?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.18.10
[linux/fpc-iii.git] / net / ipv4 / udp.c
blob24e116ddae79ce0696e3f63290385ae15e28ac18
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 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 }
226 }
227
228 return reuseport_alloc(sk);
229 }
230
231 /**
232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
233 *
234 * @sk: socket struct in question
235 * @snum: port number to look up
236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237 * with NULL address
238 */
239 int udp_lib_get_port(struct sock *sk, unsigned short snum,
240 unsigned int hash2_nulladdr)
241 {
242 struct udp_hslot *hslot, *hslot2;
243 struct udp_table *udptable = sk->sk_prot->h.udp_table;
244 int error = 1;
245 struct net *net = sock_net(sk);
246
247 if (!snum) {
248 int low, high, remaining;
249 unsigned int rand;
250 unsigned short first, last;
251 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
252
253 inet_get_local_port_range(net, &low, &high);
254 remaining = (high - low) + 1;
255
256 rand = prandom_u32();
257 first = reciprocal_scale(rand, remaining) + low;
258 /*
259 * force rand to be an odd multiple of UDP_HTABLE_SIZE
260 */
261 rand = (rand | 1) * (udptable->mask + 1);
262 last = first + udptable->mask + 1;
263 do {
264 hslot = udp_hashslot(udptable, net, first);
265 bitmap_zero(bitmap, PORTS_PER_CHAIN);
266 spin_lock_bh(&hslot->lock);
267 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
268 udptable->log);
269
270 snum = first;
271 /*
272 * Iterate on all possible values of snum for this hash.
273 * Using steps of an odd multiple of UDP_HTABLE_SIZE
274 * give us randomization and full range coverage.
275 */
276 do {
277 if (low <= snum && snum <= high &&
278 !test_bit(snum >> udptable->log, bitmap) &&
279 !inet_is_local_reserved_port(net, snum))
280 goto found;
281 snum += rand;
282 } while (snum != first);
283 spin_unlock_bh(&hslot->lock);
284 cond_resched();
285 } while (++first != last);
286 goto fail;
287 } else {
288 hslot = udp_hashslot(udptable, net, snum);
289 spin_lock_bh(&hslot->lock);
290 if (hslot->count > 10) {
291 int exist;
292 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
293
294 slot2 &= udptable->mask;
295 hash2_nulladdr &= udptable->mask;
296
297 hslot2 = udp_hashslot2(udptable, slot2);
298 if (hslot->count < hslot2->count)
299 goto scan_primary_hash;
300
301 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
302 if (!exist && (hash2_nulladdr != slot2)) {
303 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
304 exist = udp_lib_lport_inuse2(net, snum, hslot2,
305 sk);
306 }
307 if (exist)
308 goto fail_unlock;
309 else
310 goto found;
311 }
312 scan_primary_hash:
313 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
314 goto fail_unlock;
315 }
316 found:
317 inet_sk(sk)->inet_num = snum;
318 udp_sk(sk)->udp_port_hash = snum;
319 udp_sk(sk)->udp_portaddr_hash ^= snum;
320 if (sk_unhashed(sk)) {
321 if (sk->sk_reuseport &&
322 udp_reuseport_add_sock(sk, hslot)) {
323 inet_sk(sk)->inet_num = 0;
324 udp_sk(sk)->udp_port_hash = 0;
325 udp_sk(sk)->udp_portaddr_hash ^= snum;
326 goto fail_unlock;
327 }
328
329 sk_add_node_rcu(sk, &hslot->head);
330 hslot->count++;
331 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
332
333 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
334 spin_lock(&hslot2->lock);
335 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
336 sk->sk_family == AF_INET6)
337 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
338 &hslot2->head);
339 else
340 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
341 &hslot2->head);
342 hslot2->count++;
343 spin_unlock(&hslot2->lock);
344 }
345 sock_set_flag(sk, SOCK_RCU_FREE);
346 error = 0;
347 fail_unlock:
348 spin_unlock_bh(&hslot->lock);
349 fail:
350 return error;
351 }
352 EXPORT_SYMBOL(udp_lib_get_port);
353
354 int udp_v4_get_port(struct sock *sk, unsigned short snum)
355 {
356 unsigned int hash2_nulladdr =
357 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
358 unsigned int hash2_partial =
359 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
360
361 /* precompute partial secondary hash */
362 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
363 return udp_lib_get_port(sk, snum, hash2_nulladdr);
364 }
365
366 static int compute_score(struct sock *sk, struct net *net,
367 __be32 saddr, __be16 sport,
368 __be32 daddr, unsigned short hnum,
369 int dif, int sdif, bool exact_dif)
370 {
371 int score;
372 struct inet_sock *inet;
373
374 if (!net_eq(sock_net(sk), net) ||
375 udp_sk(sk)->udp_port_hash != hnum ||
376 ipv6_only_sock(sk))
377 return -1;
378
379 score = (sk->sk_family == PF_INET) ? 2 : 1;
380 inet = inet_sk(sk);
381
382 if (inet->inet_rcv_saddr) {
383 if (inet->inet_rcv_saddr != daddr)
384 return -1;
385 score += 4;
386 }
387
388 if (inet->inet_daddr) {
389 if (inet->inet_daddr != saddr)
390 return -1;
391 score += 4;
392 }
393
394 if (inet->inet_dport) {
395 if (inet->inet_dport != sport)
396 return -1;
397 score += 4;
398 }
399
400 if (sk->sk_bound_dev_if || exact_dif) {
401 bool dev_match = (sk->sk_bound_dev_if == dif ||
402 sk->sk_bound_dev_if == sdif);
403
404 if (!dev_match)
405 return -1;
406 if (sk->sk_bound_dev_if)
407 score += 4;
408 }
409
410 if (sk->sk_incoming_cpu == raw_smp_processor_id())
411 score++;
412 return score;
413 }
414
415 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
416 const __u16 lport, const __be32 faddr,
417 const __be16 fport)
418 {
419 static u32 udp_ehash_secret __read_mostly;
420
421 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
422
423 return __inet_ehashfn(laddr, lport, faddr, fport,
424 udp_ehash_secret + net_hash_mix(net));
425 }
426
427 /* called with rcu_read_lock() */
428 static struct sock *udp4_lib_lookup2(struct net *net,
429 __be32 saddr, __be16 sport,
430 __be32 daddr, unsigned int hnum,
431 int dif, int sdif, bool exact_dif,
432 struct udp_hslot *hslot2,
433 struct sk_buff *skb)
434 {
435 struct sock *sk, *result;
436 int score, badness;
437 u32 hash = 0;
438
439 result = NULL;
440 badness = 0;
441 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
442 score = compute_score(sk, net, saddr, sport,
443 daddr, hnum, dif, sdif, exact_dif);
444 if (score > badness) {
445 if (sk->sk_reuseport) {
446 hash = udp_ehashfn(net, daddr, hnum,
447 saddr, sport);
448 result = reuseport_select_sock(sk, hash, skb,
449 sizeof(struct udphdr));
450 if (result)
451 return result;
452 }
453 badness = score;
454 result = sk;
455 }
456 }
457 return result;
458 }
459
460 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
461 * harder than this. -DaveM
462 */
463 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
464 __be16 sport, __be32 daddr, __be16 dport, int dif,
465 int sdif, struct udp_table *udptable, struct sk_buff *skb)
466 {
467 struct sock *sk, *result;
468 unsigned short hnum = ntohs(dport);
469 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
470 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
471 bool exact_dif = udp_lib_exact_dif_match(net, skb);
472 int score, badness;
473 u32 hash = 0;
474
475 if (hslot->count > 10) {
476 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
477 slot2 = hash2 & udptable->mask;
478 hslot2 = &udptable->hash2[slot2];
479 if (hslot->count < hslot2->count)
480 goto begin;
481
482 result = udp4_lib_lookup2(net, saddr, sport,
483 daddr, hnum, dif, sdif,
484 exact_dif, hslot2, skb);
485 if (!result) {
486 unsigned int old_slot2 = slot2;
487 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
488 slot2 = hash2 & udptable->mask;
489 /* avoid searching the same slot again. */
490 if (unlikely(slot2 == old_slot2))
491 return result;
492
493 hslot2 = &udptable->hash2[slot2];
494 if (hslot->count < hslot2->count)
495 goto begin;
496
497 result = udp4_lib_lookup2(net, saddr, sport,
498 daddr, hnum, dif, sdif,
499 exact_dif, hslot2, skb);
500 }
501 return result;
502 }
503 begin:
504 result = NULL;
505 badness = 0;
506 sk_for_each_rcu(sk, &hslot->head) {
507 score = compute_score(sk, net, saddr, sport,
508 daddr, hnum, dif, sdif, exact_dif);
509 if (score > badness) {
510 if (sk->sk_reuseport) {
511 hash = udp_ehashfn(net, daddr, hnum,
512 saddr, sport);
513 result = reuseport_select_sock(sk, hash, skb,
514 sizeof(struct udphdr));
515 if (result)
516 return result;
517 }
518 result = sk;
519 badness = score;
520 }
521 }
522 return result;
523 }
524 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
525
526 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
527 __be16 sport, __be16 dport,
528 struct udp_table *udptable)
529 {
530 const struct iphdr *iph = ip_hdr(skb);
531
532 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
533 iph->daddr, dport, inet_iif(skb),
534 inet_sdif(skb), udptable, skb);
535 }
536
537 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
538 __be16 sport, __be16 dport)
539 {
540 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
541 }
542 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
543
544 /* Must be called under rcu_read_lock().
545 * Does increment socket refcount.
546 */
547 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
548 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
549 __be32 daddr, __be16 dport, int dif)
550 {
551 struct sock *sk;
552
553 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
554 dif, 0, &udp_table, NULL);
555 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
556 sk = NULL;
557 return sk;
558 }
559 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
560 #endif
561
562 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
563 __be16 loc_port, __be32 loc_addr,
564 __be16 rmt_port, __be32 rmt_addr,
565 int dif, int sdif, unsigned short hnum)
566 {
567 struct inet_sock *inet = inet_sk(sk);
568
569 if (!net_eq(sock_net(sk), net) ||
570 udp_sk(sk)->udp_port_hash != hnum ||
571 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
572 (inet->inet_dport != rmt_port && inet->inet_dport) ||
573 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
574 ipv6_only_sock(sk) ||
575 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
576 sk->sk_bound_dev_if != sdif))
577 return false;
578 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
579 return false;
580 return true;
581 }
582
583 /*
584 * This routine is called by the ICMP module when it gets some
585 * sort of error condition. If err < 0 then the socket should
586 * be closed and the error returned to the user. If err > 0
587 * it's just the icmp type << 8 | icmp code.
588 * Header points to the ip header of the error packet. We move
589 * on past this. Then (as it used to claim before adjustment)
590 * header points to the first 8 bytes of the udp header. We need
591 * to find the appropriate port.
592 */
593
594 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
595 {
596 struct inet_sock *inet;
597 const struct iphdr *iph = (const struct iphdr *)skb->data;
598 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
599 const int type = icmp_hdr(skb)->type;
600 const int code = icmp_hdr(skb)->code;
601 struct sock *sk;
602 int harderr;
603 int err;
604 struct net *net = dev_net(skb->dev);
605
606 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
607 iph->saddr, uh->source, skb->dev->ifindex, 0,
608 udptable, NULL);
609 if (!sk) {
610 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
611 return; /* No socket for error */
612 }
613
614 err = 0;
615 harderr = 0;
616 inet = inet_sk(sk);
617
618 switch (type) {
619 default:
620 case ICMP_TIME_EXCEEDED:
621 err = EHOSTUNREACH;
622 break;
623 case ICMP_SOURCE_QUENCH:
624 goto out;
625 case ICMP_PARAMETERPROB:
626 err = EPROTO;
627 harderr = 1;
628 break;
629 case ICMP_DEST_UNREACH:
630 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
631 ipv4_sk_update_pmtu(skb, sk, info);
632 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
633 err = EMSGSIZE;
634 harderr = 1;
635 break;
636 }
637 goto out;
638 }
639 err = EHOSTUNREACH;
640 if (code <= NR_ICMP_UNREACH) {
641 harderr = icmp_err_convert[code].fatal;
642 err = icmp_err_convert[code].errno;
643 }
644 break;
645 case ICMP_REDIRECT:
646 ipv4_sk_redirect(skb, sk);
647 goto out;
648 }
649
650 /*
651 * RFC1122: OK. Passes ICMP errors back to application, as per
652 * 4.1.3.3.
653 */
654 if (!inet->recverr) {
655 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
656 goto out;
657 } else
658 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
659
660 sk->sk_err = err;
661 sk->sk_error_report(sk);
662 out:
663 return;
664 }
665
666 void udp_err(struct sk_buff *skb, u32 info)
667 {
668 __udp4_lib_err(skb, info, &udp_table);
669 }
670
671 /*
672 * Throw away all pending data and cancel the corking. Socket is locked.
673 */
674 void udp_flush_pending_frames(struct sock *sk)
675 {
676 struct udp_sock *up = udp_sk(sk);
677
678 if (up->pending) {
679 up->len = 0;
680 up->pending = 0;
681 ip_flush_pending_frames(sk);
682 }
683 }
684 EXPORT_SYMBOL(udp_flush_pending_frames);
685
686 /**
687 * udp4_hwcsum - handle outgoing HW checksumming
688 * @skb: sk_buff containing the filled-in UDP header
689 * (checksum field must be zeroed out)
690 * @src: source IP address
691 * @dst: destination IP address
692 */
693 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
694 {
695 struct udphdr *uh = udp_hdr(skb);
696 int offset = skb_transport_offset(skb);
697 int len = skb->len - offset;
698 int hlen = len;
699 __wsum csum = 0;
700
701 if (!skb_has_frag_list(skb)) {
702 /*
703 * Only one fragment on the socket.
704 */
705 skb->csum_start = skb_transport_header(skb) - skb->head;
706 skb->csum_offset = offsetof(struct udphdr, check);
707 uh->check = ~csum_tcpudp_magic(src, dst, len,
708 IPPROTO_UDP, 0);
709 } else {
710 struct sk_buff *frags;
711
712 /*
713 * HW-checksum won't work as there are two or more
714 * fragments on the socket so that all csums of sk_buffs
715 * should be together
716 */
717 skb_walk_frags(skb, frags) {
718 csum = csum_add(csum, frags->csum);
719 hlen -= frags->len;
720 }
721
722 csum = skb_checksum(skb, offset, hlen, csum);
723 skb->ip_summed = CHECKSUM_NONE;
724
725 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
726 if (uh->check == 0)
727 uh->check = CSUM_MANGLED_0;
728 }
729 }
730 EXPORT_SYMBOL_GPL(udp4_hwcsum);
731
732 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
733 * for the simple case like when setting the checksum for a UDP tunnel.
734 */
735 void udp_set_csum(bool nocheck, struct sk_buff *skb,
736 __be32 saddr, __be32 daddr, int len)
737 {
738 struct udphdr *uh = udp_hdr(skb);
739
740 if (nocheck) {
741 uh->check = 0;
742 } else if (skb_is_gso(skb)) {
743 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
744 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
745 uh->check = 0;
746 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
747 if (uh->check == 0)
748 uh->check = CSUM_MANGLED_0;
749 } else {
750 skb->ip_summed = CHECKSUM_PARTIAL;
751 skb->csum_start = skb_transport_header(skb) - skb->head;
752 skb->csum_offset = offsetof(struct udphdr, check);
753 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
754 }
755 }
756 EXPORT_SYMBOL(udp_set_csum);
757
758 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
759 struct inet_cork *cork)
760 {
761 struct sock *sk = skb->sk;
762 struct inet_sock *inet = inet_sk(sk);
763 struct udphdr *uh;
764 int err = 0;
765 int is_udplite = IS_UDPLITE(sk);
766 int offset = skb_transport_offset(skb);
767 int len = skb->len - offset;
768 __wsum csum = 0;
769
770 /*
771 * Create a UDP header
772 */
773 uh = udp_hdr(skb);
774 uh->source = inet->inet_sport;
775 uh->dest = fl4->fl4_dport;
776 uh->len = htons(len);
777 uh->check = 0;
778
779 if (cork->gso_size) {
780 const int hlen = skb_network_header_len(skb) +
781 sizeof(struct udphdr);
782
783 if (hlen + cork->gso_size > cork->fragsize)
784 return -EINVAL;
785 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS)
786 return -EINVAL;
787 if (sk->sk_no_check_tx)
788 return -EINVAL;
789 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
790 dst_xfrm(skb_dst(skb)))
791 return -EIO;
792
793 skb_shinfo(skb)->gso_size = cork->gso_size;
794 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
795 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh),
796 cork->gso_size);
797 goto csum_partial;
798 }
799
800 if (is_udplite) /* UDP-Lite */
801 csum = udplite_csum(skb);
802
803 else if (sk->sk_no_check_tx) { /* UDP csum off */
804
805 skb->ip_summed = CHECKSUM_NONE;
806 goto send;
807
808 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
809 csum_partial:
810
811 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
812 goto send;
813
814 } else
815 csum = udp_csum(skb);
816
817 /* add protocol-dependent pseudo-header */
818 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
819 sk->sk_protocol, csum);
820 if (uh->check == 0)
821 uh->check = CSUM_MANGLED_0;
822
823 send:
824 err = ip_send_skb(sock_net(sk), skb);
825 if (err) {
826 if (err == -ENOBUFS && !inet->recverr) {
827 UDP_INC_STATS(sock_net(sk),
828 UDP_MIB_SNDBUFERRORS, is_udplite);
829 err = 0;
830 }
831 } else
832 UDP_INC_STATS(sock_net(sk),
833 UDP_MIB_OUTDATAGRAMS, is_udplite);
834 return err;
835 }
836
837 /*
838 * Push out all pending data as one UDP datagram. Socket is locked.
839 */
840 int udp_push_pending_frames(struct sock *sk)
841 {
842 struct udp_sock *up = udp_sk(sk);
843 struct inet_sock *inet = inet_sk(sk);
844 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
845 struct sk_buff *skb;
846 int err = 0;
847
848 skb = ip_finish_skb(sk, fl4);
849 if (!skb)
850 goto out;
851
852 err = udp_send_skb(skb, fl4, &inet->cork.base);
853
854 out:
855 up->len = 0;
856 up->pending = 0;
857 return err;
858 }
859 EXPORT_SYMBOL(udp_push_pending_frames);
860
861 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
862 {
863 switch (cmsg->cmsg_type) {
864 case UDP_SEGMENT:
865 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
866 return -EINVAL;
867 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
868 return 0;
869 default:
870 return -EINVAL;
871 }
872 }
873
874 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
875 {
876 struct cmsghdr *cmsg;
877 bool need_ip = false;
878 int err;
879
880 for_each_cmsghdr(cmsg, msg) {
881 if (!CMSG_OK(msg, cmsg))
882 return -EINVAL;
883
884 if (cmsg->cmsg_level != SOL_UDP) {
885 need_ip = true;
886 continue;
887 }
888
889 err = __udp_cmsg_send(cmsg, gso_size);
890 if (err)
891 return err;
892 }
893
894 return need_ip;
895 }
896 EXPORT_SYMBOL_GPL(udp_cmsg_send);
897
898 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
899 {
900 struct inet_sock *inet = inet_sk(sk);
901 struct udp_sock *up = udp_sk(sk);
902 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
903 struct flowi4 fl4_stack;
904 struct flowi4 *fl4;
905 int ulen = len;
906 struct ipcm_cookie ipc;
907 struct rtable *rt = NULL;
908 int free = 0;
909 int connected = 0;
910 __be32 daddr, faddr, saddr;
911 __be16 dport;
912 u8 tos;
913 int err, is_udplite = IS_UDPLITE(sk);
914 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
915 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
916 struct sk_buff *skb;
917 struct ip_options_data opt_copy;
918
919 if (len > 0xFFFF)
920 return -EMSGSIZE;
921
922 /*
923 * Check the flags.
924 */
925
926 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
927 return -EOPNOTSUPP;
928
929 ipc.opt = NULL;
930 ipc.tx_flags = 0;
931 ipc.ttl = 0;
932 ipc.tos = -1;
933
934 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
935
936 fl4 = &inet->cork.fl.u.ip4;
937 if (up->pending) {
938 /*
939 * There are pending frames.
940 * The socket lock must be held while it's corked.
941 */
942 lock_sock(sk);
943 if (likely(up->pending)) {
944 if (unlikely(up->pending != AF_INET)) {
945 release_sock(sk);
946 return -EINVAL;
947 }
948 goto do_append_data;
949 }
950 release_sock(sk);
951 }
952 ulen += sizeof(struct udphdr);
953
954 /*
955 * Get and verify the address.
956 */
957 if (usin) {
958 if (msg->msg_namelen < sizeof(*usin))
959 return -EINVAL;
960 if (usin->sin_family != AF_INET) {
961 if (usin->sin_family != AF_UNSPEC)
962 return -EAFNOSUPPORT;
963 }
964
965 daddr = usin->sin_addr.s_addr;
966 dport = usin->sin_port;
967 if (dport == 0)
968 return -EINVAL;
969 } else {
970 if (sk->sk_state != TCP_ESTABLISHED)
971 return -EDESTADDRREQ;
972 daddr = inet->inet_daddr;
973 dport = inet->inet_dport;
974 /* Open fast path for connected socket.
975 Route will not be used, if at least one option is set.
976 */
977 connected = 1;
978 }
979
980 ipc.sockc.tsflags = sk->sk_tsflags;
981 ipc.addr = inet->inet_saddr;
982 ipc.oif = sk->sk_bound_dev_if;
983 ipc.gso_size = up->gso_size;
984
985 if (msg->msg_controllen) {
986 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
987 if (err > 0)
988 err = ip_cmsg_send(sk, msg, &ipc,
989 sk->sk_family == AF_INET6);
990 if (unlikely(err < 0)) {
991 kfree(ipc.opt);
992 return err;
993 }
994 if (ipc.opt)
995 free = 1;
996 connected = 0;
997 }
998 if (!ipc.opt) {
999 struct ip_options_rcu *inet_opt;
1000
1001 rcu_read_lock();
1002 inet_opt = rcu_dereference(inet->inet_opt);
1003 if (inet_opt) {
1004 memcpy(&opt_copy, inet_opt,
1005 sizeof(*inet_opt) + inet_opt->opt.optlen);
1006 ipc.opt = &opt_copy.opt;
1007 }
1008 rcu_read_unlock();
1009 }
1010
1011 if (cgroup_bpf_enabled && !connected) {
1012 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1013 (struct sockaddr *)usin, &ipc.addr);
1014 if (err)
1015 goto out_free;
1016 if (usin) {
1017 if (usin->sin_port == 0) {
1018 /* BPF program set invalid port. Reject it. */
1019 err = -EINVAL;
1020 goto out_free;
1021 }
1022 daddr = usin->sin_addr.s_addr;
1023 dport = usin->sin_port;
1024 }
1025 }
1026
1027 saddr = ipc.addr;
1028 ipc.addr = faddr = daddr;
1029
1030 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
1031
1032 if (ipc.opt && ipc.opt->opt.srr) {
1033 if (!daddr) {
1034 err = -EINVAL;
1035 goto out_free;
1036 }
1037 faddr = ipc.opt->opt.faddr;
1038 connected = 0;
1039 }
1040 tos = get_rttos(&ipc, inet);
1041 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1042 (msg->msg_flags & MSG_DONTROUTE) ||
1043 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1044 tos |= RTO_ONLINK;
1045 connected = 0;
1046 }
1047
1048 if (ipv4_is_multicast(daddr)) {
1049 if (!ipc.oif)
1050 ipc.oif = inet->mc_index;
1051 if (!saddr)
1052 saddr = inet->mc_addr;
1053 connected = 0;
1054 } else if (!ipc.oif) {
1055 ipc.oif = inet->uc_index;
1056 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1057 /* oif is set, packet is to local broadcast and
1058 * and uc_index is set. oif is most likely set
1059 * by sk_bound_dev_if. If uc_index != oif check if the
1060 * oif is an L3 master and uc_index is an L3 slave.
1061 * If so, we want to allow the send using the uc_index.
1062 */
1063 if (ipc.oif != inet->uc_index &&
1064 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1065 inet->uc_index)) {
1066 ipc.oif = inet->uc_index;
1067 }
1068 }
1069
1070 if (connected)
1071 rt = (struct rtable *)sk_dst_check(sk, 0);
1072
1073 if (!rt) {
1074 struct net *net = sock_net(sk);
1075 __u8 flow_flags = inet_sk_flowi_flags(sk);
1076
1077 fl4 = &fl4_stack;
1078
1079 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1080 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1081 flow_flags,
1082 faddr, saddr, dport, inet->inet_sport,
1083 sk->sk_uid);
1084
1085 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1086 rt = ip_route_output_flow(net, fl4, sk);
1087 if (IS_ERR(rt)) {
1088 err = PTR_ERR(rt);
1089 rt = NULL;
1090 if (err == -ENETUNREACH)
1091 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1092 goto out;
1093 }
1094
1095 err = -EACCES;
1096 if ((rt->rt_flags & RTCF_BROADCAST) &&
1097 !sock_flag(sk, SOCK_BROADCAST))
1098 goto out;
1099 if (connected)
1100 sk_dst_set(sk, dst_clone(&rt->dst));
1101 }
1102
1103 if (msg->msg_flags&MSG_CONFIRM)
1104 goto do_confirm;
1105 back_from_confirm:
1106
1107 saddr = fl4->saddr;
1108 if (!ipc.addr)
1109 daddr = ipc.addr = fl4->daddr;
1110
1111 /* Lockless fast path for the non-corking case. */
1112 if (!corkreq) {
1113 struct inet_cork cork;
1114
1115 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1116 sizeof(struct udphdr), &ipc, &rt,
1117 &cork, msg->msg_flags);
1118 err = PTR_ERR(skb);
1119 if (!IS_ERR_OR_NULL(skb))
1120 err = udp_send_skb(skb, fl4, &cork);
1121 goto out;
1122 }
1123
1124 lock_sock(sk);
1125 if (unlikely(up->pending)) {
1126 /* The socket is already corked while preparing it. */
1127 /* ... which is an evident application bug. --ANK */
1128 release_sock(sk);
1129
1130 net_dbg_ratelimited("socket already corked\n");
1131 err = -EINVAL;
1132 goto out;
1133 }
1134 /*
1135 * Now cork the socket to pend data.
1136 */
1137 fl4 = &inet->cork.fl.u.ip4;
1138 fl4->daddr = daddr;
1139 fl4->saddr = saddr;
1140 fl4->fl4_dport = dport;
1141 fl4->fl4_sport = inet->inet_sport;
1142 up->pending = AF_INET;
1143
1144 do_append_data:
1145 up->len += ulen;
1146 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1147 sizeof(struct udphdr), &ipc, &rt,
1148 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1149 if (err)
1150 udp_flush_pending_frames(sk);
1151 else if (!corkreq)
1152 err = udp_push_pending_frames(sk);
1153 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1154 up->pending = 0;
1155 release_sock(sk);
1156
1157 out:
1158 ip_rt_put(rt);
1159 out_free:
1160 if (free)
1161 kfree(ipc.opt);
1162 if (!err)
1163 return len;
1164 /*
1165 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1166 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1167 * we don't have a good statistic (IpOutDiscards but it can be too many
1168 * things). We could add another new stat but at least for now that
1169 * seems like overkill.
1170 */
1171 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1172 UDP_INC_STATS(sock_net(sk),
1173 UDP_MIB_SNDBUFERRORS, is_udplite);
1174 }
1175 return err;
1176
1177 do_confirm:
1178 if (msg->msg_flags & MSG_PROBE)
1179 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1180 if (!(msg->msg_flags&MSG_PROBE) || len)
1181 goto back_from_confirm;
1182 err = 0;
1183 goto out;
1184 }
1185 EXPORT_SYMBOL(udp_sendmsg);
1186
1187 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1188 size_t size, int flags)
1189 {
1190 struct inet_sock *inet = inet_sk(sk);
1191 struct udp_sock *up = udp_sk(sk);
1192 int ret;
1193
1194 if (flags & MSG_SENDPAGE_NOTLAST)
1195 flags |= MSG_MORE;
1196
1197 if (!up->pending) {
1198 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1199
1200 /* Call udp_sendmsg to specify destination address which
1201 * sendpage interface can't pass.
1202 * This will succeed only when the socket is connected.
1203 */
1204 ret = udp_sendmsg(sk, &msg, 0);
1205 if (ret < 0)
1206 return ret;
1207 }
1208
1209 lock_sock(sk);
1210
1211 if (unlikely(!up->pending)) {
1212 release_sock(sk);
1213
1214 net_dbg_ratelimited("cork failed\n");
1215 return -EINVAL;
1216 }
1217
1218 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1219 page, offset, size, flags);
1220 if (ret == -EOPNOTSUPP) {
1221 release_sock(sk);
1222 return sock_no_sendpage(sk->sk_socket, page, offset,
1223 size, flags);
1224 }
1225 if (ret < 0) {
1226 udp_flush_pending_frames(sk);
1227 goto out;
1228 }
1229
1230 up->len += size;
1231 if (!(up->corkflag || (flags&MSG_MORE)))
1232 ret = udp_push_pending_frames(sk);
1233 if (!ret)
1234 ret = size;
1235 out:
1236 release_sock(sk);
1237 return ret;
1238 }
1239
1240 #define UDP_SKB_IS_STATELESS 0x80000000
1241
1242 static void udp_set_dev_scratch(struct sk_buff *skb)
1243 {
1244 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1245
1246 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1247 scratch->_tsize_state = skb->truesize;
1248 #if BITS_PER_LONG == 64
1249 scratch->len = skb->len;
1250 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1251 scratch->is_linear = !skb_is_nonlinear(skb);
1252 #endif
1253 /* all head states execept sp (dst, sk, nf) are always cleared by
1254 * udp_rcv() and we need to preserve secpath, if present, to eventually
1255 * process IP_CMSG_PASSSEC at recvmsg() time
1256 */
1257 if (likely(!skb_sec_path(skb)))
1258 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1259 }
1260
1261 static int udp_skb_truesize(struct sk_buff *skb)
1262 {
1263 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1264 }
1265
1266 static bool udp_skb_has_head_state(struct sk_buff *skb)
1267 {
1268 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1269 }
1270
1271 /* fully reclaim rmem/fwd memory allocated for skb */
1272 static void udp_rmem_release(struct sock *sk, int size, int partial,
1273 bool rx_queue_lock_held)
1274 {
1275 struct udp_sock *up = udp_sk(sk);
1276 struct sk_buff_head *sk_queue;
1277 int amt;
1278
1279 if (likely(partial)) {
1280 up->forward_deficit += size;
1281 size = up->forward_deficit;
1282 if (size < (sk->sk_rcvbuf >> 2))
1283 return;
1284 } else {
1285 size += up->forward_deficit;
1286 }
1287 up->forward_deficit = 0;
1288
1289 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1290 * if the called don't held it already
1291 */
1292 sk_queue = &sk->sk_receive_queue;
1293 if (!rx_queue_lock_held)
1294 spin_lock(&sk_queue->lock);
1295
1296
1297 sk->sk_forward_alloc += size;
1298 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1299 sk->sk_forward_alloc -= amt;
1300
1301 if (amt)
1302 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1303
1304 atomic_sub(size, &sk->sk_rmem_alloc);
1305
1306 /* this can save us from acquiring the rx queue lock on next receive */
1307 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1308
1309 if (!rx_queue_lock_held)
1310 spin_unlock(&sk_queue->lock);
1311 }
1312
1313 /* Note: called with reader_queue.lock held.
1314 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1315 * This avoids a cache line miss while receive_queue lock is held.
1316 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1317 */
1318 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1319 {
1320 prefetch(&skb->data);
1321 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1322 }
1323 EXPORT_SYMBOL(udp_skb_destructor);
1324
1325 /* as above, but the caller held the rx queue lock, too */
1326 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1327 {
1328 prefetch(&skb->data);
1329 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1330 }
1331
1332 /* Idea of busylocks is to let producers grab an extra spinlock
1333 * to relieve pressure on the receive_queue spinlock shared by consumer.
1334 * Under flood, this means that only one producer can be in line
1335 * trying to acquire the receive_queue spinlock.
1336 * These busylock can be allocated on a per cpu manner, instead of a
1337 * per socket one (that would consume a cache line per socket)
1338 */
1339 static int udp_busylocks_log __read_mostly;
1340 static spinlock_t *udp_busylocks __read_mostly;
1341
1342 static spinlock_t *busylock_acquire(void *ptr)
1343 {
1344 spinlock_t *busy;
1345
1346 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1347 spin_lock(busy);
1348 return busy;
1349 }
1350
1351 static void busylock_release(spinlock_t *busy)
1352 {
1353 if (busy)
1354 spin_unlock(busy);
1355 }
1356
1357 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1358 {
1359 struct sk_buff_head *list = &sk->sk_receive_queue;
1360 int rmem, delta, amt, err = -ENOMEM;
1361 spinlock_t *busy = NULL;
1362 int size;
1363
1364 /* try to avoid the costly atomic add/sub pair when the receive
1365 * queue is full; always allow at least a packet
1366 */
1367 rmem = atomic_read(&sk->sk_rmem_alloc);
1368 if (rmem > sk->sk_rcvbuf)
1369 goto drop;
1370
1371 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1372 * having linear skbs :
1373 * - Reduce memory overhead and thus increase receive queue capacity
1374 * - Less cache line misses at copyout() time
1375 * - Less work at consume_skb() (less alien page frag freeing)
1376 */
1377 if (rmem > (sk->sk_rcvbuf >> 1)) {
1378 skb_condense(skb);
1379
1380 busy = busylock_acquire(sk);
1381 }
1382 size = skb->truesize;
1383 udp_set_dev_scratch(skb);
1384
1385 /* we drop only if the receive buf is full and the receive
1386 * queue contains some other skb
1387 */
1388 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1389 if (rmem > (size + sk->sk_rcvbuf))
1390 goto uncharge_drop;
1391
1392 spin_lock(&list->lock);
1393 if (size >= sk->sk_forward_alloc) {
1394 amt = sk_mem_pages(size);
1395 delta = amt << SK_MEM_QUANTUM_SHIFT;
1396 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1397 err = -ENOBUFS;
1398 spin_unlock(&list->lock);
1399 goto uncharge_drop;
1400 }
1401
1402 sk->sk_forward_alloc += delta;
1403 }
1404
1405 sk->sk_forward_alloc -= size;
1406
1407 /* no need to setup a destructor, we will explicitly release the
1408 * forward allocated memory on dequeue
1409 */
1410 sock_skb_set_dropcount(sk, skb);
1411
1412 __skb_queue_tail(list, skb);
1413 spin_unlock(&list->lock);
1414
1415 if (!sock_flag(sk, SOCK_DEAD))
1416 sk->sk_data_ready(sk);
1417
1418 busylock_release(busy);
1419 return 0;
1420
1421 uncharge_drop:
1422 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1423
1424 drop:
1425 atomic_inc(&sk->sk_drops);
1426 busylock_release(busy);
1427 return err;
1428 }
1429 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1430
1431 void udp_destruct_sock(struct sock *sk)
1432 {
1433 /* reclaim completely the forward allocated memory */
1434 struct udp_sock *up = udp_sk(sk);
1435 unsigned int total = 0;
1436 struct sk_buff *skb;
1437
1438 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1439 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1440 total += skb->truesize;
1441 kfree_skb(skb);
1442 }
1443 udp_rmem_release(sk, total, 0, true);
1444
1445 inet_sock_destruct(sk);
1446 }
1447 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1448
1449 int udp_init_sock(struct sock *sk)
1450 {
1451 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1452 sk->sk_destruct = udp_destruct_sock;
1453 return 0;
1454 }
1455 EXPORT_SYMBOL_GPL(udp_init_sock);
1456
1457 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1458 {
1459 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1460 bool slow = lock_sock_fast(sk);
1461
1462 sk_peek_offset_bwd(sk, len);
1463 unlock_sock_fast(sk, slow);
1464 }
1465
1466 if (!skb_unref(skb))
1467 return;
1468
1469 /* In the more common cases we cleared the head states previously,
1470 * see __udp_queue_rcv_skb().
1471 */
1472 if (unlikely(udp_skb_has_head_state(skb)))
1473 skb_release_head_state(skb);
1474 __consume_stateless_skb(skb);
1475 }
1476 EXPORT_SYMBOL_GPL(skb_consume_udp);
1477
1478 static struct sk_buff *__first_packet_length(struct sock *sk,
1479 struct sk_buff_head *rcvq,
1480 int *total)
1481 {
1482 struct sk_buff *skb;
1483
1484 while ((skb = skb_peek(rcvq)) != NULL) {
1485 if (udp_lib_checksum_complete(skb)) {
1486 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1487 IS_UDPLITE(sk));
1488 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1489 IS_UDPLITE(sk));
1490 atomic_inc(&sk->sk_drops);
1491 __skb_unlink(skb, rcvq);
1492 *total += skb->truesize;
1493 kfree_skb(skb);
1494 } else {
1495 /* the csum related bits could be changed, refresh
1496 * the scratch area
1497 */
1498 udp_set_dev_scratch(skb);
1499 break;
1500 }
1501 }
1502 return skb;
1503 }
1504
1505 /**
1506 * first_packet_length - return length of first packet in receive queue
1507 * @sk: socket
1508 *
1509 * Drops all bad checksum frames, until a valid one is found.
1510 * Returns the length of found skb, or -1 if none is found.
1511 */
1512 static int first_packet_length(struct sock *sk)
1513 {
1514 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1515 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1516 struct sk_buff *skb;
1517 int total = 0;
1518 int res;
1519
1520 spin_lock_bh(&rcvq->lock);
1521 skb = __first_packet_length(sk, rcvq, &total);
1522 if (!skb && !skb_queue_empty(sk_queue)) {
1523 spin_lock(&sk_queue->lock);
1524 skb_queue_splice_tail_init(sk_queue, rcvq);
1525 spin_unlock(&sk_queue->lock);
1526
1527 skb = __first_packet_length(sk, rcvq, &total);
1528 }
1529 res = skb ? skb->len : -1;
1530 if (total)
1531 udp_rmem_release(sk, total, 1, false);
1532 spin_unlock_bh(&rcvq->lock);
1533 return res;
1534 }
1535
1536 /*
1537 * IOCTL requests applicable to the UDP protocol
1538 */
1539
1540 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1541 {
1542 switch (cmd) {
1543 case SIOCOUTQ:
1544 {
1545 int amount = sk_wmem_alloc_get(sk);
1546
1547 return put_user(amount, (int __user *)arg);
1548 }
1549
1550 case SIOCINQ:
1551 {
1552 int amount = max_t(int, 0, first_packet_length(sk));
1553
1554 return put_user(amount, (int __user *)arg);
1555 }
1556
1557 default:
1558 return -ENOIOCTLCMD;
1559 }
1560
1561 return 0;
1562 }
1563 EXPORT_SYMBOL(udp_ioctl);
1564
1565 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1566 int noblock, int *peeked, int *off, int *err)
1567 {
1568 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1569 struct sk_buff_head *queue;
1570 struct sk_buff *last;
1571 long timeo;
1572 int error;
1573
1574 queue = &udp_sk(sk)->reader_queue;
1575 flags |= noblock ? MSG_DONTWAIT : 0;
1576 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1577 do {
1578 struct sk_buff *skb;
1579
1580 error = sock_error(sk);
1581 if (error)
1582 break;
1583
1584 error = -EAGAIN;
1585 *peeked = 0;
1586 do {
1587 spin_lock_bh(&queue->lock);
1588 skb = __skb_try_recv_from_queue(sk, queue, flags,
1589 udp_skb_destructor,
1590 peeked, off, err,
1591 &last);
1592 if (skb) {
1593 spin_unlock_bh(&queue->lock);
1594 return skb;
1595 }
1596
1597 if (skb_queue_empty(sk_queue)) {
1598 spin_unlock_bh(&queue->lock);
1599 goto busy_check;
1600 }
1601
1602 /* refill the reader queue and walk it again
1603 * keep both queues locked to avoid re-acquiring
1604 * the sk_receive_queue lock if fwd memory scheduling
1605 * is needed.
1606 */
1607 spin_lock(&sk_queue->lock);
1608 skb_queue_splice_tail_init(sk_queue, queue);
1609
1610 skb = __skb_try_recv_from_queue(sk, queue, flags,
1611 udp_skb_dtor_locked,
1612 peeked, off, err,
1613 &last);
1614 spin_unlock(&sk_queue->lock);
1615 spin_unlock_bh(&queue->lock);
1616 if (skb)
1617 return skb;
1618
1619 busy_check:
1620 if (!sk_can_busy_loop(sk))
1621 break;
1622
1623 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1624 } while (!skb_queue_empty(sk_queue));
1625
1626 /* sk_queue is empty, reader_queue may contain peeked packets */
1627 } while (timeo &&
1628 !__skb_wait_for_more_packets(sk, &error, &timeo,
1629 (struct sk_buff *)sk_queue));
1630
1631 *err = error;
1632 return NULL;
1633 }
1634 EXPORT_SYMBOL_GPL(__skb_recv_udp);
1635
1636 /*
1637 * This should be easy, if there is something there we
1638 * return it, otherwise we block.
1639 */
1640
1641 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1642 int flags, int *addr_len)
1643 {
1644 struct inet_sock *inet = inet_sk(sk);
1645 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1646 struct sk_buff *skb;
1647 unsigned int ulen, copied;
1648 int peeked, peeking, off;
1649 int err;
1650 int is_udplite = IS_UDPLITE(sk);
1651 bool checksum_valid = false;
1652
1653 if (flags & MSG_ERRQUEUE)
1654 return ip_recv_error(sk, msg, len, addr_len);
1655
1656 try_again:
1657 peeking = flags & MSG_PEEK;
1658 off = sk_peek_offset(sk, flags);
1659 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1660 if (!skb)
1661 return err;
1662
1663 ulen = udp_skb_len(skb);
1664 copied = len;
1665 if (copied > ulen - off)
1666 copied = ulen - off;
1667 else if (copied < ulen)
1668 msg->msg_flags |= MSG_TRUNC;
1669
1670 /*
1671 * If checksum is needed at all, try to do it while copying the
1672 * data. If the data is truncated, or if we only want a partial
1673 * coverage checksum (UDP-Lite), do it before the copy.
1674 */
1675
1676 if (copied < ulen || peeking ||
1677 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1678 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1679 !__udp_lib_checksum_complete(skb);
1680 if (!checksum_valid)
1681 goto csum_copy_err;
1682 }
1683
1684 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1685 if (udp_skb_is_linear(skb))
1686 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1687 else
1688 err = skb_copy_datagram_msg(skb, off, msg, copied);
1689 } else {
1690 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1691
1692 if (err == -EINVAL)
1693 goto csum_copy_err;
1694 }
1695
1696 if (unlikely(err)) {
1697 if (!peeked) {
1698 atomic_inc(&sk->sk_drops);
1699 UDP_INC_STATS(sock_net(sk),
1700 UDP_MIB_INERRORS, is_udplite);
1701 }
1702 kfree_skb(skb);
1703 return err;
1704 }
1705
1706 if (!peeked)
1707 UDP_INC_STATS(sock_net(sk),
1708 UDP_MIB_INDATAGRAMS, is_udplite);
1709
1710 sock_recv_ts_and_drops(msg, sk, skb);
1711
1712 /* Copy the address. */
1713 if (sin) {
1714 sin->sin_family = AF_INET;
1715 sin->sin_port = udp_hdr(skb)->source;
1716 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1717 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1718 *addr_len = sizeof(*sin);
1719 }
1720 if (inet->cmsg_flags)
1721 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1722
1723 err = copied;
1724 if (flags & MSG_TRUNC)
1725 err = ulen;
1726
1727 skb_consume_udp(sk, skb, peeking ? -err : err);
1728 return err;
1729
1730 csum_copy_err:
1731 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1732 udp_skb_destructor)) {
1733 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1734 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1735 }
1736 kfree_skb(skb);
1737
1738 /* starting over for a new packet, but check if we need to yield */
1739 cond_resched();
1740 msg->msg_flags &= ~MSG_TRUNC;
1741 goto try_again;
1742 }
1743
1744 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1745 {
1746 /* This check is replicated from __ip4_datagram_connect() and
1747 * intended to prevent BPF program called below from accessing bytes
1748 * that are out of the bound specified by user in addr_len.
1749 */
1750 if (addr_len < sizeof(struct sockaddr_in))
1751 return -EINVAL;
1752
1753 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1754 }
1755 EXPORT_SYMBOL(udp_pre_connect);
1756
1757 int __udp_disconnect(struct sock *sk, int flags)
1758 {
1759 struct inet_sock *inet = inet_sk(sk);
1760 /*
1761 * 1003.1g - break association.
1762 */
1763
1764 sk->sk_state = TCP_CLOSE;
1765 inet->inet_daddr = 0;
1766 inet->inet_dport = 0;
1767 sock_rps_reset_rxhash(sk);
1768 sk->sk_bound_dev_if = 0;
1769 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1770 inet_reset_saddr(sk);
1771
1772 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1773 sk->sk_prot->unhash(sk);
1774 inet->inet_sport = 0;
1775 }
1776 sk_dst_reset(sk);
1777 return 0;
1778 }
1779 EXPORT_SYMBOL(__udp_disconnect);
1780
1781 int udp_disconnect(struct sock *sk, int flags)
1782 {
1783 lock_sock(sk);
1784 __udp_disconnect(sk, flags);
1785 release_sock(sk);
1786 return 0;
1787 }
1788 EXPORT_SYMBOL(udp_disconnect);
1789
1790 void udp_lib_unhash(struct sock *sk)
1791 {
1792 if (sk_hashed(sk)) {
1793 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1794 struct udp_hslot *hslot, *hslot2;
1795
1796 hslot = udp_hashslot(udptable, sock_net(sk),
1797 udp_sk(sk)->udp_port_hash);
1798 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1799
1800 spin_lock_bh(&hslot->lock);
1801 if (rcu_access_pointer(sk->sk_reuseport_cb))
1802 reuseport_detach_sock(sk);
1803 if (sk_del_node_init_rcu(sk)) {
1804 hslot->count--;
1805 inet_sk(sk)->inet_num = 0;
1806 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1807
1808 spin_lock(&hslot2->lock);
1809 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1810 hslot2->count--;
1811 spin_unlock(&hslot2->lock);
1812 }
1813 spin_unlock_bh(&hslot->lock);
1814 }
1815 }
1816 EXPORT_SYMBOL(udp_lib_unhash);
1817
1818 /*
1819 * inet_rcv_saddr was changed, we must rehash secondary hash
1820 */
1821 void udp_lib_rehash(struct sock *sk, u16 newhash)
1822 {
1823 if (sk_hashed(sk)) {
1824 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1825 struct udp_hslot *hslot, *hslot2, *nhslot2;
1826
1827 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1828 nhslot2 = udp_hashslot2(udptable, newhash);
1829 udp_sk(sk)->udp_portaddr_hash = newhash;
1830
1831 if (hslot2 != nhslot2 ||
1832 rcu_access_pointer(sk->sk_reuseport_cb)) {
1833 hslot = udp_hashslot(udptable, sock_net(sk),
1834 udp_sk(sk)->udp_port_hash);
1835 /* we must lock primary chain too */
1836 spin_lock_bh(&hslot->lock);
1837 if (rcu_access_pointer(sk->sk_reuseport_cb))
1838 reuseport_detach_sock(sk);
1839
1840 if (hslot2 != nhslot2) {
1841 spin_lock(&hslot2->lock);
1842 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1843 hslot2->count--;
1844 spin_unlock(&hslot2->lock);
1845
1846 spin_lock(&nhslot2->lock);
1847 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1848 &nhslot2->head);
1849 nhslot2->count++;
1850 spin_unlock(&nhslot2->lock);
1851 }
1852
1853 spin_unlock_bh(&hslot->lock);
1854 }
1855 }
1856 }
1857 EXPORT_SYMBOL(udp_lib_rehash);
1858
1859 static void udp_v4_rehash(struct sock *sk)
1860 {
1861 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1862 inet_sk(sk)->inet_rcv_saddr,
1863 inet_sk(sk)->inet_num);
1864 udp_lib_rehash(sk, new_hash);
1865 }
1866
1867 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1868 {
1869 int rc;
1870
1871 if (inet_sk(sk)->inet_daddr) {
1872 sock_rps_save_rxhash(sk, skb);
1873 sk_mark_napi_id(sk, skb);
1874 sk_incoming_cpu_update(sk);
1875 } else {
1876 sk_mark_napi_id_once(sk, skb);
1877 }
1878
1879 rc = __udp_enqueue_schedule_skb(sk, skb);
1880 if (rc < 0) {
1881 int is_udplite = IS_UDPLITE(sk);
1882
1883 /* Note that an ENOMEM error is charged twice */
1884 if (rc == -ENOMEM)
1885 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1886 is_udplite);
1887 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1888 kfree_skb(skb);
1889 trace_udp_fail_queue_rcv_skb(rc, sk);
1890 return -1;
1891 }
1892
1893 return 0;
1894 }
1895
1896 static DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
1897 void udp_encap_enable(void)
1898 {
1899 static_branch_enable(&udp_encap_needed_key);
1900 }
1901 EXPORT_SYMBOL(udp_encap_enable);
1902
1903 /* returns:
1904 * -1: error
1905 * 0: success
1906 * >0: "udp encap" protocol resubmission
1907 *
1908 * Note that in the success and error cases, the skb is assumed to
1909 * have either been requeued or freed.
1910 */
1911 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1912 {
1913 struct udp_sock *up = udp_sk(sk);
1914 int is_udplite = IS_UDPLITE(sk);
1915
1916 /*
1917 * Charge it to the socket, dropping if the queue is full.
1918 */
1919 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1920 goto drop;
1921 nf_reset(skb);
1922
1923 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
1924 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1925
1926 /*
1927 * This is an encapsulation socket so pass the skb to
1928 * the socket's udp_encap_rcv() hook. Otherwise, just
1929 * fall through and pass this up the UDP socket.
1930 * up->encap_rcv() returns the following value:
1931 * =0 if skb was successfully passed to the encap
1932 * handler or was discarded by it.
1933 * >0 if skb should be passed on to UDP.
1934 * <0 if skb should be resubmitted as proto -N
1935 */
1936
1937 /* if we're overly short, let UDP handle it */
1938 encap_rcv = READ_ONCE(up->encap_rcv);
1939 if (encap_rcv) {
1940 int ret;
1941
1942 /* Verify checksum before giving to encap */
1943 if (udp_lib_checksum_complete(skb))
1944 goto csum_error;
1945
1946 ret = encap_rcv(sk, skb);
1947 if (ret <= 0) {
1948 __UDP_INC_STATS(sock_net(sk),
1949 UDP_MIB_INDATAGRAMS,
1950 is_udplite);
1951 return -ret;
1952 }
1953 }
1954
1955 /* FALLTHROUGH -- it's a UDP Packet */
1956 }
1957
1958 /*
1959 * UDP-Lite specific tests, ignored on UDP sockets
1960 */
1961 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1962
1963 /*
1964 * MIB statistics other than incrementing the error count are
1965 * disabled for the following two types of errors: these depend
1966 * on the application settings, not on the functioning of the
1967 * protocol stack as such.
1968 *
1969 * RFC 3828 here recommends (sec 3.3): "There should also be a
1970 * way ... to ... at least let the receiving application block
1971 * delivery of packets with coverage values less than a value
1972 * provided by the application."
1973 */
1974 if (up->pcrlen == 0) { /* full coverage was set */
1975 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1976 UDP_SKB_CB(skb)->cscov, skb->len);
1977 goto drop;
1978 }
1979 /* The next case involves violating the min. coverage requested
1980 * by the receiver. This is subtle: if receiver wants x and x is
1981 * greater than the buffersize/MTU then receiver will complain
1982 * that it wants x while sender emits packets of smaller size y.
1983 * Therefore the above ...()->partial_cov statement is essential.
1984 */
1985 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1986 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1987 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1988 goto drop;
1989 }
1990 }
1991
1992 prefetch(&sk->sk_rmem_alloc);
1993 if (rcu_access_pointer(sk->sk_filter) &&
1994 udp_lib_checksum_complete(skb))
1995 goto csum_error;
1996
1997 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1998 goto drop;
1999
2000 udp_csum_pull_header(skb);
2001
2002 ipv4_pktinfo_prepare(sk, skb);
2003 return __udp_queue_rcv_skb(sk, skb);
2004
2005 csum_error:
2006 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2007 drop:
2008 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2009 atomic_inc(&sk->sk_drops);
2010 kfree_skb(skb);
2011 return -1;
2012 }
2013
2014 /* For TCP sockets, sk_rx_dst is protected by socket lock
2015 * For UDP, we use xchg() to guard against concurrent changes.
2016 */
2017 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2018 {
2019 struct dst_entry *old;
2020
2021 if (dst_hold_safe(dst)) {
2022 old = xchg(&sk->sk_rx_dst, dst);
2023 dst_release(old);
2024 return old != dst;
2025 }
2026 return false;
2027 }
2028 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2029
2030 /*
2031 * Multicasts and broadcasts go to each listener.
2032 *
2033 * Note: called only from the BH handler context.
2034 */
2035 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2036 struct udphdr *uh,
2037 __be32 saddr, __be32 daddr,
2038 struct udp_table *udptable,
2039 int proto)
2040 {
2041 struct sock *sk, *first = NULL;
2042 unsigned short hnum = ntohs(uh->dest);
2043 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2044 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2045 unsigned int offset = offsetof(typeof(*sk), sk_node);
2046 int dif = skb->dev->ifindex;
2047 int sdif = inet_sdif(skb);
2048 struct hlist_node *node;
2049 struct sk_buff *nskb;
2050
2051 if (use_hash2) {
2052 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2053 udptable->mask;
2054 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2055 start_lookup:
2056 hslot = &udptable->hash2[hash2];
2057 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2058 }
2059
2060 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2061 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2062 uh->source, saddr, dif, sdif, hnum))
2063 continue;
2064
2065 if (!first) {
2066 first = sk;
2067 continue;
2068 }
2069 nskb = skb_clone(skb, GFP_ATOMIC);
2070
2071 if (unlikely(!nskb)) {
2072 atomic_inc(&sk->sk_drops);
2073 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2074 IS_UDPLITE(sk));
2075 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2076 IS_UDPLITE(sk));
2077 continue;
2078 }
2079 if (udp_queue_rcv_skb(sk, nskb) > 0)
2080 consume_skb(nskb);
2081 }
2082
2083 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2084 if (use_hash2 && hash2 != hash2_any) {
2085 hash2 = hash2_any;
2086 goto start_lookup;
2087 }
2088
2089 if (first) {
2090 if (udp_queue_rcv_skb(first, skb) > 0)
2091 consume_skb(skb);
2092 } else {
2093 kfree_skb(skb);
2094 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2095 proto == IPPROTO_UDPLITE);
2096 }
2097 return 0;
2098 }
2099
2100 /* Initialize UDP checksum. If exited with zero value (success),
2101 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2102 * Otherwise, csum completion requires chacksumming packet body,
2103 * including udp header and folding it to skb->csum.
2104 */
2105 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2106 int proto)
2107 {
2108 int err;
2109
2110 UDP_SKB_CB(skb)->partial_cov = 0;
2111 UDP_SKB_CB(skb)->cscov = skb->len;
2112
2113 if (proto == IPPROTO_UDPLITE) {
2114 err = udplite_checksum_init(skb, uh);
2115 if (err)
2116 return err;
2117
2118 if (UDP_SKB_CB(skb)->partial_cov) {
2119 skb->csum = inet_compute_pseudo(skb, proto);
2120 return 0;
2121 }
2122 }
2123
2124 /* Note, we are only interested in != 0 or == 0, thus the
2125 * force to int.
2126 */
2127 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2128 inet_compute_pseudo);
2129 }
2130
2131 /*
2132 * All we need to do is get the socket, and then do a checksum.
2133 */
2134
2135 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2136 int proto)
2137 {
2138 struct sock *sk;
2139 struct udphdr *uh;
2140 unsigned short ulen;
2141 struct rtable *rt = skb_rtable(skb);
2142 __be32 saddr, daddr;
2143 struct net *net = dev_net(skb->dev);
2144
2145 /*
2146 * Validate the packet.
2147 */
2148 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2149 goto drop; /* No space for header. */
2150
2151 uh = udp_hdr(skb);
2152 ulen = ntohs(uh->len);
2153 saddr = ip_hdr(skb)->saddr;
2154 daddr = ip_hdr(skb)->daddr;
2155
2156 if (ulen > skb->len)
2157 goto short_packet;
2158
2159 if (proto == IPPROTO_UDP) {
2160 /* UDP validates ulen. */
2161 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2162 goto short_packet;
2163 uh = udp_hdr(skb);
2164 }
2165
2166 if (udp4_csum_init(skb, uh, proto))
2167 goto csum_error;
2168
2169 sk = skb_steal_sock(skb);
2170 if (sk) {
2171 struct dst_entry *dst = skb_dst(skb);
2172 int ret;
2173
2174 if (unlikely(sk->sk_rx_dst != dst))
2175 udp_sk_rx_dst_set(sk, dst);
2176
2177 ret = udp_queue_rcv_skb(sk, skb);
2178 sock_put(sk);
2179 /* a return value > 0 means to resubmit the input, but
2180 * it wants the return to be -protocol, or 0
2181 */
2182 if (ret > 0)
2183 return -ret;
2184 return 0;
2185 }
2186
2187 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2188 return __udp4_lib_mcast_deliver(net, skb, uh,
2189 saddr, daddr, udptable, proto);
2190
2191 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2192 if (sk) {
2193 int ret;
2194
2195 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2196 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2197 inet_compute_pseudo);
2198
2199 ret = udp_queue_rcv_skb(sk, skb);
2200
2201 /* a return value > 0 means to resubmit the input, but
2202 * it wants the return to be -protocol, or 0
2203 */
2204 if (ret > 0)
2205 return -ret;
2206 return 0;
2207 }
2208
2209 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2210 goto drop;
2211 nf_reset(skb);
2212
2213 /* No socket. Drop packet silently, if checksum is wrong */
2214 if (udp_lib_checksum_complete(skb))
2215 goto csum_error;
2216
2217 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2218 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2219
2220 /*
2221 * Hmm. We got an UDP packet to a port to which we
2222 * don't wanna listen. Ignore it.
2223 */
2224 kfree_skb(skb);
2225 return 0;
2226
2227 short_packet:
2228 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2229 proto == IPPROTO_UDPLITE ? "Lite" : "",
2230 &saddr, ntohs(uh->source),
2231 ulen, skb->len,
2232 &daddr, ntohs(uh->dest));
2233 goto drop;
2234
2235 csum_error:
2236 /*
2237 * RFC1122: OK. Discards the bad packet silently (as far as
2238 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2239 */
2240 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2241 proto == IPPROTO_UDPLITE ? "Lite" : "",
2242 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2243 ulen);
2244 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2245 drop:
2246 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2247 kfree_skb(skb);
2248 return 0;
2249 }
2250
2251 /* We can only early demux multicast if there is a single matching socket.
2252 * If more than one socket found returns NULL
2253 */
2254 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2255 __be16 loc_port, __be32 loc_addr,
2256 __be16 rmt_port, __be32 rmt_addr,
2257 int dif, int sdif)
2258 {
2259 struct sock *sk, *result;
2260 unsigned short hnum = ntohs(loc_port);
2261 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2262 struct udp_hslot *hslot = &udp_table.hash[slot];
2263
2264 /* Do not bother scanning a too big list */
2265 if (hslot->count > 10)
2266 return NULL;
2267
2268 result = NULL;
2269 sk_for_each_rcu(sk, &hslot->head) {
2270 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2271 rmt_port, rmt_addr, dif, sdif, hnum)) {
2272 if (result)
2273 return NULL;
2274 result = sk;
2275 }
2276 }
2277
2278 return result;
2279 }
2280
2281 /* For unicast we should only early demux connected sockets or we can
2282 * break forwarding setups. The chains here can be long so only check
2283 * if the first socket is an exact match and if not move on.
2284 */
2285 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2286 __be16 loc_port, __be32 loc_addr,
2287 __be16 rmt_port, __be32 rmt_addr,
2288 int dif, int sdif)
2289 {
2290 unsigned short hnum = ntohs(loc_port);
2291 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2292 unsigned int slot2 = hash2 & udp_table.mask;
2293 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2294 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2295 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2296 struct sock *sk;
2297
2298 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2299 if (INET_MATCH(sk, net, acookie, rmt_addr,
2300 loc_addr, ports, dif, sdif))
2301 return sk;
2302 /* Only check first socket in chain */
2303 break;
2304 }
2305 return NULL;
2306 }
2307
2308 int udp_v4_early_demux(struct sk_buff *skb)
2309 {
2310 struct net *net = dev_net(skb->dev);
2311 struct in_device *in_dev = NULL;
2312 const struct iphdr *iph;
2313 const struct udphdr *uh;
2314 struct sock *sk = NULL;
2315 struct dst_entry *dst;
2316 int dif = skb->dev->ifindex;
2317 int sdif = inet_sdif(skb);
2318 int ours;
2319
2320 /* validate the packet */
2321 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2322 return 0;
2323
2324 iph = ip_hdr(skb);
2325 uh = udp_hdr(skb);
2326
2327 if (skb->pkt_type == PACKET_MULTICAST) {
2328 in_dev = __in_dev_get_rcu(skb->dev);
2329
2330 if (!in_dev)
2331 return 0;
2332
2333 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2334 iph->protocol);
2335 if (!ours)
2336 return 0;
2337
2338 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2339 uh->source, iph->saddr,
2340 dif, sdif);
2341 } else if (skb->pkt_type == PACKET_HOST) {
2342 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2343 uh->source, iph->saddr, dif, sdif);
2344 }
2345
2346 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2347 return 0;
2348
2349 skb->sk = sk;
2350 skb->destructor = sock_efree;
2351 dst = READ_ONCE(sk->sk_rx_dst);
2352
2353 if (dst)
2354 dst = dst_check(dst, 0);
2355 if (dst) {
2356 u32 itag = 0;
2357
2358 /* set noref for now.
2359 * any place which wants to hold dst has to call
2360 * dst_hold_safe()
2361 */
2362 skb_dst_set_noref(skb, dst);
2363
2364 /* for unconnected multicast sockets we need to validate
2365 * the source on each packet
2366 */
2367 if (!inet_sk(sk)->inet_daddr && in_dev)
2368 return ip_mc_validate_source(skb, iph->daddr,
2369 iph->saddr, iph->tos,
2370 skb->dev, in_dev, &itag);
2371 }
2372 return 0;
2373 }
2374
2375 int udp_rcv(struct sk_buff *skb)
2376 {
2377 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2378 }
2379
2380 void udp_destroy_sock(struct sock *sk)
2381 {
2382 struct udp_sock *up = udp_sk(sk);
2383 bool slow = lock_sock_fast(sk);
2384 udp_flush_pending_frames(sk);
2385 unlock_sock_fast(sk, slow);
2386 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2387 void (*encap_destroy)(struct sock *sk);
2388 encap_destroy = READ_ONCE(up->encap_destroy);
2389 if (encap_destroy)
2390 encap_destroy(sk);
2391 }
2392 }
2393
2394 /*
2395 * Socket option code for UDP
2396 */
2397 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2398 char __user *optval, unsigned int optlen,
2399 int (*push_pending_frames)(struct sock *))
2400 {
2401 struct udp_sock *up = udp_sk(sk);
2402 int val, valbool;
2403 int err = 0;
2404 int is_udplite = IS_UDPLITE(sk);
2405
2406 if (optlen < sizeof(int))
2407 return -EINVAL;
2408
2409 if (get_user(val, (int __user *)optval))
2410 return -EFAULT;
2411
2412 valbool = val ? 1 : 0;
2413
2414 switch (optname) {
2415 case UDP_CORK:
2416 if (val != 0) {
2417 up->corkflag = 1;
2418 } else {
2419 up->corkflag = 0;
2420 lock_sock(sk);
2421 push_pending_frames(sk);
2422 release_sock(sk);
2423 }
2424 break;
2425
2426 case UDP_ENCAP:
2427 switch (val) {
2428 case 0:
2429 case UDP_ENCAP_ESPINUDP:
2430 case UDP_ENCAP_ESPINUDP_NON_IKE:
2431 up->encap_rcv = xfrm4_udp_encap_rcv;
2432 /* FALLTHROUGH */
2433 case UDP_ENCAP_L2TPINUDP:
2434 up->encap_type = val;
2435 udp_encap_enable();
2436 break;
2437 default:
2438 err = -ENOPROTOOPT;
2439 break;
2440 }
2441 break;
2442
2443 case UDP_NO_CHECK6_TX:
2444 up->no_check6_tx = valbool;
2445 break;
2446
2447 case UDP_NO_CHECK6_RX:
2448 up->no_check6_rx = valbool;
2449 break;
2450
2451 case UDP_SEGMENT:
2452 if (val < 0 || val > USHRT_MAX)
2453 return -EINVAL;
2454 up->gso_size = val;
2455 break;
2456
2457 /*
2458 * UDP-Lite's partial checksum coverage (RFC 3828).
2459 */
2460 /* The sender sets actual checksum coverage length via this option.
2461 * The case coverage > packet length is handled by send module. */
2462 case UDPLITE_SEND_CSCOV:
2463 if (!is_udplite) /* Disable the option on UDP sockets */
2464 return -ENOPROTOOPT;
2465 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2466 val = 8;
2467 else if (val > USHRT_MAX)
2468 val = USHRT_MAX;
2469 up->pcslen = val;
2470 up->pcflag |= UDPLITE_SEND_CC;
2471 break;
2472
2473 /* The receiver specifies a minimum checksum coverage value. To make
2474 * sense, this should be set to at least 8 (as done below). If zero is
2475 * used, this again means full checksum coverage. */
2476 case UDPLITE_RECV_CSCOV:
2477 if (!is_udplite) /* Disable the option on UDP sockets */
2478 return -ENOPROTOOPT;
2479 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2480 val = 8;
2481 else if (val > USHRT_MAX)
2482 val = USHRT_MAX;
2483 up->pcrlen = val;
2484 up->pcflag |= UDPLITE_RECV_CC;
2485 break;
2486
2487 default:
2488 err = -ENOPROTOOPT;
2489 break;
2490 }
2491
2492 return err;
2493 }
2494 EXPORT_SYMBOL(udp_lib_setsockopt);
2495
2496 int udp_setsockopt(struct sock *sk, int level, int optname,
2497 char __user *optval, unsigned int optlen)
2498 {
2499 if (level == SOL_UDP || level == SOL_UDPLITE)
2500 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2501 udp_push_pending_frames);
2502 return ip_setsockopt(sk, level, optname, optval, optlen);
2503 }
2504
2505 #ifdef CONFIG_COMPAT
2506 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2507 char __user *optval, unsigned int optlen)
2508 {
2509 if (level == SOL_UDP || level == SOL_UDPLITE)
2510 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2511 udp_push_pending_frames);
2512 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2513 }
2514 #endif
2515
2516 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2517 char __user *optval, int __user *optlen)
2518 {
2519 struct udp_sock *up = udp_sk(sk);
2520 int val, len;
2521
2522 if (get_user(len, optlen))
2523 return -EFAULT;
2524
2525 len = min_t(unsigned int, len, sizeof(int));
2526
2527 if (len < 0)
2528 return -EINVAL;
2529
2530 switch (optname) {
2531 case UDP_CORK:
2532 val = up->corkflag;
2533 break;
2534
2535 case UDP_ENCAP:
2536 val = up->encap_type;
2537 break;
2538
2539 case UDP_NO_CHECK6_TX:
2540 val = up->no_check6_tx;
2541 break;
2542
2543 case UDP_NO_CHECK6_RX:
2544 val = up->no_check6_rx;
2545 break;
2546
2547 case UDP_SEGMENT:
2548 val = up->gso_size;
2549 break;
2550
2551 /* The following two cannot be changed on UDP sockets, the return is
2552 * always 0 (which corresponds to the full checksum coverage of UDP). */
2553 case UDPLITE_SEND_CSCOV:
2554 val = up->pcslen;
2555 break;
2556
2557 case UDPLITE_RECV_CSCOV:
2558 val = up->pcrlen;
2559 break;
2560
2561 default:
2562 return -ENOPROTOOPT;
2563 }
2564
2565 if (put_user(len, optlen))
2566 return -EFAULT;
2567 if (copy_to_user(optval, &val, len))
2568 return -EFAULT;
2569 return 0;
2570 }
2571 EXPORT_SYMBOL(udp_lib_getsockopt);
2572
2573 int udp_getsockopt(struct sock *sk, int level, int optname,
2574 char __user *optval, int __user *optlen)
2575 {
2576 if (level == SOL_UDP || level == SOL_UDPLITE)
2577 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2578 return ip_getsockopt(sk, level, optname, optval, optlen);
2579 }
2580
2581 #ifdef CONFIG_COMPAT
2582 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2583 char __user *optval, int __user *optlen)
2584 {
2585 if (level == SOL_UDP || level == SOL_UDPLITE)
2586 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2587 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2588 }
2589 #endif
2590 /**
2591 * udp_poll - wait for a UDP event.
2592 * @file - file struct
2593 * @sock - socket
2594 * @wait - poll table
2595 *
2596 * This is same as datagram poll, except for the special case of
2597 * blocking sockets. If application is using a blocking fd
2598 * and a packet with checksum error is in the queue;
2599 * then it could get return from select indicating data available
2600 * but then block when reading it. Add special case code
2601 * to work around these arguably broken applications.
2602 */
2603 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2604 {
2605 __poll_t mask = datagram_poll(file, sock, wait);
2606 struct sock *sk = sock->sk;
2607
2608 if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2609 mask |= EPOLLIN | EPOLLRDNORM;
2610
2611 /* Check for false positives due to checksum errors */
2612 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2613 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2614 mask &= ~(EPOLLIN | EPOLLRDNORM);
2615
2616 return mask;
2617
2618 }
2619 EXPORT_SYMBOL(udp_poll);
2620
2621 int udp_abort(struct sock *sk, int err)
2622 {
2623 lock_sock(sk);
2624
2625 sk->sk_err = err;
2626 sk->sk_error_report(sk);
2627 __udp_disconnect(sk, 0);
2628
2629 release_sock(sk);
2630
2631 return 0;
2632 }
2633 EXPORT_SYMBOL_GPL(udp_abort);
2634
2635 struct proto udp_prot = {
2636 .name = "UDP",
2637 .owner = THIS_MODULE,
2638 .close = udp_lib_close,
2639 .pre_connect = udp_pre_connect,
2640 .connect = ip4_datagram_connect,
2641 .disconnect = udp_disconnect,
2642 .ioctl = udp_ioctl,
2643 .init = udp_init_sock,
2644 .destroy = udp_destroy_sock,
2645 .setsockopt = udp_setsockopt,
2646 .getsockopt = udp_getsockopt,
2647 .sendmsg = udp_sendmsg,
2648 .recvmsg = udp_recvmsg,
2649 .sendpage = udp_sendpage,
2650 .release_cb = ip4_datagram_release_cb,
2651 .hash = udp_lib_hash,
2652 .unhash = udp_lib_unhash,
2653 .rehash = udp_v4_rehash,
2654 .get_port = udp_v4_get_port,
2655 .memory_allocated = &udp_memory_allocated,
2656 .sysctl_mem = sysctl_udp_mem,
2657 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2658 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2659 .obj_size = sizeof(struct udp_sock),
2660 .h.udp_table = &udp_table,
2661 #ifdef CONFIG_COMPAT
2662 .compat_setsockopt = compat_udp_setsockopt,
2663 .compat_getsockopt = compat_udp_getsockopt,
2664 #endif
2665 .diag_destroy = udp_abort,
2666 };
2667 EXPORT_SYMBOL(udp_prot);
2668
2669 /* ------------------------------------------------------------------------ */
2670 #ifdef CONFIG_PROC_FS
2671
2672 static struct sock *udp_get_first(struct seq_file *seq, int start)
2673 {
2674 struct sock *sk;
2675 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2676 struct udp_iter_state *state = seq->private;
2677 struct net *net = seq_file_net(seq);
2678
2679 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2680 ++state->bucket) {
2681 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2682
2683 if (hlist_empty(&hslot->head))
2684 continue;
2685
2686 spin_lock_bh(&hslot->lock);
2687 sk_for_each(sk, &hslot->head) {
2688 if (!net_eq(sock_net(sk), net))
2689 continue;
2690 if (sk->sk_family == afinfo->family)
2691 goto found;
2692 }
2693 spin_unlock_bh(&hslot->lock);
2694 }
2695 sk = NULL;
2696 found:
2697 return sk;
2698 }
2699
2700 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2701 {
2702 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2703 struct udp_iter_state *state = seq->private;
2704 struct net *net = seq_file_net(seq);
2705
2706 do {
2707 sk = sk_next(sk);
2708 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2709
2710 if (!sk) {
2711 if (state->bucket <= afinfo->udp_table->mask)
2712 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2713 return udp_get_first(seq, state->bucket + 1);
2714 }
2715 return sk;
2716 }
2717
2718 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2719 {
2720 struct sock *sk = udp_get_first(seq, 0);
2721
2722 if (sk)
2723 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2724 --pos;
2725 return pos ? NULL : sk;
2726 }
2727
2728 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2729 {
2730 struct udp_iter_state *state = seq->private;
2731 state->bucket = MAX_UDP_PORTS;
2732
2733 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2734 }
2735 EXPORT_SYMBOL(udp_seq_start);
2736
2737 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2738 {
2739 struct sock *sk;
2740
2741 if (v == SEQ_START_TOKEN)
2742 sk = udp_get_idx(seq, 0);
2743 else
2744 sk = udp_get_next(seq, v);
2745
2746 ++*pos;
2747 return sk;
2748 }
2749 EXPORT_SYMBOL(udp_seq_next);
2750
2751 void udp_seq_stop(struct seq_file *seq, void *v)
2752 {
2753 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2754 struct udp_iter_state *state = seq->private;
2755
2756 if (state->bucket <= afinfo->udp_table->mask)
2757 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2758 }
2759 EXPORT_SYMBOL(udp_seq_stop);
2760
2761 /* ------------------------------------------------------------------------ */
2762 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2763 int bucket)
2764 {
2765 struct inet_sock *inet = inet_sk(sp);
2766 __be32 dest = inet->inet_daddr;
2767 __be32 src = inet->inet_rcv_saddr;
2768 __u16 destp = ntohs(inet->inet_dport);
2769 __u16 srcp = ntohs(inet->inet_sport);
2770
2771 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2772 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2773 bucket, src, srcp, dest, destp, sp->sk_state,
2774 sk_wmem_alloc_get(sp),
2775 udp_rqueue_get(sp),
2776 0, 0L, 0,
2777 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2778 0, sock_i_ino(sp),
2779 refcount_read(&sp->sk_refcnt), sp,
2780 atomic_read(&sp->sk_drops));
2781 }
2782
2783 int udp4_seq_show(struct seq_file *seq, void *v)
2784 {
2785 seq_setwidth(seq, 127);
2786 if (v == SEQ_START_TOKEN)
2787 seq_puts(seq, " sl local_address rem_address st tx_queue "
2788 "rx_queue tr tm->when retrnsmt uid timeout "
2789 "inode ref pointer drops");
2790 else {
2791 struct udp_iter_state *state = seq->private;
2792
2793 udp4_format_sock(v, seq, state->bucket);
2794 }
2795 seq_pad(seq, '\n');
2796 return 0;
2797 }
2798
2799 const struct seq_operations udp_seq_ops = {
2800 .start = udp_seq_start,
2801 .next = udp_seq_next,
2802 .stop = udp_seq_stop,
2803 .show = udp4_seq_show,
2804 };
2805 EXPORT_SYMBOL(udp_seq_ops);
2806
2807 static struct udp_seq_afinfo udp4_seq_afinfo = {
2808 .family = AF_INET,
2809 .udp_table = &udp_table,
2810 };
2811
2812 static int __net_init udp4_proc_init_net(struct net *net)
2813 {
2814 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2815 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2816 return -ENOMEM;
2817 return 0;
2818 }
2819
2820 static void __net_exit udp4_proc_exit_net(struct net *net)
2821 {
2822 remove_proc_entry("udp", net->proc_net);
2823 }
2824
2825 static struct pernet_operations udp4_net_ops = {
2826 .init = udp4_proc_init_net,
2827 .exit = udp4_proc_exit_net,
2828 };
2829
2830 int __init udp4_proc_init(void)
2831 {
2832 return register_pernet_subsys(&udp4_net_ops);
2833 }
2834
2835 void udp4_proc_exit(void)
2836 {
2837 unregister_pernet_subsys(&udp4_net_ops);
2838 }
2839 #endif /* CONFIG_PROC_FS */
2840
2841 static __initdata unsigned long uhash_entries;
2842 static int __init set_uhash_entries(char *str)
2843 {
2844 ssize_t ret;
2845
2846 if (!str)
2847 return 0;
2848
2849 ret = kstrtoul(str, 0, &uhash_entries);
2850 if (ret)
2851 return 0;
2852
2853 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2854 uhash_entries = UDP_HTABLE_SIZE_MIN;
2855 return 1;
2856 }
2857 __setup("uhash_entries=", set_uhash_entries);
2858
2859 void __init udp_table_init(struct udp_table *table, const char *name)
2860 {
2861 unsigned int i;
2862
2863 table->hash = alloc_large_system_hash(name,
2864 2 * sizeof(struct udp_hslot),
2865 uhash_entries,
2866 21, /* one slot per 2 MB */
2867 0,
2868 &table->log,
2869 &table->mask,
2870 UDP_HTABLE_SIZE_MIN,
2871 64 * 1024);
2872
2873 table->hash2 = table->hash + (table->mask + 1);
2874 for (i = 0; i <= table->mask; i++) {
2875 INIT_HLIST_HEAD(&table->hash[i].head);
2876 table->hash[i].count = 0;
2877 spin_lock_init(&table->hash[i].lock);
2878 }
2879 for (i = 0; i <= table->mask; i++) {
2880 INIT_HLIST_HEAD(&table->hash2[i].head);
2881 table->hash2[i].count = 0;
2882 spin_lock_init(&table->hash2[i].lock);
2883 }
2884 }
2885
2886 u32 udp_flow_hashrnd(void)
2887 {
2888 static u32 hashrnd __read_mostly;
2889
2890 net_get_random_once(&hashrnd, sizeof(hashrnd));
2891
2892 return hashrnd;
2893 }
2894 EXPORT_SYMBOL(udp_flow_hashrnd);
2895
2896 static void __udp_sysctl_init(struct net *net)
2897 {
2898 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2899 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2900
2901 #ifdef CONFIG_NET_L3_MASTER_DEV
2902 net->ipv4.sysctl_udp_l3mdev_accept = 0;
2903 #endif
2904 }
2905
2906 static int __net_init udp_sysctl_init(struct net *net)
2907 {
2908 __udp_sysctl_init(net);
2909 return 0;
2910 }
2911
2912 static struct pernet_operations __net_initdata udp_sysctl_ops = {
2913 .init = udp_sysctl_init,
2914 };
2915
2916 void __init udp_init(void)
2917 {
2918 unsigned long limit;
2919 unsigned int i;
2920
2921 udp_table_init(&udp_table, "UDP");
2922 limit = nr_free_buffer_pages() / 8;
2923 limit = max(limit, 128UL);
2924 sysctl_udp_mem[0] = limit / 4 * 3;
2925 sysctl_udp_mem[1] = limit;
2926 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2927
2928 __udp_sysctl_init(&init_net);
2929
2930 /* 16 spinlocks per cpu */
2931 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2932 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2933 GFP_KERNEL);
2934 if (!udp_busylocks)
2935 panic("UDP: failed to alloc udp_busylocks\n");
2936 for (i = 0; i < (1U << udp_busylocks_log); i++)
2937 spin_lock_init(udp_busylocks + i);
2938
2939 if (register_pernet_subsys(&udp_sysctl_ops))
2940 panic("UDP: failed to init sysctl parameters.\n");
2941 }
Linux with added FPC-III support