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

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

Version: ~ [ linux-5.13-rc2 ] ~ [ linux-5.12.4 ] ~ [ linux-5.11.21 ] ~ [ linux-5.10.37 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.119 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.190 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.232 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.268 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.268 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ 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/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/route.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <trace/events/udp.h>
112 #include <linux/static_key.h>
113 #include <trace/events/skb.h>
114 #include <net/busy_poll.h>
115 #include "udp_impl.h"
116 #include <net/sock_reuseport.h>
117 
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                                                  bool match_wildcard),
143                                unsigned int log)
144 {
145         struct sock *sk2;
146         struct hlist_nulls_node *node;
147         kuid_t uid = sock_i_uid(sk);
148 
149         sk_nulls_for_each(sk2, node, &hslot->head) {
150                 if (net_eq(sock_net(sk2), net) &&
151                     sk2 != sk &&
152                     (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
153                     (!sk2->sk_reuse || !sk->sk_reuse) &&
154                     (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
155                      sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
156                     (!sk2->sk_reuseport || !sk->sk_reuseport ||
157                      rcu_access_pointer(sk->sk_reuseport_cb) ||
158                      !uid_eq(uid, sock_i_uid(sk2))) &&
159                     saddr_comp(sk, sk2, true)) {
160                         if (!bitmap)
161                                 return 1;
162                         __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap);
163                 }
164         }
165         return 0;
166 }
167 
168 /*
169  * Note: we still hold spinlock of primary hash chain, so no other writer
170  * can insert/delete a socket with local_port == num
171  */
172 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
173                                 struct udp_hslot *hslot2,
174                                 struct sock *sk,
175                                 int (*saddr_comp)(const struct sock *sk1,
176                                                   const struct sock *sk2,
177                                                   bool match_wildcard))
178 {
179         struct sock *sk2;
180         struct hlist_nulls_node *node;
181         kuid_t uid = sock_i_uid(sk);
182         int res = 0;
183 
184         spin_lock(&hslot2->lock);
185         udp_portaddr_for_each_entry(sk2, node, &hslot2->head) {
186                 if (net_eq(sock_net(sk2), net) &&
187                     sk2 != sk &&
188                     (udp_sk(sk2)->udp_port_hash == num) &&
189                     (!sk2->sk_reuse || !sk->sk_reuse) &&
190                     (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
191                      sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
192                     (!sk2->sk_reuseport || !sk->sk_reuseport ||
193                      rcu_access_pointer(sk->sk_reuseport_cb) ||
194                      !uid_eq(uid, sock_i_uid(sk2))) &&
195                     saddr_comp(sk, sk2, true)) {
196                         res = 1;
197                         break;
198                 }
199         }
200         spin_unlock(&hslot2->lock);
201         return res;
202 }
203 
204 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot,
205                                   int (*saddr_same)(const struct sock *sk1,
206                                                     const struct sock *sk2,
207                                                     bool match_wildcard))
208 {
209         struct net *net = sock_net(sk);
210         struct hlist_nulls_node *node;
211         kuid_t uid = sock_i_uid(sk);
212         struct sock *sk2;
213 
214         sk_nulls_for_each(sk2, node, &hslot->head) {
215                 if (net_eq(sock_net(sk2), net) &&
216                     sk2 != sk &&
217                     sk2->sk_family == sk->sk_family &&
218                     ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
219                     (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
220                     (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
221                     sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
222                     (*saddr_same)(sk, sk2, false)) {
223                         return reuseport_add_sock(sk, sk2);
224                 }
225         }
226 
227         /* Initial allocation may have already happened via setsockopt */
228         if (!rcu_access_pointer(sk->sk_reuseport_cb))
229                 return reuseport_alloc(sk);
230         return 0;
231 }
232 
233 /**
234  *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
235  *
236  *  @sk:          socket struct in question
237  *  @snum:        port number to look up
238  *  @saddr_comp:  AF-dependent comparison of bound local IP addresses
239  *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
240  *                   with NULL address
241  */
242 int udp_lib_get_port(struct sock *sk, unsigned short snum,
243                      int (*saddr_comp)(const struct sock *sk1,
244                                        const struct sock *sk2,
245                                        bool match_wildcard),
246                      unsigned int hash2_nulladdr)
247 {
248         struct udp_hslot *hslot, *hslot2;
249         struct udp_table *udptable = sk->sk_prot->h.udp_table;
250         int    error = 1;
251         struct net *net = sock_net(sk);
252 
253         if (!snum) {
254                 int low, high, remaining;
255                 unsigned int rand;
256                 unsigned short first, last;
257                 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
258 
259                 inet_get_local_port_range(net, &low, &high);
260                 remaining = (high - low) + 1;
261 
262                 rand = prandom_u32();
263                 first = reciprocal_scale(rand, remaining) + low;
264                 /*
265                  * force rand to be an odd multiple of UDP_HTABLE_SIZE
266                  */
267                 rand = (rand | 1) * (udptable->mask + 1);
268                 last = first + udptable->mask + 1;
269                 do {
270                         hslot = udp_hashslot(udptable, net, first);
271                         bitmap_zero(bitmap, PORTS_PER_CHAIN);
272                         spin_lock_bh(&hslot->lock);
273                         udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
274                                             saddr_comp, udptable->log);
275 
276                         snum = first;
277                         /*
278                          * Iterate on all possible values of snum for this hash.
279                          * Using steps of an odd multiple of UDP_HTABLE_SIZE
280                          * give us randomization and full range coverage.
281                          */
282                         do {
283                                 if (low <= snum && snum <= high &&
284                                     !test_bit(snum >> udptable->log, bitmap) &&
285                                     !inet_is_local_reserved_port(net, snum))
286                                         goto found;
287                                 snum += rand;
288                         } while (snum != first);
289                         spin_unlock_bh(&hslot->lock);
290                 } while (++first != last);
291                 goto fail;
292         } else {
293                 hslot = udp_hashslot(udptable, net, snum);
294                 spin_lock_bh(&hslot->lock);
295                 if (hslot->count > 10) {
296                         int exist;
297                         unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
298 
299                         slot2          &= udptable->mask;
300                         hash2_nulladdr &= udptable->mask;
301 
302                         hslot2 = udp_hashslot2(udptable, slot2);
303                         if (hslot->count < hslot2->count)
304                                 goto scan_primary_hash;
305 
306                         exist = udp_lib_lport_inuse2(net, snum, hslot2,
307                                                      sk, saddr_comp);
308                         if (!exist && (hash2_nulladdr != slot2)) {
309                                 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
310                                 exist = udp_lib_lport_inuse2(net, snum, hslot2,
311                                                              sk, saddr_comp);
312                         }
313                         if (exist)
314                                 goto fail_unlock;
315                         else
316                                 goto found;
317                 }
318 scan_primary_hash:
319                 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
320                                         saddr_comp, 0))
321                         goto fail_unlock;
322         }
323 found:
324         inet_sk(sk)->inet_num = snum;
325         udp_sk(sk)->udp_port_hash = snum;
326         udp_sk(sk)->udp_portaddr_hash ^= snum;
327         if (sk_unhashed(sk)) {
328                 if (sk->sk_reuseport &&
329                     udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
330                         inet_sk(sk)->inet_num = 0;
331                         udp_sk(sk)->udp_port_hash = 0;
332                         udp_sk(sk)->udp_portaddr_hash ^= snum;
333                         goto fail_unlock;
334                 }
335 
336                 sk_nulls_add_node_rcu(sk, &hslot->head);
337                 hslot->count++;
338                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
339 
340                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
341                 spin_lock(&hslot2->lock);
342                 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
343                         sk->sk_family == AF_INET6)
344                         hlist_nulls_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
345                                                  &hslot2->head);
346                 else
347                         hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
348                                                  &hslot2->head);
349                 hslot2->count++;
350                 spin_unlock(&hslot2->lock);
351         }
352         error = 0;
353 fail_unlock:
354         spin_unlock_bh(&hslot->lock);
355 fail:
356         return error;
357 }
358 EXPORT_SYMBOL(udp_lib_get_port);
359 
360 /* match_wildcard == true:  0.0.0.0 equals to any IPv4 addresses
361  * match_wildcard == false: addresses must be exactly the same, i.e.
362  *                          0.0.0.0 only equals to 0.0.0.0
363  */
364 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
365                          bool match_wildcard)
366 {
367         struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
368 
369         if (!ipv6_only_sock(sk2)) {
370                 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
371                         return 1;
372                 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
373                         return match_wildcard;
374         }
375         return 0;
376 }
377 
378 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
379                               unsigned int port)
380 {
381         return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
382 }
383 
384 int udp_v4_get_port(struct sock *sk, unsigned short snum)
385 {
386         unsigned int hash2_nulladdr =
387                 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
388         unsigned int hash2_partial =
389                 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
390 
391         /* precompute partial secondary hash */
392         udp_sk(sk)->udp_portaddr_hash = hash2_partial;
393         return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
394 }
395 
396 static inline int compute_score(struct sock *sk, struct net *net,
397                                 __be32 saddr, unsigned short hnum, __be16 sport,
398                                 __be32 daddr, __be16 dport, int dif)
399 {
400         int score;
401         struct inet_sock *inet;
402 
403         if (!net_eq(sock_net(sk), net) ||
404             udp_sk(sk)->udp_port_hash != hnum ||
405             ipv6_only_sock(sk))
406                 return -1;
407 
408         score = (sk->sk_family == PF_INET) ? 2 : 1;
409         inet = inet_sk(sk);
410 
411         if (inet->inet_rcv_saddr) {
412                 if (inet->inet_rcv_saddr != daddr)
413                         return -1;
414                 score += 4;
415         }
416 
417         if (inet->inet_daddr) {
418                 if (inet->inet_daddr != saddr)
419                         return -1;
420                 score += 4;
421         }
422 
423         if (inet->inet_dport) {
424                 if (inet->inet_dport != sport)
425                         return -1;
426                 score += 4;
427         }
428 
429         if (sk->sk_bound_dev_if) {
430                 if (sk->sk_bound_dev_if != dif)
431                         return -1;
432                 score += 4;
433         }
434         if (sk->sk_incoming_cpu == raw_smp_processor_id())
435                 score++;
436         return score;
437 }
438 
439 /*
440  * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
441  */
442 static inline int compute_score2(struct sock *sk, struct net *net,
443                                  __be32 saddr, __be16 sport,
444                                  __be32 daddr, unsigned int hnum, int dif)
445 {
446         int score;
447         struct inet_sock *inet;
448 
449         if (!net_eq(sock_net(sk), net) ||
450             ipv6_only_sock(sk))
451                 return -1;
452 
453         inet = inet_sk(sk);
454 
455         if (inet->inet_rcv_saddr != daddr ||
456             inet->inet_num != hnum)
457                 return -1;
458 
459         score = (sk->sk_family == PF_INET) ? 2 : 1;
460 
461         if (inet->inet_daddr) {
462                 if (inet->inet_daddr != saddr)
463                         return -1;
464                 score += 4;
465         }
466 
467         if (inet->inet_dport) {
468                 if (inet->inet_dport != sport)
469                         return -1;
470                 score += 4;
471         }
472 
473         if (sk->sk_bound_dev_if) {
474                 if (sk->sk_bound_dev_if != dif)
475                         return -1;
476                 score += 4;
477         }
478 
479         if (sk->sk_incoming_cpu == raw_smp_processor_id())
480                 score++;
481 
482         return score;
483 }
484 
485 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
486                        const __u16 lport, const __be32 faddr,
487                        const __be16 fport)
488 {
489         static u32 udp_ehash_secret __read_mostly;
490 
491         net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
492 
493         return __inet_ehashfn(laddr, lport, faddr, fport,
494                               udp_ehash_secret + net_hash_mix(net));
495 }
496 
497 /* called with read_rcu_lock() */
498 static struct sock *udp4_lib_lookup2(struct net *net,
499                 __be32 saddr, __be16 sport,
500                 __be32 daddr, unsigned int hnum, int dif,
501                 struct udp_hslot *hslot2, unsigned int slot2,
502                 struct sk_buff *skb)
503 {
504         struct sock *sk, *result;
505         struct hlist_nulls_node *node;
506         int score, badness, matches = 0, reuseport = 0;
507         bool select_ok = true;
508         u32 hash = 0;
509 
510 begin:
511         result = NULL;
512         badness = 0;
513         udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
514                 score = compute_score2(sk, net, saddr, sport,
515                                       daddr, hnum, dif);
516                 if (score > badness) {
517                         result = sk;
518                         badness = score;
519                         reuseport = sk->sk_reuseport;
520                         if (reuseport) {
521                                 hash = udp_ehashfn(net, daddr, hnum,
522                                                    saddr, sport);
523                                 if (select_ok) {
524                                         struct sock *sk2;
525 
526                                         sk2 = reuseport_select_sock(sk, hash, skb,
527                                                         sizeof(struct udphdr));
528                                         if (sk2) {
529                                                 result = sk2;
530                                                 select_ok = false;
531                                                 goto found;
532                                         }
533                                 }
534                                 matches = 1;
535                         }
536                 } else if (score == badness && reuseport) {
537                         matches++;
538                         if (reciprocal_scale(hash, matches) == 0)
539                                 result = sk;
540                         hash = next_pseudo_random32(hash);
541                 }
542         }
543         /*
544          * if the nulls value we got at the end of this lookup is
545          * not the expected one, we must restart lookup.
546          * We probably met an item that was moved to another chain.
547          */
548         if (get_nulls_value(node) != slot2)
549                 goto begin;
550         if (result) {
551 found:
552                 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
553                         result = NULL;
554                 else if (unlikely(compute_score2(result, net, saddr, sport,
555                                   daddr, hnum, dif) < badness)) {
556                         sock_put(result);
557                         goto begin;
558                 }
559         }
560         return result;
561 }
562 
563 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
564  * harder than this. -DaveM
565  */
566 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
567                 __be16 sport, __be32 daddr, __be16 dport,
568                 int dif, struct udp_table *udptable, struct sk_buff *skb)
569 {
570         struct sock *sk, *result;
571         struct hlist_nulls_node *node;
572         unsigned short hnum = ntohs(dport);
573         unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
574         struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
575         int score, badness, matches = 0, reuseport = 0;
576         bool select_ok = true;
577         u32 hash = 0;
578 
579         rcu_read_lock();
580         if (hslot->count > 10) {
581                 hash2 = udp4_portaddr_hash(net, daddr, hnum);
582                 slot2 = hash2 & udptable->mask;
583                 hslot2 = &udptable->hash2[slot2];
584                 if (hslot->count < hslot2->count)
585                         goto begin;
586 
587                 result = udp4_lib_lookup2(net, saddr, sport,
588                                           daddr, hnum, dif,
589                                           hslot2, slot2, skb);
590                 if (!result) {
591                         hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
592                         slot2 = hash2 & udptable->mask;
593                         hslot2 = &udptable->hash2[slot2];
594                         if (hslot->count < hslot2->count)
595                                 goto begin;
596 
597                         result = udp4_lib_lookup2(net, saddr, sport,
598                                                   htonl(INADDR_ANY), hnum, dif,
599                                                   hslot2, slot2, skb);
600                 }
601                 rcu_read_unlock();
602                 return result;
603         }
604 begin:
605         result = NULL;
606         badness = 0;
607         sk_nulls_for_each_rcu(sk, node, &hslot->head) {
608                 score = compute_score(sk, net, saddr, hnum, sport,
609                                       daddr, dport, dif);
610                 if (score > badness) {
611                         result = sk;
612                         badness = score;
613                         reuseport = sk->sk_reuseport;
614                         if (reuseport) {
615                                 hash = udp_ehashfn(net, daddr, hnum,
616                                                    saddr, sport);
617                                 if (select_ok) {
618                                         struct sock *sk2;
619 
620                                         sk2 = reuseport_select_sock(sk, hash, skb,
621                                                         sizeof(struct udphdr));
622                                         if (sk2) {
623                                                 result = sk2;
624                                                 select_ok = false;
625                                                 goto found;
626                                         }
627                                 }
628                                 matches = 1;
629                         }
630                 } else if (score == badness && reuseport) {
631                         matches++;
632                         if (reciprocal_scale(hash, matches) == 0)
633                                 result = sk;
634                         hash = next_pseudo_random32(hash);
635                 }
636         }
637         /*
638          * if the nulls value we got at the end of this lookup is
639          * not the expected one, we must restart lookup.
640          * We probably met an item that was moved to another chain.
641          */
642         if (get_nulls_value(node) != slot)
643                 goto begin;
644 
645         if (result) {
646 found:
647                 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
648                         result = NULL;
649                 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
650                                   daddr, dport, dif) < badness)) {
651                         sock_put(result);
652                         goto begin;
653                 }
654         }
655         rcu_read_unlock();
656         return result;
657 }
658 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
659 
660 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
661                                                  __be16 sport, __be16 dport,
662                                                  struct udp_table *udptable)
663 {
664         const struct iphdr *iph = ip_hdr(skb);
665 
666         return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
667                                  iph->daddr, dport, inet_iif(skb),
668                                  udptable, skb);
669 }
670 
671 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
672                              __be32 daddr, __be16 dport, int dif)
673 {
674         return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif,
675                                  &udp_table, NULL);
676 }
677 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
678 
679 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
680                                        __be16 loc_port, __be32 loc_addr,
681                                        __be16 rmt_port, __be32 rmt_addr,
682                                        int dif, unsigned short hnum)
683 {
684         struct inet_sock *inet = inet_sk(sk);
685 
686         if (!net_eq(sock_net(sk), net) ||
687             udp_sk(sk)->udp_port_hash != hnum ||
688             (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
689             (inet->inet_dport != rmt_port && inet->inet_dport) ||
690             (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
691             ipv6_only_sock(sk) ||
692             (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
693                 return false;
694         if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
695                 return false;
696         return true;
697 }
698 
699 /*
700  * This routine is called by the ICMP module when it gets some
701  * sort of error condition.  If err < 0 then the socket should
702  * be closed and the error returned to the user.  If err > 0
703  * it's just the icmp type << 8 | icmp code.
704  * Header points to the ip header of the error packet. We move
705  * on past this. Then (as it used to claim before adjustment)
706  * header points to the first 8 bytes of the udp header.  We need
707  * to find the appropriate port.
708  */
709 
710 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
711 {
712         struct inet_sock *inet;
713         const struct iphdr *iph = (const struct iphdr *)skb->data;
714         struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
715         const int type = icmp_hdr(skb)->type;
716         const int code = icmp_hdr(skb)->code;
717         struct sock *sk;
718         int harderr;
719         int err;
720         struct net *net = dev_net(skb->dev);
721 
722         sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
723                         iph->saddr, uh->source, skb->dev->ifindex, udptable,
724                         NULL);
725         if (!sk) {
726                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
727                 return; /* No socket for error */
728         }
729 
730         err = 0;
731         harderr = 0;
732         inet = inet_sk(sk);
733 
734         switch (type) {
735         default:
736         case ICMP_TIME_EXCEEDED:
737                 err = EHOSTUNREACH;
738                 break;
739         case ICMP_SOURCE_QUENCH:
740                 goto out;
741         case ICMP_PARAMETERPROB:
742                 err = EPROTO;
743                 harderr = 1;
744                 break;
745         case ICMP_DEST_UNREACH:
746                 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
747                         ipv4_sk_update_pmtu(skb, sk, info);
748                         if (inet->pmtudisc != IP_PMTUDISC_DONT) {
749                                 err = EMSGSIZE;
750                                 harderr = 1;
751                                 break;
752                         }
753                         goto out;
754                 }
755                 err = EHOSTUNREACH;
756                 if (code <= NR_ICMP_UNREACH) {
757                         harderr = icmp_err_convert[code].fatal;
758                         err = icmp_err_convert[code].errno;
759                 }
760                 break;
761         case ICMP_REDIRECT:
762                 ipv4_sk_redirect(skb, sk);
763                 goto out;
764         }
765 
766         /*
767          *      RFC1122: OK.  Passes ICMP errors back to application, as per
768          *      4.1.3.3.
769          */
770         if (!inet->recverr) {
771                 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
772                         goto out;
773         } else
774                 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
775 
776         sk->sk_err = err;
777         sk->sk_error_report(sk);
778 out:
779         sock_put(sk);
780 }
781 
782 void udp_err(struct sk_buff *skb, u32 info)
783 {
784         __udp4_lib_err(skb, info, &udp_table);
785 }
786 
787 /*
788  * Throw away all pending data and cancel the corking. Socket is locked.
789  */
790 void udp_flush_pending_frames(struct sock *sk)
791 {
792         struct udp_sock *up = udp_sk(sk);
793 
794         if (up->pending) {
795                 up->len = 0;
796                 up->pending = 0;
797                 ip_flush_pending_frames(sk);
798         }
799 }
800 EXPORT_SYMBOL(udp_flush_pending_frames);
801 
802 /**
803  *      udp4_hwcsum  -  handle outgoing HW checksumming
804  *      @skb:   sk_buff containing the filled-in UDP header
805  *              (checksum field must be zeroed out)
806  *      @src:   source IP address
807  *      @dst:   destination IP address
808  */
809 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
810 {
811         struct udphdr *uh = udp_hdr(skb);
812         int offset = skb_transport_offset(skb);
813         int len = skb->len - offset;
814         int hlen = len;
815         __wsum csum = 0;
816 
817         if (!skb_has_frag_list(skb)) {
818                 /*
819                  * Only one fragment on the socket.
820                  */
821                 skb->csum_start = skb_transport_header(skb) - skb->head;
822                 skb->csum_offset = offsetof(struct udphdr, check);
823                 uh->check = ~csum_tcpudp_magic(src, dst, len,
824                                                IPPROTO_UDP, 0);
825         } else {
826                 struct sk_buff *frags;
827 
828                 /*
829                  * HW-checksum won't work as there are two or more
830                  * fragments on the socket so that all csums of sk_buffs
831                  * should be together
832                  */
833                 skb_walk_frags(skb, frags) {
834                         csum = csum_add(csum, frags->csum);
835                         hlen -= frags->len;
836                 }
837 
838                 csum = skb_checksum(skb, offset, hlen, csum);
839                 skb->ip_summed = CHECKSUM_NONE;
840 
841                 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
842                 if (uh->check == 0)
843                         uh->check = CSUM_MANGLED_0;
844         }
845 }
846 EXPORT_SYMBOL_GPL(udp4_hwcsum);
847 
848 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
849  * for the simple case like when setting the checksum for a UDP tunnel.
850  */
851 void udp_set_csum(bool nocheck, struct sk_buff *skb,
852                   __be32 saddr, __be32 daddr, int len)
853 {
854         struct udphdr *uh = udp_hdr(skb);
855 
856         if (nocheck) {
857                 uh->check = 0;
858         } else if (skb_is_gso(skb)) {
859                 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
860         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
861                 uh->check = 0;
862                 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
863                 if (uh->check == 0)
864                         uh->check = CSUM_MANGLED_0;
865         } else {
866                 skb->ip_summed = CHECKSUM_PARTIAL;
867                 skb->csum_start = skb_transport_header(skb) - skb->head;
868                 skb->csum_offset = offsetof(struct udphdr, check);
869                 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
870         }
871 }
872 EXPORT_SYMBOL(udp_set_csum);
873 
874 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
875 {
876         struct sock *sk = skb->sk;
877         struct inet_sock *inet = inet_sk(sk);
878         struct udphdr *uh;
879         int err = 0;
880         int is_udplite = IS_UDPLITE(sk);
881         int offset = skb_transport_offset(skb);
882         int len = skb->len - offset;
883         __wsum csum = 0;
884 
885         /*
886          * Create a UDP header
887          */
888         uh = udp_hdr(skb);
889         uh->source = inet->inet_sport;
890         uh->dest = fl4->fl4_dport;
891         uh->len = htons(len);
892         uh->check = 0;
893 
894         if (is_udplite)                                  /*     UDP-Lite      */
895                 csum = udplite_csum(skb);
896 
897         else if (sk->sk_no_check_tx) {   /* UDP csum disabled */
898 
899                 skb->ip_summed = CHECKSUM_NONE;
900                 goto send;
901 
902         } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
903 
904                 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
905                 goto send;
906 
907         } else
908                 csum = udp_csum(skb);
909 
910         /* add protocol-dependent pseudo-header */
911         uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
912                                       sk->sk_protocol, csum);
913         if (uh->check == 0)
914                 uh->check = CSUM_MANGLED_0;
915 
916 send:
917         err = ip_send_skb(sock_net(sk), skb);
918         if (err) {
919                 if (err == -ENOBUFS && !inet->recverr) {
920                         UDP_INC_STATS_USER(sock_net(sk),
921                                            UDP_MIB_SNDBUFERRORS, is_udplite);
922                         err = 0;
923                 }
924         } else
925                 UDP_INC_STATS_USER(sock_net(sk),
926                                    UDP_MIB_OUTDATAGRAMS, is_udplite);
927         return err;
928 }
929 
930 /*
931  * Push out all pending data as one UDP datagram. Socket is locked.
932  */
933 int udp_push_pending_frames(struct sock *sk)
934 {
935         struct udp_sock  *up = udp_sk(sk);
936         struct inet_sock *inet = inet_sk(sk);
937         struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
938         struct sk_buff *skb;
939         int err = 0;
940 
941         skb = ip_finish_skb(sk, fl4);
942         if (!skb)
943                 goto out;
944 
945         err = udp_send_skb(skb, fl4);
946 
947 out:
948         up->len = 0;
949         up->pending = 0;
950         return err;
951 }
952 EXPORT_SYMBOL(udp_push_pending_frames);
953 
954 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
955 {
956         struct inet_sock *inet = inet_sk(sk);
957         struct udp_sock *up = udp_sk(sk);
958         struct flowi4 fl4_stack;
959         struct flowi4 *fl4;
960         int ulen = len;
961         struct ipcm_cookie ipc;
962         struct rtable *rt = NULL;
963         int free = 0;
964         int connected = 0;
965         __be32 daddr, faddr, saddr;
966         __be16 dport;
967         u8  tos;
968         int err, is_udplite = IS_UDPLITE(sk);
969         int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
970         int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
971         struct sk_buff *skb;
972         struct ip_options_data opt_copy;
973 
974         if (len > 0xFFFF)
975                 return -EMSGSIZE;
976 
977         /*
978          *      Check the flags.
979          */
980 
981         if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
982                 return -EOPNOTSUPP;
983 
984         ipc.opt = NULL;
985         ipc.tx_flags = 0;
986         ipc.ttl = 0;
987         ipc.tos = -1;
988 
989         getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
990 
991         fl4 = &inet->cork.fl.u.ip4;
992         if (up->pending) {
993                 /*
994                  * There are pending frames.
995                  * The socket lock must be held while it's corked.
996                  */
997                 lock_sock(sk);
998                 if (likely(up->pending)) {
999                         if (unlikely(up->pending != AF_INET)) {
1000                                 release_sock(sk);
1001                                 return -EINVAL;
1002                         }
1003                         goto do_append_data;
1004                 }
1005                 release_sock(sk);
1006         }
1007         ulen += sizeof(struct udphdr);
1008 
1009         /*
1010          *      Get and verify the address.
1011          */
1012         if (msg->msg_name) {
1013                 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1014                 if (msg->msg_namelen < sizeof(*usin))
1015                         return -EINVAL;
1016                 if (usin->sin_family != AF_INET) {
1017                         if (usin->sin_family != AF_UNSPEC)
1018                                 return -EAFNOSUPPORT;
1019                 }
1020 
1021                 daddr = usin->sin_addr.s_addr;
1022                 dport = usin->sin_port;
1023                 if (dport == 0)
1024                         return -EINVAL;
1025         } else {
1026                 if (sk->sk_state != TCP_ESTABLISHED)
1027                         return -EDESTADDRREQ;
1028                 daddr = inet->inet_daddr;
1029                 dport = inet->inet_dport;
1030                 /* Open fast path for connected socket.
1031                    Route will not be used, if at least one option is set.
1032                  */
1033                 connected = 1;
1034         }
1035         ipc.addr = inet->inet_saddr;
1036 
1037         ipc.oif = sk->sk_bound_dev_if;
1038 
1039         sock_tx_timestamp(sk, &ipc.tx_flags);
1040 
1041         if (msg->msg_controllen) {
1042                 err = ip_cmsg_send(sock_net(sk), msg, &ipc,
1043                                    sk->sk_family == AF_INET6);
1044                 if (unlikely(err)) {
1045                         kfree(ipc.opt);
1046                         return err;
1047                 }
1048                 if (ipc.opt)
1049                         free = 1;
1050                 connected = 0;
1051         }
1052         if (!ipc.opt) {
1053                 struct ip_options_rcu *inet_opt;
1054 
1055                 rcu_read_lock();
1056                 inet_opt = rcu_dereference(inet->inet_opt);
1057                 if (inet_opt) {
1058                         memcpy(&opt_copy, inet_opt,
1059                                sizeof(*inet_opt) + inet_opt->opt.optlen);
1060                         ipc.opt = &opt_copy.opt;
1061                 }
1062                 rcu_read_unlock();
1063         }
1064 
1065         saddr = ipc.addr;
1066         ipc.addr = faddr = daddr;
1067 
1068         if (ipc.opt && ipc.opt->opt.srr) {
1069                 if (!daddr)
1070                         return -EINVAL;
1071                 faddr = ipc.opt->opt.faddr;
1072                 connected = 0;
1073         }
1074         tos = get_rttos(&ipc, inet);
1075         if (sock_flag(sk, SOCK_LOCALROUTE) ||
1076             (msg->msg_flags & MSG_DONTROUTE) ||
1077             (ipc.opt && ipc.opt->opt.is_strictroute)) {
1078                 tos |= RTO_ONLINK;
1079                 connected = 0;
1080         }
1081 
1082         if (ipv4_is_multicast(daddr)) {
1083                 if (!ipc.oif)
1084                         ipc.oif = inet->mc_index;
1085                 if (!saddr)
1086                         saddr = inet->mc_addr;
1087                 connected = 0;
1088         } else if (!ipc.oif)
1089                 ipc.oif = inet->uc_index;
1090 
1091         if (connected)
1092                 rt = (struct rtable *)sk_dst_check(sk, 0);
1093 
1094         if (!rt) {
1095                 struct net *net = sock_net(sk);
1096                 __u8 flow_flags = inet_sk_flowi_flags(sk);
1097 
1098                 fl4 = &fl4_stack;
1099 
1100                 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1101                                    RT_SCOPE_UNIVERSE, sk->sk_protocol,
1102                                    flow_flags,
1103                                    faddr, saddr, dport, inet->inet_sport);
1104 
1105                 if (!saddr && ipc.oif) {
1106                         err = l3mdev_get_saddr(net, ipc.oif, fl4);
1107                         if (err < 0)
1108                                 goto out;
1109                 }
1110 
1111                 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1112                 rt = ip_route_output_flow(net, fl4, sk);
1113                 if (IS_ERR(rt)) {
1114                         err = PTR_ERR(rt);
1115                         rt = NULL;
1116                         if (err == -ENETUNREACH)
1117                                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1118                         goto out;
1119                 }
1120 
1121                 err = -EACCES;
1122                 if ((rt->rt_flags & RTCF_BROADCAST) &&
1123                     !sock_flag(sk, SOCK_BROADCAST))
1124                         goto out;
1125                 if (connected)
1126                         sk_dst_set(sk, dst_clone(&rt->dst));
1127         }
1128 
1129         if (msg->msg_flags&MSG_CONFIRM)
1130                 goto do_confirm;
1131 back_from_confirm:
1132 
1133         saddr = fl4->saddr;
1134         if (!ipc.addr)
1135                 daddr = ipc.addr = fl4->daddr;
1136 
1137         /* Lockless fast path for the non-corking case. */
1138         if (!corkreq) {
1139                 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1140                                   sizeof(struct udphdr), &ipc, &rt,
1141                                   msg->msg_flags);
1142                 err = PTR_ERR(skb);
1143                 if (!IS_ERR_OR_NULL(skb))
1144                         err = udp_send_skb(skb, fl4);
1145                 goto out;
1146         }
1147 
1148         lock_sock(sk);
1149         if (unlikely(up->pending)) {
1150                 /* The socket is already corked while preparing it. */
1151                 /* ... which is an evident application bug. --ANK */
1152                 release_sock(sk);
1153 
1154                 net_dbg_ratelimited("cork app bug 2\n");
1155                 err = -EINVAL;
1156                 goto out;
1157         }
1158         /*
1159          *      Now cork the socket to pend data.
1160          */
1161         fl4 = &inet->cork.fl.u.ip4;
1162         fl4->daddr = daddr;
1163         fl4->saddr = saddr;
1164         fl4->fl4_dport = dport;
1165         fl4->fl4_sport = inet->inet_sport;
1166         up->pending = AF_INET;
1167 
1168 do_append_data:
1169         up->len += ulen;
1170         err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1171                              sizeof(struct udphdr), &ipc, &rt,
1172                              corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1173         if (err)
1174                 udp_flush_pending_frames(sk);
1175         else if (!corkreq)
1176                 err = udp_push_pending_frames(sk);
1177         else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1178                 up->pending = 0;
1179         release_sock(sk);
1180 
1181 out:
1182         ip_rt_put(rt);
1183         if (free)
1184                 kfree(ipc.opt);
1185         if (!err)
1186                 return len;
1187         /*
1188          * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1189          * ENOBUFS might not be good (it's not tunable per se), but otherwise
1190          * we don't have a good statistic (IpOutDiscards but it can be too many
1191          * things).  We could add another new stat but at least for now that
1192          * seems like overkill.
1193          */
1194         if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1195                 UDP_INC_STATS_USER(sock_net(sk),
1196                                 UDP_MIB_SNDBUFERRORS, is_udplite);
1197         }
1198         return err;
1199 
1200 do_confirm:
1201         dst_confirm(&rt->dst);
1202         if (!(msg->msg_flags&MSG_PROBE) || len)
1203                 goto back_from_confirm;
1204         err = 0;
1205         goto out;
1206 }
1207 EXPORT_SYMBOL(udp_sendmsg);
1208 
1209 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1210                  size_t size, int flags)
1211 {
1212         struct inet_sock *inet = inet_sk(sk);
1213         struct udp_sock *up = udp_sk(sk);
1214         int ret;
1215 
1216         if (flags & MSG_SENDPAGE_NOTLAST)
1217                 flags |= MSG_MORE;
1218 
1219         if (!up->pending) {
1220                 struct msghdr msg = {   .msg_flags = flags|MSG_MORE };
1221 
1222                 /* Call udp_sendmsg to specify destination address which
1223                  * sendpage interface can't pass.
1224                  * This will succeed only when the socket is connected.
1225                  */
1226                 ret = udp_sendmsg(sk, &msg, 0);
1227                 if (ret < 0)
1228                         return ret;
1229         }
1230 
1231         lock_sock(sk);
1232 
1233         if (unlikely(!up->pending)) {
1234                 release_sock(sk);
1235 
1236                 net_dbg_ratelimited("udp cork app bug 3\n");
1237                 return -EINVAL;
1238         }
1239 
1240         ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1241                              page, offset, size, flags);
1242         if (ret == -EOPNOTSUPP) {
1243                 release_sock(sk);
1244                 return sock_no_sendpage(sk->sk_socket, page, offset,
1245                                         size, flags);
1246         }
1247         if (ret < 0) {
1248                 udp_flush_pending_frames(sk);
1249                 goto out;
1250         }
1251 
1252         up->len += size;
1253         if (!(up->corkflag || (flags&MSG_MORE)))
1254                 ret = udp_push_pending_frames(sk);
1255         if (!ret)
1256                 ret = size;
1257 out:
1258         release_sock(sk);
1259         return ret;
1260 }
1261 
1262 /**
1263  *      first_packet_length     - return length of first packet in receive queue
1264  *      @sk: socket
1265  *
1266  *      Drops all bad checksum frames, until a valid one is found.
1267  *      Returns the length of found skb, or 0 if none is found.
1268  */
1269 static unsigned int first_packet_length(struct sock *sk)
1270 {
1271         struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1272         struct sk_buff *skb;
1273         unsigned int res;
1274 
1275         __skb_queue_head_init(&list_kill);
1276 
1277         spin_lock_bh(&rcvq->lock);
1278         while ((skb = skb_peek(rcvq)) != NULL &&
1279                 udp_lib_checksum_complete(skb)) {
1280                 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1281                                  IS_UDPLITE(sk));
1282                 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1283                                  IS_UDPLITE(sk));
1284                 atomic_inc(&sk->sk_drops);
1285                 __skb_unlink(skb, rcvq);
1286                 __skb_queue_tail(&list_kill, skb);
1287         }
1288         res = skb ? skb->len : 0;
1289         spin_unlock_bh(&rcvq->lock);
1290 
1291         if (!skb_queue_empty(&list_kill)) {
1292                 bool slow = lock_sock_fast(sk);
1293 
1294                 __skb_queue_purge(&list_kill);
1295                 sk_mem_reclaim_partial(sk);
1296                 unlock_sock_fast(sk, slow);
1297         }
1298         return res;
1299 }
1300 
1301 /*
1302  *      IOCTL requests applicable to the UDP protocol
1303  */
1304 
1305 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1306 {
1307         switch (cmd) {
1308         case SIOCOUTQ:
1309         {
1310                 int amount = sk_wmem_alloc_get(sk);
1311 
1312                 return put_user(amount, (int __user *)arg);
1313         }
1314 
1315         case SIOCINQ:
1316         {
1317                 unsigned int amount = first_packet_length(sk);
1318 
1319                 if (amount)
1320                         /*
1321                          * We will only return the amount
1322                          * of this packet since that is all
1323                          * that will be read.
1324                          */
1325                         amount -= sizeof(struct udphdr);
1326 
1327                 return put_user(amount, (int __user *)arg);
1328         }
1329 
1330         default:
1331                 return -ENOIOCTLCMD;
1332         }
1333 
1334         return 0;
1335 }
1336 EXPORT_SYMBOL(udp_ioctl);
1337 
1338 /*
1339  *      This should be easy, if there is something there we
1340  *      return it, otherwise we block.
1341  */
1342 
1343 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1344                 int flags, int *addr_len)
1345 {
1346         struct inet_sock *inet = inet_sk(sk);
1347         DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1348         struct sk_buff *skb;
1349         unsigned int ulen, copied;
1350         int peeked, off = 0;
1351         int err;
1352         int is_udplite = IS_UDPLITE(sk);
1353         bool checksum_valid = false;
1354         bool slow;
1355 
1356         if (flags & MSG_ERRQUEUE)
1357                 return ip_recv_error(sk, msg, len, addr_len);
1358 
1359 try_again:
1360         skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1361                                   &peeked, &off, &err);
1362         if (!skb)
1363                 goto out;
1364         if (ccs_socket_post_recvmsg_permission(sk, skb, flags)) {
1365                 err = -EAGAIN; /* Hope less harmful than -EPERM. */
1366                 goto out;
1367         }
1368 
1369         ulen = skb->len - sizeof(struct udphdr);
1370         copied = len;
1371         if (copied > ulen)
1372                 copied = ulen;
1373         else if (copied < ulen)
1374                 msg->msg_flags |= MSG_TRUNC;
1375 
1376         /*
1377          * If checksum is needed at all, try to do it while copying the
1378          * data.  If the data is truncated, or if we only want a partial
1379          * coverage checksum (UDP-Lite), do it before the copy.
1380          */
1381 
1382         if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1383                 checksum_valid = !udp_lib_checksum_complete(skb);
1384                 if (!checksum_valid)
1385                         goto csum_copy_err;
1386         }
1387 
1388         if (checksum_valid || skb_csum_unnecessary(skb))
1389                 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
1390                                             msg, copied);
1391         else {
1392                 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr),
1393                                                      msg);
1394 
1395                 if (err == -EINVAL)
1396                         goto csum_copy_err;
1397         }
1398 
1399         if (unlikely(err)) {
1400                 trace_kfree_skb(skb, udp_recvmsg);
1401                 if (!peeked) {
1402                         atomic_inc(&sk->sk_drops);
1403                         UDP_INC_STATS_USER(sock_net(sk),
1404                                            UDP_MIB_INERRORS, is_udplite);
1405                 }
1406                 goto out_free;
1407         }
1408 
1409         if (!peeked)
1410                 UDP_INC_STATS_USER(sock_net(sk),
1411                                 UDP_MIB_INDATAGRAMS, is_udplite);
1412 
1413         sock_recv_ts_and_drops(msg, sk, skb);
1414 
1415         /* Copy the address. */
1416         if (sin) {
1417                 sin->sin_family = AF_INET;
1418                 sin->sin_port = udp_hdr(skb)->source;
1419                 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1420                 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1421                 *addr_len = sizeof(*sin);
1422         }
1423         if (inet->cmsg_flags)
1424                 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr));
1425 
1426         err = copied;
1427         if (flags & MSG_TRUNC)
1428                 err = ulen;
1429 
1430 out_free:
1431         skb_free_datagram_locked(sk, skb);
1432 out:
1433         return err;
1434 
1435 csum_copy_err:
1436         slow = lock_sock_fast(sk);
1437         if (!skb_kill_datagram(sk, skb, flags)) {
1438                 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1439                 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1440         }
1441         unlock_sock_fast(sk, slow);
1442 
1443         /* starting over for a new packet, but check if we need to yield */
1444         cond_resched();
1445         msg->msg_flags &= ~MSG_TRUNC;
1446         goto try_again;
1447 }
1448 
1449 int udp_disconnect(struct sock *sk, int flags)
1450 {
1451         struct inet_sock *inet = inet_sk(sk);
1452         /*
1453          *      1003.1g - break association.
1454          */
1455 
1456         sk->sk_state = TCP_CLOSE;
1457         inet->inet_daddr = 0;
1458         inet->inet_dport = 0;
1459         sock_rps_reset_rxhash(sk);
1460         sk->sk_bound_dev_if = 0;
1461         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1462                 inet_reset_saddr(sk);
1463 
1464         if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1465                 sk->sk_prot->unhash(sk);
1466                 inet->inet_sport = 0;
1467         }
1468         sk_dst_reset(sk);
1469         return 0;
1470 }
1471 EXPORT_SYMBOL(udp_disconnect);
1472 
1473 void udp_lib_unhash(struct sock *sk)
1474 {
1475         if (sk_hashed(sk)) {
1476                 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1477                 struct udp_hslot *hslot, *hslot2;
1478 
1479                 hslot  = udp_hashslot(udptable, sock_net(sk),
1480                                       udp_sk(sk)->udp_port_hash);
1481                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1482 
1483                 spin_lock_bh(&hslot->lock);
1484                 if (rcu_access_pointer(sk->sk_reuseport_cb))
1485                         reuseport_detach_sock(sk);
1486                 if (sk_nulls_del_node_init_rcu(sk)) {
1487                         hslot->count--;
1488                         inet_sk(sk)->inet_num = 0;
1489                         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1490 
1491                         spin_lock(&hslot2->lock);
1492                         hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1493                         hslot2->count--;
1494                         spin_unlock(&hslot2->lock);
1495                 }
1496                 spin_unlock_bh(&hslot->lock);
1497         }
1498 }
1499 EXPORT_SYMBOL(udp_lib_unhash);
1500 
1501 /*
1502  * inet_rcv_saddr was changed, we must rehash secondary hash
1503  */
1504 void udp_lib_rehash(struct sock *sk, u16 newhash)
1505 {
1506         if (sk_hashed(sk)) {
1507                 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1508                 struct udp_hslot *hslot, *hslot2, *nhslot2;
1509 
1510                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1511                 nhslot2 = udp_hashslot2(udptable, newhash);
1512                 udp_sk(sk)->udp_portaddr_hash = newhash;
1513 
1514                 if (hslot2 != nhslot2 ||
1515                     rcu_access_pointer(sk->sk_reuseport_cb)) {
1516                         hslot = udp_hashslot(udptable, sock_net(sk),
1517                                              udp_sk(sk)->udp_port_hash);
1518                         /* we must lock primary chain too */
1519                         spin_lock_bh(&hslot->lock);
1520                         if (rcu_access_pointer(sk->sk_reuseport_cb))
1521                                 reuseport_detach_sock(sk);
1522 
1523                         if (hslot2 != nhslot2) {
1524                                 spin_lock(&hslot2->lock);
1525                                 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1526                                 hslot2->count--;
1527                                 spin_unlock(&hslot2->lock);
1528 
1529                                 spin_lock(&nhslot2->lock);
1530                                 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1531                                                          &nhslot2->head);
1532                                 nhslot2->count++;
1533                                 spin_unlock(&nhslot2->lock);
1534                         }
1535 
1536                         spin_unlock_bh(&hslot->lock);
1537                 }
1538         }
1539 }
1540 EXPORT_SYMBOL(udp_lib_rehash);
1541 
1542 static void udp_v4_rehash(struct sock *sk)
1543 {
1544         u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1545                                           inet_sk(sk)->inet_rcv_saddr,
1546                                           inet_sk(sk)->inet_num);
1547         udp_lib_rehash(sk, new_hash);
1548 }
1549 
1550 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1551 {
1552         int rc;
1553 
1554         if (inet_sk(sk)->inet_daddr) {
1555                 sock_rps_save_rxhash(sk, skb);
1556                 sk_mark_napi_id(sk, skb);
1557                 sk_incoming_cpu_update(sk);
1558         }
1559 
1560         rc = sock_queue_rcv_skb(sk, skb);
1561         if (rc < 0) {
1562                 int is_udplite = IS_UDPLITE(sk);
1563 
1564                 /* Note that an ENOMEM error is charged twice */
1565                 if (rc == -ENOMEM)
1566                         UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1567                                          is_udplite);
1568                 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1569                 kfree_skb(skb);
1570                 trace_udp_fail_queue_rcv_skb(rc, sk);
1571                 return -1;
1572         }
1573 
1574         return 0;
1575 
1576 }
1577 
1578 static struct static_key udp_encap_needed __read_mostly;
1579 void udp_encap_enable(void)
1580 {
1581         if (!static_key_enabled(&udp_encap_needed))
1582                 static_key_slow_inc(&udp_encap_needed);
1583 }
1584 EXPORT_SYMBOL(udp_encap_enable);
1585 
1586 /* returns:
1587  *  -1: error
1588  *   0: success
1589  *  >0: "udp encap" protocol resubmission
1590  *
1591  * Note that in the success and error cases, the skb is assumed to
1592  * have either been requeued or freed.
1593  */
1594 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1595 {
1596         struct udp_sock *up = udp_sk(sk);
1597         int rc;
1598         int is_udplite = IS_UDPLITE(sk);
1599 
1600         /*
1601          *      Charge it to the socket, dropping if the queue is full.
1602          */
1603         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1604                 goto drop;
1605         nf_reset(skb);
1606 
1607         if (static_key_false(&udp_encap_needed) && up->encap_type) {
1608                 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1609 
1610                 /*
1611                  * This is an encapsulation socket so pass the skb to
1612                  * the socket's udp_encap_rcv() hook. Otherwise, just
1613                  * fall through and pass this up the UDP socket.
1614                  * up->encap_rcv() returns the following value:
1615                  * =0 if skb was successfully passed to the encap
1616                  *    handler or was discarded by it.
1617                  * >0 if skb should be passed on to UDP.
1618                  * <0 if skb should be resubmitted as proto -N
1619                  */
1620 
1621                 /* if we're overly short, let UDP handle it */
1622                 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1623                 if (encap_rcv) {
1624                         int ret;
1625 
1626                         /* Verify checksum before giving to encap */
1627                         if (udp_lib_checksum_complete(skb))
1628                                 goto csum_error;
1629 
1630                         ret = encap_rcv(sk, skb);
1631                         if (ret <= 0) {
1632                                 UDP_INC_STATS_BH(sock_net(sk),
1633                                                  UDP_MIB_INDATAGRAMS,
1634                                                  is_udplite);
1635                                 return -ret;
1636                         }
1637                 }
1638 
1639                 /* FALLTHROUGH -- it's a UDP Packet */
1640         }
1641 
1642         /*
1643          *      UDP-Lite specific tests, ignored on UDP sockets
1644          */
1645         if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1646 
1647                 /*
1648                  * MIB statistics other than incrementing the error count are
1649                  * disabled for the following two types of errors: these depend
1650                  * on the application settings, not on the functioning of the
1651                  * protocol stack as such.
1652                  *
1653                  * RFC 3828 here recommends (sec 3.3): "There should also be a
1654                  * way ... to ... at least let the receiving application block
1655                  * delivery of packets with coverage values less than a value
1656                  * provided by the application."
1657                  */
1658                 if (up->pcrlen == 0) {          /* full coverage was set  */
1659                         net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1660                                             UDP_SKB_CB(skb)->cscov, skb->len);
1661                         goto drop;
1662                 }
1663                 /* The next case involves violating the min. coverage requested
1664                  * by the receiver. This is subtle: if receiver wants x and x is
1665                  * greater than the buffersize/MTU then receiver will complain
1666                  * that it wants x while sender emits packets of smaller size y.
1667                  * Therefore the above ...()->partial_cov statement is essential.
1668                  */
1669                 if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1670                         net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1671                                             UDP_SKB_CB(skb)->cscov, up->pcrlen);
1672                         goto drop;
1673                 }
1674         }
1675 
1676         if (rcu_access_pointer(sk->sk_filter) &&
1677             udp_lib_checksum_complete(skb))
1678                 goto csum_error;
1679 
1680         if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1681                 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1682                                  is_udplite);
1683                 goto drop;
1684         }
1685 
1686         rc = 0;
1687 
1688         ipv4_pktinfo_prepare(sk, skb);
1689         bh_lock_sock(sk);
1690         if (!sock_owned_by_user(sk))
1691                 rc = __udp_queue_rcv_skb(sk, skb);
1692         else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1693                 bh_unlock_sock(sk);
1694                 goto drop;
1695         }
1696         bh_unlock_sock(sk);
1697 
1698         return rc;
1699 
1700 csum_error:
1701         UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1702 drop:
1703         UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1704         atomic_inc(&sk->sk_drops);
1705         kfree_skb(skb);
1706         return -1;
1707 }
1708 
1709 static void flush_stack(struct sock **stack, unsigned int count,
1710                         struct sk_buff *skb, unsigned int final)
1711 {
1712         unsigned int i;
1713         struct sk_buff *skb1 = NULL;
1714         struct sock *sk;
1715 
1716         for (i = 0; i < count; i++) {
1717                 sk = stack[i];
1718                 if (likely(!skb1))
1719                         skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1720 
1721                 if (!skb1) {
1722                         atomic_inc(&sk->sk_drops);
1723                         UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1724                                          IS_UDPLITE(sk));
1725                         UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1726                                          IS_UDPLITE(sk));
1727                 }
1728 
1729                 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1730                         skb1 = NULL;
1731 
1732                 sock_put(sk);
1733         }
1734         if (unlikely(skb1))
1735                 kfree_skb(skb1);
1736 }
1737 
1738 /* For TCP sockets, sk_rx_dst is protected by socket lock
1739  * For UDP, we use xchg() to guard against concurrent changes.
1740  */
1741 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1742 {
1743         struct dst_entry *old;
1744 
1745         dst_hold(dst);
1746         old = xchg(&sk->sk_rx_dst, dst);
1747         dst_release(old);
1748 }
1749 
1750 /*
1751  *      Multicasts and broadcasts go to each listener.
1752  *
1753  *      Note: called only from the BH handler context.
1754  */
1755 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1756                                     struct udphdr  *uh,
1757                                     __be32 saddr, __be32 daddr,
1758                                     struct udp_table *udptable,
1759                                     int proto)
1760 {
1761         struct sock *sk, *stack[256 / sizeof(struct sock *)];
1762         struct hlist_nulls_node *node;
1763         unsigned short hnum = ntohs(uh->dest);
1764         struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1765         int dif = skb->dev->ifindex;
1766         unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
1767         unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1768         bool inner_flushed = false;
1769 
1770         if (use_hash2) {
1771                 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1772                             udp_table.mask;
1773                 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1774 start_lookup:
1775                 hslot = &udp_table.hash2[hash2];
1776                 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1777         }
1778 
1779         spin_lock(&hslot->lock);
1780         sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
1781                 if (__udp_is_mcast_sock(net, sk,
1782                                         uh->dest, daddr,
1783                                         uh->source, saddr,
1784                                         dif, hnum)) {
1785                         if (unlikely(count == ARRAY_SIZE(stack))) {
1786                                 flush_stack(stack, count, skb, ~0);
1787                                 inner_flushed = true;
1788                                 count = 0;
1789                         }
1790                         stack[count++] = sk;
1791                         sock_hold(sk);
1792                 }
1793         }
1794 
1795         spin_unlock(&hslot->lock);
1796 
1797         /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1798         if (use_hash2 && hash2 != hash2_any) {
1799                 hash2 = hash2_any;
1800                 goto start_lookup;
1801         }
1802 
1803         /*
1804          * do the slow work with no lock held
1805          */
1806         if (count) {
1807                 flush_stack(stack, count, skb, count - 1);
1808         } else {
1809                 if (!inner_flushed)
1810                         UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
1811                                          proto == IPPROTO_UDPLITE);
1812                 consume_skb(skb);
1813         }
1814         return 0;
1815 }
1816 
1817 /* Initialize UDP checksum. If exited with zero value (success),
1818  * CHECKSUM_UNNECESSARY means, that no more checks are required.
1819  * Otherwise, csum completion requires chacksumming packet body,
1820  * including udp header and folding it to skb->csum.
1821  */
1822 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1823                                  int proto)
1824 {
1825         int err;
1826 
1827         UDP_SKB_CB(skb)->partial_cov = 0;
1828         UDP_SKB_CB(skb)->cscov = skb->len;
1829 
1830         if (proto == IPPROTO_UDPLITE) {
1831                 err = udplite_checksum_init(skb, uh);
1832                 if (err)
1833                         return err;
1834         }
1835 
1836         return skb_checksum_init_zero_check(skb, proto, uh->check,
1837                                             inet_compute_pseudo);
1838 }
1839 
1840 /*
1841  *      All we need to do is get the socket, and then do a checksum.
1842  */
1843 
1844 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1845                    int proto)
1846 {
1847         struct sock *sk;
1848         struct udphdr *uh;
1849         unsigned short ulen;
1850         struct rtable *rt = skb_rtable(skb);
1851         __be32 saddr, daddr;
1852         struct net *net = dev_net(skb->dev);
1853 
1854         /*
1855          *  Validate the packet.
1856          */
1857         if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1858                 goto drop;              /* No space for header. */
1859 
1860         uh   = udp_hdr(skb);
1861         ulen = ntohs(uh->len);
1862         saddr = ip_hdr(skb)->saddr;
1863         daddr = ip_hdr(skb)->daddr;
1864 
1865         if (ulen > skb->len)
1866                 goto short_packet;
1867 
1868         if (proto == IPPROTO_UDP) {
1869                 /* UDP validates ulen. */
1870                 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1871                         goto short_packet;
1872                 uh = udp_hdr(skb);
1873         }
1874 
1875         if (udp4_csum_init(skb, uh, proto))
1876                 goto csum_error;
1877 
1878         sk = skb_steal_sock(skb);
1879         if (sk) {
1880                 struct dst_entry *dst = skb_dst(skb);
1881                 int ret;
1882 
1883                 if (unlikely(sk->sk_rx_dst != dst))
1884                         udp_sk_rx_dst_set(sk, dst);
1885 
1886                 ret = udp_queue_rcv_skb(sk, skb);
1887                 sock_put(sk);
1888                 /* a return value > 0 means to resubmit the input, but
1889                  * it wants the return to be -protocol, or 0
1890                  */
1891                 if (ret > 0)
1892                         return -ret;
1893                 return 0;
1894         }
1895 
1896         if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1897                 return __udp4_lib_mcast_deliver(net, skb, uh,
1898                                                 saddr, daddr, udptable, proto);
1899 
1900         sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1901         if (sk) {
1902                 int ret;
1903 
1904                 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1905                         skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1906                                                  inet_compute_pseudo);
1907 
1908                 ret = udp_queue_rcv_skb(sk, skb);
1909                 sock_put(sk);
1910 
1911                 /* a return value > 0 means to resubmit the input, but
1912                  * it wants the return to be -protocol, or 0
1913                  */
1914                 if (ret > 0)
1915                         return -ret;
1916                 return 0;
1917         }
1918 
1919         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1920                 goto drop;
1921         nf_reset(skb);
1922 
1923         /* No socket. Drop packet silently, if checksum is wrong */
1924         if (udp_lib_checksum_complete(skb))
1925                 goto csum_error;
1926 
1927         UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1928         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1929 
1930         /*
1931          * Hmm.  We got an UDP packet to a port to which we
1932          * don't wanna listen.  Ignore it.
1933          */
1934         kfree_skb(skb);
1935         return 0;
1936 
1937 short_packet:
1938         net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1939                             proto == IPPROTO_UDPLITE ? "Lite" : "",
1940                             &saddr, ntohs(uh->source),
1941                             ulen, skb->len,
1942                             &daddr, ntohs(uh->dest));
1943         goto drop;
1944 
1945 csum_error:
1946         /*
1947          * RFC1122: OK.  Discards the bad packet silently (as far as
1948          * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1949          */
1950         net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1951                             proto == IPPROTO_UDPLITE ? "Lite" : "",
1952                             &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1953                             ulen);
1954         UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1955 drop:
1956         UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1957         kfree_skb(skb);
1958         return 0;
1959 }
1960 
1961 /* We can only early demux multicast if there is a single matching socket.
1962  * If more than one socket found returns NULL
1963  */
1964 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1965                                                   __be16 loc_port, __be32 loc_addr,
1966                                                   __be16 rmt_port, __be32 rmt_addr,
1967                                                   int dif)
1968 {
1969         struct sock *sk, *result;
1970         struct hlist_nulls_node *node;
1971         unsigned short hnum = ntohs(loc_port);
1972         unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1973         struct udp_hslot *hslot = &udp_table.hash[slot];
1974 
1975         /* Do not bother scanning a too big list */
1976         if (hslot->count > 10)
1977                 return NULL;
1978 
1979         rcu_read_lock();
1980 begin:
1981         count = 0;
1982         result = NULL;
1983         sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1984                 if (__udp_is_mcast_sock(net, sk,
1985                                         loc_port, loc_addr,
1986                                         rmt_port, rmt_addr,
1987                                         dif, hnum)) {
1988                         result = sk;
1989                         ++count;
1990                 }
1991         }
1992         /*
1993          * if the nulls value we got at the end of this lookup is
1994          * not the expected one, we must restart lookup.
1995          * We probably met an item that was moved to another chain.
1996          */
1997         if (get_nulls_value(node) != slot)
1998                 goto begin;
1999 
2000         if (result) {
2001                 if (count != 1 ||
2002                     unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
2003                         result = NULL;
2004                 else if (unlikely(!__udp_is_mcast_sock(net, result,
2005                                                        loc_port, loc_addr,
2006                                                        rmt_port, rmt_addr,
2007                                                        dif, hnum))) {
2008                         sock_put(result);
2009                         result = NULL;
2010                 }
2011         }
2012         rcu_read_unlock();
2013         return result;
2014 }
2015 
2016 /* For unicast we should only early demux connected sockets or we can
2017  * break forwarding setups.  The chains here can be long so only check
2018  * if the first socket is an exact match and if not move on.
2019  */
2020 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2021                                             __be16 loc_port, __be32 loc_addr,
2022                                             __be16 rmt_port, __be32 rmt_addr,
2023                                             int dif)
2024 {
2025         struct sock *sk, *result;
2026         struct hlist_nulls_node *node;
2027         unsigned short hnum = ntohs(loc_port);
2028         unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2029         unsigned int slot2 = hash2 & udp_table.mask;
2030         struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2031         INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2032         const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2033 
2034         rcu_read_lock();
2035         result = NULL;
2036         udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
2037                 if (INET_MATCH(sk, net, acookie,
2038                                rmt_addr, loc_addr, ports, dif))
2039                         result = sk;
2040                 /* Only check first socket in chain */
2041                 break;
2042         }
2043 
2044         if (result) {
2045                 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
2046                         result = NULL;
2047                 else if (unlikely(!INET_MATCH(sk, net, acookie,
2048                                               rmt_addr, loc_addr,
2049                                               ports, dif))) {
2050                         sock_put(result);
2051                         result = NULL;
2052                 }
2053         }
2054         rcu_read_unlock();
2055         return result;
2056 }
2057 
2058 void udp_v4_early_demux(struct sk_buff *skb)
2059 {
2060         struct net *net = dev_net(skb->dev);
2061         const struct iphdr *iph;
2062         const struct udphdr *uh;
2063         struct sock *sk;
2064         struct dst_entry *dst;
2065         int dif = skb->dev->ifindex;
2066         int ours;
2067 
2068         /* validate the packet */
2069         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2070                 return;
2071 
2072         iph = ip_hdr(skb);
2073         uh = udp_hdr(skb);
2074 
2075         if (skb->pkt_type == PACKET_BROADCAST ||
2076             skb->pkt_type == PACKET_MULTICAST) {
2077                 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2078 
2079                 if (!in_dev)
2080                         return;
2081 
2082                 /* we are supposed to accept bcast packets */
2083                 if (skb->pkt_type == PACKET_MULTICAST) {
2084                         ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2085                                                iph->protocol);
2086                         if (!ours)
2087                                 return;
2088                 }
2089 
2090                 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2091                                                    uh->source, iph->saddr, dif);
2092         } else if (skb->pkt_type == PACKET_HOST) {
2093                 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2094                                              uh->source, iph->saddr, dif);
2095         } else {
2096                 return;
2097         }
2098 
2099         if (!sk)
2100                 return;
2101 
2102         skb->sk = sk;
2103         skb->destructor = sock_efree;
2104         dst = READ_ONCE(sk->sk_rx_dst);
2105 
2106         if (dst)
2107                 dst = dst_check(dst, 0);
2108         if (dst) {
2109                 /* DST_NOCACHE can not be used without taking a reference */
2110                 if (dst->flags & DST_NOCACHE) {
2111                         if (likely(atomic_inc_not_zero(&dst->__refcnt)))
2112                                 skb_dst_set(skb, dst);
2113                 } else {
2114                         skb_dst_set_noref(skb, dst);
2115                 }
2116         }
2117 }
2118 
2119 int udp_rcv(struct sk_buff *skb)
2120 {
2121         return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2122 }
2123 
2124 void udp_destroy_sock(struct sock *sk)
2125 {
2126         struct udp_sock *up = udp_sk(sk);
2127         bool slow = lock_sock_fast(sk);
2128         udp_flush_pending_frames(sk);
2129         unlock_sock_fast(sk, slow);
2130         if (static_key_false(&udp_encap_needed) && up->encap_type) {
2131                 void (*encap_destroy)(struct sock *sk);
2132                 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2133                 if (encap_destroy)
2134                         encap_destroy(sk);
2135         }
2136 }
2137 
2138 /*
2139  *      Socket option code for UDP
2140  */
2141 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2142                        char __user *optval, unsigned int optlen,
2143                        int (*push_pending_frames)(struct sock *))
2144 {
2145         struct udp_sock *up = udp_sk(sk);
2146         int val, valbool;
2147         int err = 0;
2148         int is_udplite = IS_UDPLITE(sk);
2149 
2150         if (optlen < sizeof(int))
2151                 return -EINVAL;
2152 
2153         if (get_user(val, (int __user *)optval))
2154                 return -EFAULT;
2155 
2156         valbool = val ? 1 : 0;
2157 
2158         switch (optname) {
2159         case UDP_CORK:
2160                 if (val != 0) {
2161                         up->corkflag = 1;
2162                 } else {
2163                         up->corkflag = 0;
2164                         lock_sock(sk);
2165                         push_pending_frames(sk);
2166                         release_sock(sk);
2167                 }
2168                 break;
2169 
2170         case UDP_ENCAP:
2171                 switch (val) {
2172                 case 0:
2173                 case UDP_ENCAP_ESPINUDP:
2174                 case UDP_ENCAP_ESPINUDP_NON_IKE:
2175                         up->encap_rcv = xfrm4_udp_encap_rcv;
2176                         /* FALLTHROUGH */
2177                 case UDP_ENCAP_L2TPINUDP:
2178                         up->encap_type = val;
2179                         udp_encap_enable();
2180                         break;
2181                 default:
2182                         err = -ENOPROTOOPT;
2183                         break;
2184                 }
2185                 break;
2186 
2187         case UDP_NO_CHECK6_TX:
2188                 up->no_check6_tx = valbool;
2189                 break;
2190 
2191         case UDP_NO_CHECK6_RX:
2192                 up->no_check6_rx = valbool;
2193                 break;
2194 
2195         /*
2196          *      UDP-Lite's partial checksum coverage (RFC 3828).
2197          */
2198         /* The sender sets actual checksum coverage length via this option.
2199          * The case coverage > packet length is handled by send module. */
2200         case UDPLITE_SEND_CSCOV:
2201                 if (!is_udplite)         /* Disable the option on UDP sockets */
2202                         return -ENOPROTOOPT;
2203                 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2204                         val = 8;
2205                 else if (val > USHRT_MAX)
2206                         val = USHRT_MAX;
2207                 up->pcslen = val;
2208                 up->pcflag |= UDPLITE_SEND_CC;
2209                 break;
2210 
2211         /* The receiver specifies a minimum checksum coverage value. To make
2212          * sense, this should be set to at least 8 (as done below). If zero is
2213          * used, this again means full checksum coverage.                     */
2214         case UDPLITE_RECV_CSCOV:
2215                 if (!is_udplite)         /* Disable the option on UDP sockets */
2216                         return -ENOPROTOOPT;
2217                 if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2218                         val = 8;
2219                 else if (val > USHRT_MAX)
2220                         val = USHRT_MAX;
2221                 up->pcrlen = val;
2222                 up->pcflag |= UDPLITE_RECV_CC;
2223                 break;
2224 
2225         default:
2226                 err = -ENOPROTOOPT;
2227                 break;
2228         }
2229 
2230         return err;
2231 }
2232 EXPORT_SYMBOL(udp_lib_setsockopt);
2233 
2234 int udp_setsockopt(struct sock *sk, int level, int optname,
2235                    char __user *optval, unsigned int optlen)
2236 {
2237         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2238                 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2239                                           udp_push_pending_frames);
2240         return ip_setsockopt(sk, level, optname, optval, optlen);
2241 }
2242 
2243 #ifdef CONFIG_COMPAT
2244 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2245                           char __user *optval, unsigned int optlen)
2246 {
2247         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2248                 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2249                                           udp_push_pending_frames);
2250         return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2251 }
2252 #endif
2253 
2254 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2255                        char __user *optval, int __user *optlen)
2256 {
2257         struct udp_sock *up = udp_sk(sk);
2258         int val, len;
2259 
2260         if (get_user(len, optlen))
2261                 return -EFAULT;
2262 
2263         len = min_t(unsigned int, len, sizeof(int));
2264 
2265         if (len < 0)
2266                 return -EINVAL;
2267 
2268         switch (optname) {
2269         case UDP_CORK:
2270                 val = up->corkflag;
2271                 break;
2272 
2273         case UDP_ENCAP:
2274                 val = up->encap_type;
2275                 break;
2276 
2277         case UDP_NO_CHECK6_TX:
2278                 val = up->no_check6_tx;
2279                 break;
2280 
2281         case UDP_NO_CHECK6_RX:
2282                 val = up->no_check6_rx;
2283                 break;
2284 
2285         /* The following two cannot be changed on UDP sockets, the return is
2286          * always 0 (which corresponds to the full checksum coverage of UDP). */
2287         case UDPLITE_SEND_CSCOV:
2288                 val = up->pcslen;
2289                 break;
2290 
2291         case UDPLITE_RECV_CSCOV:
2292                 val = up->pcrlen;
2293                 break;
2294 
2295         default:
2296                 return -ENOPROTOOPT;
2297         }
2298 
2299         if (put_user(len, optlen))
2300                 return -EFAULT;
2301         if (copy_to_user(optval, &val, len))
2302                 return -EFAULT;
2303         return 0;
2304 }
2305 EXPORT_SYMBOL(udp_lib_getsockopt);
2306 
2307 int udp_getsockopt(struct sock *sk, int level, int optname,
2308                    char __user *optval, int __user *optlen)
2309 {
2310         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2311                 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2312         return ip_getsockopt(sk, level, optname, optval, optlen);
2313 }
2314 
2315 #ifdef CONFIG_COMPAT
2316 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2317                                  char __user *optval, int __user *optlen)
2318 {
2319         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2320                 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2321         return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2322 }
2323 #endif
2324 /**
2325  *      udp_poll - wait for a UDP event.
2326  *      @file - file struct
2327  *      @sock - socket
2328  *      @wait - poll table
2329  *
2330  *      This is same as datagram poll, except for the special case of
2331  *      blocking sockets. If application is using a blocking fd
2332  *      and a packet with checksum error is in the queue;
2333  *      then it could get return from select indicating data available
2334  *      but then block when reading it. Add special case code
2335  *      to work around these arguably broken applications.
2336  */
2337 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2338 {
2339         unsigned int mask = datagram_poll(file, sock, wait);
2340         struct sock *sk = sock->sk;
2341 
2342         sock_rps_record_flow(sk);
2343 
2344         /* Check for false positives due to checksum errors */
2345         if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2346             !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2347                 mask &= ~(POLLIN | POLLRDNORM);
2348 
2349         return mask;
2350 
2351 }
2352 EXPORT_SYMBOL(udp_poll);
2353 
2354 struct proto udp_prot = {
2355         .name              = "UDP",
2356         .owner             = THIS_MODULE,
2357         .close             = udp_lib_close,
2358         .connect           = ip4_datagram_connect,
2359         .disconnect        = udp_disconnect,
2360         .ioctl             = udp_ioctl,
2361         .destroy           = udp_destroy_sock,
2362         .setsockopt        = udp_setsockopt,
2363         .getsockopt        = udp_getsockopt,
2364         .sendmsg           = udp_sendmsg,
2365         .recvmsg           = udp_recvmsg,
2366         .sendpage          = udp_sendpage,
2367         .backlog_rcv       = __udp_queue_rcv_skb,
2368         .release_cb        = ip4_datagram_release_cb,
2369         .hash              = udp_lib_hash,
2370         .unhash            = udp_lib_unhash,
2371         .rehash            = udp_v4_rehash,
2372         .get_port          = udp_v4_get_port,
2373         .memory_allocated  = &udp_memory_allocated,
2374         .sysctl_mem        = sysctl_udp_mem,
2375         .sysctl_wmem       = &sysctl_udp_wmem_min,
2376         .sysctl_rmem       = &sysctl_udp_rmem_min,
2377         .obj_size          = sizeof(struct udp_sock),
2378         .slab_flags        = SLAB_DESTROY_BY_RCU,
2379         .h.udp_table       = &udp_table,
2380 #ifdef CONFIG_COMPAT
2381         .compat_setsockopt = compat_udp_setsockopt,
2382         .compat_getsockopt = compat_udp_getsockopt,
2383 #endif
2384         .clear_sk          = sk_prot_clear_portaddr_nulls,
2385 };
2386 EXPORT_SYMBOL(udp_prot);
2387 
2388 /* ------------------------------------------------------------------------ */
2389 #ifdef CONFIG_PROC_FS
2390 
2391 static struct sock *udp_get_first(struct seq_file *seq, int start)
2392 {
2393         struct sock *sk;
2394         struct udp_iter_state *state = seq->private;
2395         struct net *net = seq_file_net(seq);
2396 
2397         for (state->bucket = start; state->bucket <= state->udp_table->mask;
2398              ++state->bucket) {
2399                 struct hlist_nulls_node *node;
2400                 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2401 
2402                 if (hlist_nulls_empty(&hslot->head))
2403                         continue;
2404 
2405                 spin_lock_bh(&hslot->lock);
2406                 sk_nulls_for_each(sk, node, &hslot->head) {
2407                         if (!net_eq(sock_net(sk), net))
2408                                 continue;
2409                         if (sk->sk_family == state->family)
2410                                 goto found;
2411                 }
2412                 spin_unlock_bh(&hslot->lock);
2413         }
2414         sk = NULL;
2415 found:
2416         return sk;
2417 }
2418 
2419 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2420 {
2421         struct udp_iter_state *state = seq->private;
2422         struct net *net = seq_file_net(seq);
2423 
2424         do {
2425                 sk = sk_nulls_next(sk);
2426         } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2427 
2428         if (!sk) {
2429                 if (state->bucket <= state->udp_table->mask)
2430                         spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2431                 return udp_get_first(seq, state->bucket + 1);
2432         }
2433         return sk;
2434 }
2435 
2436 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2437 {
2438         struct sock *sk = udp_get_first(seq, 0);
2439 
2440         if (sk)
2441                 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2442                         --pos;
2443         return pos ? NULL : sk;
2444 }
2445 
2446 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2447 {
2448         struct udp_iter_state *state = seq->private;
2449         state->bucket = MAX_UDP_PORTS;
2450 
2451         return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2452 }
2453 
2454 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2455 {
2456         struct sock *sk;
2457 
2458         if (v == SEQ_START_TOKEN)
2459                 sk = udp_get_idx(seq, 0);
2460         else
2461                 sk = udp_get_next(seq, v);
2462 
2463         ++*pos;
2464         return sk;
2465 }
2466 
2467 static void udp_seq_stop(struct seq_file *seq, void *v)
2468 {
2469         struct udp_iter_state *state = seq->private;
2470 
2471         if (state->bucket <= state->udp_table->mask)
2472                 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2473 }
2474 
2475 int udp_seq_open(struct inode *inode, struct file *file)
2476 {
2477         struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2478         struct udp_iter_state *s;
2479         int err;
2480 
2481         err = seq_open_net(inode, file, &afinfo->seq_ops,
2482                            sizeof(struct udp_iter_state));
2483         if (err < 0)
2484                 return err;
2485 
2486         s = ((struct seq_file *)file->private_data)->private;
2487         s->family               = afinfo->family;
2488         s->udp_table            = afinfo->udp_table;
2489         return err;
2490 }
2491 EXPORT_SYMBOL(udp_seq_open);
2492 
2493 /* ------------------------------------------------------------------------ */
2494 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2495 {
2496         struct proc_dir_entry *p;
2497         int rc = 0;
2498 
2499         afinfo->seq_ops.start           = udp_seq_start;
2500         afinfo->seq_ops.next            = udp_seq_next;
2501         afinfo->seq_ops.stop            = udp_seq_stop;
2502 
2503         p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2504                              afinfo->seq_fops, afinfo);
2505         if (!p)
2506                 rc = -ENOMEM;
2507         return rc;
2508 }
2509 EXPORT_SYMBOL(udp_proc_register);
2510 
2511 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2512 {
2513         remove_proc_entry(afinfo->name, net->proc_net);
2514 }
2515 EXPORT_SYMBOL(udp_proc_unregister);
2516 
2517 /* ------------------------------------------------------------------------ */
2518 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2519                 int bucket)
2520 {
2521         struct inet_sock *inet = inet_sk(sp);
2522         __be32 dest = inet->inet_daddr;
2523         __be32 src  = inet->inet_rcv_saddr;
2524         __u16 destp       = ntohs(inet->inet_dport);
2525         __u16 srcp        = ntohs(inet->inet_sport);
2526 
2527         seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2528                 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2529                 bucket, src, srcp, dest, destp, sp->sk_state,
2530                 sk_wmem_alloc_get(sp),
2531                 sk_rmem_alloc_get(sp),
2532                 0, 0L, 0,
2533                 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2534                 0, sock_i_ino(sp),
2535                 atomic_read(&sp->sk_refcnt), sp,
2536                 atomic_read(&sp->sk_drops));
2537 }
2538 
2539 int udp4_seq_show(struct seq_file *seq, void *v)
2540 {
2541         seq_setwidth(seq, 127);
2542         if (v == SEQ_START_TOKEN)
2543                 seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2544                            "rx_queue tr tm->when retrnsmt   uid  timeout "
2545                            "inode ref pointer drops");
2546         else {
2547                 struct udp_iter_state *state = seq->private;
2548 
2549                 udp4_format_sock(v, seq, state->bucket);
2550         }
2551         seq_pad(seq, '\n');
2552         return 0;
2553 }
2554 
2555 static const struct file_operations udp_afinfo_seq_fops = {
2556         .owner    = THIS_MODULE,
2557         .open     = udp_seq_open,
2558         .read     = seq_read,
2559         .llseek   = seq_lseek,
2560         .release  = seq_release_net
2561 };
2562 
2563 /* ------------------------------------------------------------------------ */
2564 static struct udp_seq_afinfo udp4_seq_afinfo = {
2565         .name           = "udp",
2566         .family         = AF_INET,
2567         .udp_table      = &udp_table,
2568         .seq_fops       = &udp_afinfo_seq_fops,
2569         .seq_ops        = {
2570                 .show           = udp4_seq_show,
2571         },
2572 };
2573 
2574 static int __net_init udp4_proc_init_net(struct net *net)
2575 {
2576         return udp_proc_register(net, &udp4_seq_afinfo);
2577 }
2578 
2579 static void __net_exit udp4_proc_exit_net(struct net *net)
2580 {
2581         udp_proc_unregister(net, &udp4_seq_afinfo);
2582 }
2583 
2584 static struct pernet_operations udp4_net_ops = {
2585         .init = udp4_proc_init_net,
2586         .exit = udp4_proc_exit_net,
2587 };
2588 
2589 int __init udp4_proc_init(void)
2590 {
2591         return register_pernet_subsys(&udp4_net_ops);
2592 }
2593 
2594 void udp4_proc_exit(void)
2595 {
2596         unregister_pernet_subsys(&udp4_net_ops);
2597 }
2598 #endif /* CONFIG_PROC_FS */
2599 
2600 static __initdata unsigned long uhash_entries;
2601 static int __init set_uhash_entries(char *str)
2602 {
2603         ssize_t ret;
2604 
2605         if (!str)
2606                 return 0;
2607 
2608         ret = kstrtoul(str, 0, &uhash_entries);
2609         if (ret)
2610                 return 0;
2611 
2612         if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2613                 uhash_entries = UDP_HTABLE_SIZE_MIN;
2614         return 1;
2615 }
2616 __setup("uhash_entries=", set_uhash_entries);
2617 
2618 void __init udp_table_init(struct udp_table *table, const char *name)
2619 {
2620         unsigned int i;
2621 
2622         table->hash = alloc_large_system_hash(name,
2623                                               2 * sizeof(struct udp_hslot),
2624                                               uhash_entries,
2625                                               21, /* one slot per 2 MB */
2626                                               0,
2627                                               &table->log,
2628                                               &table->mask,
2629                                               UDP_HTABLE_SIZE_MIN,
2630                                               64 * 1024);
2631 
2632         table->hash2 = table->hash + (table->mask + 1);
2633         for (i = 0; i <= table->mask; i++) {
2634                 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2635                 table->hash[i].count = 0;
2636                 spin_lock_init(&table->hash[i].lock);
2637         }
2638         for (i = 0; i <= table->mask; i++) {
2639                 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2640                 table->hash2[i].count = 0;
2641                 spin_lock_init(&table->hash2[i].lock);
2642         }
2643 }
2644 
2645 u32 udp_flow_hashrnd(void)
2646 {
2647         static u32 hashrnd __read_mostly;
2648 
2649         net_get_random_once(&hashrnd, sizeof(hashrnd));
2650 
2651         return hashrnd;
2652 }
2653 EXPORT_SYMBOL(udp_flow_hashrnd);
2654 
2655 void __init udp_init(void)
2656 {
2657         unsigned long limit;
2658 
2659         udp_table_init(&udp_table, "UDP");
2660         limit = nr_free_buffer_pages() / 8;
2661         limit = max(limit, 128UL);
2662         sysctl_udp_mem[0] = limit / 4 * 3;
2663         sysctl_udp_mem[1] = limit;
2664         sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2665 
2666         sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2667         sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2668 }
2669 

~ [ 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