]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/ipv4/udp.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
procfs: new helper - PDE_DATA(inode)
[mirror_ubuntu-bionic-kernel.git] / net / ipv4 / udp.c
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 <asm/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/in.h>
94 #include <linux/errno.h>
95 #include <linux/timer.h>
96 #include <linux/mm.h>
97 #include <linux/inet.h>
98 #include <linux/netdevice.h>
99 #include <linux/slab.h>
100 #include <net/tcp_states.h>
101 #include <linux/skbuff.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <net/net_namespace.h>
105 #include <net/icmp.h>
106 #include <net/route.h>
107 #include <net/checksum.h>
108 #include <net/xfrm.h>
109 #include <trace/events/udp.h>
110 #include <linux/static_key.h>
111 #include <trace/events/skb.h>
112 #include "udp_impl.h"
113
114 struct udp_table udp_table __read_mostly;
115 EXPORT_SYMBOL(udp_table);
116
117 long sysctl_udp_mem[3] __read_mostly;
118 EXPORT_SYMBOL(sysctl_udp_mem);
119
120 int sysctl_udp_rmem_min __read_mostly;
121 EXPORT_SYMBOL(sysctl_udp_rmem_min);
122
123 int sysctl_udp_wmem_min __read_mostly;
124 EXPORT_SYMBOL(sysctl_udp_wmem_min);
125
126 atomic_long_t udp_memory_allocated;
127 EXPORT_SYMBOL(udp_memory_allocated);
128
129 #define MAX_UDP_PORTS 65536
130 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
131
132 static int udp_lib_lport_inuse(struct net *net, __u16 num,
133 const struct udp_hslot *hslot,
134 unsigned long *bitmap,
135 struct sock *sk,
136 int (*saddr_comp)(const struct sock *sk1,
137 const struct sock *sk2),
138 unsigned int log)
139 {
140 struct sock *sk2;
141 struct hlist_nulls_node *node;
142 kuid_t uid = sock_i_uid(sk);
143
144 sk_nulls_for_each(sk2, node, &hslot->head)
145 if (net_eq(sock_net(sk2), net) &&
146 sk2 != sk &&
147 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
148 (!sk2->sk_reuse || !sk->sk_reuse) &&
149 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
150 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
151 (!sk2->sk_reuseport || !sk->sk_reuseport ||
152 !uid_eq(uid, sock_i_uid(sk2))) &&
153 (*saddr_comp)(sk, sk2)) {
154 if (bitmap)
155 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
156 bitmap);
157 else
158 return 1;
159 }
160 return 0;
161 }
162
163 /*
164 * Note: we still hold spinlock of primary hash chain, so no other writer
165 * can insert/delete a socket with local_port == num
166 */
167 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
168 struct udp_hslot *hslot2,
169 struct sock *sk,
170 int (*saddr_comp)(const struct sock *sk1,
171 const struct sock *sk2))
172 {
173 struct sock *sk2;
174 struct hlist_nulls_node *node;
175 kuid_t uid = sock_i_uid(sk);
176 int res = 0;
177
178 spin_lock(&hslot2->lock);
179 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
180 if (net_eq(sock_net(sk2), net) &&
181 sk2 != sk &&
182 (udp_sk(sk2)->udp_port_hash == num) &&
183 (!sk2->sk_reuse || !sk->sk_reuse) &&
184 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
185 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
186 (!sk2->sk_reuseport || !sk->sk_reuseport ||
187 !uid_eq(uid, sock_i_uid(sk2))) &&
188 (*saddr_comp)(sk, sk2)) {
189 res = 1;
190 break;
191 }
192 spin_unlock(&hslot2->lock);
193 return res;
194 }
195
196 /**
197 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
198 *
199 * @sk: socket struct in question
200 * @snum: port number to look up
201 * @saddr_comp: AF-dependent comparison of bound local IP addresses
202 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
203 * with NULL address
204 */
205 int udp_lib_get_port(struct sock *sk, unsigned short snum,
206 int (*saddr_comp)(const struct sock *sk1,
207 const struct sock *sk2),
208 unsigned int hash2_nulladdr)
209 {
210 struct udp_hslot *hslot, *hslot2;
211 struct udp_table *udptable = sk->sk_prot->h.udp_table;
212 int error = 1;
213 struct net *net = sock_net(sk);
214
215 if (!snum) {
216 int low, high, remaining;
217 unsigned int rand;
218 unsigned short first, last;
219 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
220
221 inet_get_local_port_range(&low, &high);
222 remaining = (high - low) + 1;
223
224 rand = net_random();
225 first = (((u64)rand * remaining) >> 32) + low;
226 /*
227 * force rand to be an odd multiple of UDP_HTABLE_SIZE
228 */
229 rand = (rand | 1) * (udptable->mask + 1);
230 last = first + udptable->mask + 1;
231 do {
232 hslot = udp_hashslot(udptable, net, first);
233 bitmap_zero(bitmap, PORTS_PER_CHAIN);
234 spin_lock_bh(&hslot->lock);
235 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
236 saddr_comp, udptable->log);
237
238 snum = first;
239 /*
240 * Iterate on all possible values of snum for this hash.
241 * Using steps of an odd multiple of UDP_HTABLE_SIZE
242 * give us randomization and full range coverage.
243 */
244 do {
245 if (low <= snum && snum <= high &&
246 !test_bit(snum >> udptable->log, bitmap) &&
247 !inet_is_reserved_local_port(snum))
248 goto found;
249 snum += rand;
250 } while (snum != first);
251 spin_unlock_bh(&hslot->lock);
252 } while (++first != last);
253 goto fail;
254 } else {
255 hslot = udp_hashslot(udptable, net, snum);
256 spin_lock_bh(&hslot->lock);
257 if (hslot->count > 10) {
258 int exist;
259 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
260
261 slot2 &= udptable->mask;
262 hash2_nulladdr &= udptable->mask;
263
264 hslot2 = udp_hashslot2(udptable, slot2);
265 if (hslot->count < hslot2->count)
266 goto scan_primary_hash;
267
268 exist = udp_lib_lport_inuse2(net, snum, hslot2,
269 sk, saddr_comp);
270 if (!exist && (hash2_nulladdr != slot2)) {
271 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
272 exist = udp_lib_lport_inuse2(net, snum, hslot2,
273 sk, saddr_comp);
274 }
275 if (exist)
276 goto fail_unlock;
277 else
278 goto found;
279 }
280 scan_primary_hash:
281 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
282 saddr_comp, 0))
283 goto fail_unlock;
284 }
285 found:
286 inet_sk(sk)->inet_num = snum;
287 udp_sk(sk)->udp_port_hash = snum;
288 udp_sk(sk)->udp_portaddr_hash ^= snum;
289 if (sk_unhashed(sk)) {
290 sk_nulls_add_node_rcu(sk, &hslot->head);
291 hslot->count++;
292 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
293
294 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
295 spin_lock(&hslot2->lock);
296 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
297 &hslot2->head);
298 hslot2->count++;
299 spin_unlock(&hslot2->lock);
300 }
301 error = 0;
302 fail_unlock:
303 spin_unlock_bh(&hslot->lock);
304 fail:
305 return error;
306 }
307 EXPORT_SYMBOL(udp_lib_get_port);
308
309 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
310 {
311 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
312
313 return (!ipv6_only_sock(sk2) &&
314 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
315 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
316 }
317
318 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
319 unsigned int port)
320 {
321 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
322 }
323
324 int udp_v4_get_port(struct sock *sk, unsigned short snum)
325 {
326 unsigned int hash2_nulladdr =
327 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
328 unsigned int hash2_partial =
329 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
330
331 /* precompute partial secondary hash */
332 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
333 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
334 }
335
336 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
337 unsigned short hnum,
338 __be16 sport, __be32 daddr, __be16 dport, int dif)
339 {
340 int score = -1;
341
342 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
343 !ipv6_only_sock(sk)) {
344 struct inet_sock *inet = inet_sk(sk);
345
346 score = (sk->sk_family == PF_INET ? 2 : 1);
347 if (inet->inet_rcv_saddr) {
348 if (inet->inet_rcv_saddr != daddr)
349 return -1;
350 score += 4;
351 }
352 if (inet->inet_daddr) {
353 if (inet->inet_daddr != saddr)
354 return -1;
355 score += 4;
356 }
357 if (inet->inet_dport) {
358 if (inet->inet_dport != sport)
359 return -1;
360 score += 4;
361 }
362 if (sk->sk_bound_dev_if) {
363 if (sk->sk_bound_dev_if != dif)
364 return -1;
365 score += 4;
366 }
367 }
368 return score;
369 }
370
371 /*
372 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
373 */
374 static inline int compute_score2(struct sock *sk, struct net *net,
375 __be32 saddr, __be16 sport,
376 __be32 daddr, unsigned int hnum, int dif)
377 {
378 int score = -1;
379
380 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
381 struct inet_sock *inet = inet_sk(sk);
382
383 if (inet->inet_rcv_saddr != daddr)
384 return -1;
385 if (inet->inet_num != hnum)
386 return -1;
387
388 score = (sk->sk_family == PF_INET ? 2 : 1);
389 if (inet->inet_daddr) {
390 if (inet->inet_daddr != saddr)
391 return -1;
392 score += 4;
393 }
394 if (inet->inet_dport) {
395 if (inet->inet_dport != sport)
396 return -1;
397 score += 4;
398 }
399 if (sk->sk_bound_dev_if) {
400 if (sk->sk_bound_dev_if != dif)
401 return -1;
402 score += 4;
403 }
404 }
405 return score;
406 }
407
408
409 /* called with read_rcu_lock() */
410 static struct sock *udp4_lib_lookup2(struct net *net,
411 __be32 saddr, __be16 sport,
412 __be32 daddr, unsigned int hnum, int dif,
413 struct udp_hslot *hslot2, unsigned int slot2)
414 {
415 struct sock *sk, *result;
416 struct hlist_nulls_node *node;
417 int score, badness, matches = 0, reuseport = 0;
418 u32 hash = 0;
419
420 begin:
421 result = NULL;
422 badness = 0;
423 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
424 score = compute_score2(sk, net, saddr, sport,
425 daddr, hnum, dif);
426 if (score > badness) {
427 result = sk;
428 badness = score;
429 reuseport = sk->sk_reuseport;
430 if (reuseport) {
431 hash = inet_ehashfn(net, daddr, hnum,
432 saddr, htons(sport));
433 matches = 1;
434 }
435 } else if (score == badness && reuseport) {
436 matches++;
437 if (((u64)hash * matches) >> 32 == 0)
438 result = sk;
439 hash = next_pseudo_random32(hash);
440 }
441 }
442 /*
443 * if the nulls value we got at the end of this lookup is
444 * not the expected one, we must restart lookup.
445 * We probably met an item that was moved to another chain.
446 */
447 if (get_nulls_value(node) != slot2)
448 goto begin;
449 if (result) {
450 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
451 result = NULL;
452 else if (unlikely(compute_score2(result, net, saddr, sport,
453 daddr, hnum, dif) < badness)) {
454 sock_put(result);
455 goto begin;
456 }
457 }
458 return result;
459 }
460
461 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
462 * harder than this. -DaveM
463 */
464 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
465 __be16 sport, __be32 daddr, __be16 dport,
466 int dif, struct udp_table *udptable)
467 {
468 struct sock *sk, *result;
469 struct hlist_nulls_node *node;
470 unsigned short hnum = ntohs(dport);
471 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
472 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
473 int score, badness, matches = 0, reuseport = 0;
474 u32 hash = 0;
475
476 rcu_read_lock();
477 if (hslot->count > 10) {
478 hash2 = udp4_portaddr_hash(net, daddr, hnum);
479 slot2 = hash2 & udptable->mask;
480 hslot2 = &udptable->hash2[slot2];
481 if (hslot->count < hslot2->count)
482 goto begin;
483
484 result = udp4_lib_lookup2(net, saddr, sport,
485 daddr, hnum, dif,
486 hslot2, slot2);
487 if (!result) {
488 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
489 slot2 = hash2 & udptable->mask;
490 hslot2 = &udptable->hash2[slot2];
491 if (hslot->count < hslot2->count)
492 goto begin;
493
494 result = udp4_lib_lookup2(net, saddr, sport,
495 htonl(INADDR_ANY), hnum, dif,
496 hslot2, slot2);
497 }
498 rcu_read_unlock();
499 return result;
500 }
501 begin:
502 result = NULL;
503 badness = 0;
504 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
505 score = compute_score(sk, net, saddr, hnum, sport,
506 daddr, dport, dif);
507 if (score > badness) {
508 result = sk;
509 badness = score;
510 reuseport = sk->sk_reuseport;
511 if (reuseport) {
512 hash = inet_ehashfn(net, daddr, hnum,
513 saddr, htons(sport));
514 matches = 1;
515 }
516 } else if (score == badness && reuseport) {
517 matches++;
518 if (((u64)hash * matches) >> 32 == 0)
519 result = sk;
520 hash = next_pseudo_random32(hash);
521 }
522 }
523 /*
524 * if the nulls value we got at the end of this lookup is
525 * not the expected one, we must restart lookup.
526 * We probably met an item that was moved to another chain.
527 */
528 if (get_nulls_value(node) != slot)
529 goto begin;
530
531 if (result) {
532 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
533 result = NULL;
534 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
535 daddr, dport, dif) < badness)) {
536 sock_put(result);
537 goto begin;
538 }
539 }
540 rcu_read_unlock();
541 return result;
542 }
543 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
544
545 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
546 __be16 sport, __be16 dport,
547 struct udp_table *udptable)
548 {
549 struct sock *sk;
550 const struct iphdr *iph = ip_hdr(skb);
551
552 if (unlikely(sk = skb_steal_sock(skb)))
553 return sk;
554 else
555 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
556 iph->daddr, dport, inet_iif(skb),
557 udptable);
558 }
559
560 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
561 __be32 daddr, __be16 dport, int dif)
562 {
563 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
564 }
565 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
566
567 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
568 __be16 loc_port, __be32 loc_addr,
569 __be16 rmt_port, __be32 rmt_addr,
570 int dif)
571 {
572 struct hlist_nulls_node *node;
573 struct sock *s = sk;
574 unsigned short hnum = ntohs(loc_port);
575
576 sk_nulls_for_each_from(s, node) {
577 struct inet_sock *inet = inet_sk(s);
578
579 if (!net_eq(sock_net(s), net) ||
580 udp_sk(s)->udp_port_hash != hnum ||
581 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
582 (inet->inet_dport != rmt_port && inet->inet_dport) ||
583 (inet->inet_rcv_saddr &&
584 inet->inet_rcv_saddr != loc_addr) ||
585 ipv6_only_sock(s) ||
586 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
587 continue;
588 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
589 continue;
590 goto found;
591 }
592 s = NULL;
593 found:
594 return s;
595 }
596
597 /*
598 * This routine is called by the ICMP module when it gets some
599 * sort of error condition. If err < 0 then the socket should
600 * be closed and the error returned to the user. If err > 0
601 * it's just the icmp type << 8 | icmp code.
602 * Header points to the ip header of the error packet. We move
603 * on past this. Then (as it used to claim before adjustment)
604 * header points to the first 8 bytes of the udp header. We need
605 * to find the appropriate port.
606 */
607
608 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
609 {
610 struct inet_sock *inet;
611 const struct iphdr *iph = (const struct iphdr *)skb->data;
612 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
613 const int type = icmp_hdr(skb)->type;
614 const int code = icmp_hdr(skb)->code;
615 struct sock *sk;
616 int harderr;
617 int err;
618 struct net *net = dev_net(skb->dev);
619
620 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
621 iph->saddr, uh->source, skb->dev->ifindex, udptable);
622 if (sk == NULL) {
623 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
624 return; /* No socket for error */
625 }
626
627 err = 0;
628 harderr = 0;
629 inet = inet_sk(sk);
630
631 switch (type) {
632 default:
633 case ICMP_TIME_EXCEEDED:
634 err = EHOSTUNREACH;
635 break;
636 case ICMP_SOURCE_QUENCH:
637 goto out;
638 case ICMP_PARAMETERPROB:
639 err = EPROTO;
640 harderr = 1;
641 break;
642 case ICMP_DEST_UNREACH:
643 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
644 ipv4_sk_update_pmtu(skb, sk, info);
645 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
646 err = EMSGSIZE;
647 harderr = 1;
648 break;
649 }
650 goto out;
651 }
652 err = EHOSTUNREACH;
653 if (code <= NR_ICMP_UNREACH) {
654 harderr = icmp_err_convert[code].fatal;
655 err = icmp_err_convert[code].errno;
656 }
657 break;
658 case ICMP_REDIRECT:
659 ipv4_sk_redirect(skb, sk);
660 break;
661 }
662
663 /*
664 * RFC1122: OK. Passes ICMP errors back to application, as per
665 * 4.1.3.3.
666 */
667 if (!inet->recverr) {
668 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
669 goto out;
670 } else
671 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
672
673 sk->sk_err = err;
674 sk->sk_error_report(sk);
675 out:
676 sock_put(sk);
677 }
678
679 void udp_err(struct sk_buff *skb, u32 info)
680 {
681 __udp4_lib_err(skb, info, &udp_table);
682 }
683
684 /*
685 * Throw away all pending data and cancel the corking. Socket is locked.
686 */
687 void udp_flush_pending_frames(struct sock *sk)
688 {
689 struct udp_sock *up = udp_sk(sk);
690
691 if (up->pending) {
692 up->len = 0;
693 up->pending = 0;
694 ip_flush_pending_frames(sk);
695 }
696 }
697 EXPORT_SYMBOL(udp_flush_pending_frames);
698
699 /**
700 * udp4_hwcsum - handle outgoing HW checksumming
701 * @skb: sk_buff containing the filled-in UDP header
702 * (checksum field must be zeroed out)
703 * @src: source IP address
704 * @dst: destination IP address
705 */
706 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
707 {
708 struct udphdr *uh = udp_hdr(skb);
709 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
710 int offset = skb_transport_offset(skb);
711 int len = skb->len - offset;
712 int hlen = len;
713 __wsum csum = 0;
714
715 if (!frags) {
716 /*
717 * Only one fragment on the socket.
718 */
719 skb->csum_start = skb_transport_header(skb) - skb->head;
720 skb->csum_offset = offsetof(struct udphdr, check);
721 uh->check = ~csum_tcpudp_magic(src, dst, len,
722 IPPROTO_UDP, 0);
723 } else {
724 /*
725 * HW-checksum won't work as there are two or more
726 * fragments on the socket so that all csums of sk_buffs
727 * should be together
728 */
729 do {
730 csum = csum_add(csum, frags->csum);
731 hlen -= frags->len;
732 } while ((frags = frags->next));
733
734 csum = skb_checksum(skb, offset, hlen, csum);
735 skb->ip_summed = CHECKSUM_NONE;
736
737 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
738 if (uh->check == 0)
739 uh->check = CSUM_MANGLED_0;
740 }
741 }
742
743 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
744 {
745 struct sock *sk = skb->sk;
746 struct inet_sock *inet = inet_sk(sk);
747 struct udphdr *uh;
748 int err = 0;
749 int is_udplite = IS_UDPLITE(sk);
750 int offset = skb_transport_offset(skb);
751 int len = skb->len - offset;
752 __wsum csum = 0;
753
754 /*
755 * Create a UDP header
756 */
757 uh = udp_hdr(skb);
758 uh->source = inet->inet_sport;
759 uh->dest = fl4->fl4_dport;
760 uh->len = htons(len);
761 uh->check = 0;
762
763 if (is_udplite) /* UDP-Lite */
764 csum = udplite_csum(skb);
765
766 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
767
768 skb->ip_summed = CHECKSUM_NONE;
769 goto send;
770
771 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
772
773 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
774 goto send;
775
776 } else
777 csum = udp_csum(skb);
778
779 /* add protocol-dependent pseudo-header */
780 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
781 sk->sk_protocol, csum);
782 if (uh->check == 0)
783 uh->check = CSUM_MANGLED_0;
784
785 send:
786 err = ip_send_skb(sock_net(sk), skb);
787 if (err) {
788 if (err == -ENOBUFS && !inet->recverr) {
789 UDP_INC_STATS_USER(sock_net(sk),
790 UDP_MIB_SNDBUFERRORS, is_udplite);
791 err = 0;
792 }
793 } else
794 UDP_INC_STATS_USER(sock_net(sk),
795 UDP_MIB_OUTDATAGRAMS, is_udplite);
796 return err;
797 }
798
799 /*
800 * Push out all pending data as one UDP datagram. Socket is locked.
801 */
802 static int udp_push_pending_frames(struct sock *sk)
803 {
804 struct udp_sock *up = udp_sk(sk);
805 struct inet_sock *inet = inet_sk(sk);
806 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
807 struct sk_buff *skb;
808 int err = 0;
809
810 skb = ip_finish_skb(sk, fl4);
811 if (!skb)
812 goto out;
813
814 err = udp_send_skb(skb, fl4);
815
816 out:
817 up->len = 0;
818 up->pending = 0;
819 return err;
820 }
821
822 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
823 size_t len)
824 {
825 struct inet_sock *inet = inet_sk(sk);
826 struct udp_sock *up = udp_sk(sk);
827 struct flowi4 fl4_stack;
828 struct flowi4 *fl4;
829 int ulen = len;
830 struct ipcm_cookie ipc;
831 struct rtable *rt = NULL;
832 int free = 0;
833 int connected = 0;
834 __be32 daddr, faddr, saddr;
835 __be16 dport;
836 u8 tos;
837 int err, is_udplite = IS_UDPLITE(sk);
838 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
839 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
840 struct sk_buff *skb;
841 struct ip_options_data opt_copy;
842
843 if (len > 0xFFFF)
844 return -EMSGSIZE;
845
846 /*
847 * Check the flags.
848 */
849
850 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
851 return -EOPNOTSUPP;
852
853 ipc.opt = NULL;
854 ipc.tx_flags = 0;
855
856 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
857
858 fl4 = &inet->cork.fl.u.ip4;
859 if (up->pending) {
860 /*
861 * There are pending frames.
862 * The socket lock must be held while it's corked.
863 */
864 lock_sock(sk);
865 if (likely(up->pending)) {
866 if (unlikely(up->pending != AF_INET)) {
867 release_sock(sk);
868 return -EINVAL;
869 }
870 goto do_append_data;
871 }
872 release_sock(sk);
873 }
874 ulen += sizeof(struct udphdr);
875
876 /*
877 * Get and verify the address.
878 */
879 if (msg->msg_name) {
880 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
881 if (msg->msg_namelen < sizeof(*usin))
882 return -EINVAL;
883 if (usin->sin_family != AF_INET) {
884 if (usin->sin_family != AF_UNSPEC)
885 return -EAFNOSUPPORT;
886 }
887
888 daddr = usin->sin_addr.s_addr;
889 dport = usin->sin_port;
890 if (dport == 0)
891 return -EINVAL;
892 } else {
893 if (sk->sk_state != TCP_ESTABLISHED)
894 return -EDESTADDRREQ;
895 daddr = inet->inet_daddr;
896 dport = inet->inet_dport;
897 /* Open fast path for connected socket.
898 Route will not be used, if at least one option is set.
899 */
900 connected = 1;
901 }
902 ipc.addr = inet->inet_saddr;
903
904 ipc.oif = sk->sk_bound_dev_if;
905 err = sock_tx_timestamp(sk, &ipc.tx_flags);
906 if (err)
907 return err;
908 if (msg->msg_controllen) {
909 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
910 if (err)
911 return err;
912 if (ipc.opt)
913 free = 1;
914 connected = 0;
915 }
916 if (!ipc.opt) {
917 struct ip_options_rcu *inet_opt;
918
919 rcu_read_lock();
920 inet_opt = rcu_dereference(inet->inet_opt);
921 if (inet_opt) {
922 memcpy(&opt_copy, inet_opt,
923 sizeof(*inet_opt) + inet_opt->opt.optlen);
924 ipc.opt = &opt_copy.opt;
925 }
926 rcu_read_unlock();
927 }
928
929 saddr = ipc.addr;
930 ipc.addr = faddr = daddr;
931
932 if (ipc.opt && ipc.opt->opt.srr) {
933 if (!daddr)
934 return -EINVAL;
935 faddr = ipc.opt->opt.faddr;
936 connected = 0;
937 }
938 tos = RT_TOS(inet->tos);
939 if (sock_flag(sk, SOCK_LOCALROUTE) ||
940 (msg->msg_flags & MSG_DONTROUTE) ||
941 (ipc.opt && ipc.opt->opt.is_strictroute)) {
942 tos |= RTO_ONLINK;
943 connected = 0;
944 }
945
946 if (ipv4_is_multicast(daddr)) {
947 if (!ipc.oif)
948 ipc.oif = inet->mc_index;
949 if (!saddr)
950 saddr = inet->mc_addr;
951 connected = 0;
952 } else if (!ipc.oif)
953 ipc.oif = inet->uc_index;
954
955 if (connected)
956 rt = (struct rtable *)sk_dst_check(sk, 0);
957
958 if (rt == NULL) {
959 struct net *net = sock_net(sk);
960
961 fl4 = &fl4_stack;
962 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
963 RT_SCOPE_UNIVERSE, sk->sk_protocol,
964 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
965 faddr, saddr, dport, inet->inet_sport);
966
967 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
968 rt = ip_route_output_flow(net, fl4, sk);
969 if (IS_ERR(rt)) {
970 err = PTR_ERR(rt);
971 rt = NULL;
972 if (err == -ENETUNREACH)
973 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
974 goto out;
975 }
976
977 err = -EACCES;
978 if ((rt->rt_flags & RTCF_BROADCAST) &&
979 !sock_flag(sk, SOCK_BROADCAST))
980 goto out;
981 if (connected)
982 sk_dst_set(sk, dst_clone(&rt->dst));
983 }
984
985 if (msg->msg_flags&MSG_CONFIRM)
986 goto do_confirm;
987 back_from_confirm:
988
989 saddr = fl4->saddr;
990 if (!ipc.addr)
991 daddr = ipc.addr = fl4->daddr;
992
993 /* Lockless fast path for the non-corking case. */
994 if (!corkreq) {
995 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
996 sizeof(struct udphdr), &ipc, &rt,
997 msg->msg_flags);
998 err = PTR_ERR(skb);
999 if (!IS_ERR_OR_NULL(skb))
1000 err = udp_send_skb(skb, fl4);
1001 goto out;
1002 }
1003
1004 lock_sock(sk);
1005 if (unlikely(up->pending)) {
1006 /* The socket is already corked while preparing it. */
1007 /* ... which is an evident application bug. --ANK */
1008 release_sock(sk);
1009
1010 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1011 err = -EINVAL;
1012 goto out;
1013 }
1014 /*
1015 * Now cork the socket to pend data.
1016 */
1017 fl4 = &inet->cork.fl.u.ip4;
1018 fl4->daddr = daddr;
1019 fl4->saddr = saddr;
1020 fl4->fl4_dport = dport;
1021 fl4->fl4_sport = inet->inet_sport;
1022 up->pending = AF_INET;
1023
1024 do_append_data:
1025 up->len += ulen;
1026 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1027 sizeof(struct udphdr), &ipc, &rt,
1028 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1029 if (err)
1030 udp_flush_pending_frames(sk);
1031 else if (!corkreq)
1032 err = udp_push_pending_frames(sk);
1033 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1034 up->pending = 0;
1035 release_sock(sk);
1036
1037 out:
1038 ip_rt_put(rt);
1039 if (free)
1040 kfree(ipc.opt);
1041 if (!err)
1042 return len;
1043 /*
1044 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1045 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1046 * we don't have a good statistic (IpOutDiscards but it can be too many
1047 * things). We could add another new stat but at least for now that
1048 * seems like overkill.
1049 */
1050 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1051 UDP_INC_STATS_USER(sock_net(sk),
1052 UDP_MIB_SNDBUFERRORS, is_udplite);
1053 }
1054 return err;
1055
1056 do_confirm:
1057 dst_confirm(&rt->dst);
1058 if (!(msg->msg_flags&MSG_PROBE) || len)
1059 goto back_from_confirm;
1060 err = 0;
1061 goto out;
1062 }
1063 EXPORT_SYMBOL(udp_sendmsg);
1064
1065 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1066 size_t size, int flags)
1067 {
1068 struct inet_sock *inet = inet_sk(sk);
1069 struct udp_sock *up = udp_sk(sk);
1070 int ret;
1071
1072 if (!up->pending) {
1073 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1074
1075 /* Call udp_sendmsg to specify destination address which
1076 * sendpage interface can't pass.
1077 * This will succeed only when the socket is connected.
1078 */
1079 ret = udp_sendmsg(NULL, sk, &msg, 0);
1080 if (ret < 0)
1081 return ret;
1082 }
1083
1084 lock_sock(sk);
1085
1086 if (unlikely(!up->pending)) {
1087 release_sock(sk);
1088
1089 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1090 return -EINVAL;
1091 }
1092
1093 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1094 page, offset, size, flags);
1095 if (ret == -EOPNOTSUPP) {
1096 release_sock(sk);
1097 return sock_no_sendpage(sk->sk_socket, page, offset,
1098 size, flags);
1099 }
1100 if (ret < 0) {
1101 udp_flush_pending_frames(sk);
1102 goto out;
1103 }
1104
1105 up->len += size;
1106 if (!(up->corkflag || (flags&MSG_MORE)))
1107 ret = udp_push_pending_frames(sk);
1108 if (!ret)
1109 ret = size;
1110 out:
1111 release_sock(sk);
1112 return ret;
1113 }
1114
1115
1116 /**
1117 * first_packet_length - return length of first packet in receive queue
1118 * @sk: socket
1119 *
1120 * Drops all bad checksum frames, until a valid one is found.
1121 * Returns the length of found skb, or 0 if none is found.
1122 */
1123 static unsigned int first_packet_length(struct sock *sk)
1124 {
1125 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1126 struct sk_buff *skb;
1127 unsigned int res;
1128
1129 __skb_queue_head_init(&list_kill);
1130
1131 spin_lock_bh(&rcvq->lock);
1132 while ((skb = skb_peek(rcvq)) != NULL &&
1133 udp_lib_checksum_complete(skb)) {
1134 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1135 IS_UDPLITE(sk));
1136 atomic_inc(&sk->sk_drops);
1137 __skb_unlink(skb, rcvq);
1138 __skb_queue_tail(&list_kill, skb);
1139 }
1140 res = skb ? skb->len : 0;
1141 spin_unlock_bh(&rcvq->lock);
1142
1143 if (!skb_queue_empty(&list_kill)) {
1144 bool slow = lock_sock_fast(sk);
1145
1146 __skb_queue_purge(&list_kill);
1147 sk_mem_reclaim_partial(sk);
1148 unlock_sock_fast(sk, slow);
1149 }
1150 return res;
1151 }
1152
1153 /*
1154 * IOCTL requests applicable to the UDP protocol
1155 */
1156
1157 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1158 {
1159 switch (cmd) {
1160 case SIOCOUTQ:
1161 {
1162 int amount = sk_wmem_alloc_get(sk);
1163
1164 return put_user(amount, (int __user *)arg);
1165 }
1166
1167 case SIOCINQ:
1168 {
1169 unsigned int amount = first_packet_length(sk);
1170
1171 if (amount)
1172 /*
1173 * We will only return the amount
1174 * of this packet since that is all
1175 * that will be read.
1176 */
1177 amount -= sizeof(struct udphdr);
1178
1179 return put_user(amount, (int __user *)arg);
1180 }
1181
1182 default:
1183 return -ENOIOCTLCMD;
1184 }
1185
1186 return 0;
1187 }
1188 EXPORT_SYMBOL(udp_ioctl);
1189
1190 /*
1191 * This should be easy, if there is something there we
1192 * return it, otherwise we block.
1193 */
1194
1195 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1196 size_t len, int noblock, int flags, int *addr_len)
1197 {
1198 struct inet_sock *inet = inet_sk(sk);
1199 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1200 struct sk_buff *skb;
1201 unsigned int ulen, copied;
1202 int peeked, off = 0;
1203 int err;
1204 int is_udplite = IS_UDPLITE(sk);
1205 bool slow;
1206
1207 /*
1208 * Check any passed addresses
1209 */
1210 if (addr_len)
1211 *addr_len = sizeof(*sin);
1212
1213 if (flags & MSG_ERRQUEUE)
1214 return ip_recv_error(sk, msg, len);
1215
1216 try_again:
1217 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1218 &peeked, &off, &err);
1219 if (!skb)
1220 goto out;
1221
1222 ulen = skb->len - sizeof(struct udphdr);
1223 copied = len;
1224 if (copied > ulen)
1225 copied = ulen;
1226 else if (copied < ulen)
1227 msg->msg_flags |= MSG_TRUNC;
1228
1229 /*
1230 * If checksum is needed at all, try to do it while copying the
1231 * data. If the data is truncated, or if we only want a partial
1232 * coverage checksum (UDP-Lite), do it before the copy.
1233 */
1234
1235 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1236 if (udp_lib_checksum_complete(skb))
1237 goto csum_copy_err;
1238 }
1239
1240 if (skb_csum_unnecessary(skb))
1241 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1242 msg->msg_iov, copied);
1243 else {
1244 err = skb_copy_and_csum_datagram_iovec(skb,
1245 sizeof(struct udphdr),
1246 msg->msg_iov);
1247
1248 if (err == -EINVAL)
1249 goto csum_copy_err;
1250 }
1251
1252 if (unlikely(err)) {
1253 trace_kfree_skb(skb, udp_recvmsg);
1254 if (!peeked) {
1255 atomic_inc(&sk->sk_drops);
1256 UDP_INC_STATS_USER(sock_net(sk),
1257 UDP_MIB_INERRORS, is_udplite);
1258 }
1259 goto out_free;
1260 }
1261
1262 if (!peeked)
1263 UDP_INC_STATS_USER(sock_net(sk),
1264 UDP_MIB_INDATAGRAMS, is_udplite);
1265
1266 sock_recv_ts_and_drops(msg, sk, skb);
1267
1268 /* Copy the address. */
1269 if (sin) {
1270 sin->sin_family = AF_INET;
1271 sin->sin_port = udp_hdr(skb)->source;
1272 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1273 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1274 }
1275 if (inet->cmsg_flags)
1276 ip_cmsg_recv(msg, skb);
1277
1278 err = copied;
1279 if (flags & MSG_TRUNC)
1280 err = ulen;
1281
1282 out_free:
1283 skb_free_datagram_locked(sk, skb);
1284 out:
1285 return err;
1286
1287 csum_copy_err:
1288 slow = lock_sock_fast(sk);
1289 if (!skb_kill_datagram(sk, skb, flags))
1290 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1291 unlock_sock_fast(sk, slow);
1292
1293 if (noblock)
1294 return -EAGAIN;
1295
1296 /* starting over for a new packet */
1297 msg->msg_flags &= ~MSG_TRUNC;
1298 goto try_again;
1299 }
1300
1301
1302 int udp_disconnect(struct sock *sk, int flags)
1303 {
1304 struct inet_sock *inet = inet_sk(sk);
1305 /*
1306 * 1003.1g - break association.
1307 */
1308
1309 sk->sk_state = TCP_CLOSE;
1310 inet->inet_daddr = 0;
1311 inet->inet_dport = 0;
1312 sock_rps_reset_rxhash(sk);
1313 sk->sk_bound_dev_if = 0;
1314 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1315 inet_reset_saddr(sk);
1316
1317 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1318 sk->sk_prot->unhash(sk);
1319 inet->inet_sport = 0;
1320 }
1321 sk_dst_reset(sk);
1322 return 0;
1323 }
1324 EXPORT_SYMBOL(udp_disconnect);
1325
1326 void udp_lib_unhash(struct sock *sk)
1327 {
1328 if (sk_hashed(sk)) {
1329 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1330 struct udp_hslot *hslot, *hslot2;
1331
1332 hslot = udp_hashslot(udptable, sock_net(sk),
1333 udp_sk(sk)->udp_port_hash);
1334 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1335
1336 spin_lock_bh(&hslot->lock);
1337 if (sk_nulls_del_node_init_rcu(sk)) {
1338 hslot->count--;
1339 inet_sk(sk)->inet_num = 0;
1340 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1341
1342 spin_lock(&hslot2->lock);
1343 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1344 hslot2->count--;
1345 spin_unlock(&hslot2->lock);
1346 }
1347 spin_unlock_bh(&hslot->lock);
1348 }
1349 }
1350 EXPORT_SYMBOL(udp_lib_unhash);
1351
1352 /*
1353 * inet_rcv_saddr was changed, we must rehash secondary hash
1354 */
1355 void udp_lib_rehash(struct sock *sk, u16 newhash)
1356 {
1357 if (sk_hashed(sk)) {
1358 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1359 struct udp_hslot *hslot, *hslot2, *nhslot2;
1360
1361 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1362 nhslot2 = udp_hashslot2(udptable, newhash);
1363 udp_sk(sk)->udp_portaddr_hash = newhash;
1364 if (hslot2 != nhslot2) {
1365 hslot = udp_hashslot(udptable, sock_net(sk),
1366 udp_sk(sk)->udp_port_hash);
1367 /* we must lock primary chain too */
1368 spin_lock_bh(&hslot->lock);
1369
1370 spin_lock(&hslot2->lock);
1371 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1372 hslot2->count--;
1373 spin_unlock(&hslot2->lock);
1374
1375 spin_lock(&nhslot2->lock);
1376 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1377 &nhslot2->head);
1378 nhslot2->count++;
1379 spin_unlock(&nhslot2->lock);
1380
1381 spin_unlock_bh(&hslot->lock);
1382 }
1383 }
1384 }
1385 EXPORT_SYMBOL(udp_lib_rehash);
1386
1387 static void udp_v4_rehash(struct sock *sk)
1388 {
1389 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1390 inet_sk(sk)->inet_rcv_saddr,
1391 inet_sk(sk)->inet_num);
1392 udp_lib_rehash(sk, new_hash);
1393 }
1394
1395 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1396 {
1397 int rc;
1398
1399 if (inet_sk(sk)->inet_daddr)
1400 sock_rps_save_rxhash(sk, skb);
1401
1402 rc = sock_queue_rcv_skb(sk, skb);
1403 if (rc < 0) {
1404 int is_udplite = IS_UDPLITE(sk);
1405
1406 /* Note that an ENOMEM error is charged twice */
1407 if (rc == -ENOMEM)
1408 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1409 is_udplite);
1410 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1411 kfree_skb(skb);
1412 trace_udp_fail_queue_rcv_skb(rc, sk);
1413 return -1;
1414 }
1415
1416 return 0;
1417
1418 }
1419
1420 static struct static_key udp_encap_needed __read_mostly;
1421 void udp_encap_enable(void)
1422 {
1423 if (!static_key_enabled(&udp_encap_needed))
1424 static_key_slow_inc(&udp_encap_needed);
1425 }
1426 EXPORT_SYMBOL(udp_encap_enable);
1427
1428 /* returns:
1429 * -1: error
1430 * 0: success
1431 * >0: "udp encap" protocol resubmission
1432 *
1433 * Note that in the success and error cases, the skb is assumed to
1434 * have either been requeued or freed.
1435 */
1436 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1437 {
1438 struct udp_sock *up = udp_sk(sk);
1439 int rc;
1440 int is_udplite = IS_UDPLITE(sk);
1441
1442 /*
1443 * Charge it to the socket, dropping if the queue is full.
1444 */
1445 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1446 goto drop;
1447 nf_reset(skb);
1448
1449 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1450 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1451
1452 /*
1453 * This is an encapsulation socket so pass the skb to
1454 * the socket's udp_encap_rcv() hook. Otherwise, just
1455 * fall through and pass this up the UDP socket.
1456 * up->encap_rcv() returns the following value:
1457 * =0 if skb was successfully passed to the encap
1458 * handler or was discarded by it.
1459 * >0 if skb should be passed on to UDP.
1460 * <0 if skb should be resubmitted as proto -N
1461 */
1462
1463 /* if we're overly short, let UDP handle it */
1464 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1465 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1466 int ret;
1467
1468 ret = encap_rcv(sk, skb);
1469 if (ret <= 0) {
1470 UDP_INC_STATS_BH(sock_net(sk),
1471 UDP_MIB_INDATAGRAMS,
1472 is_udplite);
1473 return -ret;
1474 }
1475 }
1476
1477 /* FALLTHROUGH -- it's a UDP Packet */
1478 }
1479
1480 /*
1481 * UDP-Lite specific tests, ignored on UDP sockets
1482 */
1483 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1484
1485 /*
1486 * MIB statistics other than incrementing the error count are
1487 * disabled for the following two types of errors: these depend
1488 * on the application settings, not on the functioning of the
1489 * protocol stack as such.
1490 *
1491 * RFC 3828 here recommends (sec 3.3): "There should also be a
1492 * way ... to ... at least let the receiving application block
1493 * delivery of packets with coverage values less than a value
1494 * provided by the application."
1495 */
1496 if (up->pcrlen == 0) { /* full coverage was set */
1497 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1498 UDP_SKB_CB(skb)->cscov, skb->len);
1499 goto drop;
1500 }
1501 /* The next case involves violating the min. coverage requested
1502 * by the receiver. This is subtle: if receiver wants x and x is
1503 * greater than the buffersize/MTU then receiver will complain
1504 * that it wants x while sender emits packets of smaller size y.
1505 * Therefore the above ...()->partial_cov statement is essential.
1506 */
1507 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1508 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1509 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1510 goto drop;
1511 }
1512 }
1513
1514 if (rcu_access_pointer(sk->sk_filter) &&
1515 udp_lib_checksum_complete(skb))
1516 goto drop;
1517
1518
1519 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1520 goto drop;
1521
1522 rc = 0;
1523
1524 ipv4_pktinfo_prepare(skb);
1525 bh_lock_sock(sk);
1526 if (!sock_owned_by_user(sk))
1527 rc = __udp_queue_rcv_skb(sk, skb);
1528 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1529 bh_unlock_sock(sk);
1530 goto drop;
1531 }
1532 bh_unlock_sock(sk);
1533
1534 return rc;
1535
1536 drop:
1537 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1538 atomic_inc(&sk->sk_drops);
1539 kfree_skb(skb);
1540 return -1;
1541 }
1542
1543
1544 static void flush_stack(struct sock **stack, unsigned int count,
1545 struct sk_buff *skb, unsigned int final)
1546 {
1547 unsigned int i;
1548 struct sk_buff *skb1 = NULL;
1549 struct sock *sk;
1550
1551 for (i = 0; i < count; i++) {
1552 sk = stack[i];
1553 if (likely(skb1 == NULL))
1554 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1555
1556 if (!skb1) {
1557 atomic_inc(&sk->sk_drops);
1558 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1559 IS_UDPLITE(sk));
1560 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1561 IS_UDPLITE(sk));
1562 }
1563
1564 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1565 skb1 = NULL;
1566 }
1567 if (unlikely(skb1))
1568 kfree_skb(skb1);
1569 }
1570
1571 /*
1572 * Multicasts and broadcasts go to each listener.
1573 *
1574 * Note: called only from the BH handler context.
1575 */
1576 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1577 struct udphdr *uh,
1578 __be32 saddr, __be32 daddr,
1579 struct udp_table *udptable)
1580 {
1581 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1582 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1583 int dif;
1584 unsigned int i, count = 0;
1585
1586 spin_lock(&hslot->lock);
1587 sk = sk_nulls_head(&hslot->head);
1588 dif = skb->dev->ifindex;
1589 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1590 while (sk) {
1591 stack[count++] = sk;
1592 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1593 daddr, uh->source, saddr, dif);
1594 if (unlikely(count == ARRAY_SIZE(stack))) {
1595 if (!sk)
1596 break;
1597 flush_stack(stack, count, skb, ~0);
1598 count = 0;
1599 }
1600 }
1601 /*
1602 * before releasing chain lock, we must take a reference on sockets
1603 */
1604 for (i = 0; i < count; i++)
1605 sock_hold(stack[i]);
1606
1607 spin_unlock(&hslot->lock);
1608
1609 /*
1610 * do the slow work with no lock held
1611 */
1612 if (count) {
1613 flush_stack(stack, count, skb, count - 1);
1614
1615 for (i = 0; i < count; i++)
1616 sock_put(stack[i]);
1617 } else {
1618 kfree_skb(skb);
1619 }
1620 return 0;
1621 }
1622
1623 /* Initialize UDP checksum. If exited with zero value (success),
1624 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1625 * Otherwise, csum completion requires chacksumming packet body,
1626 * including udp header and folding it to skb->csum.
1627 */
1628 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1629 int proto)
1630 {
1631 const struct iphdr *iph;
1632 int err;
1633
1634 UDP_SKB_CB(skb)->partial_cov = 0;
1635 UDP_SKB_CB(skb)->cscov = skb->len;
1636
1637 if (proto == IPPROTO_UDPLITE) {
1638 err = udplite_checksum_init(skb, uh);
1639 if (err)
1640 return err;
1641 }
1642
1643 iph = ip_hdr(skb);
1644 if (uh->check == 0) {
1645 skb->ip_summed = CHECKSUM_UNNECESSARY;
1646 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1647 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1648 proto, skb->csum))
1649 skb->ip_summed = CHECKSUM_UNNECESSARY;
1650 }
1651 if (!skb_csum_unnecessary(skb))
1652 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1653 skb->len, proto, 0);
1654 /* Probably, we should checksum udp header (it should be in cache
1655 * in any case) and data in tiny packets (< rx copybreak).
1656 */
1657
1658 return 0;
1659 }
1660
1661 /*
1662 * All we need to do is get the socket, and then do a checksum.
1663 */
1664
1665 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1666 int proto)
1667 {
1668 struct sock *sk;
1669 struct udphdr *uh;
1670 unsigned short ulen;
1671 struct rtable *rt = skb_rtable(skb);
1672 __be32 saddr, daddr;
1673 struct net *net = dev_net(skb->dev);
1674
1675 /*
1676 * Validate the packet.
1677 */
1678 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1679 goto drop; /* No space for header. */
1680
1681 uh = udp_hdr(skb);
1682 ulen = ntohs(uh->len);
1683 saddr = ip_hdr(skb)->saddr;
1684 daddr = ip_hdr(skb)->daddr;
1685
1686 if (ulen > skb->len)
1687 goto short_packet;
1688
1689 if (proto == IPPROTO_UDP) {
1690 /* UDP validates ulen. */
1691 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1692 goto short_packet;
1693 uh = udp_hdr(skb);
1694 }
1695
1696 if (udp4_csum_init(skb, uh, proto))
1697 goto csum_error;
1698
1699 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1700 return __udp4_lib_mcast_deliver(net, skb, uh,
1701 saddr, daddr, udptable);
1702
1703 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1704
1705 if (sk != NULL) {
1706 int ret = udp_queue_rcv_skb(sk, skb);
1707 sock_put(sk);
1708
1709 /* a return value > 0 means to resubmit the input, but
1710 * it wants the return to be -protocol, or 0
1711 */
1712 if (ret > 0)
1713 return -ret;
1714 return 0;
1715 }
1716
1717 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1718 goto drop;
1719 nf_reset(skb);
1720
1721 /* No socket. Drop packet silently, if checksum is wrong */
1722 if (udp_lib_checksum_complete(skb))
1723 goto csum_error;
1724
1725 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1726 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1727
1728 /*
1729 * Hmm. We got an UDP packet to a port to which we
1730 * don't wanna listen. Ignore it.
1731 */
1732 kfree_skb(skb);
1733 return 0;
1734
1735 short_packet:
1736 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1737 proto == IPPROTO_UDPLITE ? "Lite" : "",
1738 &saddr, ntohs(uh->source),
1739 ulen, skb->len,
1740 &daddr, ntohs(uh->dest));
1741 goto drop;
1742
1743 csum_error:
1744 /*
1745 * RFC1122: OK. Discards the bad packet silently (as far as
1746 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1747 */
1748 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1749 proto == IPPROTO_UDPLITE ? "Lite" : "",
1750 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1751 ulen);
1752 drop:
1753 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1754 kfree_skb(skb);
1755 return 0;
1756 }
1757
1758 int udp_rcv(struct sk_buff *skb)
1759 {
1760 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1761 }
1762
1763 void udp_destroy_sock(struct sock *sk)
1764 {
1765 struct udp_sock *up = udp_sk(sk);
1766 bool slow = lock_sock_fast(sk);
1767 udp_flush_pending_frames(sk);
1768 unlock_sock_fast(sk, slow);
1769 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1770 void (*encap_destroy)(struct sock *sk);
1771 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1772 if (encap_destroy)
1773 encap_destroy(sk);
1774 }
1775 }
1776
1777 /*
1778 * Socket option code for UDP
1779 */
1780 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1781 char __user *optval, unsigned int optlen,
1782 int (*push_pending_frames)(struct sock *))
1783 {
1784 struct udp_sock *up = udp_sk(sk);
1785 int val;
1786 int err = 0;
1787 int is_udplite = IS_UDPLITE(sk);
1788
1789 if (optlen < sizeof(int))
1790 return -EINVAL;
1791
1792 if (get_user(val, (int __user *)optval))
1793 return -EFAULT;
1794
1795 switch (optname) {
1796 case UDP_CORK:
1797 if (val != 0) {
1798 up->corkflag = 1;
1799 } else {
1800 up->corkflag = 0;
1801 lock_sock(sk);
1802 (*push_pending_frames)(sk);
1803 release_sock(sk);
1804 }
1805 break;
1806
1807 case UDP_ENCAP:
1808 switch (val) {
1809 case 0:
1810 case UDP_ENCAP_ESPINUDP:
1811 case UDP_ENCAP_ESPINUDP_NON_IKE:
1812 up->encap_rcv = xfrm4_udp_encap_rcv;
1813 /* FALLTHROUGH */
1814 case UDP_ENCAP_L2TPINUDP:
1815 up->encap_type = val;
1816 udp_encap_enable();
1817 break;
1818 default:
1819 err = -ENOPROTOOPT;
1820 break;
1821 }
1822 break;
1823
1824 /*
1825 * UDP-Lite's partial checksum coverage (RFC 3828).
1826 */
1827 /* The sender sets actual checksum coverage length via this option.
1828 * The case coverage > packet length is handled by send module. */
1829 case UDPLITE_SEND_CSCOV:
1830 if (!is_udplite) /* Disable the option on UDP sockets */
1831 return -ENOPROTOOPT;
1832 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1833 val = 8;
1834 else if (val > USHRT_MAX)
1835 val = USHRT_MAX;
1836 up->pcslen = val;
1837 up->pcflag |= UDPLITE_SEND_CC;
1838 break;
1839
1840 /* The receiver specifies a minimum checksum coverage value. To make
1841 * sense, this should be set to at least 8 (as done below). If zero is
1842 * used, this again means full checksum coverage. */
1843 case UDPLITE_RECV_CSCOV:
1844 if (!is_udplite) /* Disable the option on UDP sockets */
1845 return -ENOPROTOOPT;
1846 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1847 val = 8;
1848 else if (val > USHRT_MAX)
1849 val = USHRT_MAX;
1850 up->pcrlen = val;
1851 up->pcflag |= UDPLITE_RECV_CC;
1852 break;
1853
1854 default:
1855 err = -ENOPROTOOPT;
1856 break;
1857 }
1858
1859 return err;
1860 }
1861 EXPORT_SYMBOL(udp_lib_setsockopt);
1862
1863 int udp_setsockopt(struct sock *sk, int level, int optname,
1864 char __user *optval, unsigned int optlen)
1865 {
1866 if (level == SOL_UDP || level == SOL_UDPLITE)
1867 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1868 udp_push_pending_frames);
1869 return ip_setsockopt(sk, level, optname, optval, optlen);
1870 }
1871
1872 #ifdef CONFIG_COMPAT
1873 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1874 char __user *optval, unsigned int optlen)
1875 {
1876 if (level == SOL_UDP || level == SOL_UDPLITE)
1877 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1878 udp_push_pending_frames);
1879 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1880 }
1881 #endif
1882
1883 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1884 char __user *optval, int __user *optlen)
1885 {
1886 struct udp_sock *up = udp_sk(sk);
1887 int val, len;
1888
1889 if (get_user(len, optlen))
1890 return -EFAULT;
1891
1892 len = min_t(unsigned int, len, sizeof(int));
1893
1894 if (len < 0)
1895 return -EINVAL;
1896
1897 switch (optname) {
1898 case UDP_CORK:
1899 val = up->corkflag;
1900 break;
1901
1902 case UDP_ENCAP:
1903 val = up->encap_type;
1904 break;
1905
1906 /* The following two cannot be changed on UDP sockets, the return is
1907 * always 0 (which corresponds to the full checksum coverage of UDP). */
1908 case UDPLITE_SEND_CSCOV:
1909 val = up->pcslen;
1910 break;
1911
1912 case UDPLITE_RECV_CSCOV:
1913 val = up->pcrlen;
1914 break;
1915
1916 default:
1917 return -ENOPROTOOPT;
1918 }
1919
1920 if (put_user(len, optlen))
1921 return -EFAULT;
1922 if (copy_to_user(optval, &val, len))
1923 return -EFAULT;
1924 return 0;
1925 }
1926 EXPORT_SYMBOL(udp_lib_getsockopt);
1927
1928 int udp_getsockopt(struct sock *sk, int level, int optname,
1929 char __user *optval, int __user *optlen)
1930 {
1931 if (level == SOL_UDP || level == SOL_UDPLITE)
1932 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1933 return ip_getsockopt(sk, level, optname, optval, optlen);
1934 }
1935
1936 #ifdef CONFIG_COMPAT
1937 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1938 char __user *optval, int __user *optlen)
1939 {
1940 if (level == SOL_UDP || level == SOL_UDPLITE)
1941 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1942 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1943 }
1944 #endif
1945 /**
1946 * udp_poll - wait for a UDP event.
1947 * @file - file struct
1948 * @sock - socket
1949 * @wait - poll table
1950 *
1951 * This is same as datagram poll, except for the special case of
1952 * blocking sockets. If application is using a blocking fd
1953 * and a packet with checksum error is in the queue;
1954 * then it could get return from select indicating data available
1955 * but then block when reading it. Add special case code
1956 * to work around these arguably broken applications.
1957 */
1958 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1959 {
1960 unsigned int mask = datagram_poll(file, sock, wait);
1961 struct sock *sk = sock->sk;
1962
1963 /* Check for false positives due to checksum errors */
1964 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1965 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1966 mask &= ~(POLLIN | POLLRDNORM);
1967
1968 return mask;
1969
1970 }
1971 EXPORT_SYMBOL(udp_poll);
1972
1973 struct proto udp_prot = {
1974 .name = "UDP",
1975 .owner = THIS_MODULE,
1976 .close = udp_lib_close,
1977 .connect = ip4_datagram_connect,
1978 .disconnect = udp_disconnect,
1979 .ioctl = udp_ioctl,
1980 .destroy = udp_destroy_sock,
1981 .setsockopt = udp_setsockopt,
1982 .getsockopt = udp_getsockopt,
1983 .sendmsg = udp_sendmsg,
1984 .recvmsg = udp_recvmsg,
1985 .sendpage = udp_sendpage,
1986 .backlog_rcv = __udp_queue_rcv_skb,
1987 .release_cb = ip4_datagram_release_cb,
1988 .hash = udp_lib_hash,
1989 .unhash = udp_lib_unhash,
1990 .rehash = udp_v4_rehash,
1991 .get_port = udp_v4_get_port,
1992 .memory_allocated = &udp_memory_allocated,
1993 .sysctl_mem = sysctl_udp_mem,
1994 .sysctl_wmem = &sysctl_udp_wmem_min,
1995 .sysctl_rmem = &sysctl_udp_rmem_min,
1996 .obj_size = sizeof(struct udp_sock),
1997 .slab_flags = SLAB_DESTROY_BY_RCU,
1998 .h.udp_table = &udp_table,
1999 #ifdef CONFIG_COMPAT
2000 .compat_setsockopt = compat_udp_setsockopt,
2001 .compat_getsockopt = compat_udp_getsockopt,
2002 #endif
2003 .clear_sk = sk_prot_clear_portaddr_nulls,
2004 };
2005 EXPORT_SYMBOL(udp_prot);
2006
2007 /* ------------------------------------------------------------------------ */
2008 #ifdef CONFIG_PROC_FS
2009
2010 static struct sock *udp_get_first(struct seq_file *seq, int start)
2011 {
2012 struct sock *sk;
2013 struct udp_iter_state *state = seq->private;
2014 struct net *net = seq_file_net(seq);
2015
2016 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2017 ++state->bucket) {
2018 struct hlist_nulls_node *node;
2019 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2020
2021 if (hlist_nulls_empty(&hslot->head))
2022 continue;
2023
2024 spin_lock_bh(&hslot->lock);
2025 sk_nulls_for_each(sk, node, &hslot->head) {
2026 if (!net_eq(sock_net(sk), net))
2027 continue;
2028 if (sk->sk_family == state->family)
2029 goto found;
2030 }
2031 spin_unlock_bh(&hslot->lock);
2032 }
2033 sk = NULL;
2034 found:
2035 return sk;
2036 }
2037
2038 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2039 {
2040 struct udp_iter_state *state = seq->private;
2041 struct net *net = seq_file_net(seq);
2042
2043 do {
2044 sk = sk_nulls_next(sk);
2045 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2046
2047 if (!sk) {
2048 if (state->bucket <= state->udp_table->mask)
2049 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2050 return udp_get_first(seq, state->bucket + 1);
2051 }
2052 return sk;
2053 }
2054
2055 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2056 {
2057 struct sock *sk = udp_get_first(seq, 0);
2058
2059 if (sk)
2060 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2061 --pos;
2062 return pos ? NULL : sk;
2063 }
2064
2065 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2066 {
2067 struct udp_iter_state *state = seq->private;
2068 state->bucket = MAX_UDP_PORTS;
2069
2070 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2071 }
2072
2073 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2074 {
2075 struct sock *sk;
2076
2077 if (v == SEQ_START_TOKEN)
2078 sk = udp_get_idx(seq, 0);
2079 else
2080 sk = udp_get_next(seq, v);
2081
2082 ++*pos;
2083 return sk;
2084 }
2085
2086 static void udp_seq_stop(struct seq_file *seq, void *v)
2087 {
2088 struct udp_iter_state *state = seq->private;
2089
2090 if (state->bucket <= state->udp_table->mask)
2091 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2092 }
2093
2094 int udp_seq_open(struct inode *inode, struct file *file)
2095 {
2096 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2097 struct udp_iter_state *s;
2098 int err;
2099
2100 err = seq_open_net(inode, file, &afinfo->seq_ops,
2101 sizeof(struct udp_iter_state));
2102 if (err < 0)
2103 return err;
2104
2105 s = ((struct seq_file *)file->private_data)->private;
2106 s->family = afinfo->family;
2107 s->udp_table = afinfo->udp_table;
2108 return err;
2109 }
2110 EXPORT_SYMBOL(udp_seq_open);
2111
2112 /* ------------------------------------------------------------------------ */
2113 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2114 {
2115 struct proc_dir_entry *p;
2116 int rc = 0;
2117
2118 afinfo->seq_ops.start = udp_seq_start;
2119 afinfo->seq_ops.next = udp_seq_next;
2120 afinfo->seq_ops.stop = udp_seq_stop;
2121
2122 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2123 afinfo->seq_fops, afinfo);
2124 if (!p)
2125 rc = -ENOMEM;
2126 return rc;
2127 }
2128 EXPORT_SYMBOL(udp_proc_register);
2129
2130 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2131 {
2132 remove_proc_entry(afinfo->name, net->proc_net);
2133 }
2134 EXPORT_SYMBOL(udp_proc_unregister);
2135
2136 /* ------------------------------------------------------------------------ */
2137 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2138 int bucket, int *len)
2139 {
2140 struct inet_sock *inet = inet_sk(sp);
2141 __be32 dest = inet->inet_daddr;
2142 __be32 src = inet->inet_rcv_saddr;
2143 __u16 destp = ntohs(inet->inet_dport);
2144 __u16 srcp = ntohs(inet->inet_sport);
2145
2146 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2147 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2148 bucket, src, srcp, dest, destp, sp->sk_state,
2149 sk_wmem_alloc_get(sp),
2150 sk_rmem_alloc_get(sp),
2151 0, 0L, 0,
2152 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2153 0, sock_i_ino(sp),
2154 atomic_read(&sp->sk_refcnt), sp,
2155 atomic_read(&sp->sk_drops), len);
2156 }
2157
2158 int udp4_seq_show(struct seq_file *seq, void *v)
2159 {
2160 if (v == SEQ_START_TOKEN)
2161 seq_printf(seq, "%-127s\n",
2162 " sl local_address rem_address st tx_queue "
2163 "rx_queue tr tm->when retrnsmt uid timeout "
2164 "inode ref pointer drops");
2165 else {
2166 struct udp_iter_state *state = seq->private;
2167 int len;
2168
2169 udp4_format_sock(v, seq, state->bucket, &len);
2170 seq_printf(seq, "%*s\n", 127 - len, "");
2171 }
2172 return 0;
2173 }
2174
2175 static const struct file_operations udp_afinfo_seq_fops = {
2176 .owner = THIS_MODULE,
2177 .open = udp_seq_open,
2178 .read = seq_read,
2179 .llseek = seq_lseek,
2180 .release = seq_release_net
2181 };
2182
2183 /* ------------------------------------------------------------------------ */
2184 static struct udp_seq_afinfo udp4_seq_afinfo = {
2185 .name = "udp",
2186 .family = AF_INET,
2187 .udp_table = &udp_table,
2188 .seq_fops = &udp_afinfo_seq_fops,
2189 .seq_ops = {
2190 .show = udp4_seq_show,
2191 },
2192 };
2193
2194 static int __net_init udp4_proc_init_net(struct net *net)
2195 {
2196 return udp_proc_register(net, &udp4_seq_afinfo);
2197 }
2198
2199 static void __net_exit udp4_proc_exit_net(struct net *net)
2200 {
2201 udp_proc_unregister(net, &udp4_seq_afinfo);
2202 }
2203
2204 static struct pernet_operations udp4_net_ops = {
2205 .init = udp4_proc_init_net,
2206 .exit = udp4_proc_exit_net,
2207 };
2208
2209 int __init udp4_proc_init(void)
2210 {
2211 return register_pernet_subsys(&udp4_net_ops);
2212 }
2213
2214 void udp4_proc_exit(void)
2215 {
2216 unregister_pernet_subsys(&udp4_net_ops);
2217 }
2218 #endif /* CONFIG_PROC_FS */
2219
2220 static __initdata unsigned long uhash_entries;
2221 static int __init set_uhash_entries(char *str)
2222 {
2223 ssize_t ret;
2224
2225 if (!str)
2226 return 0;
2227
2228 ret = kstrtoul(str, 0, &uhash_entries);
2229 if (ret)
2230 return 0;
2231
2232 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2233 uhash_entries = UDP_HTABLE_SIZE_MIN;
2234 return 1;
2235 }
2236 __setup("uhash_entries=", set_uhash_entries);
2237
2238 void __init udp_table_init(struct udp_table *table, const char *name)
2239 {
2240 unsigned int i;
2241
2242 table->hash = alloc_large_system_hash(name,
2243 2 * sizeof(struct udp_hslot),
2244 uhash_entries,
2245 21, /* one slot per 2 MB */
2246 0,
2247 &table->log,
2248 &table->mask,
2249 UDP_HTABLE_SIZE_MIN,
2250 64 * 1024);
2251
2252 table->hash2 = table->hash + (table->mask + 1);
2253 for (i = 0; i <= table->mask; i++) {
2254 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2255 table->hash[i].count = 0;
2256 spin_lock_init(&table->hash[i].lock);
2257 }
2258 for (i = 0; i <= table->mask; i++) {
2259 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2260 table->hash2[i].count = 0;
2261 spin_lock_init(&table->hash2[i].lock);
2262 }
2263 }
2264
2265 void __init udp_init(void)
2266 {
2267 unsigned long limit;
2268
2269 udp_table_init(&udp_table, "UDP");
2270 limit = nr_free_buffer_pages() / 8;
2271 limit = max(limit, 128UL);
2272 sysctl_udp_mem[0] = limit / 4 * 3;
2273 sysctl_udp_mem[1] = limit;
2274 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2275
2276 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2277 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2278 }
2279
2280 int udp4_ufo_send_check(struct sk_buff *skb)
2281 {
2282 const struct iphdr *iph;
2283 struct udphdr *uh;
2284
2285 if (!pskb_may_pull(skb, sizeof(*uh)))
2286 return -EINVAL;
2287
2288 iph = ip_hdr(skb);
2289 uh = udp_hdr(skb);
2290
2291 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2292 IPPROTO_UDP, 0);
2293 skb->csum_start = skb_transport_header(skb) - skb->head;
2294 skb->csum_offset = offsetof(struct udphdr, check);
2295 skb->ip_summed = CHECKSUM_PARTIAL;
2296 return 0;
2297 }
2298
2299 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
2300 netdev_features_t features)
2301 {
2302 struct sk_buff *segs = ERR_PTR(-EINVAL);
2303 unsigned int mss;
2304 int offset;
2305 __wsum csum;
2306
2307 mss = skb_shinfo(skb)->gso_size;
2308 if (unlikely(skb->len <= mss))
2309 goto out;
2310
2311 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2312 /* Packet is from an untrusted source, reset gso_segs. */
2313 int type = skb_shinfo(skb)->gso_type;
2314
2315 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY |
2316 SKB_GSO_GRE) ||
2317 !(type & (SKB_GSO_UDP))))
2318 goto out;
2319
2320 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2321
2322 segs = NULL;
2323 goto out;
2324 }
2325
2326 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2327 * do checksum of UDP packets sent as multiple IP fragments.
2328 */
2329 offset = skb_checksum_start_offset(skb);
2330 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2331 offset += skb->csum_offset;
2332 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2333 skb->ip_summed = CHECKSUM_NONE;
2334
2335 /* Fragment the skb. IP headers of the fragments are updated in
2336 * inet_gso_segment()
2337 */
2338 segs = skb_segment(skb, features);
2339 out:
2340 return segs;
2341 }
2342
ubuntu-bionic-kernel mirror