1282 return busy;
1283 }
1284
1285 static void busylock_release(spinlock_t *busy)
1286 {
1287 if (busy)
1288 spin_unlock(busy);
1289 }
1290
1291 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1292 {
1293 struct sk_buff_head *list = &sk->sk_receive_queue;
1294 int rmem, delta, amt, err = -ENOMEM;
1295 spinlock_t *busy = NULL;
1296 int size;
1297
1298 /* try to avoid the costly atomic add/sub pair when the receive
1299 * queue is full; always allow at least a packet
1300 */
1301 rmem = atomic_read(&sk->sk_rmem_alloc);
1302 if (rmem > sk->sk_rcvbuf)
1303 goto drop;
1304
1305 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1306 * having linear skbs :
1307 * - Reduce memory overhead and thus increase receive queue capacity
1308 * - Less cache line misses at copyout() time
1309 * - Less work at consume_skb() (less alien page frag freeing)
1310 */
1311 if (rmem > (sk->sk_rcvbuf >> 1)) {
1312 skb_condense(skb);
1313
1314 busy = busylock_acquire(sk);
1315 }
1316 size = skb->truesize;
1317 udp_set_dev_scratch(skb);
1318
1319 /* we drop only if the receive buf is full and the receive
1320 * queue contains some other skb
1321 */
1322 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1323 if (rmem > (size + sk->sk_rcvbuf))
1324 goto uncharge_drop;
1325
1326 spin_lock(&list->lock);
1327 if (size >= sk->sk_forward_alloc) {
1328 amt = sk_mem_pages(size);
1329 delta = amt << SK_MEM_QUANTUM_SHIFT;
1330 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1331 err = -ENOBUFS;
1332 spin_unlock(&list->lock);
1333 goto uncharge_drop;
1334 }
1335
1336 sk->sk_forward_alloc += delta;
1337 }
1338
1339 sk->sk_forward_alloc -= size;
1340
1341 /* no need to setup a destructor, we will explicitly release the
1342 * forward allocated memory on dequeue
1343 */
1344 sock_skb_set_dropcount(sk, skb);
1345
1346 __skb_queue_tail(list, skb);
1347 spin_unlock(&list->lock);
1348
1349 if (!sock_flag(sk, SOCK_DEAD))
1350 sk->sk_data_ready(sk);
1351
1352 busylock_release(busy);
1353 return 0;
1354
1355 uncharge_drop:
1356 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1357
1358 drop:
1359 atomic_inc(&sk->sk_drops);
1360 busylock_release(busy);
1361 return err;
1362 }
1363 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1364
1365 void udp_destruct_sock(struct sock *sk)
1366 {
1367 /* reclaim completely the forward allocated memory */
1368 struct udp_sock *up = udp_sk(sk);
1369 unsigned int total = 0;
1370 struct sk_buff *skb;
1371
1372 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1373 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1374 total += skb->truesize;
1375 kfree_skb(skb);
1376 }
1377 udp_rmem_release(sk, total, 0, true);
1378
1379 inet_sock_destruct(sk);
1380 }
1381 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1382
1383 int udp_init_sock(struct sock *sk)
1384 {
1385 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1386 sk->sk_destruct = udp_destruct_sock;
1387 return 0;
1388 }
1389 EXPORT_SYMBOL_GPL(udp_init_sock);
1390
1391 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1392 {
1393 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1394 bool slow = lock_sock_fast(sk);
1395
1396 sk_peek_offset_bwd(sk, len);
1397 unlock_sock_fast(sk, slow);
1398 }
1399
1400 if (!skb_unref(skb))
1401 return;
1402
1403 /* In the more common cases we cleared the head states previously,
1404 * see __udp_queue_rcv_skb().
1405 */
1406 if (unlikely(udp_skb_has_head_state(skb)))
1407 skb_release_head_state(skb);
1408 __consume_stateless_skb(skb);
1409 }
1410 EXPORT_SYMBOL_GPL(skb_consume_udp);
1411
1412 static struct sk_buff *__first_packet_length(struct sock *sk,
1413 struct sk_buff_head *rcvq,
1414 int *total)
1415 {
1416 struct sk_buff *skb;
1417
1418 while ((skb = skb_peek(rcvq)) != NULL) {
1419 if (udp_lib_checksum_complete(skb)) {
1420 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1421 IS_UDPLITE(sk));
1422 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1423 IS_UDPLITE(sk));
1424 atomic_inc(&sk->sk_drops);
1425 __skb_unlink(skb, rcvq);
1426 *total += skb->truesize;
1427 kfree_skb(skb);
1428 } else {
1429 /* the csum related bits could be changed, refresh
1430 * the scratch area
1431 */
1432 udp_set_dev_scratch(skb);
1433 break;
1434 }
1435 }
1436 return skb;
1437 }
1438
1439 /**
1440 * first_packet_length - return length of first packet in receive queue
1441 * @sk: socket
1442 *
1443 * Drops all bad checksum frames, until a valid one is found.
1444 * Returns the length of found skb, or -1 if none is found.
1445 */
1446 static int first_packet_length(struct sock *sk)
1447 {
1448 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1449 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1450 struct sk_buff *skb;
1451 int total = 0;
1452 int res;
1453
1454 spin_lock_bh(&rcvq->lock);
1455 skb = __first_packet_length(sk, rcvq, &total);
1456 if (!skb && !skb_queue_empty(sk_queue)) {
1457 spin_lock(&sk_queue->lock);
1458 skb_queue_splice_tail_init(sk_queue, rcvq);
1459 spin_unlock(&sk_queue->lock);
1460
1461 skb = __first_packet_length(sk, rcvq, &total);
1462 }
1463 res = skb ? skb->len : -1;
1464 if (total)
1465 udp_rmem_release(sk, total, 1, false);
1466 spin_unlock_bh(&rcvq->lock);
1467 return res;
1468 }
1469
1470 /*
1471 * IOCTL requests applicable to the UDP protocol
1472 */
1473
1474 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1475 {
1476 switch (cmd) {
1477 case SIOCOUTQ:
1478 {
1479 int amount = sk_wmem_alloc_get(sk);
1480
1481 return put_user(amount, (int __user *)arg);
1482 }
1483
1484 case SIOCINQ:
1485 {
1486 int amount = max_t(int, 0, first_packet_length(sk));
1487
1488 return put_user(amount, (int __user *)arg);
1489 }
1490
1491 default:
1492 return -ENOIOCTLCMD;
1493 }
1494
1495 return 0;
1496 }
1497 EXPORT_SYMBOL(udp_ioctl);
1498
1499 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1500 int noblock, int *peeked, int *off, int *err)
1501 {
1502 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1503 struct sk_buff_head *queue;
1504 struct sk_buff *last;
1505 long timeo;
1506 int error;
1507
1508 queue = &udp_sk(sk)->reader_queue;
1509 flags |= noblock ? MSG_DONTWAIT : 0;
1510 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1511 do {
1512 struct sk_buff *skb;
1513
1514 error = sock_error(sk);
1515 if (error)
1516 break;
1517
1518 error = -EAGAIN;
1519 *peeked = 0;
1520 do {
1521 spin_lock_bh(&queue->lock);
1522 skb = __skb_try_recv_from_queue(sk, queue, flags,
1523 udp_skb_destructor,
1524 peeked, off, err,
1525 &last);
1526 if (skb) {
1527 spin_unlock_bh(&queue->lock);
1528 return skb;
1529 }
1530
1531 if (skb_queue_empty(sk_queue)) {
1532 spin_unlock_bh(&queue->lock);
1533 goto busy_check;
1534 }
1535
1536 /* refill the reader queue and walk it again
1537 * keep both queues locked to avoid re-acquiring
1538 * the sk_receive_queue lock if fwd memory scheduling
1539 * is needed.
1540 */
1541 spin_lock(&sk_queue->lock);
1542 skb_queue_splice_tail_init(sk_queue, queue);
1543
1544 skb = __skb_try_recv_from_queue(sk, queue, flags,
1545 udp_skb_dtor_locked,
1546 peeked, off, err,
1547 &last);
1548 spin_unlock(&sk_queue->lock);
1549 spin_unlock_bh(&queue->lock);
1550 if (skb)
1551 return skb;
1552
1553 busy_check:
1554 if (!sk_can_busy_loop(sk))
1555 break;
1556
1557 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1558 } while (!skb_queue_empty(sk_queue));
1559
1560 /* sk_queue is empty, reader_queue may contain peeked packets */
1561 } while (timeo &&
1562 !__skb_wait_for_more_packets(sk, &error, &timeo,
1563 (struct sk_buff *)sk_queue));
1564
1565 *err = error;
1566 return NULL;
1567 }
1568 EXPORT_SYMBOL(__skb_recv_udp);
1569
1570 /*
1571 * This should be easy, if there is something there we
1572 * return it, otherwise we block.
1573 */
1574
1575 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1576 int flags, int *addr_len)
1577 {
1578 struct inet_sock *inet = inet_sk(sk);
1579 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1580 struct sk_buff *skb;
1581 unsigned int ulen, copied;
1582 int peeked, peeking, off;
1583 int err;
1584 int is_udplite = IS_UDPLITE(sk);
1585 bool checksum_valid = false;
1586
1587 if (flags & MSG_ERRQUEUE)
1588 return ip_recv_error(sk, msg, len, addr_len);
1589
1590 try_again:
1591 peeking = flags & MSG_PEEK;
1592 off = sk_peek_offset(sk, flags);
1593 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1594 if (!skb)
1595 return err;
1596
1597 ulen = udp_skb_len(skb);
1598 copied = len;
1599 if (copied > ulen - off)
1600 copied = ulen - off;
1601 else if (copied < ulen)
1602 msg->msg_flags |= MSG_TRUNC;
1603
1604 /*
1605 * If checksum is needed at all, try to do it while copying the
1606 * data. If the data is truncated, or if we only want a partial
1607 * coverage checksum (UDP-Lite), do it before the copy.
1608 */
1609
1610 if (copied < ulen || peeking ||
1611 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1612 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1613 !__udp_lib_checksum_complete(skb);
1614 if (!checksum_valid)
1615 goto csum_copy_err;
1616 }
1617
1618 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1619 if (udp_skb_is_linear(skb))
1620 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1621 else
1622 err = skb_copy_datagram_msg(skb, off, msg, copied);
1623 } else {
1624 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1625
1626 if (err == -EINVAL)
1627 goto csum_copy_err;
1628 }
1629
1630 if (unlikely(err)) {
1631 if (!peeked) {
1632 atomic_inc(&sk->sk_drops);
1633 UDP_INC_STATS(sock_net(sk),
1634 UDP_MIB_INERRORS, is_udplite);
1635 }
1636 kfree_skb(skb);
1637 return err;
1638 }
1639
1640 if (!peeked)
1641 UDP_INC_STATS(sock_net(sk),
1642 UDP_MIB_INDATAGRAMS, is_udplite);
1643
1644 sock_recv_ts_and_drops(msg, sk, skb);
1645
1646 /* Copy the address. */
1647 if (sin) {
1648 sin->sin_family = AF_INET;
1649 sin->sin_port = udp_hdr(skb)->source;
1650 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1651 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1652 *addr_len = sizeof(*sin);
1653 }
1654 if (inet->cmsg_flags)
1655 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1656
1657 err = copied;
1658 if (flags & MSG_TRUNC)
1659 err = ulen;
1660
1661 skb_consume_udp(sk, skb, peeking ? -err : err);
1662 return err;
1663
1664 csum_copy_err:
1665 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1666 udp_skb_destructor)) {
1667 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1668 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1669 }
1670 kfree_skb(skb);
1671
1672 /* starting over for a new packet, but check if we need to yield */
1673 cond_resched();
1674 msg->msg_flags &= ~MSG_TRUNC;
1675 goto try_again;
1676 }
1677
1678 int __udp_disconnect(struct sock *sk, int flags)
1679 {
1680 struct inet_sock *inet = inet_sk(sk);
1681 /*
1682 * 1003.1g - break association.
1683 */
1684
1685 sk->sk_state = TCP_CLOSE;
1686 inet->inet_daddr = 0;
1687 inet->inet_dport = 0;
1688 sock_rps_reset_rxhash(sk);
1689 sk->sk_bound_dev_if = 0;
1690 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1691 inet_reset_saddr(sk);
1692
1693 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1694 sk->sk_prot->unhash(sk);
1695 inet->inet_sport = 0;
1696 }
1697 sk_dst_reset(sk);
1698 return 0;
1699 }
1700 EXPORT_SYMBOL(__udp_disconnect);
1701
1702 int udp_disconnect(struct sock *sk, int flags)
1703 {
1704 lock_sock(sk);
1705 __udp_disconnect(sk, flags);
1706 release_sock(sk);
1707 return 0;
1708 }
1709 EXPORT_SYMBOL(udp_disconnect);
1710
1711 void udp_lib_unhash(struct sock *sk)
1712 {
1713 if (sk_hashed(sk)) {
1714 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1715 struct udp_hslot *hslot, *hslot2;
1716
1717 hslot = udp_hashslot(udptable, sock_net(sk),
1718 udp_sk(sk)->udp_port_hash);
1719 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1720
1721 spin_lock_bh(&hslot->lock);
1722 if (rcu_access_pointer(sk->sk_reuseport_cb))
1723 reuseport_detach_sock(sk);
1724 if (sk_del_node_init_rcu(sk)) {
1725 hslot->count--;
1726 inet_sk(sk)->inet_num = 0;
1727 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1728
1729 spin_lock(&hslot2->lock);
1730 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1731 hslot2->count--;
1732 spin_unlock(&hslot2->lock);
1733 }
1734 spin_unlock_bh(&hslot->lock);
1735 }
1736 }
1737 EXPORT_SYMBOL(udp_lib_unhash);
1738
1739 /*
1740 * inet_rcv_saddr was changed, we must rehash secondary hash
1741 */
1742 void udp_lib_rehash(struct sock *sk, u16 newhash)
1743 {
1744 if (sk_hashed(sk)) {
1745 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1746 struct udp_hslot *hslot, *hslot2, *nhslot2;
1747
1748 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1749 nhslot2 = udp_hashslot2(udptable, newhash);
1750 udp_sk(sk)->udp_portaddr_hash = newhash;
1751
1752 if (hslot2 != nhslot2 ||
1753 rcu_access_pointer(sk->sk_reuseport_cb)) {
1754 hslot = udp_hashslot(udptable, sock_net(sk),
1755 udp_sk(sk)->udp_port_hash);
1756 /* we must lock primary chain too */
1757 spin_lock_bh(&hslot->lock);
1758 if (rcu_access_pointer(sk->sk_reuseport_cb))
1759 reuseport_detach_sock(sk);
1760
1761 if (hslot2 != nhslot2) {
1762 spin_lock(&hslot2->lock);
1763 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1764 hslot2->count--;
1765 spin_unlock(&hslot2->lock);
1766
1767 spin_lock(&nhslot2->lock);
1768 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1769 &nhslot2->head);
1770 nhslot2->count++;
1771 spin_unlock(&nhslot2->lock);
1772 }
1773
1774 spin_unlock_bh(&hslot->lock);
1775 }
1776 }
1777 }
1778 EXPORT_SYMBOL(udp_lib_rehash);
1779
1780 static void udp_v4_rehash(struct sock *sk)
1781 {
1782 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1783 inet_sk(sk)->inet_rcv_saddr,
1784 inet_sk(sk)->inet_num);
1785 udp_lib_rehash(sk, new_hash);
1786 }
1787
1788 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1789 {
1790 int rc;
1791
1792 if (inet_sk(sk)->inet_daddr) {
1793 sock_rps_save_rxhash(sk, skb);
1794 sk_mark_napi_id(sk, skb);
1795 sk_incoming_cpu_update(sk);
1796 } else {
1797 sk_mark_napi_id_once(sk, skb);
1798 }
1799
1800 rc = __udp_enqueue_schedule_skb(sk, skb);
1801 if (rc < 0) {
1802 int is_udplite = IS_UDPLITE(sk);
1803
1804 /* Note that an ENOMEM error is charged twice */
1805 if (rc == -ENOMEM)
1806 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1807 is_udplite);
1808 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1809 kfree_skb(skb);
1810 trace_udp_fail_queue_rcv_skb(rc, sk);
1811 return -1;
1812 }
1813
1814 return 0;
1815 }
1816
1817 static struct static_key udp_encap_needed __read_mostly;
1818 void udp_encap_enable(void)
1819 {
1820 static_key_enable(&udp_encap_needed);
1821 }
1822 EXPORT_SYMBOL(udp_encap_enable);
1823
1824 /* returns:
1825 * -1: error
1826 * 0: success
1827 * >0: "udp encap" protocol resubmission
1828 *
1829 * Note that in the success and error cases, the skb is assumed to
1830 * have either been requeued or freed.
1831 */
1832 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1833 {
1834 struct udp_sock *up = udp_sk(sk);
1835 int is_udplite = IS_UDPLITE(sk);
1836
1837 /*
1838 * Charge it to the socket, dropping if the queue is full.
1839 */
1840 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1841 goto drop;
1842 nf_reset(skb);
1843
1844 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1845 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1846
1847 /*
1848 * This is an encapsulation socket so pass the skb to
1849 * the socket's udp_encap_rcv() hook. Otherwise, just
1850 * fall through and pass this up the UDP socket.
1851 * up->encap_rcv() returns the following value:
1852 * =0 if skb was successfully passed to the encap
1853 * handler or was discarded by it.
1854 * >0 if skb should be passed on to UDP.
1855 * <0 if skb should be resubmitted as proto -N
1856 */
1857
1858 /* if we're overly short, let UDP handle it */
1859 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1860 if (encap_rcv) {
1861 int ret;
1862
1863 /* Verify checksum before giving to encap */
1864 if (udp_lib_checksum_complete(skb))
1865 goto csum_error;
1866
1867 ret = encap_rcv(sk, skb);
1868 if (ret <= 0) {
1869 __UDP_INC_STATS(sock_net(sk),
1870 UDP_MIB_INDATAGRAMS,
1871 is_udplite);
1872 return -ret;
1873 }
1874 }
1875
1876 /* FALLTHROUGH -- it's a UDP Packet */
1877 }
1878
1879 /*
1880 * UDP-Lite specific tests, ignored on UDP sockets
1881 */
1882 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1883
1884 /*
1885 * MIB statistics other than incrementing the error count are
1886 * disabled for the following two types of errors: these depend
1887 * on the application settings, not on the functioning of the
1888 * protocol stack as such.
1889 *
1890 * RFC 3828 here recommends (sec 3.3): "There should also be a
1891 * way ... to ... at least let the receiving application block
1892 * delivery of packets with coverage values less than a value
1893 * provided by the application."
1894 */
1895 if (up->pcrlen == 0) { /* full coverage was set */
1896 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1897 UDP_SKB_CB(skb)->cscov, skb->len);
1898 goto drop;
1899 }
1900 /* The next case involves violating the min. coverage requested
1901 * by the receiver. This is subtle: if receiver wants x and x is
1902 * greater than the buffersize/MTU then receiver will complain
1903 * that it wants x while sender emits packets of smaller size y.
1904 * Therefore the above ...()->partial_cov statement is essential.
1905 */
1906 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1907 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1908 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1909 goto drop;
1910 }
1911 }
1912
1913 prefetch(&sk->sk_rmem_alloc);
1914 if (rcu_access_pointer(sk->sk_filter) &&
1915 udp_lib_checksum_complete(skb))
1916 goto csum_error;
1917
1918 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1919 goto drop;
1920
1921 udp_csum_pull_header(skb);
1922
1923 ipv4_pktinfo_prepare(sk, skb);
1924 return __udp_queue_rcv_skb(sk, skb);
1925
1926 csum_error:
1927 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1928 drop:
1929 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1930 atomic_inc(&sk->sk_drops);
1931 kfree_skb(skb);
1932 return -1;
1933 }
1934
1935 /* For TCP sockets, sk_rx_dst is protected by socket lock
1936 * For UDP, we use xchg() to guard against concurrent changes.
1937 */
1938 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1939 {
1940 struct dst_entry *old;
1941
1942 if (dst_hold_safe(dst)) {
1943 old = xchg(&sk->sk_rx_dst, dst);
1944 dst_release(old);
1945 return old != dst;
1946 }
1947 return false;
1948 }
1949 EXPORT_SYMBOL(udp_sk_rx_dst_set);
1950
1951 /*
1952 * Multicasts and broadcasts go to each listener.
1953 *
1954 * Note: called only from the BH handler context.
1955 */
1956 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1957 struct udphdr *uh,
1958 __be32 saddr, __be32 daddr,
1959 struct udp_table *udptable,
1960 int proto)
1961 {
1962 struct sock *sk, *first = NULL;
1963 unsigned short hnum = ntohs(uh->dest);
1964 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1965 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1966 unsigned int offset = offsetof(typeof(*sk), sk_node);
1967 int dif = skb->dev->ifindex;
1968 int sdif = inet_sdif(skb);
1969 struct hlist_node *node;
1970 struct sk_buff *nskb;
1971
1972 if (use_hash2) {
1973 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1974 udptable->mask;
1975 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1976 start_lookup:
1977 hslot = &udptable->hash2[hash2];
1978 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1979 }
1980
1981 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1982 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1983 uh->source, saddr, dif, sdif, hnum))
1984 continue;
1985
1986 if (!first) {
1987 first = sk;
1988 continue;
1989 }
1990 nskb = skb_clone(skb, GFP_ATOMIC);
1991
1992 if (unlikely(!nskb)) {
1993 atomic_inc(&sk->sk_drops);
1994 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1995 IS_UDPLITE(sk));
1996 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1997 IS_UDPLITE(sk));
1998 continue;
1999 }
2000 if (udp_queue_rcv_skb(sk, nskb) > 0)
2001 consume_skb(nskb);
2002 }
2003
2004 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2005 if (use_hash2 && hash2 != hash2_any) {
2006 hash2 = hash2_any;
2007 goto start_lookup;
2008 }
2009
2010 if (first) {
2011 if (udp_queue_rcv_skb(first, skb) > 0)
2012 consume_skb(skb);
2013 } else {
2014 kfree_skb(skb);
2015 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2016 proto == IPPROTO_UDPLITE);
2017 }
2018 return 0;
2019 }
2020
2021 /* Initialize UDP checksum. If exited with zero value (success),
2022 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2023 * Otherwise, csum completion requires chacksumming packet body,
2024 * including udp header and folding it to skb->csum.
2025 */
2026 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2027 int proto)
2028 {
2029 int err;
2030
2031 UDP_SKB_CB(skb)->partial_cov = 0;
2032 UDP_SKB_CB(skb)->cscov = skb->len;
2033
2034 if (proto == IPPROTO_UDPLITE) {
2035 err = udplite_checksum_init(skb, uh);
2036 if (err)
2037 return err;
2038
2039 if (UDP_SKB_CB(skb)->partial_cov) {
2040 skb->csum = inet_compute_pseudo(skb, proto);
2041 return 0;
2042 }
2043 }
2044
2045 /* Note, we are only interested in != 0 or == 0, thus the
2046 * force to int.
2047 */
2048 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2049 inet_compute_pseudo);
2050 if (err)
2051 return err;
2052
2053 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2054 /* If SW calculated the value, we know it's bad */
2055 if (skb->csum_complete_sw)
2056 return 1;
2057
2058 /* HW says the value is bad. Let's validate that.
2059 * skb->csum is no longer the full packet checksum,
2060 * so don't treat it as such.
2061 */
2062 skb_checksum_complete_unset(skb);
2063 }
2064
2065 return 0;
2066 }
2067
2068 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2069 * return code conversion for ip layer consumption
2070 */
2071 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2072 struct udphdr *uh)
2073 {
2074 int ret;
2075
2076 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2077 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2078 inet_compute_pseudo);
2079
2080 ret = udp_queue_rcv_skb(sk, skb);
2081
2082 /* a return value > 0 means to resubmit the input, but
2083 * it wants the return to be -protocol, or 0
2084 */
2085 if (ret > 0)
2086 return -ret;
2087 return 0;
2088 }
2089
2090 /*
2091 * All we need to do is get the socket, and then do a checksum.
2092 */
2093
2094 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2095 int proto)
2096 {
2097 struct sock *sk;
2098 struct udphdr *uh;
2099 unsigned short ulen;
2100 struct rtable *rt = skb_rtable(skb);
2101 __be32 saddr, daddr;
2102 struct net *net = dev_net(skb->dev);
2103
2104 /*
2105 * Validate the packet.
2106 */
2107 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2108 goto drop; /* No space for header. */
2109
2110 uh = udp_hdr(skb);
2111 ulen = ntohs(uh->len);
2112 saddr = ip_hdr(skb)->saddr;
2113 daddr = ip_hdr(skb)->daddr;
2114
2115 if (ulen > skb->len)
2116 goto short_packet;
2117
2118 if (proto == IPPROTO_UDP) {
2119 /* UDP validates ulen. */
2120 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2121 goto short_packet;
2122 uh = udp_hdr(skb);
2123 }
2124
2125 if (udp4_csum_init(skb, uh, proto))
2126 goto csum_error;
2127
2128 sk = skb_steal_sock(skb);
2129 if (sk) {
2130 struct dst_entry *dst = skb_dst(skb);
2131 int ret;
2132
2133 if (unlikely(sk->sk_rx_dst != dst))
2134 udp_sk_rx_dst_set(sk, dst);
2135
2136 ret = udp_unicast_rcv_skb(sk, skb, uh);
2137 sock_put(sk);
2138 return ret;
2139 }
2140
2141 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2142 return __udp4_lib_mcast_deliver(net, skb, uh,
2143 saddr, daddr, udptable, proto);
2144
2145 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2146 if (sk)
2147 return udp_unicast_rcv_skb(sk, skb, uh);
2148
2149 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2150 goto drop;
2151 nf_reset(skb);
2152
2153 /* No socket. Drop packet silently, if checksum is wrong */
2154 if (udp_lib_checksum_complete(skb))
2155 goto csum_error;
2156
2157 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2158 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2159
2160 /*
2161 * Hmm. We got an UDP packet to a port to which we
2162 * don't wanna listen. Ignore it.
2163 */
2164 kfree_skb(skb);
2165 return 0;
2166
2167 short_packet:
2168 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2169 proto == IPPROTO_UDPLITE ? "Lite" : "",
2170 &saddr, ntohs(uh->source),
2171 ulen, skb->len,
2172 &daddr, ntohs(uh->dest));
2173 goto drop;
2174
2175 csum_error:
2176 /*
2177 * RFC1122: OK. Discards the bad packet silently (as far as
2178 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2179 */
2180 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2181 proto == IPPROTO_UDPLITE ? "Lite" : "",
2182 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2183 ulen);
2184 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2185 drop:
2186 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2187 kfree_skb(skb);
2188 return 0;
2189 }
2190
2191 /* We can only early demux multicast if there is a single matching socket.
2192 * If more than one socket found returns NULL
2193 */
2194 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2195 __be16 loc_port, __be32 loc_addr,
2196 __be16 rmt_port, __be32 rmt_addr,
2197 int dif, int sdif)
2198 {
2199 struct sock *sk, *result;
2200 unsigned short hnum = ntohs(loc_port);
2201 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2202 struct udp_hslot *hslot = &udp_table.hash[slot];
2203
2204 /* Do not bother scanning a too big list */
2205 if (hslot->count > 10)
2206 return NULL;
2207
2208 result = NULL;
2209 sk_for_each_rcu(sk, &hslot->head) {
2210 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2211 rmt_port, rmt_addr, dif, sdif, hnum)) {
2212 if (result)
2213 return NULL;
2214 result = sk;
2215 }
2216 }
2217
2218 return result;
2219 }
2220
2221 /* For unicast we should only early demux connected sockets or we can
2222 * break forwarding setups. The chains here can be long so only check
2223 * if the first socket is an exact match and if not move on.
2224 */
2225 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2226 __be16 loc_port, __be32 loc_addr,
2227 __be16 rmt_port, __be32 rmt_addr,
2228 int dif, int sdif)
2229 {
2230 unsigned short hnum = ntohs(loc_port);
2231 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2232 unsigned int slot2 = hash2 & udp_table.mask;
2233 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2234 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2235 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2236 struct sock *sk;
2237
2238 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2239 if (INET_MATCH(sk, net, acookie, rmt_addr,
2240 loc_addr, ports, dif, sdif))
2241 return sk;
2242 /* Only check first socket in chain */
2243 break;
2244 }
2245 return NULL;
2246 }
2247
2248 int udp_v4_early_demux(struct sk_buff *skb)
2249 {
2250 struct net *net = dev_net(skb->dev);
2251 struct in_device *in_dev = NULL;
2252 const struct iphdr *iph;
2253 const struct udphdr *uh;
2254 struct sock *sk = NULL;
2255 struct dst_entry *dst;
2256 int dif = skb->dev->ifindex;
2257 int sdif = inet_sdif(skb);
2258 int ours;
2259
2260 /* validate the packet */
2261 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2262 return 0;
2263
2264 iph = ip_hdr(skb);
2265 uh = udp_hdr(skb);
2266
2267 if (skb->pkt_type == PACKET_MULTICAST) {
2268 in_dev = __in_dev_get_rcu(skb->dev);
2269
2270 if (!in_dev)
2271 return 0;
2272
2273 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2274 iph->protocol);
2275 if (!ours)
2276 return 0;
2277
2278 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2279 uh->source, iph->saddr,
2280 dif, sdif);
2281 } else if (skb->pkt_type == PACKET_HOST) {
2282 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2283 uh->source, iph->saddr, dif, sdif);
2284 }
2285
2286 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2287 return 0;
2288
2289 skb->sk = sk;
2290 skb->destructor = sock_efree;
2291 dst = READ_ONCE(sk->sk_rx_dst);
2292
2293 if (dst)
2294 dst = dst_check(dst, 0);
2295 if (dst) {
2296 u32 itag = 0;
2297
2298 /* set noref for now.
2299 * any place which wants to hold dst has to call
2300 * dst_hold_safe()
2301 */
2302 skb_dst_set_noref(skb, dst);
2303
2304 /* for unconnected multicast sockets we need to validate
2305 * the source on each packet
2306 */
2307 if (!inet_sk(sk)->inet_daddr && in_dev)
2308 return ip_mc_validate_source(skb, iph->daddr,
2309 iph->saddr, iph->tos,
2310 skb->dev, in_dev, &itag);
2311 }
2312 return 0;
2313 }
2314
2315 int udp_rcv(struct sk_buff *skb)
2316 {
2317 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2318 }
2319
2320 void udp_destroy_sock(struct sock *sk)
2321 {
2322 struct udp_sock *up = udp_sk(sk);
2323 bool slow = lock_sock_fast(sk);
2324 udp_flush_pending_frames(sk);
2325 unlock_sock_fast(sk, slow);
2326 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2327 void (*encap_destroy)(struct sock *sk);
2328 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2329 if (encap_destroy)
2330 encap_destroy(sk);
2331 }
2332 }
2333
2334 /*
2335 * Socket option code for UDP
2336 */
2337 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2338 char __user *optval, unsigned int optlen,
2339 int (*push_pending_frames)(struct sock *))
2340 {
2341 struct udp_sock *up = udp_sk(sk);
2342 int val, valbool;
2343 int err = 0;
2344 int is_udplite = IS_UDPLITE(sk);
2345
2346 if (optlen < sizeof(int))
2347 return -EINVAL;
2348
2349 if (get_user(val, (int __user *)optval))
2350 return -EFAULT;
2351
2352 valbool = val ? 1 : 0;
2353
2354 switch (optname) {
2355 case UDP_CORK:
2356 if (val != 0) {
2357 up->corkflag = 1;
2358 } else {
2359 up->corkflag = 0;
2360 lock_sock(sk);
2361 push_pending_frames(sk);
2362 release_sock(sk);
2363 }
2364 break;
2365
2366 case UDP_ENCAP:
2367 switch (val) {
2368 case 0:
2369 case UDP_ENCAP_ESPINUDP:
2370 case UDP_ENCAP_ESPINUDP_NON_IKE:
2371 up->encap_rcv = xfrm4_udp_encap_rcv;
2372 /* FALLTHROUGH */
2373 case UDP_ENCAP_L2TPINUDP:
2374 up->encap_type = val;
2375 udp_encap_enable();
2376 break;
2377 default:
2378 err = -ENOPROTOOPT;
2379 break;
2380 }
2381 break;
2382
2383 case UDP_NO_CHECK6_TX:
2384 up->no_check6_tx = valbool;
2385 break;
2386
2387 case UDP_NO_CHECK6_RX:
2388 up->no_check6_rx = valbool;
2389 break;
2390
2391 /*
2392 * UDP-Lite's partial checksum coverage (RFC 3828).
2393 */
2394 /* The sender sets actual checksum coverage length via this option.
2395 * The case coverage > packet length is handled by send module. */
2396 case UDPLITE_SEND_CSCOV:
2397 if (!is_udplite) /* Disable the option on UDP sockets */
2398 return -ENOPROTOOPT;
2399 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2400 val = 8;
2401 else if (val > USHRT_MAX)
2402 val = USHRT_MAX;
2403 up->pcslen = val;
2404 up->pcflag |= UDPLITE_SEND_CC;
2405 break;
2406
2407 /* The receiver specifies a minimum checksum coverage value. To make
2408 * sense, this should be set to at least 8 (as done below). If zero is
2409 * used, this again means full checksum coverage. */
2410 case UDPLITE_RECV_CSCOV:
2411 if (!is_udplite) /* Disable the option on UDP sockets */
2412 return -ENOPROTOOPT;
2413 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2414 val = 8;
2415 else if (val > USHRT_MAX)
2416 val = USHRT_MAX;
2417 up->pcrlen = val;
2418 up->pcflag |= UDPLITE_RECV_CC;
2419 break;
2420
2421 default:
2422 err = -ENOPROTOOPT;
2423 break;
2424 }
2425
2426 return err;
2427 }
2428 EXPORT_SYMBOL(udp_lib_setsockopt);
2429
2430 int udp_setsockopt(struct sock *sk, int level, int optname,
2431 char __user *optval, unsigned int optlen)
2432 {
2433 if (level == SOL_UDP || level == SOL_UDPLITE)
2434 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2435 udp_push_pending_frames);
2436 return ip_setsockopt(sk, level, optname, optval, optlen);
2437 }
2438
2439 #ifdef CONFIG_COMPAT
2440 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2441 char __user *optval, unsigned int optlen)
2442 {
2443 if (level == SOL_UDP || level == SOL_UDPLITE)
2444 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2445 udp_push_pending_frames);
2446 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2447 }
2448 #endif
2449
2450 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2451 char __user *optval, int __user *optlen)
2452 {
2453 struct udp_sock *up = udp_sk(sk);
2454 int val, len;
2455
2456 if (get_user(len, optlen))
2457 return -EFAULT;
2458
2459 len = min_t(unsigned int, len, sizeof(int));
2460
2461 if (len < 0)
2462 return -EINVAL;
2463
2464 switch (optname) {
2465 case UDP_CORK:
2466 val = up->corkflag;
2467 break;
2468
2469 case UDP_ENCAP:
2470 val = up->encap_type;
2471 break;
2472
2473 case UDP_NO_CHECK6_TX:
2474 val = up->no_check6_tx;
2475 break;
2476
2477 case UDP_NO_CHECK6_RX:
2478 val = up->no_check6_rx;
2479 break;
2480
2481 /* The following two cannot be changed on UDP sockets, the return is
2482 * always 0 (which corresponds to the full checksum coverage of UDP). */
2483 case UDPLITE_SEND_CSCOV:
2484 val = up->pcslen;
2485 break;
2486
2487 case UDPLITE_RECV_CSCOV:
2488 val = up->pcrlen;
2489 break;
2490
2491 default:
2492 return -ENOPROTOOPT;
2493 }
2494
2495 if (put_user(len, optlen))
2496 return -EFAULT;
2497 if (copy_to_user(optval, &val, len))
2498 return -EFAULT;
2499 return 0;
2500 }
2501 EXPORT_SYMBOL(udp_lib_getsockopt);
2502
2503 int udp_getsockopt(struct sock *sk, int level, int optname,
2504 char __user *optval, int __user *optlen)
2505 {
2506 if (level == SOL_UDP || level == SOL_UDPLITE)
2507 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2508 return ip_getsockopt(sk, level, optname, optval, optlen);
2509 }
2510
2511 #ifdef CONFIG_COMPAT
2512 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2513 char __user *optval, int __user *optlen)
2514 {
2515 if (level == SOL_UDP || level == SOL_UDPLITE)
2516 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2517 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2518 }
2519 #endif
2520 /**
2521 * udp_poll - wait for a UDP event.
2522 * @file - file struct
2523 * @sock - socket
2524 * @wait - poll table
2525 *
2526 * This is same as datagram poll, except for the special case of
2527 * blocking sockets. If application is using a blocking fd
2528 * and a packet with checksum error is in the queue;
2529 * then it could get return from select indicating data available
2530 * but then block when reading it. Add special case code
2531 * to work around these arguably broken applications.
2532 */
2533 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2534 {
2535 unsigned int mask = datagram_poll(file, sock, wait);
2536 struct sock *sk = sock->sk;
2537
2538 if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2539 mask |= POLLIN | POLLRDNORM;
2540
2541 sock_rps_record_flow(sk);
2542
2543 /* Check for false positives due to checksum errors */
2544 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2545 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2546 mask &= ~(POLLIN | POLLRDNORM);
2547
2548 return mask;
2549
2550 }
2551 EXPORT_SYMBOL(udp_poll);
2552
2553 int udp_abort(struct sock *sk, int err)
2554 {
2555 lock_sock(sk);
2556
2557 sk->sk_err = err;
2558 sk->sk_error_report(sk);
2559 __udp_disconnect(sk, 0);
2560
2561 release_sock(sk);
2562
2563 return 0;
2564 }
2565 EXPORT_SYMBOL_GPL(udp_abort);
2566
2567 struct proto udp_prot = {
2568 .name = "UDP",
2569 .owner = THIS_MODULE,
2570 .close = udp_lib_close,
2571 .connect = ip4_datagram_connect,
2572 .disconnect = udp_disconnect,
2573 .ioctl = udp_ioctl,
2574 .init = udp_init_sock,
2575 .destroy = udp_destroy_sock,
2576 .setsockopt = udp_setsockopt,
2577 .getsockopt = udp_getsockopt,
2578 .sendmsg = udp_sendmsg,
2579 .recvmsg = udp_recvmsg,
2580 .sendpage = udp_sendpage,
2581 .release_cb = ip4_datagram_release_cb,
2582 .hash = udp_lib_hash,
2583 .unhash = udp_lib_unhash,
2584 .rehash = udp_v4_rehash,
2585 .get_port = udp_v4_get_port,
2586 .memory_allocated = &udp_memory_allocated,
2587 .sysctl_mem = sysctl_udp_mem,
2588 .sysctl_wmem = &sysctl_udp_wmem_min,
2589 .sysctl_rmem = &sysctl_udp_rmem_min,
2590 .obj_size = sizeof(struct udp_sock),
2591 .h.udp_table = &udp_table,
2592 #ifdef CONFIG_COMPAT
2593 .compat_setsockopt = compat_udp_setsockopt,
2594 .compat_getsockopt = compat_udp_getsockopt,
2595 #endif
2596 .diag_destroy = udp_abort,
2597 };
2598 EXPORT_SYMBOL(udp_prot);
2599
2600 /* ------------------------------------------------------------------------ */
2601 #ifdef CONFIG_PROC_FS
2602
2603 static struct sock *udp_get_first(struct seq_file *seq, int start)
2604 {
2605 struct sock *sk;
2606 struct udp_iter_state *state = seq->private;
2607 struct net *net = seq_file_net(seq);
2608
2609 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2610 ++state->bucket) {
2611 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2612
2613 if (hlist_empty(&hslot->head))
2614 continue;
2615
2616 spin_lock_bh(&hslot->lock);
2617 sk_for_each(sk, &hslot->head) {
2618 if (!net_eq(sock_net(sk), net))
2619 continue;
2620 if (sk->sk_family == state->family)
2621 goto found;
2622 }
2623 spin_unlock_bh(&hslot->lock);
2624 }
2625 sk = NULL;
2626 found:
2627 return sk;
2628 }
2629
2630 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2631 {
2632 struct udp_iter_state *state = seq->private;
2633 struct net *net = seq_file_net(seq);
2634
2635 do {
2636 sk = sk_next(sk);
2637 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2638
2639 if (!sk) {
2640 if (state->bucket <= state->udp_table->mask)
2641 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2642 return udp_get_first(seq, state->bucket + 1);
2643 }
2644 return sk;
2645 }
2646
2647 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2648 {
2649 struct sock *sk = udp_get_first(seq, 0);
2650
2651 if (sk)
2652 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2653 --pos;
2654 return pos ? NULL : sk;
2655 }
2656
2657 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2658 {
2659 struct udp_iter_state *state = seq->private;
2660 state->bucket = MAX_UDP_PORTS;
2661
2662 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2663 }
2664
2665 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2666 {
2667 struct sock *sk;
2668
2669 if (v == SEQ_START_TOKEN)
2670 sk = udp_get_idx(seq, 0);
2671 else
2672 sk = udp_get_next(seq, v);
2673
2674 ++*pos;
2675 return sk;
2676 }
2677
2678 static void udp_seq_stop(struct seq_file *seq, void *v)
2679 {
2680 struct udp_iter_state *state = seq->private;
2681
2682 if (state->bucket <= state->udp_table->mask)
2683 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2684 }
2685
2686 int udp_seq_open(struct inode *inode, struct file *file)
2687 {
2688 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2689 struct udp_iter_state *s;
2690 int err;
2691
2692 err = seq_open_net(inode, file, &afinfo->seq_ops,
2693 sizeof(struct udp_iter_state));
2694 if (err < 0)
2695 return err;
2696
2697 s = ((struct seq_file *)file->private_data)->private;
2698 s->family = afinfo->family;
2699 s->udp_table = afinfo->udp_table;
2700 return err;
2701 }
2702 EXPORT_SYMBOL(udp_seq_open);
2703
2704 /* ------------------------------------------------------------------------ */
2705 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2706 {
2707 struct proc_dir_entry *p;
2708 int rc = 0;
2709
2710 afinfo->seq_ops.start = udp_seq_start;
2711 afinfo->seq_ops.next = udp_seq_next;
2712 afinfo->seq_ops.stop = udp_seq_stop;
2713
2714 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2715 afinfo->seq_fops, afinfo);
2716 if (!p)
2717 rc = -ENOMEM;
2718 return rc;
2719 }
2720 EXPORT_SYMBOL(udp_proc_register);
2721
2722 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2723 {
2724 remove_proc_entry(afinfo->name, net->proc_net);
2725 }
2726 EXPORT_SYMBOL(udp_proc_unregister);
2727
2728 /* ------------------------------------------------------------------------ */
2729 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2730 int bucket)
2731 {
2732 struct inet_sock *inet = inet_sk(sp);
2733 __be32 dest = inet->inet_daddr;
2734 __be32 src = inet->inet_rcv_saddr;
2735 __u16 destp = ntohs(inet->inet_dport);
2736 __u16 srcp = ntohs(inet->inet_sport);
2737
2738 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2739 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2740 bucket, src, srcp, dest, destp, sp->sk_state,
2741 sk_wmem_alloc_get(sp),
2742 udp_rqueue_get(sp),
2743 0, 0L, 0,
2744 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2745 0, sock_i_ino(sp),
2746 refcount_read(&sp->sk_refcnt), sp,
2747 atomic_read(&sp->sk_drops));
2748 }
2749
2750 int udp4_seq_show(struct seq_file *seq, void *v)
2751 {
2752 seq_setwidth(seq, 127);
2753 if (v == SEQ_START_TOKEN)
2754 seq_puts(seq, " sl local_address rem_address st tx_queue "
2755 "rx_queue tr tm->when retrnsmt uid timeout "
2756 "inode ref pointer drops");
2757 else {
2758 struct udp_iter_state *state = seq->private;
2759
2760 udp4_format_sock(v, seq, state->bucket);
2761 }
2762 seq_pad(seq, '\n');
2763 return 0;
2764 }
2765
2766 static const struct file_operations udp_afinfo_seq_fops = {
2767 .owner = THIS_MODULE,
2768 .open = udp_seq_open,
2769 .read = seq_read,
2770 .llseek = seq_lseek,
2771 .release = seq_release_net
2772 };
2773
2774 /* ------------------------------------------------------------------------ */
2775 static struct udp_seq_afinfo udp4_seq_afinfo = {
2776 .name = "udp",
2777 .family = AF_INET,
2778 .udp_table = &udp_table,
2779 .seq_fops = &udp_afinfo_seq_fops,
2780 .seq_ops = {
2781 .show = udp4_seq_show,
2782 },
2783 };
2784
2785 static int __net_init udp4_proc_init_net(struct net *net)
2786 {
2787 return udp_proc_register(net, &udp4_seq_afinfo);
2788 }
2789
2790 static void __net_exit udp4_proc_exit_net(struct net *net)
2791 {
2792 udp_proc_unregister(net, &udp4_seq_afinfo);
2793 }
2794
2795 static struct pernet_operations udp4_net_ops = {
2796 .init = udp4_proc_init_net,
2797 .exit = udp4_proc_exit_net,
2798 };
2799
2800 int __init udp4_proc_init(void)
2801 {
2802 return register_pernet_subsys(&udp4_net_ops);
2803 }
2804
2805 void udp4_proc_exit(void)
2806 {
2807 unregister_pernet_subsys(&udp4_net_ops);
2808 }
2809 #endif /* CONFIG_PROC_FS */
2810
2811 static __initdata unsigned long uhash_entries;
2812 static int __init set_uhash_entries(char *str)
2813 {
2814 ssize_t ret;
2815
2816 if (!str)
2817 return 0;
2818
2819 ret = kstrtoul(str, 0, &uhash_entries);
2820 if (ret)
2821 return 0;
2822
2823 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2824 uhash_entries = UDP_HTABLE_SIZE_MIN;
2825 return 1;
2826 }
2827 __setup("uhash_entries=", set_uhash_entries);
2828
2829 void __init udp_table_init(struct udp_table *table, const char *name)
2830 {
2831 unsigned int i;
2832
2833 table->hash = alloc_large_system_hash(name,
2834 2 * sizeof(struct udp_hslot),
2835 uhash_entries,
2836 21, /* one slot per 2 MB */
2837 0,
2838 &table->log,
2839 &table->mask,
2840 UDP_HTABLE_SIZE_MIN,
2841 64 * 1024);
2842
2843 table->hash2 = table->hash + (table->mask + 1);
2844 for (i = 0; i <= table->mask; i++) {
2845 INIT_HLIST_HEAD(&table->hash[i].head);
2846 table->hash[i].count = 0;
2847 spin_lock_init(&table->hash[i].lock);
2848 }
2849 for (i = 0; i <= table->mask; i++) {
2850 INIT_HLIST_HEAD(&table->hash2[i].head);
2851 table->hash2[i].count = 0;
2852 spin_lock_init(&table->hash2[i].lock);
2853 }
2854 }
2855
2856 u32 udp_flow_hashrnd(void)
2857 {
2858 static u32 hashrnd __read_mostly;
2859
2860 net_get_random_once(&hashrnd, sizeof(hashrnd));
2861
2862 return hashrnd;
2863 }
2864 EXPORT_SYMBOL(udp_flow_hashrnd);
2865
2866 void __init udp_init(void)
2867 {
2868 unsigned long limit;
2869 unsigned int i;
2870
2871 udp_table_init(&udp_table, "UDP");
2872 limit = nr_free_buffer_pages() / 8;
2873 limit = max(limit, 128UL);
2874 sysctl_udp_mem[0] = limit / 4 * 3;
2875 sysctl_udp_mem[1] = limit;
2876 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2877
2878 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2879 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2880
2881 /* 16 spinlocks per cpu */
2882 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2883 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2884 GFP_KERNEL);
2885 if (!udp_busylocks)
2886 panic("UDP: failed to alloc udp_busylocks\n");
2887 for (i = 0; i < (1U << udp_busylocks_log); i++)
2888 spin_lock_init(udp_busylocks + i);
2889 }
Linux with added FPC-III support