projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ipv4: Handle PMTU in all ICMP error handlers.
[deliverable/linux.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 "udp_impl.h"
112
113 struct udp_table udp_table __read_mostly;
114 EXPORT_SYMBOL(udp_table);
115
116 long sysctl_udp_mem[3] __read_mostly;
117 EXPORT_SYMBOL(sysctl_udp_mem);
118
119 int sysctl_udp_rmem_min __read_mostly;
120 EXPORT_SYMBOL(sysctl_udp_rmem_min);
121
122 int sysctl_udp_wmem_min __read_mostly;
123 EXPORT_SYMBOL(sysctl_udp_wmem_min);
124
125 atomic_long_t udp_memory_allocated;
126 EXPORT_SYMBOL(udp_memory_allocated);
127
128 #define MAX_UDP_PORTS 65536
129 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
130
131 static int udp_lib_lport_inuse(struct net *net, __u16 num,
132 const struct udp_hslot *hslot,
133 unsigned long *bitmap,
134 struct sock *sk,
135 int (*saddr_comp)(const struct sock *sk1,
136 const struct sock *sk2),
137 unsigned int log)
138 {
139 struct sock *sk2;
140 struct hlist_nulls_node *node;
141
142 sk_nulls_for_each(sk2, node, &hslot->head)
143 if (net_eq(sock_net(sk2), net) &&
144 sk2 != sk &&
145 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
146 (!sk2->sk_reuse || !sk->sk_reuse) &&
147 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
148 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
149 (*saddr_comp)(sk, sk2)) {
150 if (bitmap)
151 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
152 bitmap);
153 else
154 return 1;
155 }
156 return 0;
157 }
158
159 /*
160 * Note: we still hold spinlock of primary hash chain, so no other writer
161 * can insert/delete a socket with local_port == num
162 */
163 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
164 struct udp_hslot *hslot2,
165 struct sock *sk,
166 int (*saddr_comp)(const struct sock *sk1,
167 const struct sock *sk2))
168 {
169 struct sock *sk2;
170 struct hlist_nulls_node *node;
171 int res = 0;
172
173 spin_lock(&hslot2->lock);
174 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
175 if (net_eq(sock_net(sk2), net) &&
176 sk2 != sk &&
177 (udp_sk(sk2)->udp_port_hash == num) &&
178 (!sk2->sk_reuse || !sk->sk_reuse) &&
179 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
180 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
181 (*saddr_comp)(sk, sk2)) {
182 res = 1;
183 break;
184 }
185 spin_unlock(&hslot2->lock);
186 return res;
187 }
188
189 /**
190 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
191 *
192 * @sk: socket struct in question
193 * @snum: port number to look up
194 * @saddr_comp: AF-dependent comparison of bound local IP addresses
195 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
196 * with NULL address
197 */
198 int udp_lib_get_port(struct sock *sk, unsigned short snum,
199 int (*saddr_comp)(const struct sock *sk1,
200 const struct sock *sk2),
201 unsigned int hash2_nulladdr)
202 {
203 struct udp_hslot *hslot, *hslot2;
204 struct udp_table *udptable = sk->sk_prot->h.udp_table;
205 int error = 1;
206 struct net *net = sock_net(sk);
207
208 if (!snum) {
209 int low, high, remaining;
210 unsigned int rand;
211 unsigned short first, last;
212 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
213
214 inet_get_local_port_range(&low, &high);
215 remaining = (high - low) + 1;
216
217 rand = net_random();
218 first = (((u64)rand * remaining) >> 32) + low;
219 /*
220 * force rand to be an odd multiple of UDP_HTABLE_SIZE
221 */
222 rand = (rand | 1) * (udptable->mask + 1);
223 last = first + udptable->mask + 1;
224 do {
225 hslot = udp_hashslot(udptable, net, first);
226 bitmap_zero(bitmap, PORTS_PER_CHAIN);
227 spin_lock_bh(&hslot->lock);
228 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
229 saddr_comp, udptable->log);
230
231 snum = first;
232 /*
233 * Iterate on all possible values of snum for this hash.
234 * Using steps of an odd multiple of UDP_HTABLE_SIZE
235 * give us randomization and full range coverage.
236 */
237 do {
238 if (low <= snum && snum <= high &&
239 !test_bit(snum >> udptable->log, bitmap) &&
240 !inet_is_reserved_local_port(snum))
241 goto found;
242 snum += rand;
243 } while (snum != first);
244 spin_unlock_bh(&hslot->lock);
245 } while (++first != last);
246 goto fail;
247 } else {
248 hslot = udp_hashslot(udptable, net, snum);
249 spin_lock_bh(&hslot->lock);
250 if (hslot->count > 10) {
251 int exist;
252 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
253
254 slot2 &= udptable->mask;
255 hash2_nulladdr &= udptable->mask;
256
257 hslot2 = udp_hashslot2(udptable, slot2);
258 if (hslot->count < hslot2->count)
259 goto scan_primary_hash;
260
261 exist = udp_lib_lport_inuse2(net, snum, hslot2,
262 sk, saddr_comp);
263 if (!exist && (hash2_nulladdr != slot2)) {
264 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
265 exist = udp_lib_lport_inuse2(net, snum, hslot2,
266 sk, saddr_comp);
267 }
268 if (exist)
269 goto fail_unlock;
270 else
271 goto found;
272 }
273 scan_primary_hash:
274 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
275 saddr_comp, 0))
276 goto fail_unlock;
277 }
278 found:
279 inet_sk(sk)->inet_num = snum;
280 udp_sk(sk)->udp_port_hash = snum;
281 udp_sk(sk)->udp_portaddr_hash ^= snum;
282 if (sk_unhashed(sk)) {
283 sk_nulls_add_node_rcu(sk, &hslot->head);
284 hslot->count++;
285 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
286
287 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
288 spin_lock(&hslot2->lock);
289 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
290 &hslot2->head);
291 hslot2->count++;
292 spin_unlock(&hslot2->lock);
293 }
294 error = 0;
295 fail_unlock:
296 spin_unlock_bh(&hslot->lock);
297 fail:
298 return error;
299 }
300 EXPORT_SYMBOL(udp_lib_get_port);
301
302 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
303 {
304 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
305
306 return (!ipv6_only_sock(sk2) &&
307 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
308 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
309 }
310
311 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
312 unsigned int port)
313 {
314 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
315 }
316
317 int udp_v4_get_port(struct sock *sk, unsigned short snum)
318 {
319 unsigned int hash2_nulladdr =
320 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
321 unsigned int hash2_partial =
322 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
323
324 /* precompute partial secondary hash */
325 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
326 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
327 }
328
329 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
330 unsigned short hnum,
331 __be16 sport, __be32 daddr, __be16 dport, int dif)
332 {
333 int score = -1;
334
335 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
336 !ipv6_only_sock(sk)) {
337 struct inet_sock *inet = inet_sk(sk);
338
339 score = (sk->sk_family == PF_INET ? 1 : 0);
340 if (inet->inet_rcv_saddr) {
341 if (inet->inet_rcv_saddr != daddr)
342 return -1;
343 score += 2;
344 }
345 if (inet->inet_daddr) {
346 if (inet->inet_daddr != saddr)
347 return -1;
348 score += 2;
349 }
350 if (inet->inet_dport) {
351 if (inet->inet_dport != sport)
352 return -1;
353 score += 2;
354 }
355 if (sk->sk_bound_dev_if) {
356 if (sk->sk_bound_dev_if != dif)
357 return -1;
358 score += 2;
359 }
360 }
361 return score;
362 }
363
364 /*
365 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
366 */
367 #define SCORE2_MAX (1 + 2 + 2 + 2)
368 static inline int compute_score2(struct sock *sk, struct net *net,
369 __be32 saddr, __be16 sport,
370 __be32 daddr, unsigned int hnum, int dif)
371 {
372 int score = -1;
373
374 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
375 struct inet_sock *inet = inet_sk(sk);
376
377 if (inet->inet_rcv_saddr != daddr)
378 return -1;
379 if (inet->inet_num != hnum)
380 return -1;
381
382 score = (sk->sk_family == PF_INET ? 1 : 0);
383 if (inet->inet_daddr) {
384 if (inet->inet_daddr != saddr)
385 return -1;
386 score += 2;
387 }
388 if (inet->inet_dport) {
389 if (inet->inet_dport != sport)
390 return -1;
391 score += 2;
392 }
393 if (sk->sk_bound_dev_if) {
394 if (sk->sk_bound_dev_if != dif)
395 return -1;
396 score += 2;
397 }
398 }
399 return score;
400 }
401
402
403 /* called with read_rcu_lock() */
404 static struct sock *udp4_lib_lookup2(struct net *net,
405 __be32 saddr, __be16 sport,
406 __be32 daddr, unsigned int hnum, int dif,
407 struct udp_hslot *hslot2, unsigned int slot2)
408 {
409 struct sock *sk, *result;
410 struct hlist_nulls_node *node;
411 int score, badness;
412
413 begin:
414 result = NULL;
415 badness = -1;
416 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
417 score = compute_score2(sk, net, saddr, sport,
418 daddr, hnum, dif);
419 if (score > badness) {
420 result = sk;
421 badness = score;
422 if (score == SCORE2_MAX)
423 goto exact_match;
424 }
425 }
426 /*
427 * if the nulls value we got at the end of this lookup is
428 * not the expected one, we must restart lookup.
429 * We probably met an item that was moved to another chain.
430 */
431 if (get_nulls_value(node) != slot2)
432 goto begin;
433
434 if (result) {
435 exact_match:
436 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
437 result = NULL;
438 else if (unlikely(compute_score2(result, net, saddr, sport,
439 daddr, hnum, dif) < badness)) {
440 sock_put(result);
441 goto begin;
442 }
443 }
444 return result;
445 }
446
447 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
448 * harder than this. -DaveM
449 */
450 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
451 __be16 sport, __be32 daddr, __be16 dport,
452 int dif, struct udp_table *udptable)
453 {
454 struct sock *sk, *result;
455 struct hlist_nulls_node *node;
456 unsigned short hnum = ntohs(dport);
457 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
458 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
459 int score, badness;
460
461 rcu_read_lock();
462 if (hslot->count > 10) {
463 hash2 = udp4_portaddr_hash(net, daddr, hnum);
464 slot2 = hash2 & udptable->mask;
465 hslot2 = &udptable->hash2[slot2];
466 if (hslot->count < hslot2->count)
467 goto begin;
468
469 result = udp4_lib_lookup2(net, saddr, sport,
470 daddr, hnum, dif,
471 hslot2, slot2);
472 if (!result) {
473 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
474 slot2 = hash2 & udptable->mask;
475 hslot2 = &udptable->hash2[slot2];
476 if (hslot->count < hslot2->count)
477 goto begin;
478
479 result = udp4_lib_lookup2(net, saddr, sport,
480 htonl(INADDR_ANY), hnum, dif,
481 hslot2, slot2);
482 }
483 rcu_read_unlock();
484 return result;
485 }
486 begin:
487 result = NULL;
488 badness = -1;
489 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
490 score = compute_score(sk, net, saddr, hnum, sport,
491 daddr, dport, dif);
492 if (score > badness) {
493 result = sk;
494 badness = score;
495 }
496 }
497 /*
498 * if the nulls value we got at the end of this lookup is
499 * not the expected one, we must restart lookup.
500 * We probably met an item that was moved to another chain.
501 */
502 if (get_nulls_value(node) != slot)
503 goto begin;
504
505 if (result) {
506 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
507 result = NULL;
508 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
509 daddr, dport, dif) < badness)) {
510 sock_put(result);
511 goto begin;
512 }
513 }
514 rcu_read_unlock();
515 return result;
516 }
517 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
518
519 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
520 __be16 sport, __be16 dport,
521 struct udp_table *udptable)
522 {
523 struct sock *sk;
524 const struct iphdr *iph = ip_hdr(skb);
525
526 if (unlikely(sk = skb_steal_sock(skb)))
527 return sk;
528 else
529 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
530 iph->daddr, dport, inet_iif(skb),
531 udptable);
532 }
533
534 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
535 __be32 daddr, __be16 dport, int dif)
536 {
537 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
538 }
539 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
540
541 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
542 __be16 loc_port, __be32 loc_addr,
543 __be16 rmt_port, __be32 rmt_addr,
544 int dif)
545 {
546 struct hlist_nulls_node *node;
547 struct sock *s = sk;
548 unsigned short hnum = ntohs(loc_port);
549
550 sk_nulls_for_each_from(s, node) {
551 struct inet_sock *inet = inet_sk(s);
552
553 if (!net_eq(sock_net(s), net) ||
554 udp_sk(s)->udp_port_hash != hnum ||
555 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
556 (inet->inet_dport != rmt_port && inet->inet_dport) ||
557 (inet->inet_rcv_saddr &&
558 inet->inet_rcv_saddr != loc_addr) ||
559 ipv6_only_sock(s) ||
560 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
561 continue;
562 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
563 continue;
564 goto found;
565 }
566 s = NULL;
567 found:
568 return s;
569 }
570
571 /*
572 * This routine is called by the ICMP module when it gets some
573 * sort of error condition. If err < 0 then the socket should
574 * be closed and the error returned to the user. If err > 0
575 * it's just the icmp type << 8 | icmp code.
576 * Header points to the ip header of the error packet. We move
577 * on past this. Then (as it used to claim before adjustment)
578 * header points to the first 8 bytes of the udp header. We need
579 * to find the appropriate port.
580 */
581
582 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
583 {
584 struct inet_sock *inet;
585 const struct iphdr *iph = (const struct iphdr *)skb->data;
586 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
587 const int type = icmp_hdr(skb)->type;
588 const int code = icmp_hdr(skb)->code;
589 struct sock *sk;
590 int harderr;
591 int err;
592 struct net *net = dev_net(skb->dev);
593
594 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
595 iph->saddr, uh->source, skb->dev->ifindex, udptable);
596 if (sk == NULL) {
597 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
598 return; /* No socket for error */
599 }
600
601 err = 0;
602 harderr = 0;
603 inet = inet_sk(sk);
604
605 switch (type) {
606 default:
607 case ICMP_TIME_EXCEEDED:
608 err = EHOSTUNREACH;
609 break;
610 case ICMP_SOURCE_QUENCH:
611 goto out;
612 case ICMP_PARAMETERPROB:
613 err = EPROTO;
614 harderr = 1;
615 break;
616 case ICMP_DEST_UNREACH:
617 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
618 ipv4_sk_update_pmtu(skb, sk, info);
619 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
620 err = EMSGSIZE;
621 harderr = 1;
622 break;
623 }
624 goto out;
625 }
626 err = EHOSTUNREACH;
627 if (code <= NR_ICMP_UNREACH) {
628 harderr = icmp_err_convert[code].fatal;
629 err = icmp_err_convert[code].errno;
630 }
631 break;
632 }
633
634 /*
635 * RFC1122: OK. Passes ICMP errors back to application, as per
636 * 4.1.3.3.
637 */
638 if (!inet->recverr) {
639 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
640 goto out;
641 } else
642 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
643
644 sk->sk_err = err;
645 sk->sk_error_report(sk);
646 out:
647 sock_put(sk);
648 }
649
650 void udp_err(struct sk_buff *skb, u32 info)
651 {
652 __udp4_lib_err(skb, info, &udp_table);
653 }
654
655 /*
656 * Throw away all pending data and cancel the corking. Socket is locked.
657 */
658 void udp_flush_pending_frames(struct sock *sk)
659 {
660 struct udp_sock *up = udp_sk(sk);
661
662 if (up->pending) {
663 up->len = 0;
664 up->pending = 0;
665 ip_flush_pending_frames(sk);
666 }
667 }
668 EXPORT_SYMBOL(udp_flush_pending_frames);
669
670 /**
671 * udp4_hwcsum - handle outgoing HW checksumming
672 * @skb: sk_buff containing the filled-in UDP header
673 * (checksum field must be zeroed out)
674 * @src: source IP address
675 * @dst: destination IP address
676 */
677 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
678 {
679 struct udphdr *uh = udp_hdr(skb);
680 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
681 int offset = skb_transport_offset(skb);
682 int len = skb->len - offset;
683 int hlen = len;
684 __wsum csum = 0;
685
686 if (!frags) {
687 /*
688 * Only one fragment on the socket.
689 */
690 skb->csum_start = skb_transport_header(skb) - skb->head;
691 skb->csum_offset = offsetof(struct udphdr, check);
692 uh->check = ~csum_tcpudp_magic(src, dst, len,
693 IPPROTO_UDP, 0);
694 } else {
695 /*
696 * HW-checksum won't work as there are two or more
697 * fragments on the socket so that all csums of sk_buffs
698 * should be together
699 */
700 do {
701 csum = csum_add(csum, frags->csum);
702 hlen -= frags->len;
703 } while ((frags = frags->next));
704
705 csum = skb_checksum(skb, offset, hlen, csum);
706 skb->ip_summed = CHECKSUM_NONE;
707
708 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
709 if (uh->check == 0)
710 uh->check = CSUM_MANGLED_0;
711 }
712 }
713
714 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
715 {
716 struct sock *sk = skb->sk;
717 struct inet_sock *inet = inet_sk(sk);
718 struct udphdr *uh;
719 int err = 0;
720 int is_udplite = IS_UDPLITE(sk);
721 int offset = skb_transport_offset(skb);
722 int len = skb->len - offset;
723 __wsum csum = 0;
724
725 /*
726 * Create a UDP header
727 */
728 uh = udp_hdr(skb);
729 uh->source = inet->inet_sport;
730 uh->dest = fl4->fl4_dport;
731 uh->len = htons(len);
732 uh->check = 0;
733
734 if (is_udplite) /* UDP-Lite */
735 csum = udplite_csum(skb);
736
737 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
738
739 skb->ip_summed = CHECKSUM_NONE;
740 goto send;
741
742 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
743
744 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
745 goto send;
746
747 } else
748 csum = udp_csum(skb);
749
750 /* add protocol-dependent pseudo-header */
751 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
752 sk->sk_protocol, csum);
753 if (uh->check == 0)
754 uh->check = CSUM_MANGLED_0;
755
756 send:
757 err = ip_send_skb(skb);
758 if (err) {
759 if (err == -ENOBUFS && !inet->recverr) {
760 UDP_INC_STATS_USER(sock_net(sk),
761 UDP_MIB_SNDBUFERRORS, is_udplite);
762 err = 0;
763 }
764 } else
765 UDP_INC_STATS_USER(sock_net(sk),
766 UDP_MIB_OUTDATAGRAMS, is_udplite);
767 return err;
768 }
769
770 /*
771 * Push out all pending data as one UDP datagram. Socket is locked.
772 */
773 static int udp_push_pending_frames(struct sock *sk)
774 {
775 struct udp_sock *up = udp_sk(sk);
776 struct inet_sock *inet = inet_sk(sk);
777 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
778 struct sk_buff *skb;
779 int err = 0;
780
781 skb = ip_finish_skb(sk, fl4);
782 if (!skb)
783 goto out;
784
785 err = udp_send_skb(skb, fl4);
786
787 out:
788 up->len = 0;
789 up->pending = 0;
790 return err;
791 }
792
793 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
794 size_t len)
795 {
796 struct inet_sock *inet = inet_sk(sk);
797 struct udp_sock *up = udp_sk(sk);
798 struct flowi4 fl4_stack;
799 struct flowi4 *fl4;
800 int ulen = len;
801 struct ipcm_cookie ipc;
802 struct rtable *rt = NULL;
803 int free = 0;
804 int connected = 0;
805 __be32 daddr, faddr, saddr;
806 __be16 dport;
807 u8 tos;
808 int err, is_udplite = IS_UDPLITE(sk);
809 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
810 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
811 struct sk_buff *skb;
812 struct ip_options_data opt_copy;
813
814 if (len > 0xFFFF)
815 return -EMSGSIZE;
816
817 /*
818 * Check the flags.
819 */
820
821 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
822 return -EOPNOTSUPP;
823
824 ipc.opt = NULL;
825 ipc.tx_flags = 0;
826
827 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
828
829 fl4 = &inet->cork.fl.u.ip4;
830 if (up->pending) {
831 /*
832 * There are pending frames.
833 * The socket lock must be held while it's corked.
834 */
835 lock_sock(sk);
836 if (likely(up->pending)) {
837 if (unlikely(up->pending != AF_INET)) {
838 release_sock(sk);
839 return -EINVAL;
840 }
841 goto do_append_data;
842 }
843 release_sock(sk);
844 }
845 ulen += sizeof(struct udphdr);
846
847 /*
848 * Get and verify the address.
849 */
850 if (msg->msg_name) {
851 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
852 if (msg->msg_namelen < sizeof(*usin))
853 return -EINVAL;
854 if (usin->sin_family != AF_INET) {
855 if (usin->sin_family != AF_UNSPEC)
856 return -EAFNOSUPPORT;
857 }
858
859 daddr = usin->sin_addr.s_addr;
860 dport = usin->sin_port;
861 if (dport == 0)
862 return -EINVAL;
863 } else {
864 if (sk->sk_state != TCP_ESTABLISHED)
865 return -EDESTADDRREQ;
866 daddr = inet->inet_daddr;
867 dport = inet->inet_dport;
868 /* Open fast path for connected socket.
869 Route will not be used, if at least one option is set.
870 */
871 connected = 1;
872 }
873 ipc.addr = inet->inet_saddr;
874
875 ipc.oif = sk->sk_bound_dev_if;
876 err = sock_tx_timestamp(sk, &ipc.tx_flags);
877 if (err)
878 return err;
879 if (msg->msg_controllen) {
880 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
881 if (err)
882 return err;
883 if (ipc.opt)
884 free = 1;
885 connected = 0;
886 }
887 if (!ipc.opt) {
888 struct ip_options_rcu *inet_opt;
889
890 rcu_read_lock();
891 inet_opt = rcu_dereference(inet->inet_opt);
892 if (inet_opt) {
893 memcpy(&opt_copy, inet_opt,
894 sizeof(*inet_opt) + inet_opt->opt.optlen);
895 ipc.opt = &opt_copy.opt;
896 }
897 rcu_read_unlock();
898 }
899
900 saddr = ipc.addr;
901 ipc.addr = faddr = daddr;
902
903 if (ipc.opt && ipc.opt->opt.srr) {
904 if (!daddr)
905 return -EINVAL;
906 faddr = ipc.opt->opt.faddr;
907 connected = 0;
908 }
909 tos = RT_TOS(inet->tos);
910 if (sock_flag(sk, SOCK_LOCALROUTE) ||
911 (msg->msg_flags & MSG_DONTROUTE) ||
912 (ipc.opt && ipc.opt->opt.is_strictroute)) {
913 tos |= RTO_ONLINK;
914 connected = 0;
915 }
916
917 if (ipv4_is_multicast(daddr)) {
918 if (!ipc.oif)
919 ipc.oif = inet->mc_index;
920 if (!saddr)
921 saddr = inet->mc_addr;
922 connected = 0;
923 } else if (!ipc.oif)
924 ipc.oif = inet->uc_index;
925
926 if (connected)
927 rt = (struct rtable *)sk_dst_check(sk, 0);
928
929 if (rt == NULL) {
930 struct net *net = sock_net(sk);
931
932 fl4 = &fl4_stack;
933 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
934 RT_SCOPE_UNIVERSE, sk->sk_protocol,
935 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
936 faddr, saddr, dport, inet->inet_sport);
937
938 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
939 rt = ip_route_output_flow(net, fl4, sk);
940 if (IS_ERR(rt)) {
941 err = PTR_ERR(rt);
942 rt = NULL;
943 if (err == -ENETUNREACH)
944 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
945 goto out;
946 }
947
948 err = -EACCES;
949 if ((rt->rt_flags & RTCF_BROADCAST) &&
950 !sock_flag(sk, SOCK_BROADCAST))
951 goto out;
952 if (connected)
953 sk_dst_set(sk, dst_clone(&rt->dst));
954 }
955
956 if (msg->msg_flags&MSG_CONFIRM)
957 goto do_confirm;
958 back_from_confirm:
959
960 saddr = fl4->saddr;
961 if (!ipc.addr)
962 daddr = ipc.addr = fl4->daddr;
963
964 /* Lockless fast path for the non-corking case. */
965 if (!corkreq) {
966 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
967 sizeof(struct udphdr), &ipc, &rt,
968 msg->msg_flags);
969 err = PTR_ERR(skb);
970 if (skb && !IS_ERR(skb))
971 err = udp_send_skb(skb, fl4);
972 goto out;
973 }
974
975 lock_sock(sk);
976 if (unlikely(up->pending)) {
977 /* The socket is already corked while preparing it. */
978 /* ... which is an evident application bug. --ANK */
979 release_sock(sk);
980
981 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
982 err = -EINVAL;
983 goto out;
984 }
985 /*
986 * Now cork the socket to pend data.
987 */
988 fl4 = &inet->cork.fl.u.ip4;
989 fl4->daddr = daddr;
990 fl4->saddr = saddr;
991 fl4->fl4_dport = dport;
992 fl4->fl4_sport = inet->inet_sport;
993 up->pending = AF_INET;
994
995 do_append_data:
996 up->len += ulen;
997 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
998 sizeof(struct udphdr), &ipc, &rt,
999 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1000 if (err)
1001 udp_flush_pending_frames(sk);
1002 else if (!corkreq)
1003 err = udp_push_pending_frames(sk);
1004 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1005 up->pending = 0;
1006 release_sock(sk);
1007
1008 out:
1009 ip_rt_put(rt);
1010 if (free)
1011 kfree(ipc.opt);
1012 if (!err)
1013 return len;
1014 /*
1015 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1016 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1017 * we don't have a good statistic (IpOutDiscards but it can be too many
1018 * things). We could add another new stat but at least for now that
1019 * seems like overkill.
1020 */
1021 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1022 UDP_INC_STATS_USER(sock_net(sk),
1023 UDP_MIB_SNDBUFERRORS, is_udplite);
1024 }
1025 return err;
1026
1027 do_confirm:
1028 dst_confirm(&rt->dst);
1029 if (!(msg->msg_flags&MSG_PROBE) || len)
1030 goto back_from_confirm;
1031 err = 0;
1032 goto out;
1033 }
1034 EXPORT_SYMBOL(udp_sendmsg);
1035
1036 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1037 size_t size, int flags)
1038 {
1039 struct inet_sock *inet = inet_sk(sk);
1040 struct udp_sock *up = udp_sk(sk);
1041 int ret;
1042
1043 if (!up->pending) {
1044 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1045
1046 /* Call udp_sendmsg to specify destination address which
1047 * sendpage interface can't pass.
1048 * This will succeed only when the socket is connected.
1049 */
1050 ret = udp_sendmsg(NULL, sk, &msg, 0);
1051 if (ret < 0)
1052 return ret;
1053 }
1054
1055 lock_sock(sk);
1056
1057 if (unlikely(!up->pending)) {
1058 release_sock(sk);
1059
1060 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1061 return -EINVAL;
1062 }
1063
1064 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1065 page, offset, size, flags);
1066 if (ret == -EOPNOTSUPP) {
1067 release_sock(sk);
1068 return sock_no_sendpage(sk->sk_socket, page, offset,
1069 size, flags);
1070 }
1071 if (ret < 0) {
1072 udp_flush_pending_frames(sk);
1073 goto out;
1074 }
1075
1076 up->len += size;
1077 if (!(up->corkflag || (flags&MSG_MORE)))
1078 ret = udp_push_pending_frames(sk);
1079 if (!ret)
1080 ret = size;
1081 out:
1082 release_sock(sk);
1083 return ret;
1084 }
1085
1086
1087 /**
1088 * first_packet_length - return length of first packet in receive queue
1089 * @sk: socket
1090 *
1091 * Drops all bad checksum frames, until a valid one is found.
1092 * Returns the length of found skb, or 0 if none is found.
1093 */
1094 static unsigned int first_packet_length(struct sock *sk)
1095 {
1096 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1097 struct sk_buff *skb;
1098 unsigned int res;
1099
1100 __skb_queue_head_init(&list_kill);
1101
1102 spin_lock_bh(&rcvq->lock);
1103 while ((skb = skb_peek(rcvq)) != NULL &&
1104 udp_lib_checksum_complete(skb)) {
1105 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1106 IS_UDPLITE(sk));
1107 atomic_inc(&sk->sk_drops);
1108 __skb_unlink(skb, rcvq);
1109 __skb_queue_tail(&list_kill, skb);
1110 }
1111 res = skb ? skb->len : 0;
1112 spin_unlock_bh(&rcvq->lock);
1113
1114 if (!skb_queue_empty(&list_kill)) {
1115 bool slow = lock_sock_fast(sk);
1116
1117 __skb_queue_purge(&list_kill);
1118 sk_mem_reclaim_partial(sk);
1119 unlock_sock_fast(sk, slow);
1120 }
1121 return res;
1122 }
1123
1124 /*
1125 * IOCTL requests applicable to the UDP protocol
1126 */
1127
1128 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1129 {
1130 switch (cmd) {
1131 case SIOCOUTQ:
1132 {
1133 int amount = sk_wmem_alloc_get(sk);
1134
1135 return put_user(amount, (int __user *)arg);
1136 }
1137
1138 case SIOCINQ:
1139 {
1140 unsigned int amount = first_packet_length(sk);
1141
1142 if (amount)
1143 /*
1144 * We will only return the amount
1145 * of this packet since that is all
1146 * that will be read.
1147 */
1148 amount -= sizeof(struct udphdr);
1149
1150 return put_user(amount, (int __user *)arg);
1151 }
1152
1153 default:
1154 return -ENOIOCTLCMD;
1155 }
1156
1157 return 0;
1158 }
1159 EXPORT_SYMBOL(udp_ioctl);
1160
1161 /*
1162 * This should be easy, if there is something there we
1163 * return it, otherwise we block.
1164 */
1165
1166 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1167 size_t len, int noblock, int flags, int *addr_len)
1168 {
1169 struct inet_sock *inet = inet_sk(sk);
1170 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1171 struct sk_buff *skb;
1172 unsigned int ulen, copied;
1173 int peeked, off = 0;
1174 int err;
1175 int is_udplite = IS_UDPLITE(sk);
1176 bool slow;
1177
1178 /*
1179 * Check any passed addresses
1180 */
1181 if (addr_len)
1182 *addr_len = sizeof(*sin);
1183
1184 if (flags & MSG_ERRQUEUE)
1185 return ip_recv_error(sk, msg, len);
1186
1187 try_again:
1188 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1189 &peeked, &off, &err);
1190 if (!skb)
1191 goto out;
1192
1193 ulen = skb->len - sizeof(struct udphdr);
1194 copied = len;
1195 if (copied > ulen)
1196 copied = ulen;
1197 else if (copied < ulen)
1198 msg->msg_flags |= MSG_TRUNC;
1199
1200 /*
1201 * If checksum is needed at all, try to do it while copying the
1202 * data. If the data is truncated, or if we only want a partial
1203 * coverage checksum (UDP-Lite), do it before the copy.
1204 */
1205
1206 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1207 if (udp_lib_checksum_complete(skb))
1208 goto csum_copy_err;
1209 }
1210
1211 if (skb_csum_unnecessary(skb))
1212 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1213 msg->msg_iov, copied);
1214 else {
1215 err = skb_copy_and_csum_datagram_iovec(skb,
1216 sizeof(struct udphdr),
1217 msg->msg_iov);
1218
1219 if (err == -EINVAL)
1220 goto csum_copy_err;
1221 }
1222
1223 if (err)
1224 goto out_free;
1225
1226 if (!peeked)
1227 UDP_INC_STATS_USER(sock_net(sk),
1228 UDP_MIB_INDATAGRAMS, is_udplite);
1229
1230 sock_recv_ts_and_drops(msg, sk, skb);
1231
1232 /* Copy the address. */
1233 if (sin) {
1234 sin->sin_family = AF_INET;
1235 sin->sin_port = udp_hdr(skb)->source;
1236 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1237 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1238 }
1239 if (inet->cmsg_flags)
1240 ip_cmsg_recv(msg, skb);
1241
1242 err = copied;
1243 if (flags & MSG_TRUNC)
1244 err = ulen;
1245
1246 out_free:
1247 skb_free_datagram_locked(sk, skb);
1248 out:
1249 return err;
1250
1251 csum_copy_err:
1252 slow = lock_sock_fast(sk);
1253 if (!skb_kill_datagram(sk, skb, flags))
1254 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1255 unlock_sock_fast(sk, slow);
1256
1257 if (noblock)
1258 return -EAGAIN;
1259
1260 /* starting over for a new packet */
1261 msg->msg_flags &= ~MSG_TRUNC;
1262 goto try_again;
1263 }
1264
1265
1266 int udp_disconnect(struct sock *sk, int flags)
1267 {
1268 struct inet_sock *inet = inet_sk(sk);
1269 /*
1270 * 1003.1g - break association.
1271 */
1272
1273 sk->sk_state = TCP_CLOSE;
1274 inet->inet_daddr = 0;
1275 inet->inet_dport = 0;
1276 sock_rps_reset_rxhash(sk);
1277 sk->sk_bound_dev_if = 0;
1278 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1279 inet_reset_saddr(sk);
1280
1281 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1282 sk->sk_prot->unhash(sk);
1283 inet->inet_sport = 0;
1284 }
1285 sk_dst_reset(sk);
1286 return 0;
1287 }
1288 EXPORT_SYMBOL(udp_disconnect);
1289
1290 void udp_lib_unhash(struct sock *sk)
1291 {
1292 if (sk_hashed(sk)) {
1293 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1294 struct udp_hslot *hslot, *hslot2;
1295
1296 hslot = udp_hashslot(udptable, sock_net(sk),
1297 udp_sk(sk)->udp_port_hash);
1298 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1299
1300 spin_lock_bh(&hslot->lock);
1301 if (sk_nulls_del_node_init_rcu(sk)) {
1302 hslot->count--;
1303 inet_sk(sk)->inet_num = 0;
1304 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1305
1306 spin_lock(&hslot2->lock);
1307 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1308 hslot2->count--;
1309 spin_unlock(&hslot2->lock);
1310 }
1311 spin_unlock_bh(&hslot->lock);
1312 }
1313 }
1314 EXPORT_SYMBOL(udp_lib_unhash);
1315
1316 /*
1317 * inet_rcv_saddr was changed, we must rehash secondary hash
1318 */
1319 void udp_lib_rehash(struct sock *sk, u16 newhash)
1320 {
1321 if (sk_hashed(sk)) {
1322 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1323 struct udp_hslot *hslot, *hslot2, *nhslot2;
1324
1325 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1326 nhslot2 = udp_hashslot2(udptable, newhash);
1327 udp_sk(sk)->udp_portaddr_hash = newhash;
1328 if (hslot2 != nhslot2) {
1329 hslot = udp_hashslot(udptable, sock_net(sk),
1330 udp_sk(sk)->udp_port_hash);
1331 /* we must lock primary chain too */
1332 spin_lock_bh(&hslot->lock);
1333
1334 spin_lock(&hslot2->lock);
1335 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1336 hslot2->count--;
1337 spin_unlock(&hslot2->lock);
1338
1339 spin_lock(&nhslot2->lock);
1340 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1341 &nhslot2->head);
1342 nhslot2->count++;
1343 spin_unlock(&nhslot2->lock);
1344
1345 spin_unlock_bh(&hslot->lock);
1346 }
1347 }
1348 }
1349 EXPORT_SYMBOL(udp_lib_rehash);
1350
1351 static void udp_v4_rehash(struct sock *sk)
1352 {
1353 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1354 inet_sk(sk)->inet_rcv_saddr,
1355 inet_sk(sk)->inet_num);
1356 udp_lib_rehash(sk, new_hash);
1357 }
1358
1359 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1360 {
1361 int rc;
1362
1363 if (inet_sk(sk)->inet_daddr)
1364 sock_rps_save_rxhash(sk, skb);
1365
1366 rc = sock_queue_rcv_skb(sk, skb);
1367 if (rc < 0) {
1368 int is_udplite = IS_UDPLITE(sk);
1369
1370 /* Note that an ENOMEM error is charged twice */
1371 if (rc == -ENOMEM)
1372 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1373 is_udplite);
1374 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1375 kfree_skb(skb);
1376 trace_udp_fail_queue_rcv_skb(rc, sk);
1377 return -1;
1378 }
1379
1380 return 0;
1381
1382 }
1383
1384 static struct static_key udp_encap_needed __read_mostly;
1385 void udp_encap_enable(void)
1386 {
1387 if (!static_key_enabled(&udp_encap_needed))
1388 static_key_slow_inc(&udp_encap_needed);
1389 }
1390 EXPORT_SYMBOL(udp_encap_enable);
1391
1392 /* returns:
1393 * -1: error
1394 * 0: success
1395 * >0: "udp encap" protocol resubmission
1396 *
1397 * Note that in the success and error cases, the skb is assumed to
1398 * have either been requeued or freed.
1399 */
1400 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1401 {
1402 struct udp_sock *up = udp_sk(sk);
1403 int rc;
1404 int is_udplite = IS_UDPLITE(sk);
1405
1406 /*
1407 * Charge it to the socket, dropping if the queue is full.
1408 */
1409 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1410 goto drop;
1411 nf_reset(skb);
1412
1413 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1414 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1415
1416 /*
1417 * This is an encapsulation socket so pass the skb to
1418 * the socket's udp_encap_rcv() hook. Otherwise, just
1419 * fall through and pass this up the UDP socket.
1420 * up->encap_rcv() returns the following value:
1421 * =0 if skb was successfully passed to the encap
1422 * handler or was discarded by it.
1423 * >0 if skb should be passed on to UDP.
1424 * <0 if skb should be resubmitted as proto -N
1425 */
1426
1427 /* if we're overly short, let UDP handle it */
1428 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1429 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1430 int ret;
1431
1432 ret = encap_rcv(sk, skb);
1433 if (ret <= 0) {
1434 UDP_INC_STATS_BH(sock_net(sk),
1435 UDP_MIB_INDATAGRAMS,
1436 is_udplite);
1437 return -ret;
1438 }
1439 }
1440
1441 /* FALLTHROUGH -- it's a UDP Packet */
1442 }
1443
1444 /*
1445 * UDP-Lite specific tests, ignored on UDP sockets
1446 */
1447 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1448
1449 /*
1450 * MIB statistics other than incrementing the error count are
1451 * disabled for the following two types of errors: these depend
1452 * on the application settings, not on the functioning of the
1453 * protocol stack as such.
1454 *
1455 * RFC 3828 here recommends (sec 3.3): "There should also be a
1456 * way ... to ... at least let the receiving application block
1457 * delivery of packets with coverage values less than a value
1458 * provided by the application."
1459 */
1460 if (up->pcrlen == 0) { /* full coverage was set */
1461 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1462 UDP_SKB_CB(skb)->cscov, skb->len);
1463 goto drop;
1464 }
1465 /* The next case involves violating the min. coverage requested
1466 * by the receiver. This is subtle: if receiver wants x and x is
1467 * greater than the buffersize/MTU then receiver will complain
1468 * that it wants x while sender emits packets of smaller size y.
1469 * Therefore the above ...()->partial_cov statement is essential.
1470 */
1471 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1472 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1473 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1474 goto drop;
1475 }
1476 }
1477
1478 if (rcu_access_pointer(sk->sk_filter) &&
1479 udp_lib_checksum_complete(skb))
1480 goto drop;
1481
1482
1483 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1484 goto drop;
1485
1486 rc = 0;
1487
1488 ipv4_pktinfo_prepare(skb);
1489 bh_lock_sock(sk);
1490 if (!sock_owned_by_user(sk))
1491 rc = __udp_queue_rcv_skb(sk, skb);
1492 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1493 bh_unlock_sock(sk);
1494 goto drop;
1495 }
1496 bh_unlock_sock(sk);
1497
1498 return rc;
1499
1500 drop:
1501 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1502 atomic_inc(&sk->sk_drops);
1503 kfree_skb(skb);
1504 return -1;
1505 }
1506
1507
1508 static void flush_stack(struct sock **stack, unsigned int count,
1509 struct sk_buff *skb, unsigned int final)
1510 {
1511 unsigned int i;
1512 struct sk_buff *skb1 = NULL;
1513 struct sock *sk;
1514
1515 for (i = 0; i < count; i++) {
1516 sk = stack[i];
1517 if (likely(skb1 == NULL))
1518 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1519
1520 if (!skb1) {
1521 atomic_inc(&sk->sk_drops);
1522 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1523 IS_UDPLITE(sk));
1524 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1525 IS_UDPLITE(sk));
1526 }
1527
1528 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1529 skb1 = NULL;
1530 }
1531 if (unlikely(skb1))
1532 kfree_skb(skb1);
1533 }
1534
1535 /*
1536 * Multicasts and broadcasts go to each listener.
1537 *
1538 * Note: called only from the BH handler context.
1539 */
1540 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1541 struct udphdr *uh,
1542 __be32 saddr, __be32 daddr,
1543 struct udp_table *udptable)
1544 {
1545 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1546 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1547 int dif;
1548 unsigned int i, count = 0;
1549
1550 spin_lock(&hslot->lock);
1551 sk = sk_nulls_head(&hslot->head);
1552 dif = skb->dev->ifindex;
1553 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1554 while (sk) {
1555 stack[count++] = sk;
1556 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1557 daddr, uh->source, saddr, dif);
1558 if (unlikely(count == ARRAY_SIZE(stack))) {
1559 if (!sk)
1560 break;
1561 flush_stack(stack, count, skb, ~0);
1562 count = 0;
1563 }
1564 }
1565 /*
1566 * before releasing chain lock, we must take a reference on sockets
1567 */
1568 for (i = 0; i < count; i++)
1569 sock_hold(stack[i]);
1570
1571 spin_unlock(&hslot->lock);
1572
1573 /*
1574 * do the slow work with no lock held
1575 */
1576 if (count) {
1577 flush_stack(stack, count, skb, count - 1);
1578
1579 for (i = 0; i < count; i++)
1580 sock_put(stack[i]);
1581 } else {
1582 kfree_skb(skb);
1583 }
1584 return 0;
1585 }
1586
1587 /* Initialize UDP checksum. If exited with zero value (success),
1588 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1589 * Otherwise, csum completion requires chacksumming packet body,
1590 * including udp header and folding it to skb->csum.
1591 */
1592 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1593 int proto)
1594 {
1595 const struct iphdr *iph;
1596 int err;
1597
1598 UDP_SKB_CB(skb)->partial_cov = 0;
1599 UDP_SKB_CB(skb)->cscov = skb->len;
1600
1601 if (proto == IPPROTO_UDPLITE) {
1602 err = udplite_checksum_init(skb, uh);
1603 if (err)
1604 return err;
1605 }
1606
1607 iph = ip_hdr(skb);
1608 if (uh->check == 0) {
1609 skb->ip_summed = CHECKSUM_UNNECESSARY;
1610 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1611 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1612 proto, skb->csum))
1613 skb->ip_summed = CHECKSUM_UNNECESSARY;
1614 }
1615 if (!skb_csum_unnecessary(skb))
1616 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1617 skb->len, proto, 0);
1618 /* Probably, we should checksum udp header (it should be in cache
1619 * in any case) and data in tiny packets (< rx copybreak).
1620 */
1621
1622 return 0;
1623 }
1624
1625 /*
1626 * All we need to do is get the socket, and then do a checksum.
1627 */
1628
1629 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1630 int proto)
1631 {
1632 struct sock *sk;
1633 struct udphdr *uh;
1634 unsigned short ulen;
1635 struct rtable *rt = skb_rtable(skb);
1636 __be32 saddr, daddr;
1637 struct net *net = dev_net(skb->dev);
1638
1639 /*
1640 * Validate the packet.
1641 */
1642 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1643 goto drop; /* No space for header. */
1644
1645 uh = udp_hdr(skb);
1646 ulen = ntohs(uh->len);
1647 saddr = ip_hdr(skb)->saddr;
1648 daddr = ip_hdr(skb)->daddr;
1649
1650 if (ulen > skb->len)
1651 goto short_packet;
1652
1653 if (proto == IPPROTO_UDP) {
1654 /* UDP validates ulen. */
1655 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1656 goto short_packet;
1657 uh = udp_hdr(skb);
1658 }
1659
1660 if (udp4_csum_init(skb, uh, proto))
1661 goto csum_error;
1662
1663 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1664 return __udp4_lib_mcast_deliver(net, skb, uh,
1665 saddr, daddr, udptable);
1666
1667 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1668
1669 if (sk != NULL) {
1670 int ret = udp_queue_rcv_skb(sk, skb);
1671 sock_put(sk);
1672
1673 /* a return value > 0 means to resubmit the input, but
1674 * it wants the return to be -protocol, or 0
1675 */
1676 if (ret > 0)
1677 return -ret;
1678 return 0;
1679 }
1680
1681 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1682 goto drop;
1683 nf_reset(skb);
1684
1685 /* No socket. Drop packet silently, if checksum is wrong */
1686 if (udp_lib_checksum_complete(skb))
1687 goto csum_error;
1688
1689 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1690 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1691
1692 /*
1693 * Hmm. We got an UDP packet to a port to which we
1694 * don't wanna listen. Ignore it.
1695 */
1696 kfree_skb(skb);
1697 return 0;
1698
1699 short_packet:
1700 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1701 proto == IPPROTO_UDPLITE ? "Lite" : "",
1702 &saddr, ntohs(uh->source),
1703 ulen, skb->len,
1704 &daddr, ntohs(uh->dest));
1705 goto drop;
1706
1707 csum_error:
1708 /*
1709 * RFC1122: OK. Discards the bad packet silently (as far as
1710 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1711 */
1712 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1713 proto == IPPROTO_UDPLITE ? "Lite" : "",
1714 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1715 ulen);
1716 drop:
1717 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1718 kfree_skb(skb);
1719 return 0;
1720 }
1721
1722 int udp_rcv(struct sk_buff *skb)
1723 {
1724 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1725 }
1726
1727 void udp_destroy_sock(struct sock *sk)
1728 {
1729 bool slow = lock_sock_fast(sk);
1730 udp_flush_pending_frames(sk);
1731 unlock_sock_fast(sk, slow);
1732 }
1733
1734 /*
1735 * Socket option code for UDP
1736 */
1737 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1738 char __user *optval, unsigned int optlen,
1739 int (*push_pending_frames)(struct sock *))
1740 {
1741 struct udp_sock *up = udp_sk(sk);
1742 int val;
1743 int err = 0;
1744 int is_udplite = IS_UDPLITE(sk);
1745
1746 if (optlen < sizeof(int))
1747 return -EINVAL;
1748
1749 if (get_user(val, (int __user *)optval))
1750 return -EFAULT;
1751
1752 switch (optname) {
1753 case UDP_CORK:
1754 if (val != 0) {
1755 up->corkflag = 1;
1756 } else {
1757 up->corkflag = 0;
1758 lock_sock(sk);
1759 (*push_pending_frames)(sk);
1760 release_sock(sk);
1761 }
1762 break;
1763
1764 case UDP_ENCAP:
1765 switch (val) {
1766 case 0:
1767 case UDP_ENCAP_ESPINUDP:
1768 case UDP_ENCAP_ESPINUDP_NON_IKE:
1769 up->encap_rcv = xfrm4_udp_encap_rcv;
1770 /* FALLTHROUGH */
1771 case UDP_ENCAP_L2TPINUDP:
1772 up->encap_type = val;
1773 udp_encap_enable();
1774 break;
1775 default:
1776 err = -ENOPROTOOPT;
1777 break;
1778 }
1779 break;
1780
1781 /*
1782 * UDP-Lite's partial checksum coverage (RFC 3828).
1783 */
1784 /* The sender sets actual checksum coverage length via this option.
1785 * The case coverage > packet length is handled by send module. */
1786 case UDPLITE_SEND_CSCOV:
1787 if (!is_udplite) /* Disable the option on UDP sockets */
1788 return -ENOPROTOOPT;
1789 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1790 val = 8;
1791 else if (val > USHRT_MAX)
1792 val = USHRT_MAX;
1793 up->pcslen = val;
1794 up->pcflag |= UDPLITE_SEND_CC;
1795 break;
1796
1797 /* The receiver specifies a minimum checksum coverage value. To make
1798 * sense, this should be set to at least 8 (as done below). If zero is
1799 * used, this again means full checksum coverage. */
1800 case UDPLITE_RECV_CSCOV:
1801 if (!is_udplite) /* Disable the option on UDP sockets */
1802 return -ENOPROTOOPT;
1803 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1804 val = 8;
1805 else if (val > USHRT_MAX)
1806 val = USHRT_MAX;
1807 up->pcrlen = val;
1808 up->pcflag |= UDPLITE_RECV_CC;
1809 break;
1810
1811 default:
1812 err = -ENOPROTOOPT;
1813 break;
1814 }
1815
1816 return err;
1817 }
1818 EXPORT_SYMBOL(udp_lib_setsockopt);
1819
1820 int udp_setsockopt(struct sock *sk, int level, int optname,
1821 char __user *optval, unsigned int optlen)
1822 {
1823 if (level == SOL_UDP || level == SOL_UDPLITE)
1824 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1825 udp_push_pending_frames);
1826 return ip_setsockopt(sk, level, optname, optval, optlen);
1827 }
1828
1829 #ifdef CONFIG_COMPAT
1830 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1831 char __user *optval, unsigned int optlen)
1832 {
1833 if (level == SOL_UDP || level == SOL_UDPLITE)
1834 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1835 udp_push_pending_frames);
1836 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1837 }
1838 #endif
1839
1840 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1841 char __user *optval, int __user *optlen)
1842 {
1843 struct udp_sock *up = udp_sk(sk);
1844 int val, len;
1845
1846 if (get_user(len, optlen))
1847 return -EFAULT;
1848
1849 len = min_t(unsigned int, len, sizeof(int));
1850
1851 if (len < 0)
1852 return -EINVAL;
1853
1854 switch (optname) {
1855 case UDP_CORK:
1856 val = up->corkflag;
1857 break;
1858
1859 case UDP_ENCAP:
1860 val = up->encap_type;
1861 break;
1862
1863 /* The following two cannot be changed on UDP sockets, the return is
1864 * always 0 (which corresponds to the full checksum coverage of UDP). */
1865 case UDPLITE_SEND_CSCOV:
1866 val = up->pcslen;
1867 break;
1868
1869 case UDPLITE_RECV_CSCOV:
1870 val = up->pcrlen;
1871 break;
1872
1873 default:
1874 return -ENOPROTOOPT;
1875 }
1876
1877 if (put_user(len, optlen))
1878 return -EFAULT;
1879 if (copy_to_user(optval, &val, len))
1880 return -EFAULT;
1881 return 0;
1882 }
1883 EXPORT_SYMBOL(udp_lib_getsockopt);
1884
1885 int udp_getsockopt(struct sock *sk, int level, int optname,
1886 char __user *optval, int __user *optlen)
1887 {
1888 if (level == SOL_UDP || level == SOL_UDPLITE)
1889 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1890 return ip_getsockopt(sk, level, optname, optval, optlen);
1891 }
1892
1893 #ifdef CONFIG_COMPAT
1894 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1895 char __user *optval, int __user *optlen)
1896 {
1897 if (level == SOL_UDP || level == SOL_UDPLITE)
1898 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1899 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1900 }
1901 #endif
1902 /**
1903 * udp_poll - wait for a UDP event.
1904 * @file - file struct
1905 * @sock - socket
1906 * @wait - poll table
1907 *
1908 * This is same as datagram poll, except for the special case of
1909 * blocking sockets. If application is using a blocking fd
1910 * and a packet with checksum error is in the queue;
1911 * then it could get return from select indicating data available
1912 * but then block when reading it. Add special case code
1913 * to work around these arguably broken applications.
1914 */
1915 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1916 {
1917 unsigned int mask = datagram_poll(file, sock, wait);
1918 struct sock *sk = sock->sk;
1919
1920 /* Check for false positives due to checksum errors */
1921 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1922 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1923 mask &= ~(POLLIN | POLLRDNORM);
1924
1925 return mask;
1926
1927 }
1928 EXPORT_SYMBOL(udp_poll);
1929
1930 struct proto udp_prot = {
1931 .name = "UDP",
1932 .owner = THIS_MODULE,
1933 .close = udp_lib_close,
1934 .connect = ip4_datagram_connect,
1935 .disconnect = udp_disconnect,
1936 .ioctl = udp_ioctl,
1937 .destroy = udp_destroy_sock,
1938 .setsockopt = udp_setsockopt,
1939 .getsockopt = udp_getsockopt,
1940 .sendmsg = udp_sendmsg,
1941 .recvmsg = udp_recvmsg,
1942 .sendpage = udp_sendpage,
1943 .backlog_rcv = __udp_queue_rcv_skb,
1944 .hash = udp_lib_hash,
1945 .unhash = udp_lib_unhash,
1946 .rehash = udp_v4_rehash,
1947 .get_port = udp_v4_get_port,
1948 .memory_allocated = &udp_memory_allocated,
1949 .sysctl_mem = sysctl_udp_mem,
1950 .sysctl_wmem = &sysctl_udp_wmem_min,
1951 .sysctl_rmem = &sysctl_udp_rmem_min,
1952 .obj_size = sizeof(struct udp_sock),
1953 .slab_flags = SLAB_DESTROY_BY_RCU,
1954 .h.udp_table = &udp_table,
1955 #ifdef CONFIG_COMPAT
1956 .compat_setsockopt = compat_udp_setsockopt,
1957 .compat_getsockopt = compat_udp_getsockopt,
1958 #endif
1959 .clear_sk = sk_prot_clear_portaddr_nulls,
1960 };
1961 EXPORT_SYMBOL(udp_prot);
1962
1963 /* ------------------------------------------------------------------------ */
1964 #ifdef CONFIG_PROC_FS
1965
1966 static struct sock *udp_get_first(struct seq_file *seq, int start)
1967 {
1968 struct sock *sk;
1969 struct udp_iter_state *state = seq->private;
1970 struct net *net = seq_file_net(seq);
1971
1972 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1973 ++state->bucket) {
1974 struct hlist_nulls_node *node;
1975 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1976
1977 if (hlist_nulls_empty(&hslot->head))
1978 continue;
1979
1980 spin_lock_bh(&hslot->lock);
1981 sk_nulls_for_each(sk, node, &hslot->head) {
1982 if (!net_eq(sock_net(sk), net))
1983 continue;
1984 if (sk->sk_family == state->family)
1985 goto found;
1986 }
1987 spin_unlock_bh(&hslot->lock);
1988 }
1989 sk = NULL;
1990 found:
1991 return sk;
1992 }
1993
1994 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1995 {
1996 struct udp_iter_state *state = seq->private;
1997 struct net *net = seq_file_net(seq);
1998
1999 do {
2000 sk = sk_nulls_next(sk);
2001 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2002
2003 if (!sk) {
2004 if (state->bucket <= state->udp_table->mask)
2005 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2006 return udp_get_first(seq, state->bucket + 1);
2007 }
2008 return sk;
2009 }
2010
2011 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2012 {
2013 struct sock *sk = udp_get_first(seq, 0);
2014
2015 if (sk)
2016 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2017 --pos;
2018 return pos ? NULL : sk;
2019 }
2020
2021 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2022 {
2023 struct udp_iter_state *state = seq->private;
2024 state->bucket = MAX_UDP_PORTS;
2025
2026 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2027 }
2028
2029 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2030 {
2031 struct sock *sk;
2032
2033 if (v == SEQ_START_TOKEN)
2034 sk = udp_get_idx(seq, 0);
2035 else
2036 sk = udp_get_next(seq, v);
2037
2038 ++*pos;
2039 return sk;
2040 }
2041
2042 static void udp_seq_stop(struct seq_file *seq, void *v)
2043 {
2044 struct udp_iter_state *state = seq->private;
2045
2046 if (state->bucket <= state->udp_table->mask)
2047 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2048 }
2049
2050 int udp_seq_open(struct inode *inode, struct file *file)
2051 {
2052 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
2053 struct udp_iter_state *s;
2054 int err;
2055
2056 err = seq_open_net(inode, file, &afinfo->seq_ops,
2057 sizeof(struct udp_iter_state));
2058 if (err < 0)
2059 return err;
2060
2061 s = ((struct seq_file *)file->private_data)->private;
2062 s->family = afinfo->family;
2063 s->udp_table = afinfo->udp_table;
2064 return err;
2065 }
2066 EXPORT_SYMBOL(udp_seq_open);
2067
2068 /* ------------------------------------------------------------------------ */
2069 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2070 {
2071 struct proc_dir_entry *p;
2072 int rc = 0;
2073
2074 afinfo->seq_ops.start = udp_seq_start;
2075 afinfo->seq_ops.next = udp_seq_next;
2076 afinfo->seq_ops.stop = udp_seq_stop;
2077
2078 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2079 afinfo->seq_fops, afinfo);
2080 if (!p)
2081 rc = -ENOMEM;
2082 return rc;
2083 }
2084 EXPORT_SYMBOL(udp_proc_register);
2085
2086 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2087 {
2088 proc_net_remove(net, afinfo->name);
2089 }
2090 EXPORT_SYMBOL(udp_proc_unregister);
2091
2092 /* ------------------------------------------------------------------------ */
2093 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2094 int bucket, int *len)
2095 {
2096 struct inet_sock *inet = inet_sk(sp);
2097 __be32 dest = inet->inet_daddr;
2098 __be32 src = inet->inet_rcv_saddr;
2099 __u16 destp = ntohs(inet->inet_dport);
2100 __u16 srcp = ntohs(inet->inet_sport);
2101
2102 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2103 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2104 bucket, src, srcp, dest, destp, sp->sk_state,
2105 sk_wmem_alloc_get(sp),
2106 sk_rmem_alloc_get(sp),
2107 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
2108 atomic_read(&sp->sk_refcnt), sp,
2109 atomic_read(&sp->sk_drops), len);
2110 }
2111
2112 int udp4_seq_show(struct seq_file *seq, void *v)
2113 {
2114 if (v == SEQ_START_TOKEN)
2115 seq_printf(seq, "%-127s\n",
2116 " sl local_address rem_address st tx_queue "
2117 "rx_queue tr tm->when retrnsmt uid timeout "
2118 "inode ref pointer drops");
2119 else {
2120 struct udp_iter_state *state = seq->private;
2121 int len;
2122
2123 udp4_format_sock(v, seq, state->bucket, &len);
2124 seq_printf(seq, "%*s\n", 127 - len, "");
2125 }
2126 return 0;
2127 }
2128
2129 static const struct file_operations udp_afinfo_seq_fops = {
2130 .owner = THIS_MODULE,
2131 .open = udp_seq_open,
2132 .read = seq_read,
2133 .llseek = seq_lseek,
2134 .release = seq_release_net
2135 };
2136
2137 /* ------------------------------------------------------------------------ */
2138 static struct udp_seq_afinfo udp4_seq_afinfo = {
2139 .name = "udp",
2140 .family = AF_INET,
2141 .udp_table = &udp_table,
2142 .seq_fops = &udp_afinfo_seq_fops,
2143 .seq_ops = {
2144 .show = udp4_seq_show,
2145 },
2146 };
2147
2148 static int __net_init udp4_proc_init_net(struct net *net)
2149 {
2150 return udp_proc_register(net, &udp4_seq_afinfo);
2151 }
2152
2153 static void __net_exit udp4_proc_exit_net(struct net *net)
2154 {
2155 udp_proc_unregister(net, &udp4_seq_afinfo);
2156 }
2157
2158 static struct pernet_operations udp4_net_ops = {
2159 .init = udp4_proc_init_net,
2160 .exit = udp4_proc_exit_net,
2161 };
2162
2163 int __init udp4_proc_init(void)
2164 {
2165 return register_pernet_subsys(&udp4_net_ops);
2166 }
2167
2168 void udp4_proc_exit(void)
2169 {
2170 unregister_pernet_subsys(&udp4_net_ops);
2171 }
2172 #endif /* CONFIG_PROC_FS */
2173
2174 static __initdata unsigned long uhash_entries;
2175 static int __init set_uhash_entries(char *str)
2176 {
2177 ssize_t ret;
2178
2179 if (!str)
2180 return 0;
2181
2182 ret = kstrtoul(str, 0, &uhash_entries);
2183 if (ret)
2184 return 0;
2185
2186 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2187 uhash_entries = UDP_HTABLE_SIZE_MIN;
2188 return 1;
2189 }
2190 __setup("uhash_entries=", set_uhash_entries);
2191
2192 void __init udp_table_init(struct udp_table *table, const char *name)
2193 {
2194 unsigned int i;
2195
2196 table->hash = alloc_large_system_hash(name,
2197 2 * sizeof(struct udp_hslot),
2198 uhash_entries,
2199 21, /* one slot per 2 MB */
2200 0,
2201 &table->log,
2202 &table->mask,
2203 UDP_HTABLE_SIZE_MIN,
2204 64 * 1024);
2205
2206 table->hash2 = table->hash + (table->mask + 1);
2207 for (i = 0; i <= table->mask; i++) {
2208 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2209 table->hash[i].count = 0;
2210 spin_lock_init(&table->hash[i].lock);
2211 }
2212 for (i = 0; i <= table->mask; i++) {
2213 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2214 table->hash2[i].count = 0;
2215 spin_lock_init(&table->hash2[i].lock);
2216 }
2217 }
2218
2219 void __init udp_init(void)
2220 {
2221 unsigned long limit;
2222
2223 udp_table_init(&udp_table, "UDP");
2224 limit = nr_free_buffer_pages() / 8;
2225 limit = max(limit, 128UL);
2226 sysctl_udp_mem[0] = limit / 4 * 3;
2227 sysctl_udp_mem[1] = limit;
2228 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2229
2230 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2231 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2232 }
2233
2234 int udp4_ufo_send_check(struct sk_buff *skb)
2235 {
2236 const struct iphdr *iph;
2237 struct udphdr *uh;
2238
2239 if (!pskb_may_pull(skb, sizeof(*uh)))
2240 return -EINVAL;
2241
2242 iph = ip_hdr(skb);
2243 uh = udp_hdr(skb);
2244
2245 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2246 IPPROTO_UDP, 0);
2247 skb->csum_start = skb_transport_header(skb) - skb->head;
2248 skb->csum_offset = offsetof(struct udphdr, check);
2249 skb->ip_summed = CHECKSUM_PARTIAL;
2250 return 0;
2251 }
2252
2253 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
2254 netdev_features_t features)
2255 {
2256 struct sk_buff *segs = ERR_PTR(-EINVAL);
2257 unsigned int mss;
2258 int offset;
2259 __wsum csum;
2260
2261 mss = skb_shinfo(skb)->gso_size;
2262 if (unlikely(skb->len <= mss))
2263 goto out;
2264
2265 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2266 /* Packet is from an untrusted source, reset gso_segs. */
2267 int type = skb_shinfo(skb)->gso_type;
2268
2269 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2270 !(type & (SKB_GSO_UDP))))
2271 goto out;
2272
2273 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2274
2275 segs = NULL;
2276 goto out;
2277 }
2278
2279 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2280 * do checksum of UDP packets sent as multiple IP fragments.
2281 */
2282 offset = skb_checksum_start_offset(skb);
2283 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2284 offset += skb->csum_offset;
2285 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2286 skb->ip_summed = CHECKSUM_NONE;
2287
2288 /* Fragment the skb. IP headers of the fragments are updated in
2289 * inet_gso_segment()
2290 */
2291 segs = skb_segment(skb, features);
2292 out:
2293 return segs;
2294 }
2295
Linux kernel - Deliverables
This page took 0.201423 seconds and 5 git commands to generate.