1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The User Datagram Protocol (UDP).
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
13 *
14 * Fixes:
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
28 * does NOT close.
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
57 * for connect.
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
64 * datagrams.
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
72 *
73 *
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
78 */
79
80 #define pr_fmt(fmt) "UDP: " fmt
81
82 #include <linux/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/bootmem.h>
85 #include <linux/highmem.h>
86 #include <linux/swap.h>
87 #include <linux/types.h>
88 #include <linux/fcntl.h>
89 #include <linux/module.h>
90 #include <linux/socket.h>
91 #include <linux/sockios.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
94 #include <linux/in.h>
95 #include <linux/errno.h>
96 #include <linux/timer.h>
97 #include <linux/mm.h>
98 #include <linux/inet.h>
99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/route.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <trace/events/udp.h>
112 #include <linux/static_key.h>
113 #include <trace/events/skb.h>
114 #include <net/busy_poll.h>
115 #include "udp_impl.h"
116 #include <net/sock_reuseport.h>
117 #include <net/addrconf.h>
118
119 struct udp_table udp_table __read_mostly;
120 EXPORT_SYMBOL(udp_table);
121
122 long sysctl_udp_mem[3] __read_mostly;
123 EXPORT_SYMBOL(sysctl_udp_mem);
124
125 int sysctl_udp_rmem_min __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_rmem_min);
127
128 int sysctl_udp_wmem_min __read_mostly;
129 EXPORT_SYMBOL(sysctl_udp_wmem_min);
130
131 atomic_long_t udp_memory_allocated;
132 EXPORT_SYMBOL(udp_memory_allocated);
133
134 #define MAX_UDP_PORTS 65536
135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
136
137 /* IPCB reference means this can not be used from early demux */
udp_lib_exact_dif_match(struct net * net,struct sk_buff * skb)138 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
139 {
140 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
141 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
142 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
143 return true;
144 #endif
145 return false;
146 }
147
udp_lib_lport_inuse(struct net * net,__u16 num,const struct udp_hslot * hslot,unsigned long * bitmap,struct sock * sk,unsigned int log)148 static int udp_lib_lport_inuse(struct net *net, __u16 num,
149 const struct udp_hslot *hslot,
150 unsigned long *bitmap,
151 struct sock *sk, unsigned int log)
152 {
153 struct sock *sk2;
154 kuid_t uid = sock_i_uid(sk);
155
156 sk_for_each(sk2, &hslot->head) {
157 if (net_eq(sock_net(sk2), net) &&
158 sk2 != sk &&
159 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
160 (!sk2->sk_reuse || !sk->sk_reuse) &&
161 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
162 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
163 inet_rcv_saddr_equal(sk, sk2, true)) {
164 if (sk2->sk_reuseport && sk->sk_reuseport &&
165 !rcu_access_pointer(sk->sk_reuseport_cb) &&
166 uid_eq(uid, sock_i_uid(sk2))) {
167 if (!bitmap)
168 return 0;
169 } else {
170 if (!bitmap)
171 return 1;
172 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
173 bitmap);
174 }
175 }
176 }
177 return 0;
178 }
179
180 /*
181 * Note: we still hold spinlock of primary hash chain, so no other writer
182 * can insert/delete a socket with local_port == num
183 */
udp_lib_lport_inuse2(struct net * net,__u16 num,struct udp_hslot * hslot2,struct sock * sk)184 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
185 struct udp_hslot *hslot2,
186 struct sock *sk)
187 {
188 struct sock *sk2;
189 kuid_t uid = sock_i_uid(sk);
190 int res = 0;
191
192 spin_lock(&hslot2->lock);
193 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
194 if (net_eq(sock_net(sk2), net) &&
195 sk2 != sk &&
196 (udp_sk(sk2)->udp_port_hash == num) &&
197 (!sk2->sk_reuse || !sk->sk_reuse) &&
198 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
199 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
200 inet_rcv_saddr_equal(sk, sk2, true)) {
201 if (sk2->sk_reuseport && sk->sk_reuseport &&
202 !rcu_access_pointer(sk->sk_reuseport_cb) &&
203 uid_eq(uid, sock_i_uid(sk2))) {
204 res = 0;
205 } else {
206 res = 1;
207 }
208 break;
209 }
210 }
211 spin_unlock(&hslot2->lock);
212 return res;
213 }
214
udp_reuseport_add_sock(struct sock * sk,struct udp_hslot * hslot)215 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
216 {
217 struct net *net = sock_net(sk);
218 kuid_t uid = sock_i_uid(sk);
219 struct sock *sk2;
220
221 sk_for_each(sk2, &hslot->head) {
222 if (net_eq(sock_net(sk2), net) &&
223 sk2 != sk &&
224 sk2->sk_family == sk->sk_family &&
225 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
226 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
227 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
228 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
229 inet_rcv_saddr_equal(sk, sk2, false)) {
230 return reuseport_add_sock(sk, sk2);
231 }
232 }
233
234 return reuseport_alloc(sk);
235 }
236
237 /**
238 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
239 *
240 * @sk: socket struct in question
241 * @snum: port number to look up
242 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
243 * with NULL address
244 */
udp_lib_get_port(struct sock * sk,unsigned short snum,unsigned int hash2_nulladdr)245 int udp_lib_get_port(struct sock *sk, unsigned short snum,
246 unsigned int hash2_nulladdr)
247 {
248 struct udp_hslot *hslot, *hslot2;
249 struct udp_table *udptable = sk->sk_prot->h.udp_table;
250 int error = 1;
251 struct net *net = sock_net(sk);
252
253 if (!snum) {
254 int low, high, remaining;
255 unsigned int rand;
256 unsigned short first, last;
257 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
258
259 inet_get_local_port_range(net, &low, &high);
260 remaining = (high - low) + 1;
261
262 rand = prandom_u32();
263 first = reciprocal_scale(rand, remaining) + low;
264 /*
265 * force rand to be an odd multiple of UDP_HTABLE_SIZE
266 */
267 rand = (rand | 1) * (udptable->mask + 1);
268 last = first + udptable->mask + 1;
269 do {
270 hslot = udp_hashslot(udptable, net, first);
271 bitmap_zero(bitmap, PORTS_PER_CHAIN);
272 spin_lock_bh(&hslot->lock);
273 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
274 udptable->log);
275
276 snum = first;
277 /*
278 * Iterate on all possible values of snum for this hash.
279 * Using steps of an odd multiple of UDP_HTABLE_SIZE
280 * give us randomization and full range coverage.
281 */
282 do {
283 if (low <= snum && snum <= high &&
284 !test_bit(snum >> udptable->log, bitmap) &&
285 !inet_is_local_reserved_port(net, snum))
286 goto found;
287 snum += rand;
288 } while (snum != first);
289 spin_unlock_bh(&hslot->lock);
290 cond_resched();
291 } while (++first != last);
292 goto fail;
293 } else {
294 hslot = udp_hashslot(udptable, net, snum);
295 spin_lock_bh(&hslot->lock);
296 if (hslot->count > 10) {
297 int exist;
298 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
299
300 slot2 &= udptable->mask;
301 hash2_nulladdr &= udptable->mask;
302
303 hslot2 = udp_hashslot2(udptable, slot2);
304 if (hslot->count < hslot2->count)
305 goto scan_primary_hash;
306
307 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
308 if (!exist && (hash2_nulladdr != slot2)) {
309 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
310 exist = udp_lib_lport_inuse2(net, snum, hslot2,
311 sk);
312 }
313 if (exist)
314 goto fail_unlock;
315 else
316 goto found;
317 }
318 scan_primary_hash:
319 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
320 goto fail_unlock;
321 }
322 found:
323 inet_sk(sk)->inet_num = snum;
324 udp_sk(sk)->udp_port_hash = snum;
325 udp_sk(sk)->udp_portaddr_hash ^= snum;
326 if (sk_unhashed(sk)) {
327 if (sk->sk_reuseport &&
328 udp_reuseport_add_sock(sk, hslot)) {
329 inet_sk(sk)->inet_num = 0;
330 udp_sk(sk)->udp_port_hash = 0;
331 udp_sk(sk)->udp_portaddr_hash ^= snum;
332 goto fail_unlock;
333 }
334
335 sk_add_node_rcu(sk, &hslot->head);
336 hslot->count++;
337 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
338
339 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
340 spin_lock(&hslot2->lock);
341 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
342 sk->sk_family == AF_INET6)
343 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
344 &hslot2->head);
345 else
346 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
347 &hslot2->head);
348 hslot2->count++;
349 spin_unlock(&hslot2->lock);
350 }
351 sock_set_flag(sk, SOCK_RCU_FREE);
352 error = 0;
353 fail_unlock:
354 spin_unlock_bh(&hslot->lock);
355 fail:
356 return error;
357 }
358 EXPORT_SYMBOL(udp_lib_get_port);
359
udp4_portaddr_hash(const struct net * net,__be32 saddr,unsigned int port)360 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
361 unsigned int port)
362 {
363 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
364 }
365
udp_v4_get_port(struct sock * sk,unsigned short snum)366 int udp_v4_get_port(struct sock *sk, unsigned short snum)
367 {
368 unsigned int hash2_nulladdr =
369 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
370 unsigned int hash2_partial =
371 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
372
373 /* precompute partial secondary hash */
374 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
375 return udp_lib_get_port(sk, snum, hash2_nulladdr);
376 }
377
compute_score(struct sock * sk,struct net * net,__be32 saddr,__be16 sport,__be32 daddr,unsigned short hnum,int dif,int sdif,bool exact_dif)378 static int compute_score(struct sock *sk, struct net *net,
379 __be32 saddr, __be16 sport,
380 __be32 daddr, unsigned short hnum,
381 int dif, int sdif, bool exact_dif)
382 {
383 int score;
384 struct inet_sock *inet;
385
386 if (!net_eq(sock_net(sk), net) ||
387 udp_sk(sk)->udp_port_hash != hnum ||
388 ipv6_only_sock(sk))
389 return -1;
390
391 score = (sk->sk_family == PF_INET) ? 2 : 1;
392 inet = inet_sk(sk);
393
394 if (inet->inet_rcv_saddr) {
395 if (inet->inet_rcv_saddr != daddr)
396 return -1;
397 score += 4;
398 }
399
400 if (inet->inet_daddr) {
401 if (inet->inet_daddr != saddr)
402 return -1;
403 score += 4;
404 }
405
406 if (inet->inet_dport) {
407 if (inet->inet_dport != sport)
408 return -1;
409 score += 4;
410 }
411
412 if (sk->sk_bound_dev_if || exact_dif) {
413 bool dev_match = (sk->sk_bound_dev_if == dif ||
414 sk->sk_bound_dev_if == sdif);
415
416 if (!dev_match)
417 return -1;
418 if (sk->sk_bound_dev_if)
419 score += 4;
420 }
421
422 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
423 score++;
424 return score;
425 }
426
udp_ehashfn(const struct net * net,const __be32 laddr,const __u16 lport,const __be32 faddr,const __be16 fport)427 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
428 const __u16 lport, const __be32 faddr,
429 const __be16 fport)
430 {
431 static u32 udp_ehash_secret __read_mostly;
432
433 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
434
435 return __inet_ehashfn(laddr, lport, faddr, fport,
436 udp_ehash_secret + net_hash_mix(net));
437 }
438
439 /* called with rcu_read_lock() */
udp4_lib_lookup2(struct net * net,__be32 saddr,__be16 sport,__be32 daddr,unsigned int hnum,int dif,int sdif,bool exact_dif,struct udp_hslot * hslot2,struct sk_buff * skb)440 static struct sock *udp4_lib_lookup2(struct net *net,
441 __be32 saddr, __be16 sport,
442 __be32 daddr, unsigned int hnum,
443 int dif, int sdif, bool exact_dif,
444 struct udp_hslot *hslot2,
445 struct sk_buff *skb)
446 {
447 struct sock *sk, *result;
448 int score, badness, matches = 0, reuseport = 0;
449 u32 hash = 0;
450
451 result = NULL;
452 badness = 0;
453 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
454 score = compute_score(sk, net, saddr, sport,
455 daddr, hnum, dif, sdif, exact_dif);
456 if (score > badness) {
457 reuseport = sk->sk_reuseport;
458 if (reuseport) {
459 hash = udp_ehashfn(net, daddr, hnum,
460 saddr, sport);
461 result = reuseport_select_sock(sk, hash, skb,
462 sizeof(struct udphdr));
463 if (result)
464 return result;
465 matches = 1;
466 }
467 badness = score;
468 result = sk;
469 } else if (score == badness && reuseport) {
470 matches++;
471 if (reciprocal_scale(hash, matches) == 0)
472 result = sk;
473 hash = next_pseudo_random32(hash);
474 }
475 }
476 return result;
477 }
478
479 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
480 * harder than this. -DaveM
481 */
__udp4_lib_lookup(struct net * net,__be32 saddr,__be16 sport,__be32 daddr,__be16 dport,int dif,int sdif,struct udp_table * udptable,struct sk_buff * skb)482 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
483 __be16 sport, __be32 daddr, __be16 dport, int dif,
484 int sdif, struct udp_table *udptable, struct sk_buff *skb)
485 {
486 struct sock *sk, *result;
487 unsigned short hnum = ntohs(dport);
488 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
489 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
490 bool exact_dif = udp_lib_exact_dif_match(net, skb);
491 int score, badness, matches = 0, reuseport = 0;
492 u32 hash = 0;
493
494 if (hslot->count > 10) {
495 hash2 = udp4_portaddr_hash(net, daddr, hnum);
496 slot2 = hash2 & udptable->mask;
497 hslot2 = &udptable->hash2[slot2];
498 if (hslot->count < hslot2->count)
499 goto begin;
500
501 result = udp4_lib_lookup2(net, saddr, sport,
502 daddr, hnum, dif, sdif,
503 exact_dif, hslot2, skb);
504 if (!result) {
505 unsigned int old_slot2 = slot2;
506 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
507 slot2 = hash2 & udptable->mask;
508 /* avoid searching the same slot again. */
509 if (unlikely(slot2 == old_slot2))
510 return result;
511
512 hslot2 = &udptable->hash2[slot2];
513 if (hslot->count < hslot2->count)
514 goto begin;
515
516 result = udp4_lib_lookup2(net, saddr, sport,
517 daddr, hnum, dif, sdif,
518 exact_dif, hslot2, skb);
519 }
520 return result;
521 }
522 begin:
523 result = NULL;
524 badness = 0;
525 sk_for_each_rcu(sk, &hslot->head) {
526 score = compute_score(sk, net, saddr, sport,
527 daddr, hnum, dif, sdif, exact_dif);
528 if (score > badness) {
529 reuseport = sk->sk_reuseport;
530 if (reuseport) {
531 hash = udp_ehashfn(net, daddr, hnum,
532 saddr, sport);
533 result = reuseport_select_sock(sk, hash, skb,
534 sizeof(struct udphdr));
535 if (result)
536 return result;
537 matches = 1;
538 }
539 result = sk;
540 badness = score;
541 } else if (score == badness && reuseport) {
542 matches++;
543 if (reciprocal_scale(hash, matches) == 0)
544 result = sk;
545 hash = next_pseudo_random32(hash);
546 }
547 }
548 return result;
549 }
550 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
551
__udp4_lib_lookup_skb(struct sk_buff * skb,__be16 sport,__be16 dport,struct udp_table * udptable)552 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
553 __be16 sport, __be16 dport,
554 struct udp_table *udptable)
555 {
556 const struct iphdr *iph = ip_hdr(skb);
557
558 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
559 iph->daddr, dport, inet_iif(skb),
560 inet_sdif(skb), udptable, skb);
561 }
562
udp4_lib_lookup_skb(struct sk_buff * skb,__be16 sport,__be16 dport)563 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
564 __be16 sport, __be16 dport)
565 {
566 const struct iphdr *iph = ip_hdr(skb);
567
568 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
569 iph->daddr, dport, inet_iif(skb),
570 inet_sdif(skb), &udp_table, NULL);
571 }
572 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
573
574 /* Must be called under rcu_read_lock().
575 * Does increment socket refcount.
576 */
577 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
578 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
579 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
udp4_lib_lookup(struct net * net,__be32 saddr,__be16 sport,__be32 daddr,__be16 dport,int dif)580 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
581 __be32 daddr, __be16 dport, int dif)
582 {
583 struct sock *sk;
584
585 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
586 dif, 0, &udp_table, NULL);
587 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
588 sk = NULL;
589 return sk;
590 }
591 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
592 #endif
593
__udp_is_mcast_sock(struct net * net,struct sock * sk,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif,unsigned short hnum)594 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
595 __be16 loc_port, __be32 loc_addr,
596 __be16 rmt_port, __be32 rmt_addr,
597 int dif, int sdif, unsigned short hnum)
598 {
599 struct inet_sock *inet = inet_sk(sk);
600
601 if (!net_eq(sock_net(sk), net) ||
602 udp_sk(sk)->udp_port_hash != hnum ||
603 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
604 (inet->inet_dport != rmt_port && inet->inet_dport) ||
605 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
606 ipv6_only_sock(sk) ||
607 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
608 sk->sk_bound_dev_if != sdif))
609 return false;
610 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
611 return false;
612 return true;
613 }
614
615 /*
616 * This routine is called by the ICMP module when it gets some
617 * sort of error condition. If err < 0 then the socket should
618 * be closed and the error returned to the user. If err > 0
619 * it's just the icmp type << 8 | icmp code.
620 * Header points to the ip header of the error packet. We move
621 * on past this. Then (as it used to claim before adjustment)
622 * header points to the first 8 bytes of the udp header. We need
623 * to find the appropriate port.
624 */
625
__udp4_lib_err(struct sk_buff * skb,u32 info,struct udp_table * udptable)626 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
627 {
628 struct inet_sock *inet;
629 const struct iphdr *iph = (const struct iphdr *)skb->data;
630 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
631 const int type = icmp_hdr(skb)->type;
632 const int code = icmp_hdr(skb)->code;
633 struct sock *sk;
634 int harderr;
635 int err;
636 struct net *net = dev_net(skb->dev);
637
638 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
639 iph->saddr, uh->source, skb->dev->ifindex, 0,
640 udptable, NULL);
641 if (!sk) {
642 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
643 return; /* No socket for error */
644 }
645
646 err = 0;
647 harderr = 0;
648 inet = inet_sk(sk);
649
650 switch (type) {
651 default:
652 case ICMP_TIME_EXCEEDED:
653 err = EHOSTUNREACH;
654 break;
655 case ICMP_SOURCE_QUENCH:
656 goto out;
657 case ICMP_PARAMETERPROB:
658 err = EPROTO;
659 harderr = 1;
660 break;
661 case ICMP_DEST_UNREACH:
662 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
663 ipv4_sk_update_pmtu(skb, sk, info);
664 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
665 err = EMSGSIZE;
666 harderr = 1;
667 break;
668 }
669 goto out;
670 }
671 err = EHOSTUNREACH;
672 if (code <= NR_ICMP_UNREACH) {
673 harderr = icmp_err_convert[code].fatal;
674 err = icmp_err_convert[code].errno;
675 }
676 break;
677 case ICMP_REDIRECT:
678 ipv4_sk_redirect(skb, sk);
679 goto out;
680 }
681
682 /*
683 * RFC1122: OK. Passes ICMP errors back to application, as per
684 * 4.1.3.3.
685 */
686 if (!inet->recverr) {
687 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
688 goto out;
689 } else
690 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
691
692 sk->sk_err = err;
693 sk->sk_error_report(sk);
694 out:
695 return;
696 }
697
udp_err(struct sk_buff * skb,u32 info)698 void udp_err(struct sk_buff *skb, u32 info)
699 {
700 __udp4_lib_err(skb, info, &udp_table);
701 }
702
703 /*
704 * Throw away all pending data and cancel the corking. Socket is locked.
705 */
udp_flush_pending_frames(struct sock * sk)706 void udp_flush_pending_frames(struct sock *sk)
707 {
708 struct udp_sock *up = udp_sk(sk);
709
710 if (up->pending) {
711 up->len = 0;
712 up->pending = 0;
713 ip_flush_pending_frames(sk);
714 }
715 }
716 EXPORT_SYMBOL(udp_flush_pending_frames);
717
718 /**
719 * udp4_hwcsum - handle outgoing HW checksumming
720 * @skb: sk_buff containing the filled-in UDP header
721 * (checksum field must be zeroed out)
722 * @src: source IP address
723 * @dst: destination IP address
724 */
udp4_hwcsum(struct sk_buff * skb,__be32 src,__be32 dst)725 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
726 {
727 struct udphdr *uh = udp_hdr(skb);
728 int offset = skb_transport_offset(skb);
729 int len = skb->len - offset;
730 int hlen = len;
731 __wsum csum = 0;
732
733 if (!skb_has_frag_list(skb)) {
734 /*
735 * Only one fragment on the socket.
736 */
737 skb->csum_start = skb_transport_header(skb) - skb->head;
738 skb->csum_offset = offsetof(struct udphdr, check);
739 uh->check = ~csum_tcpudp_magic(src, dst, len,
740 IPPROTO_UDP, 0);
741 } else {
742 struct sk_buff *frags;
743
744 /*
745 * HW-checksum won't work as there are two or more
746 * fragments on the socket so that all csums of sk_buffs
747 * should be together
748 */
749 skb_walk_frags(skb, frags) {
750 csum = csum_add(csum, frags->csum);
751 hlen -= frags->len;
752 }
753
754 csum = skb_checksum(skb, offset, hlen, csum);
755 skb->ip_summed = CHECKSUM_NONE;
756
757 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
758 if (uh->check == 0)
759 uh->check = CSUM_MANGLED_0;
760 }
761 }
762 EXPORT_SYMBOL_GPL(udp4_hwcsum);
763
764 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
765 * for the simple case like when setting the checksum for a UDP tunnel.
766 */
udp_set_csum(bool nocheck,struct sk_buff * skb,__be32 saddr,__be32 daddr,int len)767 void udp_set_csum(bool nocheck, struct sk_buff *skb,
768 __be32 saddr, __be32 daddr, int len)
769 {
770 struct udphdr *uh = udp_hdr(skb);
771
772 if (nocheck) {
773 uh->check = 0;
774 } else if (skb_is_gso(skb)) {
775 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
776 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
777 uh->check = 0;
778 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
779 if (uh->check == 0)
780 uh->check = CSUM_MANGLED_0;
781 } else {
782 skb->ip_summed = CHECKSUM_PARTIAL;
783 skb->csum_start = skb_transport_header(skb) - skb->head;
784 skb->csum_offset = offsetof(struct udphdr, check);
785 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
786 }
787 }
788 EXPORT_SYMBOL(udp_set_csum);
789
udp_send_skb(struct sk_buff * skb,struct flowi4 * fl4)790 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
791 {
792 struct sock *sk = skb->sk;
793 struct inet_sock *inet = inet_sk(sk);
794 struct udphdr *uh;
795 int err = 0;
796 int is_udplite = IS_UDPLITE(sk);
797 int offset = skb_transport_offset(skb);
798 int len = skb->len - offset;
799 __wsum csum = 0;
800
801 /*
802 * Create a UDP header
803 */
804 uh = udp_hdr(skb);
805 uh->source = inet->inet_sport;
806 uh->dest = fl4->fl4_dport;
807 uh->len = htons(len);
808 uh->check = 0;
809
810 if (is_udplite) /* UDP-Lite */
811 csum = udplite_csum(skb);
812
813 else if (sk->sk_no_check_tx) { /* UDP csum off */
814
815 skb->ip_summed = CHECKSUM_NONE;
816 goto send;
817
818 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
819
820 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
821 goto send;
822
823 } else
824 csum = udp_csum(skb);
825
826 /* add protocol-dependent pseudo-header */
827 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
828 sk->sk_protocol, csum);
829 if (uh->check == 0)
830 uh->check = CSUM_MANGLED_0;
831
832 send:
833 err = ip_send_skb(sock_net(sk), skb);
834 if (err) {
835 if (err == -ENOBUFS && !inet->recverr) {
836 UDP_INC_STATS(sock_net(sk),
837 UDP_MIB_SNDBUFERRORS, is_udplite);
838 err = 0;
839 }
840 } else
841 UDP_INC_STATS(sock_net(sk),
842 UDP_MIB_OUTDATAGRAMS, is_udplite);
843 return err;
844 }
845
846 /*
847 * Push out all pending data as one UDP datagram. Socket is locked.
848 */
udp_push_pending_frames(struct sock * sk)849 int udp_push_pending_frames(struct sock *sk)
850 {
851 struct udp_sock *up = udp_sk(sk);
852 struct inet_sock *inet = inet_sk(sk);
853 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
854 struct sk_buff *skb;
855 int err = 0;
856
857 skb = ip_finish_skb(sk, fl4);
858 if (!skb)
859 goto out;
860
861 err = udp_send_skb(skb, fl4);
862
863 out:
864 up->len = 0;
865 up->pending = 0;
866 return err;
867 }
868 EXPORT_SYMBOL(udp_push_pending_frames);
869
udp_sendmsg(struct sock * sk,struct msghdr * msg,size_t len)870 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
871 {
872 struct inet_sock *inet = inet_sk(sk);
873 struct udp_sock *up = udp_sk(sk);
874 struct flowi4 fl4_stack;
875 struct flowi4 *fl4;
876 int ulen = len;
877 struct ipcm_cookie ipc;
878 struct rtable *rt = NULL;
879 int free = 0;
880 int connected = 0;
881 __be32 daddr, faddr, saddr;
882 __be16 dport;
883 u8 tos;
884 int err, is_udplite = IS_UDPLITE(sk);
885 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
886 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
887 struct sk_buff *skb;
888 struct ip_options_data opt_copy;
889
890 if (len > 0xFFFF)
891 return -EMSGSIZE;
892
893 /*
894 * Check the flags.
895 */
896
897 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
898 return -EOPNOTSUPP;
899
900 ipc.opt = NULL;
901 ipc.tx_flags = 0;
902 ipc.ttl = 0;
903 ipc.tos = -1;
904
905 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
906
907 fl4 = &inet->cork.fl.u.ip4;
908 if (up->pending) {
909 /*
910 * There are pending frames.
911 * The socket lock must be held while it's corked.
912 */
913 lock_sock(sk);
914 if (likely(up->pending)) {
915 if (unlikely(up->pending != AF_INET)) {
916 release_sock(sk);
917 return -EINVAL;
918 }
919 goto do_append_data;
920 }
921 release_sock(sk);
922 }
923 ulen += sizeof(struct udphdr);
924
925 /*
926 * Get and verify the address.
927 */
928 if (msg->msg_name) {
929 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
930 if (msg->msg_namelen < sizeof(*usin))
931 return -EINVAL;
932 if (usin->sin_family != AF_INET) {
933 if (usin->sin_family != AF_UNSPEC)
934 return -EAFNOSUPPORT;
935 }
936
937 daddr = usin->sin_addr.s_addr;
938 dport = usin->sin_port;
939 if (dport == 0)
940 return -EINVAL;
941 } else {
942 if (sk->sk_state != TCP_ESTABLISHED)
943 return -EDESTADDRREQ;
944 daddr = inet->inet_daddr;
945 dport = inet->inet_dport;
946 /* Open fast path for connected socket.
947 Route will not be used, if at least one option is set.
948 */
949 connected = 1;
950 }
951
952 ipc.sockc.tsflags = sk->sk_tsflags;
953 ipc.addr = inet->inet_saddr;
954 ipc.oif = sk->sk_bound_dev_if;
955
956 if (msg->msg_controllen) {
957 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
958 if (unlikely(err)) {
959 kfree(ipc.opt);
960 return err;
961 }
962 if (ipc.opt)
963 free = 1;
964 connected = 0;
965 }
966 if (!ipc.opt) {
967 struct ip_options_rcu *inet_opt;
968
969 rcu_read_lock();
970 inet_opt = rcu_dereference(inet->inet_opt);
971 if (inet_opt) {
972 memcpy(&opt_copy, inet_opt,
973 sizeof(*inet_opt) + inet_opt->opt.optlen);
974 ipc.opt = &opt_copy.opt;
975 }
976 rcu_read_unlock();
977 }
978
979 saddr = ipc.addr;
980 ipc.addr = faddr = daddr;
981
982 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
983
984 if (ipc.opt && ipc.opt->opt.srr) {
985 if (!daddr) {
986 err = -EINVAL;
987 goto out_free;
988 }
989 faddr = ipc.opt->opt.faddr;
990 connected = 0;
991 }
992 tos = get_rttos(&ipc, inet);
993 if (sock_flag(sk, SOCK_LOCALROUTE) ||
994 (msg->msg_flags & MSG_DONTROUTE) ||
995 (ipc.opt && ipc.opt->opt.is_strictroute)) {
996 tos |= RTO_ONLINK;
997 connected = 0;
998 }
999
1000 if (ipv4_is_multicast(daddr)) {
1001 if (!ipc.oif)
1002 ipc.oif = inet->mc_index;
1003 if (!saddr)
1004 saddr = inet->mc_addr;
1005 connected = 0;
1006 } else if (!ipc.oif)
1007 ipc.oif = inet->uc_index;
1008
1009 if (connected)
1010 rt = (struct rtable *)sk_dst_check(sk, 0);
1011
1012 if (!rt) {
1013 struct net *net = sock_net(sk);
1014 __u8 flow_flags = inet_sk_flowi_flags(sk);
1015
1016 fl4 = &fl4_stack;
1017
1018 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1019 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1020 flow_flags,
1021 faddr, saddr, dport, inet->inet_sport,
1022 sk->sk_uid);
1023
1024 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1025 rt = ip_route_output_flow(net, fl4, sk);
1026 if (IS_ERR(rt)) {
1027 err = PTR_ERR(rt);
1028 rt = NULL;
1029 if (err == -ENETUNREACH)
1030 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1031 goto out;
1032 }
1033
1034 err = -EACCES;
1035 if ((rt->rt_flags & RTCF_BROADCAST) &&
1036 !sock_flag(sk, SOCK_BROADCAST))
1037 goto out;
1038 if (connected)
1039 sk_dst_set(sk, dst_clone(&rt->dst));
1040 }
1041
1042 if (msg->msg_flags&MSG_CONFIRM)
1043 goto do_confirm;
1044 back_from_confirm:
1045
1046 saddr = fl4->saddr;
1047 if (!ipc.addr)
1048 daddr = ipc.addr = fl4->daddr;
1049
1050 /* Lockless fast path for the non-corking case. */
1051 if (!corkreq) {
1052 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1053 sizeof(struct udphdr), &ipc, &rt,
1054 msg->msg_flags);
1055 err = PTR_ERR(skb);
1056 if (!IS_ERR_OR_NULL(skb))
1057 err = udp_send_skb(skb, fl4);
1058 goto out;
1059 }
1060
1061 lock_sock(sk);
1062 if (unlikely(up->pending)) {
1063 /* The socket is already corked while preparing it. */
1064 /* ... which is an evident application bug. --ANK */
1065 release_sock(sk);
1066
1067 net_dbg_ratelimited("socket already corked\n");
1068 err = -EINVAL;
1069 goto out;
1070 }
1071 /*
1072 * Now cork the socket to pend data.
1073 */
1074 fl4 = &inet->cork.fl.u.ip4;
1075 fl4->daddr = daddr;
1076 fl4->saddr = saddr;
1077 fl4->fl4_dport = dport;
1078 fl4->fl4_sport = inet->inet_sport;
1079 up->pending = AF_INET;
1080
1081 do_append_data:
1082 up->len += ulen;
1083 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1084 sizeof(struct udphdr), &ipc, &rt,
1085 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1086 if (err)
1087 udp_flush_pending_frames(sk);
1088 else if (!corkreq)
1089 err = udp_push_pending_frames(sk);
1090 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1091 up->pending = 0;
1092 release_sock(sk);
1093
1094 out:
1095 ip_rt_put(rt);
1096 out_free:
1097 if (free)
1098 kfree(ipc.opt);
1099 if (!err)
1100 return len;
1101 /*
1102 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1103 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1104 * we don't have a good statistic (IpOutDiscards but it can be too many
1105 * things). We could add another new stat but at least for now that
1106 * seems like overkill.
1107 */
1108 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1109 UDP_INC_STATS(sock_net(sk),
1110 UDP_MIB_SNDBUFERRORS, is_udplite);
1111 }
1112 return err;
1113
1114 do_confirm:
1115 if (msg->msg_flags & MSG_PROBE)
1116 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1117 if (!(msg->msg_flags&MSG_PROBE) || len)
1118 goto back_from_confirm;
1119 err = 0;
1120 goto out;
1121 }
1122 EXPORT_SYMBOL(udp_sendmsg);
1123
udp_sendpage(struct sock * sk,struct page * page,int offset,size_t size,int flags)1124 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1125 size_t size, int flags)
1126 {
1127 struct inet_sock *inet = inet_sk(sk);
1128 struct udp_sock *up = udp_sk(sk);
1129 int ret;
1130
1131 if (flags & MSG_SENDPAGE_NOTLAST)
1132 flags |= MSG_MORE;
1133
1134 if (!up->pending) {
1135 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1136
1137 /* Call udp_sendmsg to specify destination address which
1138 * sendpage interface can't pass.
1139 * This will succeed only when the socket is connected.
1140 */
1141 ret = udp_sendmsg(sk, &msg, 0);
1142 if (ret < 0)
1143 return ret;
1144 }
1145
1146 lock_sock(sk);
1147
1148 if (unlikely(!up->pending)) {
1149 release_sock(sk);
1150
1151 net_dbg_ratelimited("cork failed\n");
1152 return -EINVAL;
1153 }
1154
1155 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1156 page, offset, size, flags);
1157 if (ret == -EOPNOTSUPP) {
1158 release_sock(sk);
1159 return sock_no_sendpage(sk->sk_socket, page, offset,
1160 size, flags);
1161 }
1162 if (ret < 0) {
1163 udp_flush_pending_frames(sk);
1164 goto out;
1165 }
1166
1167 up->len += size;
1168 if (!(up->corkflag || (flags&MSG_MORE)))
1169 ret = udp_push_pending_frames(sk);
1170 if (!ret)
1171 ret = size;
1172 out:
1173 release_sock(sk);
1174 return ret;
1175 }
1176
1177 #define UDP_SKB_IS_STATELESS 0x80000000
1178
udp_set_dev_scratch(struct sk_buff * skb)1179 static void udp_set_dev_scratch(struct sk_buff *skb)
1180 {
1181 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1182
1183 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1184 scratch->_tsize_state = skb->truesize;
1185 #if BITS_PER_LONG == 64
1186 scratch->len = skb->len;
1187 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1188 scratch->is_linear = !skb_is_nonlinear(skb);
1189 #endif
1190 /* all head states execept sp (dst, sk, nf) are always cleared by
1191 * udp_rcv() and we need to preserve secpath, if present, to eventually
1192 * process IP_CMSG_PASSSEC at recvmsg() time
1193 */
1194 if (likely(!skb_sec_path(skb)))
1195 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1196 }
1197
udp_skb_csum_unnecessary_set(struct sk_buff * skb)1198 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1199 {
1200 /* We come here after udp_lib_checksum_complete() returned 0.
1201 * This means that __skb_checksum_complete() might have
1202 * set skb->csum_valid to 1.
1203 * On 64bit platforms, we can set csum_unnecessary
1204 * to true, but only if the skb is not shared.
1205 */
1206 #if BITS_PER_LONG == 64
1207 if (!skb_shared(skb))
1208 udp_skb_scratch(skb)->csum_unnecessary = true;
1209 #endif
1210 }
1211
udp_skb_truesize(struct sk_buff * skb)1212 static int udp_skb_truesize(struct sk_buff *skb)
1213 {
1214 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1215 }
1216
udp_skb_has_head_state(struct sk_buff * skb)1217 static bool udp_skb_has_head_state(struct sk_buff *skb)
1218 {
1219 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1220 }
1221
1222 /* fully reclaim rmem/fwd memory allocated for skb */
udp_rmem_release(struct sock * sk,int size,int partial,bool rx_queue_lock_held)1223 static void udp_rmem_release(struct sock *sk, int size, int partial,
1224 bool rx_queue_lock_held)
1225 {
1226 struct udp_sock *up = udp_sk(sk);
1227 struct sk_buff_head *sk_queue;
1228 int amt;
1229
1230 if (likely(partial)) {
1231 up->forward_deficit += size;
1232 size = up->forward_deficit;
1233 if (size < (sk->sk_rcvbuf >> 2) &&
1234 !skb_queue_empty(&up->reader_queue))
1235 return;
1236 } else {
1237 size += up->forward_deficit;
1238 }
1239 up->forward_deficit = 0;
1240
1241 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1242 * if the called don't held it already
1243 */
1244 sk_queue = &sk->sk_receive_queue;
1245 if (!rx_queue_lock_held)
1246 spin_lock(&sk_queue->lock);
1247
1248
1249 sk->sk_forward_alloc += size;
1250 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1251 sk->sk_forward_alloc -= amt;
1252
1253 if (amt)
1254 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1255
1256 atomic_sub(size, &sk->sk_rmem_alloc);
1257
1258 /* this can save us from acquiring the rx queue lock on next receive */
1259 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1260
1261 if (!rx_queue_lock_held)
1262 spin_unlock(&sk_queue->lock);
1263 }
1264
1265 /* Note: called with reader_queue.lock held.
1266 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1267 * This avoids a cache line miss while receive_queue lock is held.
1268 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1269 */
udp_skb_destructor(struct sock * sk,struct sk_buff * skb)1270 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1271 {
1272 prefetch(&skb->data);
1273 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1274 }
1275 EXPORT_SYMBOL(udp_skb_destructor);
1276
1277 /* as above, but the caller held the rx queue lock, too */
udp_skb_dtor_locked(struct sock * sk,struct sk_buff * skb)1278 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1279 {
1280 prefetch(&skb->data);
1281 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1282 }
1283
1284 /* Idea of busylocks is to let producers grab an extra spinlock
1285 * to relieve pressure on the receive_queue spinlock shared by consumer.
1286 * Under flood, this means that only one producer can be in line
1287 * trying to acquire the receive_queue spinlock.
1288 * These busylock can be allocated on a per cpu manner, instead of a
1289 * per socket one (that would consume a cache line per socket)
1290 */
1291 static int udp_busylocks_log __read_mostly;
1292 static spinlock_t *udp_busylocks __read_mostly;
1293
busylock_acquire(void * ptr)1294 static spinlock_t *busylock_acquire(void *ptr)
1295 {
1296 spinlock_t *busy;
1297
1298 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1299 spin_lock(busy);
1300 return busy;
1301 }
1302
busylock_release(spinlock_t * busy)1303 static void busylock_release(spinlock_t *busy)
1304 {
1305 if (busy)
1306 spin_unlock(busy);
1307 }
1308
__udp_enqueue_schedule_skb(struct sock * sk,struct sk_buff * skb)1309 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1310 {
1311 struct sk_buff_head *list = &sk->sk_receive_queue;
1312 int rmem, delta, amt, err = -ENOMEM;
1313 spinlock_t *busy = NULL;
1314 int size;
1315
1316 /* try to avoid the costly atomic add/sub pair when the receive
1317 * queue is full; always allow at least a packet
1318 */
1319 rmem = atomic_read(&sk->sk_rmem_alloc);
1320 if (rmem > sk->sk_rcvbuf)
1321 goto drop;
1322
1323 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1324 * having linear skbs :
1325 * - Reduce memory overhead and thus increase receive queue capacity
1326 * - Less cache line misses at copyout() time
1327 * - Less work at consume_skb() (less alien page frag freeing)
1328 */
1329 if (rmem > (sk->sk_rcvbuf >> 1)) {
1330 skb_condense(skb);
1331
1332 busy = busylock_acquire(sk);
1333 }
1334 size = skb->truesize;
1335 udp_set_dev_scratch(skb);
1336
1337 /* we drop only if the receive buf is full and the receive
1338 * queue contains some other skb
1339 */
1340 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1341 if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1342 goto uncharge_drop;
1343
1344 spin_lock(&list->lock);
1345 if (size >= sk->sk_forward_alloc) {
1346 amt = sk_mem_pages(size);
1347 delta = amt << SK_MEM_QUANTUM_SHIFT;
1348 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1349 err = -ENOBUFS;
1350 spin_unlock(&list->lock);
1351 goto uncharge_drop;
1352 }
1353
1354 sk->sk_forward_alloc += delta;
1355 }
1356
1357 sk->sk_forward_alloc -= size;
1358
1359 /* no need to setup a destructor, we will explicitly release the
1360 * forward allocated memory on dequeue
1361 */
1362 sock_skb_set_dropcount(sk, skb);
1363
1364 __skb_queue_tail(list, skb);
1365 spin_unlock(&list->lock);
1366
1367 if (!sock_flag(sk, SOCK_DEAD))
1368 sk->sk_data_ready(sk);
1369
1370 busylock_release(busy);
1371 return 0;
1372
1373 uncharge_drop:
1374 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1375
1376 drop:
1377 atomic_inc(&sk->sk_drops);
1378 busylock_release(busy);
1379 return err;
1380 }
1381 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1382
udp_destruct_sock(struct sock * sk)1383 void udp_destruct_sock(struct sock *sk)
1384 {
1385 /* reclaim completely the forward allocated memory */
1386 struct udp_sock *up = udp_sk(sk);
1387 unsigned int total = 0;
1388 struct sk_buff *skb;
1389
1390 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1391 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1392 total += skb->truesize;
1393 kfree_skb(skb);
1394 }
1395 udp_rmem_release(sk, total, 0, true);
1396
1397 inet_sock_destruct(sk);
1398 }
1399 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1400
udp_init_sock(struct sock * sk)1401 int udp_init_sock(struct sock *sk)
1402 {
1403 skb_queue_head_init(&udp_sk(sk)->reader_queue);
1404 sk->sk_destruct = udp_destruct_sock;
1405 return 0;
1406 }
1407 EXPORT_SYMBOL_GPL(udp_init_sock);
1408
skb_consume_udp(struct sock * sk,struct sk_buff * skb,int len)1409 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1410 {
1411 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1412 bool slow = lock_sock_fast(sk);
1413
1414 sk_peek_offset_bwd(sk, len);
1415 unlock_sock_fast(sk, slow);
1416 }
1417
1418 if (!skb_unref(skb))
1419 return;
1420
1421 /* In the more common cases we cleared the head states previously,
1422 * see __udp_queue_rcv_skb().
1423 */
1424 if (unlikely(udp_skb_has_head_state(skb)))
1425 skb_release_head_state(skb);
1426 __consume_stateless_skb(skb);
1427 }
1428 EXPORT_SYMBOL_GPL(skb_consume_udp);
1429
__first_packet_length(struct sock * sk,struct sk_buff_head * rcvq,int * total)1430 static struct sk_buff *__first_packet_length(struct sock *sk,
1431 struct sk_buff_head *rcvq,
1432 int *total)
1433 {
1434 struct sk_buff *skb;
1435
1436 while ((skb = skb_peek(rcvq)) != NULL) {
1437 if (udp_lib_checksum_complete(skb)) {
1438 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1439 IS_UDPLITE(sk));
1440 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1441 IS_UDPLITE(sk));
1442 atomic_inc(&sk->sk_drops);
1443 __skb_unlink(skb, rcvq);
1444 *total += skb->truesize;
1445 kfree_skb(skb);
1446 } else {
1447 udp_skb_csum_unnecessary_set(skb);
1448 break;
1449 }
1450 }
1451 return skb;
1452 }
1453
1454 /**
1455 * first_packet_length - return length of first packet in receive queue
1456 * @sk: socket
1457 *
1458 * Drops all bad checksum frames, until a valid one is found.
1459 * Returns the length of found skb, or -1 if none is found.
1460 */
first_packet_length(struct sock * sk)1461 static int first_packet_length(struct sock *sk)
1462 {
1463 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1464 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1465 struct sk_buff *skb;
1466 int total = 0;
1467 int res;
1468
1469 spin_lock_bh(&rcvq->lock);
1470 skb = __first_packet_length(sk, rcvq, &total);
1471 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1472 spin_lock(&sk_queue->lock);
1473 skb_queue_splice_tail_init(sk_queue, rcvq);
1474 spin_unlock(&sk_queue->lock);
1475
1476 skb = __first_packet_length(sk, rcvq, &total);
1477 }
1478 res = skb ? skb->len : -1;
1479 if (total)
1480 udp_rmem_release(sk, total, 1, false);
1481 spin_unlock_bh(&rcvq->lock);
1482 return res;
1483 }
1484
1485 /*
1486 * IOCTL requests applicable to the UDP protocol
1487 */
1488
udp_ioctl(struct sock * sk,int cmd,unsigned long arg)1489 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1490 {
1491 switch (cmd) {
1492 case SIOCOUTQ:
1493 {
1494 int amount = sk_wmem_alloc_get(sk);
1495
1496 return put_user(amount, (int __user *)arg);
1497 }
1498
1499 case SIOCINQ:
1500 {
1501 int amount = max_t(int, 0, first_packet_length(sk));
1502
1503 return put_user(amount, (int __user *)arg);
1504 }
1505
1506 default:
1507 return -ENOIOCTLCMD;
1508 }
1509
1510 return 0;
1511 }
1512 EXPORT_SYMBOL(udp_ioctl);
1513
__skb_recv_udp(struct sock * sk,unsigned int flags,int noblock,int * peeked,int * off,int * err)1514 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1515 int noblock, int *peeked, int *off, int *err)
1516 {
1517 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1518 struct sk_buff_head *queue;
1519 struct sk_buff *last;
1520 long timeo;
1521 int error;
1522
1523 queue = &udp_sk(sk)->reader_queue;
1524 flags |= noblock ? MSG_DONTWAIT : 0;
1525 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1526 do {
1527 struct sk_buff *skb;
1528
1529 error = sock_error(sk);
1530 if (error)
1531 break;
1532
1533 error = -EAGAIN;
1534 *peeked = 0;
1535 do {
1536 spin_lock_bh(&queue->lock);
1537 skb = __skb_try_recv_from_queue(sk, queue, flags,
1538 udp_skb_destructor,
1539 peeked, off, err,
1540 &last);
1541 if (skb) {
1542 spin_unlock_bh(&queue->lock);
1543 return skb;
1544 }
1545
1546 if (skb_queue_empty_lockless(sk_queue)) {
1547 spin_unlock_bh(&queue->lock);
1548 goto busy_check;
1549 }
1550
1551 /* refill the reader queue and walk it again
1552 * keep both queues locked to avoid re-acquiring
1553 * the sk_receive_queue lock if fwd memory scheduling
1554 * is needed.
1555 */
1556 spin_lock(&sk_queue->lock);
1557 skb_queue_splice_tail_init(sk_queue, queue);
1558
1559 skb = __skb_try_recv_from_queue(sk, queue, flags,
1560 udp_skb_dtor_locked,
1561 peeked, off, err,
1562 &last);
1563 spin_unlock(&sk_queue->lock);
1564 spin_unlock_bh(&queue->lock);
1565 if (skb)
1566 return skb;
1567
1568 busy_check:
1569 if (!sk_can_busy_loop(sk))
1570 break;
1571
1572 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1573 } while (!skb_queue_empty_lockless(sk_queue));
1574
1575 /* sk_queue is empty, reader_queue may contain peeked packets */
1576 } while (timeo &&
1577 !__skb_wait_for_more_packets(sk, &error, &timeo,
1578 (struct sk_buff *)sk_queue));
1579
1580 *err = error;
1581 return NULL;
1582 }
1583 EXPORT_SYMBOL(__skb_recv_udp);
1584
1585 /*
1586 * This should be easy, if there is something there we
1587 * return it, otherwise we block.
1588 */
1589
udp_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int noblock,int flags,int * addr_len)1590 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1591 int flags, int *addr_len)
1592 {
1593 struct inet_sock *inet = inet_sk(sk);
1594 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1595 struct sk_buff *skb;
1596 unsigned int ulen, copied;
1597 int peeked, peeking, off;
1598 int err;
1599 int is_udplite = IS_UDPLITE(sk);
1600 bool checksum_valid = false;
1601
1602 if (flags & MSG_ERRQUEUE)
1603 return ip_recv_error(sk, msg, len, addr_len);
1604
1605 try_again:
1606 peeking = flags & MSG_PEEK;
1607 off = sk_peek_offset(sk, flags);
1608 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1609 if (!skb)
1610 return err;
1611
1612 ulen = udp_skb_len(skb);
1613 copied = len;
1614 if (copied > ulen - off)
1615 copied = ulen - off;
1616 else if (copied < ulen)
1617 msg->msg_flags |= MSG_TRUNC;
1618
1619 /*
1620 * If checksum is needed at all, try to do it while copying the
1621 * data. If the data is truncated, or if we only want a partial
1622 * coverage checksum (UDP-Lite), do it before the copy.
1623 */
1624
1625 if (copied < ulen || peeking ||
1626 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1627 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1628 !__udp_lib_checksum_complete(skb);
1629 if (!checksum_valid)
1630 goto csum_copy_err;
1631 }
1632
1633 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1634 if (udp_skb_is_linear(skb))
1635 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1636 else
1637 err = skb_copy_datagram_msg(skb, off, msg, copied);
1638 } else {
1639 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1640
1641 if (err == -EINVAL)
1642 goto csum_copy_err;
1643 }
1644
1645 if (unlikely(err)) {
1646 if (!peeked) {
1647 atomic_inc(&sk->sk_drops);
1648 UDP_INC_STATS(sock_net(sk),
1649 UDP_MIB_INERRORS, is_udplite);
1650 }
1651 kfree_skb(skb);
1652 return err;
1653 }
1654
1655 if (!peeked)
1656 UDP_INC_STATS(sock_net(sk),
1657 UDP_MIB_INDATAGRAMS, is_udplite);
1658
1659 sock_recv_ts_and_drops(msg, sk, skb);
1660
1661 /* Copy the address. */
1662 if (sin) {
1663 sin->sin_family = AF_INET;
1664 sin->sin_port = udp_hdr(skb)->source;
1665 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1666 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1667 *addr_len = sizeof(*sin);
1668 }
1669 if (inet->cmsg_flags)
1670 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1671
1672 err = copied;
1673 if (flags & MSG_TRUNC)
1674 err = ulen;
1675
1676 skb_consume_udp(sk, skb, peeking ? -err : err);
1677 return err;
1678
1679 csum_copy_err:
1680 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1681 udp_skb_destructor)) {
1682 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1683 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1684 }
1685 kfree_skb(skb);
1686
1687 /* starting over for a new packet, but check if we need to yield */
1688 cond_resched();
1689 msg->msg_flags &= ~MSG_TRUNC;
1690 goto try_again;
1691 }
1692
__udp_disconnect(struct sock * sk,int flags)1693 int __udp_disconnect(struct sock *sk, int flags)
1694 {
1695 struct inet_sock *inet = inet_sk(sk);
1696 /*
1697 * 1003.1g - break association.
1698 */
1699
1700 sk->sk_state = TCP_CLOSE;
1701 inet->inet_daddr = 0;
1702 inet->inet_dport = 0;
1703 sock_rps_reset_rxhash(sk);
1704 sk->sk_bound_dev_if = 0;
1705 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1706 inet_reset_saddr(sk);
1707
1708 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1709 sk->sk_prot->unhash(sk);
1710 inet->inet_sport = 0;
1711 }
1712 sk_dst_reset(sk);
1713 return 0;
1714 }
1715 EXPORT_SYMBOL(__udp_disconnect);
1716
udp_disconnect(struct sock * sk,int flags)1717 int udp_disconnect(struct sock *sk, int flags)
1718 {
1719 lock_sock(sk);
1720 __udp_disconnect(sk, flags);
1721 release_sock(sk);
1722 return 0;
1723 }
1724 EXPORT_SYMBOL(udp_disconnect);
1725
udp_lib_unhash(struct sock * sk)1726 void udp_lib_unhash(struct sock *sk)
1727 {
1728 if (sk_hashed(sk)) {
1729 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1730 struct udp_hslot *hslot, *hslot2;
1731
1732 hslot = udp_hashslot(udptable, sock_net(sk),
1733 udp_sk(sk)->udp_port_hash);
1734 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1735
1736 spin_lock_bh(&hslot->lock);
1737 if (rcu_access_pointer(sk->sk_reuseport_cb))
1738 reuseport_detach_sock(sk);
1739 if (sk_del_node_init_rcu(sk)) {
1740 hslot->count--;
1741 inet_sk(sk)->inet_num = 0;
1742 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1743
1744 spin_lock(&hslot2->lock);
1745 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1746 hslot2->count--;
1747 spin_unlock(&hslot2->lock);
1748 }
1749 spin_unlock_bh(&hslot->lock);
1750 }
1751 }
1752 EXPORT_SYMBOL(udp_lib_unhash);
1753
1754 /*
1755 * inet_rcv_saddr was changed, we must rehash secondary hash
1756 */
udp_lib_rehash(struct sock * sk,u16 newhash)1757 void udp_lib_rehash(struct sock *sk, u16 newhash)
1758 {
1759 if (sk_hashed(sk)) {
1760 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1761 struct udp_hslot *hslot, *hslot2, *nhslot2;
1762
1763 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1764 nhslot2 = udp_hashslot2(udptable, newhash);
1765 udp_sk(sk)->udp_portaddr_hash = newhash;
1766
1767 if (hslot2 != nhslot2 ||
1768 rcu_access_pointer(sk->sk_reuseport_cb)) {
1769 hslot = udp_hashslot(udptable, sock_net(sk),
1770 udp_sk(sk)->udp_port_hash);
1771 /* we must lock primary chain too */
1772 spin_lock_bh(&hslot->lock);
1773 if (rcu_access_pointer(sk->sk_reuseport_cb))
1774 reuseport_detach_sock(sk);
1775
1776 if (hslot2 != nhslot2) {
1777 spin_lock(&hslot2->lock);
1778 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1779 hslot2->count--;
1780 spin_unlock(&hslot2->lock);
1781
1782 spin_lock(&nhslot2->lock);
1783 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1784 &nhslot2->head);
1785 nhslot2->count++;
1786 spin_unlock(&nhslot2->lock);
1787 }
1788
1789 spin_unlock_bh(&hslot->lock);
1790 }
1791 }
1792 }
1793 EXPORT_SYMBOL(udp_lib_rehash);
1794
udp_v4_rehash(struct sock * sk)1795 static void udp_v4_rehash(struct sock *sk)
1796 {
1797 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1798 inet_sk(sk)->inet_rcv_saddr,
1799 inet_sk(sk)->inet_num);
1800 udp_lib_rehash(sk, new_hash);
1801 }
1802
__udp_queue_rcv_skb(struct sock * sk,struct sk_buff * skb)1803 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1804 {
1805 int rc;
1806
1807 if (inet_sk(sk)->inet_daddr) {
1808 sock_rps_save_rxhash(sk, skb);
1809 sk_mark_napi_id(sk, skb);
1810 sk_incoming_cpu_update(sk);
1811 } else {
1812 sk_mark_napi_id_once(sk, skb);
1813 }
1814
1815 rc = __udp_enqueue_schedule_skb(sk, skb);
1816 if (rc < 0) {
1817 int is_udplite = IS_UDPLITE(sk);
1818
1819 /* Note that an ENOMEM error is charged twice */
1820 if (rc == -ENOMEM)
1821 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1822 is_udplite);
1823 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1824 kfree_skb(skb);
1825 trace_udp_fail_queue_rcv_skb(rc, sk);
1826 return -1;
1827 }
1828
1829 return 0;
1830 }
1831
1832 static struct static_key udp_encap_needed __read_mostly;
udp_encap_enable(void)1833 void udp_encap_enable(void)
1834 {
1835 static_key_enable(&udp_encap_needed);
1836 }
1837 EXPORT_SYMBOL(udp_encap_enable);
1838
1839 /* returns:
1840 * -1: error
1841 * 0: success
1842 * >0: "udp encap" protocol resubmission
1843 *
1844 * Note that in the success and error cases, the skb is assumed to
1845 * have either been requeued or freed.
1846 */
udp_queue_rcv_skb(struct sock * sk,struct sk_buff * skb)1847 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1848 {
1849 struct udp_sock *up = udp_sk(sk);
1850 int is_udplite = IS_UDPLITE(sk);
1851
1852 /*
1853 * Charge it to the socket, dropping if the queue is full.
1854 */
1855 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1856 goto drop;
1857 nf_reset(skb);
1858
1859 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1860 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1861
1862 /*
1863 * This is an encapsulation socket so pass the skb to
1864 * the socket's udp_encap_rcv() hook. Otherwise, just
1865 * fall through and pass this up the UDP socket.
1866 * up->encap_rcv() returns the following value:
1867 * =0 if skb was successfully passed to the encap
1868 * handler or was discarded by it.
1869 * >0 if skb should be passed on to UDP.
1870 * <0 if skb should be resubmitted as proto -N
1871 */
1872
1873 /* if we're overly short, let UDP handle it */
1874 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1875 if (encap_rcv) {
1876 int ret;
1877
1878 /* Verify checksum before giving to encap */
1879 if (udp_lib_checksum_complete(skb))
1880 goto csum_error;
1881
1882 ret = encap_rcv(sk, skb);
1883 if (ret <= 0) {
1884 __UDP_INC_STATS(sock_net(sk),
1885 UDP_MIB_INDATAGRAMS,
1886 is_udplite);
1887 return -ret;
1888 }
1889 }
1890
1891 /* FALLTHROUGH -- it's a UDP Packet */
1892 }
1893
1894 /*
1895 * UDP-Lite specific tests, ignored on UDP sockets
1896 */
1897 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1898
1899 /*
1900 * MIB statistics other than incrementing the error count are
1901 * disabled for the following two types of errors: these depend
1902 * on the application settings, not on the functioning of the
1903 * protocol stack as such.
1904 *
1905 * RFC 3828 here recommends (sec 3.3): "There should also be a
1906 * way ... to ... at least let the receiving application block
1907 * delivery of packets with coverage values less than a value
1908 * provided by the application."
1909 */
1910 if (up->pcrlen == 0) { /* full coverage was set */
1911 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1912 UDP_SKB_CB(skb)->cscov, skb->len);
1913 goto drop;
1914 }
1915 /* The next case involves violating the min. coverage requested
1916 * by the receiver. This is subtle: if receiver wants x and x is
1917 * greater than the buffersize/MTU then receiver will complain
1918 * that it wants x while sender emits packets of smaller size y.
1919 * Therefore the above ...()->partial_cov statement is essential.
1920 */
1921 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1922 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1923 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1924 goto drop;
1925 }
1926 }
1927
1928 prefetch(&sk->sk_rmem_alloc);
1929 if (rcu_access_pointer(sk->sk_filter) &&
1930 udp_lib_checksum_complete(skb))
1931 goto csum_error;
1932
1933 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1934 goto drop;
1935
1936 udp_csum_pull_header(skb);
1937
1938 ipv4_pktinfo_prepare(sk, skb);
1939 return __udp_queue_rcv_skb(sk, skb);
1940
1941 csum_error:
1942 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1943 drop:
1944 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1945 atomic_inc(&sk->sk_drops);
1946 kfree_skb(skb);
1947 return -1;
1948 }
1949
1950 /* For TCP sockets, sk_rx_dst is protected by socket lock
1951 * For UDP, we use xchg() to guard against concurrent changes.
1952 */
udp_sk_rx_dst_set(struct sock * sk,struct dst_entry * dst)1953 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1954 {
1955 struct dst_entry *old;
1956
1957 if (dst_hold_safe(dst)) {
1958 old = xchg(&sk->sk_rx_dst, dst);
1959 dst_release(old);
1960 return old != dst;
1961 }
1962 return false;
1963 }
1964 EXPORT_SYMBOL(udp_sk_rx_dst_set);
1965
1966 /*
1967 * Multicasts and broadcasts go to each listener.
1968 *
1969 * Note: called only from the BH handler context.
1970 */
__udp4_lib_mcast_deliver(struct net * net,struct sk_buff * skb,struct udphdr * uh,__be32 saddr,__be32 daddr,struct udp_table * udptable,int proto)1971 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1972 struct udphdr *uh,
1973 __be32 saddr, __be32 daddr,
1974 struct udp_table *udptable,
1975 int proto)
1976 {
1977 struct sock *sk, *first = NULL;
1978 unsigned short hnum = ntohs(uh->dest);
1979 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1980 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1981 unsigned int offset = offsetof(typeof(*sk), sk_node);
1982 int dif = skb->dev->ifindex;
1983 int sdif = inet_sdif(skb);
1984 struct hlist_node *node;
1985 struct sk_buff *nskb;
1986
1987 if (use_hash2) {
1988 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1989 udptable->mask;
1990 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1991 start_lookup:
1992 hslot = &udptable->hash2[hash2];
1993 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1994 }
1995
1996 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1997 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1998 uh->source, saddr, dif, sdif, hnum))
1999 continue;
2000
2001 if (!first) {
2002 first = sk;
2003 continue;
2004 }
2005 nskb = skb_clone(skb, GFP_ATOMIC);
2006
2007 if (unlikely(!nskb)) {
2008 atomic_inc(&sk->sk_drops);
2009 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2010 IS_UDPLITE(sk));
2011 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2012 IS_UDPLITE(sk));
2013 continue;
2014 }
2015 if (udp_queue_rcv_skb(sk, nskb) > 0)
2016 consume_skb(nskb);
2017 }
2018
2019 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2020 if (use_hash2 && hash2 != hash2_any) {
2021 hash2 = hash2_any;
2022 goto start_lookup;
2023 }
2024
2025 if (first) {
2026 if (udp_queue_rcv_skb(first, skb) > 0)
2027 consume_skb(skb);
2028 } else {
2029 kfree_skb(skb);
2030 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2031 proto == IPPROTO_UDPLITE);
2032 }
2033 return 0;
2034 }
2035
2036 /* Initialize UDP checksum. If exited with zero value (success),
2037 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2038 * Otherwise, csum completion requires chacksumming packet body,
2039 * including udp header and folding it to skb->csum.
2040 */
udp4_csum_init(struct sk_buff * skb,struct udphdr * uh,int proto)2041 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2042 int proto)
2043 {
2044 int err;
2045
2046 UDP_SKB_CB(skb)->partial_cov = 0;
2047 UDP_SKB_CB(skb)->cscov = skb->len;
2048
2049 if (proto == IPPROTO_UDPLITE) {
2050 err = udplite_checksum_init(skb, uh);
2051 if (err)
2052 return err;
2053
2054 if (UDP_SKB_CB(skb)->partial_cov) {
2055 skb->csum = inet_compute_pseudo(skb, proto);
2056 return 0;
2057 }
2058 }
2059
2060 /* Note, we are only interested in != 0 or == 0, thus the
2061 * force to int.
2062 */
2063 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2064 inet_compute_pseudo);
2065 if (err)
2066 return err;
2067
2068 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2069 /* If SW calculated the value, we know it's bad */
2070 if (skb->csum_complete_sw)
2071 return 1;
2072
2073 /* HW says the value is bad. Let's validate that.
2074 * skb->csum is no longer the full packet checksum,
2075 * so don't treat it as such.
2076 */
2077 skb_checksum_complete_unset(skb);
2078 }
2079
2080 return 0;
2081 }
2082
2083 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2084 * return code conversion for ip layer consumption
2085 */
udp_unicast_rcv_skb(struct sock * sk,struct sk_buff * skb,struct udphdr * uh)2086 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2087 struct udphdr *uh)
2088 {
2089 int ret;
2090
2091 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2092 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2093 inet_compute_pseudo);
2094
2095 ret = udp_queue_rcv_skb(sk, skb);
2096
2097 /* a return value > 0 means to resubmit the input, but
2098 * it wants the return to be -protocol, or 0
2099 */
2100 if (ret > 0)
2101 return -ret;
2102 return 0;
2103 }
2104
2105 /*
2106 * All we need to do is get the socket, and then do a checksum.
2107 */
2108
__udp4_lib_rcv(struct sk_buff * skb,struct udp_table * udptable,int proto)2109 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2110 int proto)
2111 {
2112 struct sock *sk;
2113 struct udphdr *uh;
2114 unsigned short ulen;
2115 struct rtable *rt = skb_rtable(skb);
2116 __be32 saddr, daddr;
2117 struct net *net = dev_net(skb->dev);
2118
2119 /*
2120 * Validate the packet.
2121 */
2122 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2123 goto drop; /* No space for header. */
2124
2125 uh = udp_hdr(skb);
2126 ulen = ntohs(uh->len);
2127 saddr = ip_hdr(skb)->saddr;
2128 daddr = ip_hdr(skb)->daddr;
2129
2130 if (ulen > skb->len)
2131 goto short_packet;
2132
2133 if (proto == IPPROTO_UDP) {
2134 /* UDP validates ulen. */
2135 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2136 goto short_packet;
2137 uh = udp_hdr(skb);
2138 }
2139
2140 if (udp4_csum_init(skb, uh, proto))
2141 goto csum_error;
2142
2143 sk = skb_steal_sock(skb);
2144 if (sk) {
2145 struct dst_entry *dst = skb_dst(skb);
2146 int ret;
2147
2148 if (unlikely(sk->sk_rx_dst != dst))
2149 udp_sk_rx_dst_set(sk, dst);
2150
2151 ret = udp_unicast_rcv_skb(sk, skb, uh);
2152 sock_put(sk);
2153 return ret;
2154 }
2155
2156 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2157 return __udp4_lib_mcast_deliver(net, skb, uh,
2158 saddr, daddr, udptable, proto);
2159
2160 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2161 if (sk)
2162 return udp_unicast_rcv_skb(sk, skb, uh);
2163
2164 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2165 goto drop;
2166 nf_reset(skb);
2167
2168 /* No socket. Drop packet silently, if checksum is wrong */
2169 if (udp_lib_checksum_complete(skb))
2170 goto csum_error;
2171
2172 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2173 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2174
2175 /*
2176 * Hmm. We got an UDP packet to a port to which we
2177 * don't wanna listen. Ignore it.
2178 */
2179 kfree_skb(skb);
2180 return 0;
2181
2182 short_packet:
2183 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2184 proto == IPPROTO_UDPLITE ? "Lite" : "",
2185 &saddr, ntohs(uh->source),
2186 ulen, skb->len,
2187 &daddr, ntohs(uh->dest));
2188 goto drop;
2189
2190 csum_error:
2191 /*
2192 * RFC1122: OK. Discards the bad packet silently (as far as
2193 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2194 */
2195 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2196 proto == IPPROTO_UDPLITE ? "Lite" : "",
2197 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2198 ulen);
2199 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2200 drop:
2201 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2202 kfree_skb(skb);
2203 return 0;
2204 }
2205
2206 /* We can only early demux multicast if there is a single matching socket.
2207 * If more than one socket found returns NULL
2208 */
__udp4_lib_mcast_demux_lookup(struct net * net,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif)2209 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2210 __be16 loc_port, __be32 loc_addr,
2211 __be16 rmt_port, __be32 rmt_addr,
2212 int dif, int sdif)
2213 {
2214 struct sock *sk, *result;
2215 unsigned short hnum = ntohs(loc_port);
2216 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2217 struct udp_hslot *hslot = &udp_table.hash[slot];
2218
2219 /* Do not bother scanning a too big list */
2220 if (hslot->count > 10)
2221 return NULL;
2222
2223 result = NULL;
2224 sk_for_each_rcu(sk, &hslot->head) {
2225 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2226 rmt_port, rmt_addr, dif, sdif, hnum)) {
2227 if (result)
2228 return NULL;
2229 result = sk;
2230 }
2231 }
2232
2233 return result;
2234 }
2235
2236 /* For unicast we should only early demux connected sockets or we can
2237 * break forwarding setups. The chains here can be long so only check
2238 * if the first socket is an exact match and if not move on.
2239 */
__udp4_lib_demux_lookup(struct net * net,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif)2240 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2241 __be16 loc_port, __be32 loc_addr,
2242 __be16 rmt_port, __be32 rmt_addr,
2243 int dif, int sdif)
2244 {
2245 unsigned short hnum = ntohs(loc_port);
2246 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2247 unsigned int slot2 = hash2 & udp_table.mask;
2248 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2249 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2250 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2251 struct sock *sk;
2252
2253 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2254 if (INET_MATCH(sk, net, acookie, rmt_addr,
2255 loc_addr, ports, dif, sdif))
2256 return sk;
2257 /* Only check first socket in chain */
2258 break;
2259 }
2260 return NULL;
2261 }
2262
udp_v4_early_demux(struct sk_buff * skb)2263 int udp_v4_early_demux(struct sk_buff *skb)
2264 {
2265 struct net *net = dev_net(skb->dev);
2266 struct in_device *in_dev = NULL;
2267 const struct iphdr *iph;
2268 const struct udphdr *uh;
2269 struct sock *sk = NULL;
2270 struct dst_entry *dst;
2271 int dif = skb->dev->ifindex;
2272 int sdif = inet_sdif(skb);
2273 int ours;
2274
2275 /* validate the packet */
2276 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2277 return 0;
2278
2279 iph = ip_hdr(skb);
2280 uh = udp_hdr(skb);
2281
2282 if (skb->pkt_type == PACKET_MULTICAST) {
2283 in_dev = __in_dev_get_rcu(skb->dev);
2284
2285 if (!in_dev)
2286 return 0;
2287
2288 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2289 iph->protocol);
2290 if (!ours)
2291 return 0;
2292
2293 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2294 uh->source, iph->saddr,
2295 dif, sdif);
2296 } else if (skb->pkt_type == PACKET_HOST) {
2297 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2298 uh->source, iph->saddr, dif, sdif);
2299 }
2300
2301 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2302 return 0;
2303
2304 skb->sk = sk;
2305 skb->destructor = sock_efree;
2306 dst = READ_ONCE(sk->sk_rx_dst);
2307
2308 if (dst)
2309 dst = dst_check(dst, 0);
2310 if (dst) {
2311 u32 itag = 0;
2312
2313 /* set noref for now.
2314 * any place which wants to hold dst has to call
2315 * dst_hold_safe()
2316 */
2317 skb_dst_set_noref(skb, dst);
2318
2319 /* for unconnected multicast sockets we need to validate
2320 * the source on each packet
2321 */
2322 if (!inet_sk(sk)->inet_daddr && in_dev)
2323 return ip_mc_validate_source(skb, iph->daddr,
2324 iph->saddr, iph->tos,
2325 skb->dev, in_dev, &itag);
2326 }
2327 return 0;
2328 }
2329
udp_rcv(struct sk_buff * skb)2330 int udp_rcv(struct sk_buff *skb)
2331 {
2332 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2333 }
2334
udp_destroy_sock(struct sock * sk)2335 void udp_destroy_sock(struct sock *sk)
2336 {
2337 struct udp_sock *up = udp_sk(sk);
2338 bool slow = lock_sock_fast(sk);
2339 udp_flush_pending_frames(sk);
2340 unlock_sock_fast(sk, slow);
2341 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2342 void (*encap_destroy)(struct sock *sk);
2343 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2344 if (encap_destroy)
2345 encap_destroy(sk);
2346 }
2347 }
2348
2349 /*
2350 * Socket option code for UDP
2351 */
udp_lib_setsockopt(struct sock * sk,int level,int optname,char __user * optval,unsigned int optlen,int (* push_pending_frames)(struct sock *))2352 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2353 char __user *optval, unsigned int optlen,
2354 int (*push_pending_frames)(struct sock *))
2355 {
2356 struct udp_sock *up = udp_sk(sk);
2357 int val, valbool;
2358 int err = 0;
2359 int is_udplite = IS_UDPLITE(sk);
2360
2361 if (optlen < sizeof(int))
2362 return -EINVAL;
2363
2364 if (get_user(val, (int __user *)optval))
2365 return -EFAULT;
2366
2367 valbool = val ? 1 : 0;
2368
2369 switch (optname) {
2370 case UDP_CORK:
2371 if (val != 0) {
2372 up->corkflag = 1;
2373 } else {
2374 up->corkflag = 0;
2375 lock_sock(sk);
2376 push_pending_frames(sk);
2377 release_sock(sk);
2378 }
2379 break;
2380
2381 case UDP_ENCAP:
2382 switch (val) {
2383 case 0:
2384 case UDP_ENCAP_ESPINUDP:
2385 case UDP_ENCAP_ESPINUDP_NON_IKE:
2386 up->encap_rcv = xfrm4_udp_encap_rcv;
2387 /* FALLTHROUGH */
2388 case UDP_ENCAP_L2TPINUDP:
2389 up->encap_type = val;
2390 udp_encap_enable();
2391 break;
2392 default:
2393 err = -ENOPROTOOPT;
2394 break;
2395 }
2396 break;
2397
2398 case UDP_NO_CHECK6_TX:
2399 up->no_check6_tx = valbool;
2400 break;
2401
2402 case UDP_NO_CHECK6_RX:
2403 up->no_check6_rx = valbool;
2404 break;
2405
2406 /*
2407 * UDP-Lite's partial checksum coverage (RFC 3828).
2408 */
2409 /* The sender sets actual checksum coverage length via this option.
2410 * The case coverage > packet length is handled by send module. */
2411 case UDPLITE_SEND_CSCOV:
2412 if (!is_udplite) /* Disable the option on UDP sockets */
2413 return -ENOPROTOOPT;
2414 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2415 val = 8;
2416 else if (val > USHRT_MAX)
2417 val = USHRT_MAX;
2418 up->pcslen = val;
2419 up->pcflag |= UDPLITE_SEND_CC;
2420 break;
2421
2422 /* The receiver specifies a minimum checksum coverage value. To make
2423 * sense, this should be set to at least 8 (as done below). If zero is
2424 * used, this again means full checksum coverage. */
2425 case UDPLITE_RECV_CSCOV:
2426 if (!is_udplite) /* Disable the option on UDP sockets */
2427 return -ENOPROTOOPT;
2428 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2429 val = 8;
2430 else if (val > USHRT_MAX)
2431 val = USHRT_MAX;
2432 up->pcrlen = val;
2433 up->pcflag |= UDPLITE_RECV_CC;
2434 break;
2435
2436 default:
2437 err = -ENOPROTOOPT;
2438 break;
2439 }
2440
2441 return err;
2442 }
2443 EXPORT_SYMBOL(udp_lib_setsockopt);
2444
udp_setsockopt(struct sock * sk,int level,int optname,char __user * optval,unsigned int optlen)2445 int udp_setsockopt(struct sock *sk, int level, int optname,
2446 char __user *optval, unsigned int optlen)
2447 {
2448 if (level == SOL_UDP || level == SOL_UDPLITE)
2449 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2450 udp_push_pending_frames);
2451 return ip_setsockopt(sk, level, optname, optval, optlen);
2452 }
2453
2454 #ifdef CONFIG_COMPAT
compat_udp_setsockopt(struct sock * sk,int level,int optname,char __user * optval,unsigned int optlen)2455 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2456 char __user *optval, unsigned int optlen)
2457 {
2458 if (level == SOL_UDP || level == SOL_UDPLITE)
2459 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2460 udp_push_pending_frames);
2461 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2462 }
2463 #endif
2464
udp_lib_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * optlen)2465 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2466 char __user *optval, int __user *optlen)
2467 {
2468 struct udp_sock *up = udp_sk(sk);
2469 int val, len;
2470
2471 if (get_user(len, optlen))
2472 return -EFAULT;
2473
2474 len = min_t(unsigned int, len, sizeof(int));
2475
2476 if (len < 0)
2477 return -EINVAL;
2478
2479 switch (optname) {
2480 case UDP_CORK:
2481 val = up->corkflag;
2482 break;
2483
2484 case UDP_ENCAP:
2485 val = up->encap_type;
2486 break;
2487
2488 case UDP_NO_CHECK6_TX:
2489 val = up->no_check6_tx;
2490 break;
2491
2492 case UDP_NO_CHECK6_RX:
2493 val = up->no_check6_rx;
2494 break;
2495
2496 /* The following two cannot be changed on UDP sockets, the return is
2497 * always 0 (which corresponds to the full checksum coverage of UDP). */
2498 case UDPLITE_SEND_CSCOV:
2499 val = up->pcslen;
2500 break;
2501
2502 case UDPLITE_RECV_CSCOV:
2503 val = up->pcrlen;
2504 break;
2505
2506 default:
2507 return -ENOPROTOOPT;
2508 }
2509
2510 if (put_user(len, optlen))
2511 return -EFAULT;
2512 if (copy_to_user(optval, &val, len))
2513 return -EFAULT;
2514 return 0;
2515 }
2516 EXPORT_SYMBOL(udp_lib_getsockopt);
2517
udp_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * optlen)2518 int udp_getsockopt(struct sock *sk, int level, int optname,
2519 char __user *optval, int __user *optlen)
2520 {
2521 if (level == SOL_UDP || level == SOL_UDPLITE)
2522 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2523 return ip_getsockopt(sk, level, optname, optval, optlen);
2524 }
2525
2526 #ifdef CONFIG_COMPAT
compat_udp_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * optlen)2527 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2528 char __user *optval, int __user *optlen)
2529 {
2530 if (level == SOL_UDP || level == SOL_UDPLITE)
2531 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2532 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2533 }
2534 #endif
2535 /**
2536 * udp_poll - wait for a UDP event.
2537 * @file - file struct
2538 * @sock - socket
2539 * @wait - poll table
2540 *
2541 * This is same as datagram poll, except for the special case of
2542 * blocking sockets. If application is using a blocking fd
2543 * and a packet with checksum error is in the queue;
2544 * then it could get return from select indicating data available
2545 * but then block when reading it. Add special case code
2546 * to work around these arguably broken applications.
2547 */
udp_poll(struct file * file,struct socket * sock,poll_table * wait)2548 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2549 {
2550 unsigned int mask = datagram_poll(file, sock, wait);
2551 struct sock *sk = sock->sk;
2552
2553 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2554 mask |= POLLIN | POLLRDNORM;
2555
2556 sock_rps_record_flow(sk);
2557
2558 /* Check for false positives due to checksum errors */
2559 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2560 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2561 mask &= ~(POLLIN | POLLRDNORM);
2562
2563 return mask;
2564
2565 }
2566 EXPORT_SYMBOL(udp_poll);
2567
udp_abort(struct sock * sk,int err)2568 int udp_abort(struct sock *sk, int err)
2569 {
2570 lock_sock(sk);
2571
2572 sk->sk_err = err;
2573 sk->sk_error_report(sk);
2574 __udp_disconnect(sk, 0);
2575
2576 release_sock(sk);
2577
2578 return 0;
2579 }
2580 EXPORT_SYMBOL_GPL(udp_abort);
2581
2582 struct proto udp_prot = {
2583 .name = "UDP",
2584 .owner = THIS_MODULE,
2585 .close = udp_lib_close,
2586 .connect = ip4_datagram_connect,
2587 .disconnect = udp_disconnect,
2588 .ioctl = udp_ioctl,
2589 .init = udp_init_sock,
2590 .destroy = udp_destroy_sock,
2591 .setsockopt = udp_setsockopt,
2592 .getsockopt = udp_getsockopt,
2593 .sendmsg = udp_sendmsg,
2594 .recvmsg = udp_recvmsg,
2595 .sendpage = udp_sendpage,
2596 .release_cb = ip4_datagram_release_cb,
2597 .hash = udp_lib_hash,
2598 .unhash = udp_lib_unhash,
2599 .rehash = udp_v4_rehash,
2600 .get_port = udp_v4_get_port,
2601 .memory_allocated = &udp_memory_allocated,
2602 .sysctl_mem = sysctl_udp_mem,
2603 .sysctl_wmem = &sysctl_udp_wmem_min,
2604 .sysctl_rmem = &sysctl_udp_rmem_min,
2605 .obj_size = sizeof(struct udp_sock),
2606 .h.udp_table = &udp_table,
2607 #ifdef CONFIG_COMPAT
2608 .compat_setsockopt = compat_udp_setsockopt,
2609 .compat_getsockopt = compat_udp_getsockopt,
2610 #endif
2611 .diag_destroy = udp_abort,
2612 };
2613 EXPORT_SYMBOL(udp_prot);
2614
2615 /* ------------------------------------------------------------------------ */
2616 #ifdef CONFIG_PROC_FS
2617
udp_get_first(struct seq_file * seq,int start)2618 static struct sock *udp_get_first(struct seq_file *seq, int start)
2619 {
2620 struct sock *sk;
2621 struct udp_iter_state *state = seq->private;
2622 struct net *net = seq_file_net(seq);
2623
2624 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2625 ++state->bucket) {
2626 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2627
2628 if (hlist_empty(&hslot->head))
2629 continue;
2630
2631 spin_lock_bh(&hslot->lock);
2632 sk_for_each(sk, &hslot->head) {
2633 if (!net_eq(sock_net(sk), net))
2634 continue;
2635 if (sk->sk_family == state->family)
2636 goto found;
2637 }
2638 spin_unlock_bh(&hslot->lock);
2639 }
2640 sk = NULL;
2641 found:
2642 return sk;
2643 }
2644
udp_get_next(struct seq_file * seq,struct sock * sk)2645 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2646 {
2647 struct udp_iter_state *state = seq->private;
2648 struct net *net = seq_file_net(seq);
2649
2650 do {
2651 sk = sk_next(sk);
2652 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2653
2654 if (!sk) {
2655 if (state->bucket <= state->udp_table->mask)
2656 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2657 return udp_get_first(seq, state->bucket + 1);
2658 }
2659 return sk;
2660 }
2661
udp_get_idx(struct seq_file * seq,loff_t pos)2662 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2663 {
2664 struct sock *sk = udp_get_first(seq, 0);
2665
2666 if (sk)
2667 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2668 --pos;
2669 return pos ? NULL : sk;
2670 }
2671
udp_seq_start(struct seq_file * seq,loff_t * pos)2672 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2673 {
2674 struct udp_iter_state *state = seq->private;
2675 state->bucket = MAX_UDP_PORTS;
2676
2677 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2678 }
2679
udp_seq_next(struct seq_file * seq,void * v,loff_t * pos)2680 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2681 {
2682 struct sock *sk;
2683
2684 if (v == SEQ_START_TOKEN)
2685 sk = udp_get_idx(seq, 0);
2686 else
2687 sk = udp_get_next(seq, v);
2688
2689 ++*pos;
2690 return sk;
2691 }
2692
udp_seq_stop(struct seq_file * seq,void * v)2693 static void udp_seq_stop(struct seq_file *seq, void *v)
2694 {
2695 struct udp_iter_state *state = seq->private;
2696
2697 if (state->bucket <= state->udp_table->mask)
2698 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2699 }
2700
udp_seq_open(struct inode * inode,struct file * file)2701 int udp_seq_open(struct inode *inode, struct file *file)
2702 {
2703 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2704 struct udp_iter_state *s;
2705 int err;
2706
2707 err = seq_open_net(inode, file, &afinfo->seq_ops,
2708 sizeof(struct udp_iter_state));
2709 if (err < 0)
2710 return err;
2711
2712 s = ((struct seq_file *)file->private_data)->private;
2713 s->family = afinfo->family;
2714 s->udp_table = afinfo->udp_table;
2715 return err;
2716 }
2717 EXPORT_SYMBOL(udp_seq_open);
2718
2719 /* ------------------------------------------------------------------------ */
udp_proc_register(struct net * net,struct udp_seq_afinfo * afinfo)2720 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2721 {
2722 struct proc_dir_entry *p;
2723 int rc = 0;
2724
2725 afinfo->seq_ops.start = udp_seq_start;
2726 afinfo->seq_ops.next = udp_seq_next;
2727 afinfo->seq_ops.stop = udp_seq_stop;
2728
2729 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2730 afinfo->seq_fops, afinfo);
2731 if (!p)
2732 rc = -ENOMEM;
2733 return rc;
2734 }
2735 EXPORT_SYMBOL(udp_proc_register);
2736
udp_proc_unregister(struct net * net,struct udp_seq_afinfo * afinfo)2737 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2738 {
2739 remove_proc_entry(afinfo->name, net->proc_net);
2740 }
2741 EXPORT_SYMBOL(udp_proc_unregister);
2742
2743 /* ------------------------------------------------------------------------ */
udp4_format_sock(struct sock * sp,struct seq_file * f,int bucket)2744 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2745 int bucket)
2746 {
2747 struct inet_sock *inet = inet_sk(sp);
2748 __be32 dest = inet->inet_daddr;
2749 __be32 src = inet->inet_rcv_saddr;
2750 __u16 destp = ntohs(inet->inet_dport);
2751 __u16 srcp = ntohs(inet->inet_sport);
2752
2753 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2754 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2755 bucket, src, srcp, dest, destp, sp->sk_state,
2756 sk_wmem_alloc_get(sp),
2757 udp_rqueue_get(sp),
2758 0, 0L, 0,
2759 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2760 0, sock_i_ino(sp),
2761 refcount_read(&sp->sk_refcnt), sp,
2762 atomic_read(&sp->sk_drops));
2763 }
2764
udp4_seq_show(struct seq_file * seq,void * v)2765 int udp4_seq_show(struct seq_file *seq, void *v)
2766 {
2767 seq_setwidth(seq, 127);
2768 if (v == SEQ_START_TOKEN)
2769 seq_puts(seq, " sl local_address rem_address st tx_queue "
2770 "rx_queue tr tm->when retrnsmt uid timeout "
2771 "inode ref pointer drops");
2772 else {
2773 struct udp_iter_state *state = seq->private;
2774
2775 udp4_format_sock(v, seq, state->bucket);
2776 }
2777 seq_pad(seq, '\n');
2778 return 0;
2779 }
2780
2781 static const struct file_operations udp_afinfo_seq_fops = {
2782 .owner = THIS_MODULE,
2783 .open = udp_seq_open,
2784 .read = seq_read,
2785 .llseek = seq_lseek,
2786 .release = seq_release_net
2787 };
2788
2789 /* ------------------------------------------------------------------------ */
2790 static struct udp_seq_afinfo udp4_seq_afinfo = {
2791 .name = "udp",
2792 .family = AF_INET,
2793 .udp_table = &udp_table,
2794 .seq_fops = &udp_afinfo_seq_fops,
2795 .seq_ops = {
2796 .show = udp4_seq_show,
2797 },
2798 };
2799
udp4_proc_init_net(struct net * net)2800 static int __net_init udp4_proc_init_net(struct net *net)
2801 {
2802 return udp_proc_register(net, &udp4_seq_afinfo);
2803 }
2804
udp4_proc_exit_net(struct net * net)2805 static void __net_exit udp4_proc_exit_net(struct net *net)
2806 {
2807 udp_proc_unregister(net, &udp4_seq_afinfo);
2808 }
2809
2810 static struct pernet_operations udp4_net_ops = {
2811 .init = udp4_proc_init_net,
2812 .exit = udp4_proc_exit_net,
2813 };
2814
udp4_proc_init(void)2815 int __init udp4_proc_init(void)
2816 {
2817 return register_pernet_subsys(&udp4_net_ops);
2818 }
2819
udp4_proc_exit(void)2820 void udp4_proc_exit(void)
2821 {
2822 unregister_pernet_subsys(&udp4_net_ops);
2823 }
2824 #endif /* CONFIG_PROC_FS */
2825
2826 static __initdata unsigned long uhash_entries;
set_uhash_entries(char * str)2827 static int __init set_uhash_entries(char *str)
2828 {
2829 ssize_t ret;
2830
2831 if (!str)
2832 return 0;
2833
2834 ret = kstrtoul(str, 0, &uhash_entries);
2835 if (ret)
2836 return 0;
2837
2838 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2839 uhash_entries = UDP_HTABLE_SIZE_MIN;
2840 return 1;
2841 }
2842 __setup("uhash_entries=", set_uhash_entries);
2843
udp_table_init(struct udp_table * table,const char * name)2844 void __init udp_table_init(struct udp_table *table, const char *name)
2845 {
2846 unsigned int i;
2847
2848 table->hash = alloc_large_system_hash(name,
2849 2 * sizeof(struct udp_hslot),
2850 uhash_entries,
2851 21, /* one slot per 2 MB */
2852 0,
2853 &table->log,
2854 &table->mask,
2855 UDP_HTABLE_SIZE_MIN,
2856 64 * 1024);
2857
2858 table->hash2 = table->hash + (table->mask + 1);
2859 for (i = 0; i <= table->mask; i++) {
2860 INIT_HLIST_HEAD(&table->hash[i].head);
2861 table->hash[i].count = 0;
2862 spin_lock_init(&table->hash[i].lock);
2863 }
2864 for (i = 0; i <= table->mask; i++) {
2865 INIT_HLIST_HEAD(&table->hash2[i].head);
2866 table->hash2[i].count = 0;
2867 spin_lock_init(&table->hash2[i].lock);
2868 }
2869 }
2870
udp_flow_hashrnd(void)2871 u32 udp_flow_hashrnd(void)
2872 {
2873 static u32 hashrnd __read_mostly;
2874
2875 net_get_random_once(&hashrnd, sizeof(hashrnd));
2876
2877 return hashrnd;
2878 }
2879 EXPORT_SYMBOL(udp_flow_hashrnd);
2880
udp_init(void)2881 void __init udp_init(void)
2882 {
2883 unsigned long limit;
2884 unsigned int i;
2885
2886 udp_table_init(&udp_table, "UDP");
2887 limit = nr_free_buffer_pages() / 8;
2888 limit = max(limit, 128UL);
2889 sysctl_udp_mem[0] = limit / 4 * 3;
2890 sysctl_udp_mem[1] = limit;
2891 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2892
2893 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2894 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2895
2896 /* 16 spinlocks per cpu */
2897 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2898 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2899 GFP_KERNEL);
2900 if (!udp_busylocks)
2901 panic("UDP: failed to alloc udp_busylocks\n");
2902 for (i = 0; i < (1U << udp_busylocks_log); i++)
2903 spin_lock_init(udp_busylocks + i);
2904 }
2905