~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/ipv4/udp.c

Version: ~ [ linux-6.0-rc1 ] ~ [ linux-5.19.1 ] ~ [ linux-5.18.17 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.60 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.136 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.210 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.255 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.290 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.325 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/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/netdevice.h>
104 #include <linux/proc_fs.h>
105 #include <linux/seq_file.h>
106 #include <net/net_namespace.h>
107 #include <net/icmp.h>
108 #include <net/inet_hashtables.h>
109 #include <net/route.h>
110 #include <net/checksum.h>
111 #include <net/xfrm.h>
112 #include <trace/events/udp.h>
113 #include <linux/static_key.h>
114 #include <trace/events/skb.h>
115 #include <net/busy_poll.h>
116 #include "udp_impl.h"
117 
118 struct udp_table udp_table __read_mostly;
119 EXPORT_SYMBOL(udp_table);
120 
121 long sysctl_udp_mem[3] __read_mostly;
122 EXPORT_SYMBOL(sysctl_udp_mem);
123 
124 int sysctl_udp_rmem_min __read_mostly;
125 EXPORT_SYMBOL(sysctl_udp_rmem_min);
126 
127 int sysctl_udp_wmem_min __read_mostly;
128 EXPORT_SYMBOL(sysctl_udp_wmem_min);
129 
130 atomic_long_t udp_memory_allocated;
131 EXPORT_SYMBOL(udp_memory_allocated);
132 
133 #define MAX_UDP_PORTS 65536
134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
135 
136 static int udp_lib_lport_inuse(struct net *net, __u16 num,
137                                const struct udp_hslot *hslot,
138                                unsigned long *bitmap,
139                                struct sock *sk,
140                                int (*saddr_comp)(const struct sock *sk1,
141                                                  const struct sock *sk2),
142                                unsigned int log)
143 {
144         struct sock *sk2;
145         struct hlist_nulls_node *node;
146         kuid_t uid = sock_i_uid(sk);
147 
148         sk_nulls_for_each(sk2, node, &hslot->head) {
149                 if (net_eq(sock_net(sk2), net) &&
150                     sk2 != sk &&
151                     (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
152                     (!sk2->sk_reuse || !sk->sk_reuse) &&
153                     (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
154                      sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
155                     (!sk2->sk_reuseport || !sk->sk_reuseport ||
156                      !uid_eq(uid, sock_i_uid(sk2))) &&
157                     saddr_comp(sk, sk2)) {
158                         if (!bitmap)
159                                 return 1;
160                         __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap);
161                 }
162         }
163         return 0;
164 }
165 
166 /*
167  * Note: we still hold spinlock of primary hash chain, so no other writer
168  * can insert/delete a socket with local_port == num
169  */
170 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
171                                 struct udp_hslot *hslot2,
172                                 struct sock *sk,
173                                 int (*saddr_comp)(const struct sock *sk1,
174                                                   const struct sock *sk2))
175 {
176         struct sock *sk2;
177         struct hlist_nulls_node *node;
178         kuid_t uid = sock_i_uid(sk);
179         int res = 0;
180 
181         spin_lock(&hslot2->lock);
182         udp_portaddr_for_each_entry(sk2, node, &hslot2->head) {
183                 if (net_eq(sock_net(sk2), net) &&
184                     sk2 != sk &&
185                     (udp_sk(sk2)->udp_port_hash == num) &&
186                     (!sk2->sk_reuse || !sk->sk_reuse) &&
187                     (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
188                      sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
189                     (!sk2->sk_reuseport || !sk->sk_reuseport ||
190                      !uid_eq(uid, sock_i_uid(sk2))) &&
191                     saddr_comp(sk, sk2)) {
192                         res = 1;
193                         break;
194                 }
195         }
196         spin_unlock(&hslot2->lock);
197         return res;
198 }
199 
200 /**
201  *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
202  *
203  *  @sk:          socket struct in question
204  *  @snum:        port number to look up
205  *  @saddr_comp:  AF-dependent comparison of bound local IP addresses
206  *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
207  *                   with NULL address
208  */
209 int udp_lib_get_port(struct sock *sk, unsigned short snum,
210                      int (*saddr_comp)(const struct sock *sk1,
211                                        const struct sock *sk2),
212                      unsigned int hash2_nulladdr)
213 {
214         struct udp_hslot *hslot, *hslot2;
215         struct udp_table *udptable = sk->sk_prot->h.udp_table;
216         int    error = 1;
217         struct net *net = sock_net(sk);
218 
219         if (!snum) {
220                 int low, high, remaining;
221                 unsigned int rand;
222                 unsigned short first, last;
223                 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
224 
225                 inet_get_local_port_range(net, &low, &high);
226                 remaining = (high - low) + 1;
227 
228                 rand = prandom_u32();
229                 first = reciprocal_scale(rand, remaining) + low;
230                 /*
231                  * force rand to be an odd multiple of UDP_HTABLE_SIZE
232                  */
233                 rand = (rand | 1) * (udptable->mask + 1);
234                 last = first + udptable->mask + 1;
235                 do {
236                         hslot = udp_hashslot(udptable, net, first);
237                         bitmap_zero(bitmap, PORTS_PER_CHAIN);
238                         spin_lock_bh(&hslot->lock);
239                         udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
240                                             saddr_comp, udptable->log);
241 
242                         snum = first;
243                         /*
244                          * Iterate on all possible values of snum for this hash.
245                          * Using steps of an odd multiple of UDP_HTABLE_SIZE
246                          * give us randomization and full range coverage.
247                          */
248                         do {
249                                 if (low <= snum && snum <= high &&
250                                     !test_bit(snum >> udptable->log, bitmap) &&
251                                     !inet_is_local_reserved_port(net, snum))
252                                         goto found;
253                                 snum += rand;
254                         } while (snum != first);
255                         spin_unlock_bh(&hslot->lock);
256                 } while (++first != last);
257                 goto fail;
258         } else {
259                 hslot = udp_hashslot(udptable, net, snum);
260                 spin_lock_bh(&hslot->lock);
261                 if (hslot->count > 10) {
262                         int exist;
263                         unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
264 
265                         slot2          &= udptable->mask;
266                         hash2_nulladdr &= udptable->mask;
267 
268                         hslot2 = udp_hashslot2(udptable, slot2);
269                         if (hslot->count < hslot2->count)
270                                 goto scan_primary_hash;
271 
272                         exist = udp_lib_lport_inuse2(net, snum, hslot2,
273                                                      sk, saddr_comp);
274                         if (!exist && (hash2_nulladdr != slot2)) {
275                                 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
276                                 exist = udp_lib_lport_inuse2(net, snum, hslot2,
277                                                              sk, saddr_comp);
278                         }
279                         if (exist)
280                                 goto fail_unlock;
281                         else
282                                 goto found;
283                 }
284 scan_primary_hash:
285                 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
286                                         saddr_comp, 0))
287                         goto fail_unlock;
288         }
289 found:
290         inet_sk(sk)->inet_num = snum;
291         udp_sk(sk)->udp_port_hash = snum;
292         udp_sk(sk)->udp_portaddr_hash ^= snum;
293         if (sk_unhashed(sk)) {
294                 sk_nulls_add_node_rcu(sk, &hslot->head);
295                 hslot->count++;
296                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
297 
298                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
299                 spin_lock(&hslot2->lock);
300                 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
301                                          &hslot2->head);
302                 hslot2->count++;
303                 spin_unlock(&hslot2->lock);
304         }
305         error = 0;
306 fail_unlock:
307         spin_unlock_bh(&hslot->lock);
308 fail:
309         return error;
310 }
311 EXPORT_SYMBOL(udp_lib_get_port);
312 
313 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
314 {
315         struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
316 
317         return  (!ipv6_only_sock(sk2)  &&
318                  (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
319                    inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
320 }
321 
322 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
323                               unsigned int port)
324 {
325         return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
326 }
327 
328 int udp_v4_get_port(struct sock *sk, unsigned short snum)
329 {
330         unsigned int hash2_nulladdr =
331                 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
332         unsigned int hash2_partial =
333                 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
334 
335         /* precompute partial secondary hash */
336         udp_sk(sk)->udp_portaddr_hash = hash2_partial;
337         return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
338 }
339 
340 static inline int compute_score(struct sock *sk, struct net *net,
341                                 __be32 saddr, unsigned short hnum, __be16 sport,
342                                 __be32 daddr, __be16 dport, int dif)
343 {
344         int score;
345         struct inet_sock *inet;
346 
347         if (!net_eq(sock_net(sk), net) ||
348             udp_sk(sk)->udp_port_hash != hnum ||
349             ipv6_only_sock(sk))
350                 return -1;
351 
352         score = (sk->sk_family == PF_INET) ? 2 : 1;
353         inet = inet_sk(sk);
354 
355         if (inet->inet_rcv_saddr) {
356                 if (inet->inet_rcv_saddr != daddr)
357                         return -1;
358                 score += 4;
359         }
360 
361         if (inet->inet_daddr) {
362                 if (inet->inet_daddr != saddr)
363                         return -1;
364                 score += 4;
365         }
366 
367         if (inet->inet_dport) {
368                 if (inet->inet_dport != sport)
369                         return -1;
370                 score += 4;
371         }
372 
373         if (sk->sk_bound_dev_if) {
374                 if (sk->sk_bound_dev_if != dif)
375                         return -1;
376                 score += 4;
377         }
378 
379         return score;
380 }
381 
382 /*
383  * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
384  */
385 static inline int compute_score2(struct sock *sk, struct net *net,
386                                  __be32 saddr, __be16 sport,
387                                  __be32 daddr, unsigned int hnum, int dif)
388 {
389         int score;
390         struct inet_sock *inet;
391 
392         if (!net_eq(sock_net(sk), net) ||
393             ipv6_only_sock(sk))
394                 return -1;
395 
396         inet = inet_sk(sk);
397 
398         if (inet->inet_rcv_saddr != daddr ||
399             inet->inet_num != hnum)
400                 return -1;
401 
402         score = (sk->sk_family == PF_INET) ? 2 : 1;
403 
404         if (inet->inet_daddr) {
405                 if (inet->inet_daddr != saddr)
406                         return -1;
407                 score += 4;
408         }
409 
410         if (inet->inet_dport) {
411                 if (inet->inet_dport != sport)
412                         return -1;
413                 score += 4;
414         }
415 
416         if (sk->sk_bound_dev_if) {
417                 if (sk->sk_bound_dev_if != dif)
418                         return -1;
419                 score += 4;
420         }
421 
422         return score;
423 }
424 
425 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
426                        const __u16 lport, const __be32 faddr,
427                        const __be16 fport)
428 {
429         static u32 udp_ehash_secret __read_mostly;
430 
431         net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
432 
433         return __inet_ehashfn(laddr, lport, faddr, fport,
434                               udp_ehash_secret + net_hash_mix(net));
435 }
436 
437 /* called with read_rcu_lock() */
438 static struct sock *udp4_lib_lookup2(struct net *net,
439                 __be32 saddr, __be16 sport,
440                 __be32 daddr, unsigned int hnum, int dif,
441                 struct udp_hslot *hslot2, unsigned int slot2)
442 {
443         struct sock *sk, *result;
444         struct hlist_nulls_node *node;
445         int score, badness, matches = 0, reuseport = 0;
446         u32 hash = 0;
447 
448 begin:
449         result = NULL;
450         badness = 0;
451         udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
452                 score = compute_score2(sk, net, saddr, sport,
453                                       daddr, hnum, dif);
454                 if (score > badness) {
455                         result = sk;
456                         badness = score;
457                         reuseport = sk->sk_reuseport;
458                         if (reuseport) {
459                                 hash = udp_ehashfn(net, daddr, hnum,
460                                                    saddr, sport);
461                                 matches = 1;
462                         }
463                 } else if (score == badness && reuseport) {
464                         matches++;
465                         if (reciprocal_scale(hash, matches) == 0)
466                                 result = sk;
467                         hash = next_pseudo_random32(hash);
468                 }
469         }
470         /*
471          * if the nulls value we got at the end of this lookup is
472          * not the expected one, we must restart lookup.
473          * We probably met an item that was moved to another chain.
474          */
475         if (get_nulls_value(node) != slot2)
476                 goto begin;
477         if (result) {
478                 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
479                         result = NULL;
480                 else if (unlikely(compute_score2(result, net, saddr, sport,
481                                   daddr, hnum, dif) < badness)) {
482                         sock_put(result);
483                         goto begin;
484                 }
485         }
486         return result;
487 }
488 
489 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
490  * harder than this. -DaveM
491  */
492 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
493                 __be16 sport, __be32 daddr, __be16 dport,
494                 int dif, struct udp_table *udptable)
495 {
496         struct sock *sk, *result;
497         struct hlist_nulls_node *node;
498         unsigned short hnum = ntohs(dport);
499         unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
500         struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
501         int score, badness, matches = 0, reuseport = 0;
502         u32 hash = 0;
503 
504         rcu_read_lock();
505         if (hslot->count > 10) {
506                 hash2 = udp4_portaddr_hash(net, daddr, hnum);
507                 slot2 = hash2 & udptable->mask;
508                 hslot2 = &udptable->hash2[slot2];
509                 if (hslot->count < hslot2->count)
510                         goto begin;
511 
512                 result = udp4_lib_lookup2(net, saddr, sport,
513                                           daddr, hnum, dif,
514                                           hslot2, slot2);
515                 if (!result) {
516                         hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
517                         slot2 = hash2 & udptable->mask;
518                         hslot2 = &udptable->hash2[slot2];
519                         if (hslot->count < hslot2->count)
520                                 goto begin;
521 
522                         result = udp4_lib_lookup2(net, saddr, sport,
523                                                   htonl(INADDR_ANY), hnum, dif,
524                                                   hslot2, slot2);
525                 }
526                 rcu_read_unlock();
527                 return result;
528         }
529 begin:
530         result = NULL;
531         badness = 0;
532         sk_nulls_for_each_rcu(sk, node, &hslot->head) {
533                 score = compute_score(sk, net, saddr, hnum, sport,
534                                       daddr, dport, dif);
535                 if (score > badness) {
536                         result = sk;
537                         badness = score;
538                         reuseport = sk->sk_reuseport;
539                         if (reuseport) {
540                                 hash = udp_ehashfn(net, daddr, hnum,
541                                                    saddr, sport);
542                                 matches = 1;
543                         }
544                 } else if (score == badness && reuseport) {
545                         matches++;
546                         if (reciprocal_scale(hash, matches) == 0)
547                                 result = sk;
548                         hash = next_pseudo_random32(hash);
549                 }
550         }
551         /*
552          * if the nulls value we got at the end of this lookup is
553          * not the expected one, we must restart lookup.
554          * We probably met an item that was moved to another chain.
555          */
556         if (get_nulls_value(node) != slot)
557                 goto begin;
558 
559         if (result) {
560                 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
561                         result = NULL;
562                 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
563                                   daddr, dport, dif) < badness)) {
564                         sock_put(result);
565                         goto begin;
566                 }
567         }
568         rcu_read_unlock();
569         return result;
570 }
571 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
572 
573 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
574                                                  __be16 sport, __be16 dport,
575                                                  struct udp_table *udptable)
576 {
577         const struct iphdr *iph = ip_hdr(skb);
578 
579         return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
580                                  iph->daddr, dport, inet_iif(skb),
581                                  udptable);
582 }
583 
584 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
585                              __be32 daddr, __be16 dport, int dif)
586 {
587         return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
588 }
589 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
590 
591 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
592                                        __be16 loc_port, __be32 loc_addr,
593                                        __be16 rmt_port, __be32 rmt_addr,
594                                        int dif, unsigned short hnum)
595 {
596         struct inet_sock *inet = inet_sk(sk);
597 
598         if (!net_eq(sock_net(sk), net) ||
599             udp_sk(sk)->udp_port_hash != hnum ||
600             (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
601             (inet->inet_dport != rmt_port && inet->inet_dport) ||
602             (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
603             ipv6_only_sock(sk) ||
604             (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
605                 return false;
606         if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
607                 return false;
608         return true;
609 }
610 
611 /*
612  * This routine is called by the ICMP module when it gets some
613  * sort of error condition.  If err < 0 then the socket should
614  * be closed and the error returned to the user.  If err > 0
615  * it's just the icmp type << 8 | icmp code.
616  * Header points to the ip header of the error packet. We move
617  * on past this. Then (as it used to claim before adjustment)
618  * header points to the first 8 bytes of the udp header.  We need
619  * to find the appropriate port.
620  */
621 
622 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
623 {
624         struct inet_sock *inet;
625         const struct iphdr *iph = (const struct iphdr *)skb->data;
626         struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
627         const int type = icmp_hdr(skb)->type;
628         const int code = icmp_hdr(skb)->code;
629         struct sock *sk;
630         int harderr;
631         int err;
632         struct net *net = dev_net(skb->dev);
633 
634         sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
635                         iph->saddr, uh->source, skb->dev->ifindex, udptable);
636         if (!sk) {
637                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
638                 return; /* No socket for error */
639         }
640 
641         err = 0;
642         harderr = 0;
643         inet = inet_sk(sk);
644 
645         switch (type) {
646         default:
647         case ICMP_TIME_EXCEEDED:
648                 err = EHOSTUNREACH;
649                 break;
650         case ICMP_SOURCE_QUENCH:
651                 goto out;
652         case ICMP_PARAMETERPROB:
653                 err = EPROTO;
654                 harderr = 1;
655                 break;
656         case ICMP_DEST_UNREACH:
657                 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
658                         ipv4_sk_update_pmtu(skb, sk, info);
659                         if (inet->pmtudisc != IP_PMTUDISC_DONT) {
660                                 err = EMSGSIZE;
661                                 harderr = 1;
662                                 break;
663                         }
664                         goto out;
665                 }
666                 err = EHOSTUNREACH;
667                 if (code <= NR_ICMP_UNREACH) {
668                         harderr = icmp_err_convert[code].fatal;
669                         err = icmp_err_convert[code].errno;
670                 }
671                 break;
672         case ICMP_REDIRECT:
673                 ipv4_sk_redirect(skb, sk);
674                 goto out;
675         }
676 
677         /*
678          *      RFC1122: OK.  Passes ICMP errors back to application, as per
679          *      4.1.3.3.
680          */
681         if (!inet->recverr) {
682                 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
683                         goto out;
684         } else
685                 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
686 
687         sk->sk_err = err;
688         sk->sk_error_report(sk);
689 out:
690         sock_put(sk);
691 }
692 
693 void udp_err(struct sk_buff *skb, u32 info)
694 {
695         __udp4_lib_err(skb, info, &udp_table);
696 }
697 
698 /*
699  * Throw away all pending data and cancel the corking. Socket is locked.
700  */
701 void udp_flush_pending_frames(struct sock *sk)
702 {
703         struct udp_sock *up = udp_sk(sk);
704 
705         if (up->pending) {
706                 up->len = 0;
707                 up->pending = 0;
708                 ip_flush_pending_frames(sk);
709         }
710 }
711 EXPORT_SYMBOL(udp_flush_pending_frames);
712 
713 /**
714  *      udp4_hwcsum  -  handle outgoing HW checksumming
715  *      @skb:   sk_buff containing the filled-in UDP header
716  *              (checksum field must be zeroed out)
717  *      @src:   source IP address
718  *      @dst:   destination IP address
719  */
720 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
721 {
722         struct udphdr *uh = udp_hdr(skb);
723         int offset = skb_transport_offset(skb);
724         int len = skb->len - offset;
725         int hlen = len;
726         __wsum csum = 0;
727 
728         if (!skb_has_frag_list(skb)) {
729                 /*
730                  * Only one fragment on the socket.
731                  */
732                 skb->csum_start = skb_transport_header(skb) - skb->head;
733                 skb->csum_offset = offsetof(struct udphdr, check);
734                 uh->check = ~csum_tcpudp_magic(src, dst, len,
735                                                IPPROTO_UDP, 0);
736         } else {
737                 struct sk_buff *frags;
738 
739                 /*
740                  * HW-checksum won't work as there are two or more
741                  * fragments on the socket so that all csums of sk_buffs
742                  * should be together
743                  */
744                 skb_walk_frags(skb, frags) {
745                         csum = csum_add(csum, frags->csum);
746                         hlen -= frags->len;
747                 }
748 
749                 csum = skb_checksum(skb, offset, hlen, csum);
750                 skb->ip_summed = CHECKSUM_NONE;
751 
752                 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
753                 if (uh->check == 0)
754                         uh->check = CSUM_MANGLED_0;
755         }
756 }
757 EXPORT_SYMBOL_GPL(udp4_hwcsum);
758 
759 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
760  * for the simple case like when setting the checksum for a UDP tunnel.
761  */
762 void udp_set_csum(bool nocheck, struct sk_buff *skb,
763                   __be32 saddr, __be32 daddr, int len)
764 {
765         struct udphdr *uh = udp_hdr(skb);
766 
767         if (nocheck)
768                 uh->check = 0;
769         else if (skb_is_gso(skb))
770                 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
771         else if (skb_dst(skb) && skb_dst(skb)->dev &&
772                  (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) {
773 
774                 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
775 
776                 skb->ip_summed = CHECKSUM_PARTIAL;
777                 skb->csum_start = skb_transport_header(skb) - skb->head;
778                 skb->csum_offset = offsetof(struct udphdr, check);
779                 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
780         } else {
781                 __wsum csum;
782 
783                 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
784 
785                 uh->check = 0;
786                 csum = skb_checksum(skb, 0, len, 0);
787                 uh->check = udp_v4_check(len, saddr, daddr, csum);
788                 if (uh->check == 0)
789                         uh->check = CSUM_MANGLED_0;
790 
791                 skb->ip_summed = CHECKSUM_UNNECESSARY;
792         }
793 }
794 EXPORT_SYMBOL(udp_set_csum);
795 
796 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
797 {
798         struct sock *sk = skb->sk;
799         struct inet_sock *inet = inet_sk(sk);
800         struct udphdr *uh;
801         int err = 0;
802         int is_udplite = IS_UDPLITE(sk);
803         int offset = skb_transport_offset(skb);
804         int len = skb->len - offset;
805         __wsum csum = 0;
806 
807         /*
808          * Create a UDP header
809          */
810         uh = udp_hdr(skb);
811         uh->source = inet->inet_sport;
812         uh->dest = fl4->fl4_dport;
813         uh->len = htons(len);
814         uh->check = 0;
815 
816         if (is_udplite)                                  /*     UDP-Lite      */
817                 csum = udplite_csum(skb);
818 
819         else if (sk->sk_no_check_tx) {   /* UDP csum disabled */
820 
821                 skb->ip_summed = CHECKSUM_NONE;
822                 goto send;
823 
824         } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
825 
826                 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
827                 goto send;
828 
829         } else
830                 csum = udp_csum(skb);
831 
832         /* add protocol-dependent pseudo-header */
833         uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
834                                       sk->sk_protocol, csum);
835         if (uh->check == 0)
836                 uh->check = CSUM_MANGLED_0;
837 
838 send:
839         err = ip_send_skb(sock_net(sk), skb);
840         if (err) {
841                 if (err == -ENOBUFS && !inet->recverr) {
842                         UDP_INC_STATS_USER(sock_net(sk),
843                                            UDP_MIB_SNDBUFERRORS, is_udplite);
844                         err = 0;
845                 }
846         } else
847                 UDP_INC_STATS_USER(sock_net(sk),
848                                    UDP_MIB_OUTDATAGRAMS, is_udplite);
849         return err;
850 }
851 
852 /*
853  * Push out all pending data as one UDP datagram. Socket is locked.
854  */
855 int udp_push_pending_frames(struct sock *sk)
856 {
857         struct udp_sock  *up = udp_sk(sk);
858         struct inet_sock *inet = inet_sk(sk);
859         struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
860         struct sk_buff *skb;
861         int err = 0;
862 
863         skb = ip_finish_skb(sk, fl4);
864         if (!skb)
865                 goto out;
866 
867         err = udp_send_skb(skb, fl4);
868 
869 out:
870         up->len = 0;
871         up->pending = 0;
872         return err;
873 }
874 EXPORT_SYMBOL(udp_push_pending_frames);
875 
876 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
877 {
878         struct inet_sock *inet = inet_sk(sk);
879         struct udp_sock *up = udp_sk(sk);
880         struct flowi4 fl4_stack;
881         struct flowi4 *fl4;
882         int ulen = len;
883         struct ipcm_cookie ipc;
884         struct rtable *rt = NULL;
885         int free = 0;
886         int connected = 0;
887         __be32 daddr, faddr, saddr;
888         __be16 dport;
889         u8  tos;
890         int err, is_udplite = IS_UDPLITE(sk);
891         int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
892         int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
893         struct sk_buff *skb;
894         struct ip_options_data opt_copy;
895 
896         if (len > 0xFFFF)
897                 return -EMSGSIZE;
898 
899         /*
900          *      Check the flags.
901          */
902 
903         if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
904                 return -EOPNOTSUPP;
905 
906         ipc.opt = NULL;
907         ipc.tx_flags = 0;
908         ipc.ttl = 0;
909         ipc.tos = -1;
910 
911         getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
912 
913         fl4 = &inet->cork.fl.u.ip4;
914         if (up->pending) {
915                 /*
916                  * There are pending frames.
917                  * The socket lock must be held while it's corked.
918                  */
919                 lock_sock(sk);
920                 if (likely(up->pending)) {
921                         if (unlikely(up->pending != AF_INET)) {
922                                 release_sock(sk);
923                                 return -EINVAL;
924                         }
925                         goto do_append_data;
926                 }
927                 release_sock(sk);
928         }
929         ulen += sizeof(struct udphdr);
930 
931         /*
932          *      Get and verify the address.
933          */
934         if (msg->msg_name) {
935                 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
936                 if (msg->msg_namelen < sizeof(*usin))
937                         return -EINVAL;
938                 if (usin->sin_family != AF_INET) {
939                         if (usin->sin_family != AF_UNSPEC)
940                                 return -EAFNOSUPPORT;
941                 }
942 
943                 daddr = usin->sin_addr.s_addr;
944                 dport = usin->sin_port;
945                 if (dport == 0)
946                         return -EINVAL;
947         } else {
948                 if (sk->sk_state != TCP_ESTABLISHED)
949                         return -EDESTADDRREQ;
950                 daddr = inet->inet_daddr;
951                 dport = inet->inet_dport;
952                 /* Open fast path for connected socket.
953                    Route will not be used, if at least one option is set.
954                  */
955                 connected = 1;
956         }
957         ipc.addr = inet->inet_saddr;
958 
959         ipc.oif = sk->sk_bound_dev_if;
960 
961         sock_tx_timestamp(sk, &ipc.tx_flags);
962 
963         if (msg->msg_controllen) {
964                 err = ip_cmsg_send(sock_net(sk), msg, &ipc,
965                                    sk->sk_family == AF_INET6);
966                 if (err)
967                         return err;
968                 if (ipc.opt)
969                         free = 1;
970                 connected = 0;
971         }
972         if (!ipc.opt) {
973                 struct ip_options_rcu *inet_opt;
974 
975                 rcu_read_lock();
976                 inet_opt = rcu_dereference(inet->inet_opt);
977                 if (inet_opt) {
978                         memcpy(&opt_copy, inet_opt,
979                                sizeof(*inet_opt) + inet_opt->opt.optlen);
980                         ipc.opt = &opt_copy.opt;
981                 }
982                 rcu_read_unlock();
983         }
984 
985         saddr = ipc.addr;
986         ipc.addr = faddr = daddr;
987 
988         if (ipc.opt && ipc.opt->opt.srr) {
989                 if (!daddr)
990                         return -EINVAL;
991                 faddr = ipc.opt->opt.faddr;
992                 connected = 0;
993         }
994         tos = get_rttos(&ipc, inet);
995         if (sock_flag(sk, SOCK_LOCALROUTE) ||
996             (msg->msg_flags & MSG_DONTROUTE) ||
997             (ipc.opt && ipc.opt->opt.is_strictroute)) {
998                 tos |= RTO_ONLINK;
999                 connected = 0;
1000         }
1001 
1002         if (ipv4_is_multicast(daddr)) {
1003                 if (!ipc.oif)
1004                         ipc.oif = inet->mc_index;
1005                 if (!saddr)
1006                         saddr = inet->mc_addr;
1007                 connected = 0;
1008         } else if (!ipc.oif)
1009                 ipc.oif = inet->uc_index;
1010 
1011         if (connected)
1012                 rt = (struct rtable *)sk_dst_check(sk, 0);
1013 
1014         if (!rt) {
1015                 struct net *net = sock_net(sk);
1016                 __u8 flow_flags = inet_sk_flowi_flags(sk);
1017 
1018                 fl4 = &fl4_stack;
1019 
1020                 /* unconnected socket. If output device is enslaved to a VRF
1021                  * device lookup source address from VRF table. This mimics
1022                  * behavior of ip_route_connect{_init}.
1023                  */
1024                 if (netif_index_is_vrf(net, ipc.oif)) {
1025                         flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1026                                            RT_SCOPE_UNIVERSE, sk->sk_protocol,
1027                                            (flow_flags | FLOWI_FLAG_VRFSRC |
1028                                             FLOWI_FLAG_SKIP_NH_OIF),
1029                                            faddr, saddr, dport,
1030                                            inet->inet_sport);
1031 
1032                         rt = ip_route_output_flow(net, fl4, sk);
1033                         if (!IS_ERR(rt)) {
1034                                 saddr = fl4->saddr;
1035                                 ip_rt_put(rt);
1036                         }
1037                 }
1038 
1039                 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1040                                    RT_SCOPE_UNIVERSE, sk->sk_protocol,
1041                                    flow_flags,
1042                                    faddr, saddr, dport, inet->inet_sport);
1043 
1044                 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1045                 rt = ip_route_output_flow(net, fl4, sk);
1046                 if (IS_ERR(rt)) {
1047                         err = PTR_ERR(rt);
1048                         rt = NULL;
1049                         if (err == -ENETUNREACH)
1050                                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1051                         goto out;
1052                 }
1053 
1054                 err = -EACCES;
1055                 if ((rt->rt_flags & RTCF_BROADCAST) &&
1056                     !sock_flag(sk, SOCK_BROADCAST))
1057                         goto out;
1058                 if (connected)
1059                         sk_dst_set(sk, dst_clone(&rt->dst));
1060         }
1061 
1062         if (msg->msg_flags&MSG_CONFIRM)
1063                 goto do_confirm;
1064 back_from_confirm:
1065 
1066         saddr = fl4->saddr;
1067         if (!ipc.addr)
1068                 daddr = ipc.addr = fl4->daddr;
1069 
1070         /* Lockless fast path for the non-corking case. */
1071         if (!corkreq) {
1072                 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1073                                   sizeof(struct udphdr), &ipc, &rt,
1074                                   msg->msg_flags);
1075                 err = PTR_ERR(skb);
1076                 if (!IS_ERR_OR_NULL(skb))
1077                         err = udp_send_skb(skb, fl4);
1078                 goto out;
1079         }
1080 
1081         lock_sock(sk);
1082         if (unlikely(up->pending)) {
1083                 /* The socket is already corked while preparing it. */
1084                 /* ... which is an evident application bug. --ANK */
1085                 release_sock(sk);
1086 
1087                 net_dbg_ratelimited("cork app bug 2\n");
1088                 err = -EINVAL;
1089                 goto out;
1090         }
1091         /*
1092          *      Now cork the socket to pend data.
1093          */
1094         fl4 = &inet->cork.fl.u.ip4;
1095         fl4->daddr = daddr;
1096         fl4->saddr = saddr;
1097         fl4->fl4_dport = dport;
1098         fl4->fl4_sport = inet->inet_sport;
1099         up->pending = AF_INET;
1100 
1101 do_append_data:
1102         up->len += ulen;
1103         err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1104                              sizeof(struct udphdr), &ipc, &rt,
1105                              corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1106         if (err)
1107                 udp_flush_pending_frames(sk);
1108         else if (!corkreq)
1109                 err = udp_push_pending_frames(sk);
1110         else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1111                 up->pending = 0;
1112         release_sock(sk);
1113 
1114 out:
1115         ip_rt_put(rt);
1116         if (free)
1117                 kfree(ipc.opt);
1118         if (!err)
1119                 return len;
1120         /*
1121          * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1122          * ENOBUFS might not be good (it's not tunable per se), but otherwise
1123          * we don't have a good statistic (IpOutDiscards but it can be too many
1124          * things).  We could add another new stat but at least for now that
1125          * seems like overkill.
1126          */
1127         if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1128                 UDP_INC_STATS_USER(sock_net(sk),
1129                                 UDP_MIB_SNDBUFERRORS, is_udplite);
1130         }
1131         return err;
1132 
1133 do_confirm:
1134         dst_confirm(&rt->dst);
1135         if (!(msg->msg_flags&MSG_PROBE) || len)
1136                 goto back_from_confirm;
1137         err = 0;
1138         goto out;
1139 }
1140 EXPORT_SYMBOL(udp_sendmsg);
1141 
1142 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1143                  size_t size, int flags)
1144 {
1145         struct inet_sock *inet = inet_sk(sk);
1146         struct udp_sock *up = udp_sk(sk);
1147         int ret;
1148 
1149         if (flags & MSG_SENDPAGE_NOTLAST)
1150                 flags |= MSG_MORE;
1151 
1152         if (!up->pending) {
1153                 struct msghdr msg = {   .msg_flags = flags|MSG_MORE };
1154 
1155                 /* Call udp_sendmsg to specify destination address which
1156                  * sendpage interface can't pass.
1157                  * This will succeed only when the socket is connected.
1158                  */
1159                 ret = udp_sendmsg(sk, &msg, 0);
1160                 if (ret < 0)
1161                         return ret;
1162         }
1163 
1164         lock_sock(sk);
1165 
1166         if (unlikely(!up->pending)) {
1167                 release_sock(sk);
1168 
1169                 net_dbg_ratelimited("udp cork app bug 3\n");
1170                 return -EINVAL;
1171         }
1172 
1173         ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1174                              page, offset, size, flags);
1175         if (ret == -EOPNOTSUPP) {
1176                 release_sock(sk);
1177                 return sock_no_sendpage(sk->sk_socket, page, offset,
1178                                         size, flags);
1179         }
1180         if (ret < 0) {
1181                 udp_flush_pending_frames(sk);
1182                 goto out;
1183         }
1184 
1185         up->len += size;
1186         if (!(up->corkflag || (flags&MSG_MORE)))
1187                 ret = udp_push_pending_frames(sk);
1188         if (!ret)
1189                 ret = size;
1190 out:
1191         release_sock(sk);
1192         return ret;
1193 }
1194 
1195 /**
1196  *      first_packet_length     - return length of first packet in receive queue
1197  *      @sk: socket
1198  *
1199  *      Drops all bad checksum frames, until a valid one is found.
1200  *      Returns the length of found skb, or 0 if none is found.
1201  */
1202 static unsigned int first_packet_length(struct sock *sk)
1203 {
1204         struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1205         struct sk_buff *skb;
1206         unsigned int res;
1207 
1208         __skb_queue_head_init(&list_kill);
1209 
1210         spin_lock_bh(&rcvq->lock);
1211         while ((skb = skb_peek(rcvq)) != NULL &&
1212                 udp_lib_checksum_complete(skb)) {
1213                 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1214                                  IS_UDPLITE(sk));
1215                 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1216                                  IS_UDPLITE(sk));
1217                 atomic_inc(&sk->sk_drops);
1218                 __skb_unlink(skb, rcvq);
1219                 __skb_queue_tail(&list_kill, skb);
1220         }
1221         res = skb ? skb->len : 0;
1222         spin_unlock_bh(&rcvq->lock);
1223 
1224         if (!skb_queue_empty(&list_kill)) {
1225                 bool slow = lock_sock_fast(sk);
1226 
1227                 __skb_queue_purge(&list_kill);
1228                 sk_mem_reclaim_partial(sk);
1229                 unlock_sock_fast(sk, slow);
1230         }
1231         return res;
1232 }
1233 
1234 /*
1235  *      IOCTL requests applicable to the UDP protocol
1236  */
1237 
1238 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1239 {
1240         switch (cmd) {
1241         case SIOCOUTQ:
1242         {
1243                 int amount = sk_wmem_alloc_get(sk);
1244 
1245                 return put_user(amount, (int __user *)arg);
1246         }
1247 
1248         case SIOCINQ:
1249         {
1250                 unsigned int amount = first_packet_length(sk);
1251 
1252                 if (amount)
1253                         /*
1254                          * We will only return the amount
1255                          * of this packet since that is all
1256                          * that will be read.
1257                          */
1258                         amount -= sizeof(struct udphdr);
1259 
1260                 return put_user(amount, (int __user *)arg);
1261         }
1262 
1263         default:
1264                 return -ENOIOCTLCMD;
1265         }
1266 
1267         return 0;
1268 }
1269 EXPORT_SYMBOL(udp_ioctl);
1270 
1271 /*
1272  *      This should be easy, if there is something there we
1273  *      return it, otherwise we block.
1274  */
1275 
1276 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1277                 int flags, int *addr_len)
1278 {
1279         struct inet_sock *inet = inet_sk(sk);
1280         DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1281         struct sk_buff *skb;
1282         unsigned int ulen, copied;
1283         int peeked, off = 0;
1284         int err;
1285         int is_udplite = IS_UDPLITE(sk);
1286         bool slow;
1287 
1288         if (flags & MSG_ERRQUEUE)
1289                 return ip_recv_error(sk, msg, len, addr_len);
1290 
1291 try_again:
1292         skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1293                                   &peeked, &off, &err);
1294         if (!skb)
1295                 goto out;
1296         if (ccs_socket_post_recvmsg_permission(sk, skb, flags)) {
1297                 err = -EAGAIN; /* Hope less harmful than -EPERM. */
1298                 goto out;
1299         }
1300 
1301         ulen = skb->len - sizeof(struct udphdr);
1302         copied = len;
1303         if (copied > ulen)
1304                 copied = ulen;
1305         else if (copied < ulen)
1306                 msg->msg_flags |= MSG_TRUNC;
1307 
1308         /*
1309          * If checksum is needed at all, try to do it while copying the
1310          * data.  If the data is truncated, or if we only want a partial
1311          * coverage checksum (UDP-Lite), do it before the copy.
1312          */
1313 
1314         if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1315                 if (udp_lib_checksum_complete(skb))
1316                         goto csum_copy_err;
1317         }
1318 
1319         if (skb_csum_unnecessary(skb))
1320                 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
1321                                             msg, copied);
1322         else {
1323                 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr),
1324                                                      msg);
1325 
1326                 if (err == -EINVAL)
1327                         goto csum_copy_err;
1328         }
1329 
1330         if (unlikely(err)) {
1331                 trace_kfree_skb(skb, udp_recvmsg);
1332                 if (!peeked) {
1333                         atomic_inc(&sk->sk_drops);
1334                         UDP_INC_STATS_USER(sock_net(sk),
1335                                            UDP_MIB_INERRORS, is_udplite);
1336                 }
1337                 goto out_free;
1338         }
1339 
1340         if (!peeked)
1341                 UDP_INC_STATS_USER(sock_net(sk),
1342                                 UDP_MIB_INDATAGRAMS, is_udplite);
1343 
1344         sock_recv_ts_and_drops(msg, sk, skb);
1345 
1346         /* Copy the address. */
1347         if (sin) {
1348                 sin->sin_family = AF_INET;
1349                 sin->sin_port = udp_hdr(skb)->source;
1350                 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1351                 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1352                 *addr_len = sizeof(*sin);
1353         }
1354         if (inet->cmsg_flags)
1355                 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr));
1356 
1357         err = copied;
1358         if (flags & MSG_TRUNC)
1359                 err = ulen;
1360 
1361 out_free:
1362         skb_free_datagram_locked(sk, skb);
1363 out:
1364         return err;
1365 
1366 csum_copy_err:
1367         slow = lock_sock_fast(sk);
1368         if (!skb_kill_datagram(sk, skb, flags)) {
1369                 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1370                 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1371         }
1372         unlock_sock_fast(sk, slow);
1373 
1374         /* starting over for a new packet, but check if we need to yield */
1375         cond_resched();
1376         msg->msg_flags &= ~MSG_TRUNC;
1377         goto try_again;
1378 }
1379 
1380 int udp_disconnect(struct sock *sk, int flags)
1381 {
1382         struct inet_sock *inet = inet_sk(sk);
1383         /*
1384          *      1003.1g - break association.
1385          */
1386 
1387         sk->sk_state = TCP_CLOSE;
1388         inet->inet_daddr = 0;
1389         inet->inet_dport = 0;
1390         sock_rps_reset_rxhash(sk);
1391         sk->sk_bound_dev_if = 0;
1392         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1393                 inet_reset_saddr(sk);
1394 
1395         if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1396                 sk->sk_prot->unhash(sk);
1397                 inet->inet_sport = 0;
1398         }
1399         sk_dst_reset(sk);
1400         return 0;
1401 }
1402 EXPORT_SYMBOL(udp_disconnect);
1403 
1404 void udp_lib_unhash(struct sock *sk)
1405 {
1406         if (sk_hashed(sk)) {
1407                 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1408                 struct udp_hslot *hslot, *hslot2;
1409 
1410                 hslot  = udp_hashslot(udptable, sock_net(sk),
1411                                       udp_sk(sk)->udp_port_hash);
1412                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1413 
1414                 spin_lock_bh(&hslot->lock);
1415                 if (sk_nulls_del_node_init_rcu(sk)) {
1416                         hslot->count--;
1417                         inet_sk(sk)->inet_num = 0;
1418                         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1419 
1420                         spin_lock(&hslot2->lock);
1421                         hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1422                         hslot2->count--;
1423                         spin_unlock(&hslot2->lock);
1424                 }
1425                 spin_unlock_bh(&hslot->lock);
1426         }
1427 }
1428 EXPORT_SYMBOL(udp_lib_unhash);
1429 
1430 /*
1431  * inet_rcv_saddr was changed, we must rehash secondary hash
1432  */
1433 void udp_lib_rehash(struct sock *sk, u16 newhash)
1434 {
1435         if (sk_hashed(sk)) {
1436                 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1437                 struct udp_hslot *hslot, *hslot2, *nhslot2;
1438 
1439                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1440                 nhslot2 = udp_hashslot2(udptable, newhash);
1441                 udp_sk(sk)->udp_portaddr_hash = newhash;
1442                 if (hslot2 != nhslot2) {
1443                         hslot = udp_hashslot(udptable, sock_net(sk),
1444                                              udp_sk(sk)->udp_port_hash);
1445                         /* we must lock primary chain too */
1446                         spin_lock_bh(&hslot->lock);
1447 
1448                         spin_lock(&hslot2->lock);
1449                         hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1450                         hslot2->count--;
1451                         spin_unlock(&hslot2->lock);
1452 
1453                         spin_lock(&nhslot2->lock);
1454                         hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1455                                                  &nhslot2->head);
1456                         nhslot2->count++;
1457                         spin_unlock(&nhslot2->lock);
1458 
1459                         spin_unlock_bh(&hslot->lock);
1460                 }
1461         }
1462 }
1463 EXPORT_SYMBOL(udp_lib_rehash);
1464 
1465 static void udp_v4_rehash(struct sock *sk)
1466 {
1467         u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1468                                           inet_sk(sk)->inet_rcv_saddr,
1469                                           inet_sk(sk)->inet_num);
1470         udp_lib_rehash(sk, new_hash);
1471 }
1472 
1473 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1474 {
1475         int rc;
1476 
1477         if (inet_sk(sk)->inet_daddr) {
1478                 sock_rps_save_rxhash(sk, skb);
1479                 sk_mark_napi_id(sk, skb);
1480                 sk_incoming_cpu_update(sk);
1481         }
1482 
1483         rc = sock_queue_rcv_skb(sk, skb);
1484         if (rc < 0) {
1485                 int is_udplite = IS_UDPLITE(sk);
1486 
1487                 /* Note that an ENOMEM error is charged twice */
1488                 if (rc == -ENOMEM)
1489                         UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1490                                          is_udplite);
1491                 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1492                 kfree_skb(skb);
1493                 trace_udp_fail_queue_rcv_skb(rc, sk);
1494                 return -1;
1495         }
1496 
1497         return 0;
1498 
1499 }
1500 
1501 static struct static_key udp_encap_needed __read_mostly;
1502 void udp_encap_enable(void)
1503 {
1504         if (!static_key_enabled(&udp_encap_needed))
1505                 static_key_slow_inc(&udp_encap_needed);
1506 }
1507 EXPORT_SYMBOL(udp_encap_enable);
1508 
1509 /* returns:
1510  *  -1: error
1511  *   0: success
1512  *  >0: "udp encap" protocol resubmission
1513  *
1514  * Note that in the success and error cases, the skb is assumed to
1515  * have either been requeued or freed.
1516  */
1517 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1518 {
1519         struct udp_sock *up = udp_sk(sk);
1520         int rc;
1521         int is_udplite = IS_UDPLITE(sk);
1522 
1523         /*
1524          *      Charge it to the socket, dropping if the queue is full.
1525          */
1526         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1527                 goto drop;
1528         nf_reset(skb);
1529 
1530         if (static_key_false(&udp_encap_needed) && up->encap_type) {
1531                 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1532 
1533                 /*
1534                  * This is an encapsulation socket so pass the skb to
1535                  * the socket's udp_encap_rcv() hook. Otherwise, just
1536                  * fall through and pass this up the UDP socket.
1537                  * up->encap_rcv() returns the following value:
1538                  * =0 if skb was successfully passed to the encap
1539                  *    handler or was discarded by it.
1540                  * >0 if skb should be passed on to UDP.
1541                  * <0 if skb should be resubmitted as proto -N
1542                  */
1543 
1544                 /* if we're overly short, let UDP handle it */
1545                 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1546                 if (skb->len > sizeof(struct udphdr) && encap_rcv) {
1547                         int ret;
1548 
1549                         /* Verify checksum before giving to encap */
1550                         if (udp_lib_checksum_complete(skb))
1551                                 goto csum_error;
1552 
1553                         ret = encap_rcv(sk, skb);
1554                         if (ret <= 0) {
1555                                 UDP_INC_STATS_BH(sock_net(sk),
1556                                                  UDP_MIB_INDATAGRAMS,
1557                                                  is_udplite);
1558                                 return -ret;
1559                         }
1560                 }
1561 
1562                 /* FALLTHROUGH -- it's a UDP Packet */
1563         }
1564 
1565         /*
1566          *      UDP-Lite specific tests, ignored on UDP sockets
1567          */
1568         if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1569 
1570                 /*
1571                  * MIB statistics other than incrementing the error count are
1572                  * disabled for the following two types of errors: these depend
1573                  * on the application settings, not on the functioning of the
1574                  * protocol stack as such.
1575                  *
1576                  * RFC 3828 here recommends (sec 3.3): "There should also be a
1577                  * way ... to ... at least let the receiving application block
1578                  * delivery of packets with coverage values less than a value
1579                  * provided by the application."
1580                  */
1581                 if (up->pcrlen == 0) {          /* full coverage was set  */
1582                         net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1583                                             UDP_SKB_CB(skb)->cscov, skb->len);
1584                         goto drop;
1585                 }
1586                 /* The next case involves violating the min. coverage requested
1587                  * by the receiver. This is subtle: if receiver wants x and x is
1588                  * greater than the buffersize/MTU then receiver will complain
1589                  * that it wants x while sender emits packets of smaller size y.
1590                  * Therefore the above ...()->partial_cov statement is essential.
1591                  */
1592                 if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1593                         net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1594                                             UDP_SKB_CB(skb)->cscov, up->pcrlen);
1595                         goto drop;
1596                 }
1597         }
1598 
1599         if (rcu_access_pointer(sk->sk_filter) &&
1600             udp_lib_checksum_complete(skb))
1601                 goto csum_error;
1602 
1603         if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1604                 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1605                                  is_udplite);
1606                 goto drop;
1607         }
1608 
1609         rc = 0;
1610 
1611         ipv4_pktinfo_prepare(sk, skb);
1612         bh_lock_sock(sk);
1613         if (!sock_owned_by_user(sk))
1614                 rc = __udp_queue_rcv_skb(sk, skb);
1615         else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1616                 bh_unlock_sock(sk);
1617                 goto drop;
1618         }
1619         bh_unlock_sock(sk);
1620 
1621         return rc;
1622 
1623 csum_error:
1624         UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1625 drop:
1626         UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1627         atomic_inc(&sk->sk_drops);
1628         kfree_skb(skb);
1629         return -1;
1630 }
1631 
1632 static void flush_stack(struct sock **stack, unsigned int count,
1633                         struct sk_buff *skb, unsigned int final)
1634 {
1635         unsigned int i;
1636         struct sk_buff *skb1 = NULL;
1637         struct sock *sk;
1638 
1639         for (i = 0; i < count; i++) {
1640                 sk = stack[i];
1641                 if (likely(!skb1))
1642                         skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1643 
1644                 if (!skb1) {
1645                         atomic_inc(&sk->sk_drops);
1646                         UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1647                                          IS_UDPLITE(sk));
1648                         UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1649                                          IS_UDPLITE(sk));
1650                 }
1651 
1652                 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1653                         skb1 = NULL;
1654 
1655                 sock_put(sk);
1656         }
1657         if (unlikely(skb1))
1658                 kfree_skb(skb1);
1659 }
1660 
1661 /* For TCP sockets, sk_rx_dst is protected by socket lock
1662  * For UDP, we use xchg() to guard against concurrent changes.
1663  */
1664 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1665 {
1666         struct dst_entry *old;
1667 
1668         dst_hold(dst);
1669         old = xchg(&sk->sk_rx_dst, dst);
1670         dst_release(old);
1671 }
1672 
1673 /*
1674  *      Multicasts and broadcasts go to each listener.
1675  *
1676  *      Note: called only from the BH handler context.
1677  */
1678 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1679                                     struct udphdr  *uh,
1680                                     __be32 saddr, __be32 daddr,
1681                                     struct udp_table *udptable,
1682                                     int proto)
1683 {
1684         struct sock *sk, *stack[256 / sizeof(struct sock *)];
1685         struct hlist_nulls_node *node;
1686         unsigned short hnum = ntohs(uh->dest);
1687         struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1688         int dif = skb->dev->ifindex;
1689         unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
1690         unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1691         bool inner_flushed = false;
1692 
1693         if (use_hash2) {
1694                 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1695                             udp_table.mask;
1696                 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1697 start_lookup:
1698                 hslot = &udp_table.hash2[hash2];
1699                 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1700         }
1701 
1702         spin_lock(&hslot->lock);
1703         sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
1704                 if (__udp_is_mcast_sock(net, sk,
1705                                         uh->dest, daddr,
1706                                         uh->source, saddr,
1707                                         dif, hnum)) {
1708                         if (unlikely(count == ARRAY_SIZE(stack))) {
1709                                 flush_stack(stack, count, skb, ~0);
1710                                 inner_flushed = true;
1711                                 count = 0;
1712                         }
1713                         stack[count++] = sk;
1714                         sock_hold(sk);
1715                 }
1716         }
1717 
1718         spin_unlock(&hslot->lock);
1719 
1720         /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1721         if (use_hash2 && hash2 != hash2_any) {
1722                 hash2 = hash2_any;
1723                 goto start_lookup;
1724         }
1725 
1726         /*
1727          * do the slow work with no lock held
1728          */
1729         if (count) {
1730                 flush_stack(stack, count, skb, count - 1);
1731         } else {
1732                 if (!inner_flushed)
1733                         UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
1734                                          proto == IPPROTO_UDPLITE);
1735                 consume_skb(skb);
1736         }
1737         return 0;
1738 }
1739 
1740 /* Initialize UDP checksum. If exited with zero value (success),
1741  * CHECKSUM_UNNECESSARY means, that no more checks are required.
1742  * Otherwise, csum completion requires chacksumming packet body,
1743  * including udp header and folding it to skb->csum.
1744  */
1745 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1746                                  int proto)
1747 {
1748         int err;
1749 
1750         UDP_SKB_CB(skb)->partial_cov = 0;
1751         UDP_SKB_CB(skb)->cscov = skb->len;
1752 
1753         if (proto == IPPROTO_UDPLITE) {
1754                 err = udplite_checksum_init(skb, uh);
1755                 if (err)
1756                         return err;
1757         }
1758 
1759         return skb_checksum_init_zero_check(skb, proto, uh->check,
1760                                             inet_compute_pseudo);
1761 }
1762 
1763 /*
1764  *      All we need to do is get the socket, and then do a checksum.
1765  */
1766 
1767 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1768                    int proto)
1769 {
1770         struct sock *sk;
1771         struct udphdr *uh;
1772         unsigned short ulen;
1773         struct rtable *rt = skb_rtable(skb);
1774         __be32 saddr, daddr;
1775         struct net *net = dev_net(skb->dev);
1776 
1777         /*
1778          *  Validate the packet.
1779          */
1780         if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1781                 goto drop;              /* No space for header. */
1782 
1783         uh   = udp_hdr(skb);
1784         ulen = ntohs(uh->len);
1785         saddr = ip_hdr(skb)->saddr;
1786         daddr = ip_hdr(skb)->daddr;
1787 
1788         if (ulen > skb->len)
1789                 goto short_packet;
1790 
1791         if (proto == IPPROTO_UDP) {
1792                 /* UDP validates ulen. */
1793                 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1794                         goto short_packet;
1795                 uh = udp_hdr(skb);
1796         }
1797 
1798         if (udp4_csum_init(skb, uh, proto))
1799                 goto csum_error;
1800 
1801         sk = skb_steal_sock(skb);
1802         if (sk) {
1803                 struct dst_entry *dst = skb_dst(skb);
1804                 int ret;
1805 
1806                 if (unlikely(sk->sk_rx_dst != dst))
1807                         udp_sk_rx_dst_set(sk, dst);
1808 
1809                 ret = udp_queue_rcv_skb(sk, skb);
1810                 sock_put(sk);
1811                 /* a return value > 0 means to resubmit the input, but
1812                  * it wants the return to be -protocol, or 0
1813                  */
1814                 if (ret > 0)
1815                         return -ret;
1816                 return 0;
1817         }
1818 
1819         if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1820                 return __udp4_lib_mcast_deliver(net, skb, uh,
1821                                                 saddr, daddr, udptable, proto);
1822 
1823         sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1824         if (sk) {
1825                 int ret;
1826 
1827                 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1828                         skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1829                                                  inet_compute_pseudo);
1830 
1831                 ret = udp_queue_rcv_skb(sk, skb);
1832                 sock_put(sk);
1833 
1834                 /* a return value > 0 means to resubmit the input, but
1835                  * it wants the return to be -protocol, or 0
1836                  */
1837                 if (ret > 0)
1838                         return -ret;
1839                 return 0;
1840         }
1841 
1842         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1843                 goto drop;
1844         nf_reset(skb);
1845 
1846         /* No socket. Drop packet silently, if checksum is wrong */
1847         if (udp_lib_checksum_complete(skb))
1848                 goto csum_error;
1849 
1850         UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1851         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1852 
1853         /*
1854          * Hmm.  We got an UDP packet to a port to which we
1855          * don't wanna listen.  Ignore it.
1856          */
1857         kfree_skb(skb);
1858         return 0;
1859 
1860 short_packet:
1861         net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1862                             proto == IPPROTO_UDPLITE ? "Lite" : "",
1863                             &saddr, ntohs(uh->source),
1864                             ulen, skb->len,
1865                             &daddr, ntohs(uh->dest));
1866         goto drop;
1867 
1868 csum_error:
1869         /*
1870          * RFC1122: OK.  Discards the bad packet silently (as far as
1871          * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1872          */
1873         net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1874                             proto == IPPROTO_UDPLITE ? "Lite" : "",
1875                             &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1876                             ulen);
1877         UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1878 drop:
1879         UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1880         kfree_skb(skb);
1881         return 0;
1882 }
1883 
1884 /* We can only early demux multicast if there is a single matching socket.
1885  * If more than one socket found returns NULL
1886  */
1887 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1888                                                   __be16 loc_port, __be32 loc_addr,
1889                                                   __be16 rmt_port, __be32 rmt_addr,
1890                                                   int dif)
1891 {
1892         struct sock *sk, *result;
1893         struct hlist_nulls_node *node;
1894         unsigned short hnum = ntohs(loc_port);
1895         unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1896         struct udp_hslot *hslot = &udp_table.hash[slot];
1897 
1898         /* Do not bother scanning a too big list */
1899         if (hslot->count > 10)
1900                 return NULL;
1901 
1902         rcu_read_lock();
1903 begin:
1904         count = 0;
1905         result = NULL;
1906         sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1907                 if (__udp_is_mcast_sock(net, sk,
1908                                         loc_port, loc_addr,
1909                                         rmt_port, rmt_addr,
1910                                         dif, hnum)) {
1911                         result = sk;
1912                         ++count;
1913                 }
1914         }
1915         /*
1916          * if the nulls value we got at the end of this lookup is
1917          * not the expected one, we must restart lookup.
1918          * We probably met an item that was moved to another chain.
1919          */
1920         if (get_nulls_value(node) != slot)
1921                 goto begin;
1922 
1923         if (result) {
1924                 if (count != 1 ||
1925                     unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1926                         result = NULL;
1927                 else if (unlikely(!__udp_is_mcast_sock(net, result,
1928                                                        loc_port, loc_addr,
1929                                                        rmt_port, rmt_addr,
1930                                                        dif, hnum))) {
1931                         sock_put(result);
1932                         result = NULL;
1933                 }
1934         }
1935         rcu_read_unlock();
1936         return result;
1937 }
1938 
1939 /* For unicast we should only early demux connected sockets or we can
1940  * break forwarding setups.  The chains here can be long so only check
1941  * if the first socket is an exact match and if not move on.
1942  */
1943 static struct sock *__udp4_lib_demux_lookup(struct net *net,
1944                                             __be16 loc_port, __be32 loc_addr,
1945                                             __be16 rmt_port, __be32 rmt_addr,
1946                                             int dif)
1947 {
1948         struct sock *sk, *result;
1949         struct hlist_nulls_node *node;
1950         unsigned short hnum = ntohs(loc_port);
1951         unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1952         unsigned int slot2 = hash2 & udp_table.mask;
1953         struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1954         INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
1955         const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1956 
1957         rcu_read_lock();
1958         result = NULL;
1959         udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
1960                 if (INET_MATCH(sk, net, acookie,
1961                                rmt_addr, loc_addr, ports, dif))
1962                         result = sk;
1963                 /* Only check first socket in chain */
1964                 break;
1965         }
1966 
1967         if (result) {
1968                 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1969                         result = NULL;
1970                 else if (unlikely(!INET_MATCH(sk, net, acookie,
1971                                               rmt_addr, loc_addr,
1972                                               ports, dif))) {
1973                         sock_put(result);
1974                         result = NULL;
1975                 }
1976         }
1977         rcu_read_unlock();
1978         return result;
1979 }
1980 
1981 void udp_v4_early_demux(struct sk_buff *skb)
1982 {
1983         struct net *net = dev_net(skb->dev);
1984         const struct iphdr *iph;
1985         const struct udphdr *uh;
1986         struct sock *sk;
1987         struct dst_entry *dst;
1988         int dif = skb->dev->ifindex;
1989         int ours;
1990 
1991         /* validate the packet */
1992         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1993                 return;
1994 
1995         iph = ip_hdr(skb);
1996         uh = udp_hdr(skb);
1997 
1998         if (skb->pkt_type == PACKET_BROADCAST ||
1999             skb->pkt_type == PACKET_MULTICAST) {
2000                 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2001 
2002                 if (!in_dev)
2003                         return;
2004 
2005                 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2006                                        iph->protocol);
2007                 if (!ours)
2008                         return;
2009                 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2010                                                    uh->source, iph->saddr, dif);
2011         } else if (skb->pkt_type == PACKET_HOST) {
2012                 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2013                                              uh->source, iph->saddr, dif);
2014         } else {
2015                 return;
2016         }
2017 
2018         if (!sk)
2019                 return;
2020 
2021         skb->sk = sk;
2022         skb->destructor = sock_efree;
2023         dst = READ_ONCE(sk->sk_rx_dst);
2024 
2025         if (dst)
2026                 dst = dst_check(dst, 0);
2027         if (dst) {
2028                 /* DST_NOCACHE can not be used without taking a reference */
2029                 if (dst->flags & DST_NOCACHE) {
2030                         if (likely(atomic_inc_not_zero(&dst->__refcnt)))
2031                                 skb_dst_set(skb, dst);
2032                 } else {
2033                         skb_dst_set_noref(skb, dst);
2034                 }
2035         }
2036 }
2037 
2038 int udp_rcv(struct sk_buff *skb)
2039 {
2040         return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2041 }
2042 
2043 void udp_destroy_sock(struct sock *sk)
2044 {
2045         struct udp_sock *up = udp_sk(sk);
2046         bool slow = lock_sock_fast(sk);
2047         udp_flush_pending_frames(sk);
2048         unlock_sock_fast(sk, slow);
2049         if (static_key_false(&udp_encap_needed) && up->encap_type) {
2050                 void (*encap_destroy)(struct sock *sk);
2051                 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2052                 if (encap_destroy)
2053                         encap_destroy(sk);
2054         }
2055 }
2056 
2057 /*
2058  *      Socket option code for UDP
2059  */
2060 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2061                        char __user *optval, unsigned int optlen,
2062                        int (*push_pending_frames)(struct sock *))
2063 {
2064         struct udp_sock *up = udp_sk(sk);
2065         int val, valbool;
2066         int err = 0;
2067         int is_udplite = IS_UDPLITE(sk);
2068 
2069         if (optlen < sizeof(int))
2070                 return -EINVAL;
2071 
2072         if (get_user(val, (int __user *)optval))
2073                 return -EFAULT;
2074 
2075         valbool = val ? 1 : 0;
2076 
2077         switch (optname) {
2078         case UDP_CORK:
2079                 if (val != 0) {
2080                         up->corkflag = 1;
2081                 } else {
2082                         up->corkflag = 0;
2083                         lock_sock(sk);
2084                         push_pending_frames(sk);
2085                         release_sock(sk);
2086                 }
2087                 break;
2088 
2089         case UDP_ENCAP:
2090                 switch (val) {
2091                 case 0:
2092                 case UDP_ENCAP_ESPINUDP:
2093                 case UDP_ENCAP_ESPINUDP_NON_IKE:
2094                         up->encap_rcv = xfrm4_udp_encap_rcv;
2095                         /* FALLTHROUGH */
2096                 case UDP_ENCAP_L2TPINUDP:
2097                         up->encap_type = val;
2098                         udp_encap_enable();
2099                         break;
2100                 default:
2101                         err = -ENOPROTOOPT;
2102                         break;
2103                 }
2104                 break;
2105 
2106         case UDP_NO_CHECK6_TX:
2107                 up->no_check6_tx = valbool;
2108                 break;
2109 
2110         case UDP_NO_CHECK6_RX:
2111                 up->no_check6_rx = valbool;
2112                 break;
2113 
2114         /*
2115          *      UDP-Lite's partial checksum coverage (RFC 3828).
2116          */
2117         /* The sender sets actual checksum coverage length via this option.
2118          * The case coverage > packet length is handled by send module. */
2119         case UDPLITE_SEND_CSCOV:
2120                 if (!is_udplite)         /* Disable the option on UDP sockets */
2121                         return -ENOPROTOOPT;
2122                 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2123                         val = 8;
2124                 else if (val > USHRT_MAX)
2125                         val = USHRT_MAX;
2126                 up->pcslen = val;
2127                 up->pcflag |= UDPLITE_SEND_CC;
2128                 break;
2129 
2130         /* The receiver specifies a minimum checksum coverage value. To make
2131          * sense, this should be set to at least 8 (as done below). If zero is
2132          * used, this again means full checksum coverage.                     */
2133         case UDPLITE_RECV_CSCOV:
2134                 if (!is_udplite)         /* Disable the option on UDP sockets */
2135                         return -ENOPROTOOPT;
2136                 if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2137                         val = 8;
2138                 else if (val > USHRT_MAX)
2139                         val = USHRT_MAX;
2140                 up->pcrlen = val;
2141                 up->pcflag |= UDPLITE_RECV_CC;
2142                 break;
2143 
2144         default:
2145                 err = -ENOPROTOOPT;
2146                 break;
2147         }
2148 
2149         return err;
2150 }
2151 EXPORT_SYMBOL(udp_lib_setsockopt);
2152 
2153 int udp_setsockopt(struct sock *sk, int level, int optname,
2154                    char __user *optval, unsigned int optlen)
2155 {
2156         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2157                 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2158                                           udp_push_pending_frames);
2159         return ip_setsockopt(sk, level, optname, optval, optlen);
2160 }
2161 
2162 #ifdef CONFIG_COMPAT
2163 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2164                           char __user *optval, unsigned int optlen)
2165 {
2166         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2167                 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2168                                           udp_push_pending_frames);
2169         return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2170 }
2171 #endif
2172 
2173 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2174                        char __user *optval, int __user *optlen)
2175 {
2176         struct udp_sock *up = udp_sk(sk);
2177         int val, len;
2178 
2179         if (get_user(len, optlen))
2180                 return -EFAULT;
2181 
2182         len = min_t(unsigned int, len, sizeof(int));
2183 
2184         if (len < 0)
2185                 return -EINVAL;
2186 
2187         switch (optname) {
2188         case UDP_CORK:
2189                 val = up->corkflag;
2190                 break;
2191 
2192         case UDP_ENCAP:
2193                 val = up->encap_type;
2194                 break;
2195 
2196         case UDP_NO_CHECK6_TX:
2197                 val = up->no_check6_tx;
2198                 break;
2199 
2200         case UDP_NO_CHECK6_RX:
2201                 val = up->no_check6_rx;
2202                 break;
2203 
2204         /* The following two cannot be changed on UDP sockets, the return is
2205          * always 0 (which corresponds to the full checksum coverage of UDP). */
2206         case UDPLITE_SEND_CSCOV:
2207                 val = up->pcslen;
2208                 break;
2209 
2210         case UDPLITE_RECV_CSCOV:
2211                 val = up->pcrlen;
2212                 break;
2213 
2214         default:
2215                 return -ENOPROTOOPT;
2216         }
2217 
2218         if (put_user(len, optlen))
2219                 return -EFAULT;
2220         if (copy_to_user(optval, &val, len))
2221                 return -EFAULT;
2222         return 0;
2223 }
2224 EXPORT_SYMBOL(udp_lib_getsockopt);
2225 
2226 int udp_getsockopt(struct sock *sk, int level, int optname,
2227                    char __user *optval, int __user *optlen)
2228 {
2229         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2230                 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2231         return ip_getsockopt(sk, level, optname, optval, optlen);
2232 }
2233 
2234 #ifdef CONFIG_COMPAT
2235 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2236                                  char __user *optval, int __user *optlen)
2237 {
2238         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2239                 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2240         return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2241 }
2242 #endif
2243 /**
2244  *      udp_poll - wait for a UDP event.
2245  *      @file - file struct
2246  *      @sock - socket
2247  *      @wait - poll table
2248  *
2249  *      This is same as datagram poll, except for the special case of
2250  *      blocking sockets. If application is using a blocking fd
2251  *      and a packet with checksum error is in the queue;
2252  *      then it could get return from select indicating data available
2253  *      but then block when reading it. Add special case code
2254  *      to work around these arguably broken applications.
2255  */
2256 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2257 {
2258         unsigned int mask = datagram_poll(file, sock, wait);
2259         struct sock *sk = sock->sk;
2260 
2261         sock_rps_record_flow(sk);
2262 
2263         /* Check for false positives due to checksum errors */
2264         if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2265             !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2266                 mask &= ~(POLLIN | POLLRDNORM);
2267 
2268         return mask;
2269 
2270 }
2271 EXPORT_SYMBOL(udp_poll);
2272 
2273 struct proto udp_prot = {
2274         .name              = "UDP",
2275         .owner             = THIS_MODULE,
2276         .close             = udp_lib_close,
2277         .connect           = ip4_datagram_connect,
2278         .disconnect        = udp_disconnect,
2279         .ioctl             = udp_ioctl,
2280         .destroy           = udp_destroy_sock,
2281         .setsockopt        = udp_setsockopt,
2282         .getsockopt        = udp_getsockopt,
2283         .sendmsg           = udp_sendmsg,
2284         .recvmsg           = udp_recvmsg,
2285         .sendpage          = udp_sendpage,
2286         .backlog_rcv       = __udp_queue_rcv_skb,
2287         .release_cb        = ip4_datagram_release_cb,
2288         .hash              = udp_lib_hash,
2289         .unhash            = udp_lib_unhash,
2290         .rehash            = udp_v4_rehash,
2291         .get_port          = udp_v4_get_port,
2292         .memory_allocated  = &udp_memory_allocated,
2293         .sysctl_mem        = sysctl_udp_mem,
2294         .sysctl_wmem       = &sysctl_udp_wmem_min,
2295         .sysctl_rmem       = &sysctl_udp_rmem_min,
2296         .obj_size          = sizeof(struct udp_sock),
2297         .slab_flags        = SLAB_DESTROY_BY_RCU,
2298         .h.udp_table       = &udp_table,
2299 #ifdef CONFIG_COMPAT
2300         .compat_setsockopt = compat_udp_setsockopt,
2301         .compat_getsockopt = compat_udp_getsockopt,
2302 #endif
2303         .clear_sk          = sk_prot_clear_portaddr_nulls,
2304 };
2305 EXPORT_SYMBOL(udp_prot);
2306 
2307 /* ------------------------------------------------------------------------ */
2308 #ifdef CONFIG_PROC_FS
2309 
2310 static struct sock *udp_get_first(struct seq_file *seq, int start)
2311 {
2312         struct sock *sk;
2313         struct udp_iter_state *state = seq->private;
2314         struct net *net = seq_file_net(seq);
2315 
2316         for (state->bucket = start; state->bucket <= state->udp_table->mask;
2317              ++state->bucket) {
2318                 struct hlist_nulls_node *node;
2319                 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2320 
2321                 if (hlist_nulls_empty(&hslot->head))
2322                         continue;
2323 
2324                 spin_lock_bh(&hslot->lock);
2325                 sk_nulls_for_each(sk, node, &hslot->head) {
2326                         if (!net_eq(sock_net(sk), net))
2327                                 continue;
2328                         if (sk->sk_family == state->family)
2329                                 goto found;
2330                 }
2331                 spin_unlock_bh(&hslot->lock);
2332         }
2333         sk = NULL;
2334 found:
2335         return sk;
2336 }
2337 
2338 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2339 {
2340         struct udp_iter_state *state = seq->private;
2341         struct net *net = seq_file_net(seq);
2342 
2343         do {
2344                 sk = sk_nulls_next(sk);
2345         } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2346 
2347         if (!sk) {
2348                 if (state->bucket <= state->udp_table->mask)
2349                         spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2350                 return udp_get_first(seq, state->bucket + 1);
2351         }
2352         return sk;
2353 }
2354 
2355 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2356 {
2357         struct sock *sk = udp_get_first(seq, 0);
2358 
2359         if (sk)
2360                 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2361                         --pos;
2362         return pos ? NULL : sk;
2363 }
2364 
2365 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2366 {
2367         struct udp_iter_state *state = seq->private;
2368         state->bucket = MAX_UDP_PORTS;
2369 
2370         return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2371 }
2372 
2373 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2374 {
2375         struct sock *sk;
2376 
2377         if (v == SEQ_START_TOKEN)
2378                 sk = udp_get_idx(seq, 0);
2379         else
2380                 sk = udp_get_next(seq, v);
2381 
2382         ++*pos;
2383         return sk;
2384 }
2385 
2386 static void udp_seq_stop(struct seq_file *seq, void *v)
2387 {
2388         struct udp_iter_state *state = seq->private;
2389 
2390         if (state->bucket <= state->udp_table->mask)
2391                 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2392 }
2393 
2394 int udp_seq_open(struct inode *inode, struct file *file)
2395 {
2396         struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2397         struct udp_iter_state *s;
2398         int err;
2399 
2400         err = seq_open_net(inode, file, &afinfo->seq_ops,
2401                            sizeof(struct udp_iter_state));
2402         if (err < 0)
2403                 return err;
2404 
2405         s = ((struct seq_file *)file->private_data)->private;
2406         s->family               = afinfo->family;
2407         s->udp_table            = afinfo->udp_table;
2408         return err;
2409 }
2410 EXPORT_SYMBOL(udp_seq_open);
2411 
2412 /* ------------------------------------------------------------------------ */
2413 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2414 {
2415         struct proc_dir_entry *p;
2416         int rc = 0;
2417 
2418         afinfo->seq_ops.start           = udp_seq_start;
2419         afinfo->seq_ops.next            = udp_seq_next;
2420         afinfo->seq_ops.stop            = udp_seq_stop;
2421 
2422         p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2423                              afinfo->seq_fops, afinfo);
2424         if (!p)
2425                 rc = -ENOMEM;
2426         return rc;
2427 }
2428 EXPORT_SYMBOL(udp_proc_register);
2429 
2430 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2431 {
2432         remove_proc_entry(afinfo->name, net->proc_net);
2433 }
2434 EXPORT_SYMBOL(udp_proc_unregister);
2435 
2436 /* ------------------------------------------------------------------------ */
2437 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2438                 int bucket)
2439 {
2440         struct inet_sock *inet = inet_sk(sp);
2441         __be32 dest = inet->inet_daddr;
2442         __be32 src  = inet->inet_rcv_saddr;
2443         __u16 destp       = ntohs(inet->inet_dport);
2444         __u16 srcp        = ntohs(inet->inet_sport);
2445 
2446         seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2447                 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2448                 bucket, src, srcp, dest, destp, sp->sk_state,
2449                 sk_wmem_alloc_get(sp),
2450                 sk_rmem_alloc_get(sp),
2451                 0, 0L, 0,
2452                 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2453                 0, sock_i_ino(sp),
2454                 atomic_read(&sp->sk_refcnt), sp,
2455                 atomic_read(&sp->sk_drops));
2456 }
2457 
2458 int udp4_seq_show(struct seq_file *seq, void *v)
2459 {
2460         seq_setwidth(seq, 127);
2461         if (v == SEQ_START_TOKEN)
2462                 seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2463                            "rx_queue tr tm->when retrnsmt   uid  timeout "
2464                            "inode ref pointer drops");
2465         else {
2466                 struct udp_iter_state *state = seq->private;
2467 
2468                 udp4_format_sock(v, seq, state->bucket);
2469         }
2470         seq_pad(seq, '\n');
2471         return 0;
2472 }
2473 
2474 static const struct file_operations udp_afinfo_seq_fops = {
2475         .owner    = THIS_MODULE,
2476         .open     = udp_seq_open,
2477         .read     = seq_read,
2478         .llseek   = seq_lseek,
2479         .release  = seq_release_net
2480 };
2481 
2482 /* ------------------------------------------------------------------------ */
2483 static struct udp_seq_afinfo udp4_seq_afinfo = {
2484         .name           = "udp",
2485         .family         = AF_INET,
2486         .udp_table      = &udp_table,
2487         .seq_fops       = &udp_afinfo_seq_fops,
2488         .seq_ops        = {
2489                 .show           = udp4_seq_show,
2490         },
2491 };
2492 
2493 static int __net_init udp4_proc_init_net(struct net *net)
2494 {
2495         return udp_proc_register(net, &udp4_seq_afinfo);
2496 }
2497 
2498 static void __net_exit udp4_proc_exit_net(struct net *net)
2499 {
2500         udp_proc_unregister(net, &udp4_seq_afinfo);
2501 }
2502 
2503 static struct pernet_operations udp4_net_ops = {
2504         .init = udp4_proc_init_net,
2505         .exit = udp4_proc_exit_net,
2506 };
2507 
2508 int __init udp4_proc_init(void)
2509 {
2510         return register_pernet_subsys(&udp4_net_ops);
2511 }
2512 
2513 void udp4_proc_exit(void)
2514 {
2515         unregister_pernet_subsys(&udp4_net_ops);
2516 }
2517 #endif /* CONFIG_PROC_FS */
2518 
2519 static __initdata unsigned long uhash_entries;
2520 static int __init set_uhash_entries(char *str)
2521 {
2522         ssize_t ret;
2523 
2524         if (!str)
2525                 return 0;
2526 
2527         ret = kstrtoul(str, 0, &uhash_entries);
2528         if (ret)
2529                 return 0;
2530 
2531         if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2532                 uhash_entries = UDP_HTABLE_SIZE_MIN;
2533         return 1;
2534 }
2535 __setup("uhash_entries=", set_uhash_entries);
2536 
2537 void __init udp_table_init(struct udp_table *table, const char *name)
2538 {
2539         unsigned int i;
2540 
2541         table->hash = alloc_large_system_hash(name,
2542                                               2 * sizeof(struct udp_hslot),
2543                                               uhash_entries,
2544                                               21, /* one slot per 2 MB */
2545                                               0,
2546                                               &table->log,
2547                                               &table->mask,
2548                                               UDP_HTABLE_SIZE_MIN,
2549                                               64 * 1024);
2550 
2551         table->hash2 = table->hash + (table->mask + 1);
2552         for (i = 0; i <= table->mask; i++) {
2553                 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2554                 table->hash[i].count = 0;
2555                 spin_lock_init(&table->hash[i].lock);
2556         }
2557         for (i = 0; i <= table->mask; i++) {
2558                 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2559                 table->hash2[i].count = 0;
2560                 spin_lock_init(&table->hash2[i].lock);
2561         }
2562 }
2563 
2564 u32 udp_flow_hashrnd(void)
2565 {
2566         static u32 hashrnd __read_mostly;
2567 
2568         net_get_random_once(&hashrnd, sizeof(hashrnd));
2569 
2570         return hashrnd;
2571 }
2572 EXPORT_SYMBOL(udp_flow_hashrnd);
2573 
2574 void __init udp_init(void)
2575 {
2576         unsigned long limit;
2577 
2578         udp_table_init(&udp_table, "UDP");
2579         limit = nr_free_buffer_pages() / 8;
2580         limit = max(limit, 128UL);
2581         sysctl_udp_mem[0] = limit / 4 * 3;
2582         sysctl_udp_mem[1] = limit;
2583         sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2584 
2585         sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2586         sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2587 }
2588 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp